Commit | Line | Data |
---|---|---|
1bc38b8f | 1 | // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) |
6061a3d6 | 2 | |
1b76c13e WN |
3 | /* |
4 | * Common eBPF ELF object loading operations. | |
5 | * | |
6 | * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org> | |
7 | * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com> | |
8 | * Copyright (C) 2015 Huawei Inc. | |
f367540c | 9 | * Copyright (C) 2017 Nicira, Inc. |
d859900c | 10 | * Copyright (C) 2019 Isovalent, Inc. |
1b76c13e WN |
11 | */ |
12 | ||
b4269954 | 13 | #ifndef _GNU_SOURCE |
531b014e | 14 | #define _GNU_SOURCE |
b4269954 | 15 | #endif |
1b76c13e | 16 | #include <stdlib.h> |
b3f59d66 WN |
17 | #include <stdio.h> |
18 | #include <stdarg.h> | |
f367540c | 19 | #include <libgen.h> |
34090915 | 20 | #include <inttypes.h> |
8ab9da57 | 21 | #include <limits.h> |
b3f59d66 | 22 | #include <string.h> |
1b76c13e | 23 | #include <unistd.h> |
cdb2f920 | 24 | #include <endian.h> |
1a5e3fb1 WN |
25 | #include <fcntl.h> |
26 | #include <errno.h> | |
113e6b7e | 27 | #include <ctype.h> |
1b76c13e | 28 | #include <asm/unistd.h> |
e28ff1a8 | 29 | #include <linux/err.h> |
cb1e5e96 | 30 | #include <linux/kernel.h> |
1b76c13e | 31 | #include <linux/bpf.h> |
38d5d3b3 | 32 | #include <linux/btf.h> |
47eff617 | 33 | #include <linux/filter.h> |
f367540c | 34 | #include <linux/limits.h> |
438363c0 | 35 | #include <linux/perf_event.h> |
9eea8faf | 36 | #include <linux/bpf_perf_event.h> |
a64af0ef | 37 | #include <linux/ring_buffer.h> |
fb84b822 | 38 | #include <sys/epoll.h> |
63f2f5ee | 39 | #include <sys/ioctl.h> |
fb84b822 | 40 | #include <sys/mman.h> |
f367540c JS |
41 | #include <sys/stat.h> |
42 | #include <sys/types.h> | |
43 | #include <sys/vfs.h> | |
ddc7c304 | 44 | #include <sys/utsname.h> |
dc3a2d25 | 45 | #include <sys/resource.h> |
1a5e3fb1 WN |
46 | #include <libelf.h> |
47 | #include <gelf.h> | |
166750bc | 48 | #include <zlib.h> |
1b76c13e WN |
49 | |
50 | #include "libbpf.h" | |
52d3352e | 51 | #include "bpf.h" |
8a138aed | 52 | #include "btf.h" |
6d41907c | 53 | #include "str_error.h" |
d7c4b398 | 54 | #include "libbpf_internal.h" |
ddc7c304 | 55 | #include "hashmap.h" |
67234743 | 56 | #include "bpf_gen_internal.h" |
c44fd845 | 57 | #include "zip.h" |
b3f59d66 | 58 | |
f367540c JS |
59 | #ifndef BPF_FS_MAGIC |
60 | #define BPF_FS_MAGIC 0xcafe4a11 | |
61 | #endif | |
62 | ||
6b434b61 AN |
63 | #define BPF_FS_DEFAULT_PATH "/sys/fs/bpf" |
64 | ||
9c0f8cbd AN |
65 | #define BPF_INSN_SZ (sizeof(struct bpf_insn)) |
66 | ||
ff466b58 AI |
67 | /* vsprintf() in __base_pr() uses nonliteral format string. It may break |
68 | * compilation if user enables corresponding warning. Disable it explicitly. | |
69 | */ | |
70 | #pragma GCC diagnostic ignored "-Wformat-nonliteral" | |
71 | ||
b3f59d66 WN |
72 | #define __printf(a, b) __attribute__((format(printf, a, b))) |
73 | ||
590a0088 | 74 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj); |
aea28a60 | 75 | static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog); |
f04deb90 | 76 | static int map_set_def_max_entries(struct bpf_map *map); |
590a0088 | 77 | |
ccde5760 DM |
78 | static const char * const attach_type_name[] = { |
79 | [BPF_CGROUP_INET_INGRESS] = "cgroup_inet_ingress", | |
80 | [BPF_CGROUP_INET_EGRESS] = "cgroup_inet_egress", | |
81 | [BPF_CGROUP_INET_SOCK_CREATE] = "cgroup_inet_sock_create", | |
82 | [BPF_CGROUP_INET_SOCK_RELEASE] = "cgroup_inet_sock_release", | |
83 | [BPF_CGROUP_SOCK_OPS] = "cgroup_sock_ops", | |
84 | [BPF_CGROUP_DEVICE] = "cgroup_device", | |
85 | [BPF_CGROUP_INET4_BIND] = "cgroup_inet4_bind", | |
86 | [BPF_CGROUP_INET6_BIND] = "cgroup_inet6_bind", | |
87 | [BPF_CGROUP_INET4_CONNECT] = "cgroup_inet4_connect", | |
88 | [BPF_CGROUP_INET6_CONNECT] = "cgroup_inet6_connect", | |
bf90438c | 89 | [BPF_CGROUP_UNIX_CONNECT] = "cgroup_unix_connect", |
ccde5760 DM |
90 | [BPF_CGROUP_INET4_POST_BIND] = "cgroup_inet4_post_bind", |
91 | [BPF_CGROUP_INET6_POST_BIND] = "cgroup_inet6_post_bind", | |
92 | [BPF_CGROUP_INET4_GETPEERNAME] = "cgroup_inet4_getpeername", | |
93 | [BPF_CGROUP_INET6_GETPEERNAME] = "cgroup_inet6_getpeername", | |
bf90438c | 94 | [BPF_CGROUP_UNIX_GETPEERNAME] = "cgroup_unix_getpeername", |
ccde5760 DM |
95 | [BPF_CGROUP_INET4_GETSOCKNAME] = "cgroup_inet4_getsockname", |
96 | [BPF_CGROUP_INET6_GETSOCKNAME] = "cgroup_inet6_getsockname", | |
bf90438c | 97 | [BPF_CGROUP_UNIX_GETSOCKNAME] = "cgroup_unix_getsockname", |
ccde5760 DM |
98 | [BPF_CGROUP_UDP4_SENDMSG] = "cgroup_udp4_sendmsg", |
99 | [BPF_CGROUP_UDP6_SENDMSG] = "cgroup_udp6_sendmsg", | |
bf90438c | 100 | [BPF_CGROUP_UNIX_SENDMSG] = "cgroup_unix_sendmsg", |
ccde5760 DM |
101 | [BPF_CGROUP_SYSCTL] = "cgroup_sysctl", |
102 | [BPF_CGROUP_UDP4_RECVMSG] = "cgroup_udp4_recvmsg", | |
103 | [BPF_CGROUP_UDP6_RECVMSG] = "cgroup_udp6_recvmsg", | |
bf90438c | 104 | [BPF_CGROUP_UNIX_RECVMSG] = "cgroup_unix_recvmsg", |
ccde5760 DM |
105 | [BPF_CGROUP_GETSOCKOPT] = "cgroup_getsockopt", |
106 | [BPF_CGROUP_SETSOCKOPT] = "cgroup_setsockopt", | |
107 | [BPF_SK_SKB_STREAM_PARSER] = "sk_skb_stream_parser", | |
108 | [BPF_SK_SKB_STREAM_VERDICT] = "sk_skb_stream_verdict", | |
109 | [BPF_SK_SKB_VERDICT] = "sk_skb_verdict", | |
110 | [BPF_SK_MSG_VERDICT] = "sk_msg_verdict", | |
111 | [BPF_LIRC_MODE2] = "lirc_mode2", | |
112 | [BPF_FLOW_DISSECTOR] = "flow_dissector", | |
113 | [BPF_TRACE_RAW_TP] = "trace_raw_tp", | |
114 | [BPF_TRACE_FENTRY] = "trace_fentry", | |
115 | [BPF_TRACE_FEXIT] = "trace_fexit", | |
116 | [BPF_MODIFY_RETURN] = "modify_return", | |
117 | [BPF_LSM_MAC] = "lsm_mac", | |
bffcf348 | 118 | [BPF_LSM_CGROUP] = "lsm_cgroup", |
ccde5760 DM |
119 | [BPF_SK_LOOKUP] = "sk_lookup", |
120 | [BPF_TRACE_ITER] = "trace_iter", | |
121 | [BPF_XDP_DEVMAP] = "xdp_devmap", | |
122 | [BPF_XDP_CPUMAP] = "xdp_cpumap", | |
123 | [BPF_XDP] = "xdp", | |
124 | [BPF_SK_REUSEPORT_SELECT] = "sk_reuseport_select", | |
125 | [BPF_SK_REUSEPORT_SELECT_OR_MIGRATE] = "sk_reuseport_select_or_migrate", | |
126 | [BPF_PERF_EVENT] = "perf_event", | |
127 | [BPF_TRACE_KPROBE_MULTI] = "trace_kprobe_multi", | |
8d1608d7 | 128 | [BPF_STRUCT_OPS] = "struct_ops", |
132328e8 | 129 | [BPF_NETFILTER] = "netfilter", |
fe20ce3a DB |
130 | [BPF_TCX_INGRESS] = "tcx_ingress", |
131 | [BPF_TCX_EGRESS] = "tcx_egress", | |
8097e460 | 132 | [BPF_TRACE_UPROBE_MULTI] = "trace_uprobe_multi", |
05c31b4a DB |
133 | [BPF_NETKIT_PRIMARY] = "netkit_primary", |
134 | [BPF_NETKIT_PEER] = "netkit_peer", | |
7b949654 | 135 | [BPF_TRACE_KPROBE_SESSION] = "trace_kprobe_session", |
ccde5760 DM |
136 | }; |
137 | ||
ba5d1b58 DM |
138 | static const char * const link_type_name[] = { |
139 | [BPF_LINK_TYPE_UNSPEC] = "unspec", | |
140 | [BPF_LINK_TYPE_RAW_TRACEPOINT] = "raw_tracepoint", | |
141 | [BPF_LINK_TYPE_TRACING] = "tracing", | |
142 | [BPF_LINK_TYPE_CGROUP] = "cgroup", | |
143 | [BPF_LINK_TYPE_ITER] = "iter", | |
144 | [BPF_LINK_TYPE_NETNS] = "netns", | |
145 | [BPF_LINK_TYPE_XDP] = "xdp", | |
146 | [BPF_LINK_TYPE_PERF_EVENT] = "perf_event", | |
147 | [BPF_LINK_TYPE_KPROBE_MULTI] = "kprobe_multi", | |
148 | [BPF_LINK_TYPE_STRUCT_OPS] = "struct_ops", | |
d0fe92fb | 149 | [BPF_LINK_TYPE_NETFILTER] = "netfilter", |
55cc3768 | 150 | [BPF_LINK_TYPE_TCX] = "tcx", |
8097e460 | 151 | [BPF_LINK_TYPE_UPROBE_MULTI] = "uprobe_multi", |
05c31b4a | 152 | [BPF_LINK_TYPE_NETKIT] = "netkit", |
849989af | 153 | [BPF_LINK_TYPE_SOCKMAP] = "sockmap", |
ba5d1b58 DM |
154 | }; |
155 | ||
3e6dc020 DM |
156 | static const char * const map_type_name[] = { |
157 | [BPF_MAP_TYPE_UNSPEC] = "unspec", | |
158 | [BPF_MAP_TYPE_HASH] = "hash", | |
159 | [BPF_MAP_TYPE_ARRAY] = "array", | |
160 | [BPF_MAP_TYPE_PROG_ARRAY] = "prog_array", | |
161 | [BPF_MAP_TYPE_PERF_EVENT_ARRAY] = "perf_event_array", | |
162 | [BPF_MAP_TYPE_PERCPU_HASH] = "percpu_hash", | |
163 | [BPF_MAP_TYPE_PERCPU_ARRAY] = "percpu_array", | |
164 | [BPF_MAP_TYPE_STACK_TRACE] = "stack_trace", | |
165 | [BPF_MAP_TYPE_CGROUP_ARRAY] = "cgroup_array", | |
166 | [BPF_MAP_TYPE_LRU_HASH] = "lru_hash", | |
167 | [BPF_MAP_TYPE_LRU_PERCPU_HASH] = "lru_percpu_hash", | |
168 | [BPF_MAP_TYPE_LPM_TRIE] = "lpm_trie", | |
169 | [BPF_MAP_TYPE_ARRAY_OF_MAPS] = "array_of_maps", | |
170 | [BPF_MAP_TYPE_HASH_OF_MAPS] = "hash_of_maps", | |
171 | [BPF_MAP_TYPE_DEVMAP] = "devmap", | |
172 | [BPF_MAP_TYPE_DEVMAP_HASH] = "devmap_hash", | |
173 | [BPF_MAP_TYPE_SOCKMAP] = "sockmap", | |
174 | [BPF_MAP_TYPE_CPUMAP] = "cpumap", | |
175 | [BPF_MAP_TYPE_XSKMAP] = "xskmap", | |
176 | [BPF_MAP_TYPE_SOCKHASH] = "sockhash", | |
177 | [BPF_MAP_TYPE_CGROUP_STORAGE] = "cgroup_storage", | |
178 | [BPF_MAP_TYPE_REUSEPORT_SOCKARRAY] = "reuseport_sockarray", | |
179 | [BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE] = "percpu_cgroup_storage", | |
180 | [BPF_MAP_TYPE_QUEUE] = "queue", | |
181 | [BPF_MAP_TYPE_STACK] = "stack", | |
182 | [BPF_MAP_TYPE_SK_STORAGE] = "sk_storage", | |
183 | [BPF_MAP_TYPE_STRUCT_OPS] = "struct_ops", | |
184 | [BPF_MAP_TYPE_RINGBUF] = "ringbuf", | |
185 | [BPF_MAP_TYPE_INODE_STORAGE] = "inode_storage", | |
186 | [BPF_MAP_TYPE_TASK_STORAGE] = "task_storage", | |
187 | [BPF_MAP_TYPE_BLOOM_FILTER] = "bloom_filter", | |
583c1f42 | 188 | [BPF_MAP_TYPE_USER_RINGBUF] = "user_ringbuf", |
4fe64af2 | 189 | [BPF_MAP_TYPE_CGRP_STORAGE] = "cgrp_storage", |
79ff13e9 | 190 | [BPF_MAP_TYPE_ARENA] = "arena", |
3e6dc020 DM |
191 | }; |
192 | ||
d18616e7 DM |
193 | static const char * const prog_type_name[] = { |
194 | [BPF_PROG_TYPE_UNSPEC] = "unspec", | |
195 | [BPF_PROG_TYPE_SOCKET_FILTER] = "socket_filter", | |
196 | [BPF_PROG_TYPE_KPROBE] = "kprobe", | |
197 | [BPF_PROG_TYPE_SCHED_CLS] = "sched_cls", | |
198 | [BPF_PROG_TYPE_SCHED_ACT] = "sched_act", | |
199 | [BPF_PROG_TYPE_TRACEPOINT] = "tracepoint", | |
200 | [BPF_PROG_TYPE_XDP] = "xdp", | |
201 | [BPF_PROG_TYPE_PERF_EVENT] = "perf_event", | |
202 | [BPF_PROG_TYPE_CGROUP_SKB] = "cgroup_skb", | |
203 | [BPF_PROG_TYPE_CGROUP_SOCK] = "cgroup_sock", | |
204 | [BPF_PROG_TYPE_LWT_IN] = "lwt_in", | |
205 | [BPF_PROG_TYPE_LWT_OUT] = "lwt_out", | |
206 | [BPF_PROG_TYPE_LWT_XMIT] = "lwt_xmit", | |
207 | [BPF_PROG_TYPE_SOCK_OPS] = "sock_ops", | |
208 | [BPF_PROG_TYPE_SK_SKB] = "sk_skb", | |
209 | [BPF_PROG_TYPE_CGROUP_DEVICE] = "cgroup_device", | |
210 | [BPF_PROG_TYPE_SK_MSG] = "sk_msg", | |
211 | [BPF_PROG_TYPE_RAW_TRACEPOINT] = "raw_tracepoint", | |
212 | [BPF_PROG_TYPE_CGROUP_SOCK_ADDR] = "cgroup_sock_addr", | |
213 | [BPF_PROG_TYPE_LWT_SEG6LOCAL] = "lwt_seg6local", | |
214 | [BPF_PROG_TYPE_LIRC_MODE2] = "lirc_mode2", | |
215 | [BPF_PROG_TYPE_SK_REUSEPORT] = "sk_reuseport", | |
216 | [BPF_PROG_TYPE_FLOW_DISSECTOR] = "flow_dissector", | |
217 | [BPF_PROG_TYPE_CGROUP_SYSCTL] = "cgroup_sysctl", | |
218 | [BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE] = "raw_tracepoint_writable", | |
219 | [BPF_PROG_TYPE_CGROUP_SOCKOPT] = "cgroup_sockopt", | |
220 | [BPF_PROG_TYPE_TRACING] = "tracing", | |
221 | [BPF_PROG_TYPE_STRUCT_OPS] = "struct_ops", | |
222 | [BPF_PROG_TYPE_EXT] = "ext", | |
223 | [BPF_PROG_TYPE_LSM] = "lsm", | |
224 | [BPF_PROG_TYPE_SK_LOOKUP] = "sk_lookup", | |
225 | [BPF_PROG_TYPE_SYSCALL] = "syscall", | |
d0fe92fb | 226 | [BPF_PROG_TYPE_NETFILTER] = "netfilter", |
d18616e7 DM |
227 | }; |
228 | ||
a8a1f7d0 SF |
229 | static int __base_pr(enum libbpf_print_level level, const char *format, |
230 | va_list args) | |
b3f59d66 | 231 | { |
6f1ae8b6 YS |
232 | if (level == LIBBPF_DEBUG) |
233 | return 0; | |
234 | ||
a8a1f7d0 | 235 | return vfprintf(stderr, format, args); |
b3f59d66 WN |
236 | } |
237 | ||
a8a1f7d0 | 238 | static libbpf_print_fn_t __libbpf_pr = __base_pr; |
b3f59d66 | 239 | |
e87fd8ba | 240 | libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn) |
b3f59d66 | 241 | { |
f1cb927c JK |
242 | libbpf_print_fn_t old_print_fn; |
243 | ||
244 | old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED); | |
e87fd8ba | 245 | |
e87fd8ba | 246 | return old_print_fn; |
b3f59d66 | 247 | } |
1a5e3fb1 | 248 | |
8461ef8b YS |
249 | __printf(2, 3) |
250 | void libbpf_print(enum libbpf_print_level level, const char *format, ...) | |
251 | { | |
252 | va_list args; | |
d7c5802f | 253 | int old_errno; |
f1cb927c | 254 | libbpf_print_fn_t print_fn; |
8461ef8b | 255 | |
f1cb927c JK |
256 | print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED); |
257 | if (!print_fn) | |
6f1ae8b6 YS |
258 | return; |
259 | ||
d7c5802f AN |
260 | old_errno = errno; |
261 | ||
8461ef8b | 262 | va_start(args, format); |
6f1ae8b6 | 263 | __libbpf_pr(level, format, args); |
8461ef8b | 264 | va_end(args); |
d7c5802f AN |
265 | |
266 | errno = old_errno; | |
8461ef8b YS |
267 | } |
268 | ||
dc3a2d25 THJ |
269 | static void pr_perm_msg(int err) |
270 | { | |
271 | struct rlimit limit; | |
272 | char buf[100]; | |
273 | ||
274 | if (err != -EPERM || geteuid() != 0) | |
275 | return; | |
276 | ||
277 | err = getrlimit(RLIMIT_MEMLOCK, &limit); | |
278 | if (err) | |
279 | return; | |
280 | ||
281 | if (limit.rlim_cur == RLIM_INFINITY) | |
282 | return; | |
283 | ||
284 | if (limit.rlim_cur < 1024) | |
b5c7d0d0 | 285 | snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur); |
dc3a2d25 THJ |
286 | else if (limit.rlim_cur < 1024*1024) |
287 | snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024); | |
288 | else | |
289 | snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024)); | |
290 | ||
291 | pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n", | |
292 | buf); | |
293 | } | |
294 | ||
6371ca3b WN |
295 | #define STRERR_BUFSIZE 128 |
296 | ||
1a5e3fb1 WN |
297 | /* Copied from tools/perf/util/util.h */ |
298 | #ifndef zfree | |
299 | # define zfree(ptr) ({ free(*ptr); *ptr = NULL; }) | |
300 | #endif | |
301 | ||
302 | #ifndef zclose | |
303 | # define zclose(fd) ({ \ | |
304 | int ___err = 0; \ | |
305 | if ((fd) >= 0) \ | |
306 | ___err = close((fd)); \ | |
307 | fd = -1; \ | |
308 | ___err; }) | |
309 | #endif | |
310 | ||
34be1646 SL |
311 | static inline __u64 ptr_to_u64(const void *ptr) |
312 | { | |
313 | return (__u64) (unsigned long) ptr; | |
314 | } | |
315 | ||
5981881d AN |
316 | int libbpf_set_strict_mode(enum libbpf_strict_mode mode) |
317 | { | |
bd054102 | 318 | /* as of v1.0 libbpf_set_strict_mode() is a no-op */ |
5981881d AN |
319 | return 0; |
320 | } | |
321 | ||
7615209f AN |
322 | __u32 libbpf_major_version(void) |
323 | { | |
324 | return LIBBPF_MAJOR_VERSION; | |
325 | } | |
326 | ||
327 | __u32 libbpf_minor_version(void) | |
328 | { | |
329 | return LIBBPF_MINOR_VERSION; | |
330 | } | |
331 | ||
332 | const char *libbpf_version_string(void) | |
333 | { | |
334 | #define __S(X) #X | |
335 | #define _S(X) __S(X) | |
336 | return "v" _S(LIBBPF_MAJOR_VERSION) "." _S(LIBBPF_MINOR_VERSION); | |
337 | #undef _S | |
338 | #undef __S | |
339 | } | |
340 | ||
166750bc AN |
341 | enum reloc_type { |
342 | RELO_LD64, | |
343 | RELO_CALL, | |
344 | RELO_DATA, | |
a18f7214 AS |
345 | RELO_EXTERN_LD64, |
346 | RELO_EXTERN_CALL, | |
53eddb5e | 347 | RELO_SUBPROG_ADDR, |
d0e92887 | 348 | RELO_CORE, |
166750bc AN |
349 | }; |
350 | ||
351 | struct reloc_desc { | |
352 | enum reloc_type type; | |
353 | int insn_idx; | |
d0e92887 AS |
354 | union { |
355 | const struct bpf_core_relo *core_relo; /* used when type == RELO_CORE */ | |
356 | struct { | |
357 | int map_idx; | |
358 | int sym_off; | |
3055ddd6 | 359 | int ext_idx; |
d0e92887 AS |
360 | }; |
361 | }; | |
166750bc AN |
362 | }; |
363 | ||
15ea31fa AN |
364 | /* stored as sec_def->cookie for all libbpf-supported SEC()s */ |
365 | enum sec_def_flags { | |
366 | SEC_NONE = 0, | |
367 | /* expected_attach_type is optional, if kernel doesn't support that */ | |
368 | SEC_EXP_ATTACH_OPT = 1, | |
369 | /* legacy, only used by libbpf_get_type_names() and | |
370 | * libbpf_attach_type_by_name(), not used by libbpf itself at all. | |
371 | * This used to be associated with cgroup (and few other) BPF programs | |
372 | * that were attachable through BPF_PROG_ATTACH command. Pretty | |
373 | * meaningless nowadays, though. | |
374 | */ | |
375 | SEC_ATTACHABLE = 2, | |
376 | SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT, | |
377 | /* attachment target is specified through BTF ID in either kernel or | |
e3ba8e4e KM |
378 | * other BPF program's BTF object |
379 | */ | |
15ea31fa AN |
380 | SEC_ATTACH_BTF = 4, |
381 | /* BPF program type allows sleeping/blocking in kernel */ | |
382 | SEC_SLEEPABLE = 8, | |
082c4bfb | 383 | /* BPF program support non-linear XDP buffer */ |
450b167f | 384 | SEC_XDP_FRAGS = 16, |
5902da6d JO |
385 | /* Setup proper attach type for usdt probes. */ |
386 | SEC_USDT = 32, | |
15ea31fa AN |
387 | }; |
388 | ||
25498a19 | 389 | struct bpf_sec_def { |
697f104d | 390 | char *sec; |
25498a19 AN |
391 | enum bpf_prog_type prog_type; |
392 | enum bpf_attach_type expected_attach_type; | |
15ea31fa | 393 | long cookie; |
697f104d | 394 | int handler_id; |
12d9466d | 395 | |
4fa5bcfe AN |
396 | libbpf_prog_setup_fn_t prog_setup_fn; |
397 | libbpf_prog_prepare_load_fn_t prog_prepare_load_fn; | |
398 | libbpf_prog_attach_fn_t prog_attach_fn; | |
25498a19 AN |
399 | }; |
400 | ||
a5b8bd47 WN |
401 | /* |
402 | * bpf_prog should be a better name but it has been used in | |
403 | * linux/filter.h. | |
404 | */ | |
405 | struct bpf_program { | |
bd054102 | 406 | char *name; |
52109584 | 407 | char *sec_name; |
c1122392 | 408 | size_t sec_idx; |
bd054102 | 409 | const struct bpf_sec_def *sec_def; |
c1122392 AN |
410 | /* this program's instruction offset (in number of instructions) |
411 | * within its containing ELF section | |
412 | */ | |
413 | size_t sec_insn_off; | |
414 | /* number of original instructions in ELF section belonging to this | |
415 | * program, not taking into account subprogram instructions possible | |
416 | * appended later during relocation | |
417 | */ | |
418 | size_t sec_insn_cnt; | |
419 | /* Offset (in number of instructions) of the start of instruction | |
420 | * belonging to this BPF program within its containing main BPF | |
421 | * program. For the entry-point (main) BPF program, this is always | |
422 | * zero. For a sub-program, this gets reset before each of main BPF | |
423 | * programs are processed and relocated and is used to determined | |
424 | * whether sub-program was already appended to the main program, and | |
425 | * if yes, at which instruction offset. | |
426 | */ | |
427 | size_t sub_insn_off; | |
428 | ||
c1122392 AN |
429 | /* instructions that belong to BPF program; insns[0] is located at |
430 | * sec_insn_off instruction within its ELF section in ELF file, so | |
431 | * when mapping ELF file instruction index to the local instruction, | |
432 | * one needs to subtract sec_insn_off; and vice versa. | |
433 | */ | |
a5b8bd47 | 434 | struct bpf_insn *insns; |
c1122392 AN |
435 | /* actual number of instruction in this BPF program's image; for |
436 | * entry-point BPF programs this includes the size of main program | |
437 | * itself plus all the used sub-programs, appended at the end | |
438 | */ | |
c3c55696 | 439 | size_t insns_cnt; |
34090915 | 440 | |
166750bc | 441 | struct reloc_desc *reloc_desc; |
34090915 | 442 | int nr_reloc; |
b3ce9079 AN |
443 | |
444 | /* BPF verifier log settings */ | |
445 | char *log_buf; | |
446 | size_t log_size; | |
447 | __u32 log_level; | |
55cffde2 | 448 | |
aa9b1ac3 | 449 | struct bpf_object *obj; |
d7be143b | 450 | |
cf90a20d | 451 | int fd; |
a3820c48 | 452 | bool autoload; |
43cb8cba | 453 | bool autoattach; |
7e2925f6 | 454 | bool sym_global; |
aea28a60 | 455 | bool mark_btf_static; |
c1122392 | 456 | enum bpf_prog_type type; |
d7be143b | 457 | enum bpf_attach_type expected_attach_type; |
7e2925f6 | 458 | int exception_cb_idx; |
cf90a20d | 459 | |
c1122392 | 460 | int prog_ifindex; |
91abb4a6 | 461 | __u32 attach_btf_obj_fd; |
12a8654b | 462 | __u32 attach_btf_id; |
e7bf94db | 463 | __u32 attach_prog_fd; |
cf90a20d | 464 | |
2993e051 YS |
465 | void *func_info; |
466 | __u32 func_info_rec_size; | |
f0187f0b | 467 | __u32 func_info_cnt; |
47eff617 | 468 | |
3d650141 MKL |
469 | void *line_info; |
470 | __u32 line_info_rec_size; | |
471 | __u32 line_info_cnt; | |
04656198 | 472 | __u32 prog_flags; |
a5b8bd47 WN |
473 | }; |
474 | ||
590a0088 MKL |
475 | struct bpf_struct_ops { |
476 | const char *tname; | |
477 | const struct btf_type *type; | |
478 | struct bpf_program **progs; | |
479 | __u32 *kern_func_off; | |
480 | /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */ | |
481 | void *data; | |
482 | /* e.g. struct bpf_struct_ops_tcp_congestion_ops in | |
483 | * btf_vmlinux's format. | |
484 | * struct bpf_struct_ops_tcp_congestion_ops { | |
485 | * [... some other kernel fields ...] | |
486 | * struct tcp_congestion_ops data; | |
487 | * } | |
488 | * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops) | |
489 | * bpf_map__init_kern_struct_ops() will populate the "kern_vdata" | |
490 | * from "data". | |
491 | */ | |
492 | void *kern_vdata; | |
493 | __u32 type_id; | |
494 | }; | |
495 | ||
ac9d1389 AN |
496 | #define DATA_SEC ".data" |
497 | #define BSS_SEC ".bss" | |
498 | #define RODATA_SEC ".rodata" | |
81bfdd08 | 499 | #define KCONFIG_SEC ".kconfig" |
1c0c7074 | 500 | #define KSYMS_SEC ".ksyms" |
590a0088 | 501 | #define STRUCT_OPS_SEC ".struct_ops" |
809a69d6 | 502 | #define STRUCT_OPS_LINK_SEC ".struct_ops.link" |
10ebe835 | 503 | #define ARENA_SEC ".addr_space.1" |
ac9d1389 | 504 | |
d859900c DB |
505 | enum libbpf_map_type { |
506 | LIBBPF_MAP_UNSPEC, | |
507 | LIBBPF_MAP_DATA, | |
508 | LIBBPF_MAP_BSS, | |
509 | LIBBPF_MAP_RODATA, | |
81bfdd08 | 510 | LIBBPF_MAP_KCONFIG, |
d859900c DB |
511 | }; |
512 | ||
146bf811 AN |
513 | struct bpf_map_def { |
514 | unsigned int type; | |
515 | unsigned int key_size; | |
516 | unsigned int value_size; | |
517 | unsigned int max_entries; | |
518 | unsigned int map_flags; | |
519 | }; | |
520 | ||
9d759a9b | 521 | struct bpf_map { |
ec41817b | 522 | struct bpf_object *obj; |
561bbcca | 523 | char *name; |
aed65917 AN |
524 | /* real_name is defined for special internal maps (.rodata*, |
525 | * .data*, .bss, .kconfig) and preserves their original ELF section | |
e3ba8e4e | 526 | * name. This is important to be able to find corresponding BTF |
aed65917 AN |
527 | * DATASEC information. |
528 | */ | |
529 | char *real_name; | |
01af3bf0 | 530 | int fd; |
db48814b AN |
531 | int sec_idx; |
532 | size_t sec_offset; | |
f0307a7e | 533 | int map_ifindex; |
addb9fc9 | 534 | int inner_map_fd; |
9d759a9b | 535 | struct bpf_map_def def; |
1bdb6c9a | 536 | __u32 numa_node; |
646f02ff | 537 | __u32 btf_var_idx; |
9e926acd | 538 | int mod_btf_fd; |
5b891af7 MKL |
539 | __u32 btf_key_type_id; |
540 | __u32 btf_value_type_id; | |
590a0088 | 541 | __u32 btf_vmlinux_value_type_id; |
d859900c | 542 | enum libbpf_map_type libbpf_type; |
eba9c5f4 | 543 | void *mmaped; |
590a0088 | 544 | struct bpf_struct_ops *st_ops; |
646f02ff AN |
545 | struct bpf_map *inner_map; |
546 | void **init_slots; | |
547 | int init_slots_sz; | |
4580b25f THJ |
548 | char *pin_path; |
549 | bool pinned; | |
ec6d5f47 | 550 | bool reused; |
ec41817b | 551 | bool autocreate; |
47512102 | 552 | __u64 map_extra; |
d859900c DB |
553 | }; |
554 | ||
166750bc AN |
555 | enum extern_type { |
556 | EXT_UNKNOWN, | |
2e33efe3 | 557 | EXT_KCFG, |
1c0c7074 | 558 | EXT_KSYM, |
2e33efe3 AN |
559 | }; |
560 | ||
561 | enum kcfg_type { | |
562 | KCFG_UNKNOWN, | |
563 | KCFG_CHAR, | |
564 | KCFG_BOOL, | |
565 | KCFG_INT, | |
566 | KCFG_TRISTATE, | |
567 | KCFG_CHAR_ARR, | |
166750bc AN |
568 | }; |
569 | ||
570 | struct extern_desc { | |
2e33efe3 | 571 | enum extern_type type; |
166750bc AN |
572 | int sym_idx; |
573 | int btf_id; | |
2e33efe3 AN |
574 | int sec_btf_id; |
575 | const char *name; | |
5964a223 | 576 | char *essent_name; |
166750bc | 577 | bool is_set; |
2e33efe3 AN |
578 | bool is_weak; |
579 | union { | |
580 | struct { | |
581 | enum kcfg_type type; | |
582 | int sz; | |
583 | int align; | |
584 | int data_off; | |
585 | bool is_signed; | |
586 | } kcfg; | |
1c0c7074 AN |
587 | struct { |
588 | unsigned long long addr; | |
d370bbe1 HL |
589 | |
590 | /* target btf_id of the corresponding kernel var. */ | |
284d2587 AN |
591 | int kernel_btf_obj_fd; |
592 | int kernel_btf_id; | |
d370bbe1 HL |
593 | |
594 | /* local btf_id of the ksym extern's type. */ | |
595 | __u32 type_id; | |
9dbe6015 KKD |
596 | /* BTF fd index to be patched in for insn->off, this is |
597 | * 0 for vmlinux BTF, index in obj->fd_array for module | |
598 | * BTF | |
599 | */ | |
600 | __s16 btf_fd_idx; | |
1c0c7074 | 601 | } ksym; |
2e33efe3 | 602 | }; |
166750bc AN |
603 | }; |
604 | ||
4f33a53d AN |
605 | struct module_btf { |
606 | struct btf *btf; | |
607 | char *name; | |
608 | __u32 id; | |
91abb4a6 | 609 | int fd; |
9dbe6015 | 610 | int fd_array_idx; |
4f33a53d AN |
611 | }; |
612 | ||
25bbbd7a AN |
613 | enum sec_type { |
614 | SEC_UNUSED = 0, | |
615 | SEC_RELO, | |
616 | SEC_BSS, | |
617 | SEC_DATA, | |
618 | SEC_RODATA, | |
240bf8a5 | 619 | SEC_ST_OPS, |
25bbbd7a AN |
620 | }; |
621 | ||
622 | struct elf_sec_desc { | |
623 | enum sec_type sec_type; | |
624 | Elf64_Shdr *shdr; | |
625 | Elf_Data *data; | |
626 | }; | |
627 | ||
29a30ff5 AN |
628 | struct elf_state { |
629 | int fd; | |
630 | const void *obj_buf; | |
631 | size_t obj_buf_sz; | |
632 | Elf *elf; | |
ad23b723 | 633 | Elf64_Ehdr *ehdr; |
29a30ff5 | 634 | Elf_Data *symbols; |
2e7ba4f8 | 635 | Elf_Data *arena_data; |
29a30ff5 AN |
636 | size_t shstrndx; /* section index for section name strings */ |
637 | size_t strtabidx; | |
25bbbd7a | 638 | struct elf_sec_desc *secs; |
51deedc9 | 639 | size_t sec_cnt; |
29a30ff5 AN |
640 | int btf_maps_shndx; |
641 | __u32 btf_maps_sec_btf_id; | |
642 | int text_shndx; | |
643 | int symbols_shndx; | |
240bf8a5 | 644 | bool has_st_ops; |
2e7ba4f8 | 645 | int arena_data_shndx; |
4f33a53d AN |
646 | }; |
647 | ||
2e4913e0 AN |
648 | struct usdt_manager; |
649 | ||
1a5e3fb1 | 650 | struct bpf_object { |
d859900c | 651 | char name[BPF_OBJ_NAME_LEN]; |
cb1e5e96 | 652 | char license[64]; |
438363c0 | 653 | __u32 kern_version; |
0b3d1efa | 654 | |
a5b8bd47 WN |
655 | struct bpf_program *programs; |
656 | size_t nr_programs; | |
9d759a9b WN |
657 | struct bpf_map *maps; |
658 | size_t nr_maps; | |
bf829271 | 659 | size_t maps_cap; |
9d759a9b | 660 | |
8601fd42 | 661 | char *kconfig; |
166750bc AN |
662 | struct extern_desc *externs; |
663 | int nr_extern; | |
81bfdd08 | 664 | int kconfig_map_idx; |
166750bc | 665 | |
52d3352e | 666 | bool loaded; |
c3c55696 | 667 | bool has_subcalls; |
25bbbd7a | 668 | bool has_rodata; |
a5b8bd47 | 669 | |
e2fa0156 AS |
670 | struct bpf_gen *gen_loader; |
671 | ||
29a30ff5 AN |
672 | /* Information when doing ELF related work. Only valid if efile.elf is not NULL */ |
673 | struct elf_state efile; | |
10931d24 | 674 | |
8a138aed | 675 | struct btf *btf; |
0f7515ca AN |
676 | struct btf_ext *btf_ext; |
677 | ||
a6ed02ca KS |
678 | /* Parse and load BTF vmlinux if any of the programs in the object need |
679 | * it at load time. | |
680 | */ | |
681 | struct btf *btf_vmlinux; | |
1373ff59 SC |
682 | /* Path to the custom BTF to be used for BPF CO-RE relocations as an |
683 | * override for vmlinux BTF. | |
684 | */ | |
685 | char *btf_custom_path; | |
0f7515ca AN |
686 | /* vmlinux BTF override for CO-RE relocations */ |
687 | struct btf *btf_vmlinux_override; | |
4f33a53d AN |
688 | /* Lazily initialized kernel module BTFs */ |
689 | struct module_btf *btf_modules; | |
690 | bool btf_modules_loaded; | |
691 | size_t btf_module_cnt; | |
692 | size_t btf_module_cap; | |
8a138aed | 693 | |
e0e3ea88 AN |
694 | /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */ |
695 | char *log_buf; | |
696 | size_t log_size; | |
697 | __u32 log_level; | |
698 | ||
9dbe6015 KKD |
699 | int *fd_array; |
700 | size_t fd_array_cap; | |
701 | size_t fd_array_cnt; | |
702 | ||
2e4913e0 AN |
703 | struct usdt_manager *usdt_man; |
704 | ||
2e7ba4f8 AN |
705 | struct bpf_map *arena_map; |
706 | void *arena_data; | |
707 | size_t arena_data_sz; | |
708 | ||
6b434b61 AN |
709 | struct kern_feature_cache *feat_cache; |
710 | char *token_path; | |
711 | int token_fd; | |
712 | ||
1a5e3fb1 WN |
713 | char path[]; |
714 | }; | |
1a5e3fb1 | 715 | |
88a82120 AN |
716 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off); |
717 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off); | |
718 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx); | |
719 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name); | |
ad23b723 | 720 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn); |
88a82120 AN |
721 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn); |
722 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn); | |
ad23b723 AN |
723 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx); |
724 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx); | |
88a82120 | 725 | |
29cd77f4 | 726 | void bpf_program__unload(struct bpf_program *prog) |
55cffde2 WN |
727 | { |
728 | if (!prog) | |
729 | return; | |
730 | ||
cf90a20d | 731 | zclose(prog->fd); |
2993e051 | 732 | |
2993e051 | 733 | zfree(&prog->func_info); |
07a09d1b | 734 | zfree(&prog->line_info); |
55cffde2 WN |
735 | } |
736 | ||
a5b8bd47 WN |
737 | static void bpf_program__exit(struct bpf_program *prog) |
738 | { | |
739 | if (!prog) | |
740 | return; | |
741 | ||
55cffde2 | 742 | bpf_program__unload(prog); |
88cda1c9 | 743 | zfree(&prog->name); |
52109584 | 744 | zfree(&prog->sec_name); |
a5b8bd47 | 745 | zfree(&prog->insns); |
34090915 WN |
746 | zfree(&prog->reloc_desc); |
747 | ||
748 | prog->nr_reloc = 0; | |
a5b8bd47 | 749 | prog->insns_cnt = 0; |
c1122392 | 750 | prog->sec_idx = -1; |
a5b8bd47 WN |
751 | } |
752 | ||
c3c55696 AN |
753 | static bool insn_is_subprog_call(const struct bpf_insn *insn) |
754 | { | |
755 | return BPF_CLASS(insn->code) == BPF_JMP && | |
756 | BPF_OP(insn->code) == BPF_CALL && | |
757 | BPF_SRC(insn->code) == BPF_K && | |
758 | insn->src_reg == BPF_PSEUDO_CALL && | |
759 | insn->dst_reg == 0 && | |
760 | insn->off == 0; | |
761 | } | |
762 | ||
aa0b8d43 MKL |
763 | static bool is_call_insn(const struct bpf_insn *insn) |
764 | { | |
765 | return insn->code == (BPF_JMP | BPF_CALL); | |
766 | } | |
767 | ||
53eddb5e YS |
768 | static bool insn_is_pseudo_func(struct bpf_insn *insn) |
769 | { | |
aa0b8d43 | 770 | return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC; |
53eddb5e YS |
771 | } |
772 | ||
a5b8bd47 | 773 | static int |
c3c55696 AN |
774 | bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog, |
775 | const char *name, size_t sec_idx, const char *sec_name, | |
776 | size_t sec_off, void *insn_data, size_t insn_data_sz) | |
a5b8bd47 | 777 | { |
c1122392 AN |
778 | if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) { |
779 | pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n", | |
780 | sec_name, name, sec_off, insn_data_sz); | |
a5b8bd47 WN |
781 | return -EINVAL; |
782 | } | |
783 | ||
1ad9cbb8 | 784 | memset(prog, 0, sizeof(*prog)); |
c3c55696 AN |
785 | prog->obj = obj; |
786 | ||
c1122392 AN |
787 | prog->sec_idx = sec_idx; |
788 | prog->sec_insn_off = sec_off / BPF_INSN_SZ; | |
789 | prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ; | |
790 | /* insns_cnt can later be increased by appending used subprograms */ | |
791 | prog->insns_cnt = prog->sec_insn_cnt; | |
a5b8bd47 | 792 | |
c1122392 | 793 | prog->type = BPF_PROG_TYPE_UNSPEC; |
cf90a20d | 794 | prog->fd = -1; |
7e2925f6 | 795 | prog->exception_cb_idx = -1; |
a3820c48 AN |
796 | |
797 | /* libbpf's convention for SEC("?abc...") is that it's just like | |
798 | * SEC("abc...") but the corresponding bpf_program starts out with | |
799 | * autoload set to false. | |
800 | */ | |
801 | if (sec_name[0] == '?') { | |
802 | prog->autoload = false; | |
803 | /* from now on forget there was ? in section name */ | |
804 | sec_name++; | |
805 | } else { | |
806 | prog->autoload = true; | |
807 | } | |
a5b8bd47 | 808 | |
43cb8cba HL |
809 | prog->autoattach = true; |
810 | ||
b3ce9079 AN |
811 | /* inherit object's log_level */ |
812 | prog->log_level = obj->log_level; | |
813 | ||
52109584 AN |
814 | prog->sec_name = strdup(sec_name); |
815 | if (!prog->sec_name) | |
c1122392 AN |
816 | goto errout; |
817 | ||
818 | prog->name = strdup(name); | |
819 | if (!prog->name) | |
a5b8bd47 | 820 | goto errout; |
a5b8bd47 | 821 | |
c1122392 AN |
822 | prog->insns = malloc(insn_data_sz); |
823 | if (!prog->insns) | |
a5b8bd47 | 824 | goto errout; |
c1122392 | 825 | memcpy(prog->insns, insn_data, insn_data_sz); |
a5b8bd47 WN |
826 | |
827 | return 0; | |
828 | errout: | |
c1122392 | 829 | pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name); |
a5b8bd47 WN |
830 | bpf_program__exit(prog); |
831 | return -ENOMEM; | |
832 | } | |
833 | ||
834 | static int | |
c1122392 AN |
835 | bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data, |
836 | const char *sec_name, int sec_idx) | |
a5b8bd47 | 837 | { |
6245947c | 838 | Elf_Data *symbols = obj->efile.symbols; |
c1122392 AN |
839 | struct bpf_program *prog, *progs; |
840 | void *data = sec_data->d_buf; | |
6245947c AN |
841 | size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms; |
842 | int nr_progs, err, i; | |
c1122392 | 843 | const char *name; |
ad23b723 | 844 | Elf64_Sym *sym; |
a5b8bd47 WN |
845 | |
846 | progs = obj->programs; | |
847 | nr_progs = obj->nr_programs; | |
ad23b723 | 848 | nr_syms = symbols->d_size / sizeof(Elf64_Sym); |
a5b8bd47 | 849 | |
6245947c | 850 | for (i = 0; i < nr_syms; i++) { |
ad23b723 AN |
851 | sym = elf_sym_by_idx(obj, i); |
852 | ||
853 | if (sym->st_shndx != sec_idx) | |
6245947c | 854 | continue; |
ad23b723 | 855 | if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC) |
6245947c | 856 | continue; |
88cda1c9 | 857 | |
ad23b723 AN |
858 | prog_sz = sym->st_size; |
859 | sec_off = sym->st_value; | |
88cda1c9 | 860 | |
ad23b723 | 861 | name = elf_sym_str(obj, sym->st_name); |
c1122392 AN |
862 | if (!name) { |
863 | pr_warn("sec '%s': failed to get symbol name for offset %zu\n", | |
864 | sec_name, sec_off); | |
865 | return -LIBBPF_ERRNO__FORMAT; | |
866 | } | |
88cda1c9 | 867 | |
c1122392 AN |
868 | if (sec_off + prog_sz > sec_sz) { |
869 | pr_warn("sec '%s': program at offset %zu crosses section boundary\n", | |
870 | sec_name, sec_off); | |
871 | return -LIBBPF_ERRNO__FORMAT; | |
872 | } | |
88cda1c9 | 873 | |
ad23b723 | 874 | if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) { |
513f485c AN |
875 | pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name); |
876 | return -ENOTSUP; | |
877 | } | |
878 | ||
c3c55696 AN |
879 | pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n", |
880 | sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz); | |
88cda1c9 | 881 | |
c3c55696 | 882 | progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs)); |
c1122392 AN |
883 | if (!progs) { |
884 | /* | |
885 | * In this case the original obj->programs | |
886 | * is still valid, so don't need special treat for | |
887 | * bpf_close_object(). | |
888 | */ | |
889 | pr_warn("sec '%s': failed to alloc memory for new program '%s'\n", | |
890 | sec_name, name); | |
891 | return -ENOMEM; | |
88cda1c9 | 892 | } |
c1122392 | 893 | obj->programs = progs; |
88cda1c9 | 894 | |
c1122392 | 895 | prog = &progs[nr_progs]; |
9a94f277 | 896 | |
c3c55696 AN |
897 | err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name, |
898 | sec_off, data + sec_off, prog_sz); | |
c1122392 AN |
899 | if (err) |
900 | return err; | |
9a94f277 | 901 | |
7e2925f6 KKD |
902 | if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL) |
903 | prog->sym_global = true; | |
904 | ||
e5670fa0 AN |
905 | /* if function is a global/weak symbol, but has restricted |
906 | * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC | |
907 | * as static to enable more permissive BPF verification mode | |
908 | * with more outside context available to BPF verifier | |
aea28a60 | 909 | */ |
7e2925f6 KKD |
910 | if (prog->sym_global && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN |
911 | || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL)) | |
aea28a60 AN |
912 | prog->mark_btf_static = true; |
913 | ||
c1122392 AN |
914 | nr_progs++; |
915 | obj->nr_programs = nr_progs; | |
88cda1c9 MKL |
916 | } |
917 | ||
918 | return 0; | |
919 | } | |
920 | ||
590a0088 MKL |
921 | static const struct btf_member * |
922 | find_member_by_offset(const struct btf_type *t, __u32 bit_offset) | |
923 | { | |
924 | struct btf_member *m; | |
925 | int i; | |
926 | ||
927 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
928 | if (btf_member_bit_offset(t, i) == bit_offset) | |
929 | return m; | |
930 | } | |
931 | ||
932 | return NULL; | |
933 | } | |
934 | ||
935 | static const struct btf_member * | |
936 | find_member_by_name(const struct btf *btf, const struct btf_type *t, | |
937 | const char *name) | |
938 | { | |
939 | struct btf_member *m; | |
940 | int i; | |
941 | ||
942 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
943 | if (!strcmp(btf__name_by_offset(btf, m->name_off), name)) | |
944 | return m; | |
945 | } | |
946 | ||
947 | return NULL; | |
948 | } | |
949 | ||
9e926acd KFL |
950 | static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name, |
951 | __u16 kind, struct btf **res_btf, | |
952 | struct module_btf **res_mod_btf); | |
953 | ||
590a0088 | 954 | #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_" |
a6ed02ca KS |
955 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
956 | const char *name, __u32 kind); | |
590a0088 MKL |
957 | |
958 | static int | |
a2a5172c | 959 | find_struct_ops_kern_types(struct bpf_object *obj, const char *tname_raw, |
9e926acd | 960 | struct module_btf **mod_btf, |
590a0088 MKL |
961 | const struct btf_type **type, __u32 *type_id, |
962 | const struct btf_type **vtype, __u32 *vtype_id, | |
963 | const struct btf_member **data_member) | |
964 | { | |
965 | const struct btf_type *kern_type, *kern_vtype; | |
966 | const struct btf_member *kern_data_member; | |
9e926acd | 967 | struct btf *btf; |
590a0088 | 968 | __s32 kern_vtype_id, kern_type_id; |
a2a5172c | 969 | char tname[256]; |
590a0088 MKL |
970 | __u32 i; |
971 | ||
a2a5172c EZ |
972 | snprintf(tname, sizeof(tname), "%.*s", |
973 | (int)bpf_core_essential_name_len(tname_raw), tname_raw); | |
974 | ||
9e926acd KFL |
975 | kern_type_id = find_ksym_btf_id(obj, tname, BTF_KIND_STRUCT, |
976 | &btf, mod_btf); | |
590a0088 MKL |
977 | if (kern_type_id < 0) { |
978 | pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n", | |
979 | tname); | |
980 | return kern_type_id; | |
981 | } | |
982 | kern_type = btf__type_by_id(btf, kern_type_id); | |
983 | ||
984 | /* Find the corresponding "map_value" type that will be used | |
985 | * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example, | |
986 | * find "struct bpf_struct_ops_tcp_congestion_ops" from the | |
987 | * btf_vmlinux. | |
988 | */ | |
a6ed02ca KS |
989 | kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX, |
990 | tname, BTF_KIND_STRUCT); | |
590a0088 | 991 | if (kern_vtype_id < 0) { |
a6ed02ca KS |
992 | pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n", |
993 | STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
994 | return kern_vtype_id; |
995 | } | |
996 | kern_vtype = btf__type_by_id(btf, kern_vtype_id); | |
997 | ||
998 | /* Find "struct tcp_congestion_ops" from | |
999 | * struct bpf_struct_ops_tcp_congestion_ops { | |
1000 | * [ ... ] | |
1001 | * struct tcp_congestion_ops data; | |
1002 | * } | |
1003 | */ | |
1004 | kern_data_member = btf_members(kern_vtype); | |
1005 | for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) { | |
1006 | if (kern_data_member->type == kern_type_id) | |
1007 | break; | |
1008 | } | |
1009 | if (i == btf_vlen(kern_vtype)) { | |
a6ed02ca KS |
1010 | pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n", |
1011 | tname, STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
1012 | return -EINVAL; |
1013 | } | |
1014 | ||
1015 | *type = kern_type; | |
1016 | *type_id = kern_type_id; | |
1017 | *vtype = kern_vtype; | |
1018 | *vtype_id = kern_vtype_id; | |
1019 | *data_member = kern_data_member; | |
1020 | ||
1021 | return 0; | |
1022 | } | |
1023 | ||
1024 | static bool bpf_map__is_struct_ops(const struct bpf_map *map) | |
1025 | { | |
1026 | return map->def.type == BPF_MAP_TYPE_STRUCT_OPS; | |
1027 | } | |
1028 | ||
69e4a9d2 KFL |
1029 | static bool is_valid_st_ops_program(struct bpf_object *obj, |
1030 | const struct bpf_program *prog) | |
1031 | { | |
1032 | int i; | |
1033 | ||
1034 | for (i = 0; i < obj->nr_programs; i++) { | |
1035 | if (&obj->programs[i] == prog) | |
1036 | return prog->type == BPF_PROG_TYPE_STRUCT_OPS; | |
1037 | } | |
1038 | ||
1039 | return false; | |
1040 | } | |
1041 | ||
fe9d049c EZ |
1042 | /* For each struct_ops program P, referenced from some struct_ops map M, |
1043 | * enable P.autoload if there are Ms for which M.autocreate is true, | |
1044 | * disable P.autoload if for all Ms M.autocreate is false. | |
1045 | * Don't change P.autoload for programs that are not referenced from any maps. | |
1046 | */ | |
1047 | static int bpf_object_adjust_struct_ops_autoload(struct bpf_object *obj) | |
1048 | { | |
1049 | struct bpf_program *prog, *slot_prog; | |
1050 | struct bpf_map *map; | |
1051 | int i, j, k, vlen; | |
1052 | ||
1053 | for (i = 0; i < obj->nr_programs; ++i) { | |
1054 | int should_load = false; | |
1055 | int use_cnt = 0; | |
1056 | ||
1057 | prog = &obj->programs[i]; | |
1058 | if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) | |
1059 | continue; | |
1060 | ||
1061 | for (j = 0; j < obj->nr_maps; ++j) { | |
1062 | map = &obj->maps[j]; | |
1063 | if (!bpf_map__is_struct_ops(map)) | |
1064 | continue; | |
1065 | ||
1066 | vlen = btf_vlen(map->st_ops->type); | |
1067 | for (k = 0; k < vlen; ++k) { | |
1068 | slot_prog = map->st_ops->progs[k]; | |
1069 | if (prog != slot_prog) | |
1070 | continue; | |
1071 | ||
1072 | use_cnt++; | |
1073 | if (map->autocreate) | |
1074 | should_load = true; | |
1075 | } | |
1076 | } | |
1077 | if (use_cnt) | |
1078 | prog->autoload = should_load; | |
1079 | } | |
1080 | ||
1081 | return 0; | |
1082 | } | |
1083 | ||
590a0088 | 1084 | /* Init the map's fields that depend on kern_btf */ |
9e926acd | 1085 | static int bpf_map__init_kern_struct_ops(struct bpf_map *map) |
590a0088 MKL |
1086 | { |
1087 | const struct btf_member *member, *kern_member, *kern_data_member; | |
1088 | const struct btf_type *type, *kern_type, *kern_vtype; | |
1089 | __u32 i, kern_type_id, kern_vtype_id, kern_data_off; | |
9e926acd KFL |
1090 | struct bpf_object *obj = map->obj; |
1091 | const struct btf *btf = obj->btf; | |
590a0088 | 1092 | struct bpf_struct_ops *st_ops; |
9e926acd KFL |
1093 | const struct btf *kern_btf; |
1094 | struct module_btf *mod_btf; | |
590a0088 MKL |
1095 | void *data, *kern_data; |
1096 | const char *tname; | |
1097 | int err; | |
1098 | ||
1099 | st_ops = map->st_ops; | |
1100 | type = st_ops->type; | |
1101 | tname = st_ops->tname; | |
9e926acd | 1102 | err = find_struct_ops_kern_types(obj, tname, &mod_btf, |
590a0088 MKL |
1103 | &kern_type, &kern_type_id, |
1104 | &kern_vtype, &kern_vtype_id, | |
1105 | &kern_data_member); | |
1106 | if (err) | |
1107 | return err; | |
1108 | ||
9e926acd KFL |
1109 | kern_btf = mod_btf ? mod_btf->btf : obj->btf_vmlinux; |
1110 | ||
590a0088 MKL |
1111 | pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n", |
1112 | map->name, st_ops->type_id, kern_type_id, kern_vtype_id); | |
1113 | ||
9e926acd | 1114 | map->mod_btf_fd = mod_btf ? mod_btf->fd : -1; |
590a0088 MKL |
1115 | map->def.value_size = kern_vtype->size; |
1116 | map->btf_vmlinux_value_type_id = kern_vtype_id; | |
1117 | ||
1118 | st_ops->kern_vdata = calloc(1, kern_vtype->size); | |
1119 | if (!st_ops->kern_vdata) | |
1120 | return -ENOMEM; | |
1121 | ||
1122 | data = st_ops->data; | |
1123 | kern_data_off = kern_data_member->offset / 8; | |
1124 | kern_data = st_ops->kern_vdata + kern_data_off; | |
1125 | ||
1126 | member = btf_members(type); | |
1127 | for (i = 0; i < btf_vlen(type); i++, member++) { | |
1128 | const struct btf_type *mtype, *kern_mtype; | |
1129 | __u32 mtype_id, kern_mtype_id; | |
1130 | void *mdata, *kern_mdata; | |
0737df6d | 1131 | struct bpf_program *prog; |
590a0088 MKL |
1132 | __s64 msize, kern_msize; |
1133 | __u32 moff, kern_moff; | |
1134 | __u32 kern_member_idx; | |
1135 | const char *mname; | |
1136 | ||
1137 | mname = btf__name_by_offset(btf, member->name_off); | |
c911fc61 KFL |
1138 | moff = member->offset / 8; |
1139 | mdata = data + moff; | |
1140 | msize = btf__resolve_size(btf, member->type); | |
1141 | if (msize < 0) { | |
1142 | pr_warn("struct_ops init_kern %s: failed to resolve the size of member %s\n", | |
1143 | map->name, mname); | |
1144 | return msize; | |
1145 | } | |
1146 | ||
590a0088 MKL |
1147 | kern_member = find_member_by_name(kern_btf, kern_type, mname); |
1148 | if (!kern_member) { | |
0737df6d AN |
1149 | if (!libbpf_is_mem_zeroed(mdata, msize)) { |
1150 | pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n", | |
c911fc61 | 1151 | map->name, mname); |
0737df6d AN |
1152 | return -ENOTSUP; |
1153 | } | |
1154 | ||
1155 | prog = st_ops->progs[i]; | |
1156 | if (prog) { | |
1157 | /* If we had declaratively set struct_ops callback, we need to | |
1158 | * first validate that it's actually a struct_ops program. | |
1159 | * And then force its autoload to false, because it doesn't have | |
1160 | * a chance of succeeding from POV of the current struct_ops map. | |
1161 | * If this program is still referenced somewhere else, though, | |
1162 | * then bpf_object_adjust_struct_ops_autoload() will update its | |
1163 | * autoload accordingly. | |
1164 | */ | |
1165 | if (!is_valid_st_ops_program(obj, prog)) { | |
1166 | pr_warn("struct_ops init_kern %s: member %s is declaratively assigned a non-struct_ops program\n", | |
1167 | map->name, mname); | |
1168 | return -EINVAL; | |
1169 | } | |
1170 | prog->autoload = false; | |
1171 | st_ops->progs[i] = NULL; | |
c911fc61 KFL |
1172 | } |
1173 | ||
0737df6d AN |
1174 | /* Skip all-zero/NULL fields if they are not present in the kernel BTF */ |
1175 | pr_info("struct_ops %s: member %s not found in kernel, skipping it as it's set to zero\n", | |
590a0088 | 1176 | map->name, mname); |
0737df6d | 1177 | continue; |
590a0088 MKL |
1178 | } |
1179 | ||
1180 | kern_member_idx = kern_member - btf_members(kern_type); | |
1181 | if (btf_member_bitfield_size(type, i) || | |
1182 | btf_member_bitfield_size(kern_type, kern_member_idx)) { | |
1183 | pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n", | |
1184 | map->name, mname); | |
1185 | return -ENOTSUP; | |
1186 | } | |
1187 | ||
590a0088 | 1188 | kern_moff = kern_member->offset / 8; |
590a0088 MKL |
1189 | kern_mdata = kern_data + kern_moff; |
1190 | ||
1191 | mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id); | |
1192 | kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type, | |
1193 | &kern_mtype_id); | |
1194 | if (BTF_INFO_KIND(mtype->info) != | |
1195 | BTF_INFO_KIND(kern_mtype->info)) { | |
1196 | pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n", | |
1197 | map->name, mname, BTF_INFO_KIND(mtype->info), | |
1198 | BTF_INFO_KIND(kern_mtype->info)); | |
1199 | return -ENOTSUP; | |
1200 | } | |
1201 | ||
1202 | if (btf_is_ptr(mtype)) { | |
69e4a9d2 KFL |
1203 | /* Update the value from the shadow type */ |
1204 | prog = *(void **)mdata; | |
1205 | st_ops->progs[i] = prog; | |
d2836ddd MKL |
1206 | if (!prog) |
1207 | continue; | |
69e4a9d2 KFL |
1208 | if (!is_valid_st_ops_program(obj, prog)) { |
1209 | pr_warn("struct_ops init_kern %s: member %s is not a struct_ops program\n", | |
1210 | map->name, mname); | |
1211 | return -ENOTSUP; | |
1212 | } | |
d2836ddd | 1213 | |
590a0088 MKL |
1214 | kern_mtype = skip_mods_and_typedefs(kern_btf, |
1215 | kern_mtype->type, | |
1216 | &kern_mtype_id); | |
d2836ddd MKL |
1217 | |
1218 | /* mtype->type must be a func_proto which was | |
1219 | * guaranteed in bpf_object__collect_st_ops_relos(), | |
1220 | * so only check kern_mtype for func_proto here. | |
1221 | */ | |
1222 | if (!btf_is_func_proto(kern_mtype)) { | |
1223 | pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n", | |
590a0088 MKL |
1224 | map->name, mname); |
1225 | return -ENOTSUP; | |
1226 | } | |
1227 | ||
9e926acd KFL |
1228 | if (mod_btf) |
1229 | prog->attach_btf_obj_fd = mod_btf->fd; | |
d9ab2f76 EZ |
1230 | |
1231 | /* if we haven't yet processed this BPF program, record proper | |
1232 | * attach_btf_id and member_idx | |
1233 | */ | |
1234 | if (!prog->attach_btf_id) { | |
1235 | prog->attach_btf_id = kern_type_id; | |
1236 | prog->expected_attach_type = kern_member_idx; | |
1237 | } | |
1238 | ||
1239 | /* struct_ops BPF prog can be re-used between multiple | |
1240 | * .struct_ops & .struct_ops.link as long as it's the | |
1241 | * same struct_ops struct definition and the same | |
1242 | * function pointer field | |
1243 | */ | |
1244 | if (prog->attach_btf_id != kern_type_id) { | |
1245 | pr_warn("struct_ops init_kern %s func ptr %s: invalid reuse of prog %s in sec %s with type %u: attach_btf_id %u != kern_type_id %u\n", | |
1246 | map->name, mname, prog->name, prog->sec_name, prog->type, | |
1247 | prog->attach_btf_id, kern_type_id); | |
1248 | return -EINVAL; | |
1249 | } | |
1250 | if (prog->expected_attach_type != kern_member_idx) { | |
1251 | pr_warn("struct_ops init_kern %s func ptr %s: invalid reuse of prog %s in sec %s with type %u: expected_attach_type %u != kern_member_idx %u\n", | |
1252 | map->name, mname, prog->name, prog->sec_name, prog->type, | |
1253 | prog->expected_attach_type, kern_member_idx); | |
1254 | return -EINVAL; | |
1255 | } | |
590a0088 MKL |
1256 | |
1257 | st_ops->kern_func_off[i] = kern_data_off + kern_moff; | |
1258 | ||
1259 | pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n", | |
1260 | map->name, mname, prog->name, moff, | |
1261 | kern_moff); | |
1262 | ||
1263 | continue; | |
1264 | } | |
1265 | ||
590a0088 | 1266 | kern_msize = btf__resolve_size(kern_btf, kern_mtype_id); |
c911fc61 | 1267 | if (kern_msize < 0 || msize != kern_msize) { |
590a0088 MKL |
1268 | pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n", |
1269 | map->name, mname, (ssize_t)msize, | |
1270 | (ssize_t)kern_msize); | |
1271 | return -ENOTSUP; | |
1272 | } | |
1273 | ||
1274 | pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n", | |
1275 | map->name, mname, (unsigned int)msize, | |
1276 | moff, kern_moff); | |
1277 | memcpy(kern_mdata, mdata, msize); | |
1278 | } | |
1279 | ||
1280 | return 0; | |
1281 | } | |
1282 | ||
1283 | static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj) | |
1284 | { | |
590a0088 MKL |
1285 | struct bpf_map *map; |
1286 | size_t i; | |
1287 | int err; | |
1288 | ||
1289 | for (i = 0; i < obj->nr_maps; i++) { | |
1290 | map = &obj->maps[i]; | |
1291 | ||
1292 | if (!bpf_map__is_struct_ops(map)) | |
1293 | continue; | |
1294 | ||
8db05261 EZ |
1295 | if (!map->autocreate) |
1296 | continue; | |
1297 | ||
9e926acd | 1298 | err = bpf_map__init_kern_struct_ops(map); |
a6ed02ca | 1299 | if (err) |
590a0088 | 1300 | return err; |
590a0088 MKL |
1301 | } |
1302 | ||
590a0088 MKL |
1303 | return 0; |
1304 | } | |
1305 | ||
809a69d6 | 1306 | static int init_struct_ops_maps(struct bpf_object *obj, const char *sec_name, |
240bf8a5 | 1307 | int shndx, Elf_Data *data) |
590a0088 MKL |
1308 | { |
1309 | const struct btf_type *type, *datasec; | |
1310 | const struct btf_var_secinfo *vsi; | |
1311 | struct bpf_struct_ops *st_ops; | |
1312 | const char *tname, *var_name; | |
1313 | __s32 type_id, datasec_id; | |
1314 | const struct btf *btf; | |
1315 | struct bpf_map *map; | |
1316 | __u32 i; | |
1317 | ||
809a69d6 | 1318 | if (shndx == -1) |
590a0088 MKL |
1319 | return 0; |
1320 | ||
1321 | btf = obj->btf; | |
809a69d6 | 1322 | datasec_id = btf__find_by_name_kind(btf, sec_name, |
590a0088 MKL |
1323 | BTF_KIND_DATASEC); |
1324 | if (datasec_id < 0) { | |
1325 | pr_warn("struct_ops init: DATASEC %s not found\n", | |
809a69d6 | 1326 | sec_name); |
590a0088 MKL |
1327 | return -EINVAL; |
1328 | } | |
1329 | ||
1330 | datasec = btf__type_by_id(btf, datasec_id); | |
1331 | vsi = btf_var_secinfos(datasec); | |
1332 | for (i = 0; i < btf_vlen(datasec); i++, vsi++) { | |
1333 | type = btf__type_by_id(obj->btf, vsi->type); | |
1334 | var_name = btf__name_by_offset(obj->btf, type->name_off); | |
1335 | ||
1336 | type_id = btf__resolve_type(obj->btf, vsi->type); | |
1337 | if (type_id < 0) { | |
1338 | pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n", | |
809a69d6 | 1339 | vsi->type, sec_name); |
590a0088 MKL |
1340 | return -EINVAL; |
1341 | } | |
1342 | ||
1343 | type = btf__type_by_id(obj->btf, type_id); | |
1344 | tname = btf__name_by_offset(obj->btf, type->name_off); | |
1345 | if (!tname[0]) { | |
1346 | pr_warn("struct_ops init: anonymous type is not supported\n"); | |
1347 | return -ENOTSUP; | |
1348 | } | |
1349 | if (!btf_is_struct(type)) { | |
1350 | pr_warn("struct_ops init: %s is not a struct\n", tname); | |
1351 | return -EINVAL; | |
1352 | } | |
1353 | ||
1354 | map = bpf_object__add_map(obj); | |
1355 | if (IS_ERR(map)) | |
1356 | return PTR_ERR(map); | |
1357 | ||
809a69d6 | 1358 | map->sec_idx = shndx; |
590a0088 MKL |
1359 | map->sec_offset = vsi->offset; |
1360 | map->name = strdup(var_name); | |
1361 | if (!map->name) | |
1362 | return -ENOMEM; | |
3644d285 | 1363 | map->btf_value_type_id = type_id; |
590a0088 | 1364 | |
5ad0ecbe EZ |
1365 | /* Follow same convention as for programs autoload: |
1366 | * SEC("?.struct_ops") means map is not created by default. | |
1367 | */ | |
1368 | if (sec_name[0] == '?') { | |
1369 | map->autocreate = false; | |
1370 | /* from now on forget there was ? in section name */ | |
1371 | sec_name++; | |
1372 | } | |
1373 | ||
590a0088 MKL |
1374 | map->def.type = BPF_MAP_TYPE_STRUCT_OPS; |
1375 | map->def.key_size = sizeof(int); | |
1376 | map->def.value_size = type->size; | |
1377 | map->def.max_entries = 1; | |
240bf8a5 | 1378 | map->def.map_flags = strcmp(sec_name, STRUCT_OPS_LINK_SEC) == 0 ? BPF_F_LINK : 0; |
590a0088 MKL |
1379 | |
1380 | map->st_ops = calloc(1, sizeof(*map->st_ops)); | |
1381 | if (!map->st_ops) | |
1382 | return -ENOMEM; | |
1383 | st_ops = map->st_ops; | |
1384 | st_ops->data = malloc(type->size); | |
1385 | st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs)); | |
1386 | st_ops->kern_func_off = malloc(btf_vlen(type) * | |
1387 | sizeof(*st_ops->kern_func_off)); | |
1388 | if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off) | |
1389 | return -ENOMEM; | |
1390 | ||
809a69d6 | 1391 | if (vsi->offset + type->size > data->d_size) { |
590a0088 | 1392 | pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n", |
809a69d6 | 1393 | var_name, sec_name); |
590a0088 MKL |
1394 | return -EINVAL; |
1395 | } | |
1396 | ||
1397 | memcpy(st_ops->data, | |
809a69d6 | 1398 | data->d_buf + vsi->offset, |
590a0088 MKL |
1399 | type->size); |
1400 | st_ops->tname = tname; | |
1401 | st_ops->type = type; | |
1402 | st_ops->type_id = type_id; | |
1403 | ||
1404 | pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n", | |
1405 | tname, type_id, var_name, vsi->offset); | |
1406 | } | |
1407 | ||
1408 | return 0; | |
1409 | } | |
1410 | ||
809a69d6 KFL |
1411 | static int bpf_object_init_struct_ops(struct bpf_object *obj) |
1412 | { | |
240bf8a5 EZ |
1413 | const char *sec_name; |
1414 | int sec_idx, err; | |
809a69d6 | 1415 | |
240bf8a5 EZ |
1416 | for (sec_idx = 0; sec_idx < obj->efile.sec_cnt; ++sec_idx) { |
1417 | struct elf_sec_desc *desc = &obj->efile.secs[sec_idx]; | |
1418 | ||
1419 | if (desc->sec_type != SEC_ST_OPS) | |
1420 | continue; | |
1421 | ||
1422 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); | |
1423 | if (!sec_name) | |
1424 | return -LIBBPF_ERRNO__FORMAT; | |
1425 | ||
1426 | err = init_struct_ops_maps(obj, sec_name, sec_idx, desc->data); | |
1427 | if (err) | |
1428 | return err; | |
1429 | } | |
1430 | ||
1431 | return 0; | |
809a69d6 KFL |
1432 | } |
1433 | ||
6c956392 | 1434 | static struct bpf_object *bpf_object__new(const char *path, |
5e61f270 | 1435 | const void *obj_buf, |
2ce8450e AN |
1436 | size_t obj_buf_sz, |
1437 | const char *obj_name) | |
1a5e3fb1 WN |
1438 | { |
1439 | struct bpf_object *obj; | |
d859900c | 1440 | char *end; |
1a5e3fb1 WN |
1441 | |
1442 | obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1); | |
1443 | if (!obj) { | |
be18010e | 1444 | pr_warn("alloc memory failed for %s\n", path); |
6371ca3b | 1445 | return ERR_PTR(-ENOMEM); |
1a5e3fb1 WN |
1446 | } |
1447 | ||
1448 | strcpy(obj->path, path); | |
2ce8450e | 1449 | if (obj_name) { |
9fc205b4 | 1450 | libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name)); |
2ce8450e AN |
1451 | } else { |
1452 | /* Using basename() GNU version which doesn't modify arg. */ | |
9fc205b4 | 1453 | libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name)); |
2ce8450e AN |
1454 | end = strchr(obj->name, '.'); |
1455 | if (end) | |
1456 | *end = 0; | |
1457 | } | |
6c956392 | 1458 | |
d859900c | 1459 | obj->efile.fd = -1; |
6c956392 | 1460 | /* |
76e1022b | 1461 | * Caller of this function should also call |
6c956392 WN |
1462 | * bpf_object__elf_finish() after data collection to return |
1463 | * obj_buf to user. If not, we should duplicate the buffer to | |
1464 | * avoid user freeing them before elf finish. | |
1465 | */ | |
1466 | obj->efile.obj_buf = obj_buf; | |
1467 | obj->efile.obj_buf_sz = obj_buf_sz; | |
abd29c93 | 1468 | obj->efile.btf_maps_shndx = -1; |
81bfdd08 | 1469 | obj->kconfig_map_idx = -1; |
6c956392 | 1470 | |
5e61f270 | 1471 | obj->kern_version = get_kernel_version(); |
52d3352e | 1472 | obj->loaded = false; |
9a208eff | 1473 | |
1a5e3fb1 WN |
1474 | return obj; |
1475 | } | |
1476 | ||
1477 | static void bpf_object__elf_finish(struct bpf_object *obj) | |
1478 | { | |
29a30ff5 | 1479 | if (!obj->efile.elf) |
1a5e3fb1 WN |
1480 | return; |
1481 | ||
b71a2ebf GC |
1482 | elf_end(obj->efile.elf); |
1483 | obj->efile.elf = NULL; | |
bec7d68c | 1484 | obj->efile.symbols = NULL; |
2e7ba4f8 | 1485 | obj->efile.arena_data = NULL; |
b62f06e8 | 1486 | |
25bbbd7a AN |
1487 | zfree(&obj->efile.secs); |
1488 | obj->efile.sec_cnt = 0; | |
1a5e3fb1 | 1489 | zclose(obj->efile.fd); |
6c956392 WN |
1490 | obj->efile.obj_buf = NULL; |
1491 | obj->efile.obj_buf_sz = 0; | |
1a5e3fb1 WN |
1492 | } |
1493 | ||
1494 | static int bpf_object__elf_init(struct bpf_object *obj) | |
1495 | { | |
ad23b723 | 1496 | Elf64_Ehdr *ehdr; |
1a5e3fb1 | 1497 | int err = 0; |
ad23b723 | 1498 | Elf *elf; |
1a5e3fb1 | 1499 | |
29a30ff5 | 1500 | if (obj->efile.elf) { |
88a82120 | 1501 | pr_warn("elf: init internal error\n"); |
6371ca3b | 1502 | return -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1503 | } |
1504 | ||
6c956392 | 1505 | if (obj->efile.obj_buf_sz > 0) { |
146bf811 | 1506 | /* obj_buf should have been validated by bpf_object__open_mem(). */ |
ad23b723 | 1507 | elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz); |
6c956392 | 1508 | } else { |
92274e24 | 1509 | obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC); |
6c956392 | 1510 | if (obj->efile.fd < 0) { |
be5c5d4e | 1511 | char errmsg[STRERR_BUFSIZE], *cp; |
1ce6a9fc | 1512 | |
be5c5d4e AN |
1513 | err = -errno; |
1514 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
88a82120 | 1515 | pr_warn("elf: failed to open %s: %s\n", obj->path, cp); |
be5c5d4e | 1516 | return err; |
6c956392 WN |
1517 | } |
1518 | ||
ad23b723 | 1519 | elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL); |
1a5e3fb1 WN |
1520 | } |
1521 | ||
ad23b723 | 1522 | if (!elf) { |
88a82120 | 1523 | pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1524 | err = -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1525 | goto errout; |
1526 | } | |
1527 | ||
ad23b723 AN |
1528 | obj->efile.elf = elf; |
1529 | ||
1530 | if (elf_kind(elf) != ELF_K_ELF) { | |
1531 | err = -LIBBPF_ERRNO__FORMAT; | |
1532 | pr_warn("elf: '%s' is not a proper ELF object\n", obj->path); | |
1533 | goto errout; | |
1534 | } | |
1535 | ||
1536 | if (gelf_getclass(elf) != ELFCLASS64) { | |
1537 | err = -LIBBPF_ERRNO__FORMAT; | |
1538 | pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path); | |
1539 | goto errout; | |
1540 | } | |
1541 | ||
1542 | obj->efile.ehdr = ehdr = elf64_getehdr(elf); | |
1543 | if (!obj->efile.ehdr) { | |
88a82120 | 1544 | pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1545 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1546 | goto errout; |
1547 | } | |
1a5e3fb1 | 1548 | |
ad23b723 | 1549 | if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) { |
88a82120 AN |
1550 | pr_warn("elf: failed to get section names section index for %s: %s\n", |
1551 | obj->path, elf_errmsg(-1)); | |
1552 | err = -LIBBPF_ERRNO__FORMAT; | |
1553 | goto errout; | |
1554 | } | |
1555 | ||
70e79866 | 1556 | /* ELF is corrupted/truncated, avoid calling elf_strptr. */ |
ad23b723 | 1557 | if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) { |
88a82120 AN |
1558 | pr_warn("elf: failed to get section names strings from %s: %s\n", |
1559 | obj->path, elf_errmsg(-1)); | |
8f3f5792 NK |
1560 | err = -LIBBPF_ERRNO__FORMAT; |
1561 | goto errout; | |
88a82120 AN |
1562 | } |
1563 | ||
9b16137a | 1564 | /* Old LLVM set e_machine to EM_NONE */ |
ad23b723 | 1565 | if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) { |
88a82120 | 1566 | pr_warn("elf: %s is not a valid eBPF object file\n", obj->path); |
6371ca3b | 1567 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1568 | goto errout; |
1569 | } | |
1570 | ||
1571 | return 0; | |
1572 | errout: | |
1573 | bpf_object__elf_finish(obj); | |
1574 | return err; | |
1575 | } | |
1576 | ||
12ef5634 | 1577 | static int bpf_object__check_endianness(struct bpf_object *obj) |
cc4228d5 | 1578 | { |
3930198d | 1579 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
ad23b723 | 1580 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB) |
12ef5634 | 1581 | return 0; |
3930198d | 1582 | #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
ad23b723 | 1583 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB) |
12ef5634 AN |
1584 | return 0; |
1585 | #else | |
1586 | # error "Unrecognized __BYTE_ORDER__" | |
1587 | #endif | |
88a82120 | 1588 | pr_warn("elf: endianness mismatch in %s.\n", obj->path); |
6371ca3b | 1589 | return -LIBBPF_ERRNO__ENDIAN; |
cc4228d5 WN |
1590 | } |
1591 | ||
cb1e5e96 | 1592 | static int |
399dc65e | 1593 | bpf_object__init_license(struct bpf_object *obj, void *data, size_t size) |
cb1e5e96 | 1594 | { |
35a85550 SHY |
1595 | if (!data) { |
1596 | pr_warn("invalid license section in %s\n", obj->path); | |
1597 | return -LIBBPF_ERRNO__FORMAT; | |
1598 | } | |
f9798239 AN |
1599 | /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't |
1600 | * go over allowed ELF data section buffer | |
1601 | */ | |
1602 | libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license))); | |
cb1e5e96 WN |
1603 | pr_debug("license of %s is %s\n", obj->path, obj->license); |
1604 | return 0; | |
1605 | } | |
1606 | ||
54b8625c JF |
1607 | static int |
1608 | bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size) | |
1609 | { | |
1610 | __u32 kver; | |
1611 | ||
35a85550 | 1612 | if (!data || size != sizeof(kver)) { |
be18010e | 1613 | pr_warn("invalid kver section in %s\n", obj->path); |
54b8625c JF |
1614 | return -LIBBPF_ERRNO__FORMAT; |
1615 | } | |
1616 | memcpy(&kver, data, sizeof(kver)); | |
1617 | obj->kern_version = kver; | |
1618 | pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version); | |
1619 | return 0; | |
1620 | } | |
1621 | ||
addb9fc9 NS |
1622 | static bool bpf_map_type__is_map_in_map(enum bpf_map_type type) |
1623 | { | |
1624 | if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1625 | type == BPF_MAP_TYPE_HASH_OF_MAPS) | |
1626 | return true; | |
1627 | return false; | |
1628 | } | |
1629 | ||
b96c07f3 | 1630 | static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size) |
1713d68b | 1631 | { |
25bbbd7a AN |
1632 | Elf_Data *data; |
1633 | Elf_Scn *scn; | |
1713d68b | 1634 | |
25bbbd7a | 1635 | if (!name) |
1713d68b | 1636 | return -EINVAL; |
88a82120 | 1637 | |
25bbbd7a AN |
1638 | scn = elf_sec_by_name(obj, name); |
1639 | data = elf_sec_data(obj, scn); | |
1640 | if (data) { | |
25bbbd7a | 1641 | *size = data->d_size; |
08894d9c | 1642 | return 0; /* found it */ |
1713d68b DB |
1643 | } |
1644 | ||
08894d9c | 1645 | return -ENOENT; |
1713d68b DB |
1646 | } |
1647 | ||
f33f742d | 1648 | static Elf64_Sym *find_elf_var_sym(const struct bpf_object *obj, const char *name) |
1713d68b DB |
1649 | { |
1650 | Elf_Data *symbols = obj->efile.symbols; | |
1651 | const char *sname; | |
1652 | size_t si; | |
1653 | ||
ad23b723 AN |
1654 | for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) { |
1655 | Elf64_Sym *sym = elf_sym_by_idx(obj, si); | |
1713d68b | 1656 | |
3a06ec0a AN |
1657 | if (ELF64_ST_TYPE(sym->st_info) != STT_OBJECT) |
1658 | continue; | |
1659 | ||
1660 | if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL && | |
1661 | ELF64_ST_BIND(sym->st_info) != STB_WEAK) | |
1713d68b DB |
1662 | continue; |
1663 | ||
ad23b723 | 1664 | sname = elf_sym_str(obj, sym->st_name); |
1713d68b | 1665 | if (!sname) { |
ad23b723 | 1666 | pr_warn("failed to get sym name string for var %s\n", name); |
f33f742d | 1667 | return ERR_PTR(-EIO); |
1713d68b | 1668 | } |
f33f742d AN |
1669 | if (strcmp(name, sname) == 0) |
1670 | return sym; | |
1713d68b DB |
1671 | } |
1672 | ||
f33f742d | 1673 | return ERR_PTR(-ENOENT); |
1713d68b DB |
1674 | } |
1675 | ||
9fa5e1a1 AN |
1676 | /* Some versions of Android don't provide memfd_create() in their libc |
1677 | * implementation, so avoid complications and just go straight to Linux | |
1678 | * syscall. | |
1679 | */ | |
1680 | static int sys_memfd_create(const char *name, unsigned flags) | |
1681 | { | |
1682 | return syscall(__NR_memfd_create, name, flags); | |
1683 | } | |
1684 | ||
ddb2ffdc ACM |
1685 | #ifndef MFD_CLOEXEC |
1686 | #define MFD_CLOEXEC 0x0001U | |
1687 | #endif | |
1688 | ||
dac645b9 AN |
1689 | static int create_placeholder_fd(void) |
1690 | { | |
1691 | int fd; | |
1692 | ||
9fa5e1a1 | 1693 | fd = ensure_good_fd(sys_memfd_create("libbpf-placeholder-fd", MFD_CLOEXEC)); |
dac645b9 AN |
1694 | if (fd < 0) |
1695 | return -errno; | |
1696 | return fd; | |
1697 | } | |
1698 | ||
bf829271 | 1699 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj) |
d859900c | 1700 | { |
69721203 AN |
1701 | struct bpf_map *map; |
1702 | int err; | |
bf829271 | 1703 | |
69721203 AN |
1704 | err = libbpf_ensure_mem((void **)&obj->maps, &obj->maps_cap, |
1705 | sizeof(*obj->maps), obj->nr_maps + 1); | |
1706 | if (err) | |
1707 | return ERR_PTR(err); | |
bf829271 | 1708 | |
69721203 | 1709 | map = &obj->maps[obj->nr_maps++]; |
ec41817b | 1710 | map->obj = obj; |
dac645b9 AN |
1711 | /* Preallocate map FD without actually creating BPF map just yet. |
1712 | * These map FD "placeholders" will be reused later without changing | |
1713 | * FD value when map is actually created in the kernel. | |
1714 | * | |
1715 | * This is useful to be able to perform BPF program relocations | |
1716 | * without having to create BPF maps before that step. This allows us | |
1717 | * to finalize and load BTF very late in BPF object's loading phase, | |
1718 | * right before BPF maps have to be created and BPF programs have to | |
1719 | * be loaded. By having these map FD placeholders we can perform all | |
1720 | * the sanitizations, relocations, and any other adjustments before we | |
1721 | * start creating actual BPF kernel objects (BTF, maps, progs). | |
1722 | */ | |
1723 | map->fd = create_placeholder_fd(); | |
1724 | if (map->fd < 0) | |
1725 | return ERR_PTR(map->fd); | |
69721203 | 1726 | map->inner_map_fd = -1; |
ec41817b | 1727 | map->autocreate = true; |
bf829271 | 1728 | |
69721203 | 1729 | return map; |
d859900c DB |
1730 | } |
1731 | ||
79ff13e9 | 1732 | static size_t array_map_mmap_sz(unsigned int value_sz, unsigned int max_entries) |
eba9c5f4 | 1733 | { |
9d0a2331 | 1734 | const long page_sz = sysconf(_SC_PAGE_SIZE); |
eba9c5f4 AN |
1735 | size_t map_sz; |
1736 | ||
9d0a2331 | 1737 | map_sz = (size_t)roundup(value_sz, 8) * max_entries; |
eba9c5f4 AN |
1738 | map_sz = roundup(map_sz, page_sz); |
1739 | return map_sz; | |
1740 | } | |
1741 | ||
79ff13e9 AS |
1742 | static size_t bpf_map_mmap_sz(const struct bpf_map *map) |
1743 | { | |
1744 | const long page_sz = sysconf(_SC_PAGE_SIZE); | |
1745 | ||
1746 | switch (map->def.type) { | |
1747 | case BPF_MAP_TYPE_ARRAY: | |
1748 | return array_map_mmap_sz(map->def.value_size, map->def.max_entries); | |
1749 | case BPF_MAP_TYPE_ARENA: | |
1750 | return page_sz * map->def.max_entries; | |
1751 | default: | |
1752 | return 0; /* not supported */ | |
1753 | } | |
1754 | } | |
1755 | ||
9d0a2331 JK |
1756 | static int bpf_map_mmap_resize(struct bpf_map *map, size_t old_sz, size_t new_sz) |
1757 | { | |
1758 | void *mmaped; | |
1759 | ||
1760 | if (!map->mmaped) | |
1761 | return -EINVAL; | |
1762 | ||
1763 | if (old_sz == new_sz) | |
1764 | return 0; | |
1765 | ||
1766 | mmaped = mmap(NULL, new_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); | |
1767 | if (mmaped == MAP_FAILED) | |
1768 | return -errno; | |
1769 | ||
1770 | memcpy(mmaped, map->mmaped, min(old_sz, new_sz)); | |
1771 | munmap(map->mmaped, old_sz); | |
1772 | map->mmaped = mmaped; | |
1773 | return 0; | |
1774 | } | |
1775 | ||
aed65917 | 1776 | static char *internal_map_name(struct bpf_object *obj, const char *real_name) |
81bfdd08 | 1777 | { |
113e6b7e | 1778 | char map_name[BPF_OBJ_NAME_LEN], *p; |
aed65917 AN |
1779 | int pfx_len, sfx_len = max((size_t)7, strlen(real_name)); |
1780 | ||
1781 | /* This is one of the more confusing parts of libbpf for various | |
1782 | * reasons, some of which are historical. The original idea for naming | |
1783 | * internal names was to include as much of BPF object name prefix as | |
1784 | * possible, so that it can be distinguished from similar internal | |
1785 | * maps of a different BPF object. | |
1786 | * As an example, let's say we have bpf_object named 'my_object_name' | |
1787 | * and internal map corresponding to '.rodata' ELF section. The final | |
1788 | * map name advertised to user and to the kernel will be | |
1789 | * 'my_objec.rodata', taking first 8 characters of object name and | |
1790 | * entire 7 characters of '.rodata'. | |
1791 | * Somewhat confusingly, if internal map ELF section name is shorter | |
1792 | * than 7 characters, e.g., '.bss', we still reserve 7 characters | |
1793 | * for the suffix, even though we only have 4 actual characters, and | |
1794 | * resulting map will be called 'my_objec.bss', not even using all 15 | |
1795 | * characters allowed by the kernel. Oh well, at least the truncated | |
1796 | * object name is somewhat consistent in this case. But if the map | |
1797 | * name is '.kconfig', we'll still have entirety of '.kconfig' added | |
1798 | * (8 chars) and thus will be left with only first 7 characters of the | |
1799 | * object name ('my_obje'). Happy guessing, user, that the final map | |
1800 | * name will be "my_obje.kconfig". | |
1801 | * Now, with libbpf starting to support arbitrarily named .rodata.* | |
1802 | * and .data.* data sections, it's possible that ELF section name is | |
1803 | * longer than allowed 15 chars, so we now need to be careful to take | |
1804 | * only up to 15 first characters of ELF name, taking no BPF object | |
1805 | * name characters at all. So '.rodata.abracadabra' will result in | |
1806 | * '.rodata.abracad' kernel and user-visible name. | |
1807 | * We need to keep this convoluted logic intact for .data, .bss and | |
1808 | * .rodata maps, but for new custom .data.custom and .rodata.custom | |
1809 | * maps we use their ELF names as is, not prepending bpf_object name | |
1810 | * in front. We still need to truncate them to 15 characters for the | |
1811 | * kernel. Full name can be recovered for such maps by using DATASEC | |
1812 | * BTF type associated with such map's value type, though. | |
1813 | */ | |
1814 | if (sfx_len >= BPF_OBJ_NAME_LEN) | |
1815 | sfx_len = BPF_OBJ_NAME_LEN - 1; | |
1816 | ||
1817 | /* if there are two or more dots in map name, it's a custom dot map */ | |
1818 | if (strchr(real_name + 1, '.') != NULL) | |
1819 | pfx_len = 0; | |
1820 | else | |
1821 | pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name)); | |
81bfdd08 AN |
1822 | |
1823 | snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name, | |
aed65917 | 1824 | sfx_len, real_name); |
81bfdd08 | 1825 | |
113e6b7e THJ |
1826 | /* sanitise map name to characters allowed by kernel */ |
1827 | for (p = map_name; *p && p < map_name + sizeof(map_name); p++) | |
1828 | if (!isalnum(*p) && *p != '_' && *p != '.') | |
1829 | *p = '_'; | |
1830 | ||
81bfdd08 AN |
1831 | return strdup(map_name); |
1832 | } | |
1833 | ||
262cfb74 | 1834 | static int |
4fcac46c AN |
1835 | map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map); |
1836 | ||
1837 | /* Internal BPF map is mmap()'able only if at least one of corresponding | |
1838 | * DATASEC's VARs are to be exposed through BPF skeleton. I.e., it's a GLOBAL | |
1839 | * variable and it's not marked as __hidden (which turns it into, effectively, | |
1840 | * a STATIC variable). | |
1841 | */ | |
1842 | static bool map_is_mmapable(struct bpf_object *obj, struct bpf_map *map) | |
1843 | { | |
1844 | const struct btf_type *t, *vt; | |
1845 | struct btf_var_secinfo *vsi; | |
1846 | int i, n; | |
1847 | ||
1848 | if (!map->btf_value_type_id) | |
1849 | return false; | |
1850 | ||
1851 | t = btf__type_by_id(obj->btf, map->btf_value_type_id); | |
1852 | if (!btf_is_datasec(t)) | |
1853 | return false; | |
1854 | ||
1855 | vsi = btf_var_secinfos(t); | |
1856 | for (i = 0, n = btf_vlen(t); i < n; i++, vsi++) { | |
1857 | vt = btf__type_by_id(obj->btf, vsi->type); | |
1858 | if (!btf_is_var(vt)) | |
1859 | continue; | |
1860 | ||
1861 | if (btf_var(vt)->linkage != BTF_VAR_STATIC) | |
1862 | return true; | |
1863 | } | |
1864 | ||
1865 | return false; | |
1866 | } | |
262cfb74 | 1867 | |
d859900c | 1868 | static int |
bf829271 | 1869 | bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type, |
aed65917 | 1870 | const char *real_name, int sec_idx, void *data, size_t data_sz) |
d859900c | 1871 | { |
bf829271 AN |
1872 | struct bpf_map_def *def; |
1873 | struct bpf_map *map; | |
9d0a2331 | 1874 | size_t mmap_sz; |
eba9c5f4 | 1875 | int err; |
bf829271 AN |
1876 | |
1877 | map = bpf_object__add_map(obj); | |
1878 | if (IS_ERR(map)) | |
1879 | return PTR_ERR(map); | |
d859900c DB |
1880 | |
1881 | map->libbpf_type = type; | |
db48814b AN |
1882 | map->sec_idx = sec_idx; |
1883 | map->sec_offset = 0; | |
aed65917 AN |
1884 | map->real_name = strdup(real_name); |
1885 | map->name = internal_map_name(obj, real_name); | |
1886 | if (!map->real_name || !map->name) { | |
1887 | zfree(&map->real_name); | |
1888 | zfree(&map->name); | |
d859900c DB |
1889 | return -ENOMEM; |
1890 | } | |
1891 | ||
bf829271 | 1892 | def = &map->def; |
d859900c DB |
1893 | def->type = BPF_MAP_TYPE_ARRAY; |
1894 | def->key_size = sizeof(int); | |
eba9c5f4 | 1895 | def->value_size = data_sz; |
d859900c | 1896 | def->max_entries = 1; |
81bfdd08 | 1897 | def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG |
2e7ba4f8 | 1898 | ? BPF_F_RDONLY_PROG : 0; |
4fcac46c AN |
1899 | |
1900 | /* failures are fine because of maps like .rodata.str1.1 */ | |
1901 | (void) map_fill_btf_type_info(obj, map); | |
1902 | ||
1903 | if (map_is_mmapable(obj, map)) | |
1904 | def->map_flags |= BPF_F_MMAPABLE; | |
7fe74b43 AN |
1905 | |
1906 | pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n", | |
81bfdd08 | 1907 | map->name, map->sec_idx, map->sec_offset, def->map_flags); |
7fe74b43 | 1908 | |
79ff13e9 | 1909 | mmap_sz = bpf_map_mmap_sz(map); |
9d0a2331 | 1910 | map->mmaped = mmap(NULL, mmap_sz, PROT_READ | PROT_WRITE, |
eba9c5f4 AN |
1911 | MAP_SHARED | MAP_ANONYMOUS, -1, 0); |
1912 | if (map->mmaped == MAP_FAILED) { | |
1913 | err = -errno; | |
1914 | map->mmaped = NULL; | |
1915 | pr_warn("failed to alloc map '%s' content buffer: %d\n", | |
1916 | map->name, err); | |
aed65917 | 1917 | zfree(&map->real_name); |
eba9c5f4 AN |
1918 | zfree(&map->name); |
1919 | return err; | |
d859900c DB |
1920 | } |
1921 | ||
166750bc | 1922 | if (data) |
eba9c5f4 AN |
1923 | memcpy(map->mmaped, data, data_sz); |
1924 | ||
e1d1dc46 | 1925 | pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name); |
d859900c DB |
1926 | return 0; |
1927 | } | |
1928 | ||
bf829271 AN |
1929 | static int bpf_object__init_global_data_maps(struct bpf_object *obj) |
1930 | { | |
25bbbd7a | 1931 | struct elf_sec_desc *sec_desc; |
aed65917 | 1932 | const char *sec_name; |
25bbbd7a | 1933 | int err = 0, sec_idx; |
bf829271 | 1934 | |
bf829271 AN |
1935 | /* |
1936 | * Populate obj->maps with libbpf internal maps. | |
1937 | */ | |
25bbbd7a AN |
1938 | for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) { |
1939 | sec_desc = &obj->efile.secs[sec_idx]; | |
1940 | ||
47ea7417 | 1941 | /* Skip recognized sections with size 0. */ |
d4e6d684 | 1942 | if (!sec_desc->data || sec_desc->data->d_size == 0) |
47ea7417 JH |
1943 | continue; |
1944 | ||
25bbbd7a AN |
1945 | switch (sec_desc->sec_type) { |
1946 | case SEC_DATA: | |
aed65917 | 1947 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1948 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA, |
aed65917 | 1949 | sec_name, sec_idx, |
25bbbd7a AN |
1950 | sec_desc->data->d_buf, |
1951 | sec_desc->data->d_size); | |
1952 | break; | |
1953 | case SEC_RODATA: | |
1954 | obj->has_rodata = true; | |
aed65917 | 1955 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1956 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA, |
aed65917 | 1957 | sec_name, sec_idx, |
25bbbd7a AN |
1958 | sec_desc->data->d_buf, |
1959 | sec_desc->data->d_size); | |
1960 | break; | |
1961 | case SEC_BSS: | |
aed65917 | 1962 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1963 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS, |
aed65917 | 1964 | sec_name, sec_idx, |
25bbbd7a AN |
1965 | NULL, |
1966 | sec_desc->data->d_size); | |
1967 | break; | |
1968 | default: | |
1969 | /* skip */ | |
1970 | break; | |
1971 | } | |
bf829271 AN |
1972 | if (err) |
1973 | return err; | |
1974 | } | |
1975 | return 0; | |
1976 | } | |
1977 | ||
166750bc AN |
1978 | |
1979 | static struct extern_desc *find_extern_by_name(const struct bpf_object *obj, | |
1980 | const void *name) | |
1981 | { | |
1982 | int i; | |
1983 | ||
1984 | for (i = 0; i < obj->nr_extern; i++) { | |
1985 | if (strcmp(obj->externs[i].name, name) == 0) | |
1986 | return &obj->externs[i]; | |
1987 | } | |
1988 | return NULL; | |
1989 | } | |
1990 | ||
c56e5977 YS |
1991 | static struct extern_desc *find_extern_by_name_with_len(const struct bpf_object *obj, |
1992 | const void *name, int len) | |
1993 | { | |
1994 | const char *ext_name; | |
1995 | int i; | |
1996 | ||
1997 | for (i = 0; i < obj->nr_extern; i++) { | |
1998 | ext_name = obj->externs[i].name; | |
1999 | if (strlen(ext_name) == len && strncmp(ext_name, name, len) == 0) | |
2000 | return &obj->externs[i]; | |
2001 | } | |
2002 | return NULL; | |
2003 | } | |
2004 | ||
2e33efe3 AN |
2005 | static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val, |
2006 | char value) | |
166750bc | 2007 | { |
2e33efe3 AN |
2008 | switch (ext->kcfg.type) { |
2009 | case KCFG_BOOL: | |
166750bc | 2010 | if (value == 'm') { |
55d00c37 | 2011 | pr_warn("extern (kcfg) '%s': value '%c' implies tristate or char type\n", |
166750bc AN |
2012 | ext->name, value); |
2013 | return -EINVAL; | |
2014 | } | |
2015 | *(bool *)ext_val = value == 'y' ? true : false; | |
2016 | break; | |
2e33efe3 | 2017 | case KCFG_TRISTATE: |
166750bc AN |
2018 | if (value == 'y') |
2019 | *(enum libbpf_tristate *)ext_val = TRI_YES; | |
2020 | else if (value == 'm') | |
2021 | *(enum libbpf_tristate *)ext_val = TRI_MODULE; | |
2022 | else /* value == 'n' */ | |
2023 | *(enum libbpf_tristate *)ext_val = TRI_NO; | |
2024 | break; | |
2e33efe3 | 2025 | case KCFG_CHAR: |
166750bc AN |
2026 | *(char *)ext_val = value; |
2027 | break; | |
2e33efe3 AN |
2028 | case KCFG_UNKNOWN: |
2029 | case KCFG_INT: | |
2030 | case KCFG_CHAR_ARR: | |
166750bc | 2031 | default: |
55d00c37 | 2032 | pr_warn("extern (kcfg) '%s': value '%c' implies bool, tristate, or char type\n", |
166750bc AN |
2033 | ext->name, value); |
2034 | return -EINVAL; | |
2035 | } | |
2036 | ext->is_set = true; | |
2037 | return 0; | |
2038 | } | |
2039 | ||
2e33efe3 AN |
2040 | static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val, |
2041 | const char *value) | |
166750bc AN |
2042 | { |
2043 | size_t len; | |
2044 | ||
2e33efe3 | 2045 | if (ext->kcfg.type != KCFG_CHAR_ARR) { |
55d00c37 AN |
2046 | pr_warn("extern (kcfg) '%s': value '%s' implies char array type\n", |
2047 | ext->name, value); | |
166750bc AN |
2048 | return -EINVAL; |
2049 | } | |
2050 | ||
2051 | len = strlen(value); | |
2052 | if (value[len - 1] != '"') { | |
2e33efe3 | 2053 | pr_warn("extern (kcfg) '%s': invalid string config '%s'\n", |
166750bc AN |
2054 | ext->name, value); |
2055 | return -EINVAL; | |
2056 | } | |
2057 | ||
2058 | /* strip quotes */ | |
2059 | len -= 2; | |
2e33efe3 | 2060 | if (len >= ext->kcfg.sz) { |
55d00c37 | 2061 | pr_warn("extern (kcfg) '%s': long string '%s' of (%zu bytes) truncated to %d bytes\n", |
2e33efe3 AN |
2062 | ext->name, value, len, ext->kcfg.sz - 1); |
2063 | len = ext->kcfg.sz - 1; | |
166750bc AN |
2064 | } |
2065 | memcpy(ext_val, value + 1, len); | |
2066 | ext_val[len] = '\0'; | |
2067 | ext->is_set = true; | |
2068 | return 0; | |
2069 | } | |
2070 | ||
2071 | static int parse_u64(const char *value, __u64 *res) | |
2072 | { | |
2073 | char *value_end; | |
2074 | int err; | |
2075 | ||
2076 | errno = 0; | |
2077 | *res = strtoull(value, &value_end, 0); | |
2078 | if (errno) { | |
2079 | err = -errno; | |
2080 | pr_warn("failed to parse '%s' as integer: %d\n", value, err); | |
2081 | return err; | |
2082 | } | |
2083 | if (*value_end) { | |
2084 | pr_warn("failed to parse '%s' as integer completely\n", value); | |
2085 | return -EINVAL; | |
2086 | } | |
2087 | return 0; | |
2088 | } | |
2089 | ||
2e33efe3 | 2090 | static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v) |
166750bc | 2091 | { |
2e33efe3 | 2092 | int bit_sz = ext->kcfg.sz * 8; |
166750bc | 2093 | |
2e33efe3 | 2094 | if (ext->kcfg.sz == 8) |
166750bc AN |
2095 | return true; |
2096 | ||
2097 | /* Validate that value stored in u64 fits in integer of `ext->sz` | |
2098 | * bytes size without any loss of information. If the target integer | |
2099 | * is signed, we rely on the following limits of integer type of | |
2100 | * Y bits and subsequent transformation: | |
2101 | * | |
2102 | * -2^(Y-1) <= X <= 2^(Y-1) - 1 | |
2103 | * 0 <= X + 2^(Y-1) <= 2^Y - 1 | |
2104 | * 0 <= X + 2^(Y-1) < 2^Y | |
2105 | * | |
2106 | * For unsigned target integer, check that all the (64 - Y) bits are | |
2107 | * zero. | |
2108 | */ | |
2e33efe3 | 2109 | if (ext->kcfg.is_signed) |
166750bc AN |
2110 | return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz); |
2111 | else | |
2112 | return (v >> bit_sz) == 0; | |
2113 | } | |
2114 | ||
2e33efe3 AN |
2115 | static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val, |
2116 | __u64 value) | |
166750bc | 2117 | { |
55d00c37 AN |
2118 | if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR && |
2119 | ext->kcfg.type != KCFG_BOOL) { | |
2120 | pr_warn("extern (kcfg) '%s': value '%llu' implies integer, char, or boolean type\n", | |
7745ff98 | 2121 | ext->name, (unsigned long long)value); |
166750bc AN |
2122 | return -EINVAL; |
2123 | } | |
55d00c37 AN |
2124 | if (ext->kcfg.type == KCFG_BOOL && value > 1) { |
2125 | pr_warn("extern (kcfg) '%s': value '%llu' isn't boolean compatible\n", | |
2126 | ext->name, (unsigned long long)value); | |
2127 | return -EINVAL; | |
2128 | ||
2129 | } | |
2e33efe3 | 2130 | if (!is_kcfg_value_in_range(ext, value)) { |
55d00c37 | 2131 | pr_warn("extern (kcfg) '%s': value '%llu' doesn't fit in %d bytes\n", |
2e33efe3 | 2132 | ext->name, (unsigned long long)value, ext->kcfg.sz); |
166750bc AN |
2133 | return -ERANGE; |
2134 | } | |
2e33efe3 | 2135 | switch (ext->kcfg.sz) { |
e3ba8e4e KM |
2136 | case 1: |
2137 | *(__u8 *)ext_val = value; | |
2138 | break; | |
2139 | case 2: | |
2140 | *(__u16 *)ext_val = value; | |
2141 | break; | |
2142 | case 4: | |
2143 | *(__u32 *)ext_val = value; | |
2144 | break; | |
2145 | case 8: | |
2146 | *(__u64 *)ext_val = value; | |
2147 | break; | |
2148 | default: | |
2149 | return -EINVAL; | |
166750bc AN |
2150 | } |
2151 | ext->is_set = true; | |
2152 | return 0; | |
2153 | } | |
2154 | ||
8601fd42 AN |
2155 | static int bpf_object__process_kconfig_line(struct bpf_object *obj, |
2156 | char *buf, void *data) | |
166750bc | 2157 | { |
166750bc | 2158 | struct extern_desc *ext; |
8601fd42 | 2159 | char *sep, *value; |
166750bc AN |
2160 | int len, err = 0; |
2161 | void *ext_val; | |
2162 | __u64 num; | |
166750bc | 2163 | |
13d35a0c | 2164 | if (!str_has_pfx(buf, "CONFIG_")) |
8601fd42 | 2165 | return 0; |
166750bc | 2166 | |
8601fd42 AN |
2167 | sep = strchr(buf, '='); |
2168 | if (!sep) { | |
2169 | pr_warn("failed to parse '%s': no separator\n", buf); | |
2170 | return -EINVAL; | |
2171 | } | |
2172 | ||
2173 | /* Trim ending '\n' */ | |
2174 | len = strlen(buf); | |
2175 | if (buf[len - 1] == '\n') | |
2176 | buf[len - 1] = '\0'; | |
2177 | /* Split on '=' and ensure that a value is present. */ | |
2178 | *sep = '\0'; | |
2179 | if (!sep[1]) { | |
2180 | *sep = '='; | |
2181 | pr_warn("failed to parse '%s': no value\n", buf); | |
2182 | return -EINVAL; | |
2183 | } | |
2184 | ||
2185 | ext = find_extern_by_name(obj, buf); | |
2186 | if (!ext || ext->is_set) | |
2187 | return 0; | |
2188 | ||
2e33efe3 | 2189 | ext_val = data + ext->kcfg.data_off; |
8601fd42 AN |
2190 | value = sep + 1; |
2191 | ||
2192 | switch (*value) { | |
2193 | case 'y': case 'n': case 'm': | |
2e33efe3 | 2194 | err = set_kcfg_value_tri(ext, ext_val, *value); |
8601fd42 AN |
2195 | break; |
2196 | case '"': | |
2e33efe3 | 2197 | err = set_kcfg_value_str(ext, ext_val, value); |
8601fd42 AN |
2198 | break; |
2199 | default: | |
2200 | /* assume integer */ | |
2201 | err = parse_u64(value, &num); | |
2202 | if (err) { | |
55d00c37 | 2203 | pr_warn("extern (kcfg) '%s': value '%s' isn't a valid integer\n", ext->name, value); |
8601fd42 AN |
2204 | return err; |
2205 | } | |
55d00c37 AN |
2206 | if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) { |
2207 | pr_warn("extern (kcfg) '%s': value '%s' implies integer type\n", ext->name, value); | |
2208 | return -EINVAL; | |
2209 | } | |
2e33efe3 | 2210 | err = set_kcfg_value_num(ext, ext_val, num); |
8601fd42 | 2211 | break; |
166750bc | 2212 | } |
8601fd42 AN |
2213 | if (err) |
2214 | return err; | |
55d00c37 | 2215 | pr_debug("extern (kcfg) '%s': set to %s\n", ext->name, value); |
8601fd42 AN |
2216 | return 0; |
2217 | } | |
2218 | ||
2219 | static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data) | |
2220 | { | |
2221 | char buf[PATH_MAX]; | |
2222 | struct utsname uts; | |
2223 | int len, err = 0; | |
2224 | gzFile file; | |
2225 | ||
2226 | uname(&uts); | |
2227 | len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release); | |
2228 | if (len < 0) | |
2229 | return -EINVAL; | |
2230 | else if (len >= PATH_MAX) | |
2231 | return -ENAMETOOLONG; | |
2232 | ||
2233 | /* gzopen also accepts uncompressed files. */ | |
8e50750f | 2234 | file = gzopen(buf, "re"); |
8601fd42 | 2235 | if (!file) |
8e50750f | 2236 | file = gzopen("/proc/config.gz", "re"); |
8601fd42 | 2237 | |
166750bc | 2238 | if (!file) { |
8601fd42 | 2239 | pr_warn("failed to open system Kconfig\n"); |
166750bc AN |
2240 | return -ENOENT; |
2241 | } | |
2242 | ||
2243 | while (gzgets(file, buf, sizeof(buf))) { | |
8601fd42 AN |
2244 | err = bpf_object__process_kconfig_line(obj, buf, data); |
2245 | if (err) { | |
2246 | pr_warn("error parsing system Kconfig line '%s': %d\n", | |
2247 | buf, err); | |
166750bc AN |
2248 | goto out; |
2249 | } | |
8601fd42 | 2250 | } |
166750bc | 2251 | |
8601fd42 AN |
2252 | out: |
2253 | gzclose(file); | |
2254 | return err; | |
2255 | } | |
166750bc | 2256 | |
8601fd42 AN |
2257 | static int bpf_object__read_kconfig_mem(struct bpf_object *obj, |
2258 | const char *config, void *data) | |
2259 | { | |
2260 | char buf[PATH_MAX]; | |
2261 | int err = 0; | |
2262 | FILE *file; | |
166750bc | 2263 | |
8601fd42 AN |
2264 | file = fmemopen((void *)config, strlen(config), "r"); |
2265 | if (!file) { | |
2266 | err = -errno; | |
2267 | pr_warn("failed to open in-memory Kconfig: %d\n", err); | |
2268 | return err; | |
2269 | } | |
2270 | ||
2271 | while (fgets(buf, sizeof(buf), file)) { | |
2272 | err = bpf_object__process_kconfig_line(obj, buf, data); | |
2273 | if (err) { | |
2274 | pr_warn("error parsing in-memory Kconfig line '%s': %d\n", | |
2275 | buf, err); | |
166750bc AN |
2276 | break; |
2277 | } | |
166750bc AN |
2278 | } |
2279 | ||
8601fd42 | 2280 | fclose(file); |
166750bc AN |
2281 | return err; |
2282 | } | |
2283 | ||
81bfdd08 | 2284 | static int bpf_object__init_kconfig_map(struct bpf_object *obj) |
166750bc | 2285 | { |
2e33efe3 | 2286 | struct extern_desc *last_ext = NULL, *ext; |
166750bc | 2287 | size_t map_sz; |
2e33efe3 | 2288 | int i, err; |
166750bc | 2289 | |
2e33efe3 AN |
2290 | for (i = 0; i < obj->nr_extern; i++) { |
2291 | ext = &obj->externs[i]; | |
2292 | if (ext->type == EXT_KCFG) | |
2293 | last_ext = ext; | |
2294 | } | |
166750bc | 2295 | |
2e33efe3 AN |
2296 | if (!last_ext) |
2297 | return 0; | |
166750bc | 2298 | |
2e33efe3 | 2299 | map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz; |
81bfdd08 | 2300 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG, |
aed65917 | 2301 | ".kconfig", obj->efile.symbols_shndx, |
166750bc AN |
2302 | NULL, map_sz); |
2303 | if (err) | |
2304 | return err; | |
2305 | ||
81bfdd08 | 2306 | obj->kconfig_map_idx = obj->nr_maps - 1; |
166750bc AN |
2307 | |
2308 | return 0; | |
2309 | } | |
2310 | ||
42869d28 | 2311 | const struct btf_type * |
ddc7c304 | 2312 | skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id) |
abd29c93 AN |
2313 | { |
2314 | const struct btf_type *t = btf__type_by_id(btf, id); | |
8837fe5d | 2315 | |
ddc7c304 AN |
2316 | if (res_id) |
2317 | *res_id = id; | |
2318 | ||
2319 | while (btf_is_mod(t) || btf_is_typedef(t)) { | |
2320 | if (res_id) | |
2321 | *res_id = t->type; | |
2322 | t = btf__type_by_id(btf, t->type); | |
abd29c93 | 2323 | } |
ddc7c304 AN |
2324 | |
2325 | return t; | |
abd29c93 AN |
2326 | } |
2327 | ||
590a0088 MKL |
2328 | static const struct btf_type * |
2329 | resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id) | |
2330 | { | |
2331 | const struct btf_type *t; | |
2332 | ||
2333 | t = skip_mods_and_typedefs(btf, id, NULL); | |
2334 | if (!btf_is_ptr(t)) | |
2335 | return NULL; | |
2336 | ||
2337 | t = skip_mods_and_typedefs(btf, t->type, res_id); | |
2338 | ||
2339 | return btf_is_func_proto(t) ? t : NULL; | |
2340 | } | |
2341 | ||
774e132e | 2342 | static const char *__btf_kind_str(__u16 kind) |
81ba0889 | 2343 | { |
774e132e | 2344 | switch (kind) { |
81ba0889 AN |
2345 | case BTF_KIND_UNKN: return "void"; |
2346 | case BTF_KIND_INT: return "int"; | |
2347 | case BTF_KIND_PTR: return "ptr"; | |
2348 | case BTF_KIND_ARRAY: return "array"; | |
2349 | case BTF_KIND_STRUCT: return "struct"; | |
2350 | case BTF_KIND_UNION: return "union"; | |
2351 | case BTF_KIND_ENUM: return "enum"; | |
2352 | case BTF_KIND_FWD: return "fwd"; | |
2353 | case BTF_KIND_TYPEDEF: return "typedef"; | |
2354 | case BTF_KIND_VOLATILE: return "volatile"; | |
2355 | case BTF_KIND_CONST: return "const"; | |
2356 | case BTF_KIND_RESTRICT: return "restrict"; | |
2357 | case BTF_KIND_FUNC: return "func"; | |
2358 | case BTF_KIND_FUNC_PROTO: return "func_proto"; | |
2359 | case BTF_KIND_VAR: return "var"; | |
2360 | case BTF_KIND_DATASEC: return "datasec"; | |
22541a9e | 2361 | case BTF_KIND_FLOAT: return "float"; |
223f903e | 2362 | case BTF_KIND_DECL_TAG: return "decl_tag"; |
2dc1e488 | 2363 | case BTF_KIND_TYPE_TAG: return "type_tag"; |
f2a62588 | 2364 | case BTF_KIND_ENUM64: return "enum64"; |
81ba0889 AN |
2365 | default: return "unknown"; |
2366 | } | |
2367 | } | |
2368 | ||
42869d28 | 2369 | const char *btf_kind_str(const struct btf_type *t) |
774e132e MKL |
2370 | { |
2371 | return __btf_kind_str(btf_kind(t)); | |
2372 | } | |
2373 | ||
ef99b02b AN |
2374 | /* |
2375 | * Fetch integer attribute of BTF map definition. Such attributes are | |
2376 | * represented using a pointer to an array, in which dimensionality of array | |
2377 | * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY]; | |
2378 | * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF | |
2379 | * type definition, while using only sizeof(void *) space in ELF data section. | |
2380 | */ | |
2381 | static bool get_map_field_int(const char *map_name, const struct btf *btf, | |
8983b731 AN |
2382 | const struct btf_member *m, __u32 *res) |
2383 | { | |
ddc7c304 | 2384 | const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL); |
abd29c93 | 2385 | const char *name = btf__name_by_offset(btf, m->name_off); |
ef99b02b AN |
2386 | const struct btf_array *arr_info; |
2387 | const struct btf_type *arr_t; | |
abd29c93 | 2388 | |
b03bc685 | 2389 | if (!btf_is_ptr(t)) { |
81ba0889 AN |
2390 | pr_warn("map '%s': attr '%s': expected PTR, got %s.\n", |
2391 | map_name, name, btf_kind_str(t)); | |
abd29c93 AN |
2392 | return false; |
2393 | } | |
ef99b02b AN |
2394 | |
2395 | arr_t = btf__type_by_id(btf, t->type); | |
2396 | if (!arr_t) { | |
be18010e KW |
2397 | pr_warn("map '%s': attr '%s': type [%u] not found.\n", |
2398 | map_name, name, t->type); | |
abd29c93 AN |
2399 | return false; |
2400 | } | |
b03bc685 | 2401 | if (!btf_is_array(arr_t)) { |
81ba0889 AN |
2402 | pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n", |
2403 | map_name, name, btf_kind_str(arr_t)); | |
abd29c93 AN |
2404 | return false; |
2405 | } | |
b03bc685 | 2406 | arr_info = btf_array(arr_t); |
ef99b02b | 2407 | *res = arr_info->nelems; |
abd29c93 AN |
2408 | return true; |
2409 | } | |
2410 | ||
d147357e AS |
2411 | static bool get_map_field_long(const char *map_name, const struct btf *btf, |
2412 | const struct btf_member *m, __u64 *res) | |
2413 | { | |
2414 | const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL); | |
2415 | const char *name = btf__name_by_offset(btf, m->name_off); | |
2416 | ||
2417 | if (btf_is_ptr(t)) { | |
2418 | __u32 res32; | |
2419 | bool ret; | |
2420 | ||
2421 | ret = get_map_field_int(map_name, btf, m, &res32); | |
2422 | if (ret) | |
2423 | *res = (__u64)res32; | |
2424 | return ret; | |
2425 | } | |
2426 | ||
2427 | if (!btf_is_enum(t) && !btf_is_enum64(t)) { | |
2428 | pr_warn("map '%s': attr '%s': expected ENUM or ENUM64, got %s.\n", | |
2429 | map_name, name, btf_kind_str(t)); | |
2430 | return false; | |
2431 | } | |
2432 | ||
2433 | if (btf_vlen(t) != 1) { | |
2434 | pr_warn("map '%s': attr '%s': invalid __ulong\n", | |
2435 | map_name, name); | |
2436 | return false; | |
2437 | } | |
2438 | ||
2439 | if (btf_is_enum(t)) { | |
2440 | const struct btf_enum *e = btf_enum(t); | |
2441 | ||
2442 | *res = e->val; | |
2443 | } else { | |
2444 | const struct btf_enum64 *e = btf_enum64(t); | |
2445 | ||
2446 | *res = btf_enum64_value(e); | |
2447 | } | |
2448 | return true; | |
2449 | } | |
2450 | ||
e588c116 WY |
2451 | static int pathname_concat(char *buf, size_t buf_sz, const char *path, const char *name) |
2452 | { | |
2453 | int len; | |
2454 | ||
2455 | len = snprintf(buf, buf_sz, "%s/%s", path, name); | |
2456 | if (len < 0) | |
2457 | return -EINVAL; | |
2458 | if (len >= buf_sz) | |
2459 | return -ENAMETOOLONG; | |
2460 | ||
2461 | return 0; | |
2462 | } | |
2463 | ||
57a00f41 THJ |
2464 | static int build_map_pin_path(struct bpf_map *map, const char *path) |
2465 | { | |
2466 | char buf[PATH_MAX]; | |
e588c116 | 2467 | int err; |
57a00f41 THJ |
2468 | |
2469 | if (!path) | |
6b434b61 | 2470 | path = BPF_FS_DEFAULT_PATH; |
57a00f41 | 2471 | |
e588c116 WY |
2472 | err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map)); |
2473 | if (err) | |
2474 | return err; | |
57a00f41 | 2475 | |
6e9cab2e | 2476 | return bpf_map__set_pin_path(map, buf); |
57a00f41 THJ |
2477 | } |
2478 | ||
146bf811 AN |
2479 | /* should match definition in bpf_helpers.h */ |
2480 | enum libbpf_pin_type { | |
2481 | LIBBPF_PIN_NONE, | |
2482 | /* PIN_BY_NAME: pin maps by name (in /sys/fs/bpf by default) */ | |
2483 | LIBBPF_PIN_BY_NAME, | |
2484 | }; | |
2485 | ||
c7ef5ec9 AN |
2486 | int parse_btf_map_def(const char *map_name, struct btf *btf, |
2487 | const struct btf_type *def_t, bool strict, | |
2488 | struct btf_map_def *map_def, struct btf_map_def *inner_def) | |
abd29c93 | 2489 | { |
41017e56 | 2490 | const struct btf_type *t; |
abd29c93 | 2491 | const struct btf_member *m; |
c7ef5ec9 | 2492 | bool is_inner = inner_def == NULL; |
abd29c93 AN |
2493 | int vlen, i; |
2494 | ||
c7ef5ec9 AN |
2495 | vlen = btf_vlen(def_t); |
2496 | m = btf_members(def_t); | |
abd29c93 | 2497 | for (i = 0; i < vlen; i++, m++) { |
c7ef5ec9 | 2498 | const char *name = btf__name_by_offset(btf, m->name_off); |
abd29c93 AN |
2499 | |
2500 | if (!name) { | |
c7ef5ec9 | 2501 | pr_warn("map '%s': invalid field #%d.\n", map_name, i); |
abd29c93 AN |
2502 | return -EINVAL; |
2503 | } | |
2504 | if (strcmp(name, "type") == 0) { | |
c7ef5ec9 | 2505 | if (!get_map_field_int(map_name, btf, m, &map_def->map_type)) |
abd29c93 | 2506 | return -EINVAL; |
c7ef5ec9 | 2507 | map_def->parts |= MAP_DEF_MAP_TYPE; |
abd29c93 | 2508 | } else if (strcmp(name, "max_entries") == 0) { |
c7ef5ec9 | 2509 | if (!get_map_field_int(map_name, btf, m, &map_def->max_entries)) |
abd29c93 | 2510 | return -EINVAL; |
c7ef5ec9 | 2511 | map_def->parts |= MAP_DEF_MAX_ENTRIES; |
abd29c93 | 2512 | } else if (strcmp(name, "map_flags") == 0) { |
c7ef5ec9 | 2513 | if (!get_map_field_int(map_name, btf, m, &map_def->map_flags)) |
abd29c93 | 2514 | return -EINVAL; |
c7ef5ec9 | 2515 | map_def->parts |= MAP_DEF_MAP_FLAGS; |
1bdb6c9a | 2516 | } else if (strcmp(name, "numa_node") == 0) { |
c7ef5ec9 | 2517 | if (!get_map_field_int(map_name, btf, m, &map_def->numa_node)) |
1bdb6c9a | 2518 | return -EINVAL; |
c7ef5ec9 | 2519 | map_def->parts |= MAP_DEF_NUMA_NODE; |
abd29c93 AN |
2520 | } else if (strcmp(name, "key_size") == 0) { |
2521 | __u32 sz; | |
2522 | ||
c7ef5ec9 | 2523 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2524 | return -EINVAL; |
c7ef5ec9 | 2525 | if (map_def->key_size && map_def->key_size != sz) { |
be18010e | 2526 | pr_warn("map '%s': conflicting key size %u != %u.\n", |
c7ef5ec9 | 2527 | map_name, map_def->key_size, sz); |
abd29c93 AN |
2528 | return -EINVAL; |
2529 | } | |
c7ef5ec9 AN |
2530 | map_def->key_size = sz; |
2531 | map_def->parts |= MAP_DEF_KEY_SIZE; | |
abd29c93 AN |
2532 | } else if (strcmp(name, "key") == 0) { |
2533 | __s64 sz; | |
2534 | ||
c7ef5ec9 | 2535 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2536 | if (!t) { |
be18010e | 2537 | pr_warn("map '%s': key type [%d] not found.\n", |
c7ef5ec9 | 2538 | map_name, m->type); |
abd29c93 AN |
2539 | return -EINVAL; |
2540 | } | |
b03bc685 | 2541 | if (!btf_is_ptr(t)) { |
81ba0889 | 2542 | pr_warn("map '%s': key spec is not PTR: %s.\n", |
c7ef5ec9 | 2543 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2544 | return -EINVAL; |
2545 | } | |
c7ef5ec9 | 2546 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2547 | if (sz < 0) { |
679152d3 | 2548 | pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n", |
c7ef5ec9 | 2549 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2550 | return sz; |
2551 | } | |
c7ef5ec9 | 2552 | if (map_def->key_size && map_def->key_size != sz) { |
679152d3 | 2553 | pr_warn("map '%s': conflicting key size %u != %zd.\n", |
c7ef5ec9 | 2554 | map_name, map_def->key_size, (ssize_t)sz); |
abd29c93 AN |
2555 | return -EINVAL; |
2556 | } | |
c7ef5ec9 AN |
2557 | map_def->key_size = sz; |
2558 | map_def->key_type_id = t->type; | |
2559 | map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE; | |
abd29c93 AN |
2560 | } else if (strcmp(name, "value_size") == 0) { |
2561 | __u32 sz; | |
2562 | ||
c7ef5ec9 | 2563 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2564 | return -EINVAL; |
c7ef5ec9 | 2565 | if (map_def->value_size && map_def->value_size != sz) { |
be18010e | 2566 | pr_warn("map '%s': conflicting value size %u != %u.\n", |
c7ef5ec9 | 2567 | map_name, map_def->value_size, sz); |
abd29c93 AN |
2568 | return -EINVAL; |
2569 | } | |
c7ef5ec9 AN |
2570 | map_def->value_size = sz; |
2571 | map_def->parts |= MAP_DEF_VALUE_SIZE; | |
abd29c93 AN |
2572 | } else if (strcmp(name, "value") == 0) { |
2573 | __s64 sz; | |
2574 | ||
c7ef5ec9 | 2575 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2576 | if (!t) { |
be18010e | 2577 | pr_warn("map '%s': value type [%d] not found.\n", |
c7ef5ec9 | 2578 | map_name, m->type); |
abd29c93 AN |
2579 | return -EINVAL; |
2580 | } | |
b03bc685 | 2581 | if (!btf_is_ptr(t)) { |
81ba0889 | 2582 | pr_warn("map '%s': value spec is not PTR: %s.\n", |
c7ef5ec9 | 2583 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2584 | return -EINVAL; |
2585 | } | |
c7ef5ec9 | 2586 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2587 | if (sz < 0) { |
679152d3 | 2588 | pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n", |
c7ef5ec9 | 2589 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2590 | return sz; |
2591 | } | |
c7ef5ec9 | 2592 | if (map_def->value_size && map_def->value_size != sz) { |
679152d3 | 2593 | pr_warn("map '%s': conflicting value size %u != %zd.\n", |
c7ef5ec9 | 2594 | map_name, map_def->value_size, (ssize_t)sz); |
abd29c93 AN |
2595 | return -EINVAL; |
2596 | } | |
c7ef5ec9 AN |
2597 | map_def->value_size = sz; |
2598 | map_def->value_type_id = t->type; | |
2599 | map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE; | |
646f02ff AN |
2600 | } |
2601 | else if (strcmp(name, "values") == 0) { | |
341ac5ff HC |
2602 | bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type); |
2603 | bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY; | |
2604 | const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value"; | |
c7ef5ec9 | 2605 | char inner_map_name[128]; |
646f02ff AN |
2606 | int err; |
2607 | ||
2608 | if (is_inner) { | |
2609 | pr_warn("map '%s': multi-level inner maps not supported.\n", | |
c7ef5ec9 | 2610 | map_name); |
646f02ff AN |
2611 | return -ENOTSUP; |
2612 | } | |
2613 | if (i != vlen - 1) { | |
2614 | pr_warn("map '%s': '%s' member should be last.\n", | |
c7ef5ec9 | 2615 | map_name, name); |
646f02ff AN |
2616 | return -EINVAL; |
2617 | } | |
341ac5ff HC |
2618 | if (!is_map_in_map && !is_prog_array) { |
2619 | pr_warn("map '%s': should be map-in-map or prog-array.\n", | |
c7ef5ec9 | 2620 | map_name); |
646f02ff AN |
2621 | return -ENOTSUP; |
2622 | } | |
c7ef5ec9 | 2623 | if (map_def->value_size && map_def->value_size != 4) { |
646f02ff | 2624 | pr_warn("map '%s': conflicting value size %u != 4.\n", |
c7ef5ec9 | 2625 | map_name, map_def->value_size); |
646f02ff AN |
2626 | return -EINVAL; |
2627 | } | |
c7ef5ec9 AN |
2628 | map_def->value_size = 4; |
2629 | t = btf__type_by_id(btf, m->type); | |
646f02ff | 2630 | if (!t) { |
341ac5ff HC |
2631 | pr_warn("map '%s': %s type [%d] not found.\n", |
2632 | map_name, desc, m->type); | |
646f02ff AN |
2633 | return -EINVAL; |
2634 | } | |
2635 | if (!btf_is_array(t) || btf_array(t)->nelems) { | |
341ac5ff HC |
2636 | pr_warn("map '%s': %s spec is not a zero-sized array.\n", |
2637 | map_name, desc); | |
646f02ff AN |
2638 | return -EINVAL; |
2639 | } | |
c7ef5ec9 | 2640 | t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL); |
646f02ff | 2641 | if (!btf_is_ptr(t)) { |
341ac5ff HC |
2642 | pr_warn("map '%s': %s def is of unexpected kind %s.\n", |
2643 | map_name, desc, btf_kind_str(t)); | |
646f02ff AN |
2644 | return -EINVAL; |
2645 | } | |
c7ef5ec9 | 2646 | t = skip_mods_and_typedefs(btf, t->type, NULL); |
341ac5ff HC |
2647 | if (is_prog_array) { |
2648 | if (!btf_is_func_proto(t)) { | |
2649 | pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n", | |
2650 | map_name, btf_kind_str(t)); | |
2651 | return -EINVAL; | |
2652 | } | |
2653 | continue; | |
2654 | } | |
646f02ff | 2655 | if (!btf_is_struct(t)) { |
81ba0889 | 2656 | pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n", |
c7ef5ec9 | 2657 | map_name, btf_kind_str(t)); |
646f02ff AN |
2658 | return -EINVAL; |
2659 | } | |
2660 | ||
c7ef5ec9 AN |
2661 | snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name); |
2662 | err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL); | |
646f02ff AN |
2663 | if (err) |
2664 | return err; | |
c7ef5ec9 AN |
2665 | |
2666 | map_def->parts |= MAP_DEF_INNER_MAP; | |
57a00f41 THJ |
2667 | } else if (strcmp(name, "pinning") == 0) { |
2668 | __u32 val; | |
57a00f41 | 2669 | |
646f02ff | 2670 | if (is_inner) { |
c7ef5ec9 | 2671 | pr_warn("map '%s': inner def can't be pinned.\n", map_name); |
646f02ff AN |
2672 | return -EINVAL; |
2673 | } | |
c7ef5ec9 | 2674 | if (!get_map_field_int(map_name, btf, m, &val)) |
57a00f41 | 2675 | return -EINVAL; |
c7ef5ec9 | 2676 | if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) { |
57a00f41 | 2677 | pr_warn("map '%s': invalid pinning value %u.\n", |
c7ef5ec9 | 2678 | map_name, val); |
57a00f41 THJ |
2679 | return -EINVAL; |
2680 | } | |
c7ef5ec9 AN |
2681 | map_def->pinning = val; |
2682 | map_def->parts |= MAP_DEF_PINNING; | |
47512102 | 2683 | } else if (strcmp(name, "map_extra") == 0) { |
d147357e | 2684 | __u64 map_extra; |
47512102 | 2685 | |
d147357e | 2686 | if (!get_map_field_long(map_name, btf, m, &map_extra)) |
47512102 JK |
2687 | return -EINVAL; |
2688 | map_def->map_extra = map_extra; | |
2689 | map_def->parts |= MAP_DEF_MAP_EXTRA; | |
abd29c93 AN |
2690 | } else { |
2691 | if (strict) { | |
c7ef5ec9 | 2692 | pr_warn("map '%s': unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2693 | return -ENOTSUP; |
2694 | } | |
c7ef5ec9 | 2695 | pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2696 | } |
2697 | } | |
2698 | ||
c7ef5ec9 AN |
2699 | if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) { |
2700 | pr_warn("map '%s': map type isn't specified.\n", map_name); | |
abd29c93 AN |
2701 | return -EINVAL; |
2702 | } | |
2703 | ||
2704 | return 0; | |
2705 | } | |
2706 | ||
597fbc46 AN |
2707 | static size_t adjust_ringbuf_sz(size_t sz) |
2708 | { | |
2709 | __u32 page_sz = sysconf(_SC_PAGE_SIZE); | |
2710 | __u32 mul; | |
2711 | ||
2712 | /* if user forgot to set any size, make sure they see error */ | |
2713 | if (sz == 0) | |
2714 | return 0; | |
2715 | /* Kernel expects BPF_MAP_TYPE_RINGBUF's max_entries to be | |
2716 | * a power-of-2 multiple of kernel's page size. If user diligently | |
2717 | * satisified these conditions, pass the size through. | |
2718 | */ | |
2719 | if ((sz % page_sz) == 0 && is_pow_of_2(sz / page_sz)) | |
2720 | return sz; | |
2721 | ||
2722 | /* Otherwise find closest (page_sz * power_of_2) product bigger than | |
2723 | * user-set size to satisfy both user size request and kernel | |
2724 | * requirements and substitute correct max_entries for map creation. | |
2725 | */ | |
2726 | for (mul = 1; mul <= UINT_MAX / page_sz; mul <<= 1) { | |
2727 | if (mul * page_sz > sz) | |
2728 | return mul * page_sz; | |
2729 | } | |
2730 | ||
2731 | /* if it's impossible to satisfy the conditions (i.e., user size is | |
2732 | * very close to UINT_MAX but is not a power-of-2 multiple of | |
2733 | * page_size) then just return original size and let kernel reject it | |
2734 | */ | |
2735 | return sz; | |
2736 | } | |
2737 | ||
b66ccae0 DV |
2738 | static bool map_is_ringbuf(const struct bpf_map *map) |
2739 | { | |
2740 | return map->def.type == BPF_MAP_TYPE_RINGBUF || | |
2741 | map->def.type == BPF_MAP_TYPE_USER_RINGBUF; | |
2742 | } | |
2743 | ||
c7ef5ec9 AN |
2744 | static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def) |
2745 | { | |
2746 | map->def.type = def->map_type; | |
2747 | map->def.key_size = def->key_size; | |
2748 | map->def.value_size = def->value_size; | |
2749 | map->def.max_entries = def->max_entries; | |
2750 | map->def.map_flags = def->map_flags; | |
47512102 | 2751 | map->map_extra = def->map_extra; |
c7ef5ec9 AN |
2752 | |
2753 | map->numa_node = def->numa_node; | |
2754 | map->btf_key_type_id = def->key_type_id; | |
2755 | map->btf_value_type_id = def->value_type_id; | |
2756 | ||
597fbc46 | 2757 | /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */ |
b66ccae0 | 2758 | if (map_is_ringbuf(map)) |
597fbc46 AN |
2759 | map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries); |
2760 | ||
c7ef5ec9 AN |
2761 | if (def->parts & MAP_DEF_MAP_TYPE) |
2762 | pr_debug("map '%s': found type = %u.\n", map->name, def->map_type); | |
2763 | ||
2764 | if (def->parts & MAP_DEF_KEY_TYPE) | |
2765 | pr_debug("map '%s': found key [%u], sz = %u.\n", | |
2766 | map->name, def->key_type_id, def->key_size); | |
2767 | else if (def->parts & MAP_DEF_KEY_SIZE) | |
2768 | pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size); | |
2769 | ||
2770 | if (def->parts & MAP_DEF_VALUE_TYPE) | |
2771 | pr_debug("map '%s': found value [%u], sz = %u.\n", | |
2772 | map->name, def->value_type_id, def->value_size); | |
2773 | else if (def->parts & MAP_DEF_VALUE_SIZE) | |
2774 | pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size); | |
2775 | ||
2776 | if (def->parts & MAP_DEF_MAX_ENTRIES) | |
2777 | pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries); | |
2778 | if (def->parts & MAP_DEF_MAP_FLAGS) | |
47512102 JK |
2779 | pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags); |
2780 | if (def->parts & MAP_DEF_MAP_EXTRA) | |
2781 | pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name, | |
2782 | (unsigned long long)def->map_extra); | |
c7ef5ec9 AN |
2783 | if (def->parts & MAP_DEF_PINNING) |
2784 | pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning); | |
2785 | if (def->parts & MAP_DEF_NUMA_NODE) | |
2786 | pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node); | |
2787 | ||
2788 | if (def->parts & MAP_DEF_INNER_MAP) | |
2789 | pr_debug("map '%s': found inner map definition.\n", map->name); | |
2790 | } | |
2791 | ||
c1cccec9 AN |
2792 | static const char *btf_var_linkage_str(__u32 linkage) |
2793 | { | |
2794 | switch (linkage) { | |
2795 | case BTF_VAR_STATIC: return "static"; | |
2796 | case BTF_VAR_GLOBAL_ALLOCATED: return "global"; | |
2797 | case BTF_VAR_GLOBAL_EXTERN: return "extern"; | |
2798 | default: return "unknown"; | |
2799 | } | |
2800 | } | |
2801 | ||
41017e56 AN |
2802 | static int bpf_object__init_user_btf_map(struct bpf_object *obj, |
2803 | const struct btf_type *sec, | |
2804 | int var_idx, int sec_idx, | |
2805 | const Elf_Data *data, bool strict, | |
2806 | const char *pin_root_path) | |
2807 | { | |
c7ef5ec9 | 2808 | struct btf_map_def map_def = {}, inner_def = {}; |
41017e56 AN |
2809 | const struct btf_type *var, *def; |
2810 | const struct btf_var_secinfo *vi; | |
2811 | const struct btf_var *var_extra; | |
2812 | const char *map_name; | |
2813 | struct bpf_map *map; | |
c7ef5ec9 | 2814 | int err; |
41017e56 AN |
2815 | |
2816 | vi = btf_var_secinfos(sec) + var_idx; | |
2817 | var = btf__type_by_id(obj->btf, vi->type); | |
2818 | var_extra = btf_var(var); | |
2819 | map_name = btf__name_by_offset(obj->btf, var->name_off); | |
2820 | ||
2821 | if (map_name == NULL || map_name[0] == '\0') { | |
2822 | pr_warn("map #%d: empty name.\n", var_idx); | |
2823 | return -EINVAL; | |
2824 | } | |
2825 | if ((__u64)vi->offset + vi->size > data->d_size) { | |
2826 | pr_warn("map '%s' BTF data is corrupted.\n", map_name); | |
2827 | return -EINVAL; | |
2828 | } | |
2829 | if (!btf_is_var(var)) { | |
81ba0889 AN |
2830 | pr_warn("map '%s': unexpected var kind %s.\n", |
2831 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2832 | return -EINVAL; |
2833 | } | |
c1cccec9 AN |
2834 | if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) { |
2835 | pr_warn("map '%s': unsupported map linkage %s.\n", | |
2836 | map_name, btf_var_linkage_str(var_extra->linkage)); | |
41017e56 AN |
2837 | return -EOPNOTSUPP; |
2838 | } | |
2839 | ||
2840 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
2841 | if (!btf_is_struct(def)) { | |
81ba0889 AN |
2842 | pr_warn("map '%s': unexpected def kind %s.\n", |
2843 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2844 | return -EINVAL; |
2845 | } | |
2846 | if (def->size > vi->size) { | |
2847 | pr_warn("map '%s': invalid def size.\n", map_name); | |
2848 | return -EINVAL; | |
2849 | } | |
2850 | ||
2851 | map = bpf_object__add_map(obj); | |
2852 | if (IS_ERR(map)) | |
2853 | return PTR_ERR(map); | |
2854 | map->name = strdup(map_name); | |
2855 | if (!map->name) { | |
2856 | pr_warn("map '%s': failed to alloc map name.\n", map_name); | |
2857 | return -ENOMEM; | |
2858 | } | |
2859 | map->libbpf_type = LIBBPF_MAP_UNSPEC; | |
2860 | map->def.type = BPF_MAP_TYPE_UNSPEC; | |
2861 | map->sec_idx = sec_idx; | |
2862 | map->sec_offset = vi->offset; | |
646f02ff | 2863 | map->btf_var_idx = var_idx; |
41017e56 AN |
2864 | pr_debug("map '%s': at sec_idx %d, offset %zu.\n", |
2865 | map_name, map->sec_idx, map->sec_offset); | |
2866 | ||
c7ef5ec9 AN |
2867 | err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def); |
2868 | if (err) | |
2869 | return err; | |
2870 | ||
2871 | fill_map_from_def(map, &map_def); | |
2872 | ||
2873 | if (map_def.pinning == LIBBPF_PIN_BY_NAME) { | |
2874 | err = build_map_pin_path(map, pin_root_path); | |
2875 | if (err) { | |
2876 | pr_warn("map '%s': couldn't build pin path.\n", map->name); | |
2877 | return err; | |
2878 | } | |
2879 | } | |
2880 | ||
2881 | if (map_def.parts & MAP_DEF_INNER_MAP) { | |
2882 | map->inner_map = calloc(1, sizeof(*map->inner_map)); | |
2883 | if (!map->inner_map) | |
2884 | return -ENOMEM; | |
dac645b9 AN |
2885 | map->inner_map->fd = create_placeholder_fd(); |
2886 | if (map->inner_map->fd < 0) | |
2887 | return map->inner_map->fd; | |
c7ef5ec9 AN |
2888 | map->inner_map->sec_idx = sec_idx; |
2889 | map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1); | |
2890 | if (!map->inner_map->name) | |
2891 | return -ENOMEM; | |
2892 | sprintf(map->inner_map->name, "%s.inner", map_name); | |
2893 | ||
2894 | fill_map_from_def(map->inner_map, &inner_def); | |
2895 | } | |
2896 | ||
4fcac46c | 2897 | err = map_fill_btf_type_info(obj, map); |
262cfb74 DK |
2898 | if (err) |
2899 | return err; | |
2900 | ||
c7ef5ec9 | 2901 | return 0; |
41017e56 AN |
2902 | } |
2903 | ||
2e7ba4f8 AN |
2904 | static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map, |
2905 | const char *sec_name, int sec_idx, | |
2906 | void *data, size_t data_sz) | |
2907 | { | |
2908 | const long page_sz = sysconf(_SC_PAGE_SIZE); | |
2909 | size_t mmap_sz; | |
2910 | ||
2911 | mmap_sz = bpf_map_mmap_sz(obj->arena_map); | |
2912 | if (roundup(data_sz, page_sz) > mmap_sz) { | |
2913 | pr_warn("elf: sec '%s': declared ARENA map size (%zu) is too small to hold global __arena variables of size %zu\n", | |
2914 | sec_name, mmap_sz, data_sz); | |
2915 | return -E2BIG; | |
2916 | } | |
2917 | ||
2918 | obj->arena_data = malloc(data_sz); | |
2919 | if (!obj->arena_data) | |
2920 | return -ENOMEM; | |
2921 | memcpy(obj->arena_data, data, data_sz); | |
2922 | obj->arena_data_sz = data_sz; | |
2923 | ||
2924 | /* make bpf_map__init_value() work for ARENA maps */ | |
2925 | map->mmaped = obj->arena_data; | |
2926 | ||
2927 | return 0; | |
2928 | } | |
2929 | ||
57a00f41 THJ |
2930 | static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict, |
2931 | const char *pin_root_path) | |
abd29c93 AN |
2932 | { |
2933 | const struct btf_type *sec = NULL; | |
2934 | int nr_types, i, vlen, err; | |
2935 | const struct btf_type *t; | |
2936 | const char *name; | |
2937 | Elf_Data *data; | |
2938 | Elf_Scn *scn; | |
2939 | ||
2940 | if (obj->efile.btf_maps_shndx < 0) | |
2941 | return 0; | |
2942 | ||
88a82120 AN |
2943 | scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx); |
2944 | data = elf_sec_data(obj, scn); | |
abd29c93 | 2945 | if (!scn || !data) { |
88a82120 AN |
2946 | pr_warn("elf: failed to get %s map definitions for %s\n", |
2947 | MAPS_ELF_SEC, obj->path); | |
abd29c93 AN |
2948 | return -EINVAL; |
2949 | } | |
2950 | ||
6a886de0 HC |
2951 | nr_types = btf__type_cnt(obj->btf); |
2952 | for (i = 1; i < nr_types; i++) { | |
abd29c93 | 2953 | t = btf__type_by_id(obj->btf, i); |
b03bc685 | 2954 | if (!btf_is_datasec(t)) |
abd29c93 AN |
2955 | continue; |
2956 | name = btf__name_by_offset(obj->btf, t->name_off); | |
2957 | if (strcmp(name, MAPS_ELF_SEC) == 0) { | |
2958 | sec = t; | |
646f02ff | 2959 | obj->efile.btf_maps_sec_btf_id = i; |
abd29c93 AN |
2960 | break; |
2961 | } | |
2962 | } | |
2963 | ||
2964 | if (!sec) { | |
be18010e | 2965 | pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC); |
abd29c93 AN |
2966 | return -ENOENT; |
2967 | } | |
2968 | ||
b03bc685 | 2969 | vlen = btf_vlen(sec); |
abd29c93 AN |
2970 | for (i = 0; i < vlen; i++) { |
2971 | err = bpf_object__init_user_btf_map(obj, sec, i, | |
2972 | obj->efile.btf_maps_shndx, | |
8983b731 AN |
2973 | data, strict, |
2974 | pin_root_path); | |
abd29c93 AN |
2975 | if (err) |
2976 | return err; | |
2977 | } | |
2978 | ||
2e7ba4f8 AN |
2979 | for (i = 0; i < obj->nr_maps; i++) { |
2980 | struct bpf_map *map = &obj->maps[i]; | |
2981 | ||
2982 | if (map->def.type != BPF_MAP_TYPE_ARENA) | |
2983 | continue; | |
2984 | ||
2985 | if (obj->arena_map) { | |
2986 | pr_warn("map '%s': only single ARENA map is supported (map '%s' is also ARENA)\n", | |
2987 | map->name, obj->arena_map->name); | |
2988 | return -EINVAL; | |
2989 | } | |
2990 | obj->arena_map = map; | |
2991 | ||
2992 | if (obj->efile.arena_data) { | |
2993 | err = init_arena_map_data(obj, map, ARENA_SEC, obj->efile.arena_data_shndx, | |
2994 | obj->efile.arena_data->d_buf, | |
2995 | obj->efile.arena_data->d_size); | |
2996 | if (err) | |
2997 | return err; | |
2998 | } | |
2999 | } | |
3000 | if (obj->efile.arena_data && !obj->arena_map) { | |
3001 | pr_warn("elf: sec '%s': to use global __arena variables the ARENA map should be explicitly declared in SEC(\".maps\")\n", | |
3002 | ARENA_SEC); | |
3003 | return -ENOENT; | |
3004 | } | |
3005 | ||
abd29c93 AN |
3006 | return 0; |
3007 | } | |
3008 | ||
0d13bfce | 3009 | static int bpf_object__init_maps(struct bpf_object *obj, |
01af3bf0 | 3010 | const struct bpf_object_open_opts *opts) |
bf829271 | 3011 | { |
166750bc AN |
3012 | const char *pin_root_path; |
3013 | bool strict; | |
bd054102 | 3014 | int err = 0; |
8837fe5d | 3015 | |
166750bc AN |
3016 | strict = !OPTS_GET(opts, relaxed_maps, false); |
3017 | pin_root_path = OPTS_GET(opts, pin_root_path, NULL); | |
bf829271 | 3018 | |
40e1bcab | 3019 | err = bpf_object__init_user_btf_maps(obj, strict, pin_root_path); |
166750bc | 3020 | err = err ?: bpf_object__init_global_data_maps(obj); |
81bfdd08 | 3021 | err = err ?: bpf_object__init_kconfig_map(obj); |
809a69d6 | 3022 | err = err ?: bpf_object_init_struct_ops(obj); |
bf829271 | 3023 | |
3b3af91c | 3024 | return err; |
561bbcca WN |
3025 | } |
3026 | ||
e3d91b0c JDB |
3027 | static bool section_have_execinstr(struct bpf_object *obj, int idx) |
3028 | { | |
ad23b723 | 3029 | Elf64_Shdr *sh; |
e3d91b0c | 3030 | |
ad23b723 AN |
3031 | sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx)); |
3032 | if (!sh) | |
e3d91b0c JDB |
3033 | return false; |
3034 | ||
ad23b723 | 3035 | return sh->sh_flags & SHF_EXECINSTR; |
e3d91b0c JDB |
3036 | } |
3037 | ||
6ebaa3fb EZ |
3038 | static bool starts_with_qmark(const char *s) |
3039 | { | |
3040 | return s && s[0] == '?'; | |
3041 | } | |
3042 | ||
0f0e55d8 AN |
3043 | static bool btf_needs_sanitization(struct bpf_object *obj) |
3044 | { | |
9ca1f56a AS |
3045 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
3046 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
3047 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
3048 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 3049 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
2dc1e488 | 3050 | bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG); |
f2a62588 | 3051 | bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64); |
6ebaa3fb | 3052 | bool has_qmark_datasec = kernel_supports(obj, FEAT_BTF_QMARK_DATASEC); |
0f0e55d8 | 3053 | |
2dc1e488 | 3054 | return !has_func || !has_datasec || !has_func_global || !has_float || |
6ebaa3fb | 3055 | !has_decl_tag || !has_type_tag || !has_enum64 || !has_qmark_datasec; |
0f0e55d8 AN |
3056 | } |
3057 | ||
f2a62588 | 3058 | static int bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf) |
d7c4b398 | 3059 | { |
9ca1f56a AS |
3060 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
3061 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
3062 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
3063 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 3064 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
2dc1e488 | 3065 | bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG); |
f2a62588 | 3066 | bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64); |
6ebaa3fb | 3067 | bool has_qmark_datasec = kernel_supports(obj, FEAT_BTF_QMARK_DATASEC); |
c49a44b3 | 3068 | int enum64_placeholder_id = 0; |
d7c4b398 AN |
3069 | struct btf_type *t; |
3070 | int i, j, vlen; | |
d7c4b398 | 3071 | |
6a886de0 | 3072 | for (i = 1; i < btf__type_cnt(btf); i++) { |
d7c4b398 | 3073 | t = (struct btf_type *)btf__type_by_id(btf, i); |
d7c4b398 | 3074 | |
223f903e YS |
3075 | if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) { |
3076 | /* replace VAR/DECL_TAG with INT */ | |
d7c4b398 | 3077 | t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0); |
1d4126c4 AN |
3078 | /* |
3079 | * using size = 1 is the safest choice, 4 will be too | |
3080 | * big and cause kernel BTF validation failure if | |
3081 | * original variable took less than 4 bytes | |
3082 | */ | |
3083 | t->size = 1; | |
708852dc | 3084 | *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8); |
b03bc685 | 3085 | } else if (!has_datasec && btf_is_datasec(t)) { |
d7c4b398 | 3086 | /* replace DATASEC with STRUCT */ |
b03bc685 AN |
3087 | const struct btf_var_secinfo *v = btf_var_secinfos(t); |
3088 | struct btf_member *m = btf_members(t); | |
d7c4b398 AN |
3089 | struct btf_type *vt; |
3090 | char *name; | |
3091 | ||
3092 | name = (char *)btf__name_by_offset(btf, t->name_off); | |
3093 | while (*name) { | |
6ebaa3fb | 3094 | if (*name == '.' || *name == '?') |
d7c4b398 AN |
3095 | *name = '_'; |
3096 | name++; | |
3097 | } | |
3098 | ||
b03bc685 | 3099 | vlen = btf_vlen(t); |
d7c4b398 AN |
3100 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen); |
3101 | for (j = 0; j < vlen; j++, v++, m++) { | |
3102 | /* order of field assignments is important */ | |
3103 | m->offset = v->offset * 8; | |
3104 | m->type = v->type; | |
3105 | /* preserve variable name as member name */ | |
3106 | vt = (void *)btf__type_by_id(btf, v->type); | |
3107 | m->name_off = vt->name_off; | |
3108 | } | |
6ebaa3fb EZ |
3109 | } else if (!has_qmark_datasec && btf_is_datasec(t) && |
3110 | starts_with_qmark(btf__name_by_offset(btf, t->name_off))) { | |
3111 | /* replace '?' prefix with '_' for DATASEC names */ | |
3112 | char *name; | |
3113 | ||
3114 | name = (char *)btf__name_by_offset(btf, t->name_off); | |
3115 | if (name[0] == '?') | |
3116 | name[0] = '_'; | |
b03bc685 | 3117 | } else if (!has_func && btf_is_func_proto(t)) { |
d7c4b398 | 3118 | /* replace FUNC_PROTO with ENUM */ |
b03bc685 | 3119 | vlen = btf_vlen(t); |
d7c4b398 AN |
3120 | t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen); |
3121 | t->size = sizeof(__u32); /* kernel enforced */ | |
b03bc685 | 3122 | } else if (!has_func && btf_is_func(t)) { |
d7c4b398 AN |
3123 | /* replace FUNC with TYPEDEF */ |
3124 | t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0); | |
2d3eb67f AS |
3125 | } else if (!has_func_global && btf_is_func(t)) { |
3126 | /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */ | |
3127 | t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0); | |
22541a9e IL |
3128 | } else if (!has_float && btf_is_float(t)) { |
3129 | /* replace FLOAT with an equally-sized empty STRUCT; | |
3130 | * since C compilers do not accept e.g. "float" as a | |
3131 | * valid struct name, make it anonymous | |
3132 | */ | |
3133 | t->name_off = 0; | |
3134 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0); | |
2dc1e488 YS |
3135 | } else if (!has_type_tag && btf_is_type_tag(t)) { |
3136 | /* replace TYPE_TAG with a CONST */ | |
3137 | t->name_off = 0; | |
3138 | t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0); | |
f2a62588 YS |
3139 | } else if (!has_enum64 && btf_is_enum(t)) { |
3140 | /* clear the kflag */ | |
3141 | t->info = btf_type_info(btf_kind(t), btf_vlen(t), false); | |
3142 | } else if (!has_enum64 && btf_is_enum64(t)) { | |
3143 | /* replace ENUM64 with a union */ | |
3144 | struct btf_member *m; | |
3145 | ||
3146 | if (enum64_placeholder_id == 0) { | |
3147 | enum64_placeholder_id = btf__add_int(btf, "enum64_placeholder", 1, 0); | |
3148 | if (enum64_placeholder_id < 0) | |
3149 | return enum64_placeholder_id; | |
3150 | ||
3151 | t = (struct btf_type *)btf__type_by_id(btf, i); | |
3152 | } | |
3153 | ||
3154 | m = btf_members(t); | |
3155 | vlen = btf_vlen(t); | |
3156 | t->info = BTF_INFO_ENC(BTF_KIND_UNION, 0, vlen); | |
3157 | for (j = 0; j < vlen; j++, m++) { | |
3158 | m->type = enum64_placeholder_id; | |
3159 | m->offset = 0; | |
3160 | } | |
e3ba8e4e | 3161 | } |
d7c4b398 | 3162 | } |
f2a62588 YS |
3163 | |
3164 | return 0; | |
d7c4b398 AN |
3165 | } |
3166 | ||
b35f14f4 | 3167 | static bool libbpf_needs_btf(const struct bpf_object *obj) |
abd29c93 | 3168 | { |
b35f14f4 | 3169 | return obj->efile.btf_maps_shndx >= 0 || |
240bf8a5 | 3170 | obj->efile.has_st_ops || |
b35f14f4 AN |
3171 | obj->nr_extern > 0; |
3172 | } | |
3173 | ||
3174 | static bool kernel_needs_btf(const struct bpf_object *obj) | |
3175 | { | |
240bf8a5 | 3176 | return obj->efile.has_st_ops; |
abd29c93 AN |
3177 | } |
3178 | ||
063183bf | 3179 | static int bpf_object__init_btf(struct bpf_object *obj, |
9c6660d0 AN |
3180 | Elf_Data *btf_data, |
3181 | Elf_Data *btf_ext_data) | |
3182 | { | |
b7d7f3e1 | 3183 | int err = -ENOENT; |
9c6660d0 AN |
3184 | |
3185 | if (btf_data) { | |
3186 | obj->btf = btf__new(btf_data->d_buf, btf_data->d_size); | |
e9fc3ce9 AN |
3187 | err = libbpf_get_error(obj->btf); |
3188 | if (err) { | |
b7d7f3e1 | 3189 | obj->btf = NULL; |
e9fc3ce9 | 3190 | pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err); |
9c6660d0 AN |
3191 | goto out; |
3192 | } | |
4c01925f AN |
3193 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
3194 | btf__set_pointer_size(obj->btf, 8); | |
9c6660d0 AN |
3195 | } |
3196 | if (btf_ext_data) { | |
11d5daa8 AN |
3197 | struct btf_ext_info *ext_segs[3]; |
3198 | int seg_num, sec_num; | |
3199 | ||
9c6660d0 AN |
3200 | if (!obj->btf) { |
3201 | pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n", | |
3202 | BTF_EXT_ELF_SEC, BTF_ELF_SEC); | |
3203 | goto out; | |
3204 | } | |
e9fc3ce9 AN |
3205 | obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size); |
3206 | err = libbpf_get_error(obj->btf_ext); | |
3207 | if (err) { | |
3208 | pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n", | |
3209 | BTF_EXT_ELF_SEC, err); | |
9c6660d0 AN |
3210 | obj->btf_ext = NULL; |
3211 | goto out; | |
3212 | } | |
11d5daa8 AN |
3213 | |
3214 | /* setup .BTF.ext to ELF section mapping */ | |
3215 | ext_segs[0] = &obj->btf_ext->func_info; | |
3216 | ext_segs[1] = &obj->btf_ext->line_info; | |
3217 | ext_segs[2] = &obj->btf_ext->core_relo_info; | |
3218 | for (seg_num = 0; seg_num < ARRAY_SIZE(ext_segs); seg_num++) { | |
3219 | struct btf_ext_info *seg = ext_segs[seg_num]; | |
3220 | const struct btf_ext_info_sec *sec; | |
3221 | const char *sec_name; | |
3222 | Elf_Scn *scn; | |
3223 | ||
3224 | if (seg->sec_cnt == 0) | |
3225 | continue; | |
3226 | ||
3227 | seg->sec_idxs = calloc(seg->sec_cnt, sizeof(*seg->sec_idxs)); | |
3228 | if (!seg->sec_idxs) { | |
3229 | err = -ENOMEM; | |
3230 | goto out; | |
3231 | } | |
3232 | ||
3233 | sec_num = 0; | |
3234 | for_each_btf_ext_sec(seg, sec) { | |
3235 | /* preventively increment index to avoid doing | |
3236 | * this before every continue below | |
3237 | */ | |
3238 | sec_num++; | |
3239 | ||
3240 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); | |
3241 | if (str_is_empty(sec_name)) | |
3242 | continue; | |
3243 | scn = elf_sec_by_name(obj, sec_name); | |
3244 | if (!scn) | |
3245 | continue; | |
3246 | ||
3247 | seg->sec_idxs[sec_num - 1] = elf_ndxscn(scn); | |
3248 | } | |
3249 | } | |
9c6660d0 AN |
3250 | } |
3251 | out: | |
b35f14f4 | 3252 | if (err && libbpf_needs_btf(obj)) { |
be18010e | 3253 | pr_warn("BTF is required, but is missing or corrupted.\n"); |
b7d7f3e1 | 3254 | return err; |
abd29c93 | 3255 | } |
9c6660d0 AN |
3256 | return 0; |
3257 | } | |
3258 | ||
b96c07f3 AN |
3259 | static int compare_vsi_off(const void *_a, const void *_b) |
3260 | { | |
3261 | const struct btf_var_secinfo *a = _a; | |
3262 | const struct btf_var_secinfo *b = _b; | |
3263 | ||
3264 | return a->offset - b->offset; | |
3265 | } | |
3266 | ||
3267 | static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf, | |
3268 | struct btf_type *t) | |
3269 | { | |
f33f742d AN |
3270 | __u32 size = 0, i, vars = btf_vlen(t); |
3271 | const char *sec_name = btf__name_by_offset(btf, t->name_off); | |
b96c07f3 | 3272 | struct btf_var_secinfo *vsi; |
4fcac46c | 3273 | bool fixup_offsets = false; |
f33f742d | 3274 | int err; |
b96c07f3 | 3275 | |
f33f742d | 3276 | if (!sec_name) { |
b96c07f3 AN |
3277 | pr_debug("No name found in string section for DATASEC kind.\n"); |
3278 | return -ENOENT; | |
3279 | } | |
3280 | ||
4fcac46c AN |
3281 | /* Extern-backing datasecs (.ksyms, .kconfig) have their size and |
3282 | * variable offsets set at the previous step. Further, not every | |
3283 | * extern BTF VAR has corresponding ELF symbol preserved, so we skip | |
3284 | * all fixups altogether for such sections and go straight to sorting | |
3285 | * VARs within their DATASEC. | |
b96c07f3 | 3286 | */ |
4fcac46c | 3287 | if (strcmp(sec_name, KCONFIG_SEC) == 0 || strcmp(sec_name, KSYMS_SEC) == 0) |
b96c07f3 AN |
3288 | goto sort_vars; |
3289 | ||
4fcac46c AN |
3290 | /* Clang leaves DATASEC size and VAR offsets as zeroes, so we need to |
3291 | * fix this up. But BPF static linker already fixes this up and fills | |
3292 | * all the sizes and offsets during static linking. So this step has | |
3293 | * to be optional. But the STV_HIDDEN handling is non-optional for any | |
3294 | * non-extern DATASEC, so the variable fixup loop below handles both | |
3295 | * functions at the same time, paying the cost of BTF VAR <-> ELF | |
3296 | * symbol matching just once. | |
3297 | */ | |
3298 | if (t->size == 0) { | |
3299 | err = find_elf_sec_sz(obj, sec_name, &size); | |
3300 | if (err || !size) { | |
3301 | pr_debug("sec '%s': failed to determine size from ELF: size %u, err %d\n", | |
3302 | sec_name, size, err); | |
3303 | return -ENOENT; | |
3304 | } | |
b96c07f3 | 3305 | |
4fcac46c AN |
3306 | t->size = size; |
3307 | fixup_offsets = true; | |
3308 | } | |
b96c07f3 AN |
3309 | |
3310 | for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) { | |
f33f742d AN |
3311 | const struct btf_type *t_var; |
3312 | struct btf_var *var; | |
3313 | const char *var_name; | |
3314 | Elf64_Sym *sym; | |
3315 | ||
b96c07f3 | 3316 | t_var = btf__type_by_id(btf, vsi->type); |
88918dc1 | 3317 | if (!t_var || !btf_is_var(t_var)) { |
f33f742d | 3318 | pr_debug("sec '%s': unexpected non-VAR type found\n", sec_name); |
b96c07f3 AN |
3319 | return -EINVAL; |
3320 | } | |
3321 | ||
88918dc1 | 3322 | var = btf_var(t_var); |
f33f742d | 3323 | if (var->linkage == BTF_VAR_STATIC || var->linkage == BTF_VAR_GLOBAL_EXTERN) |
b96c07f3 AN |
3324 | continue; |
3325 | ||
f33f742d AN |
3326 | var_name = btf__name_by_offset(btf, t_var->name_off); |
3327 | if (!var_name) { | |
3328 | pr_debug("sec '%s': failed to find name of DATASEC's member #%d\n", | |
3329 | sec_name, i); | |
b96c07f3 AN |
3330 | return -ENOENT; |
3331 | } | |
3332 | ||
f33f742d AN |
3333 | sym = find_elf_var_sym(obj, var_name); |
3334 | if (IS_ERR(sym)) { | |
3335 | pr_debug("sec '%s': failed to find ELF symbol for VAR '%s'\n", | |
3336 | sec_name, var_name); | |
b96c07f3 AN |
3337 | return -ENOENT; |
3338 | } | |
3339 | ||
4fcac46c AN |
3340 | if (fixup_offsets) |
3341 | vsi->offset = sym->st_value; | |
3342 | ||
3343 | /* if variable is a global/weak symbol, but has restricted | |
3344 | * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF VAR | |
3345 | * as static. This follows similar logic for functions (BPF | |
3346 | * subprogs) and influences libbpf's further decisions about | |
3347 | * whether to make global data BPF array maps as | |
3348 | * BPF_F_MMAPABLE. | |
3349 | */ | |
3350 | if (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN | |
3351 | || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL) | |
3352 | var->linkage = BTF_VAR_STATIC; | |
b96c07f3 AN |
3353 | } |
3354 | ||
3355 | sort_vars: | |
3356 | qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off); | |
3357 | return 0; | |
3358 | } | |
3359 | ||
f33f742d | 3360 | static int bpf_object_fixup_btf(struct bpf_object *obj) |
b96c07f3 | 3361 | { |
f33f742d | 3362 | int i, n, err = 0; |
b96c07f3 | 3363 | |
f33f742d AN |
3364 | if (!obj->btf) |
3365 | return 0; | |
3366 | ||
3367 | n = btf__type_cnt(obj->btf); | |
6a886de0 | 3368 | for (i = 1; i < n; i++) { |
f33f742d | 3369 | struct btf_type *t = btf_type_by_id(obj->btf, i); |
b96c07f3 AN |
3370 | |
3371 | /* Loader needs to fix up some of the things compiler | |
3372 | * couldn't get its hands on while emitting BTF. This | |
3373 | * is section size and global variable offset. We use | |
3374 | * the info from the ELF itself for this purpose. | |
3375 | */ | |
3376 | if (btf_is_datasec(t)) { | |
f33f742d | 3377 | err = btf_fixup_datasec(obj, obj->btf, t); |
b96c07f3 | 3378 | if (err) |
f33f742d | 3379 | return err; |
b96c07f3 AN |
3380 | } |
3381 | } | |
3382 | ||
166750bc AN |
3383 | return 0; |
3384 | } | |
3385 | ||
fe62de31 | 3386 | static bool prog_needs_vmlinux_btf(struct bpf_program *prog) |
a6ed02ca | 3387 | { |
1e092a03 KS |
3388 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS || |
3389 | prog->type == BPF_PROG_TYPE_LSM) | |
a6ed02ca KS |
3390 | return true; |
3391 | ||
3392 | /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs | |
3393 | * also need vmlinux BTF | |
3394 | */ | |
3395 | if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd) | |
3396 | return true; | |
3397 | ||
3398 | return false; | |
3399 | } | |
3400 | ||
8b7b0e5f DV |
3401 | static bool map_needs_vmlinux_btf(struct bpf_map *map) |
3402 | { | |
3403 | return bpf_map__is_struct_ops(map); | |
3404 | } | |
3405 | ||
fe62de31 | 3406 | static bool obj_needs_vmlinux_btf(const struct bpf_object *obj) |
a6ed02ca KS |
3407 | { |
3408 | struct bpf_program *prog; | |
8b7b0e5f | 3409 | struct bpf_map *map; |
fe62de31 | 3410 | int i; |
a6ed02ca | 3411 | |
1373ff59 SC |
3412 | /* CO-RE relocations need kernel BTF, only when btf_custom_path |
3413 | * is not specified | |
3414 | */ | |
3415 | if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path) | |
fe62de31 | 3416 | return true; |
192b6638 | 3417 | |
d370bbe1 HL |
3418 | /* Support for typed ksyms needs kernel BTF */ |
3419 | for (i = 0; i < obj->nr_extern; i++) { | |
3420 | const struct extern_desc *ext; | |
3421 | ||
3422 | ext = &obj->externs[i]; | |
fe62de31 AN |
3423 | if (ext->type == EXT_KSYM && ext->ksym.type_id) |
3424 | return true; | |
d370bbe1 HL |
3425 | } |
3426 | ||
a6ed02ca | 3427 | bpf_object__for_each_program(prog, obj) { |
a3820c48 | 3428 | if (!prog->autoload) |
d9297581 | 3429 | continue; |
fe62de31 AN |
3430 | if (prog_needs_vmlinux_btf(prog)) |
3431 | return true; | |
a6ed02ca KS |
3432 | } |
3433 | ||
8b7b0e5f DV |
3434 | bpf_object__for_each_map(map, obj) { |
3435 | if (map_needs_vmlinux_btf(map)) | |
3436 | return true; | |
3437 | } | |
3438 | ||
fe62de31 AN |
3439 | return false; |
3440 | } | |
3441 | ||
3442 | static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force) | |
3443 | { | |
3444 | int err; | |
3445 | ||
3446 | /* btf_vmlinux could be loaded earlier */ | |
67234743 | 3447 | if (obj->btf_vmlinux || obj->gen_loader) |
fe62de31 AN |
3448 | return 0; |
3449 | ||
3450 | if (!force && !obj_needs_vmlinux_btf(obj)) | |
192b6638 AN |
3451 | return 0; |
3452 | ||
a710eed3 | 3453 | obj->btf_vmlinux = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
3454 | err = libbpf_get_error(obj->btf_vmlinux); |
3455 | if (err) { | |
192b6638 AN |
3456 | pr_warn("Error loading vmlinux BTF: %d\n", err); |
3457 | obj->btf_vmlinux = NULL; | |
3458 | return err; | |
3459 | } | |
a6ed02ca KS |
3460 | return 0; |
3461 | } | |
3462 | ||
063183bf AN |
3463 | static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj) |
3464 | { | |
0f0e55d8 AN |
3465 | struct btf *kern_btf = obj->btf; |
3466 | bool btf_mandatory, sanitize; | |
aea28a60 | 3467 | int i, err = 0; |
063183bf AN |
3468 | |
3469 | if (!obj->btf) | |
3470 | return 0; | |
3471 | ||
9ca1f56a | 3472 | if (!kernel_supports(obj, FEAT_BTF)) { |
68b08647 AN |
3473 | if (kernel_needs_btf(obj)) { |
3474 | err = -EOPNOTSUPP; | |
3475 | goto report; | |
3476 | } | |
3477 | pr_debug("Kernel doesn't support BTF, skipping uploading it.\n"); | |
3478 | return 0; | |
3479 | } | |
3480 | ||
aea28a60 AN |
3481 | /* Even though some subprogs are global/weak, user might prefer more |
3482 | * permissive BPF verification process that BPF verifier performs for | |
3483 | * static functions, taking into account more context from the caller | |
3484 | * functions. In such case, they need to mark such subprogs with | |
3485 | * __attribute__((visibility("hidden"))) and libbpf will adjust | |
3486 | * corresponding FUNC BTF type to be marked as static and trigger more | |
3487 | * involved BPF verification process. | |
3488 | */ | |
3489 | for (i = 0; i < obj->nr_programs; i++) { | |
3490 | struct bpf_program *prog = &obj->programs[i]; | |
3491 | struct btf_type *t; | |
3492 | const char *name; | |
3493 | int j, n; | |
3494 | ||
3495 | if (!prog->mark_btf_static || !prog_is_subprog(obj, prog)) | |
3496 | continue; | |
3497 | ||
6a886de0 HC |
3498 | n = btf__type_cnt(obj->btf); |
3499 | for (j = 1; j < n; j++) { | |
aea28a60 AN |
3500 | t = btf_type_by_id(obj->btf, j); |
3501 | if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) | |
3502 | continue; | |
3503 | ||
3504 | name = btf__str_by_offset(obj->btf, t->name_off); | |
3505 | if (strcmp(name, prog->name) != 0) | |
3506 | continue; | |
3507 | ||
3508 | t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0); | |
3509 | break; | |
3510 | } | |
3511 | } | |
3512 | ||
0f0e55d8 AN |
3513 | sanitize = btf_needs_sanitization(obj); |
3514 | if (sanitize) { | |
5c3320d7 | 3515 | const void *raw_data; |
0f0e55d8 | 3516 | __u32 sz; |
063183bf | 3517 | |
0f0e55d8 | 3518 | /* clone BTF to sanitize a copy and leave the original intact */ |
6a886de0 | 3519 | raw_data = btf__raw_data(obj->btf, &sz); |
5c3320d7 | 3520 | kern_btf = btf__new(raw_data, sz); |
e9fc3ce9 AN |
3521 | err = libbpf_get_error(kern_btf); |
3522 | if (err) | |
3523 | return err; | |
04efe591 | 3524 | |
4c01925f AN |
3525 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
3526 | btf__set_pointer_size(obj->btf, 8); | |
f2a62588 YS |
3527 | err = bpf_object__sanitize_btf(obj, kern_btf); |
3528 | if (err) | |
3529 | return err; | |
063183bf | 3530 | } |
0f0e55d8 | 3531 | |
67234743 AS |
3532 | if (obj->gen_loader) { |
3533 | __u32 raw_size = 0; | |
6a886de0 | 3534 | const void *raw_data = btf__raw_data(kern_btf, &raw_size); |
67234743 AS |
3535 | |
3536 | if (!raw_data) | |
3537 | return -ENOMEM; | |
3538 | bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size); | |
3539 | /* Pretend to have valid FD to pass various fd >= 0 checks. | |
3540 | * This fd == 0 will not be used with any syscall and will be reset to -1 eventually. | |
3541 | */ | |
3542 | btf__set_fd(kern_btf, 0); | |
3543 | } else { | |
e0e3ea88 AN |
3544 | /* currently BPF_BTF_LOAD only supports log_level 1 */ |
3545 | err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size, | |
6b434b61 | 3546 | obj->log_level ? 1 : 0, obj->token_fd); |
67234743 | 3547 | } |
0f0e55d8 AN |
3548 | if (sanitize) { |
3549 | if (!err) { | |
3550 | /* move fd to libbpf's BTF */ | |
3551 | btf__set_fd(obj->btf, btf__fd(kern_btf)); | |
3552 | btf__set_fd(kern_btf, -1); | |
3553 | } | |
3554 | btf__free(kern_btf); | |
3555 | } | |
68b08647 | 3556 | report: |
0f0e55d8 AN |
3557 | if (err) { |
3558 | btf_mandatory = kernel_needs_btf(obj); | |
3559 | pr_warn("Error loading .BTF into kernel: %d. %s\n", err, | |
3560 | btf_mandatory ? "BTF is mandatory, can't proceed." | |
3561 | : "BTF is optional, ignoring."); | |
3562 | if (!btf_mandatory) | |
3563 | err = 0; | |
3564 | } | |
3565 | return err; | |
063183bf AN |
3566 | } |
3567 | ||
88a82120 AN |
3568 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off) |
3569 | { | |
3570 | const char *name; | |
3571 | ||
3572 | name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off); | |
3573 | if (!name) { | |
3574 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3575 | off, obj->path, elf_errmsg(-1)); | |
3576 | return NULL; | |
3577 | } | |
3578 | ||
3579 | return name; | |
3580 | } | |
3581 | ||
3582 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off) | |
3583 | { | |
3584 | const char *name; | |
3585 | ||
3586 | name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off); | |
3587 | if (!name) { | |
3588 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3589 | off, obj->path, elf_errmsg(-1)); | |
3590 | return NULL; | |
3591 | } | |
3592 | ||
3593 | return name; | |
3594 | } | |
3595 | ||
3596 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx) | |
3597 | { | |
3598 | Elf_Scn *scn; | |
3599 | ||
3600 | scn = elf_getscn(obj->efile.elf, idx); | |
3601 | if (!scn) { | |
3602 | pr_warn("elf: failed to get section(%zu) from %s: %s\n", | |
3603 | idx, obj->path, elf_errmsg(-1)); | |
3604 | return NULL; | |
3605 | } | |
3606 | return scn; | |
3607 | } | |
3608 | ||
3609 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name) | |
3610 | { | |
3611 | Elf_Scn *scn = NULL; | |
3612 | Elf *elf = obj->efile.elf; | |
3613 | const char *sec_name; | |
3614 | ||
3615 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
3616 | sec_name = elf_sec_name(obj, scn); | |
3617 | if (!sec_name) | |
3618 | return NULL; | |
3619 | ||
3620 | if (strcmp(sec_name, name) != 0) | |
3621 | continue; | |
3622 | ||
3623 | return scn; | |
3624 | } | |
3625 | return NULL; | |
3626 | } | |
3627 | ||
ad23b723 | 3628 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn) |
88a82120 | 3629 | { |
ad23b723 AN |
3630 | Elf64_Shdr *shdr; |
3631 | ||
88a82120 | 3632 | if (!scn) |
ad23b723 | 3633 | return NULL; |
88a82120 | 3634 | |
ad23b723 AN |
3635 | shdr = elf64_getshdr(scn); |
3636 | if (!shdr) { | |
88a82120 AN |
3637 | pr_warn("elf: failed to get section(%zu) header from %s: %s\n", |
3638 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
ad23b723 | 3639 | return NULL; |
88a82120 AN |
3640 | } |
3641 | ||
ad23b723 | 3642 | return shdr; |
88a82120 AN |
3643 | } |
3644 | ||
3645 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn) | |
3646 | { | |
3647 | const char *name; | |
ad23b723 | 3648 | Elf64_Shdr *sh; |
88a82120 AN |
3649 | |
3650 | if (!scn) | |
3651 | return NULL; | |
3652 | ||
ad23b723 AN |
3653 | sh = elf_sec_hdr(obj, scn); |
3654 | if (!sh) | |
88a82120 AN |
3655 | return NULL; |
3656 | ||
ad23b723 | 3657 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 AN |
3658 | if (!name) { |
3659 | pr_warn("elf: failed to get section(%zu) name from %s: %s\n", | |
3660 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
3661 | return NULL; | |
3662 | } | |
3663 | ||
3664 | return name; | |
3665 | } | |
3666 | ||
3667 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn) | |
3668 | { | |
3669 | Elf_Data *data; | |
3670 | ||
3671 | if (!scn) | |
3672 | return NULL; | |
3673 | ||
3674 | data = elf_getdata(scn, 0); | |
3675 | if (!data) { | |
3676 | pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n", | |
3677 | elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>", | |
3678 | obj->path, elf_errmsg(-1)); | |
3679 | return NULL; | |
3680 | } | |
3681 | ||
3682 | return data; | |
3683 | } | |
3684 | ||
ad23b723 AN |
3685 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx) |
3686 | { | |
3687 | if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym)) | |
3688 | return NULL; | |
3689 | ||
3690 | return (Elf64_Sym *)obj->efile.symbols->d_buf + idx; | |
3691 | } | |
3692 | ||
3693 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx) | |
3694 | { | |
3695 | if (idx >= data->d_size / sizeof(Elf64_Rel)) | |
3696 | return NULL; | |
3697 | ||
3698 | return (Elf64_Rel *)data->d_buf + idx; | |
3699 | } | |
3700 | ||
50e09460 AN |
3701 | static bool is_sec_name_dwarf(const char *name) |
3702 | { | |
3703 | /* approximation, but the actual list is too long */ | |
13d35a0c | 3704 | return str_has_pfx(name, ".debug_"); |
50e09460 AN |
3705 | } |
3706 | ||
ad23b723 | 3707 | static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name) |
50e09460 AN |
3708 | { |
3709 | /* no special handling of .strtab */ | |
3710 | if (hdr->sh_type == SHT_STRTAB) | |
3711 | return true; | |
3712 | ||
3713 | /* ignore .llvm_addrsig section as well */ | |
faf6ed32 | 3714 | if (hdr->sh_type == SHT_LLVM_ADDRSIG) |
50e09460 AN |
3715 | return true; |
3716 | ||
3717 | /* no subprograms will lead to an empty .text section, ignore it */ | |
3718 | if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 && | |
3719 | strcmp(name, ".text") == 0) | |
3720 | return true; | |
3721 | ||
3722 | /* DWARF sections */ | |
3723 | if (is_sec_name_dwarf(name)) | |
3724 | return true; | |
3725 | ||
13d35a0c | 3726 | if (str_has_pfx(name, ".rel")) { |
50e09460 AN |
3727 | name += sizeof(".rel") - 1; |
3728 | /* DWARF section relocations */ | |
3729 | if (is_sec_name_dwarf(name)) | |
3730 | return true; | |
3731 | ||
3732 | /* .BTF and .BTF.ext don't need relocations */ | |
3733 | if (strcmp(name, BTF_ELF_SEC) == 0 || | |
3734 | strcmp(name, BTF_EXT_ELF_SEC) == 0) | |
3735 | return true; | |
3736 | } | |
3737 | ||
3738 | return false; | |
3739 | } | |
3740 | ||
db2b8b06 AN |
3741 | static int cmp_progs(const void *_a, const void *_b) |
3742 | { | |
3743 | const struct bpf_program *a = _a; | |
3744 | const struct bpf_program *b = _b; | |
3745 | ||
3746 | if (a->sec_idx != b->sec_idx) | |
3747 | return a->sec_idx < b->sec_idx ? -1 : 1; | |
3748 | ||
3749 | /* sec_insn_off can't be the same within the section */ | |
3750 | return a->sec_insn_off < b->sec_insn_off ? -1 : 1; | |
3751 | } | |
3752 | ||
0d13bfce | 3753 | static int bpf_object__elf_collect(struct bpf_object *obj) |
29603665 | 3754 | { |
25bbbd7a | 3755 | struct elf_sec_desc *sec_desc; |
29603665 | 3756 | Elf *elf = obj->efile.elf; |
f0187f0b | 3757 | Elf_Data *btf_ext_data = NULL; |
1713d68b | 3758 | Elf_Data *btf_data = NULL; |
666810e8 | 3759 | int idx = 0, err = 0; |
0201c575 AN |
3760 | const char *name; |
3761 | Elf_Data *data; | |
3762 | Elf_Scn *scn; | |
ad23b723 | 3763 | Elf64_Shdr *sh; |
29603665 | 3764 | |
0d6988e1 | 3765 | /* ELF section indices are 0-based, but sec #0 is special "invalid" |
51deedc9 SHY |
3766 | * section. Since section count retrieved by elf_getshdrnum() does |
3767 | * include sec #0, it is already the necessary size of an array to keep | |
3768 | * all the sections. | |
25bbbd7a | 3769 | */ |
51deedc9 SHY |
3770 | if (elf_getshdrnum(obj->efile.elf, &obj->efile.sec_cnt)) { |
3771 | pr_warn("elf: failed to get the number of sections for %s: %s\n", | |
3772 | obj->path, elf_errmsg(-1)); | |
3773 | return -LIBBPF_ERRNO__FORMAT; | |
3774 | } | |
25bbbd7a AN |
3775 | obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs)); |
3776 | if (!obj->efile.secs) | |
3777 | return -ENOMEM; | |
29603665 | 3778 | |
0201c575 AN |
3779 | /* a bunch of ELF parsing functionality depends on processing symbols, |
3780 | * so do the first pass and find the symbol table | |
3781 | */ | |
3782 | scn = NULL; | |
29603665 | 3783 | while ((scn = elf_nextscn(elf, scn)) != NULL) { |
ad23b723 AN |
3784 | sh = elf_sec_hdr(obj, scn); |
3785 | if (!sh) | |
0201c575 AN |
3786 | return -LIBBPF_ERRNO__FORMAT; |
3787 | ||
ad23b723 | 3788 | if (sh->sh_type == SHT_SYMTAB) { |
0201c575 AN |
3789 | if (obj->efile.symbols) { |
3790 | pr_warn("elf: multiple symbol tables in %s\n", obj->path); | |
3791 | return -LIBBPF_ERRNO__FORMAT; | |
3792 | } | |
29603665 | 3793 | |
0201c575 AN |
3794 | data = elf_sec_data(obj, scn); |
3795 | if (!data) | |
3796 | return -LIBBPF_ERRNO__FORMAT; | |
3797 | ||
25bbbd7a AN |
3798 | idx = elf_ndxscn(scn); |
3799 | ||
0201c575 | 3800 | obj->efile.symbols = data; |
25bbbd7a | 3801 | obj->efile.symbols_shndx = idx; |
ad23b723 | 3802 | obj->efile.strtabidx = sh->sh_link; |
0201c575 AN |
3803 | } |
3804 | } | |
3805 | ||
03e601f4 THJ |
3806 | if (!obj->efile.symbols) { |
3807 | pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n", | |
3808 | obj->path); | |
3809 | return -ENOENT; | |
3810 | } | |
3811 | ||
0201c575 AN |
3812 | scn = NULL; |
3813 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
25bbbd7a AN |
3814 | idx = elf_ndxscn(scn); |
3815 | sec_desc = &obj->efile.secs[idx]; | |
88a82120 | 3816 | |
ad23b723 AN |
3817 | sh = elf_sec_hdr(obj, scn); |
3818 | if (!sh) | |
01b29d1d | 3819 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3820 | |
ad23b723 | 3821 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 | 3822 | if (!name) |
01b29d1d | 3823 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3824 | |
ad23b723 | 3825 | if (ignore_elf_section(sh, name)) |
50e09460 AN |
3826 | continue; |
3827 | ||
88a82120 AN |
3828 | data = elf_sec_data(obj, scn); |
3829 | if (!data) | |
01b29d1d | 3830 | return -LIBBPF_ERRNO__FORMAT; |
88a82120 AN |
3831 | |
3832 | pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n", | |
077c066a | 3833 | idx, name, (unsigned long)data->d_size, |
ad23b723 AN |
3834 | (int)sh->sh_link, (unsigned long)sh->sh_flags, |
3835 | (int)sh->sh_type); | |
cb1e5e96 | 3836 | |
1713d68b | 3837 | if (strcmp(name, "license") == 0) { |
88a82120 | 3838 | err = bpf_object__init_license(obj, data->d_buf, data->d_size); |
01b29d1d AN |
3839 | if (err) |
3840 | return err; | |
1713d68b | 3841 | } else if (strcmp(name, "version") == 0) { |
88a82120 | 3842 | err = bpf_object__init_kversion(obj, data->d_buf, data->d_size); |
54b8625c JF |
3843 | if (err) |
3844 | return err; | |
1713d68b | 3845 | } else if (strcmp(name, "maps") == 0) { |
e19db676 AN |
3846 | pr_warn("elf: legacy map definitions in 'maps' section are not supported by libbpf v1.0+\n"); |
3847 | return -ENOTSUP; | |
abd29c93 AN |
3848 | } else if (strcmp(name, MAPS_ELF_SEC) == 0) { |
3849 | obj->efile.btf_maps_shndx = idx; | |
1713d68b | 3850 | } else if (strcmp(name, BTF_ELF_SEC) == 0) { |
62554d52 AN |
3851 | if (sh->sh_type != SHT_PROGBITS) |
3852 | return -LIBBPF_ERRNO__FORMAT; | |
1713d68b | 3853 | btf_data = data; |
2993e051 | 3854 | } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) { |
62554d52 AN |
3855 | if (sh->sh_type != SHT_PROGBITS) |
3856 | return -LIBBPF_ERRNO__FORMAT; | |
f0187f0b | 3857 | btf_ext_data = data; |
ad23b723 | 3858 | } else if (sh->sh_type == SHT_SYMTAB) { |
0201c575 | 3859 | /* already processed during the first pass above */ |
ad23b723 AN |
3860 | } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) { |
3861 | if (sh->sh_flags & SHF_EXECINSTR) { | |
f8c7a4d4 JS |
3862 | if (strcmp(name, ".text") == 0) |
3863 | obj->efile.text_shndx = idx; | |
c1122392 | 3864 | err = bpf_object__add_programs(obj, data, name, idx); |
88a82120 | 3865 | if (err) |
01b29d1d | 3866 | return err; |
aed65917 AN |
3867 | } else if (strcmp(name, DATA_SEC) == 0 || |
3868 | str_has_pfx(name, DATA_SEC ".")) { | |
25bbbd7a AN |
3869 | sec_desc->sec_type = SEC_DATA; |
3870 | sec_desc->shdr = sh; | |
3871 | sec_desc->data = data; | |
aed65917 AN |
3872 | } else if (strcmp(name, RODATA_SEC) == 0 || |
3873 | str_has_pfx(name, RODATA_SEC ".")) { | |
25bbbd7a AN |
3874 | sec_desc->sec_type = SEC_RODATA; |
3875 | sec_desc->shdr = sh; | |
3876 | sec_desc->data = data; | |
240bf8a5 | 3877 | } else if (strcmp(name, STRUCT_OPS_SEC) == 0 || |
5ad0ecbe EZ |
3878 | strcmp(name, STRUCT_OPS_LINK_SEC) == 0 || |
3879 | strcmp(name, "?" STRUCT_OPS_SEC) == 0 || | |
3880 | strcmp(name, "?" STRUCT_OPS_LINK_SEC) == 0) { | |
240bf8a5 EZ |
3881 | sec_desc->sec_type = SEC_ST_OPS; |
3882 | sec_desc->shdr = sh; | |
3883 | sec_desc->data = data; | |
3884 | obj->efile.has_st_ops = true; | |
2e7ba4f8 AN |
3885 | } else if (strcmp(name, ARENA_SEC) == 0) { |
3886 | obj->efile.arena_data = data; | |
3887 | obj->efile.arena_data_shndx = idx; | |
d859900c | 3888 | } else { |
50e09460 AN |
3889 | pr_info("elf: skipping unrecognized data section(%d) %s\n", |
3890 | idx, name); | |
a5b8bd47 | 3891 | } |
ad23b723 | 3892 | } else if (sh->sh_type == SHT_REL) { |
25bbbd7a | 3893 | int targ_sec_idx = sh->sh_info; /* points to other section */ |
e3d91b0c | 3894 | |
b7332d28 AN |
3895 | if (sh->sh_entsize != sizeof(Elf64_Rel) || |
3896 | targ_sec_idx >= obj->efile.sec_cnt) | |
3897 | return -LIBBPF_ERRNO__FORMAT; | |
3898 | ||
e3d91b0c | 3899 | /* Only do relo for section with exec instructions */ |
25bbbd7a | 3900 | if (!section_have_execinstr(obj, targ_sec_idx) && |
646f02ff | 3901 | strcmp(name, ".rel" STRUCT_OPS_SEC) && |
809a69d6 | 3902 | strcmp(name, ".rel" STRUCT_OPS_LINK_SEC) && |
5ad0ecbe EZ |
3903 | strcmp(name, ".rel?" STRUCT_OPS_SEC) && |
3904 | strcmp(name, ".rel?" STRUCT_OPS_LINK_SEC) && | |
646f02ff | 3905 | strcmp(name, ".rel" MAPS_ELF_SEC)) { |
50e09460 | 3906 | pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n", |
25bbbd7a AN |
3907 | idx, name, targ_sec_idx, |
3908 | elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>"); | |
e3d91b0c JDB |
3909 | continue; |
3910 | } | |
b62f06e8 | 3911 | |
25bbbd7a AN |
3912 | sec_desc->sec_type = SEC_RELO; |
3913 | sec_desc->shdr = sh; | |
3914 | sec_desc->data = data; | |
dc79f035 AS |
3915 | } else if (sh->sh_type == SHT_NOBITS && (strcmp(name, BSS_SEC) == 0 || |
3916 | str_has_pfx(name, BSS_SEC "."))) { | |
25bbbd7a AN |
3917 | sec_desc->sec_type = SEC_BSS; |
3918 | sec_desc->shdr = sh; | |
3919 | sec_desc->data = data; | |
077c066a | 3920 | } else { |
2e80be60 | 3921 | pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name, |
ad23b723 | 3922 | (size_t)sh->sh_size); |
bec7d68c | 3923 | } |
29603665 | 3924 | } |
561bbcca | 3925 | |
d3a3aa0c | 3926 | if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) { |
88a82120 | 3927 | pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path); |
f102154d | 3928 | return -LIBBPF_ERRNO__FORMAT; |
77ba9a5b | 3929 | } |
db2b8b06 AN |
3930 | |
3931 | /* sort BPF programs by section name and in-section instruction offset | |
e3ba8e4e KM |
3932 | * for faster search |
3933 | */ | |
2a6a9bf2 AN |
3934 | if (obj->nr_programs) |
3935 | qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs); | |
db2b8b06 | 3936 | |
0d13bfce | 3937 | return bpf_object__init_btf(obj, btf_data, btf_ext_data); |
29603665 WN |
3938 | } |
3939 | ||
ad23b723 | 3940 | static bool sym_is_extern(const Elf64_Sym *sym) |
166750bc | 3941 | { |
ad23b723 | 3942 | int bind = ELF64_ST_BIND(sym->st_info); |
166750bc AN |
3943 | /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */ |
3944 | return sym->st_shndx == SHN_UNDEF && | |
3945 | (bind == STB_GLOBAL || bind == STB_WEAK) && | |
ad23b723 | 3946 | ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE; |
166750bc AN |
3947 | } |
3948 | ||
ad23b723 | 3949 | static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx) |
53eddb5e | 3950 | { |
ad23b723 AN |
3951 | int bind = ELF64_ST_BIND(sym->st_info); |
3952 | int type = ELF64_ST_TYPE(sym->st_info); | |
53eddb5e YS |
3953 | |
3954 | /* in .text section */ | |
3955 | if (sym->st_shndx != text_shndx) | |
3956 | return false; | |
3957 | ||
3958 | /* local function */ | |
3959 | if (bind == STB_LOCAL && type == STT_SECTION) | |
3960 | return true; | |
3961 | ||
3962 | /* global function */ | |
3963 | return bind == STB_GLOBAL && type == STT_FUNC; | |
3964 | } | |
3965 | ||
166750bc AN |
3966 | static int find_extern_btf_id(const struct btf *btf, const char *ext_name) |
3967 | { | |
3968 | const struct btf_type *t; | |
5bd022ec | 3969 | const char *tname; |
166750bc AN |
3970 | int i, n; |
3971 | ||
3972 | if (!btf) | |
3973 | return -ESRCH; | |
3974 | ||
6a886de0 HC |
3975 | n = btf__type_cnt(btf); |
3976 | for (i = 1; i < n; i++) { | |
166750bc AN |
3977 | t = btf__type_by_id(btf, i); |
3978 | ||
5bd022ec | 3979 | if (!btf_is_var(t) && !btf_is_func(t)) |
166750bc AN |
3980 | continue; |
3981 | ||
5bd022ec MKL |
3982 | tname = btf__name_by_offset(btf, t->name_off); |
3983 | if (strcmp(tname, ext_name)) | |
166750bc AN |
3984 | continue; |
3985 | ||
5bd022ec MKL |
3986 | if (btf_is_var(t) && |
3987 | btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN) | |
3988 | return -EINVAL; | |
3989 | ||
3990 | if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN) | |
166750bc AN |
3991 | return -EINVAL; |
3992 | ||
3993 | return i; | |
3994 | } | |
3995 | ||
3996 | return -ENOENT; | |
3997 | } | |
3998 | ||
2e33efe3 AN |
3999 | static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) { |
4000 | const struct btf_var_secinfo *vs; | |
4001 | const struct btf_type *t; | |
4002 | int i, j, n; | |
4003 | ||
4004 | if (!btf) | |
4005 | return -ESRCH; | |
4006 | ||
6a886de0 HC |
4007 | n = btf__type_cnt(btf); |
4008 | for (i = 1; i < n; i++) { | |
2e33efe3 AN |
4009 | t = btf__type_by_id(btf, i); |
4010 | ||
4011 | if (!btf_is_datasec(t)) | |
4012 | continue; | |
4013 | ||
4014 | vs = btf_var_secinfos(t); | |
4015 | for (j = 0; j < btf_vlen(t); j++, vs++) { | |
4016 | if (vs->type == ext_btf_id) | |
4017 | return i; | |
4018 | } | |
4019 | } | |
4020 | ||
4021 | return -ENOENT; | |
4022 | } | |
4023 | ||
4024 | static enum kcfg_type find_kcfg_type(const struct btf *btf, int id, | |
4025 | bool *is_signed) | |
166750bc AN |
4026 | { |
4027 | const struct btf_type *t; | |
4028 | const char *name; | |
4029 | ||
4030 | t = skip_mods_and_typedefs(btf, id, NULL); | |
4031 | name = btf__name_by_offset(btf, t->name_off); | |
4032 | ||
4033 | if (is_signed) | |
4034 | *is_signed = false; | |
4035 | switch (btf_kind(t)) { | |
4036 | case BTF_KIND_INT: { | |
4037 | int enc = btf_int_encoding(t); | |
4038 | ||
4039 | if (enc & BTF_INT_BOOL) | |
2e33efe3 | 4040 | return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN; |
166750bc AN |
4041 | if (is_signed) |
4042 | *is_signed = enc & BTF_INT_SIGNED; | |
4043 | if (t->size == 1) | |
2e33efe3 | 4044 | return KCFG_CHAR; |
166750bc | 4045 | if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1))) |
2e33efe3 AN |
4046 | return KCFG_UNKNOWN; |
4047 | return KCFG_INT; | |
166750bc AN |
4048 | } |
4049 | case BTF_KIND_ENUM: | |
4050 | if (t->size != 4) | |
2e33efe3 | 4051 | return KCFG_UNKNOWN; |
166750bc | 4052 | if (strcmp(name, "libbpf_tristate")) |
2e33efe3 AN |
4053 | return KCFG_UNKNOWN; |
4054 | return KCFG_TRISTATE; | |
f2a62588 YS |
4055 | case BTF_KIND_ENUM64: |
4056 | if (strcmp(name, "libbpf_tristate")) | |
4057 | return KCFG_UNKNOWN; | |
4058 | return KCFG_TRISTATE; | |
166750bc AN |
4059 | case BTF_KIND_ARRAY: |
4060 | if (btf_array(t)->nelems == 0) | |
2e33efe3 AN |
4061 | return KCFG_UNKNOWN; |
4062 | if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR) | |
4063 | return KCFG_UNKNOWN; | |
4064 | return KCFG_CHAR_ARR; | |
166750bc | 4065 | default: |
2e33efe3 | 4066 | return KCFG_UNKNOWN; |
166750bc AN |
4067 | } |
4068 | } | |
4069 | ||
4070 | static int cmp_externs(const void *_a, const void *_b) | |
4071 | { | |
4072 | const struct extern_desc *a = _a; | |
4073 | const struct extern_desc *b = _b; | |
4074 | ||
2e33efe3 AN |
4075 | if (a->type != b->type) |
4076 | return a->type < b->type ? -1 : 1; | |
4077 | ||
4078 | if (a->type == EXT_KCFG) { | |
4079 | /* descending order by alignment requirements */ | |
4080 | if (a->kcfg.align != b->kcfg.align) | |
4081 | return a->kcfg.align > b->kcfg.align ? -1 : 1; | |
4082 | /* ascending order by size, within same alignment class */ | |
4083 | if (a->kcfg.sz != b->kcfg.sz) | |
4084 | return a->kcfg.sz < b->kcfg.sz ? -1 : 1; | |
4085 | } | |
4086 | ||
166750bc AN |
4087 | /* resolve ties by name */ |
4088 | return strcmp(a->name, b->name); | |
4089 | } | |
4090 | ||
1c0c7074 AN |
4091 | static int find_int_btf_id(const struct btf *btf) |
4092 | { | |
4093 | const struct btf_type *t; | |
4094 | int i, n; | |
4095 | ||
6a886de0 HC |
4096 | n = btf__type_cnt(btf); |
4097 | for (i = 1; i < n; i++) { | |
1c0c7074 AN |
4098 | t = btf__type_by_id(btf, i); |
4099 | ||
4100 | if (btf_is_int(t) && btf_int_bits(t) == 32) | |
4101 | return i; | |
4102 | } | |
4103 | ||
4104 | return 0; | |
4105 | } | |
4106 | ||
5bd022ec MKL |
4107 | static int add_dummy_ksym_var(struct btf *btf) |
4108 | { | |
4109 | int i, int_btf_id, sec_btf_id, dummy_var_btf_id; | |
4110 | const struct btf_var_secinfo *vs; | |
4111 | const struct btf_type *sec; | |
4112 | ||
9683e577 IR |
4113 | if (!btf) |
4114 | return 0; | |
4115 | ||
5bd022ec MKL |
4116 | sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC, |
4117 | BTF_KIND_DATASEC); | |
4118 | if (sec_btf_id < 0) | |
4119 | return 0; | |
4120 | ||
4121 | sec = btf__type_by_id(btf, sec_btf_id); | |
4122 | vs = btf_var_secinfos(sec); | |
4123 | for (i = 0; i < btf_vlen(sec); i++, vs++) { | |
4124 | const struct btf_type *vt; | |
4125 | ||
4126 | vt = btf__type_by_id(btf, vs->type); | |
4127 | if (btf_is_func(vt)) | |
4128 | break; | |
4129 | } | |
4130 | ||
4131 | /* No func in ksyms sec. No need to add dummy var. */ | |
4132 | if (i == btf_vlen(sec)) | |
4133 | return 0; | |
4134 | ||
4135 | int_btf_id = find_int_btf_id(btf); | |
4136 | dummy_var_btf_id = btf__add_var(btf, | |
4137 | "dummy_ksym", | |
4138 | BTF_VAR_GLOBAL_ALLOCATED, | |
4139 | int_btf_id); | |
4140 | if (dummy_var_btf_id < 0) | |
4141 | pr_warn("cannot create a dummy_ksym var\n"); | |
4142 | ||
4143 | return dummy_var_btf_id; | |
4144 | } | |
4145 | ||
166750bc AN |
4146 | static int bpf_object__collect_externs(struct bpf_object *obj) |
4147 | { | |
1c0c7074 | 4148 | struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL; |
166750bc AN |
4149 | const struct btf_type *t; |
4150 | struct extern_desc *ext; | |
5bd022ec | 4151 | int i, n, off, dummy_var_btf_id; |
2e33efe3 | 4152 | const char *ext_name, *sec_name; |
5964a223 | 4153 | size_t ext_essent_len; |
166750bc | 4154 | Elf_Scn *scn; |
ad23b723 | 4155 | Elf64_Shdr *sh; |
166750bc AN |
4156 | |
4157 | if (!obj->efile.symbols) | |
4158 | return 0; | |
4159 | ||
88a82120 | 4160 | scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx); |
ad23b723 | 4161 | sh = elf_sec_hdr(obj, scn); |
83390787 | 4162 | if (!sh || sh->sh_entsize != sizeof(Elf64_Sym)) |
166750bc | 4163 | return -LIBBPF_ERRNO__FORMAT; |
166750bc | 4164 | |
5bd022ec MKL |
4165 | dummy_var_btf_id = add_dummy_ksym_var(obj->btf); |
4166 | if (dummy_var_btf_id < 0) | |
4167 | return dummy_var_btf_id; | |
4168 | ||
ad23b723 | 4169 | n = sh->sh_size / sh->sh_entsize; |
166750bc | 4170 | pr_debug("looking for externs among %d symbols...\n", n); |
88a82120 | 4171 | |
166750bc | 4172 | for (i = 0; i < n; i++) { |
ad23b723 | 4173 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
166750bc | 4174 | |
ad23b723 | 4175 | if (!sym) |
166750bc | 4176 | return -LIBBPF_ERRNO__FORMAT; |
ad23b723 | 4177 | if (!sym_is_extern(sym)) |
166750bc | 4178 | continue; |
ad23b723 | 4179 | ext_name = elf_sym_str(obj, sym->st_name); |
166750bc AN |
4180 | if (!ext_name || !ext_name[0]) |
4181 | continue; | |
4182 | ||
4183 | ext = obj->externs; | |
029258d7 | 4184 | ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext)); |
166750bc AN |
4185 | if (!ext) |
4186 | return -ENOMEM; | |
4187 | obj->externs = ext; | |
4188 | ext = &ext[obj->nr_extern]; | |
4189 | memset(ext, 0, sizeof(*ext)); | |
4190 | obj->nr_extern++; | |
4191 | ||
4192 | ext->btf_id = find_extern_btf_id(obj->btf, ext_name); | |
4193 | if (ext->btf_id <= 0) { | |
4194 | pr_warn("failed to find BTF for extern '%s': %d\n", | |
4195 | ext_name, ext->btf_id); | |
4196 | return ext->btf_id; | |
4197 | } | |
4198 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
4199 | ext->name = btf__name_by_offset(obj->btf, t->name_off); | |
4200 | ext->sym_idx = i; | |
ad23b723 | 4201 | ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK; |
2e33efe3 | 4202 | |
5964a223 DM |
4203 | ext_essent_len = bpf_core_essential_name_len(ext->name); |
4204 | ext->essent_name = NULL; | |
4205 | if (ext_essent_len != strlen(ext->name)) { | |
4206 | ext->essent_name = strndup(ext->name, ext_essent_len); | |
4207 | if (!ext->essent_name) | |
4208 | return -ENOMEM; | |
4209 | } | |
4210 | ||
2e33efe3 AN |
4211 | ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id); |
4212 | if (ext->sec_btf_id <= 0) { | |
4213 | pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n", | |
4214 | ext_name, ext->btf_id, ext->sec_btf_id); | |
4215 | return ext->sec_btf_id; | |
166750bc | 4216 | } |
2e33efe3 AN |
4217 | sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id); |
4218 | sec_name = btf__name_by_offset(obj->btf, sec->name_off); | |
4219 | ||
4220 | if (strcmp(sec_name, KCONFIG_SEC) == 0) { | |
5bd022ec MKL |
4221 | if (btf_is_func(t)) { |
4222 | pr_warn("extern function %s is unsupported under %s section\n", | |
4223 | ext->name, KCONFIG_SEC); | |
4224 | return -ENOTSUP; | |
4225 | } | |
2e33efe3 AN |
4226 | kcfg_sec = sec; |
4227 | ext->type = EXT_KCFG; | |
4228 | ext->kcfg.sz = btf__resolve_size(obj->btf, t->type); | |
4229 | if (ext->kcfg.sz <= 0) { | |
4230 | pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n", | |
4231 | ext_name, ext->kcfg.sz); | |
4232 | return ext->kcfg.sz; | |
4233 | } | |
4234 | ext->kcfg.align = btf__align_of(obj->btf, t->type); | |
4235 | if (ext->kcfg.align <= 0) { | |
4236 | pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n", | |
4237 | ext_name, ext->kcfg.align); | |
4238 | return -EINVAL; | |
4239 | } | |
4240 | ext->kcfg.type = find_kcfg_type(obj->btf, t->type, | |
e3ba8e4e | 4241 | &ext->kcfg.is_signed); |
2e33efe3 | 4242 | if (ext->kcfg.type == KCFG_UNKNOWN) { |
55d00c37 | 4243 | pr_warn("extern (kcfg) '%s': type is unsupported\n", ext_name); |
2e33efe3 AN |
4244 | return -ENOTSUP; |
4245 | } | |
1c0c7074 | 4246 | } else if (strcmp(sec_name, KSYMS_SEC) == 0) { |
1c0c7074 AN |
4247 | ksym_sec = sec; |
4248 | ext->type = EXT_KSYM; | |
d370bbe1 HL |
4249 | skip_mods_and_typedefs(obj->btf, t->type, |
4250 | &ext->ksym.type_id); | |
2e33efe3 AN |
4251 | } else { |
4252 | pr_warn("unrecognized extern section '%s'\n", sec_name); | |
166750bc AN |
4253 | return -ENOTSUP; |
4254 | } | |
4255 | } | |
4256 | pr_debug("collected %d externs total\n", obj->nr_extern); | |
4257 | ||
4258 | if (!obj->nr_extern) | |
4259 | return 0; | |
4260 | ||
2e33efe3 | 4261 | /* sort externs by type, for kcfg ones also by (align, size, name) */ |
166750bc | 4262 | qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs); |
166750bc | 4263 | |
1c0c7074 AN |
4264 | /* for .ksyms section, we need to turn all externs into allocated |
4265 | * variables in BTF to pass kernel verification; we do this by | |
4266 | * pretending that each extern is a 8-byte variable | |
4267 | */ | |
4268 | if (ksym_sec) { | |
4269 | /* find existing 4-byte integer type in BTF to use for fake | |
4270 | * extern variables in DATASEC | |
4271 | */ | |
4272 | int int_btf_id = find_int_btf_id(obj->btf); | |
5bd022ec MKL |
4273 | /* For extern function, a dummy_var added earlier |
4274 | * will be used to replace the vs->type and | |
4275 | * its name string will be used to refill | |
4276 | * the missing param's name. | |
4277 | */ | |
4278 | const struct btf_type *dummy_var; | |
1c0c7074 | 4279 | |
5bd022ec | 4280 | dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id); |
1c0c7074 AN |
4281 | for (i = 0; i < obj->nr_extern; i++) { |
4282 | ext = &obj->externs[i]; | |
4283 | if (ext->type != EXT_KSYM) | |
4284 | continue; | |
4285 | pr_debug("extern (ksym) #%d: symbol %d, name %s\n", | |
4286 | i, ext->sym_idx, ext->name); | |
4287 | } | |
4288 | ||
4289 | sec = ksym_sec; | |
4290 | n = btf_vlen(sec); | |
4291 | for (i = 0, off = 0; i < n; i++, off += sizeof(int)) { | |
4292 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
4293 | struct btf_type *vt; | |
4294 | ||
4295 | vt = (void *)btf__type_by_id(obj->btf, vs->type); | |
4296 | ext_name = btf__name_by_offset(obj->btf, vt->name_off); | |
4297 | ext = find_extern_by_name(obj, ext_name); | |
4298 | if (!ext) { | |
5bd022ec MKL |
4299 | pr_warn("failed to find extern definition for BTF %s '%s'\n", |
4300 | btf_kind_str(vt), ext_name); | |
1c0c7074 AN |
4301 | return -ESRCH; |
4302 | } | |
5bd022ec MKL |
4303 | if (btf_is_func(vt)) { |
4304 | const struct btf_type *func_proto; | |
4305 | struct btf_param *param; | |
4306 | int j; | |
4307 | ||
4308 | func_proto = btf__type_by_id(obj->btf, | |
4309 | vt->type); | |
4310 | param = btf_params(func_proto); | |
4311 | /* Reuse the dummy_var string if the | |
4312 | * func proto does not have param name. | |
4313 | */ | |
4314 | for (j = 0; j < btf_vlen(func_proto); j++) | |
4315 | if (param[j].type && !param[j].name_off) | |
4316 | param[j].name_off = | |
4317 | dummy_var->name_off; | |
4318 | vs->type = dummy_var_btf_id; | |
4319 | vt->info &= ~0xffff; | |
4320 | vt->info |= BTF_FUNC_GLOBAL; | |
4321 | } else { | |
4322 | btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
4323 | vt->type = int_btf_id; | |
4324 | } | |
1c0c7074 AN |
4325 | vs->offset = off; |
4326 | vs->size = sizeof(int); | |
4327 | } | |
4328 | sec->size = off; | |
4329 | } | |
4330 | ||
2e33efe3 AN |
4331 | if (kcfg_sec) { |
4332 | sec = kcfg_sec; | |
4333 | /* for kcfg externs calculate their offsets within a .kconfig map */ | |
4334 | off = 0; | |
4335 | for (i = 0; i < obj->nr_extern; i++) { | |
4336 | ext = &obj->externs[i]; | |
4337 | if (ext->type != EXT_KCFG) | |
4338 | continue; | |
166750bc | 4339 | |
2e33efe3 AN |
4340 | ext->kcfg.data_off = roundup(off, ext->kcfg.align); |
4341 | off = ext->kcfg.data_off + ext->kcfg.sz; | |
1c0c7074 | 4342 | pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n", |
2e33efe3 AN |
4343 | i, ext->sym_idx, ext->kcfg.data_off, ext->name); |
4344 | } | |
4345 | sec->size = off; | |
4346 | n = btf_vlen(sec); | |
4347 | for (i = 0; i < n; i++) { | |
4348 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
4349 | ||
4350 | t = btf__type_by_id(obj->btf, vs->type); | |
4351 | ext_name = btf__name_by_offset(obj->btf, t->name_off); | |
4352 | ext = find_extern_by_name(obj, ext_name); | |
4353 | if (!ext) { | |
4354 | pr_warn("failed to find extern definition for BTF var '%s'\n", | |
4355 | ext_name); | |
4356 | return -ESRCH; | |
4357 | } | |
4358 | btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
4359 | vs->offset = ext->kcfg.data_off; | |
166750bc | 4360 | } |
166750bc | 4361 | } |
166750bc AN |
4362 | return 0; |
4363 | } | |
4364 | ||
bd054102 AN |
4365 | static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog) |
4366 | { | |
197afc63 | 4367 | return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1; |
c3c55696 AN |
4368 | } |
4369 | ||
01af3bf0 AN |
4370 | struct bpf_program * |
4371 | bpf_object__find_program_by_name(const struct bpf_object *obj, | |
4372 | const char *name) | |
4373 | { | |
4374 | struct bpf_program *prog; | |
4375 | ||
4376 | bpf_object__for_each_program(prog, obj) { | |
c3c55696 AN |
4377 | if (prog_is_subprog(obj, prog)) |
4378 | continue; | |
01af3bf0 AN |
4379 | if (!strcmp(prog->name, name)) |
4380 | return prog; | |
4381 | } | |
e9fc3ce9 | 4382 | return errno = ENOENT, NULL; |
01af3bf0 AN |
4383 | } |
4384 | ||
d859900c DB |
4385 | static bool bpf_object__shndx_is_data(const struct bpf_object *obj, |
4386 | int shndx) | |
4387 | { | |
25bbbd7a AN |
4388 | switch (obj->efile.secs[shndx].sec_type) { |
4389 | case SEC_BSS: | |
4390 | case SEC_DATA: | |
4391 | case SEC_RODATA: | |
4392 | return true; | |
4393 | default: | |
4394 | return false; | |
4395 | } | |
d859900c DB |
4396 | } |
4397 | ||
4398 | static bool bpf_object__shndx_is_maps(const struct bpf_object *obj, | |
4399 | int shndx) | |
4400 | { | |
e19db676 | 4401 | return shndx == obj->efile.btf_maps_shndx; |
d859900c DB |
4402 | } |
4403 | ||
d859900c DB |
4404 | static enum libbpf_map_type |
4405 | bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx) | |
4406 | { | |
25bbbd7a AN |
4407 | if (shndx == obj->efile.symbols_shndx) |
4408 | return LIBBPF_MAP_KCONFIG; | |
4409 | ||
4410 | switch (obj->efile.secs[shndx].sec_type) { | |
4411 | case SEC_BSS: | |
d859900c | 4412 | return LIBBPF_MAP_BSS; |
25bbbd7a AN |
4413 | case SEC_DATA: |
4414 | return LIBBPF_MAP_DATA; | |
4415 | case SEC_RODATA: | |
d859900c | 4416 | return LIBBPF_MAP_RODATA; |
25bbbd7a | 4417 | default: |
d859900c | 4418 | return LIBBPF_MAP_UNSPEC; |
25bbbd7a | 4419 | } |
d859900c DB |
4420 | } |
4421 | ||
1f8e2bcb AN |
4422 | static int bpf_program__record_reloc(struct bpf_program *prog, |
4423 | struct reloc_desc *reloc_desc, | |
9c0f8cbd | 4424 | __u32 insn_idx, const char *sym_name, |
ad23b723 | 4425 | const Elf64_Sym *sym, const Elf64_Rel *rel) |
1f8e2bcb AN |
4426 | { |
4427 | struct bpf_insn *insn = &prog->insns[insn_idx]; | |
4428 | size_t map_idx, nr_maps = prog->obj->nr_maps; | |
4429 | struct bpf_object *obj = prog->obj; | |
4430 | __u32 shdr_idx = sym->st_shndx; | |
4431 | enum libbpf_map_type type; | |
9c0f8cbd | 4432 | const char *sym_sec_name; |
1f8e2bcb AN |
4433 | struct bpf_map *map; |
4434 | ||
aa0b8d43 | 4435 | if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) { |
9c0f8cbd AN |
4436 | pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n", |
4437 | prog->name, sym_name, insn_idx, insn->code); | |
1f8e2bcb AN |
4438 | return -LIBBPF_ERRNO__RELOC; |
4439 | } | |
166750bc AN |
4440 | |
4441 | if (sym_is_extern(sym)) { | |
ad23b723 | 4442 | int sym_idx = ELF64_R_SYM(rel->r_info); |
166750bc AN |
4443 | int i, n = obj->nr_extern; |
4444 | struct extern_desc *ext; | |
4445 | ||
4446 | for (i = 0; i < n; i++) { | |
4447 | ext = &obj->externs[i]; | |
4448 | if (ext->sym_idx == sym_idx) | |
4449 | break; | |
4450 | } | |
4451 | if (i >= n) { | |
9c0f8cbd AN |
4452 | pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n", |
4453 | prog->name, sym_name, sym_idx); | |
166750bc AN |
4454 | return -LIBBPF_ERRNO__RELOC; |
4455 | } | |
9c0f8cbd AN |
4456 | pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n", |
4457 | prog->name, i, ext->name, ext->sym_idx, insn_idx); | |
5bd022ec | 4458 | if (insn->code == (BPF_JMP | BPF_CALL)) |
a18f7214 | 4459 | reloc_desc->type = RELO_EXTERN_CALL; |
5bd022ec | 4460 | else |
a18f7214 | 4461 | reloc_desc->type = RELO_EXTERN_LD64; |
166750bc | 4462 | reloc_desc->insn_idx = insn_idx; |
3055ddd6 | 4463 | reloc_desc->ext_idx = i; |
166750bc AN |
4464 | return 0; |
4465 | } | |
4466 | ||
aa0b8d43 MKL |
4467 | /* sub-program call relocation */ |
4468 | if (is_call_insn(insn)) { | |
4469 | if (insn->src_reg != BPF_PSEUDO_CALL) { | |
4470 | pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name); | |
4471 | return -LIBBPF_ERRNO__RELOC; | |
4472 | } | |
4473 | /* text_shndx can be 0, if no default "main" program exists */ | |
4474 | if (!shdr_idx || shdr_idx != obj->efile.text_shndx) { | |
4475 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); | |
4476 | pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n", | |
4477 | prog->name, sym_name, sym_sec_name); | |
4478 | return -LIBBPF_ERRNO__RELOC; | |
4479 | } | |
4480 | if (sym->st_value % BPF_INSN_SZ) { | |
4481 | pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n", | |
4482 | prog->name, sym_name, (size_t)sym->st_value); | |
4483 | return -LIBBPF_ERRNO__RELOC; | |
4484 | } | |
4485 | reloc_desc->type = RELO_CALL; | |
4486 | reloc_desc->insn_idx = insn_idx; | |
4487 | reloc_desc->sym_off = sym->st_value; | |
4488 | return 0; | |
4489 | } | |
4490 | ||
1f8e2bcb | 4491 | if (!shdr_idx || shdr_idx >= SHN_LORESERVE) { |
9c0f8cbd AN |
4492 | pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n", |
4493 | prog->name, sym_name, shdr_idx); | |
1f8e2bcb AN |
4494 | return -LIBBPF_ERRNO__RELOC; |
4495 | } | |
4496 | ||
53eddb5e YS |
4497 | /* loading subprog addresses */ |
4498 | if (sym_is_subprog(sym, obj->efile.text_shndx)) { | |
4499 | /* global_func: sym->st_value = offset in the section, insn->imm = 0. | |
4500 | * local_func: sym->st_value = 0, insn->imm = offset in the section. | |
4501 | */ | |
4502 | if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) { | |
4503 | pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n", | |
4504 | prog->name, sym_name, (size_t)sym->st_value, insn->imm); | |
4505 | return -LIBBPF_ERRNO__RELOC; | |
4506 | } | |
4507 | ||
4508 | reloc_desc->type = RELO_SUBPROG_ADDR; | |
4509 | reloc_desc->insn_idx = insn_idx; | |
4510 | reloc_desc->sym_off = sym->st_value; | |
4511 | return 0; | |
4512 | } | |
4513 | ||
1f8e2bcb | 4514 | type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx); |
9c0f8cbd | 4515 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); |
1f8e2bcb | 4516 | |
2e7ba4f8 AN |
4517 | /* arena data relocation */ |
4518 | if (shdr_idx == obj->efile.arena_data_shndx) { | |
4519 | reloc_desc->type = RELO_DATA; | |
4520 | reloc_desc->insn_idx = insn_idx; | |
4521 | reloc_desc->map_idx = obj->arena_map - obj->maps; | |
4522 | reloc_desc->sym_off = sym->st_value; | |
4523 | return 0; | |
4524 | } | |
4525 | ||
1f8e2bcb AN |
4526 | /* generic map reference relocation */ |
4527 | if (type == LIBBPF_MAP_UNSPEC) { | |
4528 | if (!bpf_object__shndx_is_maps(obj, shdr_idx)) { | |
9c0f8cbd AN |
4529 | pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n", |
4530 | prog->name, sym_name, sym_sec_name); | |
1f8e2bcb AN |
4531 | return -LIBBPF_ERRNO__RELOC; |
4532 | } | |
4533 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { | |
4534 | map = &obj->maps[map_idx]; | |
4535 | if (map->libbpf_type != type || | |
4536 | map->sec_idx != sym->st_shndx || | |
4537 | map->sec_offset != sym->st_value) | |
4538 | continue; | |
9c0f8cbd AN |
4539 | pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n", |
4540 | prog->name, map_idx, map->name, map->sec_idx, | |
1f8e2bcb AN |
4541 | map->sec_offset, insn_idx); |
4542 | break; | |
4543 | } | |
4544 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
4545 | pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n", |
4546 | prog->name, sym_sec_name, (size_t)sym->st_value); | |
1f8e2bcb AN |
4547 | return -LIBBPF_ERRNO__RELOC; |
4548 | } | |
4549 | reloc_desc->type = RELO_LD64; | |
4550 | reloc_desc->insn_idx = insn_idx; | |
4551 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 4552 | reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */ |
1f8e2bcb AN |
4553 | return 0; |
4554 | } | |
4555 | ||
4556 | /* global data map relocation */ | |
4557 | if (!bpf_object__shndx_is_data(obj, shdr_idx)) { | |
9c0f8cbd AN |
4558 | pr_warn("prog '%s': bad data relo against section '%s'\n", |
4559 | prog->name, sym_sec_name); | |
1f8e2bcb | 4560 | return -LIBBPF_ERRNO__RELOC; |
1f8e2bcb | 4561 | } |
1f8e2bcb AN |
4562 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { |
4563 | map = &obj->maps[map_idx]; | |
25bbbd7a | 4564 | if (map->libbpf_type != type || map->sec_idx != sym->st_shndx) |
1f8e2bcb | 4565 | continue; |
9c0f8cbd AN |
4566 | pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n", |
4567 | prog->name, map_idx, map->name, map->sec_idx, | |
4568 | map->sec_offset, insn_idx); | |
1f8e2bcb AN |
4569 | break; |
4570 | } | |
4571 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
4572 | pr_warn("prog '%s': data relo failed to find map for section '%s'\n", |
4573 | prog->name, sym_sec_name); | |
1f8e2bcb AN |
4574 | return -LIBBPF_ERRNO__RELOC; |
4575 | } | |
4576 | ||
4577 | reloc_desc->type = RELO_DATA; | |
4578 | reloc_desc->insn_idx = insn_idx; | |
4579 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 4580 | reloc_desc->sym_off = sym->st_value; |
1f8e2bcb AN |
4581 | return 0; |
4582 | } | |
4583 | ||
db2b8b06 AN |
4584 | static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx) |
4585 | { | |
4586 | return insn_idx >= prog->sec_insn_off && | |
4587 | insn_idx < prog->sec_insn_off + prog->sec_insn_cnt; | |
4588 | } | |
4589 | ||
4590 | static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj, | |
4591 | size_t sec_idx, size_t insn_idx) | |
4592 | { | |
4593 | int l = 0, r = obj->nr_programs - 1, m; | |
4594 | struct bpf_program *prog; | |
4595 | ||
d0d382f9 SHY |
4596 | if (!obj->nr_programs) |
4597 | return NULL; | |
4598 | ||
db2b8b06 AN |
4599 | while (l < r) { |
4600 | m = l + (r - l + 1) / 2; | |
4601 | prog = &obj->programs[m]; | |
4602 | ||
4603 | if (prog->sec_idx < sec_idx || | |
4604 | (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx)) | |
4605 | l = m; | |
4606 | else | |
4607 | r = m - 1; | |
4608 | } | |
4609 | /* matching program could be at index l, but it still might be the | |
4610 | * wrong one, so we need to double check conditions for the last time | |
4611 | */ | |
4612 | prog = &obj->programs[l]; | |
4613 | if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx)) | |
4614 | return prog; | |
4615 | return NULL; | |
4616 | } | |
4617 | ||
34090915 | 4618 | static int |
ad23b723 | 4619 | bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data) |
34090915 | 4620 | { |
9c0f8cbd | 4621 | const char *relo_sec_name, *sec_name; |
b7332d28 | 4622 | size_t sec_idx = shdr->sh_info, sym_idx; |
c3c55696 AN |
4623 | struct bpf_program *prog; |
4624 | struct reloc_desc *relos; | |
1f8e2bcb | 4625 | int err, i, nrels; |
c3c55696 AN |
4626 | const char *sym_name; |
4627 | __u32 insn_idx; | |
6245947c AN |
4628 | Elf_Scn *scn; |
4629 | Elf_Data *scn_data; | |
ad23b723 AN |
4630 | Elf64_Sym *sym; |
4631 | Elf64_Rel *rel; | |
34090915 | 4632 | |
b7332d28 AN |
4633 | if (sec_idx >= obj->efile.sec_cnt) |
4634 | return -EINVAL; | |
4635 | ||
6245947c AN |
4636 | scn = elf_sec_by_idx(obj, sec_idx); |
4637 | scn_data = elf_sec_data(obj, scn); | |
fc3a5534 MZ |
4638 | if (!scn_data) |
4639 | return -LIBBPF_ERRNO__FORMAT; | |
6245947c | 4640 | |
9c0f8cbd | 4641 | relo_sec_name = elf_sec_str(obj, shdr->sh_name); |
6245947c | 4642 | sec_name = elf_sec_name(obj, scn); |
9c0f8cbd AN |
4643 | if (!relo_sec_name || !sec_name) |
4644 | return -EINVAL; | |
4645 | ||
4646 | pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n", | |
4647 | relo_sec_name, sec_idx, sec_name); | |
34090915 WN |
4648 | nrels = shdr->sh_size / shdr->sh_entsize; |
4649 | ||
34090915 | 4650 | for (i = 0; i < nrels; i++) { |
ad23b723 AN |
4651 | rel = elf_rel_by_idx(data, i); |
4652 | if (!rel) { | |
9c0f8cbd | 4653 | pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i); |
6371ca3b | 4654 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4655 | } |
ad23b723 | 4656 | |
b7332d28 AN |
4657 | sym_idx = ELF64_R_SYM(rel->r_info); |
4658 | sym = elf_sym_by_idx(obj, sym_idx); | |
ad23b723 | 4659 | if (!sym) { |
b7332d28 AN |
4660 | pr_warn("sec '%s': symbol #%zu not found for relo #%d\n", |
4661 | relo_sec_name, sym_idx, i); | |
4662 | return -LIBBPF_ERRNO__FORMAT; | |
4663 | } | |
4664 | ||
4665 | if (sym->st_shndx >= obj->efile.sec_cnt) { | |
4666 | pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n", | |
4667 | relo_sec_name, sym_idx, (size_t)sym->st_shndx, i); | |
6371ca3b | 4668 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4669 | } |
6245947c | 4670 | |
ad23b723 | 4671 | if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) { |
9c0f8cbd | 4672 | pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n", |
b7332d28 | 4673 | relo_sec_name, (size_t)rel->r_offset, i); |
1f8e2bcb | 4674 | return -LIBBPF_ERRNO__FORMAT; |
9c0f8cbd | 4675 | } |
d859900c | 4676 | |
ad23b723 | 4677 | insn_idx = rel->r_offset / BPF_INSN_SZ; |
c3c55696 AN |
4678 | /* relocations against static functions are recorded as |
4679 | * relocations against the section that contains a function; | |
4680 | * in such case, symbol will be STT_SECTION and sym.st_name | |
4681 | * will point to empty string (0), so fetch section name | |
4682 | * instead | |
4683 | */ | |
ad23b723 AN |
4684 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0) |
4685 | sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx)); | |
c3c55696 | 4686 | else |
ad23b723 | 4687 | sym_name = elf_sym_str(obj, sym->st_name); |
c3c55696 | 4688 | sym_name = sym_name ?: "<?"; |
d859900c | 4689 | |
9c0f8cbd AN |
4690 | pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n", |
4691 | relo_sec_name, i, insn_idx, sym_name); | |
666810e8 | 4692 | |
c3c55696 AN |
4693 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); |
4694 | if (!prog) { | |
6245947c | 4695 | pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n", |
c3c55696 | 4696 | relo_sec_name, i, sec_name, insn_idx); |
6245947c | 4697 | continue; |
c3c55696 AN |
4698 | } |
4699 | ||
4700 | relos = libbpf_reallocarray(prog->reloc_desc, | |
4701 | prog->nr_reloc + 1, sizeof(*relos)); | |
4702 | if (!relos) | |
4703 | return -ENOMEM; | |
4704 | prog->reloc_desc = relos; | |
4705 | ||
4706 | /* adjust insn_idx to local BPF program frame of reference */ | |
4707 | insn_idx -= prog->sec_insn_off; | |
4708 | err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc], | |
ad23b723 | 4709 | insn_idx, sym_name, sym, rel); |
1f8e2bcb AN |
4710 | if (err) |
4711 | return err; | |
c3c55696 AN |
4712 | |
4713 | prog->nr_reloc++; | |
34090915 WN |
4714 | } |
4715 | return 0; | |
4716 | } | |
4717 | ||
4fcac46c | 4718 | static int map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map) |
8a138aed | 4719 | { |
aaf6886d | 4720 | int id; |
8a138aed | 4721 | |
a8fee962 AN |
4722 | if (!obj->btf) |
4723 | return -ENOENT; | |
4724 | ||
590a0088 MKL |
4725 | /* if it's BTF-defined map, we don't need to search for type IDs. |
4726 | * For struct_ops map, it does not need btf_key_type_id and | |
4727 | * btf_value_type_id. | |
4728 | */ | |
aaf6886d | 4729 | if (map->sec_idx == obj->efile.btf_maps_shndx || bpf_map__is_struct_ops(map)) |
abd29c93 AN |
4730 | return 0; |
4731 | ||
aaf6886d AN |
4732 | /* |
4733 | * LLVM annotates global data differently in BTF, that is, | |
4734 | * only as '.data', '.bss' or '.rodata'. | |
4735 | */ | |
4736 | if (!bpf_map__is_internal(map)) | |
4737 | return -ENOENT; | |
4738 | ||
4739 | id = btf__find_by_name(obj->btf, map->real_name); | |
4740 | if (id < 0) | |
4741 | return id; | |
8a138aed | 4742 | |
aaf6886d AN |
4743 | map->btf_key_type_id = 0; |
4744 | map->btf_value_type_id = id; | |
8a138aed MKL |
4745 | return 0; |
4746 | } | |
4747 | ||
97eb3138 MP |
4748 | static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info) |
4749 | { | |
4750 | char file[PATH_MAX], buff[4096]; | |
4751 | FILE *fp; | |
4752 | __u32 val; | |
4753 | int err; | |
4754 | ||
4755 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
4756 | memset(info, 0, sizeof(*info)); | |
4757 | ||
59842c54 | 4758 | fp = fopen(file, "re"); |
97eb3138 MP |
4759 | if (!fp) { |
4760 | err = -errno; | |
4761 | pr_warn("failed to open %s: %d. No procfs support?\n", file, | |
4762 | err); | |
4763 | return err; | |
4764 | } | |
4765 | ||
4766 | while (fgets(buff, sizeof(buff), fp)) { | |
4767 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
4768 | info->type = val; | |
4769 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) | |
4770 | info->key_size = val; | |
4771 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) | |
4772 | info->value_size = val; | |
4773 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) | |
4774 | info->max_entries = val; | |
4775 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) | |
4776 | info->map_flags = val; | |
4777 | } | |
4778 | ||
4779 | fclose(fp); | |
4780 | ||
4781 | return 0; | |
4782 | } | |
4783 | ||
ec41817b AN |
4784 | bool bpf_map__autocreate(const struct bpf_map *map) |
4785 | { | |
4786 | return map->autocreate; | |
4787 | } | |
4788 | ||
4789 | int bpf_map__set_autocreate(struct bpf_map *map, bool autocreate) | |
4790 | { | |
4791 | if (map->obj->loaded) | |
4792 | return libbpf_err(-EBUSY); | |
4793 | ||
4794 | map->autocreate = autocreate; | |
4795 | return 0; | |
4796 | } | |
4797 | ||
26736eb9 JK |
4798 | int bpf_map__reuse_fd(struct bpf_map *map, int fd) |
4799 | { | |
813847a3 | 4800 | struct bpf_map_info info; |
bf3f0037 | 4801 | __u32 len = sizeof(info), name_len; |
26736eb9 JK |
4802 | int new_fd, err; |
4803 | char *new_name; | |
4804 | ||
813847a3 | 4805 | memset(&info, 0, len); |
629dfc66 | 4806 | err = bpf_map_get_info_by_fd(fd, &info, &len); |
97eb3138 MP |
4807 | if (err && errno == EINVAL) |
4808 | err = bpf_get_map_info_from_fdinfo(fd, &info); | |
26736eb9 | 4809 | if (err) |
e9fc3ce9 | 4810 | return libbpf_err(err); |
26736eb9 | 4811 | |
bf3f0037 AW |
4812 | name_len = strlen(info.name); |
4813 | if (name_len == BPF_OBJ_NAME_LEN - 1 && strncmp(map->name, info.name, name_len) == 0) | |
4814 | new_name = strdup(map->name); | |
4815 | else | |
4816 | new_name = strdup(info.name); | |
4817 | ||
26736eb9 | 4818 | if (!new_name) |
e9fc3ce9 | 4819 | return libbpf_err(-errno); |
26736eb9 | 4820 | |
4aadd292 AN |
4821 | /* |
4822 | * Like dup(), but make sure new FD is >= 3 and has O_CLOEXEC set. | |
4823 | * This is similar to what we do in ensure_good_fd(), but without | |
4824 | * closing original FD. | |
4825 | */ | |
4826 | new_fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); | |
d1b4574a THJ |
4827 | if (new_fd < 0) { |
4828 | err = -errno; | |
26736eb9 | 4829 | goto err_free_new_name; |
d1b4574a | 4830 | } |
26736eb9 | 4831 | |
dac645b9 AN |
4832 | err = reuse_fd(map->fd, new_fd); |
4833 | if (err) | |
4834 | goto err_free_new_name; | |
4835 | ||
26736eb9 JK |
4836 | free(map->name); |
4837 | ||
26736eb9 JK |
4838 | map->name = new_name; |
4839 | map->def.type = info.type; | |
4840 | map->def.key_size = info.key_size; | |
4841 | map->def.value_size = info.value_size; | |
4842 | map->def.max_entries = info.max_entries; | |
4843 | map->def.map_flags = info.map_flags; | |
4844 | map->btf_key_type_id = info.btf_key_type_id; | |
4845 | map->btf_value_type_id = info.btf_value_type_id; | |
ec6d5f47 | 4846 | map->reused = true; |
47512102 | 4847 | map->map_extra = info.map_extra; |
26736eb9 JK |
4848 | |
4849 | return 0; | |
4850 | ||
26736eb9 JK |
4851 | err_free_new_name: |
4852 | free(new_name); | |
e9fc3ce9 | 4853 | return libbpf_err(err); |
26736eb9 JK |
4854 | } |
4855 | ||
1bdb6c9a | 4856 | __u32 bpf_map__max_entries(const struct bpf_map *map) |
1a11a4c7 | 4857 | { |
1bdb6c9a AN |
4858 | return map->def.max_entries; |
4859 | } | |
1a11a4c7 | 4860 | |
b3278099 AN |
4861 | struct bpf_map *bpf_map__inner_map(struct bpf_map *map) |
4862 | { | |
4863 | if (!bpf_map_type__is_map_in_map(map->def.type)) | |
e9fc3ce9 | 4864 | return errno = EINVAL, NULL; |
b3278099 AN |
4865 | |
4866 | return map->inner_map; | |
4867 | } | |
4868 | ||
1bdb6c9a AN |
4869 | int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries) |
4870 | { | |
597fbc46 | 4871 | if (map->obj->loaded) |
e9fc3ce9 | 4872 | return libbpf_err(-EBUSY); |
597fbc46 | 4873 | |
1a11a4c7 | 4874 | map->def.max_entries = max_entries; |
597fbc46 AN |
4875 | |
4876 | /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */ | |
b66ccae0 | 4877 | if (map_is_ringbuf(map)) |
597fbc46 AN |
4878 | map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries); |
4879 | ||
1a11a4c7 AI |
4880 | return 0; |
4881 | } | |
4882 | ||
6b434b61 AN |
4883 | static int bpf_object_prepare_token(struct bpf_object *obj) |
4884 | { | |
4885 | const char *bpffs_path; | |
4886 | int bpffs_fd = -1, token_fd, err; | |
4887 | bool mandatory; | |
4888 | enum libbpf_print_level level; | |
4889 | ||
4890 | /* token is explicitly prevented */ | |
4891 | if (obj->token_path && obj->token_path[0] == '\0') { | |
4892 | pr_debug("object '%s': token is prevented, skipping...\n", obj->name); | |
4893 | return 0; | |
4894 | } | |
4895 | ||
4896 | mandatory = obj->token_path != NULL; | |
4897 | level = mandatory ? LIBBPF_WARN : LIBBPF_DEBUG; | |
4898 | ||
4899 | bpffs_path = obj->token_path ?: BPF_FS_DEFAULT_PATH; | |
4900 | bpffs_fd = open(bpffs_path, O_DIRECTORY, O_RDWR); | |
4901 | if (bpffs_fd < 0) { | |
4902 | err = -errno; | |
4903 | __pr(level, "object '%s': failed (%d) to open BPF FS mount at '%s'%s\n", | |
4904 | obj->name, err, bpffs_path, | |
4905 | mandatory ? "" : ", skipping optional step..."); | |
4906 | return mandatory ? err : 0; | |
4907 | } | |
4908 | ||
4909 | token_fd = bpf_token_create(bpffs_fd, 0); | |
4910 | close(bpffs_fd); | |
4911 | if (token_fd < 0) { | |
4912 | if (!mandatory && token_fd == -ENOENT) { | |
4913 | pr_debug("object '%s': BPF FS at '%s' doesn't have BPF token delegation set up, skipping...\n", | |
4914 | obj->name, bpffs_path); | |
4915 | return 0; | |
4916 | } | |
4917 | __pr(level, "object '%s': failed (%d) to create BPF token from '%s'%s\n", | |
4918 | obj->name, token_fd, bpffs_path, | |
4919 | mandatory ? "" : ", skipping optional step..."); | |
4920 | return mandatory ? token_fd : 0; | |
4921 | } | |
4922 | ||
4923 | obj->feat_cache = calloc(1, sizeof(*obj->feat_cache)); | |
4924 | if (!obj->feat_cache) { | |
4925 | close(token_fd); | |
4926 | return -ENOMEM; | |
4927 | } | |
4928 | ||
4929 | obj->token_fd = token_fd; | |
4930 | obj->feat_cache->token_fd = token_fd; | |
4931 | ||
4932 | return 0; | |
4933 | } | |
4934 | ||
47eff617 | 4935 | static int |
fd9eef1a | 4936 | bpf_object__probe_loading(struct bpf_object *obj) |
47eff617 | 4937 | { |
47eff617 SF |
4938 | char *cp, errmsg[STRERR_BUFSIZE]; |
4939 | struct bpf_insn insns[] = { | |
4940 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4941 | BPF_EXIT_INSN(), | |
4942 | }; | |
e32660ac | 4943 | int ret, insn_cnt = ARRAY_SIZE(insns); |
6b434b61 AN |
4944 | LIBBPF_OPTS(bpf_prog_load_opts, opts, |
4945 | .token_fd = obj->token_fd, | |
4946 | .prog_flags = obj->token_fd ? BPF_F_TOKEN_FD : 0, | |
4947 | ); | |
47eff617 | 4948 | |
f9bceaa5 SF |
4949 | if (obj->gen_loader) |
4950 | return 0; | |
4951 | ||
e542f2c4 AN |
4952 | ret = bump_rlimit_memlock(); |
4953 | if (ret) | |
4954 | pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret); | |
4955 | ||
47eff617 | 4956 | /* make sure basic loading works */ |
6b434b61 | 4957 | ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, &opts); |
e32660ac | 4958 | if (ret < 0) |
6b434b61 | 4959 | ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, &opts); |
47eff617 | 4960 | if (ret < 0) { |
fd9eef1a EC |
4961 | ret = errno; |
4962 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4963 | pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF " | |
4964 | "program. Make sure your kernel supports BPF " | |
4965 | "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is " | |
4966 | "set to big enough value.\n", __func__, cp, ret); | |
4967 | return -ret; | |
47eff617 SF |
4968 | } |
4969 | close(ret); | |
4970 | ||
fd9eef1a EC |
4971 | return 0; |
4972 | } | |
4973 | ||
d6dd1d49 AN |
4974 | bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id) |
4975 | { | |
8263b338 | 4976 | if (obj->gen_loader) |
d6dd1d49 AN |
4977 | /* To generate loader program assume the latest kernel |
4978 | * to avoid doing extra prog_load, map_create syscalls. | |
4979 | */ | |
4980 | return true; | |
4981 | ||
6b434b61 AN |
4982 | if (obj->token_fd) |
4983 | return feat_supported(obj->feat_cache, feat_id); | |
4984 | ||
d6dd1d49 AN |
4985 | return feat_supported(NULL, feat_id); |
4986 | } | |
4987 | ||
57a00f41 THJ |
4988 | static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd) |
4989 | { | |
813847a3 | 4990 | struct bpf_map_info map_info; |
57a00f41 | 4991 | char msg[STRERR_BUFSIZE]; |
813847a3 | 4992 | __u32 map_info_len = sizeof(map_info); |
97eb3138 | 4993 | int err; |
57a00f41 | 4994 | |
813847a3 | 4995 | memset(&map_info, 0, map_info_len); |
629dfc66 | 4996 | err = bpf_map_get_info_by_fd(map_fd, &map_info, &map_info_len); |
97eb3138 MP |
4997 | if (err && errno == EINVAL) |
4998 | err = bpf_get_map_info_from_fdinfo(map_fd, &map_info); | |
4999 | if (err) { | |
5000 | pr_warn("failed to get map info for map FD %d: %s\n", map_fd, | |
5001 | libbpf_strerror_r(errno, msg, sizeof(msg))); | |
57a00f41 THJ |
5002 | return false; |
5003 | } | |
5004 | ||
5005 | return (map_info.type == map->def.type && | |
5006 | map_info.key_size == map->def.key_size && | |
5007 | map_info.value_size == map->def.value_size && | |
5008 | map_info.max_entries == map->def.max_entries && | |
47512102 JK |
5009 | map_info.map_flags == map->def.map_flags && |
5010 | map_info.map_extra == map->map_extra); | |
57a00f41 THJ |
5011 | } |
5012 | ||
5013 | static int | |
5014 | bpf_object__reuse_map(struct bpf_map *map) | |
5015 | { | |
5016 | char *cp, errmsg[STRERR_BUFSIZE]; | |
5017 | int err, pin_fd; | |
5018 | ||
5019 | pin_fd = bpf_obj_get(map->pin_path); | |
5020 | if (pin_fd < 0) { | |
5021 | err = -errno; | |
5022 | if (err == -ENOENT) { | |
5023 | pr_debug("found no pinned map to reuse at '%s'\n", | |
5024 | map->pin_path); | |
5025 | return 0; | |
5026 | } | |
5027 | ||
5028 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
5029 | pr_warn("couldn't retrieve pinned map '%s': %s\n", | |
5030 | map->pin_path, cp); | |
5031 | return err; | |
5032 | } | |
5033 | ||
5034 | if (!map_is_reuse_compat(map, pin_fd)) { | |
5035 | pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n", | |
5036 | map->pin_path); | |
5037 | close(pin_fd); | |
5038 | return -EINVAL; | |
5039 | } | |
5040 | ||
5041 | err = bpf_map__reuse_fd(map, pin_fd); | |
d0f325c3 | 5042 | close(pin_fd); |
e3ba8e4e | 5043 | if (err) |
57a00f41 | 5044 | return err; |
e3ba8e4e | 5045 | |
57a00f41 THJ |
5046 | map->pinned = true; |
5047 | pr_debug("reused pinned map at '%s'\n", map->pin_path); | |
5048 | ||
5049 | return 0; | |
5050 | } | |
5051 | ||
d859900c DB |
5052 | static int |
5053 | bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map) | |
5054 | { | |
166750bc | 5055 | enum libbpf_map_type map_type = map->libbpf_type; |
d859900c DB |
5056 | char *cp, errmsg[STRERR_BUFSIZE]; |
5057 | int err, zero = 0; | |
d859900c | 5058 | |
67234743 AS |
5059 | if (obj->gen_loader) { |
5060 | bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps, | |
5061 | map->mmaped, map->def.value_size); | |
5062 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) | |
5063 | bpf_gen__map_freeze(obj->gen_loader, map - obj->maps); | |
5064 | return 0; | |
5065 | } | |
2e7ba4f8 | 5066 | |
eba9c5f4 AN |
5067 | err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0); |
5068 | if (err) { | |
5069 | err = -errno; | |
5070 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
5071 | pr_warn("Error setting initial map(%s) contents: %s\n", | |
5072 | map->name, cp); | |
5073 | return err; | |
5074 | } | |
d859900c | 5075 | |
81bfdd08 AN |
5076 | /* Freeze .rodata and .kconfig map as read-only from syscall side. */ |
5077 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) { | |
d859900c DB |
5078 | err = bpf_map_freeze(map->fd); |
5079 | if (err) { | |
eba9c5f4 AN |
5080 | err = -errno; |
5081 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e KW |
5082 | pr_warn("Error freezing map(%s) as read-only: %s\n", |
5083 | map->name, cp); | |
eba9c5f4 | 5084 | return err; |
d859900c DB |
5085 | } |
5086 | } | |
eba9c5f4 | 5087 | return 0; |
d859900c DB |
5088 | } |
5089 | ||
2d39d7c5 AN |
5090 | static void bpf_map__destroy(struct bpf_map *map); |
5091 | ||
f08c18e0 AN |
5092 | static bool map_is_created(const struct bpf_map *map) |
5093 | { | |
5094 | return map->obj->loaded || map->reused; | |
5095 | } | |
5096 | ||
67234743 | 5097 | static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner) |
2d39d7c5 | 5098 | { |
992c4225 | 5099 | LIBBPF_OPTS(bpf_map_create_opts, create_attr); |
2d39d7c5 | 5100 | struct bpf_map_def *def = &map->def; |
992c4225 | 5101 | const char *map_name = NULL; |
dac645b9 | 5102 | int err = 0, map_fd; |
2d39d7c5 | 5103 | |
9ca1f56a | 5104 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
992c4225 | 5105 | map_name = map->name; |
2d39d7c5 | 5106 | create_attr.map_ifindex = map->map_ifindex; |
2d39d7c5 | 5107 | create_attr.map_flags = def->map_flags; |
1bdb6c9a | 5108 | create_attr.numa_node = map->numa_node; |
47512102 | 5109 | create_attr.map_extra = map->map_extra; |
6b434b61 AN |
5110 | create_attr.token_fd = obj->token_fd; |
5111 | if (obj->token_fd) | |
5112 | create_attr.map_flags |= BPF_F_TOKEN_FD; | |
2d39d7c5 | 5113 | |
9e926acd | 5114 | if (bpf_map__is_struct_ops(map)) { |
992c4225 | 5115 | create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id; |
9e926acd KFL |
5116 | if (map->mod_btf_fd >= 0) { |
5117 | create_attr.value_type_btf_obj_fd = map->mod_btf_fd; | |
5118 | create_attr.map_flags |= BPF_F_VTYPE_BTF_OBJ_FD; | |
5119 | } | |
5120 | } | |
2d39d7c5 | 5121 | |
262cfb74 | 5122 | if (obj->btf && btf__fd(obj->btf) >= 0) { |
2d39d7c5 AN |
5123 | create_attr.btf_fd = btf__fd(obj->btf); |
5124 | create_attr.btf_key_type_id = map->btf_key_type_id; | |
5125 | create_attr.btf_value_type_id = map->btf_value_type_id; | |
5126 | } | |
5127 | ||
646f02ff AN |
5128 | if (bpf_map_type__is_map_in_map(def->type)) { |
5129 | if (map->inner_map) { | |
f04deb90 AG |
5130 | err = map_set_def_max_entries(map->inner_map); |
5131 | if (err) | |
5132 | return err; | |
67234743 | 5133 | err = bpf_object__create_map(obj, map->inner_map, true); |
646f02ff AN |
5134 | if (err) { |
5135 | pr_warn("map '%s': failed to create inner map: %d\n", | |
5136 | map->name, err); | |
5137 | return err; | |
5138 | } | |
f08c18e0 | 5139 | map->inner_map_fd = map->inner_map->fd; |
646f02ff AN |
5140 | } |
5141 | if (map->inner_map_fd >= 0) | |
5142 | create_attr.inner_map_fd = map->inner_map_fd; | |
5143 | } | |
5144 | ||
f7310523 HC |
5145 | switch (def->type) { |
5146 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
5147 | case BPF_MAP_TYPE_CGROUP_ARRAY: | |
5148 | case BPF_MAP_TYPE_STACK_TRACE: | |
5149 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
5150 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
5151 | case BPF_MAP_TYPE_DEVMAP: | |
5152 | case BPF_MAP_TYPE_DEVMAP_HASH: | |
5153 | case BPF_MAP_TYPE_CPUMAP: | |
5154 | case BPF_MAP_TYPE_XSKMAP: | |
5155 | case BPF_MAP_TYPE_SOCKMAP: | |
5156 | case BPF_MAP_TYPE_SOCKHASH: | |
5157 | case BPF_MAP_TYPE_QUEUE: | |
5158 | case BPF_MAP_TYPE_STACK: | |
79ff13e9 | 5159 | case BPF_MAP_TYPE_ARENA: |
f7310523 HC |
5160 | create_attr.btf_fd = 0; |
5161 | create_attr.btf_key_type_id = 0; | |
5162 | create_attr.btf_value_type_id = 0; | |
5163 | map->btf_key_type_id = 0; | |
5164 | map->btf_value_type_id = 0; | |
3644d285 KFL |
5165 | break; |
5166 | case BPF_MAP_TYPE_STRUCT_OPS: | |
5167 | create_attr.btf_value_type_id = 0; | |
5168 | break; | |
f7310523 HC |
5169 | default: |
5170 | break; | |
5171 | } | |
5172 | ||
67234743 | 5173 | if (obj->gen_loader) { |
992c4225 | 5174 | bpf_gen__map_create(obj->gen_loader, def->type, map_name, |
a4fbfdd7 | 5175 | def->key_size, def->value_size, def->max_entries, |
992c4225 | 5176 | &create_attr, is_inner ? -1 : map - obj->maps); |
dac645b9 AN |
5177 | /* We keep pretenting we have valid FD to pass various fd >= 0 |
5178 | * checks by just keeping original placeholder FDs in place. | |
5179 | * See bpf_object__add_map() comment. | |
5180 | * This placeholder fd will not be used with any syscall and | |
5181 | * will be reset to -1 eventually. | |
67234743 | 5182 | */ |
dac645b9 | 5183 | map_fd = map->fd; |
67234743 | 5184 | } else { |
dac645b9 AN |
5185 | map_fd = bpf_map_create(def->type, map_name, |
5186 | def->key_size, def->value_size, | |
5187 | def->max_entries, &create_attr); | |
67234743 | 5188 | } |
dac645b9 | 5189 | if (map_fd < 0 && (create_attr.btf_key_type_id || create_attr.btf_value_type_id)) { |
2d39d7c5 | 5190 | char *cp, errmsg[STRERR_BUFSIZE]; |
2d39d7c5 | 5191 | |
a21ab4c5 | 5192 | err = -errno; |
2d39d7c5 AN |
5193 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); |
5194 | pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n", | |
5195 | map->name, cp, err); | |
5196 | create_attr.btf_fd = 0; | |
5197 | create_attr.btf_key_type_id = 0; | |
5198 | create_attr.btf_value_type_id = 0; | |
5199 | map->btf_key_type_id = 0; | |
5200 | map->btf_value_type_id = 0; | |
dac645b9 AN |
5201 | map_fd = bpf_map_create(def->type, map_name, |
5202 | def->key_size, def->value_size, | |
5203 | def->max_entries, &create_attr); | |
2d39d7c5 AN |
5204 | } |
5205 | ||
646f02ff | 5206 | if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) { |
67234743 AS |
5207 | if (obj->gen_loader) |
5208 | map->inner_map->fd = -1; | |
646f02ff AN |
5209 | bpf_map__destroy(map->inner_map); |
5210 | zfree(&map->inner_map); | |
5211 | } | |
5212 | ||
dac645b9 AN |
5213 | if (map_fd < 0) |
5214 | return map_fd; | |
5215 | ||
5216 | /* obj->gen_loader case, prevent reuse_fd() from closing map_fd */ | |
5217 | if (map->fd == map_fd) | |
5218 | return 0; | |
5219 | ||
5220 | /* Keep placeholder FD value but now point it to the BPF map object. | |
5221 | * This way everything that relied on this map's FD (e.g., relocated | |
5222 | * ldimm64 instructions) will stay valid and won't need adjustments. | |
5223 | * map->fd stays valid but now point to what map_fd points to. | |
5224 | */ | |
5225 | return reuse_fd(map->fd, map_fd); | |
2d39d7c5 AN |
5226 | } |
5227 | ||
341ac5ff | 5228 | static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map) |
a0f2b7ac HL |
5229 | { |
5230 | const struct bpf_map *targ_map; | |
5231 | unsigned int i; | |
67234743 | 5232 | int fd, err = 0; |
a0f2b7ac HL |
5233 | |
5234 | for (i = 0; i < map->init_slots_sz; i++) { | |
5235 | if (!map->init_slots[i]) | |
5236 | continue; | |
5237 | ||
5238 | targ_map = map->init_slots[i]; | |
f08c18e0 | 5239 | fd = targ_map->fd; |
341ac5ff | 5240 | |
67234743 | 5241 | if (obj->gen_loader) { |
be05c944 AS |
5242 | bpf_gen__populate_outer_map(obj->gen_loader, |
5243 | map - obj->maps, i, | |
5244 | targ_map - obj->maps); | |
67234743 AS |
5245 | } else { |
5246 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
5247 | } | |
a0f2b7ac HL |
5248 | if (err) { |
5249 | err = -errno; | |
5250 | pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n", | |
341ac5ff | 5251 | map->name, i, targ_map->name, fd, err); |
a0f2b7ac HL |
5252 | return err; |
5253 | } | |
5254 | pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n", | |
5255 | map->name, i, targ_map->name, fd); | |
5256 | } | |
5257 | ||
5258 | zfree(&map->init_slots); | |
5259 | map->init_slots_sz = 0; | |
5260 | ||
5261 | return 0; | |
5262 | } | |
5263 | ||
341ac5ff HC |
5264 | static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map) |
5265 | { | |
5266 | const struct bpf_program *targ_prog; | |
5267 | unsigned int i; | |
5268 | int fd, err; | |
5269 | ||
5270 | if (obj->gen_loader) | |
5271 | return -ENOTSUP; | |
5272 | ||
5273 | for (i = 0; i < map->init_slots_sz; i++) { | |
5274 | if (!map->init_slots[i]) | |
5275 | continue; | |
5276 | ||
5277 | targ_prog = map->init_slots[i]; | |
5278 | fd = bpf_program__fd(targ_prog); | |
5279 | ||
5280 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
5281 | if (err) { | |
5282 | err = -errno; | |
5283 | pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %d\n", | |
5284 | map->name, i, targ_prog->name, fd, err); | |
5285 | return err; | |
5286 | } | |
5287 | pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n", | |
5288 | map->name, i, targ_prog->name, fd); | |
5289 | } | |
5290 | ||
5291 | zfree(&map->init_slots); | |
5292 | map->init_slots_sz = 0; | |
5293 | ||
5294 | return 0; | |
5295 | } | |
5296 | ||
5297 | static int bpf_object_init_prog_arrays(struct bpf_object *obj) | |
5298 | { | |
5299 | struct bpf_map *map; | |
5300 | int i, err; | |
5301 | ||
5302 | for (i = 0; i < obj->nr_maps; i++) { | |
5303 | map = &obj->maps[i]; | |
5304 | ||
5305 | if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY) | |
5306 | continue; | |
5307 | ||
5308 | err = init_prog_array_slots(obj, map); | |
dac645b9 | 5309 | if (err < 0) |
341ac5ff | 5310 | return err; |
341ac5ff HC |
5311 | } |
5312 | return 0; | |
5313 | } | |
5314 | ||
a4fbfdd7 ST |
5315 | static int map_set_def_max_entries(struct bpf_map *map) |
5316 | { | |
5317 | if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) { | |
5318 | int nr_cpus; | |
5319 | ||
5320 | nr_cpus = libbpf_num_possible_cpus(); | |
5321 | if (nr_cpus < 0) { | |
5322 | pr_warn("map '%s': failed to determine number of system CPUs: %d\n", | |
5323 | map->name, nr_cpus); | |
5324 | return nr_cpus; | |
5325 | } | |
5326 | pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus); | |
5327 | map->def.max_entries = nr_cpus; | |
5328 | } | |
5329 | ||
5330 | return 0; | |
5331 | } | |
5332 | ||
52d3352e WN |
5333 | static int |
5334 | bpf_object__create_maps(struct bpf_object *obj) | |
5335 | { | |
2d39d7c5 AN |
5336 | struct bpf_map *map; |
5337 | char *cp, errmsg[STRERR_BUFSIZE]; | |
5338 | unsigned int i, j; | |
8a138aed | 5339 | int err; |
043c5bb3 | 5340 | bool retried; |
52d3352e | 5341 | |
9d759a9b | 5342 | for (i = 0; i < obj->nr_maps; i++) { |
2d39d7c5 | 5343 | map = &obj->maps[i]; |
8a138aed | 5344 | |
16e0c35c AN |
5345 | /* To support old kernels, we skip creating global data maps |
5346 | * (.rodata, .data, .kconfig, etc); later on, during program | |
5347 | * loading, if we detect that at least one of the to-be-loaded | |
5348 | * programs is referencing any global data map, we'll error | |
5349 | * out with program name and relocation index logged. | |
5350 | * This approach allows to accommodate Clang emitting | |
5351 | * unnecessary .rodata.str1.1 sections for string literals, | |
5352 | * but also it allows to have CO-RE applications that use | |
5353 | * global variables in some of BPF programs, but not others. | |
5354 | * If those global variable-using programs are not loaded at | |
5355 | * runtime due to bpf_program__set_autoload(prog, false), | |
5356 | * bpf_object loading will succeed just fine even on old | |
5357 | * kernels. | |
5358 | */ | |
ec41817b AN |
5359 | if (bpf_map__is_internal(map) && !kernel_supports(obj, FEAT_GLOBAL_DATA)) |
5360 | map->autocreate = false; | |
5361 | ||
5362 | if (!map->autocreate) { | |
5363 | pr_debug("map '%s': skipped auto-creating...\n", map->name); | |
16e0c35c | 5364 | continue; |
229fae38 | 5365 | } |
16e0c35c | 5366 | |
a4fbfdd7 ST |
5367 | err = map_set_def_max_entries(map); |
5368 | if (err) | |
5369 | goto err_out; | |
5370 | ||
043c5bb3 MP |
5371 | retried = false; |
5372 | retry: | |
57a00f41 THJ |
5373 | if (map->pin_path) { |
5374 | err = bpf_object__reuse_map(map); | |
5375 | if (err) { | |
2d39d7c5 | 5376 | pr_warn("map '%s': error reusing pinned map\n", |
57a00f41 | 5377 | map->name); |
2d39d7c5 | 5378 | goto err_out; |
57a00f41 | 5379 | } |
043c5bb3 MP |
5380 | if (retried && map->fd < 0) { |
5381 | pr_warn("map '%s': cannot find pinned map\n", | |
5382 | map->name); | |
5383 | err = -ENOENT; | |
5384 | goto err_out; | |
5385 | } | |
57a00f41 THJ |
5386 | } |
5387 | ||
fa98b54b | 5388 | if (map->reused) { |
2d39d7c5 | 5389 | pr_debug("map '%s': skipping creation (preset fd=%d)\n", |
26736eb9 | 5390 | map->name, map->fd); |
2c193d32 | 5391 | } else { |
67234743 | 5392 | err = bpf_object__create_map(obj, map, false); |
2c193d32 | 5393 | if (err) |
d859900c | 5394 | goto err_out; |
d859900c | 5395 | |
2c193d32 HL |
5396 | pr_debug("map '%s': created successfully, fd=%d\n", |
5397 | map->name, map->fd); | |
646f02ff | 5398 | |
2c193d32 HL |
5399 | if (bpf_map__is_internal(map)) { |
5400 | err = bpf_object__populate_internal_map(obj, map); | |
dac645b9 | 5401 | if (err < 0) |
2c193d32 | 5402 | goto err_out; |
d859900c | 5403 | } |
79ff13e9 | 5404 | if (map->def.type == BPF_MAP_TYPE_ARENA) { |
5ab8cb89 AN |
5405 | map->mmaped = mmap((void *)(long)map->map_extra, |
5406 | bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE, | |
79ff13e9 AS |
5407 | map->map_extra ? MAP_SHARED | MAP_FIXED : MAP_SHARED, |
5408 | map->fd, 0); | |
5409 | if (map->mmaped == MAP_FAILED) { | |
5410 | err = -errno; | |
5411 | map->mmaped = NULL; | |
5412 | pr_warn("map '%s': failed to mmap arena: %d\n", | |
5413 | map->name, err); | |
5414 | return err; | |
5415 | } | |
2e7ba4f8 AN |
5416 | if (obj->arena_data) { |
5417 | memcpy(map->mmaped, obj->arena_data, obj->arena_data_sz); | |
5418 | zfree(&obj->arena_data); | |
5419 | } | |
79ff13e9 | 5420 | } |
341ac5ff HC |
5421 | if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) { |
5422 | err = init_map_in_map_slots(obj, map); | |
dac645b9 | 5423 | if (err < 0) |
646f02ff | 5424 | goto err_out; |
646f02ff | 5425 | } |
646f02ff AN |
5426 | } |
5427 | ||
57a00f41 THJ |
5428 | if (map->pin_path && !map->pinned) { |
5429 | err = bpf_map__pin(map, NULL); | |
5430 | if (err) { | |
043c5bb3 MP |
5431 | if (!retried && err == -EEXIST) { |
5432 | retried = true; | |
5433 | goto retry; | |
5434 | } | |
2d39d7c5 AN |
5435 | pr_warn("map '%s': failed to auto-pin at '%s': %d\n", |
5436 | map->name, map->pin_path, err); | |
2d39d7c5 | 5437 | goto err_out; |
57a00f41 THJ |
5438 | } |
5439 | } | |
52d3352e WN |
5440 | } |
5441 | ||
52d3352e | 5442 | return 0; |
2d39d7c5 AN |
5443 | |
5444 | err_out: | |
5445 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
5446 | pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err); | |
5447 | pr_perm_msg(err); | |
5448 | for (j = 0; j < i; j++) | |
5449 | zclose(obj->maps[j].fd); | |
5450 | return err; | |
52d3352e WN |
5451 | } |
5452 | ||
ddc7c304 AN |
5453 | static bool bpf_core_is_flavor_sep(const char *s) |
5454 | { | |
5455 | /* check X___Y name pattern, where X and Y are not underscores */ | |
5456 | return s[0] != '_' && /* X */ | |
5457 | s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */ | |
5458 | s[4] != '_'; /* Y */ | |
5459 | } | |
5460 | ||
5461 | /* Given 'some_struct_name___with_flavor' return the length of a name prefix | |
5462 | * before last triple underscore. Struct name part after last triple | |
5463 | * underscore is ignored by BPF CO-RE relocation during relocation matching. | |
5464 | */ | |
b0588390 | 5465 | size_t bpf_core_essential_name_len(const char *name) |
ddc7c304 AN |
5466 | { |
5467 | size_t n = strlen(name); | |
5468 | int i; | |
5469 | ||
5470 | for (i = n - 5; i >= 0; i--) { | |
5471 | if (bpf_core_is_flavor_sep(name + i)) | |
5472 | return i + 1; | |
5473 | } | |
5474 | return n; | |
5475 | } | |
5476 | ||
8de6cae4 | 5477 | void bpf_core_free_cands(struct bpf_core_cand_list *cands) |
ddc7c304 | 5478 | { |
8de6cae4 MV |
5479 | if (!cands) |
5480 | return; | |
5481 | ||
0f7515ca AN |
5482 | free(cands->cands); |
5483 | free(cands); | |
ddc7c304 AN |
5484 | } |
5485 | ||
8de6cae4 MV |
5486 | int bpf_core_add_cands(struct bpf_core_cand *local_cand, |
5487 | size_t local_essent_len, | |
5488 | const struct btf *targ_btf, | |
5489 | const char *targ_btf_name, | |
5490 | int targ_start_id, | |
5491 | struct bpf_core_cand_list *cands) | |
ddc7c304 | 5492 | { |
301ba4d7 | 5493 | struct bpf_core_cand *new_cands, *cand; |
03d5b991 AN |
5494 | const struct btf_type *t, *local_t; |
5495 | const char *targ_name, *local_name; | |
0f7515ca AN |
5496 | size_t targ_essent_len; |
5497 | int n, i; | |
ddc7c304 | 5498 | |
03d5b991 AN |
5499 | local_t = btf__type_by_id(local_cand->btf, local_cand->id); |
5500 | local_name = btf__str_by_offset(local_cand->btf, local_t->name_off); | |
5501 | ||
6a886de0 HC |
5502 | n = btf__type_cnt(targ_btf); |
5503 | for (i = targ_start_id; i < n; i++) { | |
ddc7c304 | 5504 | t = btf__type_by_id(targ_btf, i); |
23b2a3a8 | 5505 | if (!btf_kind_core_compat(t, local_t)) |
ddc7c304 AN |
5506 | continue; |
5507 | ||
3fc32f40 AN |
5508 | targ_name = btf__name_by_offset(targ_btf, t->name_off); |
5509 | if (str_is_empty(targ_name)) | |
d121e1d3 AN |
5510 | continue; |
5511 | ||
ddc7c304 AN |
5512 | targ_essent_len = bpf_core_essential_name_len(targ_name); |
5513 | if (targ_essent_len != local_essent_len) | |
5514 | continue; | |
5515 | ||
03d5b991 | 5516 | if (strncmp(local_name, targ_name, local_essent_len) != 0) |
0f7515ca AN |
5517 | continue; |
5518 | ||
5519 | pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n", | |
03d5b991 AN |
5520 | local_cand->id, btf_kind_str(local_t), |
5521 | local_name, i, btf_kind_str(t), targ_name, | |
0f7515ca AN |
5522 | targ_btf_name); |
5523 | new_cands = libbpf_reallocarray(cands->cands, cands->len + 1, | |
5524 | sizeof(*cands->cands)); | |
5525 | if (!new_cands) | |
5526 | return -ENOMEM; | |
5527 | ||
5528 | cand = &new_cands[cands->len]; | |
5529 | cand->btf = targ_btf; | |
0f7515ca AN |
5530 | cand->id = i; |
5531 | ||
5532 | cands->cands = new_cands; | |
5533 | cands->len++; | |
ddc7c304 | 5534 | } |
0f7515ca AN |
5535 | return 0; |
5536 | } | |
5537 | ||
4f33a53d AN |
5538 | static int load_module_btfs(struct bpf_object *obj) |
5539 | { | |
5540 | struct bpf_btf_info info; | |
5541 | struct module_btf *mod_btf; | |
5542 | struct btf *btf; | |
5543 | char name[64]; | |
5544 | __u32 id = 0, len; | |
5545 | int err, fd; | |
5546 | ||
5547 | if (obj->btf_modules_loaded) | |
5548 | return 0; | |
5549 | ||
67234743 AS |
5550 | if (obj->gen_loader) |
5551 | return 0; | |
5552 | ||
4f33a53d AN |
5553 | /* don't do this again, even if we find no module BTFs */ |
5554 | obj->btf_modules_loaded = true; | |
5555 | ||
5556 | /* kernel too old to support module BTFs */ | |
9ca1f56a | 5557 | if (!kernel_supports(obj, FEAT_MODULE_BTF)) |
4f33a53d AN |
5558 | return 0; |
5559 | ||
5560 | while (true) { | |
5561 | err = bpf_btf_get_next_id(id, &id); | |
5562 | if (err && errno == ENOENT) | |
5563 | return 0; | |
2d2c9516 AT |
5564 | if (err && errno == EPERM) { |
5565 | pr_debug("skipping module BTFs loading, missing privileges\n"); | |
5566 | return 0; | |
5567 | } | |
4f33a53d AN |
5568 | if (err) { |
5569 | err = -errno; | |
5570 | pr_warn("failed to iterate BTF objects: %d\n", err); | |
5571 | return err; | |
ddc7c304 | 5572 | } |
4f33a53d AN |
5573 | |
5574 | fd = bpf_btf_get_fd_by_id(id); | |
5575 | if (fd < 0) { | |
5576 | if (errno == ENOENT) | |
5577 | continue; /* expected race: BTF was unloaded */ | |
5578 | err = -errno; | |
5579 | pr_warn("failed to get BTF object #%d FD: %d\n", id, err); | |
5580 | return err; | |
5581 | } | |
5582 | ||
5583 | len = sizeof(info); | |
5584 | memset(&info, 0, sizeof(info)); | |
5585 | info.name = ptr_to_u64(name); | |
5586 | info.name_len = sizeof(name); | |
5587 | ||
629dfc66 | 5588 | err = bpf_btf_get_info_by_fd(fd, &info, &len); |
4f33a53d AN |
5589 | if (err) { |
5590 | err = -errno; | |
5591 | pr_warn("failed to get BTF object #%d info: %d\n", id, err); | |
91abb4a6 | 5592 | goto err_out; |
4f33a53d AN |
5593 | } |
5594 | ||
5595 | /* ignore non-module BTFs */ | |
5596 | if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) { | |
5597 | close(fd); | |
5598 | continue; | |
5599 | } | |
5600 | ||
5601 | btf = btf_get_from_fd(fd, obj->btf_vmlinux); | |
e9fc3ce9 AN |
5602 | err = libbpf_get_error(btf); |
5603 | if (err) { | |
5604 | pr_warn("failed to load module [%s]'s BTF object #%d: %d\n", | |
5605 | name, id, err); | |
91abb4a6 | 5606 | goto err_out; |
4f33a53d AN |
5607 | } |
5608 | ||
3b029e06 | 5609 | err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap, |
e3ba8e4e | 5610 | sizeof(*obj->btf_modules), obj->btf_module_cnt + 1); |
4f33a53d | 5611 | if (err) |
91abb4a6 | 5612 | goto err_out; |
4f33a53d AN |
5613 | |
5614 | mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; | |
5615 | ||
5616 | mod_btf->btf = btf; | |
5617 | mod_btf->id = id; | |
91abb4a6 | 5618 | mod_btf->fd = fd; |
4f33a53d | 5619 | mod_btf->name = strdup(name); |
91abb4a6 AN |
5620 | if (!mod_btf->name) { |
5621 | err = -ENOMEM; | |
5622 | goto err_out; | |
5623 | } | |
5624 | continue; | |
5625 | ||
5626 | err_out: | |
5627 | close(fd); | |
5628 | return err; | |
ddc7c304 | 5629 | } |
4f33a53d AN |
5630 | |
5631 | return 0; | |
5632 | } | |
5633 | ||
301ba4d7 | 5634 | static struct bpf_core_cand_list * |
0f7515ca AN |
5635 | bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id) |
5636 | { | |
301ba4d7 AS |
5637 | struct bpf_core_cand local_cand = {}; |
5638 | struct bpf_core_cand_list *cands; | |
4f33a53d | 5639 | const struct btf *main_btf; |
03d5b991 AN |
5640 | const struct btf_type *local_t; |
5641 | const char *local_name; | |
0f7515ca | 5642 | size_t local_essent_len; |
4f33a53d | 5643 | int err, i; |
0f7515ca AN |
5644 | |
5645 | local_cand.btf = local_btf; | |
03d5b991 AN |
5646 | local_cand.id = local_type_id; |
5647 | local_t = btf__type_by_id(local_btf, local_type_id); | |
5648 | if (!local_t) | |
0f7515ca AN |
5649 | return ERR_PTR(-EINVAL); |
5650 | ||
03d5b991 AN |
5651 | local_name = btf__name_by_offset(local_btf, local_t->name_off); |
5652 | if (str_is_empty(local_name)) | |
0f7515ca | 5653 | return ERR_PTR(-EINVAL); |
03d5b991 | 5654 | local_essent_len = bpf_core_essential_name_len(local_name); |
0f7515ca AN |
5655 | |
5656 | cands = calloc(1, sizeof(*cands)); | |
5657 | if (!cands) | |
5658 | return ERR_PTR(-ENOMEM); | |
5659 | ||
5660 | /* Attempt to find target candidates in vmlinux BTF first */ | |
4f33a53d AN |
5661 | main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux; |
5662 | err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands); | |
5663 | if (err) | |
5664 | goto err_out; | |
5665 | ||
5666 | /* if vmlinux BTF has any candidate, don't got for module BTFs */ | |
5667 | if (cands->len) | |
5668 | return cands; | |
5669 | ||
5670 | /* if vmlinux BTF was overridden, don't attempt to load module BTFs */ | |
5671 | if (obj->btf_vmlinux_override) | |
5672 | return cands; | |
5673 | ||
5674 | /* now look through module BTFs, trying to still find candidates */ | |
5675 | err = load_module_btfs(obj); | |
5676 | if (err) | |
5677 | goto err_out; | |
5678 | ||
5679 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
5680 | err = bpf_core_add_cands(&local_cand, local_essent_len, | |
5681 | obj->btf_modules[i].btf, | |
5682 | obj->btf_modules[i].name, | |
6a886de0 | 5683 | btf__type_cnt(obj->btf_vmlinux), |
4f33a53d AN |
5684 | cands); |
5685 | if (err) | |
5686 | goto err_out; | |
0f7515ca AN |
5687 | } |
5688 | ||
5689 | return cands; | |
ddc7c304 | 5690 | err_out: |
4f33a53d | 5691 | bpf_core_free_cands(cands); |
ddc7c304 AN |
5692 | return ERR_PTR(err); |
5693 | } | |
5694 | ||
3fc32f40 AN |
5695 | /* Check local and target types for compatibility. This check is used for |
5696 | * type-based CO-RE relocations and follow slightly different rules than | |
5697 | * field-based relocations. This function assumes that root types were already | |
5698 | * checked for name match. Beyond that initial root-level name check, names | |
5699 | * are completely ignored. Compatibility rules are as follows: | |
5700 | * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but | |
5701 | * kind should match for local and target types (i.e., STRUCT is not | |
5702 | * compatible with UNION); | |
5703 | * - for ENUMs, the size is ignored; | |
5704 | * - for INT, size and signedness are ignored; | |
5705 | * - for ARRAY, dimensionality is ignored, element types are checked for | |
5706 | * compatibility recursively; | |
5707 | * - CONST/VOLATILE/RESTRICT modifiers are ignored; | |
5708 | * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible; | |
5709 | * - FUNC_PROTOs are compatible if they have compatible signature: same | |
5710 | * number of input args and compatible return and argument types. | |
5711 | * These rules are not set in stone and probably will be adjusted as we get | |
5712 | * more experience with using BPF CO-RE relocations. | |
5713 | */ | |
b0588390 AS |
5714 | int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id, |
5715 | const struct btf *targ_btf, __u32 targ_id) | |
3fc32f40 | 5716 | { |
fd75733d | 5717 | return __bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id, 32); |
3fc32f40 AN |
5718 | } |
5719 | ||
ec6209c8 DM |
5720 | int bpf_core_types_match(const struct btf *local_btf, __u32 local_id, |
5721 | const struct btf *targ_btf, __u32 targ_id) | |
5722 | { | |
5723 | return __bpf_core_types_match(local_btf, local_id, targ_btf, targ_id, false, 32); | |
5724 | } | |
5725 | ||
c302378b | 5726 | static size_t bpf_core_hash_fn(const long key, void *ctx) |
ddc7c304 | 5727 | { |
c302378b | 5728 | return key; |
ddc7c304 AN |
5729 | } |
5730 | ||
c302378b | 5731 | static bool bpf_core_equal_fn(const long k1, const long k2, void *ctx) |
ddc7c304 AN |
5732 | { |
5733 | return k1 == k2; | |
5734 | } | |
5735 | ||
d0e92887 AS |
5736 | static int record_relo_core(struct bpf_program *prog, |
5737 | const struct bpf_core_relo *core_relo, int insn_idx) | |
5738 | { | |
5739 | struct reloc_desc *relos, *relo; | |
5740 | ||
5741 | relos = libbpf_reallocarray(prog->reloc_desc, | |
5742 | prog->nr_reloc + 1, sizeof(*relos)); | |
5743 | if (!relos) | |
5744 | return -ENOMEM; | |
5745 | relo = &relos[prog->nr_reloc]; | |
5746 | relo->type = RELO_CORE; | |
5747 | relo->insn_idx = insn_idx; | |
5748 | relo->core_relo = core_relo; | |
5749 | prog->reloc_desc = relos; | |
5750 | prog->nr_reloc++; | |
5751 | return 0; | |
5752 | } | |
5753 | ||
9fdc4273 AN |
5754 | static const struct bpf_core_relo *find_relo_core(struct bpf_program *prog, int insn_idx) |
5755 | { | |
5756 | struct reloc_desc *relo; | |
5757 | int i; | |
5758 | ||
5759 | for (i = 0; i < prog->nr_reloc; i++) { | |
5760 | relo = &prog->reloc_desc[i]; | |
5761 | if (relo->type != RELO_CORE || relo->insn_idx != insn_idx) | |
5762 | continue; | |
5763 | ||
5764 | return relo->core_relo; | |
5765 | } | |
5766 | ||
5767 | return NULL; | |
5768 | } | |
5769 | ||
adb8fa19 MV |
5770 | static int bpf_core_resolve_relo(struct bpf_program *prog, |
5771 | const struct bpf_core_relo *relo, | |
5772 | int relo_idx, | |
5773 | const struct btf *local_btf, | |
5774 | struct hashmap *cand_cache, | |
5775 | struct bpf_core_relo_res *targ_res) | |
3ee4f533 | 5776 | { |
78c1f8d0 | 5777 | struct bpf_core_spec specs_scratch[3] = {}; |
301ba4d7 | 5778 | struct bpf_core_cand_list *cands = NULL; |
3ee4f533 AS |
5779 | const char *prog_name = prog->name; |
5780 | const struct btf_type *local_type; | |
5781 | const char *local_name; | |
5782 | __u32 local_id = relo->type_id; | |
adb8fa19 | 5783 | int err; |
3ee4f533 AS |
5784 | |
5785 | local_type = btf__type_by_id(local_btf, local_id); | |
5786 | if (!local_type) | |
5787 | return -EINVAL; | |
5788 | ||
5789 | local_name = btf__name_by_offset(local_btf, local_type->name_off); | |
5790 | if (!local_name) | |
5791 | return -EINVAL; | |
5792 | ||
46334a0c | 5793 | if (relo->kind != BPF_CORE_TYPE_ID_LOCAL && |
c302378b | 5794 | !hashmap__find(cand_cache, local_id, &cands)) { |
3ee4f533 AS |
5795 | cands = bpf_core_find_cands(prog->obj, local_btf, local_id); |
5796 | if (IS_ERR(cands)) { | |
5797 | pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n", | |
5798 | prog_name, relo_idx, local_id, btf_kind_str(local_type), | |
5799 | local_name, PTR_ERR(cands)); | |
5800 | return PTR_ERR(cands); | |
5801 | } | |
c302378b | 5802 | err = hashmap__set(cand_cache, local_id, cands, NULL, NULL); |
3ee4f533 AS |
5803 | if (err) { |
5804 | bpf_core_free_cands(cands); | |
5805 | return err; | |
5806 | } | |
5807 | } | |
5808 | ||
adb8fa19 MV |
5809 | return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch, |
5810 | targ_res); | |
3ee4f533 AS |
5811 | } |
5812 | ||
ddc7c304 | 5813 | static int |
28b93c64 | 5814 | bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) |
ddc7c304 AN |
5815 | { |
5816 | const struct btf_ext_info_sec *sec; | |
adb8fa19 | 5817 | struct bpf_core_relo_res targ_res; |
28b93c64 | 5818 | const struct bpf_core_relo *rec; |
ddc7c304 AN |
5819 | const struct btf_ext_info *seg; |
5820 | struct hashmap_entry *entry; | |
5821 | struct hashmap *cand_cache = NULL; | |
5822 | struct bpf_program *prog; | |
adb8fa19 | 5823 | struct bpf_insn *insn; |
ddc7c304 | 5824 | const char *sec_name; |
11d5daa8 | 5825 | int i, err = 0, insn_idx, sec_idx, sec_num; |
ddc7c304 | 5826 | |
28b93c64 AN |
5827 | if (obj->btf_ext->core_relo_info.len == 0) |
5828 | return 0; | |
5829 | ||
0f7515ca AN |
5830 | if (targ_btf_path) { |
5831 | obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL); | |
e9fc3ce9 AN |
5832 | err = libbpf_get_error(obj->btf_vmlinux_override); |
5833 | if (err) { | |
0f7515ca AN |
5834 | pr_warn("failed to parse target BTF: %d\n", err); |
5835 | return err; | |
5836 | } | |
ddc7c304 AN |
5837 | } |
5838 | ||
5839 | cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL); | |
5840 | if (IS_ERR(cand_cache)) { | |
5841 | err = PTR_ERR(cand_cache); | |
5842 | goto out; | |
5843 | } | |
5844 | ||
28b93c64 | 5845 | seg = &obj->btf_ext->core_relo_info; |
11d5daa8 | 5846 | sec_num = 0; |
ddc7c304 | 5847 | for_each_btf_ext_sec(seg, sec) { |
11d5daa8 AN |
5848 | sec_idx = seg->sec_idxs[sec_num]; |
5849 | sec_num++; | |
5850 | ||
ddc7c304 AN |
5851 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); |
5852 | if (str_is_empty(sec_name)) { | |
5853 | err = -EINVAL; | |
5854 | goto out; | |
5855 | } | |
ddc7c304 | 5856 | |
11d5daa8 | 5857 | pr_debug("sec '%s': found %d CO-RE relocations\n", sec_name, sec->num_info); |
ddc7c304 AN |
5858 | |
5859 | for_each_btf_ext_rec(seg, sec, i, rec) { | |
adb8fa19 MV |
5860 | if (rec->insn_off % BPF_INSN_SZ) |
5861 | return -EINVAL; | |
db2b8b06 AN |
5862 | insn_idx = rec->insn_off / BPF_INSN_SZ; |
5863 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); | |
5864 | if (!prog) { | |
e89d57d9 AN |
5865 | /* When __weak subprog is "overridden" by another instance |
5866 | * of the subprog from a different object file, linker still | |
5867 | * appends all the .BTF.ext info that used to belong to that | |
5868 | * eliminated subprogram. | |
5869 | * This is similar to what x86-64 linker does for relocations. | |
5870 | * So just ignore such relocations just like we ignore | |
5871 | * subprog instructions when discovering subprograms. | |
5872 | */ | |
5873 | pr_debug("sec '%s': skipping CO-RE relocation #%d for insn #%d belonging to eliminated weak subprogram\n", | |
5874 | sec_name, i, insn_idx); | |
5875 | continue; | |
db2b8b06 | 5876 | } |
47f7cf63 AN |
5877 | /* no need to apply CO-RE relocation if the program is |
5878 | * not going to be loaded | |
5879 | */ | |
a3820c48 | 5880 | if (!prog->autoload) |
47f7cf63 | 5881 | continue; |
db2b8b06 | 5882 | |
adb8fa19 MV |
5883 | /* adjust insn_idx from section frame of reference to the local |
5884 | * program's frame of reference; (sub-)program code is not yet | |
5885 | * relocated, so it's enough to just subtract in-section offset | |
5886 | */ | |
5887 | insn_idx = insn_idx - prog->sec_insn_off; | |
5888 | if (insn_idx >= prog->insns_cnt) | |
5889 | return -EINVAL; | |
5890 | insn = &prog->insns[insn_idx]; | |
5891 | ||
185cfe83 AN |
5892 | err = record_relo_core(prog, rec, insn_idx); |
5893 | if (err) { | |
5894 | pr_warn("prog '%s': relo #%d: failed to record relocation: %d\n", | |
5895 | prog->name, i, err); | |
5896 | goto out; | |
adb8fa19 MV |
5897 | } |
5898 | ||
185cfe83 AN |
5899 | if (prog->obj->gen_loader) |
5900 | continue; | |
5901 | ||
adb8fa19 | 5902 | err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res); |
ddc7c304 | 5903 | if (err) { |
be18010e | 5904 | pr_warn("prog '%s': relo #%d: failed to relocate: %d\n", |
9c0f8cbd | 5905 | prog->name, i, err); |
ddc7c304 AN |
5906 | goto out; |
5907 | } | |
adb8fa19 MV |
5908 | |
5909 | err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res); | |
5910 | if (err) { | |
5911 | pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n", | |
5912 | prog->name, i, insn_idx, err); | |
5913 | goto out; | |
5914 | } | |
ddc7c304 AN |
5915 | } |
5916 | } | |
5917 | ||
5918 | out: | |
4f33a53d | 5919 | /* obj->btf_vmlinux and module BTFs are freed after object load */ |
0f7515ca AN |
5920 | btf__free(obj->btf_vmlinux_override); |
5921 | obj->btf_vmlinux_override = NULL; | |
5922 | ||
ddc7c304 AN |
5923 | if (!IS_ERR_OR_NULL(cand_cache)) { |
5924 | hashmap__for_each_entry(cand_cache, entry, i) { | |
c302378b | 5925 | bpf_core_free_cands(entry->pvalue); |
ddc7c304 AN |
5926 | } |
5927 | hashmap__free(cand_cache); | |
5928 | } | |
5929 | return err; | |
5930 | } | |
5931 | ||
ec41817b | 5932 | /* base map load ldimm64 special constant, used also for log fixup logic */ |
3055ddd6 AN |
5933 | #define POISON_LDIMM64_MAP_BASE 2001000000 |
5934 | #define POISON_LDIMM64_MAP_PFX "200100" | |
ec41817b AN |
5935 | |
5936 | static void poison_map_ldimm64(struct bpf_program *prog, int relo_idx, | |
5937 | int insn_idx, struct bpf_insn *insn, | |
5938 | int map_idx, const struct bpf_map *map) | |
5939 | { | |
5940 | int i; | |
5941 | ||
5942 | pr_debug("prog '%s': relo #%d: poisoning insn #%d that loads map #%d '%s'\n", | |
5943 | prog->name, relo_idx, insn_idx, map_idx, map->name); | |
5944 | ||
5945 | /* we turn single ldimm64 into two identical invalid calls */ | |
5946 | for (i = 0; i < 2; i++) { | |
5947 | insn->code = BPF_JMP | BPF_CALL; | |
5948 | insn->dst_reg = 0; | |
5949 | insn->src_reg = 0; | |
5950 | insn->off = 0; | |
5951 | /* if this instruction is reachable (not a dead code), | |
5952 | * verifier will complain with something like: | |
5953 | * invalid func unknown#2001000123 | |
5954 | * where lower 123 is map index into obj->maps[] array | |
5955 | */ | |
3055ddd6 | 5956 | insn->imm = POISON_LDIMM64_MAP_BASE + map_idx; |
ec41817b AN |
5957 | |
5958 | insn++; | |
5959 | } | |
5960 | } | |
5961 | ||
05b6f766 AN |
5962 | /* unresolved kfunc call special constant, used also for log fixup logic */ |
5963 | #define POISON_CALL_KFUNC_BASE 2002000000 | |
5964 | #define POISON_CALL_KFUNC_PFX "2002" | |
5965 | ||
5966 | static void poison_kfunc_call(struct bpf_program *prog, int relo_idx, | |
5967 | int insn_idx, struct bpf_insn *insn, | |
5968 | int ext_idx, const struct extern_desc *ext) | |
5969 | { | |
5970 | pr_debug("prog '%s': relo #%d: poisoning insn #%d that calls kfunc '%s'\n", | |
5971 | prog->name, relo_idx, insn_idx, ext->name); | |
5972 | ||
5973 | /* we turn kfunc call into invalid helper call with identifiable constant */ | |
5974 | insn->code = BPF_JMP | BPF_CALL; | |
5975 | insn->dst_reg = 0; | |
5976 | insn->src_reg = 0; | |
5977 | insn->off = 0; | |
5978 | /* if this instruction is reachable (not a dead code), | |
5979 | * verifier will complain with something like: | |
5980 | * invalid func unknown#2001000123 | |
5981 | * where lower 123 is extern index into obj->externs[] array | |
5982 | */ | |
5983 | insn->imm = POISON_CALL_KFUNC_BASE + ext_idx; | |
5984 | } | |
5985 | ||
c3c55696 AN |
5986 | /* Relocate data references within program code: |
5987 | * - map references; | |
5988 | * - global variable references; | |
5989 | * - extern references. | |
5990 | */ | |
48cca7e4 | 5991 | static int |
c3c55696 | 5992 | bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog) |
8a47a6c5 | 5993 | { |
c3c55696 | 5994 | int i; |
8a47a6c5 WN |
5995 | |
5996 | for (i = 0; i < prog->nr_reloc; i++) { | |
53f8dd43 | 5997 | struct reloc_desc *relo = &prog->reloc_desc[i]; |
166750bc | 5998 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; |
ec41817b | 5999 | const struct bpf_map *map; |
2e33efe3 | 6000 | struct extern_desc *ext; |
8a47a6c5 | 6001 | |
166750bc AN |
6002 | switch (relo->type) { |
6003 | case RELO_LD64: | |
ec41817b | 6004 | map = &obj->maps[relo->map_idx]; |
e2fa0156 AS |
6005 | if (obj->gen_loader) { |
6006 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX; | |
6007 | insn[0].imm = relo->map_idx; | |
ec41817b | 6008 | } else if (map->autocreate) { |
e2fa0156 | 6009 | insn[0].src_reg = BPF_PSEUDO_MAP_FD; |
ec41817b AN |
6010 | insn[0].imm = map->fd; |
6011 | } else { | |
6012 | poison_map_ldimm64(prog, i, relo->insn_idx, insn, | |
6013 | relo->map_idx, map); | |
e2fa0156 | 6014 | } |
166750bc AN |
6015 | break; |
6016 | case RELO_DATA: | |
ec41817b | 6017 | map = &obj->maps[relo->map_idx]; |
166750bc | 6018 | insn[1].imm = insn[0].imm + relo->sym_off; |
e2fa0156 AS |
6019 | if (obj->gen_loader) { |
6020 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
6021 | insn[0].imm = relo->map_idx; | |
ec41817b | 6022 | } else if (map->autocreate) { |
e2fa0156 | 6023 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; |
ec41817b AN |
6024 | insn[0].imm = map->fd; |
6025 | } else { | |
6026 | poison_map_ldimm64(prog, i, relo->insn_idx, insn, | |
6027 | relo->map_idx, map); | |
e2fa0156 | 6028 | } |
166750bc | 6029 | break; |
a18f7214 | 6030 | case RELO_EXTERN_LD64: |
3055ddd6 | 6031 | ext = &obj->externs[relo->ext_idx]; |
1c0c7074 | 6032 | if (ext->type == EXT_KCFG) { |
e2fa0156 AS |
6033 | if (obj->gen_loader) { |
6034 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
6035 | insn[0].imm = obj->kconfig_map_idx; | |
6036 | } else { | |
6037 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; | |
6038 | insn[0].imm = obj->maps[obj->kconfig_map_idx].fd; | |
6039 | } | |
1c0c7074 AN |
6040 | insn[1].imm = ext->kcfg.data_off; |
6041 | } else /* EXT_KSYM */ { | |
2211c825 | 6042 | if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */ |
d370bbe1 | 6043 | insn[0].src_reg = BPF_PSEUDO_BTF_ID; |
284d2587 AN |
6044 | insn[0].imm = ext->ksym.kernel_btf_id; |
6045 | insn[1].imm = ext->ksym.kernel_btf_obj_fd; | |
2211c825 | 6046 | } else { /* typeless ksyms or unresolved typed ksyms */ |
d370bbe1 HL |
6047 | insn[0].imm = (__u32)ext->ksym.addr; |
6048 | insn[1].imm = ext->ksym.addr >> 32; | |
6049 | } | |
1c0c7074 | 6050 | } |
166750bc | 6051 | break; |
a18f7214 | 6052 | case RELO_EXTERN_CALL: |
3055ddd6 | 6053 | ext = &obj->externs[relo->ext_idx]; |
5bd022ec | 6054 | insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL; |
466b2e13 KKD |
6055 | if (ext->is_set) { |
6056 | insn[0].imm = ext->ksym.kernel_btf_id; | |
6057 | insn[0].off = ext->ksym.btf_fd_idx; | |
05b6f766 AN |
6058 | } else { /* unresolved weak kfunc call */ |
6059 | poison_kfunc_call(prog, i, relo->insn_idx, insn, | |
6060 | relo->ext_idx, ext); | |
466b2e13 | 6061 | } |
5bd022ec | 6062 | break; |
53eddb5e | 6063 | case RELO_SUBPROG_ADDR: |
b1268826 AS |
6064 | if (insn[0].src_reg != BPF_PSEUDO_FUNC) { |
6065 | pr_warn("prog '%s': relo #%d: bad insn\n", | |
6066 | prog->name, i); | |
6067 | return -EINVAL; | |
6068 | } | |
6069 | /* handled already */ | |
53eddb5e | 6070 | break; |
166750bc | 6071 | case RELO_CALL: |
b1268826 | 6072 | /* handled already */ |
166750bc | 6073 | break; |
d0e92887 AS |
6074 | case RELO_CORE: |
6075 | /* will be handled by bpf_program_record_relos() */ | |
6076 | break; | |
166750bc | 6077 | default: |
9c0f8cbd AN |
6078 | pr_warn("prog '%s': relo #%d: bad relo type %d\n", |
6079 | prog->name, i, relo->type); | |
166750bc | 6080 | return -EINVAL; |
8a47a6c5 | 6081 | } |
8a47a6c5 WN |
6082 | } |
6083 | ||
c3c55696 AN |
6084 | return 0; |
6085 | } | |
6086 | ||
8505e870 AN |
6087 | static int adjust_prog_btf_ext_info(const struct bpf_object *obj, |
6088 | const struct bpf_program *prog, | |
6089 | const struct btf_ext_info *ext_info, | |
6090 | void **prog_info, __u32 *prog_rec_cnt, | |
6091 | __u32 *prog_rec_sz) | |
6092 | { | |
6093 | void *copy_start = NULL, *copy_end = NULL; | |
6094 | void *rec, *rec_end, *new_prog_info; | |
6095 | const struct btf_ext_info_sec *sec; | |
6096 | size_t old_sz, new_sz; | |
11d5daa8 | 6097 | int i, sec_num, sec_idx, off_adj; |
8505e870 | 6098 | |
11d5daa8 | 6099 | sec_num = 0; |
8505e870 | 6100 | for_each_btf_ext_sec(ext_info, sec) { |
11d5daa8 AN |
6101 | sec_idx = ext_info->sec_idxs[sec_num]; |
6102 | sec_num++; | |
6103 | if (prog->sec_idx != sec_idx) | |
8505e870 AN |
6104 | continue; |
6105 | ||
6106 | for_each_btf_ext_rec(ext_info, sec, i, rec) { | |
6107 | __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ; | |
6108 | ||
6109 | if (insn_off < prog->sec_insn_off) | |
6110 | continue; | |
6111 | if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt) | |
6112 | break; | |
6113 | ||
6114 | if (!copy_start) | |
6115 | copy_start = rec; | |
6116 | copy_end = rec + ext_info->rec_size; | |
6117 | } | |
6118 | ||
6119 | if (!copy_start) | |
6120 | return -ENOENT; | |
6121 | ||
6122 | /* append func/line info of a given (sub-)program to the main | |
6123 | * program func/line info | |
6124 | */ | |
8eb62958 | 6125 | old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size; |
8505e870 AN |
6126 | new_sz = old_sz + (copy_end - copy_start); |
6127 | new_prog_info = realloc(*prog_info, new_sz); | |
6128 | if (!new_prog_info) | |
6129 | return -ENOMEM; | |
6130 | *prog_info = new_prog_info; | |
6131 | *prog_rec_cnt = new_sz / ext_info->rec_size; | |
6132 | memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start); | |
6133 | ||
6134 | /* Kernel instruction offsets are in units of 8-byte | |
6135 | * instructions, while .BTF.ext instruction offsets generated | |
6136 | * by Clang are in units of bytes. So convert Clang offsets | |
6137 | * into kernel offsets and adjust offset according to program | |
6138 | * relocated position. | |
6139 | */ | |
6140 | off_adj = prog->sub_insn_off - prog->sec_insn_off; | |
6141 | rec = new_prog_info + old_sz; | |
6142 | rec_end = new_prog_info + new_sz; | |
6143 | for (; rec < rec_end; rec += ext_info->rec_size) { | |
6144 | __u32 *insn_off = rec; | |
6145 | ||
6146 | *insn_off = *insn_off / BPF_INSN_SZ + off_adj; | |
6147 | } | |
6148 | *prog_rec_sz = ext_info->rec_size; | |
6149 | return 0; | |
6150 | } | |
6151 | ||
6152 | return -ENOENT; | |
6153 | } | |
6154 | ||
6155 | static int | |
6156 | reloc_prog_func_and_line_info(const struct bpf_object *obj, | |
6157 | struct bpf_program *main_prog, | |
6158 | const struct bpf_program *prog) | |
6159 | { | |
6160 | int err; | |
6161 | ||
6162 | /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't | |
2f38fe68 | 6163 | * support func/line info |
8505e870 | 6164 | */ |
9ca1f56a | 6165 | if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC)) |
8505e870 AN |
6166 | return 0; |
6167 | ||
6168 | /* only attempt func info relocation if main program's func_info | |
6169 | * relocation was successful | |
6170 | */ | |
6171 | if (main_prog != prog && !main_prog->func_info) | |
6172 | goto line_info; | |
6173 | ||
6174 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info, | |
6175 | &main_prog->func_info, | |
6176 | &main_prog->func_info_cnt, | |
6177 | &main_prog->func_info_rec_size); | |
6178 | if (err) { | |
6179 | if (err != -ENOENT) { | |
6180 | pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n", | |
6181 | prog->name, err); | |
6182 | return err; | |
6183 | } | |
6184 | if (main_prog->func_info) { | |
6185 | /* | |
6186 | * Some info has already been found but has problem | |
6187 | * in the last btf_ext reloc. Must have to error out. | |
6188 | */ | |
6189 | pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name); | |
6190 | return err; | |
6191 | } | |
6192 | /* Have problem loading the very first info. Ignore the rest. */ | |
6193 | pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n", | |
6194 | prog->name); | |
6195 | } | |
6196 | ||
6197 | line_info: | |
6198 | /* don't relocate line info if main program's relocation failed */ | |
6199 | if (main_prog != prog && !main_prog->line_info) | |
6200 | return 0; | |
6201 | ||
6202 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info, | |
6203 | &main_prog->line_info, | |
6204 | &main_prog->line_info_cnt, | |
6205 | &main_prog->line_info_rec_size); | |
6206 | if (err) { | |
6207 | if (err != -ENOENT) { | |
6208 | pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n", | |
6209 | prog->name, err); | |
6210 | return err; | |
6211 | } | |
6212 | if (main_prog->line_info) { | |
6213 | /* | |
6214 | * Some info has already been found but has problem | |
6215 | * in the last btf_ext reloc. Must have to error out. | |
6216 | */ | |
6217 | pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name); | |
6218 | return err; | |
6219 | } | |
6220 | /* Have problem loading the very first info. Ignore the rest. */ | |
6221 | pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n", | |
6222 | prog->name); | |
6223 | } | |
6224 | return 0; | |
6225 | } | |
6226 | ||
c3c55696 AN |
6227 | static int cmp_relo_by_insn_idx(const void *key, const void *elem) |
6228 | { | |
6229 | size_t insn_idx = *(const size_t *)key; | |
6230 | const struct reloc_desc *relo = elem; | |
6231 | ||
6232 | if (insn_idx == relo->insn_idx) | |
6233 | return 0; | |
6234 | return insn_idx < relo->insn_idx ? -1 : 1; | |
6235 | } | |
6236 | ||
6237 | static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx) | |
6238 | { | |
2a6a9bf2 AN |
6239 | if (!prog->nr_reloc) |
6240 | return NULL; | |
c3c55696 AN |
6241 | return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc, |
6242 | sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx); | |
6243 | } | |
6244 | ||
b1268826 AS |
6245 | static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog) |
6246 | { | |
6247 | int new_cnt = main_prog->nr_reloc + subprog->nr_reloc; | |
6248 | struct reloc_desc *relos; | |
6249 | int i; | |
6250 | ||
6251 | if (main_prog == subprog) | |
6252 | return 0; | |
6253 | relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos)); | |
8a0260db AN |
6254 | /* if new count is zero, reallocarray can return a valid NULL result; |
6255 | * in this case the previous pointer will be freed, so we *have to* | |
6256 | * reassign old pointer to the new value (even if it's NULL) | |
6257 | */ | |
6258 | if (!relos && new_cnt) | |
b1268826 | 6259 | return -ENOMEM; |
2a6a9bf2 AN |
6260 | if (subprog->nr_reloc) |
6261 | memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc, | |
6262 | sizeof(*relos) * subprog->nr_reloc); | |
b1268826 AS |
6263 | |
6264 | for (i = main_prog->nr_reloc; i < new_cnt; i++) | |
6265 | relos[i].insn_idx += subprog->sub_insn_off; | |
6266 | /* After insn_idx adjustment the 'relos' array is still sorted | |
6267 | * by insn_idx and doesn't break bsearch. | |
6268 | */ | |
6269 | main_prog->reloc_desc = relos; | |
6270 | main_prog->nr_reloc = new_cnt; | |
6271 | return 0; | |
6272 | } | |
6273 | ||
6c918709 KKD |
6274 | static int |
6275 | bpf_object__append_subprog_code(struct bpf_object *obj, struct bpf_program *main_prog, | |
6276 | struct bpf_program *subprog) | |
6277 | { | |
6278 | struct bpf_insn *insns; | |
6279 | size_t new_cnt; | |
6280 | int err; | |
6281 | ||
6282 | subprog->sub_insn_off = main_prog->insns_cnt; | |
6283 | ||
6284 | new_cnt = main_prog->insns_cnt + subprog->insns_cnt; | |
6285 | insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns)); | |
6286 | if (!insns) { | |
6287 | pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name); | |
6288 | return -ENOMEM; | |
6289 | } | |
6290 | main_prog->insns = insns; | |
6291 | main_prog->insns_cnt = new_cnt; | |
6292 | ||
6293 | memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns, | |
6294 | subprog->insns_cnt * sizeof(*insns)); | |
6295 | ||
6296 | pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n", | |
6297 | main_prog->name, subprog->insns_cnt, subprog->name); | |
6298 | ||
6299 | /* The subprog insns are now appended. Append its relos too. */ | |
6300 | err = append_subprog_relos(main_prog, subprog); | |
6301 | if (err) | |
6302 | return err; | |
6303 | return 0; | |
6304 | } | |
6305 | ||
c3c55696 AN |
6306 | static int |
6307 | bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog, | |
6308 | struct bpf_program *prog) | |
6309 | { | |
7e2925f6 | 6310 | size_t sub_insn_idx, insn_idx; |
c3c55696 | 6311 | struct bpf_program *subprog; |
c3c55696 | 6312 | struct reloc_desc *relo; |
7e2925f6 | 6313 | struct bpf_insn *insn; |
c3c55696 AN |
6314 | int err; |
6315 | ||
6316 | err = reloc_prog_func_and_line_info(obj, main_prog, prog); | |
6317 | if (err) | |
6318 | return err; | |
6319 | ||
6320 | for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) { | |
6321 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
53eddb5e | 6322 | if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn)) |
c3c55696 AN |
6323 | continue; |
6324 | ||
6325 | relo = find_prog_insn_relo(prog, insn_idx); | |
a18f7214 | 6326 | if (relo && relo->type == RELO_EXTERN_CALL) |
b1268826 AS |
6327 | /* kfunc relocations will be handled later |
6328 | * in bpf_object__relocate_data() | |
6329 | */ | |
6330 | continue; | |
53eddb5e | 6331 | if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) { |
c3c55696 AN |
6332 | pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n", |
6333 | prog->name, insn_idx, relo->type); | |
6334 | return -LIBBPF_ERRNO__RELOC; | |
6335 | } | |
6336 | if (relo) { | |
6337 | /* sub-program instruction index is a combination of | |
6338 | * an offset of a symbol pointed to by relocation and | |
6339 | * call instruction's imm field; for global functions, | |
6340 | * call always has imm = -1, but for static functions | |
6341 | * relocation is against STT_SECTION and insn->imm | |
6342 | * points to a start of a static function | |
53eddb5e YS |
6343 | * |
6344 | * for subprog addr relocation, the relo->sym_off + insn->imm is | |
6345 | * the byte offset in the corresponding section. | |
c3c55696 | 6346 | */ |
53eddb5e YS |
6347 | if (relo->type == RELO_CALL) |
6348 | sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1; | |
6349 | else | |
6350 | sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ; | |
6351 | } else if (insn_is_pseudo_func(insn)) { | |
6352 | /* | |
6353 | * RELO_SUBPROG_ADDR relo is always emitted even if both | |
6354 | * functions are in the same section, so it shouldn't reach here. | |
6355 | */ | |
6356 | pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n", | |
6357 | prog->name, insn_idx); | |
6358 | return -LIBBPF_ERRNO__RELOC; | |
c3c55696 AN |
6359 | } else { |
6360 | /* if subprogram call is to a static function within | |
6361 | * the same ELF section, there won't be any relocation | |
6362 | * emitted, but it also means there is no additional | |
6363 | * offset necessary, insns->imm is relative to | |
6364 | * instruction's original position within the section | |
6365 | */ | |
6366 | sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1; | |
6367 | } | |
6368 | ||
6369 | /* we enforce that sub-programs should be in .text section */ | |
6370 | subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx); | |
6371 | if (!subprog) { | |
6372 | pr_warn("prog '%s': no .text section found yet sub-program call exists\n", | |
6373 | prog->name); | |
6374 | return -LIBBPF_ERRNO__RELOC; | |
6375 | } | |
6376 | ||
6377 | /* if it's the first call instruction calling into this | |
6378 | * subprogram (meaning this subprog hasn't been processed | |
6379 | * yet) within the context of current main program: | |
6380 | * - append it at the end of main program's instructions blog; | |
6381 | * - process is recursively, while current program is put on hold; | |
6382 | * - if that subprogram calls some other not yet processes | |
6383 | * subprogram, same thing will happen recursively until | |
6384 | * there are no more unprocesses subprograms left to append | |
6385 | * and relocate. | |
6386 | */ | |
6387 | if (subprog->sub_insn_off == 0) { | |
6c918709 | 6388 | err = bpf_object__append_subprog_code(obj, main_prog, subprog); |
b1268826 AS |
6389 | if (err) |
6390 | return err; | |
c3c55696 AN |
6391 | err = bpf_object__reloc_code(obj, main_prog, subprog); |
6392 | if (err) | |
6393 | return err; | |
6394 | } | |
6395 | ||
6396 | /* main_prog->insns memory could have been re-allocated, so | |
6397 | * calculate pointer again | |
6398 | */ | |
6399 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
6400 | /* calculate correct instruction position within current main | |
6401 | * prog; each main prog can have a different set of | |
6402 | * subprograms appended (potentially in different order as | |
6403 | * well), so position of any subprog can be different for | |
e3ba8e4e KM |
6404 | * different main programs |
6405 | */ | |
c3c55696 AN |
6406 | insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1; |
6407 | ||
c3c55696 AN |
6408 | pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n", |
6409 | prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off); | |
6410 | } | |
6411 | ||
6412 | return 0; | |
6413 | } | |
6414 | ||
6415 | /* | |
6416 | * Relocate sub-program calls. | |
6417 | * | |
6418 | * Algorithm operates as follows. Each entry-point BPF program (referred to as | |
6419 | * main prog) is processed separately. For each subprog (non-entry functions, | |
6420 | * that can be called from either entry progs or other subprogs) gets their | |
6421 | * sub_insn_off reset to zero. This serves as indicator that this subprogram | |
6422 | * hasn't been yet appended and relocated within current main prog. Once its | |
6423 | * relocated, sub_insn_off will point at the position within current main prog | |
6424 | * where given subprog was appended. This will further be used to relocate all | |
6425 | * the call instructions jumping into this subprog. | |
6426 | * | |
6427 | * We start with main program and process all call instructions. If the call | |
6428 | * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off | |
6429 | * is zero), subprog instructions are appended at the end of main program's | |
6430 | * instruction array. Then main program is "put on hold" while we recursively | |
6431 | * process newly appended subprogram. If that subprogram calls into another | |
6432 | * subprogram that hasn't been appended, new subprogram is appended again to | |
6433 | * the *main* prog's instructions (subprog's instructions are always left | |
6434 | * untouched, as they need to be in unmodified state for subsequent main progs | |
6435 | * and subprog instructions are always sent only as part of a main prog) and | |
6436 | * the process continues recursively. Once all the subprogs called from a main | |
6437 | * prog or any of its subprogs are appended (and relocated), all their | |
6438 | * positions within finalized instructions array are known, so it's easy to | |
6439 | * rewrite call instructions with correct relative offsets, corresponding to | |
6440 | * desired target subprog. | |
6441 | * | |
6442 | * Its important to realize that some subprogs might not be called from some | |
6443 | * main prog and any of its called/used subprogs. Those will keep their | |
6444 | * subprog->sub_insn_off as zero at all times and won't be appended to current | |
6445 | * main prog and won't be relocated within the context of current main prog. | |
6446 | * They might still be used from other main progs later. | |
6447 | * | |
6448 | * Visually this process can be shown as below. Suppose we have two main | |
6449 | * programs mainA and mainB and BPF object contains three subprogs: subA, | |
6450 | * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and | |
6451 | * subC both call subB: | |
6452 | * | |
6453 | * +--------+ +-------+ | |
6454 | * | v v | | |
6455 | * +--+---+ +--+-+-+ +---+--+ | |
6456 | * | subA | | subB | | subC | | |
6457 | * +--+---+ +------+ +---+--+ | |
6458 | * ^ ^ | |
6459 | * | | | |
6460 | * +---+-------+ +------+----+ | |
6461 | * | mainA | | mainB | | |
6462 | * +-----------+ +-----------+ | |
6463 | * | |
6464 | * We'll start relocating mainA, will find subA, append it and start | |
6465 | * processing sub A recursively: | |
6466 | * | |
6467 | * +-----------+------+ | |
6468 | * | mainA | subA | | |
6469 | * +-----------+------+ | |
6470 | * | |
6471 | * At this point we notice that subB is used from subA, so we append it and | |
6472 | * relocate (there are no further subcalls from subB): | |
6473 | * | |
6474 | * +-----------+------+------+ | |
6475 | * | mainA | subA | subB | | |
6476 | * +-----------+------+------+ | |
6477 | * | |
6478 | * At this point, we relocate subA calls, then go one level up and finish with | |
6479 | * relocatin mainA calls. mainA is done. | |
6480 | * | |
6481 | * For mainB process is similar but results in different order. We start with | |
6482 | * mainB and skip subA and subB, as mainB never calls them (at least | |
6483 | * directly), but we see subC is needed, so we append and start processing it: | |
6484 | * | |
6485 | * +-----------+------+ | |
6486 | * | mainB | subC | | |
6487 | * +-----------+------+ | |
6488 | * Now we see subC needs subB, so we go back to it, append and relocate it: | |
6489 | * | |
6490 | * +-----------+------+------+ | |
6491 | * | mainB | subC | subB | | |
6492 | * +-----------+------+------+ | |
6493 | * | |
6494 | * At this point we unwind recursion, relocate calls in subC, then in mainB. | |
6495 | */ | |
6496 | static int | |
6497 | bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog) | |
6498 | { | |
6499 | struct bpf_program *subprog; | |
d3d93e34 | 6500 | int i, err; |
c3c55696 | 6501 | |
c3c55696 AN |
6502 | /* mark all subprogs as not relocated (yet) within the context of |
6503 | * current main program | |
6504 | */ | |
6505 | for (i = 0; i < obj->nr_programs; i++) { | |
6506 | subprog = &obj->programs[i]; | |
6507 | if (!prog_is_subprog(obj, subprog)) | |
6508 | continue; | |
6509 | ||
6510 | subprog->sub_insn_off = 0; | |
c3c55696 AN |
6511 | } |
6512 | ||
6513 | err = bpf_object__reloc_code(obj, prog, prog); | |
6514 | if (err) | |
6515 | return err; | |
6516 | ||
8a47a6c5 WN |
6517 | return 0; |
6518 | } | |
6519 | ||
67234743 AS |
6520 | static void |
6521 | bpf_object__free_relocs(struct bpf_object *obj) | |
6522 | { | |
6523 | struct bpf_program *prog; | |
6524 | int i; | |
6525 | ||
6526 | /* free up relocation descriptors */ | |
6527 | for (i = 0; i < obj->nr_programs; i++) { | |
6528 | prog = &obj->programs[i]; | |
6529 | zfree(&prog->reloc_desc); | |
6530 | prog->nr_reloc = 0; | |
6531 | } | |
6532 | } | |
6533 | ||
d0e92887 AS |
6534 | static int cmp_relocs(const void *_a, const void *_b) |
6535 | { | |
6536 | const struct reloc_desc *a = _a; | |
6537 | const struct reloc_desc *b = _b; | |
6538 | ||
6539 | if (a->insn_idx != b->insn_idx) | |
6540 | return a->insn_idx < b->insn_idx ? -1 : 1; | |
6541 | ||
6542 | /* no two relocations should have the same insn_idx, but ... */ | |
6543 | if (a->type != b->type) | |
6544 | return a->type < b->type ? -1 : 1; | |
6545 | ||
6546 | return 0; | |
6547 | } | |
6548 | ||
6549 | static void bpf_object__sort_relos(struct bpf_object *obj) | |
6550 | { | |
6551 | int i; | |
6552 | ||
6553 | for (i = 0; i < obj->nr_programs; i++) { | |
6554 | struct bpf_program *p = &obj->programs[i]; | |
6555 | ||
6556 | if (!p->nr_reloc) | |
6557 | continue; | |
6558 | ||
6559 | qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs); | |
6560 | } | |
6561 | } | |
6562 | ||
fb03be7c AN |
6563 | static int bpf_prog_assign_exc_cb(struct bpf_object *obj, struct bpf_program *prog) |
6564 | { | |
6565 | const char *str = "exception_callback:"; | |
6566 | size_t pfx_len = strlen(str); | |
6567 | int i, j, n; | |
6568 | ||
6569 | if (!obj->btf || !kernel_supports(obj, FEAT_BTF_DECL_TAG)) | |
6570 | return 0; | |
6571 | ||
6572 | n = btf__type_cnt(obj->btf); | |
6573 | for (i = 1; i < n; i++) { | |
6574 | const char *name; | |
6575 | struct btf_type *t; | |
6576 | ||
6577 | t = btf_type_by_id(obj->btf, i); | |
6578 | if (!btf_is_decl_tag(t) || btf_decl_tag(t)->component_idx != -1) | |
6579 | continue; | |
6580 | ||
6581 | name = btf__str_by_offset(obj->btf, t->name_off); | |
6582 | if (strncmp(name, str, pfx_len) != 0) | |
6583 | continue; | |
6584 | ||
6585 | t = btf_type_by_id(obj->btf, t->type); | |
6586 | if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) { | |
6587 | pr_warn("prog '%s': exception_callback:<value> decl tag not applied to the main program\n", | |
6588 | prog->name); | |
6589 | return -EINVAL; | |
6590 | } | |
6591 | if (strcmp(prog->name, btf__str_by_offset(obj->btf, t->name_off)) != 0) | |
6592 | continue; | |
6593 | /* Multiple callbacks are specified for the same prog, | |
6594 | * the verifier will eventually return an error for this | |
6595 | * case, hence simply skip appending a subprog. | |
6596 | */ | |
6597 | if (prog->exception_cb_idx >= 0) { | |
6598 | prog->exception_cb_idx = -1; | |
6599 | break; | |
6600 | } | |
6601 | ||
6602 | name += pfx_len; | |
6603 | if (str_is_empty(name)) { | |
6604 | pr_warn("prog '%s': exception_callback:<value> decl tag contains empty value\n", | |
6605 | prog->name); | |
6606 | return -EINVAL; | |
6607 | } | |
6608 | ||
6609 | for (j = 0; j < obj->nr_programs; j++) { | |
6610 | struct bpf_program *subprog = &obj->programs[j]; | |
6611 | ||
6612 | if (!prog_is_subprog(obj, subprog)) | |
6613 | continue; | |
6614 | if (strcmp(name, subprog->name) != 0) | |
6615 | continue; | |
6616 | /* Enforce non-hidden, as from verifier point of | |
6617 | * view it expects global functions, whereas the | |
6618 | * mark_btf_static fixes up linkage as static. | |
6619 | */ | |
6620 | if (!subprog->sym_global || subprog->mark_btf_static) { | |
6621 | pr_warn("prog '%s': exception callback %s must be a global non-hidden function\n", | |
6622 | prog->name, subprog->name); | |
6623 | return -EINVAL; | |
6624 | } | |
6625 | /* Let's see if we already saw a static exception callback with the same name */ | |
6626 | if (prog->exception_cb_idx >= 0) { | |
6627 | pr_warn("prog '%s': multiple subprogs with same name as exception callback '%s'\n", | |
6628 | prog->name, subprog->name); | |
6629 | return -EINVAL; | |
6630 | } | |
6631 | prog->exception_cb_idx = j; | |
6632 | break; | |
6633 | } | |
6634 | ||
6635 | if (prog->exception_cb_idx >= 0) | |
6636 | continue; | |
6637 | ||
6638 | pr_warn("prog '%s': cannot find exception callback '%s'\n", prog->name, name); | |
6639 | return -ENOENT; | |
6640 | } | |
6641 | ||
6642 | return 0; | |
6643 | } | |
6644 | ||
2f38fe68 AN |
6645 | static struct { |
6646 | enum bpf_prog_type prog_type; | |
6647 | const char *ctx_name; | |
6648 | } global_ctx_map[] = { | |
6649 | { BPF_PROG_TYPE_CGROUP_DEVICE, "bpf_cgroup_dev_ctx" }, | |
6650 | { BPF_PROG_TYPE_CGROUP_SKB, "__sk_buff" }, | |
6651 | { BPF_PROG_TYPE_CGROUP_SOCK, "bpf_sock" }, | |
6652 | { BPF_PROG_TYPE_CGROUP_SOCK_ADDR, "bpf_sock_addr" }, | |
6653 | { BPF_PROG_TYPE_CGROUP_SOCKOPT, "bpf_sockopt" }, | |
6654 | { BPF_PROG_TYPE_CGROUP_SYSCTL, "bpf_sysctl" }, | |
6655 | { BPF_PROG_TYPE_FLOW_DISSECTOR, "__sk_buff" }, | |
6656 | { BPF_PROG_TYPE_KPROBE, "bpf_user_pt_regs_t" }, | |
6657 | { BPF_PROG_TYPE_LWT_IN, "__sk_buff" }, | |
6658 | { BPF_PROG_TYPE_LWT_OUT, "__sk_buff" }, | |
6659 | { BPF_PROG_TYPE_LWT_SEG6LOCAL, "__sk_buff" }, | |
6660 | { BPF_PROG_TYPE_LWT_XMIT, "__sk_buff" }, | |
6661 | { BPF_PROG_TYPE_NETFILTER, "bpf_nf_ctx" }, | |
6662 | { BPF_PROG_TYPE_PERF_EVENT, "bpf_perf_event_data" }, | |
6663 | { BPF_PROG_TYPE_RAW_TRACEPOINT, "bpf_raw_tracepoint_args" }, | |
6664 | { BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE, "bpf_raw_tracepoint_args" }, | |
6665 | { BPF_PROG_TYPE_SCHED_ACT, "__sk_buff" }, | |
6666 | { BPF_PROG_TYPE_SCHED_CLS, "__sk_buff" }, | |
6667 | { BPF_PROG_TYPE_SK_LOOKUP, "bpf_sk_lookup" }, | |
6668 | { BPF_PROG_TYPE_SK_MSG, "sk_msg_md" }, | |
6669 | { BPF_PROG_TYPE_SK_REUSEPORT, "sk_reuseport_md" }, | |
6670 | { BPF_PROG_TYPE_SK_SKB, "__sk_buff" }, | |
6671 | { BPF_PROG_TYPE_SOCK_OPS, "bpf_sock_ops" }, | |
6672 | { BPF_PROG_TYPE_SOCKET_FILTER, "__sk_buff" }, | |
6673 | { BPF_PROG_TYPE_XDP, "xdp_md" }, | |
6674 | /* all other program types don't have "named" context structs */ | |
6675 | }; | |
6676 | ||
9eea8faf AN |
6677 | /* forward declarations for arch-specific underlying types of bpf_user_pt_regs_t typedef, |
6678 | * for below __builtin_types_compatible_p() checks; | |
6679 | * with this approach we don't need any extra arch-specific #ifdef guards | |
6680 | */ | |
6681 | struct pt_regs; | |
6682 | struct user_pt_regs; | |
6683 | struct user_regs_struct; | |
6684 | ||
76ec90a9 AN |
6685 | static bool need_func_arg_type_fixup(const struct btf *btf, const struct bpf_program *prog, |
6686 | const char *subprog_name, int arg_idx, | |
6687 | int arg_type_id, const char *ctx_name) | |
6688 | { | |
6689 | const struct btf_type *t; | |
6690 | const char *tname; | |
6691 | ||
6692 | /* check if existing parameter already matches verifier expectations */ | |
6693 | t = skip_mods_and_typedefs(btf, arg_type_id, NULL); | |
6694 | if (!btf_is_ptr(t)) | |
6695 | goto out_warn; | |
6696 | ||
6697 | /* typedef bpf_user_pt_regs_t is a special PITA case, valid for kprobe | |
6698 | * and perf_event programs, so check this case early on and forget | |
6699 | * about it for subsequent checks | |
6700 | */ | |
6701 | while (btf_is_mod(t)) | |
6702 | t = btf__type_by_id(btf, t->type); | |
6703 | if (btf_is_typedef(t) && | |
6704 | (prog->type == BPF_PROG_TYPE_KPROBE || prog->type == BPF_PROG_TYPE_PERF_EVENT)) { | |
6705 | tname = btf__str_by_offset(btf, t->name_off) ?: "<anon>"; | |
6706 | if (strcmp(tname, "bpf_user_pt_regs_t") == 0) | |
6707 | return false; /* canonical type for kprobe/perf_event */ | |
6708 | } | |
6709 | ||
6710 | /* now we can ignore typedefs moving forward */ | |
6711 | t = skip_mods_and_typedefs(btf, t->type, NULL); | |
6712 | ||
6713 | /* if it's `void *`, definitely fix up BTF info */ | |
6714 | if (btf_is_void(t)) | |
6715 | return true; | |
6716 | ||
6717 | /* if it's already proper canonical type, no need to fix up */ | |
6718 | tname = btf__str_by_offset(btf, t->name_off) ?: "<anon>"; | |
6719 | if (btf_is_struct(t) && strcmp(tname, ctx_name) == 0) | |
6720 | return false; | |
6721 | ||
6722 | /* special cases */ | |
6723 | switch (prog->type) { | |
6724 | case BPF_PROG_TYPE_KPROBE: | |
76ec90a9 AN |
6725 | /* `struct pt_regs *` is expected, but we need to fix up */ |
6726 | if (btf_is_struct(t) && strcmp(tname, "pt_regs") == 0) | |
6727 | return true; | |
6728 | break; | |
9eea8faf AN |
6729 | case BPF_PROG_TYPE_PERF_EVENT: |
6730 | if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct pt_regs) && | |
6731 | btf_is_struct(t) && strcmp(tname, "pt_regs") == 0) | |
d7bc416a | 6732 | return true; |
9eea8faf AN |
6733 | if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct user_pt_regs) && |
6734 | btf_is_struct(t) && strcmp(tname, "user_pt_regs") == 0) | |
d7bc416a | 6735 | return true; |
9eea8faf AN |
6736 | if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct user_regs_struct) && |
6737 | btf_is_struct(t) && strcmp(tname, "user_regs_struct") == 0) | |
d7bc416a | 6738 | return true; |
9eea8faf | 6739 | break; |
76ec90a9 AN |
6740 | case BPF_PROG_TYPE_RAW_TRACEPOINT: |
6741 | case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE: | |
6742 | /* allow u64* as ctx */ | |
6743 | if (btf_is_int(t) && t->size == 8) | |
6744 | return true; | |
6745 | break; | |
6746 | default: | |
6747 | break; | |
6748 | } | |
6749 | ||
6750 | out_warn: | |
6751 | pr_warn("prog '%s': subprog '%s' arg#%d is expected to be of `struct %s *` type\n", | |
6752 | prog->name, subprog_name, arg_idx, ctx_name); | |
6753 | return false; | |
6754 | } | |
6755 | ||
2f38fe68 AN |
6756 | static int clone_func_btf_info(struct btf *btf, int orig_fn_id, struct bpf_program *prog) |
6757 | { | |
6758 | int fn_id, fn_proto_id, ret_type_id, orig_proto_id; | |
6759 | int i, err, arg_cnt, fn_name_off, linkage; | |
6760 | struct btf_type *fn_t, *fn_proto_t, *t; | |
6761 | struct btf_param *p; | |
6762 | ||
6763 | /* caller already validated FUNC -> FUNC_PROTO validity */ | |
6764 | fn_t = btf_type_by_id(btf, orig_fn_id); | |
6765 | fn_proto_t = btf_type_by_id(btf, fn_t->type); | |
6766 | ||
6767 | /* Note that each btf__add_xxx() operation invalidates | |
6768 | * all btf_type and string pointers, so we need to be | |
6769 | * very careful when cloning BTF types. BTF type | |
6770 | * pointers have to be always refetched. And to avoid | |
6771 | * problems with invalidated string pointers, we | |
6772 | * add empty strings initially, then just fix up | |
6773 | * name_off offsets in place. Offsets are stable for | |
6774 | * existing strings, so that works out. | |
6775 | */ | |
6776 | fn_name_off = fn_t->name_off; /* we are about to invalidate fn_t */ | |
6777 | linkage = btf_func_linkage(fn_t); | |
6778 | orig_proto_id = fn_t->type; /* original FUNC_PROTO ID */ | |
6779 | ret_type_id = fn_proto_t->type; /* fn_proto_t will be invalidated */ | |
6780 | arg_cnt = btf_vlen(fn_proto_t); | |
6781 | ||
6782 | /* clone FUNC_PROTO and its params */ | |
6783 | fn_proto_id = btf__add_func_proto(btf, ret_type_id); | |
6784 | if (fn_proto_id < 0) | |
6785 | return -EINVAL; | |
6786 | ||
6787 | for (i = 0; i < arg_cnt; i++) { | |
6788 | int name_off; | |
6789 | ||
6790 | /* copy original parameter data */ | |
6791 | t = btf_type_by_id(btf, orig_proto_id); | |
6792 | p = &btf_params(t)[i]; | |
6793 | name_off = p->name_off; | |
6794 | ||
6795 | err = btf__add_func_param(btf, "", p->type); | |
6796 | if (err) | |
6797 | return err; | |
6798 | ||
6799 | fn_proto_t = btf_type_by_id(btf, fn_proto_id); | |
6800 | p = &btf_params(fn_proto_t)[i]; | |
6801 | p->name_off = name_off; /* use remembered str offset */ | |
6802 | } | |
6803 | ||
6804 | /* clone FUNC now, btf__add_func() enforces non-empty name, so use | |
6805 | * entry program's name as a placeholder, which we replace immediately | |
6806 | * with original name_off | |
6807 | */ | |
6808 | fn_id = btf__add_func(btf, prog->name, linkage, fn_proto_id); | |
6809 | if (fn_id < 0) | |
6810 | return -EINVAL; | |
6811 | ||
6812 | fn_t = btf_type_by_id(btf, fn_id); | |
6813 | fn_t->name_off = fn_name_off; /* reuse original string */ | |
6814 | ||
6815 | return fn_id; | |
6816 | } | |
6817 | ||
6818 | /* Check if main program or global subprog's function prototype has `arg:ctx` | |
6819 | * argument tags, and, if necessary, substitute correct type to match what BPF | |
6820 | * verifier would expect, taking into account specific program type. This | |
6821 | * allows to support __arg_ctx tag transparently on old kernels that don't yet | |
6822 | * have a native support for it in the verifier, making user's life much | |
6823 | * easier. | |
6824 | */ | |
6825 | static int bpf_program_fixup_func_info(struct bpf_object *obj, struct bpf_program *prog) | |
6826 | { | |
76ec90a9 | 6827 | const char *ctx_name = NULL, *ctx_tag = "arg:ctx", *fn_name; |
2f38fe68 AN |
6828 | struct bpf_func_info_min *func_rec; |
6829 | struct btf_type *fn_t, *fn_proto_t; | |
6830 | struct btf *btf = obj->btf; | |
6831 | const struct btf_type *t; | |
6832 | struct btf_param *p; | |
6833 | int ptr_id = 0, struct_id, tag_id, orig_fn_id; | |
6834 | int i, n, arg_idx, arg_cnt, err, rec_idx; | |
6835 | int *orig_ids; | |
6836 | ||
6837 | /* no .BTF.ext, no problem */ | |
6838 | if (!obj->btf_ext || !prog->func_info) | |
6839 | return 0; | |
6840 | ||
01b55f4f | 6841 | /* don't do any fix ups if kernel natively supports __arg_ctx */ |
0e6d0a9d | 6842 | if (kernel_supports(obj, FEAT_ARG_CTX_TAG)) |
01b55f4f AN |
6843 | return 0; |
6844 | ||
2f38fe68 AN |
6845 | /* some BPF program types just don't have named context structs, so |
6846 | * this fallback mechanism doesn't work for them | |
6847 | */ | |
6848 | for (i = 0; i < ARRAY_SIZE(global_ctx_map); i++) { | |
6849 | if (global_ctx_map[i].prog_type != prog->type) | |
6850 | continue; | |
6851 | ctx_name = global_ctx_map[i].ctx_name; | |
6852 | break; | |
6853 | } | |
6854 | if (!ctx_name) | |
6855 | return 0; | |
6856 | ||
6857 | /* remember original func BTF IDs to detect if we already cloned them */ | |
6858 | orig_ids = calloc(prog->func_info_cnt, sizeof(*orig_ids)); | |
6859 | if (!orig_ids) | |
6860 | return -ENOMEM; | |
6861 | for (i = 0; i < prog->func_info_cnt; i++) { | |
6862 | func_rec = prog->func_info + prog->func_info_rec_size * i; | |
6863 | orig_ids[i] = func_rec->type_id; | |
6864 | } | |
6865 | ||
6866 | /* go through each DECL_TAG with "arg:ctx" and see if it points to one | |
6867 | * of our subprogs; if yes and subprog is global and needs adjustment, | |
6868 | * clone and adjust FUNC -> FUNC_PROTO combo | |
6869 | */ | |
6870 | for (i = 1, n = btf__type_cnt(btf); i < n; i++) { | |
6871 | /* only DECL_TAG with "arg:ctx" value are interesting */ | |
6872 | t = btf__type_by_id(btf, i); | |
6873 | if (!btf_is_decl_tag(t)) | |
6874 | continue; | |
6875 | if (strcmp(btf__str_by_offset(btf, t->name_off), ctx_tag) != 0) | |
6876 | continue; | |
6877 | ||
6878 | /* only global funcs need adjustment, if at all */ | |
6879 | orig_fn_id = t->type; | |
6880 | fn_t = btf_type_by_id(btf, orig_fn_id); | |
6881 | if (!btf_is_func(fn_t) || btf_func_linkage(fn_t) != BTF_FUNC_GLOBAL) | |
6882 | continue; | |
6883 | ||
6884 | /* sanity check FUNC -> FUNC_PROTO chain, just in case */ | |
6885 | fn_proto_t = btf_type_by_id(btf, fn_t->type); | |
6886 | if (!fn_proto_t || !btf_is_func_proto(fn_proto_t)) | |
6887 | continue; | |
6888 | ||
6889 | /* find corresponding func_info record */ | |
6890 | func_rec = NULL; | |
6891 | for (rec_idx = 0; rec_idx < prog->func_info_cnt; rec_idx++) { | |
6892 | if (orig_ids[rec_idx] == t->type) { | |
6893 | func_rec = prog->func_info + prog->func_info_rec_size * rec_idx; | |
6894 | break; | |
6895 | } | |
6896 | } | |
6897 | /* current main program doesn't call into this subprog */ | |
6898 | if (!func_rec) | |
6899 | continue; | |
6900 | ||
6901 | /* some more sanity checking of DECL_TAG */ | |
6902 | arg_cnt = btf_vlen(fn_proto_t); | |
6903 | arg_idx = btf_decl_tag(t)->component_idx; | |
6904 | if (arg_idx < 0 || arg_idx >= arg_cnt) | |
6905 | continue; | |
6906 | ||
76ec90a9 | 6907 | /* check if we should fix up argument type */ |
2f38fe68 | 6908 | p = &btf_params(fn_proto_t)[arg_idx]; |
76ec90a9 AN |
6909 | fn_name = btf__str_by_offset(btf, fn_t->name_off) ?: "<anon>"; |
6910 | if (!need_func_arg_type_fixup(btf, prog, fn_name, arg_idx, p->type, ctx_name)) | |
6911 | continue; | |
2f38fe68 AN |
6912 | |
6913 | /* clone fn/fn_proto, unless we already did it for another arg */ | |
6914 | if (func_rec->type_id == orig_fn_id) { | |
6915 | int fn_id; | |
6916 | ||
6917 | fn_id = clone_func_btf_info(btf, orig_fn_id, prog); | |
6918 | if (fn_id < 0) { | |
6919 | err = fn_id; | |
6920 | goto err_out; | |
6921 | } | |
6922 | ||
6923 | /* point func_info record to a cloned FUNC type */ | |
6924 | func_rec->type_id = fn_id; | |
6925 | } | |
6926 | ||
6927 | /* create PTR -> STRUCT type chain to mark PTR_TO_CTX argument; | |
6928 | * we do it just once per main BPF program, as all global | |
6929 | * funcs share the same program type, so need only PTR -> | |
6930 | * STRUCT type chain | |
6931 | */ | |
6932 | if (ptr_id == 0) { | |
6933 | struct_id = btf__add_struct(btf, ctx_name, 0); | |
6934 | ptr_id = btf__add_ptr(btf, struct_id); | |
6935 | if (ptr_id < 0 || struct_id < 0) { | |
6936 | err = -EINVAL; | |
6937 | goto err_out; | |
6938 | } | |
6939 | } | |
6940 | ||
6941 | /* for completeness, clone DECL_TAG and point it to cloned param */ | |
6942 | tag_id = btf__add_decl_tag(btf, ctx_tag, func_rec->type_id, arg_idx); | |
6943 | if (tag_id < 0) { | |
6944 | err = -EINVAL; | |
6945 | goto err_out; | |
6946 | } | |
6947 | ||
6948 | /* all the BTF manipulations invalidated pointers, refetch them */ | |
6949 | fn_t = btf_type_by_id(btf, func_rec->type_id); | |
6950 | fn_proto_t = btf_type_by_id(btf, fn_t->type); | |
6951 | ||
6952 | /* fix up type ID pointed to by param */ | |
6953 | p = &btf_params(fn_proto_t)[arg_idx]; | |
6954 | p->type = ptr_id; | |
6955 | } | |
6956 | ||
6957 | free(orig_ids); | |
6958 | return 0; | |
6959 | err_out: | |
6960 | free(orig_ids); | |
6961 | return err; | |
6962 | } | |
6963 | ||
6964 | static int bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path) | |
8a47a6c5 WN |
6965 | { |
6966 | struct bpf_program *prog; | |
b1268826 | 6967 | size_t i, j; |
8a47a6c5 WN |
6968 | int err; |
6969 | ||
ddc7c304 AN |
6970 | if (obj->btf_ext) { |
6971 | err = bpf_object__relocate_core(obj, targ_btf_path); | |
6972 | if (err) { | |
be18010e KW |
6973 | pr_warn("failed to perform CO-RE relocations: %d\n", |
6974 | err); | |
ddc7c304 AN |
6975 | return err; |
6976 | } | |
185cfe83 | 6977 | bpf_object__sort_relos(obj); |
ddc7c304 | 6978 | } |
b1268826 AS |
6979 | |
6980 | /* Before relocating calls pre-process relocations and mark | |
6981 | * few ld_imm64 instructions that points to subprogs. | |
6982 | * Otherwise bpf_object__reloc_code() later would have to consider | |
6983 | * all ld_imm64 insns as relocation candidates. That would | |
6984 | * reduce relocation speed, since amount of find_prog_insn_relo() | |
6985 | * would increase and most of them will fail to find a relo. | |
9173cac3 AN |
6986 | */ |
6987 | for (i = 0; i < obj->nr_programs; i++) { | |
6988 | prog = &obj->programs[i]; | |
b1268826 AS |
6989 | for (j = 0; j < prog->nr_reloc; j++) { |
6990 | struct reloc_desc *relo = &prog->reloc_desc[j]; | |
6991 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; | |
6992 | ||
6993 | /* mark the insn, so it's recognized by insn_is_pseudo_func() */ | |
6994 | if (relo->type == RELO_SUBPROG_ADDR) | |
6995 | insn[0].src_reg = BPF_PSEUDO_FUNC; | |
9173cac3 | 6996 | } |
9173cac3 | 6997 | } |
b1268826 AS |
6998 | |
6999 | /* relocate subprogram calls and append used subprograms to main | |
c3c55696 AN |
7000 | * programs; each copy of subprogram code needs to be relocated |
7001 | * differently for each main program, because its code location might | |
b1268826 AS |
7002 | * have changed. |
7003 | * Append subprog relos to main programs to allow data relos to be | |
7004 | * processed after text is completely relocated. | |
9173cac3 | 7005 | */ |
8a47a6c5 WN |
7006 | for (i = 0; i < obj->nr_programs; i++) { |
7007 | prog = &obj->programs[i]; | |
c3c55696 AN |
7008 | /* sub-program's sub-calls are relocated within the context of |
7009 | * its main program only | |
7010 | */ | |
7011 | if (prog_is_subprog(obj, prog)) | |
9173cac3 | 7012 | continue; |
a3820c48 | 7013 | if (!prog->autoload) |
16e0c35c | 7014 | continue; |
8a47a6c5 | 7015 | |
c3c55696 | 7016 | err = bpf_object__relocate_calls(obj, prog); |
8a47a6c5 | 7017 | if (err) { |
9c0f8cbd AN |
7018 | pr_warn("prog '%s': failed to relocate calls: %d\n", |
7019 | prog->name, err); | |
8a47a6c5 WN |
7020 | return err; |
7021 | } | |
7e2925f6 | 7022 | |
fb03be7c AN |
7023 | err = bpf_prog_assign_exc_cb(obj, prog); |
7024 | if (err) | |
7025 | return err; | |
7e2925f6 KKD |
7026 | /* Now, also append exception callback if it has not been done already. */ |
7027 | if (prog->exception_cb_idx >= 0) { | |
7028 | struct bpf_program *subprog = &obj->programs[prog->exception_cb_idx]; | |
7029 | ||
7030 | /* Calling exception callback directly is disallowed, which the | |
7031 | * verifier will reject later. In case it was processed already, | |
7032 | * we can skip this step, otherwise for all other valid cases we | |
7033 | * have to append exception callback now. | |
7034 | */ | |
7035 | if (subprog->sub_insn_off == 0) { | |
7036 | err = bpf_object__append_subprog_code(obj, prog, subprog); | |
7037 | if (err) | |
7038 | return err; | |
7039 | err = bpf_object__reloc_code(obj, prog, subprog); | |
7040 | if (err) | |
7041 | return err; | |
7042 | } | |
7043 | } | |
8a47a6c5 | 7044 | } |
b1268826 AS |
7045 | for (i = 0; i < obj->nr_programs; i++) { |
7046 | prog = &obj->programs[i]; | |
7047 | if (prog_is_subprog(obj, prog)) | |
7048 | continue; | |
a3820c48 | 7049 | if (!prog->autoload) |
16e0c35c | 7050 | continue; |
2f38fe68 AN |
7051 | |
7052 | /* Process data relos for main programs */ | |
b1268826 AS |
7053 | err = bpf_object__relocate_data(obj, prog); |
7054 | if (err) { | |
7055 | pr_warn("prog '%s': failed to relocate data references: %d\n", | |
7056 | prog->name, err); | |
7057 | return err; | |
7058 | } | |
2f38fe68 AN |
7059 | |
7060 | /* Fix up .BTF.ext information, if necessary */ | |
7061 | err = bpf_program_fixup_func_info(obj, prog); | |
7062 | if (err) { | |
7063 | pr_warn("prog '%s': failed to perform .BTF.ext fix ups: %d\n", | |
7064 | prog->name, err); | |
7065 | return err; | |
7066 | } | |
b1268826 | 7067 | } |
185cfe83 | 7068 | |
8a47a6c5 WN |
7069 | return 0; |
7070 | } | |
7071 | ||
646f02ff | 7072 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 7073 | Elf64_Shdr *shdr, Elf_Data *data); |
646f02ff AN |
7074 | |
7075 | static int bpf_object__collect_map_relos(struct bpf_object *obj, | |
ad23b723 | 7076 | Elf64_Shdr *shdr, Elf_Data *data) |
646f02ff | 7077 | { |
15728ad3 AN |
7078 | const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *); |
7079 | int i, j, nrels, new_sz; | |
063e6881 | 7080 | const struct btf_var_secinfo *vi = NULL; |
646f02ff | 7081 | const struct btf_type *sec, *var, *def; |
341ac5ff HC |
7082 | struct bpf_map *map = NULL, *targ_map = NULL; |
7083 | struct bpf_program *targ_prog = NULL; | |
7084 | bool is_prog_array, is_map_in_map; | |
646f02ff | 7085 | const struct btf_member *member; |
341ac5ff | 7086 | const char *name, *mname, *type; |
646f02ff | 7087 | unsigned int moff; |
ad23b723 AN |
7088 | Elf64_Sym *sym; |
7089 | Elf64_Rel *rel; | |
646f02ff AN |
7090 | void *tmp; |
7091 | ||
7092 | if (!obj->efile.btf_maps_sec_btf_id || !obj->btf) | |
7093 | return -EINVAL; | |
7094 | sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id); | |
7095 | if (!sec) | |
7096 | return -EINVAL; | |
7097 | ||
646f02ff AN |
7098 | nrels = shdr->sh_size / shdr->sh_entsize; |
7099 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
7100 | rel = elf_rel_by_idx(data, i); |
7101 | if (!rel) { | |
646f02ff AN |
7102 | pr_warn(".maps relo #%d: failed to get ELF relo\n", i); |
7103 | return -LIBBPF_ERRNO__FORMAT; | |
7104 | } | |
ad23b723 AN |
7105 | |
7106 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); | |
7107 | if (!sym) { | |
646f02ff | 7108 | pr_warn(".maps relo #%d: symbol %zx not found\n", |
ad23b723 | 7109 | i, (size_t)ELF64_R_SYM(rel->r_info)); |
646f02ff AN |
7110 | return -LIBBPF_ERRNO__FORMAT; |
7111 | } | |
ad23b723 | 7112 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
646f02ff | 7113 | |
ad23b723 AN |
7114 | pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n", |
7115 | i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value, | |
7116 | (size_t)rel->r_offset, sym->st_name, name); | |
646f02ff AN |
7117 | |
7118 | for (j = 0; j < obj->nr_maps; j++) { | |
7119 | map = &obj->maps[j]; | |
7120 | if (map->sec_idx != obj->efile.btf_maps_shndx) | |
7121 | continue; | |
7122 | ||
7123 | vi = btf_var_secinfos(sec) + map->btf_var_idx; | |
ad23b723 AN |
7124 | if (vi->offset <= rel->r_offset && |
7125 | rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size) | |
646f02ff AN |
7126 | break; |
7127 | } | |
7128 | if (j == obj->nr_maps) { | |
ad23b723 AN |
7129 | pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n", |
7130 | i, name, (size_t)rel->r_offset); | |
646f02ff AN |
7131 | return -EINVAL; |
7132 | } | |
7133 | ||
341ac5ff HC |
7134 | is_map_in_map = bpf_map_type__is_map_in_map(map->def.type); |
7135 | is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY; | |
7136 | type = is_map_in_map ? "map" : "prog"; | |
7137 | if (is_map_in_map) { | |
7138 | if (sym->st_shndx != obj->efile.btf_maps_shndx) { | |
7139 | pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n", | |
7140 | i, name); | |
7141 | return -LIBBPF_ERRNO__RELOC; | |
7142 | } | |
7143 | if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS && | |
7144 | map->def.key_size != sizeof(int)) { | |
7145 | pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n", | |
7146 | i, map->name, sizeof(int)); | |
7147 | return -EINVAL; | |
7148 | } | |
7149 | targ_map = bpf_object__find_map_by_name(obj, name); | |
7150 | if (!targ_map) { | |
7151 | pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n", | |
7152 | i, name); | |
7153 | return -ESRCH; | |
7154 | } | |
7155 | } else if (is_prog_array) { | |
7156 | targ_prog = bpf_object__find_program_by_name(obj, name); | |
7157 | if (!targ_prog) { | |
7158 | pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n", | |
7159 | i, name); | |
7160 | return -ESRCH; | |
7161 | } | |
7162 | if (targ_prog->sec_idx != sym->st_shndx || | |
7163 | targ_prog->sec_insn_off * 8 != sym->st_value || | |
7164 | prog_is_subprog(obj, targ_prog)) { | |
7165 | pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n", | |
7166 | i, name); | |
7167 | return -LIBBPF_ERRNO__RELOC; | |
7168 | } | |
7169 | } else { | |
646f02ff AN |
7170 | return -EINVAL; |
7171 | } | |
7172 | ||
646f02ff AN |
7173 | var = btf__type_by_id(obj->btf, vi->type); |
7174 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
7175 | if (btf_vlen(def) == 0) | |
7176 | return -EINVAL; | |
7177 | member = btf_members(def) + btf_vlen(def) - 1; | |
7178 | mname = btf__name_by_offset(obj->btf, member->name_off); | |
7179 | if (strcmp(mname, "values")) | |
7180 | return -EINVAL; | |
7181 | ||
7182 | moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8; | |
ad23b723 | 7183 | if (rel->r_offset - vi->offset < moff) |
646f02ff AN |
7184 | return -EINVAL; |
7185 | ||
ad23b723 | 7186 | moff = rel->r_offset - vi->offset - moff; |
15728ad3 AN |
7187 | /* here we use BPF pointer size, which is always 64 bit, as we |
7188 | * are parsing ELF that was built for BPF target | |
7189 | */ | |
7190 | if (moff % bpf_ptr_sz) | |
646f02ff | 7191 | return -EINVAL; |
15728ad3 | 7192 | moff /= bpf_ptr_sz; |
646f02ff AN |
7193 | if (moff >= map->init_slots_sz) { |
7194 | new_sz = moff + 1; | |
029258d7 | 7195 | tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz); |
646f02ff AN |
7196 | if (!tmp) |
7197 | return -ENOMEM; | |
7198 | map->init_slots = tmp; | |
7199 | memset(map->init_slots + map->init_slots_sz, 0, | |
15728ad3 | 7200 | (new_sz - map->init_slots_sz) * host_ptr_sz); |
646f02ff AN |
7201 | map->init_slots_sz = new_sz; |
7202 | } | |
341ac5ff | 7203 | map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog; |
646f02ff | 7204 | |
341ac5ff HC |
7205 | pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n", |
7206 | i, map->name, moff, type, name); | |
646f02ff AN |
7207 | } |
7208 | ||
7209 | return 0; | |
7210 | } | |
590a0088 | 7211 | |
c3c55696 AN |
7212 | static int bpf_object__collect_relos(struct bpf_object *obj) |
7213 | { | |
7214 | int i, err; | |
34090915 | 7215 | |
25bbbd7a AN |
7216 | for (i = 0; i < obj->efile.sec_cnt; i++) { |
7217 | struct elf_sec_desc *sec_desc = &obj->efile.secs[i]; | |
7218 | Elf64_Shdr *shdr; | |
7219 | Elf_Data *data; | |
7220 | int idx; | |
7221 | ||
7222 | if (sec_desc->sec_type != SEC_RELO) | |
7223 | continue; | |
7224 | ||
7225 | shdr = sec_desc->shdr; | |
7226 | data = sec_desc->data; | |
7227 | idx = shdr->sh_info; | |
34090915 | 7228 | |
240bf8a5 | 7229 | if (shdr->sh_type != SHT_REL || idx < 0 || idx >= obj->efile.sec_cnt) { |
be18010e | 7230 | pr_warn("internal error at %d\n", __LINE__); |
6371ca3b | 7231 | return -LIBBPF_ERRNO__INTERNAL; |
34090915 WN |
7232 | } |
7233 | ||
240bf8a5 | 7234 | if (obj->efile.secs[idx].sec_type == SEC_ST_OPS) |
646f02ff | 7235 | err = bpf_object__collect_st_ops_relos(obj, shdr, data); |
c3c55696 | 7236 | else if (idx == obj->efile.btf_maps_shndx) |
646f02ff | 7237 | err = bpf_object__collect_map_relos(obj, shdr, data); |
c3c55696 AN |
7238 | else |
7239 | err = bpf_object__collect_prog_relos(obj, shdr, data); | |
34090915 | 7240 | if (err) |
6371ca3b | 7241 | return err; |
34090915 | 7242 | } |
c3c55696 | 7243 | |
d0e92887 | 7244 | bpf_object__sort_relos(obj); |
34090915 WN |
7245 | return 0; |
7246 | } | |
7247 | ||
109cea5a AN |
7248 | static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id) |
7249 | { | |
9b2f6fec | 7250 | if (BPF_CLASS(insn->code) == BPF_JMP && |
109cea5a AN |
7251 | BPF_OP(insn->code) == BPF_CALL && |
7252 | BPF_SRC(insn->code) == BPF_K && | |
9b2f6fec AN |
7253 | insn->src_reg == 0 && |
7254 | insn->dst_reg == 0) { | |
7255 | *func_id = insn->imm; | |
109cea5a AN |
7256 | return true; |
7257 | } | |
7258 | return false; | |
7259 | } | |
7260 | ||
42869d28 | 7261 | static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog) |
109cea5a AN |
7262 | { |
7263 | struct bpf_insn *insn = prog->insns; | |
7264 | enum bpf_func_id func_id; | |
7265 | int i; | |
7266 | ||
67234743 AS |
7267 | if (obj->gen_loader) |
7268 | return 0; | |
7269 | ||
109cea5a AN |
7270 | for (i = 0; i < prog->insns_cnt; i++, insn++) { |
7271 | if (!insn_is_helper_call(insn, &func_id)) | |
7272 | continue; | |
7273 | ||
7274 | /* on kernels that don't yet support | |
7275 | * bpf_probe_read_{kernel,user}[_str] helpers, fall back | |
7276 | * to bpf_probe_read() which works well for old kernels | |
7277 | */ | |
7278 | switch (func_id) { | |
7279 | case BPF_FUNC_probe_read_kernel: | |
7280 | case BPF_FUNC_probe_read_user: | |
9ca1f56a | 7281 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
7282 | insn->imm = BPF_FUNC_probe_read; |
7283 | break; | |
7284 | case BPF_FUNC_probe_read_kernel_str: | |
7285 | case BPF_FUNC_probe_read_user_str: | |
9ca1f56a | 7286 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
7287 | insn->imm = BPF_FUNC_probe_read_str; |
7288 | break; | |
7289 | default: | |
7290 | break; | |
7291 | } | |
7292 | } | |
7293 | return 0; | |
7294 | } | |
7295 | ||
15ea31fa AN |
7296 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
7297 | int *btf_obj_fd, int *btf_type_id); | |
12d9466d | 7298 | |
4fa5bcfe AN |
7299 | /* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */ |
7300 | static int libbpf_prepare_prog_load(struct bpf_program *prog, | |
7301 | struct bpf_prog_load_opts *opts, long cookie) | |
12d9466d | 7302 | { |
15ea31fa AN |
7303 | enum sec_def_flags def = cookie; |
7304 | ||
12d9466d | 7305 | /* old kernels might not support specifying expected_attach_type */ |
15ea31fa | 7306 | if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE)) |
d10ef2b8 | 7307 | opts->expected_attach_type = 0; |
12d9466d | 7308 | |
15ea31fa | 7309 | if (def & SEC_SLEEPABLE) |
d10ef2b8 | 7310 | opts->prog_flags |= BPF_F_SLEEPABLE; |
12d9466d | 7311 | |
082c4bfb LB |
7312 | if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS)) |
7313 | opts->prog_flags |= BPF_F_XDP_HAS_FRAGS; | |
7314 | ||
5902da6d JO |
7315 | /* special check for usdt to use uprobe_multi link */ |
7316 | if ((def & SEC_USDT) && kernel_supports(prog->obj, FEAT_UPROBE_MULTI_LINK)) | |
7317 | prog->expected_attach_type = BPF_TRACE_UPROBE_MULTI; | |
7318 | ||
cc7d8f2c | 7319 | if ((def & SEC_ATTACH_BTF) && !prog->attach_btf_id) { |
12d9466d | 7320 | int btf_obj_fd = 0, btf_type_id = 0, err; |
15ea31fa | 7321 | const char *attach_name; |
12d9466d | 7322 | |
cc7d8f2c AN |
7323 | attach_name = strchr(prog->sec_name, '/'); |
7324 | if (!attach_name) { | |
7325 | /* if BPF program is annotated with just SEC("fentry") | |
7326 | * (or similar) without declaratively specifying | |
7327 | * target, then it is expected that target will be | |
7328 | * specified with bpf_program__set_attach_target() at | |
7329 | * runtime before BPF object load step. If not, then | |
7330 | * there is nothing to load into the kernel as BPF | |
7331 | * verifier won't be able to validate BPF program | |
7332 | * correctness anyways. | |
7333 | */ | |
7334 | pr_warn("prog '%s': no BTF-based attach target is specified, use bpf_program__set_attach_target()\n", | |
7335 | prog->name); | |
7336 | return -EINVAL; | |
7337 | } | |
7338 | attach_name++; /* skip over / */ | |
7339 | ||
15ea31fa | 7340 | err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id); |
12d9466d AN |
7341 | if (err) |
7342 | return err; | |
7343 | ||
7344 | /* cache resolved BTF FD and BTF type ID in the prog */ | |
7345 | prog->attach_btf_obj_fd = btf_obj_fd; | |
7346 | prog->attach_btf_id = btf_type_id; | |
7347 | ||
7348 | /* but by now libbpf common logic is not utilizing | |
7349 | * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because | |
d10ef2b8 AN |
7350 | * this callback is called after opts were populated by |
7351 | * libbpf, so this callback has to update opts explicitly here | |
12d9466d | 7352 | */ |
d10ef2b8 AN |
7353 | opts->attach_btf_obj_fd = btf_obj_fd; |
7354 | opts->attach_btf_id = btf_type_id; | |
12d9466d AN |
7355 | } |
7356 | return 0; | |
7357 | } | |
7358 | ||
9fdc4273 AN |
7359 | static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz); |
7360 | ||
cf90a20d AN |
7361 | static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog, |
7362 | struct bpf_insn *insns, int insns_cnt, | |
7363 | const char *license, __u32 kern_version, int *prog_fd) | |
55cffde2 | 7364 | { |
d10ef2b8 AN |
7365 | LIBBPF_OPTS(bpf_prog_load_opts, load_attr); |
7366 | const char *prog_name = NULL; | |
1ce6a9fc | 7367 | char *cp, errmsg[STRERR_BUFSIZE]; |
8395f320 | 7368 | size_t log_buf_size = 0; |
b3ce9079 | 7369 | char *log_buf = NULL, *tmp; |
b3ce9079 AN |
7370 | bool own_log_buf = true; |
7371 | __u32 log_level = prog->log_level; | |
9bf48fa1 | 7372 | int ret, err; |
55cffde2 | 7373 | |
80b2b5c3 AM |
7374 | if (prog->type == BPF_PROG_TYPE_UNSPEC) { |
7375 | /* | |
7376 | * The program type must be set. Most likely we couldn't find a proper | |
7377 | * section definition at load time, and thus we didn't infer the type. | |
7378 | */ | |
7379 | pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n", | |
7380 | prog->name, prog->sec_name); | |
7381 | return -EINVAL; | |
7382 | } | |
7383 | ||
fba01a06 AN |
7384 | if (!insns || !insns_cnt) |
7385 | return -EINVAL; | |
7386 | ||
25bbbd7a | 7387 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
d10ef2b8 | 7388 | prog_name = prog->name; |
12d9466d AN |
7389 | load_attr.attach_prog_fd = prog->attach_prog_fd; |
7390 | load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd; | |
6aef10a4 AN |
7391 | load_attr.attach_btf_id = prog->attach_btf_id; |
7392 | load_attr.kern_version = kern_version; | |
7393 | load_attr.prog_ifindex = prog->prog_ifindex; | |
7394 | ||
0f0e55d8 | 7395 | /* specify func_info/line_info only if kernel supports them */ |
9bf48fa1 QM |
7396 | if (obj->btf && btf__fd(obj->btf) >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) { |
7397 | load_attr.prog_btf_fd = btf__fd(obj->btf); | |
0f0e55d8 AN |
7398 | load_attr.func_info = prog->func_info; |
7399 | load_attr.func_info_rec_size = prog->func_info_rec_size; | |
7400 | load_attr.func_info_cnt = prog->func_info_cnt; | |
7401 | load_attr.line_info = prog->line_info; | |
7402 | load_attr.line_info_rec_size = prog->line_info_rec_size; | |
7403 | load_attr.line_info_cnt = prog->line_info_cnt; | |
7404 | } | |
b3ce9079 | 7405 | load_attr.log_level = log_level; |
04656198 | 7406 | load_attr.prog_flags = prog->prog_flags; |
25bbbd7a | 7407 | load_attr.fd_array = obj->fd_array; |
55cffde2 | 7408 | |
6b434b61 AN |
7409 | load_attr.token_fd = obj->token_fd; |
7410 | if (obj->token_fd) | |
7411 | load_attr.prog_flags |= BPF_F_TOKEN_FD; | |
7412 | ||
12d9466d | 7413 | /* adjust load_attr if sec_def provides custom preload callback */ |
4fa5bcfe AN |
7414 | if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) { |
7415 | err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie); | |
12d9466d AN |
7416 | if (err < 0) { |
7417 | pr_warn("prog '%s': failed to prepare load attributes: %d\n", | |
7418 | prog->name, err); | |
7419 | return err; | |
7420 | } | |
b63b3c49 JO |
7421 | insns = prog->insns; |
7422 | insns_cnt = prog->insns_cnt; | |
12d9466d AN |
7423 | } |
7424 | ||
5902da6d JO |
7425 | /* allow prog_prepare_load_fn to change expected_attach_type */ |
7426 | load_attr.expected_attach_type = prog->expected_attach_type; | |
7427 | ||
25bbbd7a | 7428 | if (obj->gen_loader) { |
d10ef2b8 AN |
7429 | bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name, |
7430 | license, insns, insns_cnt, &load_attr, | |
25bbbd7a | 7431 | prog - obj->programs); |
be2f2d16 | 7432 | *prog_fd = -1; |
67234743 AS |
7433 | return 0; |
7434 | } | |
8395f320 | 7435 | |
b3ce9079 | 7436 | retry_load: |
bb412cf1 | 7437 | /* if log_level is zero, we don't request logs initially even if |
b3ce9079 AN |
7438 | * custom log_buf is specified; if the program load fails, then we'll |
7439 | * bump log_level to 1 and use either custom log_buf or we'll allocate | |
7440 | * our own and retry the load to get details on what failed | |
7441 | */ | |
7442 | if (log_level) { | |
7443 | if (prog->log_buf) { | |
7444 | log_buf = prog->log_buf; | |
7445 | log_buf_size = prog->log_size; | |
7446 | own_log_buf = false; | |
7447 | } else if (obj->log_buf) { | |
7448 | log_buf = obj->log_buf; | |
7449 | log_buf_size = obj->log_size; | |
7450 | own_log_buf = false; | |
7451 | } else { | |
7452 | log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2); | |
7453 | tmp = realloc(log_buf, log_buf_size); | |
7454 | if (!tmp) { | |
7455 | ret = -ENOMEM; | |
7456 | goto out; | |
7457 | } | |
7458 | log_buf = tmp; | |
7459 | log_buf[0] = '\0'; | |
7460 | own_log_buf = true; | |
7461 | } | |
8395f320 | 7462 | } |
55cffde2 | 7463 | |
6aef10a4 | 7464 | load_attr.log_buf = log_buf; |
d10ef2b8 | 7465 | load_attr.log_size = log_buf_size; |
b3ce9079 | 7466 | load_attr.log_level = log_level; |
55cffde2 | 7467 | |
b3ce9079 | 7468 | ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr); |
55cffde2 | 7469 | if (ret >= 0) { |
b3ce9079 | 7470 | if (log_level && own_log_buf) { |
ad9a7f96 AN |
7471 | pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n", |
7472 | prog->name, log_buf); | |
7473 | } | |
5d23328d | 7474 | |
25bbbd7a AN |
7475 | if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) { |
7476 | struct bpf_map *map; | |
7477 | int i; | |
7478 | ||
7479 | for (i = 0; i < obj->nr_maps; i++) { | |
7480 | map = &prog->obj->maps[i]; | |
7481 | if (map->libbpf_type != LIBBPF_MAP_RODATA) | |
7482 | continue; | |
5d23328d | 7483 | |
f08c18e0 | 7484 | if (bpf_prog_bind_map(ret, map->fd, NULL)) { |
25bbbd7a | 7485 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
ad9a7f96 AN |
7486 | pr_warn("prog '%s': failed to bind map '%s': %s\n", |
7487 | prog->name, map->real_name, cp); | |
25bbbd7a AN |
7488 | /* Don't fail hard if can't bind rodata. */ |
7489 | } | |
5d23328d YZ |
7490 | } |
7491 | } | |
7492 | ||
be2f2d16 | 7493 | *prog_fd = ret; |
55cffde2 WN |
7494 | ret = 0; |
7495 | goto out; | |
7496 | } | |
7497 | ||
b3ce9079 AN |
7498 | if (log_level == 0) { |
7499 | log_level = 1; | |
da11b417 AS |
7500 | goto retry_load; |
7501 | } | |
b3ce9079 AN |
7502 | /* On ENOSPC, increase log buffer size and retry, unless custom |
7503 | * log_buf is specified. | |
7504 | * Be careful to not overflow u32, though. Kernel's log buf size limit | |
7505 | * isn't part of UAPI so it can always be bumped to full 4GB. So don't | |
7506 | * multiply by 2 unless we are sure we'll fit within 32 bits. | |
7507 | * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2). | |
7508 | */ | |
7509 | if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2) | |
7510 | goto retry_load; | |
2eda2145 AN |
7511 | |
7512 | ret = -errno; | |
9fdc4273 AN |
7513 | |
7514 | /* post-process verifier log to improve error descriptions */ | |
7515 | fixup_verifier_log(prog, log_buf, log_buf_size); | |
7516 | ||
24d6a808 | 7517 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
ad9a7f96 | 7518 | pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, cp); |
dc3a2d25 | 7519 | pr_perm_msg(ret); |
55cffde2 | 7520 | |
b3ce9079 | 7521 | if (own_log_buf && log_buf && log_buf[0] != '\0') { |
ad9a7f96 AN |
7522 | pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n", |
7523 | prog->name, log_buf); | |
7524 | } | |
55cffde2 WN |
7525 | |
7526 | out: | |
b3ce9079 AN |
7527 | if (own_log_buf) |
7528 | free(log_buf); | |
55cffde2 WN |
7529 | return ret; |
7530 | } | |
7531 | ||
9fdc4273 AN |
7532 | static char *find_prev_line(char *buf, char *cur) |
7533 | { | |
7534 | char *p; | |
7535 | ||
7536 | if (cur == buf) /* end of a log buf */ | |
7537 | return NULL; | |
7538 | ||
7539 | p = cur - 1; | |
7540 | while (p - 1 >= buf && *(p - 1) != '\n') | |
7541 | p--; | |
7542 | ||
7543 | return p; | |
7544 | } | |
7545 | ||
7546 | static void patch_log(char *buf, size_t buf_sz, size_t log_sz, | |
7547 | char *orig, size_t orig_sz, const char *patch) | |
7548 | { | |
7549 | /* size of the remaining log content to the right from the to-be-replaced part */ | |
7550 | size_t rem_sz = (buf + log_sz) - (orig + orig_sz); | |
7551 | size_t patch_sz = strlen(patch); | |
7552 | ||
7553 | if (patch_sz != orig_sz) { | |
7554 | /* If patch line(s) are longer than original piece of verifier log, | |
7555 | * shift log contents by (patch_sz - orig_sz) bytes to the right | |
7556 | * starting from after to-be-replaced part of the log. | |
7557 | * | |
7558 | * If patch line(s) are shorter than original piece of verifier log, | |
7559 | * shift log contents by (orig_sz - patch_sz) bytes to the left | |
7560 | * starting from after to-be-replaced part of the log | |
7561 | * | |
7562 | * We need to be careful about not overflowing available | |
7563 | * buf_sz capacity. If that's the case, we'll truncate the end | |
7564 | * of the original log, as necessary. | |
7565 | */ | |
7566 | if (patch_sz > orig_sz) { | |
7567 | if (orig + patch_sz >= buf + buf_sz) { | |
7568 | /* patch is big enough to cover remaining space completely */ | |
7569 | patch_sz -= (orig + patch_sz) - (buf + buf_sz) + 1; | |
7570 | rem_sz = 0; | |
7571 | } else if (patch_sz - orig_sz > buf_sz - log_sz) { | |
7572 | /* patch causes part of remaining log to be truncated */ | |
7573 | rem_sz -= (patch_sz - orig_sz) - (buf_sz - log_sz); | |
7574 | } | |
7575 | } | |
7576 | /* shift remaining log to the right by calculated amount */ | |
7577 | memmove(orig + patch_sz, orig + orig_sz, rem_sz); | |
7578 | } | |
7579 | ||
7580 | memcpy(orig, patch, patch_sz); | |
7581 | } | |
7582 | ||
7583 | static void fixup_log_failed_core_relo(struct bpf_program *prog, | |
7584 | char *buf, size_t buf_sz, size_t log_sz, | |
7585 | char *line1, char *line2, char *line3) | |
7586 | { | |
7587 | /* Expected log for failed and not properly guarded CO-RE relocation: | |
7588 | * line1 -> 123: (85) call unknown#195896080 | |
7589 | * line2 -> invalid func unknown#195896080 | |
7590 | * line3 -> <anything else or end of buffer> | |
7591 | * | |
7592 | * "123" is the index of the instruction that was poisoned. We extract | |
7593 | * instruction index to find corresponding CO-RE relocation and | |
7594 | * replace this part of the log with more relevant information about | |
7595 | * failed CO-RE relocation. | |
7596 | */ | |
7597 | const struct bpf_core_relo *relo; | |
7598 | struct bpf_core_spec spec; | |
7599 | char patch[512], spec_buf[256]; | |
b198881d | 7600 | int insn_idx, err, spec_len; |
9fdc4273 AN |
7601 | |
7602 | if (sscanf(line1, "%d: (%*d) call unknown#195896080\n", &insn_idx) != 1) | |
7603 | return; | |
7604 | ||
7605 | relo = find_relo_core(prog, insn_idx); | |
7606 | if (!relo) | |
7607 | return; | |
7608 | ||
7609 | err = bpf_core_parse_spec(prog->name, prog->obj->btf, relo, &spec); | |
7610 | if (err) | |
7611 | return; | |
7612 | ||
b198881d | 7613 | spec_len = bpf_core_format_spec(spec_buf, sizeof(spec_buf), &spec); |
9fdc4273 AN |
7614 | snprintf(patch, sizeof(patch), |
7615 | "%d: <invalid CO-RE relocation>\n" | |
b198881d AN |
7616 | "failed to resolve CO-RE relocation %s%s\n", |
7617 | insn_idx, spec_buf, spec_len >= sizeof(spec_buf) ? "..." : ""); | |
9fdc4273 AN |
7618 | |
7619 | patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch); | |
7620 | } | |
7621 | ||
ec41817b AN |
7622 | static void fixup_log_missing_map_load(struct bpf_program *prog, |
7623 | char *buf, size_t buf_sz, size_t log_sz, | |
7624 | char *line1, char *line2, char *line3) | |
7625 | { | |
3055ddd6 | 7626 | /* Expected log for failed and not properly guarded map reference: |
ec41817b AN |
7627 | * line1 -> 123: (85) call unknown#2001000345 |
7628 | * line2 -> invalid func unknown#2001000345 | |
7629 | * line3 -> <anything else or end of buffer> | |
7630 | * | |
7631 | * "123" is the index of the instruction that was poisoned. | |
3055ddd6 | 7632 | * "345" in "2001000345" is a map index in obj->maps to fetch map name. |
ec41817b AN |
7633 | */ |
7634 | struct bpf_object *obj = prog->obj; | |
7635 | const struct bpf_map *map; | |
7636 | int insn_idx, map_idx; | |
7637 | char patch[128]; | |
7638 | ||
7639 | if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &map_idx) != 2) | |
7640 | return; | |
7641 | ||
3055ddd6 | 7642 | map_idx -= POISON_LDIMM64_MAP_BASE; |
ec41817b AN |
7643 | if (map_idx < 0 || map_idx >= obj->nr_maps) |
7644 | return; | |
7645 | map = &obj->maps[map_idx]; | |
7646 | ||
7647 | snprintf(patch, sizeof(patch), | |
7648 | "%d: <invalid BPF map reference>\n" | |
7649 | "BPF map '%s' is referenced but wasn't created\n", | |
7650 | insn_idx, map->name); | |
7651 | ||
7652 | patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch); | |
7653 | } | |
7654 | ||
05b6f766 AN |
7655 | static void fixup_log_missing_kfunc_call(struct bpf_program *prog, |
7656 | char *buf, size_t buf_sz, size_t log_sz, | |
7657 | char *line1, char *line2, char *line3) | |
7658 | { | |
7659 | /* Expected log for failed and not properly guarded kfunc call: | |
7660 | * line1 -> 123: (85) call unknown#2002000345 | |
7661 | * line2 -> invalid func unknown#2002000345 | |
7662 | * line3 -> <anything else or end of buffer> | |
7663 | * | |
7664 | * "123" is the index of the instruction that was poisoned. | |
7665 | * "345" in "2002000345" is an extern index in obj->externs to fetch kfunc name. | |
7666 | */ | |
7667 | struct bpf_object *obj = prog->obj; | |
7668 | const struct extern_desc *ext; | |
7669 | int insn_idx, ext_idx; | |
7670 | char patch[128]; | |
7671 | ||
7672 | if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &ext_idx) != 2) | |
7673 | return; | |
7674 | ||
7675 | ext_idx -= POISON_CALL_KFUNC_BASE; | |
7676 | if (ext_idx < 0 || ext_idx >= obj->nr_extern) | |
7677 | return; | |
7678 | ext = &obj->externs[ext_idx]; | |
7679 | ||
7680 | snprintf(patch, sizeof(patch), | |
7681 | "%d: <invalid kfunc call>\n" | |
7682 | "kfunc '%s' is referenced but wasn't resolved\n", | |
7683 | insn_idx, ext->name); | |
7684 | ||
7685 | patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch); | |
7686 | } | |
7687 | ||
9fdc4273 AN |
7688 | static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz) |
7689 | { | |
7690 | /* look for familiar error patterns in last N lines of the log */ | |
7691 | const size_t max_last_line_cnt = 10; | |
7692 | char *prev_line, *cur_line, *next_line; | |
7693 | size_t log_sz; | |
7694 | int i; | |
7695 | ||
7696 | if (!buf) | |
7697 | return; | |
7698 | ||
7699 | log_sz = strlen(buf) + 1; | |
7700 | next_line = buf + log_sz - 1; | |
7701 | ||
7702 | for (i = 0; i < max_last_line_cnt; i++, next_line = cur_line) { | |
7703 | cur_line = find_prev_line(buf, next_line); | |
7704 | if (!cur_line) | |
7705 | return; | |
7706 | ||
9fdc4273 AN |
7707 | if (str_has_pfx(cur_line, "invalid func unknown#195896080\n")) { |
7708 | prev_line = find_prev_line(buf, cur_line); | |
7709 | if (!prev_line) | |
7710 | continue; | |
7711 | ||
3055ddd6 | 7712 | /* failed CO-RE relocation case */ |
9fdc4273 AN |
7713 | fixup_log_failed_core_relo(prog, buf, buf_sz, log_sz, |
7714 | prev_line, cur_line, next_line); | |
7715 | return; | |
3055ddd6 | 7716 | } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_LDIMM64_MAP_PFX)) { |
ec41817b AN |
7717 | prev_line = find_prev_line(buf, cur_line); |
7718 | if (!prev_line) | |
7719 | continue; | |
7720 | ||
3055ddd6 | 7721 | /* reference to uncreated BPF map */ |
ec41817b AN |
7722 | fixup_log_missing_map_load(prog, buf, buf_sz, log_sz, |
7723 | prev_line, cur_line, next_line); | |
7724 | return; | |
05b6f766 AN |
7725 | } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_CALL_KFUNC_PFX)) { |
7726 | prev_line = find_prev_line(buf, cur_line); | |
7727 | if (!prev_line) | |
7728 | continue; | |
7729 | ||
7730 | /* reference to unresolved kfunc */ | |
7731 | fixup_log_missing_kfunc_call(prog, buf, buf_sz, log_sz, | |
7732 | prev_line, cur_line, next_line); | |
7733 | return; | |
9fdc4273 AN |
7734 | } |
7735 | } | |
7736 | } | |
7737 | ||
d0e92887 | 7738 | static int bpf_program_record_relos(struct bpf_program *prog) |
67234743 AS |
7739 | { |
7740 | struct bpf_object *obj = prog->obj; | |
7741 | int i; | |
7742 | ||
7743 | for (i = 0; i < prog->nr_reloc; i++) { | |
7744 | struct reloc_desc *relo = &prog->reloc_desc[i]; | |
3055ddd6 | 7745 | struct extern_desc *ext = &obj->externs[relo->ext_idx]; |
708cdc57 | 7746 | int kind; |
67234743 AS |
7747 | |
7748 | switch (relo->type) { | |
a18f7214 | 7749 | case RELO_EXTERN_LD64: |
67234743 AS |
7750 | if (ext->type != EXT_KSYM) |
7751 | continue; | |
708cdc57 AS |
7752 | kind = btf_is_var(btf__type_by_id(obj->btf, ext->btf_id)) ? |
7753 | BTF_KIND_VAR : BTF_KIND_FUNC; | |
c24941cd KKD |
7754 | bpf_gen__record_extern(obj->gen_loader, ext->name, |
7755 | ext->is_weak, !ext->ksym.type_id, | |
708cdc57 | 7756 | true, kind, relo->insn_idx); |
67234743 | 7757 | break; |
a18f7214 | 7758 | case RELO_EXTERN_CALL: |
c24941cd | 7759 | bpf_gen__record_extern(obj->gen_loader, ext->name, |
708cdc57 | 7760 | ext->is_weak, false, false, BTF_KIND_FUNC, |
67234743 AS |
7761 | relo->insn_idx); |
7762 | break; | |
d0e92887 AS |
7763 | case RELO_CORE: { |
7764 | struct bpf_core_relo cr = { | |
7765 | .insn_off = relo->insn_idx * 8, | |
7766 | .type_id = relo->core_relo->type_id, | |
7767 | .access_str_off = relo->core_relo->access_str_off, | |
7768 | .kind = relo->core_relo->kind, | |
7769 | }; | |
7770 | ||
7771 | bpf_gen__record_relo_core(obj->gen_loader, &cr); | |
7772 | break; | |
7773 | } | |
67234743 AS |
7774 | default: |
7775 | continue; | |
7776 | } | |
7777 | } | |
7778 | return 0; | |
7779 | } | |
7780 | ||
55cffde2 | 7781 | static int |
60276f98 | 7782 | bpf_object__load_progs(struct bpf_object *obj, int log_level) |
55cffde2 | 7783 | { |
d9297581 | 7784 | struct bpf_program *prog; |
55cffde2 WN |
7785 | size_t i; |
7786 | int err; | |
7787 | ||
109cea5a AN |
7788 | for (i = 0; i < obj->nr_programs; i++) { |
7789 | prog = &obj->programs[i]; | |
7790 | err = bpf_object__sanitize_prog(obj, prog); | |
7791 | if (err) | |
7792 | return err; | |
7793 | } | |
7794 | ||
55cffde2 | 7795 | for (i = 0; i < obj->nr_programs; i++) { |
d9297581 | 7796 | prog = &obj->programs[i]; |
c3c55696 | 7797 | if (prog_is_subprog(obj, prog)) |
48cca7e4 | 7798 | continue; |
a3820c48 | 7799 | if (!prog->autoload) { |
9c0f8cbd | 7800 | pr_debug("prog '%s': skipped loading\n", prog->name); |
d9297581 AN |
7801 | continue; |
7802 | } | |
7803 | prog->log_level |= log_level; | |
cf90a20d AN |
7804 | |
7805 | if (obj->gen_loader) | |
7806 | bpf_program_record_relos(prog); | |
7807 | ||
7808 | err = bpf_object_load_prog(obj, prog, prog->insns, prog->insns_cnt, | |
7809 | obj->license, obj->kern_version, &prog->fd); | |
7810 | if (err) { | |
7811 | pr_warn("prog '%s': failed to load: %d\n", prog->name, err); | |
55cffde2 | 7812 | return err; |
cf90a20d | 7813 | } |
55cffde2 | 7814 | } |
185cfe83 AN |
7815 | |
7816 | bpf_object__free_relocs(obj); | |
55cffde2 WN |
7817 | return 0; |
7818 | } | |
7819 | ||
25498a19 AN |
7820 | static const struct bpf_sec_def *find_sec_def(const char *sec_name); |
7821 | ||
91b4d1d1 AN |
7822 | static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts) |
7823 | { | |
7824 | struct bpf_program *prog; | |
12d9466d | 7825 | int err; |
91b4d1d1 AN |
7826 | |
7827 | bpf_object__for_each_program(prog, obj) { | |
7828 | prog->sec_def = find_sec_def(prog->sec_name); | |
7829 | if (!prog->sec_def) { | |
7830 | /* couldn't guess, but user might manually specify */ | |
7831 | pr_debug("prog '%s': unrecognized ELF section name '%s'\n", | |
7832 | prog->name, prog->sec_name); | |
7833 | continue; | |
7834 | } | |
7835 | ||
df286716 GS |
7836 | prog->type = prog->sec_def->prog_type; |
7837 | prog->expected_attach_type = prog->sec_def->expected_attach_type; | |
91b4d1d1 | 7838 | |
12d9466d AN |
7839 | /* sec_def can have custom callback which should be called |
7840 | * after bpf_program is initialized to adjust its properties | |
7841 | */ | |
4fa5bcfe AN |
7842 | if (prog->sec_def->prog_setup_fn) { |
7843 | err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie); | |
12d9466d AN |
7844 | if (err < 0) { |
7845 | pr_warn("prog '%s': failed to initialize: %d\n", | |
7846 | prog->name, err); | |
7847 | return err; | |
7848 | } | |
7849 | } | |
91b4d1d1 AN |
7850 | } |
7851 | ||
7852 | return 0; | |
7853 | } | |
7854 | ||
ad9a7f96 AN |
7855 | static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz, |
7856 | const struct bpf_object_open_opts *opts) | |
1a5e3fb1 | 7857 | { |
6b434b61 | 7858 | const char *obj_name, *kconfig, *btf_tmp_path, *token_path; |
1a5e3fb1 | 7859 | struct bpf_object *obj; |
291ee02b | 7860 | char tmp_name[64]; |
d17aff80 | 7861 | int err; |
e0e3ea88 AN |
7862 | char *log_buf; |
7863 | size_t log_size; | |
7864 | __u32 log_level; | |
1a5e3fb1 WN |
7865 | |
7866 | if (elf_version(EV_CURRENT) == EV_NONE) { | |
be18010e KW |
7867 | pr_warn("failed to init libelf for %s\n", |
7868 | path ? : "(mem buf)"); | |
6371ca3b | 7869 | return ERR_PTR(-LIBBPF_ERRNO__LIBELF); |
1a5e3fb1 WN |
7870 | } |
7871 | ||
291ee02b AN |
7872 | if (!OPTS_VALID(opts, bpf_object_open_opts)) |
7873 | return ERR_PTR(-EINVAL); | |
7874 | ||
1aace10f | 7875 | obj_name = OPTS_GET(opts, object_name, NULL); |
291ee02b AN |
7876 | if (obj_buf) { |
7877 | if (!obj_name) { | |
7878 | snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx", | |
7879 | (unsigned long)obj_buf, | |
7880 | (unsigned long)obj_buf_sz); | |
7881 | obj_name = tmp_name; | |
7882 | } | |
7883 | path = obj_name; | |
7884 | pr_debug("loading object '%s' from buffer\n", obj_name); | |
7885 | } | |
7886 | ||
e0e3ea88 AN |
7887 | log_buf = OPTS_GET(opts, kernel_log_buf, NULL); |
7888 | log_size = OPTS_GET(opts, kernel_log_size, 0); | |
7889 | log_level = OPTS_GET(opts, kernel_log_level, 0); | |
7890 | if (log_size > UINT_MAX) | |
7891 | return ERR_PTR(-EINVAL); | |
7892 | if (log_size && !log_buf) | |
7893 | return ERR_PTR(-EINVAL); | |
7894 | ||
6b434b61 | 7895 | token_path = OPTS_GET(opts, bpf_token_path, NULL); |
cac270ad AN |
7896 | /* if user didn't specify bpf_token_path explicitly, check if |
7897 | * LIBBPF_BPF_TOKEN_PATH envvar was set and treat it as bpf_token_path | |
7898 | * option | |
7899 | */ | |
7900 | if (!token_path) | |
7901 | token_path = getenv("LIBBPF_BPF_TOKEN_PATH"); | |
6b434b61 AN |
7902 | if (token_path && strlen(token_path) >= PATH_MAX) |
7903 | return ERR_PTR(-ENAMETOOLONG); | |
7904 | ||
2ce8450e | 7905 | obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name); |
6371ca3b WN |
7906 | if (IS_ERR(obj)) |
7907 | return obj; | |
1a5e3fb1 | 7908 | |
e0e3ea88 AN |
7909 | obj->log_buf = log_buf; |
7910 | obj->log_size = log_size; | |
7911 | obj->log_level = log_level; | |
7912 | ||
6b434b61 AN |
7913 | if (token_path) { |
7914 | obj->token_path = strdup(token_path); | |
7915 | if (!obj->token_path) { | |
7916 | err = -ENOMEM; | |
7917 | goto out; | |
7918 | } | |
7919 | } | |
7920 | ||
1373ff59 SC |
7921 | btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL); |
7922 | if (btf_tmp_path) { | |
7923 | if (strlen(btf_tmp_path) >= PATH_MAX) { | |
7924 | err = -ENAMETOOLONG; | |
7925 | goto out; | |
7926 | } | |
7927 | obj->btf_custom_path = strdup(btf_tmp_path); | |
7928 | if (!obj->btf_custom_path) { | |
7929 | err = -ENOMEM; | |
7930 | goto out; | |
7931 | } | |
7932 | } | |
7933 | ||
8601fd42 AN |
7934 | kconfig = OPTS_GET(opts, kconfig, NULL); |
7935 | if (kconfig) { | |
7936 | obj->kconfig = strdup(kconfig); | |
18353c87 SC |
7937 | if (!obj->kconfig) { |
7938 | err = -ENOMEM; | |
7939 | goto out; | |
7940 | } | |
166750bc | 7941 | } |
291ee02b | 7942 | |
0d13bfce AN |
7943 | err = bpf_object__elf_init(obj); |
7944 | err = err ? : bpf_object__check_endianness(obj); | |
7945 | err = err ? : bpf_object__elf_collect(obj); | |
166750bc | 7946 | err = err ? : bpf_object__collect_externs(obj); |
f33f742d | 7947 | err = err ? : bpf_object_fixup_btf(obj); |
0d13bfce | 7948 | err = err ? : bpf_object__init_maps(obj, opts); |
91b4d1d1 | 7949 | err = err ? : bpf_object_init_progs(obj, opts); |
c3c55696 | 7950 | err = err ? : bpf_object__collect_relos(obj); |
0d13bfce AN |
7951 | if (err) |
7952 | goto out; | |
dd4436bb | 7953 | |
91b4d1d1 | 7954 | bpf_object__elf_finish(obj); |
dd4436bb | 7955 | |
1a5e3fb1 WN |
7956 | return obj; |
7957 | out: | |
7958 | bpf_object__close(obj); | |
6371ca3b | 7959 | return ERR_PTR(err); |
1a5e3fb1 WN |
7960 | } |
7961 | ||
2ce8450e | 7962 | struct bpf_object * |
01af3bf0 | 7963 | bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts) |
2ce8450e | 7964 | { |
2ce8450e | 7965 | if (!path) |
e9fc3ce9 | 7966 | return libbpf_err_ptr(-EINVAL); |
2ce8450e AN |
7967 | |
7968 | pr_debug("loading %s\n", path); | |
7969 | ||
ad9a7f96 | 7970 | return libbpf_ptr(bpf_object_open(path, NULL, 0, opts)); |
2ce8450e AN |
7971 | } |
7972 | ||
146bf811 AN |
7973 | struct bpf_object *bpf_object__open(const char *path) |
7974 | { | |
7975 | return bpf_object__open_file(path, NULL); | |
7976 | } | |
7977 | ||
2ce8450e AN |
7978 | struct bpf_object * |
7979 | bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz, | |
01af3bf0 | 7980 | const struct bpf_object_open_opts *opts) |
6c956392 | 7981 | { |
2ce8450e | 7982 | if (!obj_buf || obj_buf_sz == 0) |
e9fc3ce9 | 7983 | return libbpf_err_ptr(-EINVAL); |
6c956392 | 7984 | |
ad9a7f96 | 7985 | return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, opts)); |
2ce8450e AN |
7986 | } |
7987 | ||
4a404a7e | 7988 | static int bpf_object_unload(struct bpf_object *obj) |
52d3352e WN |
7989 | { |
7990 | size_t i; | |
7991 | ||
7992 | if (!obj) | |
e9fc3ce9 | 7993 | return libbpf_err(-EINVAL); |
52d3352e | 7994 | |
590a0088 | 7995 | for (i = 0; i < obj->nr_maps; i++) { |
9d759a9b | 7996 | zclose(obj->maps[i].fd); |
590a0088 MKL |
7997 | if (obj->maps[i].st_ops) |
7998 | zfree(&obj->maps[i].st_ops->kern_vdata); | |
7999 | } | |
52d3352e | 8000 | |
55cffde2 WN |
8001 | for (i = 0; i < obj->nr_programs; i++) |
8002 | bpf_program__unload(&obj->programs[i]); | |
8003 | ||
52d3352e WN |
8004 | return 0; |
8005 | } | |
8006 | ||
0d13bfce AN |
8007 | static int bpf_object__sanitize_maps(struct bpf_object *obj) |
8008 | { | |
8009 | struct bpf_map *m; | |
8010 | ||
8011 | bpf_object__for_each_map(m, obj) { | |
8012 | if (!bpf_map__is_internal(m)) | |
8013 | continue; | |
9ca1f56a | 8014 | if (!kernel_supports(obj, FEAT_ARRAY_MMAP)) |
6920b086 | 8015 | m->def.map_flags &= ~BPF_F_MMAPABLE; |
0d13bfce AN |
8016 | } |
8017 | ||
8018 | return 0; | |
8019 | } | |
8020 | ||
ad2b0528 YS |
8021 | typedef int (*kallsyms_cb_t)(unsigned long long sym_addr, char sym_type, |
8022 | const char *sym_name, void *ctx); | |
8023 | ||
8024 | static int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *ctx) | |
1c0c7074 AN |
8025 | { |
8026 | char sym_type, sym_name[500]; | |
8027 | unsigned long long sym_addr; | |
1c0c7074 AN |
8028 | int ret, err = 0; |
8029 | FILE *f; | |
8030 | ||
59842c54 | 8031 | f = fopen("/proc/kallsyms", "re"); |
1c0c7074 AN |
8032 | if (!f) { |
8033 | err = -errno; | |
8034 | pr_warn("failed to open /proc/kallsyms: %d\n", err); | |
8035 | return err; | |
8036 | } | |
8037 | ||
8038 | while (true) { | |
8039 | ret = fscanf(f, "%llx %c %499s%*[^\n]\n", | |
8040 | &sym_addr, &sym_type, sym_name); | |
8041 | if (ret == EOF && feof(f)) | |
8042 | break; | |
8043 | if (ret != 3) { | |
135c783f | 8044 | pr_warn("failed to read kallsyms entry: %d\n", ret); |
1c0c7074 | 8045 | err = -EINVAL; |
85153ac0 | 8046 | break; |
1c0c7074 AN |
8047 | } |
8048 | ||
85153ac0 JO |
8049 | err = cb(sym_addr, sym_type, sym_name, ctx); |
8050 | if (err) | |
8051 | break; | |
1c0c7074 AN |
8052 | } |
8053 | ||
1c0c7074 AN |
8054 | fclose(f); |
8055 | return err; | |
8056 | } | |
8057 | ||
85153ac0 JO |
8058 | static int kallsyms_cb(unsigned long long sym_addr, char sym_type, |
8059 | const char *sym_name, void *ctx) | |
8060 | { | |
8061 | struct bpf_object *obj = ctx; | |
8062 | const struct btf_type *t; | |
8063 | struct extern_desc *ext; | |
c56e5977 | 8064 | char *res; |
85153ac0 | 8065 | |
c56e5977 YS |
8066 | res = strstr(sym_name, ".llvm."); |
8067 | if (sym_type == 'd' && res) | |
8068 | ext = find_extern_by_name_with_len(obj, sym_name, res - sym_name); | |
8069 | else | |
8070 | ext = find_extern_by_name(obj, sym_name); | |
85153ac0 JO |
8071 | if (!ext || ext->type != EXT_KSYM) |
8072 | return 0; | |
8073 | ||
8074 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
8075 | if (!btf_is_var(t)) | |
8076 | return 0; | |
8077 | ||
8078 | if (ext->is_set && ext->ksym.addr != sym_addr) { | |
55d00c37 | 8079 | pr_warn("extern (ksym) '%s': resolution is ambiguous: 0x%llx or 0x%llx\n", |
85153ac0 JO |
8080 | sym_name, ext->ksym.addr, sym_addr); |
8081 | return -EINVAL; | |
8082 | } | |
8083 | if (!ext->is_set) { | |
8084 | ext->is_set = true; | |
8085 | ext->ksym.addr = sym_addr; | |
55d00c37 | 8086 | pr_debug("extern (ksym) '%s': set to 0x%llx\n", sym_name, sym_addr); |
85153ac0 JO |
8087 | } |
8088 | return 0; | |
8089 | } | |
8090 | ||
8091 | static int bpf_object__read_kallsyms_file(struct bpf_object *obj) | |
8092 | { | |
8093 | return libbpf_kallsyms_parse(kallsyms_cb, obj); | |
8094 | } | |
8095 | ||
774e132e MKL |
8096 | static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name, |
8097 | __u16 kind, struct btf **res_btf, | |
9dbe6015 | 8098 | struct module_btf **res_mod_btf) |
d370bbe1 | 8099 | { |
9dbe6015 | 8100 | struct module_btf *mod_btf; |
284d2587 | 8101 | struct btf *btf; |
9dbe6015 | 8102 | int i, id, err; |
d370bbe1 | 8103 | |
933d1aa3 | 8104 | btf = obj->btf_vmlinux; |
9dbe6015 | 8105 | mod_btf = NULL; |
774e132e MKL |
8106 | id = btf__find_by_name_kind(btf, ksym_name, kind); |
8107 | ||
933d1aa3 MKL |
8108 | if (id == -ENOENT) { |
8109 | err = load_module_btfs(obj); | |
8110 | if (err) | |
8111 | return err; | |
d370bbe1 | 8112 | |
933d1aa3 | 8113 | for (i = 0; i < obj->btf_module_cnt; i++) { |
9dbe6015 KKD |
8114 | /* we assume module_btf's BTF FD is always >0 */ |
8115 | mod_btf = &obj->btf_modules[i]; | |
8116 | btf = mod_btf->btf; | |
8117 | id = btf__find_by_name_kind_own(btf, ksym_name, kind); | |
933d1aa3 MKL |
8118 | if (id != -ENOENT) |
8119 | break; | |
8120 | } | |
8121 | } | |
2211c825 | 8122 | if (id <= 0) |
933d1aa3 | 8123 | return -ESRCH; |
d370bbe1 | 8124 | |
774e132e | 8125 | *res_btf = btf; |
9dbe6015 | 8126 | *res_mod_btf = mod_btf; |
774e132e MKL |
8127 | return id; |
8128 | } | |
8129 | ||
8130 | static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj, | |
8131 | struct extern_desc *ext) | |
8132 | { | |
8133 | const struct btf_type *targ_var, *targ_type; | |
8134 | __u32 targ_type_id, local_type_id; | |
9dbe6015 | 8135 | struct module_btf *mod_btf = NULL; |
774e132e | 8136 | const char *targ_var_name; |
774e132e | 8137 | struct btf *btf = NULL; |
9dbe6015 | 8138 | int id, err; |
774e132e | 8139 | |
9dbe6015 | 8140 | id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf); |
466b2e13 KKD |
8141 | if (id < 0) { |
8142 | if (id == -ESRCH && ext->is_weak) | |
8143 | return 0; | |
2211c825 HL |
8144 | pr_warn("extern (var ksym) '%s': not found in kernel BTF\n", |
8145 | ext->name); | |
774e132e | 8146 | return id; |
2211c825 | 8147 | } |
774e132e | 8148 | |
933d1aa3 MKL |
8149 | /* find local type_id */ |
8150 | local_type_id = ext->ksym.type_id; | |
284d2587 | 8151 | |
933d1aa3 MKL |
8152 | /* find target type_id */ |
8153 | targ_var = btf__type_by_id(btf, id); | |
8154 | targ_var_name = btf__name_by_offset(btf, targ_var->name_off); | |
8155 | targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id); | |
d370bbe1 | 8156 | |
933d1aa3 MKL |
8157 | err = bpf_core_types_are_compat(obj->btf, local_type_id, |
8158 | btf, targ_type_id); | |
8159 | if (err <= 0) { | |
8160 | const struct btf_type *local_type; | |
8161 | const char *targ_name, *local_name; | |
d370bbe1 | 8162 | |
933d1aa3 MKL |
8163 | local_type = btf__type_by_id(obj->btf, local_type_id); |
8164 | local_name = btf__name_by_offset(obj->btf, local_type->name_off); | |
8165 | targ_name = btf__name_by_offset(btf, targ_type->name_off); | |
d370bbe1 | 8166 | |
933d1aa3 MKL |
8167 | pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n", |
8168 | ext->name, local_type_id, | |
8169 | btf_kind_str(local_type), local_name, targ_type_id, | |
8170 | btf_kind_str(targ_type), targ_name); | |
8171 | return -EINVAL; | |
8172 | } | |
d370bbe1 | 8173 | |
933d1aa3 | 8174 | ext->is_set = true; |
9dbe6015 | 8175 | ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0; |
933d1aa3 MKL |
8176 | ext->ksym.kernel_btf_id = id; |
8177 | pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n", | |
8178 | ext->name, id, btf_kind_str(targ_var), targ_var_name); | |
d370bbe1 | 8179 | |
933d1aa3 MKL |
8180 | return 0; |
8181 | } | |
d370bbe1 | 8182 | |
5bd022ec MKL |
8183 | static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj, |
8184 | struct extern_desc *ext) | |
8185 | { | |
8186 | int local_func_proto_id, kfunc_proto_id, kfunc_id; | |
9dbe6015 | 8187 | struct module_btf *mod_btf = NULL; |
5bd022ec MKL |
8188 | const struct btf_type *kern_func; |
8189 | struct btf *kern_btf = NULL; | |
9dbe6015 | 8190 | int ret; |
5bd022ec MKL |
8191 | |
8192 | local_func_proto_id = ext->ksym.type_id; | |
8193 | ||
5964a223 DM |
8194 | kfunc_id = find_ksym_btf_id(obj, ext->essent_name ?: ext->name, BTF_KIND_FUNC, &kern_btf, |
8195 | &mod_btf); | |
5bd022ec | 8196 | if (kfunc_id < 0) { |
466b2e13 KKD |
8197 | if (kfunc_id == -ESRCH && ext->is_weak) |
8198 | return 0; | |
8199 | pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n", | |
5bd022ec MKL |
8200 | ext->name); |
8201 | return kfunc_id; | |
8202 | } | |
8203 | ||
5bd022ec MKL |
8204 | kern_func = btf__type_by_id(kern_btf, kfunc_id); |
8205 | kfunc_proto_id = kern_func->type; | |
8206 | ||
8207 | ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id, | |
8208 | kern_btf, kfunc_proto_id); | |
8209 | if (ret <= 0) { | |
5964a223 DM |
8210 | if (ext->is_weak) |
8211 | return 0; | |
8212 | ||
f709160d AN |
8213 | pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with %s [%d]\n", |
8214 | ext->name, local_func_proto_id, | |
8215 | mod_btf ? mod_btf->name : "vmlinux", kfunc_proto_id); | |
5bd022ec MKL |
8216 | return -EINVAL; |
8217 | } | |
8218 | ||
9dbe6015 KKD |
8219 | /* set index for module BTF fd in fd_array, if unset */ |
8220 | if (mod_btf && !mod_btf->fd_array_idx) { | |
8221 | /* insn->off is s16 */ | |
8222 | if (obj->fd_array_cnt == INT16_MAX) { | |
8223 | pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n", | |
8224 | ext->name, mod_btf->fd_array_idx); | |
8225 | return -E2BIG; | |
8226 | } | |
8227 | /* Cannot use index 0 for module BTF fd */ | |
8228 | if (!obj->fd_array_cnt) | |
8229 | obj->fd_array_cnt = 1; | |
8230 | ||
8231 | ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int), | |
8232 | obj->fd_array_cnt + 1); | |
8233 | if (ret) | |
8234 | return ret; | |
8235 | mod_btf->fd_array_idx = obj->fd_array_cnt; | |
8236 | /* we assume module BTF FD is always >0 */ | |
8237 | obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd; | |
8238 | } | |
8239 | ||
5bd022ec | 8240 | ext->is_set = true; |
5bd022ec | 8241 | ext->ksym.kernel_btf_id = kfunc_id; |
9dbe6015 | 8242 | ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0; |
5fc13ad5 AS |
8243 | /* Also set kernel_btf_obj_fd to make sure that bpf_object__relocate_data() |
8244 | * populates FD into ld_imm64 insn when it's used to point to kfunc. | |
8245 | * {kernel_btf_id, btf_fd_idx} -> fixup bpf_call. | |
8246 | * {kernel_btf_id, kernel_btf_obj_fd} -> fixup ld_imm64. | |
8247 | */ | |
8248 | ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0; | |
f709160d AN |
8249 | pr_debug("extern (func ksym) '%s': resolved to %s [%d]\n", |
8250 | ext->name, mod_btf ? mod_btf->name : "vmlinux", kfunc_id); | |
5bd022ec MKL |
8251 | |
8252 | return 0; | |
8253 | } | |
8254 | ||
933d1aa3 MKL |
8255 | static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj) |
8256 | { | |
5bd022ec | 8257 | const struct btf_type *t; |
933d1aa3 MKL |
8258 | struct extern_desc *ext; |
8259 | int i, err; | |
8260 | ||
8261 | for (i = 0; i < obj->nr_extern; i++) { | |
8262 | ext = &obj->externs[i]; | |
8263 | if (ext->type != EXT_KSYM || !ext->ksym.type_id) | |
8264 | continue; | |
8265 | ||
67234743 AS |
8266 | if (obj->gen_loader) { |
8267 | ext->is_set = true; | |
8268 | ext->ksym.kernel_btf_obj_fd = 0; | |
8269 | ext->ksym.kernel_btf_id = 0; | |
8270 | continue; | |
8271 | } | |
5bd022ec MKL |
8272 | t = btf__type_by_id(obj->btf, ext->btf_id); |
8273 | if (btf_is_var(t)) | |
8274 | err = bpf_object__resolve_ksym_var_btf_id(obj, ext); | |
8275 | else | |
8276 | err = bpf_object__resolve_ksym_func_btf_id(obj, ext); | |
933d1aa3 MKL |
8277 | if (err) |
8278 | return err; | |
d370bbe1 HL |
8279 | } |
8280 | return 0; | |
8281 | } | |
8282 | ||
166750bc | 8283 | static int bpf_object__resolve_externs(struct bpf_object *obj, |
8601fd42 | 8284 | const char *extra_kconfig) |
166750bc | 8285 | { |
1c0c7074 | 8286 | bool need_config = false, need_kallsyms = false; |
d370bbe1 | 8287 | bool need_vmlinux_btf = false; |
166750bc | 8288 | struct extern_desc *ext; |
2e33efe3 | 8289 | void *kcfg_data = NULL; |
166750bc | 8290 | int err, i; |
166750bc AN |
8291 | |
8292 | if (obj->nr_extern == 0) | |
8293 | return 0; | |
8294 | ||
2e33efe3 AN |
8295 | if (obj->kconfig_map_idx >= 0) |
8296 | kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped; | |
166750bc AN |
8297 | |
8298 | for (i = 0; i < obj->nr_extern; i++) { | |
8299 | ext = &obj->externs[i]; | |
8300 | ||
55d00c37 AN |
8301 | if (ext->type == EXT_KSYM) { |
8302 | if (ext->ksym.type_id) | |
8303 | need_vmlinux_btf = true; | |
8304 | else | |
8305 | need_kallsyms = true; | |
8306 | continue; | |
8307 | } else if (ext->type == EXT_KCFG) { | |
8308 | void *ext_ptr = kcfg_data + ext->kcfg.data_off; | |
8309 | __u64 value = 0; | |
166750bc | 8310 | |
55d00c37 AN |
8311 | /* Kconfig externs need actual /proc/config.gz */ |
8312 | if (str_has_pfx(ext->name, "CONFIG_")) { | |
8313 | need_config = true; | |
8314 | continue; | |
8315 | } | |
8316 | ||
8317 | /* Virtual kcfg externs are customly handled by libbpf */ | |
8318 | if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) { | |
8319 | value = get_kernel_version(); | |
8320 | if (!value) { | |
8321 | pr_warn("extern (kcfg) '%s': failed to get kernel version\n", ext->name); | |
8322 | return -EINVAL; | |
8323 | } | |
8324 | } else if (strcmp(ext->name, "LINUX_HAS_BPF_COOKIE") == 0) { | |
8325 | value = kernel_supports(obj, FEAT_BPF_COOKIE); | |
6f5d467d AN |
8326 | } else if (strcmp(ext->name, "LINUX_HAS_SYSCALL_WRAPPER") == 0) { |
8327 | value = kernel_supports(obj, FEAT_SYSCALL_WRAPPER); | |
55d00c37 AN |
8328 | } else if (!str_has_pfx(ext->name, "LINUX_") || !ext->is_weak) { |
8329 | /* Currently libbpf supports only CONFIG_ and LINUX_ prefixed | |
8330 | * __kconfig externs, where LINUX_ ones are virtual and filled out | |
8331 | * customly by libbpf (their values don't come from Kconfig). | |
8332 | * If LINUX_xxx variable is not recognized by libbpf, but is marked | |
8333 | * __weak, it defaults to zero value, just like for CONFIG_xxx | |
8334 | * externs. | |
8335 | */ | |
8336 | pr_warn("extern (kcfg) '%s': unrecognized virtual extern\n", ext->name); | |
166750bc AN |
8337 | return -EINVAL; |
8338 | } | |
55d00c37 AN |
8339 | |
8340 | err = set_kcfg_value_num(ext, ext_ptr, value); | |
166750bc AN |
8341 | if (err) |
8342 | return err; | |
55d00c37 AN |
8343 | pr_debug("extern (kcfg) '%s': set to 0x%llx\n", |
8344 | ext->name, (long long)value); | |
166750bc | 8345 | } else { |
55d00c37 | 8346 | pr_warn("extern '%s': unrecognized extern kind\n", ext->name); |
166750bc AN |
8347 | return -EINVAL; |
8348 | } | |
8349 | } | |
8601fd42 | 8350 | if (need_config && extra_kconfig) { |
2e33efe3 | 8351 | err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data); |
8601fd42 AN |
8352 | if (err) |
8353 | return -EINVAL; | |
8354 | need_config = false; | |
8355 | for (i = 0; i < obj->nr_extern; i++) { | |
8356 | ext = &obj->externs[i]; | |
2e33efe3 | 8357 | if (ext->type == EXT_KCFG && !ext->is_set) { |
8601fd42 AN |
8358 | need_config = true; |
8359 | break; | |
8360 | } | |
8361 | } | |
8362 | } | |
166750bc | 8363 | if (need_config) { |
2e33efe3 | 8364 | err = bpf_object__read_kconfig_file(obj, kcfg_data); |
166750bc AN |
8365 | if (err) |
8366 | return -EINVAL; | |
8367 | } | |
1c0c7074 AN |
8368 | if (need_kallsyms) { |
8369 | err = bpf_object__read_kallsyms_file(obj); | |
8370 | if (err) | |
8371 | return -EINVAL; | |
8372 | } | |
d370bbe1 HL |
8373 | if (need_vmlinux_btf) { |
8374 | err = bpf_object__resolve_ksyms_btf_id(obj); | |
8375 | if (err) | |
8376 | return -EINVAL; | |
8377 | } | |
166750bc AN |
8378 | for (i = 0; i < obj->nr_extern; i++) { |
8379 | ext = &obj->externs[i]; | |
8380 | ||
8381 | if (!ext->is_set && !ext->is_weak) { | |
55d00c37 | 8382 | pr_warn("extern '%s' (strong): not resolved\n", ext->name); |
166750bc AN |
8383 | return -ESRCH; |
8384 | } else if (!ext->is_set) { | |
55d00c37 | 8385 | pr_debug("extern '%s' (weak): not resolved, defaulting to zero\n", |
166750bc AN |
8386 | ext->name); |
8387 | } | |
8388 | } | |
8389 | ||
8390 | return 0; | |
8391 | } | |
8392 | ||
8d1608d7 KFL |
8393 | static void bpf_map_prepare_vdata(const struct bpf_map *map) |
8394 | { | |
8395 | struct bpf_struct_ops *st_ops; | |
8396 | __u32 i; | |
8397 | ||
8398 | st_ops = map->st_ops; | |
8399 | for (i = 0; i < btf_vlen(st_ops->type); i++) { | |
8400 | struct bpf_program *prog = st_ops->progs[i]; | |
8401 | void *kern_data; | |
8402 | int prog_fd; | |
8403 | ||
8404 | if (!prog) | |
8405 | continue; | |
8406 | ||
8407 | prog_fd = bpf_program__fd(prog); | |
8408 | kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i]; | |
8409 | *(unsigned long *)kern_data = prog_fd; | |
8410 | } | |
8411 | } | |
8412 | ||
8413 | static int bpf_object_prepare_struct_ops(struct bpf_object *obj) | |
8414 | { | |
8db05261 | 8415 | struct bpf_map *map; |
8d1608d7 KFL |
8416 | int i; |
8417 | ||
8db05261 EZ |
8418 | for (i = 0; i < obj->nr_maps; i++) { |
8419 | map = &obj->maps[i]; | |
8420 | ||
8421 | if (!bpf_map__is_struct_ops(map)) | |
8422 | continue; | |
8423 | ||
8424 | if (!map->autocreate) | |
8425 | continue; | |
8426 | ||
8427 | bpf_map_prepare_vdata(map); | |
8428 | } | |
8d1608d7 KFL |
8429 | |
8430 | return 0; | |
8431 | } | |
8432 | ||
e7b924ca | 8433 | static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path) |
52d3352e | 8434 | { |
ec6d5f47 | 8435 | int err, i; |
6371ca3b | 8436 | |
52d3352e | 8437 | if (!obj) |
e9fc3ce9 | 8438 | return libbpf_err(-EINVAL); |
52d3352e WN |
8439 | |
8440 | if (obj->loaded) { | |
d9297581 | 8441 | pr_warn("object '%s': load can't be attempted twice\n", obj->name); |
e9fc3ce9 | 8442 | return libbpf_err(-EINVAL); |
52d3352e WN |
8443 | } |
8444 | ||
67234743 | 8445 | if (obj->gen_loader) |
be315829 | 8446 | bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps); |
67234743 | 8447 | |
6b434b61 AN |
8448 | err = bpf_object_prepare_token(obj); |
8449 | err = err ? : bpf_object__probe_loading(obj); | |
fe62de31 | 8450 | err = err ? : bpf_object__load_vmlinux_btf(obj, false); |
8601fd42 | 8451 | err = err ? : bpf_object__resolve_externs(obj, obj->kconfig); |
0d13bfce | 8452 | err = err ? : bpf_object__sanitize_maps(obj); |
590a0088 | 8453 | err = err ? : bpf_object__init_kern_struct_ops_maps(obj); |
fe9d049c | 8454 | err = err ? : bpf_object_adjust_struct_ops_autoload(obj); |
e7b924ca | 8455 | err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path); |
1004742d | 8456 | err = err ? : bpf_object__sanitize_and_load_btf(obj); |
dac645b9 | 8457 | err = err ? : bpf_object__create_maps(obj); |
e7b924ca | 8458 | err = err ? : bpf_object__load_progs(obj, extra_log_level); |
341ac5ff | 8459 | err = err ? : bpf_object_init_prog_arrays(obj); |
8d1608d7 | 8460 | err = err ? : bpf_object_prepare_struct_ops(obj); |
a6ed02ca | 8461 | |
67234743 AS |
8462 | if (obj->gen_loader) { |
8463 | /* reset FDs */ | |
4729445b KKD |
8464 | if (obj->btf) |
8465 | btf__set_fd(obj->btf, -1); | |
67234743 | 8466 | if (!err) |
ba05fd36 | 8467 | err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps); |
67234743 AS |
8468 | } |
8469 | ||
9dbe6015 KKD |
8470 | /* clean up fd_array */ |
8471 | zfree(&obj->fd_array); | |
8472 | ||
4f33a53d AN |
8473 | /* clean up module BTFs */ |
8474 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
91abb4a6 | 8475 | close(obj->btf_modules[i].fd); |
4f33a53d AN |
8476 | btf__free(obj->btf_modules[i].btf); |
8477 | free(obj->btf_modules[i].name); | |
8478 | } | |
8479 | free(obj->btf_modules); | |
8480 | ||
8481 | /* clean up vmlinux BTF */ | |
a6ed02ca KS |
8482 | btf__free(obj->btf_vmlinux); |
8483 | obj->btf_vmlinux = NULL; | |
8484 | ||
d9297581 AN |
8485 | obj->loaded = true; /* doesn't matter if successfully or not */ |
8486 | ||
0d13bfce AN |
8487 | if (err) |
8488 | goto out; | |
52d3352e WN |
8489 | |
8490 | return 0; | |
8491 | out: | |
ec6d5f47 THJ |
8492 | /* unpin any maps that were auto-pinned during load */ |
8493 | for (i = 0; i < obj->nr_maps; i++) | |
8494 | if (obj->maps[i].pinned && !obj->maps[i].reused) | |
8495 | bpf_map__unpin(&obj->maps[i], NULL); | |
8496 | ||
4a404a7e | 8497 | bpf_object_unload(obj); |
be18010e | 8498 | pr_warn("failed to load object '%s'\n", obj->path); |
e9fc3ce9 | 8499 | return libbpf_err(err); |
52d3352e WN |
8500 | } |
8501 | ||
e7b924ca AN |
8502 | int bpf_object__load(struct bpf_object *obj) |
8503 | { | |
8504 | return bpf_object_load(obj, 0, NULL); | |
60276f98 QM |
8505 | } |
8506 | ||
196f8487 THJ |
8507 | static int make_parent_dir(const char *path) |
8508 | { | |
8509 | char *cp, errmsg[STRERR_BUFSIZE]; | |
8510 | char *dname, *dir; | |
8511 | int err = 0; | |
8512 | ||
8513 | dname = strdup(path); | |
8514 | if (dname == NULL) | |
8515 | return -ENOMEM; | |
8516 | ||
8517 | dir = dirname(dname); | |
8518 | if (mkdir(dir, 0700) && errno != EEXIST) | |
8519 | err = -errno; | |
8520 | ||
8521 | free(dname); | |
8522 | if (err) { | |
8523 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
8524 | pr_warn("failed to mkdir %s: %s\n", path, cp); | |
8525 | } | |
8526 | return err; | |
8527 | } | |
8528 | ||
f367540c JS |
8529 | static int check_path(const char *path) |
8530 | { | |
1ce6a9fc | 8531 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
8532 | struct statfs st_fs; |
8533 | char *dname, *dir; | |
8534 | int err = 0; | |
8535 | ||
8536 | if (path == NULL) | |
8537 | return -EINVAL; | |
8538 | ||
8539 | dname = strdup(path); | |
8540 | if (dname == NULL) | |
8541 | return -ENOMEM; | |
8542 | ||
8543 | dir = dirname(dname); | |
8544 | if (statfs(dir, &st_fs)) { | |
24d6a808 | 8545 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
be18010e | 8546 | pr_warn("failed to statfs %s: %s\n", dir, cp); |
f367540c JS |
8547 | err = -errno; |
8548 | } | |
8549 | free(dname); | |
8550 | ||
8551 | if (!err && st_fs.f_type != BPF_FS_MAGIC) { | |
be18010e | 8552 | pr_warn("specified path %s is not on BPF FS\n", path); |
f367540c JS |
8553 | err = -EINVAL; |
8554 | } | |
8555 | ||
8556 | return err; | |
8557 | } | |
8558 | ||
cf90a20d | 8559 | int bpf_program__pin(struct bpf_program *prog, const char *path) |
f367540c | 8560 | { |
1ce6a9fc | 8561 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
8562 | int err; |
8563 | ||
cf90a20d AN |
8564 | if (prog->fd < 0) { |
8565 | pr_warn("prog '%s': can't pin program that wasn't loaded\n", prog->name); | |
8566 | return libbpf_err(-EINVAL); | |
8567 | } | |
8568 | ||
196f8487 THJ |
8569 | err = make_parent_dir(path); |
8570 | if (err) | |
e9fc3ce9 | 8571 | return libbpf_err(err); |
196f8487 | 8572 | |
f367540c JS |
8573 | err = check_path(path); |
8574 | if (err) | |
e9fc3ce9 | 8575 | return libbpf_err(err); |
f367540c | 8576 | |
cf90a20d | 8577 | if (bpf_obj_pin(prog->fd, path)) { |
23ab656b THJ |
8578 | err = -errno; |
8579 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
cf90a20d | 8580 | pr_warn("prog '%s': failed to pin at '%s': %s\n", prog->name, path, cp); |
e9fc3ce9 | 8581 | return libbpf_err(err); |
f367540c | 8582 | } |
f367540c | 8583 | |
cf90a20d | 8584 | pr_debug("prog '%s': pinned at '%s'\n", prog->name, path); |
f367540c JS |
8585 | return 0; |
8586 | } | |
8587 | ||
cf90a20d | 8588 | int bpf_program__unpin(struct bpf_program *prog, const char *path) |
0c19a9fb SF |
8589 | { |
8590 | int err; | |
8591 | ||
cf90a20d AN |
8592 | if (prog->fd < 0) { |
8593 | pr_warn("prog '%s': can't unpin program that wasn't loaded\n", prog->name); | |
e9fc3ce9 | 8594 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
8595 | } |
8596 | ||
f367540c JS |
8597 | err = check_path(path); |
8598 | if (err) | |
e9fc3ce9 | 8599 | return libbpf_err(err); |
f367540c | 8600 | |
cf90a20d | 8601 | err = unlink(path); |
0c19a9fb | 8602 | if (err) |
e9fc3ce9 | 8603 | return libbpf_err(-errno); |
0c19a9fb | 8604 | |
cf90a20d | 8605 | pr_debug("prog '%s': unpinned from '%s'\n", prog->name, path); |
f367540c JS |
8606 | return 0; |
8607 | } | |
8608 | ||
b6989f35 JS |
8609 | int bpf_map__pin(struct bpf_map *map, const char *path) |
8610 | { | |
1ce6a9fc | 8611 | char *cp, errmsg[STRERR_BUFSIZE]; |
b6989f35 JS |
8612 | int err; |
8613 | ||
b6989f35 | 8614 | if (map == NULL) { |
be18010e | 8615 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 8616 | return libbpf_err(-EINVAL); |
b6989f35 JS |
8617 | } |
8618 | ||
7b30c296 MY |
8619 | if (map->fd < 0) { |
8620 | pr_warn("map '%s': can't pin BPF map without FD (was it created?)\n", map->name); | |
8621 | return libbpf_err(-EINVAL); | |
8622 | } | |
8623 | ||
4580b25f THJ |
8624 | if (map->pin_path) { |
8625 | if (path && strcmp(path, map->pin_path)) { | |
8626 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
8627 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 8628 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8629 | } else if (map->pinned) { |
8630 | pr_debug("map '%s' already pinned at '%s'; not re-pinning\n", | |
8631 | bpf_map__name(map), map->pin_path); | |
8632 | return 0; | |
8633 | } | |
8634 | } else { | |
8635 | if (!path) { | |
8636 | pr_warn("missing a path to pin map '%s' at\n", | |
8637 | bpf_map__name(map)); | |
e9fc3ce9 | 8638 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8639 | } else if (map->pinned) { |
8640 | pr_warn("map '%s' already pinned\n", bpf_map__name(map)); | |
e9fc3ce9 | 8641 | return libbpf_err(-EEXIST); |
4580b25f THJ |
8642 | } |
8643 | ||
8644 | map->pin_path = strdup(path); | |
8645 | if (!map->pin_path) { | |
8646 | err = -errno; | |
8647 | goto out_err; | |
8648 | } | |
b6989f35 JS |
8649 | } |
8650 | ||
196f8487 THJ |
8651 | err = make_parent_dir(map->pin_path); |
8652 | if (err) | |
e9fc3ce9 | 8653 | return libbpf_err(err); |
196f8487 | 8654 | |
4580b25f THJ |
8655 | err = check_path(map->pin_path); |
8656 | if (err) | |
e9fc3ce9 | 8657 | return libbpf_err(err); |
4580b25f THJ |
8658 | |
8659 | if (bpf_obj_pin(map->fd, map->pin_path)) { | |
8660 | err = -errno; | |
8661 | goto out_err; | |
8662 | } | |
8663 | ||
8664 | map->pinned = true; | |
8665 | pr_debug("pinned map '%s'\n", map->pin_path); | |
0c19a9fb | 8666 | |
b6989f35 | 8667 | return 0; |
4580b25f THJ |
8668 | |
8669 | out_err: | |
8670 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
8671 | pr_warn("failed to pin map: %s\n", cp); | |
e9fc3ce9 | 8672 | return libbpf_err(err); |
b6989f35 JS |
8673 | } |
8674 | ||
0c19a9fb SF |
8675 | int bpf_map__unpin(struct bpf_map *map, const char *path) |
8676 | { | |
8677 | int err; | |
8678 | ||
0c19a9fb | 8679 | if (map == NULL) { |
be18010e | 8680 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 8681 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
8682 | } |
8683 | ||
4580b25f THJ |
8684 | if (map->pin_path) { |
8685 | if (path && strcmp(path, map->pin_path)) { | |
8686 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
8687 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 8688 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8689 | } |
8690 | path = map->pin_path; | |
8691 | } else if (!path) { | |
8692 | pr_warn("no path to unpin map '%s' from\n", | |
8693 | bpf_map__name(map)); | |
e9fc3ce9 | 8694 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8695 | } |
8696 | ||
8697 | err = check_path(path); | |
8698 | if (err) | |
e9fc3ce9 | 8699 | return libbpf_err(err); |
4580b25f | 8700 | |
0c19a9fb SF |
8701 | err = unlink(path); |
8702 | if (err != 0) | |
e9fc3ce9 | 8703 | return libbpf_err(-errno); |
4580b25f THJ |
8704 | |
8705 | map->pinned = false; | |
8706 | pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path); | |
0c19a9fb SF |
8707 | |
8708 | return 0; | |
8709 | } | |
8710 | ||
4580b25f THJ |
8711 | int bpf_map__set_pin_path(struct bpf_map *map, const char *path) |
8712 | { | |
8713 | char *new = NULL; | |
8714 | ||
8715 | if (path) { | |
8716 | new = strdup(path); | |
8717 | if (!new) | |
e9fc3ce9 | 8718 | return libbpf_err(-errno); |
4580b25f THJ |
8719 | } |
8720 | ||
8721 | free(map->pin_path); | |
8722 | map->pin_path = new; | |
8723 | return 0; | |
8724 | } | |
8725 | ||
20eccf29 AN |
8726 | __alias(bpf_map__pin_path) |
8727 | const char *bpf_map__get_pin_path(const struct bpf_map *map); | |
4580b25f | 8728 | |
e244d34d EL |
8729 | const char *bpf_map__pin_path(const struct bpf_map *map) |
8730 | { | |
8731 | return map->pin_path; | |
8732 | } | |
8733 | ||
4580b25f THJ |
8734 | bool bpf_map__is_pinned(const struct bpf_map *map) |
8735 | { | |
8736 | return map->pinned; | |
8737 | } | |
8738 | ||
9cf309c5 THJ |
8739 | static void sanitize_pin_path(char *s) |
8740 | { | |
8741 | /* bpffs disallows periods in path names */ | |
8742 | while (*s) { | |
8743 | if (*s == '.') | |
8744 | *s = '_'; | |
8745 | s++; | |
8746 | } | |
8747 | } | |
8748 | ||
0c19a9fb | 8749 | int bpf_object__pin_maps(struct bpf_object *obj, const char *path) |
d5148d85 | 8750 | { |
d5148d85 JS |
8751 | struct bpf_map *map; |
8752 | int err; | |
8753 | ||
8754 | if (!obj) | |
e9fc3ce9 | 8755 | return libbpf_err(-ENOENT); |
d5148d85 JS |
8756 | |
8757 | if (!obj->loaded) { | |
be18010e | 8758 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 8759 | return libbpf_err(-ENOENT); |
d5148d85 JS |
8760 | } |
8761 | ||
f74a53d9 | 8762 | bpf_object__for_each_map(map, obj) { |
4580b25f | 8763 | char *pin_path = NULL; |
0c19a9fb | 8764 | char buf[PATH_MAX]; |
0c19a9fb | 8765 | |
ec41817b | 8766 | if (!map->autocreate) |
229fae38 SC |
8767 | continue; |
8768 | ||
4580b25f | 8769 | if (path) { |
e588c116 WY |
8770 | err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map)); |
8771 | if (err) | |
4580b25f | 8772 | goto err_unpin_maps; |
9cf309c5 | 8773 | sanitize_pin_path(buf); |
4580b25f THJ |
8774 | pin_path = buf; |
8775 | } else if (!map->pin_path) { | |
8776 | continue; | |
0c19a9fb SF |
8777 | } |
8778 | ||
4580b25f | 8779 | err = bpf_map__pin(map, pin_path); |
0c19a9fb SF |
8780 | if (err) |
8781 | goto err_unpin_maps; | |
8782 | } | |
8783 | ||
8784 | return 0; | |
8785 | ||
8786 | err_unpin_maps: | |
bcc40fc0 | 8787 | while ((map = bpf_object__prev_map(obj, map))) { |
4580b25f | 8788 | if (!map->pin_path) |
0c19a9fb SF |
8789 | continue; |
8790 | ||
4580b25f | 8791 | bpf_map__unpin(map, NULL); |
0c19a9fb SF |
8792 | } |
8793 | ||
e9fc3ce9 | 8794 | return libbpf_err(err); |
0c19a9fb SF |
8795 | } |
8796 | ||
8797 | int bpf_object__unpin_maps(struct bpf_object *obj, const char *path) | |
8798 | { | |
8799 | struct bpf_map *map; | |
8800 | int err; | |
8801 | ||
8802 | if (!obj) | |
e9fc3ce9 | 8803 | return libbpf_err(-ENOENT); |
0c19a9fb | 8804 | |
f74a53d9 | 8805 | bpf_object__for_each_map(map, obj) { |
4580b25f | 8806 | char *pin_path = NULL; |
d5148d85 | 8807 | char buf[PATH_MAX]; |
d5148d85 | 8808 | |
4580b25f | 8809 | if (path) { |
e588c116 WY |
8810 | err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map)); |
8811 | if (err) | |
8812 | return libbpf_err(err); | |
9cf309c5 | 8813 | sanitize_pin_path(buf); |
4580b25f THJ |
8814 | pin_path = buf; |
8815 | } else if (!map->pin_path) { | |
8816 | continue; | |
8817 | } | |
d5148d85 | 8818 | |
4580b25f | 8819 | err = bpf_map__unpin(map, pin_path); |
d5148d85 | 8820 | if (err) |
e9fc3ce9 | 8821 | return libbpf_err(err); |
d5148d85 JS |
8822 | } |
8823 | ||
0c19a9fb SF |
8824 | return 0; |
8825 | } | |
8826 | ||
8827 | int bpf_object__pin_programs(struct bpf_object *obj, const char *path) | |
8828 | { | |
8829 | struct bpf_program *prog; | |
e588c116 | 8830 | char buf[PATH_MAX]; |
0c19a9fb SF |
8831 | int err; |
8832 | ||
8833 | if (!obj) | |
e9fc3ce9 | 8834 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
8835 | |
8836 | if (!obj->loaded) { | |
be18010e | 8837 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 8838 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
8839 | } |
8840 | ||
0c19a9fb | 8841 | bpf_object__for_each_program(prog, obj) { |
e588c116 WY |
8842 | err = pathname_concat(buf, sizeof(buf), path, prog->name); |
8843 | if (err) | |
0c19a9fb | 8844 | goto err_unpin_programs; |
0c19a9fb SF |
8845 | |
8846 | err = bpf_program__pin(prog, buf); | |
8847 | if (err) | |
8848 | goto err_unpin_programs; | |
8849 | } | |
8850 | ||
8851 | return 0; | |
8852 | ||
8853 | err_unpin_programs: | |
bcc40fc0 | 8854 | while ((prog = bpf_object__prev_program(obj, prog))) { |
e588c116 | 8855 | if (pathname_concat(buf, sizeof(buf), path, prog->name)) |
0c19a9fb SF |
8856 | continue; |
8857 | ||
8858 | bpf_program__unpin(prog, buf); | |
8859 | } | |
8860 | ||
e9fc3ce9 | 8861 | return libbpf_err(err); |
0c19a9fb SF |
8862 | } |
8863 | ||
8864 | int bpf_object__unpin_programs(struct bpf_object *obj, const char *path) | |
8865 | { | |
8866 | struct bpf_program *prog; | |
8867 | int err; | |
8868 | ||
8869 | if (!obj) | |
e9fc3ce9 | 8870 | return libbpf_err(-ENOENT); |
0c19a9fb | 8871 | |
d5148d85 JS |
8872 | bpf_object__for_each_program(prog, obj) { |
8873 | char buf[PATH_MAX]; | |
d5148d85 | 8874 | |
e588c116 WY |
8875 | err = pathname_concat(buf, sizeof(buf), path, prog->name); |
8876 | if (err) | |
8877 | return libbpf_err(err); | |
d5148d85 | 8878 | |
0c19a9fb | 8879 | err = bpf_program__unpin(prog, buf); |
d5148d85 | 8880 | if (err) |
e9fc3ce9 | 8881 | return libbpf_err(err); |
d5148d85 JS |
8882 | } |
8883 | ||
8884 | return 0; | |
8885 | } | |
8886 | ||
0c19a9fb SF |
8887 | int bpf_object__pin(struct bpf_object *obj, const char *path) |
8888 | { | |
8889 | int err; | |
8890 | ||
8891 | err = bpf_object__pin_maps(obj, path); | |
8892 | if (err) | |
e9fc3ce9 | 8893 | return libbpf_err(err); |
0c19a9fb SF |
8894 | |
8895 | err = bpf_object__pin_programs(obj, path); | |
8896 | if (err) { | |
8897 | bpf_object__unpin_maps(obj, path); | |
e9fc3ce9 | 8898 | return libbpf_err(err); |
0c19a9fb SF |
8899 | } |
8900 | ||
8901 | return 0; | |
8902 | } | |
8903 | ||
068ca522 DX |
8904 | int bpf_object__unpin(struct bpf_object *obj, const char *path) |
8905 | { | |
8906 | int err; | |
8907 | ||
8908 | err = bpf_object__unpin_programs(obj, path); | |
8909 | if (err) | |
8910 | return libbpf_err(err); | |
8911 | ||
8912 | err = bpf_object__unpin_maps(obj, path); | |
8913 | if (err) | |
8914 | return libbpf_err(err); | |
8915 | ||
8916 | return 0; | |
8917 | } | |
8918 | ||
2d39d7c5 AN |
8919 | static void bpf_map__destroy(struct bpf_map *map) |
8920 | { | |
646f02ff AN |
8921 | if (map->inner_map) { |
8922 | bpf_map__destroy(map->inner_map); | |
8923 | zfree(&map->inner_map); | |
8924 | } | |
8925 | ||
8926 | zfree(&map->init_slots); | |
8927 | map->init_slots_sz = 0; | |
8928 | ||
2e7ba4f8 AN |
8929 | if (map->mmaped && map->mmaped != map->obj->arena_data) |
8930 | munmap(map->mmaped, bpf_map_mmap_sz(map)); | |
8931 | map->mmaped = NULL; | |
2d39d7c5 AN |
8932 | |
8933 | if (map->st_ops) { | |
8934 | zfree(&map->st_ops->data); | |
8935 | zfree(&map->st_ops->progs); | |
8936 | zfree(&map->st_ops->kern_func_off); | |
8937 | zfree(&map->st_ops); | |
8938 | } | |
8939 | ||
8940 | zfree(&map->name); | |
aed65917 | 8941 | zfree(&map->real_name); |
2d39d7c5 AN |
8942 | zfree(&map->pin_path); |
8943 | ||
8944 | if (map->fd >= 0) | |
8945 | zclose(map->fd); | |
8946 | } | |
8947 | ||
1a5e3fb1 WN |
8948 | void bpf_object__close(struct bpf_object *obj) |
8949 | { | |
a5b8bd47 WN |
8950 | size_t i; |
8951 | ||
50450fc7 | 8952 | if (IS_ERR_OR_NULL(obj)) |
1a5e3fb1 WN |
8953 | return; |
8954 | ||
2e4913e0 AN |
8955 | usdt_manager_free(obj->usdt_man); |
8956 | obj->usdt_man = NULL; | |
8957 | ||
67234743 | 8958 | bpf_gen__free(obj->gen_loader); |
1a5e3fb1 | 8959 | bpf_object__elf_finish(obj); |
4a404a7e | 8960 | bpf_object_unload(obj); |
8a138aed | 8961 | btf__free(obj->btf); |
29d67fde | 8962 | btf__free(obj->btf_vmlinux); |
2993e051 | 8963 | btf_ext__free(obj->btf_ext); |
1a5e3fb1 | 8964 | |
2d39d7c5 AN |
8965 | for (i = 0; i < obj->nr_maps; i++) |
8966 | bpf_map__destroy(&obj->maps[i]); | |
d859900c | 8967 | |
1373ff59 | 8968 | zfree(&obj->btf_custom_path); |
8601fd42 | 8969 | zfree(&obj->kconfig); |
5964a223 DM |
8970 | |
8971 | for (i = 0; i < obj->nr_extern; i++) | |
8972 | zfree(&obj->externs[i].essent_name); | |
8973 | ||
166750bc AN |
8974 | zfree(&obj->externs); |
8975 | obj->nr_extern = 0; | |
8976 | ||
9d759a9b WN |
8977 | zfree(&obj->maps); |
8978 | obj->nr_maps = 0; | |
a5b8bd47 WN |
8979 | |
8980 | if (obj->programs && obj->nr_programs) { | |
8981 | for (i = 0; i < obj->nr_programs; i++) | |
8982 | bpf_program__exit(&obj->programs[i]); | |
8983 | } | |
8984 | zfree(&obj->programs); | |
8985 | ||
6b434b61 AN |
8986 | zfree(&obj->feat_cache); |
8987 | zfree(&obj->token_path); | |
8988 | if (obj->token_fd > 0) | |
8989 | close(obj->token_fd); | |
8990 | ||
2e7ba4f8 AN |
8991 | zfree(&obj->arena_data); |
8992 | ||
1a5e3fb1 WN |
8993 | free(obj); |
8994 | } | |
aa9b1ac3 | 8995 | |
a324aae3 | 8996 | const char *bpf_object__name(const struct bpf_object *obj) |
acf860ae | 8997 | { |
e9fc3ce9 | 8998 | return obj ? obj->name : libbpf_err_ptr(-EINVAL); |
acf860ae WN |
8999 | } |
9000 | ||
a324aae3 | 9001 | unsigned int bpf_object__kversion(const struct bpf_object *obj) |
45825d8a | 9002 | { |
a7fe0450 | 9003 | return obj ? obj->kern_version : 0; |
45825d8a WN |
9004 | } |
9005 | ||
a324aae3 | 9006 | struct btf *bpf_object__btf(const struct bpf_object *obj) |
789f6bab AI |
9007 | { |
9008 | return obj ? obj->btf : NULL; | |
9009 | } | |
9010 | ||
8a138aed MKL |
9011 | int bpf_object__btf_fd(const struct bpf_object *obj) |
9012 | { | |
9013 | return obj->btf ? btf__fd(obj->btf) : -1; | |
9014 | } | |
9015 | ||
155f556d RDT |
9016 | int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version) |
9017 | { | |
9018 | if (obj->loaded) | |
e9fc3ce9 | 9019 | return libbpf_err(-EINVAL); |
155f556d RDT |
9020 | |
9021 | obj->kern_version = kern_version; | |
9022 | ||
9023 | return 0; | |
9024 | } | |
9025 | ||
67234743 AS |
9026 | int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts) |
9027 | { | |
9028 | struct bpf_gen *gen; | |
9029 | ||
9030 | if (!opts) | |
9031 | return -EFAULT; | |
9032 | if (!OPTS_VALID(opts, gen_loader_opts)) | |
9033 | return -EINVAL; | |
9034 | gen = calloc(sizeof(*gen), 1); | |
9035 | if (!gen) | |
9036 | return -ENOMEM; | |
9037 | gen->opts = opts; | |
9038 | obj->gen_loader = gen; | |
9039 | return 0; | |
9040 | } | |
9041 | ||
eac7d845 | 9042 | static struct bpf_program * |
a324aae3 AN |
9043 | __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj, |
9044 | bool forward) | |
aa9b1ac3 | 9045 | { |
a83d6e76 | 9046 | size_t nr_programs = obj->nr_programs; |
0c19a9fb | 9047 | ssize_t idx; |
aa9b1ac3 | 9048 | |
a83d6e76 | 9049 | if (!nr_programs) |
aa9b1ac3 | 9050 | return NULL; |
aa9b1ac3 | 9051 | |
a83d6e76 MKL |
9052 | if (!p) |
9053 | /* Iter from the beginning */ | |
9054 | return forward ? &obj->programs[0] : | |
9055 | &obj->programs[nr_programs - 1]; | |
9056 | ||
0c19a9fb | 9057 | if (p->obj != obj) { |
be18010e | 9058 | pr_warn("error: program handler doesn't match object\n"); |
e9fc3ce9 | 9059 | return errno = EINVAL, NULL; |
aa9b1ac3 WN |
9060 | } |
9061 | ||
a83d6e76 | 9062 | idx = (p - obj->programs) + (forward ? 1 : -1); |
0c19a9fb | 9063 | if (idx >= obj->nr_programs || idx < 0) |
aa9b1ac3 WN |
9064 | return NULL; |
9065 | return &obj->programs[idx]; | |
9066 | } | |
9067 | ||
2088a3a7 HC |
9068 | struct bpf_program * |
9069 | bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev) | |
eac7d845 JK |
9070 | { |
9071 | struct bpf_program *prog = prev; | |
9072 | ||
9073 | do { | |
a83d6e76 | 9074 | prog = __bpf_program__iter(prog, obj, true); |
c3c55696 | 9075 | } while (prog && prog_is_subprog(obj, prog)); |
0c19a9fb SF |
9076 | |
9077 | return prog; | |
9078 | } | |
9079 | ||
2088a3a7 HC |
9080 | struct bpf_program * |
9081 | bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next) | |
0c19a9fb SF |
9082 | { |
9083 | struct bpf_program *prog = next; | |
9084 | ||
0c19a9fb | 9085 | do { |
a83d6e76 | 9086 | prog = __bpf_program__iter(prog, obj, false); |
c3c55696 | 9087 | } while (prog && prog_is_subprog(obj, prog)); |
eac7d845 JK |
9088 | |
9089 | return prog; | |
9090 | } | |
9091 | ||
9aba3613 JK |
9092 | void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex) |
9093 | { | |
9094 | prog->prog_ifindex = ifindex; | |
9095 | } | |
9096 | ||
01af3bf0 AN |
9097 | const char *bpf_program__name(const struct bpf_program *prog) |
9098 | { | |
9099 | return prog->name; | |
9100 | } | |
9101 | ||
52109584 AN |
9102 | const char *bpf_program__section_name(const struct bpf_program *prog) |
9103 | { | |
9104 | return prog->sec_name; | |
9105 | } | |
9106 | ||
d9297581 AN |
9107 | bool bpf_program__autoload(const struct bpf_program *prog) |
9108 | { | |
a3820c48 | 9109 | return prog->autoload; |
d9297581 AN |
9110 | } |
9111 | ||
9112 | int bpf_program__set_autoload(struct bpf_program *prog, bool autoload) | |
9113 | { | |
9114 | if (prog->obj->loaded) | |
e9fc3ce9 | 9115 | return libbpf_err(-EINVAL); |
d9297581 | 9116 | |
a3820c48 | 9117 | prog->autoload = autoload; |
d9297581 AN |
9118 | return 0; |
9119 | } | |
9120 | ||
43cb8cba HL |
9121 | bool bpf_program__autoattach(const struct bpf_program *prog) |
9122 | { | |
9123 | return prog->autoattach; | |
9124 | } | |
9125 | ||
9126 | void bpf_program__set_autoattach(struct bpf_program *prog, bool autoattach) | |
9127 | { | |
9128 | prog->autoattach = autoattach; | |
9129 | } | |
9130 | ||
65a7fa2e AN |
9131 | const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog) |
9132 | { | |
9133 | return prog->insns; | |
9134 | } | |
9135 | ||
9136 | size_t bpf_program__insn_cnt(const struct bpf_program *prog) | |
9137 | { | |
9138 | return prog->insns_cnt; | |
9139 | } | |
9140 | ||
b63b3c49 JO |
9141 | int bpf_program__set_insns(struct bpf_program *prog, |
9142 | struct bpf_insn *new_insns, size_t new_insn_cnt) | |
9143 | { | |
9144 | struct bpf_insn *insns; | |
9145 | ||
9146 | if (prog->obj->loaded) | |
9147 | return -EBUSY; | |
9148 | ||
9149 | insns = libbpf_reallocarray(prog->insns, new_insn_cnt, sizeof(*insns)); | |
8a0260db AN |
9150 | /* NULL is a valid return from reallocarray if the new count is zero */ |
9151 | if (!insns && new_insn_cnt) { | |
b63b3c49 JO |
9152 | pr_warn("prog '%s': failed to realloc prog code\n", prog->name); |
9153 | return -ENOMEM; | |
9154 | } | |
9155 | memcpy(insns, new_insns, new_insn_cnt * sizeof(*insns)); | |
9156 | ||
9157 | prog->insns = insns; | |
9158 | prog->insns_cnt = new_insn_cnt; | |
9159 | return 0; | |
9160 | } | |
9161 | ||
cf90a20d | 9162 | int bpf_program__fd(const struct bpf_program *prog) |
b580563e | 9163 | { |
1e960043 | 9164 | if (!prog) |
e9fc3ce9 | 9165 | return libbpf_err(-EINVAL); |
1e960043 | 9166 | |
cf90a20d | 9167 | if (prog->fd < 0) |
e9fc3ce9 | 9168 | return libbpf_err(-ENOENT); |
b580563e | 9169 | |
cf90a20d | 9170 | return prog->fd; |
aa9b1ac3 | 9171 | } |
9d759a9b | 9172 | |
20eccf29 AN |
9173 | __alias(bpf_program__type) |
9174 | enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog); | |
9175 | ||
9176 | enum bpf_prog_type bpf_program__type(const struct bpf_program *prog) | |
f1eead9e AN |
9177 | { |
9178 | return prog->type; | |
9179 | } | |
9180 | ||
c628747c AN |
9181 | static size_t custom_sec_def_cnt; |
9182 | static struct bpf_sec_def *custom_sec_defs; | |
9183 | static struct bpf_sec_def custom_fallback_def; | |
9184 | static bool has_custom_fallback_def; | |
9185 | static int last_custom_sec_def_handler_id; | |
9186 | ||
93442f13 | 9187 | int bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type) |
5f44e4c8 | 9188 | { |
93442f13 GS |
9189 | if (prog->obj->loaded) |
9190 | return libbpf_err(-EBUSY); | |
9191 | ||
c628747c AN |
9192 | /* if type is not changed, do nothing */ |
9193 | if (prog->type == type) | |
9194 | return 0; | |
9195 | ||
5f44e4c8 | 9196 | prog->type = type; |
c628747c AN |
9197 | |
9198 | /* If a program type was changed, we need to reset associated SEC() | |
9199 | * handler, as it will be invalid now. The only exception is a generic | |
9200 | * fallback handler, which by definition is program type-agnostic and | |
9201 | * is a catch-all custom handler, optionally set by the application, | |
9202 | * so should be able to handle any type of BPF program. | |
9203 | */ | |
9204 | if (prog->sec_def != &custom_fallback_def) | |
9205 | prog->sec_def = NULL; | |
93442f13 | 9206 | return 0; |
5f44e4c8 WN |
9207 | } |
9208 | ||
20eccf29 AN |
9209 | __alias(bpf_program__expected_attach_type) |
9210 | enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog); | |
9211 | ||
9212 | enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog) | |
f1eead9e AN |
9213 | { |
9214 | return prog->expected_attach_type; | |
9215 | } | |
9216 | ||
93442f13 | 9217 | int bpf_program__set_expected_attach_type(struct bpf_program *prog, |
16962b24 | 9218 | enum bpf_attach_type type) |
d7be143b | 9219 | { |
93442f13 GS |
9220 | if (prog->obj->loaded) |
9221 | return libbpf_err(-EBUSY); | |
9222 | ||
d7be143b | 9223 | prog->expected_attach_type = type; |
93442f13 | 9224 | return 0; |
d7be143b AI |
9225 | } |
9226 | ||
a6ca7158 AN |
9227 | __u32 bpf_program__flags(const struct bpf_program *prog) |
9228 | { | |
9229 | return prog->prog_flags; | |
9230 | } | |
9231 | ||
8cccee9e | 9232 | int bpf_program__set_flags(struct bpf_program *prog, __u32 flags) |
a6ca7158 AN |
9233 | { |
9234 | if (prog->obj->loaded) | |
9235 | return libbpf_err(-EBUSY); | |
9236 | ||
8cccee9e | 9237 | prog->prog_flags = flags; |
a6ca7158 AN |
9238 | return 0; |
9239 | } | |
9240 | ||
dbdd2c7f AN |
9241 | __u32 bpf_program__log_level(const struct bpf_program *prog) |
9242 | { | |
9243 | return prog->log_level; | |
9244 | } | |
9245 | ||
9246 | int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level) | |
9247 | { | |
9248 | if (prog->obj->loaded) | |
9249 | return libbpf_err(-EBUSY); | |
9250 | ||
9251 | prog->log_level = log_level; | |
9252 | return 0; | |
9253 | } | |
9254 | ||
b3ce9079 AN |
9255 | const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size) |
9256 | { | |
9257 | *log_size = prog->log_size; | |
9258 | return prog->log_buf; | |
9259 | } | |
9260 | ||
9261 | int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size) | |
9262 | { | |
9263 | if (log_size && !log_buf) | |
9264 | return -EINVAL; | |
9265 | if (prog->log_size > UINT_MAX) | |
9266 | return -EINVAL; | |
9267 | if (prog->obj->loaded) | |
9268 | return -EBUSY; | |
9269 | ||
9270 | prog->log_buf = log_buf; | |
9271 | prog->log_size = log_size; | |
a6ca7158 AN |
9272 | return 0; |
9273 | } | |
9274 | ||
15ea31fa | 9275 | #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \ |
697f104d | 9276 | .sec = (char *)sec_pfx, \ |
d7a18ea7 | 9277 | .prog_type = BPF_PROG_TYPE_##ptype, \ |
15ea31fa AN |
9278 | .expected_attach_type = atype, \ |
9279 | .cookie = (long)(flags), \ | |
4fa5bcfe | 9280 | .prog_prepare_load_fn = libbpf_prepare_prog_load, \ |
d7a18ea7 AN |
9281 | __VA_ARGS__ \ |
9282 | } | |
9283 | ||
4fa5bcfe | 9284 | static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
39f8dc43 | 9285 | static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
708ac5be | 9286 | static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
2e4913e0 | 9287 | static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
4fa5bcfe AN |
9288 | static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
9289 | static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
9290 | static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
ddc6b049 | 9291 | static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
2ca178f0 | 9292 | static int attach_kprobe_session(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
5bfdd32d | 9293 | static int attach_uprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
4fa5bcfe AN |
9294 | static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
9295 | static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
d7a18ea7 | 9296 | |
d7a18ea7 | 9297 | static const struct bpf_sec_def section_defs[] = { |
450b167f AN |
9298 | SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE), |
9299 | SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE), | |
9300 | SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE), | |
9af8efc4 | 9301 | SEC_DEF("kprobe+", KPROBE, 0, SEC_NONE, attach_kprobe), |
39f8dc43 | 9302 | SEC_DEF("uprobe+", KPROBE, 0, SEC_NONE, attach_uprobe), |
c4cac71f | 9303 | SEC_DEF("uprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe), |
9af8efc4 | 9304 | SEC_DEF("kretprobe+", KPROBE, 0, SEC_NONE, attach_kprobe), |
39f8dc43 | 9305 | SEC_DEF("uretprobe+", KPROBE, 0, SEC_NONE, attach_uprobe), |
c4cac71f | 9306 | SEC_DEF("uretprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe), |
9af8efc4 AN |
9307 | SEC_DEF("kprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi), |
9308 | SEC_DEF("kretprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi), | |
2ca178f0 | 9309 | SEC_DEF("kprobe.session+", KPROBE, BPF_TRACE_KPROBE_SESSION, SEC_NONE, attach_kprobe_session), |
5bfdd32d JO |
9310 | SEC_DEF("uprobe.multi+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_NONE, attach_uprobe_multi), |
9311 | SEC_DEF("uretprobe.multi+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_NONE, attach_uprobe_multi), | |
9312 | SEC_DEF("uprobe.multi.s+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_SLEEPABLE, attach_uprobe_multi), | |
9313 | SEC_DEF("uretprobe.multi.s+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_SLEEPABLE, attach_uprobe_multi), | |
708ac5be AN |
9314 | SEC_DEF("ksyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall), |
9315 | SEC_DEF("kretsyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall), | |
5902da6d JO |
9316 | SEC_DEF("usdt+", KPROBE, 0, SEC_USDT, attach_usdt), |
9317 | SEC_DEF("usdt.s+", KPROBE, 0, SEC_USDT | SEC_SLEEPABLE, attach_usdt), | |
fe20ce3a DB |
9318 | SEC_DEF("tc/ingress", SCHED_CLS, BPF_TCX_INGRESS, SEC_NONE), /* alias for tcx */ |
9319 | SEC_DEF("tc/egress", SCHED_CLS, BPF_TCX_EGRESS, SEC_NONE), /* alias for tcx */ | |
9320 | SEC_DEF("tcx/ingress", SCHED_CLS, BPF_TCX_INGRESS, SEC_NONE), | |
9321 | SEC_DEF("tcx/egress", SCHED_CLS, BPF_TCX_EGRESS, SEC_NONE), | |
9322 | SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE), /* deprecated / legacy, use tcx */ | |
9323 | SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE), /* deprecated / legacy, use tcx */ | |
9324 | SEC_DEF("action", SCHED_ACT, 0, SEC_NONE), /* deprecated / legacy, use tcx */ | |
05c31b4a DB |
9325 | SEC_DEF("netkit/primary", SCHED_CLS, BPF_NETKIT_PRIMARY, SEC_NONE), |
9326 | SEC_DEF("netkit/peer", SCHED_CLS, BPF_NETKIT_PEER, SEC_NONE), | |
9af8efc4 AN |
9327 | SEC_DEF("tracepoint+", TRACEPOINT, 0, SEC_NONE, attach_tp), |
9328 | SEC_DEF("tp+", TRACEPOINT, 0, SEC_NONE, attach_tp), | |
9329 | SEC_DEF("raw_tracepoint+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
9330 | SEC_DEF("raw_tp+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
9331 | SEC_DEF("raw_tracepoint.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), | |
9332 | SEC_DEF("raw_tp.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), | |
cc7d8f2c AN |
9333 | SEC_DEF("tp_btf+", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace), |
9334 | SEC_DEF("fentry+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace), | |
9335 | SEC_DEF("fmod_ret+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace), | |
9336 | SEC_DEF("fexit+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace), | |
9337 | SEC_DEF("fentry.s+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
9338 | SEC_DEF("fmod_ret.s+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
9339 | SEC_DEF("fexit.s+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
9340 | SEC_DEF("freplace+", EXT, 0, SEC_ATTACH_BTF, attach_trace), | |
9341 | SEC_DEF("lsm+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm), | |
9342 | SEC_DEF("lsm.s+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm), | |
bffcf348 | 9343 | SEC_DEF("lsm_cgroup+", LSM, BPF_LSM_CGROUP, SEC_ATTACH_BTF), |
cc7d8f2c AN |
9344 | SEC_DEF("iter+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter), |
9345 | SEC_DEF("iter.s+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter), | |
15ea31fa | 9346 | SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE), |
082c4bfb | 9347 | SEC_DEF("xdp.frags/devmap", XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS), |
4a4d4cee | 9348 | SEC_DEF("xdp/devmap", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE), |
082c4bfb | 9349 | SEC_DEF("xdp.frags/cpumap", XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS), |
4a4d4cee | 9350 | SEC_DEF("xdp/cpumap", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE), |
082c4bfb | 9351 | SEC_DEF("xdp.frags", XDP, BPF_XDP, SEC_XDP_FRAGS), |
450b167f AN |
9352 | SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT), |
9353 | SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE), | |
9354 | SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE), | |
9355 | SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE), | |
9356 | SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE), | |
9357 | SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE), | |
9358 | SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT), | |
9359 | SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT), | |
9360 | SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT), | |
61df5756 | 9361 | SEC_DEF("sk_skb/verdict", SK_SKB, BPF_SK_SKB_VERDICT, SEC_ATTACHABLE_OPT), |
450b167f AN |
9362 | SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE), |
9363 | SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT), | |
9364 | SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT), | |
9365 | SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT), | |
9366 | SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT), | |
9367 | SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT), | |
9368 | SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE), | |
9369 | SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE), | |
9370 | SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE), | |
9371 | SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT), | |
9372 | SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE), | |
9373 | SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE), | |
9374 | SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE), | |
9375 | SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE), | |
9376 | SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE), | |
9377 | SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE), | |
bf90438c | 9378 | SEC_DEF("cgroup/connect_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_CONNECT, SEC_ATTACHABLE), |
450b167f AN |
9379 | SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE), |
9380 | SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE), | |
bf90438c | 9381 | SEC_DEF("cgroup/sendmsg_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_SENDMSG, SEC_ATTACHABLE), |
450b167f AN |
9382 | SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE), |
9383 | SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE), | |
bf90438c | 9384 | SEC_DEF("cgroup/recvmsg_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_RECVMSG, SEC_ATTACHABLE), |
450b167f AN |
9385 | SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE), |
9386 | SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE), | |
bf90438c | 9387 | SEC_DEF("cgroup/getpeername_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_GETPEERNAME, SEC_ATTACHABLE), |
450b167f AN |
9388 | SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE), |
9389 | SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE), | |
bf90438c | 9390 | SEC_DEF("cgroup/getsockname_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_GETSOCKNAME, SEC_ATTACHABLE), |
450b167f AN |
9391 | SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE), |
9392 | SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE), | |
9393 | SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE), | |
9394 | SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT), | |
dd94d45c | 9395 | SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), |
913b2255 | 9396 | SEC_DEF("struct_ops.s+", STRUCT_OPS, 0, SEC_SLEEPABLE), |
450b167f | 9397 | SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE), |
132328e8 | 9398 | SEC_DEF("netfilter", NETFILTER, BPF_NETFILTER, SEC_NONE), |
583c9009 | 9399 | }; |
d7be143b | 9400 | |
697f104d AN |
9401 | int libbpf_register_prog_handler(const char *sec, |
9402 | enum bpf_prog_type prog_type, | |
9403 | enum bpf_attach_type exp_attach_type, | |
9404 | const struct libbpf_prog_handler_opts *opts) | |
d7a18ea7 | 9405 | { |
697f104d | 9406 | struct bpf_sec_def *sec_def; |
d7a18ea7 | 9407 | |
697f104d AN |
9408 | if (!OPTS_VALID(opts, libbpf_prog_handler_opts)) |
9409 | return libbpf_err(-EINVAL); | |
dd94d45c | 9410 | |
697f104d AN |
9411 | if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */ |
9412 | return libbpf_err(-E2BIG); | |
dd94d45c | 9413 | |
697f104d AN |
9414 | if (sec) { |
9415 | sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1, | |
9416 | sizeof(*sec_def)); | |
9417 | if (!sec_def) | |
9418 | return libbpf_err(-ENOMEM); | |
dd94d45c | 9419 | |
697f104d AN |
9420 | custom_sec_defs = sec_def; |
9421 | sec_def = &custom_sec_defs[custom_sec_def_cnt]; | |
9422 | } else { | |
9423 | if (has_custom_fallback_def) | |
9424 | return libbpf_err(-EBUSY); | |
dd94d45c | 9425 | |
697f104d AN |
9426 | sec_def = &custom_fallback_def; |
9427 | } | |
9428 | ||
9429 | sec_def->sec = sec ? strdup(sec) : NULL; | |
9430 | if (sec && !sec_def->sec) | |
9431 | return libbpf_err(-ENOMEM); | |
9432 | ||
9433 | sec_def->prog_type = prog_type; | |
9434 | sec_def->expected_attach_type = exp_attach_type; | |
9435 | sec_def->cookie = OPTS_GET(opts, cookie, 0); | |
9436 | ||
9437 | sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL); | |
9438 | sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL); | |
9439 | sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL); | |
9440 | ||
9441 | sec_def->handler_id = ++last_custom_sec_def_handler_id; | |
9442 | ||
9443 | if (sec) | |
9444 | custom_sec_def_cnt++; | |
9445 | else | |
9446 | has_custom_fallback_def = true; | |
9447 | ||
9448 | return sec_def->handler_id; | |
9449 | } | |
9450 | ||
9451 | int libbpf_unregister_prog_handler(int handler_id) | |
9452 | { | |
9453 | struct bpf_sec_def *sec_defs; | |
9454 | int i; | |
9455 | ||
9456 | if (handler_id <= 0) | |
9457 | return libbpf_err(-EINVAL); | |
9458 | ||
9459 | if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) { | |
9460 | memset(&custom_fallback_def, 0, sizeof(custom_fallback_def)); | |
9461 | has_custom_fallback_def = false; | |
9462 | return 0; | |
9463 | } | |
9464 | ||
9465 | for (i = 0; i < custom_sec_def_cnt; i++) { | |
9466 | if (custom_sec_defs[i].handler_id == handler_id) | |
9467 | break; | |
9468 | } | |
9469 | ||
9470 | if (i == custom_sec_def_cnt) | |
9471 | return libbpf_err(-ENOENT); | |
9472 | ||
9473 | free(custom_sec_defs[i].sec); | |
9474 | for (i = i + 1; i < custom_sec_def_cnt; i++) | |
9475 | custom_sec_defs[i - 1] = custom_sec_defs[i]; | |
9476 | custom_sec_def_cnt--; | |
9477 | ||
9478 | /* try to shrink the array, but it's ok if we couldn't */ | |
9479 | sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs)); | |
8a0260db AN |
9480 | /* if new count is zero, reallocarray can return a valid NULL result; |
9481 | * in this case the previous pointer will be freed, so we *have to* | |
9482 | * reassign old pointer to the new value (even if it's NULL) | |
9483 | */ | |
9484 | if (sec_defs || custom_sec_def_cnt == 0) | |
697f104d AN |
9485 | custom_sec_defs = sec_defs; |
9486 | ||
9487 | return 0; | |
9488 | } | |
9489 | ||
450b167f | 9490 | static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name) |
697f104d AN |
9491 | { |
9492 | size_t len = strlen(sec_def->sec); | |
9493 | ||
9494 | /* "type/" always has to have proper SEC("type/extras") form */ | |
9495 | if (sec_def->sec[len - 1] == '/') { | |
9496 | if (str_has_pfx(sec_name, sec_def->sec)) | |
9497 | return true; | |
9498 | return false; | |
9499 | } | |
9500 | ||
9501 | /* "type+" means it can be either exact SEC("type") or | |
9502 | * well-formed SEC("type/extras") with proper '/' separator | |
9503 | */ | |
9504 | if (sec_def->sec[len - 1] == '+') { | |
9505 | len--; | |
9506 | /* not even a prefix */ | |
9507 | if (strncmp(sec_name, sec_def->sec, len) != 0) | |
9508 | return false; | |
9509 | /* exact match or has '/' separator */ | |
9510 | if (sec_name[len] == '\0' || sec_name[len] == '/') | |
9511 | return true; | |
9512 | return false; | |
9513 | } | |
9514 | ||
697f104d AN |
9515 | return strcmp(sec_name, sec_def->sec) == 0; |
9516 | } | |
9517 | ||
9518 | static const struct bpf_sec_def *find_sec_def(const char *sec_name) | |
9519 | { | |
9520 | const struct bpf_sec_def *sec_def; | |
9521 | int i, n; | |
697f104d AN |
9522 | |
9523 | n = custom_sec_def_cnt; | |
9524 | for (i = 0; i < n; i++) { | |
9525 | sec_def = &custom_sec_defs[i]; | |
450b167f | 9526 | if (sec_def_matches(sec_def, sec_name)) |
697f104d AN |
9527 | return sec_def; |
9528 | } | |
9529 | ||
9530 | n = ARRAY_SIZE(section_defs); | |
9531 | for (i = 0; i < n; i++) { | |
9532 | sec_def = §ion_defs[i]; | |
450b167f | 9533 | if (sec_def_matches(sec_def, sec_name)) |
dd94d45c | 9534 | return sec_def; |
d7a18ea7 | 9535 | } |
697f104d AN |
9536 | |
9537 | if (has_custom_fallback_def) | |
9538 | return &custom_fallback_def; | |
9539 | ||
d7a18ea7 AN |
9540 | return NULL; |
9541 | } | |
9542 | ||
697f104d AN |
9543 | #define MAX_TYPE_NAME_SIZE 32 |
9544 | ||
c76e4c22 TS |
9545 | static char *libbpf_get_type_names(bool attach_type) |
9546 | { | |
d7a18ea7 | 9547 | int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE; |
c76e4c22 TS |
9548 | char *buf; |
9549 | ||
9550 | buf = malloc(len); | |
9551 | if (!buf) | |
9552 | return NULL; | |
9553 | ||
9554 | buf[0] = '\0'; | |
9555 | /* Forge string buf with all available names */ | |
d7a18ea7 | 9556 | for (i = 0; i < ARRAY_SIZE(section_defs); i++) { |
15ea31fa AN |
9557 | const struct bpf_sec_def *sec_def = §ion_defs[i]; |
9558 | ||
9559 | if (attach_type) { | |
4fa5bcfe | 9560 | if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load) |
15ea31fa AN |
9561 | continue; |
9562 | ||
9563 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
9564 | continue; | |
9565 | } | |
c76e4c22 | 9566 | |
d7a18ea7 | 9567 | if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) { |
c76e4c22 TS |
9568 | free(buf); |
9569 | return NULL; | |
9570 | } | |
9571 | strcat(buf, " "); | |
d7a18ea7 | 9572 | strcat(buf, section_defs[i].sec); |
c76e4c22 TS |
9573 | } |
9574 | ||
9575 | return buf; | |
9576 | } | |
9577 | ||
b60df2a0 JK |
9578 | int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type, |
9579 | enum bpf_attach_type *expected_attach_type) | |
583c9009 | 9580 | { |
d7a18ea7 | 9581 | const struct bpf_sec_def *sec_def; |
c76e4c22 | 9582 | char *type_names; |
583c9009 | 9583 | |
b60df2a0 | 9584 | if (!name) |
e9fc3ce9 | 9585 | return libbpf_err(-EINVAL); |
583c9009 | 9586 | |
d7a18ea7 AN |
9587 | sec_def = find_sec_def(name); |
9588 | if (sec_def) { | |
9589 | *prog_type = sec_def->prog_type; | |
9590 | *expected_attach_type = sec_def->expected_attach_type; | |
b60df2a0 JK |
9591 | return 0; |
9592 | } | |
d7a18ea7 | 9593 | |
4a3d6c6a | 9594 | pr_debug("failed to guess program type from ELF section '%s'\n", name); |
c76e4c22 TS |
9595 | type_names = libbpf_get_type_names(false); |
9596 | if (type_names != NULL) { | |
3f519353 | 9597 | pr_debug("supported section(type) names are:%s\n", type_names); |
c76e4c22 TS |
9598 | free(type_names); |
9599 | } | |
9600 | ||
e9fc3ce9 | 9601 | return libbpf_err(-ESRCH); |
b60df2a0 | 9602 | } |
583c9009 | 9603 | |
ccde5760 DM |
9604 | const char *libbpf_bpf_attach_type_str(enum bpf_attach_type t) |
9605 | { | |
9606 | if (t < 0 || t >= ARRAY_SIZE(attach_type_name)) | |
9607 | return NULL; | |
9608 | ||
9609 | return attach_type_name[t]; | |
9610 | } | |
9611 | ||
ba5d1b58 DM |
9612 | const char *libbpf_bpf_link_type_str(enum bpf_link_type t) |
9613 | { | |
9614 | if (t < 0 || t >= ARRAY_SIZE(link_type_name)) | |
9615 | return NULL; | |
9616 | ||
9617 | return link_type_name[t]; | |
9618 | } | |
9619 | ||
3e6dc020 DM |
9620 | const char *libbpf_bpf_map_type_str(enum bpf_map_type t) |
9621 | { | |
9622 | if (t < 0 || t >= ARRAY_SIZE(map_type_name)) | |
9623 | return NULL; | |
9624 | ||
9625 | return map_type_name[t]; | |
9626 | } | |
9627 | ||
d18616e7 DM |
9628 | const char *libbpf_bpf_prog_type_str(enum bpf_prog_type t) |
9629 | { | |
9630 | if (t < 0 || t >= ARRAY_SIZE(prog_type_name)) | |
9631 | return NULL; | |
9632 | ||
9633 | return prog_type_name[t]; | |
9634 | } | |
9635 | ||
590a0088 | 9636 | static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj, |
809a69d6 | 9637 | int sec_idx, |
590a0088 MKL |
9638 | size_t offset) |
9639 | { | |
9640 | struct bpf_map *map; | |
9641 | size_t i; | |
9642 | ||
9643 | for (i = 0; i < obj->nr_maps; i++) { | |
9644 | map = &obj->maps[i]; | |
9645 | if (!bpf_map__is_struct_ops(map)) | |
9646 | continue; | |
809a69d6 KFL |
9647 | if (map->sec_idx == sec_idx && |
9648 | map->sec_offset <= offset && | |
590a0088 MKL |
9649 | offset - map->sec_offset < map->def.value_size) |
9650 | return map; | |
9651 | } | |
9652 | ||
9653 | return NULL; | |
9654 | } | |
9655 | ||
69e4a9d2 KFL |
9656 | /* Collect the reloc from ELF, populate the st_ops->progs[], and update |
9657 | * st_ops->data for shadow type. | |
9658 | */ | |
646f02ff | 9659 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 9660 | Elf64_Shdr *shdr, Elf_Data *data) |
590a0088 MKL |
9661 | { |
9662 | const struct btf_member *member; | |
9663 | struct bpf_struct_ops *st_ops; | |
9664 | struct bpf_program *prog; | |
9665 | unsigned int shdr_idx; | |
9666 | const struct btf *btf; | |
9667 | struct bpf_map *map; | |
7e06aad5 | 9668 | unsigned int moff, insn_idx; |
590a0088 | 9669 | const char *name; |
1d1a3bcf | 9670 | __u32 member_idx; |
ad23b723 AN |
9671 | Elf64_Sym *sym; |
9672 | Elf64_Rel *rel; | |
590a0088 MKL |
9673 | int i, nrels; |
9674 | ||
590a0088 MKL |
9675 | btf = obj->btf; |
9676 | nrels = shdr->sh_size / shdr->sh_entsize; | |
9677 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
9678 | rel = elf_rel_by_idx(data, i); |
9679 | if (!rel) { | |
590a0088 MKL |
9680 | pr_warn("struct_ops reloc: failed to get %d reloc\n", i); |
9681 | return -LIBBPF_ERRNO__FORMAT; | |
9682 | } | |
9683 | ||
ad23b723 AN |
9684 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); |
9685 | if (!sym) { | |
590a0088 | 9686 | pr_warn("struct_ops reloc: symbol %zx not found\n", |
ad23b723 | 9687 | (size_t)ELF64_R_SYM(rel->r_info)); |
590a0088 MKL |
9688 | return -LIBBPF_ERRNO__FORMAT; |
9689 | } | |
9690 | ||
ad23b723 | 9691 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
809a69d6 | 9692 | map = find_struct_ops_map_by_offset(obj, shdr->sh_info, rel->r_offset); |
590a0088 | 9693 | if (!map) { |
ad23b723 AN |
9694 | pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n", |
9695 | (size_t)rel->r_offset); | |
590a0088 MKL |
9696 | return -EINVAL; |
9697 | } | |
9698 | ||
ad23b723 AN |
9699 | moff = rel->r_offset - map->sec_offset; |
9700 | shdr_idx = sym->st_shndx; | |
590a0088 | 9701 | st_ops = map->st_ops; |
ad23b723 | 9702 | pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel->r_offset %zu map->sec_offset %zu name %d (\'%s\')\n", |
590a0088 | 9703 | map->name, |
ad23b723 AN |
9704 | (long long)(rel->r_info >> 32), |
9705 | (long long)sym->st_value, | |
9706 | shdr_idx, (size_t)rel->r_offset, | |
9707 | map->sec_offset, sym->st_name, name); | |
590a0088 MKL |
9708 | |
9709 | if (shdr_idx >= SHN_LORESERVE) { | |
ad23b723 AN |
9710 | pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n", |
9711 | map->name, (size_t)rel->r_offset, shdr_idx); | |
590a0088 MKL |
9712 | return -LIBBPF_ERRNO__RELOC; |
9713 | } | |
ad23b723 | 9714 | if (sym->st_value % BPF_INSN_SZ) { |
7e06aad5 | 9715 | pr_warn("struct_ops reloc %s: invalid target program offset %llu\n", |
ad23b723 | 9716 | map->name, (unsigned long long)sym->st_value); |
7e06aad5 AN |
9717 | return -LIBBPF_ERRNO__FORMAT; |
9718 | } | |
ad23b723 | 9719 | insn_idx = sym->st_value / BPF_INSN_SZ; |
590a0088 MKL |
9720 | |
9721 | member = find_member_by_offset(st_ops->type, moff * 8); | |
9722 | if (!member) { | |
9723 | pr_warn("struct_ops reloc %s: cannot find member at moff %u\n", | |
9724 | map->name, moff); | |
9725 | return -EINVAL; | |
9726 | } | |
9727 | member_idx = member - btf_members(st_ops->type); | |
9728 | name = btf__name_by_offset(btf, member->name_off); | |
9729 | ||
9730 | if (!resolve_func_ptr(btf, member->type, NULL)) { | |
9731 | pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n", | |
9732 | map->name, name); | |
9733 | return -EINVAL; | |
9734 | } | |
9735 | ||
7e06aad5 | 9736 | prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx); |
590a0088 MKL |
9737 | if (!prog) { |
9738 | pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n", | |
9739 | map->name, shdr_idx, name); | |
9740 | return -EINVAL; | |
9741 | } | |
9742 | ||
91b4d1d1 AN |
9743 | /* prevent the use of BPF prog with invalid type */ |
9744 | if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) { | |
9745 | pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n", | |
9746 | map->name, prog->name); | |
9747 | return -EINVAL; | |
9748 | } | |
590a0088 | 9749 | |
590a0088 | 9750 | st_ops->progs[member_idx] = prog; |
69e4a9d2 KFL |
9751 | |
9752 | /* st_ops->data will be exposed to users, being returned by | |
9753 | * bpf_map__initial_value() as a pointer to the shadow | |
9754 | * type. All function pointers in the original struct type | |
9755 | * should be converted to a pointer to struct bpf_program | |
9756 | * in the shadow type. | |
9757 | */ | |
9758 | *((struct bpf_program **)(st_ops->data + moff)) = prog; | |
590a0088 MKL |
9759 | } |
9760 | ||
9761 | return 0; | |
590a0088 MKL |
9762 | } |
9763 | ||
a6ed02ca | 9764 | #define BTF_TRACE_PREFIX "btf_trace_" |
1e092a03 | 9765 | #define BTF_LSM_PREFIX "bpf_lsm_" |
21aef70e | 9766 | #define BTF_ITER_PREFIX "bpf_iter_" |
a6ed02ca KS |
9767 | #define BTF_MAX_NAME_SIZE 128 |
9768 | ||
67234743 AS |
9769 | void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type, |
9770 | const char **prefix, int *kind) | |
9771 | { | |
9772 | switch (attach_type) { | |
9773 | case BPF_TRACE_RAW_TP: | |
9774 | *prefix = BTF_TRACE_PREFIX; | |
9775 | *kind = BTF_KIND_TYPEDEF; | |
9776 | break; | |
9777 | case BPF_LSM_MAC: | |
bffcf348 | 9778 | case BPF_LSM_CGROUP: |
67234743 AS |
9779 | *prefix = BTF_LSM_PREFIX; |
9780 | *kind = BTF_KIND_FUNC; | |
9781 | break; | |
9782 | case BPF_TRACE_ITER: | |
9783 | *prefix = BTF_ITER_PREFIX; | |
9784 | *kind = BTF_KIND_FUNC; | |
9785 | break; | |
9786 | default: | |
9787 | *prefix = ""; | |
9788 | *kind = BTF_KIND_FUNC; | |
9789 | } | |
9790 | } | |
9791 | ||
a6ed02ca KS |
9792 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
9793 | const char *name, __u32 kind) | |
9794 | { | |
9795 | char btf_type_name[BTF_MAX_NAME_SIZE]; | |
9796 | int ret; | |
9797 | ||
9798 | ret = snprintf(btf_type_name, sizeof(btf_type_name), | |
9799 | "%s%s", prefix, name); | |
9800 | /* snprintf returns the number of characters written excluding the | |
c139e40a | 9801 | * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it |
a6ed02ca KS |
9802 | * indicates truncation. |
9803 | */ | |
9804 | if (ret < 0 || ret >= sizeof(btf_type_name)) | |
9805 | return -ENAMETOOLONG; | |
9806 | return btf__find_by_name_kind(btf, btf_type_name, kind); | |
9807 | } | |
9808 | ||
91abb4a6 AN |
9809 | static inline int find_attach_btf_id(struct btf *btf, const char *name, |
9810 | enum bpf_attach_type attach_type) | |
a6ed02ca | 9811 | { |
67234743 AS |
9812 | const char *prefix; |
9813 | int kind; | |
a6ed02ca | 9814 | |
67234743 AS |
9815 | btf_get_kernel_prefix_kind(attach_type, &prefix, &kind); |
9816 | return find_btf_by_prefix_kind(btf, prefix, name, kind); | |
a6ed02ca KS |
9817 | } |
9818 | ||
b8c54ea4 AS |
9819 | int libbpf_find_vmlinux_btf_id(const char *name, |
9820 | enum bpf_attach_type attach_type) | |
12a8654b | 9821 | { |
a6ed02ca | 9822 | struct btf *btf; |
3521ffa2 | 9823 | int err; |
12a8654b | 9824 | |
a710eed3 | 9825 | btf = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
9826 | err = libbpf_get_error(btf); |
9827 | if (err) { | |
12a8654b | 9828 | pr_warn("vmlinux BTF is not found\n"); |
e9fc3ce9 | 9829 | return libbpf_err(err); |
12a8654b AS |
9830 | } |
9831 | ||
91abb4a6 AN |
9832 | err = find_attach_btf_id(btf, name, attach_type); |
9833 | if (err <= 0) | |
9834 | pr_warn("%s is not found in vmlinux BTF\n", name); | |
9835 | ||
3521ffa2 | 9836 | btf__free(btf); |
e9fc3ce9 | 9837 | return libbpf_err(err); |
b8c54ea4 AS |
9838 | } |
9839 | ||
e7bf94db AS |
9840 | static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd) |
9841 | { | |
813847a3 | 9842 | struct bpf_prog_info info; |
ebc7b50a | 9843 | __u32 info_len = sizeof(info); |
6cc93e2f | 9844 | struct btf *btf; |
6d2d73cd | 9845 | int err; |
e7bf94db | 9846 | |
813847a3 | 9847 | memset(&info, 0, info_len); |
629dfc66 | 9848 | err = bpf_prog_get_info_by_fd(attach_prog_fd, &info, &info_len); |
e9fc3ce9 | 9849 | if (err) { |
629dfc66 | 9850 | pr_warn("failed bpf_prog_get_info_by_fd for FD %d: %d\n", |
ebc7b50a | 9851 | attach_prog_fd, err); |
e9fc3ce9 | 9852 | return err; |
e7bf94db | 9853 | } |
6d2d73cd QM |
9854 | |
9855 | err = -EINVAL; | |
ebc7b50a | 9856 | if (!info.btf_id) { |
e7bf94db AS |
9857 | pr_warn("The target program doesn't have BTF\n"); |
9858 | goto out; | |
9859 | } | |
ebc7b50a DM |
9860 | btf = btf__load_from_kernel_by_id(info.btf_id); |
9861 | err = libbpf_get_error(btf); | |
9862 | if (err) { | |
9863 | pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err); | |
e7bf94db AS |
9864 | goto out; |
9865 | } | |
9866 | err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC); | |
9867 | btf__free(btf); | |
9868 | if (err <= 0) { | |
9869 | pr_warn("%s is not found in prog's BTF\n", name); | |
9870 | goto out; | |
9871 | } | |
9872 | out: | |
e7bf94db AS |
9873 | return err; |
9874 | } | |
9875 | ||
91abb4a6 AN |
9876 | static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name, |
9877 | enum bpf_attach_type attach_type, | |
9878 | int *btf_obj_fd, int *btf_type_id) | |
9879 | { | |
8f8a0242 VM |
9880 | int ret, i, mod_len; |
9881 | const char *fn_name, *mod_name = NULL; | |
91abb4a6 | 9882 | |
8f8a0242 VM |
9883 | fn_name = strchr(attach_name, ':'); |
9884 | if (fn_name) { | |
9885 | mod_name = attach_name; | |
9886 | mod_len = fn_name - mod_name; | |
9887 | fn_name++; | |
9888 | } | |
9889 | ||
9890 | if (!mod_name || strncmp(mod_name, "vmlinux", mod_len) == 0) { | |
9891 | ret = find_attach_btf_id(obj->btf_vmlinux, | |
9892 | mod_name ? fn_name : attach_name, | |
9893 | attach_type); | |
9894 | if (ret > 0) { | |
9895 | *btf_obj_fd = 0; /* vmlinux BTF */ | |
9896 | *btf_type_id = ret; | |
9897 | return 0; | |
9898 | } | |
9899 | if (ret != -ENOENT) | |
9900 | return ret; | |
91abb4a6 | 9901 | } |
91abb4a6 AN |
9902 | |
9903 | ret = load_module_btfs(obj); | |
9904 | if (ret) | |
9905 | return ret; | |
9906 | ||
9907 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
9908 | const struct module_btf *mod = &obj->btf_modules[i]; | |
9909 | ||
8f8a0242 VM |
9910 | if (mod_name && strncmp(mod->name, mod_name, mod_len) != 0) |
9911 | continue; | |
9912 | ||
9913 | ret = find_attach_btf_id(mod->btf, | |
9914 | mod_name ? fn_name : attach_name, | |
9915 | attach_type); | |
91abb4a6 AN |
9916 | if (ret > 0) { |
9917 | *btf_obj_fd = mod->fd; | |
9918 | *btf_type_id = ret; | |
9919 | return 0; | |
9920 | } | |
9921 | if (ret == -ENOENT) | |
9922 | continue; | |
9923 | ||
9924 | return ret; | |
9925 | } | |
9926 | ||
9927 | return -ESRCH; | |
9928 | } | |
9929 | ||
15ea31fa AN |
9930 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
9931 | int *btf_obj_fd, int *btf_type_id) | |
b8c54ea4 | 9932 | { |
a6ed02ca KS |
9933 | enum bpf_attach_type attach_type = prog->expected_attach_type; |
9934 | __u32 attach_prog_fd = prog->attach_prog_fd; | |
b6291a6f | 9935 | int err = 0; |
b8c54ea4 | 9936 | |
91abb4a6 | 9937 | /* BPF program's BTF ID */ |
749c202c AN |
9938 | if (prog->type == BPF_PROG_TYPE_EXT || attach_prog_fd) { |
9939 | if (!attach_prog_fd) { | |
9940 | pr_warn("prog '%s': attach program FD is not set\n", prog->name); | |
9941 | return -EINVAL; | |
9942 | } | |
91abb4a6 AN |
9943 | err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd); |
9944 | if (err < 0) { | |
749c202c AN |
9945 | pr_warn("prog '%s': failed to find BPF program (FD %d) BTF ID for '%s': %d\n", |
9946 | prog->name, attach_prog_fd, attach_name, err); | |
91abb4a6 AN |
9947 | return err; |
9948 | } | |
9949 | *btf_obj_fd = 0; | |
9950 | *btf_type_id = err; | |
9951 | return 0; | |
9952 | } | |
9953 | ||
9954 | /* kernel/module BTF ID */ | |
67234743 AS |
9955 | if (prog->obj->gen_loader) { |
9956 | bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type); | |
9957 | *btf_obj_fd = 0; | |
9958 | *btf_type_id = 1; | |
9959 | } else { | |
9e926acd KFL |
9960 | err = find_kernel_btf_id(prog->obj, attach_name, |
9961 | attach_type, btf_obj_fd, | |
9962 | btf_type_id); | |
67234743 | 9963 | } |
91abb4a6 | 9964 | if (err) { |
749c202c AN |
9965 | pr_warn("prog '%s': failed to find kernel BTF type ID of '%s': %d\n", |
9966 | prog->name, attach_name, err); | |
b8c54ea4 | 9967 | return err; |
12a8654b | 9968 | } |
91abb4a6 | 9969 | return 0; |
12a8654b AS |
9970 | } |
9971 | ||
956b620f AI |
9972 | int libbpf_attach_type_by_name(const char *name, |
9973 | enum bpf_attach_type *attach_type) | |
9974 | { | |
c76e4c22 | 9975 | char *type_names; |
b6291a6f | 9976 | const struct bpf_sec_def *sec_def; |
956b620f AI |
9977 | |
9978 | if (!name) | |
e9fc3ce9 | 9979 | return libbpf_err(-EINVAL); |
956b620f | 9980 | |
b6291a6f AN |
9981 | sec_def = find_sec_def(name); |
9982 | if (!sec_def) { | |
9983 | pr_debug("failed to guess attach type based on ELF section name '%s'\n", name); | |
9984 | type_names = libbpf_get_type_names(true); | |
9985 | if (type_names != NULL) { | |
9986 | pr_debug("attachable section(type) names are:%s\n", type_names); | |
9987 | free(type_names); | |
9988 | } | |
9989 | ||
9990 | return libbpf_err(-EINVAL); | |
c76e4c22 TS |
9991 | } |
9992 | ||
4fa5bcfe | 9993 | if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load) |
15ea31fa AN |
9994 | return libbpf_err(-EINVAL); |
9995 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
b6291a6f AN |
9996 | return libbpf_err(-EINVAL); |
9997 | ||
9998 | *attach_type = sec_def->expected_attach_type; | |
9999 | return 0; | |
956b620f AI |
10000 | } |
10001 | ||
a324aae3 | 10002 | int bpf_map__fd(const struct bpf_map *map) |
9d759a9b | 10003 | { |
f08c18e0 AN |
10004 | if (!map) |
10005 | return libbpf_err(-EINVAL); | |
10006 | if (!map_is_created(map)) | |
10007 | return -1; | |
10008 | return map->fd; | |
9d759a9b WN |
10009 | } |
10010 | ||
aed65917 AN |
10011 | static bool map_uses_real_name(const struct bpf_map *map) |
10012 | { | |
10013 | /* Since libbpf started to support custom .data.* and .rodata.* maps, | |
10014 | * their user-visible name differs from kernel-visible name. Users see | |
10015 | * such map's corresponding ELF section name as a map name. | |
10016 | * This check distinguishes .data/.rodata from .data.* and .rodata.* | |
10017 | * maps to know which name has to be returned to the user. | |
10018 | */ | |
10019 | if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0) | |
10020 | return true; | |
10021 | if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0) | |
10022 | return true; | |
10023 | return false; | |
10024 | } | |
10025 | ||
a324aae3 | 10026 | const char *bpf_map__name(const struct bpf_map *map) |
561bbcca | 10027 | { |
aed65917 AN |
10028 | if (!map) |
10029 | return NULL; | |
10030 | ||
10031 | if (map_uses_real_name(map)) | |
10032 | return map->real_name; | |
10033 | ||
10034 | return map->name; | |
561bbcca WN |
10035 | } |
10036 | ||
1bdb6c9a AN |
10037 | enum bpf_map_type bpf_map__type(const struct bpf_map *map) |
10038 | { | |
10039 | return map->def.type; | |
10040 | } | |
10041 | ||
10042 | int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type) | |
10043 | { | |
f08c18e0 | 10044 | if (map_is_created(map)) |
e9fc3ce9 | 10045 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
10046 | map->def.type = type; |
10047 | return 0; | |
10048 | } | |
10049 | ||
10050 | __u32 bpf_map__map_flags(const struct bpf_map *map) | |
10051 | { | |
10052 | return map->def.map_flags; | |
10053 | } | |
10054 | ||
10055 | int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags) | |
10056 | { | |
f08c18e0 | 10057 | if (map_is_created(map)) |
e9fc3ce9 | 10058 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
10059 | map->def.map_flags = flags; |
10060 | return 0; | |
10061 | } | |
10062 | ||
47512102 JK |
10063 | __u64 bpf_map__map_extra(const struct bpf_map *map) |
10064 | { | |
10065 | return map->map_extra; | |
10066 | } | |
10067 | ||
10068 | int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra) | |
10069 | { | |
f08c18e0 | 10070 | if (map_is_created(map)) |
47512102 JK |
10071 | return libbpf_err(-EBUSY); |
10072 | map->map_extra = map_extra; | |
10073 | return 0; | |
10074 | } | |
10075 | ||
1bdb6c9a AN |
10076 | __u32 bpf_map__numa_node(const struct bpf_map *map) |
10077 | { | |
10078 | return map->numa_node; | |
10079 | } | |
10080 | ||
10081 | int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node) | |
10082 | { | |
f08c18e0 | 10083 | if (map_is_created(map)) |
e9fc3ce9 | 10084 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
10085 | map->numa_node = numa_node; |
10086 | return 0; | |
10087 | } | |
10088 | ||
10089 | __u32 bpf_map__key_size(const struct bpf_map *map) | |
10090 | { | |
10091 | return map->def.key_size; | |
10092 | } | |
10093 | ||
10094 | int bpf_map__set_key_size(struct bpf_map *map, __u32 size) | |
10095 | { | |
f08c18e0 | 10096 | if (map_is_created(map)) |
e9fc3ce9 | 10097 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
10098 | map->def.key_size = size; |
10099 | return 0; | |
10100 | } | |
10101 | ||
10102 | __u32 bpf_map__value_size(const struct bpf_map *map) | |
10103 | { | |
10104 | return map->def.value_size; | |
10105 | } | |
10106 | ||
9d0a2331 JK |
10107 | static int map_btf_datasec_resize(struct bpf_map *map, __u32 size) |
10108 | { | |
10109 | struct btf *btf; | |
10110 | struct btf_type *datasec_type, *var_type; | |
10111 | struct btf_var_secinfo *var; | |
10112 | const struct btf_type *array_type; | |
10113 | const struct btf_array *array; | |
4c857a71 JK |
10114 | int vlen, element_sz, new_array_id; |
10115 | __u32 nr_elements; | |
9d0a2331 JK |
10116 | |
10117 | /* check btf existence */ | |
10118 | btf = bpf_object__btf(map->obj); | |
10119 | if (!btf) | |
10120 | return -ENOENT; | |
10121 | ||
10122 | /* verify map is datasec */ | |
10123 | datasec_type = btf_type_by_id(btf, bpf_map__btf_value_type_id(map)); | |
10124 | if (!btf_is_datasec(datasec_type)) { | |
10125 | pr_warn("map '%s': cannot be resized, map value type is not a datasec\n", | |
10126 | bpf_map__name(map)); | |
10127 | return -EINVAL; | |
10128 | } | |
10129 | ||
10130 | /* verify datasec has at least one var */ | |
10131 | vlen = btf_vlen(datasec_type); | |
10132 | if (vlen == 0) { | |
10133 | pr_warn("map '%s': cannot be resized, map value datasec is empty\n", | |
10134 | bpf_map__name(map)); | |
10135 | return -EINVAL; | |
10136 | } | |
10137 | ||
10138 | /* verify last var in the datasec is an array */ | |
10139 | var = &btf_var_secinfos(datasec_type)[vlen - 1]; | |
10140 | var_type = btf_type_by_id(btf, var->type); | |
10141 | array_type = skip_mods_and_typedefs(btf, var_type->type, NULL); | |
10142 | if (!btf_is_array(array_type)) { | |
10143 | pr_warn("map '%s': cannot be resized, last var must be an array\n", | |
10144 | bpf_map__name(map)); | |
10145 | return -EINVAL; | |
10146 | } | |
10147 | ||
10148 | /* verify request size aligns with array */ | |
10149 | array = btf_array(array_type); | |
10150 | element_sz = btf__resolve_size(btf, array->type); | |
10151 | if (element_sz <= 0 || (size - var->offset) % element_sz != 0) { | |
10152 | pr_warn("map '%s': cannot be resized, element size (%d) doesn't align with new total size (%u)\n", | |
10153 | bpf_map__name(map), element_sz, size); | |
10154 | return -EINVAL; | |
10155 | } | |
10156 | ||
10157 | /* create a new array based on the existing array, but with new length */ | |
10158 | nr_elements = (size - var->offset) / element_sz; | |
10159 | new_array_id = btf__add_array(btf, array->index_type, array->type, nr_elements); | |
10160 | if (new_array_id < 0) | |
10161 | return new_array_id; | |
10162 | ||
10163 | /* adding a new btf type invalidates existing pointers to btf objects, | |
10164 | * so refresh pointers before proceeding | |
10165 | */ | |
10166 | datasec_type = btf_type_by_id(btf, map->btf_value_type_id); | |
10167 | var = &btf_var_secinfos(datasec_type)[vlen - 1]; | |
10168 | var_type = btf_type_by_id(btf, var->type); | |
10169 | ||
10170 | /* finally update btf info */ | |
10171 | datasec_type->size = size; | |
10172 | var->size = size - var->offset; | |
10173 | var_type->type = new_array_id; | |
10174 | ||
10175 | return 0; | |
10176 | } | |
10177 | ||
1bdb6c9a AN |
10178 | int bpf_map__set_value_size(struct bpf_map *map, __u32 size) |
10179 | { | |
f08c18e0 | 10180 | if (map->obj->loaded || map->reused) |
e9fc3ce9 | 10181 | return libbpf_err(-EBUSY); |
9d0a2331 JK |
10182 | |
10183 | if (map->mmaped) { | |
9d0a2331 | 10184 | size_t mmap_old_sz, mmap_new_sz; |
79ff13e9 AS |
10185 | int err; |
10186 | ||
10187 | if (map->def.type != BPF_MAP_TYPE_ARRAY) | |
10188 | return -EOPNOTSUPP; | |
9d0a2331 | 10189 | |
79ff13e9 AS |
10190 | mmap_old_sz = bpf_map_mmap_sz(map); |
10191 | mmap_new_sz = array_map_mmap_sz(size, map->def.max_entries); | |
9d0a2331 JK |
10192 | err = bpf_map_mmap_resize(map, mmap_old_sz, mmap_new_sz); |
10193 | if (err) { | |
10194 | pr_warn("map '%s': failed to resize memory-mapped region: %d\n", | |
10195 | bpf_map__name(map), err); | |
10196 | return err; | |
10197 | } | |
10198 | err = map_btf_datasec_resize(map, size); | |
10199 | if (err && err != -ENOENT) { | |
10200 | pr_warn("map '%s': failed to adjust resized BTF, clearing BTF key/value info: %d\n", | |
10201 | bpf_map__name(map), err); | |
10202 | map->btf_value_type_id = 0; | |
10203 | map->btf_key_type_id = 0; | |
10204 | } | |
10205 | } | |
10206 | ||
1bdb6c9a AN |
10207 | map->def.value_size = size; |
10208 | return 0; | |
10209 | } | |
10210 | ||
5b891af7 | 10211 | __u32 bpf_map__btf_key_type_id(const struct bpf_map *map) |
8a138aed | 10212 | { |
61746dbe | 10213 | return map ? map->btf_key_type_id : 0; |
8a138aed MKL |
10214 | } |
10215 | ||
5b891af7 | 10216 | __u32 bpf_map__btf_value_type_id(const struct bpf_map *map) |
8a138aed | 10217 | { |
61746dbe | 10218 | return map ? map->btf_value_type_id : 0; |
8a138aed MKL |
10219 | } |
10220 | ||
e2842be5 THJ |
10221 | int bpf_map__set_initial_value(struct bpf_map *map, |
10222 | const void *data, size_t size) | |
10223 | { | |
2e7ba4f8 AN |
10224 | size_t actual_sz; |
10225 | ||
f08c18e0 AN |
10226 | if (map->obj->loaded || map->reused) |
10227 | return libbpf_err(-EBUSY); | |
10228 | ||
2e7ba4f8 AN |
10229 | if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG) |
10230 | return libbpf_err(-EINVAL); | |
10231 | ||
10232 | if (map->def.type == BPF_MAP_TYPE_ARENA) | |
10233 | actual_sz = map->obj->arena_data_sz; | |
10234 | else | |
10235 | actual_sz = map->def.value_size; | |
10236 | if (size != actual_sz) | |
e9fc3ce9 | 10237 | return libbpf_err(-EINVAL); |
e2842be5 THJ |
10238 | |
10239 | memcpy(map->mmaped, data, size); | |
10240 | return 0; | |
10241 | } | |
10242 | ||
2e7ba4f8 | 10243 | void *bpf_map__initial_value(const struct bpf_map *map, size_t *psize) |
7723256b | 10244 | { |
69e4a9d2 KFL |
10245 | if (bpf_map__is_struct_ops(map)) { |
10246 | if (psize) | |
10247 | *psize = map->def.value_size; | |
10248 | return map->st_ops->data; | |
10249 | } | |
10250 | ||
7723256b AS |
10251 | if (!map->mmaped) |
10252 | return NULL; | |
2e7ba4f8 AN |
10253 | |
10254 | if (map->def.type == BPF_MAP_TYPE_ARENA) | |
10255 | *psize = map->obj->arena_data_sz; | |
10256 | else | |
10257 | *psize = map->def.value_size; | |
10258 | ||
7723256b AS |
10259 | return map->mmaped; |
10260 | } | |
10261 | ||
a324aae3 | 10262 | bool bpf_map__is_internal(const struct bpf_map *map) |
d859900c DB |
10263 | { |
10264 | return map->libbpf_type != LIBBPF_MAP_UNSPEC; | |
10265 | } | |
10266 | ||
1bdb6c9a AN |
10267 | __u32 bpf_map__ifindex(const struct bpf_map *map) |
10268 | { | |
10269 | return map->map_ifindex; | |
10270 | } | |
10271 | ||
10272 | int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex) | |
9aba3613 | 10273 | { |
f08c18e0 | 10274 | if (map_is_created(map)) |
e9fc3ce9 | 10275 | return libbpf_err(-EBUSY); |
9aba3613 | 10276 | map->map_ifindex = ifindex; |
1bdb6c9a | 10277 | return 0; |
9aba3613 JK |
10278 | } |
10279 | ||
addb9fc9 NS |
10280 | int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd) |
10281 | { | |
10282 | if (!bpf_map_type__is_map_in_map(map->def.type)) { | |
be18010e | 10283 | pr_warn("error: unsupported map type\n"); |
e9fc3ce9 | 10284 | return libbpf_err(-EINVAL); |
addb9fc9 NS |
10285 | } |
10286 | if (map->inner_map_fd != -1) { | |
be18010e | 10287 | pr_warn("error: inner_map_fd already specified\n"); |
e9fc3ce9 | 10288 | return libbpf_err(-EINVAL); |
addb9fc9 | 10289 | } |
8f7b239e AN |
10290 | if (map->inner_map) { |
10291 | bpf_map__destroy(map->inner_map); | |
10292 | zfree(&map->inner_map); | |
10293 | } | |
addb9fc9 NS |
10294 | map->inner_map_fd = fd; |
10295 | return 0; | |
10296 | } | |
10297 | ||
0c19a9fb | 10298 | static struct bpf_map * |
a324aae3 | 10299 | __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i) |
9d759a9b | 10300 | { |
0c19a9fb | 10301 | ssize_t idx; |
9d759a9b WN |
10302 | struct bpf_map *s, *e; |
10303 | ||
10304 | if (!obj || !obj->maps) | |
e9fc3ce9 | 10305 | return errno = EINVAL, NULL; |
9d759a9b WN |
10306 | |
10307 | s = obj->maps; | |
10308 | e = obj->maps + obj->nr_maps; | |
10309 | ||
0c19a9fb | 10310 | if ((m < s) || (m >= e)) { |
be18010e KW |
10311 | pr_warn("error in %s: map handler doesn't belong to object\n", |
10312 | __func__); | |
e9fc3ce9 | 10313 | return errno = EINVAL, NULL; |
9d759a9b WN |
10314 | } |
10315 | ||
0c19a9fb SF |
10316 | idx = (m - obj->maps) + i; |
10317 | if (idx >= obj->nr_maps || idx < 0) | |
9d759a9b WN |
10318 | return NULL; |
10319 | return &obj->maps[idx]; | |
10320 | } | |
561bbcca | 10321 | |
2088a3a7 HC |
10322 | struct bpf_map * |
10323 | bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev) | |
0c19a9fb SF |
10324 | { |
10325 | if (prev == NULL) | |
10326 | return obj->maps; | |
10327 | ||
10328 | return __bpf_map__iter(prev, obj, 1); | |
10329 | } | |
10330 | ||
2088a3a7 HC |
10331 | struct bpf_map * |
10332 | bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next) | |
0c19a9fb SF |
10333 | { |
10334 | if (next == NULL) { | |
10335 | if (!obj->nr_maps) | |
10336 | return NULL; | |
10337 | return obj->maps + obj->nr_maps - 1; | |
10338 | } | |
10339 | ||
10340 | return __bpf_map__iter(next, obj, -1); | |
10341 | } | |
10342 | ||
561bbcca | 10343 | struct bpf_map * |
a324aae3 | 10344 | bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name) |
561bbcca WN |
10345 | { |
10346 | struct bpf_map *pos; | |
10347 | ||
f74a53d9 | 10348 | bpf_object__for_each_map(pos, obj) { |
26071635 AN |
10349 | /* if it's a special internal map name (which always starts |
10350 | * with dot) then check if that special name matches the | |
10351 | * real map name (ELF section name) | |
10352 | */ | |
10353 | if (name[0] == '.') { | |
10354 | if (pos->real_name && strcmp(pos->real_name, name) == 0) | |
10355 | return pos; | |
10356 | continue; | |
10357 | } | |
10358 | /* otherwise map name has to be an exact match */ | |
aed65917 AN |
10359 | if (map_uses_real_name(pos)) { |
10360 | if (strcmp(pos->real_name, name) == 0) | |
10361 | return pos; | |
10362 | continue; | |
10363 | } | |
10364 | if (strcmp(pos->name, name) == 0) | |
561bbcca WN |
10365 | return pos; |
10366 | } | |
e9fc3ce9 | 10367 | return errno = ENOENT, NULL; |
561bbcca | 10368 | } |
5a6acad1 | 10369 | |
f3cea32d | 10370 | int |
a324aae3 | 10371 | bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name) |
f3cea32d MF |
10372 | { |
10373 | return bpf_map__fd(bpf_object__find_map_by_name(obj, name)); | |
10374 | } | |
10375 | ||
737d0646 AN |
10376 | static int validate_map_op(const struct bpf_map *map, size_t key_sz, |
10377 | size_t value_sz, bool check_value_sz) | |
10378 | { | |
f08c18e0 | 10379 | if (!map_is_created(map)) /* map is not yet created */ |
737d0646 AN |
10380 | return -ENOENT; |
10381 | ||
10382 | if (map->def.key_size != key_sz) { | |
10383 | pr_warn("map '%s': unexpected key size %zu provided, expected %u\n", | |
10384 | map->name, key_sz, map->def.key_size); | |
10385 | return -EINVAL; | |
10386 | } | |
10387 | ||
7b30c296 MY |
10388 | if (map->fd < 0) { |
10389 | pr_warn("map '%s': can't use BPF map without FD (was it created?)\n", map->name); | |
10390 | return -EINVAL; | |
10391 | } | |
10392 | ||
737d0646 AN |
10393 | if (!check_value_sz) |
10394 | return 0; | |
10395 | ||
10396 | switch (map->def.type) { | |
10397 | case BPF_MAP_TYPE_PERCPU_ARRAY: | |
10398 | case BPF_MAP_TYPE_PERCPU_HASH: | |
10399 | case BPF_MAP_TYPE_LRU_PERCPU_HASH: | |
10400 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: { | |
10401 | int num_cpu = libbpf_num_possible_cpus(); | |
10402 | size_t elem_sz = roundup(map->def.value_size, 8); | |
10403 | ||
10404 | if (value_sz != num_cpu * elem_sz) { | |
10405 | pr_warn("map '%s': unexpected value size %zu provided for per-CPU map, expected %d * %zu = %zd\n", | |
10406 | map->name, value_sz, num_cpu, elem_sz, num_cpu * elem_sz); | |
10407 | return -EINVAL; | |
10408 | } | |
10409 | break; | |
10410 | } | |
10411 | default: | |
10412 | if (map->def.value_size != value_sz) { | |
10413 | pr_warn("map '%s': unexpected value size %zu provided, expected %u\n", | |
10414 | map->name, value_sz, map->def.value_size); | |
10415 | return -EINVAL; | |
10416 | } | |
10417 | break; | |
10418 | } | |
10419 | return 0; | |
10420 | } | |
10421 | ||
10422 | int bpf_map__lookup_elem(const struct bpf_map *map, | |
10423 | const void *key, size_t key_sz, | |
10424 | void *value, size_t value_sz, __u64 flags) | |
10425 | { | |
10426 | int err; | |
10427 | ||
10428 | err = validate_map_op(map, key_sz, value_sz, true); | |
10429 | if (err) | |
10430 | return libbpf_err(err); | |
10431 | ||
10432 | return bpf_map_lookup_elem_flags(map->fd, key, value, flags); | |
10433 | } | |
10434 | ||
10435 | int bpf_map__update_elem(const struct bpf_map *map, | |
10436 | const void *key, size_t key_sz, | |
10437 | const void *value, size_t value_sz, __u64 flags) | |
10438 | { | |
10439 | int err; | |
10440 | ||
10441 | err = validate_map_op(map, key_sz, value_sz, true); | |
10442 | if (err) | |
10443 | return libbpf_err(err); | |
10444 | ||
10445 | return bpf_map_update_elem(map->fd, key, value, flags); | |
10446 | } | |
10447 | ||
10448 | int bpf_map__delete_elem(const struct bpf_map *map, | |
10449 | const void *key, size_t key_sz, __u64 flags) | |
10450 | { | |
10451 | int err; | |
10452 | ||
10453 | err = validate_map_op(map, key_sz, 0, false /* check_value_sz */); | |
10454 | if (err) | |
10455 | return libbpf_err(err); | |
10456 | ||
10457 | return bpf_map_delete_elem_flags(map->fd, key, flags); | |
10458 | } | |
10459 | ||
10460 | int bpf_map__lookup_and_delete_elem(const struct bpf_map *map, | |
10461 | const void *key, size_t key_sz, | |
10462 | void *value, size_t value_sz, __u64 flags) | |
10463 | { | |
10464 | int err; | |
10465 | ||
10466 | err = validate_map_op(map, key_sz, value_sz, true); | |
10467 | if (err) | |
10468 | return libbpf_err(err); | |
10469 | ||
10470 | return bpf_map_lookup_and_delete_elem_flags(map->fd, key, value, flags); | |
10471 | } | |
10472 | ||
10473 | int bpf_map__get_next_key(const struct bpf_map *map, | |
10474 | const void *cur_key, void *next_key, size_t key_sz) | |
10475 | { | |
10476 | int err; | |
10477 | ||
10478 | err = validate_map_op(map, key_sz, 0, false /* check_value_sz */); | |
10479 | if (err) | |
10480 | return libbpf_err(err); | |
10481 | ||
10482 | return bpf_map_get_next_key(map->fd, cur_key, next_key); | |
10483 | } | |
10484 | ||
e28ff1a8 JS |
10485 | long libbpf_get_error(const void *ptr) |
10486 | { | |
e9fc3ce9 AN |
10487 | if (!IS_ERR_OR_NULL(ptr)) |
10488 | return 0; | |
10489 | ||
10490 | if (IS_ERR(ptr)) | |
10491 | errno = -PTR_ERR(ptr); | |
10492 | ||
10493 | /* If ptr == NULL, then errno should be already set by the failing | |
10494 | * API, because libbpf never returns NULL on success and it now always | |
10495 | * sets errno on error. So no extra errno handling for ptr == NULL | |
10496 | * case. | |
10497 | */ | |
10498 | return -errno; | |
e28ff1a8 | 10499 | } |
6f6d33f3 | 10500 | |
cc4f864b AN |
10501 | /* Replace link's underlying BPF program with the new one */ |
10502 | int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog) | |
10503 | { | |
e9fc3ce9 | 10504 | int ret; |
7b30c296 MY |
10505 | int prog_fd = bpf_program__fd(prog); |
10506 | ||
10507 | if (prog_fd < 0) { | |
10508 | pr_warn("prog '%s': can't use BPF program without FD (was it loaded?)\n", | |
10509 | prog->name); | |
10510 | return libbpf_err(-EINVAL); | |
10511 | } | |
c139e40a | 10512 | |
7b30c296 | 10513 | ret = bpf_link_update(bpf_link__fd(link), prog_fd, NULL); |
e9fc3ce9 | 10514 | return libbpf_err_errno(ret); |
cc4f864b AN |
10515 | } |
10516 | ||
d6958706 AN |
10517 | /* Release "ownership" of underlying BPF resource (typically, BPF program |
10518 | * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected | |
10519 | * link, when destructed through bpf_link__destroy() call won't attempt to | |
10520 | * detach/unregisted that BPF resource. This is useful in situations where, | |
10521 | * say, attached BPF program has to outlive userspace program that attached it | |
10522 | * in the system. Depending on type of BPF program, though, there might be | |
10523 | * additional steps (like pinning BPF program in BPF FS) necessary to ensure | |
10524 | * exit of userspace program doesn't trigger automatic detachment and clean up | |
10525 | * inside the kernel. | |
10526 | */ | |
10527 | void bpf_link__disconnect(struct bpf_link *link) | |
10528 | { | |
10529 | link->disconnected = true; | |
10530 | } | |
10531 | ||
1c2e9efc AN |
10532 | int bpf_link__destroy(struct bpf_link *link) |
10533 | { | |
d6958706 | 10534 | int err = 0; |
1c2e9efc | 10535 | |
50450fc7 | 10536 | if (IS_ERR_OR_NULL(link)) |
1c2e9efc AN |
10537 | return 0; |
10538 | ||
d6958706 AN |
10539 | if (!link->disconnected && link->detach) |
10540 | err = link->detach(link); | |
c016b68e AN |
10541 | if (link->pin_path) |
10542 | free(link->pin_path); | |
d88b71d4 AN |
10543 | if (link->dealloc) |
10544 | link->dealloc(link); | |
10545 | else | |
10546 | free(link); | |
1c2e9efc | 10547 | |
e9fc3ce9 | 10548 | return libbpf_err(err); |
1c2e9efc AN |
10549 | } |
10550 | ||
c016b68e AN |
10551 | int bpf_link__fd(const struct bpf_link *link) |
10552 | { | |
10553 | return link->fd; | |
10554 | } | |
10555 | ||
10556 | const char *bpf_link__pin_path(const struct bpf_link *link) | |
10557 | { | |
10558 | return link->pin_path; | |
10559 | } | |
10560 | ||
10561 | static int bpf_link__detach_fd(struct bpf_link *link) | |
10562 | { | |
e9fc3ce9 | 10563 | return libbpf_err_errno(close(link->fd)); |
c016b68e AN |
10564 | } |
10565 | ||
10566 | struct bpf_link *bpf_link__open(const char *path) | |
10567 | { | |
10568 | struct bpf_link *link; | |
10569 | int fd; | |
10570 | ||
10571 | fd = bpf_obj_get(path); | |
10572 | if (fd < 0) { | |
10573 | fd = -errno; | |
10574 | pr_warn("failed to open link at %s: %d\n", path, fd); | |
e9fc3ce9 | 10575 | return libbpf_err_ptr(fd); |
c016b68e AN |
10576 | } |
10577 | ||
10578 | link = calloc(1, sizeof(*link)); | |
10579 | if (!link) { | |
10580 | close(fd); | |
e9fc3ce9 | 10581 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
10582 | } |
10583 | link->detach = &bpf_link__detach_fd; | |
10584 | link->fd = fd; | |
10585 | ||
10586 | link->pin_path = strdup(path); | |
10587 | if (!link->pin_path) { | |
10588 | bpf_link__destroy(link); | |
e9fc3ce9 | 10589 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
10590 | } |
10591 | ||
10592 | return link; | |
10593 | } | |
10594 | ||
2e49527e AN |
10595 | int bpf_link__detach(struct bpf_link *link) |
10596 | { | |
10597 | return bpf_link_detach(link->fd) ? -errno : 0; | |
10598 | } | |
10599 | ||
c016b68e AN |
10600 | int bpf_link__pin(struct bpf_link *link, const char *path) |
10601 | { | |
10602 | int err; | |
10603 | ||
10604 | if (link->pin_path) | |
e9fc3ce9 | 10605 | return libbpf_err(-EBUSY); |
c016b68e AN |
10606 | err = make_parent_dir(path); |
10607 | if (err) | |
e9fc3ce9 | 10608 | return libbpf_err(err); |
c016b68e AN |
10609 | err = check_path(path); |
10610 | if (err) | |
e9fc3ce9 | 10611 | return libbpf_err(err); |
c016b68e AN |
10612 | |
10613 | link->pin_path = strdup(path); | |
10614 | if (!link->pin_path) | |
e9fc3ce9 | 10615 | return libbpf_err(-ENOMEM); |
c016b68e AN |
10616 | |
10617 | if (bpf_obj_pin(link->fd, link->pin_path)) { | |
10618 | err = -errno; | |
10619 | zfree(&link->pin_path); | |
e9fc3ce9 | 10620 | return libbpf_err(err); |
c016b68e AN |
10621 | } |
10622 | ||
10623 | pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path); | |
10624 | return 0; | |
10625 | } | |
10626 | ||
10627 | int bpf_link__unpin(struct bpf_link *link) | |
10628 | { | |
10629 | int err; | |
10630 | ||
10631 | if (!link->pin_path) | |
e9fc3ce9 | 10632 | return libbpf_err(-EINVAL); |
c016b68e AN |
10633 | |
10634 | err = unlink(link->pin_path); | |
10635 | if (err != 0) | |
af0efa05 | 10636 | return -errno; |
c016b68e AN |
10637 | |
10638 | pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path); | |
10639 | zfree(&link->pin_path); | |
10640 | return 0; | |
10641 | } | |
63f2f5ee | 10642 | |
668ace0e AN |
10643 | struct bpf_link_perf { |
10644 | struct bpf_link link; | |
10645 | int perf_event_fd; | |
ca304b40 RDT |
10646 | /* legacy kprobe support: keep track of probe identifier and type */ |
10647 | char *legacy_probe_name; | |
46ed5fc3 | 10648 | bool legacy_is_kprobe; |
ca304b40 | 10649 | bool legacy_is_retprobe; |
668ace0e AN |
10650 | }; |
10651 | ||
46ed5fc3 | 10652 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe); |
cc10623c | 10653 | static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe); |
46ed5fc3 | 10654 | |
668ace0e | 10655 | static int bpf_link_perf_detach(struct bpf_link *link) |
63f2f5ee | 10656 | { |
668ace0e AN |
10657 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); |
10658 | int err = 0; | |
63f2f5ee | 10659 | |
668ace0e | 10660 | if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0) |
63f2f5ee AN |
10661 | err = -errno; |
10662 | ||
668ace0e AN |
10663 | if (perf_link->perf_event_fd != link->fd) |
10664 | close(perf_link->perf_event_fd); | |
c016b68e | 10665 | close(link->fd); |
668ace0e | 10666 | |
cc10623c | 10667 | /* legacy uprobe/kprobe needs to be removed after perf event fd closure */ |
46ed5fc3 AN |
10668 | if (perf_link->legacy_probe_name) { |
10669 | if (perf_link->legacy_is_kprobe) { | |
10670 | err = remove_kprobe_event_legacy(perf_link->legacy_probe_name, | |
10671 | perf_link->legacy_is_retprobe); | |
cc10623c AN |
10672 | } else { |
10673 | err = remove_uprobe_event_legacy(perf_link->legacy_probe_name, | |
10674 | perf_link->legacy_is_retprobe); | |
46ed5fc3 AN |
10675 | } |
10676 | } | |
ca304b40 RDT |
10677 | |
10678 | return err; | |
63f2f5ee AN |
10679 | } |
10680 | ||
668ace0e AN |
10681 | static void bpf_link_perf_dealloc(struct bpf_link *link) |
10682 | { | |
10683 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
10684 | ||
ca304b40 | 10685 | free(perf_link->legacy_probe_name); |
668ace0e AN |
10686 | free(perf_link); |
10687 | } | |
10688 | ||
942025c9 | 10689 | struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd, |
47faff37 | 10690 | const struct bpf_perf_event_opts *opts) |
63f2f5ee AN |
10691 | { |
10692 | char errmsg[STRERR_BUFSIZE]; | |
668ace0e AN |
10693 | struct bpf_link_perf *link; |
10694 | int prog_fd, link_fd = -1, err; | |
f8b299bc | 10695 | bool force_ioctl_attach; |
63f2f5ee | 10696 | |
47faff37 AN |
10697 | if (!OPTS_VALID(opts, bpf_perf_event_opts)) |
10698 | return libbpf_err_ptr(-EINVAL); | |
10699 | ||
63f2f5ee | 10700 | if (pfd < 0) { |
52109584 AN |
10701 | pr_warn("prog '%s': invalid perf event FD %d\n", |
10702 | prog->name, pfd); | |
e9fc3ce9 | 10703 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
10704 | } |
10705 | prog_fd = bpf_program__fd(prog); | |
10706 | if (prog_fd < 0) { | |
7b30c296 | 10707 | pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n", |
52109584 | 10708 | prog->name); |
e9fc3ce9 | 10709 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
10710 | } |
10711 | ||
d6958706 | 10712 | link = calloc(1, sizeof(*link)); |
63f2f5ee | 10713 | if (!link) |
e9fc3ce9 | 10714 | return libbpf_err_ptr(-ENOMEM); |
668ace0e AN |
10715 | link->link.detach = &bpf_link_perf_detach; |
10716 | link->link.dealloc = &bpf_link_perf_dealloc; | |
10717 | link->perf_event_fd = pfd; | |
63f2f5ee | 10718 | |
f8b299bc MD |
10719 | force_ioctl_attach = OPTS_GET(opts, force_ioctl_attach, false); |
10720 | if (kernel_supports(prog->obj, FEAT_PERF_LINK) && !force_ioctl_attach) { | |
47faff37 AN |
10721 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts, |
10722 | .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0)); | |
10723 | ||
10724 | link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts); | |
668ace0e AN |
10725 | if (link_fd < 0) { |
10726 | err = -errno; | |
10727 | pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n", | |
10728 | prog->name, pfd, | |
10729 | err, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10730 | goto err_out; | |
10731 | } | |
10732 | link->link.fd = link_fd; | |
10733 | } else { | |
47faff37 AN |
10734 | if (OPTS_GET(opts, bpf_cookie, 0)) { |
10735 | pr_warn("prog '%s': user context value is not supported\n", prog->name); | |
10736 | err = -EOPNOTSUPP; | |
10737 | goto err_out; | |
10738 | } | |
10739 | ||
668ace0e AN |
10740 | if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) { |
10741 | err = -errno; | |
10742 | pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n", | |
10743 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10744 | if (err == -EPROTO) | |
10745 | pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n", | |
10746 | prog->name, pfd); | |
10747 | goto err_out; | |
10748 | } | |
10749 | link->link.fd = pfd; | |
63f2f5ee AN |
10750 | } |
10751 | if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
10752 | err = -errno; | |
668ace0e | 10753 | pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n", |
52109584 | 10754 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
668ace0e | 10755 | goto err_out; |
63f2f5ee | 10756 | } |
668ace0e AN |
10757 | |
10758 | return &link->link; | |
10759 | err_out: | |
10760 | if (link_fd >= 0) | |
10761 | close(link_fd); | |
10762 | free(link); | |
10763 | return libbpf_err_ptr(err); | |
63f2f5ee AN |
10764 | } |
10765 | ||
942025c9 | 10766 | struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd) |
47faff37 AN |
10767 | { |
10768 | return bpf_program__attach_perf_event_opts(prog, pfd, NULL); | |
10769 | } | |
10770 | ||
b2650027 AN |
10771 | /* |
10772 | * this function is expected to parse integer in the range of [0, 2^31-1] from | |
10773 | * given file using scanf format string fmt. If actual parsed value is | |
10774 | * negative, the result might be indistinguishable from error | |
10775 | */ | |
10776 | static int parse_uint_from_file(const char *file, const char *fmt) | |
10777 | { | |
10778 | char buf[STRERR_BUFSIZE]; | |
10779 | int err, ret; | |
10780 | FILE *f; | |
10781 | ||
59842c54 | 10782 | f = fopen(file, "re"); |
b2650027 AN |
10783 | if (!f) { |
10784 | err = -errno; | |
10785 | pr_debug("failed to open '%s': %s\n", file, | |
10786 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
10787 | return err; | |
10788 | } | |
10789 | err = fscanf(f, fmt, &ret); | |
10790 | if (err != 1) { | |
10791 | err = err == EOF ? -EIO : -errno; | |
10792 | pr_debug("failed to parse '%s': %s\n", file, | |
10793 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
10794 | fclose(f); | |
10795 | return err; | |
10796 | } | |
10797 | fclose(f); | |
10798 | return ret; | |
10799 | } | |
10800 | ||
10801 | static int determine_kprobe_perf_type(void) | |
10802 | { | |
10803 | const char *file = "/sys/bus/event_source/devices/kprobe/type"; | |
10804 | ||
10805 | return parse_uint_from_file(file, "%d\n"); | |
10806 | } | |
10807 | ||
10808 | static int determine_uprobe_perf_type(void) | |
10809 | { | |
10810 | const char *file = "/sys/bus/event_source/devices/uprobe/type"; | |
10811 | ||
10812 | return parse_uint_from_file(file, "%d\n"); | |
10813 | } | |
10814 | ||
10815 | static int determine_kprobe_retprobe_bit(void) | |
10816 | { | |
10817 | const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe"; | |
10818 | ||
10819 | return parse_uint_from_file(file, "config:%d\n"); | |
10820 | } | |
10821 | ||
10822 | static int determine_uprobe_retprobe_bit(void) | |
10823 | { | |
10824 | const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe"; | |
10825 | ||
10826 | return parse_uint_from_file(file, "config:%d\n"); | |
10827 | } | |
10828 | ||
5e3b8356 AN |
10829 | #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32 |
10830 | #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32 | |
10831 | ||
b2650027 | 10832 | static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name, |
5e3b8356 | 10833 | uint64_t offset, int pid, size_t ref_ctr_off) |
b2650027 | 10834 | { |
813847a3 AN |
10835 | const size_t attr_sz = sizeof(struct perf_event_attr); |
10836 | struct perf_event_attr attr; | |
b2650027 | 10837 | char errmsg[STRERR_BUFSIZE]; |
708ac5be | 10838 | int type, pfd; |
b2650027 | 10839 | |
1520e846 | 10840 | if ((__u64)ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS)) |
5e3b8356 AN |
10841 | return -EINVAL; |
10842 | ||
813847a3 AN |
10843 | memset(&attr, 0, attr_sz); |
10844 | ||
b2650027 AN |
10845 | type = uprobe ? determine_uprobe_perf_type() |
10846 | : determine_kprobe_perf_type(); | |
10847 | if (type < 0) { | |
be18010e KW |
10848 | pr_warn("failed to determine %s perf type: %s\n", |
10849 | uprobe ? "uprobe" : "kprobe", | |
10850 | libbpf_strerror_r(type, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10851 | return type; |
10852 | } | |
10853 | if (retprobe) { | |
10854 | int bit = uprobe ? determine_uprobe_retprobe_bit() | |
10855 | : determine_kprobe_retprobe_bit(); | |
10856 | ||
10857 | if (bit < 0) { | |
be18010e KW |
10858 | pr_warn("failed to determine %s retprobe bit: %s\n", |
10859 | uprobe ? "uprobe" : "kprobe", | |
10860 | libbpf_strerror_r(bit, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10861 | return bit; |
10862 | } | |
10863 | attr.config |= 1 << bit; | |
10864 | } | |
813847a3 | 10865 | attr.size = attr_sz; |
b2650027 | 10866 | attr.type = type; |
5e3b8356 | 10867 | attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT; |
36db2a94 AN |
10868 | attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */ |
10869 | attr.config2 = offset; /* kprobe_addr or probe_offset */ | |
b2650027 AN |
10870 | |
10871 | /* pid filter is meaningful only for uprobes */ | |
10872 | pfd = syscall(__NR_perf_event_open, &attr, | |
10873 | pid < 0 ? -1 : pid /* pid */, | |
10874 | pid == -1 ? 0 : -1 /* cpu */, | |
10875 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
708ac5be | 10876 | return pfd >= 0 ? pfd : -errno; |
b2650027 AN |
10877 | } |
10878 | ||
46ed5fc3 AN |
10879 | static int append_to_file(const char *file, const char *fmt, ...) |
10880 | { | |
10881 | int fd, n, err = 0; | |
10882 | va_list ap; | |
01dc26c9 LP |
10883 | char buf[1024]; |
10884 | ||
10885 | va_start(ap, fmt); | |
10886 | n = vsnprintf(buf, sizeof(buf), fmt, ap); | |
10887 | va_end(ap); | |
10888 | ||
10889 | if (n < 0 || n >= sizeof(buf)) | |
10890 | return -EINVAL; | |
46ed5fc3 | 10891 | |
92274e24 | 10892 | fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0); |
46ed5fc3 AN |
10893 | if (fd < 0) |
10894 | return -errno; | |
10895 | ||
01dc26c9 | 10896 | if (write(fd, buf, n) < 0) |
46ed5fc3 AN |
10897 | err = -errno; |
10898 | ||
10899 | close(fd); | |
10900 | return err; | |
10901 | } | |
10902 | ||
a1ac9fd6 AN |
10903 | #define DEBUGFS "/sys/kernel/debug/tracing" |
10904 | #define TRACEFS "/sys/kernel/tracing" | |
10905 | ||
10906 | static bool use_debugfs(void) | |
10907 | { | |
10908 | static int has_debugfs = -1; | |
10909 | ||
10910 | if (has_debugfs < 0) | |
6a4ab886 | 10911 | has_debugfs = faccessat(AT_FDCWD, DEBUGFS, F_OK, AT_EACCESS) == 0; |
a1ac9fd6 AN |
10912 | |
10913 | return has_debugfs == 1; | |
10914 | } | |
10915 | ||
10916 | static const char *tracefs_path(void) | |
10917 | { | |
10918 | return use_debugfs() ? DEBUGFS : TRACEFS; | |
10919 | } | |
10920 | ||
10921 | static const char *tracefs_kprobe_events(void) | |
10922 | { | |
10923 | return use_debugfs() ? DEBUGFS"/kprobe_events" : TRACEFS"/kprobe_events"; | |
10924 | } | |
10925 | ||
10926 | static const char *tracefs_uprobe_events(void) | |
10927 | { | |
10928 | return use_debugfs() ? DEBUGFS"/uprobe_events" : TRACEFS"/uprobe_events"; | |
10929 | } | |
10930 | ||
8a3fe76f JL |
10931 | static const char *tracefs_available_filter_functions(void) |
10932 | { | |
10933 | return use_debugfs() ? DEBUGFS"/available_filter_functions" | |
10934 | : TRACEFS"/available_filter_functions"; | |
10935 | } | |
10936 | ||
56baeeba JL |
10937 | static const char *tracefs_available_filter_functions_addrs(void) |
10938 | { | |
10939 | return use_debugfs() ? DEBUGFS"/available_filter_functions_addrs" | |
10940 | : TRACEFS"/available_filter_functions_addrs"; | |
10941 | } | |
10942 | ||
46ed5fc3 AN |
10943 | static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz, |
10944 | const char *kfunc_name, size_t offset) | |
10945 | { | |
51a33c60 | 10946 | static int index = 0; |
2fa07453 | 10947 | int i; |
51a33c60 QW |
10948 | |
10949 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset, | |
10950 | __sync_fetch_and_add(&index, 1)); | |
2fa07453 MD |
10951 | |
10952 | /* sanitize binary_path in the probe name */ | |
10953 | for (i = 0; buf[i]; i++) { | |
10954 | if (!isalnum(buf[i])) | |
10955 | buf[i] = '_'; | |
10956 | } | |
46ed5fc3 AN |
10957 | } |
10958 | ||
10959 | static int add_kprobe_event_legacy(const char *probe_name, bool retprobe, | |
10960 | const char *kfunc_name, size_t offset) | |
10961 | { | |
a1ac9fd6 | 10962 | return append_to_file(tracefs_kprobe_events(), "%c:%s/%s %s+0x%zx", |
46ed5fc3 AN |
10963 | retprobe ? 'r' : 'p', |
10964 | retprobe ? "kretprobes" : "kprobes", | |
10965 | probe_name, kfunc_name, offset); | |
10966 | } | |
10967 | ||
10968 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe) | |
10969 | { | |
a1ac9fd6 AN |
10970 | return append_to_file(tracefs_kprobe_events(), "-:%s/%s", |
10971 | retprobe ? "kretprobes" : "kprobes", probe_name); | |
46ed5fc3 AN |
10972 | } |
10973 | ||
10974 | static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
10975 | { | |
10976 | char file[256]; | |
10977 | ||
a1ac9fd6 AN |
10978 | snprintf(file, sizeof(file), "%s/events/%s/%s/id", |
10979 | tracefs_path(), retprobe ? "kretprobes" : "kprobes", probe_name); | |
46ed5fc3 AN |
10980 | |
10981 | return parse_uint_from_file(file, "%d\n"); | |
10982 | } | |
10983 | ||
10984 | static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe, | |
10985 | const char *kfunc_name, size_t offset, int pid) | |
ca304b40 | 10986 | { |
813847a3 AN |
10987 | const size_t attr_sz = sizeof(struct perf_event_attr); |
10988 | struct perf_event_attr attr; | |
ca304b40 RDT |
10989 | char errmsg[STRERR_BUFSIZE]; |
10990 | int type, pfd, err; | |
10991 | ||
46ed5fc3 | 10992 | err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset); |
ca304b40 | 10993 | if (err < 0) { |
46ed5fc3 AN |
10994 | pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n", |
10995 | kfunc_name, offset, | |
ca304b40 RDT |
10996 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10997 | return err; | |
10998 | } | |
46ed5fc3 | 10999 | type = determine_kprobe_perf_type_legacy(probe_name, retprobe); |
ca304b40 | 11000 | if (type < 0) { |
80940293 | 11001 | err = type; |
46ed5fc3 AN |
11002 | pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n", |
11003 | kfunc_name, offset, | |
80940293 CW |
11004 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
11005 | goto err_clean_legacy; | |
ca304b40 | 11006 | } |
813847a3 AN |
11007 | |
11008 | memset(&attr, 0, attr_sz); | |
11009 | attr.size = attr_sz; | |
ca304b40 RDT |
11010 | attr.config = type; |
11011 | attr.type = PERF_TYPE_TRACEPOINT; | |
11012 | ||
11013 | pfd = syscall(__NR_perf_event_open, &attr, | |
11014 | pid < 0 ? -1 : pid, /* pid */ | |
11015 | pid == -1 ? 0 : -1, /* cpu */ | |
11016 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
11017 | if (pfd < 0) { | |
11018 | err = -errno; | |
11019 | pr_warn("legacy kprobe perf_event_open() failed: %s\n", | |
11020 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
80940293 | 11021 | goto err_clean_legacy; |
ca304b40 RDT |
11022 | } |
11023 | return pfd; | |
80940293 CW |
11024 | |
11025 | err_clean_legacy: | |
11026 | /* Clear the newly added legacy kprobe_event */ | |
11027 | remove_kprobe_event_legacy(probe_name, retprobe); | |
11028 | return err; | |
ca304b40 RDT |
11029 | } |
11030 | ||
708ac5be AN |
11031 | static const char *arch_specific_syscall_pfx(void) |
11032 | { | |
11033 | #if defined(__x86_64__) | |
11034 | return "x64"; | |
11035 | #elif defined(__i386__) | |
11036 | return "ia32"; | |
11037 | #elif defined(__s390x__) | |
11038 | return "s390x"; | |
11039 | #elif defined(__s390__) | |
11040 | return "s390"; | |
11041 | #elif defined(__arm__) | |
11042 | return "arm"; | |
11043 | #elif defined(__aarch64__) | |
11044 | return "arm64"; | |
11045 | #elif defined(__mips__) | |
11046 | return "mips"; | |
11047 | #elif defined(__riscv) | |
11048 | return "riscv"; | |
64893e83 DM |
11049 | #elif defined(__powerpc__) |
11050 | return "powerpc"; | |
11051 | #elif defined(__powerpc64__) | |
11052 | return "powerpc64"; | |
708ac5be AN |
11053 | #else |
11054 | return NULL; | |
11055 | #endif | |
11056 | } | |
11057 | ||
f3dcee93 | 11058 | int probe_kern_syscall_wrapper(int token_fd) |
708ac5be AN |
11059 | { |
11060 | char syscall_name[64]; | |
11061 | const char *ksys_pfx; | |
11062 | ||
11063 | ksys_pfx = arch_specific_syscall_pfx(); | |
11064 | if (!ksys_pfx) | |
11065 | return 0; | |
11066 | ||
11067 | snprintf(syscall_name, sizeof(syscall_name), "__%s_sys_bpf", ksys_pfx); | |
11068 | ||
11069 | if (determine_kprobe_perf_type() >= 0) { | |
11070 | int pfd; | |
11071 | ||
11072 | pfd = perf_event_open_probe(false, false, syscall_name, 0, getpid(), 0); | |
11073 | if (pfd >= 0) | |
11074 | close(pfd); | |
11075 | ||
11076 | return pfd >= 0 ? 1 : 0; | |
11077 | } else { /* legacy mode */ | |
11078 | char probe_name[128]; | |
11079 | ||
11080 | gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), syscall_name, 0); | |
11081 | if (add_kprobe_event_legacy(probe_name, false, syscall_name, 0) < 0) | |
11082 | return 0; | |
11083 | ||
11084 | (void)remove_kprobe_event_legacy(probe_name, false); | |
11085 | return 1; | |
11086 | } | |
11087 | } | |
11088 | ||
da97553e | 11089 | struct bpf_link * |
942025c9 | 11090 | bpf_program__attach_kprobe_opts(const struct bpf_program *prog, |
ac0ed488 | 11091 | const char *func_name, |
47faff37 | 11092 | const struct bpf_kprobe_opts *opts) |
b2650027 | 11093 | { |
47faff37 | 11094 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f8b299bc | 11095 | enum probe_attach_mode attach_mode; |
b2650027 | 11096 | char errmsg[STRERR_BUFSIZE]; |
ca304b40 | 11097 | char *legacy_probe = NULL; |
b2650027 | 11098 | struct bpf_link *link; |
46ed5fc3 | 11099 | size_t offset; |
ca304b40 | 11100 | bool retprobe, legacy; |
b2650027 AN |
11101 | int pfd, err; |
11102 | ||
da97553e JO |
11103 | if (!OPTS_VALID(opts, bpf_kprobe_opts)) |
11104 | return libbpf_err_ptr(-EINVAL); | |
11105 | ||
f8b299bc | 11106 | attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT); |
da97553e JO |
11107 | retprobe = OPTS_GET(opts, retprobe, false); |
11108 | offset = OPTS_GET(opts, offset, 0); | |
47faff37 | 11109 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
da97553e | 11110 | |
ca304b40 | 11111 | legacy = determine_kprobe_perf_type() < 0; |
f8b299bc MD |
11112 | switch (attach_mode) { |
11113 | case PROBE_ATTACH_MODE_LEGACY: | |
11114 | legacy = true; | |
11115 | pe_opts.force_ioctl_attach = true; | |
11116 | break; | |
11117 | case PROBE_ATTACH_MODE_PERF: | |
11118 | if (legacy) | |
11119 | return libbpf_err_ptr(-ENOTSUP); | |
11120 | pe_opts.force_ioctl_attach = true; | |
11121 | break; | |
11122 | case PROBE_ATTACH_MODE_LINK: | |
11123 | if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK)) | |
11124 | return libbpf_err_ptr(-ENOTSUP); | |
11125 | break; | |
11126 | case PROBE_ATTACH_MODE_DEFAULT: | |
11127 | break; | |
11128 | default: | |
11129 | return libbpf_err_ptr(-EINVAL); | |
11130 | } | |
11131 | ||
ca304b40 RDT |
11132 | if (!legacy) { |
11133 | pfd = perf_event_open_probe(false /* uprobe */, retprobe, | |
11134 | func_name, offset, | |
11135 | -1 /* pid */, 0 /* ref_ctr_off */); | |
11136 | } else { | |
46ed5fc3 AN |
11137 | char probe_name[256]; |
11138 | ||
11139 | gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
11140 | func_name, offset); | |
11141 | ||
71cff670 | 11142 | legacy_probe = strdup(probe_name); |
ca304b40 RDT |
11143 | if (!legacy_probe) |
11144 | return libbpf_err_ptr(-ENOMEM); | |
11145 | ||
46ed5fc3 | 11146 | pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name, |
ca304b40 RDT |
11147 | offset, -1 /* pid */); |
11148 | } | |
b2650027 | 11149 | if (pfd < 0) { |
46ed5fc3 AN |
11150 | err = -errno; |
11151 | pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n", | |
11152 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
11153 | func_name, offset, | |
303a2572 AN |
11154 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
11155 | goto err_out; | |
b2650027 | 11156 | } |
47faff37 | 11157 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
11158 | err = libbpf_get_error(link); |
11159 | if (err) { | |
b2650027 | 11160 | close(pfd); |
46ed5fc3 AN |
11161 | pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n", |
11162 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
11163 | func_name, offset, | |
be18010e | 11164 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
80940293 | 11165 | goto err_clean_legacy; |
b2650027 | 11166 | } |
ca304b40 RDT |
11167 | if (legacy) { |
11168 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
11169 | ||
11170 | perf_link->legacy_probe_name = legacy_probe; | |
46ed5fc3 | 11171 | perf_link->legacy_is_kprobe = true; |
ca304b40 RDT |
11172 | perf_link->legacy_is_retprobe = retprobe; |
11173 | } | |
11174 | ||
b2650027 | 11175 | return link; |
80940293 CW |
11176 | |
11177 | err_clean_legacy: | |
11178 | if (legacy) | |
11179 | remove_kprobe_event_legacy(legacy_probe, retprobe); | |
303a2572 AN |
11180 | err_out: |
11181 | free(legacy_probe); | |
11182 | return libbpf_err_ptr(err); | |
b2650027 AN |
11183 | } |
11184 | ||
942025c9 | 11185 | struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog, |
ac0ed488 JO |
11186 | bool retprobe, |
11187 | const char *func_name) | |
11188 | { | |
da97553e | 11189 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts, |
ac0ed488 | 11190 | .retprobe = retprobe, |
da97553e | 11191 | ); |
ac0ed488 JO |
11192 | |
11193 | return bpf_program__attach_kprobe_opts(prog, func_name, &opts); | |
11194 | } | |
11195 | ||
708ac5be AN |
11196 | struct bpf_link *bpf_program__attach_ksyscall(const struct bpf_program *prog, |
11197 | const char *syscall_name, | |
11198 | const struct bpf_ksyscall_opts *opts) | |
11199 | { | |
11200 | LIBBPF_OPTS(bpf_kprobe_opts, kprobe_opts); | |
11201 | char func_name[128]; | |
11202 | ||
11203 | if (!OPTS_VALID(opts, bpf_ksyscall_opts)) | |
11204 | return libbpf_err_ptr(-EINVAL); | |
11205 | ||
11206 | if (kernel_supports(prog->obj, FEAT_SYSCALL_WRAPPER)) { | |
64893e83 DM |
11207 | /* arch_specific_syscall_pfx() should never return NULL here |
11208 | * because it is guarded by kernel_supports(). However, since | |
11209 | * compiler does not know that we have an explicit conditional | |
11210 | * as well. | |
11211 | */ | |
708ac5be | 11212 | snprintf(func_name, sizeof(func_name), "__%s_sys_%s", |
64893e83 | 11213 | arch_specific_syscall_pfx() ? : "", syscall_name); |
708ac5be AN |
11214 | } else { |
11215 | snprintf(func_name, sizeof(func_name), "__se_sys_%s", syscall_name); | |
11216 | } | |
11217 | ||
11218 | kprobe_opts.retprobe = OPTS_GET(opts, retprobe, false); | |
11219 | kprobe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); | |
11220 | ||
11221 | return bpf_program__attach_kprobe_opts(prog, func_name, &kprobe_opts); | |
11222 | } | |
11223 | ||
ddc6b049 | 11224 | /* Adapted from perf/util/string.c */ |
e613d1d0 | 11225 | bool glob_match(const char *str, const char *pat) |
ddc6b049 JO |
11226 | { |
11227 | while (*str && *pat && *pat != '*') { | |
11228 | if (*pat == '?') { /* Matches any single character */ | |
11229 | str++; | |
11230 | pat++; | |
11231 | continue; | |
11232 | } | |
11233 | if (*str != *pat) | |
11234 | return false; | |
11235 | str++; | |
11236 | pat++; | |
11237 | } | |
11238 | /* Check wild card */ | |
11239 | if (*pat == '*') { | |
11240 | while (*pat == '*') | |
11241 | pat++; | |
11242 | if (!*pat) /* Tail wild card matches all */ | |
11243 | return true; | |
11244 | while (*str) | |
11245 | if (glob_match(str++, pat)) | |
11246 | return true; | |
11247 | } | |
11248 | return !*str && !*pat; | |
11249 | } | |
11250 | ||
11251 | struct kprobe_multi_resolve { | |
11252 | const char *pattern; | |
11253 | unsigned long *addrs; | |
11254 | size_t cap; | |
11255 | size_t cnt; | |
11256 | }; | |
11257 | ||
8a3fe76f JL |
11258 | struct avail_kallsyms_data { |
11259 | char **syms; | |
11260 | size_t cnt; | |
11261 | struct kprobe_multi_resolve *res; | |
11262 | }; | |
11263 | ||
11264 | static int avail_func_cmp(const void *a, const void *b) | |
11265 | { | |
11266 | return strcmp(*(const char **)a, *(const char **)b); | |
11267 | } | |
11268 | ||
11269 | static int avail_kallsyms_cb(unsigned long long sym_addr, char sym_type, | |
11270 | const char *sym_name, void *ctx) | |
ddc6b049 | 11271 | { |
8a3fe76f JL |
11272 | struct avail_kallsyms_data *data = ctx; |
11273 | struct kprobe_multi_resolve *res = data->res; | |
ddc6b049 JO |
11274 | int err; |
11275 | ||
8a3fe76f | 11276 | if (!bsearch(&sym_name, data->syms, data->cnt, sizeof(*data->syms), avail_func_cmp)) |
ddc6b049 JO |
11277 | return 0; |
11278 | ||
8a3fe76f | 11279 | err = libbpf_ensure_mem((void **)&res->addrs, &res->cap, sizeof(*res->addrs), res->cnt + 1); |
ddc6b049 JO |
11280 | if (err) |
11281 | return err; | |
11282 | ||
8a3fe76f | 11283 | res->addrs[res->cnt++] = (unsigned long)sym_addr; |
ddc6b049 JO |
11284 | return 0; |
11285 | } | |
11286 | ||
8a3fe76f JL |
11287 | static int libbpf_available_kallsyms_parse(struct kprobe_multi_resolve *res) |
11288 | { | |
11289 | const char *available_functions_file = tracefs_available_filter_functions(); | |
11290 | struct avail_kallsyms_data data; | |
11291 | char sym_name[500]; | |
11292 | FILE *f; | |
11293 | int err = 0, ret, i; | |
11294 | char **syms = NULL; | |
11295 | size_t cap = 0, cnt = 0; | |
11296 | ||
11297 | f = fopen(available_functions_file, "re"); | |
11298 | if (!f) { | |
11299 | err = -errno; | |
11300 | pr_warn("failed to open %s: %d\n", available_functions_file, err); | |
11301 | return err; | |
11302 | } | |
11303 | ||
11304 | while (true) { | |
11305 | char *name; | |
11306 | ||
11307 | ret = fscanf(f, "%499s%*[^\n]\n", sym_name); | |
11308 | if (ret == EOF && feof(f)) | |
11309 | break; | |
11310 | ||
11311 | if (ret != 1) { | |
11312 | pr_warn("failed to parse available_filter_functions entry: %d\n", ret); | |
11313 | err = -EINVAL; | |
11314 | goto cleanup; | |
11315 | } | |
11316 | ||
11317 | if (!glob_match(sym_name, res->pattern)) | |
11318 | continue; | |
11319 | ||
11320 | err = libbpf_ensure_mem((void **)&syms, &cap, sizeof(*syms), cnt + 1); | |
11321 | if (err) | |
11322 | goto cleanup; | |
11323 | ||
11324 | name = strdup(sym_name); | |
11325 | if (!name) { | |
11326 | err = -errno; | |
11327 | goto cleanup; | |
11328 | } | |
11329 | ||
11330 | syms[cnt++] = name; | |
11331 | } | |
11332 | ||
11333 | /* no entries found, bail out */ | |
11334 | if (cnt == 0) { | |
11335 | err = -ENOENT; | |
11336 | goto cleanup; | |
11337 | } | |
11338 | ||
11339 | /* sort available functions */ | |
11340 | qsort(syms, cnt, sizeof(*syms), avail_func_cmp); | |
11341 | ||
11342 | data.syms = syms; | |
11343 | data.res = res; | |
11344 | data.cnt = cnt; | |
11345 | libbpf_kallsyms_parse(avail_kallsyms_cb, &data); | |
11346 | ||
11347 | if (res->cnt == 0) | |
11348 | err = -ENOENT; | |
11349 | ||
11350 | cleanup: | |
11351 | for (i = 0; i < cnt; i++) | |
11352 | free((char *)syms[i]); | |
11353 | free(syms); | |
11354 | ||
11355 | fclose(f); | |
11356 | return err; | |
11357 | } | |
11358 | ||
56baeeba JL |
11359 | static bool has_available_filter_functions_addrs(void) |
11360 | { | |
11361 | return access(tracefs_available_filter_functions_addrs(), R_OK) != -1; | |
11362 | } | |
11363 | ||
11364 | static int libbpf_available_kprobes_parse(struct kprobe_multi_resolve *res) | |
11365 | { | |
11366 | const char *available_path = tracefs_available_filter_functions_addrs(); | |
11367 | char sym_name[500]; | |
11368 | FILE *f; | |
11369 | int ret, err = 0; | |
11370 | unsigned long long sym_addr; | |
11371 | ||
11372 | f = fopen(available_path, "re"); | |
11373 | if (!f) { | |
11374 | err = -errno; | |
11375 | pr_warn("failed to open %s: %d\n", available_path, err); | |
11376 | return err; | |
11377 | } | |
11378 | ||
11379 | while (true) { | |
11380 | ret = fscanf(f, "%llx %499s%*[^\n]\n", &sym_addr, sym_name); | |
11381 | if (ret == EOF && feof(f)) | |
11382 | break; | |
11383 | ||
11384 | if (ret != 2) { | |
11385 | pr_warn("failed to parse available_filter_functions_addrs entry: %d\n", | |
11386 | ret); | |
11387 | err = -EINVAL; | |
11388 | goto cleanup; | |
11389 | } | |
11390 | ||
11391 | if (!glob_match(sym_name, res->pattern)) | |
11392 | continue; | |
11393 | ||
11394 | err = libbpf_ensure_mem((void **)&res->addrs, &res->cap, | |
11395 | sizeof(*res->addrs), res->cnt + 1); | |
11396 | if (err) | |
11397 | goto cleanup; | |
11398 | ||
11399 | res->addrs[res->cnt++] = (unsigned long)sym_addr; | |
11400 | } | |
11401 | ||
11402 | if (res->cnt == 0) | |
11403 | err = -ENOENT; | |
11404 | ||
11405 | cleanup: | |
11406 | fclose(f); | |
11407 | return err; | |
11408 | } | |
11409 | ||
ddc6b049 JO |
11410 | struct bpf_link * |
11411 | bpf_program__attach_kprobe_multi_opts(const struct bpf_program *prog, | |
11412 | const char *pattern, | |
11413 | const struct bpf_kprobe_multi_opts *opts) | |
11414 | { | |
11415 | LIBBPF_OPTS(bpf_link_create_opts, lopts); | |
11416 | struct kprobe_multi_resolve res = { | |
11417 | .pattern = pattern, | |
11418 | }; | |
2ca178f0 | 11419 | enum bpf_attach_type attach_type; |
ddc6b049 JO |
11420 | struct bpf_link *link = NULL; |
11421 | char errmsg[STRERR_BUFSIZE]; | |
11422 | const unsigned long *addrs; | |
11423 | int err, link_fd, prog_fd; | |
2ca178f0 | 11424 | bool retprobe, session; |
ddc6b049 JO |
11425 | const __u64 *cookies; |
11426 | const char **syms; | |
ddc6b049 JO |
11427 | size_t cnt; |
11428 | ||
11429 | if (!OPTS_VALID(opts, bpf_kprobe_multi_opts)) | |
11430 | return libbpf_err_ptr(-EINVAL); | |
11431 | ||
7b30c296 MY |
11432 | prog_fd = bpf_program__fd(prog); |
11433 | if (prog_fd < 0) { | |
11434 | pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n", | |
11435 | prog->name); | |
11436 | return libbpf_err_ptr(-EINVAL); | |
11437 | } | |
11438 | ||
ddc6b049 JO |
11439 | syms = OPTS_GET(opts, syms, false); |
11440 | addrs = OPTS_GET(opts, addrs, false); | |
11441 | cnt = OPTS_GET(opts, cnt, false); | |
11442 | cookies = OPTS_GET(opts, cookies, false); | |
11443 | ||
11444 | if (!pattern && !addrs && !syms) | |
11445 | return libbpf_err_ptr(-EINVAL); | |
11446 | if (pattern && (addrs || syms || cookies || cnt)) | |
11447 | return libbpf_err_ptr(-EINVAL); | |
11448 | if (!pattern && !cnt) | |
11449 | return libbpf_err_ptr(-EINVAL); | |
11450 | if (addrs && syms) | |
11451 | return libbpf_err_ptr(-EINVAL); | |
11452 | ||
11453 | if (pattern) { | |
56baeeba JL |
11454 | if (has_available_filter_functions_addrs()) |
11455 | err = libbpf_available_kprobes_parse(&res); | |
11456 | else | |
11457 | err = libbpf_available_kallsyms_parse(&res); | |
ddc6b049 JO |
11458 | if (err) |
11459 | goto error; | |
ddc6b049 JO |
11460 | addrs = res.addrs; |
11461 | cnt = res.cnt; | |
11462 | } | |
11463 | ||
11464 | retprobe = OPTS_GET(opts, retprobe, false); | |
2ca178f0 JO |
11465 | session = OPTS_GET(opts, session, false); |
11466 | ||
11467 | if (retprobe && session) | |
11468 | return libbpf_err_ptr(-EINVAL); | |
11469 | ||
11470 | attach_type = session ? BPF_TRACE_KPROBE_SESSION : BPF_TRACE_KPROBE_MULTI; | |
ddc6b049 JO |
11471 | |
11472 | lopts.kprobe_multi.syms = syms; | |
11473 | lopts.kprobe_multi.addrs = addrs; | |
11474 | lopts.kprobe_multi.cookies = cookies; | |
11475 | lopts.kprobe_multi.cnt = cnt; | |
11476 | lopts.kprobe_multi.flags = retprobe ? BPF_F_KPROBE_MULTI_RETURN : 0; | |
11477 | ||
11478 | link = calloc(1, sizeof(*link)); | |
11479 | if (!link) { | |
11480 | err = -ENOMEM; | |
11481 | goto error; | |
11482 | } | |
11483 | link->detach = &bpf_link__detach_fd; | |
11484 | ||
2ca178f0 | 11485 | link_fd = bpf_link_create(prog_fd, 0, attach_type, &lopts); |
ddc6b049 JO |
11486 | if (link_fd < 0) { |
11487 | err = -errno; | |
11488 | pr_warn("prog '%s': failed to attach: %s\n", | |
11489 | prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
11490 | goto error; | |
11491 | } | |
11492 | link->fd = link_fd; | |
11493 | free(res.addrs); | |
11494 | return link; | |
11495 | ||
11496 | error: | |
11497 | free(link); | |
11498 | free(res.addrs); | |
11499 | return libbpf_err_ptr(err); | |
11500 | } | |
11501 | ||
4fa5bcfe | 11502 | static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 11503 | { |
da97553e | 11504 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts); |
a2488b5f | 11505 | unsigned long offset = 0; |
d7a18ea7 | 11506 | const char *func_name; |
a2488b5f | 11507 | char *func; |
4fa5bcfe | 11508 | int n; |
d7a18ea7 | 11509 | |
9af8efc4 AN |
11510 | *link = NULL; |
11511 | ||
11512 | /* no auto-attach for SEC("kprobe") and SEC("kretprobe") */ | |
11513 | if (strcmp(prog->sec_name, "kprobe") == 0 || strcmp(prog->sec_name, "kretprobe") == 0) | |
11514 | return 0; | |
11515 | ||
13d35a0c AN |
11516 | opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/"); |
11517 | if (opts.retprobe) | |
11518 | func_name = prog->sec_name + sizeof("kretprobe/") - 1; | |
11519 | else | |
11520 | func_name = prog->sec_name + sizeof("kprobe/") - 1; | |
d7a18ea7 | 11521 | |
e3f9bc35 | 11522 | n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset); |
a2488b5f | 11523 | if (n < 1) { |
a2488b5f | 11524 | pr_warn("kprobe name is invalid: %s\n", func_name); |
4fa5bcfe | 11525 | return -EINVAL; |
a2488b5f AM |
11526 | } |
11527 | if (opts.retprobe && offset != 0) { | |
1f71a468 | 11528 | free(func); |
a2488b5f | 11529 | pr_warn("kretprobes do not support offset specification\n"); |
4fa5bcfe | 11530 | return -EINVAL; |
a2488b5f | 11531 | } |
d7a18ea7 | 11532 | |
a2488b5f | 11533 | opts.offset = offset; |
4fa5bcfe | 11534 | *link = bpf_program__attach_kprobe_opts(prog, func, &opts); |
a2488b5f | 11535 | free(func); |
4fa5bcfe | 11536 | return libbpf_get_error(*link); |
d7a18ea7 AN |
11537 | } |
11538 | ||
708ac5be AN |
11539 | static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
11540 | { | |
11541 | LIBBPF_OPTS(bpf_ksyscall_opts, opts); | |
11542 | const char *syscall_name; | |
11543 | ||
11544 | *link = NULL; | |
11545 | ||
11546 | /* no auto-attach for SEC("ksyscall") and SEC("kretsyscall") */ | |
11547 | if (strcmp(prog->sec_name, "ksyscall") == 0 || strcmp(prog->sec_name, "kretsyscall") == 0) | |
11548 | return 0; | |
11549 | ||
11550 | opts.retprobe = str_has_pfx(prog->sec_name, "kretsyscall/"); | |
11551 | if (opts.retprobe) | |
11552 | syscall_name = prog->sec_name + sizeof("kretsyscall/") - 1; | |
11553 | else | |
11554 | syscall_name = prog->sec_name + sizeof("ksyscall/") - 1; | |
11555 | ||
11556 | *link = bpf_program__attach_ksyscall(prog, syscall_name, &opts); | |
11557 | return *link ? 0 : -errno; | |
11558 | } | |
11559 | ||
ddc6b049 JO |
11560 | static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
11561 | { | |
11562 | LIBBPF_OPTS(bpf_kprobe_multi_opts, opts); | |
11563 | const char *spec; | |
11564 | char *pattern; | |
11565 | int n; | |
11566 | ||
9af8efc4 AN |
11567 | *link = NULL; |
11568 | ||
11569 | /* no auto-attach for SEC("kprobe.multi") and SEC("kretprobe.multi") */ | |
11570 | if (strcmp(prog->sec_name, "kprobe.multi") == 0 || | |
11571 | strcmp(prog->sec_name, "kretprobe.multi") == 0) | |
11572 | return 0; | |
11573 | ||
ddc6b049 JO |
11574 | opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe.multi/"); |
11575 | if (opts.retprobe) | |
11576 | spec = prog->sec_name + sizeof("kretprobe.multi/") - 1; | |
11577 | else | |
11578 | spec = prog->sec_name + sizeof("kprobe.multi/") - 1; | |
11579 | ||
11580 | n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern); | |
11581 | if (n < 1) { | |
7c13ef16 | 11582 | pr_warn("kprobe multi pattern is invalid: %s\n", spec); |
ddc6b049 JO |
11583 | return -EINVAL; |
11584 | } | |
11585 | ||
11586 | *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts); | |
11587 | free(pattern); | |
11588 | return libbpf_get_error(*link); | |
11589 | } | |
11590 | ||
2ca178f0 JO |
11591 | static int attach_kprobe_session(const struct bpf_program *prog, long cookie, |
11592 | struct bpf_link **link) | |
11593 | { | |
11594 | LIBBPF_OPTS(bpf_kprobe_multi_opts, opts, .session = true); | |
11595 | const char *spec; | |
11596 | char *pattern; | |
11597 | int n; | |
11598 | ||
11599 | *link = NULL; | |
11600 | ||
11601 | /* no auto-attach for SEC("kprobe.session") */ | |
11602 | if (strcmp(prog->sec_name, "kprobe.session") == 0) | |
11603 | return 0; | |
11604 | ||
11605 | spec = prog->sec_name + sizeof("kprobe.session/") - 1; | |
11606 | n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern); | |
11607 | if (n < 1) { | |
5a3941f8 | 11608 | pr_warn("kprobe session pattern is invalid: %s\n", spec); |
2ca178f0 JO |
11609 | return -EINVAL; |
11610 | } | |
11611 | ||
11612 | *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts); | |
11613 | free(pattern); | |
11614 | return *link ? 0 : -errno; | |
11615 | } | |
11616 | ||
5bfdd32d JO |
11617 | static int attach_uprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
11618 | { | |
11619 | char *probe_type = NULL, *binary_path = NULL, *func_name = NULL; | |
11620 | LIBBPF_OPTS(bpf_uprobe_multi_opts, opts); | |
11621 | int n, ret = -EINVAL; | |
11622 | ||
11623 | *link = NULL; | |
11624 | ||
2147c8d0 | 11625 | n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[^\n]", |
5bfdd32d JO |
11626 | &probe_type, &binary_path, &func_name); |
11627 | switch (n) { | |
11628 | case 1: | |
11629 | /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */ | |
11630 | ret = 0; | |
11631 | break; | |
11632 | case 3: | |
11633 | opts.retprobe = strcmp(probe_type, "uretprobe.multi") == 0; | |
11634 | *link = bpf_program__attach_uprobe_multi(prog, -1, binary_path, func_name, &opts); | |
11635 | ret = libbpf_get_error(*link); | |
11636 | break; | |
11637 | default: | |
11638 | pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name, | |
11639 | prog->sec_name); | |
11640 | break; | |
11641 | } | |
11642 | free(probe_type); | |
11643 | free(binary_path); | |
11644 | free(func_name); | |
11645 | return ret; | |
11646 | } | |
11647 | ||
cc10623c AN |
11648 | static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz, |
11649 | const char *binary_path, uint64_t offset) | |
11650 | { | |
11651 | int i; | |
11652 | ||
11653 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset); | |
11654 | ||
11655 | /* sanitize binary_path in the probe name */ | |
11656 | for (i = 0; buf[i]; i++) { | |
11657 | if (!isalnum(buf[i])) | |
11658 | buf[i] = '_'; | |
11659 | } | |
11660 | } | |
11661 | ||
11662 | static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe, | |
11663 | const char *binary_path, size_t offset) | |
11664 | { | |
a1ac9fd6 | 11665 | return append_to_file(tracefs_uprobe_events(), "%c:%s/%s %s:0x%zx", |
cc10623c AN |
11666 | retprobe ? 'r' : 'p', |
11667 | retprobe ? "uretprobes" : "uprobes", | |
11668 | probe_name, binary_path, offset); | |
11669 | } | |
11670 | ||
11671 | static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe) | |
11672 | { | |
a1ac9fd6 AN |
11673 | return append_to_file(tracefs_uprobe_events(), "-:%s/%s", |
11674 | retprobe ? "uretprobes" : "uprobes", probe_name); | |
cc10623c AN |
11675 | } |
11676 | ||
11677 | static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
11678 | { | |
11679 | char file[512]; | |
11680 | ||
a1ac9fd6 AN |
11681 | snprintf(file, sizeof(file), "%s/events/%s/%s/id", |
11682 | tracefs_path(), retprobe ? "uretprobes" : "uprobes", probe_name); | |
cc10623c AN |
11683 | |
11684 | return parse_uint_from_file(file, "%d\n"); | |
11685 | } | |
11686 | ||
11687 | static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe, | |
11688 | const char *binary_path, size_t offset, int pid) | |
11689 | { | |
813847a3 | 11690 | const size_t attr_sz = sizeof(struct perf_event_attr); |
cc10623c AN |
11691 | struct perf_event_attr attr; |
11692 | int type, pfd, err; | |
11693 | ||
11694 | err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset); | |
11695 | if (err < 0) { | |
11696 | pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n", | |
11697 | binary_path, (size_t)offset, err); | |
11698 | return err; | |
11699 | } | |
11700 | type = determine_uprobe_perf_type_legacy(probe_name, retprobe); | |
11701 | if (type < 0) { | |
2655144f | 11702 | err = type; |
cc10623c | 11703 | pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n", |
2655144f CW |
11704 | binary_path, offset, err); |
11705 | goto err_clean_legacy; | |
cc10623c AN |
11706 | } |
11707 | ||
813847a3 AN |
11708 | memset(&attr, 0, attr_sz); |
11709 | attr.size = attr_sz; | |
cc10623c AN |
11710 | attr.config = type; |
11711 | attr.type = PERF_TYPE_TRACEPOINT; | |
11712 | ||
11713 | pfd = syscall(__NR_perf_event_open, &attr, | |
11714 | pid < 0 ? -1 : pid, /* pid */ | |
11715 | pid == -1 ? 0 : -1, /* cpu */ | |
11716 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
11717 | if (pfd < 0) { | |
11718 | err = -errno; | |
11719 | pr_warn("legacy uprobe perf_event_open() failed: %d\n", err); | |
2655144f | 11720 | goto err_clean_legacy; |
cc10623c AN |
11721 | } |
11722 | return pfd; | |
2655144f CW |
11723 | |
11724 | err_clean_legacy: | |
11725 | /* Clear the newly added legacy uprobe_event */ | |
11726 | remove_uprobe_event_legacy(probe_name, retprobe); | |
11727 | return err; | |
cc10623c AN |
11728 | } |
11729 | ||
c44fd845 DM |
11730 | /* Find offset of function name in archive specified by path. Currently |
11731 | * supported are .zip files that do not compress their contents, as used on | |
11732 | * Android in the form of APKs, for example. "file_name" is the name of the ELF | |
11733 | * file inside the archive. "func_name" matches symbol name or name@@LIB for | |
11734 | * library functions. | |
11735 | * | |
11736 | * An overview of the APK format specifically provided here: | |
11737 | * https://en.wikipedia.org/w/index.php?title=Apk_(file_format)&oldid=1139099120#Package_contents | |
11738 | */ | |
11739 | static long elf_find_func_offset_from_archive(const char *archive_path, const char *file_name, | |
11740 | const char *func_name) | |
11741 | { | |
11742 | struct zip_archive *archive; | |
11743 | struct zip_entry entry; | |
11744 | long ret; | |
11745 | Elf *elf; | |
11746 | ||
11747 | archive = zip_archive_open(archive_path); | |
11748 | if (IS_ERR(archive)) { | |
11749 | ret = PTR_ERR(archive); | |
11750 | pr_warn("zip: failed to open %s: %ld\n", archive_path, ret); | |
11751 | return ret; | |
11752 | } | |
11753 | ||
11754 | ret = zip_archive_find_entry(archive, file_name, &entry); | |
11755 | if (ret) { | |
11756 | pr_warn("zip: could not find archive member %s in %s: %ld\n", file_name, | |
11757 | archive_path, ret); | |
11758 | goto out; | |
11759 | } | |
11760 | pr_debug("zip: found entry for %s in %s at 0x%lx\n", file_name, archive_path, | |
11761 | (unsigned long)entry.data_offset); | |
11762 | ||
11763 | if (entry.compression) { | |
11764 | pr_warn("zip: entry %s of %s is compressed and cannot be handled\n", file_name, | |
11765 | archive_path); | |
11766 | ret = -LIBBPF_ERRNO__FORMAT; | |
11767 | goto out; | |
11768 | } | |
11769 | ||
11770 | elf = elf_memory((void *)entry.data, entry.data_length); | |
11771 | if (!elf) { | |
11772 | pr_warn("elf: could not read elf file %s from %s: %s\n", file_name, archive_path, | |
11773 | elf_errmsg(-1)); | |
11774 | ret = -LIBBPF_ERRNO__LIBELF; | |
11775 | goto out; | |
11776 | } | |
11777 | ||
11778 | ret = elf_find_func_offset(elf, file_name, func_name); | |
11779 | if (ret > 0) { | |
11780 | pr_debug("elf: symbol address match for %s of %s in %s: 0x%x + 0x%lx = 0x%lx\n", | |
11781 | func_name, file_name, archive_path, entry.data_offset, ret, | |
11782 | ret + entry.data_offset); | |
11783 | ret += entry.data_offset; | |
11784 | } | |
11785 | elf_end(elf); | |
11786 | ||
11787 | out: | |
11788 | zip_archive_close(archive); | |
11789 | return ret; | |
11790 | } | |
11791 | ||
56818931 IL |
11792 | static const char *arch_specific_lib_paths(void) |
11793 | { | |
11794 | /* | |
11795 | * Based on https://packages.debian.org/sid/libc6. | |
11796 | * | |
11797 | * Assume that the traced program is built for the same architecture | |
11798 | * as libbpf, which should cover the vast majority of cases. | |
11799 | */ | |
11800 | #if defined(__x86_64__) | |
11801 | return "/lib/x86_64-linux-gnu"; | |
11802 | #elif defined(__i386__) | |
11803 | return "/lib/i386-linux-gnu"; | |
11804 | #elif defined(__s390x__) | |
11805 | return "/lib/s390x-linux-gnu"; | |
11806 | #elif defined(__s390__) | |
11807 | return "/lib/s390-linux-gnu"; | |
11808 | #elif defined(__arm__) && defined(__SOFTFP__) | |
11809 | return "/lib/arm-linux-gnueabi"; | |
11810 | #elif defined(__arm__) && !defined(__SOFTFP__) | |
11811 | return "/lib/arm-linux-gnueabihf"; | |
11812 | #elif defined(__aarch64__) | |
11813 | return "/lib/aarch64-linux-gnu"; | |
11814 | #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 64 | |
11815 | return "/lib/mips64el-linux-gnuabi64"; | |
11816 | #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 32 | |
11817 | return "/lib/mipsel-linux-gnu"; | |
11818 | #elif defined(__powerpc64__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ | |
11819 | return "/lib/powerpc64le-linux-gnu"; | |
11820 | #elif defined(__sparc__) && defined(__arch64__) | |
11821 | return "/lib/sparc64-linux-gnu"; | |
11822 | #elif defined(__riscv) && __riscv_xlen == 64 | |
11823 | return "/lib/riscv64-linux-gnu"; | |
11824 | #else | |
11825 | return NULL; | |
11826 | #endif | |
11827 | } | |
11828 | ||
1ce3a60e AM |
11829 | /* Get full path to program/shared library. */ |
11830 | static int resolve_full_path(const char *file, char *result, size_t result_sz) | |
11831 | { | |
56818931 | 11832 | const char *search_paths[3] = {}; |
9e32084e | 11833 | int i, perm; |
1ce3a60e | 11834 | |
a1c9d61b | 11835 | if (str_has_sfx(file, ".so") || strstr(file, ".so.")) { |
1ce3a60e AM |
11836 | search_paths[0] = getenv("LD_LIBRARY_PATH"); |
11837 | search_paths[1] = "/usr/lib64:/usr/lib"; | |
56818931 | 11838 | search_paths[2] = arch_specific_lib_paths(); |
9e32084e | 11839 | perm = R_OK; |
1ce3a60e AM |
11840 | } else { |
11841 | search_paths[0] = getenv("PATH"); | |
11842 | search_paths[1] = "/usr/bin:/usr/sbin"; | |
9e32084e | 11843 | perm = R_OK | X_OK; |
1ce3a60e AM |
11844 | } |
11845 | ||
11846 | for (i = 0; i < ARRAY_SIZE(search_paths); i++) { | |
11847 | const char *s; | |
11848 | ||
11849 | if (!search_paths[i]) | |
11850 | continue; | |
11851 | for (s = search_paths[i]; s != NULL; s = strchr(s, ':')) { | |
11852 | char *next_path; | |
11853 | int seg_len; | |
11854 | ||
11855 | if (s[0] == ':') | |
11856 | s++; | |
11857 | next_path = strchr(s, ':'); | |
11858 | seg_len = next_path ? next_path - s : strlen(s); | |
11859 | if (!seg_len) | |
11860 | continue; | |
11861 | snprintf(result, result_sz, "%.*s/%s", seg_len, s, file); | |
9e32084e | 11862 | /* ensure it has required permissions */ |
6a4ab886 | 11863 | if (faccessat(AT_FDCWD, result, perm, AT_EACCESS) < 0) |
1ce3a60e AM |
11864 | continue; |
11865 | pr_debug("resolved '%s' to '%s'\n", file, result); | |
11866 | return 0; | |
11867 | } | |
11868 | } | |
11869 | return -ENOENT; | |
11870 | } | |
11871 | ||
3140cf12 JO |
11872 | struct bpf_link * |
11873 | bpf_program__attach_uprobe_multi(const struct bpf_program *prog, | |
11874 | pid_t pid, | |
11875 | const char *path, | |
11876 | const char *func_pattern, | |
11877 | const struct bpf_uprobe_multi_opts *opts) | |
11878 | { | |
11879 | const unsigned long *ref_ctr_offsets = NULL, *offsets = NULL; | |
11880 | LIBBPF_OPTS(bpf_link_create_opts, lopts); | |
11881 | unsigned long *resolved_offsets = NULL; | |
11882 | int err = 0, link_fd, prog_fd; | |
11883 | struct bpf_link *link = NULL; | |
11884 | char errmsg[STRERR_BUFSIZE]; | |
11885 | char full_path[PATH_MAX]; | |
11886 | const __u64 *cookies; | |
11887 | const char **syms; | |
11888 | size_t cnt; | |
11889 | ||
11890 | if (!OPTS_VALID(opts, bpf_uprobe_multi_opts)) | |
11891 | return libbpf_err_ptr(-EINVAL); | |
11892 | ||
7b30c296 MY |
11893 | prog_fd = bpf_program__fd(prog); |
11894 | if (prog_fd < 0) { | |
11895 | pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n", | |
11896 | prog->name); | |
11897 | return libbpf_err_ptr(-EINVAL); | |
11898 | } | |
11899 | ||
3140cf12 JO |
11900 | syms = OPTS_GET(opts, syms, NULL); |
11901 | offsets = OPTS_GET(opts, offsets, NULL); | |
11902 | ref_ctr_offsets = OPTS_GET(opts, ref_ctr_offsets, NULL); | |
11903 | cookies = OPTS_GET(opts, cookies, NULL); | |
11904 | cnt = OPTS_GET(opts, cnt, 0); | |
11905 | ||
11906 | /* | |
11907 | * User can specify 2 mutually exclusive set of inputs: | |
11908 | * | |
11909 | * 1) use only path/func_pattern/pid arguments | |
11910 | * | |
11911 | * 2) use path/pid with allowed combinations of: | |
11912 | * syms/offsets/ref_ctr_offsets/cookies/cnt | |
11913 | * | |
11914 | * - syms and offsets are mutually exclusive | |
11915 | * - ref_ctr_offsets and cookies are optional | |
11916 | * | |
11917 | * Any other usage results in error. | |
11918 | */ | |
11919 | ||
11920 | if (!path) | |
11921 | return libbpf_err_ptr(-EINVAL); | |
11922 | if (!func_pattern && cnt == 0) | |
11923 | return libbpf_err_ptr(-EINVAL); | |
11924 | ||
11925 | if (func_pattern) { | |
11926 | if (syms || offsets || ref_ctr_offsets || cookies || cnt) | |
11927 | return libbpf_err_ptr(-EINVAL); | |
11928 | } else { | |
11929 | if (!!syms == !!offsets) | |
11930 | return libbpf_err_ptr(-EINVAL); | |
11931 | } | |
11932 | ||
11933 | if (func_pattern) { | |
11934 | if (!strchr(path, '/')) { | |
11935 | err = resolve_full_path(path, full_path, sizeof(full_path)); | |
11936 | if (err) { | |
11937 | pr_warn("prog '%s': failed to resolve full path for '%s': %d\n", | |
11938 | prog->name, path, err); | |
11939 | return libbpf_err_ptr(err); | |
11940 | } | |
11941 | path = full_path; | |
11942 | } | |
11943 | ||
11944 | err = elf_resolve_pattern_offsets(path, func_pattern, | |
11945 | &resolved_offsets, &cnt); | |
11946 | if (err < 0) | |
11947 | return libbpf_err_ptr(err); | |
11948 | offsets = resolved_offsets; | |
11949 | } else if (syms) { | |
48f0dfd8 | 11950 | err = elf_resolve_syms_offsets(path, cnt, syms, &resolved_offsets, STT_FUNC); |
3140cf12 JO |
11951 | if (err < 0) |
11952 | return libbpf_err_ptr(err); | |
11953 | offsets = resolved_offsets; | |
11954 | } | |
11955 | ||
11956 | lopts.uprobe_multi.path = path; | |
11957 | lopts.uprobe_multi.offsets = offsets; | |
11958 | lopts.uprobe_multi.ref_ctr_offsets = ref_ctr_offsets; | |
11959 | lopts.uprobe_multi.cookies = cookies; | |
11960 | lopts.uprobe_multi.cnt = cnt; | |
11961 | lopts.uprobe_multi.flags = OPTS_GET(opts, retprobe, false) ? BPF_F_UPROBE_MULTI_RETURN : 0; | |
11962 | ||
11963 | if (pid == 0) | |
11964 | pid = getpid(); | |
11965 | if (pid > 0) | |
11966 | lopts.uprobe_multi.pid = pid; | |
11967 | ||
11968 | link = calloc(1, sizeof(*link)); | |
11969 | if (!link) { | |
11970 | err = -ENOMEM; | |
11971 | goto error; | |
11972 | } | |
11973 | link->detach = &bpf_link__detach_fd; | |
11974 | ||
3140cf12 JO |
11975 | link_fd = bpf_link_create(prog_fd, 0, BPF_TRACE_UPROBE_MULTI, &lopts); |
11976 | if (link_fd < 0) { | |
11977 | err = -errno; | |
11978 | pr_warn("prog '%s': failed to attach multi-uprobe: %s\n", | |
11979 | prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
11980 | goto error; | |
11981 | } | |
11982 | link->fd = link_fd; | |
11983 | free(resolved_offsets); | |
11984 | return link; | |
11985 | ||
11986 | error: | |
11987 | free(resolved_offsets); | |
11988 | free(link); | |
11989 | return libbpf_err_ptr(err); | |
11990 | } | |
11991 | ||
47faff37 | 11992 | LIBBPF_API struct bpf_link * |
942025c9 | 11993 | bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid, |
47faff37 AN |
11994 | const char *binary_path, size_t func_offset, |
11995 | const struct bpf_uprobe_opts *opts) | |
b2650027 | 11996 | { |
c44fd845 | 11997 | const char *archive_path = NULL, *archive_sep = NULL; |
cc10623c | 11998 | char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL; |
c44fd845 | 11999 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f8b299bc | 12000 | enum probe_attach_mode attach_mode; |
c44fd845 | 12001 | char full_path[PATH_MAX]; |
b2650027 | 12002 | struct bpf_link *link; |
5e3b8356 | 12003 | size_t ref_ctr_off; |
b2650027 | 12004 | int pfd, err; |
cc10623c | 12005 | bool retprobe, legacy; |
433966e3 | 12006 | const char *func_name; |
47faff37 AN |
12007 | |
12008 | if (!OPTS_VALID(opts, bpf_uprobe_opts)) | |
12009 | return libbpf_err_ptr(-EINVAL); | |
12010 | ||
f8b299bc | 12011 | attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT); |
47faff37 | 12012 | retprobe = OPTS_GET(opts, retprobe, false); |
5e3b8356 | 12013 | ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0); |
47faff37 | 12014 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
b2650027 | 12015 | |
8ed2f5a6 HC |
12016 | if (!binary_path) |
12017 | return libbpf_err_ptr(-EINVAL); | |
12018 | ||
c44fd845 DM |
12019 | /* Check if "binary_path" refers to an archive. */ |
12020 | archive_sep = strstr(binary_path, "!/"); | |
12021 | if (archive_sep) { | |
12022 | full_path[0] = '\0'; | |
12023 | libbpf_strlcpy(full_path, binary_path, | |
12024 | min(sizeof(full_path), (size_t)(archive_sep - binary_path + 1))); | |
12025 | archive_path = full_path; | |
12026 | binary_path = archive_sep + 2; | |
12027 | } else if (!strchr(binary_path, '/')) { | |
12028 | err = resolve_full_path(binary_path, full_path, sizeof(full_path)); | |
1ce3a60e | 12029 | if (err) { |
2e4913e0 AN |
12030 | pr_warn("prog '%s': failed to resolve full path for '%s': %d\n", |
12031 | prog->name, binary_path, err); | |
1ce3a60e AM |
12032 | return libbpf_err_ptr(err); |
12033 | } | |
c44fd845 | 12034 | binary_path = full_path; |
1ce3a60e | 12035 | } |
433966e3 AM |
12036 | func_name = OPTS_GET(opts, func_name, NULL); |
12037 | if (func_name) { | |
12038 | long sym_off; | |
12039 | ||
c44fd845 DM |
12040 | if (archive_path) { |
12041 | sym_off = elf_find_func_offset_from_archive(archive_path, binary_path, | |
12042 | func_name); | |
12043 | binary_path = archive_path; | |
12044 | } else { | |
12045 | sym_off = elf_find_func_offset_from_file(binary_path, func_name); | |
12046 | } | |
433966e3 AM |
12047 | if (sym_off < 0) |
12048 | return libbpf_err_ptr(sym_off); | |
12049 | func_offset += sym_off; | |
12050 | } | |
1ce3a60e | 12051 | |
cc10623c | 12052 | legacy = determine_uprobe_perf_type() < 0; |
f8b299bc MD |
12053 | switch (attach_mode) { |
12054 | case PROBE_ATTACH_MODE_LEGACY: | |
12055 | legacy = true; | |
12056 | pe_opts.force_ioctl_attach = true; | |
12057 | break; | |
12058 | case PROBE_ATTACH_MODE_PERF: | |
12059 | if (legacy) | |
12060 | return libbpf_err_ptr(-ENOTSUP); | |
12061 | pe_opts.force_ioctl_attach = true; | |
12062 | break; | |
12063 | case PROBE_ATTACH_MODE_LINK: | |
12064 | if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK)) | |
12065 | return libbpf_err_ptr(-ENOTSUP); | |
12066 | break; | |
12067 | case PROBE_ATTACH_MODE_DEFAULT: | |
12068 | break; | |
12069 | default: | |
12070 | return libbpf_err_ptr(-EINVAL); | |
12071 | } | |
12072 | ||
cc10623c AN |
12073 | if (!legacy) { |
12074 | pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path, | |
12075 | func_offset, pid, ref_ctr_off); | |
12076 | } else { | |
1ce3a60e | 12077 | char probe_name[PATH_MAX + 64]; |
cc10623c AN |
12078 | |
12079 | if (ref_ctr_off) | |
12080 | return libbpf_err_ptr(-EINVAL); | |
12081 | ||
12082 | gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
12083 | binary_path, func_offset); | |
12084 | ||
12085 | legacy_probe = strdup(probe_name); | |
12086 | if (!legacy_probe) | |
12087 | return libbpf_err_ptr(-ENOMEM); | |
12088 | ||
12089 | pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe, | |
12090 | binary_path, func_offset, pid); | |
12091 | } | |
b2650027 | 12092 | if (pfd < 0) { |
cc10623c | 12093 | err = -errno; |
52109584 AN |
12094 | pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n", |
12095 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e | 12096 | binary_path, func_offset, |
cc10623c AN |
12097 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
12098 | goto err_out; | |
b2650027 | 12099 | } |
cc10623c | 12100 | |
47faff37 | 12101 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
12102 | err = libbpf_get_error(link); |
12103 | if (err) { | |
b2650027 | 12104 | close(pfd); |
52109584 AN |
12105 | pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n", |
12106 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e KW |
12107 | binary_path, func_offset, |
12108 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
2655144f | 12109 | goto err_clean_legacy; |
cc10623c AN |
12110 | } |
12111 | if (legacy) { | |
12112 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
12113 | ||
12114 | perf_link->legacy_probe_name = legacy_probe; | |
12115 | perf_link->legacy_is_kprobe = false; | |
12116 | perf_link->legacy_is_retprobe = retprobe; | |
b2650027 AN |
12117 | } |
12118 | return link; | |
2655144f CW |
12119 | |
12120 | err_clean_legacy: | |
12121 | if (legacy) | |
12122 | remove_uprobe_event_legacy(legacy_probe, retprobe); | |
cc10623c AN |
12123 | err_out: |
12124 | free(legacy_probe); | |
12125 | return libbpf_err_ptr(err); | |
b2650027 AN |
12126 | } |
12127 | ||
39f8dc43 AM |
12128 | /* Format of u[ret]probe section definition supporting auto-attach: |
12129 | * u[ret]probe/binary:function[+offset] | |
12130 | * | |
12131 | * binary can be an absolute/relative path or a filename; the latter is resolved to a | |
12132 | * full binary path via bpf_program__attach_uprobe_opts. | |
12133 | * | |
12134 | * Specifying uprobe+ ensures we carry out strict matching; either "uprobe" must be | |
12135 | * specified (and auto-attach is not possible) or the above format is specified for | |
12136 | * auto-attach. | |
12137 | */ | |
12138 | static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link) | |
12139 | { | |
12140 | DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts); | |
2147c8d0 HC |
12141 | char *probe_type = NULL, *binary_path = NULL, *func_name = NULL, *func_off; |
12142 | int n, c, ret = -EINVAL; | |
90db26e6 | 12143 | long offset = 0; |
39f8dc43 AM |
12144 | |
12145 | *link = NULL; | |
12146 | ||
2147c8d0 HC |
12147 | n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[^\n]", |
12148 | &probe_type, &binary_path, &func_name); | |
90db26e6 AM |
12149 | switch (n) { |
12150 | case 1: | |
12151 | /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */ | |
12152 | ret = 0; | |
12153 | break; | |
12154 | case 2: | |
12155 | pr_warn("prog '%s': section '%s' missing ':function[+offset]' specification\n", | |
12156 | prog->name, prog->sec_name); | |
12157 | break; | |
12158 | case 3: | |
2147c8d0 HC |
12159 | /* check if user specifies `+offset`, if yes, this should be |
12160 | * the last part of the string, make sure sscanf read to EOL | |
12161 | */ | |
12162 | func_off = strrchr(func_name, '+'); | |
12163 | if (func_off) { | |
12164 | n = sscanf(func_off, "+%li%n", &offset, &c); | |
12165 | if (n == 1 && *(func_off + c) == '\0') | |
12166 | func_off[0] = '\0'; | |
12167 | else | |
12168 | offset = 0; | |
12169 | } | |
c4cac71f DK |
12170 | opts.retprobe = strcmp(probe_type, "uretprobe") == 0 || |
12171 | strcmp(probe_type, "uretprobe.s") == 0; | |
90db26e6 AM |
12172 | if (opts.retprobe && offset != 0) { |
12173 | pr_warn("prog '%s': uretprobes do not support offset specification\n", | |
12174 | prog->name); | |
12175 | break; | |
12176 | } | |
12177 | opts.func_name = func_name; | |
12178 | *link = bpf_program__attach_uprobe_opts(prog, -1, binary_path, offset, &opts); | |
12179 | ret = libbpf_get_error(*link); | |
12180 | break; | |
12181 | default: | |
12182 | pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name, | |
39f8dc43 | 12183 | prog->sec_name); |
90db26e6 | 12184 | break; |
39f8dc43 | 12185 | } |
90db26e6 AM |
12186 | free(probe_type); |
12187 | free(binary_path); | |
12188 | free(func_name); | |
39f8dc43 | 12189 | |
90db26e6 | 12190 | return ret; |
39f8dc43 AM |
12191 | } |
12192 | ||
942025c9 | 12193 | struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog, |
47faff37 AN |
12194 | bool retprobe, pid_t pid, |
12195 | const char *binary_path, | |
12196 | size_t func_offset) | |
12197 | { | |
12198 | DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe); | |
12199 | ||
12200 | return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts); | |
12201 | } | |
12202 | ||
2e4913e0 AN |
12203 | struct bpf_link *bpf_program__attach_usdt(const struct bpf_program *prog, |
12204 | pid_t pid, const char *binary_path, | |
12205 | const char *usdt_provider, const char *usdt_name, | |
12206 | const struct bpf_usdt_opts *opts) | |
12207 | { | |
12208 | char resolved_path[512]; | |
12209 | struct bpf_object *obj = prog->obj; | |
12210 | struct bpf_link *link; | |
5af25a41 | 12211 | __u64 usdt_cookie; |
2e4913e0 AN |
12212 | int err; |
12213 | ||
12214 | if (!OPTS_VALID(opts, bpf_uprobe_opts)) | |
12215 | return libbpf_err_ptr(-EINVAL); | |
12216 | ||
12217 | if (bpf_program__fd(prog) < 0) { | |
7b30c296 | 12218 | pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n", |
2e4913e0 AN |
12219 | prog->name); |
12220 | return libbpf_err_ptr(-EINVAL); | |
12221 | } | |
12222 | ||
8ed2f5a6 HC |
12223 | if (!binary_path) |
12224 | return libbpf_err_ptr(-EINVAL); | |
12225 | ||
2e4913e0 AN |
12226 | if (!strchr(binary_path, '/')) { |
12227 | err = resolve_full_path(binary_path, resolved_path, sizeof(resolved_path)); | |
12228 | if (err) { | |
12229 | pr_warn("prog '%s': failed to resolve full path for '%s': %d\n", | |
12230 | prog->name, binary_path, err); | |
12231 | return libbpf_err_ptr(err); | |
12232 | } | |
12233 | binary_path = resolved_path; | |
12234 | } | |
12235 | ||
12236 | /* USDT manager is instantiated lazily on first USDT attach. It will | |
12237 | * be destroyed together with BPF object in bpf_object__close(). | |
12238 | */ | |
12239 | if (IS_ERR(obj->usdt_man)) | |
12240 | return libbpf_ptr(obj->usdt_man); | |
12241 | if (!obj->usdt_man) { | |
12242 | obj->usdt_man = usdt_manager_new(obj); | |
12243 | if (IS_ERR(obj->usdt_man)) | |
12244 | return libbpf_ptr(obj->usdt_man); | |
12245 | } | |
12246 | ||
12247 | usdt_cookie = OPTS_GET(opts, usdt_cookie, 0); | |
12248 | link = usdt_manager_attach_usdt(obj->usdt_man, prog, pid, binary_path, | |
e3ba8e4e | 12249 | usdt_provider, usdt_name, usdt_cookie); |
2e4913e0 AN |
12250 | err = libbpf_get_error(link); |
12251 | if (err) | |
12252 | return libbpf_err_ptr(err); | |
12253 | return link; | |
12254 | } | |
12255 | ||
12256 | static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link) | |
12257 | { | |
12258 | char *path = NULL, *provider = NULL, *name = NULL; | |
12259 | const char *sec_name; | |
12260 | int n, err; | |
12261 | ||
12262 | sec_name = bpf_program__section_name(prog); | |
12263 | if (strcmp(sec_name, "usdt") == 0) { | |
12264 | /* no auto-attach for just SEC("usdt") */ | |
12265 | *link = NULL; | |
12266 | return 0; | |
12267 | } | |
12268 | ||
12269 | n = sscanf(sec_name, "usdt/%m[^:]:%m[^:]:%m[^:]", &path, &provider, &name); | |
12270 | if (n != 3) { | |
12271 | pr_warn("invalid section '%s', expected SEC(\"usdt/<path>:<provider>:<name>\")\n", | |
12272 | sec_name); | |
12273 | err = -EINVAL; | |
12274 | } else { | |
12275 | *link = bpf_program__attach_usdt(prog, -1 /* any process */, path, | |
12276 | provider, name, NULL); | |
12277 | err = libbpf_get_error(*link); | |
12278 | } | |
12279 | free(path); | |
12280 | free(provider); | |
12281 | free(name); | |
12282 | return err; | |
12283 | } | |
12284 | ||
f6de59c1 AN |
12285 | static int determine_tracepoint_id(const char *tp_category, |
12286 | const char *tp_name) | |
12287 | { | |
12288 | char file[PATH_MAX]; | |
12289 | int ret; | |
12290 | ||
a1ac9fd6 AN |
12291 | ret = snprintf(file, sizeof(file), "%s/events/%s/%s/id", |
12292 | tracefs_path(), tp_category, tp_name); | |
f6de59c1 AN |
12293 | if (ret < 0) |
12294 | return -errno; | |
12295 | if (ret >= sizeof(file)) { | |
12296 | pr_debug("tracepoint %s/%s path is too long\n", | |
12297 | tp_category, tp_name); | |
12298 | return -E2BIG; | |
12299 | } | |
12300 | return parse_uint_from_file(file, "%d\n"); | |
12301 | } | |
12302 | ||
12303 | static int perf_event_open_tracepoint(const char *tp_category, | |
12304 | const char *tp_name) | |
12305 | { | |
813847a3 AN |
12306 | const size_t attr_sz = sizeof(struct perf_event_attr); |
12307 | struct perf_event_attr attr; | |
f6de59c1 AN |
12308 | char errmsg[STRERR_BUFSIZE]; |
12309 | int tp_id, pfd, err; | |
12310 | ||
12311 | tp_id = determine_tracepoint_id(tp_category, tp_name); | |
12312 | if (tp_id < 0) { | |
be18010e KW |
12313 | pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n", |
12314 | tp_category, tp_name, | |
12315 | libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
12316 | return tp_id; |
12317 | } | |
12318 | ||
813847a3 | 12319 | memset(&attr, 0, attr_sz); |
f6de59c1 | 12320 | attr.type = PERF_TYPE_TRACEPOINT; |
813847a3 | 12321 | attr.size = attr_sz; |
f6de59c1 AN |
12322 | attr.config = tp_id; |
12323 | ||
12324 | pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */, | |
12325 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
12326 | if (pfd < 0) { | |
12327 | err = -errno; | |
be18010e KW |
12328 | pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n", |
12329 | tp_category, tp_name, | |
12330 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
12331 | return err; |
12332 | } | |
12333 | return pfd; | |
12334 | } | |
12335 | ||
942025c9 | 12336 | struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog, |
47faff37 AN |
12337 | const char *tp_category, |
12338 | const char *tp_name, | |
12339 | const struct bpf_tracepoint_opts *opts) | |
f6de59c1 | 12340 | { |
47faff37 | 12341 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f6de59c1 AN |
12342 | char errmsg[STRERR_BUFSIZE]; |
12343 | struct bpf_link *link; | |
12344 | int pfd, err; | |
12345 | ||
47faff37 AN |
12346 | if (!OPTS_VALID(opts, bpf_tracepoint_opts)) |
12347 | return libbpf_err_ptr(-EINVAL); | |
12348 | ||
12349 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); | |
12350 | ||
f6de59c1 AN |
12351 | pfd = perf_event_open_tracepoint(tp_category, tp_name); |
12352 | if (pfd < 0) { | |
52109584 AN |
12353 | pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n", |
12354 | prog->name, tp_category, tp_name, | |
be18010e | 12355 | libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 12356 | return libbpf_err_ptr(pfd); |
f6de59c1 | 12357 | } |
47faff37 | 12358 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
12359 | err = libbpf_get_error(link); |
12360 | if (err) { | |
f6de59c1 | 12361 | close(pfd); |
52109584 AN |
12362 | pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n", |
12363 | prog->name, tp_category, tp_name, | |
be18010e | 12364 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 12365 | return libbpf_err_ptr(err); |
f6de59c1 AN |
12366 | } |
12367 | return link; | |
12368 | } | |
12369 | ||
942025c9 | 12370 | struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog, |
47faff37 AN |
12371 | const char *tp_category, |
12372 | const char *tp_name) | |
12373 | { | |
12374 | return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL); | |
12375 | } | |
12376 | ||
4fa5bcfe | 12377 | static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 AN |
12378 | { |
12379 | char *sec_name, *tp_cat, *tp_name; | |
d7a18ea7 | 12380 | |
9af8efc4 AN |
12381 | *link = NULL; |
12382 | ||
12383 | /* no auto-attach for SEC("tp") or SEC("tracepoint") */ | |
12384 | if (strcmp(prog->sec_name, "tp") == 0 || strcmp(prog->sec_name, "tracepoint") == 0) | |
12385 | return 0; | |
12386 | ||
ac6a6586 AN |
12387 | sec_name = strdup(prog->sec_name); |
12388 | if (!sec_name) | |
12389 | return -ENOMEM; | |
12390 | ||
13d35a0c AN |
12391 | /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */ |
12392 | if (str_has_pfx(prog->sec_name, "tp/")) | |
12393 | tp_cat = sec_name + sizeof("tp/") - 1; | |
12394 | else | |
12395 | tp_cat = sec_name + sizeof("tracepoint/") - 1; | |
d7a18ea7 AN |
12396 | tp_name = strchr(tp_cat, '/'); |
12397 | if (!tp_name) { | |
e9fc3ce9 | 12398 | free(sec_name); |
4fa5bcfe | 12399 | return -EINVAL; |
d7a18ea7 AN |
12400 | } |
12401 | *tp_name = '\0'; | |
12402 | tp_name++; | |
12403 | ||
4fa5bcfe | 12404 | *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name); |
d7a18ea7 | 12405 | free(sec_name); |
4fa5bcfe | 12406 | return libbpf_get_error(*link); |
d7a18ea7 AN |
12407 | } |
12408 | ||
36ffb202 AN |
12409 | struct bpf_link * |
12410 | bpf_program__attach_raw_tracepoint_opts(const struct bpf_program *prog, | |
12411 | const char *tp_name, | |
12412 | struct bpf_raw_tracepoint_opts *opts) | |
84bf5e1f | 12413 | { |
36ffb202 | 12414 | LIBBPF_OPTS(bpf_raw_tp_opts, raw_opts); |
84bf5e1f | 12415 | char errmsg[STRERR_BUFSIZE]; |
c016b68e | 12416 | struct bpf_link *link; |
84bf5e1f AN |
12417 | int prog_fd, pfd; |
12418 | ||
36ffb202 AN |
12419 | if (!OPTS_VALID(opts, bpf_raw_tracepoint_opts)) |
12420 | return libbpf_err_ptr(-EINVAL); | |
12421 | ||
84bf5e1f AN |
12422 | prog_fd = bpf_program__fd(prog); |
12423 | if (prog_fd < 0) { | |
52109584 | 12424 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 12425 | return libbpf_err_ptr(-EINVAL); |
84bf5e1f AN |
12426 | } |
12427 | ||
d6958706 | 12428 | link = calloc(1, sizeof(*link)); |
84bf5e1f | 12429 | if (!link) |
e9fc3ce9 | 12430 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 12431 | link->detach = &bpf_link__detach_fd; |
84bf5e1f | 12432 | |
36ffb202 AN |
12433 | raw_opts.tp_name = tp_name; |
12434 | raw_opts.cookie = OPTS_GET(opts, cookie, 0); | |
12435 | pfd = bpf_raw_tracepoint_open_opts(prog_fd, &raw_opts); | |
84bf5e1f AN |
12436 | if (pfd < 0) { |
12437 | pfd = -errno; | |
12438 | free(link); | |
52109584 AN |
12439 | pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n", |
12440 | prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 12441 | return libbpf_err_ptr(pfd); |
84bf5e1f AN |
12442 | } |
12443 | link->fd = pfd; | |
c016b68e | 12444 | return link; |
84bf5e1f AN |
12445 | } |
12446 | ||
36ffb202 AN |
12447 | struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog, |
12448 | const char *tp_name) | |
12449 | { | |
12450 | return bpf_program__attach_raw_tracepoint_opts(prog, tp_name, NULL); | |
12451 | } | |
12452 | ||
4fa5bcfe | 12453 | static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 12454 | { |
ccaf12d6 | 12455 | static const char *const prefixes[] = { |
9af8efc4 AN |
12456 | "raw_tp", |
12457 | "raw_tracepoint", | |
12458 | "raw_tp.w", | |
12459 | "raw_tracepoint.w", | |
ccaf12d6 HT |
12460 | }; |
12461 | size_t i; | |
12462 | const char *tp_name = NULL; | |
13d35a0c | 12463 | |
9af8efc4 AN |
12464 | *link = NULL; |
12465 | ||
ccaf12d6 | 12466 | for (i = 0; i < ARRAY_SIZE(prefixes); i++) { |
9af8efc4 AN |
12467 | size_t pfx_len; |
12468 | ||
12469 | if (!str_has_pfx(prog->sec_name, prefixes[i])) | |
12470 | continue; | |
12471 | ||
12472 | pfx_len = strlen(prefixes[i]); | |
12473 | /* no auto-attach case of, e.g., SEC("raw_tp") */ | |
12474 | if (prog->sec_name[pfx_len] == '\0') | |
12475 | return 0; | |
12476 | ||
12477 | if (prog->sec_name[pfx_len] != '/') | |
12478 | continue; | |
12479 | ||
12480 | tp_name = prog->sec_name + pfx_len + 1; | |
12481 | break; | |
ccaf12d6 | 12482 | } |
9af8efc4 | 12483 | |
ccaf12d6 HT |
12484 | if (!tp_name) { |
12485 | pr_warn("prog '%s': invalid section name '%s'\n", | |
12486 | prog->name, prog->sec_name); | |
4fa5bcfe | 12487 | return -EINVAL; |
ccaf12d6 | 12488 | } |
d7a18ea7 | 12489 | |
4fa5bcfe | 12490 | *link = bpf_program__attach_raw_tracepoint(prog, tp_name); |
5fd2a60a | 12491 | return libbpf_get_error(*link); |
d7a18ea7 AN |
12492 | } |
12493 | ||
1e092a03 | 12494 | /* Common logic for all BPF program types that attach to a btf_id */ |
129b9c5e KFL |
12495 | static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog, |
12496 | const struct bpf_trace_opts *opts) | |
b8c54ea4 | 12497 | { |
129b9c5e | 12498 | LIBBPF_OPTS(bpf_link_create_opts, link_opts); |
b8c54ea4 | 12499 | char errmsg[STRERR_BUFSIZE]; |
c016b68e | 12500 | struct bpf_link *link; |
b8c54ea4 AS |
12501 | int prog_fd, pfd; |
12502 | ||
129b9c5e KFL |
12503 | if (!OPTS_VALID(opts, bpf_trace_opts)) |
12504 | return libbpf_err_ptr(-EINVAL); | |
12505 | ||
b8c54ea4 AS |
12506 | prog_fd = bpf_program__fd(prog); |
12507 | if (prog_fd < 0) { | |
52109584 | 12508 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 12509 | return libbpf_err_ptr(-EINVAL); |
b8c54ea4 AS |
12510 | } |
12511 | ||
d6958706 | 12512 | link = calloc(1, sizeof(*link)); |
b8c54ea4 | 12513 | if (!link) |
e9fc3ce9 | 12514 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 12515 | link->detach = &bpf_link__detach_fd; |
b8c54ea4 | 12516 | |
8462e0b4 | 12517 | /* libbpf is smart enough to redirect to BPF_RAW_TRACEPOINT_OPEN on old kernels */ |
129b9c5e KFL |
12518 | link_opts.tracing.cookie = OPTS_GET(opts, cookie, 0); |
12519 | pfd = bpf_link_create(prog_fd, 0, bpf_program__expected_attach_type(prog), &link_opts); | |
b8c54ea4 AS |
12520 | if (pfd < 0) { |
12521 | pfd = -errno; | |
12522 | free(link); | |
52109584 AN |
12523 | pr_warn("prog '%s': failed to attach: %s\n", |
12524 | prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 12525 | return libbpf_err_ptr(pfd); |
b8c54ea4 AS |
12526 | } |
12527 | link->fd = pfd; | |
003fed59 | 12528 | return link; |
b8c54ea4 AS |
12529 | } |
12530 | ||
942025c9 | 12531 | struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog) |
1e092a03 | 12532 | { |
129b9c5e KFL |
12533 | return bpf_program__attach_btf_id(prog, NULL); |
12534 | } | |
12535 | ||
12536 | struct bpf_link *bpf_program__attach_trace_opts(const struct bpf_program *prog, | |
12537 | const struct bpf_trace_opts *opts) | |
12538 | { | |
12539 | return bpf_program__attach_btf_id(prog, opts); | |
1e092a03 KS |
12540 | } |
12541 | ||
942025c9 | 12542 | struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog) |
1e092a03 | 12543 | { |
129b9c5e | 12544 | return bpf_program__attach_btf_id(prog, NULL); |
1e092a03 KS |
12545 | } |
12546 | ||
4fa5bcfe | 12547 | static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 12548 | { |
4fa5bcfe AN |
12549 | *link = bpf_program__attach_trace(prog); |
12550 | return libbpf_get_error(*link); | |
d7a18ea7 AN |
12551 | } |
12552 | ||
4fa5bcfe | 12553 | static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
1e092a03 | 12554 | { |
4fa5bcfe AN |
12555 | *link = bpf_program__attach_lsm(prog); |
12556 | return libbpf_get_error(*link); | |
1e092a03 KS |
12557 | } |
12558 | ||
d60d81ac | 12559 | static struct bpf_link * |
55cc3768 DB |
12560 | bpf_program_attach_fd(const struct bpf_program *prog, |
12561 | int target_fd, const char *target_name, | |
12562 | const struct bpf_link_create_opts *opts) | |
cc4f864b | 12563 | { |
cc4f864b AN |
12564 | enum bpf_attach_type attach_type; |
12565 | char errmsg[STRERR_BUFSIZE]; | |
12566 | struct bpf_link *link; | |
12567 | int prog_fd, link_fd; | |
12568 | ||
12569 | prog_fd = bpf_program__fd(prog); | |
12570 | if (prog_fd < 0) { | |
52109584 | 12571 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 12572 | return libbpf_err_ptr(-EINVAL); |
cc4f864b AN |
12573 | } |
12574 | ||
12575 | link = calloc(1, sizeof(*link)); | |
12576 | if (!link) | |
e9fc3ce9 | 12577 | return libbpf_err_ptr(-ENOMEM); |
cc4f864b AN |
12578 | link->detach = &bpf_link__detach_fd; |
12579 | ||
20eccf29 | 12580 | attach_type = bpf_program__expected_attach_type(prog); |
55cc3768 | 12581 | link_fd = bpf_link_create(prog_fd, target_fd, attach_type, opts); |
cc4f864b AN |
12582 | if (link_fd < 0) { |
12583 | link_fd = -errno; | |
12584 | free(link); | |
52109584 AN |
12585 | pr_warn("prog '%s': failed to attach to %s: %s\n", |
12586 | prog->name, target_name, | |
cc4f864b | 12587 | libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 12588 | return libbpf_err_ptr(link_fd); |
cc4f864b AN |
12589 | } |
12590 | link->fd = link_fd; | |
12591 | return link; | |
12592 | } | |
12593 | ||
d60d81ac | 12594 | struct bpf_link * |
942025c9 | 12595 | bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd) |
d60d81ac | 12596 | { |
55cc3768 | 12597 | return bpf_program_attach_fd(prog, cgroup_fd, "cgroup", NULL); |
d60d81ac JS |
12598 | } |
12599 | ||
12600 | struct bpf_link * | |
942025c9 | 12601 | bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd) |
d60d81ac | 12602 | { |
55cc3768 | 12603 | return bpf_program_attach_fd(prog, netns_fd, "netns", NULL); |
d60d81ac JS |
12604 | } |
12605 | ||
849989af YS |
12606 | struct bpf_link * |
12607 | bpf_program__attach_sockmap(const struct bpf_program *prog, int map_fd) | |
12608 | { | |
12609 | return bpf_program_attach_fd(prog, map_fd, "sockmap", NULL); | |
12610 | } | |
12611 | ||
942025c9 | 12612 | struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex) |
dc8698ca AN |
12613 | { |
12614 | /* target_fd/target_ifindex use the same field in LINK_CREATE */ | |
55cc3768 DB |
12615 | return bpf_program_attach_fd(prog, ifindex, "xdp", NULL); |
12616 | } | |
12617 | ||
12618 | struct bpf_link * | |
12619 | bpf_program__attach_tcx(const struct bpf_program *prog, int ifindex, | |
12620 | const struct bpf_tcx_opts *opts) | |
12621 | { | |
12622 | LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); | |
12623 | __u32 relative_id; | |
12624 | int relative_fd; | |
12625 | ||
12626 | if (!OPTS_VALID(opts, bpf_tcx_opts)) | |
12627 | return libbpf_err_ptr(-EINVAL); | |
12628 | ||
12629 | relative_id = OPTS_GET(opts, relative_id, 0); | |
12630 | relative_fd = OPTS_GET(opts, relative_fd, 0); | |
12631 | ||
12632 | /* validate we don't have unexpected combinations of non-zero fields */ | |
12633 | if (!ifindex) { | |
12634 | pr_warn("prog '%s': target netdevice ifindex cannot be zero\n", | |
12635 | prog->name); | |
12636 | return libbpf_err_ptr(-EINVAL); | |
12637 | } | |
12638 | if (relative_fd && relative_id) { | |
12639 | pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n", | |
12640 | prog->name); | |
12641 | return libbpf_err_ptr(-EINVAL); | |
12642 | } | |
12643 | ||
12644 | link_create_opts.tcx.expected_revision = OPTS_GET(opts, expected_revision, 0); | |
12645 | link_create_opts.tcx.relative_fd = relative_fd; | |
12646 | link_create_opts.tcx.relative_id = relative_id; | |
12647 | link_create_opts.flags = OPTS_GET(opts, flags, 0); | |
12648 | ||
12649 | /* target_fd/target_ifindex use the same field in LINK_CREATE */ | |
12650 | return bpf_program_attach_fd(prog, ifindex, "tcx", &link_create_opts); | |
a5359091 THJ |
12651 | } |
12652 | ||
05c31b4a DB |
12653 | struct bpf_link * |
12654 | bpf_program__attach_netkit(const struct bpf_program *prog, int ifindex, | |
12655 | const struct bpf_netkit_opts *opts) | |
12656 | { | |
12657 | LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); | |
12658 | __u32 relative_id; | |
12659 | int relative_fd; | |
12660 | ||
12661 | if (!OPTS_VALID(opts, bpf_netkit_opts)) | |
12662 | return libbpf_err_ptr(-EINVAL); | |
12663 | ||
12664 | relative_id = OPTS_GET(opts, relative_id, 0); | |
12665 | relative_fd = OPTS_GET(opts, relative_fd, 0); | |
12666 | ||
12667 | /* validate we don't have unexpected combinations of non-zero fields */ | |
12668 | if (!ifindex) { | |
12669 | pr_warn("prog '%s': target netdevice ifindex cannot be zero\n", | |
12670 | prog->name); | |
12671 | return libbpf_err_ptr(-EINVAL); | |
12672 | } | |
12673 | if (relative_fd && relative_id) { | |
12674 | pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n", | |
12675 | prog->name); | |
12676 | return libbpf_err_ptr(-EINVAL); | |
12677 | } | |
12678 | ||
12679 | link_create_opts.netkit.expected_revision = OPTS_GET(opts, expected_revision, 0); | |
12680 | link_create_opts.netkit.relative_fd = relative_fd; | |
12681 | link_create_opts.netkit.relative_id = relative_id; | |
12682 | link_create_opts.flags = OPTS_GET(opts, flags, 0); | |
12683 | ||
12684 | return bpf_program_attach_fd(prog, ifindex, "netkit", &link_create_opts); | |
12685 | } | |
12686 | ||
942025c9 | 12687 | struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog, |
a5359091 THJ |
12688 | int target_fd, |
12689 | const char *attach_func_name) | |
12690 | { | |
12691 | int btf_id; | |
12692 | ||
12693 | if (!!target_fd != !!attach_func_name) { | |
12694 | pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n", | |
12695 | prog->name); | |
e9fc3ce9 | 12696 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
12697 | } |
12698 | ||
12699 | if (prog->type != BPF_PROG_TYPE_EXT) { | |
12700 | pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace", | |
12701 | prog->name); | |
e9fc3ce9 | 12702 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
12703 | } |
12704 | ||
12705 | if (target_fd) { | |
55cc3768 DB |
12706 | LIBBPF_OPTS(bpf_link_create_opts, target_opts); |
12707 | ||
a5359091 THJ |
12708 | btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd); |
12709 | if (btf_id < 0) | |
e9fc3ce9 | 12710 | return libbpf_err_ptr(btf_id); |
a5359091 | 12711 | |
55cc3768 DB |
12712 | target_opts.target_btf_id = btf_id; |
12713 | ||
12714 | return bpf_program_attach_fd(prog, target_fd, "freplace", | |
12715 | &target_opts); | |
a5359091 THJ |
12716 | } else { |
12717 | /* no target, so use raw_tracepoint_open for compatibility | |
12718 | * with old kernels | |
12719 | */ | |
12720 | return bpf_program__attach_trace(prog); | |
12721 | } | |
dc8698ca AN |
12722 | } |
12723 | ||
c09add2f | 12724 | struct bpf_link * |
942025c9 | 12725 | bpf_program__attach_iter(const struct bpf_program *prog, |
c09add2f YS |
12726 | const struct bpf_iter_attach_opts *opts) |
12727 | { | |
cd31039a | 12728 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); |
c09add2f YS |
12729 | char errmsg[STRERR_BUFSIZE]; |
12730 | struct bpf_link *link; | |
12731 | int prog_fd, link_fd; | |
cd31039a | 12732 | __u32 target_fd = 0; |
c09add2f YS |
12733 | |
12734 | if (!OPTS_VALID(opts, bpf_iter_attach_opts)) | |
e9fc3ce9 | 12735 | return libbpf_err_ptr(-EINVAL); |
c09add2f | 12736 | |
74fc097d YS |
12737 | link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0); |
12738 | link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0); | |
cd31039a | 12739 | |
c09add2f YS |
12740 | prog_fd = bpf_program__fd(prog); |
12741 | if (prog_fd < 0) { | |
52109584 | 12742 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 12743 | return libbpf_err_ptr(-EINVAL); |
c09add2f YS |
12744 | } |
12745 | ||
12746 | link = calloc(1, sizeof(*link)); | |
12747 | if (!link) | |
e9fc3ce9 | 12748 | return libbpf_err_ptr(-ENOMEM); |
c09add2f YS |
12749 | link->detach = &bpf_link__detach_fd; |
12750 | ||
cd31039a YS |
12751 | link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER, |
12752 | &link_create_opts); | |
c09add2f YS |
12753 | if (link_fd < 0) { |
12754 | link_fd = -errno; | |
12755 | free(link); | |
52109584 AN |
12756 | pr_warn("prog '%s': failed to attach to iterator: %s\n", |
12757 | prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 12758 | return libbpf_err_ptr(link_fd); |
c09add2f YS |
12759 | } |
12760 | link->fd = link_fd; | |
12761 | return link; | |
12762 | } | |
12763 | ||
4fa5bcfe | 12764 | static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
e9fc3ce9 | 12765 | { |
4fa5bcfe AN |
12766 | *link = bpf_program__attach_iter(prog, NULL); |
12767 | return libbpf_get_error(*link); | |
e9fc3ce9 AN |
12768 | } |
12769 | ||
52364abb FW |
12770 | struct bpf_link *bpf_program__attach_netfilter(const struct bpf_program *prog, |
12771 | const struct bpf_netfilter_opts *opts) | |
12772 | { | |
12773 | LIBBPF_OPTS(bpf_link_create_opts, lopts); | |
12774 | struct bpf_link *link; | |
12775 | int prog_fd, link_fd; | |
12776 | ||
12777 | if (!OPTS_VALID(opts, bpf_netfilter_opts)) | |
12778 | return libbpf_err_ptr(-EINVAL); | |
12779 | ||
12780 | prog_fd = bpf_program__fd(prog); | |
12781 | if (prog_fd < 0) { | |
12782 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); | |
12783 | return libbpf_err_ptr(-EINVAL); | |
12784 | } | |
12785 | ||
12786 | link = calloc(1, sizeof(*link)); | |
12787 | if (!link) | |
12788 | return libbpf_err_ptr(-ENOMEM); | |
12789 | ||
12790 | link->detach = &bpf_link__detach_fd; | |
12791 | ||
12792 | lopts.netfilter.pf = OPTS_GET(opts, pf, 0); | |
12793 | lopts.netfilter.hooknum = OPTS_GET(opts, hooknum, 0); | |
12794 | lopts.netfilter.priority = OPTS_GET(opts, priority, 0); | |
12795 | lopts.netfilter.flags = OPTS_GET(opts, flags, 0); | |
12796 | ||
12797 | link_fd = bpf_link_create(prog_fd, 0, BPF_NETFILTER, &lopts); | |
12798 | if (link_fd < 0) { | |
12799 | char errmsg[STRERR_BUFSIZE]; | |
12800 | ||
12801 | link_fd = -errno; | |
12802 | free(link); | |
12803 | pr_warn("prog '%s': failed to attach to netfilter: %s\n", | |
12804 | prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); | |
12805 | return libbpf_err_ptr(link_fd); | |
12806 | } | |
12807 | link->fd = link_fd; | |
12808 | ||
12809 | return link; | |
12810 | } | |
12811 | ||
942025c9 | 12812 | struct bpf_link *bpf_program__attach(const struct bpf_program *prog) |
d7a18ea7 | 12813 | { |
4fa5bcfe AN |
12814 | struct bpf_link *link = NULL; |
12815 | int err; | |
12816 | ||
12817 | if (!prog->sec_def || !prog->sec_def->prog_attach_fn) | |
12818 | return libbpf_err_ptr(-EOPNOTSUPP); | |
d7a18ea7 | 12819 | |
7b30c296 MY |
12820 | if (bpf_program__fd(prog) < 0) { |
12821 | pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n", | |
12822 | prog->name); | |
12823 | return libbpf_err_ptr(-EINVAL); | |
12824 | } | |
12825 | ||
4fa5bcfe AN |
12826 | err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link); |
12827 | if (err) | |
12828 | return libbpf_err_ptr(err); | |
12829 | ||
12830 | /* When calling bpf_program__attach() explicitly, auto-attach support | |
12831 | * is expected to work, so NULL returned link is considered an error. | |
12832 | * This is different for skeleton's attach, see comment in | |
12833 | * bpf_object__attach_skeleton(). | |
12834 | */ | |
12835 | if (!link) | |
12836 | return libbpf_err_ptr(-EOPNOTSUPP); | |
12837 | ||
12838 | return link; | |
d7a18ea7 AN |
12839 | } |
12840 | ||
8d1608d7 KFL |
12841 | struct bpf_link_struct_ops { |
12842 | struct bpf_link link; | |
12843 | int map_fd; | |
12844 | }; | |
12845 | ||
590a0088 MKL |
12846 | static int bpf_link__detach_struct_ops(struct bpf_link *link) |
12847 | { | |
8d1608d7 | 12848 | struct bpf_link_struct_ops *st_link; |
590a0088 MKL |
12849 | __u32 zero = 0; |
12850 | ||
8d1608d7 | 12851 | st_link = container_of(link, struct bpf_link_struct_ops, link); |
590a0088 | 12852 | |
8d1608d7 KFL |
12853 | if (st_link->map_fd < 0) |
12854 | /* w/o a real link */ | |
12855 | return bpf_map_delete_elem(link->fd, &zero); | |
12856 | ||
12857 | return close(link->fd); | |
590a0088 MKL |
12858 | } |
12859 | ||
942025c9 | 12860 | struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map) |
590a0088 | 12861 | { |
8d1608d7 KFL |
12862 | struct bpf_link_struct_ops *link; |
12863 | __u32 zero = 0; | |
12864 | int err, fd; | |
590a0088 | 12865 | |
7b30c296 | 12866 | if (!bpf_map__is_struct_ops(map)) |
e9fc3ce9 | 12867 | return libbpf_err_ptr(-EINVAL); |
590a0088 | 12868 | |
7b30c296 MY |
12869 | if (map->fd < 0) { |
12870 | pr_warn("map '%s': can't attach BPF map without FD (was it created?)\n", map->name); | |
12871 | return libbpf_err_ptr(-EINVAL); | |
12872 | } | |
12873 | ||
590a0088 MKL |
12874 | link = calloc(1, sizeof(*link)); |
12875 | if (!link) | |
e9fc3ce9 | 12876 | return libbpf_err_ptr(-EINVAL); |
590a0088 | 12877 | |
8d1608d7 KFL |
12878 | /* kern_vdata should be prepared during the loading phase. */ |
12879 | err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0); | |
12880 | /* It can be EBUSY if the map has been used to create or | |
12881 | * update a link before. We don't allow updating the value of | |
12882 | * a struct_ops once it is set. That ensures that the value | |
12883 | * never changed. So, it is safe to skip EBUSY. | |
12884 | */ | |
12885 | if (err && (!(map->def.map_flags & BPF_F_LINK) || err != -EBUSY)) { | |
12886 | free(link); | |
12887 | return libbpf_err_ptr(err); | |
12888 | } | |
590a0088 | 12889 | |
8d1608d7 | 12890 | link->link.detach = bpf_link__detach_struct_ops; |
590a0088 | 12891 | |
8d1608d7 KFL |
12892 | if (!(map->def.map_flags & BPF_F_LINK)) { |
12893 | /* w/o a real link */ | |
12894 | link->link.fd = map->fd; | |
12895 | link->map_fd = -1; | |
12896 | return &link->link; | |
590a0088 MKL |
12897 | } |
12898 | ||
8d1608d7 KFL |
12899 | fd = bpf_link_create(map->fd, 0, BPF_STRUCT_OPS, NULL); |
12900 | if (fd < 0) { | |
590a0088 | 12901 | free(link); |
8d1608d7 | 12902 | return libbpf_err_ptr(fd); |
590a0088 MKL |
12903 | } |
12904 | ||
8d1608d7 KFL |
12905 | link->link.fd = fd; |
12906 | link->map_fd = map->fd; | |
590a0088 | 12907 | |
8d1608d7 | 12908 | return &link->link; |
590a0088 MKL |
12909 | } |
12910 | ||
912dd4b0 KFL |
12911 | /* |
12912 | * Swap the back struct_ops of a link with a new struct_ops map. | |
12913 | */ | |
12914 | int bpf_link__update_map(struct bpf_link *link, const struct bpf_map *map) | |
12915 | { | |
12916 | struct bpf_link_struct_ops *st_ops_link; | |
12917 | __u32 zero = 0; | |
12918 | int err; | |
12919 | ||
7b30c296 MY |
12920 | if (!bpf_map__is_struct_ops(map)) |
12921 | return -EINVAL; | |
12922 | ||
12923 | if (map->fd < 0) { | |
12924 | pr_warn("map '%s': can't use BPF map without FD (was it created?)\n", map->name); | |
912dd4b0 | 12925 | return -EINVAL; |
7b30c296 | 12926 | } |
912dd4b0 KFL |
12927 | |
12928 | st_ops_link = container_of(link, struct bpf_link_struct_ops, link); | |
12929 | /* Ensure the type of a link is correct */ | |
12930 | if (st_ops_link->map_fd < 0) | |
12931 | return -EINVAL; | |
12932 | ||
12933 | err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0); | |
12934 | /* It can be EBUSY if the map has been used to create or | |
12935 | * update a link before. We don't allow updating the value of | |
12936 | * a struct_ops once it is set. That ensures that the value | |
12937 | * never changed. So, it is safe to skip EBUSY. | |
12938 | */ | |
12939 | if (err && err != -EBUSY) | |
12940 | return err; | |
12941 | ||
12942 | err = bpf_link_update(link->fd, map->fd, NULL); | |
12943 | if (err < 0) | |
12944 | return err; | |
12945 | ||
12946 | st_ops_link->map_fd = map->fd; | |
12947 | ||
12948 | return 0; | |
12949 | } | |
12950 | ||
22dd7a58 AN |
12951 | typedef enum bpf_perf_event_ret (*bpf_perf_event_print_t)(struct perf_event_header *hdr, |
12952 | void *private_data); | |
12953 | ||
7218c28c CL |
12954 | static enum bpf_perf_event_ret |
12955 | perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size, | |
12956 | void **copy_mem, size_t *copy_size, | |
12957 | bpf_perf_event_print_t fn, void *private_data) | |
d0cabbb0 | 12958 | { |
3dca2115 | 12959 | struct perf_event_mmap_page *header = mmap_mem; |
a64af0ef | 12960 | __u64 data_head = ring_buffer_read_head(header); |
d0cabbb0 | 12961 | __u64 data_tail = header->data_tail; |
3dca2115 DB |
12962 | void *base = ((__u8 *)header) + page_size; |
12963 | int ret = LIBBPF_PERF_EVENT_CONT; | |
12964 | struct perf_event_header *ehdr; | |
12965 | size_t ehdr_size; | |
12966 | ||
12967 | while (data_head != data_tail) { | |
12968 | ehdr = base + (data_tail & (mmap_size - 1)); | |
12969 | ehdr_size = ehdr->size; | |
12970 | ||
12971 | if (((void *)ehdr) + ehdr_size > base + mmap_size) { | |
12972 | void *copy_start = ehdr; | |
12973 | size_t len_first = base + mmap_size - copy_start; | |
12974 | size_t len_secnd = ehdr_size - len_first; | |
12975 | ||
12976 | if (*copy_size < ehdr_size) { | |
12977 | free(*copy_mem); | |
12978 | *copy_mem = malloc(ehdr_size); | |
12979 | if (!*copy_mem) { | |
12980 | *copy_size = 0; | |
d0cabbb0 JK |
12981 | ret = LIBBPF_PERF_EVENT_ERROR; |
12982 | break; | |
12983 | } | |
3dca2115 | 12984 | *copy_size = ehdr_size; |
d0cabbb0 JK |
12985 | } |
12986 | ||
3dca2115 DB |
12987 | memcpy(*copy_mem, copy_start, len_first); |
12988 | memcpy(*copy_mem + len_first, base, len_secnd); | |
12989 | ehdr = *copy_mem; | |
d0cabbb0 JK |
12990 | } |
12991 | ||
3dca2115 DB |
12992 | ret = fn(ehdr, private_data); |
12993 | data_tail += ehdr_size; | |
d0cabbb0 JK |
12994 | if (ret != LIBBPF_PERF_EVENT_CONT) |
12995 | break; | |
d0cabbb0 JK |
12996 | } |
12997 | ||
a64af0ef | 12998 | ring_buffer_write_tail(header, data_tail); |
e9fc3ce9 | 12999 | return libbpf_err(ret); |
d0cabbb0 | 13000 | } |
34be1646 | 13001 | |
fb84b822 AN |
13002 | struct perf_buffer; |
13003 | ||
13004 | struct perf_buffer_params { | |
13005 | struct perf_event_attr *attr; | |
13006 | /* if event_cb is specified, it takes precendence */ | |
13007 | perf_buffer_event_fn event_cb; | |
13008 | /* sample_cb and lost_cb are higher-level common-case callbacks */ | |
13009 | perf_buffer_sample_fn sample_cb; | |
13010 | perf_buffer_lost_fn lost_cb; | |
13011 | void *ctx; | |
13012 | int cpu_cnt; | |
13013 | int *cpus; | |
13014 | int *map_keys; | |
13015 | }; | |
13016 | ||
13017 | struct perf_cpu_buf { | |
13018 | struct perf_buffer *pb; | |
13019 | void *base; /* mmap()'ed memory */ | |
13020 | void *buf; /* for reconstructing segmented data */ | |
13021 | size_t buf_size; | |
13022 | int fd; | |
13023 | int cpu; | |
13024 | int map_key; | |
13025 | }; | |
13026 | ||
13027 | struct perf_buffer { | |
13028 | perf_buffer_event_fn event_cb; | |
13029 | perf_buffer_sample_fn sample_cb; | |
13030 | perf_buffer_lost_fn lost_cb; | |
13031 | void *ctx; /* passed into callbacks */ | |
13032 | ||
13033 | size_t page_size; | |
13034 | size_t mmap_size; | |
13035 | struct perf_cpu_buf **cpu_bufs; | |
13036 | struct epoll_event *events; | |
783b8f01 | 13037 | int cpu_cnt; /* number of allocated CPU buffers */ |
fb84b822 AN |
13038 | int epoll_fd; /* perf event FD */ |
13039 | int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */ | |
13040 | }; | |
13041 | ||
13042 | static void perf_buffer__free_cpu_buf(struct perf_buffer *pb, | |
13043 | struct perf_cpu_buf *cpu_buf) | |
13044 | { | |
13045 | if (!cpu_buf) | |
13046 | return; | |
13047 | if (cpu_buf->base && | |
13048 | munmap(cpu_buf->base, pb->mmap_size + pb->page_size)) | |
be18010e | 13049 | pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu); |
fb84b822 AN |
13050 | if (cpu_buf->fd >= 0) { |
13051 | ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0); | |
13052 | close(cpu_buf->fd); | |
13053 | } | |
13054 | free(cpu_buf->buf); | |
13055 | free(cpu_buf); | |
13056 | } | |
13057 | ||
13058 | void perf_buffer__free(struct perf_buffer *pb) | |
13059 | { | |
13060 | int i; | |
13061 | ||
50450fc7 | 13062 | if (IS_ERR_OR_NULL(pb)) |
fb84b822 AN |
13063 | return; |
13064 | if (pb->cpu_bufs) { | |
601b05ca | 13065 | for (i = 0; i < pb->cpu_cnt; i++) { |
fb84b822 AN |
13066 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; |
13067 | ||
601b05ca EC |
13068 | if (!cpu_buf) |
13069 | continue; | |
13070 | ||
fb84b822 AN |
13071 | bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key); |
13072 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
13073 | } | |
13074 | free(pb->cpu_bufs); | |
13075 | } | |
13076 | if (pb->epoll_fd >= 0) | |
13077 | close(pb->epoll_fd); | |
13078 | free(pb->events); | |
13079 | free(pb); | |
13080 | } | |
13081 | ||
13082 | static struct perf_cpu_buf * | |
13083 | perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr, | |
13084 | int cpu, int map_key) | |
13085 | { | |
13086 | struct perf_cpu_buf *cpu_buf; | |
13087 | char msg[STRERR_BUFSIZE]; | |
13088 | int err; | |
13089 | ||
13090 | cpu_buf = calloc(1, sizeof(*cpu_buf)); | |
13091 | if (!cpu_buf) | |
13092 | return ERR_PTR(-ENOMEM); | |
13093 | ||
13094 | cpu_buf->pb = pb; | |
13095 | cpu_buf->cpu = cpu; | |
13096 | cpu_buf->map_key = map_key; | |
13097 | ||
13098 | cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu, | |
13099 | -1, PERF_FLAG_FD_CLOEXEC); | |
13100 | if (cpu_buf->fd < 0) { | |
13101 | err = -errno; | |
be18010e KW |
13102 | pr_warn("failed to open perf buffer event on cpu #%d: %s\n", |
13103 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13104 | goto error; |
13105 | } | |
13106 | ||
13107 | cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size, | |
13108 | PROT_READ | PROT_WRITE, MAP_SHARED, | |
13109 | cpu_buf->fd, 0); | |
13110 | if (cpu_buf->base == MAP_FAILED) { | |
13111 | cpu_buf->base = NULL; | |
13112 | err = -errno; | |
be18010e KW |
13113 | pr_warn("failed to mmap perf buffer on cpu #%d: %s\n", |
13114 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13115 | goto error; |
13116 | } | |
13117 | ||
13118 | if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
13119 | err = -errno; | |
be18010e KW |
13120 | pr_warn("failed to enable perf buffer event on cpu #%d: %s\n", |
13121 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13122 | goto error; |
13123 | } | |
13124 | ||
13125 | return cpu_buf; | |
13126 | ||
13127 | error: | |
13128 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
13129 | return (struct perf_cpu_buf *)ERR_PTR(err); | |
13130 | } | |
13131 | ||
13132 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
13133 | struct perf_buffer_params *p); | |
13134 | ||
22dd7a58 AN |
13135 | struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt, |
13136 | perf_buffer_sample_fn sample_cb, | |
13137 | perf_buffer_lost_fn lost_cb, | |
13138 | void *ctx, | |
13139 | const struct perf_buffer_opts *opts) | |
fb84b822 | 13140 | { |
813847a3 | 13141 | const size_t attr_sz = sizeof(struct perf_event_attr); |
fb84b822 | 13142 | struct perf_buffer_params p = {}; |
813847a3 | 13143 | struct perf_event_attr attr; |
ab8684b8 | 13144 | __u32 sample_period; |
41788934 AN |
13145 | |
13146 | if (!OPTS_VALID(opts, perf_buffer_opts)) | |
13147 | return libbpf_err_ptr(-EINVAL); | |
4be6e05c | 13148 | |
ab8684b8 JD |
13149 | sample_period = OPTS_GET(opts, sample_period, 1); |
13150 | if (!sample_period) | |
13151 | sample_period = 1; | |
13152 | ||
813847a3 AN |
13153 | memset(&attr, 0, attr_sz); |
13154 | attr.size = attr_sz; | |
65bb2e0f | 13155 | attr.config = PERF_COUNT_SW_BPF_OUTPUT; |
4be6e05c ACM |
13156 | attr.type = PERF_TYPE_SOFTWARE; |
13157 | attr.sample_type = PERF_SAMPLE_RAW; | |
ab8684b8 JD |
13158 | attr.sample_period = sample_period; |
13159 | attr.wakeup_events = sample_period; | |
fb84b822 AN |
13160 | |
13161 | p.attr = &attr; | |
41788934 AN |
13162 | p.sample_cb = sample_cb; |
13163 | p.lost_cb = lost_cb; | |
13164 | p.ctx = ctx; | |
fb84b822 | 13165 | |
e9fc3ce9 | 13166 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
13167 | } |
13168 | ||
22dd7a58 AN |
13169 | struct perf_buffer *perf_buffer__new_raw(int map_fd, size_t page_cnt, |
13170 | struct perf_event_attr *attr, | |
13171 | perf_buffer_event_fn event_cb, void *ctx, | |
13172 | const struct perf_buffer_raw_opts *opts) | |
fb84b822 AN |
13173 | { |
13174 | struct perf_buffer_params p = {}; | |
13175 | ||
41332d6e | 13176 | if (!attr) |
41788934 AN |
13177 | return libbpf_err_ptr(-EINVAL); |
13178 | ||
13179 | if (!OPTS_VALID(opts, perf_buffer_raw_opts)) | |
13180 | return libbpf_err_ptr(-EINVAL); | |
13181 | ||
13182 | p.attr = attr; | |
13183 | p.event_cb = event_cb; | |
13184 | p.ctx = ctx; | |
13185 | p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0); | |
13186 | p.cpus = OPTS_GET(opts, cpus, NULL); | |
13187 | p.map_keys = OPTS_GET(opts, map_keys, NULL); | |
fb84b822 | 13188 | |
e9fc3ce9 | 13189 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
13190 | } |
13191 | ||
13192 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
13193 | struct perf_buffer_params *p) | |
13194 | { | |
783b8f01 | 13195 | const char *online_cpus_file = "/sys/devices/system/cpu/online"; |
0e289487 | 13196 | struct bpf_map_info map; |
fb84b822 AN |
13197 | char msg[STRERR_BUFSIZE]; |
13198 | struct perf_buffer *pb; | |
783b8f01 | 13199 | bool *online = NULL; |
fb84b822 | 13200 | __u32 map_info_len; |
783b8f01 | 13201 | int err, i, j, n; |
fb84b822 | 13202 | |
41332d6e | 13203 | if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) { |
be18010e KW |
13204 | pr_warn("page count should be power of two, but is %zu\n", |
13205 | page_cnt); | |
fb84b822 AN |
13206 | return ERR_PTR(-EINVAL); |
13207 | } | |
13208 | ||
0e289487 AN |
13209 | /* best-effort sanity checks */ |
13210 | memset(&map, 0, sizeof(map)); | |
fb84b822 | 13211 | map_info_len = sizeof(map); |
629dfc66 | 13212 | err = bpf_map_get_info_by_fd(map_fd, &map, &map_info_len); |
fb84b822 AN |
13213 | if (err) { |
13214 | err = -errno; | |
0e289487 AN |
13215 | /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return |
13216 | * -EBADFD, -EFAULT, or -E2BIG on real error | |
13217 | */ | |
13218 | if (err != -EINVAL) { | |
13219 | pr_warn("failed to get map info for map FD %d: %s\n", | |
13220 | map_fd, libbpf_strerror_r(err, msg, sizeof(msg))); | |
13221 | return ERR_PTR(err); | |
13222 | } | |
13223 | pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n", | |
13224 | map_fd); | |
13225 | } else { | |
13226 | if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) { | |
13227 | pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n", | |
13228 | map.name); | |
13229 | return ERR_PTR(-EINVAL); | |
13230 | } | |
fb84b822 AN |
13231 | } |
13232 | ||
13233 | pb = calloc(1, sizeof(*pb)); | |
13234 | if (!pb) | |
13235 | return ERR_PTR(-ENOMEM); | |
13236 | ||
13237 | pb->event_cb = p->event_cb; | |
13238 | pb->sample_cb = p->sample_cb; | |
13239 | pb->lost_cb = p->lost_cb; | |
13240 | pb->ctx = p->ctx; | |
13241 | ||
13242 | pb->page_size = getpagesize(); | |
13243 | pb->mmap_size = pb->page_size * page_cnt; | |
13244 | pb->map_fd = map_fd; | |
13245 | ||
13246 | pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
13247 | if (pb->epoll_fd < 0) { | |
13248 | err = -errno; | |
be18010e KW |
13249 | pr_warn("failed to create epoll instance: %s\n", |
13250 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13251 | goto error; |
13252 | } | |
13253 | ||
13254 | if (p->cpu_cnt > 0) { | |
13255 | pb->cpu_cnt = p->cpu_cnt; | |
13256 | } else { | |
13257 | pb->cpu_cnt = libbpf_num_possible_cpus(); | |
13258 | if (pb->cpu_cnt < 0) { | |
13259 | err = pb->cpu_cnt; | |
13260 | goto error; | |
13261 | } | |
0e289487 | 13262 | if (map.max_entries && map.max_entries < pb->cpu_cnt) |
fb84b822 AN |
13263 | pb->cpu_cnt = map.max_entries; |
13264 | } | |
13265 | ||
13266 | pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events)); | |
13267 | if (!pb->events) { | |
13268 | err = -ENOMEM; | |
be18010e | 13269 | pr_warn("failed to allocate events: out of memory\n"); |
fb84b822 AN |
13270 | goto error; |
13271 | } | |
13272 | pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs)); | |
13273 | if (!pb->cpu_bufs) { | |
13274 | err = -ENOMEM; | |
be18010e | 13275 | pr_warn("failed to allocate buffers: out of memory\n"); |
fb84b822 AN |
13276 | goto error; |
13277 | } | |
13278 | ||
783b8f01 AN |
13279 | err = parse_cpu_mask_file(online_cpus_file, &online, &n); |
13280 | if (err) { | |
13281 | pr_warn("failed to get online CPU mask: %d\n", err); | |
13282 | goto error; | |
13283 | } | |
13284 | ||
13285 | for (i = 0, j = 0; i < pb->cpu_cnt; i++) { | |
fb84b822 AN |
13286 | struct perf_cpu_buf *cpu_buf; |
13287 | int cpu, map_key; | |
13288 | ||
13289 | cpu = p->cpu_cnt > 0 ? p->cpus[i] : i; | |
13290 | map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i; | |
13291 | ||
783b8f01 AN |
13292 | /* in case user didn't explicitly requested particular CPUs to |
13293 | * be attached to, skip offline/not present CPUs | |
13294 | */ | |
13295 | if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu])) | |
13296 | continue; | |
13297 | ||
fb84b822 AN |
13298 | cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key); |
13299 | if (IS_ERR(cpu_buf)) { | |
13300 | err = PTR_ERR(cpu_buf); | |
13301 | goto error; | |
13302 | } | |
13303 | ||
783b8f01 | 13304 | pb->cpu_bufs[j] = cpu_buf; |
fb84b822 AN |
13305 | |
13306 | err = bpf_map_update_elem(pb->map_fd, &map_key, | |
13307 | &cpu_buf->fd, 0); | |
13308 | if (err) { | |
13309 | err = -errno; | |
be18010e KW |
13310 | pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n", |
13311 | cpu, map_key, cpu_buf->fd, | |
13312 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13313 | goto error; |
13314 | } | |
13315 | ||
783b8f01 AN |
13316 | pb->events[j].events = EPOLLIN; |
13317 | pb->events[j].data.ptr = cpu_buf; | |
fb84b822 | 13318 | if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd, |
783b8f01 | 13319 | &pb->events[j]) < 0) { |
fb84b822 | 13320 | err = -errno; |
be18010e KW |
13321 | pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n", |
13322 | cpu, cpu_buf->fd, | |
13323 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13324 | goto error; |
13325 | } | |
783b8f01 | 13326 | j++; |
fb84b822 | 13327 | } |
783b8f01 AN |
13328 | pb->cpu_cnt = j; |
13329 | free(online); | |
fb84b822 AN |
13330 | |
13331 | return pb; | |
13332 | ||
13333 | error: | |
783b8f01 | 13334 | free(online); |
fb84b822 AN |
13335 | if (pb) |
13336 | perf_buffer__free(pb); | |
13337 | return ERR_PTR(err); | |
13338 | } | |
13339 | ||
13340 | struct perf_sample_raw { | |
13341 | struct perf_event_header header; | |
13342 | uint32_t size; | |
385bbf7b | 13343 | char data[]; |
fb84b822 AN |
13344 | }; |
13345 | ||
13346 | struct perf_sample_lost { | |
13347 | struct perf_event_header header; | |
13348 | uint64_t id; | |
13349 | uint64_t lost; | |
13350 | uint64_t sample_id; | |
13351 | }; | |
13352 | ||
13353 | static enum bpf_perf_event_ret | |
13354 | perf_buffer__process_record(struct perf_event_header *e, void *ctx) | |
13355 | { | |
13356 | struct perf_cpu_buf *cpu_buf = ctx; | |
13357 | struct perf_buffer *pb = cpu_buf->pb; | |
13358 | void *data = e; | |
13359 | ||
13360 | /* user wants full control over parsing perf event */ | |
13361 | if (pb->event_cb) | |
13362 | return pb->event_cb(pb->ctx, cpu_buf->cpu, e); | |
13363 | ||
13364 | switch (e->type) { | |
13365 | case PERF_RECORD_SAMPLE: { | |
13366 | struct perf_sample_raw *s = data; | |
13367 | ||
13368 | if (pb->sample_cb) | |
13369 | pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size); | |
13370 | break; | |
13371 | } | |
13372 | case PERF_RECORD_LOST: { | |
13373 | struct perf_sample_lost *s = data; | |
13374 | ||
13375 | if (pb->lost_cb) | |
13376 | pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost); | |
13377 | break; | |
13378 | } | |
13379 | default: | |
be18010e | 13380 | pr_warn("unknown perf sample type %d\n", e->type); |
fb84b822 AN |
13381 | return LIBBPF_PERF_EVENT_ERROR; |
13382 | } | |
13383 | return LIBBPF_PERF_EVENT_CONT; | |
13384 | } | |
13385 | ||
13386 | static int perf_buffer__process_records(struct perf_buffer *pb, | |
13387 | struct perf_cpu_buf *cpu_buf) | |
13388 | { | |
13389 | enum bpf_perf_event_ret ret; | |
13390 | ||
7218c28c CL |
13391 | ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size, |
13392 | pb->page_size, &cpu_buf->buf, | |
13393 | &cpu_buf->buf_size, | |
13394 | perf_buffer__process_record, cpu_buf); | |
fb84b822 AN |
13395 | if (ret != LIBBPF_PERF_EVENT_CONT) |
13396 | return ret; | |
13397 | return 0; | |
13398 | } | |
13399 | ||
dca5612f AN |
13400 | int perf_buffer__epoll_fd(const struct perf_buffer *pb) |
13401 | { | |
13402 | return pb->epoll_fd; | |
13403 | } | |
13404 | ||
fb84b822 AN |
13405 | int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms) |
13406 | { | |
13407 | int i, cnt, err; | |
13408 | ||
13409 | cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms); | |
e9fc3ce9 | 13410 | if (cnt < 0) |
af0efa05 | 13411 | return -errno; |
e9fc3ce9 | 13412 | |
fb84b822 AN |
13413 | for (i = 0; i < cnt; i++) { |
13414 | struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr; | |
13415 | ||
13416 | err = perf_buffer__process_records(pb, cpu_buf); | |
13417 | if (err) { | |
be18010e | 13418 | pr_warn("error while processing records: %d\n", err); |
e9fc3ce9 | 13419 | return libbpf_err(err); |
fb84b822 AN |
13420 | } |
13421 | } | |
e9fc3ce9 | 13422 | return cnt; |
fb84b822 AN |
13423 | } |
13424 | ||
dca5612f AN |
13425 | /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer |
13426 | * manager. | |
13427 | */ | |
13428 | size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb) | |
13429 | { | |
13430 | return pb->cpu_cnt; | |
13431 | } | |
13432 | ||
13433 | /* | |
13434 | * Return perf_event FD of a ring buffer in *buf_idx* slot of | |
13435 | * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using | |
13436 | * select()/poll()/epoll() Linux syscalls. | |
13437 | */ | |
13438 | int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx) | |
13439 | { | |
13440 | struct perf_cpu_buf *cpu_buf; | |
13441 | ||
13442 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 13443 | return libbpf_err(-EINVAL); |
dca5612f AN |
13444 | |
13445 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
13446 | if (!cpu_buf) | |
e9fc3ce9 | 13447 | return libbpf_err(-ENOENT); |
dca5612f AN |
13448 | |
13449 | return cpu_buf->fd; | |
13450 | } | |
13451 | ||
9ff5efde JD |
13452 | int perf_buffer__buffer(struct perf_buffer *pb, int buf_idx, void **buf, size_t *buf_size) |
13453 | { | |
13454 | struct perf_cpu_buf *cpu_buf; | |
13455 | ||
13456 | if (buf_idx >= pb->cpu_cnt) | |
13457 | return libbpf_err(-EINVAL); | |
13458 | ||
13459 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
13460 | if (!cpu_buf) | |
13461 | return libbpf_err(-ENOENT); | |
13462 | ||
13463 | *buf = cpu_buf->base; | |
13464 | *buf_size = pb->mmap_size; | |
13465 | return 0; | |
13466 | } | |
13467 | ||
dca5612f AN |
13468 | /* |
13469 | * Consume data from perf ring buffer corresponding to slot *buf_idx* in | |
13470 | * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to | |
13471 | * consume, do nothing and return success. | |
13472 | * Returns: | |
13473 | * - 0 on success; | |
13474 | * - <0 on failure. | |
13475 | */ | |
13476 | int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx) | |
13477 | { | |
13478 | struct perf_cpu_buf *cpu_buf; | |
13479 | ||
13480 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 13481 | return libbpf_err(-EINVAL); |
dca5612f AN |
13482 | |
13483 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
13484 | if (!cpu_buf) | |
e9fc3ce9 | 13485 | return libbpf_err(-ENOENT); |
dca5612f AN |
13486 | |
13487 | return perf_buffer__process_records(pb, cpu_buf); | |
13488 | } | |
13489 | ||
272d51af EC |
13490 | int perf_buffer__consume(struct perf_buffer *pb) |
13491 | { | |
13492 | int i, err; | |
13493 | ||
13494 | for (i = 0; i < pb->cpu_cnt; i++) { | |
13495 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; | |
13496 | ||
13497 | if (!cpu_buf) | |
13498 | continue; | |
13499 | ||
13500 | err = perf_buffer__process_records(pb, cpu_buf); | |
13501 | if (err) { | |
dca5612f | 13502 | pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err); |
e9fc3ce9 | 13503 | return libbpf_err(err); |
272d51af EC |
13504 | } |
13505 | } | |
13506 | return 0; | |
13507 | } | |
13508 | ||
ff26ce5c EC |
13509 | int bpf_program__set_attach_target(struct bpf_program *prog, |
13510 | int attach_prog_fd, | |
13511 | const char *attach_func_name) | |
13512 | { | |
fe62de31 | 13513 | int btf_obj_fd = 0, btf_id = 0, err; |
ff26ce5c | 13514 | |
2d5ec1c6 | 13515 | if (!prog || attach_prog_fd < 0) |
e9fc3ce9 | 13516 | return libbpf_err(-EINVAL); |
ff26ce5c | 13517 | |
fe62de31 | 13518 | if (prog->obj->loaded) |
e9fc3ce9 | 13519 | return libbpf_err(-EINVAL); |
fe62de31 | 13520 | |
2d5ec1c6 AN |
13521 | if (attach_prog_fd && !attach_func_name) { |
13522 | /* remember attach_prog_fd and let bpf_program__load() find | |
13523 | * BTF ID during the program load | |
13524 | */ | |
13525 | prog->attach_prog_fd = attach_prog_fd; | |
13526 | return 0; | |
13527 | } | |
13528 | ||
fe62de31 | 13529 | if (attach_prog_fd) { |
ff26ce5c EC |
13530 | btf_id = libbpf_find_prog_btf_id(attach_func_name, |
13531 | attach_prog_fd); | |
fe62de31 | 13532 | if (btf_id < 0) |
e9fc3ce9 | 13533 | return libbpf_err(btf_id); |
fe62de31 | 13534 | } else { |
2d5ec1c6 AN |
13535 | if (!attach_func_name) |
13536 | return libbpf_err(-EINVAL); | |
13537 | ||
fe62de31 AN |
13538 | /* load btf_vmlinux, if not yet */ |
13539 | err = bpf_object__load_vmlinux_btf(prog->obj, true); | |
13540 | if (err) | |
e9fc3ce9 | 13541 | return libbpf_err(err); |
fe62de31 AN |
13542 | err = find_kernel_btf_id(prog->obj, attach_func_name, |
13543 | prog->expected_attach_type, | |
13544 | &btf_obj_fd, &btf_id); | |
13545 | if (err) | |
e9fc3ce9 | 13546 | return libbpf_err(err); |
fe62de31 | 13547 | } |
ff26ce5c EC |
13548 | |
13549 | prog->attach_btf_id = btf_id; | |
fe62de31 | 13550 | prog->attach_btf_obj_fd = btf_obj_fd; |
ff26ce5c EC |
13551 | prog->attach_prog_fd = attach_prog_fd; |
13552 | return 0; | |
13553 | } | |
13554 | ||
6803ee25 | 13555 | int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz) |
6446b315 | 13556 | { |
6803ee25 AN |
13557 | int err = 0, n, len, start, end = -1; |
13558 | bool *tmp; | |
6446b315 | 13559 | |
6803ee25 AN |
13560 | *mask = NULL; |
13561 | *mask_sz = 0; | |
13562 | ||
13563 | /* Each sub string separated by ',' has format \d+-\d+ or \d+ */ | |
13564 | while (*s) { | |
13565 | if (*s == ',' || *s == '\n') { | |
13566 | s++; | |
13567 | continue; | |
13568 | } | |
13569 | n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len); | |
13570 | if (n <= 0 || n > 2) { | |
13571 | pr_warn("Failed to get CPU range %s: %d\n", s, n); | |
13572 | err = -EINVAL; | |
13573 | goto cleanup; | |
13574 | } else if (n == 1) { | |
13575 | end = start; | |
13576 | } | |
13577 | if (start < 0 || start > end) { | |
13578 | pr_warn("Invalid CPU range [%d,%d] in %s\n", | |
13579 | start, end, s); | |
13580 | err = -EINVAL; | |
13581 | goto cleanup; | |
13582 | } | |
13583 | tmp = realloc(*mask, end + 1); | |
13584 | if (!tmp) { | |
13585 | err = -ENOMEM; | |
13586 | goto cleanup; | |
13587 | } | |
13588 | *mask = tmp; | |
13589 | memset(tmp + *mask_sz, 0, start - *mask_sz); | |
13590 | memset(tmp + start, 1, end - start + 1); | |
13591 | *mask_sz = end + 1; | |
13592 | s += len; | |
13593 | } | |
13594 | if (!*mask_sz) { | |
13595 | pr_warn("Empty CPU range\n"); | |
13596 | return -EINVAL; | |
13597 | } | |
13598 | return 0; | |
13599 | cleanup: | |
13600 | free(*mask); | |
13601 | *mask = NULL; | |
13602 | return err; | |
13603 | } | |
13604 | ||
13605 | int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz) | |
13606 | { | |
13607 | int fd, err = 0, len; | |
13608 | char buf[128]; | |
6446b315 | 13609 | |
92274e24 | 13610 | fd = open(fcpu, O_RDONLY | O_CLOEXEC); |
6446b315 | 13611 | if (fd < 0) { |
6803ee25 AN |
13612 | err = -errno; |
13613 | pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err); | |
13614 | return err; | |
6446b315 HL |
13615 | } |
13616 | len = read(fd, buf, sizeof(buf)); | |
13617 | close(fd); | |
13618 | if (len <= 0) { | |
6803ee25 AN |
13619 | err = len ? -errno : -EINVAL; |
13620 | pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err); | |
13621 | return err; | |
6446b315 | 13622 | } |
6803ee25 AN |
13623 | if (len >= sizeof(buf)) { |
13624 | pr_warn("CPU mask is too big in file %s\n", fcpu); | |
13625 | return -E2BIG; | |
6446b315 HL |
13626 | } |
13627 | buf[len] = '\0'; | |
13628 | ||
6803ee25 AN |
13629 | return parse_cpu_mask_str(buf, mask, mask_sz); |
13630 | } | |
13631 | ||
13632 | int libbpf_num_possible_cpus(void) | |
13633 | { | |
13634 | static const char *fcpu = "/sys/devices/system/cpu/possible"; | |
13635 | static int cpus; | |
13636 | int err, n, i, tmp_cpus; | |
13637 | bool *mask; | |
13638 | ||
13639 | tmp_cpus = READ_ONCE(cpus); | |
13640 | if (tmp_cpus > 0) | |
13641 | return tmp_cpus; | |
13642 | ||
13643 | err = parse_cpu_mask_file(fcpu, &mask, &n); | |
13644 | if (err) | |
e9fc3ce9 | 13645 | return libbpf_err(err); |
6803ee25 AN |
13646 | |
13647 | tmp_cpus = 0; | |
13648 | for (i = 0; i < n; i++) { | |
13649 | if (mask[i]) | |
13650 | tmp_cpus++; | |
6446b315 | 13651 | } |
6803ee25 | 13652 | free(mask); |
56fbc241 TC |
13653 | |
13654 | WRITE_ONCE(cpus, tmp_cpus); | |
13655 | return tmp_cpus; | |
6446b315 | 13656 | } |
d66562fb | 13657 | |
430025e5 DK |
13658 | static int populate_skeleton_maps(const struct bpf_object *obj, |
13659 | struct bpf_map_skeleton *maps, | |
13660 | size_t map_cnt) | |
13661 | { | |
13662 | int i; | |
13663 | ||
13664 | for (i = 0; i < map_cnt; i++) { | |
13665 | struct bpf_map **map = maps[i].map; | |
13666 | const char *name = maps[i].name; | |
13667 | void **mmaped = maps[i].mmaped; | |
13668 | ||
13669 | *map = bpf_object__find_map_by_name(obj, name); | |
13670 | if (!*map) { | |
13671 | pr_warn("failed to find skeleton map '%s'\n", name); | |
13672 | return -ESRCH; | |
13673 | } | |
13674 | ||
13675 | /* externs shouldn't be pre-setup from user code */ | |
13676 | if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG) | |
13677 | *mmaped = (*map)->mmaped; | |
13678 | } | |
13679 | return 0; | |
13680 | } | |
13681 | ||
13682 | static int populate_skeleton_progs(const struct bpf_object *obj, | |
13683 | struct bpf_prog_skeleton *progs, | |
13684 | size_t prog_cnt) | |
13685 | { | |
13686 | int i; | |
13687 | ||
13688 | for (i = 0; i < prog_cnt; i++) { | |
13689 | struct bpf_program **prog = progs[i].prog; | |
13690 | const char *name = progs[i].name; | |
13691 | ||
13692 | *prog = bpf_object__find_program_by_name(obj, name); | |
13693 | if (!*prog) { | |
13694 | pr_warn("failed to find skeleton program '%s'\n", name); | |
13695 | return -ESRCH; | |
13696 | } | |
13697 | } | |
13698 | return 0; | |
13699 | } | |
13700 | ||
d66562fb AN |
13701 | int bpf_object__open_skeleton(struct bpf_object_skeleton *s, |
13702 | const struct bpf_object_open_opts *opts) | |
13703 | { | |
13704 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts, | |
13705 | .object_name = s->name, | |
13706 | ); | |
13707 | struct bpf_object *obj; | |
430025e5 | 13708 | int err; |
d66562fb AN |
13709 | |
13710 | /* Attempt to preserve opts->object_name, unless overriden by user | |
13711 | * explicitly. Overwriting object name for skeletons is discouraged, | |
13712 | * as it breaks global data maps, because they contain object name | |
13713 | * prefix as their own map name prefix. When skeleton is generated, | |
13714 | * bpftool is making an assumption that this name will stay the same. | |
13715 | */ | |
13716 | if (opts) { | |
13717 | memcpy(&skel_opts, opts, sizeof(*opts)); | |
13718 | if (!opts->object_name) | |
13719 | skel_opts.object_name = s->name; | |
13720 | } | |
13721 | ||
13722 | obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts); | |
e9fc3ce9 AN |
13723 | err = libbpf_get_error(obj); |
13724 | if (err) { | |
13725 | pr_warn("failed to initialize skeleton BPF object '%s': %d\n", | |
13726 | s->name, err); | |
13727 | return libbpf_err(err); | |
d66562fb AN |
13728 | } |
13729 | ||
13730 | *s->obj = obj; | |
430025e5 DK |
13731 | err = populate_skeleton_maps(obj, s->maps, s->map_cnt); |
13732 | if (err) { | |
13733 | pr_warn("failed to populate skeleton maps for '%s': %d\n", s->name, err); | |
13734 | return libbpf_err(err); | |
13735 | } | |
13736 | ||
13737 | err = populate_skeleton_progs(obj, s->progs, s->prog_cnt); | |
13738 | if (err) { | |
13739 | pr_warn("failed to populate skeleton progs for '%s': %d\n", s->name, err); | |
13740 | return libbpf_err(err); | |
13741 | } | |
d66562fb | 13742 | |
430025e5 DK |
13743 | return 0; |
13744 | } | |
d66562fb | 13745 | |
430025e5 DK |
13746 | int bpf_object__open_subskeleton(struct bpf_object_subskeleton *s) |
13747 | { | |
13748 | int err, len, var_idx, i; | |
13749 | const char *var_name; | |
13750 | const struct bpf_map *map; | |
13751 | struct btf *btf; | |
13752 | __u32 map_type_id; | |
13753 | const struct btf_type *map_type, *var_type; | |
13754 | const struct bpf_var_skeleton *var_skel; | |
13755 | struct btf_var_secinfo *var; | |
d66562fb | 13756 | |
430025e5 DK |
13757 | if (!s->obj) |
13758 | return libbpf_err(-EINVAL); | |
13759 | ||
13760 | btf = bpf_object__btf(s->obj); | |
13761 | if (!btf) { | |
13762 | pr_warn("subskeletons require BTF at runtime (object %s)\n", | |
e3ba8e4e | 13763 | bpf_object__name(s->obj)); |
430025e5 | 13764 | return libbpf_err(-errno); |
d66562fb AN |
13765 | } |
13766 | ||
430025e5 DK |
13767 | err = populate_skeleton_maps(s->obj, s->maps, s->map_cnt); |
13768 | if (err) { | |
13769 | pr_warn("failed to populate subskeleton maps: %d\n", err); | |
13770 | return libbpf_err(err); | |
13771 | } | |
d66562fb | 13772 | |
430025e5 DK |
13773 | err = populate_skeleton_progs(s->obj, s->progs, s->prog_cnt); |
13774 | if (err) { | |
13775 | pr_warn("failed to populate subskeleton maps: %d\n", err); | |
13776 | return libbpf_err(err); | |
d66562fb AN |
13777 | } |
13778 | ||
430025e5 DK |
13779 | for (var_idx = 0; var_idx < s->var_cnt; var_idx++) { |
13780 | var_skel = &s->vars[var_idx]; | |
13781 | map = *var_skel->map; | |
13782 | map_type_id = bpf_map__btf_value_type_id(map); | |
13783 | map_type = btf__type_by_id(btf, map_type_id); | |
13784 | ||
13785 | if (!btf_is_datasec(map_type)) { | |
13786 | pr_warn("type for map '%1$s' is not a datasec: %2$s", | |
13787 | bpf_map__name(map), | |
13788 | __btf_kind_str(btf_kind(map_type))); | |
13789 | return libbpf_err(-EINVAL); | |
13790 | } | |
13791 | ||
13792 | len = btf_vlen(map_type); | |
13793 | var = btf_var_secinfos(map_type); | |
13794 | for (i = 0; i < len; i++, var++) { | |
13795 | var_type = btf__type_by_id(btf, var->type); | |
13796 | var_name = btf__name_by_offset(btf, var_type->name_off); | |
13797 | if (strcmp(var_name, var_skel->name) == 0) { | |
13798 | *var_skel->addr = map->mmaped + var->offset; | |
13799 | break; | |
13800 | } | |
13801 | } | |
13802 | } | |
d66562fb AN |
13803 | return 0; |
13804 | } | |
13805 | ||
430025e5 DK |
13806 | void bpf_object__destroy_subskeleton(struct bpf_object_subskeleton *s) |
13807 | { | |
13808 | if (!s) | |
13809 | return; | |
13810 | free(s->maps); | |
13811 | free(s->progs); | |
13812 | free(s->vars); | |
13813 | free(s); | |
13814 | } | |
13815 | ||
d66562fb AN |
13816 | int bpf_object__load_skeleton(struct bpf_object_skeleton *s) |
13817 | { | |
13818 | int i, err; | |
13819 | ||
13820 | err = bpf_object__load(*s->obj); | |
13821 | if (err) { | |
13822 | pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err); | |
e9fc3ce9 | 13823 | return libbpf_err(err); |
d66562fb AN |
13824 | } |
13825 | ||
13826 | for (i = 0; i < s->map_cnt; i++) { | |
13827 | struct bpf_map *map = *s->maps[i].map; | |
79ff13e9 | 13828 | size_t mmap_sz = bpf_map_mmap_sz(map); |
f08c18e0 | 13829 | int prot, map_fd = map->fd; |
d66562fb | 13830 | void **mmaped = s->maps[i].mmaped; |
d66562fb AN |
13831 | |
13832 | if (!mmaped) | |
13833 | continue; | |
13834 | ||
13835 | if (!(map->def.map_flags & BPF_F_MMAPABLE)) { | |
13836 | *mmaped = NULL; | |
13837 | continue; | |
13838 | } | |
13839 | ||
2e7ba4f8 AN |
13840 | if (map->def.type == BPF_MAP_TYPE_ARENA) { |
13841 | *mmaped = map->mmaped; | |
13842 | continue; | |
13843 | } | |
13844 | ||
d66562fb AN |
13845 | if (map->def.map_flags & BPF_F_RDONLY_PROG) |
13846 | prot = PROT_READ; | |
13847 | else | |
13848 | prot = PROT_READ | PROT_WRITE; | |
13849 | ||
13850 | /* Remap anonymous mmap()-ed "map initialization image" as | |
13851 | * a BPF map-backed mmap()-ed memory, but preserving the same | |
13852 | * memory address. This will cause kernel to change process' | |
13853 | * page table to point to a different piece of kernel memory, | |
13854 | * but from userspace point of view memory address (and its | |
13855 | * contents, being identical at this point) will stay the | |
13856 | * same. This mapping will be released by bpf_object__close() | |
13857 | * as per normal clean up procedure, so we don't need to worry | |
13858 | * about it from skeleton's clean up perspective. | |
13859 | */ | |
9d0a2331 | 13860 | *mmaped = mmap(map->mmaped, mmap_sz, prot, MAP_SHARED | MAP_FIXED, map_fd, 0); |
2ad97d47 | 13861 | if (*mmaped == MAP_FAILED) { |
d66562fb AN |
13862 | err = -errno; |
13863 | *mmaped = NULL; | |
13864 | pr_warn("failed to re-mmap() map '%s': %d\n", | |
13865 | bpf_map__name(map), err); | |
e9fc3ce9 | 13866 | return libbpf_err(err); |
d66562fb AN |
13867 | } |
13868 | } | |
13869 | ||
13870 | return 0; | |
13871 | } | |
13872 | ||
13873 | int bpf_object__attach_skeleton(struct bpf_object_skeleton *s) | |
13874 | { | |
e9fc3ce9 | 13875 | int i, err; |
d66562fb AN |
13876 | |
13877 | for (i = 0; i < s->prog_cnt; i++) { | |
13878 | struct bpf_program *prog = *s->progs[i].prog; | |
13879 | struct bpf_link **link = s->progs[i].link; | |
d66562fb | 13880 | |
43cb8cba | 13881 | if (!prog->autoload || !prog->autoattach) |
d9297581 AN |
13882 | continue; |
13883 | ||
5532dfd4 | 13884 | /* auto-attaching not supported for this program */ |
4fa5bcfe | 13885 | if (!prog->sec_def || !prog->sec_def->prog_attach_fn) |
d66562fb AN |
13886 | continue; |
13887 | ||
4fa5bcfe AN |
13888 | /* if user already set the link manually, don't attempt auto-attach */ |
13889 | if (*link) | |
13890 | continue; | |
13891 | ||
13892 | err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link); | |
e9fc3ce9 | 13893 | if (err) { |
4fa5bcfe | 13894 | pr_warn("prog '%s': failed to auto-attach: %d\n", |
e9fc3ce9 AN |
13895 | bpf_program__name(prog), err); |
13896 | return libbpf_err(err); | |
d66562fb | 13897 | } |
4fa5bcfe AN |
13898 | |
13899 | /* It's possible that for some SEC() definitions auto-attach | |
13900 | * is supported in some cases (e.g., if definition completely | |
13901 | * specifies target information), but is not in other cases. | |
13902 | * SEC("uprobe") is one such case. If user specified target | |
13903 | * binary and function name, such BPF program can be | |
13904 | * auto-attached. But if not, it shouldn't trigger skeleton's | |
13905 | * attach to fail. It should just be skipped. | |
13906 | * attach_fn signals such case with returning 0 (no error) and | |
13907 | * setting link to NULL. | |
13908 | */ | |
d66562fb AN |
13909 | } |
13910 | ||
13911 | return 0; | |
13912 | } | |
13913 | ||
13914 | void bpf_object__detach_skeleton(struct bpf_object_skeleton *s) | |
13915 | { | |
13916 | int i; | |
13917 | ||
13918 | for (i = 0; i < s->prog_cnt; i++) { | |
13919 | struct bpf_link **link = s->progs[i].link; | |
13920 | ||
50450fc7 | 13921 | bpf_link__destroy(*link); |
d66562fb AN |
13922 | *link = NULL; |
13923 | } | |
13924 | } | |
13925 | ||
13926 | void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s) | |
13927 | { | |
a32ea51a YS |
13928 | if (!s) |
13929 | return; | |
13930 | ||
d66562fb AN |
13931 | if (s->progs) |
13932 | bpf_object__detach_skeleton(s); | |
13933 | if (s->obj) | |
13934 | bpf_object__close(*s->obj); | |
13935 | free(s->maps); | |
13936 | free(s->progs); | |
13937 | free(s); | |
13938 | } |