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; | |
1131 | __s64 msize, kern_msize; | |
1132 | __u32 moff, kern_moff; | |
1133 | __u32 kern_member_idx; | |
1134 | const char *mname; | |
1135 | ||
1136 | mname = btf__name_by_offset(btf, member->name_off); | |
c911fc61 KFL |
1137 | moff = member->offset / 8; |
1138 | mdata = data + moff; | |
1139 | msize = btf__resolve_size(btf, member->type); | |
1140 | if (msize < 0) { | |
1141 | pr_warn("struct_ops init_kern %s: failed to resolve the size of member %s\n", | |
1142 | map->name, mname); | |
1143 | return msize; | |
1144 | } | |
1145 | ||
590a0088 MKL |
1146 | kern_member = find_member_by_name(kern_btf, kern_type, mname); |
1147 | if (!kern_member) { | |
c911fc61 KFL |
1148 | /* Skip all zeros or null fields if they are not |
1149 | * presented in the kernel BTF. | |
1150 | */ | |
1151 | if (libbpf_is_mem_zeroed(mdata, msize)) { | |
f973fccd | 1152 | st_ops->progs[i] = NULL; |
c911fc61 KFL |
1153 | pr_info("struct_ops %s: member %s not found in kernel, skipping it as it's set to zero\n", |
1154 | map->name, mname); | |
1155 | continue; | |
1156 | } | |
1157 | ||
590a0088 MKL |
1158 | pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n", |
1159 | map->name, mname); | |
1160 | return -ENOTSUP; | |
1161 | } | |
1162 | ||
1163 | kern_member_idx = kern_member - btf_members(kern_type); | |
1164 | if (btf_member_bitfield_size(type, i) || | |
1165 | btf_member_bitfield_size(kern_type, kern_member_idx)) { | |
1166 | pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n", | |
1167 | map->name, mname); | |
1168 | return -ENOTSUP; | |
1169 | } | |
1170 | ||
590a0088 | 1171 | kern_moff = kern_member->offset / 8; |
590a0088 MKL |
1172 | kern_mdata = kern_data + kern_moff; |
1173 | ||
1174 | mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id); | |
1175 | kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type, | |
1176 | &kern_mtype_id); | |
1177 | if (BTF_INFO_KIND(mtype->info) != | |
1178 | BTF_INFO_KIND(kern_mtype->info)) { | |
1179 | pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n", | |
1180 | map->name, mname, BTF_INFO_KIND(mtype->info), | |
1181 | BTF_INFO_KIND(kern_mtype->info)); | |
1182 | return -ENOTSUP; | |
1183 | } | |
1184 | ||
1185 | if (btf_is_ptr(mtype)) { | |
1186 | struct bpf_program *prog; | |
1187 | ||
69e4a9d2 KFL |
1188 | /* Update the value from the shadow type */ |
1189 | prog = *(void **)mdata; | |
1190 | st_ops->progs[i] = prog; | |
d2836ddd MKL |
1191 | if (!prog) |
1192 | continue; | |
69e4a9d2 KFL |
1193 | if (!is_valid_st_ops_program(obj, prog)) { |
1194 | pr_warn("struct_ops init_kern %s: member %s is not a struct_ops program\n", | |
1195 | map->name, mname); | |
1196 | return -ENOTSUP; | |
1197 | } | |
d2836ddd | 1198 | |
590a0088 MKL |
1199 | kern_mtype = skip_mods_and_typedefs(kern_btf, |
1200 | kern_mtype->type, | |
1201 | &kern_mtype_id); | |
d2836ddd MKL |
1202 | |
1203 | /* mtype->type must be a func_proto which was | |
1204 | * guaranteed in bpf_object__collect_st_ops_relos(), | |
1205 | * so only check kern_mtype for func_proto here. | |
1206 | */ | |
1207 | if (!btf_is_func_proto(kern_mtype)) { | |
1208 | pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n", | |
590a0088 MKL |
1209 | map->name, mname); |
1210 | return -ENOTSUP; | |
1211 | } | |
1212 | ||
9e926acd KFL |
1213 | if (mod_btf) |
1214 | prog->attach_btf_obj_fd = mod_btf->fd; | |
d9ab2f76 EZ |
1215 | |
1216 | /* if we haven't yet processed this BPF program, record proper | |
1217 | * attach_btf_id and member_idx | |
1218 | */ | |
1219 | if (!prog->attach_btf_id) { | |
1220 | prog->attach_btf_id = kern_type_id; | |
1221 | prog->expected_attach_type = kern_member_idx; | |
1222 | } | |
1223 | ||
1224 | /* struct_ops BPF prog can be re-used between multiple | |
1225 | * .struct_ops & .struct_ops.link as long as it's the | |
1226 | * same struct_ops struct definition and the same | |
1227 | * function pointer field | |
1228 | */ | |
1229 | if (prog->attach_btf_id != kern_type_id) { | |
1230 | 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", | |
1231 | map->name, mname, prog->name, prog->sec_name, prog->type, | |
1232 | prog->attach_btf_id, kern_type_id); | |
1233 | return -EINVAL; | |
1234 | } | |
1235 | if (prog->expected_attach_type != kern_member_idx) { | |
1236 | 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", | |
1237 | map->name, mname, prog->name, prog->sec_name, prog->type, | |
1238 | prog->expected_attach_type, kern_member_idx); | |
1239 | return -EINVAL; | |
1240 | } | |
590a0088 MKL |
1241 | |
1242 | st_ops->kern_func_off[i] = kern_data_off + kern_moff; | |
1243 | ||
1244 | pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n", | |
1245 | map->name, mname, prog->name, moff, | |
1246 | kern_moff); | |
1247 | ||
1248 | continue; | |
1249 | } | |
1250 | ||
590a0088 | 1251 | kern_msize = btf__resolve_size(kern_btf, kern_mtype_id); |
c911fc61 | 1252 | if (kern_msize < 0 || msize != kern_msize) { |
590a0088 MKL |
1253 | pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n", |
1254 | map->name, mname, (ssize_t)msize, | |
1255 | (ssize_t)kern_msize); | |
1256 | return -ENOTSUP; | |
1257 | } | |
1258 | ||
1259 | pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n", | |
1260 | map->name, mname, (unsigned int)msize, | |
1261 | moff, kern_moff); | |
1262 | memcpy(kern_mdata, mdata, msize); | |
1263 | } | |
1264 | ||
1265 | return 0; | |
1266 | } | |
1267 | ||
1268 | static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj) | |
1269 | { | |
590a0088 MKL |
1270 | struct bpf_map *map; |
1271 | size_t i; | |
1272 | int err; | |
1273 | ||
1274 | for (i = 0; i < obj->nr_maps; i++) { | |
1275 | map = &obj->maps[i]; | |
1276 | ||
1277 | if (!bpf_map__is_struct_ops(map)) | |
1278 | continue; | |
1279 | ||
8db05261 EZ |
1280 | if (!map->autocreate) |
1281 | continue; | |
1282 | ||
9e926acd | 1283 | err = bpf_map__init_kern_struct_ops(map); |
a6ed02ca | 1284 | if (err) |
590a0088 | 1285 | return err; |
590a0088 MKL |
1286 | } |
1287 | ||
590a0088 MKL |
1288 | return 0; |
1289 | } | |
1290 | ||
809a69d6 | 1291 | static int init_struct_ops_maps(struct bpf_object *obj, const char *sec_name, |
240bf8a5 | 1292 | int shndx, Elf_Data *data) |
590a0088 MKL |
1293 | { |
1294 | const struct btf_type *type, *datasec; | |
1295 | const struct btf_var_secinfo *vsi; | |
1296 | struct bpf_struct_ops *st_ops; | |
1297 | const char *tname, *var_name; | |
1298 | __s32 type_id, datasec_id; | |
1299 | const struct btf *btf; | |
1300 | struct bpf_map *map; | |
1301 | __u32 i; | |
1302 | ||
809a69d6 | 1303 | if (shndx == -1) |
590a0088 MKL |
1304 | return 0; |
1305 | ||
1306 | btf = obj->btf; | |
809a69d6 | 1307 | datasec_id = btf__find_by_name_kind(btf, sec_name, |
590a0088 MKL |
1308 | BTF_KIND_DATASEC); |
1309 | if (datasec_id < 0) { | |
1310 | pr_warn("struct_ops init: DATASEC %s not found\n", | |
809a69d6 | 1311 | sec_name); |
590a0088 MKL |
1312 | return -EINVAL; |
1313 | } | |
1314 | ||
1315 | datasec = btf__type_by_id(btf, datasec_id); | |
1316 | vsi = btf_var_secinfos(datasec); | |
1317 | for (i = 0; i < btf_vlen(datasec); i++, vsi++) { | |
1318 | type = btf__type_by_id(obj->btf, vsi->type); | |
1319 | var_name = btf__name_by_offset(obj->btf, type->name_off); | |
1320 | ||
1321 | type_id = btf__resolve_type(obj->btf, vsi->type); | |
1322 | if (type_id < 0) { | |
1323 | pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n", | |
809a69d6 | 1324 | vsi->type, sec_name); |
590a0088 MKL |
1325 | return -EINVAL; |
1326 | } | |
1327 | ||
1328 | type = btf__type_by_id(obj->btf, type_id); | |
1329 | tname = btf__name_by_offset(obj->btf, type->name_off); | |
1330 | if (!tname[0]) { | |
1331 | pr_warn("struct_ops init: anonymous type is not supported\n"); | |
1332 | return -ENOTSUP; | |
1333 | } | |
1334 | if (!btf_is_struct(type)) { | |
1335 | pr_warn("struct_ops init: %s is not a struct\n", tname); | |
1336 | return -EINVAL; | |
1337 | } | |
1338 | ||
1339 | map = bpf_object__add_map(obj); | |
1340 | if (IS_ERR(map)) | |
1341 | return PTR_ERR(map); | |
1342 | ||
809a69d6 | 1343 | map->sec_idx = shndx; |
590a0088 MKL |
1344 | map->sec_offset = vsi->offset; |
1345 | map->name = strdup(var_name); | |
1346 | if (!map->name) | |
1347 | return -ENOMEM; | |
3644d285 | 1348 | map->btf_value_type_id = type_id; |
590a0088 | 1349 | |
5ad0ecbe EZ |
1350 | /* Follow same convention as for programs autoload: |
1351 | * SEC("?.struct_ops") means map is not created by default. | |
1352 | */ | |
1353 | if (sec_name[0] == '?') { | |
1354 | map->autocreate = false; | |
1355 | /* from now on forget there was ? in section name */ | |
1356 | sec_name++; | |
1357 | } | |
1358 | ||
590a0088 MKL |
1359 | map->def.type = BPF_MAP_TYPE_STRUCT_OPS; |
1360 | map->def.key_size = sizeof(int); | |
1361 | map->def.value_size = type->size; | |
1362 | map->def.max_entries = 1; | |
240bf8a5 | 1363 | map->def.map_flags = strcmp(sec_name, STRUCT_OPS_LINK_SEC) == 0 ? BPF_F_LINK : 0; |
590a0088 MKL |
1364 | |
1365 | map->st_ops = calloc(1, sizeof(*map->st_ops)); | |
1366 | if (!map->st_ops) | |
1367 | return -ENOMEM; | |
1368 | st_ops = map->st_ops; | |
1369 | st_ops->data = malloc(type->size); | |
1370 | st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs)); | |
1371 | st_ops->kern_func_off = malloc(btf_vlen(type) * | |
1372 | sizeof(*st_ops->kern_func_off)); | |
1373 | if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off) | |
1374 | return -ENOMEM; | |
1375 | ||
809a69d6 | 1376 | if (vsi->offset + type->size > data->d_size) { |
590a0088 | 1377 | pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n", |
809a69d6 | 1378 | var_name, sec_name); |
590a0088 MKL |
1379 | return -EINVAL; |
1380 | } | |
1381 | ||
1382 | memcpy(st_ops->data, | |
809a69d6 | 1383 | data->d_buf + vsi->offset, |
590a0088 MKL |
1384 | type->size); |
1385 | st_ops->tname = tname; | |
1386 | st_ops->type = type; | |
1387 | st_ops->type_id = type_id; | |
1388 | ||
1389 | pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n", | |
1390 | tname, type_id, var_name, vsi->offset); | |
1391 | } | |
1392 | ||
1393 | return 0; | |
1394 | } | |
1395 | ||
809a69d6 KFL |
1396 | static int bpf_object_init_struct_ops(struct bpf_object *obj) |
1397 | { | |
240bf8a5 EZ |
1398 | const char *sec_name; |
1399 | int sec_idx, err; | |
809a69d6 | 1400 | |
240bf8a5 EZ |
1401 | for (sec_idx = 0; sec_idx < obj->efile.sec_cnt; ++sec_idx) { |
1402 | struct elf_sec_desc *desc = &obj->efile.secs[sec_idx]; | |
1403 | ||
1404 | if (desc->sec_type != SEC_ST_OPS) | |
1405 | continue; | |
1406 | ||
1407 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); | |
1408 | if (!sec_name) | |
1409 | return -LIBBPF_ERRNO__FORMAT; | |
1410 | ||
1411 | err = init_struct_ops_maps(obj, sec_name, sec_idx, desc->data); | |
1412 | if (err) | |
1413 | return err; | |
1414 | } | |
1415 | ||
1416 | return 0; | |
809a69d6 KFL |
1417 | } |
1418 | ||
6c956392 | 1419 | static struct bpf_object *bpf_object__new(const char *path, |
5e61f270 | 1420 | const void *obj_buf, |
2ce8450e AN |
1421 | size_t obj_buf_sz, |
1422 | const char *obj_name) | |
1a5e3fb1 WN |
1423 | { |
1424 | struct bpf_object *obj; | |
d859900c | 1425 | char *end; |
1a5e3fb1 WN |
1426 | |
1427 | obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1); | |
1428 | if (!obj) { | |
be18010e | 1429 | pr_warn("alloc memory failed for %s\n", path); |
6371ca3b | 1430 | return ERR_PTR(-ENOMEM); |
1a5e3fb1 WN |
1431 | } |
1432 | ||
1433 | strcpy(obj->path, path); | |
2ce8450e | 1434 | if (obj_name) { |
9fc205b4 | 1435 | libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name)); |
2ce8450e AN |
1436 | } else { |
1437 | /* Using basename() GNU version which doesn't modify arg. */ | |
9fc205b4 | 1438 | libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name)); |
2ce8450e AN |
1439 | end = strchr(obj->name, '.'); |
1440 | if (end) | |
1441 | *end = 0; | |
1442 | } | |
6c956392 | 1443 | |
d859900c | 1444 | obj->efile.fd = -1; |
6c956392 | 1445 | /* |
76e1022b | 1446 | * Caller of this function should also call |
6c956392 WN |
1447 | * bpf_object__elf_finish() after data collection to return |
1448 | * obj_buf to user. If not, we should duplicate the buffer to | |
1449 | * avoid user freeing them before elf finish. | |
1450 | */ | |
1451 | obj->efile.obj_buf = obj_buf; | |
1452 | obj->efile.obj_buf_sz = obj_buf_sz; | |
abd29c93 | 1453 | obj->efile.btf_maps_shndx = -1; |
81bfdd08 | 1454 | obj->kconfig_map_idx = -1; |
6c956392 | 1455 | |
5e61f270 | 1456 | obj->kern_version = get_kernel_version(); |
52d3352e | 1457 | obj->loaded = false; |
9a208eff | 1458 | |
1a5e3fb1 WN |
1459 | return obj; |
1460 | } | |
1461 | ||
1462 | static void bpf_object__elf_finish(struct bpf_object *obj) | |
1463 | { | |
29a30ff5 | 1464 | if (!obj->efile.elf) |
1a5e3fb1 WN |
1465 | return; |
1466 | ||
b71a2ebf GC |
1467 | elf_end(obj->efile.elf); |
1468 | obj->efile.elf = NULL; | |
bec7d68c | 1469 | obj->efile.symbols = NULL; |
2e7ba4f8 | 1470 | obj->efile.arena_data = NULL; |
b62f06e8 | 1471 | |
25bbbd7a AN |
1472 | zfree(&obj->efile.secs); |
1473 | obj->efile.sec_cnt = 0; | |
1a5e3fb1 | 1474 | zclose(obj->efile.fd); |
6c956392 WN |
1475 | obj->efile.obj_buf = NULL; |
1476 | obj->efile.obj_buf_sz = 0; | |
1a5e3fb1 WN |
1477 | } |
1478 | ||
1479 | static int bpf_object__elf_init(struct bpf_object *obj) | |
1480 | { | |
ad23b723 | 1481 | Elf64_Ehdr *ehdr; |
1a5e3fb1 | 1482 | int err = 0; |
ad23b723 | 1483 | Elf *elf; |
1a5e3fb1 | 1484 | |
29a30ff5 | 1485 | if (obj->efile.elf) { |
88a82120 | 1486 | pr_warn("elf: init internal error\n"); |
6371ca3b | 1487 | return -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1488 | } |
1489 | ||
6c956392 | 1490 | if (obj->efile.obj_buf_sz > 0) { |
146bf811 | 1491 | /* obj_buf should have been validated by bpf_object__open_mem(). */ |
ad23b723 | 1492 | elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz); |
6c956392 | 1493 | } else { |
92274e24 | 1494 | obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC); |
6c956392 | 1495 | if (obj->efile.fd < 0) { |
be5c5d4e | 1496 | char errmsg[STRERR_BUFSIZE], *cp; |
1ce6a9fc | 1497 | |
be5c5d4e AN |
1498 | err = -errno; |
1499 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
88a82120 | 1500 | pr_warn("elf: failed to open %s: %s\n", obj->path, cp); |
be5c5d4e | 1501 | return err; |
6c956392 WN |
1502 | } |
1503 | ||
ad23b723 | 1504 | elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL); |
1a5e3fb1 WN |
1505 | } |
1506 | ||
ad23b723 | 1507 | if (!elf) { |
88a82120 | 1508 | pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1509 | err = -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1510 | goto errout; |
1511 | } | |
1512 | ||
ad23b723 AN |
1513 | obj->efile.elf = elf; |
1514 | ||
1515 | if (elf_kind(elf) != ELF_K_ELF) { | |
1516 | err = -LIBBPF_ERRNO__FORMAT; | |
1517 | pr_warn("elf: '%s' is not a proper ELF object\n", obj->path); | |
1518 | goto errout; | |
1519 | } | |
1520 | ||
1521 | if (gelf_getclass(elf) != ELFCLASS64) { | |
1522 | err = -LIBBPF_ERRNO__FORMAT; | |
1523 | pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path); | |
1524 | goto errout; | |
1525 | } | |
1526 | ||
1527 | obj->efile.ehdr = ehdr = elf64_getehdr(elf); | |
1528 | if (!obj->efile.ehdr) { | |
88a82120 | 1529 | pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1530 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1531 | goto errout; |
1532 | } | |
1a5e3fb1 | 1533 | |
ad23b723 | 1534 | if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) { |
88a82120 AN |
1535 | pr_warn("elf: failed to get section names section index for %s: %s\n", |
1536 | obj->path, elf_errmsg(-1)); | |
1537 | err = -LIBBPF_ERRNO__FORMAT; | |
1538 | goto errout; | |
1539 | } | |
1540 | ||
70e79866 | 1541 | /* ELF is corrupted/truncated, avoid calling elf_strptr. */ |
ad23b723 | 1542 | if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) { |
88a82120 AN |
1543 | pr_warn("elf: failed to get section names strings from %s: %s\n", |
1544 | obj->path, elf_errmsg(-1)); | |
8f3f5792 NK |
1545 | err = -LIBBPF_ERRNO__FORMAT; |
1546 | goto errout; | |
88a82120 AN |
1547 | } |
1548 | ||
9b16137a | 1549 | /* Old LLVM set e_machine to EM_NONE */ |
ad23b723 | 1550 | if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) { |
88a82120 | 1551 | pr_warn("elf: %s is not a valid eBPF object file\n", obj->path); |
6371ca3b | 1552 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1553 | goto errout; |
1554 | } | |
1555 | ||
1556 | return 0; | |
1557 | errout: | |
1558 | bpf_object__elf_finish(obj); | |
1559 | return err; | |
1560 | } | |
1561 | ||
12ef5634 | 1562 | static int bpf_object__check_endianness(struct bpf_object *obj) |
cc4228d5 | 1563 | { |
3930198d | 1564 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
ad23b723 | 1565 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB) |
12ef5634 | 1566 | return 0; |
3930198d | 1567 | #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
ad23b723 | 1568 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB) |
12ef5634 AN |
1569 | return 0; |
1570 | #else | |
1571 | # error "Unrecognized __BYTE_ORDER__" | |
1572 | #endif | |
88a82120 | 1573 | pr_warn("elf: endianness mismatch in %s.\n", obj->path); |
6371ca3b | 1574 | return -LIBBPF_ERRNO__ENDIAN; |
cc4228d5 WN |
1575 | } |
1576 | ||
cb1e5e96 | 1577 | static int |
399dc65e | 1578 | bpf_object__init_license(struct bpf_object *obj, void *data, size_t size) |
cb1e5e96 | 1579 | { |
35a85550 SHY |
1580 | if (!data) { |
1581 | pr_warn("invalid license section in %s\n", obj->path); | |
1582 | return -LIBBPF_ERRNO__FORMAT; | |
1583 | } | |
f9798239 AN |
1584 | /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't |
1585 | * go over allowed ELF data section buffer | |
1586 | */ | |
1587 | libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license))); | |
cb1e5e96 WN |
1588 | pr_debug("license of %s is %s\n", obj->path, obj->license); |
1589 | return 0; | |
1590 | } | |
1591 | ||
54b8625c JF |
1592 | static int |
1593 | bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size) | |
1594 | { | |
1595 | __u32 kver; | |
1596 | ||
35a85550 | 1597 | if (!data || size != sizeof(kver)) { |
be18010e | 1598 | pr_warn("invalid kver section in %s\n", obj->path); |
54b8625c JF |
1599 | return -LIBBPF_ERRNO__FORMAT; |
1600 | } | |
1601 | memcpy(&kver, data, sizeof(kver)); | |
1602 | obj->kern_version = kver; | |
1603 | pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version); | |
1604 | return 0; | |
1605 | } | |
1606 | ||
addb9fc9 NS |
1607 | static bool bpf_map_type__is_map_in_map(enum bpf_map_type type) |
1608 | { | |
1609 | if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1610 | type == BPF_MAP_TYPE_HASH_OF_MAPS) | |
1611 | return true; | |
1612 | return false; | |
1613 | } | |
1614 | ||
b96c07f3 | 1615 | static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size) |
1713d68b | 1616 | { |
25bbbd7a AN |
1617 | Elf_Data *data; |
1618 | Elf_Scn *scn; | |
1713d68b | 1619 | |
25bbbd7a | 1620 | if (!name) |
1713d68b | 1621 | return -EINVAL; |
88a82120 | 1622 | |
25bbbd7a AN |
1623 | scn = elf_sec_by_name(obj, name); |
1624 | data = elf_sec_data(obj, scn); | |
1625 | if (data) { | |
25bbbd7a | 1626 | *size = data->d_size; |
08894d9c | 1627 | return 0; /* found it */ |
1713d68b DB |
1628 | } |
1629 | ||
08894d9c | 1630 | return -ENOENT; |
1713d68b DB |
1631 | } |
1632 | ||
f33f742d | 1633 | static Elf64_Sym *find_elf_var_sym(const struct bpf_object *obj, const char *name) |
1713d68b DB |
1634 | { |
1635 | Elf_Data *symbols = obj->efile.symbols; | |
1636 | const char *sname; | |
1637 | size_t si; | |
1638 | ||
ad23b723 AN |
1639 | for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) { |
1640 | Elf64_Sym *sym = elf_sym_by_idx(obj, si); | |
1713d68b | 1641 | |
3a06ec0a AN |
1642 | if (ELF64_ST_TYPE(sym->st_info) != STT_OBJECT) |
1643 | continue; | |
1644 | ||
1645 | if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL && | |
1646 | ELF64_ST_BIND(sym->st_info) != STB_WEAK) | |
1713d68b DB |
1647 | continue; |
1648 | ||
ad23b723 | 1649 | sname = elf_sym_str(obj, sym->st_name); |
1713d68b | 1650 | if (!sname) { |
ad23b723 | 1651 | pr_warn("failed to get sym name string for var %s\n", name); |
f33f742d | 1652 | return ERR_PTR(-EIO); |
1713d68b | 1653 | } |
f33f742d AN |
1654 | if (strcmp(name, sname) == 0) |
1655 | return sym; | |
1713d68b DB |
1656 | } |
1657 | ||
f33f742d | 1658 | return ERR_PTR(-ENOENT); |
1713d68b DB |
1659 | } |
1660 | ||
9fa5e1a1 AN |
1661 | /* Some versions of Android don't provide memfd_create() in their libc |
1662 | * implementation, so avoid complications and just go straight to Linux | |
1663 | * syscall. | |
1664 | */ | |
1665 | static int sys_memfd_create(const char *name, unsigned flags) | |
1666 | { | |
1667 | return syscall(__NR_memfd_create, name, flags); | |
1668 | } | |
1669 | ||
ddb2ffdc ACM |
1670 | #ifndef MFD_CLOEXEC |
1671 | #define MFD_CLOEXEC 0x0001U | |
1672 | #endif | |
1673 | ||
dac645b9 AN |
1674 | static int create_placeholder_fd(void) |
1675 | { | |
1676 | int fd; | |
1677 | ||
9fa5e1a1 | 1678 | fd = ensure_good_fd(sys_memfd_create("libbpf-placeholder-fd", MFD_CLOEXEC)); |
dac645b9 AN |
1679 | if (fd < 0) |
1680 | return -errno; | |
1681 | return fd; | |
1682 | } | |
1683 | ||
bf829271 | 1684 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj) |
d859900c | 1685 | { |
69721203 AN |
1686 | struct bpf_map *map; |
1687 | int err; | |
bf829271 | 1688 | |
69721203 AN |
1689 | err = libbpf_ensure_mem((void **)&obj->maps, &obj->maps_cap, |
1690 | sizeof(*obj->maps), obj->nr_maps + 1); | |
1691 | if (err) | |
1692 | return ERR_PTR(err); | |
bf829271 | 1693 | |
69721203 | 1694 | map = &obj->maps[obj->nr_maps++]; |
ec41817b | 1695 | map->obj = obj; |
dac645b9 AN |
1696 | /* Preallocate map FD without actually creating BPF map just yet. |
1697 | * These map FD "placeholders" will be reused later without changing | |
1698 | * FD value when map is actually created in the kernel. | |
1699 | * | |
1700 | * This is useful to be able to perform BPF program relocations | |
1701 | * without having to create BPF maps before that step. This allows us | |
1702 | * to finalize and load BTF very late in BPF object's loading phase, | |
1703 | * right before BPF maps have to be created and BPF programs have to | |
1704 | * be loaded. By having these map FD placeholders we can perform all | |
1705 | * the sanitizations, relocations, and any other adjustments before we | |
1706 | * start creating actual BPF kernel objects (BTF, maps, progs). | |
1707 | */ | |
1708 | map->fd = create_placeholder_fd(); | |
1709 | if (map->fd < 0) | |
1710 | return ERR_PTR(map->fd); | |
69721203 | 1711 | map->inner_map_fd = -1; |
ec41817b | 1712 | map->autocreate = true; |
bf829271 | 1713 | |
69721203 | 1714 | return map; |
d859900c DB |
1715 | } |
1716 | ||
79ff13e9 | 1717 | static size_t array_map_mmap_sz(unsigned int value_sz, unsigned int max_entries) |
eba9c5f4 | 1718 | { |
9d0a2331 | 1719 | const long page_sz = sysconf(_SC_PAGE_SIZE); |
eba9c5f4 AN |
1720 | size_t map_sz; |
1721 | ||
9d0a2331 | 1722 | map_sz = (size_t)roundup(value_sz, 8) * max_entries; |
eba9c5f4 AN |
1723 | map_sz = roundup(map_sz, page_sz); |
1724 | return map_sz; | |
1725 | } | |
1726 | ||
79ff13e9 AS |
1727 | static size_t bpf_map_mmap_sz(const struct bpf_map *map) |
1728 | { | |
1729 | const long page_sz = sysconf(_SC_PAGE_SIZE); | |
1730 | ||
1731 | switch (map->def.type) { | |
1732 | case BPF_MAP_TYPE_ARRAY: | |
1733 | return array_map_mmap_sz(map->def.value_size, map->def.max_entries); | |
1734 | case BPF_MAP_TYPE_ARENA: | |
1735 | return page_sz * map->def.max_entries; | |
1736 | default: | |
1737 | return 0; /* not supported */ | |
1738 | } | |
1739 | } | |
1740 | ||
9d0a2331 JK |
1741 | static int bpf_map_mmap_resize(struct bpf_map *map, size_t old_sz, size_t new_sz) |
1742 | { | |
1743 | void *mmaped; | |
1744 | ||
1745 | if (!map->mmaped) | |
1746 | return -EINVAL; | |
1747 | ||
1748 | if (old_sz == new_sz) | |
1749 | return 0; | |
1750 | ||
1751 | mmaped = mmap(NULL, new_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); | |
1752 | if (mmaped == MAP_FAILED) | |
1753 | return -errno; | |
1754 | ||
1755 | memcpy(mmaped, map->mmaped, min(old_sz, new_sz)); | |
1756 | munmap(map->mmaped, old_sz); | |
1757 | map->mmaped = mmaped; | |
1758 | return 0; | |
1759 | } | |
1760 | ||
aed65917 | 1761 | static char *internal_map_name(struct bpf_object *obj, const char *real_name) |
81bfdd08 | 1762 | { |
113e6b7e | 1763 | char map_name[BPF_OBJ_NAME_LEN], *p; |
aed65917 AN |
1764 | int pfx_len, sfx_len = max((size_t)7, strlen(real_name)); |
1765 | ||
1766 | /* This is one of the more confusing parts of libbpf for various | |
1767 | * reasons, some of which are historical. The original idea for naming | |
1768 | * internal names was to include as much of BPF object name prefix as | |
1769 | * possible, so that it can be distinguished from similar internal | |
1770 | * maps of a different BPF object. | |
1771 | * As an example, let's say we have bpf_object named 'my_object_name' | |
1772 | * and internal map corresponding to '.rodata' ELF section. The final | |
1773 | * map name advertised to user and to the kernel will be | |
1774 | * 'my_objec.rodata', taking first 8 characters of object name and | |
1775 | * entire 7 characters of '.rodata'. | |
1776 | * Somewhat confusingly, if internal map ELF section name is shorter | |
1777 | * than 7 characters, e.g., '.bss', we still reserve 7 characters | |
1778 | * for the suffix, even though we only have 4 actual characters, and | |
1779 | * resulting map will be called 'my_objec.bss', not even using all 15 | |
1780 | * characters allowed by the kernel. Oh well, at least the truncated | |
1781 | * object name is somewhat consistent in this case. But if the map | |
1782 | * name is '.kconfig', we'll still have entirety of '.kconfig' added | |
1783 | * (8 chars) and thus will be left with only first 7 characters of the | |
1784 | * object name ('my_obje'). Happy guessing, user, that the final map | |
1785 | * name will be "my_obje.kconfig". | |
1786 | * Now, with libbpf starting to support arbitrarily named .rodata.* | |
1787 | * and .data.* data sections, it's possible that ELF section name is | |
1788 | * longer than allowed 15 chars, so we now need to be careful to take | |
1789 | * only up to 15 first characters of ELF name, taking no BPF object | |
1790 | * name characters at all. So '.rodata.abracadabra' will result in | |
1791 | * '.rodata.abracad' kernel and user-visible name. | |
1792 | * We need to keep this convoluted logic intact for .data, .bss and | |
1793 | * .rodata maps, but for new custom .data.custom and .rodata.custom | |
1794 | * maps we use their ELF names as is, not prepending bpf_object name | |
1795 | * in front. We still need to truncate them to 15 characters for the | |
1796 | * kernel. Full name can be recovered for such maps by using DATASEC | |
1797 | * BTF type associated with such map's value type, though. | |
1798 | */ | |
1799 | if (sfx_len >= BPF_OBJ_NAME_LEN) | |
1800 | sfx_len = BPF_OBJ_NAME_LEN - 1; | |
1801 | ||
1802 | /* if there are two or more dots in map name, it's a custom dot map */ | |
1803 | if (strchr(real_name + 1, '.') != NULL) | |
1804 | pfx_len = 0; | |
1805 | else | |
1806 | pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name)); | |
81bfdd08 AN |
1807 | |
1808 | snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name, | |
aed65917 | 1809 | sfx_len, real_name); |
81bfdd08 | 1810 | |
113e6b7e THJ |
1811 | /* sanitise map name to characters allowed by kernel */ |
1812 | for (p = map_name; *p && p < map_name + sizeof(map_name); p++) | |
1813 | if (!isalnum(*p) && *p != '_' && *p != '.') | |
1814 | *p = '_'; | |
1815 | ||
81bfdd08 AN |
1816 | return strdup(map_name); |
1817 | } | |
1818 | ||
262cfb74 | 1819 | static int |
4fcac46c AN |
1820 | map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map); |
1821 | ||
1822 | /* Internal BPF map is mmap()'able only if at least one of corresponding | |
1823 | * DATASEC's VARs are to be exposed through BPF skeleton. I.e., it's a GLOBAL | |
1824 | * variable and it's not marked as __hidden (which turns it into, effectively, | |
1825 | * a STATIC variable). | |
1826 | */ | |
1827 | static bool map_is_mmapable(struct bpf_object *obj, struct bpf_map *map) | |
1828 | { | |
1829 | const struct btf_type *t, *vt; | |
1830 | struct btf_var_secinfo *vsi; | |
1831 | int i, n; | |
1832 | ||
1833 | if (!map->btf_value_type_id) | |
1834 | return false; | |
1835 | ||
1836 | t = btf__type_by_id(obj->btf, map->btf_value_type_id); | |
1837 | if (!btf_is_datasec(t)) | |
1838 | return false; | |
1839 | ||
1840 | vsi = btf_var_secinfos(t); | |
1841 | for (i = 0, n = btf_vlen(t); i < n; i++, vsi++) { | |
1842 | vt = btf__type_by_id(obj->btf, vsi->type); | |
1843 | if (!btf_is_var(vt)) | |
1844 | continue; | |
1845 | ||
1846 | if (btf_var(vt)->linkage != BTF_VAR_STATIC) | |
1847 | return true; | |
1848 | } | |
1849 | ||
1850 | return false; | |
1851 | } | |
262cfb74 | 1852 | |
d859900c | 1853 | static int |
bf829271 | 1854 | bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type, |
aed65917 | 1855 | const char *real_name, int sec_idx, void *data, size_t data_sz) |
d859900c | 1856 | { |
bf829271 AN |
1857 | struct bpf_map_def *def; |
1858 | struct bpf_map *map; | |
9d0a2331 | 1859 | size_t mmap_sz; |
eba9c5f4 | 1860 | int err; |
bf829271 AN |
1861 | |
1862 | map = bpf_object__add_map(obj); | |
1863 | if (IS_ERR(map)) | |
1864 | return PTR_ERR(map); | |
d859900c DB |
1865 | |
1866 | map->libbpf_type = type; | |
db48814b AN |
1867 | map->sec_idx = sec_idx; |
1868 | map->sec_offset = 0; | |
aed65917 AN |
1869 | map->real_name = strdup(real_name); |
1870 | map->name = internal_map_name(obj, real_name); | |
1871 | if (!map->real_name || !map->name) { | |
1872 | zfree(&map->real_name); | |
1873 | zfree(&map->name); | |
d859900c DB |
1874 | return -ENOMEM; |
1875 | } | |
1876 | ||
bf829271 | 1877 | def = &map->def; |
d859900c DB |
1878 | def->type = BPF_MAP_TYPE_ARRAY; |
1879 | def->key_size = sizeof(int); | |
eba9c5f4 | 1880 | def->value_size = data_sz; |
d859900c | 1881 | def->max_entries = 1; |
81bfdd08 | 1882 | def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG |
2e7ba4f8 | 1883 | ? BPF_F_RDONLY_PROG : 0; |
4fcac46c AN |
1884 | |
1885 | /* failures are fine because of maps like .rodata.str1.1 */ | |
1886 | (void) map_fill_btf_type_info(obj, map); | |
1887 | ||
1888 | if (map_is_mmapable(obj, map)) | |
1889 | def->map_flags |= BPF_F_MMAPABLE; | |
7fe74b43 AN |
1890 | |
1891 | pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n", | |
81bfdd08 | 1892 | map->name, map->sec_idx, map->sec_offset, def->map_flags); |
7fe74b43 | 1893 | |
79ff13e9 | 1894 | mmap_sz = bpf_map_mmap_sz(map); |
9d0a2331 | 1895 | map->mmaped = mmap(NULL, mmap_sz, PROT_READ | PROT_WRITE, |
eba9c5f4 AN |
1896 | MAP_SHARED | MAP_ANONYMOUS, -1, 0); |
1897 | if (map->mmaped == MAP_FAILED) { | |
1898 | err = -errno; | |
1899 | map->mmaped = NULL; | |
1900 | pr_warn("failed to alloc map '%s' content buffer: %d\n", | |
1901 | map->name, err); | |
aed65917 | 1902 | zfree(&map->real_name); |
eba9c5f4 AN |
1903 | zfree(&map->name); |
1904 | return err; | |
d859900c DB |
1905 | } |
1906 | ||
166750bc | 1907 | if (data) |
eba9c5f4 AN |
1908 | memcpy(map->mmaped, data, data_sz); |
1909 | ||
e1d1dc46 | 1910 | pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name); |
d859900c DB |
1911 | return 0; |
1912 | } | |
1913 | ||
bf829271 AN |
1914 | static int bpf_object__init_global_data_maps(struct bpf_object *obj) |
1915 | { | |
25bbbd7a | 1916 | struct elf_sec_desc *sec_desc; |
aed65917 | 1917 | const char *sec_name; |
25bbbd7a | 1918 | int err = 0, sec_idx; |
bf829271 | 1919 | |
bf829271 AN |
1920 | /* |
1921 | * Populate obj->maps with libbpf internal maps. | |
1922 | */ | |
25bbbd7a AN |
1923 | for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) { |
1924 | sec_desc = &obj->efile.secs[sec_idx]; | |
1925 | ||
47ea7417 | 1926 | /* Skip recognized sections with size 0. */ |
d4e6d684 | 1927 | if (!sec_desc->data || sec_desc->data->d_size == 0) |
47ea7417 JH |
1928 | continue; |
1929 | ||
25bbbd7a AN |
1930 | switch (sec_desc->sec_type) { |
1931 | case SEC_DATA: | |
aed65917 | 1932 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1933 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA, |
aed65917 | 1934 | sec_name, sec_idx, |
25bbbd7a AN |
1935 | sec_desc->data->d_buf, |
1936 | sec_desc->data->d_size); | |
1937 | break; | |
1938 | case SEC_RODATA: | |
1939 | obj->has_rodata = true; | |
aed65917 | 1940 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1941 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA, |
aed65917 | 1942 | sec_name, sec_idx, |
25bbbd7a AN |
1943 | sec_desc->data->d_buf, |
1944 | sec_desc->data->d_size); | |
1945 | break; | |
1946 | case SEC_BSS: | |
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_BSS, |
aed65917 | 1949 | sec_name, sec_idx, |
25bbbd7a AN |
1950 | NULL, |
1951 | sec_desc->data->d_size); | |
1952 | break; | |
1953 | default: | |
1954 | /* skip */ | |
1955 | break; | |
1956 | } | |
bf829271 AN |
1957 | if (err) |
1958 | return err; | |
1959 | } | |
1960 | return 0; | |
1961 | } | |
1962 | ||
166750bc AN |
1963 | |
1964 | static struct extern_desc *find_extern_by_name(const struct bpf_object *obj, | |
1965 | const void *name) | |
1966 | { | |
1967 | int i; | |
1968 | ||
1969 | for (i = 0; i < obj->nr_extern; i++) { | |
1970 | if (strcmp(obj->externs[i].name, name) == 0) | |
1971 | return &obj->externs[i]; | |
1972 | } | |
1973 | return NULL; | |
1974 | } | |
1975 | ||
c56e5977 YS |
1976 | static struct extern_desc *find_extern_by_name_with_len(const struct bpf_object *obj, |
1977 | const void *name, int len) | |
1978 | { | |
1979 | const char *ext_name; | |
1980 | int i; | |
1981 | ||
1982 | for (i = 0; i < obj->nr_extern; i++) { | |
1983 | ext_name = obj->externs[i].name; | |
1984 | if (strlen(ext_name) == len && strncmp(ext_name, name, len) == 0) | |
1985 | return &obj->externs[i]; | |
1986 | } | |
1987 | return NULL; | |
1988 | } | |
1989 | ||
2e33efe3 AN |
1990 | static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val, |
1991 | char value) | |
166750bc | 1992 | { |
2e33efe3 AN |
1993 | switch (ext->kcfg.type) { |
1994 | case KCFG_BOOL: | |
166750bc | 1995 | if (value == 'm') { |
55d00c37 | 1996 | pr_warn("extern (kcfg) '%s': value '%c' implies tristate or char type\n", |
166750bc AN |
1997 | ext->name, value); |
1998 | return -EINVAL; | |
1999 | } | |
2000 | *(bool *)ext_val = value == 'y' ? true : false; | |
2001 | break; | |
2e33efe3 | 2002 | case KCFG_TRISTATE: |
166750bc AN |
2003 | if (value == 'y') |
2004 | *(enum libbpf_tristate *)ext_val = TRI_YES; | |
2005 | else if (value == 'm') | |
2006 | *(enum libbpf_tristate *)ext_val = TRI_MODULE; | |
2007 | else /* value == 'n' */ | |
2008 | *(enum libbpf_tristate *)ext_val = TRI_NO; | |
2009 | break; | |
2e33efe3 | 2010 | case KCFG_CHAR: |
166750bc AN |
2011 | *(char *)ext_val = value; |
2012 | break; | |
2e33efe3 AN |
2013 | case KCFG_UNKNOWN: |
2014 | case KCFG_INT: | |
2015 | case KCFG_CHAR_ARR: | |
166750bc | 2016 | default: |
55d00c37 | 2017 | pr_warn("extern (kcfg) '%s': value '%c' implies bool, tristate, or char type\n", |
166750bc AN |
2018 | ext->name, value); |
2019 | return -EINVAL; | |
2020 | } | |
2021 | ext->is_set = true; | |
2022 | return 0; | |
2023 | } | |
2024 | ||
2e33efe3 AN |
2025 | static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val, |
2026 | const char *value) | |
166750bc AN |
2027 | { |
2028 | size_t len; | |
2029 | ||
2e33efe3 | 2030 | if (ext->kcfg.type != KCFG_CHAR_ARR) { |
55d00c37 AN |
2031 | pr_warn("extern (kcfg) '%s': value '%s' implies char array type\n", |
2032 | ext->name, value); | |
166750bc AN |
2033 | return -EINVAL; |
2034 | } | |
2035 | ||
2036 | len = strlen(value); | |
2037 | if (value[len - 1] != '"') { | |
2e33efe3 | 2038 | pr_warn("extern (kcfg) '%s': invalid string config '%s'\n", |
166750bc AN |
2039 | ext->name, value); |
2040 | return -EINVAL; | |
2041 | } | |
2042 | ||
2043 | /* strip quotes */ | |
2044 | len -= 2; | |
2e33efe3 | 2045 | if (len >= ext->kcfg.sz) { |
55d00c37 | 2046 | pr_warn("extern (kcfg) '%s': long string '%s' of (%zu bytes) truncated to %d bytes\n", |
2e33efe3 AN |
2047 | ext->name, value, len, ext->kcfg.sz - 1); |
2048 | len = ext->kcfg.sz - 1; | |
166750bc AN |
2049 | } |
2050 | memcpy(ext_val, value + 1, len); | |
2051 | ext_val[len] = '\0'; | |
2052 | ext->is_set = true; | |
2053 | return 0; | |
2054 | } | |
2055 | ||
2056 | static int parse_u64(const char *value, __u64 *res) | |
2057 | { | |
2058 | char *value_end; | |
2059 | int err; | |
2060 | ||
2061 | errno = 0; | |
2062 | *res = strtoull(value, &value_end, 0); | |
2063 | if (errno) { | |
2064 | err = -errno; | |
2065 | pr_warn("failed to parse '%s' as integer: %d\n", value, err); | |
2066 | return err; | |
2067 | } | |
2068 | if (*value_end) { | |
2069 | pr_warn("failed to parse '%s' as integer completely\n", value); | |
2070 | return -EINVAL; | |
2071 | } | |
2072 | return 0; | |
2073 | } | |
2074 | ||
2e33efe3 | 2075 | static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v) |
166750bc | 2076 | { |
2e33efe3 | 2077 | int bit_sz = ext->kcfg.sz * 8; |
166750bc | 2078 | |
2e33efe3 | 2079 | if (ext->kcfg.sz == 8) |
166750bc AN |
2080 | return true; |
2081 | ||
2082 | /* Validate that value stored in u64 fits in integer of `ext->sz` | |
2083 | * bytes size without any loss of information. If the target integer | |
2084 | * is signed, we rely on the following limits of integer type of | |
2085 | * Y bits and subsequent transformation: | |
2086 | * | |
2087 | * -2^(Y-1) <= X <= 2^(Y-1) - 1 | |
2088 | * 0 <= X + 2^(Y-1) <= 2^Y - 1 | |
2089 | * 0 <= X + 2^(Y-1) < 2^Y | |
2090 | * | |
2091 | * For unsigned target integer, check that all the (64 - Y) bits are | |
2092 | * zero. | |
2093 | */ | |
2e33efe3 | 2094 | if (ext->kcfg.is_signed) |
166750bc AN |
2095 | return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz); |
2096 | else | |
2097 | return (v >> bit_sz) == 0; | |
2098 | } | |
2099 | ||
2e33efe3 AN |
2100 | static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val, |
2101 | __u64 value) | |
166750bc | 2102 | { |
55d00c37 AN |
2103 | if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR && |
2104 | ext->kcfg.type != KCFG_BOOL) { | |
2105 | pr_warn("extern (kcfg) '%s': value '%llu' implies integer, char, or boolean type\n", | |
7745ff98 | 2106 | ext->name, (unsigned long long)value); |
166750bc AN |
2107 | return -EINVAL; |
2108 | } | |
55d00c37 AN |
2109 | if (ext->kcfg.type == KCFG_BOOL && value > 1) { |
2110 | pr_warn("extern (kcfg) '%s': value '%llu' isn't boolean compatible\n", | |
2111 | ext->name, (unsigned long long)value); | |
2112 | return -EINVAL; | |
2113 | ||
2114 | } | |
2e33efe3 | 2115 | if (!is_kcfg_value_in_range(ext, value)) { |
55d00c37 | 2116 | pr_warn("extern (kcfg) '%s': value '%llu' doesn't fit in %d bytes\n", |
2e33efe3 | 2117 | ext->name, (unsigned long long)value, ext->kcfg.sz); |
166750bc AN |
2118 | return -ERANGE; |
2119 | } | |
2e33efe3 | 2120 | switch (ext->kcfg.sz) { |
e3ba8e4e KM |
2121 | case 1: |
2122 | *(__u8 *)ext_val = value; | |
2123 | break; | |
2124 | case 2: | |
2125 | *(__u16 *)ext_val = value; | |
2126 | break; | |
2127 | case 4: | |
2128 | *(__u32 *)ext_val = value; | |
2129 | break; | |
2130 | case 8: | |
2131 | *(__u64 *)ext_val = value; | |
2132 | break; | |
2133 | default: | |
2134 | return -EINVAL; | |
166750bc AN |
2135 | } |
2136 | ext->is_set = true; | |
2137 | return 0; | |
2138 | } | |
2139 | ||
8601fd42 AN |
2140 | static int bpf_object__process_kconfig_line(struct bpf_object *obj, |
2141 | char *buf, void *data) | |
166750bc | 2142 | { |
166750bc | 2143 | struct extern_desc *ext; |
8601fd42 | 2144 | char *sep, *value; |
166750bc AN |
2145 | int len, err = 0; |
2146 | void *ext_val; | |
2147 | __u64 num; | |
166750bc | 2148 | |
13d35a0c | 2149 | if (!str_has_pfx(buf, "CONFIG_")) |
8601fd42 | 2150 | return 0; |
166750bc | 2151 | |
8601fd42 AN |
2152 | sep = strchr(buf, '='); |
2153 | if (!sep) { | |
2154 | pr_warn("failed to parse '%s': no separator\n", buf); | |
2155 | return -EINVAL; | |
2156 | } | |
2157 | ||
2158 | /* Trim ending '\n' */ | |
2159 | len = strlen(buf); | |
2160 | if (buf[len - 1] == '\n') | |
2161 | buf[len - 1] = '\0'; | |
2162 | /* Split on '=' and ensure that a value is present. */ | |
2163 | *sep = '\0'; | |
2164 | if (!sep[1]) { | |
2165 | *sep = '='; | |
2166 | pr_warn("failed to parse '%s': no value\n", buf); | |
2167 | return -EINVAL; | |
2168 | } | |
2169 | ||
2170 | ext = find_extern_by_name(obj, buf); | |
2171 | if (!ext || ext->is_set) | |
2172 | return 0; | |
2173 | ||
2e33efe3 | 2174 | ext_val = data + ext->kcfg.data_off; |
8601fd42 AN |
2175 | value = sep + 1; |
2176 | ||
2177 | switch (*value) { | |
2178 | case 'y': case 'n': case 'm': | |
2e33efe3 | 2179 | err = set_kcfg_value_tri(ext, ext_val, *value); |
8601fd42 AN |
2180 | break; |
2181 | case '"': | |
2e33efe3 | 2182 | err = set_kcfg_value_str(ext, ext_val, value); |
8601fd42 AN |
2183 | break; |
2184 | default: | |
2185 | /* assume integer */ | |
2186 | err = parse_u64(value, &num); | |
2187 | if (err) { | |
55d00c37 | 2188 | pr_warn("extern (kcfg) '%s': value '%s' isn't a valid integer\n", ext->name, value); |
8601fd42 AN |
2189 | return err; |
2190 | } | |
55d00c37 AN |
2191 | if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) { |
2192 | pr_warn("extern (kcfg) '%s': value '%s' implies integer type\n", ext->name, value); | |
2193 | return -EINVAL; | |
2194 | } | |
2e33efe3 | 2195 | err = set_kcfg_value_num(ext, ext_val, num); |
8601fd42 | 2196 | break; |
166750bc | 2197 | } |
8601fd42 AN |
2198 | if (err) |
2199 | return err; | |
55d00c37 | 2200 | pr_debug("extern (kcfg) '%s': set to %s\n", ext->name, value); |
8601fd42 AN |
2201 | return 0; |
2202 | } | |
2203 | ||
2204 | static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data) | |
2205 | { | |
2206 | char buf[PATH_MAX]; | |
2207 | struct utsname uts; | |
2208 | int len, err = 0; | |
2209 | gzFile file; | |
2210 | ||
2211 | uname(&uts); | |
2212 | len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release); | |
2213 | if (len < 0) | |
2214 | return -EINVAL; | |
2215 | else if (len >= PATH_MAX) | |
2216 | return -ENAMETOOLONG; | |
2217 | ||
2218 | /* gzopen also accepts uncompressed files. */ | |
8e50750f | 2219 | file = gzopen(buf, "re"); |
8601fd42 | 2220 | if (!file) |
8e50750f | 2221 | file = gzopen("/proc/config.gz", "re"); |
8601fd42 | 2222 | |
166750bc | 2223 | if (!file) { |
8601fd42 | 2224 | pr_warn("failed to open system Kconfig\n"); |
166750bc AN |
2225 | return -ENOENT; |
2226 | } | |
2227 | ||
2228 | while (gzgets(file, buf, sizeof(buf))) { | |
8601fd42 AN |
2229 | err = bpf_object__process_kconfig_line(obj, buf, data); |
2230 | if (err) { | |
2231 | pr_warn("error parsing system Kconfig line '%s': %d\n", | |
2232 | buf, err); | |
166750bc AN |
2233 | goto out; |
2234 | } | |
8601fd42 | 2235 | } |
166750bc | 2236 | |
8601fd42 AN |
2237 | out: |
2238 | gzclose(file); | |
2239 | return err; | |
2240 | } | |
166750bc | 2241 | |
8601fd42 AN |
2242 | static int bpf_object__read_kconfig_mem(struct bpf_object *obj, |
2243 | const char *config, void *data) | |
2244 | { | |
2245 | char buf[PATH_MAX]; | |
2246 | int err = 0; | |
2247 | FILE *file; | |
166750bc | 2248 | |
8601fd42 AN |
2249 | file = fmemopen((void *)config, strlen(config), "r"); |
2250 | if (!file) { | |
2251 | err = -errno; | |
2252 | pr_warn("failed to open in-memory Kconfig: %d\n", err); | |
2253 | return err; | |
2254 | } | |
2255 | ||
2256 | while (fgets(buf, sizeof(buf), file)) { | |
2257 | err = bpf_object__process_kconfig_line(obj, buf, data); | |
2258 | if (err) { | |
2259 | pr_warn("error parsing in-memory Kconfig line '%s': %d\n", | |
2260 | buf, err); | |
166750bc AN |
2261 | break; |
2262 | } | |
166750bc AN |
2263 | } |
2264 | ||
8601fd42 | 2265 | fclose(file); |
166750bc AN |
2266 | return err; |
2267 | } | |
2268 | ||
81bfdd08 | 2269 | static int bpf_object__init_kconfig_map(struct bpf_object *obj) |
166750bc | 2270 | { |
2e33efe3 | 2271 | struct extern_desc *last_ext = NULL, *ext; |
166750bc | 2272 | size_t map_sz; |
2e33efe3 | 2273 | int i, err; |
166750bc | 2274 | |
2e33efe3 AN |
2275 | for (i = 0; i < obj->nr_extern; i++) { |
2276 | ext = &obj->externs[i]; | |
2277 | if (ext->type == EXT_KCFG) | |
2278 | last_ext = ext; | |
2279 | } | |
166750bc | 2280 | |
2e33efe3 AN |
2281 | if (!last_ext) |
2282 | return 0; | |
166750bc | 2283 | |
2e33efe3 | 2284 | map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz; |
81bfdd08 | 2285 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG, |
aed65917 | 2286 | ".kconfig", obj->efile.symbols_shndx, |
166750bc AN |
2287 | NULL, map_sz); |
2288 | if (err) | |
2289 | return err; | |
2290 | ||
81bfdd08 | 2291 | obj->kconfig_map_idx = obj->nr_maps - 1; |
166750bc AN |
2292 | |
2293 | return 0; | |
2294 | } | |
2295 | ||
42869d28 | 2296 | const struct btf_type * |
ddc7c304 | 2297 | skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id) |
abd29c93 AN |
2298 | { |
2299 | const struct btf_type *t = btf__type_by_id(btf, id); | |
8837fe5d | 2300 | |
ddc7c304 AN |
2301 | if (res_id) |
2302 | *res_id = id; | |
2303 | ||
2304 | while (btf_is_mod(t) || btf_is_typedef(t)) { | |
2305 | if (res_id) | |
2306 | *res_id = t->type; | |
2307 | t = btf__type_by_id(btf, t->type); | |
abd29c93 | 2308 | } |
ddc7c304 AN |
2309 | |
2310 | return t; | |
abd29c93 AN |
2311 | } |
2312 | ||
590a0088 MKL |
2313 | static const struct btf_type * |
2314 | resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id) | |
2315 | { | |
2316 | const struct btf_type *t; | |
2317 | ||
2318 | t = skip_mods_and_typedefs(btf, id, NULL); | |
2319 | if (!btf_is_ptr(t)) | |
2320 | return NULL; | |
2321 | ||
2322 | t = skip_mods_and_typedefs(btf, t->type, res_id); | |
2323 | ||
2324 | return btf_is_func_proto(t) ? t : NULL; | |
2325 | } | |
2326 | ||
774e132e | 2327 | static const char *__btf_kind_str(__u16 kind) |
81ba0889 | 2328 | { |
774e132e | 2329 | switch (kind) { |
81ba0889 AN |
2330 | case BTF_KIND_UNKN: return "void"; |
2331 | case BTF_KIND_INT: return "int"; | |
2332 | case BTF_KIND_PTR: return "ptr"; | |
2333 | case BTF_KIND_ARRAY: return "array"; | |
2334 | case BTF_KIND_STRUCT: return "struct"; | |
2335 | case BTF_KIND_UNION: return "union"; | |
2336 | case BTF_KIND_ENUM: return "enum"; | |
2337 | case BTF_KIND_FWD: return "fwd"; | |
2338 | case BTF_KIND_TYPEDEF: return "typedef"; | |
2339 | case BTF_KIND_VOLATILE: return "volatile"; | |
2340 | case BTF_KIND_CONST: return "const"; | |
2341 | case BTF_KIND_RESTRICT: return "restrict"; | |
2342 | case BTF_KIND_FUNC: return "func"; | |
2343 | case BTF_KIND_FUNC_PROTO: return "func_proto"; | |
2344 | case BTF_KIND_VAR: return "var"; | |
2345 | case BTF_KIND_DATASEC: return "datasec"; | |
22541a9e | 2346 | case BTF_KIND_FLOAT: return "float"; |
223f903e | 2347 | case BTF_KIND_DECL_TAG: return "decl_tag"; |
2dc1e488 | 2348 | case BTF_KIND_TYPE_TAG: return "type_tag"; |
f2a62588 | 2349 | case BTF_KIND_ENUM64: return "enum64"; |
81ba0889 AN |
2350 | default: return "unknown"; |
2351 | } | |
2352 | } | |
2353 | ||
42869d28 | 2354 | const char *btf_kind_str(const struct btf_type *t) |
774e132e MKL |
2355 | { |
2356 | return __btf_kind_str(btf_kind(t)); | |
2357 | } | |
2358 | ||
ef99b02b AN |
2359 | /* |
2360 | * Fetch integer attribute of BTF map definition. Such attributes are | |
2361 | * represented using a pointer to an array, in which dimensionality of array | |
2362 | * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY]; | |
2363 | * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF | |
2364 | * type definition, while using only sizeof(void *) space in ELF data section. | |
2365 | */ | |
2366 | static bool get_map_field_int(const char *map_name, const struct btf *btf, | |
8983b731 AN |
2367 | const struct btf_member *m, __u32 *res) |
2368 | { | |
ddc7c304 | 2369 | const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL); |
abd29c93 | 2370 | const char *name = btf__name_by_offset(btf, m->name_off); |
ef99b02b AN |
2371 | const struct btf_array *arr_info; |
2372 | const struct btf_type *arr_t; | |
abd29c93 | 2373 | |
b03bc685 | 2374 | if (!btf_is_ptr(t)) { |
81ba0889 AN |
2375 | pr_warn("map '%s': attr '%s': expected PTR, got %s.\n", |
2376 | map_name, name, btf_kind_str(t)); | |
abd29c93 AN |
2377 | return false; |
2378 | } | |
ef99b02b AN |
2379 | |
2380 | arr_t = btf__type_by_id(btf, t->type); | |
2381 | if (!arr_t) { | |
be18010e KW |
2382 | pr_warn("map '%s': attr '%s': type [%u] not found.\n", |
2383 | map_name, name, t->type); | |
abd29c93 AN |
2384 | return false; |
2385 | } | |
b03bc685 | 2386 | if (!btf_is_array(arr_t)) { |
81ba0889 AN |
2387 | pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n", |
2388 | map_name, name, btf_kind_str(arr_t)); | |
abd29c93 AN |
2389 | return false; |
2390 | } | |
b03bc685 | 2391 | arr_info = btf_array(arr_t); |
ef99b02b | 2392 | *res = arr_info->nelems; |
abd29c93 AN |
2393 | return true; |
2394 | } | |
2395 | ||
d147357e AS |
2396 | static bool get_map_field_long(const char *map_name, const struct btf *btf, |
2397 | const struct btf_member *m, __u64 *res) | |
2398 | { | |
2399 | const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL); | |
2400 | const char *name = btf__name_by_offset(btf, m->name_off); | |
2401 | ||
2402 | if (btf_is_ptr(t)) { | |
2403 | __u32 res32; | |
2404 | bool ret; | |
2405 | ||
2406 | ret = get_map_field_int(map_name, btf, m, &res32); | |
2407 | if (ret) | |
2408 | *res = (__u64)res32; | |
2409 | return ret; | |
2410 | } | |
2411 | ||
2412 | if (!btf_is_enum(t) && !btf_is_enum64(t)) { | |
2413 | pr_warn("map '%s': attr '%s': expected ENUM or ENUM64, got %s.\n", | |
2414 | map_name, name, btf_kind_str(t)); | |
2415 | return false; | |
2416 | } | |
2417 | ||
2418 | if (btf_vlen(t) != 1) { | |
2419 | pr_warn("map '%s': attr '%s': invalid __ulong\n", | |
2420 | map_name, name); | |
2421 | return false; | |
2422 | } | |
2423 | ||
2424 | if (btf_is_enum(t)) { | |
2425 | const struct btf_enum *e = btf_enum(t); | |
2426 | ||
2427 | *res = e->val; | |
2428 | } else { | |
2429 | const struct btf_enum64 *e = btf_enum64(t); | |
2430 | ||
2431 | *res = btf_enum64_value(e); | |
2432 | } | |
2433 | return true; | |
2434 | } | |
2435 | ||
e588c116 WY |
2436 | static int pathname_concat(char *buf, size_t buf_sz, const char *path, const char *name) |
2437 | { | |
2438 | int len; | |
2439 | ||
2440 | len = snprintf(buf, buf_sz, "%s/%s", path, name); | |
2441 | if (len < 0) | |
2442 | return -EINVAL; | |
2443 | if (len >= buf_sz) | |
2444 | return -ENAMETOOLONG; | |
2445 | ||
2446 | return 0; | |
2447 | } | |
2448 | ||
57a00f41 THJ |
2449 | static int build_map_pin_path(struct bpf_map *map, const char *path) |
2450 | { | |
2451 | char buf[PATH_MAX]; | |
e588c116 | 2452 | int err; |
57a00f41 THJ |
2453 | |
2454 | if (!path) | |
6b434b61 | 2455 | path = BPF_FS_DEFAULT_PATH; |
57a00f41 | 2456 | |
e588c116 WY |
2457 | err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map)); |
2458 | if (err) | |
2459 | return err; | |
57a00f41 | 2460 | |
6e9cab2e | 2461 | return bpf_map__set_pin_path(map, buf); |
57a00f41 THJ |
2462 | } |
2463 | ||
146bf811 AN |
2464 | /* should match definition in bpf_helpers.h */ |
2465 | enum libbpf_pin_type { | |
2466 | LIBBPF_PIN_NONE, | |
2467 | /* PIN_BY_NAME: pin maps by name (in /sys/fs/bpf by default) */ | |
2468 | LIBBPF_PIN_BY_NAME, | |
2469 | }; | |
2470 | ||
c7ef5ec9 AN |
2471 | int parse_btf_map_def(const char *map_name, struct btf *btf, |
2472 | const struct btf_type *def_t, bool strict, | |
2473 | struct btf_map_def *map_def, struct btf_map_def *inner_def) | |
abd29c93 | 2474 | { |
41017e56 | 2475 | const struct btf_type *t; |
abd29c93 | 2476 | const struct btf_member *m; |
c7ef5ec9 | 2477 | bool is_inner = inner_def == NULL; |
abd29c93 AN |
2478 | int vlen, i; |
2479 | ||
c7ef5ec9 AN |
2480 | vlen = btf_vlen(def_t); |
2481 | m = btf_members(def_t); | |
abd29c93 | 2482 | for (i = 0; i < vlen; i++, m++) { |
c7ef5ec9 | 2483 | const char *name = btf__name_by_offset(btf, m->name_off); |
abd29c93 AN |
2484 | |
2485 | if (!name) { | |
c7ef5ec9 | 2486 | pr_warn("map '%s': invalid field #%d.\n", map_name, i); |
abd29c93 AN |
2487 | return -EINVAL; |
2488 | } | |
2489 | if (strcmp(name, "type") == 0) { | |
c7ef5ec9 | 2490 | if (!get_map_field_int(map_name, btf, m, &map_def->map_type)) |
abd29c93 | 2491 | return -EINVAL; |
c7ef5ec9 | 2492 | map_def->parts |= MAP_DEF_MAP_TYPE; |
abd29c93 | 2493 | } else if (strcmp(name, "max_entries") == 0) { |
c7ef5ec9 | 2494 | if (!get_map_field_int(map_name, btf, m, &map_def->max_entries)) |
abd29c93 | 2495 | return -EINVAL; |
c7ef5ec9 | 2496 | map_def->parts |= MAP_DEF_MAX_ENTRIES; |
abd29c93 | 2497 | } else if (strcmp(name, "map_flags") == 0) { |
c7ef5ec9 | 2498 | if (!get_map_field_int(map_name, btf, m, &map_def->map_flags)) |
abd29c93 | 2499 | return -EINVAL; |
c7ef5ec9 | 2500 | map_def->parts |= MAP_DEF_MAP_FLAGS; |
1bdb6c9a | 2501 | } else if (strcmp(name, "numa_node") == 0) { |
c7ef5ec9 | 2502 | if (!get_map_field_int(map_name, btf, m, &map_def->numa_node)) |
1bdb6c9a | 2503 | return -EINVAL; |
c7ef5ec9 | 2504 | map_def->parts |= MAP_DEF_NUMA_NODE; |
abd29c93 AN |
2505 | } else if (strcmp(name, "key_size") == 0) { |
2506 | __u32 sz; | |
2507 | ||
c7ef5ec9 | 2508 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2509 | return -EINVAL; |
c7ef5ec9 | 2510 | if (map_def->key_size && map_def->key_size != sz) { |
be18010e | 2511 | pr_warn("map '%s': conflicting key size %u != %u.\n", |
c7ef5ec9 | 2512 | map_name, map_def->key_size, sz); |
abd29c93 AN |
2513 | return -EINVAL; |
2514 | } | |
c7ef5ec9 AN |
2515 | map_def->key_size = sz; |
2516 | map_def->parts |= MAP_DEF_KEY_SIZE; | |
abd29c93 AN |
2517 | } else if (strcmp(name, "key") == 0) { |
2518 | __s64 sz; | |
2519 | ||
c7ef5ec9 | 2520 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2521 | if (!t) { |
be18010e | 2522 | pr_warn("map '%s': key type [%d] not found.\n", |
c7ef5ec9 | 2523 | map_name, m->type); |
abd29c93 AN |
2524 | return -EINVAL; |
2525 | } | |
b03bc685 | 2526 | if (!btf_is_ptr(t)) { |
81ba0889 | 2527 | pr_warn("map '%s': key spec is not PTR: %s.\n", |
c7ef5ec9 | 2528 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2529 | return -EINVAL; |
2530 | } | |
c7ef5ec9 | 2531 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2532 | if (sz < 0) { |
679152d3 | 2533 | pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n", |
c7ef5ec9 | 2534 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2535 | return sz; |
2536 | } | |
c7ef5ec9 | 2537 | if (map_def->key_size && map_def->key_size != sz) { |
679152d3 | 2538 | pr_warn("map '%s': conflicting key size %u != %zd.\n", |
c7ef5ec9 | 2539 | map_name, map_def->key_size, (ssize_t)sz); |
abd29c93 AN |
2540 | return -EINVAL; |
2541 | } | |
c7ef5ec9 AN |
2542 | map_def->key_size = sz; |
2543 | map_def->key_type_id = t->type; | |
2544 | map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE; | |
abd29c93 AN |
2545 | } else if (strcmp(name, "value_size") == 0) { |
2546 | __u32 sz; | |
2547 | ||
c7ef5ec9 | 2548 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2549 | return -EINVAL; |
c7ef5ec9 | 2550 | if (map_def->value_size && map_def->value_size != sz) { |
be18010e | 2551 | pr_warn("map '%s': conflicting value size %u != %u.\n", |
c7ef5ec9 | 2552 | map_name, map_def->value_size, sz); |
abd29c93 AN |
2553 | return -EINVAL; |
2554 | } | |
c7ef5ec9 AN |
2555 | map_def->value_size = sz; |
2556 | map_def->parts |= MAP_DEF_VALUE_SIZE; | |
abd29c93 AN |
2557 | } else if (strcmp(name, "value") == 0) { |
2558 | __s64 sz; | |
2559 | ||
c7ef5ec9 | 2560 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2561 | if (!t) { |
be18010e | 2562 | pr_warn("map '%s': value type [%d] not found.\n", |
c7ef5ec9 | 2563 | map_name, m->type); |
abd29c93 AN |
2564 | return -EINVAL; |
2565 | } | |
b03bc685 | 2566 | if (!btf_is_ptr(t)) { |
81ba0889 | 2567 | pr_warn("map '%s': value spec is not PTR: %s.\n", |
c7ef5ec9 | 2568 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2569 | return -EINVAL; |
2570 | } | |
c7ef5ec9 | 2571 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2572 | if (sz < 0) { |
679152d3 | 2573 | pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n", |
c7ef5ec9 | 2574 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2575 | return sz; |
2576 | } | |
c7ef5ec9 | 2577 | if (map_def->value_size && map_def->value_size != sz) { |
679152d3 | 2578 | pr_warn("map '%s': conflicting value size %u != %zd.\n", |
c7ef5ec9 | 2579 | map_name, map_def->value_size, (ssize_t)sz); |
abd29c93 AN |
2580 | return -EINVAL; |
2581 | } | |
c7ef5ec9 AN |
2582 | map_def->value_size = sz; |
2583 | map_def->value_type_id = t->type; | |
2584 | map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE; | |
646f02ff AN |
2585 | } |
2586 | else if (strcmp(name, "values") == 0) { | |
341ac5ff HC |
2587 | bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type); |
2588 | bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY; | |
2589 | const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value"; | |
c7ef5ec9 | 2590 | char inner_map_name[128]; |
646f02ff AN |
2591 | int err; |
2592 | ||
2593 | if (is_inner) { | |
2594 | pr_warn("map '%s': multi-level inner maps not supported.\n", | |
c7ef5ec9 | 2595 | map_name); |
646f02ff AN |
2596 | return -ENOTSUP; |
2597 | } | |
2598 | if (i != vlen - 1) { | |
2599 | pr_warn("map '%s': '%s' member should be last.\n", | |
c7ef5ec9 | 2600 | map_name, name); |
646f02ff AN |
2601 | return -EINVAL; |
2602 | } | |
341ac5ff HC |
2603 | if (!is_map_in_map && !is_prog_array) { |
2604 | pr_warn("map '%s': should be map-in-map or prog-array.\n", | |
c7ef5ec9 | 2605 | map_name); |
646f02ff AN |
2606 | return -ENOTSUP; |
2607 | } | |
c7ef5ec9 | 2608 | if (map_def->value_size && map_def->value_size != 4) { |
646f02ff | 2609 | pr_warn("map '%s': conflicting value size %u != 4.\n", |
c7ef5ec9 | 2610 | map_name, map_def->value_size); |
646f02ff AN |
2611 | return -EINVAL; |
2612 | } | |
c7ef5ec9 AN |
2613 | map_def->value_size = 4; |
2614 | t = btf__type_by_id(btf, m->type); | |
646f02ff | 2615 | if (!t) { |
341ac5ff HC |
2616 | pr_warn("map '%s': %s type [%d] not found.\n", |
2617 | map_name, desc, m->type); | |
646f02ff AN |
2618 | return -EINVAL; |
2619 | } | |
2620 | if (!btf_is_array(t) || btf_array(t)->nelems) { | |
341ac5ff HC |
2621 | pr_warn("map '%s': %s spec is not a zero-sized array.\n", |
2622 | map_name, desc); | |
646f02ff AN |
2623 | return -EINVAL; |
2624 | } | |
c7ef5ec9 | 2625 | t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL); |
646f02ff | 2626 | if (!btf_is_ptr(t)) { |
341ac5ff HC |
2627 | pr_warn("map '%s': %s def is of unexpected kind %s.\n", |
2628 | map_name, desc, btf_kind_str(t)); | |
646f02ff AN |
2629 | return -EINVAL; |
2630 | } | |
c7ef5ec9 | 2631 | t = skip_mods_and_typedefs(btf, t->type, NULL); |
341ac5ff HC |
2632 | if (is_prog_array) { |
2633 | if (!btf_is_func_proto(t)) { | |
2634 | pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n", | |
2635 | map_name, btf_kind_str(t)); | |
2636 | return -EINVAL; | |
2637 | } | |
2638 | continue; | |
2639 | } | |
646f02ff | 2640 | if (!btf_is_struct(t)) { |
81ba0889 | 2641 | pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n", |
c7ef5ec9 | 2642 | map_name, btf_kind_str(t)); |
646f02ff AN |
2643 | return -EINVAL; |
2644 | } | |
2645 | ||
c7ef5ec9 AN |
2646 | snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name); |
2647 | err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL); | |
646f02ff AN |
2648 | if (err) |
2649 | return err; | |
c7ef5ec9 AN |
2650 | |
2651 | map_def->parts |= MAP_DEF_INNER_MAP; | |
57a00f41 THJ |
2652 | } else if (strcmp(name, "pinning") == 0) { |
2653 | __u32 val; | |
57a00f41 | 2654 | |
646f02ff | 2655 | if (is_inner) { |
c7ef5ec9 | 2656 | pr_warn("map '%s': inner def can't be pinned.\n", map_name); |
646f02ff AN |
2657 | return -EINVAL; |
2658 | } | |
c7ef5ec9 | 2659 | if (!get_map_field_int(map_name, btf, m, &val)) |
57a00f41 | 2660 | return -EINVAL; |
c7ef5ec9 | 2661 | if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) { |
57a00f41 | 2662 | pr_warn("map '%s': invalid pinning value %u.\n", |
c7ef5ec9 | 2663 | map_name, val); |
57a00f41 THJ |
2664 | return -EINVAL; |
2665 | } | |
c7ef5ec9 AN |
2666 | map_def->pinning = val; |
2667 | map_def->parts |= MAP_DEF_PINNING; | |
47512102 | 2668 | } else if (strcmp(name, "map_extra") == 0) { |
d147357e | 2669 | __u64 map_extra; |
47512102 | 2670 | |
d147357e | 2671 | if (!get_map_field_long(map_name, btf, m, &map_extra)) |
47512102 JK |
2672 | return -EINVAL; |
2673 | map_def->map_extra = map_extra; | |
2674 | map_def->parts |= MAP_DEF_MAP_EXTRA; | |
abd29c93 AN |
2675 | } else { |
2676 | if (strict) { | |
c7ef5ec9 | 2677 | pr_warn("map '%s': unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2678 | return -ENOTSUP; |
2679 | } | |
c7ef5ec9 | 2680 | pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2681 | } |
2682 | } | |
2683 | ||
c7ef5ec9 AN |
2684 | if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) { |
2685 | pr_warn("map '%s': map type isn't specified.\n", map_name); | |
abd29c93 AN |
2686 | return -EINVAL; |
2687 | } | |
2688 | ||
2689 | return 0; | |
2690 | } | |
2691 | ||
597fbc46 AN |
2692 | static size_t adjust_ringbuf_sz(size_t sz) |
2693 | { | |
2694 | __u32 page_sz = sysconf(_SC_PAGE_SIZE); | |
2695 | __u32 mul; | |
2696 | ||
2697 | /* if user forgot to set any size, make sure they see error */ | |
2698 | if (sz == 0) | |
2699 | return 0; | |
2700 | /* Kernel expects BPF_MAP_TYPE_RINGBUF's max_entries to be | |
2701 | * a power-of-2 multiple of kernel's page size. If user diligently | |
2702 | * satisified these conditions, pass the size through. | |
2703 | */ | |
2704 | if ((sz % page_sz) == 0 && is_pow_of_2(sz / page_sz)) | |
2705 | return sz; | |
2706 | ||
2707 | /* Otherwise find closest (page_sz * power_of_2) product bigger than | |
2708 | * user-set size to satisfy both user size request and kernel | |
2709 | * requirements and substitute correct max_entries for map creation. | |
2710 | */ | |
2711 | for (mul = 1; mul <= UINT_MAX / page_sz; mul <<= 1) { | |
2712 | if (mul * page_sz > sz) | |
2713 | return mul * page_sz; | |
2714 | } | |
2715 | ||
2716 | /* if it's impossible to satisfy the conditions (i.e., user size is | |
2717 | * very close to UINT_MAX but is not a power-of-2 multiple of | |
2718 | * page_size) then just return original size and let kernel reject it | |
2719 | */ | |
2720 | return sz; | |
2721 | } | |
2722 | ||
b66ccae0 DV |
2723 | static bool map_is_ringbuf(const struct bpf_map *map) |
2724 | { | |
2725 | return map->def.type == BPF_MAP_TYPE_RINGBUF || | |
2726 | map->def.type == BPF_MAP_TYPE_USER_RINGBUF; | |
2727 | } | |
2728 | ||
c7ef5ec9 AN |
2729 | static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def) |
2730 | { | |
2731 | map->def.type = def->map_type; | |
2732 | map->def.key_size = def->key_size; | |
2733 | map->def.value_size = def->value_size; | |
2734 | map->def.max_entries = def->max_entries; | |
2735 | map->def.map_flags = def->map_flags; | |
47512102 | 2736 | map->map_extra = def->map_extra; |
c7ef5ec9 AN |
2737 | |
2738 | map->numa_node = def->numa_node; | |
2739 | map->btf_key_type_id = def->key_type_id; | |
2740 | map->btf_value_type_id = def->value_type_id; | |
2741 | ||
597fbc46 | 2742 | /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */ |
b66ccae0 | 2743 | if (map_is_ringbuf(map)) |
597fbc46 AN |
2744 | map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries); |
2745 | ||
c7ef5ec9 AN |
2746 | if (def->parts & MAP_DEF_MAP_TYPE) |
2747 | pr_debug("map '%s': found type = %u.\n", map->name, def->map_type); | |
2748 | ||
2749 | if (def->parts & MAP_DEF_KEY_TYPE) | |
2750 | pr_debug("map '%s': found key [%u], sz = %u.\n", | |
2751 | map->name, def->key_type_id, def->key_size); | |
2752 | else if (def->parts & MAP_DEF_KEY_SIZE) | |
2753 | pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size); | |
2754 | ||
2755 | if (def->parts & MAP_DEF_VALUE_TYPE) | |
2756 | pr_debug("map '%s': found value [%u], sz = %u.\n", | |
2757 | map->name, def->value_type_id, def->value_size); | |
2758 | else if (def->parts & MAP_DEF_VALUE_SIZE) | |
2759 | pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size); | |
2760 | ||
2761 | if (def->parts & MAP_DEF_MAX_ENTRIES) | |
2762 | pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries); | |
2763 | if (def->parts & MAP_DEF_MAP_FLAGS) | |
47512102 JK |
2764 | pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags); |
2765 | if (def->parts & MAP_DEF_MAP_EXTRA) | |
2766 | pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name, | |
2767 | (unsigned long long)def->map_extra); | |
c7ef5ec9 AN |
2768 | if (def->parts & MAP_DEF_PINNING) |
2769 | pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning); | |
2770 | if (def->parts & MAP_DEF_NUMA_NODE) | |
2771 | pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node); | |
2772 | ||
2773 | if (def->parts & MAP_DEF_INNER_MAP) | |
2774 | pr_debug("map '%s': found inner map definition.\n", map->name); | |
2775 | } | |
2776 | ||
c1cccec9 AN |
2777 | static const char *btf_var_linkage_str(__u32 linkage) |
2778 | { | |
2779 | switch (linkage) { | |
2780 | case BTF_VAR_STATIC: return "static"; | |
2781 | case BTF_VAR_GLOBAL_ALLOCATED: return "global"; | |
2782 | case BTF_VAR_GLOBAL_EXTERN: return "extern"; | |
2783 | default: return "unknown"; | |
2784 | } | |
2785 | } | |
2786 | ||
41017e56 AN |
2787 | static int bpf_object__init_user_btf_map(struct bpf_object *obj, |
2788 | const struct btf_type *sec, | |
2789 | int var_idx, int sec_idx, | |
2790 | const Elf_Data *data, bool strict, | |
2791 | const char *pin_root_path) | |
2792 | { | |
c7ef5ec9 | 2793 | struct btf_map_def map_def = {}, inner_def = {}; |
41017e56 AN |
2794 | const struct btf_type *var, *def; |
2795 | const struct btf_var_secinfo *vi; | |
2796 | const struct btf_var *var_extra; | |
2797 | const char *map_name; | |
2798 | struct bpf_map *map; | |
c7ef5ec9 | 2799 | int err; |
41017e56 AN |
2800 | |
2801 | vi = btf_var_secinfos(sec) + var_idx; | |
2802 | var = btf__type_by_id(obj->btf, vi->type); | |
2803 | var_extra = btf_var(var); | |
2804 | map_name = btf__name_by_offset(obj->btf, var->name_off); | |
2805 | ||
2806 | if (map_name == NULL || map_name[0] == '\0') { | |
2807 | pr_warn("map #%d: empty name.\n", var_idx); | |
2808 | return -EINVAL; | |
2809 | } | |
2810 | if ((__u64)vi->offset + vi->size > data->d_size) { | |
2811 | pr_warn("map '%s' BTF data is corrupted.\n", map_name); | |
2812 | return -EINVAL; | |
2813 | } | |
2814 | if (!btf_is_var(var)) { | |
81ba0889 AN |
2815 | pr_warn("map '%s': unexpected var kind %s.\n", |
2816 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2817 | return -EINVAL; |
2818 | } | |
c1cccec9 AN |
2819 | if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) { |
2820 | pr_warn("map '%s': unsupported map linkage %s.\n", | |
2821 | map_name, btf_var_linkage_str(var_extra->linkage)); | |
41017e56 AN |
2822 | return -EOPNOTSUPP; |
2823 | } | |
2824 | ||
2825 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
2826 | if (!btf_is_struct(def)) { | |
81ba0889 AN |
2827 | pr_warn("map '%s': unexpected def kind %s.\n", |
2828 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2829 | return -EINVAL; |
2830 | } | |
2831 | if (def->size > vi->size) { | |
2832 | pr_warn("map '%s': invalid def size.\n", map_name); | |
2833 | return -EINVAL; | |
2834 | } | |
2835 | ||
2836 | map = bpf_object__add_map(obj); | |
2837 | if (IS_ERR(map)) | |
2838 | return PTR_ERR(map); | |
2839 | map->name = strdup(map_name); | |
2840 | if (!map->name) { | |
2841 | pr_warn("map '%s': failed to alloc map name.\n", map_name); | |
2842 | return -ENOMEM; | |
2843 | } | |
2844 | map->libbpf_type = LIBBPF_MAP_UNSPEC; | |
2845 | map->def.type = BPF_MAP_TYPE_UNSPEC; | |
2846 | map->sec_idx = sec_idx; | |
2847 | map->sec_offset = vi->offset; | |
646f02ff | 2848 | map->btf_var_idx = var_idx; |
41017e56 AN |
2849 | pr_debug("map '%s': at sec_idx %d, offset %zu.\n", |
2850 | map_name, map->sec_idx, map->sec_offset); | |
2851 | ||
c7ef5ec9 AN |
2852 | err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def); |
2853 | if (err) | |
2854 | return err; | |
2855 | ||
2856 | fill_map_from_def(map, &map_def); | |
2857 | ||
2858 | if (map_def.pinning == LIBBPF_PIN_BY_NAME) { | |
2859 | err = build_map_pin_path(map, pin_root_path); | |
2860 | if (err) { | |
2861 | pr_warn("map '%s': couldn't build pin path.\n", map->name); | |
2862 | return err; | |
2863 | } | |
2864 | } | |
2865 | ||
2866 | if (map_def.parts & MAP_DEF_INNER_MAP) { | |
2867 | map->inner_map = calloc(1, sizeof(*map->inner_map)); | |
2868 | if (!map->inner_map) | |
2869 | return -ENOMEM; | |
dac645b9 AN |
2870 | map->inner_map->fd = create_placeholder_fd(); |
2871 | if (map->inner_map->fd < 0) | |
2872 | return map->inner_map->fd; | |
c7ef5ec9 AN |
2873 | map->inner_map->sec_idx = sec_idx; |
2874 | map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1); | |
2875 | if (!map->inner_map->name) | |
2876 | return -ENOMEM; | |
2877 | sprintf(map->inner_map->name, "%s.inner", map_name); | |
2878 | ||
2879 | fill_map_from_def(map->inner_map, &inner_def); | |
2880 | } | |
2881 | ||
4fcac46c | 2882 | err = map_fill_btf_type_info(obj, map); |
262cfb74 DK |
2883 | if (err) |
2884 | return err; | |
2885 | ||
c7ef5ec9 | 2886 | return 0; |
41017e56 AN |
2887 | } |
2888 | ||
2e7ba4f8 AN |
2889 | static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map, |
2890 | const char *sec_name, int sec_idx, | |
2891 | void *data, size_t data_sz) | |
2892 | { | |
2893 | const long page_sz = sysconf(_SC_PAGE_SIZE); | |
2894 | size_t mmap_sz; | |
2895 | ||
2896 | mmap_sz = bpf_map_mmap_sz(obj->arena_map); | |
2897 | if (roundup(data_sz, page_sz) > mmap_sz) { | |
2898 | pr_warn("elf: sec '%s': declared ARENA map size (%zu) is too small to hold global __arena variables of size %zu\n", | |
2899 | sec_name, mmap_sz, data_sz); | |
2900 | return -E2BIG; | |
2901 | } | |
2902 | ||
2903 | obj->arena_data = malloc(data_sz); | |
2904 | if (!obj->arena_data) | |
2905 | return -ENOMEM; | |
2906 | memcpy(obj->arena_data, data, data_sz); | |
2907 | obj->arena_data_sz = data_sz; | |
2908 | ||
2909 | /* make bpf_map__init_value() work for ARENA maps */ | |
2910 | map->mmaped = obj->arena_data; | |
2911 | ||
2912 | return 0; | |
2913 | } | |
2914 | ||
57a00f41 THJ |
2915 | static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict, |
2916 | const char *pin_root_path) | |
abd29c93 AN |
2917 | { |
2918 | const struct btf_type *sec = NULL; | |
2919 | int nr_types, i, vlen, err; | |
2920 | const struct btf_type *t; | |
2921 | const char *name; | |
2922 | Elf_Data *data; | |
2923 | Elf_Scn *scn; | |
2924 | ||
2925 | if (obj->efile.btf_maps_shndx < 0) | |
2926 | return 0; | |
2927 | ||
88a82120 AN |
2928 | scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx); |
2929 | data = elf_sec_data(obj, scn); | |
abd29c93 | 2930 | if (!scn || !data) { |
88a82120 AN |
2931 | pr_warn("elf: failed to get %s map definitions for %s\n", |
2932 | MAPS_ELF_SEC, obj->path); | |
abd29c93 AN |
2933 | return -EINVAL; |
2934 | } | |
2935 | ||
6a886de0 HC |
2936 | nr_types = btf__type_cnt(obj->btf); |
2937 | for (i = 1; i < nr_types; i++) { | |
abd29c93 | 2938 | t = btf__type_by_id(obj->btf, i); |
b03bc685 | 2939 | if (!btf_is_datasec(t)) |
abd29c93 AN |
2940 | continue; |
2941 | name = btf__name_by_offset(obj->btf, t->name_off); | |
2942 | if (strcmp(name, MAPS_ELF_SEC) == 0) { | |
2943 | sec = t; | |
646f02ff | 2944 | obj->efile.btf_maps_sec_btf_id = i; |
abd29c93 AN |
2945 | break; |
2946 | } | |
2947 | } | |
2948 | ||
2949 | if (!sec) { | |
be18010e | 2950 | pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC); |
abd29c93 AN |
2951 | return -ENOENT; |
2952 | } | |
2953 | ||
b03bc685 | 2954 | vlen = btf_vlen(sec); |
abd29c93 AN |
2955 | for (i = 0; i < vlen; i++) { |
2956 | err = bpf_object__init_user_btf_map(obj, sec, i, | |
2957 | obj->efile.btf_maps_shndx, | |
8983b731 AN |
2958 | data, strict, |
2959 | pin_root_path); | |
abd29c93 AN |
2960 | if (err) |
2961 | return err; | |
2962 | } | |
2963 | ||
2e7ba4f8 AN |
2964 | for (i = 0; i < obj->nr_maps; i++) { |
2965 | struct bpf_map *map = &obj->maps[i]; | |
2966 | ||
2967 | if (map->def.type != BPF_MAP_TYPE_ARENA) | |
2968 | continue; | |
2969 | ||
2970 | if (obj->arena_map) { | |
2971 | pr_warn("map '%s': only single ARENA map is supported (map '%s' is also ARENA)\n", | |
2972 | map->name, obj->arena_map->name); | |
2973 | return -EINVAL; | |
2974 | } | |
2975 | obj->arena_map = map; | |
2976 | ||
2977 | if (obj->efile.arena_data) { | |
2978 | err = init_arena_map_data(obj, map, ARENA_SEC, obj->efile.arena_data_shndx, | |
2979 | obj->efile.arena_data->d_buf, | |
2980 | obj->efile.arena_data->d_size); | |
2981 | if (err) | |
2982 | return err; | |
2983 | } | |
2984 | } | |
2985 | if (obj->efile.arena_data && !obj->arena_map) { | |
2986 | pr_warn("elf: sec '%s': to use global __arena variables the ARENA map should be explicitly declared in SEC(\".maps\")\n", | |
2987 | ARENA_SEC); | |
2988 | return -ENOENT; | |
2989 | } | |
2990 | ||
abd29c93 AN |
2991 | return 0; |
2992 | } | |
2993 | ||
0d13bfce | 2994 | static int bpf_object__init_maps(struct bpf_object *obj, |
01af3bf0 | 2995 | const struct bpf_object_open_opts *opts) |
bf829271 | 2996 | { |
166750bc AN |
2997 | const char *pin_root_path; |
2998 | bool strict; | |
bd054102 | 2999 | int err = 0; |
8837fe5d | 3000 | |
166750bc AN |
3001 | strict = !OPTS_GET(opts, relaxed_maps, false); |
3002 | pin_root_path = OPTS_GET(opts, pin_root_path, NULL); | |
bf829271 | 3003 | |
40e1bcab | 3004 | err = bpf_object__init_user_btf_maps(obj, strict, pin_root_path); |
166750bc | 3005 | err = err ?: bpf_object__init_global_data_maps(obj); |
81bfdd08 | 3006 | err = err ?: bpf_object__init_kconfig_map(obj); |
809a69d6 | 3007 | err = err ?: bpf_object_init_struct_ops(obj); |
bf829271 | 3008 | |
3b3af91c | 3009 | return err; |
561bbcca WN |
3010 | } |
3011 | ||
e3d91b0c JDB |
3012 | static bool section_have_execinstr(struct bpf_object *obj, int idx) |
3013 | { | |
ad23b723 | 3014 | Elf64_Shdr *sh; |
e3d91b0c | 3015 | |
ad23b723 AN |
3016 | sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx)); |
3017 | if (!sh) | |
e3d91b0c JDB |
3018 | return false; |
3019 | ||
ad23b723 | 3020 | return sh->sh_flags & SHF_EXECINSTR; |
e3d91b0c JDB |
3021 | } |
3022 | ||
6ebaa3fb EZ |
3023 | static bool starts_with_qmark(const char *s) |
3024 | { | |
3025 | return s && s[0] == '?'; | |
3026 | } | |
3027 | ||
0f0e55d8 AN |
3028 | static bool btf_needs_sanitization(struct bpf_object *obj) |
3029 | { | |
9ca1f56a AS |
3030 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
3031 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
3032 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
3033 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 3034 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
2dc1e488 | 3035 | bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG); |
f2a62588 | 3036 | bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64); |
6ebaa3fb | 3037 | bool has_qmark_datasec = kernel_supports(obj, FEAT_BTF_QMARK_DATASEC); |
0f0e55d8 | 3038 | |
2dc1e488 | 3039 | return !has_func || !has_datasec || !has_func_global || !has_float || |
6ebaa3fb | 3040 | !has_decl_tag || !has_type_tag || !has_enum64 || !has_qmark_datasec; |
0f0e55d8 AN |
3041 | } |
3042 | ||
f2a62588 | 3043 | static int bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf) |
d7c4b398 | 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); |
c49a44b3 | 3053 | int enum64_placeholder_id = 0; |
d7c4b398 AN |
3054 | struct btf_type *t; |
3055 | int i, j, vlen; | |
d7c4b398 | 3056 | |
6a886de0 | 3057 | for (i = 1; i < btf__type_cnt(btf); i++) { |
d7c4b398 | 3058 | t = (struct btf_type *)btf__type_by_id(btf, i); |
d7c4b398 | 3059 | |
223f903e YS |
3060 | if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) { |
3061 | /* replace VAR/DECL_TAG with INT */ | |
d7c4b398 | 3062 | t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0); |
1d4126c4 AN |
3063 | /* |
3064 | * using size = 1 is the safest choice, 4 will be too | |
3065 | * big and cause kernel BTF validation failure if | |
3066 | * original variable took less than 4 bytes | |
3067 | */ | |
3068 | t->size = 1; | |
708852dc | 3069 | *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8); |
b03bc685 | 3070 | } else if (!has_datasec && btf_is_datasec(t)) { |
d7c4b398 | 3071 | /* replace DATASEC with STRUCT */ |
b03bc685 AN |
3072 | const struct btf_var_secinfo *v = btf_var_secinfos(t); |
3073 | struct btf_member *m = btf_members(t); | |
d7c4b398 AN |
3074 | struct btf_type *vt; |
3075 | char *name; | |
3076 | ||
3077 | name = (char *)btf__name_by_offset(btf, t->name_off); | |
3078 | while (*name) { | |
6ebaa3fb | 3079 | if (*name == '.' || *name == '?') |
d7c4b398 AN |
3080 | *name = '_'; |
3081 | name++; | |
3082 | } | |
3083 | ||
b03bc685 | 3084 | vlen = btf_vlen(t); |
d7c4b398 AN |
3085 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen); |
3086 | for (j = 0; j < vlen; j++, v++, m++) { | |
3087 | /* order of field assignments is important */ | |
3088 | m->offset = v->offset * 8; | |
3089 | m->type = v->type; | |
3090 | /* preserve variable name as member name */ | |
3091 | vt = (void *)btf__type_by_id(btf, v->type); | |
3092 | m->name_off = vt->name_off; | |
3093 | } | |
6ebaa3fb EZ |
3094 | } else if (!has_qmark_datasec && btf_is_datasec(t) && |
3095 | starts_with_qmark(btf__name_by_offset(btf, t->name_off))) { | |
3096 | /* replace '?' prefix with '_' for DATASEC names */ | |
3097 | char *name; | |
3098 | ||
3099 | name = (char *)btf__name_by_offset(btf, t->name_off); | |
3100 | if (name[0] == '?') | |
3101 | name[0] = '_'; | |
b03bc685 | 3102 | } else if (!has_func && btf_is_func_proto(t)) { |
d7c4b398 | 3103 | /* replace FUNC_PROTO with ENUM */ |
b03bc685 | 3104 | vlen = btf_vlen(t); |
d7c4b398 AN |
3105 | t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen); |
3106 | t->size = sizeof(__u32); /* kernel enforced */ | |
b03bc685 | 3107 | } else if (!has_func && btf_is_func(t)) { |
d7c4b398 AN |
3108 | /* replace FUNC with TYPEDEF */ |
3109 | t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0); | |
2d3eb67f AS |
3110 | } else if (!has_func_global && btf_is_func(t)) { |
3111 | /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */ | |
3112 | t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0); | |
22541a9e IL |
3113 | } else if (!has_float && btf_is_float(t)) { |
3114 | /* replace FLOAT with an equally-sized empty STRUCT; | |
3115 | * since C compilers do not accept e.g. "float" as a | |
3116 | * valid struct name, make it anonymous | |
3117 | */ | |
3118 | t->name_off = 0; | |
3119 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0); | |
2dc1e488 YS |
3120 | } else if (!has_type_tag && btf_is_type_tag(t)) { |
3121 | /* replace TYPE_TAG with a CONST */ | |
3122 | t->name_off = 0; | |
3123 | t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0); | |
f2a62588 YS |
3124 | } else if (!has_enum64 && btf_is_enum(t)) { |
3125 | /* clear the kflag */ | |
3126 | t->info = btf_type_info(btf_kind(t), btf_vlen(t), false); | |
3127 | } else if (!has_enum64 && btf_is_enum64(t)) { | |
3128 | /* replace ENUM64 with a union */ | |
3129 | struct btf_member *m; | |
3130 | ||
3131 | if (enum64_placeholder_id == 0) { | |
3132 | enum64_placeholder_id = btf__add_int(btf, "enum64_placeholder", 1, 0); | |
3133 | if (enum64_placeholder_id < 0) | |
3134 | return enum64_placeholder_id; | |
3135 | ||
3136 | t = (struct btf_type *)btf__type_by_id(btf, i); | |
3137 | } | |
3138 | ||
3139 | m = btf_members(t); | |
3140 | vlen = btf_vlen(t); | |
3141 | t->info = BTF_INFO_ENC(BTF_KIND_UNION, 0, vlen); | |
3142 | for (j = 0; j < vlen; j++, m++) { | |
3143 | m->type = enum64_placeholder_id; | |
3144 | m->offset = 0; | |
3145 | } | |
e3ba8e4e | 3146 | } |
d7c4b398 | 3147 | } |
f2a62588 YS |
3148 | |
3149 | return 0; | |
d7c4b398 AN |
3150 | } |
3151 | ||
b35f14f4 | 3152 | static bool libbpf_needs_btf(const struct bpf_object *obj) |
abd29c93 | 3153 | { |
b35f14f4 | 3154 | return obj->efile.btf_maps_shndx >= 0 || |
240bf8a5 | 3155 | obj->efile.has_st_ops || |
b35f14f4 AN |
3156 | obj->nr_extern > 0; |
3157 | } | |
3158 | ||
3159 | static bool kernel_needs_btf(const struct bpf_object *obj) | |
3160 | { | |
240bf8a5 | 3161 | return obj->efile.has_st_ops; |
abd29c93 AN |
3162 | } |
3163 | ||
063183bf | 3164 | static int bpf_object__init_btf(struct bpf_object *obj, |
9c6660d0 AN |
3165 | Elf_Data *btf_data, |
3166 | Elf_Data *btf_ext_data) | |
3167 | { | |
b7d7f3e1 | 3168 | int err = -ENOENT; |
9c6660d0 AN |
3169 | |
3170 | if (btf_data) { | |
3171 | obj->btf = btf__new(btf_data->d_buf, btf_data->d_size); | |
e9fc3ce9 AN |
3172 | err = libbpf_get_error(obj->btf); |
3173 | if (err) { | |
b7d7f3e1 | 3174 | obj->btf = NULL; |
e9fc3ce9 | 3175 | pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err); |
9c6660d0 AN |
3176 | goto out; |
3177 | } | |
4c01925f AN |
3178 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
3179 | btf__set_pointer_size(obj->btf, 8); | |
9c6660d0 AN |
3180 | } |
3181 | if (btf_ext_data) { | |
11d5daa8 AN |
3182 | struct btf_ext_info *ext_segs[3]; |
3183 | int seg_num, sec_num; | |
3184 | ||
9c6660d0 AN |
3185 | if (!obj->btf) { |
3186 | pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n", | |
3187 | BTF_EXT_ELF_SEC, BTF_ELF_SEC); | |
3188 | goto out; | |
3189 | } | |
e9fc3ce9 AN |
3190 | obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size); |
3191 | err = libbpf_get_error(obj->btf_ext); | |
3192 | if (err) { | |
3193 | pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n", | |
3194 | BTF_EXT_ELF_SEC, err); | |
9c6660d0 AN |
3195 | obj->btf_ext = NULL; |
3196 | goto out; | |
3197 | } | |
11d5daa8 AN |
3198 | |
3199 | /* setup .BTF.ext to ELF section mapping */ | |
3200 | ext_segs[0] = &obj->btf_ext->func_info; | |
3201 | ext_segs[1] = &obj->btf_ext->line_info; | |
3202 | ext_segs[2] = &obj->btf_ext->core_relo_info; | |
3203 | for (seg_num = 0; seg_num < ARRAY_SIZE(ext_segs); seg_num++) { | |
3204 | struct btf_ext_info *seg = ext_segs[seg_num]; | |
3205 | const struct btf_ext_info_sec *sec; | |
3206 | const char *sec_name; | |
3207 | Elf_Scn *scn; | |
3208 | ||
3209 | if (seg->sec_cnt == 0) | |
3210 | continue; | |
3211 | ||
3212 | seg->sec_idxs = calloc(seg->sec_cnt, sizeof(*seg->sec_idxs)); | |
3213 | if (!seg->sec_idxs) { | |
3214 | err = -ENOMEM; | |
3215 | goto out; | |
3216 | } | |
3217 | ||
3218 | sec_num = 0; | |
3219 | for_each_btf_ext_sec(seg, sec) { | |
3220 | /* preventively increment index to avoid doing | |
3221 | * this before every continue below | |
3222 | */ | |
3223 | sec_num++; | |
3224 | ||
3225 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); | |
3226 | if (str_is_empty(sec_name)) | |
3227 | continue; | |
3228 | scn = elf_sec_by_name(obj, sec_name); | |
3229 | if (!scn) | |
3230 | continue; | |
3231 | ||
3232 | seg->sec_idxs[sec_num - 1] = elf_ndxscn(scn); | |
3233 | } | |
3234 | } | |
9c6660d0 AN |
3235 | } |
3236 | out: | |
b35f14f4 | 3237 | if (err && libbpf_needs_btf(obj)) { |
be18010e | 3238 | pr_warn("BTF is required, but is missing or corrupted.\n"); |
b7d7f3e1 | 3239 | return err; |
abd29c93 | 3240 | } |
9c6660d0 AN |
3241 | return 0; |
3242 | } | |
3243 | ||
b96c07f3 AN |
3244 | static int compare_vsi_off(const void *_a, const void *_b) |
3245 | { | |
3246 | const struct btf_var_secinfo *a = _a; | |
3247 | const struct btf_var_secinfo *b = _b; | |
3248 | ||
3249 | return a->offset - b->offset; | |
3250 | } | |
3251 | ||
3252 | static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf, | |
3253 | struct btf_type *t) | |
3254 | { | |
f33f742d AN |
3255 | __u32 size = 0, i, vars = btf_vlen(t); |
3256 | const char *sec_name = btf__name_by_offset(btf, t->name_off); | |
b96c07f3 | 3257 | struct btf_var_secinfo *vsi; |
4fcac46c | 3258 | bool fixup_offsets = false; |
f33f742d | 3259 | int err; |
b96c07f3 | 3260 | |
f33f742d | 3261 | if (!sec_name) { |
b96c07f3 AN |
3262 | pr_debug("No name found in string section for DATASEC kind.\n"); |
3263 | return -ENOENT; | |
3264 | } | |
3265 | ||
4fcac46c AN |
3266 | /* Extern-backing datasecs (.ksyms, .kconfig) have their size and |
3267 | * variable offsets set at the previous step. Further, not every | |
3268 | * extern BTF VAR has corresponding ELF symbol preserved, so we skip | |
3269 | * all fixups altogether for such sections and go straight to sorting | |
3270 | * VARs within their DATASEC. | |
b96c07f3 | 3271 | */ |
4fcac46c | 3272 | if (strcmp(sec_name, KCONFIG_SEC) == 0 || strcmp(sec_name, KSYMS_SEC) == 0) |
b96c07f3 AN |
3273 | goto sort_vars; |
3274 | ||
4fcac46c AN |
3275 | /* Clang leaves DATASEC size and VAR offsets as zeroes, so we need to |
3276 | * fix this up. But BPF static linker already fixes this up and fills | |
3277 | * all the sizes and offsets during static linking. So this step has | |
3278 | * to be optional. But the STV_HIDDEN handling is non-optional for any | |
3279 | * non-extern DATASEC, so the variable fixup loop below handles both | |
3280 | * functions at the same time, paying the cost of BTF VAR <-> ELF | |
3281 | * symbol matching just once. | |
3282 | */ | |
3283 | if (t->size == 0) { | |
3284 | err = find_elf_sec_sz(obj, sec_name, &size); | |
3285 | if (err || !size) { | |
3286 | pr_debug("sec '%s': failed to determine size from ELF: size %u, err %d\n", | |
3287 | sec_name, size, err); | |
3288 | return -ENOENT; | |
3289 | } | |
b96c07f3 | 3290 | |
4fcac46c AN |
3291 | t->size = size; |
3292 | fixup_offsets = true; | |
3293 | } | |
b96c07f3 AN |
3294 | |
3295 | for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) { | |
f33f742d AN |
3296 | const struct btf_type *t_var; |
3297 | struct btf_var *var; | |
3298 | const char *var_name; | |
3299 | Elf64_Sym *sym; | |
3300 | ||
b96c07f3 | 3301 | t_var = btf__type_by_id(btf, vsi->type); |
88918dc1 | 3302 | if (!t_var || !btf_is_var(t_var)) { |
f33f742d | 3303 | pr_debug("sec '%s': unexpected non-VAR type found\n", sec_name); |
b96c07f3 AN |
3304 | return -EINVAL; |
3305 | } | |
3306 | ||
88918dc1 | 3307 | var = btf_var(t_var); |
f33f742d | 3308 | if (var->linkage == BTF_VAR_STATIC || var->linkage == BTF_VAR_GLOBAL_EXTERN) |
b96c07f3 AN |
3309 | continue; |
3310 | ||
f33f742d AN |
3311 | var_name = btf__name_by_offset(btf, t_var->name_off); |
3312 | if (!var_name) { | |
3313 | pr_debug("sec '%s': failed to find name of DATASEC's member #%d\n", | |
3314 | sec_name, i); | |
b96c07f3 AN |
3315 | return -ENOENT; |
3316 | } | |
3317 | ||
f33f742d AN |
3318 | sym = find_elf_var_sym(obj, var_name); |
3319 | if (IS_ERR(sym)) { | |
3320 | pr_debug("sec '%s': failed to find ELF symbol for VAR '%s'\n", | |
3321 | sec_name, var_name); | |
b96c07f3 AN |
3322 | return -ENOENT; |
3323 | } | |
3324 | ||
4fcac46c AN |
3325 | if (fixup_offsets) |
3326 | vsi->offset = sym->st_value; | |
3327 | ||
3328 | /* if variable is a global/weak symbol, but has restricted | |
3329 | * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF VAR | |
3330 | * as static. This follows similar logic for functions (BPF | |
3331 | * subprogs) and influences libbpf's further decisions about | |
3332 | * whether to make global data BPF array maps as | |
3333 | * BPF_F_MMAPABLE. | |
3334 | */ | |
3335 | if (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN | |
3336 | || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL) | |
3337 | var->linkage = BTF_VAR_STATIC; | |
b96c07f3 AN |
3338 | } |
3339 | ||
3340 | sort_vars: | |
3341 | qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off); | |
3342 | return 0; | |
3343 | } | |
3344 | ||
f33f742d | 3345 | static int bpf_object_fixup_btf(struct bpf_object *obj) |
b96c07f3 | 3346 | { |
f33f742d | 3347 | int i, n, err = 0; |
b96c07f3 | 3348 | |
f33f742d AN |
3349 | if (!obj->btf) |
3350 | return 0; | |
3351 | ||
3352 | n = btf__type_cnt(obj->btf); | |
6a886de0 | 3353 | for (i = 1; i < n; i++) { |
f33f742d | 3354 | struct btf_type *t = btf_type_by_id(obj->btf, i); |
b96c07f3 AN |
3355 | |
3356 | /* Loader needs to fix up some of the things compiler | |
3357 | * couldn't get its hands on while emitting BTF. This | |
3358 | * is section size and global variable offset. We use | |
3359 | * the info from the ELF itself for this purpose. | |
3360 | */ | |
3361 | if (btf_is_datasec(t)) { | |
f33f742d | 3362 | err = btf_fixup_datasec(obj, obj->btf, t); |
b96c07f3 | 3363 | if (err) |
f33f742d | 3364 | return err; |
b96c07f3 AN |
3365 | } |
3366 | } | |
3367 | ||
166750bc AN |
3368 | return 0; |
3369 | } | |
3370 | ||
fe62de31 | 3371 | static bool prog_needs_vmlinux_btf(struct bpf_program *prog) |
a6ed02ca | 3372 | { |
1e092a03 KS |
3373 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS || |
3374 | prog->type == BPF_PROG_TYPE_LSM) | |
a6ed02ca KS |
3375 | return true; |
3376 | ||
3377 | /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs | |
3378 | * also need vmlinux BTF | |
3379 | */ | |
3380 | if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd) | |
3381 | return true; | |
3382 | ||
3383 | return false; | |
3384 | } | |
3385 | ||
8b7b0e5f DV |
3386 | static bool map_needs_vmlinux_btf(struct bpf_map *map) |
3387 | { | |
3388 | return bpf_map__is_struct_ops(map); | |
3389 | } | |
3390 | ||
fe62de31 | 3391 | static bool obj_needs_vmlinux_btf(const struct bpf_object *obj) |
a6ed02ca KS |
3392 | { |
3393 | struct bpf_program *prog; | |
8b7b0e5f | 3394 | struct bpf_map *map; |
fe62de31 | 3395 | int i; |
a6ed02ca | 3396 | |
1373ff59 SC |
3397 | /* CO-RE relocations need kernel BTF, only when btf_custom_path |
3398 | * is not specified | |
3399 | */ | |
3400 | if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path) | |
fe62de31 | 3401 | return true; |
192b6638 | 3402 | |
d370bbe1 HL |
3403 | /* Support for typed ksyms needs kernel BTF */ |
3404 | for (i = 0; i < obj->nr_extern; i++) { | |
3405 | const struct extern_desc *ext; | |
3406 | ||
3407 | ext = &obj->externs[i]; | |
fe62de31 AN |
3408 | if (ext->type == EXT_KSYM && ext->ksym.type_id) |
3409 | return true; | |
d370bbe1 HL |
3410 | } |
3411 | ||
a6ed02ca | 3412 | bpf_object__for_each_program(prog, obj) { |
a3820c48 | 3413 | if (!prog->autoload) |
d9297581 | 3414 | continue; |
fe62de31 AN |
3415 | if (prog_needs_vmlinux_btf(prog)) |
3416 | return true; | |
a6ed02ca KS |
3417 | } |
3418 | ||
8b7b0e5f DV |
3419 | bpf_object__for_each_map(map, obj) { |
3420 | if (map_needs_vmlinux_btf(map)) | |
3421 | return true; | |
3422 | } | |
3423 | ||
fe62de31 AN |
3424 | return false; |
3425 | } | |
3426 | ||
3427 | static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force) | |
3428 | { | |
3429 | int err; | |
3430 | ||
3431 | /* btf_vmlinux could be loaded earlier */ | |
67234743 | 3432 | if (obj->btf_vmlinux || obj->gen_loader) |
fe62de31 AN |
3433 | return 0; |
3434 | ||
3435 | if (!force && !obj_needs_vmlinux_btf(obj)) | |
192b6638 AN |
3436 | return 0; |
3437 | ||
a710eed3 | 3438 | obj->btf_vmlinux = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
3439 | err = libbpf_get_error(obj->btf_vmlinux); |
3440 | if (err) { | |
192b6638 AN |
3441 | pr_warn("Error loading vmlinux BTF: %d\n", err); |
3442 | obj->btf_vmlinux = NULL; | |
3443 | return err; | |
3444 | } | |
a6ed02ca KS |
3445 | return 0; |
3446 | } | |
3447 | ||
063183bf AN |
3448 | static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj) |
3449 | { | |
0f0e55d8 AN |
3450 | struct btf *kern_btf = obj->btf; |
3451 | bool btf_mandatory, sanitize; | |
aea28a60 | 3452 | int i, err = 0; |
063183bf AN |
3453 | |
3454 | if (!obj->btf) | |
3455 | return 0; | |
3456 | ||
9ca1f56a | 3457 | if (!kernel_supports(obj, FEAT_BTF)) { |
68b08647 AN |
3458 | if (kernel_needs_btf(obj)) { |
3459 | err = -EOPNOTSUPP; | |
3460 | goto report; | |
3461 | } | |
3462 | pr_debug("Kernel doesn't support BTF, skipping uploading it.\n"); | |
3463 | return 0; | |
3464 | } | |
3465 | ||
aea28a60 AN |
3466 | /* Even though some subprogs are global/weak, user might prefer more |
3467 | * permissive BPF verification process that BPF verifier performs for | |
3468 | * static functions, taking into account more context from the caller | |
3469 | * functions. In such case, they need to mark such subprogs with | |
3470 | * __attribute__((visibility("hidden"))) and libbpf will adjust | |
3471 | * corresponding FUNC BTF type to be marked as static and trigger more | |
3472 | * involved BPF verification process. | |
3473 | */ | |
3474 | for (i = 0; i < obj->nr_programs; i++) { | |
3475 | struct bpf_program *prog = &obj->programs[i]; | |
3476 | struct btf_type *t; | |
3477 | const char *name; | |
3478 | int j, n; | |
3479 | ||
3480 | if (!prog->mark_btf_static || !prog_is_subprog(obj, prog)) | |
3481 | continue; | |
3482 | ||
6a886de0 HC |
3483 | n = btf__type_cnt(obj->btf); |
3484 | for (j = 1; j < n; j++) { | |
aea28a60 AN |
3485 | t = btf_type_by_id(obj->btf, j); |
3486 | if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) | |
3487 | continue; | |
3488 | ||
3489 | name = btf__str_by_offset(obj->btf, t->name_off); | |
3490 | if (strcmp(name, prog->name) != 0) | |
3491 | continue; | |
3492 | ||
3493 | t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0); | |
3494 | break; | |
3495 | } | |
3496 | } | |
3497 | ||
0f0e55d8 AN |
3498 | sanitize = btf_needs_sanitization(obj); |
3499 | if (sanitize) { | |
5c3320d7 | 3500 | const void *raw_data; |
0f0e55d8 | 3501 | __u32 sz; |
063183bf | 3502 | |
0f0e55d8 | 3503 | /* clone BTF to sanitize a copy and leave the original intact */ |
6a886de0 | 3504 | raw_data = btf__raw_data(obj->btf, &sz); |
5c3320d7 | 3505 | kern_btf = btf__new(raw_data, sz); |
e9fc3ce9 AN |
3506 | err = libbpf_get_error(kern_btf); |
3507 | if (err) | |
3508 | return err; | |
04efe591 | 3509 | |
4c01925f AN |
3510 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
3511 | btf__set_pointer_size(obj->btf, 8); | |
f2a62588 YS |
3512 | err = bpf_object__sanitize_btf(obj, kern_btf); |
3513 | if (err) | |
3514 | return err; | |
063183bf | 3515 | } |
0f0e55d8 | 3516 | |
67234743 AS |
3517 | if (obj->gen_loader) { |
3518 | __u32 raw_size = 0; | |
6a886de0 | 3519 | const void *raw_data = btf__raw_data(kern_btf, &raw_size); |
67234743 AS |
3520 | |
3521 | if (!raw_data) | |
3522 | return -ENOMEM; | |
3523 | bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size); | |
3524 | /* Pretend to have valid FD to pass various fd >= 0 checks. | |
3525 | * This fd == 0 will not be used with any syscall and will be reset to -1 eventually. | |
3526 | */ | |
3527 | btf__set_fd(kern_btf, 0); | |
3528 | } else { | |
e0e3ea88 AN |
3529 | /* currently BPF_BTF_LOAD only supports log_level 1 */ |
3530 | err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size, | |
6b434b61 | 3531 | obj->log_level ? 1 : 0, obj->token_fd); |
67234743 | 3532 | } |
0f0e55d8 AN |
3533 | if (sanitize) { |
3534 | if (!err) { | |
3535 | /* move fd to libbpf's BTF */ | |
3536 | btf__set_fd(obj->btf, btf__fd(kern_btf)); | |
3537 | btf__set_fd(kern_btf, -1); | |
3538 | } | |
3539 | btf__free(kern_btf); | |
3540 | } | |
68b08647 | 3541 | report: |
0f0e55d8 AN |
3542 | if (err) { |
3543 | btf_mandatory = kernel_needs_btf(obj); | |
3544 | pr_warn("Error loading .BTF into kernel: %d. %s\n", err, | |
3545 | btf_mandatory ? "BTF is mandatory, can't proceed." | |
3546 | : "BTF is optional, ignoring."); | |
3547 | if (!btf_mandatory) | |
3548 | err = 0; | |
3549 | } | |
3550 | return err; | |
063183bf AN |
3551 | } |
3552 | ||
88a82120 AN |
3553 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off) |
3554 | { | |
3555 | const char *name; | |
3556 | ||
3557 | name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off); | |
3558 | if (!name) { | |
3559 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3560 | off, obj->path, elf_errmsg(-1)); | |
3561 | return NULL; | |
3562 | } | |
3563 | ||
3564 | return name; | |
3565 | } | |
3566 | ||
3567 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off) | |
3568 | { | |
3569 | const char *name; | |
3570 | ||
3571 | name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off); | |
3572 | if (!name) { | |
3573 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3574 | off, obj->path, elf_errmsg(-1)); | |
3575 | return NULL; | |
3576 | } | |
3577 | ||
3578 | return name; | |
3579 | } | |
3580 | ||
3581 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx) | |
3582 | { | |
3583 | Elf_Scn *scn; | |
3584 | ||
3585 | scn = elf_getscn(obj->efile.elf, idx); | |
3586 | if (!scn) { | |
3587 | pr_warn("elf: failed to get section(%zu) from %s: %s\n", | |
3588 | idx, obj->path, elf_errmsg(-1)); | |
3589 | return NULL; | |
3590 | } | |
3591 | return scn; | |
3592 | } | |
3593 | ||
3594 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name) | |
3595 | { | |
3596 | Elf_Scn *scn = NULL; | |
3597 | Elf *elf = obj->efile.elf; | |
3598 | const char *sec_name; | |
3599 | ||
3600 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
3601 | sec_name = elf_sec_name(obj, scn); | |
3602 | if (!sec_name) | |
3603 | return NULL; | |
3604 | ||
3605 | if (strcmp(sec_name, name) != 0) | |
3606 | continue; | |
3607 | ||
3608 | return scn; | |
3609 | } | |
3610 | return NULL; | |
3611 | } | |
3612 | ||
ad23b723 | 3613 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn) |
88a82120 | 3614 | { |
ad23b723 AN |
3615 | Elf64_Shdr *shdr; |
3616 | ||
88a82120 | 3617 | if (!scn) |
ad23b723 | 3618 | return NULL; |
88a82120 | 3619 | |
ad23b723 AN |
3620 | shdr = elf64_getshdr(scn); |
3621 | if (!shdr) { | |
88a82120 AN |
3622 | pr_warn("elf: failed to get section(%zu) header from %s: %s\n", |
3623 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
ad23b723 | 3624 | return NULL; |
88a82120 AN |
3625 | } |
3626 | ||
ad23b723 | 3627 | return shdr; |
88a82120 AN |
3628 | } |
3629 | ||
3630 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn) | |
3631 | { | |
3632 | const char *name; | |
ad23b723 | 3633 | Elf64_Shdr *sh; |
88a82120 AN |
3634 | |
3635 | if (!scn) | |
3636 | return NULL; | |
3637 | ||
ad23b723 AN |
3638 | sh = elf_sec_hdr(obj, scn); |
3639 | if (!sh) | |
88a82120 AN |
3640 | return NULL; |
3641 | ||
ad23b723 | 3642 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 AN |
3643 | if (!name) { |
3644 | pr_warn("elf: failed to get section(%zu) name from %s: %s\n", | |
3645 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
3646 | return NULL; | |
3647 | } | |
3648 | ||
3649 | return name; | |
3650 | } | |
3651 | ||
3652 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn) | |
3653 | { | |
3654 | Elf_Data *data; | |
3655 | ||
3656 | if (!scn) | |
3657 | return NULL; | |
3658 | ||
3659 | data = elf_getdata(scn, 0); | |
3660 | if (!data) { | |
3661 | pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n", | |
3662 | elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>", | |
3663 | obj->path, elf_errmsg(-1)); | |
3664 | return NULL; | |
3665 | } | |
3666 | ||
3667 | return data; | |
3668 | } | |
3669 | ||
ad23b723 AN |
3670 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx) |
3671 | { | |
3672 | if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym)) | |
3673 | return NULL; | |
3674 | ||
3675 | return (Elf64_Sym *)obj->efile.symbols->d_buf + idx; | |
3676 | } | |
3677 | ||
3678 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx) | |
3679 | { | |
3680 | if (idx >= data->d_size / sizeof(Elf64_Rel)) | |
3681 | return NULL; | |
3682 | ||
3683 | return (Elf64_Rel *)data->d_buf + idx; | |
3684 | } | |
3685 | ||
50e09460 AN |
3686 | static bool is_sec_name_dwarf(const char *name) |
3687 | { | |
3688 | /* approximation, but the actual list is too long */ | |
13d35a0c | 3689 | return str_has_pfx(name, ".debug_"); |
50e09460 AN |
3690 | } |
3691 | ||
ad23b723 | 3692 | static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name) |
50e09460 AN |
3693 | { |
3694 | /* no special handling of .strtab */ | |
3695 | if (hdr->sh_type == SHT_STRTAB) | |
3696 | return true; | |
3697 | ||
3698 | /* ignore .llvm_addrsig section as well */ | |
faf6ed32 | 3699 | if (hdr->sh_type == SHT_LLVM_ADDRSIG) |
50e09460 AN |
3700 | return true; |
3701 | ||
3702 | /* no subprograms will lead to an empty .text section, ignore it */ | |
3703 | if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 && | |
3704 | strcmp(name, ".text") == 0) | |
3705 | return true; | |
3706 | ||
3707 | /* DWARF sections */ | |
3708 | if (is_sec_name_dwarf(name)) | |
3709 | return true; | |
3710 | ||
13d35a0c | 3711 | if (str_has_pfx(name, ".rel")) { |
50e09460 AN |
3712 | name += sizeof(".rel") - 1; |
3713 | /* DWARF section relocations */ | |
3714 | if (is_sec_name_dwarf(name)) | |
3715 | return true; | |
3716 | ||
3717 | /* .BTF and .BTF.ext don't need relocations */ | |
3718 | if (strcmp(name, BTF_ELF_SEC) == 0 || | |
3719 | strcmp(name, BTF_EXT_ELF_SEC) == 0) | |
3720 | return true; | |
3721 | } | |
3722 | ||
3723 | return false; | |
3724 | } | |
3725 | ||
db2b8b06 AN |
3726 | static int cmp_progs(const void *_a, const void *_b) |
3727 | { | |
3728 | const struct bpf_program *a = _a; | |
3729 | const struct bpf_program *b = _b; | |
3730 | ||
3731 | if (a->sec_idx != b->sec_idx) | |
3732 | return a->sec_idx < b->sec_idx ? -1 : 1; | |
3733 | ||
3734 | /* sec_insn_off can't be the same within the section */ | |
3735 | return a->sec_insn_off < b->sec_insn_off ? -1 : 1; | |
3736 | } | |
3737 | ||
0d13bfce | 3738 | static int bpf_object__elf_collect(struct bpf_object *obj) |
29603665 | 3739 | { |
25bbbd7a | 3740 | struct elf_sec_desc *sec_desc; |
29603665 | 3741 | Elf *elf = obj->efile.elf; |
f0187f0b | 3742 | Elf_Data *btf_ext_data = NULL; |
1713d68b | 3743 | Elf_Data *btf_data = NULL; |
666810e8 | 3744 | int idx = 0, err = 0; |
0201c575 AN |
3745 | const char *name; |
3746 | Elf_Data *data; | |
3747 | Elf_Scn *scn; | |
ad23b723 | 3748 | Elf64_Shdr *sh; |
29603665 | 3749 | |
0d6988e1 | 3750 | /* ELF section indices are 0-based, but sec #0 is special "invalid" |
51deedc9 SHY |
3751 | * section. Since section count retrieved by elf_getshdrnum() does |
3752 | * include sec #0, it is already the necessary size of an array to keep | |
3753 | * all the sections. | |
25bbbd7a | 3754 | */ |
51deedc9 SHY |
3755 | if (elf_getshdrnum(obj->efile.elf, &obj->efile.sec_cnt)) { |
3756 | pr_warn("elf: failed to get the number of sections for %s: %s\n", | |
3757 | obj->path, elf_errmsg(-1)); | |
3758 | return -LIBBPF_ERRNO__FORMAT; | |
3759 | } | |
25bbbd7a AN |
3760 | obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs)); |
3761 | if (!obj->efile.secs) | |
3762 | return -ENOMEM; | |
29603665 | 3763 | |
0201c575 AN |
3764 | /* a bunch of ELF parsing functionality depends on processing symbols, |
3765 | * so do the first pass and find the symbol table | |
3766 | */ | |
3767 | scn = NULL; | |
29603665 | 3768 | while ((scn = elf_nextscn(elf, scn)) != NULL) { |
ad23b723 AN |
3769 | sh = elf_sec_hdr(obj, scn); |
3770 | if (!sh) | |
0201c575 AN |
3771 | return -LIBBPF_ERRNO__FORMAT; |
3772 | ||
ad23b723 | 3773 | if (sh->sh_type == SHT_SYMTAB) { |
0201c575 AN |
3774 | if (obj->efile.symbols) { |
3775 | pr_warn("elf: multiple symbol tables in %s\n", obj->path); | |
3776 | return -LIBBPF_ERRNO__FORMAT; | |
3777 | } | |
29603665 | 3778 | |
0201c575 AN |
3779 | data = elf_sec_data(obj, scn); |
3780 | if (!data) | |
3781 | return -LIBBPF_ERRNO__FORMAT; | |
3782 | ||
25bbbd7a AN |
3783 | idx = elf_ndxscn(scn); |
3784 | ||
0201c575 | 3785 | obj->efile.symbols = data; |
25bbbd7a | 3786 | obj->efile.symbols_shndx = idx; |
ad23b723 | 3787 | obj->efile.strtabidx = sh->sh_link; |
0201c575 AN |
3788 | } |
3789 | } | |
3790 | ||
03e601f4 THJ |
3791 | if (!obj->efile.symbols) { |
3792 | pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n", | |
3793 | obj->path); | |
3794 | return -ENOENT; | |
3795 | } | |
3796 | ||
0201c575 AN |
3797 | scn = NULL; |
3798 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
25bbbd7a AN |
3799 | idx = elf_ndxscn(scn); |
3800 | sec_desc = &obj->efile.secs[idx]; | |
88a82120 | 3801 | |
ad23b723 AN |
3802 | sh = elf_sec_hdr(obj, scn); |
3803 | if (!sh) | |
01b29d1d | 3804 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3805 | |
ad23b723 | 3806 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 | 3807 | if (!name) |
01b29d1d | 3808 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3809 | |
ad23b723 | 3810 | if (ignore_elf_section(sh, name)) |
50e09460 AN |
3811 | continue; |
3812 | ||
88a82120 AN |
3813 | data = elf_sec_data(obj, scn); |
3814 | if (!data) | |
01b29d1d | 3815 | return -LIBBPF_ERRNO__FORMAT; |
88a82120 AN |
3816 | |
3817 | pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n", | |
077c066a | 3818 | idx, name, (unsigned long)data->d_size, |
ad23b723 AN |
3819 | (int)sh->sh_link, (unsigned long)sh->sh_flags, |
3820 | (int)sh->sh_type); | |
cb1e5e96 | 3821 | |
1713d68b | 3822 | if (strcmp(name, "license") == 0) { |
88a82120 | 3823 | err = bpf_object__init_license(obj, data->d_buf, data->d_size); |
01b29d1d AN |
3824 | if (err) |
3825 | return err; | |
1713d68b | 3826 | } else if (strcmp(name, "version") == 0) { |
88a82120 | 3827 | err = bpf_object__init_kversion(obj, data->d_buf, data->d_size); |
54b8625c JF |
3828 | if (err) |
3829 | return err; | |
1713d68b | 3830 | } else if (strcmp(name, "maps") == 0) { |
e19db676 AN |
3831 | pr_warn("elf: legacy map definitions in 'maps' section are not supported by libbpf v1.0+\n"); |
3832 | return -ENOTSUP; | |
abd29c93 AN |
3833 | } else if (strcmp(name, MAPS_ELF_SEC) == 0) { |
3834 | obj->efile.btf_maps_shndx = idx; | |
1713d68b | 3835 | } else if (strcmp(name, BTF_ELF_SEC) == 0) { |
62554d52 AN |
3836 | if (sh->sh_type != SHT_PROGBITS) |
3837 | return -LIBBPF_ERRNO__FORMAT; | |
1713d68b | 3838 | btf_data = data; |
2993e051 | 3839 | } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) { |
62554d52 AN |
3840 | if (sh->sh_type != SHT_PROGBITS) |
3841 | return -LIBBPF_ERRNO__FORMAT; | |
f0187f0b | 3842 | btf_ext_data = data; |
ad23b723 | 3843 | } else if (sh->sh_type == SHT_SYMTAB) { |
0201c575 | 3844 | /* already processed during the first pass above */ |
ad23b723 AN |
3845 | } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) { |
3846 | if (sh->sh_flags & SHF_EXECINSTR) { | |
f8c7a4d4 JS |
3847 | if (strcmp(name, ".text") == 0) |
3848 | obj->efile.text_shndx = idx; | |
c1122392 | 3849 | err = bpf_object__add_programs(obj, data, name, idx); |
88a82120 | 3850 | if (err) |
01b29d1d | 3851 | return err; |
aed65917 AN |
3852 | } else if (strcmp(name, DATA_SEC) == 0 || |
3853 | str_has_pfx(name, DATA_SEC ".")) { | |
25bbbd7a AN |
3854 | sec_desc->sec_type = SEC_DATA; |
3855 | sec_desc->shdr = sh; | |
3856 | sec_desc->data = data; | |
aed65917 AN |
3857 | } else if (strcmp(name, RODATA_SEC) == 0 || |
3858 | str_has_pfx(name, RODATA_SEC ".")) { | |
25bbbd7a AN |
3859 | sec_desc->sec_type = SEC_RODATA; |
3860 | sec_desc->shdr = sh; | |
3861 | sec_desc->data = data; | |
240bf8a5 | 3862 | } else if (strcmp(name, STRUCT_OPS_SEC) == 0 || |
5ad0ecbe EZ |
3863 | strcmp(name, STRUCT_OPS_LINK_SEC) == 0 || |
3864 | strcmp(name, "?" STRUCT_OPS_SEC) == 0 || | |
3865 | strcmp(name, "?" STRUCT_OPS_LINK_SEC) == 0) { | |
240bf8a5 EZ |
3866 | sec_desc->sec_type = SEC_ST_OPS; |
3867 | sec_desc->shdr = sh; | |
3868 | sec_desc->data = data; | |
3869 | obj->efile.has_st_ops = true; | |
2e7ba4f8 AN |
3870 | } else if (strcmp(name, ARENA_SEC) == 0) { |
3871 | obj->efile.arena_data = data; | |
3872 | obj->efile.arena_data_shndx = idx; | |
d859900c | 3873 | } else { |
50e09460 AN |
3874 | pr_info("elf: skipping unrecognized data section(%d) %s\n", |
3875 | idx, name); | |
a5b8bd47 | 3876 | } |
ad23b723 | 3877 | } else if (sh->sh_type == SHT_REL) { |
25bbbd7a | 3878 | int targ_sec_idx = sh->sh_info; /* points to other section */ |
e3d91b0c | 3879 | |
b7332d28 AN |
3880 | if (sh->sh_entsize != sizeof(Elf64_Rel) || |
3881 | targ_sec_idx >= obj->efile.sec_cnt) | |
3882 | return -LIBBPF_ERRNO__FORMAT; | |
3883 | ||
e3d91b0c | 3884 | /* Only do relo for section with exec instructions */ |
25bbbd7a | 3885 | if (!section_have_execinstr(obj, targ_sec_idx) && |
646f02ff | 3886 | strcmp(name, ".rel" STRUCT_OPS_SEC) && |
809a69d6 | 3887 | strcmp(name, ".rel" STRUCT_OPS_LINK_SEC) && |
5ad0ecbe EZ |
3888 | strcmp(name, ".rel?" STRUCT_OPS_SEC) && |
3889 | strcmp(name, ".rel?" STRUCT_OPS_LINK_SEC) && | |
646f02ff | 3890 | strcmp(name, ".rel" MAPS_ELF_SEC)) { |
50e09460 | 3891 | pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n", |
25bbbd7a AN |
3892 | idx, name, targ_sec_idx, |
3893 | elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>"); | |
e3d91b0c JDB |
3894 | continue; |
3895 | } | |
b62f06e8 | 3896 | |
25bbbd7a AN |
3897 | sec_desc->sec_type = SEC_RELO; |
3898 | sec_desc->shdr = sh; | |
3899 | sec_desc->data = data; | |
dc79f035 AS |
3900 | } else if (sh->sh_type == SHT_NOBITS && (strcmp(name, BSS_SEC) == 0 || |
3901 | str_has_pfx(name, BSS_SEC "."))) { | |
25bbbd7a AN |
3902 | sec_desc->sec_type = SEC_BSS; |
3903 | sec_desc->shdr = sh; | |
3904 | sec_desc->data = data; | |
077c066a | 3905 | } else { |
2e80be60 | 3906 | pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name, |
ad23b723 | 3907 | (size_t)sh->sh_size); |
bec7d68c | 3908 | } |
29603665 | 3909 | } |
561bbcca | 3910 | |
d3a3aa0c | 3911 | if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) { |
88a82120 | 3912 | pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path); |
f102154d | 3913 | return -LIBBPF_ERRNO__FORMAT; |
77ba9a5b | 3914 | } |
db2b8b06 AN |
3915 | |
3916 | /* sort BPF programs by section name and in-section instruction offset | |
e3ba8e4e KM |
3917 | * for faster search |
3918 | */ | |
2a6a9bf2 AN |
3919 | if (obj->nr_programs) |
3920 | qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs); | |
db2b8b06 | 3921 | |
0d13bfce | 3922 | return bpf_object__init_btf(obj, btf_data, btf_ext_data); |
29603665 WN |
3923 | } |
3924 | ||
ad23b723 | 3925 | static bool sym_is_extern(const Elf64_Sym *sym) |
166750bc | 3926 | { |
ad23b723 | 3927 | int bind = ELF64_ST_BIND(sym->st_info); |
166750bc AN |
3928 | /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */ |
3929 | return sym->st_shndx == SHN_UNDEF && | |
3930 | (bind == STB_GLOBAL || bind == STB_WEAK) && | |
ad23b723 | 3931 | ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE; |
166750bc AN |
3932 | } |
3933 | ||
ad23b723 | 3934 | static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx) |
53eddb5e | 3935 | { |
ad23b723 AN |
3936 | int bind = ELF64_ST_BIND(sym->st_info); |
3937 | int type = ELF64_ST_TYPE(sym->st_info); | |
53eddb5e YS |
3938 | |
3939 | /* in .text section */ | |
3940 | if (sym->st_shndx != text_shndx) | |
3941 | return false; | |
3942 | ||
3943 | /* local function */ | |
3944 | if (bind == STB_LOCAL && type == STT_SECTION) | |
3945 | return true; | |
3946 | ||
3947 | /* global function */ | |
3948 | return bind == STB_GLOBAL && type == STT_FUNC; | |
3949 | } | |
3950 | ||
166750bc AN |
3951 | static int find_extern_btf_id(const struct btf *btf, const char *ext_name) |
3952 | { | |
3953 | const struct btf_type *t; | |
5bd022ec | 3954 | const char *tname; |
166750bc AN |
3955 | int i, n; |
3956 | ||
3957 | if (!btf) | |
3958 | return -ESRCH; | |
3959 | ||
6a886de0 HC |
3960 | n = btf__type_cnt(btf); |
3961 | for (i = 1; i < n; i++) { | |
166750bc AN |
3962 | t = btf__type_by_id(btf, i); |
3963 | ||
5bd022ec | 3964 | if (!btf_is_var(t) && !btf_is_func(t)) |
166750bc AN |
3965 | continue; |
3966 | ||
5bd022ec MKL |
3967 | tname = btf__name_by_offset(btf, t->name_off); |
3968 | if (strcmp(tname, ext_name)) | |
166750bc AN |
3969 | continue; |
3970 | ||
5bd022ec MKL |
3971 | if (btf_is_var(t) && |
3972 | btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN) | |
3973 | return -EINVAL; | |
3974 | ||
3975 | if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN) | |
166750bc AN |
3976 | return -EINVAL; |
3977 | ||
3978 | return i; | |
3979 | } | |
3980 | ||
3981 | return -ENOENT; | |
3982 | } | |
3983 | ||
2e33efe3 AN |
3984 | static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) { |
3985 | const struct btf_var_secinfo *vs; | |
3986 | const struct btf_type *t; | |
3987 | int i, j, n; | |
3988 | ||
3989 | if (!btf) | |
3990 | return -ESRCH; | |
3991 | ||
6a886de0 HC |
3992 | n = btf__type_cnt(btf); |
3993 | for (i = 1; i < n; i++) { | |
2e33efe3 AN |
3994 | t = btf__type_by_id(btf, i); |
3995 | ||
3996 | if (!btf_is_datasec(t)) | |
3997 | continue; | |
3998 | ||
3999 | vs = btf_var_secinfos(t); | |
4000 | for (j = 0; j < btf_vlen(t); j++, vs++) { | |
4001 | if (vs->type == ext_btf_id) | |
4002 | return i; | |
4003 | } | |
4004 | } | |
4005 | ||
4006 | return -ENOENT; | |
4007 | } | |
4008 | ||
4009 | static enum kcfg_type find_kcfg_type(const struct btf *btf, int id, | |
4010 | bool *is_signed) | |
166750bc AN |
4011 | { |
4012 | const struct btf_type *t; | |
4013 | const char *name; | |
4014 | ||
4015 | t = skip_mods_and_typedefs(btf, id, NULL); | |
4016 | name = btf__name_by_offset(btf, t->name_off); | |
4017 | ||
4018 | if (is_signed) | |
4019 | *is_signed = false; | |
4020 | switch (btf_kind(t)) { | |
4021 | case BTF_KIND_INT: { | |
4022 | int enc = btf_int_encoding(t); | |
4023 | ||
4024 | if (enc & BTF_INT_BOOL) | |
2e33efe3 | 4025 | return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN; |
166750bc AN |
4026 | if (is_signed) |
4027 | *is_signed = enc & BTF_INT_SIGNED; | |
4028 | if (t->size == 1) | |
2e33efe3 | 4029 | return KCFG_CHAR; |
166750bc | 4030 | if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1))) |
2e33efe3 AN |
4031 | return KCFG_UNKNOWN; |
4032 | return KCFG_INT; | |
166750bc AN |
4033 | } |
4034 | case BTF_KIND_ENUM: | |
4035 | if (t->size != 4) | |
2e33efe3 | 4036 | return KCFG_UNKNOWN; |
166750bc | 4037 | if (strcmp(name, "libbpf_tristate")) |
2e33efe3 AN |
4038 | return KCFG_UNKNOWN; |
4039 | return KCFG_TRISTATE; | |
f2a62588 YS |
4040 | case BTF_KIND_ENUM64: |
4041 | if (strcmp(name, "libbpf_tristate")) | |
4042 | return KCFG_UNKNOWN; | |
4043 | return KCFG_TRISTATE; | |
166750bc AN |
4044 | case BTF_KIND_ARRAY: |
4045 | if (btf_array(t)->nelems == 0) | |
2e33efe3 AN |
4046 | return KCFG_UNKNOWN; |
4047 | if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR) | |
4048 | return KCFG_UNKNOWN; | |
4049 | return KCFG_CHAR_ARR; | |
166750bc | 4050 | default: |
2e33efe3 | 4051 | return KCFG_UNKNOWN; |
166750bc AN |
4052 | } |
4053 | } | |
4054 | ||
4055 | static int cmp_externs(const void *_a, const void *_b) | |
4056 | { | |
4057 | const struct extern_desc *a = _a; | |
4058 | const struct extern_desc *b = _b; | |
4059 | ||
2e33efe3 AN |
4060 | if (a->type != b->type) |
4061 | return a->type < b->type ? -1 : 1; | |
4062 | ||
4063 | if (a->type == EXT_KCFG) { | |
4064 | /* descending order by alignment requirements */ | |
4065 | if (a->kcfg.align != b->kcfg.align) | |
4066 | return a->kcfg.align > b->kcfg.align ? -1 : 1; | |
4067 | /* ascending order by size, within same alignment class */ | |
4068 | if (a->kcfg.sz != b->kcfg.sz) | |
4069 | return a->kcfg.sz < b->kcfg.sz ? -1 : 1; | |
4070 | } | |
4071 | ||
166750bc AN |
4072 | /* resolve ties by name */ |
4073 | return strcmp(a->name, b->name); | |
4074 | } | |
4075 | ||
1c0c7074 AN |
4076 | static int find_int_btf_id(const struct btf *btf) |
4077 | { | |
4078 | const struct btf_type *t; | |
4079 | int i, n; | |
4080 | ||
6a886de0 HC |
4081 | n = btf__type_cnt(btf); |
4082 | for (i = 1; i < n; i++) { | |
1c0c7074 AN |
4083 | t = btf__type_by_id(btf, i); |
4084 | ||
4085 | if (btf_is_int(t) && btf_int_bits(t) == 32) | |
4086 | return i; | |
4087 | } | |
4088 | ||
4089 | return 0; | |
4090 | } | |
4091 | ||
5bd022ec MKL |
4092 | static int add_dummy_ksym_var(struct btf *btf) |
4093 | { | |
4094 | int i, int_btf_id, sec_btf_id, dummy_var_btf_id; | |
4095 | const struct btf_var_secinfo *vs; | |
4096 | const struct btf_type *sec; | |
4097 | ||
9683e577 IR |
4098 | if (!btf) |
4099 | return 0; | |
4100 | ||
5bd022ec MKL |
4101 | sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC, |
4102 | BTF_KIND_DATASEC); | |
4103 | if (sec_btf_id < 0) | |
4104 | return 0; | |
4105 | ||
4106 | sec = btf__type_by_id(btf, sec_btf_id); | |
4107 | vs = btf_var_secinfos(sec); | |
4108 | for (i = 0; i < btf_vlen(sec); i++, vs++) { | |
4109 | const struct btf_type *vt; | |
4110 | ||
4111 | vt = btf__type_by_id(btf, vs->type); | |
4112 | if (btf_is_func(vt)) | |
4113 | break; | |
4114 | } | |
4115 | ||
4116 | /* No func in ksyms sec. No need to add dummy var. */ | |
4117 | if (i == btf_vlen(sec)) | |
4118 | return 0; | |
4119 | ||
4120 | int_btf_id = find_int_btf_id(btf); | |
4121 | dummy_var_btf_id = btf__add_var(btf, | |
4122 | "dummy_ksym", | |
4123 | BTF_VAR_GLOBAL_ALLOCATED, | |
4124 | int_btf_id); | |
4125 | if (dummy_var_btf_id < 0) | |
4126 | pr_warn("cannot create a dummy_ksym var\n"); | |
4127 | ||
4128 | return dummy_var_btf_id; | |
4129 | } | |
4130 | ||
166750bc AN |
4131 | static int bpf_object__collect_externs(struct bpf_object *obj) |
4132 | { | |
1c0c7074 | 4133 | struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL; |
166750bc AN |
4134 | const struct btf_type *t; |
4135 | struct extern_desc *ext; | |
5bd022ec | 4136 | int i, n, off, dummy_var_btf_id; |
2e33efe3 | 4137 | const char *ext_name, *sec_name; |
5964a223 | 4138 | size_t ext_essent_len; |
166750bc | 4139 | Elf_Scn *scn; |
ad23b723 | 4140 | Elf64_Shdr *sh; |
166750bc AN |
4141 | |
4142 | if (!obj->efile.symbols) | |
4143 | return 0; | |
4144 | ||
88a82120 | 4145 | scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx); |
ad23b723 | 4146 | sh = elf_sec_hdr(obj, scn); |
83390787 | 4147 | if (!sh || sh->sh_entsize != sizeof(Elf64_Sym)) |
166750bc | 4148 | return -LIBBPF_ERRNO__FORMAT; |
166750bc | 4149 | |
5bd022ec MKL |
4150 | dummy_var_btf_id = add_dummy_ksym_var(obj->btf); |
4151 | if (dummy_var_btf_id < 0) | |
4152 | return dummy_var_btf_id; | |
4153 | ||
ad23b723 | 4154 | n = sh->sh_size / sh->sh_entsize; |
166750bc | 4155 | pr_debug("looking for externs among %d symbols...\n", n); |
88a82120 | 4156 | |
166750bc | 4157 | for (i = 0; i < n; i++) { |
ad23b723 | 4158 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
166750bc | 4159 | |
ad23b723 | 4160 | if (!sym) |
166750bc | 4161 | return -LIBBPF_ERRNO__FORMAT; |
ad23b723 | 4162 | if (!sym_is_extern(sym)) |
166750bc | 4163 | continue; |
ad23b723 | 4164 | ext_name = elf_sym_str(obj, sym->st_name); |
166750bc AN |
4165 | if (!ext_name || !ext_name[0]) |
4166 | continue; | |
4167 | ||
4168 | ext = obj->externs; | |
029258d7 | 4169 | ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext)); |
166750bc AN |
4170 | if (!ext) |
4171 | return -ENOMEM; | |
4172 | obj->externs = ext; | |
4173 | ext = &ext[obj->nr_extern]; | |
4174 | memset(ext, 0, sizeof(*ext)); | |
4175 | obj->nr_extern++; | |
4176 | ||
4177 | ext->btf_id = find_extern_btf_id(obj->btf, ext_name); | |
4178 | if (ext->btf_id <= 0) { | |
4179 | pr_warn("failed to find BTF for extern '%s': %d\n", | |
4180 | ext_name, ext->btf_id); | |
4181 | return ext->btf_id; | |
4182 | } | |
4183 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
4184 | ext->name = btf__name_by_offset(obj->btf, t->name_off); | |
4185 | ext->sym_idx = i; | |
ad23b723 | 4186 | ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK; |
2e33efe3 | 4187 | |
5964a223 DM |
4188 | ext_essent_len = bpf_core_essential_name_len(ext->name); |
4189 | ext->essent_name = NULL; | |
4190 | if (ext_essent_len != strlen(ext->name)) { | |
4191 | ext->essent_name = strndup(ext->name, ext_essent_len); | |
4192 | if (!ext->essent_name) | |
4193 | return -ENOMEM; | |
4194 | } | |
4195 | ||
2e33efe3 AN |
4196 | ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id); |
4197 | if (ext->sec_btf_id <= 0) { | |
4198 | pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n", | |
4199 | ext_name, ext->btf_id, ext->sec_btf_id); | |
4200 | return ext->sec_btf_id; | |
166750bc | 4201 | } |
2e33efe3 AN |
4202 | sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id); |
4203 | sec_name = btf__name_by_offset(obj->btf, sec->name_off); | |
4204 | ||
4205 | if (strcmp(sec_name, KCONFIG_SEC) == 0) { | |
5bd022ec MKL |
4206 | if (btf_is_func(t)) { |
4207 | pr_warn("extern function %s is unsupported under %s section\n", | |
4208 | ext->name, KCONFIG_SEC); | |
4209 | return -ENOTSUP; | |
4210 | } | |
2e33efe3 AN |
4211 | kcfg_sec = sec; |
4212 | ext->type = EXT_KCFG; | |
4213 | ext->kcfg.sz = btf__resolve_size(obj->btf, t->type); | |
4214 | if (ext->kcfg.sz <= 0) { | |
4215 | pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n", | |
4216 | ext_name, ext->kcfg.sz); | |
4217 | return ext->kcfg.sz; | |
4218 | } | |
4219 | ext->kcfg.align = btf__align_of(obj->btf, t->type); | |
4220 | if (ext->kcfg.align <= 0) { | |
4221 | pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n", | |
4222 | ext_name, ext->kcfg.align); | |
4223 | return -EINVAL; | |
4224 | } | |
4225 | ext->kcfg.type = find_kcfg_type(obj->btf, t->type, | |
e3ba8e4e | 4226 | &ext->kcfg.is_signed); |
2e33efe3 | 4227 | if (ext->kcfg.type == KCFG_UNKNOWN) { |
55d00c37 | 4228 | pr_warn("extern (kcfg) '%s': type is unsupported\n", ext_name); |
2e33efe3 AN |
4229 | return -ENOTSUP; |
4230 | } | |
1c0c7074 | 4231 | } else if (strcmp(sec_name, KSYMS_SEC) == 0) { |
1c0c7074 AN |
4232 | ksym_sec = sec; |
4233 | ext->type = EXT_KSYM; | |
d370bbe1 HL |
4234 | skip_mods_and_typedefs(obj->btf, t->type, |
4235 | &ext->ksym.type_id); | |
2e33efe3 AN |
4236 | } else { |
4237 | pr_warn("unrecognized extern section '%s'\n", sec_name); | |
166750bc AN |
4238 | return -ENOTSUP; |
4239 | } | |
4240 | } | |
4241 | pr_debug("collected %d externs total\n", obj->nr_extern); | |
4242 | ||
4243 | if (!obj->nr_extern) | |
4244 | return 0; | |
4245 | ||
2e33efe3 | 4246 | /* sort externs by type, for kcfg ones also by (align, size, name) */ |
166750bc | 4247 | qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs); |
166750bc | 4248 | |
1c0c7074 AN |
4249 | /* for .ksyms section, we need to turn all externs into allocated |
4250 | * variables in BTF to pass kernel verification; we do this by | |
4251 | * pretending that each extern is a 8-byte variable | |
4252 | */ | |
4253 | if (ksym_sec) { | |
4254 | /* find existing 4-byte integer type in BTF to use for fake | |
4255 | * extern variables in DATASEC | |
4256 | */ | |
4257 | int int_btf_id = find_int_btf_id(obj->btf); | |
5bd022ec MKL |
4258 | /* For extern function, a dummy_var added earlier |
4259 | * will be used to replace the vs->type and | |
4260 | * its name string will be used to refill | |
4261 | * the missing param's name. | |
4262 | */ | |
4263 | const struct btf_type *dummy_var; | |
1c0c7074 | 4264 | |
5bd022ec | 4265 | dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id); |
1c0c7074 AN |
4266 | for (i = 0; i < obj->nr_extern; i++) { |
4267 | ext = &obj->externs[i]; | |
4268 | if (ext->type != EXT_KSYM) | |
4269 | continue; | |
4270 | pr_debug("extern (ksym) #%d: symbol %d, name %s\n", | |
4271 | i, ext->sym_idx, ext->name); | |
4272 | } | |
4273 | ||
4274 | sec = ksym_sec; | |
4275 | n = btf_vlen(sec); | |
4276 | for (i = 0, off = 0; i < n; i++, off += sizeof(int)) { | |
4277 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
4278 | struct btf_type *vt; | |
4279 | ||
4280 | vt = (void *)btf__type_by_id(obj->btf, vs->type); | |
4281 | ext_name = btf__name_by_offset(obj->btf, vt->name_off); | |
4282 | ext = find_extern_by_name(obj, ext_name); | |
4283 | if (!ext) { | |
5bd022ec MKL |
4284 | pr_warn("failed to find extern definition for BTF %s '%s'\n", |
4285 | btf_kind_str(vt), ext_name); | |
1c0c7074 AN |
4286 | return -ESRCH; |
4287 | } | |
5bd022ec MKL |
4288 | if (btf_is_func(vt)) { |
4289 | const struct btf_type *func_proto; | |
4290 | struct btf_param *param; | |
4291 | int j; | |
4292 | ||
4293 | func_proto = btf__type_by_id(obj->btf, | |
4294 | vt->type); | |
4295 | param = btf_params(func_proto); | |
4296 | /* Reuse the dummy_var string if the | |
4297 | * func proto does not have param name. | |
4298 | */ | |
4299 | for (j = 0; j < btf_vlen(func_proto); j++) | |
4300 | if (param[j].type && !param[j].name_off) | |
4301 | param[j].name_off = | |
4302 | dummy_var->name_off; | |
4303 | vs->type = dummy_var_btf_id; | |
4304 | vt->info &= ~0xffff; | |
4305 | vt->info |= BTF_FUNC_GLOBAL; | |
4306 | } else { | |
4307 | btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
4308 | vt->type = int_btf_id; | |
4309 | } | |
1c0c7074 AN |
4310 | vs->offset = off; |
4311 | vs->size = sizeof(int); | |
4312 | } | |
4313 | sec->size = off; | |
4314 | } | |
4315 | ||
2e33efe3 AN |
4316 | if (kcfg_sec) { |
4317 | sec = kcfg_sec; | |
4318 | /* for kcfg externs calculate their offsets within a .kconfig map */ | |
4319 | off = 0; | |
4320 | for (i = 0; i < obj->nr_extern; i++) { | |
4321 | ext = &obj->externs[i]; | |
4322 | if (ext->type != EXT_KCFG) | |
4323 | continue; | |
166750bc | 4324 | |
2e33efe3 AN |
4325 | ext->kcfg.data_off = roundup(off, ext->kcfg.align); |
4326 | off = ext->kcfg.data_off + ext->kcfg.sz; | |
1c0c7074 | 4327 | pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n", |
2e33efe3 AN |
4328 | i, ext->sym_idx, ext->kcfg.data_off, ext->name); |
4329 | } | |
4330 | sec->size = off; | |
4331 | n = btf_vlen(sec); | |
4332 | for (i = 0; i < n; i++) { | |
4333 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
4334 | ||
4335 | t = btf__type_by_id(obj->btf, vs->type); | |
4336 | ext_name = btf__name_by_offset(obj->btf, t->name_off); | |
4337 | ext = find_extern_by_name(obj, ext_name); | |
4338 | if (!ext) { | |
4339 | pr_warn("failed to find extern definition for BTF var '%s'\n", | |
4340 | ext_name); | |
4341 | return -ESRCH; | |
4342 | } | |
4343 | btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
4344 | vs->offset = ext->kcfg.data_off; | |
166750bc | 4345 | } |
166750bc | 4346 | } |
166750bc AN |
4347 | return 0; |
4348 | } | |
4349 | ||
bd054102 AN |
4350 | static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog) |
4351 | { | |
197afc63 | 4352 | return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1; |
c3c55696 AN |
4353 | } |
4354 | ||
01af3bf0 AN |
4355 | struct bpf_program * |
4356 | bpf_object__find_program_by_name(const struct bpf_object *obj, | |
4357 | const char *name) | |
4358 | { | |
4359 | struct bpf_program *prog; | |
4360 | ||
4361 | bpf_object__for_each_program(prog, obj) { | |
c3c55696 AN |
4362 | if (prog_is_subprog(obj, prog)) |
4363 | continue; | |
01af3bf0 AN |
4364 | if (!strcmp(prog->name, name)) |
4365 | return prog; | |
4366 | } | |
e9fc3ce9 | 4367 | return errno = ENOENT, NULL; |
01af3bf0 AN |
4368 | } |
4369 | ||
d859900c DB |
4370 | static bool bpf_object__shndx_is_data(const struct bpf_object *obj, |
4371 | int shndx) | |
4372 | { | |
25bbbd7a AN |
4373 | switch (obj->efile.secs[shndx].sec_type) { |
4374 | case SEC_BSS: | |
4375 | case SEC_DATA: | |
4376 | case SEC_RODATA: | |
4377 | return true; | |
4378 | default: | |
4379 | return false; | |
4380 | } | |
d859900c DB |
4381 | } |
4382 | ||
4383 | static bool bpf_object__shndx_is_maps(const struct bpf_object *obj, | |
4384 | int shndx) | |
4385 | { | |
e19db676 | 4386 | return shndx == obj->efile.btf_maps_shndx; |
d859900c DB |
4387 | } |
4388 | ||
d859900c DB |
4389 | static enum libbpf_map_type |
4390 | bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx) | |
4391 | { | |
25bbbd7a AN |
4392 | if (shndx == obj->efile.symbols_shndx) |
4393 | return LIBBPF_MAP_KCONFIG; | |
4394 | ||
4395 | switch (obj->efile.secs[shndx].sec_type) { | |
4396 | case SEC_BSS: | |
d859900c | 4397 | return LIBBPF_MAP_BSS; |
25bbbd7a AN |
4398 | case SEC_DATA: |
4399 | return LIBBPF_MAP_DATA; | |
4400 | case SEC_RODATA: | |
d859900c | 4401 | return LIBBPF_MAP_RODATA; |
25bbbd7a | 4402 | default: |
d859900c | 4403 | return LIBBPF_MAP_UNSPEC; |
25bbbd7a | 4404 | } |
d859900c DB |
4405 | } |
4406 | ||
1f8e2bcb AN |
4407 | static int bpf_program__record_reloc(struct bpf_program *prog, |
4408 | struct reloc_desc *reloc_desc, | |
9c0f8cbd | 4409 | __u32 insn_idx, const char *sym_name, |
ad23b723 | 4410 | const Elf64_Sym *sym, const Elf64_Rel *rel) |
1f8e2bcb AN |
4411 | { |
4412 | struct bpf_insn *insn = &prog->insns[insn_idx]; | |
4413 | size_t map_idx, nr_maps = prog->obj->nr_maps; | |
4414 | struct bpf_object *obj = prog->obj; | |
4415 | __u32 shdr_idx = sym->st_shndx; | |
4416 | enum libbpf_map_type type; | |
9c0f8cbd | 4417 | const char *sym_sec_name; |
1f8e2bcb AN |
4418 | struct bpf_map *map; |
4419 | ||
aa0b8d43 | 4420 | if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) { |
9c0f8cbd AN |
4421 | pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n", |
4422 | prog->name, sym_name, insn_idx, insn->code); | |
1f8e2bcb AN |
4423 | return -LIBBPF_ERRNO__RELOC; |
4424 | } | |
166750bc AN |
4425 | |
4426 | if (sym_is_extern(sym)) { | |
ad23b723 | 4427 | int sym_idx = ELF64_R_SYM(rel->r_info); |
166750bc AN |
4428 | int i, n = obj->nr_extern; |
4429 | struct extern_desc *ext; | |
4430 | ||
4431 | for (i = 0; i < n; i++) { | |
4432 | ext = &obj->externs[i]; | |
4433 | if (ext->sym_idx == sym_idx) | |
4434 | break; | |
4435 | } | |
4436 | if (i >= n) { | |
9c0f8cbd AN |
4437 | pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n", |
4438 | prog->name, sym_name, sym_idx); | |
166750bc AN |
4439 | return -LIBBPF_ERRNO__RELOC; |
4440 | } | |
9c0f8cbd AN |
4441 | pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n", |
4442 | prog->name, i, ext->name, ext->sym_idx, insn_idx); | |
5bd022ec | 4443 | if (insn->code == (BPF_JMP | BPF_CALL)) |
a18f7214 | 4444 | reloc_desc->type = RELO_EXTERN_CALL; |
5bd022ec | 4445 | else |
a18f7214 | 4446 | reloc_desc->type = RELO_EXTERN_LD64; |
166750bc | 4447 | reloc_desc->insn_idx = insn_idx; |
3055ddd6 | 4448 | reloc_desc->ext_idx = i; |
166750bc AN |
4449 | return 0; |
4450 | } | |
4451 | ||
aa0b8d43 MKL |
4452 | /* sub-program call relocation */ |
4453 | if (is_call_insn(insn)) { | |
4454 | if (insn->src_reg != BPF_PSEUDO_CALL) { | |
4455 | pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name); | |
4456 | return -LIBBPF_ERRNO__RELOC; | |
4457 | } | |
4458 | /* text_shndx can be 0, if no default "main" program exists */ | |
4459 | if (!shdr_idx || shdr_idx != obj->efile.text_shndx) { | |
4460 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); | |
4461 | pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n", | |
4462 | prog->name, sym_name, sym_sec_name); | |
4463 | return -LIBBPF_ERRNO__RELOC; | |
4464 | } | |
4465 | if (sym->st_value % BPF_INSN_SZ) { | |
4466 | pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n", | |
4467 | prog->name, sym_name, (size_t)sym->st_value); | |
4468 | return -LIBBPF_ERRNO__RELOC; | |
4469 | } | |
4470 | reloc_desc->type = RELO_CALL; | |
4471 | reloc_desc->insn_idx = insn_idx; | |
4472 | reloc_desc->sym_off = sym->st_value; | |
4473 | return 0; | |
4474 | } | |
4475 | ||
1f8e2bcb | 4476 | if (!shdr_idx || shdr_idx >= SHN_LORESERVE) { |
9c0f8cbd AN |
4477 | pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n", |
4478 | prog->name, sym_name, shdr_idx); | |
1f8e2bcb AN |
4479 | return -LIBBPF_ERRNO__RELOC; |
4480 | } | |
4481 | ||
53eddb5e YS |
4482 | /* loading subprog addresses */ |
4483 | if (sym_is_subprog(sym, obj->efile.text_shndx)) { | |
4484 | /* global_func: sym->st_value = offset in the section, insn->imm = 0. | |
4485 | * local_func: sym->st_value = 0, insn->imm = offset in the section. | |
4486 | */ | |
4487 | if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) { | |
4488 | pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n", | |
4489 | prog->name, sym_name, (size_t)sym->st_value, insn->imm); | |
4490 | return -LIBBPF_ERRNO__RELOC; | |
4491 | } | |
4492 | ||
4493 | reloc_desc->type = RELO_SUBPROG_ADDR; | |
4494 | reloc_desc->insn_idx = insn_idx; | |
4495 | reloc_desc->sym_off = sym->st_value; | |
4496 | return 0; | |
4497 | } | |
4498 | ||
1f8e2bcb | 4499 | type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx); |
9c0f8cbd | 4500 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); |
1f8e2bcb | 4501 | |
2e7ba4f8 AN |
4502 | /* arena data relocation */ |
4503 | if (shdr_idx == obj->efile.arena_data_shndx) { | |
4504 | reloc_desc->type = RELO_DATA; | |
4505 | reloc_desc->insn_idx = insn_idx; | |
4506 | reloc_desc->map_idx = obj->arena_map - obj->maps; | |
4507 | reloc_desc->sym_off = sym->st_value; | |
4508 | return 0; | |
4509 | } | |
4510 | ||
1f8e2bcb AN |
4511 | /* generic map reference relocation */ |
4512 | if (type == LIBBPF_MAP_UNSPEC) { | |
4513 | if (!bpf_object__shndx_is_maps(obj, shdr_idx)) { | |
9c0f8cbd AN |
4514 | pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n", |
4515 | prog->name, sym_name, sym_sec_name); | |
1f8e2bcb AN |
4516 | return -LIBBPF_ERRNO__RELOC; |
4517 | } | |
4518 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { | |
4519 | map = &obj->maps[map_idx]; | |
4520 | if (map->libbpf_type != type || | |
4521 | map->sec_idx != sym->st_shndx || | |
4522 | map->sec_offset != sym->st_value) | |
4523 | continue; | |
9c0f8cbd AN |
4524 | pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n", |
4525 | prog->name, map_idx, map->name, map->sec_idx, | |
1f8e2bcb AN |
4526 | map->sec_offset, insn_idx); |
4527 | break; | |
4528 | } | |
4529 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
4530 | pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n", |
4531 | prog->name, sym_sec_name, (size_t)sym->st_value); | |
1f8e2bcb AN |
4532 | return -LIBBPF_ERRNO__RELOC; |
4533 | } | |
4534 | reloc_desc->type = RELO_LD64; | |
4535 | reloc_desc->insn_idx = insn_idx; | |
4536 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 4537 | reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */ |
1f8e2bcb AN |
4538 | return 0; |
4539 | } | |
4540 | ||
4541 | /* global data map relocation */ | |
4542 | if (!bpf_object__shndx_is_data(obj, shdr_idx)) { | |
9c0f8cbd AN |
4543 | pr_warn("prog '%s': bad data relo against section '%s'\n", |
4544 | prog->name, sym_sec_name); | |
1f8e2bcb | 4545 | return -LIBBPF_ERRNO__RELOC; |
1f8e2bcb | 4546 | } |
1f8e2bcb AN |
4547 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { |
4548 | map = &obj->maps[map_idx]; | |
25bbbd7a | 4549 | if (map->libbpf_type != type || map->sec_idx != sym->st_shndx) |
1f8e2bcb | 4550 | continue; |
9c0f8cbd AN |
4551 | pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n", |
4552 | prog->name, map_idx, map->name, map->sec_idx, | |
4553 | map->sec_offset, insn_idx); | |
1f8e2bcb AN |
4554 | break; |
4555 | } | |
4556 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
4557 | pr_warn("prog '%s': data relo failed to find map for section '%s'\n", |
4558 | prog->name, sym_sec_name); | |
1f8e2bcb AN |
4559 | return -LIBBPF_ERRNO__RELOC; |
4560 | } | |
4561 | ||
4562 | reloc_desc->type = RELO_DATA; | |
4563 | reloc_desc->insn_idx = insn_idx; | |
4564 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 4565 | reloc_desc->sym_off = sym->st_value; |
1f8e2bcb AN |
4566 | return 0; |
4567 | } | |
4568 | ||
db2b8b06 AN |
4569 | static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx) |
4570 | { | |
4571 | return insn_idx >= prog->sec_insn_off && | |
4572 | insn_idx < prog->sec_insn_off + prog->sec_insn_cnt; | |
4573 | } | |
4574 | ||
4575 | static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj, | |
4576 | size_t sec_idx, size_t insn_idx) | |
4577 | { | |
4578 | int l = 0, r = obj->nr_programs - 1, m; | |
4579 | struct bpf_program *prog; | |
4580 | ||
d0d382f9 SHY |
4581 | if (!obj->nr_programs) |
4582 | return NULL; | |
4583 | ||
db2b8b06 AN |
4584 | while (l < r) { |
4585 | m = l + (r - l + 1) / 2; | |
4586 | prog = &obj->programs[m]; | |
4587 | ||
4588 | if (prog->sec_idx < sec_idx || | |
4589 | (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx)) | |
4590 | l = m; | |
4591 | else | |
4592 | r = m - 1; | |
4593 | } | |
4594 | /* matching program could be at index l, but it still might be the | |
4595 | * wrong one, so we need to double check conditions for the last time | |
4596 | */ | |
4597 | prog = &obj->programs[l]; | |
4598 | if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx)) | |
4599 | return prog; | |
4600 | return NULL; | |
4601 | } | |
4602 | ||
34090915 | 4603 | static int |
ad23b723 | 4604 | bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data) |
34090915 | 4605 | { |
9c0f8cbd | 4606 | const char *relo_sec_name, *sec_name; |
b7332d28 | 4607 | size_t sec_idx = shdr->sh_info, sym_idx; |
c3c55696 AN |
4608 | struct bpf_program *prog; |
4609 | struct reloc_desc *relos; | |
1f8e2bcb | 4610 | int err, i, nrels; |
c3c55696 AN |
4611 | const char *sym_name; |
4612 | __u32 insn_idx; | |
6245947c AN |
4613 | Elf_Scn *scn; |
4614 | Elf_Data *scn_data; | |
ad23b723 AN |
4615 | Elf64_Sym *sym; |
4616 | Elf64_Rel *rel; | |
34090915 | 4617 | |
b7332d28 AN |
4618 | if (sec_idx >= obj->efile.sec_cnt) |
4619 | return -EINVAL; | |
4620 | ||
6245947c AN |
4621 | scn = elf_sec_by_idx(obj, sec_idx); |
4622 | scn_data = elf_sec_data(obj, scn); | |
fc3a5534 MZ |
4623 | if (!scn_data) |
4624 | return -LIBBPF_ERRNO__FORMAT; | |
6245947c | 4625 | |
9c0f8cbd | 4626 | relo_sec_name = elf_sec_str(obj, shdr->sh_name); |
6245947c | 4627 | sec_name = elf_sec_name(obj, scn); |
9c0f8cbd AN |
4628 | if (!relo_sec_name || !sec_name) |
4629 | return -EINVAL; | |
4630 | ||
4631 | pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n", | |
4632 | relo_sec_name, sec_idx, sec_name); | |
34090915 WN |
4633 | nrels = shdr->sh_size / shdr->sh_entsize; |
4634 | ||
34090915 | 4635 | for (i = 0; i < nrels; i++) { |
ad23b723 AN |
4636 | rel = elf_rel_by_idx(data, i); |
4637 | if (!rel) { | |
9c0f8cbd | 4638 | pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i); |
6371ca3b | 4639 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4640 | } |
ad23b723 | 4641 | |
b7332d28 AN |
4642 | sym_idx = ELF64_R_SYM(rel->r_info); |
4643 | sym = elf_sym_by_idx(obj, sym_idx); | |
ad23b723 | 4644 | if (!sym) { |
b7332d28 AN |
4645 | pr_warn("sec '%s': symbol #%zu not found for relo #%d\n", |
4646 | relo_sec_name, sym_idx, i); | |
4647 | return -LIBBPF_ERRNO__FORMAT; | |
4648 | } | |
4649 | ||
4650 | if (sym->st_shndx >= obj->efile.sec_cnt) { | |
4651 | pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n", | |
4652 | relo_sec_name, sym_idx, (size_t)sym->st_shndx, i); | |
6371ca3b | 4653 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4654 | } |
6245947c | 4655 | |
ad23b723 | 4656 | if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) { |
9c0f8cbd | 4657 | pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n", |
b7332d28 | 4658 | relo_sec_name, (size_t)rel->r_offset, i); |
1f8e2bcb | 4659 | return -LIBBPF_ERRNO__FORMAT; |
9c0f8cbd | 4660 | } |
d859900c | 4661 | |
ad23b723 | 4662 | insn_idx = rel->r_offset / BPF_INSN_SZ; |
c3c55696 AN |
4663 | /* relocations against static functions are recorded as |
4664 | * relocations against the section that contains a function; | |
4665 | * in such case, symbol will be STT_SECTION and sym.st_name | |
4666 | * will point to empty string (0), so fetch section name | |
4667 | * instead | |
4668 | */ | |
ad23b723 AN |
4669 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0) |
4670 | sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx)); | |
c3c55696 | 4671 | else |
ad23b723 | 4672 | sym_name = elf_sym_str(obj, sym->st_name); |
c3c55696 | 4673 | sym_name = sym_name ?: "<?"; |
d859900c | 4674 | |
9c0f8cbd AN |
4675 | pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n", |
4676 | relo_sec_name, i, insn_idx, sym_name); | |
666810e8 | 4677 | |
c3c55696 AN |
4678 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); |
4679 | if (!prog) { | |
6245947c | 4680 | pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n", |
c3c55696 | 4681 | relo_sec_name, i, sec_name, insn_idx); |
6245947c | 4682 | continue; |
c3c55696 AN |
4683 | } |
4684 | ||
4685 | relos = libbpf_reallocarray(prog->reloc_desc, | |
4686 | prog->nr_reloc + 1, sizeof(*relos)); | |
4687 | if (!relos) | |
4688 | return -ENOMEM; | |
4689 | prog->reloc_desc = relos; | |
4690 | ||
4691 | /* adjust insn_idx to local BPF program frame of reference */ | |
4692 | insn_idx -= prog->sec_insn_off; | |
4693 | err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc], | |
ad23b723 | 4694 | insn_idx, sym_name, sym, rel); |
1f8e2bcb AN |
4695 | if (err) |
4696 | return err; | |
c3c55696 AN |
4697 | |
4698 | prog->nr_reloc++; | |
34090915 WN |
4699 | } |
4700 | return 0; | |
4701 | } | |
4702 | ||
4fcac46c | 4703 | static int map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map) |
8a138aed | 4704 | { |
aaf6886d | 4705 | int id; |
8a138aed | 4706 | |
a8fee962 AN |
4707 | if (!obj->btf) |
4708 | return -ENOENT; | |
4709 | ||
590a0088 MKL |
4710 | /* if it's BTF-defined map, we don't need to search for type IDs. |
4711 | * For struct_ops map, it does not need btf_key_type_id and | |
4712 | * btf_value_type_id. | |
4713 | */ | |
aaf6886d | 4714 | if (map->sec_idx == obj->efile.btf_maps_shndx || bpf_map__is_struct_ops(map)) |
abd29c93 AN |
4715 | return 0; |
4716 | ||
aaf6886d AN |
4717 | /* |
4718 | * LLVM annotates global data differently in BTF, that is, | |
4719 | * only as '.data', '.bss' or '.rodata'. | |
4720 | */ | |
4721 | if (!bpf_map__is_internal(map)) | |
4722 | return -ENOENT; | |
4723 | ||
4724 | id = btf__find_by_name(obj->btf, map->real_name); | |
4725 | if (id < 0) | |
4726 | return id; | |
8a138aed | 4727 | |
aaf6886d AN |
4728 | map->btf_key_type_id = 0; |
4729 | map->btf_value_type_id = id; | |
8a138aed MKL |
4730 | return 0; |
4731 | } | |
4732 | ||
97eb3138 MP |
4733 | static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info) |
4734 | { | |
4735 | char file[PATH_MAX], buff[4096]; | |
4736 | FILE *fp; | |
4737 | __u32 val; | |
4738 | int err; | |
4739 | ||
4740 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
4741 | memset(info, 0, sizeof(*info)); | |
4742 | ||
59842c54 | 4743 | fp = fopen(file, "re"); |
97eb3138 MP |
4744 | if (!fp) { |
4745 | err = -errno; | |
4746 | pr_warn("failed to open %s: %d. No procfs support?\n", file, | |
4747 | err); | |
4748 | return err; | |
4749 | } | |
4750 | ||
4751 | while (fgets(buff, sizeof(buff), fp)) { | |
4752 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
4753 | info->type = val; | |
4754 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) | |
4755 | info->key_size = val; | |
4756 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) | |
4757 | info->value_size = val; | |
4758 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) | |
4759 | info->max_entries = val; | |
4760 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) | |
4761 | info->map_flags = val; | |
4762 | } | |
4763 | ||
4764 | fclose(fp); | |
4765 | ||
4766 | return 0; | |
4767 | } | |
4768 | ||
ec41817b AN |
4769 | bool bpf_map__autocreate(const struct bpf_map *map) |
4770 | { | |
4771 | return map->autocreate; | |
4772 | } | |
4773 | ||
4774 | int bpf_map__set_autocreate(struct bpf_map *map, bool autocreate) | |
4775 | { | |
4776 | if (map->obj->loaded) | |
4777 | return libbpf_err(-EBUSY); | |
4778 | ||
4779 | map->autocreate = autocreate; | |
4780 | return 0; | |
4781 | } | |
4782 | ||
26736eb9 JK |
4783 | int bpf_map__reuse_fd(struct bpf_map *map, int fd) |
4784 | { | |
813847a3 | 4785 | struct bpf_map_info info; |
bf3f0037 | 4786 | __u32 len = sizeof(info), name_len; |
26736eb9 JK |
4787 | int new_fd, err; |
4788 | char *new_name; | |
4789 | ||
813847a3 | 4790 | memset(&info, 0, len); |
629dfc66 | 4791 | err = bpf_map_get_info_by_fd(fd, &info, &len); |
97eb3138 MP |
4792 | if (err && errno == EINVAL) |
4793 | err = bpf_get_map_info_from_fdinfo(fd, &info); | |
26736eb9 | 4794 | if (err) |
e9fc3ce9 | 4795 | return libbpf_err(err); |
26736eb9 | 4796 | |
bf3f0037 AW |
4797 | name_len = strlen(info.name); |
4798 | if (name_len == BPF_OBJ_NAME_LEN - 1 && strncmp(map->name, info.name, name_len) == 0) | |
4799 | new_name = strdup(map->name); | |
4800 | else | |
4801 | new_name = strdup(info.name); | |
4802 | ||
26736eb9 | 4803 | if (!new_name) |
e9fc3ce9 | 4804 | return libbpf_err(-errno); |
26736eb9 | 4805 | |
4aadd292 AN |
4806 | /* |
4807 | * Like dup(), but make sure new FD is >= 3 and has O_CLOEXEC set. | |
4808 | * This is similar to what we do in ensure_good_fd(), but without | |
4809 | * closing original FD. | |
4810 | */ | |
4811 | new_fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); | |
d1b4574a THJ |
4812 | if (new_fd < 0) { |
4813 | err = -errno; | |
26736eb9 | 4814 | goto err_free_new_name; |
d1b4574a | 4815 | } |
26736eb9 | 4816 | |
dac645b9 AN |
4817 | err = reuse_fd(map->fd, new_fd); |
4818 | if (err) | |
4819 | goto err_free_new_name; | |
4820 | ||
26736eb9 JK |
4821 | free(map->name); |
4822 | ||
26736eb9 JK |
4823 | map->name = new_name; |
4824 | map->def.type = info.type; | |
4825 | map->def.key_size = info.key_size; | |
4826 | map->def.value_size = info.value_size; | |
4827 | map->def.max_entries = info.max_entries; | |
4828 | map->def.map_flags = info.map_flags; | |
4829 | map->btf_key_type_id = info.btf_key_type_id; | |
4830 | map->btf_value_type_id = info.btf_value_type_id; | |
ec6d5f47 | 4831 | map->reused = true; |
47512102 | 4832 | map->map_extra = info.map_extra; |
26736eb9 JK |
4833 | |
4834 | return 0; | |
4835 | ||
26736eb9 JK |
4836 | err_free_new_name: |
4837 | free(new_name); | |
e9fc3ce9 | 4838 | return libbpf_err(err); |
26736eb9 JK |
4839 | } |
4840 | ||
1bdb6c9a | 4841 | __u32 bpf_map__max_entries(const struct bpf_map *map) |
1a11a4c7 | 4842 | { |
1bdb6c9a AN |
4843 | return map->def.max_entries; |
4844 | } | |
1a11a4c7 | 4845 | |
b3278099 AN |
4846 | struct bpf_map *bpf_map__inner_map(struct bpf_map *map) |
4847 | { | |
4848 | if (!bpf_map_type__is_map_in_map(map->def.type)) | |
e9fc3ce9 | 4849 | return errno = EINVAL, NULL; |
b3278099 AN |
4850 | |
4851 | return map->inner_map; | |
4852 | } | |
4853 | ||
1bdb6c9a AN |
4854 | int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries) |
4855 | { | |
597fbc46 | 4856 | if (map->obj->loaded) |
e9fc3ce9 | 4857 | return libbpf_err(-EBUSY); |
597fbc46 | 4858 | |
1a11a4c7 | 4859 | map->def.max_entries = max_entries; |
597fbc46 AN |
4860 | |
4861 | /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */ | |
b66ccae0 | 4862 | if (map_is_ringbuf(map)) |
597fbc46 AN |
4863 | map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries); |
4864 | ||
1a11a4c7 AI |
4865 | return 0; |
4866 | } | |
4867 | ||
6b434b61 AN |
4868 | static int bpf_object_prepare_token(struct bpf_object *obj) |
4869 | { | |
4870 | const char *bpffs_path; | |
4871 | int bpffs_fd = -1, token_fd, err; | |
4872 | bool mandatory; | |
4873 | enum libbpf_print_level level; | |
4874 | ||
4875 | /* token is explicitly prevented */ | |
4876 | if (obj->token_path && obj->token_path[0] == '\0') { | |
4877 | pr_debug("object '%s': token is prevented, skipping...\n", obj->name); | |
4878 | return 0; | |
4879 | } | |
4880 | ||
4881 | mandatory = obj->token_path != NULL; | |
4882 | level = mandatory ? LIBBPF_WARN : LIBBPF_DEBUG; | |
4883 | ||
4884 | bpffs_path = obj->token_path ?: BPF_FS_DEFAULT_PATH; | |
4885 | bpffs_fd = open(bpffs_path, O_DIRECTORY, O_RDWR); | |
4886 | if (bpffs_fd < 0) { | |
4887 | err = -errno; | |
4888 | __pr(level, "object '%s': failed (%d) to open BPF FS mount at '%s'%s\n", | |
4889 | obj->name, err, bpffs_path, | |
4890 | mandatory ? "" : ", skipping optional step..."); | |
4891 | return mandatory ? err : 0; | |
4892 | } | |
4893 | ||
4894 | token_fd = bpf_token_create(bpffs_fd, 0); | |
4895 | close(bpffs_fd); | |
4896 | if (token_fd < 0) { | |
4897 | if (!mandatory && token_fd == -ENOENT) { | |
4898 | pr_debug("object '%s': BPF FS at '%s' doesn't have BPF token delegation set up, skipping...\n", | |
4899 | obj->name, bpffs_path); | |
4900 | return 0; | |
4901 | } | |
4902 | __pr(level, "object '%s': failed (%d) to create BPF token from '%s'%s\n", | |
4903 | obj->name, token_fd, bpffs_path, | |
4904 | mandatory ? "" : ", skipping optional step..."); | |
4905 | return mandatory ? token_fd : 0; | |
4906 | } | |
4907 | ||
4908 | obj->feat_cache = calloc(1, sizeof(*obj->feat_cache)); | |
4909 | if (!obj->feat_cache) { | |
4910 | close(token_fd); | |
4911 | return -ENOMEM; | |
4912 | } | |
4913 | ||
4914 | obj->token_fd = token_fd; | |
4915 | obj->feat_cache->token_fd = token_fd; | |
4916 | ||
4917 | return 0; | |
4918 | } | |
4919 | ||
47eff617 | 4920 | static int |
fd9eef1a | 4921 | bpf_object__probe_loading(struct bpf_object *obj) |
47eff617 | 4922 | { |
47eff617 SF |
4923 | char *cp, errmsg[STRERR_BUFSIZE]; |
4924 | struct bpf_insn insns[] = { | |
4925 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4926 | BPF_EXIT_INSN(), | |
4927 | }; | |
e32660ac | 4928 | int ret, insn_cnt = ARRAY_SIZE(insns); |
6b434b61 AN |
4929 | LIBBPF_OPTS(bpf_prog_load_opts, opts, |
4930 | .token_fd = obj->token_fd, | |
4931 | .prog_flags = obj->token_fd ? BPF_F_TOKEN_FD : 0, | |
4932 | ); | |
47eff617 | 4933 | |
f9bceaa5 SF |
4934 | if (obj->gen_loader) |
4935 | return 0; | |
4936 | ||
e542f2c4 AN |
4937 | ret = bump_rlimit_memlock(); |
4938 | if (ret) | |
4939 | pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret); | |
4940 | ||
47eff617 | 4941 | /* make sure basic loading works */ |
6b434b61 | 4942 | ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, &opts); |
e32660ac | 4943 | if (ret < 0) |
6b434b61 | 4944 | ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, &opts); |
47eff617 | 4945 | if (ret < 0) { |
fd9eef1a EC |
4946 | ret = errno; |
4947 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4948 | pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF " | |
4949 | "program. Make sure your kernel supports BPF " | |
4950 | "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is " | |
4951 | "set to big enough value.\n", __func__, cp, ret); | |
4952 | return -ret; | |
47eff617 SF |
4953 | } |
4954 | close(ret); | |
4955 | ||
fd9eef1a EC |
4956 | return 0; |
4957 | } | |
4958 | ||
d6dd1d49 AN |
4959 | bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id) |
4960 | { | |
8263b338 | 4961 | if (obj->gen_loader) |
d6dd1d49 AN |
4962 | /* To generate loader program assume the latest kernel |
4963 | * to avoid doing extra prog_load, map_create syscalls. | |
4964 | */ | |
4965 | return true; | |
4966 | ||
6b434b61 AN |
4967 | if (obj->token_fd) |
4968 | return feat_supported(obj->feat_cache, feat_id); | |
4969 | ||
d6dd1d49 AN |
4970 | return feat_supported(NULL, feat_id); |
4971 | } | |
4972 | ||
57a00f41 THJ |
4973 | static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd) |
4974 | { | |
813847a3 | 4975 | struct bpf_map_info map_info; |
57a00f41 | 4976 | char msg[STRERR_BUFSIZE]; |
813847a3 | 4977 | __u32 map_info_len = sizeof(map_info); |
97eb3138 | 4978 | int err; |
57a00f41 | 4979 | |
813847a3 | 4980 | memset(&map_info, 0, map_info_len); |
629dfc66 | 4981 | err = bpf_map_get_info_by_fd(map_fd, &map_info, &map_info_len); |
97eb3138 MP |
4982 | if (err && errno == EINVAL) |
4983 | err = bpf_get_map_info_from_fdinfo(map_fd, &map_info); | |
4984 | if (err) { | |
4985 | pr_warn("failed to get map info for map FD %d: %s\n", map_fd, | |
4986 | libbpf_strerror_r(errno, msg, sizeof(msg))); | |
57a00f41 THJ |
4987 | return false; |
4988 | } | |
4989 | ||
4990 | return (map_info.type == map->def.type && | |
4991 | map_info.key_size == map->def.key_size && | |
4992 | map_info.value_size == map->def.value_size && | |
4993 | map_info.max_entries == map->def.max_entries && | |
47512102 JK |
4994 | map_info.map_flags == map->def.map_flags && |
4995 | map_info.map_extra == map->map_extra); | |
57a00f41 THJ |
4996 | } |
4997 | ||
4998 | static int | |
4999 | bpf_object__reuse_map(struct bpf_map *map) | |
5000 | { | |
5001 | char *cp, errmsg[STRERR_BUFSIZE]; | |
5002 | int err, pin_fd; | |
5003 | ||
5004 | pin_fd = bpf_obj_get(map->pin_path); | |
5005 | if (pin_fd < 0) { | |
5006 | err = -errno; | |
5007 | if (err == -ENOENT) { | |
5008 | pr_debug("found no pinned map to reuse at '%s'\n", | |
5009 | map->pin_path); | |
5010 | return 0; | |
5011 | } | |
5012 | ||
5013 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
5014 | pr_warn("couldn't retrieve pinned map '%s': %s\n", | |
5015 | map->pin_path, cp); | |
5016 | return err; | |
5017 | } | |
5018 | ||
5019 | if (!map_is_reuse_compat(map, pin_fd)) { | |
5020 | pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n", | |
5021 | map->pin_path); | |
5022 | close(pin_fd); | |
5023 | return -EINVAL; | |
5024 | } | |
5025 | ||
5026 | err = bpf_map__reuse_fd(map, pin_fd); | |
d0f325c3 | 5027 | close(pin_fd); |
e3ba8e4e | 5028 | if (err) |
57a00f41 | 5029 | return err; |
e3ba8e4e | 5030 | |
57a00f41 THJ |
5031 | map->pinned = true; |
5032 | pr_debug("reused pinned map at '%s'\n", map->pin_path); | |
5033 | ||
5034 | return 0; | |
5035 | } | |
5036 | ||
d859900c DB |
5037 | static int |
5038 | bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map) | |
5039 | { | |
166750bc | 5040 | enum libbpf_map_type map_type = map->libbpf_type; |
d859900c DB |
5041 | char *cp, errmsg[STRERR_BUFSIZE]; |
5042 | int err, zero = 0; | |
d859900c | 5043 | |
67234743 AS |
5044 | if (obj->gen_loader) { |
5045 | bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps, | |
5046 | map->mmaped, map->def.value_size); | |
5047 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) | |
5048 | bpf_gen__map_freeze(obj->gen_loader, map - obj->maps); | |
5049 | return 0; | |
5050 | } | |
2e7ba4f8 | 5051 | |
eba9c5f4 AN |
5052 | err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0); |
5053 | if (err) { | |
5054 | err = -errno; | |
5055 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
5056 | pr_warn("Error setting initial map(%s) contents: %s\n", | |
5057 | map->name, cp); | |
5058 | return err; | |
5059 | } | |
d859900c | 5060 | |
81bfdd08 AN |
5061 | /* Freeze .rodata and .kconfig map as read-only from syscall side. */ |
5062 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) { | |
d859900c DB |
5063 | err = bpf_map_freeze(map->fd); |
5064 | if (err) { | |
eba9c5f4 AN |
5065 | err = -errno; |
5066 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e KW |
5067 | pr_warn("Error freezing map(%s) as read-only: %s\n", |
5068 | map->name, cp); | |
eba9c5f4 | 5069 | return err; |
d859900c DB |
5070 | } |
5071 | } | |
eba9c5f4 | 5072 | return 0; |
d859900c DB |
5073 | } |
5074 | ||
2d39d7c5 AN |
5075 | static void bpf_map__destroy(struct bpf_map *map); |
5076 | ||
f08c18e0 AN |
5077 | static bool map_is_created(const struct bpf_map *map) |
5078 | { | |
5079 | return map->obj->loaded || map->reused; | |
5080 | } | |
5081 | ||
67234743 | 5082 | static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner) |
2d39d7c5 | 5083 | { |
992c4225 | 5084 | LIBBPF_OPTS(bpf_map_create_opts, create_attr); |
2d39d7c5 | 5085 | struct bpf_map_def *def = &map->def; |
992c4225 | 5086 | const char *map_name = NULL; |
dac645b9 | 5087 | int err = 0, map_fd; |
2d39d7c5 | 5088 | |
9ca1f56a | 5089 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
992c4225 | 5090 | map_name = map->name; |
2d39d7c5 | 5091 | create_attr.map_ifindex = map->map_ifindex; |
2d39d7c5 | 5092 | create_attr.map_flags = def->map_flags; |
1bdb6c9a | 5093 | create_attr.numa_node = map->numa_node; |
47512102 | 5094 | create_attr.map_extra = map->map_extra; |
6b434b61 AN |
5095 | create_attr.token_fd = obj->token_fd; |
5096 | if (obj->token_fd) | |
5097 | create_attr.map_flags |= BPF_F_TOKEN_FD; | |
2d39d7c5 | 5098 | |
9e926acd | 5099 | if (bpf_map__is_struct_ops(map)) { |
992c4225 | 5100 | create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id; |
9e926acd KFL |
5101 | if (map->mod_btf_fd >= 0) { |
5102 | create_attr.value_type_btf_obj_fd = map->mod_btf_fd; | |
5103 | create_attr.map_flags |= BPF_F_VTYPE_BTF_OBJ_FD; | |
5104 | } | |
5105 | } | |
2d39d7c5 | 5106 | |
262cfb74 | 5107 | if (obj->btf && btf__fd(obj->btf) >= 0) { |
2d39d7c5 AN |
5108 | create_attr.btf_fd = btf__fd(obj->btf); |
5109 | create_attr.btf_key_type_id = map->btf_key_type_id; | |
5110 | create_attr.btf_value_type_id = map->btf_value_type_id; | |
5111 | } | |
5112 | ||
646f02ff AN |
5113 | if (bpf_map_type__is_map_in_map(def->type)) { |
5114 | if (map->inner_map) { | |
f04deb90 AG |
5115 | err = map_set_def_max_entries(map->inner_map); |
5116 | if (err) | |
5117 | return err; | |
67234743 | 5118 | err = bpf_object__create_map(obj, map->inner_map, true); |
646f02ff AN |
5119 | if (err) { |
5120 | pr_warn("map '%s': failed to create inner map: %d\n", | |
5121 | map->name, err); | |
5122 | return err; | |
5123 | } | |
f08c18e0 | 5124 | map->inner_map_fd = map->inner_map->fd; |
646f02ff AN |
5125 | } |
5126 | if (map->inner_map_fd >= 0) | |
5127 | create_attr.inner_map_fd = map->inner_map_fd; | |
5128 | } | |
5129 | ||
f7310523 HC |
5130 | switch (def->type) { |
5131 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
5132 | case BPF_MAP_TYPE_CGROUP_ARRAY: | |
5133 | case BPF_MAP_TYPE_STACK_TRACE: | |
5134 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
5135 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
5136 | case BPF_MAP_TYPE_DEVMAP: | |
5137 | case BPF_MAP_TYPE_DEVMAP_HASH: | |
5138 | case BPF_MAP_TYPE_CPUMAP: | |
5139 | case BPF_MAP_TYPE_XSKMAP: | |
5140 | case BPF_MAP_TYPE_SOCKMAP: | |
5141 | case BPF_MAP_TYPE_SOCKHASH: | |
5142 | case BPF_MAP_TYPE_QUEUE: | |
5143 | case BPF_MAP_TYPE_STACK: | |
79ff13e9 | 5144 | case BPF_MAP_TYPE_ARENA: |
f7310523 HC |
5145 | create_attr.btf_fd = 0; |
5146 | create_attr.btf_key_type_id = 0; | |
5147 | create_attr.btf_value_type_id = 0; | |
5148 | map->btf_key_type_id = 0; | |
5149 | map->btf_value_type_id = 0; | |
3644d285 KFL |
5150 | break; |
5151 | case BPF_MAP_TYPE_STRUCT_OPS: | |
5152 | create_attr.btf_value_type_id = 0; | |
5153 | break; | |
f7310523 HC |
5154 | default: |
5155 | break; | |
5156 | } | |
5157 | ||
67234743 | 5158 | if (obj->gen_loader) { |
992c4225 | 5159 | bpf_gen__map_create(obj->gen_loader, def->type, map_name, |
a4fbfdd7 | 5160 | def->key_size, def->value_size, def->max_entries, |
992c4225 | 5161 | &create_attr, is_inner ? -1 : map - obj->maps); |
dac645b9 AN |
5162 | /* We keep pretenting we have valid FD to pass various fd >= 0 |
5163 | * checks by just keeping original placeholder FDs in place. | |
5164 | * See bpf_object__add_map() comment. | |
5165 | * This placeholder fd will not be used with any syscall and | |
5166 | * will be reset to -1 eventually. | |
67234743 | 5167 | */ |
dac645b9 | 5168 | map_fd = map->fd; |
67234743 | 5169 | } else { |
dac645b9 AN |
5170 | map_fd = bpf_map_create(def->type, map_name, |
5171 | def->key_size, def->value_size, | |
5172 | def->max_entries, &create_attr); | |
67234743 | 5173 | } |
dac645b9 | 5174 | if (map_fd < 0 && (create_attr.btf_key_type_id || create_attr.btf_value_type_id)) { |
2d39d7c5 | 5175 | char *cp, errmsg[STRERR_BUFSIZE]; |
2d39d7c5 | 5176 | |
a21ab4c5 | 5177 | err = -errno; |
2d39d7c5 AN |
5178 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); |
5179 | pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n", | |
5180 | map->name, cp, err); | |
5181 | create_attr.btf_fd = 0; | |
5182 | create_attr.btf_key_type_id = 0; | |
5183 | create_attr.btf_value_type_id = 0; | |
5184 | map->btf_key_type_id = 0; | |
5185 | map->btf_value_type_id = 0; | |
dac645b9 AN |
5186 | map_fd = bpf_map_create(def->type, map_name, |
5187 | def->key_size, def->value_size, | |
5188 | def->max_entries, &create_attr); | |
2d39d7c5 AN |
5189 | } |
5190 | ||
646f02ff | 5191 | if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) { |
67234743 AS |
5192 | if (obj->gen_loader) |
5193 | map->inner_map->fd = -1; | |
646f02ff AN |
5194 | bpf_map__destroy(map->inner_map); |
5195 | zfree(&map->inner_map); | |
5196 | } | |
5197 | ||
dac645b9 AN |
5198 | if (map_fd < 0) |
5199 | return map_fd; | |
5200 | ||
5201 | /* obj->gen_loader case, prevent reuse_fd() from closing map_fd */ | |
5202 | if (map->fd == map_fd) | |
5203 | return 0; | |
5204 | ||
5205 | /* Keep placeholder FD value but now point it to the BPF map object. | |
5206 | * This way everything that relied on this map's FD (e.g., relocated | |
5207 | * ldimm64 instructions) will stay valid and won't need adjustments. | |
5208 | * map->fd stays valid but now point to what map_fd points to. | |
5209 | */ | |
5210 | return reuse_fd(map->fd, map_fd); | |
2d39d7c5 AN |
5211 | } |
5212 | ||
341ac5ff | 5213 | static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map) |
a0f2b7ac HL |
5214 | { |
5215 | const struct bpf_map *targ_map; | |
5216 | unsigned int i; | |
67234743 | 5217 | int fd, err = 0; |
a0f2b7ac HL |
5218 | |
5219 | for (i = 0; i < map->init_slots_sz; i++) { | |
5220 | if (!map->init_slots[i]) | |
5221 | continue; | |
5222 | ||
5223 | targ_map = map->init_slots[i]; | |
f08c18e0 | 5224 | fd = targ_map->fd; |
341ac5ff | 5225 | |
67234743 | 5226 | if (obj->gen_loader) { |
be05c944 AS |
5227 | bpf_gen__populate_outer_map(obj->gen_loader, |
5228 | map - obj->maps, i, | |
5229 | targ_map - obj->maps); | |
67234743 AS |
5230 | } else { |
5231 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
5232 | } | |
a0f2b7ac HL |
5233 | if (err) { |
5234 | err = -errno; | |
5235 | pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n", | |
341ac5ff | 5236 | map->name, i, targ_map->name, fd, err); |
a0f2b7ac HL |
5237 | return err; |
5238 | } | |
5239 | pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n", | |
5240 | map->name, i, targ_map->name, fd); | |
5241 | } | |
5242 | ||
5243 | zfree(&map->init_slots); | |
5244 | map->init_slots_sz = 0; | |
5245 | ||
5246 | return 0; | |
5247 | } | |
5248 | ||
341ac5ff HC |
5249 | static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map) |
5250 | { | |
5251 | const struct bpf_program *targ_prog; | |
5252 | unsigned int i; | |
5253 | int fd, err; | |
5254 | ||
5255 | if (obj->gen_loader) | |
5256 | return -ENOTSUP; | |
5257 | ||
5258 | for (i = 0; i < map->init_slots_sz; i++) { | |
5259 | if (!map->init_slots[i]) | |
5260 | continue; | |
5261 | ||
5262 | targ_prog = map->init_slots[i]; | |
5263 | fd = bpf_program__fd(targ_prog); | |
5264 | ||
5265 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
5266 | if (err) { | |
5267 | err = -errno; | |
5268 | pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %d\n", | |
5269 | map->name, i, targ_prog->name, fd, err); | |
5270 | return err; | |
5271 | } | |
5272 | pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n", | |
5273 | map->name, i, targ_prog->name, fd); | |
5274 | } | |
5275 | ||
5276 | zfree(&map->init_slots); | |
5277 | map->init_slots_sz = 0; | |
5278 | ||
5279 | return 0; | |
5280 | } | |
5281 | ||
5282 | static int bpf_object_init_prog_arrays(struct bpf_object *obj) | |
5283 | { | |
5284 | struct bpf_map *map; | |
5285 | int i, err; | |
5286 | ||
5287 | for (i = 0; i < obj->nr_maps; i++) { | |
5288 | map = &obj->maps[i]; | |
5289 | ||
5290 | if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY) | |
5291 | continue; | |
5292 | ||
5293 | err = init_prog_array_slots(obj, map); | |
dac645b9 | 5294 | if (err < 0) |
341ac5ff | 5295 | return err; |
341ac5ff HC |
5296 | } |
5297 | return 0; | |
5298 | } | |
5299 | ||
a4fbfdd7 ST |
5300 | static int map_set_def_max_entries(struct bpf_map *map) |
5301 | { | |
5302 | if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) { | |
5303 | int nr_cpus; | |
5304 | ||
5305 | nr_cpus = libbpf_num_possible_cpus(); | |
5306 | if (nr_cpus < 0) { | |
5307 | pr_warn("map '%s': failed to determine number of system CPUs: %d\n", | |
5308 | map->name, nr_cpus); | |
5309 | return nr_cpus; | |
5310 | } | |
5311 | pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus); | |
5312 | map->def.max_entries = nr_cpus; | |
5313 | } | |
5314 | ||
5315 | return 0; | |
5316 | } | |
5317 | ||
52d3352e WN |
5318 | static int |
5319 | bpf_object__create_maps(struct bpf_object *obj) | |
5320 | { | |
2d39d7c5 AN |
5321 | struct bpf_map *map; |
5322 | char *cp, errmsg[STRERR_BUFSIZE]; | |
5323 | unsigned int i, j; | |
8a138aed | 5324 | int err; |
043c5bb3 | 5325 | bool retried; |
52d3352e | 5326 | |
9d759a9b | 5327 | for (i = 0; i < obj->nr_maps; i++) { |
2d39d7c5 | 5328 | map = &obj->maps[i]; |
8a138aed | 5329 | |
16e0c35c AN |
5330 | /* To support old kernels, we skip creating global data maps |
5331 | * (.rodata, .data, .kconfig, etc); later on, during program | |
5332 | * loading, if we detect that at least one of the to-be-loaded | |
5333 | * programs is referencing any global data map, we'll error | |
5334 | * out with program name and relocation index logged. | |
5335 | * This approach allows to accommodate Clang emitting | |
5336 | * unnecessary .rodata.str1.1 sections for string literals, | |
5337 | * but also it allows to have CO-RE applications that use | |
5338 | * global variables in some of BPF programs, but not others. | |
5339 | * If those global variable-using programs are not loaded at | |
5340 | * runtime due to bpf_program__set_autoload(prog, false), | |
5341 | * bpf_object loading will succeed just fine even on old | |
5342 | * kernels. | |
5343 | */ | |
ec41817b AN |
5344 | if (bpf_map__is_internal(map) && !kernel_supports(obj, FEAT_GLOBAL_DATA)) |
5345 | map->autocreate = false; | |
5346 | ||
5347 | if (!map->autocreate) { | |
5348 | pr_debug("map '%s': skipped auto-creating...\n", map->name); | |
16e0c35c | 5349 | continue; |
229fae38 | 5350 | } |
16e0c35c | 5351 | |
a4fbfdd7 ST |
5352 | err = map_set_def_max_entries(map); |
5353 | if (err) | |
5354 | goto err_out; | |
5355 | ||
043c5bb3 MP |
5356 | retried = false; |
5357 | retry: | |
57a00f41 THJ |
5358 | if (map->pin_path) { |
5359 | err = bpf_object__reuse_map(map); | |
5360 | if (err) { | |
2d39d7c5 | 5361 | pr_warn("map '%s': error reusing pinned map\n", |
57a00f41 | 5362 | map->name); |
2d39d7c5 | 5363 | goto err_out; |
57a00f41 | 5364 | } |
043c5bb3 MP |
5365 | if (retried && map->fd < 0) { |
5366 | pr_warn("map '%s': cannot find pinned map\n", | |
5367 | map->name); | |
5368 | err = -ENOENT; | |
5369 | goto err_out; | |
5370 | } | |
57a00f41 THJ |
5371 | } |
5372 | ||
fa98b54b | 5373 | if (map->reused) { |
2d39d7c5 | 5374 | pr_debug("map '%s': skipping creation (preset fd=%d)\n", |
26736eb9 | 5375 | map->name, map->fd); |
2c193d32 | 5376 | } else { |
67234743 | 5377 | err = bpf_object__create_map(obj, map, false); |
2c193d32 | 5378 | if (err) |
d859900c | 5379 | goto err_out; |
d859900c | 5380 | |
2c193d32 HL |
5381 | pr_debug("map '%s': created successfully, fd=%d\n", |
5382 | map->name, map->fd); | |
646f02ff | 5383 | |
2c193d32 HL |
5384 | if (bpf_map__is_internal(map)) { |
5385 | err = bpf_object__populate_internal_map(obj, map); | |
dac645b9 | 5386 | if (err < 0) |
2c193d32 | 5387 | goto err_out; |
d859900c | 5388 | } |
79ff13e9 | 5389 | if (map->def.type == BPF_MAP_TYPE_ARENA) { |
5ab8cb89 AN |
5390 | map->mmaped = mmap((void *)(long)map->map_extra, |
5391 | bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE, | |
79ff13e9 AS |
5392 | map->map_extra ? MAP_SHARED | MAP_FIXED : MAP_SHARED, |
5393 | map->fd, 0); | |
5394 | if (map->mmaped == MAP_FAILED) { | |
5395 | err = -errno; | |
5396 | map->mmaped = NULL; | |
5397 | pr_warn("map '%s': failed to mmap arena: %d\n", | |
5398 | map->name, err); | |
5399 | return err; | |
5400 | } | |
2e7ba4f8 AN |
5401 | if (obj->arena_data) { |
5402 | memcpy(map->mmaped, obj->arena_data, obj->arena_data_sz); | |
5403 | zfree(&obj->arena_data); | |
5404 | } | |
79ff13e9 | 5405 | } |
341ac5ff HC |
5406 | if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) { |
5407 | err = init_map_in_map_slots(obj, map); | |
dac645b9 | 5408 | if (err < 0) |
646f02ff | 5409 | goto err_out; |
646f02ff | 5410 | } |
646f02ff AN |
5411 | } |
5412 | ||
57a00f41 THJ |
5413 | if (map->pin_path && !map->pinned) { |
5414 | err = bpf_map__pin(map, NULL); | |
5415 | if (err) { | |
043c5bb3 MP |
5416 | if (!retried && err == -EEXIST) { |
5417 | retried = true; | |
5418 | goto retry; | |
5419 | } | |
2d39d7c5 AN |
5420 | pr_warn("map '%s': failed to auto-pin at '%s': %d\n", |
5421 | map->name, map->pin_path, err); | |
2d39d7c5 | 5422 | goto err_out; |
57a00f41 THJ |
5423 | } |
5424 | } | |
52d3352e WN |
5425 | } |
5426 | ||
52d3352e | 5427 | return 0; |
2d39d7c5 AN |
5428 | |
5429 | err_out: | |
5430 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
5431 | pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err); | |
5432 | pr_perm_msg(err); | |
5433 | for (j = 0; j < i; j++) | |
5434 | zclose(obj->maps[j].fd); | |
5435 | return err; | |
52d3352e WN |
5436 | } |
5437 | ||
ddc7c304 AN |
5438 | static bool bpf_core_is_flavor_sep(const char *s) |
5439 | { | |
5440 | /* check X___Y name pattern, where X and Y are not underscores */ | |
5441 | return s[0] != '_' && /* X */ | |
5442 | s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */ | |
5443 | s[4] != '_'; /* Y */ | |
5444 | } | |
5445 | ||
5446 | /* Given 'some_struct_name___with_flavor' return the length of a name prefix | |
5447 | * before last triple underscore. Struct name part after last triple | |
5448 | * underscore is ignored by BPF CO-RE relocation during relocation matching. | |
5449 | */ | |
b0588390 | 5450 | size_t bpf_core_essential_name_len(const char *name) |
ddc7c304 AN |
5451 | { |
5452 | size_t n = strlen(name); | |
5453 | int i; | |
5454 | ||
5455 | for (i = n - 5; i >= 0; i--) { | |
5456 | if (bpf_core_is_flavor_sep(name + i)) | |
5457 | return i + 1; | |
5458 | } | |
5459 | return n; | |
5460 | } | |
5461 | ||
8de6cae4 | 5462 | void bpf_core_free_cands(struct bpf_core_cand_list *cands) |
ddc7c304 | 5463 | { |
8de6cae4 MV |
5464 | if (!cands) |
5465 | return; | |
5466 | ||
0f7515ca AN |
5467 | free(cands->cands); |
5468 | free(cands); | |
ddc7c304 AN |
5469 | } |
5470 | ||
8de6cae4 MV |
5471 | int bpf_core_add_cands(struct bpf_core_cand *local_cand, |
5472 | size_t local_essent_len, | |
5473 | const struct btf *targ_btf, | |
5474 | const char *targ_btf_name, | |
5475 | int targ_start_id, | |
5476 | struct bpf_core_cand_list *cands) | |
ddc7c304 | 5477 | { |
301ba4d7 | 5478 | struct bpf_core_cand *new_cands, *cand; |
03d5b991 AN |
5479 | const struct btf_type *t, *local_t; |
5480 | const char *targ_name, *local_name; | |
0f7515ca AN |
5481 | size_t targ_essent_len; |
5482 | int n, i; | |
ddc7c304 | 5483 | |
03d5b991 AN |
5484 | local_t = btf__type_by_id(local_cand->btf, local_cand->id); |
5485 | local_name = btf__str_by_offset(local_cand->btf, local_t->name_off); | |
5486 | ||
6a886de0 HC |
5487 | n = btf__type_cnt(targ_btf); |
5488 | for (i = targ_start_id; i < n; i++) { | |
ddc7c304 | 5489 | t = btf__type_by_id(targ_btf, i); |
23b2a3a8 | 5490 | if (!btf_kind_core_compat(t, local_t)) |
ddc7c304 AN |
5491 | continue; |
5492 | ||
3fc32f40 AN |
5493 | targ_name = btf__name_by_offset(targ_btf, t->name_off); |
5494 | if (str_is_empty(targ_name)) | |
d121e1d3 AN |
5495 | continue; |
5496 | ||
ddc7c304 AN |
5497 | targ_essent_len = bpf_core_essential_name_len(targ_name); |
5498 | if (targ_essent_len != local_essent_len) | |
5499 | continue; | |
5500 | ||
03d5b991 | 5501 | if (strncmp(local_name, targ_name, local_essent_len) != 0) |
0f7515ca AN |
5502 | continue; |
5503 | ||
5504 | pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n", | |
03d5b991 AN |
5505 | local_cand->id, btf_kind_str(local_t), |
5506 | local_name, i, btf_kind_str(t), targ_name, | |
0f7515ca AN |
5507 | targ_btf_name); |
5508 | new_cands = libbpf_reallocarray(cands->cands, cands->len + 1, | |
5509 | sizeof(*cands->cands)); | |
5510 | if (!new_cands) | |
5511 | return -ENOMEM; | |
5512 | ||
5513 | cand = &new_cands[cands->len]; | |
5514 | cand->btf = targ_btf; | |
0f7515ca AN |
5515 | cand->id = i; |
5516 | ||
5517 | cands->cands = new_cands; | |
5518 | cands->len++; | |
ddc7c304 | 5519 | } |
0f7515ca AN |
5520 | return 0; |
5521 | } | |
5522 | ||
4f33a53d AN |
5523 | static int load_module_btfs(struct bpf_object *obj) |
5524 | { | |
5525 | struct bpf_btf_info info; | |
5526 | struct module_btf *mod_btf; | |
5527 | struct btf *btf; | |
5528 | char name[64]; | |
5529 | __u32 id = 0, len; | |
5530 | int err, fd; | |
5531 | ||
5532 | if (obj->btf_modules_loaded) | |
5533 | return 0; | |
5534 | ||
67234743 AS |
5535 | if (obj->gen_loader) |
5536 | return 0; | |
5537 | ||
4f33a53d AN |
5538 | /* don't do this again, even if we find no module BTFs */ |
5539 | obj->btf_modules_loaded = true; | |
5540 | ||
5541 | /* kernel too old to support module BTFs */ | |
9ca1f56a | 5542 | if (!kernel_supports(obj, FEAT_MODULE_BTF)) |
4f33a53d AN |
5543 | return 0; |
5544 | ||
5545 | while (true) { | |
5546 | err = bpf_btf_get_next_id(id, &id); | |
5547 | if (err && errno == ENOENT) | |
5548 | return 0; | |
2d2c9516 AT |
5549 | if (err && errno == EPERM) { |
5550 | pr_debug("skipping module BTFs loading, missing privileges\n"); | |
5551 | return 0; | |
5552 | } | |
4f33a53d AN |
5553 | if (err) { |
5554 | err = -errno; | |
5555 | pr_warn("failed to iterate BTF objects: %d\n", err); | |
5556 | return err; | |
ddc7c304 | 5557 | } |
4f33a53d AN |
5558 | |
5559 | fd = bpf_btf_get_fd_by_id(id); | |
5560 | if (fd < 0) { | |
5561 | if (errno == ENOENT) | |
5562 | continue; /* expected race: BTF was unloaded */ | |
5563 | err = -errno; | |
5564 | pr_warn("failed to get BTF object #%d FD: %d\n", id, err); | |
5565 | return err; | |
5566 | } | |
5567 | ||
5568 | len = sizeof(info); | |
5569 | memset(&info, 0, sizeof(info)); | |
5570 | info.name = ptr_to_u64(name); | |
5571 | info.name_len = sizeof(name); | |
5572 | ||
629dfc66 | 5573 | err = bpf_btf_get_info_by_fd(fd, &info, &len); |
4f33a53d AN |
5574 | if (err) { |
5575 | err = -errno; | |
5576 | pr_warn("failed to get BTF object #%d info: %d\n", id, err); | |
91abb4a6 | 5577 | goto err_out; |
4f33a53d AN |
5578 | } |
5579 | ||
5580 | /* ignore non-module BTFs */ | |
5581 | if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) { | |
5582 | close(fd); | |
5583 | continue; | |
5584 | } | |
5585 | ||
5586 | btf = btf_get_from_fd(fd, obj->btf_vmlinux); | |
e9fc3ce9 AN |
5587 | err = libbpf_get_error(btf); |
5588 | if (err) { | |
5589 | pr_warn("failed to load module [%s]'s BTF object #%d: %d\n", | |
5590 | name, id, err); | |
91abb4a6 | 5591 | goto err_out; |
4f33a53d AN |
5592 | } |
5593 | ||
3b029e06 | 5594 | err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap, |
e3ba8e4e | 5595 | sizeof(*obj->btf_modules), obj->btf_module_cnt + 1); |
4f33a53d | 5596 | if (err) |
91abb4a6 | 5597 | goto err_out; |
4f33a53d AN |
5598 | |
5599 | mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; | |
5600 | ||
5601 | mod_btf->btf = btf; | |
5602 | mod_btf->id = id; | |
91abb4a6 | 5603 | mod_btf->fd = fd; |
4f33a53d | 5604 | mod_btf->name = strdup(name); |
91abb4a6 AN |
5605 | if (!mod_btf->name) { |
5606 | err = -ENOMEM; | |
5607 | goto err_out; | |
5608 | } | |
5609 | continue; | |
5610 | ||
5611 | err_out: | |
5612 | close(fd); | |
5613 | return err; | |
ddc7c304 | 5614 | } |
4f33a53d AN |
5615 | |
5616 | return 0; | |
5617 | } | |
5618 | ||
301ba4d7 | 5619 | static struct bpf_core_cand_list * |
0f7515ca AN |
5620 | bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id) |
5621 | { | |
301ba4d7 AS |
5622 | struct bpf_core_cand local_cand = {}; |
5623 | struct bpf_core_cand_list *cands; | |
4f33a53d | 5624 | const struct btf *main_btf; |
03d5b991 AN |
5625 | const struct btf_type *local_t; |
5626 | const char *local_name; | |
0f7515ca | 5627 | size_t local_essent_len; |
4f33a53d | 5628 | int err, i; |
0f7515ca AN |
5629 | |
5630 | local_cand.btf = local_btf; | |
03d5b991 AN |
5631 | local_cand.id = local_type_id; |
5632 | local_t = btf__type_by_id(local_btf, local_type_id); | |
5633 | if (!local_t) | |
0f7515ca AN |
5634 | return ERR_PTR(-EINVAL); |
5635 | ||
03d5b991 AN |
5636 | local_name = btf__name_by_offset(local_btf, local_t->name_off); |
5637 | if (str_is_empty(local_name)) | |
0f7515ca | 5638 | return ERR_PTR(-EINVAL); |
03d5b991 | 5639 | local_essent_len = bpf_core_essential_name_len(local_name); |
0f7515ca AN |
5640 | |
5641 | cands = calloc(1, sizeof(*cands)); | |
5642 | if (!cands) | |
5643 | return ERR_PTR(-ENOMEM); | |
5644 | ||
5645 | /* Attempt to find target candidates in vmlinux BTF first */ | |
4f33a53d AN |
5646 | main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux; |
5647 | err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands); | |
5648 | if (err) | |
5649 | goto err_out; | |
5650 | ||
5651 | /* if vmlinux BTF has any candidate, don't got for module BTFs */ | |
5652 | if (cands->len) | |
5653 | return cands; | |
5654 | ||
5655 | /* if vmlinux BTF was overridden, don't attempt to load module BTFs */ | |
5656 | if (obj->btf_vmlinux_override) | |
5657 | return cands; | |
5658 | ||
5659 | /* now look through module BTFs, trying to still find candidates */ | |
5660 | err = load_module_btfs(obj); | |
5661 | if (err) | |
5662 | goto err_out; | |
5663 | ||
5664 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
5665 | err = bpf_core_add_cands(&local_cand, local_essent_len, | |
5666 | obj->btf_modules[i].btf, | |
5667 | obj->btf_modules[i].name, | |
6a886de0 | 5668 | btf__type_cnt(obj->btf_vmlinux), |
4f33a53d AN |
5669 | cands); |
5670 | if (err) | |
5671 | goto err_out; | |
0f7515ca AN |
5672 | } |
5673 | ||
5674 | return cands; | |
ddc7c304 | 5675 | err_out: |
4f33a53d | 5676 | bpf_core_free_cands(cands); |
ddc7c304 AN |
5677 | return ERR_PTR(err); |
5678 | } | |
5679 | ||
3fc32f40 AN |
5680 | /* Check local and target types for compatibility. This check is used for |
5681 | * type-based CO-RE relocations and follow slightly different rules than | |
5682 | * field-based relocations. This function assumes that root types were already | |
5683 | * checked for name match. Beyond that initial root-level name check, names | |
5684 | * are completely ignored. Compatibility rules are as follows: | |
5685 | * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but | |
5686 | * kind should match for local and target types (i.e., STRUCT is not | |
5687 | * compatible with UNION); | |
5688 | * - for ENUMs, the size is ignored; | |
5689 | * - for INT, size and signedness are ignored; | |
5690 | * - for ARRAY, dimensionality is ignored, element types are checked for | |
5691 | * compatibility recursively; | |
5692 | * - CONST/VOLATILE/RESTRICT modifiers are ignored; | |
5693 | * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible; | |
5694 | * - FUNC_PROTOs are compatible if they have compatible signature: same | |
5695 | * number of input args and compatible return and argument types. | |
5696 | * These rules are not set in stone and probably will be adjusted as we get | |
5697 | * more experience with using BPF CO-RE relocations. | |
5698 | */ | |
b0588390 AS |
5699 | int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id, |
5700 | const struct btf *targ_btf, __u32 targ_id) | |
3fc32f40 | 5701 | { |
fd75733d | 5702 | return __bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id, 32); |
3fc32f40 AN |
5703 | } |
5704 | ||
ec6209c8 DM |
5705 | int bpf_core_types_match(const struct btf *local_btf, __u32 local_id, |
5706 | const struct btf *targ_btf, __u32 targ_id) | |
5707 | { | |
5708 | return __bpf_core_types_match(local_btf, local_id, targ_btf, targ_id, false, 32); | |
5709 | } | |
5710 | ||
c302378b | 5711 | static size_t bpf_core_hash_fn(const long key, void *ctx) |
ddc7c304 | 5712 | { |
c302378b | 5713 | return key; |
ddc7c304 AN |
5714 | } |
5715 | ||
c302378b | 5716 | static bool bpf_core_equal_fn(const long k1, const long k2, void *ctx) |
ddc7c304 AN |
5717 | { |
5718 | return k1 == k2; | |
5719 | } | |
5720 | ||
d0e92887 AS |
5721 | static int record_relo_core(struct bpf_program *prog, |
5722 | const struct bpf_core_relo *core_relo, int insn_idx) | |
5723 | { | |
5724 | struct reloc_desc *relos, *relo; | |
5725 | ||
5726 | relos = libbpf_reallocarray(prog->reloc_desc, | |
5727 | prog->nr_reloc + 1, sizeof(*relos)); | |
5728 | if (!relos) | |
5729 | return -ENOMEM; | |
5730 | relo = &relos[prog->nr_reloc]; | |
5731 | relo->type = RELO_CORE; | |
5732 | relo->insn_idx = insn_idx; | |
5733 | relo->core_relo = core_relo; | |
5734 | prog->reloc_desc = relos; | |
5735 | prog->nr_reloc++; | |
5736 | return 0; | |
5737 | } | |
5738 | ||
9fdc4273 AN |
5739 | static const struct bpf_core_relo *find_relo_core(struct bpf_program *prog, int insn_idx) |
5740 | { | |
5741 | struct reloc_desc *relo; | |
5742 | int i; | |
5743 | ||
5744 | for (i = 0; i < prog->nr_reloc; i++) { | |
5745 | relo = &prog->reloc_desc[i]; | |
5746 | if (relo->type != RELO_CORE || relo->insn_idx != insn_idx) | |
5747 | continue; | |
5748 | ||
5749 | return relo->core_relo; | |
5750 | } | |
5751 | ||
5752 | return NULL; | |
5753 | } | |
5754 | ||
adb8fa19 MV |
5755 | static int bpf_core_resolve_relo(struct bpf_program *prog, |
5756 | const struct bpf_core_relo *relo, | |
5757 | int relo_idx, | |
5758 | const struct btf *local_btf, | |
5759 | struct hashmap *cand_cache, | |
5760 | struct bpf_core_relo_res *targ_res) | |
3ee4f533 | 5761 | { |
78c1f8d0 | 5762 | struct bpf_core_spec specs_scratch[3] = {}; |
301ba4d7 | 5763 | struct bpf_core_cand_list *cands = NULL; |
3ee4f533 AS |
5764 | const char *prog_name = prog->name; |
5765 | const struct btf_type *local_type; | |
5766 | const char *local_name; | |
5767 | __u32 local_id = relo->type_id; | |
adb8fa19 | 5768 | int err; |
3ee4f533 AS |
5769 | |
5770 | local_type = btf__type_by_id(local_btf, local_id); | |
5771 | if (!local_type) | |
5772 | return -EINVAL; | |
5773 | ||
5774 | local_name = btf__name_by_offset(local_btf, local_type->name_off); | |
5775 | if (!local_name) | |
5776 | return -EINVAL; | |
5777 | ||
46334a0c | 5778 | if (relo->kind != BPF_CORE_TYPE_ID_LOCAL && |
c302378b | 5779 | !hashmap__find(cand_cache, local_id, &cands)) { |
3ee4f533 AS |
5780 | cands = bpf_core_find_cands(prog->obj, local_btf, local_id); |
5781 | if (IS_ERR(cands)) { | |
5782 | pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n", | |
5783 | prog_name, relo_idx, local_id, btf_kind_str(local_type), | |
5784 | local_name, PTR_ERR(cands)); | |
5785 | return PTR_ERR(cands); | |
5786 | } | |
c302378b | 5787 | err = hashmap__set(cand_cache, local_id, cands, NULL, NULL); |
3ee4f533 AS |
5788 | if (err) { |
5789 | bpf_core_free_cands(cands); | |
5790 | return err; | |
5791 | } | |
5792 | } | |
5793 | ||
adb8fa19 MV |
5794 | return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch, |
5795 | targ_res); | |
3ee4f533 AS |
5796 | } |
5797 | ||
ddc7c304 | 5798 | static int |
28b93c64 | 5799 | bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) |
ddc7c304 AN |
5800 | { |
5801 | const struct btf_ext_info_sec *sec; | |
adb8fa19 | 5802 | struct bpf_core_relo_res targ_res; |
28b93c64 | 5803 | const struct bpf_core_relo *rec; |
ddc7c304 AN |
5804 | const struct btf_ext_info *seg; |
5805 | struct hashmap_entry *entry; | |
5806 | struct hashmap *cand_cache = NULL; | |
5807 | struct bpf_program *prog; | |
adb8fa19 | 5808 | struct bpf_insn *insn; |
ddc7c304 | 5809 | const char *sec_name; |
11d5daa8 | 5810 | int i, err = 0, insn_idx, sec_idx, sec_num; |
ddc7c304 | 5811 | |
28b93c64 AN |
5812 | if (obj->btf_ext->core_relo_info.len == 0) |
5813 | return 0; | |
5814 | ||
0f7515ca AN |
5815 | if (targ_btf_path) { |
5816 | obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL); | |
e9fc3ce9 AN |
5817 | err = libbpf_get_error(obj->btf_vmlinux_override); |
5818 | if (err) { | |
0f7515ca AN |
5819 | pr_warn("failed to parse target BTF: %d\n", err); |
5820 | return err; | |
5821 | } | |
ddc7c304 AN |
5822 | } |
5823 | ||
5824 | cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL); | |
5825 | if (IS_ERR(cand_cache)) { | |
5826 | err = PTR_ERR(cand_cache); | |
5827 | goto out; | |
5828 | } | |
5829 | ||
28b93c64 | 5830 | seg = &obj->btf_ext->core_relo_info; |
11d5daa8 | 5831 | sec_num = 0; |
ddc7c304 | 5832 | for_each_btf_ext_sec(seg, sec) { |
11d5daa8 AN |
5833 | sec_idx = seg->sec_idxs[sec_num]; |
5834 | sec_num++; | |
5835 | ||
ddc7c304 AN |
5836 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); |
5837 | if (str_is_empty(sec_name)) { | |
5838 | err = -EINVAL; | |
5839 | goto out; | |
5840 | } | |
ddc7c304 | 5841 | |
11d5daa8 | 5842 | pr_debug("sec '%s': found %d CO-RE relocations\n", sec_name, sec->num_info); |
ddc7c304 AN |
5843 | |
5844 | for_each_btf_ext_rec(seg, sec, i, rec) { | |
adb8fa19 MV |
5845 | if (rec->insn_off % BPF_INSN_SZ) |
5846 | return -EINVAL; | |
db2b8b06 AN |
5847 | insn_idx = rec->insn_off / BPF_INSN_SZ; |
5848 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); | |
5849 | if (!prog) { | |
e89d57d9 AN |
5850 | /* When __weak subprog is "overridden" by another instance |
5851 | * of the subprog from a different object file, linker still | |
5852 | * appends all the .BTF.ext info that used to belong to that | |
5853 | * eliminated subprogram. | |
5854 | * This is similar to what x86-64 linker does for relocations. | |
5855 | * So just ignore such relocations just like we ignore | |
5856 | * subprog instructions when discovering subprograms. | |
5857 | */ | |
5858 | pr_debug("sec '%s': skipping CO-RE relocation #%d for insn #%d belonging to eliminated weak subprogram\n", | |
5859 | sec_name, i, insn_idx); | |
5860 | continue; | |
db2b8b06 | 5861 | } |
47f7cf63 AN |
5862 | /* no need to apply CO-RE relocation if the program is |
5863 | * not going to be loaded | |
5864 | */ | |
a3820c48 | 5865 | if (!prog->autoload) |
47f7cf63 | 5866 | continue; |
db2b8b06 | 5867 | |
adb8fa19 MV |
5868 | /* adjust insn_idx from section frame of reference to the local |
5869 | * program's frame of reference; (sub-)program code is not yet | |
5870 | * relocated, so it's enough to just subtract in-section offset | |
5871 | */ | |
5872 | insn_idx = insn_idx - prog->sec_insn_off; | |
5873 | if (insn_idx >= prog->insns_cnt) | |
5874 | return -EINVAL; | |
5875 | insn = &prog->insns[insn_idx]; | |
5876 | ||
185cfe83 AN |
5877 | err = record_relo_core(prog, rec, insn_idx); |
5878 | if (err) { | |
5879 | pr_warn("prog '%s': relo #%d: failed to record relocation: %d\n", | |
5880 | prog->name, i, err); | |
5881 | goto out; | |
adb8fa19 MV |
5882 | } |
5883 | ||
185cfe83 AN |
5884 | if (prog->obj->gen_loader) |
5885 | continue; | |
5886 | ||
adb8fa19 | 5887 | err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res); |
ddc7c304 | 5888 | if (err) { |
be18010e | 5889 | pr_warn("prog '%s': relo #%d: failed to relocate: %d\n", |
9c0f8cbd | 5890 | prog->name, i, err); |
ddc7c304 AN |
5891 | goto out; |
5892 | } | |
adb8fa19 MV |
5893 | |
5894 | err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res); | |
5895 | if (err) { | |
5896 | pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n", | |
5897 | prog->name, i, insn_idx, err); | |
5898 | goto out; | |
5899 | } | |
ddc7c304 AN |
5900 | } |
5901 | } | |
5902 | ||
5903 | out: | |
4f33a53d | 5904 | /* obj->btf_vmlinux and module BTFs are freed after object load */ |
0f7515ca AN |
5905 | btf__free(obj->btf_vmlinux_override); |
5906 | obj->btf_vmlinux_override = NULL; | |
5907 | ||
ddc7c304 AN |
5908 | if (!IS_ERR_OR_NULL(cand_cache)) { |
5909 | hashmap__for_each_entry(cand_cache, entry, i) { | |
c302378b | 5910 | bpf_core_free_cands(entry->pvalue); |
ddc7c304 AN |
5911 | } |
5912 | hashmap__free(cand_cache); | |
5913 | } | |
5914 | return err; | |
5915 | } | |
5916 | ||
ec41817b | 5917 | /* base map load ldimm64 special constant, used also for log fixup logic */ |
3055ddd6 AN |
5918 | #define POISON_LDIMM64_MAP_BASE 2001000000 |
5919 | #define POISON_LDIMM64_MAP_PFX "200100" | |
ec41817b AN |
5920 | |
5921 | static void poison_map_ldimm64(struct bpf_program *prog, int relo_idx, | |
5922 | int insn_idx, struct bpf_insn *insn, | |
5923 | int map_idx, const struct bpf_map *map) | |
5924 | { | |
5925 | int i; | |
5926 | ||
5927 | pr_debug("prog '%s': relo #%d: poisoning insn #%d that loads map #%d '%s'\n", | |
5928 | prog->name, relo_idx, insn_idx, map_idx, map->name); | |
5929 | ||
5930 | /* we turn single ldimm64 into two identical invalid calls */ | |
5931 | for (i = 0; i < 2; i++) { | |
5932 | insn->code = BPF_JMP | BPF_CALL; | |
5933 | insn->dst_reg = 0; | |
5934 | insn->src_reg = 0; | |
5935 | insn->off = 0; | |
5936 | /* if this instruction is reachable (not a dead code), | |
5937 | * verifier will complain with something like: | |
5938 | * invalid func unknown#2001000123 | |
5939 | * where lower 123 is map index into obj->maps[] array | |
5940 | */ | |
3055ddd6 | 5941 | insn->imm = POISON_LDIMM64_MAP_BASE + map_idx; |
ec41817b AN |
5942 | |
5943 | insn++; | |
5944 | } | |
5945 | } | |
5946 | ||
05b6f766 AN |
5947 | /* unresolved kfunc call special constant, used also for log fixup logic */ |
5948 | #define POISON_CALL_KFUNC_BASE 2002000000 | |
5949 | #define POISON_CALL_KFUNC_PFX "2002" | |
5950 | ||
5951 | static void poison_kfunc_call(struct bpf_program *prog, int relo_idx, | |
5952 | int insn_idx, struct bpf_insn *insn, | |
5953 | int ext_idx, const struct extern_desc *ext) | |
5954 | { | |
5955 | pr_debug("prog '%s': relo #%d: poisoning insn #%d that calls kfunc '%s'\n", | |
5956 | prog->name, relo_idx, insn_idx, ext->name); | |
5957 | ||
5958 | /* we turn kfunc call into invalid helper call with identifiable constant */ | |
5959 | insn->code = BPF_JMP | BPF_CALL; | |
5960 | insn->dst_reg = 0; | |
5961 | insn->src_reg = 0; | |
5962 | insn->off = 0; | |
5963 | /* if this instruction is reachable (not a dead code), | |
5964 | * verifier will complain with something like: | |
5965 | * invalid func unknown#2001000123 | |
5966 | * where lower 123 is extern index into obj->externs[] array | |
5967 | */ | |
5968 | insn->imm = POISON_CALL_KFUNC_BASE + ext_idx; | |
5969 | } | |
5970 | ||
c3c55696 AN |
5971 | /* Relocate data references within program code: |
5972 | * - map references; | |
5973 | * - global variable references; | |
5974 | * - extern references. | |
5975 | */ | |
48cca7e4 | 5976 | static int |
c3c55696 | 5977 | bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog) |
8a47a6c5 | 5978 | { |
c3c55696 | 5979 | int i; |
8a47a6c5 WN |
5980 | |
5981 | for (i = 0; i < prog->nr_reloc; i++) { | |
53f8dd43 | 5982 | struct reloc_desc *relo = &prog->reloc_desc[i]; |
166750bc | 5983 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; |
ec41817b | 5984 | const struct bpf_map *map; |
2e33efe3 | 5985 | struct extern_desc *ext; |
8a47a6c5 | 5986 | |
166750bc AN |
5987 | switch (relo->type) { |
5988 | case RELO_LD64: | |
ec41817b | 5989 | map = &obj->maps[relo->map_idx]; |
e2fa0156 AS |
5990 | if (obj->gen_loader) { |
5991 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX; | |
5992 | insn[0].imm = relo->map_idx; | |
ec41817b | 5993 | } else if (map->autocreate) { |
e2fa0156 | 5994 | insn[0].src_reg = BPF_PSEUDO_MAP_FD; |
ec41817b AN |
5995 | insn[0].imm = map->fd; |
5996 | } else { | |
5997 | poison_map_ldimm64(prog, i, relo->insn_idx, insn, | |
5998 | relo->map_idx, map); | |
e2fa0156 | 5999 | } |
166750bc AN |
6000 | break; |
6001 | case RELO_DATA: | |
ec41817b | 6002 | map = &obj->maps[relo->map_idx]; |
166750bc | 6003 | insn[1].imm = insn[0].imm + relo->sym_off; |
e2fa0156 AS |
6004 | if (obj->gen_loader) { |
6005 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
6006 | insn[0].imm = relo->map_idx; | |
ec41817b | 6007 | } else if (map->autocreate) { |
e2fa0156 | 6008 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; |
ec41817b AN |
6009 | insn[0].imm = map->fd; |
6010 | } else { | |
6011 | poison_map_ldimm64(prog, i, relo->insn_idx, insn, | |
6012 | relo->map_idx, map); | |
e2fa0156 | 6013 | } |
166750bc | 6014 | break; |
a18f7214 | 6015 | case RELO_EXTERN_LD64: |
3055ddd6 | 6016 | ext = &obj->externs[relo->ext_idx]; |
1c0c7074 | 6017 | if (ext->type == EXT_KCFG) { |
e2fa0156 AS |
6018 | if (obj->gen_loader) { |
6019 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
6020 | insn[0].imm = obj->kconfig_map_idx; | |
6021 | } else { | |
6022 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; | |
6023 | insn[0].imm = obj->maps[obj->kconfig_map_idx].fd; | |
6024 | } | |
1c0c7074 AN |
6025 | insn[1].imm = ext->kcfg.data_off; |
6026 | } else /* EXT_KSYM */ { | |
2211c825 | 6027 | if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */ |
d370bbe1 | 6028 | insn[0].src_reg = BPF_PSEUDO_BTF_ID; |
284d2587 AN |
6029 | insn[0].imm = ext->ksym.kernel_btf_id; |
6030 | insn[1].imm = ext->ksym.kernel_btf_obj_fd; | |
2211c825 | 6031 | } else { /* typeless ksyms or unresolved typed ksyms */ |
d370bbe1 HL |
6032 | insn[0].imm = (__u32)ext->ksym.addr; |
6033 | insn[1].imm = ext->ksym.addr >> 32; | |
6034 | } | |
1c0c7074 | 6035 | } |
166750bc | 6036 | break; |
a18f7214 | 6037 | case RELO_EXTERN_CALL: |
3055ddd6 | 6038 | ext = &obj->externs[relo->ext_idx]; |
5bd022ec | 6039 | insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL; |
466b2e13 KKD |
6040 | if (ext->is_set) { |
6041 | insn[0].imm = ext->ksym.kernel_btf_id; | |
6042 | insn[0].off = ext->ksym.btf_fd_idx; | |
05b6f766 AN |
6043 | } else { /* unresolved weak kfunc call */ |
6044 | poison_kfunc_call(prog, i, relo->insn_idx, insn, | |
6045 | relo->ext_idx, ext); | |
466b2e13 | 6046 | } |
5bd022ec | 6047 | break; |
53eddb5e | 6048 | case RELO_SUBPROG_ADDR: |
b1268826 AS |
6049 | if (insn[0].src_reg != BPF_PSEUDO_FUNC) { |
6050 | pr_warn("prog '%s': relo #%d: bad insn\n", | |
6051 | prog->name, i); | |
6052 | return -EINVAL; | |
6053 | } | |
6054 | /* handled already */ | |
53eddb5e | 6055 | break; |
166750bc | 6056 | case RELO_CALL: |
b1268826 | 6057 | /* handled already */ |
166750bc | 6058 | break; |
d0e92887 AS |
6059 | case RELO_CORE: |
6060 | /* will be handled by bpf_program_record_relos() */ | |
6061 | break; | |
166750bc | 6062 | default: |
9c0f8cbd AN |
6063 | pr_warn("prog '%s': relo #%d: bad relo type %d\n", |
6064 | prog->name, i, relo->type); | |
166750bc | 6065 | return -EINVAL; |
8a47a6c5 | 6066 | } |
8a47a6c5 WN |
6067 | } |
6068 | ||
c3c55696 AN |
6069 | return 0; |
6070 | } | |
6071 | ||
8505e870 AN |
6072 | static int adjust_prog_btf_ext_info(const struct bpf_object *obj, |
6073 | const struct bpf_program *prog, | |
6074 | const struct btf_ext_info *ext_info, | |
6075 | void **prog_info, __u32 *prog_rec_cnt, | |
6076 | __u32 *prog_rec_sz) | |
6077 | { | |
6078 | void *copy_start = NULL, *copy_end = NULL; | |
6079 | void *rec, *rec_end, *new_prog_info; | |
6080 | const struct btf_ext_info_sec *sec; | |
6081 | size_t old_sz, new_sz; | |
11d5daa8 | 6082 | int i, sec_num, sec_idx, off_adj; |
8505e870 | 6083 | |
11d5daa8 | 6084 | sec_num = 0; |
8505e870 | 6085 | for_each_btf_ext_sec(ext_info, sec) { |
11d5daa8 AN |
6086 | sec_idx = ext_info->sec_idxs[sec_num]; |
6087 | sec_num++; | |
6088 | if (prog->sec_idx != sec_idx) | |
8505e870 AN |
6089 | continue; |
6090 | ||
6091 | for_each_btf_ext_rec(ext_info, sec, i, rec) { | |
6092 | __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ; | |
6093 | ||
6094 | if (insn_off < prog->sec_insn_off) | |
6095 | continue; | |
6096 | if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt) | |
6097 | break; | |
6098 | ||
6099 | if (!copy_start) | |
6100 | copy_start = rec; | |
6101 | copy_end = rec + ext_info->rec_size; | |
6102 | } | |
6103 | ||
6104 | if (!copy_start) | |
6105 | return -ENOENT; | |
6106 | ||
6107 | /* append func/line info of a given (sub-)program to the main | |
6108 | * program func/line info | |
6109 | */ | |
8eb62958 | 6110 | old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size; |
8505e870 AN |
6111 | new_sz = old_sz + (copy_end - copy_start); |
6112 | new_prog_info = realloc(*prog_info, new_sz); | |
6113 | if (!new_prog_info) | |
6114 | return -ENOMEM; | |
6115 | *prog_info = new_prog_info; | |
6116 | *prog_rec_cnt = new_sz / ext_info->rec_size; | |
6117 | memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start); | |
6118 | ||
6119 | /* Kernel instruction offsets are in units of 8-byte | |
6120 | * instructions, while .BTF.ext instruction offsets generated | |
6121 | * by Clang are in units of bytes. So convert Clang offsets | |
6122 | * into kernel offsets and adjust offset according to program | |
6123 | * relocated position. | |
6124 | */ | |
6125 | off_adj = prog->sub_insn_off - prog->sec_insn_off; | |
6126 | rec = new_prog_info + old_sz; | |
6127 | rec_end = new_prog_info + new_sz; | |
6128 | for (; rec < rec_end; rec += ext_info->rec_size) { | |
6129 | __u32 *insn_off = rec; | |
6130 | ||
6131 | *insn_off = *insn_off / BPF_INSN_SZ + off_adj; | |
6132 | } | |
6133 | *prog_rec_sz = ext_info->rec_size; | |
6134 | return 0; | |
6135 | } | |
6136 | ||
6137 | return -ENOENT; | |
6138 | } | |
6139 | ||
6140 | static int | |
6141 | reloc_prog_func_and_line_info(const struct bpf_object *obj, | |
6142 | struct bpf_program *main_prog, | |
6143 | const struct bpf_program *prog) | |
6144 | { | |
6145 | int err; | |
6146 | ||
6147 | /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't | |
2f38fe68 | 6148 | * support func/line info |
8505e870 | 6149 | */ |
9ca1f56a | 6150 | if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC)) |
8505e870 AN |
6151 | return 0; |
6152 | ||
6153 | /* only attempt func info relocation if main program's func_info | |
6154 | * relocation was successful | |
6155 | */ | |
6156 | if (main_prog != prog && !main_prog->func_info) | |
6157 | goto line_info; | |
6158 | ||
6159 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info, | |
6160 | &main_prog->func_info, | |
6161 | &main_prog->func_info_cnt, | |
6162 | &main_prog->func_info_rec_size); | |
6163 | if (err) { | |
6164 | if (err != -ENOENT) { | |
6165 | pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n", | |
6166 | prog->name, err); | |
6167 | return err; | |
6168 | } | |
6169 | if (main_prog->func_info) { | |
6170 | /* | |
6171 | * Some info has already been found but has problem | |
6172 | * in the last btf_ext reloc. Must have to error out. | |
6173 | */ | |
6174 | pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name); | |
6175 | return err; | |
6176 | } | |
6177 | /* Have problem loading the very first info. Ignore the rest. */ | |
6178 | pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n", | |
6179 | prog->name); | |
6180 | } | |
6181 | ||
6182 | line_info: | |
6183 | /* don't relocate line info if main program's relocation failed */ | |
6184 | if (main_prog != prog && !main_prog->line_info) | |
6185 | return 0; | |
6186 | ||
6187 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info, | |
6188 | &main_prog->line_info, | |
6189 | &main_prog->line_info_cnt, | |
6190 | &main_prog->line_info_rec_size); | |
6191 | if (err) { | |
6192 | if (err != -ENOENT) { | |
6193 | pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n", | |
6194 | prog->name, err); | |
6195 | return err; | |
6196 | } | |
6197 | if (main_prog->line_info) { | |
6198 | /* | |
6199 | * Some info has already been found but has problem | |
6200 | * in the last btf_ext reloc. Must have to error out. | |
6201 | */ | |
6202 | pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name); | |
6203 | return err; | |
6204 | } | |
6205 | /* Have problem loading the very first info. Ignore the rest. */ | |
6206 | pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n", | |
6207 | prog->name); | |
6208 | } | |
6209 | return 0; | |
6210 | } | |
6211 | ||
c3c55696 AN |
6212 | static int cmp_relo_by_insn_idx(const void *key, const void *elem) |
6213 | { | |
6214 | size_t insn_idx = *(const size_t *)key; | |
6215 | const struct reloc_desc *relo = elem; | |
6216 | ||
6217 | if (insn_idx == relo->insn_idx) | |
6218 | return 0; | |
6219 | return insn_idx < relo->insn_idx ? -1 : 1; | |
6220 | } | |
6221 | ||
6222 | static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx) | |
6223 | { | |
2a6a9bf2 AN |
6224 | if (!prog->nr_reloc) |
6225 | return NULL; | |
c3c55696 AN |
6226 | return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc, |
6227 | sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx); | |
6228 | } | |
6229 | ||
b1268826 AS |
6230 | static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog) |
6231 | { | |
6232 | int new_cnt = main_prog->nr_reloc + subprog->nr_reloc; | |
6233 | struct reloc_desc *relos; | |
6234 | int i; | |
6235 | ||
6236 | if (main_prog == subprog) | |
6237 | return 0; | |
6238 | relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos)); | |
8a0260db AN |
6239 | /* if new count is zero, reallocarray can return a valid NULL result; |
6240 | * in this case the previous pointer will be freed, so we *have to* | |
6241 | * reassign old pointer to the new value (even if it's NULL) | |
6242 | */ | |
6243 | if (!relos && new_cnt) | |
b1268826 | 6244 | return -ENOMEM; |
2a6a9bf2 AN |
6245 | if (subprog->nr_reloc) |
6246 | memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc, | |
6247 | sizeof(*relos) * subprog->nr_reloc); | |
b1268826 AS |
6248 | |
6249 | for (i = main_prog->nr_reloc; i < new_cnt; i++) | |
6250 | relos[i].insn_idx += subprog->sub_insn_off; | |
6251 | /* After insn_idx adjustment the 'relos' array is still sorted | |
6252 | * by insn_idx and doesn't break bsearch. | |
6253 | */ | |
6254 | main_prog->reloc_desc = relos; | |
6255 | main_prog->nr_reloc = new_cnt; | |
6256 | return 0; | |
6257 | } | |
6258 | ||
6c918709 KKD |
6259 | static int |
6260 | bpf_object__append_subprog_code(struct bpf_object *obj, struct bpf_program *main_prog, | |
6261 | struct bpf_program *subprog) | |
6262 | { | |
6263 | struct bpf_insn *insns; | |
6264 | size_t new_cnt; | |
6265 | int err; | |
6266 | ||
6267 | subprog->sub_insn_off = main_prog->insns_cnt; | |
6268 | ||
6269 | new_cnt = main_prog->insns_cnt + subprog->insns_cnt; | |
6270 | insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns)); | |
6271 | if (!insns) { | |
6272 | pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name); | |
6273 | return -ENOMEM; | |
6274 | } | |
6275 | main_prog->insns = insns; | |
6276 | main_prog->insns_cnt = new_cnt; | |
6277 | ||
6278 | memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns, | |
6279 | subprog->insns_cnt * sizeof(*insns)); | |
6280 | ||
6281 | pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n", | |
6282 | main_prog->name, subprog->insns_cnt, subprog->name); | |
6283 | ||
6284 | /* The subprog insns are now appended. Append its relos too. */ | |
6285 | err = append_subprog_relos(main_prog, subprog); | |
6286 | if (err) | |
6287 | return err; | |
6288 | return 0; | |
6289 | } | |
6290 | ||
c3c55696 AN |
6291 | static int |
6292 | bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog, | |
6293 | struct bpf_program *prog) | |
6294 | { | |
7e2925f6 | 6295 | size_t sub_insn_idx, insn_idx; |
c3c55696 | 6296 | struct bpf_program *subprog; |
c3c55696 | 6297 | struct reloc_desc *relo; |
7e2925f6 | 6298 | struct bpf_insn *insn; |
c3c55696 AN |
6299 | int err; |
6300 | ||
6301 | err = reloc_prog_func_and_line_info(obj, main_prog, prog); | |
6302 | if (err) | |
6303 | return err; | |
6304 | ||
6305 | for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) { | |
6306 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
53eddb5e | 6307 | if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn)) |
c3c55696 AN |
6308 | continue; |
6309 | ||
6310 | relo = find_prog_insn_relo(prog, insn_idx); | |
a18f7214 | 6311 | if (relo && relo->type == RELO_EXTERN_CALL) |
b1268826 AS |
6312 | /* kfunc relocations will be handled later |
6313 | * in bpf_object__relocate_data() | |
6314 | */ | |
6315 | continue; | |
53eddb5e | 6316 | if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) { |
c3c55696 AN |
6317 | pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n", |
6318 | prog->name, insn_idx, relo->type); | |
6319 | return -LIBBPF_ERRNO__RELOC; | |
6320 | } | |
6321 | if (relo) { | |
6322 | /* sub-program instruction index is a combination of | |
6323 | * an offset of a symbol pointed to by relocation and | |
6324 | * call instruction's imm field; for global functions, | |
6325 | * call always has imm = -1, but for static functions | |
6326 | * relocation is against STT_SECTION and insn->imm | |
6327 | * points to a start of a static function | |
53eddb5e YS |
6328 | * |
6329 | * for subprog addr relocation, the relo->sym_off + insn->imm is | |
6330 | * the byte offset in the corresponding section. | |
c3c55696 | 6331 | */ |
53eddb5e YS |
6332 | if (relo->type == RELO_CALL) |
6333 | sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1; | |
6334 | else | |
6335 | sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ; | |
6336 | } else if (insn_is_pseudo_func(insn)) { | |
6337 | /* | |
6338 | * RELO_SUBPROG_ADDR relo is always emitted even if both | |
6339 | * functions are in the same section, so it shouldn't reach here. | |
6340 | */ | |
6341 | pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n", | |
6342 | prog->name, insn_idx); | |
6343 | return -LIBBPF_ERRNO__RELOC; | |
c3c55696 AN |
6344 | } else { |
6345 | /* if subprogram call is to a static function within | |
6346 | * the same ELF section, there won't be any relocation | |
6347 | * emitted, but it also means there is no additional | |
6348 | * offset necessary, insns->imm is relative to | |
6349 | * instruction's original position within the section | |
6350 | */ | |
6351 | sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1; | |
6352 | } | |
6353 | ||
6354 | /* we enforce that sub-programs should be in .text section */ | |
6355 | subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx); | |
6356 | if (!subprog) { | |
6357 | pr_warn("prog '%s': no .text section found yet sub-program call exists\n", | |
6358 | prog->name); | |
6359 | return -LIBBPF_ERRNO__RELOC; | |
6360 | } | |
6361 | ||
6362 | /* if it's the first call instruction calling into this | |
6363 | * subprogram (meaning this subprog hasn't been processed | |
6364 | * yet) within the context of current main program: | |
6365 | * - append it at the end of main program's instructions blog; | |
6366 | * - process is recursively, while current program is put on hold; | |
6367 | * - if that subprogram calls some other not yet processes | |
6368 | * subprogram, same thing will happen recursively until | |
6369 | * there are no more unprocesses subprograms left to append | |
6370 | * and relocate. | |
6371 | */ | |
6372 | if (subprog->sub_insn_off == 0) { | |
6c918709 | 6373 | err = bpf_object__append_subprog_code(obj, main_prog, subprog); |
b1268826 AS |
6374 | if (err) |
6375 | return err; | |
c3c55696 AN |
6376 | err = bpf_object__reloc_code(obj, main_prog, subprog); |
6377 | if (err) | |
6378 | return err; | |
6379 | } | |
6380 | ||
6381 | /* main_prog->insns memory could have been re-allocated, so | |
6382 | * calculate pointer again | |
6383 | */ | |
6384 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
6385 | /* calculate correct instruction position within current main | |
6386 | * prog; each main prog can have a different set of | |
6387 | * subprograms appended (potentially in different order as | |
6388 | * well), so position of any subprog can be different for | |
e3ba8e4e KM |
6389 | * different main programs |
6390 | */ | |
c3c55696 AN |
6391 | insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1; |
6392 | ||
c3c55696 AN |
6393 | pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n", |
6394 | prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off); | |
6395 | } | |
6396 | ||
6397 | return 0; | |
6398 | } | |
6399 | ||
6400 | /* | |
6401 | * Relocate sub-program calls. | |
6402 | * | |
6403 | * Algorithm operates as follows. Each entry-point BPF program (referred to as | |
6404 | * main prog) is processed separately. For each subprog (non-entry functions, | |
6405 | * that can be called from either entry progs or other subprogs) gets their | |
6406 | * sub_insn_off reset to zero. This serves as indicator that this subprogram | |
6407 | * hasn't been yet appended and relocated within current main prog. Once its | |
6408 | * relocated, sub_insn_off will point at the position within current main prog | |
6409 | * where given subprog was appended. This will further be used to relocate all | |
6410 | * the call instructions jumping into this subprog. | |
6411 | * | |
6412 | * We start with main program and process all call instructions. If the call | |
6413 | * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off | |
6414 | * is zero), subprog instructions are appended at the end of main program's | |
6415 | * instruction array. Then main program is "put on hold" while we recursively | |
6416 | * process newly appended subprogram. If that subprogram calls into another | |
6417 | * subprogram that hasn't been appended, new subprogram is appended again to | |
6418 | * the *main* prog's instructions (subprog's instructions are always left | |
6419 | * untouched, as they need to be in unmodified state for subsequent main progs | |
6420 | * and subprog instructions are always sent only as part of a main prog) and | |
6421 | * the process continues recursively. Once all the subprogs called from a main | |
6422 | * prog or any of its subprogs are appended (and relocated), all their | |
6423 | * positions within finalized instructions array are known, so it's easy to | |
6424 | * rewrite call instructions with correct relative offsets, corresponding to | |
6425 | * desired target subprog. | |
6426 | * | |
6427 | * Its important to realize that some subprogs might not be called from some | |
6428 | * main prog and any of its called/used subprogs. Those will keep their | |
6429 | * subprog->sub_insn_off as zero at all times and won't be appended to current | |
6430 | * main prog and won't be relocated within the context of current main prog. | |
6431 | * They might still be used from other main progs later. | |
6432 | * | |
6433 | * Visually this process can be shown as below. Suppose we have two main | |
6434 | * programs mainA and mainB and BPF object contains three subprogs: subA, | |
6435 | * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and | |
6436 | * subC both call subB: | |
6437 | * | |
6438 | * +--------+ +-------+ | |
6439 | * | v v | | |
6440 | * +--+---+ +--+-+-+ +---+--+ | |
6441 | * | subA | | subB | | subC | | |
6442 | * +--+---+ +------+ +---+--+ | |
6443 | * ^ ^ | |
6444 | * | | | |
6445 | * +---+-------+ +------+----+ | |
6446 | * | mainA | | mainB | | |
6447 | * +-----------+ +-----------+ | |
6448 | * | |
6449 | * We'll start relocating mainA, will find subA, append it and start | |
6450 | * processing sub A recursively: | |
6451 | * | |
6452 | * +-----------+------+ | |
6453 | * | mainA | subA | | |
6454 | * +-----------+------+ | |
6455 | * | |
6456 | * At this point we notice that subB is used from subA, so we append it and | |
6457 | * relocate (there are no further subcalls from subB): | |
6458 | * | |
6459 | * +-----------+------+------+ | |
6460 | * | mainA | subA | subB | | |
6461 | * +-----------+------+------+ | |
6462 | * | |
6463 | * At this point, we relocate subA calls, then go one level up and finish with | |
6464 | * relocatin mainA calls. mainA is done. | |
6465 | * | |
6466 | * For mainB process is similar but results in different order. We start with | |
6467 | * mainB and skip subA and subB, as mainB never calls them (at least | |
6468 | * directly), but we see subC is needed, so we append and start processing it: | |
6469 | * | |
6470 | * +-----------+------+ | |
6471 | * | mainB | subC | | |
6472 | * +-----------+------+ | |
6473 | * Now we see subC needs subB, so we go back to it, append and relocate it: | |
6474 | * | |
6475 | * +-----------+------+------+ | |
6476 | * | mainB | subC | subB | | |
6477 | * +-----------+------+------+ | |
6478 | * | |
6479 | * At this point we unwind recursion, relocate calls in subC, then in mainB. | |
6480 | */ | |
6481 | static int | |
6482 | bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog) | |
6483 | { | |
6484 | struct bpf_program *subprog; | |
d3d93e34 | 6485 | int i, err; |
c3c55696 | 6486 | |
c3c55696 AN |
6487 | /* mark all subprogs as not relocated (yet) within the context of |
6488 | * current main program | |
6489 | */ | |
6490 | for (i = 0; i < obj->nr_programs; i++) { | |
6491 | subprog = &obj->programs[i]; | |
6492 | if (!prog_is_subprog(obj, subprog)) | |
6493 | continue; | |
6494 | ||
6495 | subprog->sub_insn_off = 0; | |
c3c55696 AN |
6496 | } |
6497 | ||
6498 | err = bpf_object__reloc_code(obj, prog, prog); | |
6499 | if (err) | |
6500 | return err; | |
6501 | ||
8a47a6c5 WN |
6502 | return 0; |
6503 | } | |
6504 | ||
67234743 AS |
6505 | static void |
6506 | bpf_object__free_relocs(struct bpf_object *obj) | |
6507 | { | |
6508 | struct bpf_program *prog; | |
6509 | int i; | |
6510 | ||
6511 | /* free up relocation descriptors */ | |
6512 | for (i = 0; i < obj->nr_programs; i++) { | |
6513 | prog = &obj->programs[i]; | |
6514 | zfree(&prog->reloc_desc); | |
6515 | prog->nr_reloc = 0; | |
6516 | } | |
6517 | } | |
6518 | ||
d0e92887 AS |
6519 | static int cmp_relocs(const void *_a, const void *_b) |
6520 | { | |
6521 | const struct reloc_desc *a = _a; | |
6522 | const struct reloc_desc *b = _b; | |
6523 | ||
6524 | if (a->insn_idx != b->insn_idx) | |
6525 | return a->insn_idx < b->insn_idx ? -1 : 1; | |
6526 | ||
6527 | /* no two relocations should have the same insn_idx, but ... */ | |
6528 | if (a->type != b->type) | |
6529 | return a->type < b->type ? -1 : 1; | |
6530 | ||
6531 | return 0; | |
6532 | } | |
6533 | ||
6534 | static void bpf_object__sort_relos(struct bpf_object *obj) | |
6535 | { | |
6536 | int i; | |
6537 | ||
6538 | for (i = 0; i < obj->nr_programs; i++) { | |
6539 | struct bpf_program *p = &obj->programs[i]; | |
6540 | ||
6541 | if (!p->nr_reloc) | |
6542 | continue; | |
6543 | ||
6544 | qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs); | |
6545 | } | |
6546 | } | |
6547 | ||
fb03be7c AN |
6548 | static int bpf_prog_assign_exc_cb(struct bpf_object *obj, struct bpf_program *prog) |
6549 | { | |
6550 | const char *str = "exception_callback:"; | |
6551 | size_t pfx_len = strlen(str); | |
6552 | int i, j, n; | |
6553 | ||
6554 | if (!obj->btf || !kernel_supports(obj, FEAT_BTF_DECL_TAG)) | |
6555 | return 0; | |
6556 | ||
6557 | n = btf__type_cnt(obj->btf); | |
6558 | for (i = 1; i < n; i++) { | |
6559 | const char *name; | |
6560 | struct btf_type *t; | |
6561 | ||
6562 | t = btf_type_by_id(obj->btf, i); | |
6563 | if (!btf_is_decl_tag(t) || btf_decl_tag(t)->component_idx != -1) | |
6564 | continue; | |
6565 | ||
6566 | name = btf__str_by_offset(obj->btf, t->name_off); | |
6567 | if (strncmp(name, str, pfx_len) != 0) | |
6568 | continue; | |
6569 | ||
6570 | t = btf_type_by_id(obj->btf, t->type); | |
6571 | if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) { | |
6572 | pr_warn("prog '%s': exception_callback:<value> decl tag not applied to the main program\n", | |
6573 | prog->name); | |
6574 | return -EINVAL; | |
6575 | } | |
6576 | if (strcmp(prog->name, btf__str_by_offset(obj->btf, t->name_off)) != 0) | |
6577 | continue; | |
6578 | /* Multiple callbacks are specified for the same prog, | |
6579 | * the verifier will eventually return an error for this | |
6580 | * case, hence simply skip appending a subprog. | |
6581 | */ | |
6582 | if (prog->exception_cb_idx >= 0) { | |
6583 | prog->exception_cb_idx = -1; | |
6584 | break; | |
6585 | } | |
6586 | ||
6587 | name += pfx_len; | |
6588 | if (str_is_empty(name)) { | |
6589 | pr_warn("prog '%s': exception_callback:<value> decl tag contains empty value\n", | |
6590 | prog->name); | |
6591 | return -EINVAL; | |
6592 | } | |
6593 | ||
6594 | for (j = 0; j < obj->nr_programs; j++) { | |
6595 | struct bpf_program *subprog = &obj->programs[j]; | |
6596 | ||
6597 | if (!prog_is_subprog(obj, subprog)) | |
6598 | continue; | |
6599 | if (strcmp(name, subprog->name) != 0) | |
6600 | continue; | |
6601 | /* Enforce non-hidden, as from verifier point of | |
6602 | * view it expects global functions, whereas the | |
6603 | * mark_btf_static fixes up linkage as static. | |
6604 | */ | |
6605 | if (!subprog->sym_global || subprog->mark_btf_static) { | |
6606 | pr_warn("prog '%s': exception callback %s must be a global non-hidden function\n", | |
6607 | prog->name, subprog->name); | |
6608 | return -EINVAL; | |
6609 | } | |
6610 | /* Let's see if we already saw a static exception callback with the same name */ | |
6611 | if (prog->exception_cb_idx >= 0) { | |
6612 | pr_warn("prog '%s': multiple subprogs with same name as exception callback '%s'\n", | |
6613 | prog->name, subprog->name); | |
6614 | return -EINVAL; | |
6615 | } | |
6616 | prog->exception_cb_idx = j; | |
6617 | break; | |
6618 | } | |
6619 | ||
6620 | if (prog->exception_cb_idx >= 0) | |
6621 | continue; | |
6622 | ||
6623 | pr_warn("prog '%s': cannot find exception callback '%s'\n", prog->name, name); | |
6624 | return -ENOENT; | |
6625 | } | |
6626 | ||
6627 | return 0; | |
6628 | } | |
6629 | ||
2f38fe68 AN |
6630 | static struct { |
6631 | enum bpf_prog_type prog_type; | |
6632 | const char *ctx_name; | |
6633 | } global_ctx_map[] = { | |
6634 | { BPF_PROG_TYPE_CGROUP_DEVICE, "bpf_cgroup_dev_ctx" }, | |
6635 | { BPF_PROG_TYPE_CGROUP_SKB, "__sk_buff" }, | |
6636 | { BPF_PROG_TYPE_CGROUP_SOCK, "bpf_sock" }, | |
6637 | { BPF_PROG_TYPE_CGROUP_SOCK_ADDR, "bpf_sock_addr" }, | |
6638 | { BPF_PROG_TYPE_CGROUP_SOCKOPT, "bpf_sockopt" }, | |
6639 | { BPF_PROG_TYPE_CGROUP_SYSCTL, "bpf_sysctl" }, | |
6640 | { BPF_PROG_TYPE_FLOW_DISSECTOR, "__sk_buff" }, | |
6641 | { BPF_PROG_TYPE_KPROBE, "bpf_user_pt_regs_t" }, | |
6642 | { BPF_PROG_TYPE_LWT_IN, "__sk_buff" }, | |
6643 | { BPF_PROG_TYPE_LWT_OUT, "__sk_buff" }, | |
6644 | { BPF_PROG_TYPE_LWT_SEG6LOCAL, "__sk_buff" }, | |
6645 | { BPF_PROG_TYPE_LWT_XMIT, "__sk_buff" }, | |
6646 | { BPF_PROG_TYPE_NETFILTER, "bpf_nf_ctx" }, | |
6647 | { BPF_PROG_TYPE_PERF_EVENT, "bpf_perf_event_data" }, | |
6648 | { BPF_PROG_TYPE_RAW_TRACEPOINT, "bpf_raw_tracepoint_args" }, | |
6649 | { BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE, "bpf_raw_tracepoint_args" }, | |
6650 | { BPF_PROG_TYPE_SCHED_ACT, "__sk_buff" }, | |
6651 | { BPF_PROG_TYPE_SCHED_CLS, "__sk_buff" }, | |
6652 | { BPF_PROG_TYPE_SK_LOOKUP, "bpf_sk_lookup" }, | |
6653 | { BPF_PROG_TYPE_SK_MSG, "sk_msg_md" }, | |
6654 | { BPF_PROG_TYPE_SK_REUSEPORT, "sk_reuseport_md" }, | |
6655 | { BPF_PROG_TYPE_SK_SKB, "__sk_buff" }, | |
6656 | { BPF_PROG_TYPE_SOCK_OPS, "bpf_sock_ops" }, | |
6657 | { BPF_PROG_TYPE_SOCKET_FILTER, "__sk_buff" }, | |
6658 | { BPF_PROG_TYPE_XDP, "xdp_md" }, | |
6659 | /* all other program types don't have "named" context structs */ | |
6660 | }; | |
6661 | ||
9eea8faf AN |
6662 | /* forward declarations for arch-specific underlying types of bpf_user_pt_regs_t typedef, |
6663 | * for below __builtin_types_compatible_p() checks; | |
6664 | * with this approach we don't need any extra arch-specific #ifdef guards | |
6665 | */ | |
6666 | struct pt_regs; | |
6667 | struct user_pt_regs; | |
6668 | struct user_regs_struct; | |
6669 | ||
76ec90a9 AN |
6670 | static bool need_func_arg_type_fixup(const struct btf *btf, const struct bpf_program *prog, |
6671 | const char *subprog_name, int arg_idx, | |
6672 | int arg_type_id, const char *ctx_name) | |
6673 | { | |
6674 | const struct btf_type *t; | |
6675 | const char *tname; | |
6676 | ||
6677 | /* check if existing parameter already matches verifier expectations */ | |
6678 | t = skip_mods_and_typedefs(btf, arg_type_id, NULL); | |
6679 | if (!btf_is_ptr(t)) | |
6680 | goto out_warn; | |
6681 | ||
6682 | /* typedef bpf_user_pt_regs_t is a special PITA case, valid for kprobe | |
6683 | * and perf_event programs, so check this case early on and forget | |
6684 | * about it for subsequent checks | |
6685 | */ | |
6686 | while (btf_is_mod(t)) | |
6687 | t = btf__type_by_id(btf, t->type); | |
6688 | if (btf_is_typedef(t) && | |
6689 | (prog->type == BPF_PROG_TYPE_KPROBE || prog->type == BPF_PROG_TYPE_PERF_EVENT)) { | |
6690 | tname = btf__str_by_offset(btf, t->name_off) ?: "<anon>"; | |
6691 | if (strcmp(tname, "bpf_user_pt_regs_t") == 0) | |
6692 | return false; /* canonical type for kprobe/perf_event */ | |
6693 | } | |
6694 | ||
6695 | /* now we can ignore typedefs moving forward */ | |
6696 | t = skip_mods_and_typedefs(btf, t->type, NULL); | |
6697 | ||
6698 | /* if it's `void *`, definitely fix up BTF info */ | |
6699 | if (btf_is_void(t)) | |
6700 | return true; | |
6701 | ||
6702 | /* if it's already proper canonical type, no need to fix up */ | |
6703 | tname = btf__str_by_offset(btf, t->name_off) ?: "<anon>"; | |
6704 | if (btf_is_struct(t) && strcmp(tname, ctx_name) == 0) | |
6705 | return false; | |
6706 | ||
6707 | /* special cases */ | |
6708 | switch (prog->type) { | |
6709 | case BPF_PROG_TYPE_KPROBE: | |
76ec90a9 AN |
6710 | /* `struct pt_regs *` is expected, but we need to fix up */ |
6711 | if (btf_is_struct(t) && strcmp(tname, "pt_regs") == 0) | |
6712 | return true; | |
6713 | break; | |
9eea8faf AN |
6714 | case BPF_PROG_TYPE_PERF_EVENT: |
6715 | if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct pt_regs) && | |
6716 | btf_is_struct(t) && strcmp(tname, "pt_regs") == 0) | |
d7bc416a | 6717 | return true; |
9eea8faf AN |
6718 | if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct user_pt_regs) && |
6719 | btf_is_struct(t) && strcmp(tname, "user_pt_regs") == 0) | |
d7bc416a | 6720 | return true; |
9eea8faf AN |
6721 | if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct user_regs_struct) && |
6722 | btf_is_struct(t) && strcmp(tname, "user_regs_struct") == 0) | |
d7bc416a | 6723 | return true; |
9eea8faf | 6724 | break; |
76ec90a9 AN |
6725 | case BPF_PROG_TYPE_RAW_TRACEPOINT: |
6726 | case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE: | |
6727 | /* allow u64* as ctx */ | |
6728 | if (btf_is_int(t) && t->size == 8) | |
6729 | return true; | |
6730 | break; | |
6731 | default: | |
6732 | break; | |
6733 | } | |
6734 | ||
6735 | out_warn: | |
6736 | pr_warn("prog '%s': subprog '%s' arg#%d is expected to be of `struct %s *` type\n", | |
6737 | prog->name, subprog_name, arg_idx, ctx_name); | |
6738 | return false; | |
6739 | } | |
6740 | ||
2f38fe68 AN |
6741 | static int clone_func_btf_info(struct btf *btf, int orig_fn_id, struct bpf_program *prog) |
6742 | { | |
6743 | int fn_id, fn_proto_id, ret_type_id, orig_proto_id; | |
6744 | int i, err, arg_cnt, fn_name_off, linkage; | |
6745 | struct btf_type *fn_t, *fn_proto_t, *t; | |
6746 | struct btf_param *p; | |
6747 | ||
6748 | /* caller already validated FUNC -> FUNC_PROTO validity */ | |
6749 | fn_t = btf_type_by_id(btf, orig_fn_id); | |
6750 | fn_proto_t = btf_type_by_id(btf, fn_t->type); | |
6751 | ||
6752 | /* Note that each btf__add_xxx() operation invalidates | |
6753 | * all btf_type and string pointers, so we need to be | |
6754 | * very careful when cloning BTF types. BTF type | |
6755 | * pointers have to be always refetched. And to avoid | |
6756 | * problems with invalidated string pointers, we | |
6757 | * add empty strings initially, then just fix up | |
6758 | * name_off offsets in place. Offsets are stable for | |
6759 | * existing strings, so that works out. | |
6760 | */ | |
6761 | fn_name_off = fn_t->name_off; /* we are about to invalidate fn_t */ | |
6762 | linkage = btf_func_linkage(fn_t); | |
6763 | orig_proto_id = fn_t->type; /* original FUNC_PROTO ID */ | |
6764 | ret_type_id = fn_proto_t->type; /* fn_proto_t will be invalidated */ | |
6765 | arg_cnt = btf_vlen(fn_proto_t); | |
6766 | ||
6767 | /* clone FUNC_PROTO and its params */ | |
6768 | fn_proto_id = btf__add_func_proto(btf, ret_type_id); | |
6769 | if (fn_proto_id < 0) | |
6770 | return -EINVAL; | |
6771 | ||
6772 | for (i = 0; i < arg_cnt; i++) { | |
6773 | int name_off; | |
6774 | ||
6775 | /* copy original parameter data */ | |
6776 | t = btf_type_by_id(btf, orig_proto_id); | |
6777 | p = &btf_params(t)[i]; | |
6778 | name_off = p->name_off; | |
6779 | ||
6780 | err = btf__add_func_param(btf, "", p->type); | |
6781 | if (err) | |
6782 | return err; | |
6783 | ||
6784 | fn_proto_t = btf_type_by_id(btf, fn_proto_id); | |
6785 | p = &btf_params(fn_proto_t)[i]; | |
6786 | p->name_off = name_off; /* use remembered str offset */ | |
6787 | } | |
6788 | ||
6789 | /* clone FUNC now, btf__add_func() enforces non-empty name, so use | |
6790 | * entry program's name as a placeholder, which we replace immediately | |
6791 | * with original name_off | |
6792 | */ | |
6793 | fn_id = btf__add_func(btf, prog->name, linkage, fn_proto_id); | |
6794 | if (fn_id < 0) | |
6795 | return -EINVAL; | |
6796 | ||
6797 | fn_t = btf_type_by_id(btf, fn_id); | |
6798 | fn_t->name_off = fn_name_off; /* reuse original string */ | |
6799 | ||
6800 | return fn_id; | |
6801 | } | |
6802 | ||
6803 | /* Check if main program or global subprog's function prototype has `arg:ctx` | |
6804 | * argument tags, and, if necessary, substitute correct type to match what BPF | |
6805 | * verifier would expect, taking into account specific program type. This | |
6806 | * allows to support __arg_ctx tag transparently on old kernels that don't yet | |
6807 | * have a native support for it in the verifier, making user's life much | |
6808 | * easier. | |
6809 | */ | |
6810 | static int bpf_program_fixup_func_info(struct bpf_object *obj, struct bpf_program *prog) | |
6811 | { | |
76ec90a9 | 6812 | const char *ctx_name = NULL, *ctx_tag = "arg:ctx", *fn_name; |
2f38fe68 AN |
6813 | struct bpf_func_info_min *func_rec; |
6814 | struct btf_type *fn_t, *fn_proto_t; | |
6815 | struct btf *btf = obj->btf; | |
6816 | const struct btf_type *t; | |
6817 | struct btf_param *p; | |
6818 | int ptr_id = 0, struct_id, tag_id, orig_fn_id; | |
6819 | int i, n, arg_idx, arg_cnt, err, rec_idx; | |
6820 | int *orig_ids; | |
6821 | ||
6822 | /* no .BTF.ext, no problem */ | |
6823 | if (!obj->btf_ext || !prog->func_info) | |
6824 | return 0; | |
6825 | ||
01b55f4f | 6826 | /* don't do any fix ups if kernel natively supports __arg_ctx */ |
0e6d0a9d | 6827 | if (kernel_supports(obj, FEAT_ARG_CTX_TAG)) |
01b55f4f AN |
6828 | return 0; |
6829 | ||
2f38fe68 AN |
6830 | /* some BPF program types just don't have named context structs, so |
6831 | * this fallback mechanism doesn't work for them | |
6832 | */ | |
6833 | for (i = 0; i < ARRAY_SIZE(global_ctx_map); i++) { | |
6834 | if (global_ctx_map[i].prog_type != prog->type) | |
6835 | continue; | |
6836 | ctx_name = global_ctx_map[i].ctx_name; | |
6837 | break; | |
6838 | } | |
6839 | if (!ctx_name) | |
6840 | return 0; | |
6841 | ||
6842 | /* remember original func BTF IDs to detect if we already cloned them */ | |
6843 | orig_ids = calloc(prog->func_info_cnt, sizeof(*orig_ids)); | |
6844 | if (!orig_ids) | |
6845 | return -ENOMEM; | |
6846 | for (i = 0; i < prog->func_info_cnt; i++) { | |
6847 | func_rec = prog->func_info + prog->func_info_rec_size * i; | |
6848 | orig_ids[i] = func_rec->type_id; | |
6849 | } | |
6850 | ||
6851 | /* go through each DECL_TAG with "arg:ctx" and see if it points to one | |
6852 | * of our subprogs; if yes and subprog is global and needs adjustment, | |
6853 | * clone and adjust FUNC -> FUNC_PROTO combo | |
6854 | */ | |
6855 | for (i = 1, n = btf__type_cnt(btf); i < n; i++) { | |
6856 | /* only DECL_TAG with "arg:ctx" value are interesting */ | |
6857 | t = btf__type_by_id(btf, i); | |
6858 | if (!btf_is_decl_tag(t)) | |
6859 | continue; | |
6860 | if (strcmp(btf__str_by_offset(btf, t->name_off), ctx_tag) != 0) | |
6861 | continue; | |
6862 | ||
6863 | /* only global funcs need adjustment, if at all */ | |
6864 | orig_fn_id = t->type; | |
6865 | fn_t = btf_type_by_id(btf, orig_fn_id); | |
6866 | if (!btf_is_func(fn_t) || btf_func_linkage(fn_t) != BTF_FUNC_GLOBAL) | |
6867 | continue; | |
6868 | ||
6869 | /* sanity check FUNC -> FUNC_PROTO chain, just in case */ | |
6870 | fn_proto_t = btf_type_by_id(btf, fn_t->type); | |
6871 | if (!fn_proto_t || !btf_is_func_proto(fn_proto_t)) | |
6872 | continue; | |
6873 | ||
6874 | /* find corresponding func_info record */ | |
6875 | func_rec = NULL; | |
6876 | for (rec_idx = 0; rec_idx < prog->func_info_cnt; rec_idx++) { | |
6877 | if (orig_ids[rec_idx] == t->type) { | |
6878 | func_rec = prog->func_info + prog->func_info_rec_size * rec_idx; | |
6879 | break; | |
6880 | } | |
6881 | } | |
6882 | /* current main program doesn't call into this subprog */ | |
6883 | if (!func_rec) | |
6884 | continue; | |
6885 | ||
6886 | /* some more sanity checking of DECL_TAG */ | |
6887 | arg_cnt = btf_vlen(fn_proto_t); | |
6888 | arg_idx = btf_decl_tag(t)->component_idx; | |
6889 | if (arg_idx < 0 || arg_idx >= arg_cnt) | |
6890 | continue; | |
6891 | ||
76ec90a9 | 6892 | /* check if we should fix up argument type */ |
2f38fe68 | 6893 | p = &btf_params(fn_proto_t)[arg_idx]; |
76ec90a9 AN |
6894 | fn_name = btf__str_by_offset(btf, fn_t->name_off) ?: "<anon>"; |
6895 | if (!need_func_arg_type_fixup(btf, prog, fn_name, arg_idx, p->type, ctx_name)) | |
6896 | continue; | |
2f38fe68 AN |
6897 | |
6898 | /* clone fn/fn_proto, unless we already did it for another arg */ | |
6899 | if (func_rec->type_id == orig_fn_id) { | |
6900 | int fn_id; | |
6901 | ||
6902 | fn_id = clone_func_btf_info(btf, orig_fn_id, prog); | |
6903 | if (fn_id < 0) { | |
6904 | err = fn_id; | |
6905 | goto err_out; | |
6906 | } | |
6907 | ||
6908 | /* point func_info record to a cloned FUNC type */ | |
6909 | func_rec->type_id = fn_id; | |
6910 | } | |
6911 | ||
6912 | /* create PTR -> STRUCT type chain to mark PTR_TO_CTX argument; | |
6913 | * we do it just once per main BPF program, as all global | |
6914 | * funcs share the same program type, so need only PTR -> | |
6915 | * STRUCT type chain | |
6916 | */ | |
6917 | if (ptr_id == 0) { | |
6918 | struct_id = btf__add_struct(btf, ctx_name, 0); | |
6919 | ptr_id = btf__add_ptr(btf, struct_id); | |
6920 | if (ptr_id < 0 || struct_id < 0) { | |
6921 | err = -EINVAL; | |
6922 | goto err_out; | |
6923 | } | |
6924 | } | |
6925 | ||
6926 | /* for completeness, clone DECL_TAG and point it to cloned param */ | |
6927 | tag_id = btf__add_decl_tag(btf, ctx_tag, func_rec->type_id, arg_idx); | |
6928 | if (tag_id < 0) { | |
6929 | err = -EINVAL; | |
6930 | goto err_out; | |
6931 | } | |
6932 | ||
6933 | /* all the BTF manipulations invalidated pointers, refetch them */ | |
6934 | fn_t = btf_type_by_id(btf, func_rec->type_id); | |
6935 | fn_proto_t = btf_type_by_id(btf, fn_t->type); | |
6936 | ||
6937 | /* fix up type ID pointed to by param */ | |
6938 | p = &btf_params(fn_proto_t)[arg_idx]; | |
6939 | p->type = ptr_id; | |
6940 | } | |
6941 | ||
6942 | free(orig_ids); | |
6943 | return 0; | |
6944 | err_out: | |
6945 | free(orig_ids); | |
6946 | return err; | |
6947 | } | |
6948 | ||
6949 | static int bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path) | |
8a47a6c5 WN |
6950 | { |
6951 | struct bpf_program *prog; | |
b1268826 | 6952 | size_t i, j; |
8a47a6c5 WN |
6953 | int err; |
6954 | ||
ddc7c304 AN |
6955 | if (obj->btf_ext) { |
6956 | err = bpf_object__relocate_core(obj, targ_btf_path); | |
6957 | if (err) { | |
be18010e KW |
6958 | pr_warn("failed to perform CO-RE relocations: %d\n", |
6959 | err); | |
ddc7c304 AN |
6960 | return err; |
6961 | } | |
185cfe83 | 6962 | bpf_object__sort_relos(obj); |
ddc7c304 | 6963 | } |
b1268826 AS |
6964 | |
6965 | /* Before relocating calls pre-process relocations and mark | |
6966 | * few ld_imm64 instructions that points to subprogs. | |
6967 | * Otherwise bpf_object__reloc_code() later would have to consider | |
6968 | * all ld_imm64 insns as relocation candidates. That would | |
6969 | * reduce relocation speed, since amount of find_prog_insn_relo() | |
6970 | * would increase and most of them will fail to find a relo. | |
9173cac3 AN |
6971 | */ |
6972 | for (i = 0; i < obj->nr_programs; i++) { | |
6973 | prog = &obj->programs[i]; | |
b1268826 AS |
6974 | for (j = 0; j < prog->nr_reloc; j++) { |
6975 | struct reloc_desc *relo = &prog->reloc_desc[j]; | |
6976 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; | |
6977 | ||
6978 | /* mark the insn, so it's recognized by insn_is_pseudo_func() */ | |
6979 | if (relo->type == RELO_SUBPROG_ADDR) | |
6980 | insn[0].src_reg = BPF_PSEUDO_FUNC; | |
9173cac3 | 6981 | } |
9173cac3 | 6982 | } |
b1268826 AS |
6983 | |
6984 | /* relocate subprogram calls and append used subprograms to main | |
c3c55696 AN |
6985 | * programs; each copy of subprogram code needs to be relocated |
6986 | * differently for each main program, because its code location might | |
b1268826 AS |
6987 | * have changed. |
6988 | * Append subprog relos to main programs to allow data relos to be | |
6989 | * processed after text is completely relocated. | |
9173cac3 | 6990 | */ |
8a47a6c5 WN |
6991 | for (i = 0; i < obj->nr_programs; i++) { |
6992 | prog = &obj->programs[i]; | |
c3c55696 AN |
6993 | /* sub-program's sub-calls are relocated within the context of |
6994 | * its main program only | |
6995 | */ | |
6996 | if (prog_is_subprog(obj, prog)) | |
9173cac3 | 6997 | continue; |
a3820c48 | 6998 | if (!prog->autoload) |
16e0c35c | 6999 | continue; |
8a47a6c5 | 7000 | |
c3c55696 | 7001 | err = bpf_object__relocate_calls(obj, prog); |
8a47a6c5 | 7002 | if (err) { |
9c0f8cbd AN |
7003 | pr_warn("prog '%s': failed to relocate calls: %d\n", |
7004 | prog->name, err); | |
8a47a6c5 WN |
7005 | return err; |
7006 | } | |
7e2925f6 | 7007 | |
fb03be7c AN |
7008 | err = bpf_prog_assign_exc_cb(obj, prog); |
7009 | if (err) | |
7010 | return err; | |
7e2925f6 KKD |
7011 | /* Now, also append exception callback if it has not been done already. */ |
7012 | if (prog->exception_cb_idx >= 0) { | |
7013 | struct bpf_program *subprog = &obj->programs[prog->exception_cb_idx]; | |
7014 | ||
7015 | /* Calling exception callback directly is disallowed, which the | |
7016 | * verifier will reject later. In case it was processed already, | |
7017 | * we can skip this step, otherwise for all other valid cases we | |
7018 | * have to append exception callback now. | |
7019 | */ | |
7020 | if (subprog->sub_insn_off == 0) { | |
7021 | err = bpf_object__append_subprog_code(obj, prog, subprog); | |
7022 | if (err) | |
7023 | return err; | |
7024 | err = bpf_object__reloc_code(obj, prog, subprog); | |
7025 | if (err) | |
7026 | return err; | |
7027 | } | |
7028 | } | |
8a47a6c5 | 7029 | } |
b1268826 AS |
7030 | for (i = 0; i < obj->nr_programs; i++) { |
7031 | prog = &obj->programs[i]; | |
7032 | if (prog_is_subprog(obj, prog)) | |
7033 | continue; | |
a3820c48 | 7034 | if (!prog->autoload) |
16e0c35c | 7035 | continue; |
2f38fe68 AN |
7036 | |
7037 | /* Process data relos for main programs */ | |
b1268826 AS |
7038 | err = bpf_object__relocate_data(obj, prog); |
7039 | if (err) { | |
7040 | pr_warn("prog '%s': failed to relocate data references: %d\n", | |
7041 | prog->name, err); | |
7042 | return err; | |
7043 | } | |
2f38fe68 AN |
7044 | |
7045 | /* Fix up .BTF.ext information, if necessary */ | |
7046 | err = bpf_program_fixup_func_info(obj, prog); | |
7047 | if (err) { | |
7048 | pr_warn("prog '%s': failed to perform .BTF.ext fix ups: %d\n", | |
7049 | prog->name, err); | |
7050 | return err; | |
7051 | } | |
b1268826 | 7052 | } |
185cfe83 | 7053 | |
8a47a6c5 WN |
7054 | return 0; |
7055 | } | |
7056 | ||
646f02ff | 7057 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 7058 | Elf64_Shdr *shdr, Elf_Data *data); |
646f02ff AN |
7059 | |
7060 | static int bpf_object__collect_map_relos(struct bpf_object *obj, | |
ad23b723 | 7061 | Elf64_Shdr *shdr, Elf_Data *data) |
646f02ff | 7062 | { |
15728ad3 AN |
7063 | const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *); |
7064 | int i, j, nrels, new_sz; | |
063e6881 | 7065 | const struct btf_var_secinfo *vi = NULL; |
646f02ff | 7066 | const struct btf_type *sec, *var, *def; |
341ac5ff HC |
7067 | struct bpf_map *map = NULL, *targ_map = NULL; |
7068 | struct bpf_program *targ_prog = NULL; | |
7069 | bool is_prog_array, is_map_in_map; | |
646f02ff | 7070 | const struct btf_member *member; |
341ac5ff | 7071 | const char *name, *mname, *type; |
646f02ff | 7072 | unsigned int moff; |
ad23b723 AN |
7073 | Elf64_Sym *sym; |
7074 | Elf64_Rel *rel; | |
646f02ff AN |
7075 | void *tmp; |
7076 | ||
7077 | if (!obj->efile.btf_maps_sec_btf_id || !obj->btf) | |
7078 | return -EINVAL; | |
7079 | sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id); | |
7080 | if (!sec) | |
7081 | return -EINVAL; | |
7082 | ||
646f02ff AN |
7083 | nrels = shdr->sh_size / shdr->sh_entsize; |
7084 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
7085 | rel = elf_rel_by_idx(data, i); |
7086 | if (!rel) { | |
646f02ff AN |
7087 | pr_warn(".maps relo #%d: failed to get ELF relo\n", i); |
7088 | return -LIBBPF_ERRNO__FORMAT; | |
7089 | } | |
ad23b723 AN |
7090 | |
7091 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); | |
7092 | if (!sym) { | |
646f02ff | 7093 | pr_warn(".maps relo #%d: symbol %zx not found\n", |
ad23b723 | 7094 | i, (size_t)ELF64_R_SYM(rel->r_info)); |
646f02ff AN |
7095 | return -LIBBPF_ERRNO__FORMAT; |
7096 | } | |
ad23b723 | 7097 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
646f02ff | 7098 | |
ad23b723 AN |
7099 | pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n", |
7100 | i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value, | |
7101 | (size_t)rel->r_offset, sym->st_name, name); | |
646f02ff AN |
7102 | |
7103 | for (j = 0; j < obj->nr_maps; j++) { | |
7104 | map = &obj->maps[j]; | |
7105 | if (map->sec_idx != obj->efile.btf_maps_shndx) | |
7106 | continue; | |
7107 | ||
7108 | vi = btf_var_secinfos(sec) + map->btf_var_idx; | |
ad23b723 AN |
7109 | if (vi->offset <= rel->r_offset && |
7110 | rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size) | |
646f02ff AN |
7111 | break; |
7112 | } | |
7113 | if (j == obj->nr_maps) { | |
ad23b723 AN |
7114 | pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n", |
7115 | i, name, (size_t)rel->r_offset); | |
646f02ff AN |
7116 | return -EINVAL; |
7117 | } | |
7118 | ||
341ac5ff HC |
7119 | is_map_in_map = bpf_map_type__is_map_in_map(map->def.type); |
7120 | is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY; | |
7121 | type = is_map_in_map ? "map" : "prog"; | |
7122 | if (is_map_in_map) { | |
7123 | if (sym->st_shndx != obj->efile.btf_maps_shndx) { | |
7124 | pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n", | |
7125 | i, name); | |
7126 | return -LIBBPF_ERRNO__RELOC; | |
7127 | } | |
7128 | if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS && | |
7129 | map->def.key_size != sizeof(int)) { | |
7130 | pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n", | |
7131 | i, map->name, sizeof(int)); | |
7132 | return -EINVAL; | |
7133 | } | |
7134 | targ_map = bpf_object__find_map_by_name(obj, name); | |
7135 | if (!targ_map) { | |
7136 | pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n", | |
7137 | i, name); | |
7138 | return -ESRCH; | |
7139 | } | |
7140 | } else if (is_prog_array) { | |
7141 | targ_prog = bpf_object__find_program_by_name(obj, name); | |
7142 | if (!targ_prog) { | |
7143 | pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n", | |
7144 | i, name); | |
7145 | return -ESRCH; | |
7146 | } | |
7147 | if (targ_prog->sec_idx != sym->st_shndx || | |
7148 | targ_prog->sec_insn_off * 8 != sym->st_value || | |
7149 | prog_is_subprog(obj, targ_prog)) { | |
7150 | pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n", | |
7151 | i, name); | |
7152 | return -LIBBPF_ERRNO__RELOC; | |
7153 | } | |
7154 | } else { | |
646f02ff AN |
7155 | return -EINVAL; |
7156 | } | |
7157 | ||
646f02ff AN |
7158 | var = btf__type_by_id(obj->btf, vi->type); |
7159 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
7160 | if (btf_vlen(def) == 0) | |
7161 | return -EINVAL; | |
7162 | member = btf_members(def) + btf_vlen(def) - 1; | |
7163 | mname = btf__name_by_offset(obj->btf, member->name_off); | |
7164 | if (strcmp(mname, "values")) | |
7165 | return -EINVAL; | |
7166 | ||
7167 | moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8; | |
ad23b723 | 7168 | if (rel->r_offset - vi->offset < moff) |
646f02ff AN |
7169 | return -EINVAL; |
7170 | ||
ad23b723 | 7171 | moff = rel->r_offset - vi->offset - moff; |
15728ad3 AN |
7172 | /* here we use BPF pointer size, which is always 64 bit, as we |
7173 | * are parsing ELF that was built for BPF target | |
7174 | */ | |
7175 | if (moff % bpf_ptr_sz) | |
646f02ff | 7176 | return -EINVAL; |
15728ad3 | 7177 | moff /= bpf_ptr_sz; |
646f02ff AN |
7178 | if (moff >= map->init_slots_sz) { |
7179 | new_sz = moff + 1; | |
029258d7 | 7180 | tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz); |
646f02ff AN |
7181 | if (!tmp) |
7182 | return -ENOMEM; | |
7183 | map->init_slots = tmp; | |
7184 | memset(map->init_slots + map->init_slots_sz, 0, | |
15728ad3 | 7185 | (new_sz - map->init_slots_sz) * host_ptr_sz); |
646f02ff AN |
7186 | map->init_slots_sz = new_sz; |
7187 | } | |
341ac5ff | 7188 | map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog; |
646f02ff | 7189 | |
341ac5ff HC |
7190 | pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n", |
7191 | i, map->name, moff, type, name); | |
646f02ff AN |
7192 | } |
7193 | ||
7194 | return 0; | |
7195 | } | |
590a0088 | 7196 | |
c3c55696 AN |
7197 | static int bpf_object__collect_relos(struct bpf_object *obj) |
7198 | { | |
7199 | int i, err; | |
34090915 | 7200 | |
25bbbd7a AN |
7201 | for (i = 0; i < obj->efile.sec_cnt; i++) { |
7202 | struct elf_sec_desc *sec_desc = &obj->efile.secs[i]; | |
7203 | Elf64_Shdr *shdr; | |
7204 | Elf_Data *data; | |
7205 | int idx; | |
7206 | ||
7207 | if (sec_desc->sec_type != SEC_RELO) | |
7208 | continue; | |
7209 | ||
7210 | shdr = sec_desc->shdr; | |
7211 | data = sec_desc->data; | |
7212 | idx = shdr->sh_info; | |
34090915 | 7213 | |
240bf8a5 | 7214 | if (shdr->sh_type != SHT_REL || idx < 0 || idx >= obj->efile.sec_cnt) { |
be18010e | 7215 | pr_warn("internal error at %d\n", __LINE__); |
6371ca3b | 7216 | return -LIBBPF_ERRNO__INTERNAL; |
34090915 WN |
7217 | } |
7218 | ||
240bf8a5 | 7219 | if (obj->efile.secs[idx].sec_type == SEC_ST_OPS) |
646f02ff | 7220 | err = bpf_object__collect_st_ops_relos(obj, shdr, data); |
c3c55696 | 7221 | else if (idx == obj->efile.btf_maps_shndx) |
646f02ff | 7222 | err = bpf_object__collect_map_relos(obj, shdr, data); |
c3c55696 AN |
7223 | else |
7224 | err = bpf_object__collect_prog_relos(obj, shdr, data); | |
34090915 | 7225 | if (err) |
6371ca3b | 7226 | return err; |
34090915 | 7227 | } |
c3c55696 | 7228 | |
d0e92887 | 7229 | bpf_object__sort_relos(obj); |
34090915 WN |
7230 | return 0; |
7231 | } | |
7232 | ||
109cea5a AN |
7233 | static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id) |
7234 | { | |
9b2f6fec | 7235 | if (BPF_CLASS(insn->code) == BPF_JMP && |
109cea5a AN |
7236 | BPF_OP(insn->code) == BPF_CALL && |
7237 | BPF_SRC(insn->code) == BPF_K && | |
9b2f6fec AN |
7238 | insn->src_reg == 0 && |
7239 | insn->dst_reg == 0) { | |
7240 | *func_id = insn->imm; | |
109cea5a AN |
7241 | return true; |
7242 | } | |
7243 | return false; | |
7244 | } | |
7245 | ||
42869d28 | 7246 | static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog) |
109cea5a AN |
7247 | { |
7248 | struct bpf_insn *insn = prog->insns; | |
7249 | enum bpf_func_id func_id; | |
7250 | int i; | |
7251 | ||
67234743 AS |
7252 | if (obj->gen_loader) |
7253 | return 0; | |
7254 | ||
109cea5a AN |
7255 | for (i = 0; i < prog->insns_cnt; i++, insn++) { |
7256 | if (!insn_is_helper_call(insn, &func_id)) | |
7257 | continue; | |
7258 | ||
7259 | /* on kernels that don't yet support | |
7260 | * bpf_probe_read_{kernel,user}[_str] helpers, fall back | |
7261 | * to bpf_probe_read() which works well for old kernels | |
7262 | */ | |
7263 | switch (func_id) { | |
7264 | case BPF_FUNC_probe_read_kernel: | |
7265 | case BPF_FUNC_probe_read_user: | |
9ca1f56a | 7266 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
7267 | insn->imm = BPF_FUNC_probe_read; |
7268 | break; | |
7269 | case BPF_FUNC_probe_read_kernel_str: | |
7270 | case BPF_FUNC_probe_read_user_str: | |
9ca1f56a | 7271 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
7272 | insn->imm = BPF_FUNC_probe_read_str; |
7273 | break; | |
7274 | default: | |
7275 | break; | |
7276 | } | |
7277 | } | |
7278 | return 0; | |
7279 | } | |
7280 | ||
15ea31fa AN |
7281 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
7282 | int *btf_obj_fd, int *btf_type_id); | |
12d9466d | 7283 | |
4fa5bcfe AN |
7284 | /* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */ |
7285 | static int libbpf_prepare_prog_load(struct bpf_program *prog, | |
7286 | struct bpf_prog_load_opts *opts, long cookie) | |
12d9466d | 7287 | { |
15ea31fa AN |
7288 | enum sec_def_flags def = cookie; |
7289 | ||
12d9466d | 7290 | /* old kernels might not support specifying expected_attach_type */ |
15ea31fa | 7291 | if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE)) |
d10ef2b8 | 7292 | opts->expected_attach_type = 0; |
12d9466d | 7293 | |
15ea31fa | 7294 | if (def & SEC_SLEEPABLE) |
d10ef2b8 | 7295 | opts->prog_flags |= BPF_F_SLEEPABLE; |
12d9466d | 7296 | |
082c4bfb LB |
7297 | if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS)) |
7298 | opts->prog_flags |= BPF_F_XDP_HAS_FRAGS; | |
7299 | ||
5902da6d JO |
7300 | /* special check for usdt to use uprobe_multi link */ |
7301 | if ((def & SEC_USDT) && kernel_supports(prog->obj, FEAT_UPROBE_MULTI_LINK)) | |
7302 | prog->expected_attach_type = BPF_TRACE_UPROBE_MULTI; | |
7303 | ||
cc7d8f2c | 7304 | if ((def & SEC_ATTACH_BTF) && !prog->attach_btf_id) { |
12d9466d | 7305 | int btf_obj_fd = 0, btf_type_id = 0, err; |
15ea31fa | 7306 | const char *attach_name; |
12d9466d | 7307 | |
cc7d8f2c AN |
7308 | attach_name = strchr(prog->sec_name, '/'); |
7309 | if (!attach_name) { | |
7310 | /* if BPF program is annotated with just SEC("fentry") | |
7311 | * (or similar) without declaratively specifying | |
7312 | * target, then it is expected that target will be | |
7313 | * specified with bpf_program__set_attach_target() at | |
7314 | * runtime before BPF object load step. If not, then | |
7315 | * there is nothing to load into the kernel as BPF | |
7316 | * verifier won't be able to validate BPF program | |
7317 | * correctness anyways. | |
7318 | */ | |
7319 | pr_warn("prog '%s': no BTF-based attach target is specified, use bpf_program__set_attach_target()\n", | |
7320 | prog->name); | |
7321 | return -EINVAL; | |
7322 | } | |
7323 | attach_name++; /* skip over / */ | |
7324 | ||
15ea31fa | 7325 | err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id); |
12d9466d AN |
7326 | if (err) |
7327 | return err; | |
7328 | ||
7329 | /* cache resolved BTF FD and BTF type ID in the prog */ | |
7330 | prog->attach_btf_obj_fd = btf_obj_fd; | |
7331 | prog->attach_btf_id = btf_type_id; | |
7332 | ||
7333 | /* but by now libbpf common logic is not utilizing | |
7334 | * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because | |
d10ef2b8 AN |
7335 | * this callback is called after opts were populated by |
7336 | * libbpf, so this callback has to update opts explicitly here | |
12d9466d | 7337 | */ |
d10ef2b8 AN |
7338 | opts->attach_btf_obj_fd = btf_obj_fd; |
7339 | opts->attach_btf_id = btf_type_id; | |
12d9466d AN |
7340 | } |
7341 | return 0; | |
7342 | } | |
7343 | ||
9fdc4273 AN |
7344 | static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz); |
7345 | ||
cf90a20d AN |
7346 | static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog, |
7347 | struct bpf_insn *insns, int insns_cnt, | |
7348 | const char *license, __u32 kern_version, int *prog_fd) | |
55cffde2 | 7349 | { |
d10ef2b8 AN |
7350 | LIBBPF_OPTS(bpf_prog_load_opts, load_attr); |
7351 | const char *prog_name = NULL; | |
1ce6a9fc | 7352 | char *cp, errmsg[STRERR_BUFSIZE]; |
8395f320 | 7353 | size_t log_buf_size = 0; |
b3ce9079 | 7354 | char *log_buf = NULL, *tmp; |
b3ce9079 AN |
7355 | bool own_log_buf = true; |
7356 | __u32 log_level = prog->log_level; | |
9bf48fa1 | 7357 | int ret, err; |
55cffde2 | 7358 | |
80b2b5c3 AM |
7359 | if (prog->type == BPF_PROG_TYPE_UNSPEC) { |
7360 | /* | |
7361 | * The program type must be set. Most likely we couldn't find a proper | |
7362 | * section definition at load time, and thus we didn't infer the type. | |
7363 | */ | |
7364 | pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n", | |
7365 | prog->name, prog->sec_name); | |
7366 | return -EINVAL; | |
7367 | } | |
7368 | ||
fba01a06 AN |
7369 | if (!insns || !insns_cnt) |
7370 | return -EINVAL; | |
7371 | ||
25bbbd7a | 7372 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
d10ef2b8 | 7373 | prog_name = prog->name; |
12d9466d AN |
7374 | load_attr.attach_prog_fd = prog->attach_prog_fd; |
7375 | load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd; | |
6aef10a4 AN |
7376 | load_attr.attach_btf_id = prog->attach_btf_id; |
7377 | load_attr.kern_version = kern_version; | |
7378 | load_attr.prog_ifindex = prog->prog_ifindex; | |
7379 | ||
0f0e55d8 | 7380 | /* specify func_info/line_info only if kernel supports them */ |
9bf48fa1 QM |
7381 | if (obj->btf && btf__fd(obj->btf) >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) { |
7382 | load_attr.prog_btf_fd = btf__fd(obj->btf); | |
0f0e55d8 AN |
7383 | load_attr.func_info = prog->func_info; |
7384 | load_attr.func_info_rec_size = prog->func_info_rec_size; | |
7385 | load_attr.func_info_cnt = prog->func_info_cnt; | |
7386 | load_attr.line_info = prog->line_info; | |
7387 | load_attr.line_info_rec_size = prog->line_info_rec_size; | |
7388 | load_attr.line_info_cnt = prog->line_info_cnt; | |
7389 | } | |
b3ce9079 | 7390 | load_attr.log_level = log_level; |
04656198 | 7391 | load_attr.prog_flags = prog->prog_flags; |
25bbbd7a | 7392 | load_attr.fd_array = obj->fd_array; |
55cffde2 | 7393 | |
6b434b61 AN |
7394 | load_attr.token_fd = obj->token_fd; |
7395 | if (obj->token_fd) | |
7396 | load_attr.prog_flags |= BPF_F_TOKEN_FD; | |
7397 | ||
12d9466d | 7398 | /* adjust load_attr if sec_def provides custom preload callback */ |
4fa5bcfe AN |
7399 | if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) { |
7400 | err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie); | |
12d9466d AN |
7401 | if (err < 0) { |
7402 | pr_warn("prog '%s': failed to prepare load attributes: %d\n", | |
7403 | prog->name, err); | |
7404 | return err; | |
7405 | } | |
b63b3c49 JO |
7406 | insns = prog->insns; |
7407 | insns_cnt = prog->insns_cnt; | |
12d9466d AN |
7408 | } |
7409 | ||
5902da6d JO |
7410 | /* allow prog_prepare_load_fn to change expected_attach_type */ |
7411 | load_attr.expected_attach_type = prog->expected_attach_type; | |
7412 | ||
25bbbd7a | 7413 | if (obj->gen_loader) { |
d10ef2b8 AN |
7414 | bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name, |
7415 | license, insns, insns_cnt, &load_attr, | |
25bbbd7a | 7416 | prog - obj->programs); |
be2f2d16 | 7417 | *prog_fd = -1; |
67234743 AS |
7418 | return 0; |
7419 | } | |
8395f320 | 7420 | |
b3ce9079 | 7421 | retry_load: |
bb412cf1 | 7422 | /* if log_level is zero, we don't request logs initially even if |
b3ce9079 AN |
7423 | * custom log_buf is specified; if the program load fails, then we'll |
7424 | * bump log_level to 1 and use either custom log_buf or we'll allocate | |
7425 | * our own and retry the load to get details on what failed | |
7426 | */ | |
7427 | if (log_level) { | |
7428 | if (prog->log_buf) { | |
7429 | log_buf = prog->log_buf; | |
7430 | log_buf_size = prog->log_size; | |
7431 | own_log_buf = false; | |
7432 | } else if (obj->log_buf) { | |
7433 | log_buf = obj->log_buf; | |
7434 | log_buf_size = obj->log_size; | |
7435 | own_log_buf = false; | |
7436 | } else { | |
7437 | log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2); | |
7438 | tmp = realloc(log_buf, log_buf_size); | |
7439 | if (!tmp) { | |
7440 | ret = -ENOMEM; | |
7441 | goto out; | |
7442 | } | |
7443 | log_buf = tmp; | |
7444 | log_buf[0] = '\0'; | |
7445 | own_log_buf = true; | |
7446 | } | |
8395f320 | 7447 | } |
55cffde2 | 7448 | |
6aef10a4 | 7449 | load_attr.log_buf = log_buf; |
d10ef2b8 | 7450 | load_attr.log_size = log_buf_size; |
b3ce9079 | 7451 | load_attr.log_level = log_level; |
55cffde2 | 7452 | |
b3ce9079 | 7453 | ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr); |
55cffde2 | 7454 | if (ret >= 0) { |
b3ce9079 | 7455 | if (log_level && own_log_buf) { |
ad9a7f96 AN |
7456 | pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n", |
7457 | prog->name, log_buf); | |
7458 | } | |
5d23328d | 7459 | |
25bbbd7a AN |
7460 | if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) { |
7461 | struct bpf_map *map; | |
7462 | int i; | |
7463 | ||
7464 | for (i = 0; i < obj->nr_maps; i++) { | |
7465 | map = &prog->obj->maps[i]; | |
7466 | if (map->libbpf_type != LIBBPF_MAP_RODATA) | |
7467 | continue; | |
5d23328d | 7468 | |
f08c18e0 | 7469 | if (bpf_prog_bind_map(ret, map->fd, NULL)) { |
25bbbd7a | 7470 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
ad9a7f96 AN |
7471 | pr_warn("prog '%s': failed to bind map '%s': %s\n", |
7472 | prog->name, map->real_name, cp); | |
25bbbd7a AN |
7473 | /* Don't fail hard if can't bind rodata. */ |
7474 | } | |
5d23328d YZ |
7475 | } |
7476 | } | |
7477 | ||
be2f2d16 | 7478 | *prog_fd = ret; |
55cffde2 WN |
7479 | ret = 0; |
7480 | goto out; | |
7481 | } | |
7482 | ||
b3ce9079 AN |
7483 | if (log_level == 0) { |
7484 | log_level = 1; | |
da11b417 AS |
7485 | goto retry_load; |
7486 | } | |
b3ce9079 AN |
7487 | /* On ENOSPC, increase log buffer size and retry, unless custom |
7488 | * log_buf is specified. | |
7489 | * Be careful to not overflow u32, though. Kernel's log buf size limit | |
7490 | * isn't part of UAPI so it can always be bumped to full 4GB. So don't | |
7491 | * multiply by 2 unless we are sure we'll fit within 32 bits. | |
7492 | * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2). | |
7493 | */ | |
7494 | if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2) | |
7495 | goto retry_load; | |
2eda2145 AN |
7496 | |
7497 | ret = -errno; | |
9fdc4273 AN |
7498 | |
7499 | /* post-process verifier log to improve error descriptions */ | |
7500 | fixup_verifier_log(prog, log_buf, log_buf_size); | |
7501 | ||
24d6a808 | 7502 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
ad9a7f96 | 7503 | pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, cp); |
dc3a2d25 | 7504 | pr_perm_msg(ret); |
55cffde2 | 7505 | |
b3ce9079 | 7506 | if (own_log_buf && log_buf && log_buf[0] != '\0') { |
ad9a7f96 AN |
7507 | pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n", |
7508 | prog->name, log_buf); | |
7509 | } | |
55cffde2 WN |
7510 | |
7511 | out: | |
b3ce9079 AN |
7512 | if (own_log_buf) |
7513 | free(log_buf); | |
55cffde2 WN |
7514 | return ret; |
7515 | } | |
7516 | ||
9fdc4273 AN |
7517 | static char *find_prev_line(char *buf, char *cur) |
7518 | { | |
7519 | char *p; | |
7520 | ||
7521 | if (cur == buf) /* end of a log buf */ | |
7522 | return NULL; | |
7523 | ||
7524 | p = cur - 1; | |
7525 | while (p - 1 >= buf && *(p - 1) != '\n') | |
7526 | p--; | |
7527 | ||
7528 | return p; | |
7529 | } | |
7530 | ||
7531 | static void patch_log(char *buf, size_t buf_sz, size_t log_sz, | |
7532 | char *orig, size_t orig_sz, const char *patch) | |
7533 | { | |
7534 | /* size of the remaining log content to the right from the to-be-replaced part */ | |
7535 | size_t rem_sz = (buf + log_sz) - (orig + orig_sz); | |
7536 | size_t patch_sz = strlen(patch); | |
7537 | ||
7538 | if (patch_sz != orig_sz) { | |
7539 | /* If patch line(s) are longer than original piece of verifier log, | |
7540 | * shift log contents by (patch_sz - orig_sz) bytes to the right | |
7541 | * starting from after to-be-replaced part of the log. | |
7542 | * | |
7543 | * If patch line(s) are shorter than original piece of verifier log, | |
7544 | * shift log contents by (orig_sz - patch_sz) bytes to the left | |
7545 | * starting from after to-be-replaced part of the log | |
7546 | * | |
7547 | * We need to be careful about not overflowing available | |
7548 | * buf_sz capacity. If that's the case, we'll truncate the end | |
7549 | * of the original log, as necessary. | |
7550 | */ | |
7551 | if (patch_sz > orig_sz) { | |
7552 | if (orig + patch_sz >= buf + buf_sz) { | |
7553 | /* patch is big enough to cover remaining space completely */ | |
7554 | patch_sz -= (orig + patch_sz) - (buf + buf_sz) + 1; | |
7555 | rem_sz = 0; | |
7556 | } else if (patch_sz - orig_sz > buf_sz - log_sz) { | |
7557 | /* patch causes part of remaining log to be truncated */ | |
7558 | rem_sz -= (patch_sz - orig_sz) - (buf_sz - log_sz); | |
7559 | } | |
7560 | } | |
7561 | /* shift remaining log to the right by calculated amount */ | |
7562 | memmove(orig + patch_sz, orig + orig_sz, rem_sz); | |
7563 | } | |
7564 | ||
7565 | memcpy(orig, patch, patch_sz); | |
7566 | } | |
7567 | ||
7568 | static void fixup_log_failed_core_relo(struct bpf_program *prog, | |
7569 | char *buf, size_t buf_sz, size_t log_sz, | |
7570 | char *line1, char *line2, char *line3) | |
7571 | { | |
7572 | /* Expected log for failed and not properly guarded CO-RE relocation: | |
7573 | * line1 -> 123: (85) call unknown#195896080 | |
7574 | * line2 -> invalid func unknown#195896080 | |
7575 | * line3 -> <anything else or end of buffer> | |
7576 | * | |
7577 | * "123" is the index of the instruction that was poisoned. We extract | |
7578 | * instruction index to find corresponding CO-RE relocation and | |
7579 | * replace this part of the log with more relevant information about | |
7580 | * failed CO-RE relocation. | |
7581 | */ | |
7582 | const struct bpf_core_relo *relo; | |
7583 | struct bpf_core_spec spec; | |
7584 | char patch[512], spec_buf[256]; | |
b198881d | 7585 | int insn_idx, err, spec_len; |
9fdc4273 AN |
7586 | |
7587 | if (sscanf(line1, "%d: (%*d) call unknown#195896080\n", &insn_idx) != 1) | |
7588 | return; | |
7589 | ||
7590 | relo = find_relo_core(prog, insn_idx); | |
7591 | if (!relo) | |
7592 | return; | |
7593 | ||
7594 | err = bpf_core_parse_spec(prog->name, prog->obj->btf, relo, &spec); | |
7595 | if (err) | |
7596 | return; | |
7597 | ||
b198881d | 7598 | spec_len = bpf_core_format_spec(spec_buf, sizeof(spec_buf), &spec); |
9fdc4273 AN |
7599 | snprintf(patch, sizeof(patch), |
7600 | "%d: <invalid CO-RE relocation>\n" | |
b198881d AN |
7601 | "failed to resolve CO-RE relocation %s%s\n", |
7602 | insn_idx, spec_buf, spec_len >= sizeof(spec_buf) ? "..." : ""); | |
9fdc4273 AN |
7603 | |
7604 | patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch); | |
7605 | } | |
7606 | ||
ec41817b AN |
7607 | static void fixup_log_missing_map_load(struct bpf_program *prog, |
7608 | char *buf, size_t buf_sz, size_t log_sz, | |
7609 | char *line1, char *line2, char *line3) | |
7610 | { | |
3055ddd6 | 7611 | /* Expected log for failed and not properly guarded map reference: |
ec41817b AN |
7612 | * line1 -> 123: (85) call unknown#2001000345 |
7613 | * line2 -> invalid func unknown#2001000345 | |
7614 | * line3 -> <anything else or end of buffer> | |
7615 | * | |
7616 | * "123" is the index of the instruction that was poisoned. | |
3055ddd6 | 7617 | * "345" in "2001000345" is a map index in obj->maps to fetch map name. |
ec41817b AN |
7618 | */ |
7619 | struct bpf_object *obj = prog->obj; | |
7620 | const struct bpf_map *map; | |
7621 | int insn_idx, map_idx; | |
7622 | char patch[128]; | |
7623 | ||
7624 | if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &map_idx) != 2) | |
7625 | return; | |
7626 | ||
3055ddd6 | 7627 | map_idx -= POISON_LDIMM64_MAP_BASE; |
ec41817b AN |
7628 | if (map_idx < 0 || map_idx >= obj->nr_maps) |
7629 | return; | |
7630 | map = &obj->maps[map_idx]; | |
7631 | ||
7632 | snprintf(patch, sizeof(patch), | |
7633 | "%d: <invalid BPF map reference>\n" | |
7634 | "BPF map '%s' is referenced but wasn't created\n", | |
7635 | insn_idx, map->name); | |
7636 | ||
7637 | patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch); | |
7638 | } | |
7639 | ||
05b6f766 AN |
7640 | static void fixup_log_missing_kfunc_call(struct bpf_program *prog, |
7641 | char *buf, size_t buf_sz, size_t log_sz, | |
7642 | char *line1, char *line2, char *line3) | |
7643 | { | |
7644 | /* Expected log for failed and not properly guarded kfunc call: | |
7645 | * line1 -> 123: (85) call unknown#2002000345 | |
7646 | * line2 -> invalid func unknown#2002000345 | |
7647 | * line3 -> <anything else or end of buffer> | |
7648 | * | |
7649 | * "123" is the index of the instruction that was poisoned. | |
7650 | * "345" in "2002000345" is an extern index in obj->externs to fetch kfunc name. | |
7651 | */ | |
7652 | struct bpf_object *obj = prog->obj; | |
7653 | const struct extern_desc *ext; | |
7654 | int insn_idx, ext_idx; | |
7655 | char patch[128]; | |
7656 | ||
7657 | if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &ext_idx) != 2) | |
7658 | return; | |
7659 | ||
7660 | ext_idx -= POISON_CALL_KFUNC_BASE; | |
7661 | if (ext_idx < 0 || ext_idx >= obj->nr_extern) | |
7662 | return; | |
7663 | ext = &obj->externs[ext_idx]; | |
7664 | ||
7665 | snprintf(patch, sizeof(patch), | |
7666 | "%d: <invalid kfunc call>\n" | |
7667 | "kfunc '%s' is referenced but wasn't resolved\n", | |
7668 | insn_idx, ext->name); | |
7669 | ||
7670 | patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch); | |
7671 | } | |
7672 | ||
9fdc4273 AN |
7673 | static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz) |
7674 | { | |
7675 | /* look for familiar error patterns in last N lines of the log */ | |
7676 | const size_t max_last_line_cnt = 10; | |
7677 | char *prev_line, *cur_line, *next_line; | |
7678 | size_t log_sz; | |
7679 | int i; | |
7680 | ||
7681 | if (!buf) | |
7682 | return; | |
7683 | ||
7684 | log_sz = strlen(buf) + 1; | |
7685 | next_line = buf + log_sz - 1; | |
7686 | ||
7687 | for (i = 0; i < max_last_line_cnt; i++, next_line = cur_line) { | |
7688 | cur_line = find_prev_line(buf, next_line); | |
7689 | if (!cur_line) | |
7690 | return; | |
7691 | ||
9fdc4273 AN |
7692 | if (str_has_pfx(cur_line, "invalid func unknown#195896080\n")) { |
7693 | prev_line = find_prev_line(buf, cur_line); | |
7694 | if (!prev_line) | |
7695 | continue; | |
7696 | ||
3055ddd6 | 7697 | /* failed CO-RE relocation case */ |
9fdc4273 AN |
7698 | fixup_log_failed_core_relo(prog, buf, buf_sz, log_sz, |
7699 | prev_line, cur_line, next_line); | |
7700 | return; | |
3055ddd6 | 7701 | } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_LDIMM64_MAP_PFX)) { |
ec41817b AN |
7702 | prev_line = find_prev_line(buf, cur_line); |
7703 | if (!prev_line) | |
7704 | continue; | |
7705 | ||
3055ddd6 | 7706 | /* reference to uncreated BPF map */ |
ec41817b AN |
7707 | fixup_log_missing_map_load(prog, buf, buf_sz, log_sz, |
7708 | prev_line, cur_line, next_line); | |
7709 | return; | |
05b6f766 AN |
7710 | } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_CALL_KFUNC_PFX)) { |
7711 | prev_line = find_prev_line(buf, cur_line); | |
7712 | if (!prev_line) | |
7713 | continue; | |
7714 | ||
7715 | /* reference to unresolved kfunc */ | |
7716 | fixup_log_missing_kfunc_call(prog, buf, buf_sz, log_sz, | |
7717 | prev_line, cur_line, next_line); | |
7718 | return; | |
9fdc4273 AN |
7719 | } |
7720 | } | |
7721 | } | |
7722 | ||
d0e92887 | 7723 | static int bpf_program_record_relos(struct bpf_program *prog) |
67234743 AS |
7724 | { |
7725 | struct bpf_object *obj = prog->obj; | |
7726 | int i; | |
7727 | ||
7728 | for (i = 0; i < prog->nr_reloc; i++) { | |
7729 | struct reloc_desc *relo = &prog->reloc_desc[i]; | |
3055ddd6 | 7730 | struct extern_desc *ext = &obj->externs[relo->ext_idx]; |
708cdc57 | 7731 | int kind; |
67234743 AS |
7732 | |
7733 | switch (relo->type) { | |
a18f7214 | 7734 | case RELO_EXTERN_LD64: |
67234743 AS |
7735 | if (ext->type != EXT_KSYM) |
7736 | continue; | |
708cdc57 AS |
7737 | kind = btf_is_var(btf__type_by_id(obj->btf, ext->btf_id)) ? |
7738 | BTF_KIND_VAR : BTF_KIND_FUNC; | |
c24941cd KKD |
7739 | bpf_gen__record_extern(obj->gen_loader, ext->name, |
7740 | ext->is_weak, !ext->ksym.type_id, | |
708cdc57 | 7741 | true, kind, relo->insn_idx); |
67234743 | 7742 | break; |
a18f7214 | 7743 | case RELO_EXTERN_CALL: |
c24941cd | 7744 | bpf_gen__record_extern(obj->gen_loader, ext->name, |
708cdc57 | 7745 | ext->is_weak, false, false, BTF_KIND_FUNC, |
67234743 AS |
7746 | relo->insn_idx); |
7747 | break; | |
d0e92887 AS |
7748 | case RELO_CORE: { |
7749 | struct bpf_core_relo cr = { | |
7750 | .insn_off = relo->insn_idx * 8, | |
7751 | .type_id = relo->core_relo->type_id, | |
7752 | .access_str_off = relo->core_relo->access_str_off, | |
7753 | .kind = relo->core_relo->kind, | |
7754 | }; | |
7755 | ||
7756 | bpf_gen__record_relo_core(obj->gen_loader, &cr); | |
7757 | break; | |
7758 | } | |
67234743 AS |
7759 | default: |
7760 | continue; | |
7761 | } | |
7762 | } | |
7763 | return 0; | |
7764 | } | |
7765 | ||
55cffde2 | 7766 | static int |
60276f98 | 7767 | bpf_object__load_progs(struct bpf_object *obj, int log_level) |
55cffde2 | 7768 | { |
d9297581 | 7769 | struct bpf_program *prog; |
55cffde2 WN |
7770 | size_t i; |
7771 | int err; | |
7772 | ||
109cea5a AN |
7773 | for (i = 0; i < obj->nr_programs; i++) { |
7774 | prog = &obj->programs[i]; | |
7775 | err = bpf_object__sanitize_prog(obj, prog); | |
7776 | if (err) | |
7777 | return err; | |
7778 | } | |
7779 | ||
55cffde2 | 7780 | for (i = 0; i < obj->nr_programs; i++) { |
d9297581 | 7781 | prog = &obj->programs[i]; |
c3c55696 | 7782 | if (prog_is_subprog(obj, prog)) |
48cca7e4 | 7783 | continue; |
a3820c48 | 7784 | if (!prog->autoload) { |
9c0f8cbd | 7785 | pr_debug("prog '%s': skipped loading\n", prog->name); |
d9297581 AN |
7786 | continue; |
7787 | } | |
7788 | prog->log_level |= log_level; | |
cf90a20d AN |
7789 | |
7790 | if (obj->gen_loader) | |
7791 | bpf_program_record_relos(prog); | |
7792 | ||
7793 | err = bpf_object_load_prog(obj, prog, prog->insns, prog->insns_cnt, | |
7794 | obj->license, obj->kern_version, &prog->fd); | |
7795 | if (err) { | |
7796 | pr_warn("prog '%s': failed to load: %d\n", prog->name, err); | |
55cffde2 | 7797 | return err; |
cf90a20d | 7798 | } |
55cffde2 | 7799 | } |
185cfe83 AN |
7800 | |
7801 | bpf_object__free_relocs(obj); | |
55cffde2 WN |
7802 | return 0; |
7803 | } | |
7804 | ||
25498a19 AN |
7805 | static const struct bpf_sec_def *find_sec_def(const char *sec_name); |
7806 | ||
91b4d1d1 AN |
7807 | static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts) |
7808 | { | |
7809 | struct bpf_program *prog; | |
12d9466d | 7810 | int err; |
91b4d1d1 AN |
7811 | |
7812 | bpf_object__for_each_program(prog, obj) { | |
7813 | prog->sec_def = find_sec_def(prog->sec_name); | |
7814 | if (!prog->sec_def) { | |
7815 | /* couldn't guess, but user might manually specify */ | |
7816 | pr_debug("prog '%s': unrecognized ELF section name '%s'\n", | |
7817 | prog->name, prog->sec_name); | |
7818 | continue; | |
7819 | } | |
7820 | ||
df286716 GS |
7821 | prog->type = prog->sec_def->prog_type; |
7822 | prog->expected_attach_type = prog->sec_def->expected_attach_type; | |
91b4d1d1 | 7823 | |
12d9466d AN |
7824 | /* sec_def can have custom callback which should be called |
7825 | * after bpf_program is initialized to adjust its properties | |
7826 | */ | |
4fa5bcfe AN |
7827 | if (prog->sec_def->prog_setup_fn) { |
7828 | err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie); | |
12d9466d AN |
7829 | if (err < 0) { |
7830 | pr_warn("prog '%s': failed to initialize: %d\n", | |
7831 | prog->name, err); | |
7832 | return err; | |
7833 | } | |
7834 | } | |
91b4d1d1 AN |
7835 | } |
7836 | ||
7837 | return 0; | |
7838 | } | |
7839 | ||
ad9a7f96 AN |
7840 | static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz, |
7841 | const struct bpf_object_open_opts *opts) | |
1a5e3fb1 | 7842 | { |
6b434b61 | 7843 | const char *obj_name, *kconfig, *btf_tmp_path, *token_path; |
1a5e3fb1 | 7844 | struct bpf_object *obj; |
291ee02b | 7845 | char tmp_name[64]; |
d17aff80 | 7846 | int err; |
e0e3ea88 AN |
7847 | char *log_buf; |
7848 | size_t log_size; | |
7849 | __u32 log_level; | |
1a5e3fb1 WN |
7850 | |
7851 | if (elf_version(EV_CURRENT) == EV_NONE) { | |
be18010e KW |
7852 | pr_warn("failed to init libelf for %s\n", |
7853 | path ? : "(mem buf)"); | |
6371ca3b | 7854 | return ERR_PTR(-LIBBPF_ERRNO__LIBELF); |
1a5e3fb1 WN |
7855 | } |
7856 | ||
291ee02b AN |
7857 | if (!OPTS_VALID(opts, bpf_object_open_opts)) |
7858 | return ERR_PTR(-EINVAL); | |
7859 | ||
1aace10f | 7860 | obj_name = OPTS_GET(opts, object_name, NULL); |
291ee02b AN |
7861 | if (obj_buf) { |
7862 | if (!obj_name) { | |
7863 | snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx", | |
7864 | (unsigned long)obj_buf, | |
7865 | (unsigned long)obj_buf_sz); | |
7866 | obj_name = tmp_name; | |
7867 | } | |
7868 | path = obj_name; | |
7869 | pr_debug("loading object '%s' from buffer\n", obj_name); | |
7870 | } | |
7871 | ||
e0e3ea88 AN |
7872 | log_buf = OPTS_GET(opts, kernel_log_buf, NULL); |
7873 | log_size = OPTS_GET(opts, kernel_log_size, 0); | |
7874 | log_level = OPTS_GET(opts, kernel_log_level, 0); | |
7875 | if (log_size > UINT_MAX) | |
7876 | return ERR_PTR(-EINVAL); | |
7877 | if (log_size && !log_buf) | |
7878 | return ERR_PTR(-EINVAL); | |
7879 | ||
6b434b61 | 7880 | token_path = OPTS_GET(opts, bpf_token_path, NULL); |
cac270ad AN |
7881 | /* if user didn't specify bpf_token_path explicitly, check if |
7882 | * LIBBPF_BPF_TOKEN_PATH envvar was set and treat it as bpf_token_path | |
7883 | * option | |
7884 | */ | |
7885 | if (!token_path) | |
7886 | token_path = getenv("LIBBPF_BPF_TOKEN_PATH"); | |
6b434b61 AN |
7887 | if (token_path && strlen(token_path) >= PATH_MAX) |
7888 | return ERR_PTR(-ENAMETOOLONG); | |
7889 | ||
2ce8450e | 7890 | obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name); |
6371ca3b WN |
7891 | if (IS_ERR(obj)) |
7892 | return obj; | |
1a5e3fb1 | 7893 | |
e0e3ea88 AN |
7894 | obj->log_buf = log_buf; |
7895 | obj->log_size = log_size; | |
7896 | obj->log_level = log_level; | |
7897 | ||
6b434b61 AN |
7898 | if (token_path) { |
7899 | obj->token_path = strdup(token_path); | |
7900 | if (!obj->token_path) { | |
7901 | err = -ENOMEM; | |
7902 | goto out; | |
7903 | } | |
7904 | } | |
7905 | ||
1373ff59 SC |
7906 | btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL); |
7907 | if (btf_tmp_path) { | |
7908 | if (strlen(btf_tmp_path) >= PATH_MAX) { | |
7909 | err = -ENAMETOOLONG; | |
7910 | goto out; | |
7911 | } | |
7912 | obj->btf_custom_path = strdup(btf_tmp_path); | |
7913 | if (!obj->btf_custom_path) { | |
7914 | err = -ENOMEM; | |
7915 | goto out; | |
7916 | } | |
7917 | } | |
7918 | ||
8601fd42 AN |
7919 | kconfig = OPTS_GET(opts, kconfig, NULL); |
7920 | if (kconfig) { | |
7921 | obj->kconfig = strdup(kconfig); | |
18353c87 SC |
7922 | if (!obj->kconfig) { |
7923 | err = -ENOMEM; | |
7924 | goto out; | |
7925 | } | |
166750bc | 7926 | } |
291ee02b | 7927 | |
0d13bfce AN |
7928 | err = bpf_object__elf_init(obj); |
7929 | err = err ? : bpf_object__check_endianness(obj); | |
7930 | err = err ? : bpf_object__elf_collect(obj); | |
166750bc | 7931 | err = err ? : bpf_object__collect_externs(obj); |
f33f742d | 7932 | err = err ? : bpf_object_fixup_btf(obj); |
0d13bfce | 7933 | err = err ? : bpf_object__init_maps(obj, opts); |
91b4d1d1 | 7934 | err = err ? : bpf_object_init_progs(obj, opts); |
c3c55696 | 7935 | err = err ? : bpf_object__collect_relos(obj); |
0d13bfce AN |
7936 | if (err) |
7937 | goto out; | |
dd4436bb | 7938 | |
91b4d1d1 | 7939 | bpf_object__elf_finish(obj); |
dd4436bb | 7940 | |
1a5e3fb1 WN |
7941 | return obj; |
7942 | out: | |
7943 | bpf_object__close(obj); | |
6371ca3b | 7944 | return ERR_PTR(err); |
1a5e3fb1 WN |
7945 | } |
7946 | ||
2ce8450e | 7947 | struct bpf_object * |
01af3bf0 | 7948 | bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts) |
2ce8450e | 7949 | { |
2ce8450e | 7950 | if (!path) |
e9fc3ce9 | 7951 | return libbpf_err_ptr(-EINVAL); |
2ce8450e AN |
7952 | |
7953 | pr_debug("loading %s\n", path); | |
7954 | ||
ad9a7f96 | 7955 | return libbpf_ptr(bpf_object_open(path, NULL, 0, opts)); |
2ce8450e AN |
7956 | } |
7957 | ||
146bf811 AN |
7958 | struct bpf_object *bpf_object__open(const char *path) |
7959 | { | |
7960 | return bpf_object__open_file(path, NULL); | |
7961 | } | |
7962 | ||
2ce8450e AN |
7963 | struct bpf_object * |
7964 | bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz, | |
01af3bf0 | 7965 | const struct bpf_object_open_opts *opts) |
6c956392 | 7966 | { |
2ce8450e | 7967 | if (!obj_buf || obj_buf_sz == 0) |
e9fc3ce9 | 7968 | return libbpf_err_ptr(-EINVAL); |
6c956392 | 7969 | |
ad9a7f96 | 7970 | return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, opts)); |
2ce8450e AN |
7971 | } |
7972 | ||
4a404a7e | 7973 | static int bpf_object_unload(struct bpf_object *obj) |
52d3352e WN |
7974 | { |
7975 | size_t i; | |
7976 | ||
7977 | if (!obj) | |
e9fc3ce9 | 7978 | return libbpf_err(-EINVAL); |
52d3352e | 7979 | |
590a0088 | 7980 | for (i = 0; i < obj->nr_maps; i++) { |
9d759a9b | 7981 | zclose(obj->maps[i].fd); |
590a0088 MKL |
7982 | if (obj->maps[i].st_ops) |
7983 | zfree(&obj->maps[i].st_ops->kern_vdata); | |
7984 | } | |
52d3352e | 7985 | |
55cffde2 WN |
7986 | for (i = 0; i < obj->nr_programs; i++) |
7987 | bpf_program__unload(&obj->programs[i]); | |
7988 | ||
52d3352e WN |
7989 | return 0; |
7990 | } | |
7991 | ||
0d13bfce AN |
7992 | static int bpf_object__sanitize_maps(struct bpf_object *obj) |
7993 | { | |
7994 | struct bpf_map *m; | |
7995 | ||
7996 | bpf_object__for_each_map(m, obj) { | |
7997 | if (!bpf_map__is_internal(m)) | |
7998 | continue; | |
9ca1f56a | 7999 | if (!kernel_supports(obj, FEAT_ARRAY_MMAP)) |
6920b086 | 8000 | m->def.map_flags &= ~BPF_F_MMAPABLE; |
0d13bfce AN |
8001 | } |
8002 | ||
8003 | return 0; | |
8004 | } | |
8005 | ||
ad2b0528 YS |
8006 | typedef int (*kallsyms_cb_t)(unsigned long long sym_addr, char sym_type, |
8007 | const char *sym_name, void *ctx); | |
8008 | ||
8009 | static int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *ctx) | |
1c0c7074 AN |
8010 | { |
8011 | char sym_type, sym_name[500]; | |
8012 | unsigned long long sym_addr; | |
1c0c7074 AN |
8013 | int ret, err = 0; |
8014 | FILE *f; | |
8015 | ||
59842c54 | 8016 | f = fopen("/proc/kallsyms", "re"); |
1c0c7074 AN |
8017 | if (!f) { |
8018 | err = -errno; | |
8019 | pr_warn("failed to open /proc/kallsyms: %d\n", err); | |
8020 | return err; | |
8021 | } | |
8022 | ||
8023 | while (true) { | |
8024 | ret = fscanf(f, "%llx %c %499s%*[^\n]\n", | |
8025 | &sym_addr, &sym_type, sym_name); | |
8026 | if (ret == EOF && feof(f)) | |
8027 | break; | |
8028 | if (ret != 3) { | |
135c783f | 8029 | pr_warn("failed to read kallsyms entry: %d\n", ret); |
1c0c7074 | 8030 | err = -EINVAL; |
85153ac0 | 8031 | break; |
1c0c7074 AN |
8032 | } |
8033 | ||
85153ac0 JO |
8034 | err = cb(sym_addr, sym_type, sym_name, ctx); |
8035 | if (err) | |
8036 | break; | |
1c0c7074 AN |
8037 | } |
8038 | ||
1c0c7074 AN |
8039 | fclose(f); |
8040 | return err; | |
8041 | } | |
8042 | ||
85153ac0 JO |
8043 | static int kallsyms_cb(unsigned long long sym_addr, char sym_type, |
8044 | const char *sym_name, void *ctx) | |
8045 | { | |
8046 | struct bpf_object *obj = ctx; | |
8047 | const struct btf_type *t; | |
8048 | struct extern_desc *ext; | |
c56e5977 | 8049 | char *res; |
85153ac0 | 8050 | |
c56e5977 YS |
8051 | res = strstr(sym_name, ".llvm."); |
8052 | if (sym_type == 'd' && res) | |
8053 | ext = find_extern_by_name_with_len(obj, sym_name, res - sym_name); | |
8054 | else | |
8055 | ext = find_extern_by_name(obj, sym_name); | |
85153ac0 JO |
8056 | if (!ext || ext->type != EXT_KSYM) |
8057 | return 0; | |
8058 | ||
8059 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
8060 | if (!btf_is_var(t)) | |
8061 | return 0; | |
8062 | ||
8063 | if (ext->is_set && ext->ksym.addr != sym_addr) { | |
55d00c37 | 8064 | pr_warn("extern (ksym) '%s': resolution is ambiguous: 0x%llx or 0x%llx\n", |
85153ac0 JO |
8065 | sym_name, ext->ksym.addr, sym_addr); |
8066 | return -EINVAL; | |
8067 | } | |
8068 | if (!ext->is_set) { | |
8069 | ext->is_set = true; | |
8070 | ext->ksym.addr = sym_addr; | |
55d00c37 | 8071 | pr_debug("extern (ksym) '%s': set to 0x%llx\n", sym_name, sym_addr); |
85153ac0 JO |
8072 | } |
8073 | return 0; | |
8074 | } | |
8075 | ||
8076 | static int bpf_object__read_kallsyms_file(struct bpf_object *obj) | |
8077 | { | |
8078 | return libbpf_kallsyms_parse(kallsyms_cb, obj); | |
8079 | } | |
8080 | ||
774e132e MKL |
8081 | static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name, |
8082 | __u16 kind, struct btf **res_btf, | |
9dbe6015 | 8083 | struct module_btf **res_mod_btf) |
d370bbe1 | 8084 | { |
9dbe6015 | 8085 | struct module_btf *mod_btf; |
284d2587 | 8086 | struct btf *btf; |
9dbe6015 | 8087 | int i, id, err; |
d370bbe1 | 8088 | |
933d1aa3 | 8089 | btf = obj->btf_vmlinux; |
9dbe6015 | 8090 | mod_btf = NULL; |
774e132e MKL |
8091 | id = btf__find_by_name_kind(btf, ksym_name, kind); |
8092 | ||
933d1aa3 MKL |
8093 | if (id == -ENOENT) { |
8094 | err = load_module_btfs(obj); | |
8095 | if (err) | |
8096 | return err; | |
d370bbe1 | 8097 | |
933d1aa3 | 8098 | for (i = 0; i < obj->btf_module_cnt; i++) { |
9dbe6015 KKD |
8099 | /* we assume module_btf's BTF FD is always >0 */ |
8100 | mod_btf = &obj->btf_modules[i]; | |
8101 | btf = mod_btf->btf; | |
8102 | id = btf__find_by_name_kind_own(btf, ksym_name, kind); | |
933d1aa3 MKL |
8103 | if (id != -ENOENT) |
8104 | break; | |
8105 | } | |
8106 | } | |
2211c825 | 8107 | if (id <= 0) |
933d1aa3 | 8108 | return -ESRCH; |
d370bbe1 | 8109 | |
774e132e | 8110 | *res_btf = btf; |
9dbe6015 | 8111 | *res_mod_btf = mod_btf; |
774e132e MKL |
8112 | return id; |
8113 | } | |
8114 | ||
8115 | static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj, | |
8116 | struct extern_desc *ext) | |
8117 | { | |
8118 | const struct btf_type *targ_var, *targ_type; | |
8119 | __u32 targ_type_id, local_type_id; | |
9dbe6015 | 8120 | struct module_btf *mod_btf = NULL; |
774e132e | 8121 | const char *targ_var_name; |
774e132e | 8122 | struct btf *btf = NULL; |
9dbe6015 | 8123 | int id, err; |
774e132e | 8124 | |
9dbe6015 | 8125 | id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf); |
466b2e13 KKD |
8126 | if (id < 0) { |
8127 | if (id == -ESRCH && ext->is_weak) | |
8128 | return 0; | |
2211c825 HL |
8129 | pr_warn("extern (var ksym) '%s': not found in kernel BTF\n", |
8130 | ext->name); | |
774e132e | 8131 | return id; |
2211c825 | 8132 | } |
774e132e | 8133 | |
933d1aa3 MKL |
8134 | /* find local type_id */ |
8135 | local_type_id = ext->ksym.type_id; | |
284d2587 | 8136 | |
933d1aa3 MKL |
8137 | /* find target type_id */ |
8138 | targ_var = btf__type_by_id(btf, id); | |
8139 | targ_var_name = btf__name_by_offset(btf, targ_var->name_off); | |
8140 | targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id); | |
d370bbe1 | 8141 | |
933d1aa3 MKL |
8142 | err = bpf_core_types_are_compat(obj->btf, local_type_id, |
8143 | btf, targ_type_id); | |
8144 | if (err <= 0) { | |
8145 | const struct btf_type *local_type; | |
8146 | const char *targ_name, *local_name; | |
d370bbe1 | 8147 | |
933d1aa3 MKL |
8148 | local_type = btf__type_by_id(obj->btf, local_type_id); |
8149 | local_name = btf__name_by_offset(obj->btf, local_type->name_off); | |
8150 | targ_name = btf__name_by_offset(btf, targ_type->name_off); | |
d370bbe1 | 8151 | |
933d1aa3 MKL |
8152 | pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n", |
8153 | ext->name, local_type_id, | |
8154 | btf_kind_str(local_type), local_name, targ_type_id, | |
8155 | btf_kind_str(targ_type), targ_name); | |
8156 | return -EINVAL; | |
8157 | } | |
d370bbe1 | 8158 | |
933d1aa3 | 8159 | ext->is_set = true; |
9dbe6015 | 8160 | ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0; |
933d1aa3 MKL |
8161 | ext->ksym.kernel_btf_id = id; |
8162 | pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n", | |
8163 | ext->name, id, btf_kind_str(targ_var), targ_var_name); | |
d370bbe1 | 8164 | |
933d1aa3 MKL |
8165 | return 0; |
8166 | } | |
d370bbe1 | 8167 | |
5bd022ec MKL |
8168 | static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj, |
8169 | struct extern_desc *ext) | |
8170 | { | |
8171 | int local_func_proto_id, kfunc_proto_id, kfunc_id; | |
9dbe6015 | 8172 | struct module_btf *mod_btf = NULL; |
5bd022ec MKL |
8173 | const struct btf_type *kern_func; |
8174 | struct btf *kern_btf = NULL; | |
9dbe6015 | 8175 | int ret; |
5bd022ec MKL |
8176 | |
8177 | local_func_proto_id = ext->ksym.type_id; | |
8178 | ||
5964a223 DM |
8179 | kfunc_id = find_ksym_btf_id(obj, ext->essent_name ?: ext->name, BTF_KIND_FUNC, &kern_btf, |
8180 | &mod_btf); | |
5bd022ec | 8181 | if (kfunc_id < 0) { |
466b2e13 KKD |
8182 | if (kfunc_id == -ESRCH && ext->is_weak) |
8183 | return 0; | |
8184 | pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n", | |
5bd022ec MKL |
8185 | ext->name); |
8186 | return kfunc_id; | |
8187 | } | |
8188 | ||
5bd022ec MKL |
8189 | kern_func = btf__type_by_id(kern_btf, kfunc_id); |
8190 | kfunc_proto_id = kern_func->type; | |
8191 | ||
8192 | ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id, | |
8193 | kern_btf, kfunc_proto_id); | |
8194 | if (ret <= 0) { | |
5964a223 DM |
8195 | if (ext->is_weak) |
8196 | return 0; | |
8197 | ||
f709160d AN |
8198 | pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with %s [%d]\n", |
8199 | ext->name, local_func_proto_id, | |
8200 | mod_btf ? mod_btf->name : "vmlinux", kfunc_proto_id); | |
5bd022ec MKL |
8201 | return -EINVAL; |
8202 | } | |
8203 | ||
9dbe6015 KKD |
8204 | /* set index for module BTF fd in fd_array, if unset */ |
8205 | if (mod_btf && !mod_btf->fd_array_idx) { | |
8206 | /* insn->off is s16 */ | |
8207 | if (obj->fd_array_cnt == INT16_MAX) { | |
8208 | pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n", | |
8209 | ext->name, mod_btf->fd_array_idx); | |
8210 | return -E2BIG; | |
8211 | } | |
8212 | /* Cannot use index 0 for module BTF fd */ | |
8213 | if (!obj->fd_array_cnt) | |
8214 | obj->fd_array_cnt = 1; | |
8215 | ||
8216 | ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int), | |
8217 | obj->fd_array_cnt + 1); | |
8218 | if (ret) | |
8219 | return ret; | |
8220 | mod_btf->fd_array_idx = obj->fd_array_cnt; | |
8221 | /* we assume module BTF FD is always >0 */ | |
8222 | obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd; | |
8223 | } | |
8224 | ||
5bd022ec | 8225 | ext->is_set = true; |
5bd022ec | 8226 | ext->ksym.kernel_btf_id = kfunc_id; |
9dbe6015 | 8227 | ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0; |
5fc13ad5 AS |
8228 | /* Also set kernel_btf_obj_fd to make sure that bpf_object__relocate_data() |
8229 | * populates FD into ld_imm64 insn when it's used to point to kfunc. | |
8230 | * {kernel_btf_id, btf_fd_idx} -> fixup bpf_call. | |
8231 | * {kernel_btf_id, kernel_btf_obj_fd} -> fixup ld_imm64. | |
8232 | */ | |
8233 | ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0; | |
f709160d AN |
8234 | pr_debug("extern (func ksym) '%s': resolved to %s [%d]\n", |
8235 | ext->name, mod_btf ? mod_btf->name : "vmlinux", kfunc_id); | |
5bd022ec MKL |
8236 | |
8237 | return 0; | |
8238 | } | |
8239 | ||
933d1aa3 MKL |
8240 | static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj) |
8241 | { | |
5bd022ec | 8242 | const struct btf_type *t; |
933d1aa3 MKL |
8243 | struct extern_desc *ext; |
8244 | int i, err; | |
8245 | ||
8246 | for (i = 0; i < obj->nr_extern; i++) { | |
8247 | ext = &obj->externs[i]; | |
8248 | if (ext->type != EXT_KSYM || !ext->ksym.type_id) | |
8249 | continue; | |
8250 | ||
67234743 AS |
8251 | if (obj->gen_loader) { |
8252 | ext->is_set = true; | |
8253 | ext->ksym.kernel_btf_obj_fd = 0; | |
8254 | ext->ksym.kernel_btf_id = 0; | |
8255 | continue; | |
8256 | } | |
5bd022ec MKL |
8257 | t = btf__type_by_id(obj->btf, ext->btf_id); |
8258 | if (btf_is_var(t)) | |
8259 | err = bpf_object__resolve_ksym_var_btf_id(obj, ext); | |
8260 | else | |
8261 | err = bpf_object__resolve_ksym_func_btf_id(obj, ext); | |
933d1aa3 MKL |
8262 | if (err) |
8263 | return err; | |
d370bbe1 HL |
8264 | } |
8265 | return 0; | |
8266 | } | |
8267 | ||
166750bc | 8268 | static int bpf_object__resolve_externs(struct bpf_object *obj, |
8601fd42 | 8269 | const char *extra_kconfig) |
166750bc | 8270 | { |
1c0c7074 | 8271 | bool need_config = false, need_kallsyms = false; |
d370bbe1 | 8272 | bool need_vmlinux_btf = false; |
166750bc | 8273 | struct extern_desc *ext; |
2e33efe3 | 8274 | void *kcfg_data = NULL; |
166750bc | 8275 | int err, i; |
166750bc AN |
8276 | |
8277 | if (obj->nr_extern == 0) | |
8278 | return 0; | |
8279 | ||
2e33efe3 AN |
8280 | if (obj->kconfig_map_idx >= 0) |
8281 | kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped; | |
166750bc AN |
8282 | |
8283 | for (i = 0; i < obj->nr_extern; i++) { | |
8284 | ext = &obj->externs[i]; | |
8285 | ||
55d00c37 AN |
8286 | if (ext->type == EXT_KSYM) { |
8287 | if (ext->ksym.type_id) | |
8288 | need_vmlinux_btf = true; | |
8289 | else | |
8290 | need_kallsyms = true; | |
8291 | continue; | |
8292 | } else if (ext->type == EXT_KCFG) { | |
8293 | void *ext_ptr = kcfg_data + ext->kcfg.data_off; | |
8294 | __u64 value = 0; | |
166750bc | 8295 | |
55d00c37 AN |
8296 | /* Kconfig externs need actual /proc/config.gz */ |
8297 | if (str_has_pfx(ext->name, "CONFIG_")) { | |
8298 | need_config = true; | |
8299 | continue; | |
8300 | } | |
8301 | ||
8302 | /* Virtual kcfg externs are customly handled by libbpf */ | |
8303 | if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) { | |
8304 | value = get_kernel_version(); | |
8305 | if (!value) { | |
8306 | pr_warn("extern (kcfg) '%s': failed to get kernel version\n", ext->name); | |
8307 | return -EINVAL; | |
8308 | } | |
8309 | } else if (strcmp(ext->name, "LINUX_HAS_BPF_COOKIE") == 0) { | |
8310 | value = kernel_supports(obj, FEAT_BPF_COOKIE); | |
6f5d467d AN |
8311 | } else if (strcmp(ext->name, "LINUX_HAS_SYSCALL_WRAPPER") == 0) { |
8312 | value = kernel_supports(obj, FEAT_SYSCALL_WRAPPER); | |
55d00c37 AN |
8313 | } else if (!str_has_pfx(ext->name, "LINUX_") || !ext->is_weak) { |
8314 | /* Currently libbpf supports only CONFIG_ and LINUX_ prefixed | |
8315 | * __kconfig externs, where LINUX_ ones are virtual and filled out | |
8316 | * customly by libbpf (their values don't come from Kconfig). | |
8317 | * If LINUX_xxx variable is not recognized by libbpf, but is marked | |
8318 | * __weak, it defaults to zero value, just like for CONFIG_xxx | |
8319 | * externs. | |
8320 | */ | |
8321 | pr_warn("extern (kcfg) '%s': unrecognized virtual extern\n", ext->name); | |
166750bc AN |
8322 | return -EINVAL; |
8323 | } | |
55d00c37 AN |
8324 | |
8325 | err = set_kcfg_value_num(ext, ext_ptr, value); | |
166750bc AN |
8326 | if (err) |
8327 | return err; | |
55d00c37 AN |
8328 | pr_debug("extern (kcfg) '%s': set to 0x%llx\n", |
8329 | ext->name, (long long)value); | |
166750bc | 8330 | } else { |
55d00c37 | 8331 | pr_warn("extern '%s': unrecognized extern kind\n", ext->name); |
166750bc AN |
8332 | return -EINVAL; |
8333 | } | |
8334 | } | |
8601fd42 | 8335 | if (need_config && extra_kconfig) { |
2e33efe3 | 8336 | err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data); |
8601fd42 AN |
8337 | if (err) |
8338 | return -EINVAL; | |
8339 | need_config = false; | |
8340 | for (i = 0; i < obj->nr_extern; i++) { | |
8341 | ext = &obj->externs[i]; | |
2e33efe3 | 8342 | if (ext->type == EXT_KCFG && !ext->is_set) { |
8601fd42 AN |
8343 | need_config = true; |
8344 | break; | |
8345 | } | |
8346 | } | |
8347 | } | |
166750bc | 8348 | if (need_config) { |
2e33efe3 | 8349 | err = bpf_object__read_kconfig_file(obj, kcfg_data); |
166750bc AN |
8350 | if (err) |
8351 | return -EINVAL; | |
8352 | } | |
1c0c7074 AN |
8353 | if (need_kallsyms) { |
8354 | err = bpf_object__read_kallsyms_file(obj); | |
8355 | if (err) | |
8356 | return -EINVAL; | |
8357 | } | |
d370bbe1 HL |
8358 | if (need_vmlinux_btf) { |
8359 | err = bpf_object__resolve_ksyms_btf_id(obj); | |
8360 | if (err) | |
8361 | return -EINVAL; | |
8362 | } | |
166750bc AN |
8363 | for (i = 0; i < obj->nr_extern; i++) { |
8364 | ext = &obj->externs[i]; | |
8365 | ||
8366 | if (!ext->is_set && !ext->is_weak) { | |
55d00c37 | 8367 | pr_warn("extern '%s' (strong): not resolved\n", ext->name); |
166750bc AN |
8368 | return -ESRCH; |
8369 | } else if (!ext->is_set) { | |
55d00c37 | 8370 | pr_debug("extern '%s' (weak): not resolved, defaulting to zero\n", |
166750bc AN |
8371 | ext->name); |
8372 | } | |
8373 | } | |
8374 | ||
8375 | return 0; | |
8376 | } | |
8377 | ||
8d1608d7 KFL |
8378 | static void bpf_map_prepare_vdata(const struct bpf_map *map) |
8379 | { | |
8380 | struct bpf_struct_ops *st_ops; | |
8381 | __u32 i; | |
8382 | ||
8383 | st_ops = map->st_ops; | |
8384 | for (i = 0; i < btf_vlen(st_ops->type); i++) { | |
8385 | struct bpf_program *prog = st_ops->progs[i]; | |
8386 | void *kern_data; | |
8387 | int prog_fd; | |
8388 | ||
8389 | if (!prog) | |
8390 | continue; | |
8391 | ||
8392 | prog_fd = bpf_program__fd(prog); | |
8393 | kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i]; | |
8394 | *(unsigned long *)kern_data = prog_fd; | |
8395 | } | |
8396 | } | |
8397 | ||
8398 | static int bpf_object_prepare_struct_ops(struct bpf_object *obj) | |
8399 | { | |
8db05261 | 8400 | struct bpf_map *map; |
8d1608d7 KFL |
8401 | int i; |
8402 | ||
8db05261 EZ |
8403 | for (i = 0; i < obj->nr_maps; i++) { |
8404 | map = &obj->maps[i]; | |
8405 | ||
8406 | if (!bpf_map__is_struct_ops(map)) | |
8407 | continue; | |
8408 | ||
8409 | if (!map->autocreate) | |
8410 | continue; | |
8411 | ||
8412 | bpf_map_prepare_vdata(map); | |
8413 | } | |
8d1608d7 KFL |
8414 | |
8415 | return 0; | |
8416 | } | |
8417 | ||
e7b924ca | 8418 | static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path) |
52d3352e | 8419 | { |
ec6d5f47 | 8420 | int err, i; |
6371ca3b | 8421 | |
52d3352e | 8422 | if (!obj) |
e9fc3ce9 | 8423 | return libbpf_err(-EINVAL); |
52d3352e WN |
8424 | |
8425 | if (obj->loaded) { | |
d9297581 | 8426 | pr_warn("object '%s': load can't be attempted twice\n", obj->name); |
e9fc3ce9 | 8427 | return libbpf_err(-EINVAL); |
52d3352e WN |
8428 | } |
8429 | ||
67234743 | 8430 | if (obj->gen_loader) |
be315829 | 8431 | bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps); |
67234743 | 8432 | |
6b434b61 AN |
8433 | err = bpf_object_prepare_token(obj); |
8434 | err = err ? : bpf_object__probe_loading(obj); | |
fe62de31 | 8435 | err = err ? : bpf_object__load_vmlinux_btf(obj, false); |
8601fd42 | 8436 | err = err ? : bpf_object__resolve_externs(obj, obj->kconfig); |
0d13bfce | 8437 | err = err ? : bpf_object__sanitize_maps(obj); |
590a0088 | 8438 | err = err ? : bpf_object__init_kern_struct_ops_maps(obj); |
fe9d049c | 8439 | err = err ? : bpf_object_adjust_struct_ops_autoload(obj); |
e7b924ca | 8440 | err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path); |
1004742d | 8441 | err = err ? : bpf_object__sanitize_and_load_btf(obj); |
dac645b9 | 8442 | err = err ? : bpf_object__create_maps(obj); |
e7b924ca | 8443 | err = err ? : bpf_object__load_progs(obj, extra_log_level); |
341ac5ff | 8444 | err = err ? : bpf_object_init_prog_arrays(obj); |
8d1608d7 | 8445 | err = err ? : bpf_object_prepare_struct_ops(obj); |
a6ed02ca | 8446 | |
67234743 AS |
8447 | if (obj->gen_loader) { |
8448 | /* reset FDs */ | |
4729445b KKD |
8449 | if (obj->btf) |
8450 | btf__set_fd(obj->btf, -1); | |
67234743 | 8451 | if (!err) |
ba05fd36 | 8452 | err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps); |
67234743 AS |
8453 | } |
8454 | ||
9dbe6015 KKD |
8455 | /* clean up fd_array */ |
8456 | zfree(&obj->fd_array); | |
8457 | ||
4f33a53d AN |
8458 | /* clean up module BTFs */ |
8459 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
91abb4a6 | 8460 | close(obj->btf_modules[i].fd); |
4f33a53d AN |
8461 | btf__free(obj->btf_modules[i].btf); |
8462 | free(obj->btf_modules[i].name); | |
8463 | } | |
8464 | free(obj->btf_modules); | |
8465 | ||
8466 | /* clean up vmlinux BTF */ | |
a6ed02ca KS |
8467 | btf__free(obj->btf_vmlinux); |
8468 | obj->btf_vmlinux = NULL; | |
8469 | ||
d9297581 AN |
8470 | obj->loaded = true; /* doesn't matter if successfully or not */ |
8471 | ||
0d13bfce AN |
8472 | if (err) |
8473 | goto out; | |
52d3352e WN |
8474 | |
8475 | return 0; | |
8476 | out: | |
ec6d5f47 THJ |
8477 | /* unpin any maps that were auto-pinned during load */ |
8478 | for (i = 0; i < obj->nr_maps; i++) | |
8479 | if (obj->maps[i].pinned && !obj->maps[i].reused) | |
8480 | bpf_map__unpin(&obj->maps[i], NULL); | |
8481 | ||
4a404a7e | 8482 | bpf_object_unload(obj); |
be18010e | 8483 | pr_warn("failed to load object '%s'\n", obj->path); |
e9fc3ce9 | 8484 | return libbpf_err(err); |
52d3352e WN |
8485 | } |
8486 | ||
e7b924ca AN |
8487 | int bpf_object__load(struct bpf_object *obj) |
8488 | { | |
8489 | return bpf_object_load(obj, 0, NULL); | |
60276f98 QM |
8490 | } |
8491 | ||
196f8487 THJ |
8492 | static int make_parent_dir(const char *path) |
8493 | { | |
8494 | char *cp, errmsg[STRERR_BUFSIZE]; | |
8495 | char *dname, *dir; | |
8496 | int err = 0; | |
8497 | ||
8498 | dname = strdup(path); | |
8499 | if (dname == NULL) | |
8500 | return -ENOMEM; | |
8501 | ||
8502 | dir = dirname(dname); | |
8503 | if (mkdir(dir, 0700) && errno != EEXIST) | |
8504 | err = -errno; | |
8505 | ||
8506 | free(dname); | |
8507 | if (err) { | |
8508 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
8509 | pr_warn("failed to mkdir %s: %s\n", path, cp); | |
8510 | } | |
8511 | return err; | |
8512 | } | |
8513 | ||
f367540c JS |
8514 | static int check_path(const char *path) |
8515 | { | |
1ce6a9fc | 8516 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
8517 | struct statfs st_fs; |
8518 | char *dname, *dir; | |
8519 | int err = 0; | |
8520 | ||
8521 | if (path == NULL) | |
8522 | return -EINVAL; | |
8523 | ||
8524 | dname = strdup(path); | |
8525 | if (dname == NULL) | |
8526 | return -ENOMEM; | |
8527 | ||
8528 | dir = dirname(dname); | |
8529 | if (statfs(dir, &st_fs)) { | |
24d6a808 | 8530 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
be18010e | 8531 | pr_warn("failed to statfs %s: %s\n", dir, cp); |
f367540c JS |
8532 | err = -errno; |
8533 | } | |
8534 | free(dname); | |
8535 | ||
8536 | if (!err && st_fs.f_type != BPF_FS_MAGIC) { | |
be18010e | 8537 | pr_warn("specified path %s is not on BPF FS\n", path); |
f367540c JS |
8538 | err = -EINVAL; |
8539 | } | |
8540 | ||
8541 | return err; | |
8542 | } | |
8543 | ||
cf90a20d | 8544 | int bpf_program__pin(struct bpf_program *prog, const char *path) |
f367540c | 8545 | { |
1ce6a9fc | 8546 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
8547 | int err; |
8548 | ||
cf90a20d AN |
8549 | if (prog->fd < 0) { |
8550 | pr_warn("prog '%s': can't pin program that wasn't loaded\n", prog->name); | |
8551 | return libbpf_err(-EINVAL); | |
8552 | } | |
8553 | ||
196f8487 THJ |
8554 | err = make_parent_dir(path); |
8555 | if (err) | |
e9fc3ce9 | 8556 | return libbpf_err(err); |
196f8487 | 8557 | |
f367540c JS |
8558 | err = check_path(path); |
8559 | if (err) | |
e9fc3ce9 | 8560 | return libbpf_err(err); |
f367540c | 8561 | |
cf90a20d | 8562 | if (bpf_obj_pin(prog->fd, path)) { |
23ab656b THJ |
8563 | err = -errno; |
8564 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
cf90a20d | 8565 | pr_warn("prog '%s': failed to pin at '%s': %s\n", prog->name, path, cp); |
e9fc3ce9 | 8566 | return libbpf_err(err); |
f367540c | 8567 | } |
f367540c | 8568 | |
cf90a20d | 8569 | pr_debug("prog '%s': pinned at '%s'\n", prog->name, path); |
f367540c JS |
8570 | return 0; |
8571 | } | |
8572 | ||
cf90a20d | 8573 | int bpf_program__unpin(struct bpf_program *prog, const char *path) |
0c19a9fb SF |
8574 | { |
8575 | int err; | |
8576 | ||
cf90a20d AN |
8577 | if (prog->fd < 0) { |
8578 | pr_warn("prog '%s': can't unpin program that wasn't loaded\n", prog->name); | |
e9fc3ce9 | 8579 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
8580 | } |
8581 | ||
f367540c JS |
8582 | err = check_path(path); |
8583 | if (err) | |
e9fc3ce9 | 8584 | return libbpf_err(err); |
f367540c | 8585 | |
cf90a20d | 8586 | err = unlink(path); |
0c19a9fb | 8587 | if (err) |
e9fc3ce9 | 8588 | return libbpf_err(-errno); |
0c19a9fb | 8589 | |
cf90a20d | 8590 | pr_debug("prog '%s': unpinned from '%s'\n", prog->name, path); |
f367540c JS |
8591 | return 0; |
8592 | } | |
8593 | ||
b6989f35 JS |
8594 | int bpf_map__pin(struct bpf_map *map, const char *path) |
8595 | { | |
1ce6a9fc | 8596 | char *cp, errmsg[STRERR_BUFSIZE]; |
b6989f35 JS |
8597 | int err; |
8598 | ||
b6989f35 | 8599 | if (map == NULL) { |
be18010e | 8600 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 8601 | return libbpf_err(-EINVAL); |
b6989f35 JS |
8602 | } |
8603 | ||
7b30c296 MY |
8604 | if (map->fd < 0) { |
8605 | pr_warn("map '%s': can't pin BPF map without FD (was it created?)\n", map->name); | |
8606 | return libbpf_err(-EINVAL); | |
8607 | } | |
8608 | ||
4580b25f THJ |
8609 | if (map->pin_path) { |
8610 | if (path && strcmp(path, map->pin_path)) { | |
8611 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
8612 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 8613 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8614 | } else if (map->pinned) { |
8615 | pr_debug("map '%s' already pinned at '%s'; not re-pinning\n", | |
8616 | bpf_map__name(map), map->pin_path); | |
8617 | return 0; | |
8618 | } | |
8619 | } else { | |
8620 | if (!path) { | |
8621 | pr_warn("missing a path to pin map '%s' at\n", | |
8622 | bpf_map__name(map)); | |
e9fc3ce9 | 8623 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8624 | } else if (map->pinned) { |
8625 | pr_warn("map '%s' already pinned\n", bpf_map__name(map)); | |
e9fc3ce9 | 8626 | return libbpf_err(-EEXIST); |
4580b25f THJ |
8627 | } |
8628 | ||
8629 | map->pin_path = strdup(path); | |
8630 | if (!map->pin_path) { | |
8631 | err = -errno; | |
8632 | goto out_err; | |
8633 | } | |
b6989f35 JS |
8634 | } |
8635 | ||
196f8487 THJ |
8636 | err = make_parent_dir(map->pin_path); |
8637 | if (err) | |
e9fc3ce9 | 8638 | return libbpf_err(err); |
196f8487 | 8639 | |
4580b25f THJ |
8640 | err = check_path(map->pin_path); |
8641 | if (err) | |
e9fc3ce9 | 8642 | return libbpf_err(err); |
4580b25f THJ |
8643 | |
8644 | if (bpf_obj_pin(map->fd, map->pin_path)) { | |
8645 | err = -errno; | |
8646 | goto out_err; | |
8647 | } | |
8648 | ||
8649 | map->pinned = true; | |
8650 | pr_debug("pinned map '%s'\n", map->pin_path); | |
0c19a9fb | 8651 | |
b6989f35 | 8652 | return 0; |
4580b25f THJ |
8653 | |
8654 | out_err: | |
8655 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
8656 | pr_warn("failed to pin map: %s\n", cp); | |
e9fc3ce9 | 8657 | return libbpf_err(err); |
b6989f35 JS |
8658 | } |
8659 | ||
0c19a9fb SF |
8660 | int bpf_map__unpin(struct bpf_map *map, const char *path) |
8661 | { | |
8662 | int err; | |
8663 | ||
0c19a9fb | 8664 | if (map == NULL) { |
be18010e | 8665 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 8666 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
8667 | } |
8668 | ||
4580b25f THJ |
8669 | if (map->pin_path) { |
8670 | if (path && strcmp(path, map->pin_path)) { | |
8671 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
8672 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 8673 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8674 | } |
8675 | path = map->pin_path; | |
8676 | } else if (!path) { | |
8677 | pr_warn("no path to unpin map '%s' from\n", | |
8678 | bpf_map__name(map)); | |
e9fc3ce9 | 8679 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8680 | } |
8681 | ||
8682 | err = check_path(path); | |
8683 | if (err) | |
e9fc3ce9 | 8684 | return libbpf_err(err); |
4580b25f | 8685 | |
0c19a9fb SF |
8686 | err = unlink(path); |
8687 | if (err != 0) | |
e9fc3ce9 | 8688 | return libbpf_err(-errno); |
4580b25f THJ |
8689 | |
8690 | map->pinned = false; | |
8691 | pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path); | |
0c19a9fb SF |
8692 | |
8693 | return 0; | |
8694 | } | |
8695 | ||
4580b25f THJ |
8696 | int bpf_map__set_pin_path(struct bpf_map *map, const char *path) |
8697 | { | |
8698 | char *new = NULL; | |
8699 | ||
8700 | if (path) { | |
8701 | new = strdup(path); | |
8702 | if (!new) | |
e9fc3ce9 | 8703 | return libbpf_err(-errno); |
4580b25f THJ |
8704 | } |
8705 | ||
8706 | free(map->pin_path); | |
8707 | map->pin_path = new; | |
8708 | return 0; | |
8709 | } | |
8710 | ||
20eccf29 AN |
8711 | __alias(bpf_map__pin_path) |
8712 | const char *bpf_map__get_pin_path(const struct bpf_map *map); | |
4580b25f | 8713 | |
e244d34d EL |
8714 | const char *bpf_map__pin_path(const struct bpf_map *map) |
8715 | { | |
8716 | return map->pin_path; | |
8717 | } | |
8718 | ||
4580b25f THJ |
8719 | bool bpf_map__is_pinned(const struct bpf_map *map) |
8720 | { | |
8721 | return map->pinned; | |
8722 | } | |
8723 | ||
9cf309c5 THJ |
8724 | static void sanitize_pin_path(char *s) |
8725 | { | |
8726 | /* bpffs disallows periods in path names */ | |
8727 | while (*s) { | |
8728 | if (*s == '.') | |
8729 | *s = '_'; | |
8730 | s++; | |
8731 | } | |
8732 | } | |
8733 | ||
0c19a9fb | 8734 | int bpf_object__pin_maps(struct bpf_object *obj, const char *path) |
d5148d85 | 8735 | { |
d5148d85 JS |
8736 | struct bpf_map *map; |
8737 | int err; | |
8738 | ||
8739 | if (!obj) | |
e9fc3ce9 | 8740 | return libbpf_err(-ENOENT); |
d5148d85 JS |
8741 | |
8742 | if (!obj->loaded) { | |
be18010e | 8743 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 8744 | return libbpf_err(-ENOENT); |
d5148d85 JS |
8745 | } |
8746 | ||
f74a53d9 | 8747 | bpf_object__for_each_map(map, obj) { |
4580b25f | 8748 | char *pin_path = NULL; |
0c19a9fb | 8749 | char buf[PATH_MAX]; |
0c19a9fb | 8750 | |
ec41817b | 8751 | if (!map->autocreate) |
229fae38 SC |
8752 | continue; |
8753 | ||
4580b25f | 8754 | if (path) { |
e588c116 WY |
8755 | err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map)); |
8756 | if (err) | |
4580b25f | 8757 | goto err_unpin_maps; |
9cf309c5 | 8758 | sanitize_pin_path(buf); |
4580b25f THJ |
8759 | pin_path = buf; |
8760 | } else if (!map->pin_path) { | |
8761 | continue; | |
0c19a9fb SF |
8762 | } |
8763 | ||
4580b25f | 8764 | err = bpf_map__pin(map, pin_path); |
0c19a9fb SF |
8765 | if (err) |
8766 | goto err_unpin_maps; | |
8767 | } | |
8768 | ||
8769 | return 0; | |
8770 | ||
8771 | err_unpin_maps: | |
bcc40fc0 | 8772 | while ((map = bpf_object__prev_map(obj, map))) { |
4580b25f | 8773 | if (!map->pin_path) |
0c19a9fb SF |
8774 | continue; |
8775 | ||
4580b25f | 8776 | bpf_map__unpin(map, NULL); |
0c19a9fb SF |
8777 | } |
8778 | ||
e9fc3ce9 | 8779 | return libbpf_err(err); |
0c19a9fb SF |
8780 | } |
8781 | ||
8782 | int bpf_object__unpin_maps(struct bpf_object *obj, const char *path) | |
8783 | { | |
8784 | struct bpf_map *map; | |
8785 | int err; | |
8786 | ||
8787 | if (!obj) | |
e9fc3ce9 | 8788 | return libbpf_err(-ENOENT); |
0c19a9fb | 8789 | |
f74a53d9 | 8790 | bpf_object__for_each_map(map, obj) { |
4580b25f | 8791 | char *pin_path = NULL; |
d5148d85 | 8792 | char buf[PATH_MAX]; |
d5148d85 | 8793 | |
4580b25f | 8794 | if (path) { |
e588c116 WY |
8795 | err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map)); |
8796 | if (err) | |
8797 | return libbpf_err(err); | |
9cf309c5 | 8798 | sanitize_pin_path(buf); |
4580b25f THJ |
8799 | pin_path = buf; |
8800 | } else if (!map->pin_path) { | |
8801 | continue; | |
8802 | } | |
d5148d85 | 8803 | |
4580b25f | 8804 | err = bpf_map__unpin(map, pin_path); |
d5148d85 | 8805 | if (err) |
e9fc3ce9 | 8806 | return libbpf_err(err); |
d5148d85 JS |
8807 | } |
8808 | ||
0c19a9fb SF |
8809 | return 0; |
8810 | } | |
8811 | ||
8812 | int bpf_object__pin_programs(struct bpf_object *obj, const char *path) | |
8813 | { | |
8814 | struct bpf_program *prog; | |
e588c116 | 8815 | char buf[PATH_MAX]; |
0c19a9fb SF |
8816 | int err; |
8817 | ||
8818 | if (!obj) | |
e9fc3ce9 | 8819 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
8820 | |
8821 | if (!obj->loaded) { | |
be18010e | 8822 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 8823 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
8824 | } |
8825 | ||
0c19a9fb | 8826 | bpf_object__for_each_program(prog, obj) { |
e588c116 WY |
8827 | err = pathname_concat(buf, sizeof(buf), path, prog->name); |
8828 | if (err) | |
0c19a9fb | 8829 | goto err_unpin_programs; |
0c19a9fb SF |
8830 | |
8831 | err = bpf_program__pin(prog, buf); | |
8832 | if (err) | |
8833 | goto err_unpin_programs; | |
8834 | } | |
8835 | ||
8836 | return 0; | |
8837 | ||
8838 | err_unpin_programs: | |
bcc40fc0 | 8839 | while ((prog = bpf_object__prev_program(obj, prog))) { |
e588c116 | 8840 | if (pathname_concat(buf, sizeof(buf), path, prog->name)) |
0c19a9fb SF |
8841 | continue; |
8842 | ||
8843 | bpf_program__unpin(prog, buf); | |
8844 | } | |
8845 | ||
e9fc3ce9 | 8846 | return libbpf_err(err); |
0c19a9fb SF |
8847 | } |
8848 | ||
8849 | int bpf_object__unpin_programs(struct bpf_object *obj, const char *path) | |
8850 | { | |
8851 | struct bpf_program *prog; | |
8852 | int err; | |
8853 | ||
8854 | if (!obj) | |
e9fc3ce9 | 8855 | return libbpf_err(-ENOENT); |
0c19a9fb | 8856 | |
d5148d85 JS |
8857 | bpf_object__for_each_program(prog, obj) { |
8858 | char buf[PATH_MAX]; | |
d5148d85 | 8859 | |
e588c116 WY |
8860 | err = pathname_concat(buf, sizeof(buf), path, prog->name); |
8861 | if (err) | |
8862 | return libbpf_err(err); | |
d5148d85 | 8863 | |
0c19a9fb | 8864 | err = bpf_program__unpin(prog, buf); |
d5148d85 | 8865 | if (err) |
e9fc3ce9 | 8866 | return libbpf_err(err); |
d5148d85 JS |
8867 | } |
8868 | ||
8869 | return 0; | |
8870 | } | |
8871 | ||
0c19a9fb SF |
8872 | int bpf_object__pin(struct bpf_object *obj, const char *path) |
8873 | { | |
8874 | int err; | |
8875 | ||
8876 | err = bpf_object__pin_maps(obj, path); | |
8877 | if (err) | |
e9fc3ce9 | 8878 | return libbpf_err(err); |
0c19a9fb SF |
8879 | |
8880 | err = bpf_object__pin_programs(obj, path); | |
8881 | if (err) { | |
8882 | bpf_object__unpin_maps(obj, path); | |
e9fc3ce9 | 8883 | return libbpf_err(err); |
0c19a9fb SF |
8884 | } |
8885 | ||
8886 | return 0; | |
8887 | } | |
8888 | ||
068ca522 DX |
8889 | int bpf_object__unpin(struct bpf_object *obj, const char *path) |
8890 | { | |
8891 | int err; | |
8892 | ||
8893 | err = bpf_object__unpin_programs(obj, path); | |
8894 | if (err) | |
8895 | return libbpf_err(err); | |
8896 | ||
8897 | err = bpf_object__unpin_maps(obj, path); | |
8898 | if (err) | |
8899 | return libbpf_err(err); | |
8900 | ||
8901 | return 0; | |
8902 | } | |
8903 | ||
2d39d7c5 AN |
8904 | static void bpf_map__destroy(struct bpf_map *map) |
8905 | { | |
646f02ff AN |
8906 | if (map->inner_map) { |
8907 | bpf_map__destroy(map->inner_map); | |
8908 | zfree(&map->inner_map); | |
8909 | } | |
8910 | ||
8911 | zfree(&map->init_slots); | |
8912 | map->init_slots_sz = 0; | |
8913 | ||
2e7ba4f8 AN |
8914 | if (map->mmaped && map->mmaped != map->obj->arena_data) |
8915 | munmap(map->mmaped, bpf_map_mmap_sz(map)); | |
8916 | map->mmaped = NULL; | |
2d39d7c5 AN |
8917 | |
8918 | if (map->st_ops) { | |
8919 | zfree(&map->st_ops->data); | |
8920 | zfree(&map->st_ops->progs); | |
8921 | zfree(&map->st_ops->kern_func_off); | |
8922 | zfree(&map->st_ops); | |
8923 | } | |
8924 | ||
8925 | zfree(&map->name); | |
aed65917 | 8926 | zfree(&map->real_name); |
2d39d7c5 AN |
8927 | zfree(&map->pin_path); |
8928 | ||
8929 | if (map->fd >= 0) | |
8930 | zclose(map->fd); | |
8931 | } | |
8932 | ||
1a5e3fb1 WN |
8933 | void bpf_object__close(struct bpf_object *obj) |
8934 | { | |
a5b8bd47 WN |
8935 | size_t i; |
8936 | ||
50450fc7 | 8937 | if (IS_ERR_OR_NULL(obj)) |
1a5e3fb1 WN |
8938 | return; |
8939 | ||
2e4913e0 AN |
8940 | usdt_manager_free(obj->usdt_man); |
8941 | obj->usdt_man = NULL; | |
8942 | ||
67234743 | 8943 | bpf_gen__free(obj->gen_loader); |
1a5e3fb1 | 8944 | bpf_object__elf_finish(obj); |
4a404a7e | 8945 | bpf_object_unload(obj); |
8a138aed | 8946 | btf__free(obj->btf); |
29d67fde | 8947 | btf__free(obj->btf_vmlinux); |
2993e051 | 8948 | btf_ext__free(obj->btf_ext); |
1a5e3fb1 | 8949 | |
2d39d7c5 AN |
8950 | for (i = 0; i < obj->nr_maps; i++) |
8951 | bpf_map__destroy(&obj->maps[i]); | |
d859900c | 8952 | |
1373ff59 | 8953 | zfree(&obj->btf_custom_path); |
8601fd42 | 8954 | zfree(&obj->kconfig); |
5964a223 DM |
8955 | |
8956 | for (i = 0; i < obj->nr_extern; i++) | |
8957 | zfree(&obj->externs[i].essent_name); | |
8958 | ||
166750bc AN |
8959 | zfree(&obj->externs); |
8960 | obj->nr_extern = 0; | |
8961 | ||
9d759a9b WN |
8962 | zfree(&obj->maps); |
8963 | obj->nr_maps = 0; | |
a5b8bd47 WN |
8964 | |
8965 | if (obj->programs && obj->nr_programs) { | |
8966 | for (i = 0; i < obj->nr_programs; i++) | |
8967 | bpf_program__exit(&obj->programs[i]); | |
8968 | } | |
8969 | zfree(&obj->programs); | |
8970 | ||
6b434b61 AN |
8971 | zfree(&obj->feat_cache); |
8972 | zfree(&obj->token_path); | |
8973 | if (obj->token_fd > 0) | |
8974 | close(obj->token_fd); | |
8975 | ||
2e7ba4f8 AN |
8976 | zfree(&obj->arena_data); |
8977 | ||
1a5e3fb1 WN |
8978 | free(obj); |
8979 | } | |
aa9b1ac3 | 8980 | |
a324aae3 | 8981 | const char *bpf_object__name(const struct bpf_object *obj) |
acf860ae | 8982 | { |
e9fc3ce9 | 8983 | return obj ? obj->name : libbpf_err_ptr(-EINVAL); |
acf860ae WN |
8984 | } |
8985 | ||
a324aae3 | 8986 | unsigned int bpf_object__kversion(const struct bpf_object *obj) |
45825d8a | 8987 | { |
a7fe0450 | 8988 | return obj ? obj->kern_version : 0; |
45825d8a WN |
8989 | } |
8990 | ||
a324aae3 | 8991 | struct btf *bpf_object__btf(const struct bpf_object *obj) |
789f6bab AI |
8992 | { |
8993 | return obj ? obj->btf : NULL; | |
8994 | } | |
8995 | ||
8a138aed MKL |
8996 | int bpf_object__btf_fd(const struct bpf_object *obj) |
8997 | { | |
8998 | return obj->btf ? btf__fd(obj->btf) : -1; | |
8999 | } | |
9000 | ||
155f556d RDT |
9001 | int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version) |
9002 | { | |
9003 | if (obj->loaded) | |
e9fc3ce9 | 9004 | return libbpf_err(-EINVAL); |
155f556d RDT |
9005 | |
9006 | obj->kern_version = kern_version; | |
9007 | ||
9008 | return 0; | |
9009 | } | |
9010 | ||
67234743 AS |
9011 | int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts) |
9012 | { | |
9013 | struct bpf_gen *gen; | |
9014 | ||
9015 | if (!opts) | |
9016 | return -EFAULT; | |
9017 | if (!OPTS_VALID(opts, gen_loader_opts)) | |
9018 | return -EINVAL; | |
9019 | gen = calloc(sizeof(*gen), 1); | |
9020 | if (!gen) | |
9021 | return -ENOMEM; | |
9022 | gen->opts = opts; | |
9023 | obj->gen_loader = gen; | |
9024 | return 0; | |
9025 | } | |
9026 | ||
eac7d845 | 9027 | static struct bpf_program * |
a324aae3 AN |
9028 | __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj, |
9029 | bool forward) | |
aa9b1ac3 | 9030 | { |
a83d6e76 | 9031 | size_t nr_programs = obj->nr_programs; |
0c19a9fb | 9032 | ssize_t idx; |
aa9b1ac3 | 9033 | |
a83d6e76 | 9034 | if (!nr_programs) |
aa9b1ac3 | 9035 | return NULL; |
aa9b1ac3 | 9036 | |
a83d6e76 MKL |
9037 | if (!p) |
9038 | /* Iter from the beginning */ | |
9039 | return forward ? &obj->programs[0] : | |
9040 | &obj->programs[nr_programs - 1]; | |
9041 | ||
0c19a9fb | 9042 | if (p->obj != obj) { |
be18010e | 9043 | pr_warn("error: program handler doesn't match object\n"); |
e9fc3ce9 | 9044 | return errno = EINVAL, NULL; |
aa9b1ac3 WN |
9045 | } |
9046 | ||
a83d6e76 | 9047 | idx = (p - obj->programs) + (forward ? 1 : -1); |
0c19a9fb | 9048 | if (idx >= obj->nr_programs || idx < 0) |
aa9b1ac3 WN |
9049 | return NULL; |
9050 | return &obj->programs[idx]; | |
9051 | } | |
9052 | ||
2088a3a7 HC |
9053 | struct bpf_program * |
9054 | bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev) | |
eac7d845 JK |
9055 | { |
9056 | struct bpf_program *prog = prev; | |
9057 | ||
9058 | do { | |
a83d6e76 | 9059 | prog = __bpf_program__iter(prog, obj, true); |
c3c55696 | 9060 | } while (prog && prog_is_subprog(obj, prog)); |
0c19a9fb SF |
9061 | |
9062 | return prog; | |
9063 | } | |
9064 | ||
2088a3a7 HC |
9065 | struct bpf_program * |
9066 | bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next) | |
0c19a9fb SF |
9067 | { |
9068 | struct bpf_program *prog = next; | |
9069 | ||
0c19a9fb | 9070 | do { |
a83d6e76 | 9071 | prog = __bpf_program__iter(prog, obj, false); |
c3c55696 | 9072 | } while (prog && prog_is_subprog(obj, prog)); |
eac7d845 JK |
9073 | |
9074 | return prog; | |
9075 | } | |
9076 | ||
9aba3613 JK |
9077 | void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex) |
9078 | { | |
9079 | prog->prog_ifindex = ifindex; | |
9080 | } | |
9081 | ||
01af3bf0 AN |
9082 | const char *bpf_program__name(const struct bpf_program *prog) |
9083 | { | |
9084 | return prog->name; | |
9085 | } | |
9086 | ||
52109584 AN |
9087 | const char *bpf_program__section_name(const struct bpf_program *prog) |
9088 | { | |
9089 | return prog->sec_name; | |
9090 | } | |
9091 | ||
d9297581 AN |
9092 | bool bpf_program__autoload(const struct bpf_program *prog) |
9093 | { | |
a3820c48 | 9094 | return prog->autoload; |
d9297581 AN |
9095 | } |
9096 | ||
9097 | int bpf_program__set_autoload(struct bpf_program *prog, bool autoload) | |
9098 | { | |
9099 | if (prog->obj->loaded) | |
e9fc3ce9 | 9100 | return libbpf_err(-EINVAL); |
d9297581 | 9101 | |
a3820c48 | 9102 | prog->autoload = autoload; |
d9297581 AN |
9103 | return 0; |
9104 | } | |
9105 | ||
43cb8cba HL |
9106 | bool bpf_program__autoattach(const struct bpf_program *prog) |
9107 | { | |
9108 | return prog->autoattach; | |
9109 | } | |
9110 | ||
9111 | void bpf_program__set_autoattach(struct bpf_program *prog, bool autoattach) | |
9112 | { | |
9113 | prog->autoattach = autoattach; | |
9114 | } | |
9115 | ||
65a7fa2e AN |
9116 | const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog) |
9117 | { | |
9118 | return prog->insns; | |
9119 | } | |
9120 | ||
9121 | size_t bpf_program__insn_cnt(const struct bpf_program *prog) | |
9122 | { | |
9123 | return prog->insns_cnt; | |
9124 | } | |
9125 | ||
b63b3c49 JO |
9126 | int bpf_program__set_insns(struct bpf_program *prog, |
9127 | struct bpf_insn *new_insns, size_t new_insn_cnt) | |
9128 | { | |
9129 | struct bpf_insn *insns; | |
9130 | ||
9131 | if (prog->obj->loaded) | |
9132 | return -EBUSY; | |
9133 | ||
9134 | insns = libbpf_reallocarray(prog->insns, new_insn_cnt, sizeof(*insns)); | |
8a0260db AN |
9135 | /* NULL is a valid return from reallocarray if the new count is zero */ |
9136 | if (!insns && new_insn_cnt) { | |
b63b3c49 JO |
9137 | pr_warn("prog '%s': failed to realloc prog code\n", prog->name); |
9138 | return -ENOMEM; | |
9139 | } | |
9140 | memcpy(insns, new_insns, new_insn_cnt * sizeof(*insns)); | |
9141 | ||
9142 | prog->insns = insns; | |
9143 | prog->insns_cnt = new_insn_cnt; | |
9144 | return 0; | |
9145 | } | |
9146 | ||
cf90a20d | 9147 | int bpf_program__fd(const struct bpf_program *prog) |
b580563e | 9148 | { |
1e960043 | 9149 | if (!prog) |
e9fc3ce9 | 9150 | return libbpf_err(-EINVAL); |
1e960043 | 9151 | |
cf90a20d | 9152 | if (prog->fd < 0) |
e9fc3ce9 | 9153 | return libbpf_err(-ENOENT); |
b580563e | 9154 | |
cf90a20d | 9155 | return prog->fd; |
aa9b1ac3 | 9156 | } |
9d759a9b | 9157 | |
20eccf29 AN |
9158 | __alias(bpf_program__type) |
9159 | enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog); | |
9160 | ||
9161 | enum bpf_prog_type bpf_program__type(const struct bpf_program *prog) | |
f1eead9e AN |
9162 | { |
9163 | return prog->type; | |
9164 | } | |
9165 | ||
c628747c AN |
9166 | static size_t custom_sec_def_cnt; |
9167 | static struct bpf_sec_def *custom_sec_defs; | |
9168 | static struct bpf_sec_def custom_fallback_def; | |
9169 | static bool has_custom_fallback_def; | |
9170 | static int last_custom_sec_def_handler_id; | |
9171 | ||
93442f13 | 9172 | int bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type) |
5f44e4c8 | 9173 | { |
93442f13 GS |
9174 | if (prog->obj->loaded) |
9175 | return libbpf_err(-EBUSY); | |
9176 | ||
c628747c AN |
9177 | /* if type is not changed, do nothing */ |
9178 | if (prog->type == type) | |
9179 | return 0; | |
9180 | ||
5f44e4c8 | 9181 | prog->type = type; |
c628747c AN |
9182 | |
9183 | /* If a program type was changed, we need to reset associated SEC() | |
9184 | * handler, as it will be invalid now. The only exception is a generic | |
9185 | * fallback handler, which by definition is program type-agnostic and | |
9186 | * is a catch-all custom handler, optionally set by the application, | |
9187 | * so should be able to handle any type of BPF program. | |
9188 | */ | |
9189 | if (prog->sec_def != &custom_fallback_def) | |
9190 | prog->sec_def = NULL; | |
93442f13 | 9191 | return 0; |
5f44e4c8 WN |
9192 | } |
9193 | ||
20eccf29 AN |
9194 | __alias(bpf_program__expected_attach_type) |
9195 | enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog); | |
9196 | ||
9197 | enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog) | |
f1eead9e AN |
9198 | { |
9199 | return prog->expected_attach_type; | |
9200 | } | |
9201 | ||
93442f13 | 9202 | int bpf_program__set_expected_attach_type(struct bpf_program *prog, |
16962b24 | 9203 | enum bpf_attach_type type) |
d7be143b | 9204 | { |
93442f13 GS |
9205 | if (prog->obj->loaded) |
9206 | return libbpf_err(-EBUSY); | |
9207 | ||
d7be143b | 9208 | prog->expected_attach_type = type; |
93442f13 | 9209 | return 0; |
d7be143b AI |
9210 | } |
9211 | ||
a6ca7158 AN |
9212 | __u32 bpf_program__flags(const struct bpf_program *prog) |
9213 | { | |
9214 | return prog->prog_flags; | |
9215 | } | |
9216 | ||
8cccee9e | 9217 | int bpf_program__set_flags(struct bpf_program *prog, __u32 flags) |
a6ca7158 AN |
9218 | { |
9219 | if (prog->obj->loaded) | |
9220 | return libbpf_err(-EBUSY); | |
9221 | ||
8cccee9e | 9222 | prog->prog_flags = flags; |
a6ca7158 AN |
9223 | return 0; |
9224 | } | |
9225 | ||
dbdd2c7f AN |
9226 | __u32 bpf_program__log_level(const struct bpf_program *prog) |
9227 | { | |
9228 | return prog->log_level; | |
9229 | } | |
9230 | ||
9231 | int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level) | |
9232 | { | |
9233 | if (prog->obj->loaded) | |
9234 | return libbpf_err(-EBUSY); | |
9235 | ||
9236 | prog->log_level = log_level; | |
9237 | return 0; | |
9238 | } | |
9239 | ||
b3ce9079 AN |
9240 | const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size) |
9241 | { | |
9242 | *log_size = prog->log_size; | |
9243 | return prog->log_buf; | |
9244 | } | |
9245 | ||
9246 | int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size) | |
9247 | { | |
9248 | if (log_size && !log_buf) | |
9249 | return -EINVAL; | |
9250 | if (prog->log_size > UINT_MAX) | |
9251 | return -EINVAL; | |
9252 | if (prog->obj->loaded) | |
9253 | return -EBUSY; | |
9254 | ||
9255 | prog->log_buf = log_buf; | |
9256 | prog->log_size = log_size; | |
a6ca7158 AN |
9257 | return 0; |
9258 | } | |
9259 | ||
15ea31fa | 9260 | #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \ |
697f104d | 9261 | .sec = (char *)sec_pfx, \ |
d7a18ea7 | 9262 | .prog_type = BPF_PROG_TYPE_##ptype, \ |
15ea31fa AN |
9263 | .expected_attach_type = atype, \ |
9264 | .cookie = (long)(flags), \ | |
4fa5bcfe | 9265 | .prog_prepare_load_fn = libbpf_prepare_prog_load, \ |
d7a18ea7 AN |
9266 | __VA_ARGS__ \ |
9267 | } | |
9268 | ||
4fa5bcfe | 9269 | static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
39f8dc43 | 9270 | static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
708ac5be | 9271 | static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
2e4913e0 | 9272 | static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
4fa5bcfe AN |
9273 | static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
9274 | static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
9275 | static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
ddc6b049 | 9276 | static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
2ca178f0 | 9277 | static int attach_kprobe_session(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
5bfdd32d | 9278 | static int attach_uprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
4fa5bcfe AN |
9279 | static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
9280 | static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
d7a18ea7 | 9281 | |
d7a18ea7 | 9282 | static const struct bpf_sec_def section_defs[] = { |
450b167f AN |
9283 | SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE), |
9284 | SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE), | |
9285 | SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE), | |
9af8efc4 | 9286 | SEC_DEF("kprobe+", KPROBE, 0, SEC_NONE, attach_kprobe), |
39f8dc43 | 9287 | SEC_DEF("uprobe+", KPROBE, 0, SEC_NONE, attach_uprobe), |
c4cac71f | 9288 | SEC_DEF("uprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe), |
9af8efc4 | 9289 | SEC_DEF("kretprobe+", KPROBE, 0, SEC_NONE, attach_kprobe), |
39f8dc43 | 9290 | SEC_DEF("uretprobe+", KPROBE, 0, SEC_NONE, attach_uprobe), |
c4cac71f | 9291 | SEC_DEF("uretprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe), |
9af8efc4 AN |
9292 | SEC_DEF("kprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi), |
9293 | SEC_DEF("kretprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi), | |
2ca178f0 | 9294 | SEC_DEF("kprobe.session+", KPROBE, BPF_TRACE_KPROBE_SESSION, SEC_NONE, attach_kprobe_session), |
5bfdd32d JO |
9295 | SEC_DEF("uprobe.multi+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_NONE, attach_uprobe_multi), |
9296 | SEC_DEF("uretprobe.multi+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_NONE, attach_uprobe_multi), | |
9297 | SEC_DEF("uprobe.multi.s+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_SLEEPABLE, attach_uprobe_multi), | |
9298 | SEC_DEF("uretprobe.multi.s+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_SLEEPABLE, attach_uprobe_multi), | |
708ac5be AN |
9299 | SEC_DEF("ksyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall), |
9300 | SEC_DEF("kretsyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall), | |
5902da6d JO |
9301 | SEC_DEF("usdt+", KPROBE, 0, SEC_USDT, attach_usdt), |
9302 | SEC_DEF("usdt.s+", KPROBE, 0, SEC_USDT | SEC_SLEEPABLE, attach_usdt), | |
fe20ce3a DB |
9303 | SEC_DEF("tc/ingress", SCHED_CLS, BPF_TCX_INGRESS, SEC_NONE), /* alias for tcx */ |
9304 | SEC_DEF("tc/egress", SCHED_CLS, BPF_TCX_EGRESS, SEC_NONE), /* alias for tcx */ | |
9305 | SEC_DEF("tcx/ingress", SCHED_CLS, BPF_TCX_INGRESS, SEC_NONE), | |
9306 | SEC_DEF("tcx/egress", SCHED_CLS, BPF_TCX_EGRESS, SEC_NONE), | |
9307 | SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE), /* deprecated / legacy, use tcx */ | |
9308 | SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE), /* deprecated / legacy, use tcx */ | |
9309 | SEC_DEF("action", SCHED_ACT, 0, SEC_NONE), /* deprecated / legacy, use tcx */ | |
05c31b4a DB |
9310 | SEC_DEF("netkit/primary", SCHED_CLS, BPF_NETKIT_PRIMARY, SEC_NONE), |
9311 | SEC_DEF("netkit/peer", SCHED_CLS, BPF_NETKIT_PEER, SEC_NONE), | |
9af8efc4 AN |
9312 | SEC_DEF("tracepoint+", TRACEPOINT, 0, SEC_NONE, attach_tp), |
9313 | SEC_DEF("tp+", TRACEPOINT, 0, SEC_NONE, attach_tp), | |
9314 | SEC_DEF("raw_tracepoint+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
9315 | SEC_DEF("raw_tp+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
9316 | SEC_DEF("raw_tracepoint.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), | |
9317 | SEC_DEF("raw_tp.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), | |
cc7d8f2c AN |
9318 | SEC_DEF("tp_btf+", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace), |
9319 | SEC_DEF("fentry+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace), | |
9320 | SEC_DEF("fmod_ret+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace), | |
9321 | SEC_DEF("fexit+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace), | |
9322 | SEC_DEF("fentry.s+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
9323 | SEC_DEF("fmod_ret.s+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
9324 | SEC_DEF("fexit.s+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
9325 | SEC_DEF("freplace+", EXT, 0, SEC_ATTACH_BTF, attach_trace), | |
9326 | SEC_DEF("lsm+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm), | |
9327 | SEC_DEF("lsm.s+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm), | |
bffcf348 | 9328 | SEC_DEF("lsm_cgroup+", LSM, BPF_LSM_CGROUP, SEC_ATTACH_BTF), |
cc7d8f2c AN |
9329 | SEC_DEF("iter+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter), |
9330 | SEC_DEF("iter.s+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter), | |
15ea31fa | 9331 | SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE), |
082c4bfb | 9332 | SEC_DEF("xdp.frags/devmap", XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS), |
4a4d4cee | 9333 | SEC_DEF("xdp/devmap", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE), |
082c4bfb | 9334 | SEC_DEF("xdp.frags/cpumap", XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS), |
4a4d4cee | 9335 | SEC_DEF("xdp/cpumap", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE), |
082c4bfb | 9336 | SEC_DEF("xdp.frags", XDP, BPF_XDP, SEC_XDP_FRAGS), |
450b167f AN |
9337 | SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT), |
9338 | SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE), | |
9339 | SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE), | |
9340 | SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE), | |
9341 | SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE), | |
9342 | SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE), | |
9343 | SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT), | |
9344 | SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT), | |
9345 | SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT), | |
61df5756 | 9346 | SEC_DEF("sk_skb/verdict", SK_SKB, BPF_SK_SKB_VERDICT, SEC_ATTACHABLE_OPT), |
450b167f AN |
9347 | SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE), |
9348 | SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT), | |
9349 | SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT), | |
9350 | SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT), | |
9351 | SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT), | |
9352 | SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT), | |
9353 | SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE), | |
9354 | SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE), | |
9355 | SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE), | |
9356 | SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT), | |
9357 | SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE), | |
9358 | SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE), | |
9359 | SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE), | |
9360 | SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE), | |
9361 | SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE), | |
9362 | SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE), | |
bf90438c | 9363 | SEC_DEF("cgroup/connect_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_CONNECT, SEC_ATTACHABLE), |
450b167f AN |
9364 | SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE), |
9365 | SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE), | |
bf90438c | 9366 | SEC_DEF("cgroup/sendmsg_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_SENDMSG, SEC_ATTACHABLE), |
450b167f AN |
9367 | SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE), |
9368 | SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE), | |
bf90438c | 9369 | SEC_DEF("cgroup/recvmsg_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_RECVMSG, SEC_ATTACHABLE), |
450b167f AN |
9370 | SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE), |
9371 | SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE), | |
bf90438c | 9372 | SEC_DEF("cgroup/getpeername_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_GETPEERNAME, SEC_ATTACHABLE), |
450b167f AN |
9373 | SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE), |
9374 | SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE), | |
bf90438c | 9375 | SEC_DEF("cgroup/getsockname_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_GETSOCKNAME, SEC_ATTACHABLE), |
450b167f AN |
9376 | SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE), |
9377 | SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE), | |
9378 | SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE), | |
9379 | SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT), | |
dd94d45c | 9380 | SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), |
913b2255 | 9381 | SEC_DEF("struct_ops.s+", STRUCT_OPS, 0, SEC_SLEEPABLE), |
450b167f | 9382 | SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE), |
132328e8 | 9383 | SEC_DEF("netfilter", NETFILTER, BPF_NETFILTER, SEC_NONE), |
583c9009 | 9384 | }; |
d7be143b | 9385 | |
697f104d AN |
9386 | int libbpf_register_prog_handler(const char *sec, |
9387 | enum bpf_prog_type prog_type, | |
9388 | enum bpf_attach_type exp_attach_type, | |
9389 | const struct libbpf_prog_handler_opts *opts) | |
d7a18ea7 | 9390 | { |
697f104d | 9391 | struct bpf_sec_def *sec_def; |
d7a18ea7 | 9392 | |
697f104d AN |
9393 | if (!OPTS_VALID(opts, libbpf_prog_handler_opts)) |
9394 | return libbpf_err(-EINVAL); | |
dd94d45c | 9395 | |
697f104d AN |
9396 | if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */ |
9397 | return libbpf_err(-E2BIG); | |
dd94d45c | 9398 | |
697f104d AN |
9399 | if (sec) { |
9400 | sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1, | |
9401 | sizeof(*sec_def)); | |
9402 | if (!sec_def) | |
9403 | return libbpf_err(-ENOMEM); | |
dd94d45c | 9404 | |
697f104d AN |
9405 | custom_sec_defs = sec_def; |
9406 | sec_def = &custom_sec_defs[custom_sec_def_cnt]; | |
9407 | } else { | |
9408 | if (has_custom_fallback_def) | |
9409 | return libbpf_err(-EBUSY); | |
dd94d45c | 9410 | |
697f104d AN |
9411 | sec_def = &custom_fallback_def; |
9412 | } | |
9413 | ||
9414 | sec_def->sec = sec ? strdup(sec) : NULL; | |
9415 | if (sec && !sec_def->sec) | |
9416 | return libbpf_err(-ENOMEM); | |
9417 | ||
9418 | sec_def->prog_type = prog_type; | |
9419 | sec_def->expected_attach_type = exp_attach_type; | |
9420 | sec_def->cookie = OPTS_GET(opts, cookie, 0); | |
9421 | ||
9422 | sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL); | |
9423 | sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL); | |
9424 | sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL); | |
9425 | ||
9426 | sec_def->handler_id = ++last_custom_sec_def_handler_id; | |
9427 | ||
9428 | if (sec) | |
9429 | custom_sec_def_cnt++; | |
9430 | else | |
9431 | has_custom_fallback_def = true; | |
9432 | ||
9433 | return sec_def->handler_id; | |
9434 | } | |
9435 | ||
9436 | int libbpf_unregister_prog_handler(int handler_id) | |
9437 | { | |
9438 | struct bpf_sec_def *sec_defs; | |
9439 | int i; | |
9440 | ||
9441 | if (handler_id <= 0) | |
9442 | return libbpf_err(-EINVAL); | |
9443 | ||
9444 | if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) { | |
9445 | memset(&custom_fallback_def, 0, sizeof(custom_fallback_def)); | |
9446 | has_custom_fallback_def = false; | |
9447 | return 0; | |
9448 | } | |
9449 | ||
9450 | for (i = 0; i < custom_sec_def_cnt; i++) { | |
9451 | if (custom_sec_defs[i].handler_id == handler_id) | |
9452 | break; | |
9453 | } | |
9454 | ||
9455 | if (i == custom_sec_def_cnt) | |
9456 | return libbpf_err(-ENOENT); | |
9457 | ||
9458 | free(custom_sec_defs[i].sec); | |
9459 | for (i = i + 1; i < custom_sec_def_cnt; i++) | |
9460 | custom_sec_defs[i - 1] = custom_sec_defs[i]; | |
9461 | custom_sec_def_cnt--; | |
9462 | ||
9463 | /* try to shrink the array, but it's ok if we couldn't */ | |
9464 | sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs)); | |
8a0260db AN |
9465 | /* if new count is zero, reallocarray can return a valid NULL result; |
9466 | * in this case the previous pointer will be freed, so we *have to* | |
9467 | * reassign old pointer to the new value (even if it's NULL) | |
9468 | */ | |
9469 | if (sec_defs || custom_sec_def_cnt == 0) | |
697f104d AN |
9470 | custom_sec_defs = sec_defs; |
9471 | ||
9472 | return 0; | |
9473 | } | |
9474 | ||
450b167f | 9475 | static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name) |
697f104d AN |
9476 | { |
9477 | size_t len = strlen(sec_def->sec); | |
9478 | ||
9479 | /* "type/" always has to have proper SEC("type/extras") form */ | |
9480 | if (sec_def->sec[len - 1] == '/') { | |
9481 | if (str_has_pfx(sec_name, sec_def->sec)) | |
9482 | return true; | |
9483 | return false; | |
9484 | } | |
9485 | ||
9486 | /* "type+" means it can be either exact SEC("type") or | |
9487 | * well-formed SEC("type/extras") with proper '/' separator | |
9488 | */ | |
9489 | if (sec_def->sec[len - 1] == '+') { | |
9490 | len--; | |
9491 | /* not even a prefix */ | |
9492 | if (strncmp(sec_name, sec_def->sec, len) != 0) | |
9493 | return false; | |
9494 | /* exact match or has '/' separator */ | |
9495 | if (sec_name[len] == '\0' || sec_name[len] == '/') | |
9496 | return true; | |
9497 | return false; | |
9498 | } | |
9499 | ||
697f104d AN |
9500 | return strcmp(sec_name, sec_def->sec) == 0; |
9501 | } | |
9502 | ||
9503 | static const struct bpf_sec_def *find_sec_def(const char *sec_name) | |
9504 | { | |
9505 | const struct bpf_sec_def *sec_def; | |
9506 | int i, n; | |
697f104d AN |
9507 | |
9508 | n = custom_sec_def_cnt; | |
9509 | for (i = 0; i < n; i++) { | |
9510 | sec_def = &custom_sec_defs[i]; | |
450b167f | 9511 | if (sec_def_matches(sec_def, sec_name)) |
697f104d AN |
9512 | return sec_def; |
9513 | } | |
9514 | ||
9515 | n = ARRAY_SIZE(section_defs); | |
9516 | for (i = 0; i < n; i++) { | |
9517 | sec_def = §ion_defs[i]; | |
450b167f | 9518 | if (sec_def_matches(sec_def, sec_name)) |
dd94d45c | 9519 | return sec_def; |
d7a18ea7 | 9520 | } |
697f104d AN |
9521 | |
9522 | if (has_custom_fallback_def) | |
9523 | return &custom_fallback_def; | |
9524 | ||
d7a18ea7 AN |
9525 | return NULL; |
9526 | } | |
9527 | ||
697f104d AN |
9528 | #define MAX_TYPE_NAME_SIZE 32 |
9529 | ||
c76e4c22 TS |
9530 | static char *libbpf_get_type_names(bool attach_type) |
9531 | { | |
d7a18ea7 | 9532 | int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE; |
c76e4c22 TS |
9533 | char *buf; |
9534 | ||
9535 | buf = malloc(len); | |
9536 | if (!buf) | |
9537 | return NULL; | |
9538 | ||
9539 | buf[0] = '\0'; | |
9540 | /* Forge string buf with all available names */ | |
d7a18ea7 | 9541 | for (i = 0; i < ARRAY_SIZE(section_defs); i++) { |
15ea31fa AN |
9542 | const struct bpf_sec_def *sec_def = §ion_defs[i]; |
9543 | ||
9544 | if (attach_type) { | |
4fa5bcfe | 9545 | if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load) |
15ea31fa AN |
9546 | continue; |
9547 | ||
9548 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
9549 | continue; | |
9550 | } | |
c76e4c22 | 9551 | |
d7a18ea7 | 9552 | if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) { |
c76e4c22 TS |
9553 | free(buf); |
9554 | return NULL; | |
9555 | } | |
9556 | strcat(buf, " "); | |
d7a18ea7 | 9557 | strcat(buf, section_defs[i].sec); |
c76e4c22 TS |
9558 | } |
9559 | ||
9560 | return buf; | |
9561 | } | |
9562 | ||
b60df2a0 JK |
9563 | int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type, |
9564 | enum bpf_attach_type *expected_attach_type) | |
583c9009 | 9565 | { |
d7a18ea7 | 9566 | const struct bpf_sec_def *sec_def; |
c76e4c22 | 9567 | char *type_names; |
583c9009 | 9568 | |
b60df2a0 | 9569 | if (!name) |
e9fc3ce9 | 9570 | return libbpf_err(-EINVAL); |
583c9009 | 9571 | |
d7a18ea7 AN |
9572 | sec_def = find_sec_def(name); |
9573 | if (sec_def) { | |
9574 | *prog_type = sec_def->prog_type; | |
9575 | *expected_attach_type = sec_def->expected_attach_type; | |
b60df2a0 JK |
9576 | return 0; |
9577 | } | |
d7a18ea7 | 9578 | |
4a3d6c6a | 9579 | pr_debug("failed to guess program type from ELF section '%s'\n", name); |
c76e4c22 TS |
9580 | type_names = libbpf_get_type_names(false); |
9581 | if (type_names != NULL) { | |
3f519353 | 9582 | pr_debug("supported section(type) names are:%s\n", type_names); |
c76e4c22 TS |
9583 | free(type_names); |
9584 | } | |
9585 | ||
e9fc3ce9 | 9586 | return libbpf_err(-ESRCH); |
b60df2a0 | 9587 | } |
583c9009 | 9588 | |
ccde5760 DM |
9589 | const char *libbpf_bpf_attach_type_str(enum bpf_attach_type t) |
9590 | { | |
9591 | if (t < 0 || t >= ARRAY_SIZE(attach_type_name)) | |
9592 | return NULL; | |
9593 | ||
9594 | return attach_type_name[t]; | |
9595 | } | |
9596 | ||
ba5d1b58 DM |
9597 | const char *libbpf_bpf_link_type_str(enum bpf_link_type t) |
9598 | { | |
9599 | if (t < 0 || t >= ARRAY_SIZE(link_type_name)) | |
9600 | return NULL; | |
9601 | ||
9602 | return link_type_name[t]; | |
9603 | } | |
9604 | ||
3e6dc020 DM |
9605 | const char *libbpf_bpf_map_type_str(enum bpf_map_type t) |
9606 | { | |
9607 | if (t < 0 || t >= ARRAY_SIZE(map_type_name)) | |
9608 | return NULL; | |
9609 | ||
9610 | return map_type_name[t]; | |
9611 | } | |
9612 | ||
d18616e7 DM |
9613 | const char *libbpf_bpf_prog_type_str(enum bpf_prog_type t) |
9614 | { | |
9615 | if (t < 0 || t >= ARRAY_SIZE(prog_type_name)) | |
9616 | return NULL; | |
9617 | ||
9618 | return prog_type_name[t]; | |
9619 | } | |
9620 | ||
590a0088 | 9621 | static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj, |
809a69d6 | 9622 | int sec_idx, |
590a0088 MKL |
9623 | size_t offset) |
9624 | { | |
9625 | struct bpf_map *map; | |
9626 | size_t i; | |
9627 | ||
9628 | for (i = 0; i < obj->nr_maps; i++) { | |
9629 | map = &obj->maps[i]; | |
9630 | if (!bpf_map__is_struct_ops(map)) | |
9631 | continue; | |
809a69d6 KFL |
9632 | if (map->sec_idx == sec_idx && |
9633 | map->sec_offset <= offset && | |
590a0088 MKL |
9634 | offset - map->sec_offset < map->def.value_size) |
9635 | return map; | |
9636 | } | |
9637 | ||
9638 | return NULL; | |
9639 | } | |
9640 | ||
69e4a9d2 KFL |
9641 | /* Collect the reloc from ELF, populate the st_ops->progs[], and update |
9642 | * st_ops->data for shadow type. | |
9643 | */ | |
646f02ff | 9644 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 9645 | Elf64_Shdr *shdr, Elf_Data *data) |
590a0088 MKL |
9646 | { |
9647 | const struct btf_member *member; | |
9648 | struct bpf_struct_ops *st_ops; | |
9649 | struct bpf_program *prog; | |
9650 | unsigned int shdr_idx; | |
9651 | const struct btf *btf; | |
9652 | struct bpf_map *map; | |
7e06aad5 | 9653 | unsigned int moff, insn_idx; |
590a0088 | 9654 | const char *name; |
1d1a3bcf | 9655 | __u32 member_idx; |
ad23b723 AN |
9656 | Elf64_Sym *sym; |
9657 | Elf64_Rel *rel; | |
590a0088 MKL |
9658 | int i, nrels; |
9659 | ||
590a0088 MKL |
9660 | btf = obj->btf; |
9661 | nrels = shdr->sh_size / shdr->sh_entsize; | |
9662 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
9663 | rel = elf_rel_by_idx(data, i); |
9664 | if (!rel) { | |
590a0088 MKL |
9665 | pr_warn("struct_ops reloc: failed to get %d reloc\n", i); |
9666 | return -LIBBPF_ERRNO__FORMAT; | |
9667 | } | |
9668 | ||
ad23b723 AN |
9669 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); |
9670 | if (!sym) { | |
590a0088 | 9671 | pr_warn("struct_ops reloc: symbol %zx not found\n", |
ad23b723 | 9672 | (size_t)ELF64_R_SYM(rel->r_info)); |
590a0088 MKL |
9673 | return -LIBBPF_ERRNO__FORMAT; |
9674 | } | |
9675 | ||
ad23b723 | 9676 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
809a69d6 | 9677 | map = find_struct_ops_map_by_offset(obj, shdr->sh_info, rel->r_offset); |
590a0088 | 9678 | if (!map) { |
ad23b723 AN |
9679 | pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n", |
9680 | (size_t)rel->r_offset); | |
590a0088 MKL |
9681 | return -EINVAL; |
9682 | } | |
9683 | ||
ad23b723 AN |
9684 | moff = rel->r_offset - map->sec_offset; |
9685 | shdr_idx = sym->st_shndx; | |
590a0088 | 9686 | st_ops = map->st_ops; |
ad23b723 | 9687 | 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 | 9688 | map->name, |
ad23b723 AN |
9689 | (long long)(rel->r_info >> 32), |
9690 | (long long)sym->st_value, | |
9691 | shdr_idx, (size_t)rel->r_offset, | |
9692 | map->sec_offset, sym->st_name, name); | |
590a0088 MKL |
9693 | |
9694 | if (shdr_idx >= SHN_LORESERVE) { | |
ad23b723 AN |
9695 | pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n", |
9696 | map->name, (size_t)rel->r_offset, shdr_idx); | |
590a0088 MKL |
9697 | return -LIBBPF_ERRNO__RELOC; |
9698 | } | |
ad23b723 | 9699 | if (sym->st_value % BPF_INSN_SZ) { |
7e06aad5 | 9700 | pr_warn("struct_ops reloc %s: invalid target program offset %llu\n", |
ad23b723 | 9701 | map->name, (unsigned long long)sym->st_value); |
7e06aad5 AN |
9702 | return -LIBBPF_ERRNO__FORMAT; |
9703 | } | |
ad23b723 | 9704 | insn_idx = sym->st_value / BPF_INSN_SZ; |
590a0088 MKL |
9705 | |
9706 | member = find_member_by_offset(st_ops->type, moff * 8); | |
9707 | if (!member) { | |
9708 | pr_warn("struct_ops reloc %s: cannot find member at moff %u\n", | |
9709 | map->name, moff); | |
9710 | return -EINVAL; | |
9711 | } | |
9712 | member_idx = member - btf_members(st_ops->type); | |
9713 | name = btf__name_by_offset(btf, member->name_off); | |
9714 | ||
9715 | if (!resolve_func_ptr(btf, member->type, NULL)) { | |
9716 | pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n", | |
9717 | map->name, name); | |
9718 | return -EINVAL; | |
9719 | } | |
9720 | ||
7e06aad5 | 9721 | prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx); |
590a0088 MKL |
9722 | if (!prog) { |
9723 | pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n", | |
9724 | map->name, shdr_idx, name); | |
9725 | return -EINVAL; | |
9726 | } | |
9727 | ||
91b4d1d1 AN |
9728 | /* prevent the use of BPF prog with invalid type */ |
9729 | if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) { | |
9730 | pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n", | |
9731 | map->name, prog->name); | |
9732 | return -EINVAL; | |
9733 | } | |
590a0088 | 9734 | |
590a0088 | 9735 | st_ops->progs[member_idx] = prog; |
69e4a9d2 KFL |
9736 | |
9737 | /* st_ops->data will be exposed to users, being returned by | |
9738 | * bpf_map__initial_value() as a pointer to the shadow | |
9739 | * type. All function pointers in the original struct type | |
9740 | * should be converted to a pointer to struct bpf_program | |
9741 | * in the shadow type. | |
9742 | */ | |
9743 | *((struct bpf_program **)(st_ops->data + moff)) = prog; | |
590a0088 MKL |
9744 | } |
9745 | ||
9746 | return 0; | |
590a0088 MKL |
9747 | } |
9748 | ||
a6ed02ca | 9749 | #define BTF_TRACE_PREFIX "btf_trace_" |
1e092a03 | 9750 | #define BTF_LSM_PREFIX "bpf_lsm_" |
21aef70e | 9751 | #define BTF_ITER_PREFIX "bpf_iter_" |
a6ed02ca KS |
9752 | #define BTF_MAX_NAME_SIZE 128 |
9753 | ||
67234743 AS |
9754 | void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type, |
9755 | const char **prefix, int *kind) | |
9756 | { | |
9757 | switch (attach_type) { | |
9758 | case BPF_TRACE_RAW_TP: | |
9759 | *prefix = BTF_TRACE_PREFIX; | |
9760 | *kind = BTF_KIND_TYPEDEF; | |
9761 | break; | |
9762 | case BPF_LSM_MAC: | |
bffcf348 | 9763 | case BPF_LSM_CGROUP: |
67234743 AS |
9764 | *prefix = BTF_LSM_PREFIX; |
9765 | *kind = BTF_KIND_FUNC; | |
9766 | break; | |
9767 | case BPF_TRACE_ITER: | |
9768 | *prefix = BTF_ITER_PREFIX; | |
9769 | *kind = BTF_KIND_FUNC; | |
9770 | break; | |
9771 | default: | |
9772 | *prefix = ""; | |
9773 | *kind = BTF_KIND_FUNC; | |
9774 | } | |
9775 | } | |
9776 | ||
a6ed02ca KS |
9777 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
9778 | const char *name, __u32 kind) | |
9779 | { | |
9780 | char btf_type_name[BTF_MAX_NAME_SIZE]; | |
9781 | int ret; | |
9782 | ||
9783 | ret = snprintf(btf_type_name, sizeof(btf_type_name), | |
9784 | "%s%s", prefix, name); | |
9785 | /* snprintf returns the number of characters written excluding the | |
c139e40a | 9786 | * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it |
a6ed02ca KS |
9787 | * indicates truncation. |
9788 | */ | |
9789 | if (ret < 0 || ret >= sizeof(btf_type_name)) | |
9790 | return -ENAMETOOLONG; | |
9791 | return btf__find_by_name_kind(btf, btf_type_name, kind); | |
9792 | } | |
9793 | ||
91abb4a6 AN |
9794 | static inline int find_attach_btf_id(struct btf *btf, const char *name, |
9795 | enum bpf_attach_type attach_type) | |
a6ed02ca | 9796 | { |
67234743 AS |
9797 | const char *prefix; |
9798 | int kind; | |
a6ed02ca | 9799 | |
67234743 AS |
9800 | btf_get_kernel_prefix_kind(attach_type, &prefix, &kind); |
9801 | return find_btf_by_prefix_kind(btf, prefix, name, kind); | |
a6ed02ca KS |
9802 | } |
9803 | ||
b8c54ea4 AS |
9804 | int libbpf_find_vmlinux_btf_id(const char *name, |
9805 | enum bpf_attach_type attach_type) | |
12a8654b | 9806 | { |
a6ed02ca | 9807 | struct btf *btf; |
3521ffa2 | 9808 | int err; |
12a8654b | 9809 | |
a710eed3 | 9810 | btf = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
9811 | err = libbpf_get_error(btf); |
9812 | if (err) { | |
12a8654b | 9813 | pr_warn("vmlinux BTF is not found\n"); |
e9fc3ce9 | 9814 | return libbpf_err(err); |
12a8654b AS |
9815 | } |
9816 | ||
91abb4a6 AN |
9817 | err = find_attach_btf_id(btf, name, attach_type); |
9818 | if (err <= 0) | |
9819 | pr_warn("%s is not found in vmlinux BTF\n", name); | |
9820 | ||
3521ffa2 | 9821 | btf__free(btf); |
e9fc3ce9 | 9822 | return libbpf_err(err); |
b8c54ea4 AS |
9823 | } |
9824 | ||
e7bf94db AS |
9825 | static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd) |
9826 | { | |
813847a3 | 9827 | struct bpf_prog_info info; |
ebc7b50a | 9828 | __u32 info_len = sizeof(info); |
6cc93e2f | 9829 | struct btf *btf; |
6d2d73cd | 9830 | int err; |
e7bf94db | 9831 | |
813847a3 | 9832 | memset(&info, 0, info_len); |
629dfc66 | 9833 | err = bpf_prog_get_info_by_fd(attach_prog_fd, &info, &info_len); |
e9fc3ce9 | 9834 | if (err) { |
629dfc66 | 9835 | pr_warn("failed bpf_prog_get_info_by_fd for FD %d: %d\n", |
ebc7b50a | 9836 | attach_prog_fd, err); |
e9fc3ce9 | 9837 | return err; |
e7bf94db | 9838 | } |
6d2d73cd QM |
9839 | |
9840 | err = -EINVAL; | |
ebc7b50a | 9841 | if (!info.btf_id) { |
e7bf94db AS |
9842 | pr_warn("The target program doesn't have BTF\n"); |
9843 | goto out; | |
9844 | } | |
ebc7b50a DM |
9845 | btf = btf__load_from_kernel_by_id(info.btf_id); |
9846 | err = libbpf_get_error(btf); | |
9847 | if (err) { | |
9848 | pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err); | |
e7bf94db AS |
9849 | goto out; |
9850 | } | |
9851 | err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC); | |
9852 | btf__free(btf); | |
9853 | if (err <= 0) { | |
9854 | pr_warn("%s is not found in prog's BTF\n", name); | |
9855 | goto out; | |
9856 | } | |
9857 | out: | |
e7bf94db AS |
9858 | return err; |
9859 | } | |
9860 | ||
91abb4a6 AN |
9861 | static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name, |
9862 | enum bpf_attach_type attach_type, | |
9863 | int *btf_obj_fd, int *btf_type_id) | |
9864 | { | |
8f8a0242 VM |
9865 | int ret, i, mod_len; |
9866 | const char *fn_name, *mod_name = NULL; | |
91abb4a6 | 9867 | |
8f8a0242 VM |
9868 | fn_name = strchr(attach_name, ':'); |
9869 | if (fn_name) { | |
9870 | mod_name = attach_name; | |
9871 | mod_len = fn_name - mod_name; | |
9872 | fn_name++; | |
9873 | } | |
9874 | ||
9875 | if (!mod_name || strncmp(mod_name, "vmlinux", mod_len) == 0) { | |
9876 | ret = find_attach_btf_id(obj->btf_vmlinux, | |
9877 | mod_name ? fn_name : attach_name, | |
9878 | attach_type); | |
9879 | if (ret > 0) { | |
9880 | *btf_obj_fd = 0; /* vmlinux BTF */ | |
9881 | *btf_type_id = ret; | |
9882 | return 0; | |
9883 | } | |
9884 | if (ret != -ENOENT) | |
9885 | return ret; | |
91abb4a6 | 9886 | } |
91abb4a6 AN |
9887 | |
9888 | ret = load_module_btfs(obj); | |
9889 | if (ret) | |
9890 | return ret; | |
9891 | ||
9892 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
9893 | const struct module_btf *mod = &obj->btf_modules[i]; | |
9894 | ||
8f8a0242 VM |
9895 | if (mod_name && strncmp(mod->name, mod_name, mod_len) != 0) |
9896 | continue; | |
9897 | ||
9898 | ret = find_attach_btf_id(mod->btf, | |
9899 | mod_name ? fn_name : attach_name, | |
9900 | attach_type); | |
91abb4a6 AN |
9901 | if (ret > 0) { |
9902 | *btf_obj_fd = mod->fd; | |
9903 | *btf_type_id = ret; | |
9904 | return 0; | |
9905 | } | |
9906 | if (ret == -ENOENT) | |
9907 | continue; | |
9908 | ||
9909 | return ret; | |
9910 | } | |
9911 | ||
9912 | return -ESRCH; | |
9913 | } | |
9914 | ||
15ea31fa AN |
9915 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
9916 | int *btf_obj_fd, int *btf_type_id) | |
b8c54ea4 | 9917 | { |
a6ed02ca KS |
9918 | enum bpf_attach_type attach_type = prog->expected_attach_type; |
9919 | __u32 attach_prog_fd = prog->attach_prog_fd; | |
b6291a6f | 9920 | int err = 0; |
b8c54ea4 | 9921 | |
91abb4a6 | 9922 | /* BPF program's BTF ID */ |
749c202c AN |
9923 | if (prog->type == BPF_PROG_TYPE_EXT || attach_prog_fd) { |
9924 | if (!attach_prog_fd) { | |
9925 | pr_warn("prog '%s': attach program FD is not set\n", prog->name); | |
9926 | return -EINVAL; | |
9927 | } | |
91abb4a6 AN |
9928 | err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd); |
9929 | if (err < 0) { | |
749c202c AN |
9930 | pr_warn("prog '%s': failed to find BPF program (FD %d) BTF ID for '%s': %d\n", |
9931 | prog->name, attach_prog_fd, attach_name, err); | |
91abb4a6 AN |
9932 | return err; |
9933 | } | |
9934 | *btf_obj_fd = 0; | |
9935 | *btf_type_id = err; | |
9936 | return 0; | |
9937 | } | |
9938 | ||
9939 | /* kernel/module BTF ID */ | |
67234743 AS |
9940 | if (prog->obj->gen_loader) { |
9941 | bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type); | |
9942 | *btf_obj_fd = 0; | |
9943 | *btf_type_id = 1; | |
9944 | } else { | |
9e926acd KFL |
9945 | err = find_kernel_btf_id(prog->obj, attach_name, |
9946 | attach_type, btf_obj_fd, | |
9947 | btf_type_id); | |
67234743 | 9948 | } |
91abb4a6 | 9949 | if (err) { |
749c202c AN |
9950 | pr_warn("prog '%s': failed to find kernel BTF type ID of '%s': %d\n", |
9951 | prog->name, attach_name, err); | |
b8c54ea4 | 9952 | return err; |
12a8654b | 9953 | } |
91abb4a6 | 9954 | return 0; |
12a8654b AS |
9955 | } |
9956 | ||
956b620f AI |
9957 | int libbpf_attach_type_by_name(const char *name, |
9958 | enum bpf_attach_type *attach_type) | |
9959 | { | |
c76e4c22 | 9960 | char *type_names; |
b6291a6f | 9961 | const struct bpf_sec_def *sec_def; |
956b620f AI |
9962 | |
9963 | if (!name) | |
e9fc3ce9 | 9964 | return libbpf_err(-EINVAL); |
956b620f | 9965 | |
b6291a6f AN |
9966 | sec_def = find_sec_def(name); |
9967 | if (!sec_def) { | |
9968 | pr_debug("failed to guess attach type based on ELF section name '%s'\n", name); | |
9969 | type_names = libbpf_get_type_names(true); | |
9970 | if (type_names != NULL) { | |
9971 | pr_debug("attachable section(type) names are:%s\n", type_names); | |
9972 | free(type_names); | |
9973 | } | |
9974 | ||
9975 | return libbpf_err(-EINVAL); | |
c76e4c22 TS |
9976 | } |
9977 | ||
4fa5bcfe | 9978 | if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load) |
15ea31fa AN |
9979 | return libbpf_err(-EINVAL); |
9980 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
b6291a6f AN |
9981 | return libbpf_err(-EINVAL); |
9982 | ||
9983 | *attach_type = sec_def->expected_attach_type; | |
9984 | return 0; | |
956b620f AI |
9985 | } |
9986 | ||
a324aae3 | 9987 | int bpf_map__fd(const struct bpf_map *map) |
9d759a9b | 9988 | { |
f08c18e0 AN |
9989 | if (!map) |
9990 | return libbpf_err(-EINVAL); | |
9991 | if (!map_is_created(map)) | |
9992 | return -1; | |
9993 | return map->fd; | |
9d759a9b WN |
9994 | } |
9995 | ||
aed65917 AN |
9996 | static bool map_uses_real_name(const struct bpf_map *map) |
9997 | { | |
9998 | /* Since libbpf started to support custom .data.* and .rodata.* maps, | |
9999 | * their user-visible name differs from kernel-visible name. Users see | |
10000 | * such map's corresponding ELF section name as a map name. | |
10001 | * This check distinguishes .data/.rodata from .data.* and .rodata.* | |
10002 | * maps to know which name has to be returned to the user. | |
10003 | */ | |
10004 | if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0) | |
10005 | return true; | |
10006 | if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0) | |
10007 | return true; | |
10008 | return false; | |
10009 | } | |
10010 | ||
a324aae3 | 10011 | const char *bpf_map__name(const struct bpf_map *map) |
561bbcca | 10012 | { |
aed65917 AN |
10013 | if (!map) |
10014 | return NULL; | |
10015 | ||
10016 | if (map_uses_real_name(map)) | |
10017 | return map->real_name; | |
10018 | ||
10019 | return map->name; | |
561bbcca WN |
10020 | } |
10021 | ||
1bdb6c9a AN |
10022 | enum bpf_map_type bpf_map__type(const struct bpf_map *map) |
10023 | { | |
10024 | return map->def.type; | |
10025 | } | |
10026 | ||
10027 | int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type) | |
10028 | { | |
f08c18e0 | 10029 | if (map_is_created(map)) |
e9fc3ce9 | 10030 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
10031 | map->def.type = type; |
10032 | return 0; | |
10033 | } | |
10034 | ||
10035 | __u32 bpf_map__map_flags(const struct bpf_map *map) | |
10036 | { | |
10037 | return map->def.map_flags; | |
10038 | } | |
10039 | ||
10040 | int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags) | |
10041 | { | |
f08c18e0 | 10042 | if (map_is_created(map)) |
e9fc3ce9 | 10043 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
10044 | map->def.map_flags = flags; |
10045 | return 0; | |
10046 | } | |
10047 | ||
47512102 JK |
10048 | __u64 bpf_map__map_extra(const struct bpf_map *map) |
10049 | { | |
10050 | return map->map_extra; | |
10051 | } | |
10052 | ||
10053 | int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra) | |
10054 | { | |
f08c18e0 | 10055 | if (map_is_created(map)) |
47512102 JK |
10056 | return libbpf_err(-EBUSY); |
10057 | map->map_extra = map_extra; | |
10058 | return 0; | |
10059 | } | |
10060 | ||
1bdb6c9a AN |
10061 | __u32 bpf_map__numa_node(const struct bpf_map *map) |
10062 | { | |
10063 | return map->numa_node; | |
10064 | } | |
10065 | ||
10066 | int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node) | |
10067 | { | |
f08c18e0 | 10068 | if (map_is_created(map)) |
e9fc3ce9 | 10069 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
10070 | map->numa_node = numa_node; |
10071 | return 0; | |
10072 | } | |
10073 | ||
10074 | __u32 bpf_map__key_size(const struct bpf_map *map) | |
10075 | { | |
10076 | return map->def.key_size; | |
10077 | } | |
10078 | ||
10079 | int bpf_map__set_key_size(struct bpf_map *map, __u32 size) | |
10080 | { | |
f08c18e0 | 10081 | if (map_is_created(map)) |
e9fc3ce9 | 10082 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
10083 | map->def.key_size = size; |
10084 | return 0; | |
10085 | } | |
10086 | ||
10087 | __u32 bpf_map__value_size(const struct bpf_map *map) | |
10088 | { | |
10089 | return map->def.value_size; | |
10090 | } | |
10091 | ||
9d0a2331 JK |
10092 | static int map_btf_datasec_resize(struct bpf_map *map, __u32 size) |
10093 | { | |
10094 | struct btf *btf; | |
10095 | struct btf_type *datasec_type, *var_type; | |
10096 | struct btf_var_secinfo *var; | |
10097 | const struct btf_type *array_type; | |
10098 | const struct btf_array *array; | |
4c857a71 JK |
10099 | int vlen, element_sz, new_array_id; |
10100 | __u32 nr_elements; | |
9d0a2331 JK |
10101 | |
10102 | /* check btf existence */ | |
10103 | btf = bpf_object__btf(map->obj); | |
10104 | if (!btf) | |
10105 | return -ENOENT; | |
10106 | ||
10107 | /* verify map is datasec */ | |
10108 | datasec_type = btf_type_by_id(btf, bpf_map__btf_value_type_id(map)); | |
10109 | if (!btf_is_datasec(datasec_type)) { | |
10110 | pr_warn("map '%s': cannot be resized, map value type is not a datasec\n", | |
10111 | bpf_map__name(map)); | |
10112 | return -EINVAL; | |
10113 | } | |
10114 | ||
10115 | /* verify datasec has at least one var */ | |
10116 | vlen = btf_vlen(datasec_type); | |
10117 | if (vlen == 0) { | |
10118 | pr_warn("map '%s': cannot be resized, map value datasec is empty\n", | |
10119 | bpf_map__name(map)); | |
10120 | return -EINVAL; | |
10121 | } | |
10122 | ||
10123 | /* verify last var in the datasec is an array */ | |
10124 | var = &btf_var_secinfos(datasec_type)[vlen - 1]; | |
10125 | var_type = btf_type_by_id(btf, var->type); | |
10126 | array_type = skip_mods_and_typedefs(btf, var_type->type, NULL); | |
10127 | if (!btf_is_array(array_type)) { | |
10128 | pr_warn("map '%s': cannot be resized, last var must be an array\n", | |
10129 | bpf_map__name(map)); | |
10130 | return -EINVAL; | |
10131 | } | |
10132 | ||
10133 | /* verify request size aligns with array */ | |
10134 | array = btf_array(array_type); | |
10135 | element_sz = btf__resolve_size(btf, array->type); | |
10136 | if (element_sz <= 0 || (size - var->offset) % element_sz != 0) { | |
10137 | pr_warn("map '%s': cannot be resized, element size (%d) doesn't align with new total size (%u)\n", | |
10138 | bpf_map__name(map), element_sz, size); | |
10139 | return -EINVAL; | |
10140 | } | |
10141 | ||
10142 | /* create a new array based on the existing array, but with new length */ | |
10143 | nr_elements = (size - var->offset) / element_sz; | |
10144 | new_array_id = btf__add_array(btf, array->index_type, array->type, nr_elements); | |
10145 | if (new_array_id < 0) | |
10146 | return new_array_id; | |
10147 | ||
10148 | /* adding a new btf type invalidates existing pointers to btf objects, | |
10149 | * so refresh pointers before proceeding | |
10150 | */ | |
10151 | datasec_type = btf_type_by_id(btf, map->btf_value_type_id); | |
10152 | var = &btf_var_secinfos(datasec_type)[vlen - 1]; | |
10153 | var_type = btf_type_by_id(btf, var->type); | |
10154 | ||
10155 | /* finally update btf info */ | |
10156 | datasec_type->size = size; | |
10157 | var->size = size - var->offset; | |
10158 | var_type->type = new_array_id; | |
10159 | ||
10160 | return 0; | |
10161 | } | |
10162 | ||
1bdb6c9a AN |
10163 | int bpf_map__set_value_size(struct bpf_map *map, __u32 size) |
10164 | { | |
f08c18e0 | 10165 | if (map->obj->loaded || map->reused) |
e9fc3ce9 | 10166 | return libbpf_err(-EBUSY); |
9d0a2331 JK |
10167 | |
10168 | if (map->mmaped) { | |
9d0a2331 | 10169 | size_t mmap_old_sz, mmap_new_sz; |
79ff13e9 AS |
10170 | int err; |
10171 | ||
10172 | if (map->def.type != BPF_MAP_TYPE_ARRAY) | |
10173 | return -EOPNOTSUPP; | |
9d0a2331 | 10174 | |
79ff13e9 AS |
10175 | mmap_old_sz = bpf_map_mmap_sz(map); |
10176 | mmap_new_sz = array_map_mmap_sz(size, map->def.max_entries); | |
9d0a2331 JK |
10177 | err = bpf_map_mmap_resize(map, mmap_old_sz, mmap_new_sz); |
10178 | if (err) { | |
10179 | pr_warn("map '%s': failed to resize memory-mapped region: %d\n", | |
10180 | bpf_map__name(map), err); | |
10181 | return err; | |
10182 | } | |
10183 | err = map_btf_datasec_resize(map, size); | |
10184 | if (err && err != -ENOENT) { | |
10185 | pr_warn("map '%s': failed to adjust resized BTF, clearing BTF key/value info: %d\n", | |
10186 | bpf_map__name(map), err); | |
10187 | map->btf_value_type_id = 0; | |
10188 | map->btf_key_type_id = 0; | |
10189 | } | |
10190 | } | |
10191 | ||
1bdb6c9a AN |
10192 | map->def.value_size = size; |
10193 | return 0; | |
10194 | } | |
10195 | ||
5b891af7 | 10196 | __u32 bpf_map__btf_key_type_id(const struct bpf_map *map) |
8a138aed | 10197 | { |
61746dbe | 10198 | return map ? map->btf_key_type_id : 0; |
8a138aed MKL |
10199 | } |
10200 | ||
5b891af7 | 10201 | __u32 bpf_map__btf_value_type_id(const struct bpf_map *map) |
8a138aed | 10202 | { |
61746dbe | 10203 | return map ? map->btf_value_type_id : 0; |
8a138aed MKL |
10204 | } |
10205 | ||
e2842be5 THJ |
10206 | int bpf_map__set_initial_value(struct bpf_map *map, |
10207 | const void *data, size_t size) | |
10208 | { | |
2e7ba4f8 AN |
10209 | size_t actual_sz; |
10210 | ||
f08c18e0 AN |
10211 | if (map->obj->loaded || map->reused) |
10212 | return libbpf_err(-EBUSY); | |
10213 | ||
2e7ba4f8 AN |
10214 | if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG) |
10215 | return libbpf_err(-EINVAL); | |
10216 | ||
10217 | if (map->def.type == BPF_MAP_TYPE_ARENA) | |
10218 | actual_sz = map->obj->arena_data_sz; | |
10219 | else | |
10220 | actual_sz = map->def.value_size; | |
10221 | if (size != actual_sz) | |
e9fc3ce9 | 10222 | return libbpf_err(-EINVAL); |
e2842be5 THJ |
10223 | |
10224 | memcpy(map->mmaped, data, size); | |
10225 | return 0; | |
10226 | } | |
10227 | ||
2e7ba4f8 | 10228 | void *bpf_map__initial_value(const struct bpf_map *map, size_t *psize) |
7723256b | 10229 | { |
69e4a9d2 KFL |
10230 | if (bpf_map__is_struct_ops(map)) { |
10231 | if (psize) | |
10232 | *psize = map->def.value_size; | |
10233 | return map->st_ops->data; | |
10234 | } | |
10235 | ||
7723256b AS |
10236 | if (!map->mmaped) |
10237 | return NULL; | |
2e7ba4f8 AN |
10238 | |
10239 | if (map->def.type == BPF_MAP_TYPE_ARENA) | |
10240 | *psize = map->obj->arena_data_sz; | |
10241 | else | |
10242 | *psize = map->def.value_size; | |
10243 | ||
7723256b AS |
10244 | return map->mmaped; |
10245 | } | |
10246 | ||
a324aae3 | 10247 | bool bpf_map__is_internal(const struct bpf_map *map) |
d859900c DB |
10248 | { |
10249 | return map->libbpf_type != LIBBPF_MAP_UNSPEC; | |
10250 | } | |
10251 | ||
1bdb6c9a AN |
10252 | __u32 bpf_map__ifindex(const struct bpf_map *map) |
10253 | { | |
10254 | return map->map_ifindex; | |
10255 | } | |
10256 | ||
10257 | int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex) | |
9aba3613 | 10258 | { |
f08c18e0 | 10259 | if (map_is_created(map)) |
e9fc3ce9 | 10260 | return libbpf_err(-EBUSY); |
9aba3613 | 10261 | map->map_ifindex = ifindex; |
1bdb6c9a | 10262 | return 0; |
9aba3613 JK |
10263 | } |
10264 | ||
addb9fc9 NS |
10265 | int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd) |
10266 | { | |
10267 | if (!bpf_map_type__is_map_in_map(map->def.type)) { | |
be18010e | 10268 | pr_warn("error: unsupported map type\n"); |
e9fc3ce9 | 10269 | return libbpf_err(-EINVAL); |
addb9fc9 NS |
10270 | } |
10271 | if (map->inner_map_fd != -1) { | |
be18010e | 10272 | pr_warn("error: inner_map_fd already specified\n"); |
e9fc3ce9 | 10273 | return libbpf_err(-EINVAL); |
addb9fc9 | 10274 | } |
8f7b239e AN |
10275 | if (map->inner_map) { |
10276 | bpf_map__destroy(map->inner_map); | |
10277 | zfree(&map->inner_map); | |
10278 | } | |
addb9fc9 NS |
10279 | map->inner_map_fd = fd; |
10280 | return 0; | |
10281 | } | |
10282 | ||
0c19a9fb | 10283 | static struct bpf_map * |
a324aae3 | 10284 | __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i) |
9d759a9b | 10285 | { |
0c19a9fb | 10286 | ssize_t idx; |
9d759a9b WN |
10287 | struct bpf_map *s, *e; |
10288 | ||
10289 | if (!obj || !obj->maps) | |
e9fc3ce9 | 10290 | return errno = EINVAL, NULL; |
9d759a9b WN |
10291 | |
10292 | s = obj->maps; | |
10293 | e = obj->maps + obj->nr_maps; | |
10294 | ||
0c19a9fb | 10295 | if ((m < s) || (m >= e)) { |
be18010e KW |
10296 | pr_warn("error in %s: map handler doesn't belong to object\n", |
10297 | __func__); | |
e9fc3ce9 | 10298 | return errno = EINVAL, NULL; |
9d759a9b WN |
10299 | } |
10300 | ||
0c19a9fb SF |
10301 | idx = (m - obj->maps) + i; |
10302 | if (idx >= obj->nr_maps || idx < 0) | |
9d759a9b WN |
10303 | return NULL; |
10304 | return &obj->maps[idx]; | |
10305 | } | |
561bbcca | 10306 | |
2088a3a7 HC |
10307 | struct bpf_map * |
10308 | bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev) | |
0c19a9fb SF |
10309 | { |
10310 | if (prev == NULL) | |
10311 | return obj->maps; | |
10312 | ||
10313 | return __bpf_map__iter(prev, obj, 1); | |
10314 | } | |
10315 | ||
2088a3a7 HC |
10316 | struct bpf_map * |
10317 | bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next) | |
0c19a9fb SF |
10318 | { |
10319 | if (next == NULL) { | |
10320 | if (!obj->nr_maps) | |
10321 | return NULL; | |
10322 | return obj->maps + obj->nr_maps - 1; | |
10323 | } | |
10324 | ||
10325 | return __bpf_map__iter(next, obj, -1); | |
10326 | } | |
10327 | ||
561bbcca | 10328 | struct bpf_map * |
a324aae3 | 10329 | bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name) |
561bbcca WN |
10330 | { |
10331 | struct bpf_map *pos; | |
10332 | ||
f74a53d9 | 10333 | bpf_object__for_each_map(pos, obj) { |
26071635 AN |
10334 | /* if it's a special internal map name (which always starts |
10335 | * with dot) then check if that special name matches the | |
10336 | * real map name (ELF section name) | |
10337 | */ | |
10338 | if (name[0] == '.') { | |
10339 | if (pos->real_name && strcmp(pos->real_name, name) == 0) | |
10340 | return pos; | |
10341 | continue; | |
10342 | } | |
10343 | /* otherwise map name has to be an exact match */ | |
aed65917 AN |
10344 | if (map_uses_real_name(pos)) { |
10345 | if (strcmp(pos->real_name, name) == 0) | |
10346 | return pos; | |
10347 | continue; | |
10348 | } | |
10349 | if (strcmp(pos->name, name) == 0) | |
561bbcca WN |
10350 | return pos; |
10351 | } | |
e9fc3ce9 | 10352 | return errno = ENOENT, NULL; |
561bbcca | 10353 | } |
5a6acad1 | 10354 | |
f3cea32d | 10355 | int |
a324aae3 | 10356 | bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name) |
f3cea32d MF |
10357 | { |
10358 | return bpf_map__fd(bpf_object__find_map_by_name(obj, name)); | |
10359 | } | |
10360 | ||
737d0646 AN |
10361 | static int validate_map_op(const struct bpf_map *map, size_t key_sz, |
10362 | size_t value_sz, bool check_value_sz) | |
10363 | { | |
f08c18e0 | 10364 | if (!map_is_created(map)) /* map is not yet created */ |
737d0646 AN |
10365 | return -ENOENT; |
10366 | ||
10367 | if (map->def.key_size != key_sz) { | |
10368 | pr_warn("map '%s': unexpected key size %zu provided, expected %u\n", | |
10369 | map->name, key_sz, map->def.key_size); | |
10370 | return -EINVAL; | |
10371 | } | |
10372 | ||
7b30c296 MY |
10373 | if (map->fd < 0) { |
10374 | pr_warn("map '%s': can't use BPF map without FD (was it created?)\n", map->name); | |
10375 | return -EINVAL; | |
10376 | } | |
10377 | ||
737d0646 AN |
10378 | if (!check_value_sz) |
10379 | return 0; | |
10380 | ||
10381 | switch (map->def.type) { | |
10382 | case BPF_MAP_TYPE_PERCPU_ARRAY: | |
10383 | case BPF_MAP_TYPE_PERCPU_HASH: | |
10384 | case BPF_MAP_TYPE_LRU_PERCPU_HASH: | |
10385 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: { | |
10386 | int num_cpu = libbpf_num_possible_cpus(); | |
10387 | size_t elem_sz = roundup(map->def.value_size, 8); | |
10388 | ||
10389 | if (value_sz != num_cpu * elem_sz) { | |
10390 | pr_warn("map '%s': unexpected value size %zu provided for per-CPU map, expected %d * %zu = %zd\n", | |
10391 | map->name, value_sz, num_cpu, elem_sz, num_cpu * elem_sz); | |
10392 | return -EINVAL; | |
10393 | } | |
10394 | break; | |
10395 | } | |
10396 | default: | |
10397 | if (map->def.value_size != value_sz) { | |
10398 | pr_warn("map '%s': unexpected value size %zu provided, expected %u\n", | |
10399 | map->name, value_sz, map->def.value_size); | |
10400 | return -EINVAL; | |
10401 | } | |
10402 | break; | |
10403 | } | |
10404 | return 0; | |
10405 | } | |
10406 | ||
10407 | int bpf_map__lookup_elem(const struct bpf_map *map, | |
10408 | const void *key, size_t key_sz, | |
10409 | void *value, size_t value_sz, __u64 flags) | |
10410 | { | |
10411 | int err; | |
10412 | ||
10413 | err = validate_map_op(map, key_sz, value_sz, true); | |
10414 | if (err) | |
10415 | return libbpf_err(err); | |
10416 | ||
10417 | return bpf_map_lookup_elem_flags(map->fd, key, value, flags); | |
10418 | } | |
10419 | ||
10420 | int bpf_map__update_elem(const struct bpf_map *map, | |
10421 | const void *key, size_t key_sz, | |
10422 | const void *value, size_t value_sz, __u64 flags) | |
10423 | { | |
10424 | int err; | |
10425 | ||
10426 | err = validate_map_op(map, key_sz, value_sz, true); | |
10427 | if (err) | |
10428 | return libbpf_err(err); | |
10429 | ||
10430 | return bpf_map_update_elem(map->fd, key, value, flags); | |
10431 | } | |
10432 | ||
10433 | int bpf_map__delete_elem(const struct bpf_map *map, | |
10434 | const void *key, size_t key_sz, __u64 flags) | |
10435 | { | |
10436 | int err; | |
10437 | ||
10438 | err = validate_map_op(map, key_sz, 0, false /* check_value_sz */); | |
10439 | if (err) | |
10440 | return libbpf_err(err); | |
10441 | ||
10442 | return bpf_map_delete_elem_flags(map->fd, key, flags); | |
10443 | } | |
10444 | ||
10445 | int bpf_map__lookup_and_delete_elem(const struct bpf_map *map, | |
10446 | const void *key, size_t key_sz, | |
10447 | void *value, size_t value_sz, __u64 flags) | |
10448 | { | |
10449 | int err; | |
10450 | ||
10451 | err = validate_map_op(map, key_sz, value_sz, true); | |
10452 | if (err) | |
10453 | return libbpf_err(err); | |
10454 | ||
10455 | return bpf_map_lookup_and_delete_elem_flags(map->fd, key, value, flags); | |
10456 | } | |
10457 | ||
10458 | int bpf_map__get_next_key(const struct bpf_map *map, | |
10459 | const void *cur_key, void *next_key, size_t key_sz) | |
10460 | { | |
10461 | int err; | |
10462 | ||
10463 | err = validate_map_op(map, key_sz, 0, false /* check_value_sz */); | |
10464 | if (err) | |
10465 | return libbpf_err(err); | |
10466 | ||
10467 | return bpf_map_get_next_key(map->fd, cur_key, next_key); | |
10468 | } | |
10469 | ||
e28ff1a8 JS |
10470 | long libbpf_get_error(const void *ptr) |
10471 | { | |
e9fc3ce9 AN |
10472 | if (!IS_ERR_OR_NULL(ptr)) |
10473 | return 0; | |
10474 | ||
10475 | if (IS_ERR(ptr)) | |
10476 | errno = -PTR_ERR(ptr); | |
10477 | ||
10478 | /* If ptr == NULL, then errno should be already set by the failing | |
10479 | * API, because libbpf never returns NULL on success and it now always | |
10480 | * sets errno on error. So no extra errno handling for ptr == NULL | |
10481 | * case. | |
10482 | */ | |
10483 | return -errno; | |
e28ff1a8 | 10484 | } |
6f6d33f3 | 10485 | |
cc4f864b AN |
10486 | /* Replace link's underlying BPF program with the new one */ |
10487 | int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog) | |
10488 | { | |
e9fc3ce9 | 10489 | int ret; |
7b30c296 MY |
10490 | int prog_fd = bpf_program__fd(prog); |
10491 | ||
10492 | if (prog_fd < 0) { | |
10493 | pr_warn("prog '%s': can't use BPF program without FD (was it loaded?)\n", | |
10494 | prog->name); | |
10495 | return libbpf_err(-EINVAL); | |
10496 | } | |
c139e40a | 10497 | |
7b30c296 | 10498 | ret = bpf_link_update(bpf_link__fd(link), prog_fd, NULL); |
e9fc3ce9 | 10499 | return libbpf_err_errno(ret); |
cc4f864b AN |
10500 | } |
10501 | ||
d6958706 AN |
10502 | /* Release "ownership" of underlying BPF resource (typically, BPF program |
10503 | * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected | |
10504 | * link, when destructed through bpf_link__destroy() call won't attempt to | |
10505 | * detach/unregisted that BPF resource. This is useful in situations where, | |
10506 | * say, attached BPF program has to outlive userspace program that attached it | |
10507 | * in the system. Depending on type of BPF program, though, there might be | |
10508 | * additional steps (like pinning BPF program in BPF FS) necessary to ensure | |
10509 | * exit of userspace program doesn't trigger automatic detachment and clean up | |
10510 | * inside the kernel. | |
10511 | */ | |
10512 | void bpf_link__disconnect(struct bpf_link *link) | |
10513 | { | |
10514 | link->disconnected = true; | |
10515 | } | |
10516 | ||
1c2e9efc AN |
10517 | int bpf_link__destroy(struct bpf_link *link) |
10518 | { | |
d6958706 | 10519 | int err = 0; |
1c2e9efc | 10520 | |
50450fc7 | 10521 | if (IS_ERR_OR_NULL(link)) |
1c2e9efc AN |
10522 | return 0; |
10523 | ||
d6958706 AN |
10524 | if (!link->disconnected && link->detach) |
10525 | err = link->detach(link); | |
c016b68e AN |
10526 | if (link->pin_path) |
10527 | free(link->pin_path); | |
d88b71d4 AN |
10528 | if (link->dealloc) |
10529 | link->dealloc(link); | |
10530 | else | |
10531 | free(link); | |
1c2e9efc | 10532 | |
e9fc3ce9 | 10533 | return libbpf_err(err); |
1c2e9efc AN |
10534 | } |
10535 | ||
c016b68e AN |
10536 | int bpf_link__fd(const struct bpf_link *link) |
10537 | { | |
10538 | return link->fd; | |
10539 | } | |
10540 | ||
10541 | const char *bpf_link__pin_path(const struct bpf_link *link) | |
10542 | { | |
10543 | return link->pin_path; | |
10544 | } | |
10545 | ||
10546 | static int bpf_link__detach_fd(struct bpf_link *link) | |
10547 | { | |
e9fc3ce9 | 10548 | return libbpf_err_errno(close(link->fd)); |
c016b68e AN |
10549 | } |
10550 | ||
10551 | struct bpf_link *bpf_link__open(const char *path) | |
10552 | { | |
10553 | struct bpf_link *link; | |
10554 | int fd; | |
10555 | ||
10556 | fd = bpf_obj_get(path); | |
10557 | if (fd < 0) { | |
10558 | fd = -errno; | |
10559 | pr_warn("failed to open link at %s: %d\n", path, fd); | |
e9fc3ce9 | 10560 | return libbpf_err_ptr(fd); |
c016b68e AN |
10561 | } |
10562 | ||
10563 | link = calloc(1, sizeof(*link)); | |
10564 | if (!link) { | |
10565 | close(fd); | |
e9fc3ce9 | 10566 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
10567 | } |
10568 | link->detach = &bpf_link__detach_fd; | |
10569 | link->fd = fd; | |
10570 | ||
10571 | link->pin_path = strdup(path); | |
10572 | if (!link->pin_path) { | |
10573 | bpf_link__destroy(link); | |
e9fc3ce9 | 10574 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
10575 | } |
10576 | ||
10577 | return link; | |
10578 | } | |
10579 | ||
2e49527e AN |
10580 | int bpf_link__detach(struct bpf_link *link) |
10581 | { | |
10582 | return bpf_link_detach(link->fd) ? -errno : 0; | |
10583 | } | |
10584 | ||
c016b68e AN |
10585 | int bpf_link__pin(struct bpf_link *link, const char *path) |
10586 | { | |
10587 | int err; | |
10588 | ||
10589 | if (link->pin_path) | |
e9fc3ce9 | 10590 | return libbpf_err(-EBUSY); |
c016b68e AN |
10591 | err = make_parent_dir(path); |
10592 | if (err) | |
e9fc3ce9 | 10593 | return libbpf_err(err); |
c016b68e AN |
10594 | err = check_path(path); |
10595 | if (err) | |
e9fc3ce9 | 10596 | return libbpf_err(err); |
c016b68e AN |
10597 | |
10598 | link->pin_path = strdup(path); | |
10599 | if (!link->pin_path) | |
e9fc3ce9 | 10600 | return libbpf_err(-ENOMEM); |
c016b68e AN |
10601 | |
10602 | if (bpf_obj_pin(link->fd, link->pin_path)) { | |
10603 | err = -errno; | |
10604 | zfree(&link->pin_path); | |
e9fc3ce9 | 10605 | return libbpf_err(err); |
c016b68e AN |
10606 | } |
10607 | ||
10608 | pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path); | |
10609 | return 0; | |
10610 | } | |
10611 | ||
10612 | int bpf_link__unpin(struct bpf_link *link) | |
10613 | { | |
10614 | int err; | |
10615 | ||
10616 | if (!link->pin_path) | |
e9fc3ce9 | 10617 | return libbpf_err(-EINVAL); |
c016b68e AN |
10618 | |
10619 | err = unlink(link->pin_path); | |
10620 | if (err != 0) | |
af0efa05 | 10621 | return -errno; |
c016b68e AN |
10622 | |
10623 | pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path); | |
10624 | zfree(&link->pin_path); | |
10625 | return 0; | |
10626 | } | |
63f2f5ee | 10627 | |
668ace0e AN |
10628 | struct bpf_link_perf { |
10629 | struct bpf_link link; | |
10630 | int perf_event_fd; | |
ca304b40 RDT |
10631 | /* legacy kprobe support: keep track of probe identifier and type */ |
10632 | char *legacy_probe_name; | |
46ed5fc3 | 10633 | bool legacy_is_kprobe; |
ca304b40 | 10634 | bool legacy_is_retprobe; |
668ace0e AN |
10635 | }; |
10636 | ||
46ed5fc3 | 10637 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe); |
cc10623c | 10638 | static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe); |
46ed5fc3 | 10639 | |
668ace0e | 10640 | static int bpf_link_perf_detach(struct bpf_link *link) |
63f2f5ee | 10641 | { |
668ace0e AN |
10642 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); |
10643 | int err = 0; | |
63f2f5ee | 10644 | |
668ace0e | 10645 | if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0) |
63f2f5ee AN |
10646 | err = -errno; |
10647 | ||
668ace0e AN |
10648 | if (perf_link->perf_event_fd != link->fd) |
10649 | close(perf_link->perf_event_fd); | |
c016b68e | 10650 | close(link->fd); |
668ace0e | 10651 | |
cc10623c | 10652 | /* legacy uprobe/kprobe needs to be removed after perf event fd closure */ |
46ed5fc3 AN |
10653 | if (perf_link->legacy_probe_name) { |
10654 | if (perf_link->legacy_is_kprobe) { | |
10655 | err = remove_kprobe_event_legacy(perf_link->legacy_probe_name, | |
10656 | perf_link->legacy_is_retprobe); | |
cc10623c AN |
10657 | } else { |
10658 | err = remove_uprobe_event_legacy(perf_link->legacy_probe_name, | |
10659 | perf_link->legacy_is_retprobe); | |
46ed5fc3 AN |
10660 | } |
10661 | } | |
ca304b40 RDT |
10662 | |
10663 | return err; | |
63f2f5ee AN |
10664 | } |
10665 | ||
668ace0e AN |
10666 | static void bpf_link_perf_dealloc(struct bpf_link *link) |
10667 | { | |
10668 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
10669 | ||
ca304b40 | 10670 | free(perf_link->legacy_probe_name); |
668ace0e AN |
10671 | free(perf_link); |
10672 | } | |
10673 | ||
942025c9 | 10674 | struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd, |
47faff37 | 10675 | const struct bpf_perf_event_opts *opts) |
63f2f5ee AN |
10676 | { |
10677 | char errmsg[STRERR_BUFSIZE]; | |
668ace0e AN |
10678 | struct bpf_link_perf *link; |
10679 | int prog_fd, link_fd = -1, err; | |
f8b299bc | 10680 | bool force_ioctl_attach; |
63f2f5ee | 10681 | |
47faff37 AN |
10682 | if (!OPTS_VALID(opts, bpf_perf_event_opts)) |
10683 | return libbpf_err_ptr(-EINVAL); | |
10684 | ||
63f2f5ee | 10685 | if (pfd < 0) { |
52109584 AN |
10686 | pr_warn("prog '%s': invalid perf event FD %d\n", |
10687 | prog->name, pfd); | |
e9fc3ce9 | 10688 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
10689 | } |
10690 | prog_fd = bpf_program__fd(prog); | |
10691 | if (prog_fd < 0) { | |
7b30c296 | 10692 | pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n", |
52109584 | 10693 | prog->name); |
e9fc3ce9 | 10694 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
10695 | } |
10696 | ||
d6958706 | 10697 | link = calloc(1, sizeof(*link)); |
63f2f5ee | 10698 | if (!link) |
e9fc3ce9 | 10699 | return libbpf_err_ptr(-ENOMEM); |
668ace0e AN |
10700 | link->link.detach = &bpf_link_perf_detach; |
10701 | link->link.dealloc = &bpf_link_perf_dealloc; | |
10702 | link->perf_event_fd = pfd; | |
63f2f5ee | 10703 | |
f8b299bc MD |
10704 | force_ioctl_attach = OPTS_GET(opts, force_ioctl_attach, false); |
10705 | if (kernel_supports(prog->obj, FEAT_PERF_LINK) && !force_ioctl_attach) { | |
47faff37 AN |
10706 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts, |
10707 | .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0)); | |
10708 | ||
10709 | link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts); | |
668ace0e AN |
10710 | if (link_fd < 0) { |
10711 | err = -errno; | |
10712 | pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n", | |
10713 | prog->name, pfd, | |
10714 | err, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10715 | goto err_out; | |
10716 | } | |
10717 | link->link.fd = link_fd; | |
10718 | } else { | |
47faff37 AN |
10719 | if (OPTS_GET(opts, bpf_cookie, 0)) { |
10720 | pr_warn("prog '%s': user context value is not supported\n", prog->name); | |
10721 | err = -EOPNOTSUPP; | |
10722 | goto err_out; | |
10723 | } | |
10724 | ||
668ace0e AN |
10725 | if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) { |
10726 | err = -errno; | |
10727 | pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n", | |
10728 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10729 | if (err == -EPROTO) | |
10730 | pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n", | |
10731 | prog->name, pfd); | |
10732 | goto err_out; | |
10733 | } | |
10734 | link->link.fd = pfd; | |
63f2f5ee AN |
10735 | } |
10736 | if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
10737 | err = -errno; | |
668ace0e | 10738 | pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n", |
52109584 | 10739 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
668ace0e | 10740 | goto err_out; |
63f2f5ee | 10741 | } |
668ace0e AN |
10742 | |
10743 | return &link->link; | |
10744 | err_out: | |
10745 | if (link_fd >= 0) | |
10746 | close(link_fd); | |
10747 | free(link); | |
10748 | return libbpf_err_ptr(err); | |
63f2f5ee AN |
10749 | } |
10750 | ||
942025c9 | 10751 | struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd) |
47faff37 AN |
10752 | { |
10753 | return bpf_program__attach_perf_event_opts(prog, pfd, NULL); | |
10754 | } | |
10755 | ||
b2650027 AN |
10756 | /* |
10757 | * this function is expected to parse integer in the range of [0, 2^31-1] from | |
10758 | * given file using scanf format string fmt. If actual parsed value is | |
10759 | * negative, the result might be indistinguishable from error | |
10760 | */ | |
10761 | static int parse_uint_from_file(const char *file, const char *fmt) | |
10762 | { | |
10763 | char buf[STRERR_BUFSIZE]; | |
10764 | int err, ret; | |
10765 | FILE *f; | |
10766 | ||
59842c54 | 10767 | f = fopen(file, "re"); |
b2650027 AN |
10768 | if (!f) { |
10769 | err = -errno; | |
10770 | pr_debug("failed to open '%s': %s\n", file, | |
10771 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
10772 | return err; | |
10773 | } | |
10774 | err = fscanf(f, fmt, &ret); | |
10775 | if (err != 1) { | |
10776 | err = err == EOF ? -EIO : -errno; | |
10777 | pr_debug("failed to parse '%s': %s\n", file, | |
10778 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
10779 | fclose(f); | |
10780 | return err; | |
10781 | } | |
10782 | fclose(f); | |
10783 | return ret; | |
10784 | } | |
10785 | ||
10786 | static int determine_kprobe_perf_type(void) | |
10787 | { | |
10788 | const char *file = "/sys/bus/event_source/devices/kprobe/type"; | |
10789 | ||
10790 | return parse_uint_from_file(file, "%d\n"); | |
10791 | } | |
10792 | ||
10793 | static int determine_uprobe_perf_type(void) | |
10794 | { | |
10795 | const char *file = "/sys/bus/event_source/devices/uprobe/type"; | |
10796 | ||
10797 | return parse_uint_from_file(file, "%d\n"); | |
10798 | } | |
10799 | ||
10800 | static int determine_kprobe_retprobe_bit(void) | |
10801 | { | |
10802 | const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe"; | |
10803 | ||
10804 | return parse_uint_from_file(file, "config:%d\n"); | |
10805 | } | |
10806 | ||
10807 | static int determine_uprobe_retprobe_bit(void) | |
10808 | { | |
10809 | const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe"; | |
10810 | ||
10811 | return parse_uint_from_file(file, "config:%d\n"); | |
10812 | } | |
10813 | ||
5e3b8356 AN |
10814 | #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32 |
10815 | #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32 | |
10816 | ||
b2650027 | 10817 | static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name, |
5e3b8356 | 10818 | uint64_t offset, int pid, size_t ref_ctr_off) |
b2650027 | 10819 | { |
813847a3 AN |
10820 | const size_t attr_sz = sizeof(struct perf_event_attr); |
10821 | struct perf_event_attr attr; | |
b2650027 | 10822 | char errmsg[STRERR_BUFSIZE]; |
708ac5be | 10823 | int type, pfd; |
b2650027 | 10824 | |
1520e846 | 10825 | if ((__u64)ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS)) |
5e3b8356 AN |
10826 | return -EINVAL; |
10827 | ||
813847a3 AN |
10828 | memset(&attr, 0, attr_sz); |
10829 | ||
b2650027 AN |
10830 | type = uprobe ? determine_uprobe_perf_type() |
10831 | : determine_kprobe_perf_type(); | |
10832 | if (type < 0) { | |
be18010e KW |
10833 | pr_warn("failed to determine %s perf type: %s\n", |
10834 | uprobe ? "uprobe" : "kprobe", | |
10835 | libbpf_strerror_r(type, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10836 | return type; |
10837 | } | |
10838 | if (retprobe) { | |
10839 | int bit = uprobe ? determine_uprobe_retprobe_bit() | |
10840 | : determine_kprobe_retprobe_bit(); | |
10841 | ||
10842 | if (bit < 0) { | |
be18010e KW |
10843 | pr_warn("failed to determine %s retprobe bit: %s\n", |
10844 | uprobe ? "uprobe" : "kprobe", | |
10845 | libbpf_strerror_r(bit, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10846 | return bit; |
10847 | } | |
10848 | attr.config |= 1 << bit; | |
10849 | } | |
813847a3 | 10850 | attr.size = attr_sz; |
b2650027 | 10851 | attr.type = type; |
5e3b8356 | 10852 | attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT; |
36db2a94 AN |
10853 | attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */ |
10854 | attr.config2 = offset; /* kprobe_addr or probe_offset */ | |
b2650027 AN |
10855 | |
10856 | /* pid filter is meaningful only for uprobes */ | |
10857 | pfd = syscall(__NR_perf_event_open, &attr, | |
10858 | pid < 0 ? -1 : pid /* pid */, | |
10859 | pid == -1 ? 0 : -1 /* cpu */, | |
10860 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
708ac5be | 10861 | return pfd >= 0 ? pfd : -errno; |
b2650027 AN |
10862 | } |
10863 | ||
46ed5fc3 AN |
10864 | static int append_to_file(const char *file, const char *fmt, ...) |
10865 | { | |
10866 | int fd, n, err = 0; | |
10867 | va_list ap; | |
01dc26c9 LP |
10868 | char buf[1024]; |
10869 | ||
10870 | va_start(ap, fmt); | |
10871 | n = vsnprintf(buf, sizeof(buf), fmt, ap); | |
10872 | va_end(ap); | |
10873 | ||
10874 | if (n < 0 || n >= sizeof(buf)) | |
10875 | return -EINVAL; | |
46ed5fc3 | 10876 | |
92274e24 | 10877 | fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0); |
46ed5fc3 AN |
10878 | if (fd < 0) |
10879 | return -errno; | |
10880 | ||
01dc26c9 | 10881 | if (write(fd, buf, n) < 0) |
46ed5fc3 AN |
10882 | err = -errno; |
10883 | ||
10884 | close(fd); | |
10885 | return err; | |
10886 | } | |
10887 | ||
a1ac9fd6 AN |
10888 | #define DEBUGFS "/sys/kernel/debug/tracing" |
10889 | #define TRACEFS "/sys/kernel/tracing" | |
10890 | ||
10891 | static bool use_debugfs(void) | |
10892 | { | |
10893 | static int has_debugfs = -1; | |
10894 | ||
10895 | if (has_debugfs < 0) | |
6a4ab886 | 10896 | has_debugfs = faccessat(AT_FDCWD, DEBUGFS, F_OK, AT_EACCESS) == 0; |
a1ac9fd6 AN |
10897 | |
10898 | return has_debugfs == 1; | |
10899 | } | |
10900 | ||
10901 | static const char *tracefs_path(void) | |
10902 | { | |
10903 | return use_debugfs() ? DEBUGFS : TRACEFS; | |
10904 | } | |
10905 | ||
10906 | static const char *tracefs_kprobe_events(void) | |
10907 | { | |
10908 | return use_debugfs() ? DEBUGFS"/kprobe_events" : TRACEFS"/kprobe_events"; | |
10909 | } | |
10910 | ||
10911 | static const char *tracefs_uprobe_events(void) | |
10912 | { | |
10913 | return use_debugfs() ? DEBUGFS"/uprobe_events" : TRACEFS"/uprobe_events"; | |
10914 | } | |
10915 | ||
8a3fe76f JL |
10916 | static const char *tracefs_available_filter_functions(void) |
10917 | { | |
10918 | return use_debugfs() ? DEBUGFS"/available_filter_functions" | |
10919 | : TRACEFS"/available_filter_functions"; | |
10920 | } | |
10921 | ||
56baeeba JL |
10922 | static const char *tracefs_available_filter_functions_addrs(void) |
10923 | { | |
10924 | return use_debugfs() ? DEBUGFS"/available_filter_functions_addrs" | |
10925 | : TRACEFS"/available_filter_functions_addrs"; | |
10926 | } | |
10927 | ||
46ed5fc3 AN |
10928 | static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz, |
10929 | const char *kfunc_name, size_t offset) | |
10930 | { | |
51a33c60 | 10931 | static int index = 0; |
2fa07453 | 10932 | int i; |
51a33c60 QW |
10933 | |
10934 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset, | |
10935 | __sync_fetch_and_add(&index, 1)); | |
2fa07453 MD |
10936 | |
10937 | /* sanitize binary_path in the probe name */ | |
10938 | for (i = 0; buf[i]; i++) { | |
10939 | if (!isalnum(buf[i])) | |
10940 | buf[i] = '_'; | |
10941 | } | |
46ed5fc3 AN |
10942 | } |
10943 | ||
10944 | static int add_kprobe_event_legacy(const char *probe_name, bool retprobe, | |
10945 | const char *kfunc_name, size_t offset) | |
10946 | { | |
a1ac9fd6 | 10947 | return append_to_file(tracefs_kprobe_events(), "%c:%s/%s %s+0x%zx", |
46ed5fc3 AN |
10948 | retprobe ? 'r' : 'p', |
10949 | retprobe ? "kretprobes" : "kprobes", | |
10950 | probe_name, kfunc_name, offset); | |
10951 | } | |
10952 | ||
10953 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe) | |
10954 | { | |
a1ac9fd6 AN |
10955 | return append_to_file(tracefs_kprobe_events(), "-:%s/%s", |
10956 | retprobe ? "kretprobes" : "kprobes", probe_name); | |
46ed5fc3 AN |
10957 | } |
10958 | ||
10959 | static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
10960 | { | |
10961 | char file[256]; | |
10962 | ||
a1ac9fd6 AN |
10963 | snprintf(file, sizeof(file), "%s/events/%s/%s/id", |
10964 | tracefs_path(), retprobe ? "kretprobes" : "kprobes", probe_name); | |
46ed5fc3 AN |
10965 | |
10966 | return parse_uint_from_file(file, "%d\n"); | |
10967 | } | |
10968 | ||
10969 | static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe, | |
10970 | const char *kfunc_name, size_t offset, int pid) | |
ca304b40 | 10971 | { |
813847a3 AN |
10972 | const size_t attr_sz = sizeof(struct perf_event_attr); |
10973 | struct perf_event_attr attr; | |
ca304b40 RDT |
10974 | char errmsg[STRERR_BUFSIZE]; |
10975 | int type, pfd, err; | |
10976 | ||
46ed5fc3 | 10977 | err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset); |
ca304b40 | 10978 | if (err < 0) { |
46ed5fc3 AN |
10979 | pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n", |
10980 | kfunc_name, offset, | |
ca304b40 RDT |
10981 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10982 | return err; | |
10983 | } | |
46ed5fc3 | 10984 | type = determine_kprobe_perf_type_legacy(probe_name, retprobe); |
ca304b40 | 10985 | if (type < 0) { |
80940293 | 10986 | err = type; |
46ed5fc3 AN |
10987 | pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n", |
10988 | kfunc_name, offset, | |
80940293 CW |
10989 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10990 | goto err_clean_legacy; | |
ca304b40 | 10991 | } |
813847a3 AN |
10992 | |
10993 | memset(&attr, 0, attr_sz); | |
10994 | attr.size = attr_sz; | |
ca304b40 RDT |
10995 | attr.config = type; |
10996 | attr.type = PERF_TYPE_TRACEPOINT; | |
10997 | ||
10998 | pfd = syscall(__NR_perf_event_open, &attr, | |
10999 | pid < 0 ? -1 : pid, /* pid */ | |
11000 | pid == -1 ? 0 : -1, /* cpu */ | |
11001 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
11002 | if (pfd < 0) { | |
11003 | err = -errno; | |
11004 | pr_warn("legacy kprobe perf_event_open() failed: %s\n", | |
11005 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
80940293 | 11006 | goto err_clean_legacy; |
ca304b40 RDT |
11007 | } |
11008 | return pfd; | |
80940293 CW |
11009 | |
11010 | err_clean_legacy: | |
11011 | /* Clear the newly added legacy kprobe_event */ | |
11012 | remove_kprobe_event_legacy(probe_name, retprobe); | |
11013 | return err; | |
ca304b40 RDT |
11014 | } |
11015 | ||
708ac5be AN |
11016 | static const char *arch_specific_syscall_pfx(void) |
11017 | { | |
11018 | #if defined(__x86_64__) | |
11019 | return "x64"; | |
11020 | #elif defined(__i386__) | |
11021 | return "ia32"; | |
11022 | #elif defined(__s390x__) | |
11023 | return "s390x"; | |
11024 | #elif defined(__s390__) | |
11025 | return "s390"; | |
11026 | #elif defined(__arm__) | |
11027 | return "arm"; | |
11028 | #elif defined(__aarch64__) | |
11029 | return "arm64"; | |
11030 | #elif defined(__mips__) | |
11031 | return "mips"; | |
11032 | #elif defined(__riscv) | |
11033 | return "riscv"; | |
64893e83 DM |
11034 | #elif defined(__powerpc__) |
11035 | return "powerpc"; | |
11036 | #elif defined(__powerpc64__) | |
11037 | return "powerpc64"; | |
708ac5be AN |
11038 | #else |
11039 | return NULL; | |
11040 | #endif | |
11041 | } | |
11042 | ||
f3dcee93 | 11043 | int probe_kern_syscall_wrapper(int token_fd) |
708ac5be AN |
11044 | { |
11045 | char syscall_name[64]; | |
11046 | const char *ksys_pfx; | |
11047 | ||
11048 | ksys_pfx = arch_specific_syscall_pfx(); | |
11049 | if (!ksys_pfx) | |
11050 | return 0; | |
11051 | ||
11052 | snprintf(syscall_name, sizeof(syscall_name), "__%s_sys_bpf", ksys_pfx); | |
11053 | ||
11054 | if (determine_kprobe_perf_type() >= 0) { | |
11055 | int pfd; | |
11056 | ||
11057 | pfd = perf_event_open_probe(false, false, syscall_name, 0, getpid(), 0); | |
11058 | if (pfd >= 0) | |
11059 | close(pfd); | |
11060 | ||
11061 | return pfd >= 0 ? 1 : 0; | |
11062 | } else { /* legacy mode */ | |
11063 | char probe_name[128]; | |
11064 | ||
11065 | gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), syscall_name, 0); | |
11066 | if (add_kprobe_event_legacy(probe_name, false, syscall_name, 0) < 0) | |
11067 | return 0; | |
11068 | ||
11069 | (void)remove_kprobe_event_legacy(probe_name, false); | |
11070 | return 1; | |
11071 | } | |
11072 | } | |
11073 | ||
da97553e | 11074 | struct bpf_link * |
942025c9 | 11075 | bpf_program__attach_kprobe_opts(const struct bpf_program *prog, |
ac0ed488 | 11076 | const char *func_name, |
47faff37 | 11077 | const struct bpf_kprobe_opts *opts) |
b2650027 | 11078 | { |
47faff37 | 11079 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f8b299bc | 11080 | enum probe_attach_mode attach_mode; |
b2650027 | 11081 | char errmsg[STRERR_BUFSIZE]; |
ca304b40 | 11082 | char *legacy_probe = NULL; |
b2650027 | 11083 | struct bpf_link *link; |
46ed5fc3 | 11084 | size_t offset; |
ca304b40 | 11085 | bool retprobe, legacy; |
b2650027 AN |
11086 | int pfd, err; |
11087 | ||
da97553e JO |
11088 | if (!OPTS_VALID(opts, bpf_kprobe_opts)) |
11089 | return libbpf_err_ptr(-EINVAL); | |
11090 | ||
f8b299bc | 11091 | attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT); |
da97553e JO |
11092 | retprobe = OPTS_GET(opts, retprobe, false); |
11093 | offset = OPTS_GET(opts, offset, 0); | |
47faff37 | 11094 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
da97553e | 11095 | |
ca304b40 | 11096 | legacy = determine_kprobe_perf_type() < 0; |
f8b299bc MD |
11097 | switch (attach_mode) { |
11098 | case PROBE_ATTACH_MODE_LEGACY: | |
11099 | legacy = true; | |
11100 | pe_opts.force_ioctl_attach = true; | |
11101 | break; | |
11102 | case PROBE_ATTACH_MODE_PERF: | |
11103 | if (legacy) | |
11104 | return libbpf_err_ptr(-ENOTSUP); | |
11105 | pe_opts.force_ioctl_attach = true; | |
11106 | break; | |
11107 | case PROBE_ATTACH_MODE_LINK: | |
11108 | if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK)) | |
11109 | return libbpf_err_ptr(-ENOTSUP); | |
11110 | break; | |
11111 | case PROBE_ATTACH_MODE_DEFAULT: | |
11112 | break; | |
11113 | default: | |
11114 | return libbpf_err_ptr(-EINVAL); | |
11115 | } | |
11116 | ||
ca304b40 RDT |
11117 | if (!legacy) { |
11118 | pfd = perf_event_open_probe(false /* uprobe */, retprobe, | |
11119 | func_name, offset, | |
11120 | -1 /* pid */, 0 /* ref_ctr_off */); | |
11121 | } else { | |
46ed5fc3 AN |
11122 | char probe_name[256]; |
11123 | ||
11124 | gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
11125 | func_name, offset); | |
11126 | ||
71cff670 | 11127 | legacy_probe = strdup(probe_name); |
ca304b40 RDT |
11128 | if (!legacy_probe) |
11129 | return libbpf_err_ptr(-ENOMEM); | |
11130 | ||
46ed5fc3 | 11131 | pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name, |
ca304b40 RDT |
11132 | offset, -1 /* pid */); |
11133 | } | |
b2650027 | 11134 | if (pfd < 0) { |
46ed5fc3 AN |
11135 | err = -errno; |
11136 | pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n", | |
11137 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
11138 | func_name, offset, | |
303a2572 AN |
11139 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
11140 | goto err_out; | |
b2650027 | 11141 | } |
47faff37 | 11142 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
11143 | err = libbpf_get_error(link); |
11144 | if (err) { | |
b2650027 | 11145 | close(pfd); |
46ed5fc3 AN |
11146 | pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n", |
11147 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
11148 | func_name, offset, | |
be18010e | 11149 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
80940293 | 11150 | goto err_clean_legacy; |
b2650027 | 11151 | } |
ca304b40 RDT |
11152 | if (legacy) { |
11153 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
11154 | ||
11155 | perf_link->legacy_probe_name = legacy_probe; | |
46ed5fc3 | 11156 | perf_link->legacy_is_kprobe = true; |
ca304b40 RDT |
11157 | perf_link->legacy_is_retprobe = retprobe; |
11158 | } | |
11159 | ||
b2650027 | 11160 | return link; |
80940293 CW |
11161 | |
11162 | err_clean_legacy: | |
11163 | if (legacy) | |
11164 | remove_kprobe_event_legacy(legacy_probe, retprobe); | |
303a2572 AN |
11165 | err_out: |
11166 | free(legacy_probe); | |
11167 | return libbpf_err_ptr(err); | |
b2650027 AN |
11168 | } |
11169 | ||
942025c9 | 11170 | struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog, |
ac0ed488 JO |
11171 | bool retprobe, |
11172 | const char *func_name) | |
11173 | { | |
da97553e | 11174 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts, |
ac0ed488 | 11175 | .retprobe = retprobe, |
da97553e | 11176 | ); |
ac0ed488 JO |
11177 | |
11178 | return bpf_program__attach_kprobe_opts(prog, func_name, &opts); | |
11179 | } | |
11180 | ||
708ac5be AN |
11181 | struct bpf_link *bpf_program__attach_ksyscall(const struct bpf_program *prog, |
11182 | const char *syscall_name, | |
11183 | const struct bpf_ksyscall_opts *opts) | |
11184 | { | |
11185 | LIBBPF_OPTS(bpf_kprobe_opts, kprobe_opts); | |
11186 | char func_name[128]; | |
11187 | ||
11188 | if (!OPTS_VALID(opts, bpf_ksyscall_opts)) | |
11189 | return libbpf_err_ptr(-EINVAL); | |
11190 | ||
11191 | if (kernel_supports(prog->obj, FEAT_SYSCALL_WRAPPER)) { | |
64893e83 DM |
11192 | /* arch_specific_syscall_pfx() should never return NULL here |
11193 | * because it is guarded by kernel_supports(). However, since | |
11194 | * compiler does not know that we have an explicit conditional | |
11195 | * as well. | |
11196 | */ | |
708ac5be | 11197 | snprintf(func_name, sizeof(func_name), "__%s_sys_%s", |
64893e83 | 11198 | arch_specific_syscall_pfx() ? : "", syscall_name); |
708ac5be AN |
11199 | } else { |
11200 | snprintf(func_name, sizeof(func_name), "__se_sys_%s", syscall_name); | |
11201 | } | |
11202 | ||
11203 | kprobe_opts.retprobe = OPTS_GET(opts, retprobe, false); | |
11204 | kprobe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); | |
11205 | ||
11206 | return bpf_program__attach_kprobe_opts(prog, func_name, &kprobe_opts); | |
11207 | } | |
11208 | ||
ddc6b049 | 11209 | /* Adapted from perf/util/string.c */ |
e613d1d0 | 11210 | bool glob_match(const char *str, const char *pat) |
ddc6b049 JO |
11211 | { |
11212 | while (*str && *pat && *pat != '*') { | |
11213 | if (*pat == '?') { /* Matches any single character */ | |
11214 | str++; | |
11215 | pat++; | |
11216 | continue; | |
11217 | } | |
11218 | if (*str != *pat) | |
11219 | return false; | |
11220 | str++; | |
11221 | pat++; | |
11222 | } | |
11223 | /* Check wild card */ | |
11224 | if (*pat == '*') { | |
11225 | while (*pat == '*') | |
11226 | pat++; | |
11227 | if (!*pat) /* Tail wild card matches all */ | |
11228 | return true; | |
11229 | while (*str) | |
11230 | if (glob_match(str++, pat)) | |
11231 | return true; | |
11232 | } | |
11233 | return !*str && !*pat; | |
11234 | } | |
11235 | ||
11236 | struct kprobe_multi_resolve { | |
11237 | const char *pattern; | |
11238 | unsigned long *addrs; | |
11239 | size_t cap; | |
11240 | size_t cnt; | |
11241 | }; | |
11242 | ||
8a3fe76f JL |
11243 | struct avail_kallsyms_data { |
11244 | char **syms; | |
11245 | size_t cnt; | |
11246 | struct kprobe_multi_resolve *res; | |
11247 | }; | |
11248 | ||
11249 | static int avail_func_cmp(const void *a, const void *b) | |
11250 | { | |
11251 | return strcmp(*(const char **)a, *(const char **)b); | |
11252 | } | |
11253 | ||
11254 | static int avail_kallsyms_cb(unsigned long long sym_addr, char sym_type, | |
11255 | const char *sym_name, void *ctx) | |
ddc6b049 | 11256 | { |
8a3fe76f JL |
11257 | struct avail_kallsyms_data *data = ctx; |
11258 | struct kprobe_multi_resolve *res = data->res; | |
ddc6b049 JO |
11259 | int err; |
11260 | ||
8a3fe76f | 11261 | if (!bsearch(&sym_name, data->syms, data->cnt, sizeof(*data->syms), avail_func_cmp)) |
ddc6b049 JO |
11262 | return 0; |
11263 | ||
8a3fe76f | 11264 | err = libbpf_ensure_mem((void **)&res->addrs, &res->cap, sizeof(*res->addrs), res->cnt + 1); |
ddc6b049 JO |
11265 | if (err) |
11266 | return err; | |
11267 | ||
8a3fe76f | 11268 | res->addrs[res->cnt++] = (unsigned long)sym_addr; |
ddc6b049 JO |
11269 | return 0; |
11270 | } | |
11271 | ||
8a3fe76f JL |
11272 | static int libbpf_available_kallsyms_parse(struct kprobe_multi_resolve *res) |
11273 | { | |
11274 | const char *available_functions_file = tracefs_available_filter_functions(); | |
11275 | struct avail_kallsyms_data data; | |
11276 | char sym_name[500]; | |
11277 | FILE *f; | |
11278 | int err = 0, ret, i; | |
11279 | char **syms = NULL; | |
11280 | size_t cap = 0, cnt = 0; | |
11281 | ||
11282 | f = fopen(available_functions_file, "re"); | |
11283 | if (!f) { | |
11284 | err = -errno; | |
11285 | pr_warn("failed to open %s: %d\n", available_functions_file, err); | |
11286 | return err; | |
11287 | } | |
11288 | ||
11289 | while (true) { | |
11290 | char *name; | |
11291 | ||
11292 | ret = fscanf(f, "%499s%*[^\n]\n", sym_name); | |
11293 | if (ret == EOF && feof(f)) | |
11294 | break; | |
11295 | ||
11296 | if (ret != 1) { | |
11297 | pr_warn("failed to parse available_filter_functions entry: %d\n", ret); | |
11298 | err = -EINVAL; | |
11299 | goto cleanup; | |
11300 | } | |
11301 | ||
11302 | if (!glob_match(sym_name, res->pattern)) | |
11303 | continue; | |
11304 | ||
11305 | err = libbpf_ensure_mem((void **)&syms, &cap, sizeof(*syms), cnt + 1); | |
11306 | if (err) | |
11307 | goto cleanup; | |
11308 | ||
11309 | name = strdup(sym_name); | |
11310 | if (!name) { | |
11311 | err = -errno; | |
11312 | goto cleanup; | |
11313 | } | |
11314 | ||
11315 | syms[cnt++] = name; | |
11316 | } | |
11317 | ||
11318 | /* no entries found, bail out */ | |
11319 | if (cnt == 0) { | |
11320 | err = -ENOENT; | |
11321 | goto cleanup; | |
11322 | } | |
11323 | ||
11324 | /* sort available functions */ | |
11325 | qsort(syms, cnt, sizeof(*syms), avail_func_cmp); | |
11326 | ||
11327 | data.syms = syms; | |
11328 | data.res = res; | |
11329 | data.cnt = cnt; | |
11330 | libbpf_kallsyms_parse(avail_kallsyms_cb, &data); | |
11331 | ||
11332 | if (res->cnt == 0) | |
11333 | err = -ENOENT; | |
11334 | ||
11335 | cleanup: | |
11336 | for (i = 0; i < cnt; i++) | |
11337 | free((char *)syms[i]); | |
11338 | free(syms); | |
11339 | ||
11340 | fclose(f); | |
11341 | return err; | |
11342 | } | |
11343 | ||
56baeeba JL |
11344 | static bool has_available_filter_functions_addrs(void) |
11345 | { | |
11346 | return access(tracefs_available_filter_functions_addrs(), R_OK) != -1; | |
11347 | } | |
11348 | ||
11349 | static int libbpf_available_kprobes_parse(struct kprobe_multi_resolve *res) | |
11350 | { | |
11351 | const char *available_path = tracefs_available_filter_functions_addrs(); | |
11352 | char sym_name[500]; | |
11353 | FILE *f; | |
11354 | int ret, err = 0; | |
11355 | unsigned long long sym_addr; | |
11356 | ||
11357 | f = fopen(available_path, "re"); | |
11358 | if (!f) { | |
11359 | err = -errno; | |
11360 | pr_warn("failed to open %s: %d\n", available_path, err); | |
11361 | return err; | |
11362 | } | |
11363 | ||
11364 | while (true) { | |
11365 | ret = fscanf(f, "%llx %499s%*[^\n]\n", &sym_addr, sym_name); | |
11366 | if (ret == EOF && feof(f)) | |
11367 | break; | |
11368 | ||
11369 | if (ret != 2) { | |
11370 | pr_warn("failed to parse available_filter_functions_addrs entry: %d\n", | |
11371 | ret); | |
11372 | err = -EINVAL; | |
11373 | goto cleanup; | |
11374 | } | |
11375 | ||
11376 | if (!glob_match(sym_name, res->pattern)) | |
11377 | continue; | |
11378 | ||
11379 | err = libbpf_ensure_mem((void **)&res->addrs, &res->cap, | |
11380 | sizeof(*res->addrs), res->cnt + 1); | |
11381 | if (err) | |
11382 | goto cleanup; | |
11383 | ||
11384 | res->addrs[res->cnt++] = (unsigned long)sym_addr; | |
11385 | } | |
11386 | ||
11387 | if (res->cnt == 0) | |
11388 | err = -ENOENT; | |
11389 | ||
11390 | cleanup: | |
11391 | fclose(f); | |
11392 | return err; | |
11393 | } | |
11394 | ||
ddc6b049 JO |
11395 | struct bpf_link * |
11396 | bpf_program__attach_kprobe_multi_opts(const struct bpf_program *prog, | |
11397 | const char *pattern, | |
11398 | const struct bpf_kprobe_multi_opts *opts) | |
11399 | { | |
11400 | LIBBPF_OPTS(bpf_link_create_opts, lopts); | |
11401 | struct kprobe_multi_resolve res = { | |
11402 | .pattern = pattern, | |
11403 | }; | |
2ca178f0 | 11404 | enum bpf_attach_type attach_type; |
ddc6b049 JO |
11405 | struct bpf_link *link = NULL; |
11406 | char errmsg[STRERR_BUFSIZE]; | |
11407 | const unsigned long *addrs; | |
11408 | int err, link_fd, prog_fd; | |
2ca178f0 | 11409 | bool retprobe, session; |
ddc6b049 JO |
11410 | const __u64 *cookies; |
11411 | const char **syms; | |
ddc6b049 JO |
11412 | size_t cnt; |
11413 | ||
11414 | if (!OPTS_VALID(opts, bpf_kprobe_multi_opts)) | |
11415 | return libbpf_err_ptr(-EINVAL); | |
11416 | ||
7b30c296 MY |
11417 | prog_fd = bpf_program__fd(prog); |
11418 | if (prog_fd < 0) { | |
11419 | pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n", | |
11420 | prog->name); | |
11421 | return libbpf_err_ptr(-EINVAL); | |
11422 | } | |
11423 | ||
ddc6b049 JO |
11424 | syms = OPTS_GET(opts, syms, false); |
11425 | addrs = OPTS_GET(opts, addrs, false); | |
11426 | cnt = OPTS_GET(opts, cnt, false); | |
11427 | cookies = OPTS_GET(opts, cookies, false); | |
11428 | ||
11429 | if (!pattern && !addrs && !syms) | |
11430 | return libbpf_err_ptr(-EINVAL); | |
11431 | if (pattern && (addrs || syms || cookies || cnt)) | |
11432 | return libbpf_err_ptr(-EINVAL); | |
11433 | if (!pattern && !cnt) | |
11434 | return libbpf_err_ptr(-EINVAL); | |
11435 | if (addrs && syms) | |
11436 | return libbpf_err_ptr(-EINVAL); | |
11437 | ||
11438 | if (pattern) { | |
56baeeba JL |
11439 | if (has_available_filter_functions_addrs()) |
11440 | err = libbpf_available_kprobes_parse(&res); | |
11441 | else | |
11442 | err = libbpf_available_kallsyms_parse(&res); | |
ddc6b049 JO |
11443 | if (err) |
11444 | goto error; | |
ddc6b049 JO |
11445 | addrs = res.addrs; |
11446 | cnt = res.cnt; | |
11447 | } | |
11448 | ||
11449 | retprobe = OPTS_GET(opts, retprobe, false); | |
2ca178f0 JO |
11450 | session = OPTS_GET(opts, session, false); |
11451 | ||
11452 | if (retprobe && session) | |
11453 | return libbpf_err_ptr(-EINVAL); | |
11454 | ||
11455 | attach_type = session ? BPF_TRACE_KPROBE_SESSION : BPF_TRACE_KPROBE_MULTI; | |
ddc6b049 JO |
11456 | |
11457 | lopts.kprobe_multi.syms = syms; | |
11458 | lopts.kprobe_multi.addrs = addrs; | |
11459 | lopts.kprobe_multi.cookies = cookies; | |
11460 | lopts.kprobe_multi.cnt = cnt; | |
11461 | lopts.kprobe_multi.flags = retprobe ? BPF_F_KPROBE_MULTI_RETURN : 0; | |
11462 | ||
11463 | link = calloc(1, sizeof(*link)); | |
11464 | if (!link) { | |
11465 | err = -ENOMEM; | |
11466 | goto error; | |
11467 | } | |
11468 | link->detach = &bpf_link__detach_fd; | |
11469 | ||
2ca178f0 | 11470 | link_fd = bpf_link_create(prog_fd, 0, attach_type, &lopts); |
ddc6b049 JO |
11471 | if (link_fd < 0) { |
11472 | err = -errno; | |
11473 | pr_warn("prog '%s': failed to attach: %s\n", | |
11474 | prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
11475 | goto error; | |
11476 | } | |
11477 | link->fd = link_fd; | |
11478 | free(res.addrs); | |
11479 | return link; | |
11480 | ||
11481 | error: | |
11482 | free(link); | |
11483 | free(res.addrs); | |
11484 | return libbpf_err_ptr(err); | |
11485 | } | |
11486 | ||
4fa5bcfe | 11487 | static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 11488 | { |
da97553e | 11489 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts); |
a2488b5f | 11490 | unsigned long offset = 0; |
d7a18ea7 | 11491 | const char *func_name; |
a2488b5f | 11492 | char *func; |
4fa5bcfe | 11493 | int n; |
d7a18ea7 | 11494 | |
9af8efc4 AN |
11495 | *link = NULL; |
11496 | ||
11497 | /* no auto-attach for SEC("kprobe") and SEC("kretprobe") */ | |
11498 | if (strcmp(prog->sec_name, "kprobe") == 0 || strcmp(prog->sec_name, "kretprobe") == 0) | |
11499 | return 0; | |
11500 | ||
13d35a0c AN |
11501 | opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/"); |
11502 | if (opts.retprobe) | |
11503 | func_name = prog->sec_name + sizeof("kretprobe/") - 1; | |
11504 | else | |
11505 | func_name = prog->sec_name + sizeof("kprobe/") - 1; | |
d7a18ea7 | 11506 | |
e3f9bc35 | 11507 | n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset); |
a2488b5f | 11508 | if (n < 1) { |
a2488b5f | 11509 | pr_warn("kprobe name is invalid: %s\n", func_name); |
4fa5bcfe | 11510 | return -EINVAL; |
a2488b5f AM |
11511 | } |
11512 | if (opts.retprobe && offset != 0) { | |
1f71a468 | 11513 | free(func); |
a2488b5f | 11514 | pr_warn("kretprobes do not support offset specification\n"); |
4fa5bcfe | 11515 | return -EINVAL; |
a2488b5f | 11516 | } |
d7a18ea7 | 11517 | |
a2488b5f | 11518 | opts.offset = offset; |
4fa5bcfe | 11519 | *link = bpf_program__attach_kprobe_opts(prog, func, &opts); |
a2488b5f | 11520 | free(func); |
4fa5bcfe | 11521 | return libbpf_get_error(*link); |
d7a18ea7 AN |
11522 | } |
11523 | ||
708ac5be AN |
11524 | static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
11525 | { | |
11526 | LIBBPF_OPTS(bpf_ksyscall_opts, opts); | |
11527 | const char *syscall_name; | |
11528 | ||
11529 | *link = NULL; | |
11530 | ||
11531 | /* no auto-attach for SEC("ksyscall") and SEC("kretsyscall") */ | |
11532 | if (strcmp(prog->sec_name, "ksyscall") == 0 || strcmp(prog->sec_name, "kretsyscall") == 0) | |
11533 | return 0; | |
11534 | ||
11535 | opts.retprobe = str_has_pfx(prog->sec_name, "kretsyscall/"); | |
11536 | if (opts.retprobe) | |
11537 | syscall_name = prog->sec_name + sizeof("kretsyscall/") - 1; | |
11538 | else | |
11539 | syscall_name = prog->sec_name + sizeof("ksyscall/") - 1; | |
11540 | ||
11541 | *link = bpf_program__attach_ksyscall(prog, syscall_name, &opts); | |
11542 | return *link ? 0 : -errno; | |
11543 | } | |
11544 | ||
ddc6b049 JO |
11545 | static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
11546 | { | |
11547 | LIBBPF_OPTS(bpf_kprobe_multi_opts, opts); | |
11548 | const char *spec; | |
11549 | char *pattern; | |
11550 | int n; | |
11551 | ||
9af8efc4 AN |
11552 | *link = NULL; |
11553 | ||
11554 | /* no auto-attach for SEC("kprobe.multi") and SEC("kretprobe.multi") */ | |
11555 | if (strcmp(prog->sec_name, "kprobe.multi") == 0 || | |
11556 | strcmp(prog->sec_name, "kretprobe.multi") == 0) | |
11557 | return 0; | |
11558 | ||
ddc6b049 JO |
11559 | opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe.multi/"); |
11560 | if (opts.retprobe) | |
11561 | spec = prog->sec_name + sizeof("kretprobe.multi/") - 1; | |
11562 | else | |
11563 | spec = prog->sec_name + sizeof("kprobe.multi/") - 1; | |
11564 | ||
11565 | n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern); | |
11566 | if (n < 1) { | |
11567 | pr_warn("kprobe multi pattern is invalid: %s\n", pattern); | |
11568 | return -EINVAL; | |
11569 | } | |
11570 | ||
11571 | *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts); | |
11572 | free(pattern); | |
11573 | return libbpf_get_error(*link); | |
11574 | } | |
11575 | ||
2ca178f0 JO |
11576 | static int attach_kprobe_session(const struct bpf_program *prog, long cookie, |
11577 | struct bpf_link **link) | |
11578 | { | |
11579 | LIBBPF_OPTS(bpf_kprobe_multi_opts, opts, .session = true); | |
11580 | const char *spec; | |
11581 | char *pattern; | |
11582 | int n; | |
11583 | ||
11584 | *link = NULL; | |
11585 | ||
11586 | /* no auto-attach for SEC("kprobe.session") */ | |
11587 | if (strcmp(prog->sec_name, "kprobe.session") == 0) | |
11588 | return 0; | |
11589 | ||
11590 | spec = prog->sec_name + sizeof("kprobe.session/") - 1; | |
11591 | n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern); | |
11592 | if (n < 1) { | |
11593 | pr_warn("kprobe session pattern is invalid: %s\n", pattern); | |
11594 | return -EINVAL; | |
11595 | } | |
11596 | ||
11597 | *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts); | |
11598 | free(pattern); | |
11599 | return *link ? 0 : -errno; | |
11600 | } | |
11601 | ||
5bfdd32d JO |
11602 | static int attach_uprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
11603 | { | |
11604 | char *probe_type = NULL, *binary_path = NULL, *func_name = NULL; | |
11605 | LIBBPF_OPTS(bpf_uprobe_multi_opts, opts); | |
11606 | int n, ret = -EINVAL; | |
11607 | ||
11608 | *link = NULL; | |
11609 | ||
2147c8d0 | 11610 | n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[^\n]", |
5bfdd32d JO |
11611 | &probe_type, &binary_path, &func_name); |
11612 | switch (n) { | |
11613 | case 1: | |
11614 | /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */ | |
11615 | ret = 0; | |
11616 | break; | |
11617 | case 3: | |
11618 | opts.retprobe = strcmp(probe_type, "uretprobe.multi") == 0; | |
11619 | *link = bpf_program__attach_uprobe_multi(prog, -1, binary_path, func_name, &opts); | |
11620 | ret = libbpf_get_error(*link); | |
11621 | break; | |
11622 | default: | |
11623 | pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name, | |
11624 | prog->sec_name); | |
11625 | break; | |
11626 | } | |
11627 | free(probe_type); | |
11628 | free(binary_path); | |
11629 | free(func_name); | |
11630 | return ret; | |
11631 | } | |
11632 | ||
cc10623c AN |
11633 | static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz, |
11634 | const char *binary_path, uint64_t offset) | |
11635 | { | |
11636 | int i; | |
11637 | ||
11638 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset); | |
11639 | ||
11640 | /* sanitize binary_path in the probe name */ | |
11641 | for (i = 0; buf[i]; i++) { | |
11642 | if (!isalnum(buf[i])) | |
11643 | buf[i] = '_'; | |
11644 | } | |
11645 | } | |
11646 | ||
11647 | static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe, | |
11648 | const char *binary_path, size_t offset) | |
11649 | { | |
a1ac9fd6 | 11650 | return append_to_file(tracefs_uprobe_events(), "%c:%s/%s %s:0x%zx", |
cc10623c AN |
11651 | retprobe ? 'r' : 'p', |
11652 | retprobe ? "uretprobes" : "uprobes", | |
11653 | probe_name, binary_path, offset); | |
11654 | } | |
11655 | ||
11656 | static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe) | |
11657 | { | |
a1ac9fd6 AN |
11658 | return append_to_file(tracefs_uprobe_events(), "-:%s/%s", |
11659 | retprobe ? "uretprobes" : "uprobes", probe_name); | |
cc10623c AN |
11660 | } |
11661 | ||
11662 | static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
11663 | { | |
11664 | char file[512]; | |
11665 | ||
a1ac9fd6 AN |
11666 | snprintf(file, sizeof(file), "%s/events/%s/%s/id", |
11667 | tracefs_path(), retprobe ? "uretprobes" : "uprobes", probe_name); | |
cc10623c AN |
11668 | |
11669 | return parse_uint_from_file(file, "%d\n"); | |
11670 | } | |
11671 | ||
11672 | static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe, | |
11673 | const char *binary_path, size_t offset, int pid) | |
11674 | { | |
813847a3 | 11675 | const size_t attr_sz = sizeof(struct perf_event_attr); |
cc10623c AN |
11676 | struct perf_event_attr attr; |
11677 | int type, pfd, err; | |
11678 | ||
11679 | err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset); | |
11680 | if (err < 0) { | |
11681 | pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n", | |
11682 | binary_path, (size_t)offset, err); | |
11683 | return err; | |
11684 | } | |
11685 | type = determine_uprobe_perf_type_legacy(probe_name, retprobe); | |
11686 | if (type < 0) { | |
2655144f | 11687 | err = type; |
cc10623c | 11688 | pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n", |
2655144f CW |
11689 | binary_path, offset, err); |
11690 | goto err_clean_legacy; | |
cc10623c AN |
11691 | } |
11692 | ||
813847a3 AN |
11693 | memset(&attr, 0, attr_sz); |
11694 | attr.size = attr_sz; | |
cc10623c AN |
11695 | attr.config = type; |
11696 | attr.type = PERF_TYPE_TRACEPOINT; | |
11697 | ||
11698 | pfd = syscall(__NR_perf_event_open, &attr, | |
11699 | pid < 0 ? -1 : pid, /* pid */ | |
11700 | pid == -1 ? 0 : -1, /* cpu */ | |
11701 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
11702 | if (pfd < 0) { | |
11703 | err = -errno; | |
11704 | pr_warn("legacy uprobe perf_event_open() failed: %d\n", err); | |
2655144f | 11705 | goto err_clean_legacy; |
cc10623c AN |
11706 | } |
11707 | return pfd; | |
2655144f CW |
11708 | |
11709 | err_clean_legacy: | |
11710 | /* Clear the newly added legacy uprobe_event */ | |
11711 | remove_uprobe_event_legacy(probe_name, retprobe); | |
11712 | return err; | |
cc10623c AN |
11713 | } |
11714 | ||
c44fd845 DM |
11715 | /* Find offset of function name in archive specified by path. Currently |
11716 | * supported are .zip files that do not compress their contents, as used on | |
11717 | * Android in the form of APKs, for example. "file_name" is the name of the ELF | |
11718 | * file inside the archive. "func_name" matches symbol name or name@@LIB for | |
11719 | * library functions. | |
11720 | * | |
11721 | * An overview of the APK format specifically provided here: | |
11722 | * https://en.wikipedia.org/w/index.php?title=Apk_(file_format)&oldid=1139099120#Package_contents | |
11723 | */ | |
11724 | static long elf_find_func_offset_from_archive(const char *archive_path, const char *file_name, | |
11725 | const char *func_name) | |
11726 | { | |
11727 | struct zip_archive *archive; | |
11728 | struct zip_entry entry; | |
11729 | long ret; | |
11730 | Elf *elf; | |
11731 | ||
11732 | archive = zip_archive_open(archive_path); | |
11733 | if (IS_ERR(archive)) { | |
11734 | ret = PTR_ERR(archive); | |
11735 | pr_warn("zip: failed to open %s: %ld\n", archive_path, ret); | |
11736 | return ret; | |
11737 | } | |
11738 | ||
11739 | ret = zip_archive_find_entry(archive, file_name, &entry); | |
11740 | if (ret) { | |
11741 | pr_warn("zip: could not find archive member %s in %s: %ld\n", file_name, | |
11742 | archive_path, ret); | |
11743 | goto out; | |
11744 | } | |
11745 | pr_debug("zip: found entry for %s in %s at 0x%lx\n", file_name, archive_path, | |
11746 | (unsigned long)entry.data_offset); | |
11747 | ||
11748 | if (entry.compression) { | |
11749 | pr_warn("zip: entry %s of %s is compressed and cannot be handled\n", file_name, | |
11750 | archive_path); | |
11751 | ret = -LIBBPF_ERRNO__FORMAT; | |
11752 | goto out; | |
11753 | } | |
11754 | ||
11755 | elf = elf_memory((void *)entry.data, entry.data_length); | |
11756 | if (!elf) { | |
11757 | pr_warn("elf: could not read elf file %s from %s: %s\n", file_name, archive_path, | |
11758 | elf_errmsg(-1)); | |
11759 | ret = -LIBBPF_ERRNO__LIBELF; | |
11760 | goto out; | |
11761 | } | |
11762 | ||
11763 | ret = elf_find_func_offset(elf, file_name, func_name); | |
11764 | if (ret > 0) { | |
11765 | pr_debug("elf: symbol address match for %s of %s in %s: 0x%x + 0x%lx = 0x%lx\n", | |
11766 | func_name, file_name, archive_path, entry.data_offset, ret, | |
11767 | ret + entry.data_offset); | |
11768 | ret += entry.data_offset; | |
11769 | } | |
11770 | elf_end(elf); | |
11771 | ||
11772 | out: | |
11773 | zip_archive_close(archive); | |
11774 | return ret; | |
11775 | } | |
11776 | ||
56818931 IL |
11777 | static const char *arch_specific_lib_paths(void) |
11778 | { | |
11779 | /* | |
11780 | * Based on https://packages.debian.org/sid/libc6. | |
11781 | * | |
11782 | * Assume that the traced program is built for the same architecture | |
11783 | * as libbpf, which should cover the vast majority of cases. | |
11784 | */ | |
11785 | #if defined(__x86_64__) | |
11786 | return "/lib/x86_64-linux-gnu"; | |
11787 | #elif defined(__i386__) | |
11788 | return "/lib/i386-linux-gnu"; | |
11789 | #elif defined(__s390x__) | |
11790 | return "/lib/s390x-linux-gnu"; | |
11791 | #elif defined(__s390__) | |
11792 | return "/lib/s390-linux-gnu"; | |
11793 | #elif defined(__arm__) && defined(__SOFTFP__) | |
11794 | return "/lib/arm-linux-gnueabi"; | |
11795 | #elif defined(__arm__) && !defined(__SOFTFP__) | |
11796 | return "/lib/arm-linux-gnueabihf"; | |
11797 | #elif defined(__aarch64__) | |
11798 | return "/lib/aarch64-linux-gnu"; | |
11799 | #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 64 | |
11800 | return "/lib/mips64el-linux-gnuabi64"; | |
11801 | #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 32 | |
11802 | return "/lib/mipsel-linux-gnu"; | |
11803 | #elif defined(__powerpc64__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ | |
11804 | return "/lib/powerpc64le-linux-gnu"; | |
11805 | #elif defined(__sparc__) && defined(__arch64__) | |
11806 | return "/lib/sparc64-linux-gnu"; | |
11807 | #elif defined(__riscv) && __riscv_xlen == 64 | |
11808 | return "/lib/riscv64-linux-gnu"; | |
11809 | #else | |
11810 | return NULL; | |
11811 | #endif | |
11812 | } | |
11813 | ||
1ce3a60e AM |
11814 | /* Get full path to program/shared library. */ |
11815 | static int resolve_full_path(const char *file, char *result, size_t result_sz) | |
11816 | { | |
56818931 | 11817 | const char *search_paths[3] = {}; |
9e32084e | 11818 | int i, perm; |
1ce3a60e | 11819 | |
a1c9d61b | 11820 | if (str_has_sfx(file, ".so") || strstr(file, ".so.")) { |
1ce3a60e AM |
11821 | search_paths[0] = getenv("LD_LIBRARY_PATH"); |
11822 | search_paths[1] = "/usr/lib64:/usr/lib"; | |
56818931 | 11823 | search_paths[2] = arch_specific_lib_paths(); |
9e32084e | 11824 | perm = R_OK; |
1ce3a60e AM |
11825 | } else { |
11826 | search_paths[0] = getenv("PATH"); | |
11827 | search_paths[1] = "/usr/bin:/usr/sbin"; | |
9e32084e | 11828 | perm = R_OK | X_OK; |
1ce3a60e AM |
11829 | } |
11830 | ||
11831 | for (i = 0; i < ARRAY_SIZE(search_paths); i++) { | |
11832 | const char *s; | |
11833 | ||
11834 | if (!search_paths[i]) | |
11835 | continue; | |
11836 | for (s = search_paths[i]; s != NULL; s = strchr(s, ':')) { | |
11837 | char *next_path; | |
11838 | int seg_len; | |
11839 | ||
11840 | if (s[0] == ':') | |
11841 | s++; | |
11842 | next_path = strchr(s, ':'); | |
11843 | seg_len = next_path ? next_path - s : strlen(s); | |
11844 | if (!seg_len) | |
11845 | continue; | |
11846 | snprintf(result, result_sz, "%.*s/%s", seg_len, s, file); | |
9e32084e | 11847 | /* ensure it has required permissions */ |
6a4ab886 | 11848 | if (faccessat(AT_FDCWD, result, perm, AT_EACCESS) < 0) |
1ce3a60e AM |
11849 | continue; |
11850 | pr_debug("resolved '%s' to '%s'\n", file, result); | |
11851 | return 0; | |
11852 | } | |
11853 | } | |
11854 | return -ENOENT; | |
11855 | } | |
11856 | ||
3140cf12 JO |
11857 | struct bpf_link * |
11858 | bpf_program__attach_uprobe_multi(const struct bpf_program *prog, | |
11859 | pid_t pid, | |
11860 | const char *path, | |
11861 | const char *func_pattern, | |
11862 | const struct bpf_uprobe_multi_opts *opts) | |
11863 | { | |
11864 | const unsigned long *ref_ctr_offsets = NULL, *offsets = NULL; | |
11865 | LIBBPF_OPTS(bpf_link_create_opts, lopts); | |
11866 | unsigned long *resolved_offsets = NULL; | |
11867 | int err = 0, link_fd, prog_fd; | |
11868 | struct bpf_link *link = NULL; | |
11869 | char errmsg[STRERR_BUFSIZE]; | |
11870 | char full_path[PATH_MAX]; | |
11871 | const __u64 *cookies; | |
11872 | const char **syms; | |
11873 | size_t cnt; | |
11874 | ||
11875 | if (!OPTS_VALID(opts, bpf_uprobe_multi_opts)) | |
11876 | return libbpf_err_ptr(-EINVAL); | |
11877 | ||
7b30c296 MY |
11878 | prog_fd = bpf_program__fd(prog); |
11879 | if (prog_fd < 0) { | |
11880 | pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n", | |
11881 | prog->name); | |
11882 | return libbpf_err_ptr(-EINVAL); | |
11883 | } | |
11884 | ||
3140cf12 JO |
11885 | syms = OPTS_GET(opts, syms, NULL); |
11886 | offsets = OPTS_GET(opts, offsets, NULL); | |
11887 | ref_ctr_offsets = OPTS_GET(opts, ref_ctr_offsets, NULL); | |
11888 | cookies = OPTS_GET(opts, cookies, NULL); | |
11889 | cnt = OPTS_GET(opts, cnt, 0); | |
11890 | ||
11891 | /* | |
11892 | * User can specify 2 mutually exclusive set of inputs: | |
11893 | * | |
11894 | * 1) use only path/func_pattern/pid arguments | |
11895 | * | |
11896 | * 2) use path/pid with allowed combinations of: | |
11897 | * syms/offsets/ref_ctr_offsets/cookies/cnt | |
11898 | * | |
11899 | * - syms and offsets are mutually exclusive | |
11900 | * - ref_ctr_offsets and cookies are optional | |
11901 | * | |
11902 | * Any other usage results in error. | |
11903 | */ | |
11904 | ||
11905 | if (!path) | |
11906 | return libbpf_err_ptr(-EINVAL); | |
11907 | if (!func_pattern && cnt == 0) | |
11908 | return libbpf_err_ptr(-EINVAL); | |
11909 | ||
11910 | if (func_pattern) { | |
11911 | if (syms || offsets || ref_ctr_offsets || cookies || cnt) | |
11912 | return libbpf_err_ptr(-EINVAL); | |
11913 | } else { | |
11914 | if (!!syms == !!offsets) | |
11915 | return libbpf_err_ptr(-EINVAL); | |
11916 | } | |
11917 | ||
11918 | if (func_pattern) { | |
11919 | if (!strchr(path, '/')) { | |
11920 | err = resolve_full_path(path, full_path, sizeof(full_path)); | |
11921 | if (err) { | |
11922 | pr_warn("prog '%s': failed to resolve full path for '%s': %d\n", | |
11923 | prog->name, path, err); | |
11924 | return libbpf_err_ptr(err); | |
11925 | } | |
11926 | path = full_path; | |
11927 | } | |
11928 | ||
11929 | err = elf_resolve_pattern_offsets(path, func_pattern, | |
11930 | &resolved_offsets, &cnt); | |
11931 | if (err < 0) | |
11932 | return libbpf_err_ptr(err); | |
11933 | offsets = resolved_offsets; | |
11934 | } else if (syms) { | |
48f0dfd8 | 11935 | err = elf_resolve_syms_offsets(path, cnt, syms, &resolved_offsets, STT_FUNC); |
3140cf12 JO |
11936 | if (err < 0) |
11937 | return libbpf_err_ptr(err); | |
11938 | offsets = resolved_offsets; | |
11939 | } | |
11940 | ||
11941 | lopts.uprobe_multi.path = path; | |
11942 | lopts.uprobe_multi.offsets = offsets; | |
11943 | lopts.uprobe_multi.ref_ctr_offsets = ref_ctr_offsets; | |
11944 | lopts.uprobe_multi.cookies = cookies; | |
11945 | lopts.uprobe_multi.cnt = cnt; | |
11946 | lopts.uprobe_multi.flags = OPTS_GET(opts, retprobe, false) ? BPF_F_UPROBE_MULTI_RETURN : 0; | |
11947 | ||
11948 | if (pid == 0) | |
11949 | pid = getpid(); | |
11950 | if (pid > 0) | |
11951 | lopts.uprobe_multi.pid = pid; | |
11952 | ||
11953 | link = calloc(1, sizeof(*link)); | |
11954 | if (!link) { | |
11955 | err = -ENOMEM; | |
11956 | goto error; | |
11957 | } | |
11958 | link->detach = &bpf_link__detach_fd; | |
11959 | ||
3140cf12 JO |
11960 | link_fd = bpf_link_create(prog_fd, 0, BPF_TRACE_UPROBE_MULTI, &lopts); |
11961 | if (link_fd < 0) { | |
11962 | err = -errno; | |
11963 | pr_warn("prog '%s': failed to attach multi-uprobe: %s\n", | |
11964 | prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
11965 | goto error; | |
11966 | } | |
11967 | link->fd = link_fd; | |
11968 | free(resolved_offsets); | |
11969 | return link; | |
11970 | ||
11971 | error: | |
11972 | free(resolved_offsets); | |
11973 | free(link); | |
11974 | return libbpf_err_ptr(err); | |
11975 | } | |
11976 | ||
47faff37 | 11977 | LIBBPF_API struct bpf_link * |
942025c9 | 11978 | bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid, |
47faff37 AN |
11979 | const char *binary_path, size_t func_offset, |
11980 | const struct bpf_uprobe_opts *opts) | |
b2650027 | 11981 | { |
c44fd845 | 11982 | const char *archive_path = NULL, *archive_sep = NULL; |
cc10623c | 11983 | char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL; |
c44fd845 | 11984 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f8b299bc | 11985 | enum probe_attach_mode attach_mode; |
c44fd845 | 11986 | char full_path[PATH_MAX]; |
b2650027 | 11987 | struct bpf_link *link; |
5e3b8356 | 11988 | size_t ref_ctr_off; |
b2650027 | 11989 | int pfd, err; |
cc10623c | 11990 | bool retprobe, legacy; |
433966e3 | 11991 | const char *func_name; |
47faff37 AN |
11992 | |
11993 | if (!OPTS_VALID(opts, bpf_uprobe_opts)) | |
11994 | return libbpf_err_ptr(-EINVAL); | |
11995 | ||
f8b299bc | 11996 | attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT); |
47faff37 | 11997 | retprobe = OPTS_GET(opts, retprobe, false); |
5e3b8356 | 11998 | ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0); |
47faff37 | 11999 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
b2650027 | 12000 | |
8ed2f5a6 HC |
12001 | if (!binary_path) |
12002 | return libbpf_err_ptr(-EINVAL); | |
12003 | ||
c44fd845 DM |
12004 | /* Check if "binary_path" refers to an archive. */ |
12005 | archive_sep = strstr(binary_path, "!/"); | |
12006 | if (archive_sep) { | |
12007 | full_path[0] = '\0'; | |
12008 | libbpf_strlcpy(full_path, binary_path, | |
12009 | min(sizeof(full_path), (size_t)(archive_sep - binary_path + 1))); | |
12010 | archive_path = full_path; | |
12011 | binary_path = archive_sep + 2; | |
12012 | } else if (!strchr(binary_path, '/')) { | |
12013 | err = resolve_full_path(binary_path, full_path, sizeof(full_path)); | |
1ce3a60e | 12014 | if (err) { |
2e4913e0 AN |
12015 | pr_warn("prog '%s': failed to resolve full path for '%s': %d\n", |
12016 | prog->name, binary_path, err); | |
1ce3a60e AM |
12017 | return libbpf_err_ptr(err); |
12018 | } | |
c44fd845 | 12019 | binary_path = full_path; |
1ce3a60e | 12020 | } |
433966e3 AM |
12021 | func_name = OPTS_GET(opts, func_name, NULL); |
12022 | if (func_name) { | |
12023 | long sym_off; | |
12024 | ||
c44fd845 DM |
12025 | if (archive_path) { |
12026 | sym_off = elf_find_func_offset_from_archive(archive_path, binary_path, | |
12027 | func_name); | |
12028 | binary_path = archive_path; | |
12029 | } else { | |
12030 | sym_off = elf_find_func_offset_from_file(binary_path, func_name); | |
12031 | } | |
433966e3 AM |
12032 | if (sym_off < 0) |
12033 | return libbpf_err_ptr(sym_off); | |
12034 | func_offset += sym_off; | |
12035 | } | |
1ce3a60e | 12036 | |
cc10623c | 12037 | legacy = determine_uprobe_perf_type() < 0; |
f8b299bc MD |
12038 | switch (attach_mode) { |
12039 | case PROBE_ATTACH_MODE_LEGACY: | |
12040 | legacy = true; | |
12041 | pe_opts.force_ioctl_attach = true; | |
12042 | break; | |
12043 | case PROBE_ATTACH_MODE_PERF: | |
12044 | if (legacy) | |
12045 | return libbpf_err_ptr(-ENOTSUP); | |
12046 | pe_opts.force_ioctl_attach = true; | |
12047 | break; | |
12048 | case PROBE_ATTACH_MODE_LINK: | |
12049 | if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK)) | |
12050 | return libbpf_err_ptr(-ENOTSUP); | |
12051 | break; | |
12052 | case PROBE_ATTACH_MODE_DEFAULT: | |
12053 | break; | |
12054 | default: | |
12055 | return libbpf_err_ptr(-EINVAL); | |
12056 | } | |
12057 | ||
cc10623c AN |
12058 | if (!legacy) { |
12059 | pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path, | |
12060 | func_offset, pid, ref_ctr_off); | |
12061 | } else { | |
1ce3a60e | 12062 | char probe_name[PATH_MAX + 64]; |
cc10623c AN |
12063 | |
12064 | if (ref_ctr_off) | |
12065 | return libbpf_err_ptr(-EINVAL); | |
12066 | ||
12067 | gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
12068 | binary_path, func_offset); | |
12069 | ||
12070 | legacy_probe = strdup(probe_name); | |
12071 | if (!legacy_probe) | |
12072 | return libbpf_err_ptr(-ENOMEM); | |
12073 | ||
12074 | pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe, | |
12075 | binary_path, func_offset, pid); | |
12076 | } | |
b2650027 | 12077 | if (pfd < 0) { |
cc10623c | 12078 | err = -errno; |
52109584 AN |
12079 | pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n", |
12080 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e | 12081 | binary_path, func_offset, |
cc10623c AN |
12082 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
12083 | goto err_out; | |
b2650027 | 12084 | } |
cc10623c | 12085 | |
47faff37 | 12086 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
12087 | err = libbpf_get_error(link); |
12088 | if (err) { | |
b2650027 | 12089 | close(pfd); |
52109584 AN |
12090 | pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n", |
12091 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e KW |
12092 | binary_path, func_offset, |
12093 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
2655144f | 12094 | goto err_clean_legacy; |
cc10623c AN |
12095 | } |
12096 | if (legacy) { | |
12097 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
12098 | ||
12099 | perf_link->legacy_probe_name = legacy_probe; | |
12100 | perf_link->legacy_is_kprobe = false; | |
12101 | perf_link->legacy_is_retprobe = retprobe; | |
b2650027 AN |
12102 | } |
12103 | return link; | |
2655144f CW |
12104 | |
12105 | err_clean_legacy: | |
12106 | if (legacy) | |
12107 | remove_uprobe_event_legacy(legacy_probe, retprobe); | |
cc10623c AN |
12108 | err_out: |
12109 | free(legacy_probe); | |
12110 | return libbpf_err_ptr(err); | |
b2650027 AN |
12111 | } |
12112 | ||
39f8dc43 AM |
12113 | /* Format of u[ret]probe section definition supporting auto-attach: |
12114 | * u[ret]probe/binary:function[+offset] | |
12115 | * | |
12116 | * binary can be an absolute/relative path or a filename; the latter is resolved to a | |
12117 | * full binary path via bpf_program__attach_uprobe_opts. | |
12118 | * | |
12119 | * Specifying uprobe+ ensures we carry out strict matching; either "uprobe" must be | |
12120 | * specified (and auto-attach is not possible) or the above format is specified for | |
12121 | * auto-attach. | |
12122 | */ | |
12123 | static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link) | |
12124 | { | |
12125 | DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts); | |
2147c8d0 HC |
12126 | char *probe_type = NULL, *binary_path = NULL, *func_name = NULL, *func_off; |
12127 | int n, c, ret = -EINVAL; | |
90db26e6 | 12128 | long offset = 0; |
39f8dc43 AM |
12129 | |
12130 | *link = NULL; | |
12131 | ||
2147c8d0 HC |
12132 | n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[^\n]", |
12133 | &probe_type, &binary_path, &func_name); | |
90db26e6 AM |
12134 | switch (n) { |
12135 | case 1: | |
12136 | /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */ | |
12137 | ret = 0; | |
12138 | break; | |
12139 | case 2: | |
12140 | pr_warn("prog '%s': section '%s' missing ':function[+offset]' specification\n", | |
12141 | prog->name, prog->sec_name); | |
12142 | break; | |
12143 | case 3: | |
2147c8d0 HC |
12144 | /* check if user specifies `+offset`, if yes, this should be |
12145 | * the last part of the string, make sure sscanf read to EOL | |
12146 | */ | |
12147 | func_off = strrchr(func_name, '+'); | |
12148 | if (func_off) { | |
12149 | n = sscanf(func_off, "+%li%n", &offset, &c); | |
12150 | if (n == 1 && *(func_off + c) == '\0') | |
12151 | func_off[0] = '\0'; | |
12152 | else | |
12153 | offset = 0; | |
12154 | } | |
c4cac71f DK |
12155 | opts.retprobe = strcmp(probe_type, "uretprobe") == 0 || |
12156 | strcmp(probe_type, "uretprobe.s") == 0; | |
90db26e6 AM |
12157 | if (opts.retprobe && offset != 0) { |
12158 | pr_warn("prog '%s': uretprobes do not support offset specification\n", | |
12159 | prog->name); | |
12160 | break; | |
12161 | } | |
12162 | opts.func_name = func_name; | |
12163 | *link = bpf_program__attach_uprobe_opts(prog, -1, binary_path, offset, &opts); | |
12164 | ret = libbpf_get_error(*link); | |
12165 | break; | |
12166 | default: | |
12167 | pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name, | |
39f8dc43 | 12168 | prog->sec_name); |
90db26e6 | 12169 | break; |
39f8dc43 | 12170 | } |
90db26e6 AM |
12171 | free(probe_type); |
12172 | free(binary_path); | |
12173 | free(func_name); | |
39f8dc43 | 12174 | |
90db26e6 | 12175 | return ret; |
39f8dc43 AM |
12176 | } |
12177 | ||
942025c9 | 12178 | struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog, |
47faff37 AN |
12179 | bool retprobe, pid_t pid, |
12180 | const char *binary_path, | |
12181 | size_t func_offset) | |
12182 | { | |
12183 | DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe); | |
12184 | ||
12185 | return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts); | |
12186 | } | |
12187 | ||
2e4913e0 AN |
12188 | struct bpf_link *bpf_program__attach_usdt(const struct bpf_program *prog, |
12189 | pid_t pid, const char *binary_path, | |
12190 | const char *usdt_provider, const char *usdt_name, | |
12191 | const struct bpf_usdt_opts *opts) | |
12192 | { | |
12193 | char resolved_path[512]; | |
12194 | struct bpf_object *obj = prog->obj; | |
12195 | struct bpf_link *link; | |
5af25a41 | 12196 | __u64 usdt_cookie; |
2e4913e0 AN |
12197 | int err; |
12198 | ||
12199 | if (!OPTS_VALID(opts, bpf_uprobe_opts)) | |
12200 | return libbpf_err_ptr(-EINVAL); | |
12201 | ||
12202 | if (bpf_program__fd(prog) < 0) { | |
7b30c296 | 12203 | pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n", |
2e4913e0 AN |
12204 | prog->name); |
12205 | return libbpf_err_ptr(-EINVAL); | |
12206 | } | |
12207 | ||
8ed2f5a6 HC |
12208 | if (!binary_path) |
12209 | return libbpf_err_ptr(-EINVAL); | |
12210 | ||
2e4913e0 AN |
12211 | if (!strchr(binary_path, '/')) { |
12212 | err = resolve_full_path(binary_path, resolved_path, sizeof(resolved_path)); | |
12213 | if (err) { | |
12214 | pr_warn("prog '%s': failed to resolve full path for '%s': %d\n", | |
12215 | prog->name, binary_path, err); | |
12216 | return libbpf_err_ptr(err); | |
12217 | } | |
12218 | binary_path = resolved_path; | |
12219 | } | |
12220 | ||
12221 | /* USDT manager is instantiated lazily on first USDT attach. It will | |
12222 | * be destroyed together with BPF object in bpf_object__close(). | |
12223 | */ | |
12224 | if (IS_ERR(obj->usdt_man)) | |
12225 | return libbpf_ptr(obj->usdt_man); | |
12226 | if (!obj->usdt_man) { | |
12227 | obj->usdt_man = usdt_manager_new(obj); | |
12228 | if (IS_ERR(obj->usdt_man)) | |
12229 | return libbpf_ptr(obj->usdt_man); | |
12230 | } | |
12231 | ||
12232 | usdt_cookie = OPTS_GET(opts, usdt_cookie, 0); | |
12233 | link = usdt_manager_attach_usdt(obj->usdt_man, prog, pid, binary_path, | |
e3ba8e4e | 12234 | usdt_provider, usdt_name, usdt_cookie); |
2e4913e0 AN |
12235 | err = libbpf_get_error(link); |
12236 | if (err) | |
12237 | return libbpf_err_ptr(err); | |
12238 | return link; | |
12239 | } | |
12240 | ||
12241 | static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link) | |
12242 | { | |
12243 | char *path = NULL, *provider = NULL, *name = NULL; | |
12244 | const char *sec_name; | |
12245 | int n, err; | |
12246 | ||
12247 | sec_name = bpf_program__section_name(prog); | |
12248 | if (strcmp(sec_name, "usdt") == 0) { | |
12249 | /* no auto-attach for just SEC("usdt") */ | |
12250 | *link = NULL; | |
12251 | return 0; | |
12252 | } | |
12253 | ||
12254 | n = sscanf(sec_name, "usdt/%m[^:]:%m[^:]:%m[^:]", &path, &provider, &name); | |
12255 | if (n != 3) { | |
12256 | pr_warn("invalid section '%s', expected SEC(\"usdt/<path>:<provider>:<name>\")\n", | |
12257 | sec_name); | |
12258 | err = -EINVAL; | |
12259 | } else { | |
12260 | *link = bpf_program__attach_usdt(prog, -1 /* any process */, path, | |
12261 | provider, name, NULL); | |
12262 | err = libbpf_get_error(*link); | |
12263 | } | |
12264 | free(path); | |
12265 | free(provider); | |
12266 | free(name); | |
12267 | return err; | |
12268 | } | |
12269 | ||
f6de59c1 AN |
12270 | static int determine_tracepoint_id(const char *tp_category, |
12271 | const char *tp_name) | |
12272 | { | |
12273 | char file[PATH_MAX]; | |
12274 | int ret; | |
12275 | ||
a1ac9fd6 AN |
12276 | ret = snprintf(file, sizeof(file), "%s/events/%s/%s/id", |
12277 | tracefs_path(), tp_category, tp_name); | |
f6de59c1 AN |
12278 | if (ret < 0) |
12279 | return -errno; | |
12280 | if (ret >= sizeof(file)) { | |
12281 | pr_debug("tracepoint %s/%s path is too long\n", | |
12282 | tp_category, tp_name); | |
12283 | return -E2BIG; | |
12284 | } | |
12285 | return parse_uint_from_file(file, "%d\n"); | |
12286 | } | |
12287 | ||
12288 | static int perf_event_open_tracepoint(const char *tp_category, | |
12289 | const char *tp_name) | |
12290 | { | |
813847a3 AN |
12291 | const size_t attr_sz = sizeof(struct perf_event_attr); |
12292 | struct perf_event_attr attr; | |
f6de59c1 AN |
12293 | char errmsg[STRERR_BUFSIZE]; |
12294 | int tp_id, pfd, err; | |
12295 | ||
12296 | tp_id = determine_tracepoint_id(tp_category, tp_name); | |
12297 | if (tp_id < 0) { | |
be18010e KW |
12298 | pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n", |
12299 | tp_category, tp_name, | |
12300 | libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
12301 | return tp_id; |
12302 | } | |
12303 | ||
813847a3 | 12304 | memset(&attr, 0, attr_sz); |
f6de59c1 | 12305 | attr.type = PERF_TYPE_TRACEPOINT; |
813847a3 | 12306 | attr.size = attr_sz; |
f6de59c1 AN |
12307 | attr.config = tp_id; |
12308 | ||
12309 | pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */, | |
12310 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
12311 | if (pfd < 0) { | |
12312 | err = -errno; | |
be18010e KW |
12313 | pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n", |
12314 | tp_category, tp_name, | |
12315 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
12316 | return err; |
12317 | } | |
12318 | return pfd; | |
12319 | } | |
12320 | ||
942025c9 | 12321 | struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog, |
47faff37 AN |
12322 | const char *tp_category, |
12323 | const char *tp_name, | |
12324 | const struct bpf_tracepoint_opts *opts) | |
f6de59c1 | 12325 | { |
47faff37 | 12326 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f6de59c1 AN |
12327 | char errmsg[STRERR_BUFSIZE]; |
12328 | struct bpf_link *link; | |
12329 | int pfd, err; | |
12330 | ||
47faff37 AN |
12331 | if (!OPTS_VALID(opts, bpf_tracepoint_opts)) |
12332 | return libbpf_err_ptr(-EINVAL); | |
12333 | ||
12334 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); | |
12335 | ||
f6de59c1 AN |
12336 | pfd = perf_event_open_tracepoint(tp_category, tp_name); |
12337 | if (pfd < 0) { | |
52109584 AN |
12338 | pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n", |
12339 | prog->name, tp_category, tp_name, | |
be18010e | 12340 | libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 12341 | return libbpf_err_ptr(pfd); |
f6de59c1 | 12342 | } |
47faff37 | 12343 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
12344 | err = libbpf_get_error(link); |
12345 | if (err) { | |
f6de59c1 | 12346 | close(pfd); |
52109584 AN |
12347 | pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n", |
12348 | prog->name, tp_category, tp_name, | |
be18010e | 12349 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 12350 | return libbpf_err_ptr(err); |
f6de59c1 AN |
12351 | } |
12352 | return link; | |
12353 | } | |
12354 | ||
942025c9 | 12355 | struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog, |
47faff37 AN |
12356 | const char *tp_category, |
12357 | const char *tp_name) | |
12358 | { | |
12359 | return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL); | |
12360 | } | |
12361 | ||
4fa5bcfe | 12362 | static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 AN |
12363 | { |
12364 | char *sec_name, *tp_cat, *tp_name; | |
d7a18ea7 | 12365 | |
9af8efc4 AN |
12366 | *link = NULL; |
12367 | ||
12368 | /* no auto-attach for SEC("tp") or SEC("tracepoint") */ | |
12369 | if (strcmp(prog->sec_name, "tp") == 0 || strcmp(prog->sec_name, "tracepoint") == 0) | |
12370 | return 0; | |
12371 | ||
ac6a6586 AN |
12372 | sec_name = strdup(prog->sec_name); |
12373 | if (!sec_name) | |
12374 | return -ENOMEM; | |
12375 | ||
13d35a0c AN |
12376 | /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */ |
12377 | if (str_has_pfx(prog->sec_name, "tp/")) | |
12378 | tp_cat = sec_name + sizeof("tp/") - 1; | |
12379 | else | |
12380 | tp_cat = sec_name + sizeof("tracepoint/") - 1; | |
d7a18ea7 AN |
12381 | tp_name = strchr(tp_cat, '/'); |
12382 | if (!tp_name) { | |
e9fc3ce9 | 12383 | free(sec_name); |
4fa5bcfe | 12384 | return -EINVAL; |
d7a18ea7 AN |
12385 | } |
12386 | *tp_name = '\0'; | |
12387 | tp_name++; | |
12388 | ||
4fa5bcfe | 12389 | *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name); |
d7a18ea7 | 12390 | free(sec_name); |
4fa5bcfe | 12391 | return libbpf_get_error(*link); |
d7a18ea7 AN |
12392 | } |
12393 | ||
36ffb202 AN |
12394 | struct bpf_link * |
12395 | bpf_program__attach_raw_tracepoint_opts(const struct bpf_program *prog, | |
12396 | const char *tp_name, | |
12397 | struct bpf_raw_tracepoint_opts *opts) | |
84bf5e1f | 12398 | { |
36ffb202 | 12399 | LIBBPF_OPTS(bpf_raw_tp_opts, raw_opts); |
84bf5e1f | 12400 | char errmsg[STRERR_BUFSIZE]; |
c016b68e | 12401 | struct bpf_link *link; |
84bf5e1f AN |
12402 | int prog_fd, pfd; |
12403 | ||
36ffb202 AN |
12404 | if (!OPTS_VALID(opts, bpf_raw_tracepoint_opts)) |
12405 | return libbpf_err_ptr(-EINVAL); | |
12406 | ||
84bf5e1f AN |
12407 | prog_fd = bpf_program__fd(prog); |
12408 | if (prog_fd < 0) { | |
52109584 | 12409 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 12410 | return libbpf_err_ptr(-EINVAL); |
84bf5e1f AN |
12411 | } |
12412 | ||
d6958706 | 12413 | link = calloc(1, sizeof(*link)); |
84bf5e1f | 12414 | if (!link) |
e9fc3ce9 | 12415 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 12416 | link->detach = &bpf_link__detach_fd; |
84bf5e1f | 12417 | |
36ffb202 AN |
12418 | raw_opts.tp_name = tp_name; |
12419 | raw_opts.cookie = OPTS_GET(opts, cookie, 0); | |
12420 | pfd = bpf_raw_tracepoint_open_opts(prog_fd, &raw_opts); | |
84bf5e1f AN |
12421 | if (pfd < 0) { |
12422 | pfd = -errno; | |
12423 | free(link); | |
52109584 AN |
12424 | pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n", |
12425 | prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 12426 | return libbpf_err_ptr(pfd); |
84bf5e1f AN |
12427 | } |
12428 | link->fd = pfd; | |
c016b68e | 12429 | return link; |
84bf5e1f AN |
12430 | } |
12431 | ||
36ffb202 AN |
12432 | struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog, |
12433 | const char *tp_name) | |
12434 | { | |
12435 | return bpf_program__attach_raw_tracepoint_opts(prog, tp_name, NULL); | |
12436 | } | |
12437 | ||
4fa5bcfe | 12438 | static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 12439 | { |
ccaf12d6 | 12440 | static const char *const prefixes[] = { |
9af8efc4 AN |
12441 | "raw_tp", |
12442 | "raw_tracepoint", | |
12443 | "raw_tp.w", | |
12444 | "raw_tracepoint.w", | |
ccaf12d6 HT |
12445 | }; |
12446 | size_t i; | |
12447 | const char *tp_name = NULL; | |
13d35a0c | 12448 | |
9af8efc4 AN |
12449 | *link = NULL; |
12450 | ||
ccaf12d6 | 12451 | for (i = 0; i < ARRAY_SIZE(prefixes); i++) { |
9af8efc4 AN |
12452 | size_t pfx_len; |
12453 | ||
12454 | if (!str_has_pfx(prog->sec_name, prefixes[i])) | |
12455 | continue; | |
12456 | ||
12457 | pfx_len = strlen(prefixes[i]); | |
12458 | /* no auto-attach case of, e.g., SEC("raw_tp") */ | |
12459 | if (prog->sec_name[pfx_len] == '\0') | |
12460 | return 0; | |
12461 | ||
12462 | if (prog->sec_name[pfx_len] != '/') | |
12463 | continue; | |
12464 | ||
12465 | tp_name = prog->sec_name + pfx_len + 1; | |
12466 | break; | |
ccaf12d6 | 12467 | } |
9af8efc4 | 12468 | |
ccaf12d6 HT |
12469 | if (!tp_name) { |
12470 | pr_warn("prog '%s': invalid section name '%s'\n", | |
12471 | prog->name, prog->sec_name); | |
4fa5bcfe | 12472 | return -EINVAL; |
ccaf12d6 | 12473 | } |
d7a18ea7 | 12474 | |
4fa5bcfe | 12475 | *link = bpf_program__attach_raw_tracepoint(prog, tp_name); |
5fd2a60a | 12476 | return libbpf_get_error(*link); |
d7a18ea7 AN |
12477 | } |
12478 | ||
1e092a03 | 12479 | /* Common logic for all BPF program types that attach to a btf_id */ |
129b9c5e KFL |
12480 | static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog, |
12481 | const struct bpf_trace_opts *opts) | |
b8c54ea4 | 12482 | { |
129b9c5e | 12483 | LIBBPF_OPTS(bpf_link_create_opts, link_opts); |
b8c54ea4 | 12484 | char errmsg[STRERR_BUFSIZE]; |
c016b68e | 12485 | struct bpf_link *link; |
b8c54ea4 AS |
12486 | int prog_fd, pfd; |
12487 | ||
129b9c5e KFL |
12488 | if (!OPTS_VALID(opts, bpf_trace_opts)) |
12489 | return libbpf_err_ptr(-EINVAL); | |
12490 | ||
b8c54ea4 AS |
12491 | prog_fd = bpf_program__fd(prog); |
12492 | if (prog_fd < 0) { | |
52109584 | 12493 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 12494 | return libbpf_err_ptr(-EINVAL); |
b8c54ea4 AS |
12495 | } |
12496 | ||
d6958706 | 12497 | link = calloc(1, sizeof(*link)); |
b8c54ea4 | 12498 | if (!link) |
e9fc3ce9 | 12499 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 12500 | link->detach = &bpf_link__detach_fd; |
b8c54ea4 | 12501 | |
8462e0b4 | 12502 | /* libbpf is smart enough to redirect to BPF_RAW_TRACEPOINT_OPEN on old kernels */ |
129b9c5e KFL |
12503 | link_opts.tracing.cookie = OPTS_GET(opts, cookie, 0); |
12504 | pfd = bpf_link_create(prog_fd, 0, bpf_program__expected_attach_type(prog), &link_opts); | |
b8c54ea4 AS |
12505 | if (pfd < 0) { |
12506 | pfd = -errno; | |
12507 | free(link); | |
52109584 AN |
12508 | pr_warn("prog '%s': failed to attach: %s\n", |
12509 | prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 12510 | return libbpf_err_ptr(pfd); |
b8c54ea4 AS |
12511 | } |
12512 | link->fd = pfd; | |
003fed59 | 12513 | return link; |
b8c54ea4 AS |
12514 | } |
12515 | ||
942025c9 | 12516 | struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog) |
1e092a03 | 12517 | { |
129b9c5e KFL |
12518 | return bpf_program__attach_btf_id(prog, NULL); |
12519 | } | |
12520 | ||
12521 | struct bpf_link *bpf_program__attach_trace_opts(const struct bpf_program *prog, | |
12522 | const struct bpf_trace_opts *opts) | |
12523 | { | |
12524 | return bpf_program__attach_btf_id(prog, opts); | |
1e092a03 KS |
12525 | } |
12526 | ||
942025c9 | 12527 | struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog) |
1e092a03 | 12528 | { |
129b9c5e | 12529 | return bpf_program__attach_btf_id(prog, NULL); |
1e092a03 KS |
12530 | } |
12531 | ||
4fa5bcfe | 12532 | static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 12533 | { |
4fa5bcfe AN |
12534 | *link = bpf_program__attach_trace(prog); |
12535 | return libbpf_get_error(*link); | |
d7a18ea7 AN |
12536 | } |
12537 | ||
4fa5bcfe | 12538 | static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
1e092a03 | 12539 | { |
4fa5bcfe AN |
12540 | *link = bpf_program__attach_lsm(prog); |
12541 | return libbpf_get_error(*link); | |
1e092a03 KS |
12542 | } |
12543 | ||
d60d81ac | 12544 | static struct bpf_link * |
55cc3768 DB |
12545 | bpf_program_attach_fd(const struct bpf_program *prog, |
12546 | int target_fd, const char *target_name, | |
12547 | const struct bpf_link_create_opts *opts) | |
cc4f864b | 12548 | { |
cc4f864b AN |
12549 | enum bpf_attach_type attach_type; |
12550 | char errmsg[STRERR_BUFSIZE]; | |
12551 | struct bpf_link *link; | |
12552 | int prog_fd, link_fd; | |
12553 | ||
12554 | prog_fd = bpf_program__fd(prog); | |
12555 | if (prog_fd < 0) { | |
52109584 | 12556 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 12557 | return libbpf_err_ptr(-EINVAL); |
cc4f864b AN |
12558 | } |
12559 | ||
12560 | link = calloc(1, sizeof(*link)); | |
12561 | if (!link) | |
e9fc3ce9 | 12562 | return libbpf_err_ptr(-ENOMEM); |
cc4f864b AN |
12563 | link->detach = &bpf_link__detach_fd; |
12564 | ||
20eccf29 | 12565 | attach_type = bpf_program__expected_attach_type(prog); |
55cc3768 | 12566 | link_fd = bpf_link_create(prog_fd, target_fd, attach_type, opts); |
cc4f864b AN |
12567 | if (link_fd < 0) { |
12568 | link_fd = -errno; | |
12569 | free(link); | |
52109584 AN |
12570 | pr_warn("prog '%s': failed to attach to %s: %s\n", |
12571 | prog->name, target_name, | |
cc4f864b | 12572 | libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 12573 | return libbpf_err_ptr(link_fd); |
cc4f864b AN |
12574 | } |
12575 | link->fd = link_fd; | |
12576 | return link; | |
12577 | } | |
12578 | ||
d60d81ac | 12579 | struct bpf_link * |
942025c9 | 12580 | bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd) |
d60d81ac | 12581 | { |
55cc3768 | 12582 | return bpf_program_attach_fd(prog, cgroup_fd, "cgroup", NULL); |
d60d81ac JS |
12583 | } |
12584 | ||
12585 | struct bpf_link * | |
942025c9 | 12586 | bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd) |
d60d81ac | 12587 | { |
55cc3768 | 12588 | return bpf_program_attach_fd(prog, netns_fd, "netns", NULL); |
d60d81ac JS |
12589 | } |
12590 | ||
849989af YS |
12591 | struct bpf_link * |
12592 | bpf_program__attach_sockmap(const struct bpf_program *prog, int map_fd) | |
12593 | { | |
12594 | return bpf_program_attach_fd(prog, map_fd, "sockmap", NULL); | |
12595 | } | |
12596 | ||
942025c9 | 12597 | struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex) |
dc8698ca AN |
12598 | { |
12599 | /* target_fd/target_ifindex use the same field in LINK_CREATE */ | |
55cc3768 DB |
12600 | return bpf_program_attach_fd(prog, ifindex, "xdp", NULL); |
12601 | } | |
12602 | ||
12603 | struct bpf_link * | |
12604 | bpf_program__attach_tcx(const struct bpf_program *prog, int ifindex, | |
12605 | const struct bpf_tcx_opts *opts) | |
12606 | { | |
12607 | LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); | |
12608 | __u32 relative_id; | |
12609 | int relative_fd; | |
12610 | ||
12611 | if (!OPTS_VALID(opts, bpf_tcx_opts)) | |
12612 | return libbpf_err_ptr(-EINVAL); | |
12613 | ||
12614 | relative_id = OPTS_GET(opts, relative_id, 0); | |
12615 | relative_fd = OPTS_GET(opts, relative_fd, 0); | |
12616 | ||
12617 | /* validate we don't have unexpected combinations of non-zero fields */ | |
12618 | if (!ifindex) { | |
12619 | pr_warn("prog '%s': target netdevice ifindex cannot be zero\n", | |
12620 | prog->name); | |
12621 | return libbpf_err_ptr(-EINVAL); | |
12622 | } | |
12623 | if (relative_fd && relative_id) { | |
12624 | pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n", | |
12625 | prog->name); | |
12626 | return libbpf_err_ptr(-EINVAL); | |
12627 | } | |
12628 | ||
12629 | link_create_opts.tcx.expected_revision = OPTS_GET(opts, expected_revision, 0); | |
12630 | link_create_opts.tcx.relative_fd = relative_fd; | |
12631 | link_create_opts.tcx.relative_id = relative_id; | |
12632 | link_create_opts.flags = OPTS_GET(opts, flags, 0); | |
12633 | ||
12634 | /* target_fd/target_ifindex use the same field in LINK_CREATE */ | |
12635 | return bpf_program_attach_fd(prog, ifindex, "tcx", &link_create_opts); | |
a5359091 THJ |
12636 | } |
12637 | ||
05c31b4a DB |
12638 | struct bpf_link * |
12639 | bpf_program__attach_netkit(const struct bpf_program *prog, int ifindex, | |
12640 | const struct bpf_netkit_opts *opts) | |
12641 | { | |
12642 | LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); | |
12643 | __u32 relative_id; | |
12644 | int relative_fd; | |
12645 | ||
12646 | if (!OPTS_VALID(opts, bpf_netkit_opts)) | |
12647 | return libbpf_err_ptr(-EINVAL); | |
12648 | ||
12649 | relative_id = OPTS_GET(opts, relative_id, 0); | |
12650 | relative_fd = OPTS_GET(opts, relative_fd, 0); | |
12651 | ||
12652 | /* validate we don't have unexpected combinations of non-zero fields */ | |
12653 | if (!ifindex) { | |
12654 | pr_warn("prog '%s': target netdevice ifindex cannot be zero\n", | |
12655 | prog->name); | |
12656 | return libbpf_err_ptr(-EINVAL); | |
12657 | } | |
12658 | if (relative_fd && relative_id) { | |
12659 | pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n", | |
12660 | prog->name); | |
12661 | return libbpf_err_ptr(-EINVAL); | |
12662 | } | |
12663 | ||
12664 | link_create_opts.netkit.expected_revision = OPTS_GET(opts, expected_revision, 0); | |
12665 | link_create_opts.netkit.relative_fd = relative_fd; | |
12666 | link_create_opts.netkit.relative_id = relative_id; | |
12667 | link_create_opts.flags = OPTS_GET(opts, flags, 0); | |
12668 | ||
12669 | return bpf_program_attach_fd(prog, ifindex, "netkit", &link_create_opts); | |
12670 | } | |
12671 | ||
942025c9 | 12672 | struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog, |
a5359091 THJ |
12673 | int target_fd, |
12674 | const char *attach_func_name) | |
12675 | { | |
12676 | int btf_id; | |
12677 | ||
12678 | if (!!target_fd != !!attach_func_name) { | |
12679 | pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n", | |
12680 | prog->name); | |
e9fc3ce9 | 12681 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
12682 | } |
12683 | ||
12684 | if (prog->type != BPF_PROG_TYPE_EXT) { | |
12685 | pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace", | |
12686 | prog->name); | |
e9fc3ce9 | 12687 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
12688 | } |
12689 | ||
12690 | if (target_fd) { | |
55cc3768 DB |
12691 | LIBBPF_OPTS(bpf_link_create_opts, target_opts); |
12692 | ||
a5359091 THJ |
12693 | btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd); |
12694 | if (btf_id < 0) | |
e9fc3ce9 | 12695 | return libbpf_err_ptr(btf_id); |
a5359091 | 12696 | |
55cc3768 DB |
12697 | target_opts.target_btf_id = btf_id; |
12698 | ||
12699 | return bpf_program_attach_fd(prog, target_fd, "freplace", | |
12700 | &target_opts); | |
a5359091 THJ |
12701 | } else { |
12702 | /* no target, so use raw_tracepoint_open for compatibility | |
12703 | * with old kernels | |
12704 | */ | |
12705 | return bpf_program__attach_trace(prog); | |
12706 | } | |
dc8698ca AN |
12707 | } |
12708 | ||
c09add2f | 12709 | struct bpf_link * |
942025c9 | 12710 | bpf_program__attach_iter(const struct bpf_program *prog, |
c09add2f YS |
12711 | const struct bpf_iter_attach_opts *opts) |
12712 | { | |
cd31039a | 12713 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); |
c09add2f YS |
12714 | char errmsg[STRERR_BUFSIZE]; |
12715 | struct bpf_link *link; | |
12716 | int prog_fd, link_fd; | |
cd31039a | 12717 | __u32 target_fd = 0; |
c09add2f YS |
12718 | |
12719 | if (!OPTS_VALID(opts, bpf_iter_attach_opts)) | |
e9fc3ce9 | 12720 | return libbpf_err_ptr(-EINVAL); |
c09add2f | 12721 | |
74fc097d YS |
12722 | link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0); |
12723 | link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0); | |
cd31039a | 12724 | |
c09add2f YS |
12725 | prog_fd = bpf_program__fd(prog); |
12726 | if (prog_fd < 0) { | |
52109584 | 12727 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 12728 | return libbpf_err_ptr(-EINVAL); |
c09add2f YS |
12729 | } |
12730 | ||
12731 | link = calloc(1, sizeof(*link)); | |
12732 | if (!link) | |
e9fc3ce9 | 12733 | return libbpf_err_ptr(-ENOMEM); |
c09add2f YS |
12734 | link->detach = &bpf_link__detach_fd; |
12735 | ||
cd31039a YS |
12736 | link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER, |
12737 | &link_create_opts); | |
c09add2f YS |
12738 | if (link_fd < 0) { |
12739 | link_fd = -errno; | |
12740 | free(link); | |
52109584 AN |
12741 | pr_warn("prog '%s': failed to attach to iterator: %s\n", |
12742 | prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 12743 | return libbpf_err_ptr(link_fd); |
c09add2f YS |
12744 | } |
12745 | link->fd = link_fd; | |
12746 | return link; | |
12747 | } | |
12748 | ||
4fa5bcfe | 12749 | static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
e9fc3ce9 | 12750 | { |
4fa5bcfe AN |
12751 | *link = bpf_program__attach_iter(prog, NULL); |
12752 | return libbpf_get_error(*link); | |
e9fc3ce9 AN |
12753 | } |
12754 | ||
52364abb FW |
12755 | struct bpf_link *bpf_program__attach_netfilter(const struct bpf_program *prog, |
12756 | const struct bpf_netfilter_opts *opts) | |
12757 | { | |
12758 | LIBBPF_OPTS(bpf_link_create_opts, lopts); | |
12759 | struct bpf_link *link; | |
12760 | int prog_fd, link_fd; | |
12761 | ||
12762 | if (!OPTS_VALID(opts, bpf_netfilter_opts)) | |
12763 | return libbpf_err_ptr(-EINVAL); | |
12764 | ||
12765 | prog_fd = bpf_program__fd(prog); | |
12766 | if (prog_fd < 0) { | |
12767 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); | |
12768 | return libbpf_err_ptr(-EINVAL); | |
12769 | } | |
12770 | ||
12771 | link = calloc(1, sizeof(*link)); | |
12772 | if (!link) | |
12773 | return libbpf_err_ptr(-ENOMEM); | |
12774 | ||
12775 | link->detach = &bpf_link__detach_fd; | |
12776 | ||
12777 | lopts.netfilter.pf = OPTS_GET(opts, pf, 0); | |
12778 | lopts.netfilter.hooknum = OPTS_GET(opts, hooknum, 0); | |
12779 | lopts.netfilter.priority = OPTS_GET(opts, priority, 0); | |
12780 | lopts.netfilter.flags = OPTS_GET(opts, flags, 0); | |
12781 | ||
12782 | link_fd = bpf_link_create(prog_fd, 0, BPF_NETFILTER, &lopts); | |
12783 | if (link_fd < 0) { | |
12784 | char errmsg[STRERR_BUFSIZE]; | |
12785 | ||
12786 | link_fd = -errno; | |
12787 | free(link); | |
12788 | pr_warn("prog '%s': failed to attach to netfilter: %s\n", | |
12789 | prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); | |
12790 | return libbpf_err_ptr(link_fd); | |
12791 | } | |
12792 | link->fd = link_fd; | |
12793 | ||
12794 | return link; | |
12795 | } | |
12796 | ||
942025c9 | 12797 | struct bpf_link *bpf_program__attach(const struct bpf_program *prog) |
d7a18ea7 | 12798 | { |
4fa5bcfe AN |
12799 | struct bpf_link *link = NULL; |
12800 | int err; | |
12801 | ||
12802 | if (!prog->sec_def || !prog->sec_def->prog_attach_fn) | |
12803 | return libbpf_err_ptr(-EOPNOTSUPP); | |
d7a18ea7 | 12804 | |
7b30c296 MY |
12805 | if (bpf_program__fd(prog) < 0) { |
12806 | pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n", | |
12807 | prog->name); | |
12808 | return libbpf_err_ptr(-EINVAL); | |
12809 | } | |
12810 | ||
4fa5bcfe AN |
12811 | err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link); |
12812 | if (err) | |
12813 | return libbpf_err_ptr(err); | |
12814 | ||
12815 | /* When calling bpf_program__attach() explicitly, auto-attach support | |
12816 | * is expected to work, so NULL returned link is considered an error. | |
12817 | * This is different for skeleton's attach, see comment in | |
12818 | * bpf_object__attach_skeleton(). | |
12819 | */ | |
12820 | if (!link) | |
12821 | return libbpf_err_ptr(-EOPNOTSUPP); | |
12822 | ||
12823 | return link; | |
d7a18ea7 AN |
12824 | } |
12825 | ||
8d1608d7 KFL |
12826 | struct bpf_link_struct_ops { |
12827 | struct bpf_link link; | |
12828 | int map_fd; | |
12829 | }; | |
12830 | ||
590a0088 MKL |
12831 | static int bpf_link__detach_struct_ops(struct bpf_link *link) |
12832 | { | |
8d1608d7 | 12833 | struct bpf_link_struct_ops *st_link; |
590a0088 MKL |
12834 | __u32 zero = 0; |
12835 | ||
8d1608d7 | 12836 | st_link = container_of(link, struct bpf_link_struct_ops, link); |
590a0088 | 12837 | |
8d1608d7 KFL |
12838 | if (st_link->map_fd < 0) |
12839 | /* w/o a real link */ | |
12840 | return bpf_map_delete_elem(link->fd, &zero); | |
12841 | ||
12842 | return close(link->fd); | |
590a0088 MKL |
12843 | } |
12844 | ||
942025c9 | 12845 | struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map) |
590a0088 | 12846 | { |
8d1608d7 KFL |
12847 | struct bpf_link_struct_ops *link; |
12848 | __u32 zero = 0; | |
12849 | int err, fd; | |
590a0088 | 12850 | |
7b30c296 | 12851 | if (!bpf_map__is_struct_ops(map)) |
e9fc3ce9 | 12852 | return libbpf_err_ptr(-EINVAL); |
590a0088 | 12853 | |
7b30c296 MY |
12854 | if (map->fd < 0) { |
12855 | pr_warn("map '%s': can't attach BPF map without FD (was it created?)\n", map->name); | |
12856 | return libbpf_err_ptr(-EINVAL); | |
12857 | } | |
12858 | ||
590a0088 MKL |
12859 | link = calloc(1, sizeof(*link)); |
12860 | if (!link) | |
e9fc3ce9 | 12861 | return libbpf_err_ptr(-EINVAL); |
590a0088 | 12862 | |
8d1608d7 KFL |
12863 | /* kern_vdata should be prepared during the loading phase. */ |
12864 | err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0); | |
12865 | /* It can be EBUSY if the map has been used to create or | |
12866 | * update a link before. We don't allow updating the value of | |
12867 | * a struct_ops once it is set. That ensures that the value | |
12868 | * never changed. So, it is safe to skip EBUSY. | |
12869 | */ | |
12870 | if (err && (!(map->def.map_flags & BPF_F_LINK) || err != -EBUSY)) { | |
12871 | free(link); | |
12872 | return libbpf_err_ptr(err); | |
12873 | } | |
590a0088 | 12874 | |
8d1608d7 | 12875 | link->link.detach = bpf_link__detach_struct_ops; |
590a0088 | 12876 | |
8d1608d7 KFL |
12877 | if (!(map->def.map_flags & BPF_F_LINK)) { |
12878 | /* w/o a real link */ | |
12879 | link->link.fd = map->fd; | |
12880 | link->map_fd = -1; | |
12881 | return &link->link; | |
590a0088 MKL |
12882 | } |
12883 | ||
8d1608d7 KFL |
12884 | fd = bpf_link_create(map->fd, 0, BPF_STRUCT_OPS, NULL); |
12885 | if (fd < 0) { | |
590a0088 | 12886 | free(link); |
8d1608d7 | 12887 | return libbpf_err_ptr(fd); |
590a0088 MKL |
12888 | } |
12889 | ||
8d1608d7 KFL |
12890 | link->link.fd = fd; |
12891 | link->map_fd = map->fd; | |
590a0088 | 12892 | |
8d1608d7 | 12893 | return &link->link; |
590a0088 MKL |
12894 | } |
12895 | ||
912dd4b0 KFL |
12896 | /* |
12897 | * Swap the back struct_ops of a link with a new struct_ops map. | |
12898 | */ | |
12899 | int bpf_link__update_map(struct bpf_link *link, const struct bpf_map *map) | |
12900 | { | |
12901 | struct bpf_link_struct_ops *st_ops_link; | |
12902 | __u32 zero = 0; | |
12903 | int err; | |
12904 | ||
7b30c296 MY |
12905 | if (!bpf_map__is_struct_ops(map)) |
12906 | return -EINVAL; | |
12907 | ||
12908 | if (map->fd < 0) { | |
12909 | pr_warn("map '%s': can't use BPF map without FD (was it created?)\n", map->name); | |
912dd4b0 | 12910 | return -EINVAL; |
7b30c296 | 12911 | } |
912dd4b0 KFL |
12912 | |
12913 | st_ops_link = container_of(link, struct bpf_link_struct_ops, link); | |
12914 | /* Ensure the type of a link is correct */ | |
12915 | if (st_ops_link->map_fd < 0) | |
12916 | return -EINVAL; | |
12917 | ||
12918 | err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0); | |
12919 | /* It can be EBUSY if the map has been used to create or | |
12920 | * update a link before. We don't allow updating the value of | |
12921 | * a struct_ops once it is set. That ensures that the value | |
12922 | * never changed. So, it is safe to skip EBUSY. | |
12923 | */ | |
12924 | if (err && err != -EBUSY) | |
12925 | return err; | |
12926 | ||
12927 | err = bpf_link_update(link->fd, map->fd, NULL); | |
12928 | if (err < 0) | |
12929 | return err; | |
12930 | ||
12931 | st_ops_link->map_fd = map->fd; | |
12932 | ||
12933 | return 0; | |
12934 | } | |
12935 | ||
22dd7a58 AN |
12936 | typedef enum bpf_perf_event_ret (*bpf_perf_event_print_t)(struct perf_event_header *hdr, |
12937 | void *private_data); | |
12938 | ||
7218c28c CL |
12939 | static enum bpf_perf_event_ret |
12940 | perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size, | |
12941 | void **copy_mem, size_t *copy_size, | |
12942 | bpf_perf_event_print_t fn, void *private_data) | |
d0cabbb0 | 12943 | { |
3dca2115 | 12944 | struct perf_event_mmap_page *header = mmap_mem; |
a64af0ef | 12945 | __u64 data_head = ring_buffer_read_head(header); |
d0cabbb0 | 12946 | __u64 data_tail = header->data_tail; |
3dca2115 DB |
12947 | void *base = ((__u8 *)header) + page_size; |
12948 | int ret = LIBBPF_PERF_EVENT_CONT; | |
12949 | struct perf_event_header *ehdr; | |
12950 | size_t ehdr_size; | |
12951 | ||
12952 | while (data_head != data_tail) { | |
12953 | ehdr = base + (data_tail & (mmap_size - 1)); | |
12954 | ehdr_size = ehdr->size; | |
12955 | ||
12956 | if (((void *)ehdr) + ehdr_size > base + mmap_size) { | |
12957 | void *copy_start = ehdr; | |
12958 | size_t len_first = base + mmap_size - copy_start; | |
12959 | size_t len_secnd = ehdr_size - len_first; | |
12960 | ||
12961 | if (*copy_size < ehdr_size) { | |
12962 | free(*copy_mem); | |
12963 | *copy_mem = malloc(ehdr_size); | |
12964 | if (!*copy_mem) { | |
12965 | *copy_size = 0; | |
d0cabbb0 JK |
12966 | ret = LIBBPF_PERF_EVENT_ERROR; |
12967 | break; | |
12968 | } | |
3dca2115 | 12969 | *copy_size = ehdr_size; |
d0cabbb0 JK |
12970 | } |
12971 | ||
3dca2115 DB |
12972 | memcpy(*copy_mem, copy_start, len_first); |
12973 | memcpy(*copy_mem + len_first, base, len_secnd); | |
12974 | ehdr = *copy_mem; | |
d0cabbb0 JK |
12975 | } |
12976 | ||
3dca2115 DB |
12977 | ret = fn(ehdr, private_data); |
12978 | data_tail += ehdr_size; | |
d0cabbb0 JK |
12979 | if (ret != LIBBPF_PERF_EVENT_CONT) |
12980 | break; | |
d0cabbb0 JK |
12981 | } |
12982 | ||
a64af0ef | 12983 | ring_buffer_write_tail(header, data_tail); |
e9fc3ce9 | 12984 | return libbpf_err(ret); |
d0cabbb0 | 12985 | } |
34be1646 | 12986 | |
fb84b822 AN |
12987 | struct perf_buffer; |
12988 | ||
12989 | struct perf_buffer_params { | |
12990 | struct perf_event_attr *attr; | |
12991 | /* if event_cb is specified, it takes precendence */ | |
12992 | perf_buffer_event_fn event_cb; | |
12993 | /* sample_cb and lost_cb are higher-level common-case callbacks */ | |
12994 | perf_buffer_sample_fn sample_cb; | |
12995 | perf_buffer_lost_fn lost_cb; | |
12996 | void *ctx; | |
12997 | int cpu_cnt; | |
12998 | int *cpus; | |
12999 | int *map_keys; | |
13000 | }; | |
13001 | ||
13002 | struct perf_cpu_buf { | |
13003 | struct perf_buffer *pb; | |
13004 | void *base; /* mmap()'ed memory */ | |
13005 | void *buf; /* for reconstructing segmented data */ | |
13006 | size_t buf_size; | |
13007 | int fd; | |
13008 | int cpu; | |
13009 | int map_key; | |
13010 | }; | |
13011 | ||
13012 | struct perf_buffer { | |
13013 | perf_buffer_event_fn event_cb; | |
13014 | perf_buffer_sample_fn sample_cb; | |
13015 | perf_buffer_lost_fn lost_cb; | |
13016 | void *ctx; /* passed into callbacks */ | |
13017 | ||
13018 | size_t page_size; | |
13019 | size_t mmap_size; | |
13020 | struct perf_cpu_buf **cpu_bufs; | |
13021 | struct epoll_event *events; | |
783b8f01 | 13022 | int cpu_cnt; /* number of allocated CPU buffers */ |
fb84b822 AN |
13023 | int epoll_fd; /* perf event FD */ |
13024 | int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */ | |
13025 | }; | |
13026 | ||
13027 | static void perf_buffer__free_cpu_buf(struct perf_buffer *pb, | |
13028 | struct perf_cpu_buf *cpu_buf) | |
13029 | { | |
13030 | if (!cpu_buf) | |
13031 | return; | |
13032 | if (cpu_buf->base && | |
13033 | munmap(cpu_buf->base, pb->mmap_size + pb->page_size)) | |
be18010e | 13034 | pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu); |
fb84b822 AN |
13035 | if (cpu_buf->fd >= 0) { |
13036 | ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0); | |
13037 | close(cpu_buf->fd); | |
13038 | } | |
13039 | free(cpu_buf->buf); | |
13040 | free(cpu_buf); | |
13041 | } | |
13042 | ||
13043 | void perf_buffer__free(struct perf_buffer *pb) | |
13044 | { | |
13045 | int i; | |
13046 | ||
50450fc7 | 13047 | if (IS_ERR_OR_NULL(pb)) |
fb84b822 AN |
13048 | return; |
13049 | if (pb->cpu_bufs) { | |
601b05ca | 13050 | for (i = 0; i < pb->cpu_cnt; i++) { |
fb84b822 AN |
13051 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; |
13052 | ||
601b05ca EC |
13053 | if (!cpu_buf) |
13054 | continue; | |
13055 | ||
fb84b822 AN |
13056 | bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key); |
13057 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
13058 | } | |
13059 | free(pb->cpu_bufs); | |
13060 | } | |
13061 | if (pb->epoll_fd >= 0) | |
13062 | close(pb->epoll_fd); | |
13063 | free(pb->events); | |
13064 | free(pb); | |
13065 | } | |
13066 | ||
13067 | static struct perf_cpu_buf * | |
13068 | perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr, | |
13069 | int cpu, int map_key) | |
13070 | { | |
13071 | struct perf_cpu_buf *cpu_buf; | |
13072 | char msg[STRERR_BUFSIZE]; | |
13073 | int err; | |
13074 | ||
13075 | cpu_buf = calloc(1, sizeof(*cpu_buf)); | |
13076 | if (!cpu_buf) | |
13077 | return ERR_PTR(-ENOMEM); | |
13078 | ||
13079 | cpu_buf->pb = pb; | |
13080 | cpu_buf->cpu = cpu; | |
13081 | cpu_buf->map_key = map_key; | |
13082 | ||
13083 | cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu, | |
13084 | -1, PERF_FLAG_FD_CLOEXEC); | |
13085 | if (cpu_buf->fd < 0) { | |
13086 | err = -errno; | |
be18010e KW |
13087 | pr_warn("failed to open perf buffer event on cpu #%d: %s\n", |
13088 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13089 | goto error; |
13090 | } | |
13091 | ||
13092 | cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size, | |
13093 | PROT_READ | PROT_WRITE, MAP_SHARED, | |
13094 | cpu_buf->fd, 0); | |
13095 | if (cpu_buf->base == MAP_FAILED) { | |
13096 | cpu_buf->base = NULL; | |
13097 | err = -errno; | |
be18010e KW |
13098 | pr_warn("failed to mmap perf buffer on cpu #%d: %s\n", |
13099 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13100 | goto error; |
13101 | } | |
13102 | ||
13103 | if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
13104 | err = -errno; | |
be18010e KW |
13105 | pr_warn("failed to enable perf buffer event on cpu #%d: %s\n", |
13106 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13107 | goto error; |
13108 | } | |
13109 | ||
13110 | return cpu_buf; | |
13111 | ||
13112 | error: | |
13113 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
13114 | return (struct perf_cpu_buf *)ERR_PTR(err); | |
13115 | } | |
13116 | ||
13117 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
13118 | struct perf_buffer_params *p); | |
13119 | ||
22dd7a58 AN |
13120 | struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt, |
13121 | perf_buffer_sample_fn sample_cb, | |
13122 | perf_buffer_lost_fn lost_cb, | |
13123 | void *ctx, | |
13124 | const struct perf_buffer_opts *opts) | |
fb84b822 | 13125 | { |
813847a3 | 13126 | const size_t attr_sz = sizeof(struct perf_event_attr); |
fb84b822 | 13127 | struct perf_buffer_params p = {}; |
813847a3 | 13128 | struct perf_event_attr attr; |
ab8684b8 | 13129 | __u32 sample_period; |
41788934 AN |
13130 | |
13131 | if (!OPTS_VALID(opts, perf_buffer_opts)) | |
13132 | return libbpf_err_ptr(-EINVAL); | |
4be6e05c | 13133 | |
ab8684b8 JD |
13134 | sample_period = OPTS_GET(opts, sample_period, 1); |
13135 | if (!sample_period) | |
13136 | sample_period = 1; | |
13137 | ||
813847a3 AN |
13138 | memset(&attr, 0, attr_sz); |
13139 | attr.size = attr_sz; | |
65bb2e0f | 13140 | attr.config = PERF_COUNT_SW_BPF_OUTPUT; |
4be6e05c ACM |
13141 | attr.type = PERF_TYPE_SOFTWARE; |
13142 | attr.sample_type = PERF_SAMPLE_RAW; | |
ab8684b8 JD |
13143 | attr.sample_period = sample_period; |
13144 | attr.wakeup_events = sample_period; | |
fb84b822 AN |
13145 | |
13146 | p.attr = &attr; | |
41788934 AN |
13147 | p.sample_cb = sample_cb; |
13148 | p.lost_cb = lost_cb; | |
13149 | p.ctx = ctx; | |
fb84b822 | 13150 | |
e9fc3ce9 | 13151 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
13152 | } |
13153 | ||
22dd7a58 AN |
13154 | struct perf_buffer *perf_buffer__new_raw(int map_fd, size_t page_cnt, |
13155 | struct perf_event_attr *attr, | |
13156 | perf_buffer_event_fn event_cb, void *ctx, | |
13157 | const struct perf_buffer_raw_opts *opts) | |
fb84b822 AN |
13158 | { |
13159 | struct perf_buffer_params p = {}; | |
13160 | ||
41332d6e | 13161 | if (!attr) |
41788934 AN |
13162 | return libbpf_err_ptr(-EINVAL); |
13163 | ||
13164 | if (!OPTS_VALID(opts, perf_buffer_raw_opts)) | |
13165 | return libbpf_err_ptr(-EINVAL); | |
13166 | ||
13167 | p.attr = attr; | |
13168 | p.event_cb = event_cb; | |
13169 | p.ctx = ctx; | |
13170 | p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0); | |
13171 | p.cpus = OPTS_GET(opts, cpus, NULL); | |
13172 | p.map_keys = OPTS_GET(opts, map_keys, NULL); | |
fb84b822 | 13173 | |
e9fc3ce9 | 13174 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
13175 | } |
13176 | ||
13177 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
13178 | struct perf_buffer_params *p) | |
13179 | { | |
783b8f01 | 13180 | const char *online_cpus_file = "/sys/devices/system/cpu/online"; |
0e289487 | 13181 | struct bpf_map_info map; |
fb84b822 AN |
13182 | char msg[STRERR_BUFSIZE]; |
13183 | struct perf_buffer *pb; | |
783b8f01 | 13184 | bool *online = NULL; |
fb84b822 | 13185 | __u32 map_info_len; |
783b8f01 | 13186 | int err, i, j, n; |
fb84b822 | 13187 | |
41332d6e | 13188 | if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) { |
be18010e KW |
13189 | pr_warn("page count should be power of two, but is %zu\n", |
13190 | page_cnt); | |
fb84b822 AN |
13191 | return ERR_PTR(-EINVAL); |
13192 | } | |
13193 | ||
0e289487 AN |
13194 | /* best-effort sanity checks */ |
13195 | memset(&map, 0, sizeof(map)); | |
fb84b822 | 13196 | map_info_len = sizeof(map); |
629dfc66 | 13197 | err = bpf_map_get_info_by_fd(map_fd, &map, &map_info_len); |
fb84b822 AN |
13198 | if (err) { |
13199 | err = -errno; | |
0e289487 AN |
13200 | /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return |
13201 | * -EBADFD, -EFAULT, or -E2BIG on real error | |
13202 | */ | |
13203 | if (err != -EINVAL) { | |
13204 | pr_warn("failed to get map info for map FD %d: %s\n", | |
13205 | map_fd, libbpf_strerror_r(err, msg, sizeof(msg))); | |
13206 | return ERR_PTR(err); | |
13207 | } | |
13208 | pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n", | |
13209 | map_fd); | |
13210 | } else { | |
13211 | if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) { | |
13212 | pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n", | |
13213 | map.name); | |
13214 | return ERR_PTR(-EINVAL); | |
13215 | } | |
fb84b822 AN |
13216 | } |
13217 | ||
13218 | pb = calloc(1, sizeof(*pb)); | |
13219 | if (!pb) | |
13220 | return ERR_PTR(-ENOMEM); | |
13221 | ||
13222 | pb->event_cb = p->event_cb; | |
13223 | pb->sample_cb = p->sample_cb; | |
13224 | pb->lost_cb = p->lost_cb; | |
13225 | pb->ctx = p->ctx; | |
13226 | ||
13227 | pb->page_size = getpagesize(); | |
13228 | pb->mmap_size = pb->page_size * page_cnt; | |
13229 | pb->map_fd = map_fd; | |
13230 | ||
13231 | pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
13232 | if (pb->epoll_fd < 0) { | |
13233 | err = -errno; | |
be18010e KW |
13234 | pr_warn("failed to create epoll instance: %s\n", |
13235 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13236 | goto error; |
13237 | } | |
13238 | ||
13239 | if (p->cpu_cnt > 0) { | |
13240 | pb->cpu_cnt = p->cpu_cnt; | |
13241 | } else { | |
13242 | pb->cpu_cnt = libbpf_num_possible_cpus(); | |
13243 | if (pb->cpu_cnt < 0) { | |
13244 | err = pb->cpu_cnt; | |
13245 | goto error; | |
13246 | } | |
0e289487 | 13247 | if (map.max_entries && map.max_entries < pb->cpu_cnt) |
fb84b822 AN |
13248 | pb->cpu_cnt = map.max_entries; |
13249 | } | |
13250 | ||
13251 | pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events)); | |
13252 | if (!pb->events) { | |
13253 | err = -ENOMEM; | |
be18010e | 13254 | pr_warn("failed to allocate events: out of memory\n"); |
fb84b822 AN |
13255 | goto error; |
13256 | } | |
13257 | pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs)); | |
13258 | if (!pb->cpu_bufs) { | |
13259 | err = -ENOMEM; | |
be18010e | 13260 | pr_warn("failed to allocate buffers: out of memory\n"); |
fb84b822 AN |
13261 | goto error; |
13262 | } | |
13263 | ||
783b8f01 AN |
13264 | err = parse_cpu_mask_file(online_cpus_file, &online, &n); |
13265 | if (err) { | |
13266 | pr_warn("failed to get online CPU mask: %d\n", err); | |
13267 | goto error; | |
13268 | } | |
13269 | ||
13270 | for (i = 0, j = 0; i < pb->cpu_cnt; i++) { | |
fb84b822 AN |
13271 | struct perf_cpu_buf *cpu_buf; |
13272 | int cpu, map_key; | |
13273 | ||
13274 | cpu = p->cpu_cnt > 0 ? p->cpus[i] : i; | |
13275 | map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i; | |
13276 | ||
783b8f01 AN |
13277 | /* in case user didn't explicitly requested particular CPUs to |
13278 | * be attached to, skip offline/not present CPUs | |
13279 | */ | |
13280 | if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu])) | |
13281 | continue; | |
13282 | ||
fb84b822 AN |
13283 | cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key); |
13284 | if (IS_ERR(cpu_buf)) { | |
13285 | err = PTR_ERR(cpu_buf); | |
13286 | goto error; | |
13287 | } | |
13288 | ||
783b8f01 | 13289 | pb->cpu_bufs[j] = cpu_buf; |
fb84b822 AN |
13290 | |
13291 | err = bpf_map_update_elem(pb->map_fd, &map_key, | |
13292 | &cpu_buf->fd, 0); | |
13293 | if (err) { | |
13294 | err = -errno; | |
be18010e KW |
13295 | pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n", |
13296 | cpu, map_key, cpu_buf->fd, | |
13297 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13298 | goto error; |
13299 | } | |
13300 | ||
783b8f01 AN |
13301 | pb->events[j].events = EPOLLIN; |
13302 | pb->events[j].data.ptr = cpu_buf; | |
fb84b822 | 13303 | if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd, |
783b8f01 | 13304 | &pb->events[j]) < 0) { |
fb84b822 | 13305 | err = -errno; |
be18010e KW |
13306 | pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n", |
13307 | cpu, cpu_buf->fd, | |
13308 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
13309 | goto error; |
13310 | } | |
783b8f01 | 13311 | j++; |
fb84b822 | 13312 | } |
783b8f01 AN |
13313 | pb->cpu_cnt = j; |
13314 | free(online); | |
fb84b822 AN |
13315 | |
13316 | return pb; | |
13317 | ||
13318 | error: | |
783b8f01 | 13319 | free(online); |
fb84b822 AN |
13320 | if (pb) |
13321 | perf_buffer__free(pb); | |
13322 | return ERR_PTR(err); | |
13323 | } | |
13324 | ||
13325 | struct perf_sample_raw { | |
13326 | struct perf_event_header header; | |
13327 | uint32_t size; | |
385bbf7b | 13328 | char data[]; |
fb84b822 AN |
13329 | }; |
13330 | ||
13331 | struct perf_sample_lost { | |
13332 | struct perf_event_header header; | |
13333 | uint64_t id; | |
13334 | uint64_t lost; | |
13335 | uint64_t sample_id; | |
13336 | }; | |
13337 | ||
13338 | static enum bpf_perf_event_ret | |
13339 | perf_buffer__process_record(struct perf_event_header *e, void *ctx) | |
13340 | { | |
13341 | struct perf_cpu_buf *cpu_buf = ctx; | |
13342 | struct perf_buffer *pb = cpu_buf->pb; | |
13343 | void *data = e; | |
13344 | ||
13345 | /* user wants full control over parsing perf event */ | |
13346 | if (pb->event_cb) | |
13347 | return pb->event_cb(pb->ctx, cpu_buf->cpu, e); | |
13348 | ||
13349 | switch (e->type) { | |
13350 | case PERF_RECORD_SAMPLE: { | |
13351 | struct perf_sample_raw *s = data; | |
13352 | ||
13353 | if (pb->sample_cb) | |
13354 | pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size); | |
13355 | break; | |
13356 | } | |
13357 | case PERF_RECORD_LOST: { | |
13358 | struct perf_sample_lost *s = data; | |
13359 | ||
13360 | if (pb->lost_cb) | |
13361 | pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost); | |
13362 | break; | |
13363 | } | |
13364 | default: | |
be18010e | 13365 | pr_warn("unknown perf sample type %d\n", e->type); |
fb84b822 AN |
13366 | return LIBBPF_PERF_EVENT_ERROR; |
13367 | } | |
13368 | return LIBBPF_PERF_EVENT_CONT; | |
13369 | } | |
13370 | ||
13371 | static int perf_buffer__process_records(struct perf_buffer *pb, | |
13372 | struct perf_cpu_buf *cpu_buf) | |
13373 | { | |
13374 | enum bpf_perf_event_ret ret; | |
13375 | ||
7218c28c CL |
13376 | ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size, |
13377 | pb->page_size, &cpu_buf->buf, | |
13378 | &cpu_buf->buf_size, | |
13379 | perf_buffer__process_record, cpu_buf); | |
fb84b822 AN |
13380 | if (ret != LIBBPF_PERF_EVENT_CONT) |
13381 | return ret; | |
13382 | return 0; | |
13383 | } | |
13384 | ||
dca5612f AN |
13385 | int perf_buffer__epoll_fd(const struct perf_buffer *pb) |
13386 | { | |
13387 | return pb->epoll_fd; | |
13388 | } | |
13389 | ||
fb84b822 AN |
13390 | int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms) |
13391 | { | |
13392 | int i, cnt, err; | |
13393 | ||
13394 | cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms); | |
e9fc3ce9 | 13395 | if (cnt < 0) |
af0efa05 | 13396 | return -errno; |
e9fc3ce9 | 13397 | |
fb84b822 AN |
13398 | for (i = 0; i < cnt; i++) { |
13399 | struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr; | |
13400 | ||
13401 | err = perf_buffer__process_records(pb, cpu_buf); | |
13402 | if (err) { | |
be18010e | 13403 | pr_warn("error while processing records: %d\n", err); |
e9fc3ce9 | 13404 | return libbpf_err(err); |
fb84b822 AN |
13405 | } |
13406 | } | |
e9fc3ce9 | 13407 | return cnt; |
fb84b822 AN |
13408 | } |
13409 | ||
dca5612f AN |
13410 | /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer |
13411 | * manager. | |
13412 | */ | |
13413 | size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb) | |
13414 | { | |
13415 | return pb->cpu_cnt; | |
13416 | } | |
13417 | ||
13418 | /* | |
13419 | * Return perf_event FD of a ring buffer in *buf_idx* slot of | |
13420 | * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using | |
13421 | * select()/poll()/epoll() Linux syscalls. | |
13422 | */ | |
13423 | int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx) | |
13424 | { | |
13425 | struct perf_cpu_buf *cpu_buf; | |
13426 | ||
13427 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 13428 | return libbpf_err(-EINVAL); |
dca5612f AN |
13429 | |
13430 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
13431 | if (!cpu_buf) | |
e9fc3ce9 | 13432 | return libbpf_err(-ENOENT); |
dca5612f AN |
13433 | |
13434 | return cpu_buf->fd; | |
13435 | } | |
13436 | ||
9ff5efde JD |
13437 | int perf_buffer__buffer(struct perf_buffer *pb, int buf_idx, void **buf, size_t *buf_size) |
13438 | { | |
13439 | struct perf_cpu_buf *cpu_buf; | |
13440 | ||
13441 | if (buf_idx >= pb->cpu_cnt) | |
13442 | return libbpf_err(-EINVAL); | |
13443 | ||
13444 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
13445 | if (!cpu_buf) | |
13446 | return libbpf_err(-ENOENT); | |
13447 | ||
13448 | *buf = cpu_buf->base; | |
13449 | *buf_size = pb->mmap_size; | |
13450 | return 0; | |
13451 | } | |
13452 | ||
dca5612f AN |
13453 | /* |
13454 | * Consume data from perf ring buffer corresponding to slot *buf_idx* in | |
13455 | * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to | |
13456 | * consume, do nothing and return success. | |
13457 | * Returns: | |
13458 | * - 0 on success; | |
13459 | * - <0 on failure. | |
13460 | */ | |
13461 | int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx) | |
13462 | { | |
13463 | struct perf_cpu_buf *cpu_buf; | |
13464 | ||
13465 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 13466 | return libbpf_err(-EINVAL); |
dca5612f AN |
13467 | |
13468 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
13469 | if (!cpu_buf) | |
e9fc3ce9 | 13470 | return libbpf_err(-ENOENT); |
dca5612f AN |
13471 | |
13472 | return perf_buffer__process_records(pb, cpu_buf); | |
13473 | } | |
13474 | ||
272d51af EC |
13475 | int perf_buffer__consume(struct perf_buffer *pb) |
13476 | { | |
13477 | int i, err; | |
13478 | ||
13479 | for (i = 0; i < pb->cpu_cnt; i++) { | |
13480 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; | |
13481 | ||
13482 | if (!cpu_buf) | |
13483 | continue; | |
13484 | ||
13485 | err = perf_buffer__process_records(pb, cpu_buf); | |
13486 | if (err) { | |
dca5612f | 13487 | pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err); |
e9fc3ce9 | 13488 | return libbpf_err(err); |
272d51af EC |
13489 | } |
13490 | } | |
13491 | return 0; | |
13492 | } | |
13493 | ||
ff26ce5c EC |
13494 | int bpf_program__set_attach_target(struct bpf_program *prog, |
13495 | int attach_prog_fd, | |
13496 | const char *attach_func_name) | |
13497 | { | |
fe62de31 | 13498 | int btf_obj_fd = 0, btf_id = 0, err; |
ff26ce5c | 13499 | |
2d5ec1c6 | 13500 | if (!prog || attach_prog_fd < 0) |
e9fc3ce9 | 13501 | return libbpf_err(-EINVAL); |
ff26ce5c | 13502 | |
fe62de31 | 13503 | if (prog->obj->loaded) |
e9fc3ce9 | 13504 | return libbpf_err(-EINVAL); |
fe62de31 | 13505 | |
2d5ec1c6 AN |
13506 | if (attach_prog_fd && !attach_func_name) { |
13507 | /* remember attach_prog_fd and let bpf_program__load() find | |
13508 | * BTF ID during the program load | |
13509 | */ | |
13510 | prog->attach_prog_fd = attach_prog_fd; | |
13511 | return 0; | |
13512 | } | |
13513 | ||
fe62de31 | 13514 | if (attach_prog_fd) { |
ff26ce5c EC |
13515 | btf_id = libbpf_find_prog_btf_id(attach_func_name, |
13516 | attach_prog_fd); | |
fe62de31 | 13517 | if (btf_id < 0) |
e9fc3ce9 | 13518 | return libbpf_err(btf_id); |
fe62de31 | 13519 | } else { |
2d5ec1c6 AN |
13520 | if (!attach_func_name) |
13521 | return libbpf_err(-EINVAL); | |
13522 | ||
fe62de31 AN |
13523 | /* load btf_vmlinux, if not yet */ |
13524 | err = bpf_object__load_vmlinux_btf(prog->obj, true); | |
13525 | if (err) | |
e9fc3ce9 | 13526 | return libbpf_err(err); |
fe62de31 AN |
13527 | err = find_kernel_btf_id(prog->obj, attach_func_name, |
13528 | prog->expected_attach_type, | |
13529 | &btf_obj_fd, &btf_id); | |
13530 | if (err) | |
e9fc3ce9 | 13531 | return libbpf_err(err); |
fe62de31 | 13532 | } |
ff26ce5c EC |
13533 | |
13534 | prog->attach_btf_id = btf_id; | |
fe62de31 | 13535 | prog->attach_btf_obj_fd = btf_obj_fd; |
ff26ce5c EC |
13536 | prog->attach_prog_fd = attach_prog_fd; |
13537 | return 0; | |
13538 | } | |
13539 | ||
6803ee25 | 13540 | int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz) |
6446b315 | 13541 | { |
6803ee25 AN |
13542 | int err = 0, n, len, start, end = -1; |
13543 | bool *tmp; | |
6446b315 | 13544 | |
6803ee25 AN |
13545 | *mask = NULL; |
13546 | *mask_sz = 0; | |
13547 | ||
13548 | /* Each sub string separated by ',' has format \d+-\d+ or \d+ */ | |
13549 | while (*s) { | |
13550 | if (*s == ',' || *s == '\n') { | |
13551 | s++; | |
13552 | continue; | |
13553 | } | |
13554 | n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len); | |
13555 | if (n <= 0 || n > 2) { | |
13556 | pr_warn("Failed to get CPU range %s: %d\n", s, n); | |
13557 | err = -EINVAL; | |
13558 | goto cleanup; | |
13559 | } else if (n == 1) { | |
13560 | end = start; | |
13561 | } | |
13562 | if (start < 0 || start > end) { | |
13563 | pr_warn("Invalid CPU range [%d,%d] in %s\n", | |
13564 | start, end, s); | |
13565 | err = -EINVAL; | |
13566 | goto cleanup; | |
13567 | } | |
13568 | tmp = realloc(*mask, end + 1); | |
13569 | if (!tmp) { | |
13570 | err = -ENOMEM; | |
13571 | goto cleanup; | |
13572 | } | |
13573 | *mask = tmp; | |
13574 | memset(tmp + *mask_sz, 0, start - *mask_sz); | |
13575 | memset(tmp + start, 1, end - start + 1); | |
13576 | *mask_sz = end + 1; | |
13577 | s += len; | |
13578 | } | |
13579 | if (!*mask_sz) { | |
13580 | pr_warn("Empty CPU range\n"); | |
13581 | return -EINVAL; | |
13582 | } | |
13583 | return 0; | |
13584 | cleanup: | |
13585 | free(*mask); | |
13586 | *mask = NULL; | |
13587 | return err; | |
13588 | } | |
13589 | ||
13590 | int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz) | |
13591 | { | |
13592 | int fd, err = 0, len; | |
13593 | char buf[128]; | |
6446b315 | 13594 | |
92274e24 | 13595 | fd = open(fcpu, O_RDONLY | O_CLOEXEC); |
6446b315 | 13596 | if (fd < 0) { |
6803ee25 AN |
13597 | err = -errno; |
13598 | pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err); | |
13599 | return err; | |
6446b315 HL |
13600 | } |
13601 | len = read(fd, buf, sizeof(buf)); | |
13602 | close(fd); | |
13603 | if (len <= 0) { | |
6803ee25 AN |
13604 | err = len ? -errno : -EINVAL; |
13605 | pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err); | |
13606 | return err; | |
6446b315 | 13607 | } |
6803ee25 AN |
13608 | if (len >= sizeof(buf)) { |
13609 | pr_warn("CPU mask is too big in file %s\n", fcpu); | |
13610 | return -E2BIG; | |
6446b315 HL |
13611 | } |
13612 | buf[len] = '\0'; | |
13613 | ||
6803ee25 AN |
13614 | return parse_cpu_mask_str(buf, mask, mask_sz); |
13615 | } | |
13616 | ||
13617 | int libbpf_num_possible_cpus(void) | |
13618 | { | |
13619 | static const char *fcpu = "/sys/devices/system/cpu/possible"; | |
13620 | static int cpus; | |
13621 | int err, n, i, tmp_cpus; | |
13622 | bool *mask; | |
13623 | ||
13624 | tmp_cpus = READ_ONCE(cpus); | |
13625 | if (tmp_cpus > 0) | |
13626 | return tmp_cpus; | |
13627 | ||
13628 | err = parse_cpu_mask_file(fcpu, &mask, &n); | |
13629 | if (err) | |
e9fc3ce9 | 13630 | return libbpf_err(err); |
6803ee25 AN |
13631 | |
13632 | tmp_cpus = 0; | |
13633 | for (i = 0; i < n; i++) { | |
13634 | if (mask[i]) | |
13635 | tmp_cpus++; | |
6446b315 | 13636 | } |
6803ee25 | 13637 | free(mask); |
56fbc241 TC |
13638 | |
13639 | WRITE_ONCE(cpus, tmp_cpus); | |
13640 | return tmp_cpus; | |
6446b315 | 13641 | } |
d66562fb | 13642 | |
430025e5 DK |
13643 | static int populate_skeleton_maps(const struct bpf_object *obj, |
13644 | struct bpf_map_skeleton *maps, | |
13645 | size_t map_cnt) | |
13646 | { | |
13647 | int i; | |
13648 | ||
13649 | for (i = 0; i < map_cnt; i++) { | |
13650 | struct bpf_map **map = maps[i].map; | |
13651 | const char *name = maps[i].name; | |
13652 | void **mmaped = maps[i].mmaped; | |
13653 | ||
13654 | *map = bpf_object__find_map_by_name(obj, name); | |
13655 | if (!*map) { | |
13656 | pr_warn("failed to find skeleton map '%s'\n", name); | |
13657 | return -ESRCH; | |
13658 | } | |
13659 | ||
13660 | /* externs shouldn't be pre-setup from user code */ | |
13661 | if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG) | |
13662 | *mmaped = (*map)->mmaped; | |
13663 | } | |
13664 | return 0; | |
13665 | } | |
13666 | ||
13667 | static int populate_skeleton_progs(const struct bpf_object *obj, | |
13668 | struct bpf_prog_skeleton *progs, | |
13669 | size_t prog_cnt) | |
13670 | { | |
13671 | int i; | |
13672 | ||
13673 | for (i = 0; i < prog_cnt; i++) { | |
13674 | struct bpf_program **prog = progs[i].prog; | |
13675 | const char *name = progs[i].name; | |
13676 | ||
13677 | *prog = bpf_object__find_program_by_name(obj, name); | |
13678 | if (!*prog) { | |
13679 | pr_warn("failed to find skeleton program '%s'\n", name); | |
13680 | return -ESRCH; | |
13681 | } | |
13682 | } | |
13683 | return 0; | |
13684 | } | |
13685 | ||
d66562fb AN |
13686 | int bpf_object__open_skeleton(struct bpf_object_skeleton *s, |
13687 | const struct bpf_object_open_opts *opts) | |
13688 | { | |
13689 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts, | |
13690 | .object_name = s->name, | |
13691 | ); | |
13692 | struct bpf_object *obj; | |
430025e5 | 13693 | int err; |
d66562fb AN |
13694 | |
13695 | /* Attempt to preserve opts->object_name, unless overriden by user | |
13696 | * explicitly. Overwriting object name for skeletons is discouraged, | |
13697 | * as it breaks global data maps, because they contain object name | |
13698 | * prefix as their own map name prefix. When skeleton is generated, | |
13699 | * bpftool is making an assumption that this name will stay the same. | |
13700 | */ | |
13701 | if (opts) { | |
13702 | memcpy(&skel_opts, opts, sizeof(*opts)); | |
13703 | if (!opts->object_name) | |
13704 | skel_opts.object_name = s->name; | |
13705 | } | |
13706 | ||
13707 | obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts); | |
e9fc3ce9 AN |
13708 | err = libbpf_get_error(obj); |
13709 | if (err) { | |
13710 | pr_warn("failed to initialize skeleton BPF object '%s': %d\n", | |
13711 | s->name, err); | |
13712 | return libbpf_err(err); | |
d66562fb AN |
13713 | } |
13714 | ||
13715 | *s->obj = obj; | |
430025e5 DK |
13716 | err = populate_skeleton_maps(obj, s->maps, s->map_cnt); |
13717 | if (err) { | |
13718 | pr_warn("failed to populate skeleton maps for '%s': %d\n", s->name, err); | |
13719 | return libbpf_err(err); | |
13720 | } | |
13721 | ||
13722 | err = populate_skeleton_progs(obj, s->progs, s->prog_cnt); | |
13723 | if (err) { | |
13724 | pr_warn("failed to populate skeleton progs for '%s': %d\n", s->name, err); | |
13725 | return libbpf_err(err); | |
13726 | } | |
d66562fb | 13727 | |
430025e5 DK |
13728 | return 0; |
13729 | } | |
d66562fb | 13730 | |
430025e5 DK |
13731 | int bpf_object__open_subskeleton(struct bpf_object_subskeleton *s) |
13732 | { | |
13733 | int err, len, var_idx, i; | |
13734 | const char *var_name; | |
13735 | const struct bpf_map *map; | |
13736 | struct btf *btf; | |
13737 | __u32 map_type_id; | |
13738 | const struct btf_type *map_type, *var_type; | |
13739 | const struct bpf_var_skeleton *var_skel; | |
13740 | struct btf_var_secinfo *var; | |
d66562fb | 13741 | |
430025e5 DK |
13742 | if (!s->obj) |
13743 | return libbpf_err(-EINVAL); | |
13744 | ||
13745 | btf = bpf_object__btf(s->obj); | |
13746 | if (!btf) { | |
13747 | pr_warn("subskeletons require BTF at runtime (object %s)\n", | |
e3ba8e4e | 13748 | bpf_object__name(s->obj)); |
430025e5 | 13749 | return libbpf_err(-errno); |
d66562fb AN |
13750 | } |
13751 | ||
430025e5 DK |
13752 | err = populate_skeleton_maps(s->obj, s->maps, s->map_cnt); |
13753 | if (err) { | |
13754 | pr_warn("failed to populate subskeleton maps: %d\n", err); | |
13755 | return libbpf_err(err); | |
13756 | } | |
d66562fb | 13757 | |
430025e5 DK |
13758 | err = populate_skeleton_progs(s->obj, s->progs, s->prog_cnt); |
13759 | if (err) { | |
13760 | pr_warn("failed to populate subskeleton maps: %d\n", err); | |
13761 | return libbpf_err(err); | |
d66562fb AN |
13762 | } |
13763 | ||
430025e5 DK |
13764 | for (var_idx = 0; var_idx < s->var_cnt; var_idx++) { |
13765 | var_skel = &s->vars[var_idx]; | |
13766 | map = *var_skel->map; | |
13767 | map_type_id = bpf_map__btf_value_type_id(map); | |
13768 | map_type = btf__type_by_id(btf, map_type_id); | |
13769 | ||
13770 | if (!btf_is_datasec(map_type)) { | |
13771 | pr_warn("type for map '%1$s' is not a datasec: %2$s", | |
13772 | bpf_map__name(map), | |
13773 | __btf_kind_str(btf_kind(map_type))); | |
13774 | return libbpf_err(-EINVAL); | |
13775 | } | |
13776 | ||
13777 | len = btf_vlen(map_type); | |
13778 | var = btf_var_secinfos(map_type); | |
13779 | for (i = 0; i < len; i++, var++) { | |
13780 | var_type = btf__type_by_id(btf, var->type); | |
13781 | var_name = btf__name_by_offset(btf, var_type->name_off); | |
13782 | if (strcmp(var_name, var_skel->name) == 0) { | |
13783 | *var_skel->addr = map->mmaped + var->offset; | |
13784 | break; | |
13785 | } | |
13786 | } | |
13787 | } | |
d66562fb AN |
13788 | return 0; |
13789 | } | |
13790 | ||
430025e5 DK |
13791 | void bpf_object__destroy_subskeleton(struct bpf_object_subskeleton *s) |
13792 | { | |
13793 | if (!s) | |
13794 | return; | |
13795 | free(s->maps); | |
13796 | free(s->progs); | |
13797 | free(s->vars); | |
13798 | free(s); | |
13799 | } | |
13800 | ||
d66562fb AN |
13801 | int bpf_object__load_skeleton(struct bpf_object_skeleton *s) |
13802 | { | |
13803 | int i, err; | |
13804 | ||
13805 | err = bpf_object__load(*s->obj); | |
13806 | if (err) { | |
13807 | pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err); | |
e9fc3ce9 | 13808 | return libbpf_err(err); |
d66562fb AN |
13809 | } |
13810 | ||
13811 | for (i = 0; i < s->map_cnt; i++) { | |
13812 | struct bpf_map *map = *s->maps[i].map; | |
79ff13e9 | 13813 | size_t mmap_sz = bpf_map_mmap_sz(map); |
f08c18e0 | 13814 | int prot, map_fd = map->fd; |
d66562fb | 13815 | void **mmaped = s->maps[i].mmaped; |
d66562fb AN |
13816 | |
13817 | if (!mmaped) | |
13818 | continue; | |
13819 | ||
13820 | if (!(map->def.map_flags & BPF_F_MMAPABLE)) { | |
13821 | *mmaped = NULL; | |
13822 | continue; | |
13823 | } | |
13824 | ||
2e7ba4f8 AN |
13825 | if (map->def.type == BPF_MAP_TYPE_ARENA) { |
13826 | *mmaped = map->mmaped; | |
13827 | continue; | |
13828 | } | |
13829 | ||
d66562fb AN |
13830 | if (map->def.map_flags & BPF_F_RDONLY_PROG) |
13831 | prot = PROT_READ; | |
13832 | else | |
13833 | prot = PROT_READ | PROT_WRITE; | |
13834 | ||
13835 | /* Remap anonymous mmap()-ed "map initialization image" as | |
13836 | * a BPF map-backed mmap()-ed memory, but preserving the same | |
13837 | * memory address. This will cause kernel to change process' | |
13838 | * page table to point to a different piece of kernel memory, | |
13839 | * but from userspace point of view memory address (and its | |
13840 | * contents, being identical at this point) will stay the | |
13841 | * same. This mapping will be released by bpf_object__close() | |
13842 | * as per normal clean up procedure, so we don't need to worry | |
13843 | * about it from skeleton's clean up perspective. | |
13844 | */ | |
9d0a2331 | 13845 | *mmaped = mmap(map->mmaped, mmap_sz, prot, MAP_SHARED | MAP_FIXED, map_fd, 0); |
2ad97d47 | 13846 | if (*mmaped == MAP_FAILED) { |
d66562fb AN |
13847 | err = -errno; |
13848 | *mmaped = NULL; | |
13849 | pr_warn("failed to re-mmap() map '%s': %d\n", | |
13850 | bpf_map__name(map), err); | |
e9fc3ce9 | 13851 | return libbpf_err(err); |
d66562fb AN |
13852 | } |
13853 | } | |
13854 | ||
13855 | return 0; | |
13856 | } | |
13857 | ||
13858 | int bpf_object__attach_skeleton(struct bpf_object_skeleton *s) | |
13859 | { | |
e9fc3ce9 | 13860 | int i, err; |
d66562fb AN |
13861 | |
13862 | for (i = 0; i < s->prog_cnt; i++) { | |
13863 | struct bpf_program *prog = *s->progs[i].prog; | |
13864 | struct bpf_link **link = s->progs[i].link; | |
d66562fb | 13865 | |
43cb8cba | 13866 | if (!prog->autoload || !prog->autoattach) |
d9297581 AN |
13867 | continue; |
13868 | ||
5532dfd4 | 13869 | /* auto-attaching not supported for this program */ |
4fa5bcfe | 13870 | if (!prog->sec_def || !prog->sec_def->prog_attach_fn) |
d66562fb AN |
13871 | continue; |
13872 | ||
4fa5bcfe AN |
13873 | /* if user already set the link manually, don't attempt auto-attach */ |
13874 | if (*link) | |
13875 | continue; | |
13876 | ||
13877 | err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link); | |
e9fc3ce9 | 13878 | if (err) { |
4fa5bcfe | 13879 | pr_warn("prog '%s': failed to auto-attach: %d\n", |
e9fc3ce9 AN |
13880 | bpf_program__name(prog), err); |
13881 | return libbpf_err(err); | |
d66562fb | 13882 | } |
4fa5bcfe AN |
13883 | |
13884 | /* It's possible that for some SEC() definitions auto-attach | |
13885 | * is supported in some cases (e.g., if definition completely | |
13886 | * specifies target information), but is not in other cases. | |
13887 | * SEC("uprobe") is one such case. If user specified target | |
13888 | * binary and function name, such BPF program can be | |
13889 | * auto-attached. But if not, it shouldn't trigger skeleton's | |
13890 | * attach to fail. It should just be skipped. | |
13891 | * attach_fn signals such case with returning 0 (no error) and | |
13892 | * setting link to NULL. | |
13893 | */ | |
d66562fb AN |
13894 | } |
13895 | ||
13896 | return 0; | |
13897 | } | |
13898 | ||
13899 | void bpf_object__detach_skeleton(struct bpf_object_skeleton *s) | |
13900 | { | |
13901 | int i; | |
13902 | ||
13903 | for (i = 0; i < s->prog_cnt; i++) { | |
13904 | struct bpf_link **link = s->progs[i].link; | |
13905 | ||
50450fc7 | 13906 | bpf_link__destroy(*link); |
d66562fb AN |
13907 | *link = NULL; |
13908 | } | |
13909 | } | |
13910 | ||
13911 | void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s) | |
13912 | { | |
a32ea51a YS |
13913 | if (!s) |
13914 | return; | |
13915 | ||
d66562fb AN |
13916 | if (s->progs) |
13917 | bpf_object__detach_skeleton(s); | |
13918 | if (s->obj) | |
13919 | bpf_object__close(*s->obj); | |
13920 | free(s->maps); | |
13921 | free(s->progs); | |
13922 | free(s); | |
13923 | } |