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> |
a64af0ef | 36 | #include <linux/ring_buffer.h> |
fb84b822 | 37 | #include <sys/epoll.h> |
63f2f5ee | 38 | #include <sys/ioctl.h> |
fb84b822 | 39 | #include <sys/mman.h> |
f367540c JS |
40 | #include <sys/stat.h> |
41 | #include <sys/types.h> | |
42 | #include <sys/vfs.h> | |
ddc7c304 | 43 | #include <sys/utsname.h> |
dc3a2d25 | 44 | #include <sys/resource.h> |
1a5e3fb1 WN |
45 | #include <libelf.h> |
46 | #include <gelf.h> | |
166750bc | 47 | #include <zlib.h> |
1b76c13e WN |
48 | |
49 | #include "libbpf.h" | |
52d3352e | 50 | #include "bpf.h" |
8a138aed | 51 | #include "btf.h" |
6d41907c | 52 | #include "str_error.h" |
d7c4b398 | 53 | #include "libbpf_internal.h" |
ddc7c304 | 54 | #include "hashmap.h" |
67234743 | 55 | #include "bpf_gen_internal.h" |
c44fd845 | 56 | #include "zip.h" |
b3f59d66 | 57 | |
f367540c JS |
58 | #ifndef BPF_FS_MAGIC |
59 | #define BPF_FS_MAGIC 0xcafe4a11 | |
60 | #endif | |
61 | ||
9c0f8cbd AN |
62 | #define BPF_INSN_SZ (sizeof(struct bpf_insn)) |
63 | ||
ff466b58 AI |
64 | /* vsprintf() in __base_pr() uses nonliteral format string. It may break |
65 | * compilation if user enables corresponding warning. Disable it explicitly. | |
66 | */ | |
67 | #pragma GCC diagnostic ignored "-Wformat-nonliteral" | |
68 | ||
b3f59d66 WN |
69 | #define __printf(a, b) __attribute__((format(printf, a, b))) |
70 | ||
590a0088 | 71 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj); |
aea28a60 | 72 | static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog); |
590a0088 | 73 | |
ccde5760 DM |
74 | static const char * const attach_type_name[] = { |
75 | [BPF_CGROUP_INET_INGRESS] = "cgroup_inet_ingress", | |
76 | [BPF_CGROUP_INET_EGRESS] = "cgroup_inet_egress", | |
77 | [BPF_CGROUP_INET_SOCK_CREATE] = "cgroup_inet_sock_create", | |
78 | [BPF_CGROUP_INET_SOCK_RELEASE] = "cgroup_inet_sock_release", | |
79 | [BPF_CGROUP_SOCK_OPS] = "cgroup_sock_ops", | |
80 | [BPF_CGROUP_DEVICE] = "cgroup_device", | |
81 | [BPF_CGROUP_INET4_BIND] = "cgroup_inet4_bind", | |
82 | [BPF_CGROUP_INET6_BIND] = "cgroup_inet6_bind", | |
83 | [BPF_CGROUP_INET4_CONNECT] = "cgroup_inet4_connect", | |
84 | [BPF_CGROUP_INET6_CONNECT] = "cgroup_inet6_connect", | |
85 | [BPF_CGROUP_INET4_POST_BIND] = "cgroup_inet4_post_bind", | |
86 | [BPF_CGROUP_INET6_POST_BIND] = "cgroup_inet6_post_bind", | |
87 | [BPF_CGROUP_INET4_GETPEERNAME] = "cgroup_inet4_getpeername", | |
88 | [BPF_CGROUP_INET6_GETPEERNAME] = "cgroup_inet6_getpeername", | |
89 | [BPF_CGROUP_INET4_GETSOCKNAME] = "cgroup_inet4_getsockname", | |
90 | [BPF_CGROUP_INET6_GETSOCKNAME] = "cgroup_inet6_getsockname", | |
91 | [BPF_CGROUP_UDP4_SENDMSG] = "cgroup_udp4_sendmsg", | |
92 | [BPF_CGROUP_UDP6_SENDMSG] = "cgroup_udp6_sendmsg", | |
93 | [BPF_CGROUP_SYSCTL] = "cgroup_sysctl", | |
94 | [BPF_CGROUP_UDP4_RECVMSG] = "cgroup_udp4_recvmsg", | |
95 | [BPF_CGROUP_UDP6_RECVMSG] = "cgroup_udp6_recvmsg", | |
96 | [BPF_CGROUP_GETSOCKOPT] = "cgroup_getsockopt", | |
97 | [BPF_CGROUP_SETSOCKOPT] = "cgroup_setsockopt", | |
98 | [BPF_SK_SKB_STREAM_PARSER] = "sk_skb_stream_parser", | |
99 | [BPF_SK_SKB_STREAM_VERDICT] = "sk_skb_stream_verdict", | |
100 | [BPF_SK_SKB_VERDICT] = "sk_skb_verdict", | |
101 | [BPF_SK_MSG_VERDICT] = "sk_msg_verdict", | |
102 | [BPF_LIRC_MODE2] = "lirc_mode2", | |
103 | [BPF_FLOW_DISSECTOR] = "flow_dissector", | |
104 | [BPF_TRACE_RAW_TP] = "trace_raw_tp", | |
105 | [BPF_TRACE_FENTRY] = "trace_fentry", | |
106 | [BPF_TRACE_FEXIT] = "trace_fexit", | |
107 | [BPF_MODIFY_RETURN] = "modify_return", | |
108 | [BPF_LSM_MAC] = "lsm_mac", | |
bffcf348 | 109 | [BPF_LSM_CGROUP] = "lsm_cgroup", |
ccde5760 DM |
110 | [BPF_SK_LOOKUP] = "sk_lookup", |
111 | [BPF_TRACE_ITER] = "trace_iter", | |
112 | [BPF_XDP_DEVMAP] = "xdp_devmap", | |
113 | [BPF_XDP_CPUMAP] = "xdp_cpumap", | |
114 | [BPF_XDP] = "xdp", | |
115 | [BPF_SK_REUSEPORT_SELECT] = "sk_reuseport_select", | |
116 | [BPF_SK_REUSEPORT_SELECT_OR_MIGRATE] = "sk_reuseport_select_or_migrate", | |
117 | [BPF_PERF_EVENT] = "perf_event", | |
118 | [BPF_TRACE_KPROBE_MULTI] = "trace_kprobe_multi", | |
8d1608d7 | 119 | [BPF_STRUCT_OPS] = "struct_ops", |
ccde5760 DM |
120 | }; |
121 | ||
ba5d1b58 DM |
122 | static const char * const link_type_name[] = { |
123 | [BPF_LINK_TYPE_UNSPEC] = "unspec", | |
124 | [BPF_LINK_TYPE_RAW_TRACEPOINT] = "raw_tracepoint", | |
125 | [BPF_LINK_TYPE_TRACING] = "tracing", | |
126 | [BPF_LINK_TYPE_CGROUP] = "cgroup", | |
127 | [BPF_LINK_TYPE_ITER] = "iter", | |
128 | [BPF_LINK_TYPE_NETNS] = "netns", | |
129 | [BPF_LINK_TYPE_XDP] = "xdp", | |
130 | [BPF_LINK_TYPE_PERF_EVENT] = "perf_event", | |
131 | [BPF_LINK_TYPE_KPROBE_MULTI] = "kprobe_multi", | |
132 | [BPF_LINK_TYPE_STRUCT_OPS] = "struct_ops", | |
d0fe92fb | 133 | [BPF_LINK_TYPE_NETFILTER] = "netfilter", |
ba5d1b58 DM |
134 | }; |
135 | ||
3e6dc020 DM |
136 | static const char * const map_type_name[] = { |
137 | [BPF_MAP_TYPE_UNSPEC] = "unspec", | |
138 | [BPF_MAP_TYPE_HASH] = "hash", | |
139 | [BPF_MAP_TYPE_ARRAY] = "array", | |
140 | [BPF_MAP_TYPE_PROG_ARRAY] = "prog_array", | |
141 | [BPF_MAP_TYPE_PERF_EVENT_ARRAY] = "perf_event_array", | |
142 | [BPF_MAP_TYPE_PERCPU_HASH] = "percpu_hash", | |
143 | [BPF_MAP_TYPE_PERCPU_ARRAY] = "percpu_array", | |
144 | [BPF_MAP_TYPE_STACK_TRACE] = "stack_trace", | |
145 | [BPF_MAP_TYPE_CGROUP_ARRAY] = "cgroup_array", | |
146 | [BPF_MAP_TYPE_LRU_HASH] = "lru_hash", | |
147 | [BPF_MAP_TYPE_LRU_PERCPU_HASH] = "lru_percpu_hash", | |
148 | [BPF_MAP_TYPE_LPM_TRIE] = "lpm_trie", | |
149 | [BPF_MAP_TYPE_ARRAY_OF_MAPS] = "array_of_maps", | |
150 | [BPF_MAP_TYPE_HASH_OF_MAPS] = "hash_of_maps", | |
151 | [BPF_MAP_TYPE_DEVMAP] = "devmap", | |
152 | [BPF_MAP_TYPE_DEVMAP_HASH] = "devmap_hash", | |
153 | [BPF_MAP_TYPE_SOCKMAP] = "sockmap", | |
154 | [BPF_MAP_TYPE_CPUMAP] = "cpumap", | |
155 | [BPF_MAP_TYPE_XSKMAP] = "xskmap", | |
156 | [BPF_MAP_TYPE_SOCKHASH] = "sockhash", | |
157 | [BPF_MAP_TYPE_CGROUP_STORAGE] = "cgroup_storage", | |
158 | [BPF_MAP_TYPE_REUSEPORT_SOCKARRAY] = "reuseport_sockarray", | |
159 | [BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE] = "percpu_cgroup_storage", | |
160 | [BPF_MAP_TYPE_QUEUE] = "queue", | |
161 | [BPF_MAP_TYPE_STACK] = "stack", | |
162 | [BPF_MAP_TYPE_SK_STORAGE] = "sk_storage", | |
163 | [BPF_MAP_TYPE_STRUCT_OPS] = "struct_ops", | |
164 | [BPF_MAP_TYPE_RINGBUF] = "ringbuf", | |
165 | [BPF_MAP_TYPE_INODE_STORAGE] = "inode_storage", | |
166 | [BPF_MAP_TYPE_TASK_STORAGE] = "task_storage", | |
167 | [BPF_MAP_TYPE_BLOOM_FILTER] = "bloom_filter", | |
583c1f42 | 168 | [BPF_MAP_TYPE_USER_RINGBUF] = "user_ringbuf", |
4fe64af2 | 169 | [BPF_MAP_TYPE_CGRP_STORAGE] = "cgrp_storage", |
3e6dc020 DM |
170 | }; |
171 | ||
d18616e7 DM |
172 | static const char * const prog_type_name[] = { |
173 | [BPF_PROG_TYPE_UNSPEC] = "unspec", | |
174 | [BPF_PROG_TYPE_SOCKET_FILTER] = "socket_filter", | |
175 | [BPF_PROG_TYPE_KPROBE] = "kprobe", | |
176 | [BPF_PROG_TYPE_SCHED_CLS] = "sched_cls", | |
177 | [BPF_PROG_TYPE_SCHED_ACT] = "sched_act", | |
178 | [BPF_PROG_TYPE_TRACEPOINT] = "tracepoint", | |
179 | [BPF_PROG_TYPE_XDP] = "xdp", | |
180 | [BPF_PROG_TYPE_PERF_EVENT] = "perf_event", | |
181 | [BPF_PROG_TYPE_CGROUP_SKB] = "cgroup_skb", | |
182 | [BPF_PROG_TYPE_CGROUP_SOCK] = "cgroup_sock", | |
183 | [BPF_PROG_TYPE_LWT_IN] = "lwt_in", | |
184 | [BPF_PROG_TYPE_LWT_OUT] = "lwt_out", | |
185 | [BPF_PROG_TYPE_LWT_XMIT] = "lwt_xmit", | |
186 | [BPF_PROG_TYPE_SOCK_OPS] = "sock_ops", | |
187 | [BPF_PROG_TYPE_SK_SKB] = "sk_skb", | |
188 | [BPF_PROG_TYPE_CGROUP_DEVICE] = "cgroup_device", | |
189 | [BPF_PROG_TYPE_SK_MSG] = "sk_msg", | |
190 | [BPF_PROG_TYPE_RAW_TRACEPOINT] = "raw_tracepoint", | |
191 | [BPF_PROG_TYPE_CGROUP_SOCK_ADDR] = "cgroup_sock_addr", | |
192 | [BPF_PROG_TYPE_LWT_SEG6LOCAL] = "lwt_seg6local", | |
193 | [BPF_PROG_TYPE_LIRC_MODE2] = "lirc_mode2", | |
194 | [BPF_PROG_TYPE_SK_REUSEPORT] = "sk_reuseport", | |
195 | [BPF_PROG_TYPE_FLOW_DISSECTOR] = "flow_dissector", | |
196 | [BPF_PROG_TYPE_CGROUP_SYSCTL] = "cgroup_sysctl", | |
197 | [BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE] = "raw_tracepoint_writable", | |
198 | [BPF_PROG_TYPE_CGROUP_SOCKOPT] = "cgroup_sockopt", | |
199 | [BPF_PROG_TYPE_TRACING] = "tracing", | |
200 | [BPF_PROG_TYPE_STRUCT_OPS] = "struct_ops", | |
201 | [BPF_PROG_TYPE_EXT] = "ext", | |
202 | [BPF_PROG_TYPE_LSM] = "lsm", | |
203 | [BPF_PROG_TYPE_SK_LOOKUP] = "sk_lookup", | |
204 | [BPF_PROG_TYPE_SYSCALL] = "syscall", | |
d0fe92fb | 205 | [BPF_PROG_TYPE_NETFILTER] = "netfilter", |
d18616e7 DM |
206 | }; |
207 | ||
a8a1f7d0 SF |
208 | static int __base_pr(enum libbpf_print_level level, const char *format, |
209 | va_list args) | |
b3f59d66 | 210 | { |
6f1ae8b6 YS |
211 | if (level == LIBBPF_DEBUG) |
212 | return 0; | |
213 | ||
a8a1f7d0 | 214 | return vfprintf(stderr, format, args); |
b3f59d66 WN |
215 | } |
216 | ||
a8a1f7d0 | 217 | static libbpf_print_fn_t __libbpf_pr = __base_pr; |
b3f59d66 | 218 | |
e87fd8ba | 219 | libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn) |
b3f59d66 | 220 | { |
f1cb927c JK |
221 | libbpf_print_fn_t old_print_fn; |
222 | ||
223 | old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED); | |
e87fd8ba | 224 | |
e87fd8ba | 225 | return old_print_fn; |
b3f59d66 | 226 | } |
1a5e3fb1 | 227 | |
8461ef8b YS |
228 | __printf(2, 3) |
229 | void libbpf_print(enum libbpf_print_level level, const char *format, ...) | |
230 | { | |
231 | va_list args; | |
d7c5802f | 232 | int old_errno; |
f1cb927c | 233 | libbpf_print_fn_t print_fn; |
8461ef8b | 234 | |
f1cb927c JK |
235 | print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED); |
236 | if (!print_fn) | |
6f1ae8b6 YS |
237 | return; |
238 | ||
d7c5802f AN |
239 | old_errno = errno; |
240 | ||
8461ef8b | 241 | va_start(args, format); |
6f1ae8b6 | 242 | __libbpf_pr(level, format, args); |
8461ef8b | 243 | va_end(args); |
d7c5802f AN |
244 | |
245 | errno = old_errno; | |
8461ef8b YS |
246 | } |
247 | ||
dc3a2d25 THJ |
248 | static void pr_perm_msg(int err) |
249 | { | |
250 | struct rlimit limit; | |
251 | char buf[100]; | |
252 | ||
253 | if (err != -EPERM || geteuid() != 0) | |
254 | return; | |
255 | ||
256 | err = getrlimit(RLIMIT_MEMLOCK, &limit); | |
257 | if (err) | |
258 | return; | |
259 | ||
260 | if (limit.rlim_cur == RLIM_INFINITY) | |
261 | return; | |
262 | ||
263 | if (limit.rlim_cur < 1024) | |
b5c7d0d0 | 264 | snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur); |
dc3a2d25 THJ |
265 | else if (limit.rlim_cur < 1024*1024) |
266 | snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024); | |
267 | else | |
268 | snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024)); | |
269 | ||
270 | pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n", | |
271 | buf); | |
272 | } | |
273 | ||
6371ca3b WN |
274 | #define STRERR_BUFSIZE 128 |
275 | ||
1a5e3fb1 WN |
276 | /* Copied from tools/perf/util/util.h */ |
277 | #ifndef zfree | |
278 | # define zfree(ptr) ({ free(*ptr); *ptr = NULL; }) | |
279 | #endif | |
280 | ||
281 | #ifndef zclose | |
282 | # define zclose(fd) ({ \ | |
283 | int ___err = 0; \ | |
284 | if ((fd) >= 0) \ | |
285 | ___err = close((fd)); \ | |
286 | fd = -1; \ | |
287 | ___err; }) | |
288 | #endif | |
289 | ||
34be1646 SL |
290 | static inline __u64 ptr_to_u64(const void *ptr) |
291 | { | |
292 | return (__u64) (unsigned long) ptr; | |
293 | } | |
294 | ||
5981881d AN |
295 | int libbpf_set_strict_mode(enum libbpf_strict_mode mode) |
296 | { | |
bd054102 | 297 | /* as of v1.0 libbpf_set_strict_mode() is a no-op */ |
5981881d AN |
298 | return 0; |
299 | } | |
300 | ||
7615209f AN |
301 | __u32 libbpf_major_version(void) |
302 | { | |
303 | return LIBBPF_MAJOR_VERSION; | |
304 | } | |
305 | ||
306 | __u32 libbpf_minor_version(void) | |
307 | { | |
308 | return LIBBPF_MINOR_VERSION; | |
309 | } | |
310 | ||
311 | const char *libbpf_version_string(void) | |
312 | { | |
313 | #define __S(X) #X | |
314 | #define _S(X) __S(X) | |
315 | return "v" _S(LIBBPF_MAJOR_VERSION) "." _S(LIBBPF_MINOR_VERSION); | |
316 | #undef _S | |
317 | #undef __S | |
318 | } | |
319 | ||
166750bc AN |
320 | enum reloc_type { |
321 | RELO_LD64, | |
322 | RELO_CALL, | |
323 | RELO_DATA, | |
a18f7214 AS |
324 | RELO_EXTERN_LD64, |
325 | RELO_EXTERN_CALL, | |
53eddb5e | 326 | RELO_SUBPROG_ADDR, |
d0e92887 | 327 | RELO_CORE, |
166750bc AN |
328 | }; |
329 | ||
330 | struct reloc_desc { | |
331 | enum reloc_type type; | |
332 | int insn_idx; | |
d0e92887 AS |
333 | union { |
334 | const struct bpf_core_relo *core_relo; /* used when type == RELO_CORE */ | |
335 | struct { | |
336 | int map_idx; | |
337 | int sym_off; | |
3055ddd6 | 338 | int ext_idx; |
d0e92887 AS |
339 | }; |
340 | }; | |
166750bc AN |
341 | }; |
342 | ||
15ea31fa AN |
343 | /* stored as sec_def->cookie for all libbpf-supported SEC()s */ |
344 | enum sec_def_flags { | |
345 | SEC_NONE = 0, | |
346 | /* expected_attach_type is optional, if kernel doesn't support that */ | |
347 | SEC_EXP_ATTACH_OPT = 1, | |
348 | /* legacy, only used by libbpf_get_type_names() and | |
349 | * libbpf_attach_type_by_name(), not used by libbpf itself at all. | |
350 | * This used to be associated with cgroup (and few other) BPF programs | |
351 | * that were attachable through BPF_PROG_ATTACH command. Pretty | |
352 | * meaningless nowadays, though. | |
353 | */ | |
354 | SEC_ATTACHABLE = 2, | |
355 | SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT, | |
356 | /* attachment target is specified through BTF ID in either kernel or | |
e3ba8e4e KM |
357 | * other BPF program's BTF object |
358 | */ | |
15ea31fa AN |
359 | SEC_ATTACH_BTF = 4, |
360 | /* BPF program type allows sleeping/blocking in kernel */ | |
361 | SEC_SLEEPABLE = 8, | |
082c4bfb | 362 | /* BPF program support non-linear XDP buffer */ |
450b167f | 363 | SEC_XDP_FRAGS = 16, |
15ea31fa AN |
364 | }; |
365 | ||
25498a19 | 366 | struct bpf_sec_def { |
697f104d | 367 | char *sec; |
25498a19 AN |
368 | enum bpf_prog_type prog_type; |
369 | enum bpf_attach_type expected_attach_type; | |
15ea31fa | 370 | long cookie; |
697f104d | 371 | int handler_id; |
12d9466d | 372 | |
4fa5bcfe AN |
373 | libbpf_prog_setup_fn_t prog_setup_fn; |
374 | libbpf_prog_prepare_load_fn_t prog_prepare_load_fn; | |
375 | libbpf_prog_attach_fn_t prog_attach_fn; | |
25498a19 AN |
376 | }; |
377 | ||
a5b8bd47 WN |
378 | /* |
379 | * bpf_prog should be a better name but it has been used in | |
380 | * linux/filter.h. | |
381 | */ | |
382 | struct bpf_program { | |
bd054102 | 383 | char *name; |
52109584 | 384 | char *sec_name; |
c1122392 | 385 | size_t sec_idx; |
bd054102 | 386 | const struct bpf_sec_def *sec_def; |
c1122392 AN |
387 | /* this program's instruction offset (in number of instructions) |
388 | * within its containing ELF section | |
389 | */ | |
390 | size_t sec_insn_off; | |
391 | /* number of original instructions in ELF section belonging to this | |
392 | * program, not taking into account subprogram instructions possible | |
393 | * appended later during relocation | |
394 | */ | |
395 | size_t sec_insn_cnt; | |
396 | /* Offset (in number of instructions) of the start of instruction | |
397 | * belonging to this BPF program within its containing main BPF | |
398 | * program. For the entry-point (main) BPF program, this is always | |
399 | * zero. For a sub-program, this gets reset before each of main BPF | |
400 | * programs are processed and relocated and is used to determined | |
401 | * whether sub-program was already appended to the main program, and | |
402 | * if yes, at which instruction offset. | |
403 | */ | |
404 | size_t sub_insn_off; | |
405 | ||
c1122392 AN |
406 | /* instructions that belong to BPF program; insns[0] is located at |
407 | * sec_insn_off instruction within its ELF section in ELF file, so | |
408 | * when mapping ELF file instruction index to the local instruction, | |
409 | * one needs to subtract sec_insn_off; and vice versa. | |
410 | */ | |
a5b8bd47 | 411 | struct bpf_insn *insns; |
c1122392 AN |
412 | /* actual number of instruction in this BPF program's image; for |
413 | * entry-point BPF programs this includes the size of main program | |
414 | * itself plus all the used sub-programs, appended at the end | |
415 | */ | |
c3c55696 | 416 | size_t insns_cnt; |
34090915 | 417 | |
166750bc | 418 | struct reloc_desc *reloc_desc; |
34090915 | 419 | int nr_reloc; |
b3ce9079 AN |
420 | |
421 | /* BPF verifier log settings */ | |
422 | char *log_buf; | |
423 | size_t log_size; | |
424 | __u32 log_level; | |
55cffde2 | 425 | |
aa9b1ac3 | 426 | struct bpf_object *obj; |
d7be143b | 427 | |
cf90a20d | 428 | int fd; |
a3820c48 | 429 | bool autoload; |
43cb8cba | 430 | bool autoattach; |
aea28a60 | 431 | bool mark_btf_static; |
c1122392 | 432 | enum bpf_prog_type type; |
d7be143b | 433 | enum bpf_attach_type expected_attach_type; |
cf90a20d | 434 | |
c1122392 | 435 | int prog_ifindex; |
91abb4a6 | 436 | __u32 attach_btf_obj_fd; |
12a8654b | 437 | __u32 attach_btf_id; |
e7bf94db | 438 | __u32 attach_prog_fd; |
cf90a20d | 439 | |
2993e051 YS |
440 | void *func_info; |
441 | __u32 func_info_rec_size; | |
f0187f0b | 442 | __u32 func_info_cnt; |
47eff617 | 443 | |
3d650141 MKL |
444 | void *line_info; |
445 | __u32 line_info_rec_size; | |
446 | __u32 line_info_cnt; | |
04656198 | 447 | __u32 prog_flags; |
a5b8bd47 WN |
448 | }; |
449 | ||
590a0088 MKL |
450 | struct bpf_struct_ops { |
451 | const char *tname; | |
452 | const struct btf_type *type; | |
453 | struct bpf_program **progs; | |
454 | __u32 *kern_func_off; | |
455 | /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */ | |
456 | void *data; | |
457 | /* e.g. struct bpf_struct_ops_tcp_congestion_ops in | |
458 | * btf_vmlinux's format. | |
459 | * struct bpf_struct_ops_tcp_congestion_ops { | |
460 | * [... some other kernel fields ...] | |
461 | * struct tcp_congestion_ops data; | |
462 | * } | |
463 | * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops) | |
464 | * bpf_map__init_kern_struct_ops() will populate the "kern_vdata" | |
465 | * from "data". | |
466 | */ | |
467 | void *kern_vdata; | |
468 | __u32 type_id; | |
469 | }; | |
470 | ||
ac9d1389 AN |
471 | #define DATA_SEC ".data" |
472 | #define BSS_SEC ".bss" | |
473 | #define RODATA_SEC ".rodata" | |
81bfdd08 | 474 | #define KCONFIG_SEC ".kconfig" |
1c0c7074 | 475 | #define KSYMS_SEC ".ksyms" |
590a0088 | 476 | #define STRUCT_OPS_SEC ".struct_ops" |
809a69d6 | 477 | #define STRUCT_OPS_LINK_SEC ".struct_ops.link" |
ac9d1389 | 478 | |
d859900c DB |
479 | enum libbpf_map_type { |
480 | LIBBPF_MAP_UNSPEC, | |
481 | LIBBPF_MAP_DATA, | |
482 | LIBBPF_MAP_BSS, | |
483 | LIBBPF_MAP_RODATA, | |
81bfdd08 | 484 | LIBBPF_MAP_KCONFIG, |
d859900c DB |
485 | }; |
486 | ||
146bf811 AN |
487 | struct bpf_map_def { |
488 | unsigned int type; | |
489 | unsigned int key_size; | |
490 | unsigned int value_size; | |
491 | unsigned int max_entries; | |
492 | unsigned int map_flags; | |
493 | }; | |
494 | ||
9d759a9b | 495 | struct bpf_map { |
ec41817b | 496 | struct bpf_object *obj; |
561bbcca | 497 | char *name; |
aed65917 AN |
498 | /* real_name is defined for special internal maps (.rodata*, |
499 | * .data*, .bss, .kconfig) and preserves their original ELF section | |
e3ba8e4e | 500 | * name. This is important to be able to find corresponding BTF |
aed65917 AN |
501 | * DATASEC information. |
502 | */ | |
503 | char *real_name; | |
01af3bf0 | 504 | int fd; |
db48814b AN |
505 | int sec_idx; |
506 | size_t sec_offset; | |
f0307a7e | 507 | int map_ifindex; |
addb9fc9 | 508 | int inner_map_fd; |
9d759a9b | 509 | struct bpf_map_def def; |
1bdb6c9a | 510 | __u32 numa_node; |
646f02ff | 511 | __u32 btf_var_idx; |
5b891af7 MKL |
512 | __u32 btf_key_type_id; |
513 | __u32 btf_value_type_id; | |
590a0088 | 514 | __u32 btf_vmlinux_value_type_id; |
d859900c | 515 | enum libbpf_map_type libbpf_type; |
eba9c5f4 | 516 | void *mmaped; |
590a0088 | 517 | struct bpf_struct_ops *st_ops; |
646f02ff AN |
518 | struct bpf_map *inner_map; |
519 | void **init_slots; | |
520 | int init_slots_sz; | |
4580b25f THJ |
521 | char *pin_path; |
522 | bool pinned; | |
ec6d5f47 | 523 | bool reused; |
ec41817b | 524 | bool autocreate; |
47512102 | 525 | __u64 map_extra; |
d859900c DB |
526 | }; |
527 | ||
166750bc AN |
528 | enum extern_type { |
529 | EXT_UNKNOWN, | |
2e33efe3 | 530 | EXT_KCFG, |
1c0c7074 | 531 | EXT_KSYM, |
2e33efe3 AN |
532 | }; |
533 | ||
534 | enum kcfg_type { | |
535 | KCFG_UNKNOWN, | |
536 | KCFG_CHAR, | |
537 | KCFG_BOOL, | |
538 | KCFG_INT, | |
539 | KCFG_TRISTATE, | |
540 | KCFG_CHAR_ARR, | |
166750bc AN |
541 | }; |
542 | ||
543 | struct extern_desc { | |
2e33efe3 | 544 | enum extern_type type; |
166750bc AN |
545 | int sym_idx; |
546 | int btf_id; | |
2e33efe3 AN |
547 | int sec_btf_id; |
548 | const char *name; | |
166750bc | 549 | bool is_set; |
2e33efe3 AN |
550 | bool is_weak; |
551 | union { | |
552 | struct { | |
553 | enum kcfg_type type; | |
554 | int sz; | |
555 | int align; | |
556 | int data_off; | |
557 | bool is_signed; | |
558 | } kcfg; | |
1c0c7074 AN |
559 | struct { |
560 | unsigned long long addr; | |
d370bbe1 HL |
561 | |
562 | /* target btf_id of the corresponding kernel var. */ | |
284d2587 AN |
563 | int kernel_btf_obj_fd; |
564 | int kernel_btf_id; | |
d370bbe1 HL |
565 | |
566 | /* local btf_id of the ksym extern's type. */ | |
567 | __u32 type_id; | |
9dbe6015 KKD |
568 | /* BTF fd index to be patched in for insn->off, this is |
569 | * 0 for vmlinux BTF, index in obj->fd_array for module | |
570 | * BTF | |
571 | */ | |
572 | __s16 btf_fd_idx; | |
1c0c7074 | 573 | } ksym; |
2e33efe3 | 574 | }; |
166750bc AN |
575 | }; |
576 | ||
4f33a53d AN |
577 | struct module_btf { |
578 | struct btf *btf; | |
579 | char *name; | |
580 | __u32 id; | |
91abb4a6 | 581 | int fd; |
9dbe6015 | 582 | int fd_array_idx; |
4f33a53d AN |
583 | }; |
584 | ||
25bbbd7a AN |
585 | enum sec_type { |
586 | SEC_UNUSED = 0, | |
587 | SEC_RELO, | |
588 | SEC_BSS, | |
589 | SEC_DATA, | |
590 | SEC_RODATA, | |
591 | }; | |
592 | ||
593 | struct elf_sec_desc { | |
594 | enum sec_type sec_type; | |
595 | Elf64_Shdr *shdr; | |
596 | Elf_Data *data; | |
597 | }; | |
598 | ||
29a30ff5 AN |
599 | struct elf_state { |
600 | int fd; | |
601 | const void *obj_buf; | |
602 | size_t obj_buf_sz; | |
603 | Elf *elf; | |
ad23b723 | 604 | Elf64_Ehdr *ehdr; |
29a30ff5 | 605 | Elf_Data *symbols; |
29a30ff5 | 606 | Elf_Data *st_ops_data; |
809a69d6 | 607 | Elf_Data *st_ops_link_data; |
29a30ff5 AN |
608 | size_t shstrndx; /* section index for section name strings */ |
609 | size_t strtabidx; | |
25bbbd7a | 610 | struct elf_sec_desc *secs; |
51deedc9 | 611 | size_t sec_cnt; |
29a30ff5 AN |
612 | int btf_maps_shndx; |
613 | __u32 btf_maps_sec_btf_id; | |
614 | int text_shndx; | |
615 | int symbols_shndx; | |
29a30ff5 | 616 | int st_ops_shndx; |
809a69d6 | 617 | int st_ops_link_shndx; |
4f33a53d AN |
618 | }; |
619 | ||
2e4913e0 AN |
620 | struct usdt_manager; |
621 | ||
1a5e3fb1 | 622 | struct bpf_object { |
d859900c | 623 | char name[BPF_OBJ_NAME_LEN]; |
cb1e5e96 | 624 | char license[64]; |
438363c0 | 625 | __u32 kern_version; |
0b3d1efa | 626 | |
a5b8bd47 WN |
627 | struct bpf_program *programs; |
628 | size_t nr_programs; | |
9d759a9b WN |
629 | struct bpf_map *maps; |
630 | size_t nr_maps; | |
bf829271 | 631 | size_t maps_cap; |
9d759a9b | 632 | |
8601fd42 | 633 | char *kconfig; |
166750bc AN |
634 | struct extern_desc *externs; |
635 | int nr_extern; | |
81bfdd08 | 636 | int kconfig_map_idx; |
166750bc | 637 | |
52d3352e | 638 | bool loaded; |
c3c55696 | 639 | bool has_subcalls; |
25bbbd7a | 640 | bool has_rodata; |
a5b8bd47 | 641 | |
e2fa0156 AS |
642 | struct bpf_gen *gen_loader; |
643 | ||
29a30ff5 AN |
644 | /* Information when doing ELF related work. Only valid if efile.elf is not NULL */ |
645 | struct elf_state efile; | |
10931d24 | 646 | |
8a138aed | 647 | struct btf *btf; |
0f7515ca AN |
648 | struct btf_ext *btf_ext; |
649 | ||
a6ed02ca KS |
650 | /* Parse and load BTF vmlinux if any of the programs in the object need |
651 | * it at load time. | |
652 | */ | |
653 | struct btf *btf_vmlinux; | |
1373ff59 SC |
654 | /* Path to the custom BTF to be used for BPF CO-RE relocations as an |
655 | * override for vmlinux BTF. | |
656 | */ | |
657 | char *btf_custom_path; | |
0f7515ca AN |
658 | /* vmlinux BTF override for CO-RE relocations */ |
659 | struct btf *btf_vmlinux_override; | |
4f33a53d AN |
660 | /* Lazily initialized kernel module BTFs */ |
661 | struct module_btf *btf_modules; | |
662 | bool btf_modules_loaded; | |
663 | size_t btf_module_cnt; | |
664 | size_t btf_module_cap; | |
8a138aed | 665 | |
e0e3ea88 AN |
666 | /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */ |
667 | char *log_buf; | |
668 | size_t log_size; | |
669 | __u32 log_level; | |
670 | ||
9dbe6015 KKD |
671 | int *fd_array; |
672 | size_t fd_array_cap; | |
673 | size_t fd_array_cnt; | |
674 | ||
2e4913e0 AN |
675 | struct usdt_manager *usdt_man; |
676 | ||
1a5e3fb1 WN |
677 | char path[]; |
678 | }; | |
1a5e3fb1 | 679 | |
88a82120 AN |
680 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off); |
681 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off); | |
682 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx); | |
683 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name); | |
ad23b723 | 684 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn); |
88a82120 AN |
685 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn); |
686 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn); | |
ad23b723 AN |
687 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx); |
688 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx); | |
88a82120 | 689 | |
29cd77f4 | 690 | void bpf_program__unload(struct bpf_program *prog) |
55cffde2 WN |
691 | { |
692 | if (!prog) | |
693 | return; | |
694 | ||
cf90a20d | 695 | zclose(prog->fd); |
2993e051 | 696 | |
2993e051 | 697 | zfree(&prog->func_info); |
07a09d1b | 698 | zfree(&prog->line_info); |
55cffde2 WN |
699 | } |
700 | ||
a5b8bd47 WN |
701 | static void bpf_program__exit(struct bpf_program *prog) |
702 | { | |
703 | if (!prog) | |
704 | return; | |
705 | ||
55cffde2 | 706 | bpf_program__unload(prog); |
88cda1c9 | 707 | zfree(&prog->name); |
52109584 | 708 | zfree(&prog->sec_name); |
a5b8bd47 | 709 | zfree(&prog->insns); |
34090915 WN |
710 | zfree(&prog->reloc_desc); |
711 | ||
712 | prog->nr_reloc = 0; | |
a5b8bd47 | 713 | prog->insns_cnt = 0; |
c1122392 | 714 | prog->sec_idx = -1; |
a5b8bd47 WN |
715 | } |
716 | ||
c3c55696 AN |
717 | static bool insn_is_subprog_call(const struct bpf_insn *insn) |
718 | { | |
719 | return BPF_CLASS(insn->code) == BPF_JMP && | |
720 | BPF_OP(insn->code) == BPF_CALL && | |
721 | BPF_SRC(insn->code) == BPF_K && | |
722 | insn->src_reg == BPF_PSEUDO_CALL && | |
723 | insn->dst_reg == 0 && | |
724 | insn->off == 0; | |
725 | } | |
726 | ||
aa0b8d43 MKL |
727 | static bool is_call_insn(const struct bpf_insn *insn) |
728 | { | |
729 | return insn->code == (BPF_JMP | BPF_CALL); | |
730 | } | |
731 | ||
53eddb5e YS |
732 | static bool insn_is_pseudo_func(struct bpf_insn *insn) |
733 | { | |
aa0b8d43 | 734 | return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC; |
53eddb5e YS |
735 | } |
736 | ||
a5b8bd47 | 737 | static int |
c3c55696 AN |
738 | bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog, |
739 | const char *name, size_t sec_idx, const char *sec_name, | |
740 | size_t sec_off, void *insn_data, size_t insn_data_sz) | |
a5b8bd47 | 741 | { |
c1122392 AN |
742 | if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) { |
743 | pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n", | |
744 | sec_name, name, sec_off, insn_data_sz); | |
a5b8bd47 WN |
745 | return -EINVAL; |
746 | } | |
747 | ||
1ad9cbb8 | 748 | memset(prog, 0, sizeof(*prog)); |
c3c55696 AN |
749 | prog->obj = obj; |
750 | ||
c1122392 AN |
751 | prog->sec_idx = sec_idx; |
752 | prog->sec_insn_off = sec_off / BPF_INSN_SZ; | |
753 | prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ; | |
754 | /* insns_cnt can later be increased by appending used subprograms */ | |
755 | prog->insns_cnt = prog->sec_insn_cnt; | |
a5b8bd47 | 756 | |
c1122392 | 757 | prog->type = BPF_PROG_TYPE_UNSPEC; |
cf90a20d | 758 | prog->fd = -1; |
a3820c48 AN |
759 | |
760 | /* libbpf's convention for SEC("?abc...") is that it's just like | |
761 | * SEC("abc...") but the corresponding bpf_program starts out with | |
762 | * autoload set to false. | |
763 | */ | |
764 | if (sec_name[0] == '?') { | |
765 | prog->autoload = false; | |
766 | /* from now on forget there was ? in section name */ | |
767 | sec_name++; | |
768 | } else { | |
769 | prog->autoload = true; | |
770 | } | |
a5b8bd47 | 771 | |
43cb8cba HL |
772 | prog->autoattach = true; |
773 | ||
b3ce9079 AN |
774 | /* inherit object's log_level */ |
775 | prog->log_level = obj->log_level; | |
776 | ||
52109584 AN |
777 | prog->sec_name = strdup(sec_name); |
778 | if (!prog->sec_name) | |
c1122392 AN |
779 | goto errout; |
780 | ||
781 | prog->name = strdup(name); | |
782 | if (!prog->name) | |
a5b8bd47 | 783 | goto errout; |
a5b8bd47 | 784 | |
c1122392 AN |
785 | prog->insns = malloc(insn_data_sz); |
786 | if (!prog->insns) | |
a5b8bd47 | 787 | goto errout; |
c1122392 | 788 | memcpy(prog->insns, insn_data, insn_data_sz); |
a5b8bd47 WN |
789 | |
790 | return 0; | |
791 | errout: | |
c1122392 | 792 | pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name); |
a5b8bd47 WN |
793 | bpf_program__exit(prog); |
794 | return -ENOMEM; | |
795 | } | |
796 | ||
797 | static int | |
c1122392 AN |
798 | bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data, |
799 | const char *sec_name, int sec_idx) | |
a5b8bd47 | 800 | { |
6245947c | 801 | Elf_Data *symbols = obj->efile.symbols; |
c1122392 AN |
802 | struct bpf_program *prog, *progs; |
803 | void *data = sec_data->d_buf; | |
6245947c AN |
804 | size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms; |
805 | int nr_progs, err, i; | |
c1122392 | 806 | const char *name; |
ad23b723 | 807 | Elf64_Sym *sym; |
a5b8bd47 WN |
808 | |
809 | progs = obj->programs; | |
810 | nr_progs = obj->nr_programs; | |
ad23b723 | 811 | nr_syms = symbols->d_size / sizeof(Elf64_Sym); |
a5b8bd47 | 812 | |
6245947c | 813 | for (i = 0; i < nr_syms; i++) { |
ad23b723 AN |
814 | sym = elf_sym_by_idx(obj, i); |
815 | ||
816 | if (sym->st_shndx != sec_idx) | |
6245947c | 817 | continue; |
ad23b723 | 818 | if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC) |
6245947c | 819 | continue; |
88cda1c9 | 820 | |
ad23b723 AN |
821 | prog_sz = sym->st_size; |
822 | sec_off = sym->st_value; | |
88cda1c9 | 823 | |
ad23b723 | 824 | name = elf_sym_str(obj, sym->st_name); |
c1122392 AN |
825 | if (!name) { |
826 | pr_warn("sec '%s': failed to get symbol name for offset %zu\n", | |
827 | sec_name, sec_off); | |
828 | return -LIBBPF_ERRNO__FORMAT; | |
829 | } | |
88cda1c9 | 830 | |
c1122392 AN |
831 | if (sec_off + prog_sz > sec_sz) { |
832 | pr_warn("sec '%s': program at offset %zu crosses section boundary\n", | |
833 | sec_name, sec_off); | |
834 | return -LIBBPF_ERRNO__FORMAT; | |
835 | } | |
88cda1c9 | 836 | |
ad23b723 | 837 | if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) { |
513f485c AN |
838 | pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name); |
839 | return -ENOTSUP; | |
840 | } | |
841 | ||
c3c55696 AN |
842 | pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n", |
843 | sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz); | |
88cda1c9 | 844 | |
c3c55696 | 845 | progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs)); |
c1122392 AN |
846 | if (!progs) { |
847 | /* | |
848 | * In this case the original obj->programs | |
849 | * is still valid, so don't need special treat for | |
850 | * bpf_close_object(). | |
851 | */ | |
852 | pr_warn("sec '%s': failed to alloc memory for new program '%s'\n", | |
853 | sec_name, name); | |
854 | return -ENOMEM; | |
88cda1c9 | 855 | } |
c1122392 | 856 | obj->programs = progs; |
88cda1c9 | 857 | |
c1122392 | 858 | prog = &progs[nr_progs]; |
9a94f277 | 859 | |
c3c55696 AN |
860 | err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name, |
861 | sec_off, data + sec_off, prog_sz); | |
c1122392 AN |
862 | if (err) |
863 | return err; | |
9a94f277 | 864 | |
e5670fa0 AN |
865 | /* if function is a global/weak symbol, but has restricted |
866 | * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC | |
867 | * as static to enable more permissive BPF verification mode | |
868 | * with more outside context available to BPF verifier | |
aea28a60 | 869 | */ |
ad23b723 AN |
870 | if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL |
871 | && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN | |
872 | || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL)) | |
aea28a60 AN |
873 | prog->mark_btf_static = true; |
874 | ||
c1122392 AN |
875 | nr_progs++; |
876 | obj->nr_programs = nr_progs; | |
88cda1c9 MKL |
877 | } |
878 | ||
879 | return 0; | |
880 | } | |
881 | ||
590a0088 MKL |
882 | static const struct btf_member * |
883 | find_member_by_offset(const struct btf_type *t, __u32 bit_offset) | |
884 | { | |
885 | struct btf_member *m; | |
886 | int i; | |
887 | ||
888 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
889 | if (btf_member_bit_offset(t, i) == bit_offset) | |
890 | return m; | |
891 | } | |
892 | ||
893 | return NULL; | |
894 | } | |
895 | ||
896 | static const struct btf_member * | |
897 | find_member_by_name(const struct btf *btf, const struct btf_type *t, | |
898 | const char *name) | |
899 | { | |
900 | struct btf_member *m; | |
901 | int i; | |
902 | ||
903 | for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) { | |
904 | if (!strcmp(btf__name_by_offset(btf, m->name_off), name)) | |
905 | return m; | |
906 | } | |
907 | ||
908 | return NULL; | |
909 | } | |
910 | ||
911 | #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_" | |
a6ed02ca KS |
912 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
913 | const char *name, __u32 kind); | |
590a0088 MKL |
914 | |
915 | static int | |
916 | find_struct_ops_kern_types(const struct btf *btf, const char *tname, | |
917 | const struct btf_type **type, __u32 *type_id, | |
918 | const struct btf_type **vtype, __u32 *vtype_id, | |
919 | const struct btf_member **data_member) | |
920 | { | |
921 | const struct btf_type *kern_type, *kern_vtype; | |
922 | const struct btf_member *kern_data_member; | |
923 | __s32 kern_vtype_id, kern_type_id; | |
590a0088 MKL |
924 | __u32 i; |
925 | ||
926 | kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT); | |
927 | if (kern_type_id < 0) { | |
928 | pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n", | |
929 | tname); | |
930 | return kern_type_id; | |
931 | } | |
932 | kern_type = btf__type_by_id(btf, kern_type_id); | |
933 | ||
934 | /* Find the corresponding "map_value" type that will be used | |
935 | * in map_update(BPF_MAP_TYPE_STRUCT_OPS). For example, | |
936 | * find "struct bpf_struct_ops_tcp_congestion_ops" from the | |
937 | * btf_vmlinux. | |
938 | */ | |
a6ed02ca KS |
939 | kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX, |
940 | tname, BTF_KIND_STRUCT); | |
590a0088 | 941 | if (kern_vtype_id < 0) { |
a6ed02ca KS |
942 | pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n", |
943 | STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
944 | return kern_vtype_id; |
945 | } | |
946 | kern_vtype = btf__type_by_id(btf, kern_vtype_id); | |
947 | ||
948 | /* Find "struct tcp_congestion_ops" from | |
949 | * struct bpf_struct_ops_tcp_congestion_ops { | |
950 | * [ ... ] | |
951 | * struct tcp_congestion_ops data; | |
952 | * } | |
953 | */ | |
954 | kern_data_member = btf_members(kern_vtype); | |
955 | for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) { | |
956 | if (kern_data_member->type == kern_type_id) | |
957 | break; | |
958 | } | |
959 | if (i == btf_vlen(kern_vtype)) { | |
a6ed02ca KS |
960 | pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n", |
961 | tname, STRUCT_OPS_VALUE_PREFIX, tname); | |
590a0088 MKL |
962 | return -EINVAL; |
963 | } | |
964 | ||
965 | *type = kern_type; | |
966 | *type_id = kern_type_id; | |
967 | *vtype = kern_vtype; | |
968 | *vtype_id = kern_vtype_id; | |
969 | *data_member = kern_data_member; | |
970 | ||
971 | return 0; | |
972 | } | |
973 | ||
974 | static bool bpf_map__is_struct_ops(const struct bpf_map *map) | |
975 | { | |
976 | return map->def.type == BPF_MAP_TYPE_STRUCT_OPS; | |
977 | } | |
978 | ||
979 | /* Init the map's fields that depend on kern_btf */ | |
980 | static int bpf_map__init_kern_struct_ops(struct bpf_map *map, | |
981 | const struct btf *btf, | |
982 | const struct btf *kern_btf) | |
983 | { | |
984 | const struct btf_member *member, *kern_member, *kern_data_member; | |
985 | const struct btf_type *type, *kern_type, *kern_vtype; | |
986 | __u32 i, kern_type_id, kern_vtype_id, kern_data_off; | |
987 | struct bpf_struct_ops *st_ops; | |
988 | void *data, *kern_data; | |
989 | const char *tname; | |
990 | int err; | |
991 | ||
992 | st_ops = map->st_ops; | |
993 | type = st_ops->type; | |
994 | tname = st_ops->tname; | |
995 | err = find_struct_ops_kern_types(kern_btf, tname, | |
996 | &kern_type, &kern_type_id, | |
997 | &kern_vtype, &kern_vtype_id, | |
998 | &kern_data_member); | |
999 | if (err) | |
1000 | return err; | |
1001 | ||
1002 | pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n", | |
1003 | map->name, st_ops->type_id, kern_type_id, kern_vtype_id); | |
1004 | ||
1005 | map->def.value_size = kern_vtype->size; | |
1006 | map->btf_vmlinux_value_type_id = kern_vtype_id; | |
1007 | ||
1008 | st_ops->kern_vdata = calloc(1, kern_vtype->size); | |
1009 | if (!st_ops->kern_vdata) | |
1010 | return -ENOMEM; | |
1011 | ||
1012 | data = st_ops->data; | |
1013 | kern_data_off = kern_data_member->offset / 8; | |
1014 | kern_data = st_ops->kern_vdata + kern_data_off; | |
1015 | ||
1016 | member = btf_members(type); | |
1017 | for (i = 0; i < btf_vlen(type); i++, member++) { | |
1018 | const struct btf_type *mtype, *kern_mtype; | |
1019 | __u32 mtype_id, kern_mtype_id; | |
1020 | void *mdata, *kern_mdata; | |
1021 | __s64 msize, kern_msize; | |
1022 | __u32 moff, kern_moff; | |
1023 | __u32 kern_member_idx; | |
1024 | const char *mname; | |
1025 | ||
1026 | mname = btf__name_by_offset(btf, member->name_off); | |
1027 | kern_member = find_member_by_name(kern_btf, kern_type, mname); | |
1028 | if (!kern_member) { | |
1029 | pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n", | |
1030 | map->name, mname); | |
1031 | return -ENOTSUP; | |
1032 | } | |
1033 | ||
1034 | kern_member_idx = kern_member - btf_members(kern_type); | |
1035 | if (btf_member_bitfield_size(type, i) || | |
1036 | btf_member_bitfield_size(kern_type, kern_member_idx)) { | |
1037 | pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n", | |
1038 | map->name, mname); | |
1039 | return -ENOTSUP; | |
1040 | } | |
1041 | ||
1042 | moff = member->offset / 8; | |
1043 | kern_moff = kern_member->offset / 8; | |
1044 | ||
1045 | mdata = data + moff; | |
1046 | kern_mdata = kern_data + kern_moff; | |
1047 | ||
1048 | mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id); | |
1049 | kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type, | |
1050 | &kern_mtype_id); | |
1051 | if (BTF_INFO_KIND(mtype->info) != | |
1052 | BTF_INFO_KIND(kern_mtype->info)) { | |
1053 | pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n", | |
1054 | map->name, mname, BTF_INFO_KIND(mtype->info), | |
1055 | BTF_INFO_KIND(kern_mtype->info)); | |
1056 | return -ENOTSUP; | |
1057 | } | |
1058 | ||
1059 | if (btf_is_ptr(mtype)) { | |
1060 | struct bpf_program *prog; | |
1061 | ||
d2836ddd MKL |
1062 | prog = st_ops->progs[i]; |
1063 | if (!prog) | |
1064 | continue; | |
1065 | ||
590a0088 MKL |
1066 | kern_mtype = skip_mods_and_typedefs(kern_btf, |
1067 | kern_mtype->type, | |
1068 | &kern_mtype_id); | |
d2836ddd MKL |
1069 | |
1070 | /* mtype->type must be a func_proto which was | |
1071 | * guaranteed in bpf_object__collect_st_ops_relos(), | |
1072 | * so only check kern_mtype for func_proto here. | |
1073 | */ | |
1074 | if (!btf_is_func_proto(kern_mtype)) { | |
1075 | pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n", | |
590a0088 MKL |
1076 | map->name, mname); |
1077 | return -ENOTSUP; | |
1078 | } | |
1079 | ||
590a0088 MKL |
1080 | prog->attach_btf_id = kern_type_id; |
1081 | prog->expected_attach_type = kern_member_idx; | |
1082 | ||
1083 | st_ops->kern_func_off[i] = kern_data_off + kern_moff; | |
1084 | ||
1085 | pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n", | |
1086 | map->name, mname, prog->name, moff, | |
1087 | kern_moff); | |
1088 | ||
1089 | continue; | |
1090 | } | |
1091 | ||
1092 | msize = btf__resolve_size(btf, mtype_id); | |
1093 | kern_msize = btf__resolve_size(kern_btf, kern_mtype_id); | |
1094 | if (msize < 0 || kern_msize < 0 || msize != kern_msize) { | |
1095 | pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n", | |
1096 | map->name, mname, (ssize_t)msize, | |
1097 | (ssize_t)kern_msize); | |
1098 | return -ENOTSUP; | |
1099 | } | |
1100 | ||
1101 | pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n", | |
1102 | map->name, mname, (unsigned int)msize, | |
1103 | moff, kern_moff); | |
1104 | memcpy(kern_mdata, mdata, msize); | |
1105 | } | |
1106 | ||
1107 | return 0; | |
1108 | } | |
1109 | ||
1110 | static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj) | |
1111 | { | |
590a0088 MKL |
1112 | struct bpf_map *map; |
1113 | size_t i; | |
1114 | int err; | |
1115 | ||
1116 | for (i = 0; i < obj->nr_maps; i++) { | |
1117 | map = &obj->maps[i]; | |
1118 | ||
1119 | if (!bpf_map__is_struct_ops(map)) | |
1120 | continue; | |
1121 | ||
a6ed02ca KS |
1122 | err = bpf_map__init_kern_struct_ops(map, obj->btf, |
1123 | obj->btf_vmlinux); | |
1124 | if (err) | |
590a0088 | 1125 | return err; |
590a0088 MKL |
1126 | } |
1127 | ||
590a0088 MKL |
1128 | return 0; |
1129 | } | |
1130 | ||
809a69d6 KFL |
1131 | static int init_struct_ops_maps(struct bpf_object *obj, const char *sec_name, |
1132 | int shndx, Elf_Data *data, __u32 map_flags) | |
590a0088 MKL |
1133 | { |
1134 | const struct btf_type *type, *datasec; | |
1135 | const struct btf_var_secinfo *vsi; | |
1136 | struct bpf_struct_ops *st_ops; | |
1137 | const char *tname, *var_name; | |
1138 | __s32 type_id, datasec_id; | |
1139 | const struct btf *btf; | |
1140 | struct bpf_map *map; | |
1141 | __u32 i; | |
1142 | ||
809a69d6 | 1143 | if (shndx == -1) |
590a0088 MKL |
1144 | return 0; |
1145 | ||
1146 | btf = obj->btf; | |
809a69d6 | 1147 | datasec_id = btf__find_by_name_kind(btf, sec_name, |
590a0088 MKL |
1148 | BTF_KIND_DATASEC); |
1149 | if (datasec_id < 0) { | |
1150 | pr_warn("struct_ops init: DATASEC %s not found\n", | |
809a69d6 | 1151 | sec_name); |
590a0088 MKL |
1152 | return -EINVAL; |
1153 | } | |
1154 | ||
1155 | datasec = btf__type_by_id(btf, datasec_id); | |
1156 | vsi = btf_var_secinfos(datasec); | |
1157 | for (i = 0; i < btf_vlen(datasec); i++, vsi++) { | |
1158 | type = btf__type_by_id(obj->btf, vsi->type); | |
1159 | var_name = btf__name_by_offset(obj->btf, type->name_off); | |
1160 | ||
1161 | type_id = btf__resolve_type(obj->btf, vsi->type); | |
1162 | if (type_id < 0) { | |
1163 | pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n", | |
809a69d6 | 1164 | vsi->type, sec_name); |
590a0088 MKL |
1165 | return -EINVAL; |
1166 | } | |
1167 | ||
1168 | type = btf__type_by_id(obj->btf, type_id); | |
1169 | tname = btf__name_by_offset(obj->btf, type->name_off); | |
1170 | if (!tname[0]) { | |
1171 | pr_warn("struct_ops init: anonymous type is not supported\n"); | |
1172 | return -ENOTSUP; | |
1173 | } | |
1174 | if (!btf_is_struct(type)) { | |
1175 | pr_warn("struct_ops init: %s is not a struct\n", tname); | |
1176 | return -EINVAL; | |
1177 | } | |
1178 | ||
1179 | map = bpf_object__add_map(obj); | |
1180 | if (IS_ERR(map)) | |
1181 | return PTR_ERR(map); | |
1182 | ||
809a69d6 | 1183 | map->sec_idx = shndx; |
590a0088 MKL |
1184 | map->sec_offset = vsi->offset; |
1185 | map->name = strdup(var_name); | |
1186 | if (!map->name) | |
1187 | return -ENOMEM; | |
1188 | ||
1189 | map->def.type = BPF_MAP_TYPE_STRUCT_OPS; | |
1190 | map->def.key_size = sizeof(int); | |
1191 | map->def.value_size = type->size; | |
1192 | map->def.max_entries = 1; | |
809a69d6 | 1193 | map->def.map_flags = map_flags; |
590a0088 MKL |
1194 | |
1195 | map->st_ops = calloc(1, sizeof(*map->st_ops)); | |
1196 | if (!map->st_ops) | |
1197 | return -ENOMEM; | |
1198 | st_ops = map->st_ops; | |
1199 | st_ops->data = malloc(type->size); | |
1200 | st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs)); | |
1201 | st_ops->kern_func_off = malloc(btf_vlen(type) * | |
1202 | sizeof(*st_ops->kern_func_off)); | |
1203 | if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off) | |
1204 | return -ENOMEM; | |
1205 | ||
809a69d6 | 1206 | if (vsi->offset + type->size > data->d_size) { |
590a0088 | 1207 | pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n", |
809a69d6 | 1208 | var_name, sec_name); |
590a0088 MKL |
1209 | return -EINVAL; |
1210 | } | |
1211 | ||
1212 | memcpy(st_ops->data, | |
809a69d6 | 1213 | data->d_buf + vsi->offset, |
590a0088 MKL |
1214 | type->size); |
1215 | st_ops->tname = tname; | |
1216 | st_ops->type = type; | |
1217 | st_ops->type_id = type_id; | |
1218 | ||
1219 | pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n", | |
1220 | tname, type_id, var_name, vsi->offset); | |
1221 | } | |
1222 | ||
1223 | return 0; | |
1224 | } | |
1225 | ||
809a69d6 KFL |
1226 | static int bpf_object_init_struct_ops(struct bpf_object *obj) |
1227 | { | |
1228 | int err; | |
1229 | ||
1230 | err = init_struct_ops_maps(obj, STRUCT_OPS_SEC, obj->efile.st_ops_shndx, | |
1231 | obj->efile.st_ops_data, 0); | |
1232 | err = err ?: init_struct_ops_maps(obj, STRUCT_OPS_LINK_SEC, | |
1233 | obj->efile.st_ops_link_shndx, | |
1234 | obj->efile.st_ops_link_data, | |
1235 | BPF_F_LINK); | |
1236 | return err; | |
1237 | } | |
1238 | ||
6c956392 | 1239 | static struct bpf_object *bpf_object__new(const char *path, |
5e61f270 | 1240 | const void *obj_buf, |
2ce8450e AN |
1241 | size_t obj_buf_sz, |
1242 | const char *obj_name) | |
1a5e3fb1 WN |
1243 | { |
1244 | struct bpf_object *obj; | |
d859900c | 1245 | char *end; |
1a5e3fb1 WN |
1246 | |
1247 | obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1); | |
1248 | if (!obj) { | |
be18010e | 1249 | pr_warn("alloc memory failed for %s\n", path); |
6371ca3b | 1250 | return ERR_PTR(-ENOMEM); |
1a5e3fb1 WN |
1251 | } |
1252 | ||
1253 | strcpy(obj->path, path); | |
2ce8450e | 1254 | if (obj_name) { |
9fc205b4 | 1255 | libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name)); |
2ce8450e AN |
1256 | } else { |
1257 | /* Using basename() GNU version which doesn't modify arg. */ | |
9fc205b4 | 1258 | libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name)); |
2ce8450e AN |
1259 | end = strchr(obj->name, '.'); |
1260 | if (end) | |
1261 | *end = 0; | |
1262 | } | |
6c956392 | 1263 | |
d859900c | 1264 | obj->efile.fd = -1; |
6c956392 | 1265 | /* |
76e1022b | 1266 | * Caller of this function should also call |
6c956392 WN |
1267 | * bpf_object__elf_finish() after data collection to return |
1268 | * obj_buf to user. If not, we should duplicate the buffer to | |
1269 | * avoid user freeing them before elf finish. | |
1270 | */ | |
1271 | obj->efile.obj_buf = obj_buf; | |
1272 | obj->efile.obj_buf_sz = obj_buf_sz; | |
abd29c93 | 1273 | obj->efile.btf_maps_shndx = -1; |
590a0088 | 1274 | obj->efile.st_ops_shndx = -1; |
809a69d6 | 1275 | obj->efile.st_ops_link_shndx = -1; |
81bfdd08 | 1276 | obj->kconfig_map_idx = -1; |
6c956392 | 1277 | |
5e61f270 | 1278 | obj->kern_version = get_kernel_version(); |
52d3352e | 1279 | obj->loaded = false; |
9a208eff | 1280 | |
1a5e3fb1 WN |
1281 | return obj; |
1282 | } | |
1283 | ||
1284 | static void bpf_object__elf_finish(struct bpf_object *obj) | |
1285 | { | |
29a30ff5 | 1286 | if (!obj->efile.elf) |
1a5e3fb1 WN |
1287 | return; |
1288 | ||
b71a2ebf GC |
1289 | elf_end(obj->efile.elf); |
1290 | obj->efile.elf = NULL; | |
bec7d68c | 1291 | obj->efile.symbols = NULL; |
590a0088 | 1292 | obj->efile.st_ops_data = NULL; |
809a69d6 | 1293 | obj->efile.st_ops_link_data = NULL; |
b62f06e8 | 1294 | |
25bbbd7a AN |
1295 | zfree(&obj->efile.secs); |
1296 | obj->efile.sec_cnt = 0; | |
1a5e3fb1 | 1297 | zclose(obj->efile.fd); |
6c956392 WN |
1298 | obj->efile.obj_buf = NULL; |
1299 | obj->efile.obj_buf_sz = 0; | |
1a5e3fb1 WN |
1300 | } |
1301 | ||
1302 | static int bpf_object__elf_init(struct bpf_object *obj) | |
1303 | { | |
ad23b723 | 1304 | Elf64_Ehdr *ehdr; |
1a5e3fb1 | 1305 | int err = 0; |
ad23b723 | 1306 | Elf *elf; |
1a5e3fb1 | 1307 | |
29a30ff5 | 1308 | if (obj->efile.elf) { |
88a82120 | 1309 | pr_warn("elf: init internal error\n"); |
6371ca3b | 1310 | return -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1311 | } |
1312 | ||
6c956392 | 1313 | if (obj->efile.obj_buf_sz > 0) { |
146bf811 | 1314 | /* obj_buf should have been validated by bpf_object__open_mem(). */ |
ad23b723 | 1315 | elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz); |
6c956392 | 1316 | } else { |
92274e24 | 1317 | obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC); |
6c956392 | 1318 | if (obj->efile.fd < 0) { |
be5c5d4e | 1319 | char errmsg[STRERR_BUFSIZE], *cp; |
1ce6a9fc | 1320 | |
be5c5d4e AN |
1321 | err = -errno; |
1322 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
88a82120 | 1323 | pr_warn("elf: failed to open %s: %s\n", obj->path, cp); |
be5c5d4e | 1324 | return err; |
6c956392 WN |
1325 | } |
1326 | ||
ad23b723 | 1327 | elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL); |
1a5e3fb1 WN |
1328 | } |
1329 | ||
ad23b723 | 1330 | if (!elf) { |
88a82120 | 1331 | pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1332 | err = -LIBBPF_ERRNO__LIBELF; |
1a5e3fb1 WN |
1333 | goto errout; |
1334 | } | |
1335 | ||
ad23b723 AN |
1336 | obj->efile.elf = elf; |
1337 | ||
1338 | if (elf_kind(elf) != ELF_K_ELF) { | |
1339 | err = -LIBBPF_ERRNO__FORMAT; | |
1340 | pr_warn("elf: '%s' is not a proper ELF object\n", obj->path); | |
1341 | goto errout; | |
1342 | } | |
1343 | ||
1344 | if (gelf_getclass(elf) != ELFCLASS64) { | |
1345 | err = -LIBBPF_ERRNO__FORMAT; | |
1346 | pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path); | |
1347 | goto errout; | |
1348 | } | |
1349 | ||
1350 | obj->efile.ehdr = ehdr = elf64_getehdr(elf); | |
1351 | if (!obj->efile.ehdr) { | |
88a82120 | 1352 | pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1)); |
6371ca3b | 1353 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1354 | goto errout; |
1355 | } | |
1a5e3fb1 | 1356 | |
ad23b723 | 1357 | if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) { |
88a82120 AN |
1358 | pr_warn("elf: failed to get section names section index for %s: %s\n", |
1359 | obj->path, elf_errmsg(-1)); | |
1360 | err = -LIBBPF_ERRNO__FORMAT; | |
1361 | goto errout; | |
1362 | } | |
1363 | ||
70e79866 | 1364 | /* ELF is corrupted/truncated, avoid calling elf_strptr. */ |
ad23b723 | 1365 | if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) { |
88a82120 AN |
1366 | pr_warn("elf: failed to get section names strings from %s: %s\n", |
1367 | obj->path, elf_errmsg(-1)); | |
8f3f5792 NK |
1368 | err = -LIBBPF_ERRNO__FORMAT; |
1369 | goto errout; | |
88a82120 AN |
1370 | } |
1371 | ||
9b16137a | 1372 | /* Old LLVM set e_machine to EM_NONE */ |
ad23b723 | 1373 | if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) { |
88a82120 | 1374 | pr_warn("elf: %s is not a valid eBPF object file\n", obj->path); |
6371ca3b | 1375 | err = -LIBBPF_ERRNO__FORMAT; |
1a5e3fb1 WN |
1376 | goto errout; |
1377 | } | |
1378 | ||
1379 | return 0; | |
1380 | errout: | |
1381 | bpf_object__elf_finish(obj); | |
1382 | return err; | |
1383 | } | |
1384 | ||
12ef5634 | 1385 | static int bpf_object__check_endianness(struct bpf_object *obj) |
cc4228d5 | 1386 | { |
3930198d | 1387 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
ad23b723 | 1388 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB) |
12ef5634 | 1389 | return 0; |
3930198d | 1390 | #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
ad23b723 | 1391 | if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB) |
12ef5634 AN |
1392 | return 0; |
1393 | #else | |
1394 | # error "Unrecognized __BYTE_ORDER__" | |
1395 | #endif | |
88a82120 | 1396 | pr_warn("elf: endianness mismatch in %s.\n", obj->path); |
6371ca3b | 1397 | return -LIBBPF_ERRNO__ENDIAN; |
cc4228d5 WN |
1398 | } |
1399 | ||
cb1e5e96 | 1400 | static int |
399dc65e | 1401 | bpf_object__init_license(struct bpf_object *obj, void *data, size_t size) |
cb1e5e96 | 1402 | { |
35a85550 SHY |
1403 | if (!data) { |
1404 | pr_warn("invalid license section in %s\n", obj->path); | |
1405 | return -LIBBPF_ERRNO__FORMAT; | |
1406 | } | |
f9798239 AN |
1407 | /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't |
1408 | * go over allowed ELF data section buffer | |
1409 | */ | |
1410 | libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license))); | |
cb1e5e96 WN |
1411 | pr_debug("license of %s is %s\n", obj->path, obj->license); |
1412 | return 0; | |
1413 | } | |
1414 | ||
54b8625c JF |
1415 | static int |
1416 | bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size) | |
1417 | { | |
1418 | __u32 kver; | |
1419 | ||
35a85550 | 1420 | if (!data || size != sizeof(kver)) { |
be18010e | 1421 | pr_warn("invalid kver section in %s\n", obj->path); |
54b8625c JF |
1422 | return -LIBBPF_ERRNO__FORMAT; |
1423 | } | |
1424 | memcpy(&kver, data, sizeof(kver)); | |
1425 | obj->kern_version = kver; | |
1426 | pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version); | |
1427 | return 0; | |
1428 | } | |
1429 | ||
addb9fc9 NS |
1430 | static bool bpf_map_type__is_map_in_map(enum bpf_map_type type) |
1431 | { | |
1432 | if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1433 | type == BPF_MAP_TYPE_HASH_OF_MAPS) | |
1434 | return true; | |
1435 | return false; | |
1436 | } | |
1437 | ||
b96c07f3 | 1438 | static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size) |
1713d68b | 1439 | { |
25bbbd7a AN |
1440 | Elf_Data *data; |
1441 | Elf_Scn *scn; | |
1713d68b | 1442 | |
25bbbd7a | 1443 | if (!name) |
1713d68b | 1444 | return -EINVAL; |
88a82120 | 1445 | |
25bbbd7a AN |
1446 | scn = elf_sec_by_name(obj, name); |
1447 | data = elf_sec_data(obj, scn); | |
1448 | if (data) { | |
25bbbd7a | 1449 | *size = data->d_size; |
08894d9c | 1450 | return 0; /* found it */ |
1713d68b DB |
1451 | } |
1452 | ||
08894d9c | 1453 | return -ENOENT; |
1713d68b DB |
1454 | } |
1455 | ||
f33f742d | 1456 | static Elf64_Sym *find_elf_var_sym(const struct bpf_object *obj, const char *name) |
1713d68b DB |
1457 | { |
1458 | Elf_Data *symbols = obj->efile.symbols; | |
1459 | const char *sname; | |
1460 | size_t si; | |
1461 | ||
ad23b723 AN |
1462 | for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) { |
1463 | Elf64_Sym *sym = elf_sym_by_idx(obj, si); | |
1713d68b | 1464 | |
3a06ec0a AN |
1465 | if (ELF64_ST_TYPE(sym->st_info) != STT_OBJECT) |
1466 | continue; | |
1467 | ||
1468 | if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL && | |
1469 | ELF64_ST_BIND(sym->st_info) != STB_WEAK) | |
1713d68b DB |
1470 | continue; |
1471 | ||
ad23b723 | 1472 | sname = elf_sym_str(obj, sym->st_name); |
1713d68b | 1473 | if (!sname) { |
ad23b723 | 1474 | pr_warn("failed to get sym name string for var %s\n", name); |
f33f742d | 1475 | return ERR_PTR(-EIO); |
1713d68b | 1476 | } |
f33f742d AN |
1477 | if (strcmp(name, sname) == 0) |
1478 | return sym; | |
1713d68b DB |
1479 | } |
1480 | ||
f33f742d | 1481 | return ERR_PTR(-ENOENT); |
1713d68b DB |
1482 | } |
1483 | ||
bf829271 | 1484 | static struct bpf_map *bpf_object__add_map(struct bpf_object *obj) |
d859900c | 1485 | { |
69721203 AN |
1486 | struct bpf_map *map; |
1487 | int err; | |
bf829271 | 1488 | |
69721203 AN |
1489 | err = libbpf_ensure_mem((void **)&obj->maps, &obj->maps_cap, |
1490 | sizeof(*obj->maps), obj->nr_maps + 1); | |
1491 | if (err) | |
1492 | return ERR_PTR(err); | |
bf829271 | 1493 | |
69721203 | 1494 | map = &obj->maps[obj->nr_maps++]; |
ec41817b | 1495 | map->obj = obj; |
69721203 AN |
1496 | map->fd = -1; |
1497 | map->inner_map_fd = -1; | |
ec41817b | 1498 | map->autocreate = true; |
bf829271 | 1499 | |
69721203 | 1500 | return map; |
d859900c DB |
1501 | } |
1502 | ||
9d0a2331 | 1503 | static size_t bpf_map_mmap_sz(unsigned int value_sz, unsigned int max_entries) |
eba9c5f4 | 1504 | { |
9d0a2331 | 1505 | const long page_sz = sysconf(_SC_PAGE_SIZE); |
eba9c5f4 AN |
1506 | size_t map_sz; |
1507 | ||
9d0a2331 | 1508 | map_sz = (size_t)roundup(value_sz, 8) * max_entries; |
eba9c5f4 AN |
1509 | map_sz = roundup(map_sz, page_sz); |
1510 | return map_sz; | |
1511 | } | |
1512 | ||
9d0a2331 JK |
1513 | static int bpf_map_mmap_resize(struct bpf_map *map, size_t old_sz, size_t new_sz) |
1514 | { | |
1515 | void *mmaped; | |
1516 | ||
1517 | if (!map->mmaped) | |
1518 | return -EINVAL; | |
1519 | ||
1520 | if (old_sz == new_sz) | |
1521 | return 0; | |
1522 | ||
1523 | mmaped = mmap(NULL, new_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); | |
1524 | if (mmaped == MAP_FAILED) | |
1525 | return -errno; | |
1526 | ||
1527 | memcpy(mmaped, map->mmaped, min(old_sz, new_sz)); | |
1528 | munmap(map->mmaped, old_sz); | |
1529 | map->mmaped = mmaped; | |
1530 | return 0; | |
1531 | } | |
1532 | ||
aed65917 | 1533 | static char *internal_map_name(struct bpf_object *obj, const char *real_name) |
81bfdd08 | 1534 | { |
113e6b7e | 1535 | char map_name[BPF_OBJ_NAME_LEN], *p; |
aed65917 AN |
1536 | int pfx_len, sfx_len = max((size_t)7, strlen(real_name)); |
1537 | ||
1538 | /* This is one of the more confusing parts of libbpf for various | |
1539 | * reasons, some of which are historical. The original idea for naming | |
1540 | * internal names was to include as much of BPF object name prefix as | |
1541 | * possible, so that it can be distinguished from similar internal | |
1542 | * maps of a different BPF object. | |
1543 | * As an example, let's say we have bpf_object named 'my_object_name' | |
1544 | * and internal map corresponding to '.rodata' ELF section. The final | |
1545 | * map name advertised to user and to the kernel will be | |
1546 | * 'my_objec.rodata', taking first 8 characters of object name and | |
1547 | * entire 7 characters of '.rodata'. | |
1548 | * Somewhat confusingly, if internal map ELF section name is shorter | |
1549 | * than 7 characters, e.g., '.bss', we still reserve 7 characters | |
1550 | * for the suffix, even though we only have 4 actual characters, and | |
1551 | * resulting map will be called 'my_objec.bss', not even using all 15 | |
1552 | * characters allowed by the kernel. Oh well, at least the truncated | |
1553 | * object name is somewhat consistent in this case. But if the map | |
1554 | * name is '.kconfig', we'll still have entirety of '.kconfig' added | |
1555 | * (8 chars) and thus will be left with only first 7 characters of the | |
1556 | * object name ('my_obje'). Happy guessing, user, that the final map | |
1557 | * name will be "my_obje.kconfig". | |
1558 | * Now, with libbpf starting to support arbitrarily named .rodata.* | |
1559 | * and .data.* data sections, it's possible that ELF section name is | |
1560 | * longer than allowed 15 chars, so we now need to be careful to take | |
1561 | * only up to 15 first characters of ELF name, taking no BPF object | |
1562 | * name characters at all. So '.rodata.abracadabra' will result in | |
1563 | * '.rodata.abracad' kernel and user-visible name. | |
1564 | * We need to keep this convoluted logic intact for .data, .bss and | |
1565 | * .rodata maps, but for new custom .data.custom and .rodata.custom | |
1566 | * maps we use their ELF names as is, not prepending bpf_object name | |
1567 | * in front. We still need to truncate them to 15 characters for the | |
1568 | * kernel. Full name can be recovered for such maps by using DATASEC | |
1569 | * BTF type associated with such map's value type, though. | |
1570 | */ | |
1571 | if (sfx_len >= BPF_OBJ_NAME_LEN) | |
1572 | sfx_len = BPF_OBJ_NAME_LEN - 1; | |
1573 | ||
1574 | /* if there are two or more dots in map name, it's a custom dot map */ | |
1575 | if (strchr(real_name + 1, '.') != NULL) | |
1576 | pfx_len = 0; | |
1577 | else | |
1578 | pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name)); | |
81bfdd08 AN |
1579 | |
1580 | snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name, | |
aed65917 | 1581 | sfx_len, real_name); |
81bfdd08 | 1582 | |
113e6b7e THJ |
1583 | /* sanitise map name to characters allowed by kernel */ |
1584 | for (p = map_name; *p && p < map_name + sizeof(map_name); p++) | |
1585 | if (!isalnum(*p) && *p != '_' && *p != '.') | |
1586 | *p = '_'; | |
1587 | ||
81bfdd08 AN |
1588 | return strdup(map_name); |
1589 | } | |
1590 | ||
262cfb74 | 1591 | static int |
4fcac46c AN |
1592 | map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map); |
1593 | ||
1594 | /* Internal BPF map is mmap()'able only if at least one of corresponding | |
1595 | * DATASEC's VARs are to be exposed through BPF skeleton. I.e., it's a GLOBAL | |
1596 | * variable and it's not marked as __hidden (which turns it into, effectively, | |
1597 | * a STATIC variable). | |
1598 | */ | |
1599 | static bool map_is_mmapable(struct bpf_object *obj, struct bpf_map *map) | |
1600 | { | |
1601 | const struct btf_type *t, *vt; | |
1602 | struct btf_var_secinfo *vsi; | |
1603 | int i, n; | |
1604 | ||
1605 | if (!map->btf_value_type_id) | |
1606 | return false; | |
1607 | ||
1608 | t = btf__type_by_id(obj->btf, map->btf_value_type_id); | |
1609 | if (!btf_is_datasec(t)) | |
1610 | return false; | |
1611 | ||
1612 | vsi = btf_var_secinfos(t); | |
1613 | for (i = 0, n = btf_vlen(t); i < n; i++, vsi++) { | |
1614 | vt = btf__type_by_id(obj->btf, vsi->type); | |
1615 | if (!btf_is_var(vt)) | |
1616 | continue; | |
1617 | ||
1618 | if (btf_var(vt)->linkage != BTF_VAR_STATIC) | |
1619 | return true; | |
1620 | } | |
1621 | ||
1622 | return false; | |
1623 | } | |
262cfb74 | 1624 | |
d859900c | 1625 | static int |
bf829271 | 1626 | bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type, |
aed65917 | 1627 | const char *real_name, int sec_idx, void *data, size_t data_sz) |
d859900c | 1628 | { |
bf829271 AN |
1629 | struct bpf_map_def *def; |
1630 | struct bpf_map *map; | |
9d0a2331 | 1631 | size_t mmap_sz; |
eba9c5f4 | 1632 | int err; |
bf829271 AN |
1633 | |
1634 | map = bpf_object__add_map(obj); | |
1635 | if (IS_ERR(map)) | |
1636 | return PTR_ERR(map); | |
d859900c DB |
1637 | |
1638 | map->libbpf_type = type; | |
db48814b AN |
1639 | map->sec_idx = sec_idx; |
1640 | map->sec_offset = 0; | |
aed65917 AN |
1641 | map->real_name = strdup(real_name); |
1642 | map->name = internal_map_name(obj, real_name); | |
1643 | if (!map->real_name || !map->name) { | |
1644 | zfree(&map->real_name); | |
1645 | zfree(&map->name); | |
d859900c DB |
1646 | return -ENOMEM; |
1647 | } | |
1648 | ||
bf829271 | 1649 | def = &map->def; |
d859900c DB |
1650 | def->type = BPF_MAP_TYPE_ARRAY; |
1651 | def->key_size = sizeof(int); | |
eba9c5f4 | 1652 | def->value_size = data_sz; |
d859900c | 1653 | def->max_entries = 1; |
81bfdd08 | 1654 | def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG |
166750bc | 1655 | ? BPF_F_RDONLY_PROG : 0; |
4fcac46c AN |
1656 | |
1657 | /* failures are fine because of maps like .rodata.str1.1 */ | |
1658 | (void) map_fill_btf_type_info(obj, map); | |
1659 | ||
1660 | if (map_is_mmapable(obj, map)) | |
1661 | def->map_flags |= BPF_F_MMAPABLE; | |
7fe74b43 AN |
1662 | |
1663 | pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n", | |
81bfdd08 | 1664 | map->name, map->sec_idx, map->sec_offset, def->map_flags); |
7fe74b43 | 1665 | |
9d0a2331 JK |
1666 | mmap_sz = bpf_map_mmap_sz(map->def.value_size, map->def.max_entries); |
1667 | map->mmaped = mmap(NULL, mmap_sz, PROT_READ | PROT_WRITE, | |
eba9c5f4 AN |
1668 | MAP_SHARED | MAP_ANONYMOUS, -1, 0); |
1669 | if (map->mmaped == MAP_FAILED) { | |
1670 | err = -errno; | |
1671 | map->mmaped = NULL; | |
1672 | pr_warn("failed to alloc map '%s' content buffer: %d\n", | |
1673 | map->name, err); | |
aed65917 | 1674 | zfree(&map->real_name); |
eba9c5f4 AN |
1675 | zfree(&map->name); |
1676 | return err; | |
d859900c DB |
1677 | } |
1678 | ||
166750bc | 1679 | if (data) |
eba9c5f4 AN |
1680 | memcpy(map->mmaped, data, data_sz); |
1681 | ||
e1d1dc46 | 1682 | pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name); |
d859900c DB |
1683 | return 0; |
1684 | } | |
1685 | ||
bf829271 AN |
1686 | static int bpf_object__init_global_data_maps(struct bpf_object *obj) |
1687 | { | |
25bbbd7a | 1688 | struct elf_sec_desc *sec_desc; |
aed65917 | 1689 | const char *sec_name; |
25bbbd7a | 1690 | int err = 0, sec_idx; |
bf829271 | 1691 | |
bf829271 AN |
1692 | /* |
1693 | * Populate obj->maps with libbpf internal maps. | |
1694 | */ | |
25bbbd7a AN |
1695 | for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) { |
1696 | sec_desc = &obj->efile.secs[sec_idx]; | |
1697 | ||
47ea7417 | 1698 | /* Skip recognized sections with size 0. */ |
d4e6d684 | 1699 | if (!sec_desc->data || sec_desc->data->d_size == 0) |
47ea7417 JH |
1700 | continue; |
1701 | ||
25bbbd7a AN |
1702 | switch (sec_desc->sec_type) { |
1703 | case SEC_DATA: | |
aed65917 | 1704 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1705 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA, |
aed65917 | 1706 | sec_name, sec_idx, |
25bbbd7a AN |
1707 | sec_desc->data->d_buf, |
1708 | sec_desc->data->d_size); | |
1709 | break; | |
1710 | case SEC_RODATA: | |
1711 | obj->has_rodata = true; | |
aed65917 | 1712 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1713 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA, |
aed65917 | 1714 | sec_name, sec_idx, |
25bbbd7a AN |
1715 | sec_desc->data->d_buf, |
1716 | sec_desc->data->d_size); | |
1717 | break; | |
1718 | case SEC_BSS: | |
aed65917 | 1719 | sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx)); |
25bbbd7a | 1720 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS, |
aed65917 | 1721 | sec_name, sec_idx, |
25bbbd7a AN |
1722 | NULL, |
1723 | sec_desc->data->d_size); | |
1724 | break; | |
1725 | default: | |
1726 | /* skip */ | |
1727 | break; | |
1728 | } | |
bf829271 AN |
1729 | if (err) |
1730 | return err; | |
1731 | } | |
1732 | return 0; | |
1733 | } | |
1734 | ||
166750bc AN |
1735 | |
1736 | static struct extern_desc *find_extern_by_name(const struct bpf_object *obj, | |
1737 | const void *name) | |
1738 | { | |
1739 | int i; | |
1740 | ||
1741 | for (i = 0; i < obj->nr_extern; i++) { | |
1742 | if (strcmp(obj->externs[i].name, name) == 0) | |
1743 | return &obj->externs[i]; | |
1744 | } | |
1745 | return NULL; | |
1746 | } | |
1747 | ||
2e33efe3 AN |
1748 | static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val, |
1749 | char value) | |
166750bc | 1750 | { |
2e33efe3 AN |
1751 | switch (ext->kcfg.type) { |
1752 | case KCFG_BOOL: | |
166750bc | 1753 | if (value == 'm') { |
55d00c37 | 1754 | pr_warn("extern (kcfg) '%s': value '%c' implies tristate or char type\n", |
166750bc AN |
1755 | ext->name, value); |
1756 | return -EINVAL; | |
1757 | } | |
1758 | *(bool *)ext_val = value == 'y' ? true : false; | |
1759 | break; | |
2e33efe3 | 1760 | case KCFG_TRISTATE: |
166750bc AN |
1761 | if (value == 'y') |
1762 | *(enum libbpf_tristate *)ext_val = TRI_YES; | |
1763 | else if (value == 'm') | |
1764 | *(enum libbpf_tristate *)ext_val = TRI_MODULE; | |
1765 | else /* value == 'n' */ | |
1766 | *(enum libbpf_tristate *)ext_val = TRI_NO; | |
1767 | break; | |
2e33efe3 | 1768 | case KCFG_CHAR: |
166750bc AN |
1769 | *(char *)ext_val = value; |
1770 | break; | |
2e33efe3 AN |
1771 | case KCFG_UNKNOWN: |
1772 | case KCFG_INT: | |
1773 | case KCFG_CHAR_ARR: | |
166750bc | 1774 | default: |
55d00c37 | 1775 | pr_warn("extern (kcfg) '%s': value '%c' implies bool, tristate, or char type\n", |
166750bc AN |
1776 | ext->name, value); |
1777 | return -EINVAL; | |
1778 | } | |
1779 | ext->is_set = true; | |
1780 | return 0; | |
1781 | } | |
1782 | ||
2e33efe3 AN |
1783 | static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val, |
1784 | const char *value) | |
166750bc AN |
1785 | { |
1786 | size_t len; | |
1787 | ||
2e33efe3 | 1788 | if (ext->kcfg.type != KCFG_CHAR_ARR) { |
55d00c37 AN |
1789 | pr_warn("extern (kcfg) '%s': value '%s' implies char array type\n", |
1790 | ext->name, value); | |
166750bc AN |
1791 | return -EINVAL; |
1792 | } | |
1793 | ||
1794 | len = strlen(value); | |
1795 | if (value[len - 1] != '"') { | |
2e33efe3 | 1796 | pr_warn("extern (kcfg) '%s': invalid string config '%s'\n", |
166750bc AN |
1797 | ext->name, value); |
1798 | return -EINVAL; | |
1799 | } | |
1800 | ||
1801 | /* strip quotes */ | |
1802 | len -= 2; | |
2e33efe3 | 1803 | if (len >= ext->kcfg.sz) { |
55d00c37 | 1804 | pr_warn("extern (kcfg) '%s': long string '%s' of (%zu bytes) truncated to %d bytes\n", |
2e33efe3 AN |
1805 | ext->name, value, len, ext->kcfg.sz - 1); |
1806 | len = ext->kcfg.sz - 1; | |
166750bc AN |
1807 | } |
1808 | memcpy(ext_val, value + 1, len); | |
1809 | ext_val[len] = '\0'; | |
1810 | ext->is_set = true; | |
1811 | return 0; | |
1812 | } | |
1813 | ||
1814 | static int parse_u64(const char *value, __u64 *res) | |
1815 | { | |
1816 | char *value_end; | |
1817 | int err; | |
1818 | ||
1819 | errno = 0; | |
1820 | *res = strtoull(value, &value_end, 0); | |
1821 | if (errno) { | |
1822 | err = -errno; | |
1823 | pr_warn("failed to parse '%s' as integer: %d\n", value, err); | |
1824 | return err; | |
1825 | } | |
1826 | if (*value_end) { | |
1827 | pr_warn("failed to parse '%s' as integer completely\n", value); | |
1828 | return -EINVAL; | |
1829 | } | |
1830 | return 0; | |
1831 | } | |
1832 | ||
2e33efe3 | 1833 | static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v) |
166750bc | 1834 | { |
2e33efe3 | 1835 | int bit_sz = ext->kcfg.sz * 8; |
166750bc | 1836 | |
2e33efe3 | 1837 | if (ext->kcfg.sz == 8) |
166750bc AN |
1838 | return true; |
1839 | ||
1840 | /* Validate that value stored in u64 fits in integer of `ext->sz` | |
1841 | * bytes size without any loss of information. If the target integer | |
1842 | * is signed, we rely on the following limits of integer type of | |
1843 | * Y bits and subsequent transformation: | |
1844 | * | |
1845 | * -2^(Y-1) <= X <= 2^(Y-1) - 1 | |
1846 | * 0 <= X + 2^(Y-1) <= 2^Y - 1 | |
1847 | * 0 <= X + 2^(Y-1) < 2^Y | |
1848 | * | |
1849 | * For unsigned target integer, check that all the (64 - Y) bits are | |
1850 | * zero. | |
1851 | */ | |
2e33efe3 | 1852 | if (ext->kcfg.is_signed) |
166750bc AN |
1853 | return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz); |
1854 | else | |
1855 | return (v >> bit_sz) == 0; | |
1856 | } | |
1857 | ||
2e33efe3 AN |
1858 | static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val, |
1859 | __u64 value) | |
166750bc | 1860 | { |
55d00c37 AN |
1861 | if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR && |
1862 | ext->kcfg.type != KCFG_BOOL) { | |
1863 | pr_warn("extern (kcfg) '%s': value '%llu' implies integer, char, or boolean type\n", | |
7745ff98 | 1864 | ext->name, (unsigned long long)value); |
166750bc AN |
1865 | return -EINVAL; |
1866 | } | |
55d00c37 AN |
1867 | if (ext->kcfg.type == KCFG_BOOL && value > 1) { |
1868 | pr_warn("extern (kcfg) '%s': value '%llu' isn't boolean compatible\n", | |
1869 | ext->name, (unsigned long long)value); | |
1870 | return -EINVAL; | |
1871 | ||
1872 | } | |
2e33efe3 | 1873 | if (!is_kcfg_value_in_range(ext, value)) { |
55d00c37 | 1874 | pr_warn("extern (kcfg) '%s': value '%llu' doesn't fit in %d bytes\n", |
2e33efe3 | 1875 | ext->name, (unsigned long long)value, ext->kcfg.sz); |
166750bc AN |
1876 | return -ERANGE; |
1877 | } | |
2e33efe3 | 1878 | switch (ext->kcfg.sz) { |
e3ba8e4e KM |
1879 | case 1: |
1880 | *(__u8 *)ext_val = value; | |
1881 | break; | |
1882 | case 2: | |
1883 | *(__u16 *)ext_val = value; | |
1884 | break; | |
1885 | case 4: | |
1886 | *(__u32 *)ext_val = value; | |
1887 | break; | |
1888 | case 8: | |
1889 | *(__u64 *)ext_val = value; | |
1890 | break; | |
1891 | default: | |
1892 | return -EINVAL; | |
166750bc AN |
1893 | } |
1894 | ext->is_set = true; | |
1895 | return 0; | |
1896 | } | |
1897 | ||
8601fd42 AN |
1898 | static int bpf_object__process_kconfig_line(struct bpf_object *obj, |
1899 | char *buf, void *data) | |
166750bc | 1900 | { |
166750bc | 1901 | struct extern_desc *ext; |
8601fd42 | 1902 | char *sep, *value; |
166750bc AN |
1903 | int len, err = 0; |
1904 | void *ext_val; | |
1905 | __u64 num; | |
166750bc | 1906 | |
13d35a0c | 1907 | if (!str_has_pfx(buf, "CONFIG_")) |
8601fd42 | 1908 | return 0; |
166750bc | 1909 | |
8601fd42 AN |
1910 | sep = strchr(buf, '='); |
1911 | if (!sep) { | |
1912 | pr_warn("failed to parse '%s': no separator\n", buf); | |
1913 | return -EINVAL; | |
1914 | } | |
1915 | ||
1916 | /* Trim ending '\n' */ | |
1917 | len = strlen(buf); | |
1918 | if (buf[len - 1] == '\n') | |
1919 | buf[len - 1] = '\0'; | |
1920 | /* Split on '=' and ensure that a value is present. */ | |
1921 | *sep = '\0'; | |
1922 | if (!sep[1]) { | |
1923 | *sep = '='; | |
1924 | pr_warn("failed to parse '%s': no value\n", buf); | |
1925 | return -EINVAL; | |
1926 | } | |
1927 | ||
1928 | ext = find_extern_by_name(obj, buf); | |
1929 | if (!ext || ext->is_set) | |
1930 | return 0; | |
1931 | ||
2e33efe3 | 1932 | ext_val = data + ext->kcfg.data_off; |
8601fd42 AN |
1933 | value = sep + 1; |
1934 | ||
1935 | switch (*value) { | |
1936 | case 'y': case 'n': case 'm': | |
2e33efe3 | 1937 | err = set_kcfg_value_tri(ext, ext_val, *value); |
8601fd42 AN |
1938 | break; |
1939 | case '"': | |
2e33efe3 | 1940 | err = set_kcfg_value_str(ext, ext_val, value); |
8601fd42 AN |
1941 | break; |
1942 | default: | |
1943 | /* assume integer */ | |
1944 | err = parse_u64(value, &num); | |
1945 | if (err) { | |
55d00c37 | 1946 | pr_warn("extern (kcfg) '%s': value '%s' isn't a valid integer\n", ext->name, value); |
8601fd42 AN |
1947 | return err; |
1948 | } | |
55d00c37 AN |
1949 | if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) { |
1950 | pr_warn("extern (kcfg) '%s': value '%s' implies integer type\n", ext->name, value); | |
1951 | return -EINVAL; | |
1952 | } | |
2e33efe3 | 1953 | err = set_kcfg_value_num(ext, ext_val, num); |
8601fd42 | 1954 | break; |
166750bc | 1955 | } |
8601fd42 AN |
1956 | if (err) |
1957 | return err; | |
55d00c37 | 1958 | pr_debug("extern (kcfg) '%s': set to %s\n", ext->name, value); |
8601fd42 AN |
1959 | return 0; |
1960 | } | |
1961 | ||
1962 | static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data) | |
1963 | { | |
1964 | char buf[PATH_MAX]; | |
1965 | struct utsname uts; | |
1966 | int len, err = 0; | |
1967 | gzFile file; | |
1968 | ||
1969 | uname(&uts); | |
1970 | len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release); | |
1971 | if (len < 0) | |
1972 | return -EINVAL; | |
1973 | else if (len >= PATH_MAX) | |
1974 | return -ENAMETOOLONG; | |
1975 | ||
1976 | /* gzopen also accepts uncompressed files. */ | |
1977 | file = gzopen(buf, "r"); | |
1978 | if (!file) | |
1979 | file = gzopen("/proc/config.gz", "r"); | |
1980 | ||
166750bc | 1981 | if (!file) { |
8601fd42 | 1982 | pr_warn("failed to open system Kconfig\n"); |
166750bc AN |
1983 | return -ENOENT; |
1984 | } | |
1985 | ||
1986 | while (gzgets(file, buf, sizeof(buf))) { | |
8601fd42 AN |
1987 | err = bpf_object__process_kconfig_line(obj, buf, data); |
1988 | if (err) { | |
1989 | pr_warn("error parsing system Kconfig line '%s': %d\n", | |
1990 | buf, err); | |
166750bc AN |
1991 | goto out; |
1992 | } | |
8601fd42 | 1993 | } |
166750bc | 1994 | |
8601fd42 AN |
1995 | out: |
1996 | gzclose(file); | |
1997 | return err; | |
1998 | } | |
166750bc | 1999 | |
8601fd42 AN |
2000 | static int bpf_object__read_kconfig_mem(struct bpf_object *obj, |
2001 | const char *config, void *data) | |
2002 | { | |
2003 | char buf[PATH_MAX]; | |
2004 | int err = 0; | |
2005 | FILE *file; | |
166750bc | 2006 | |
8601fd42 AN |
2007 | file = fmemopen((void *)config, strlen(config), "r"); |
2008 | if (!file) { | |
2009 | err = -errno; | |
2010 | pr_warn("failed to open in-memory Kconfig: %d\n", err); | |
2011 | return err; | |
2012 | } | |
2013 | ||
2014 | while (fgets(buf, sizeof(buf), file)) { | |
2015 | err = bpf_object__process_kconfig_line(obj, buf, data); | |
2016 | if (err) { | |
2017 | pr_warn("error parsing in-memory Kconfig line '%s': %d\n", | |
2018 | buf, err); | |
166750bc AN |
2019 | break; |
2020 | } | |
166750bc AN |
2021 | } |
2022 | ||
8601fd42 | 2023 | fclose(file); |
166750bc AN |
2024 | return err; |
2025 | } | |
2026 | ||
81bfdd08 | 2027 | static int bpf_object__init_kconfig_map(struct bpf_object *obj) |
166750bc | 2028 | { |
2e33efe3 | 2029 | struct extern_desc *last_ext = NULL, *ext; |
166750bc | 2030 | size_t map_sz; |
2e33efe3 | 2031 | int i, err; |
166750bc | 2032 | |
2e33efe3 AN |
2033 | for (i = 0; i < obj->nr_extern; i++) { |
2034 | ext = &obj->externs[i]; | |
2035 | if (ext->type == EXT_KCFG) | |
2036 | last_ext = ext; | |
2037 | } | |
166750bc | 2038 | |
2e33efe3 AN |
2039 | if (!last_ext) |
2040 | return 0; | |
166750bc | 2041 | |
2e33efe3 | 2042 | map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz; |
81bfdd08 | 2043 | err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG, |
aed65917 | 2044 | ".kconfig", obj->efile.symbols_shndx, |
166750bc AN |
2045 | NULL, map_sz); |
2046 | if (err) | |
2047 | return err; | |
2048 | ||
81bfdd08 | 2049 | obj->kconfig_map_idx = obj->nr_maps - 1; |
166750bc AN |
2050 | |
2051 | return 0; | |
2052 | } | |
2053 | ||
42869d28 | 2054 | const struct btf_type * |
ddc7c304 | 2055 | skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id) |
abd29c93 AN |
2056 | { |
2057 | const struct btf_type *t = btf__type_by_id(btf, id); | |
8837fe5d | 2058 | |
ddc7c304 AN |
2059 | if (res_id) |
2060 | *res_id = id; | |
2061 | ||
2062 | while (btf_is_mod(t) || btf_is_typedef(t)) { | |
2063 | if (res_id) | |
2064 | *res_id = t->type; | |
2065 | t = btf__type_by_id(btf, t->type); | |
abd29c93 | 2066 | } |
ddc7c304 AN |
2067 | |
2068 | return t; | |
abd29c93 AN |
2069 | } |
2070 | ||
590a0088 MKL |
2071 | static const struct btf_type * |
2072 | resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id) | |
2073 | { | |
2074 | const struct btf_type *t; | |
2075 | ||
2076 | t = skip_mods_and_typedefs(btf, id, NULL); | |
2077 | if (!btf_is_ptr(t)) | |
2078 | return NULL; | |
2079 | ||
2080 | t = skip_mods_and_typedefs(btf, t->type, res_id); | |
2081 | ||
2082 | return btf_is_func_proto(t) ? t : NULL; | |
2083 | } | |
2084 | ||
774e132e | 2085 | static const char *__btf_kind_str(__u16 kind) |
81ba0889 | 2086 | { |
774e132e | 2087 | switch (kind) { |
81ba0889 AN |
2088 | case BTF_KIND_UNKN: return "void"; |
2089 | case BTF_KIND_INT: return "int"; | |
2090 | case BTF_KIND_PTR: return "ptr"; | |
2091 | case BTF_KIND_ARRAY: return "array"; | |
2092 | case BTF_KIND_STRUCT: return "struct"; | |
2093 | case BTF_KIND_UNION: return "union"; | |
2094 | case BTF_KIND_ENUM: return "enum"; | |
2095 | case BTF_KIND_FWD: return "fwd"; | |
2096 | case BTF_KIND_TYPEDEF: return "typedef"; | |
2097 | case BTF_KIND_VOLATILE: return "volatile"; | |
2098 | case BTF_KIND_CONST: return "const"; | |
2099 | case BTF_KIND_RESTRICT: return "restrict"; | |
2100 | case BTF_KIND_FUNC: return "func"; | |
2101 | case BTF_KIND_FUNC_PROTO: return "func_proto"; | |
2102 | case BTF_KIND_VAR: return "var"; | |
2103 | case BTF_KIND_DATASEC: return "datasec"; | |
22541a9e | 2104 | case BTF_KIND_FLOAT: return "float"; |
223f903e | 2105 | case BTF_KIND_DECL_TAG: return "decl_tag"; |
2dc1e488 | 2106 | case BTF_KIND_TYPE_TAG: return "type_tag"; |
f2a62588 | 2107 | case BTF_KIND_ENUM64: return "enum64"; |
81ba0889 AN |
2108 | default: return "unknown"; |
2109 | } | |
2110 | } | |
2111 | ||
42869d28 | 2112 | const char *btf_kind_str(const struct btf_type *t) |
774e132e MKL |
2113 | { |
2114 | return __btf_kind_str(btf_kind(t)); | |
2115 | } | |
2116 | ||
ef99b02b AN |
2117 | /* |
2118 | * Fetch integer attribute of BTF map definition. Such attributes are | |
2119 | * represented using a pointer to an array, in which dimensionality of array | |
2120 | * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY]; | |
2121 | * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF | |
2122 | * type definition, while using only sizeof(void *) space in ELF data section. | |
2123 | */ | |
2124 | static bool get_map_field_int(const char *map_name, const struct btf *btf, | |
8983b731 AN |
2125 | const struct btf_member *m, __u32 *res) |
2126 | { | |
ddc7c304 | 2127 | const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL); |
abd29c93 | 2128 | const char *name = btf__name_by_offset(btf, m->name_off); |
ef99b02b AN |
2129 | const struct btf_array *arr_info; |
2130 | const struct btf_type *arr_t; | |
abd29c93 | 2131 | |
b03bc685 | 2132 | if (!btf_is_ptr(t)) { |
81ba0889 AN |
2133 | pr_warn("map '%s': attr '%s': expected PTR, got %s.\n", |
2134 | map_name, name, btf_kind_str(t)); | |
abd29c93 AN |
2135 | return false; |
2136 | } | |
ef99b02b AN |
2137 | |
2138 | arr_t = btf__type_by_id(btf, t->type); | |
2139 | if (!arr_t) { | |
be18010e KW |
2140 | pr_warn("map '%s': attr '%s': type [%u] not found.\n", |
2141 | map_name, name, t->type); | |
abd29c93 AN |
2142 | return false; |
2143 | } | |
b03bc685 | 2144 | if (!btf_is_array(arr_t)) { |
81ba0889 AN |
2145 | pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n", |
2146 | map_name, name, btf_kind_str(arr_t)); | |
abd29c93 AN |
2147 | return false; |
2148 | } | |
b03bc685 | 2149 | arr_info = btf_array(arr_t); |
ef99b02b | 2150 | *res = arr_info->nelems; |
abd29c93 AN |
2151 | return true; |
2152 | } | |
2153 | ||
e588c116 WY |
2154 | static int pathname_concat(char *buf, size_t buf_sz, const char *path, const char *name) |
2155 | { | |
2156 | int len; | |
2157 | ||
2158 | len = snprintf(buf, buf_sz, "%s/%s", path, name); | |
2159 | if (len < 0) | |
2160 | return -EINVAL; | |
2161 | if (len >= buf_sz) | |
2162 | return -ENAMETOOLONG; | |
2163 | ||
2164 | return 0; | |
2165 | } | |
2166 | ||
57a00f41 THJ |
2167 | static int build_map_pin_path(struct bpf_map *map, const char *path) |
2168 | { | |
2169 | char buf[PATH_MAX]; | |
e588c116 | 2170 | int err; |
57a00f41 THJ |
2171 | |
2172 | if (!path) | |
2173 | path = "/sys/fs/bpf"; | |
2174 | ||
e588c116 WY |
2175 | err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map)); |
2176 | if (err) | |
2177 | return err; | |
57a00f41 | 2178 | |
6e9cab2e | 2179 | return bpf_map__set_pin_path(map, buf); |
57a00f41 THJ |
2180 | } |
2181 | ||
146bf811 AN |
2182 | /* should match definition in bpf_helpers.h */ |
2183 | enum libbpf_pin_type { | |
2184 | LIBBPF_PIN_NONE, | |
2185 | /* PIN_BY_NAME: pin maps by name (in /sys/fs/bpf by default) */ | |
2186 | LIBBPF_PIN_BY_NAME, | |
2187 | }; | |
2188 | ||
c7ef5ec9 AN |
2189 | int parse_btf_map_def(const char *map_name, struct btf *btf, |
2190 | const struct btf_type *def_t, bool strict, | |
2191 | struct btf_map_def *map_def, struct btf_map_def *inner_def) | |
abd29c93 | 2192 | { |
41017e56 | 2193 | const struct btf_type *t; |
abd29c93 | 2194 | const struct btf_member *m; |
c7ef5ec9 | 2195 | bool is_inner = inner_def == NULL; |
abd29c93 AN |
2196 | int vlen, i; |
2197 | ||
c7ef5ec9 AN |
2198 | vlen = btf_vlen(def_t); |
2199 | m = btf_members(def_t); | |
abd29c93 | 2200 | for (i = 0; i < vlen; i++, m++) { |
c7ef5ec9 | 2201 | const char *name = btf__name_by_offset(btf, m->name_off); |
abd29c93 AN |
2202 | |
2203 | if (!name) { | |
c7ef5ec9 | 2204 | pr_warn("map '%s': invalid field #%d.\n", map_name, i); |
abd29c93 AN |
2205 | return -EINVAL; |
2206 | } | |
2207 | if (strcmp(name, "type") == 0) { | |
c7ef5ec9 | 2208 | if (!get_map_field_int(map_name, btf, m, &map_def->map_type)) |
abd29c93 | 2209 | return -EINVAL; |
c7ef5ec9 | 2210 | map_def->parts |= MAP_DEF_MAP_TYPE; |
abd29c93 | 2211 | } else if (strcmp(name, "max_entries") == 0) { |
c7ef5ec9 | 2212 | if (!get_map_field_int(map_name, btf, m, &map_def->max_entries)) |
abd29c93 | 2213 | return -EINVAL; |
c7ef5ec9 | 2214 | map_def->parts |= MAP_DEF_MAX_ENTRIES; |
abd29c93 | 2215 | } else if (strcmp(name, "map_flags") == 0) { |
c7ef5ec9 | 2216 | if (!get_map_field_int(map_name, btf, m, &map_def->map_flags)) |
abd29c93 | 2217 | return -EINVAL; |
c7ef5ec9 | 2218 | map_def->parts |= MAP_DEF_MAP_FLAGS; |
1bdb6c9a | 2219 | } else if (strcmp(name, "numa_node") == 0) { |
c7ef5ec9 | 2220 | if (!get_map_field_int(map_name, btf, m, &map_def->numa_node)) |
1bdb6c9a | 2221 | return -EINVAL; |
c7ef5ec9 | 2222 | map_def->parts |= MAP_DEF_NUMA_NODE; |
abd29c93 AN |
2223 | } else if (strcmp(name, "key_size") == 0) { |
2224 | __u32 sz; | |
2225 | ||
c7ef5ec9 | 2226 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2227 | return -EINVAL; |
c7ef5ec9 | 2228 | if (map_def->key_size && map_def->key_size != sz) { |
be18010e | 2229 | pr_warn("map '%s': conflicting key size %u != %u.\n", |
c7ef5ec9 | 2230 | map_name, map_def->key_size, sz); |
abd29c93 AN |
2231 | return -EINVAL; |
2232 | } | |
c7ef5ec9 AN |
2233 | map_def->key_size = sz; |
2234 | map_def->parts |= MAP_DEF_KEY_SIZE; | |
abd29c93 AN |
2235 | } else if (strcmp(name, "key") == 0) { |
2236 | __s64 sz; | |
2237 | ||
c7ef5ec9 | 2238 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2239 | if (!t) { |
be18010e | 2240 | pr_warn("map '%s': key type [%d] not found.\n", |
c7ef5ec9 | 2241 | map_name, m->type); |
abd29c93 AN |
2242 | return -EINVAL; |
2243 | } | |
b03bc685 | 2244 | if (!btf_is_ptr(t)) { |
81ba0889 | 2245 | pr_warn("map '%s': key spec is not PTR: %s.\n", |
c7ef5ec9 | 2246 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2247 | return -EINVAL; |
2248 | } | |
c7ef5ec9 | 2249 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2250 | if (sz < 0) { |
679152d3 | 2251 | pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n", |
c7ef5ec9 | 2252 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2253 | return sz; |
2254 | } | |
c7ef5ec9 | 2255 | if (map_def->key_size && map_def->key_size != sz) { |
679152d3 | 2256 | pr_warn("map '%s': conflicting key size %u != %zd.\n", |
c7ef5ec9 | 2257 | map_name, map_def->key_size, (ssize_t)sz); |
abd29c93 AN |
2258 | return -EINVAL; |
2259 | } | |
c7ef5ec9 AN |
2260 | map_def->key_size = sz; |
2261 | map_def->key_type_id = t->type; | |
2262 | map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE; | |
abd29c93 AN |
2263 | } else if (strcmp(name, "value_size") == 0) { |
2264 | __u32 sz; | |
2265 | ||
c7ef5ec9 | 2266 | if (!get_map_field_int(map_name, btf, m, &sz)) |
abd29c93 | 2267 | return -EINVAL; |
c7ef5ec9 | 2268 | if (map_def->value_size && map_def->value_size != sz) { |
be18010e | 2269 | pr_warn("map '%s': conflicting value size %u != %u.\n", |
c7ef5ec9 | 2270 | map_name, map_def->value_size, sz); |
abd29c93 AN |
2271 | return -EINVAL; |
2272 | } | |
c7ef5ec9 AN |
2273 | map_def->value_size = sz; |
2274 | map_def->parts |= MAP_DEF_VALUE_SIZE; | |
abd29c93 AN |
2275 | } else if (strcmp(name, "value") == 0) { |
2276 | __s64 sz; | |
2277 | ||
c7ef5ec9 | 2278 | t = btf__type_by_id(btf, m->type); |
abd29c93 | 2279 | if (!t) { |
be18010e | 2280 | pr_warn("map '%s': value type [%d] not found.\n", |
c7ef5ec9 | 2281 | map_name, m->type); |
abd29c93 AN |
2282 | return -EINVAL; |
2283 | } | |
b03bc685 | 2284 | if (!btf_is_ptr(t)) { |
81ba0889 | 2285 | pr_warn("map '%s': value spec is not PTR: %s.\n", |
c7ef5ec9 | 2286 | map_name, btf_kind_str(t)); |
abd29c93 AN |
2287 | return -EINVAL; |
2288 | } | |
c7ef5ec9 | 2289 | sz = btf__resolve_size(btf, t->type); |
abd29c93 | 2290 | if (sz < 0) { |
679152d3 | 2291 | pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n", |
c7ef5ec9 | 2292 | map_name, t->type, (ssize_t)sz); |
abd29c93 AN |
2293 | return sz; |
2294 | } | |
c7ef5ec9 | 2295 | if (map_def->value_size && map_def->value_size != sz) { |
679152d3 | 2296 | pr_warn("map '%s': conflicting value size %u != %zd.\n", |
c7ef5ec9 | 2297 | map_name, map_def->value_size, (ssize_t)sz); |
abd29c93 AN |
2298 | return -EINVAL; |
2299 | } | |
c7ef5ec9 AN |
2300 | map_def->value_size = sz; |
2301 | map_def->value_type_id = t->type; | |
2302 | map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE; | |
646f02ff AN |
2303 | } |
2304 | else if (strcmp(name, "values") == 0) { | |
341ac5ff HC |
2305 | bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type); |
2306 | bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY; | |
2307 | const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value"; | |
c7ef5ec9 | 2308 | char inner_map_name[128]; |
646f02ff AN |
2309 | int err; |
2310 | ||
2311 | if (is_inner) { | |
2312 | pr_warn("map '%s': multi-level inner maps not supported.\n", | |
c7ef5ec9 | 2313 | map_name); |
646f02ff AN |
2314 | return -ENOTSUP; |
2315 | } | |
2316 | if (i != vlen - 1) { | |
2317 | pr_warn("map '%s': '%s' member should be last.\n", | |
c7ef5ec9 | 2318 | map_name, name); |
646f02ff AN |
2319 | return -EINVAL; |
2320 | } | |
341ac5ff HC |
2321 | if (!is_map_in_map && !is_prog_array) { |
2322 | pr_warn("map '%s': should be map-in-map or prog-array.\n", | |
c7ef5ec9 | 2323 | map_name); |
646f02ff AN |
2324 | return -ENOTSUP; |
2325 | } | |
c7ef5ec9 | 2326 | if (map_def->value_size && map_def->value_size != 4) { |
646f02ff | 2327 | pr_warn("map '%s': conflicting value size %u != 4.\n", |
c7ef5ec9 | 2328 | map_name, map_def->value_size); |
646f02ff AN |
2329 | return -EINVAL; |
2330 | } | |
c7ef5ec9 AN |
2331 | map_def->value_size = 4; |
2332 | t = btf__type_by_id(btf, m->type); | |
646f02ff | 2333 | if (!t) { |
341ac5ff HC |
2334 | pr_warn("map '%s': %s type [%d] not found.\n", |
2335 | map_name, desc, m->type); | |
646f02ff AN |
2336 | return -EINVAL; |
2337 | } | |
2338 | if (!btf_is_array(t) || btf_array(t)->nelems) { | |
341ac5ff HC |
2339 | pr_warn("map '%s': %s spec is not a zero-sized array.\n", |
2340 | map_name, desc); | |
646f02ff AN |
2341 | return -EINVAL; |
2342 | } | |
c7ef5ec9 | 2343 | t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL); |
646f02ff | 2344 | if (!btf_is_ptr(t)) { |
341ac5ff HC |
2345 | pr_warn("map '%s': %s def is of unexpected kind %s.\n", |
2346 | map_name, desc, btf_kind_str(t)); | |
646f02ff AN |
2347 | return -EINVAL; |
2348 | } | |
c7ef5ec9 | 2349 | t = skip_mods_and_typedefs(btf, t->type, NULL); |
341ac5ff HC |
2350 | if (is_prog_array) { |
2351 | if (!btf_is_func_proto(t)) { | |
2352 | pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n", | |
2353 | map_name, btf_kind_str(t)); | |
2354 | return -EINVAL; | |
2355 | } | |
2356 | continue; | |
2357 | } | |
646f02ff | 2358 | if (!btf_is_struct(t)) { |
81ba0889 | 2359 | pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n", |
c7ef5ec9 | 2360 | map_name, btf_kind_str(t)); |
646f02ff AN |
2361 | return -EINVAL; |
2362 | } | |
2363 | ||
c7ef5ec9 AN |
2364 | snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name); |
2365 | err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL); | |
646f02ff AN |
2366 | if (err) |
2367 | return err; | |
c7ef5ec9 AN |
2368 | |
2369 | map_def->parts |= MAP_DEF_INNER_MAP; | |
57a00f41 THJ |
2370 | } else if (strcmp(name, "pinning") == 0) { |
2371 | __u32 val; | |
57a00f41 | 2372 | |
646f02ff | 2373 | if (is_inner) { |
c7ef5ec9 | 2374 | pr_warn("map '%s': inner def can't be pinned.\n", map_name); |
646f02ff AN |
2375 | return -EINVAL; |
2376 | } | |
c7ef5ec9 | 2377 | if (!get_map_field_int(map_name, btf, m, &val)) |
57a00f41 | 2378 | return -EINVAL; |
c7ef5ec9 | 2379 | if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) { |
57a00f41 | 2380 | pr_warn("map '%s': invalid pinning value %u.\n", |
c7ef5ec9 | 2381 | map_name, val); |
57a00f41 THJ |
2382 | return -EINVAL; |
2383 | } | |
c7ef5ec9 AN |
2384 | map_def->pinning = val; |
2385 | map_def->parts |= MAP_DEF_PINNING; | |
47512102 JK |
2386 | } else if (strcmp(name, "map_extra") == 0) { |
2387 | __u32 map_extra; | |
2388 | ||
2389 | if (!get_map_field_int(map_name, btf, m, &map_extra)) | |
2390 | return -EINVAL; | |
2391 | map_def->map_extra = map_extra; | |
2392 | map_def->parts |= MAP_DEF_MAP_EXTRA; | |
abd29c93 AN |
2393 | } else { |
2394 | if (strict) { | |
c7ef5ec9 | 2395 | pr_warn("map '%s': unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2396 | return -ENOTSUP; |
2397 | } | |
c7ef5ec9 | 2398 | pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name); |
abd29c93 AN |
2399 | } |
2400 | } | |
2401 | ||
c7ef5ec9 AN |
2402 | if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) { |
2403 | pr_warn("map '%s': map type isn't specified.\n", map_name); | |
abd29c93 AN |
2404 | return -EINVAL; |
2405 | } | |
2406 | ||
2407 | return 0; | |
2408 | } | |
2409 | ||
597fbc46 AN |
2410 | static size_t adjust_ringbuf_sz(size_t sz) |
2411 | { | |
2412 | __u32 page_sz = sysconf(_SC_PAGE_SIZE); | |
2413 | __u32 mul; | |
2414 | ||
2415 | /* if user forgot to set any size, make sure they see error */ | |
2416 | if (sz == 0) | |
2417 | return 0; | |
2418 | /* Kernel expects BPF_MAP_TYPE_RINGBUF's max_entries to be | |
2419 | * a power-of-2 multiple of kernel's page size. If user diligently | |
2420 | * satisified these conditions, pass the size through. | |
2421 | */ | |
2422 | if ((sz % page_sz) == 0 && is_pow_of_2(sz / page_sz)) | |
2423 | return sz; | |
2424 | ||
2425 | /* Otherwise find closest (page_sz * power_of_2) product bigger than | |
2426 | * user-set size to satisfy both user size request and kernel | |
2427 | * requirements and substitute correct max_entries for map creation. | |
2428 | */ | |
2429 | for (mul = 1; mul <= UINT_MAX / page_sz; mul <<= 1) { | |
2430 | if (mul * page_sz > sz) | |
2431 | return mul * page_sz; | |
2432 | } | |
2433 | ||
2434 | /* if it's impossible to satisfy the conditions (i.e., user size is | |
2435 | * very close to UINT_MAX but is not a power-of-2 multiple of | |
2436 | * page_size) then just return original size and let kernel reject it | |
2437 | */ | |
2438 | return sz; | |
2439 | } | |
2440 | ||
b66ccae0 DV |
2441 | static bool map_is_ringbuf(const struct bpf_map *map) |
2442 | { | |
2443 | return map->def.type == BPF_MAP_TYPE_RINGBUF || | |
2444 | map->def.type == BPF_MAP_TYPE_USER_RINGBUF; | |
2445 | } | |
2446 | ||
c7ef5ec9 AN |
2447 | static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def) |
2448 | { | |
2449 | map->def.type = def->map_type; | |
2450 | map->def.key_size = def->key_size; | |
2451 | map->def.value_size = def->value_size; | |
2452 | map->def.max_entries = def->max_entries; | |
2453 | map->def.map_flags = def->map_flags; | |
47512102 | 2454 | map->map_extra = def->map_extra; |
c7ef5ec9 AN |
2455 | |
2456 | map->numa_node = def->numa_node; | |
2457 | map->btf_key_type_id = def->key_type_id; | |
2458 | map->btf_value_type_id = def->value_type_id; | |
2459 | ||
597fbc46 | 2460 | /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */ |
b66ccae0 | 2461 | if (map_is_ringbuf(map)) |
597fbc46 AN |
2462 | map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries); |
2463 | ||
c7ef5ec9 AN |
2464 | if (def->parts & MAP_DEF_MAP_TYPE) |
2465 | pr_debug("map '%s': found type = %u.\n", map->name, def->map_type); | |
2466 | ||
2467 | if (def->parts & MAP_DEF_KEY_TYPE) | |
2468 | pr_debug("map '%s': found key [%u], sz = %u.\n", | |
2469 | map->name, def->key_type_id, def->key_size); | |
2470 | else if (def->parts & MAP_DEF_KEY_SIZE) | |
2471 | pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size); | |
2472 | ||
2473 | if (def->parts & MAP_DEF_VALUE_TYPE) | |
2474 | pr_debug("map '%s': found value [%u], sz = %u.\n", | |
2475 | map->name, def->value_type_id, def->value_size); | |
2476 | else if (def->parts & MAP_DEF_VALUE_SIZE) | |
2477 | pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size); | |
2478 | ||
2479 | if (def->parts & MAP_DEF_MAX_ENTRIES) | |
2480 | pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries); | |
2481 | if (def->parts & MAP_DEF_MAP_FLAGS) | |
47512102 JK |
2482 | pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags); |
2483 | if (def->parts & MAP_DEF_MAP_EXTRA) | |
2484 | pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name, | |
2485 | (unsigned long long)def->map_extra); | |
c7ef5ec9 AN |
2486 | if (def->parts & MAP_DEF_PINNING) |
2487 | pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning); | |
2488 | if (def->parts & MAP_DEF_NUMA_NODE) | |
2489 | pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node); | |
2490 | ||
2491 | if (def->parts & MAP_DEF_INNER_MAP) | |
2492 | pr_debug("map '%s': found inner map definition.\n", map->name); | |
2493 | } | |
2494 | ||
c1cccec9 AN |
2495 | static const char *btf_var_linkage_str(__u32 linkage) |
2496 | { | |
2497 | switch (linkage) { | |
2498 | case BTF_VAR_STATIC: return "static"; | |
2499 | case BTF_VAR_GLOBAL_ALLOCATED: return "global"; | |
2500 | case BTF_VAR_GLOBAL_EXTERN: return "extern"; | |
2501 | default: return "unknown"; | |
2502 | } | |
2503 | } | |
2504 | ||
41017e56 AN |
2505 | static int bpf_object__init_user_btf_map(struct bpf_object *obj, |
2506 | const struct btf_type *sec, | |
2507 | int var_idx, int sec_idx, | |
2508 | const Elf_Data *data, bool strict, | |
2509 | const char *pin_root_path) | |
2510 | { | |
c7ef5ec9 | 2511 | struct btf_map_def map_def = {}, inner_def = {}; |
41017e56 AN |
2512 | const struct btf_type *var, *def; |
2513 | const struct btf_var_secinfo *vi; | |
2514 | const struct btf_var *var_extra; | |
2515 | const char *map_name; | |
2516 | struct bpf_map *map; | |
c7ef5ec9 | 2517 | int err; |
41017e56 AN |
2518 | |
2519 | vi = btf_var_secinfos(sec) + var_idx; | |
2520 | var = btf__type_by_id(obj->btf, vi->type); | |
2521 | var_extra = btf_var(var); | |
2522 | map_name = btf__name_by_offset(obj->btf, var->name_off); | |
2523 | ||
2524 | if (map_name == NULL || map_name[0] == '\0') { | |
2525 | pr_warn("map #%d: empty name.\n", var_idx); | |
2526 | return -EINVAL; | |
2527 | } | |
2528 | if ((__u64)vi->offset + vi->size > data->d_size) { | |
2529 | pr_warn("map '%s' BTF data is corrupted.\n", map_name); | |
2530 | return -EINVAL; | |
2531 | } | |
2532 | if (!btf_is_var(var)) { | |
81ba0889 AN |
2533 | pr_warn("map '%s': unexpected var kind %s.\n", |
2534 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2535 | return -EINVAL; |
2536 | } | |
c1cccec9 AN |
2537 | if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) { |
2538 | pr_warn("map '%s': unsupported map linkage %s.\n", | |
2539 | map_name, btf_var_linkage_str(var_extra->linkage)); | |
41017e56 AN |
2540 | return -EOPNOTSUPP; |
2541 | } | |
2542 | ||
2543 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
2544 | if (!btf_is_struct(def)) { | |
81ba0889 AN |
2545 | pr_warn("map '%s': unexpected def kind %s.\n", |
2546 | map_name, btf_kind_str(var)); | |
41017e56 AN |
2547 | return -EINVAL; |
2548 | } | |
2549 | if (def->size > vi->size) { | |
2550 | pr_warn("map '%s': invalid def size.\n", map_name); | |
2551 | return -EINVAL; | |
2552 | } | |
2553 | ||
2554 | map = bpf_object__add_map(obj); | |
2555 | if (IS_ERR(map)) | |
2556 | return PTR_ERR(map); | |
2557 | map->name = strdup(map_name); | |
2558 | if (!map->name) { | |
2559 | pr_warn("map '%s': failed to alloc map name.\n", map_name); | |
2560 | return -ENOMEM; | |
2561 | } | |
2562 | map->libbpf_type = LIBBPF_MAP_UNSPEC; | |
2563 | map->def.type = BPF_MAP_TYPE_UNSPEC; | |
2564 | map->sec_idx = sec_idx; | |
2565 | map->sec_offset = vi->offset; | |
646f02ff | 2566 | map->btf_var_idx = var_idx; |
41017e56 AN |
2567 | pr_debug("map '%s': at sec_idx %d, offset %zu.\n", |
2568 | map_name, map->sec_idx, map->sec_offset); | |
2569 | ||
c7ef5ec9 AN |
2570 | err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def); |
2571 | if (err) | |
2572 | return err; | |
2573 | ||
2574 | fill_map_from_def(map, &map_def); | |
2575 | ||
2576 | if (map_def.pinning == LIBBPF_PIN_BY_NAME) { | |
2577 | err = build_map_pin_path(map, pin_root_path); | |
2578 | if (err) { | |
2579 | pr_warn("map '%s': couldn't build pin path.\n", map->name); | |
2580 | return err; | |
2581 | } | |
2582 | } | |
2583 | ||
2584 | if (map_def.parts & MAP_DEF_INNER_MAP) { | |
2585 | map->inner_map = calloc(1, sizeof(*map->inner_map)); | |
2586 | if (!map->inner_map) | |
2587 | return -ENOMEM; | |
2588 | map->inner_map->fd = -1; | |
2589 | map->inner_map->sec_idx = sec_idx; | |
2590 | map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1); | |
2591 | if (!map->inner_map->name) | |
2592 | return -ENOMEM; | |
2593 | sprintf(map->inner_map->name, "%s.inner", map_name); | |
2594 | ||
2595 | fill_map_from_def(map->inner_map, &inner_def); | |
2596 | } | |
2597 | ||
4fcac46c | 2598 | err = map_fill_btf_type_info(obj, map); |
262cfb74 DK |
2599 | if (err) |
2600 | return err; | |
2601 | ||
c7ef5ec9 | 2602 | return 0; |
41017e56 AN |
2603 | } |
2604 | ||
57a00f41 THJ |
2605 | static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict, |
2606 | const char *pin_root_path) | |
abd29c93 AN |
2607 | { |
2608 | const struct btf_type *sec = NULL; | |
2609 | int nr_types, i, vlen, err; | |
2610 | const struct btf_type *t; | |
2611 | const char *name; | |
2612 | Elf_Data *data; | |
2613 | Elf_Scn *scn; | |
2614 | ||
2615 | if (obj->efile.btf_maps_shndx < 0) | |
2616 | return 0; | |
2617 | ||
88a82120 AN |
2618 | scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx); |
2619 | data = elf_sec_data(obj, scn); | |
abd29c93 | 2620 | if (!scn || !data) { |
88a82120 AN |
2621 | pr_warn("elf: failed to get %s map definitions for %s\n", |
2622 | MAPS_ELF_SEC, obj->path); | |
abd29c93 AN |
2623 | return -EINVAL; |
2624 | } | |
2625 | ||
6a886de0 HC |
2626 | nr_types = btf__type_cnt(obj->btf); |
2627 | for (i = 1; i < nr_types; i++) { | |
abd29c93 | 2628 | t = btf__type_by_id(obj->btf, i); |
b03bc685 | 2629 | if (!btf_is_datasec(t)) |
abd29c93 AN |
2630 | continue; |
2631 | name = btf__name_by_offset(obj->btf, t->name_off); | |
2632 | if (strcmp(name, MAPS_ELF_SEC) == 0) { | |
2633 | sec = t; | |
646f02ff | 2634 | obj->efile.btf_maps_sec_btf_id = i; |
abd29c93 AN |
2635 | break; |
2636 | } | |
2637 | } | |
2638 | ||
2639 | if (!sec) { | |
be18010e | 2640 | pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC); |
abd29c93 AN |
2641 | return -ENOENT; |
2642 | } | |
2643 | ||
b03bc685 | 2644 | vlen = btf_vlen(sec); |
abd29c93 AN |
2645 | for (i = 0; i < vlen; i++) { |
2646 | err = bpf_object__init_user_btf_map(obj, sec, i, | |
2647 | obj->efile.btf_maps_shndx, | |
8983b731 AN |
2648 | data, strict, |
2649 | pin_root_path); | |
abd29c93 AN |
2650 | if (err) |
2651 | return err; | |
2652 | } | |
2653 | ||
2654 | return 0; | |
2655 | } | |
2656 | ||
0d13bfce | 2657 | static int bpf_object__init_maps(struct bpf_object *obj, |
01af3bf0 | 2658 | const struct bpf_object_open_opts *opts) |
bf829271 | 2659 | { |
166750bc AN |
2660 | const char *pin_root_path; |
2661 | bool strict; | |
bd054102 | 2662 | int err = 0; |
8837fe5d | 2663 | |
166750bc AN |
2664 | strict = !OPTS_GET(opts, relaxed_maps, false); |
2665 | pin_root_path = OPTS_GET(opts, pin_root_path, NULL); | |
bf829271 | 2666 | |
40e1bcab | 2667 | err = bpf_object__init_user_btf_maps(obj, strict, pin_root_path); |
166750bc | 2668 | err = err ?: bpf_object__init_global_data_maps(obj); |
81bfdd08 | 2669 | err = err ?: bpf_object__init_kconfig_map(obj); |
809a69d6 | 2670 | err = err ?: bpf_object_init_struct_ops(obj); |
bf829271 | 2671 | |
3b3af91c | 2672 | return err; |
561bbcca WN |
2673 | } |
2674 | ||
e3d91b0c JDB |
2675 | static bool section_have_execinstr(struct bpf_object *obj, int idx) |
2676 | { | |
ad23b723 | 2677 | Elf64_Shdr *sh; |
e3d91b0c | 2678 | |
ad23b723 AN |
2679 | sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx)); |
2680 | if (!sh) | |
e3d91b0c JDB |
2681 | return false; |
2682 | ||
ad23b723 | 2683 | return sh->sh_flags & SHF_EXECINSTR; |
e3d91b0c JDB |
2684 | } |
2685 | ||
0f0e55d8 AN |
2686 | static bool btf_needs_sanitization(struct bpf_object *obj) |
2687 | { | |
9ca1f56a AS |
2688 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
2689 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
2690 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
2691 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 2692 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
2dc1e488 | 2693 | bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG); |
f2a62588 | 2694 | bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64); |
0f0e55d8 | 2695 | |
2dc1e488 | 2696 | return !has_func || !has_datasec || !has_func_global || !has_float || |
f2a62588 | 2697 | !has_decl_tag || !has_type_tag || !has_enum64; |
0f0e55d8 AN |
2698 | } |
2699 | ||
f2a62588 | 2700 | static int bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf) |
d7c4b398 | 2701 | { |
9ca1f56a AS |
2702 | bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC); |
2703 | bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC); | |
2704 | bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT); | |
2705 | bool has_func = kernel_supports(obj, FEAT_BTF_FUNC); | |
223f903e | 2706 | bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG); |
2dc1e488 | 2707 | bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG); |
f2a62588 | 2708 | bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64); |
c49a44b3 | 2709 | int enum64_placeholder_id = 0; |
d7c4b398 AN |
2710 | struct btf_type *t; |
2711 | int i, j, vlen; | |
d7c4b398 | 2712 | |
6a886de0 | 2713 | for (i = 1; i < btf__type_cnt(btf); i++) { |
d7c4b398 | 2714 | t = (struct btf_type *)btf__type_by_id(btf, i); |
d7c4b398 | 2715 | |
223f903e YS |
2716 | if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) { |
2717 | /* replace VAR/DECL_TAG with INT */ | |
d7c4b398 | 2718 | t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0); |
1d4126c4 AN |
2719 | /* |
2720 | * using size = 1 is the safest choice, 4 will be too | |
2721 | * big and cause kernel BTF validation failure if | |
2722 | * original variable took less than 4 bytes | |
2723 | */ | |
2724 | t->size = 1; | |
708852dc | 2725 | *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8); |
b03bc685 | 2726 | } else if (!has_datasec && btf_is_datasec(t)) { |
d7c4b398 | 2727 | /* replace DATASEC with STRUCT */ |
b03bc685 AN |
2728 | const struct btf_var_secinfo *v = btf_var_secinfos(t); |
2729 | struct btf_member *m = btf_members(t); | |
d7c4b398 AN |
2730 | struct btf_type *vt; |
2731 | char *name; | |
2732 | ||
2733 | name = (char *)btf__name_by_offset(btf, t->name_off); | |
2734 | while (*name) { | |
2735 | if (*name == '.') | |
2736 | *name = '_'; | |
2737 | name++; | |
2738 | } | |
2739 | ||
b03bc685 | 2740 | vlen = btf_vlen(t); |
d7c4b398 AN |
2741 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen); |
2742 | for (j = 0; j < vlen; j++, v++, m++) { | |
2743 | /* order of field assignments is important */ | |
2744 | m->offset = v->offset * 8; | |
2745 | m->type = v->type; | |
2746 | /* preserve variable name as member name */ | |
2747 | vt = (void *)btf__type_by_id(btf, v->type); | |
2748 | m->name_off = vt->name_off; | |
2749 | } | |
b03bc685 | 2750 | } else if (!has_func && btf_is_func_proto(t)) { |
d7c4b398 | 2751 | /* replace FUNC_PROTO with ENUM */ |
b03bc685 | 2752 | vlen = btf_vlen(t); |
d7c4b398 AN |
2753 | t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen); |
2754 | t->size = sizeof(__u32); /* kernel enforced */ | |
b03bc685 | 2755 | } else if (!has_func && btf_is_func(t)) { |
d7c4b398 AN |
2756 | /* replace FUNC with TYPEDEF */ |
2757 | t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0); | |
2d3eb67f AS |
2758 | } else if (!has_func_global && btf_is_func(t)) { |
2759 | /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */ | |
2760 | t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0); | |
22541a9e IL |
2761 | } else if (!has_float && btf_is_float(t)) { |
2762 | /* replace FLOAT with an equally-sized empty STRUCT; | |
2763 | * since C compilers do not accept e.g. "float" as a | |
2764 | * valid struct name, make it anonymous | |
2765 | */ | |
2766 | t->name_off = 0; | |
2767 | t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0); | |
2dc1e488 YS |
2768 | } else if (!has_type_tag && btf_is_type_tag(t)) { |
2769 | /* replace TYPE_TAG with a CONST */ | |
2770 | t->name_off = 0; | |
2771 | t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0); | |
f2a62588 YS |
2772 | } else if (!has_enum64 && btf_is_enum(t)) { |
2773 | /* clear the kflag */ | |
2774 | t->info = btf_type_info(btf_kind(t), btf_vlen(t), false); | |
2775 | } else if (!has_enum64 && btf_is_enum64(t)) { | |
2776 | /* replace ENUM64 with a union */ | |
2777 | struct btf_member *m; | |
2778 | ||
2779 | if (enum64_placeholder_id == 0) { | |
2780 | enum64_placeholder_id = btf__add_int(btf, "enum64_placeholder", 1, 0); | |
2781 | if (enum64_placeholder_id < 0) | |
2782 | return enum64_placeholder_id; | |
2783 | ||
2784 | t = (struct btf_type *)btf__type_by_id(btf, i); | |
2785 | } | |
2786 | ||
2787 | m = btf_members(t); | |
2788 | vlen = btf_vlen(t); | |
2789 | t->info = BTF_INFO_ENC(BTF_KIND_UNION, 0, vlen); | |
2790 | for (j = 0; j < vlen; j++, m++) { | |
2791 | m->type = enum64_placeholder_id; | |
2792 | m->offset = 0; | |
2793 | } | |
e3ba8e4e | 2794 | } |
d7c4b398 | 2795 | } |
f2a62588 YS |
2796 | |
2797 | return 0; | |
d7c4b398 AN |
2798 | } |
2799 | ||
b35f14f4 | 2800 | static bool libbpf_needs_btf(const struct bpf_object *obj) |
abd29c93 | 2801 | { |
b35f14f4 AN |
2802 | return obj->efile.btf_maps_shndx >= 0 || |
2803 | obj->efile.st_ops_shndx >= 0 || | |
809a69d6 | 2804 | obj->efile.st_ops_link_shndx >= 0 || |
b35f14f4 AN |
2805 | obj->nr_extern > 0; |
2806 | } | |
2807 | ||
2808 | static bool kernel_needs_btf(const struct bpf_object *obj) | |
2809 | { | |
809a69d6 | 2810 | return obj->efile.st_ops_shndx >= 0 || obj->efile.st_ops_link_shndx >= 0; |
abd29c93 AN |
2811 | } |
2812 | ||
063183bf | 2813 | static int bpf_object__init_btf(struct bpf_object *obj, |
9c6660d0 AN |
2814 | Elf_Data *btf_data, |
2815 | Elf_Data *btf_ext_data) | |
2816 | { | |
b7d7f3e1 | 2817 | int err = -ENOENT; |
9c6660d0 AN |
2818 | |
2819 | if (btf_data) { | |
2820 | obj->btf = btf__new(btf_data->d_buf, btf_data->d_size); | |
e9fc3ce9 AN |
2821 | err = libbpf_get_error(obj->btf); |
2822 | if (err) { | |
b7d7f3e1 | 2823 | obj->btf = NULL; |
e9fc3ce9 | 2824 | pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err); |
9c6660d0 AN |
2825 | goto out; |
2826 | } | |
4c01925f AN |
2827 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
2828 | btf__set_pointer_size(obj->btf, 8); | |
9c6660d0 AN |
2829 | } |
2830 | if (btf_ext_data) { | |
11d5daa8 AN |
2831 | struct btf_ext_info *ext_segs[3]; |
2832 | int seg_num, sec_num; | |
2833 | ||
9c6660d0 AN |
2834 | if (!obj->btf) { |
2835 | pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n", | |
2836 | BTF_EXT_ELF_SEC, BTF_ELF_SEC); | |
2837 | goto out; | |
2838 | } | |
e9fc3ce9 AN |
2839 | obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size); |
2840 | err = libbpf_get_error(obj->btf_ext); | |
2841 | if (err) { | |
2842 | pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n", | |
2843 | BTF_EXT_ELF_SEC, err); | |
9c6660d0 AN |
2844 | obj->btf_ext = NULL; |
2845 | goto out; | |
2846 | } | |
11d5daa8 AN |
2847 | |
2848 | /* setup .BTF.ext to ELF section mapping */ | |
2849 | ext_segs[0] = &obj->btf_ext->func_info; | |
2850 | ext_segs[1] = &obj->btf_ext->line_info; | |
2851 | ext_segs[2] = &obj->btf_ext->core_relo_info; | |
2852 | for (seg_num = 0; seg_num < ARRAY_SIZE(ext_segs); seg_num++) { | |
2853 | struct btf_ext_info *seg = ext_segs[seg_num]; | |
2854 | const struct btf_ext_info_sec *sec; | |
2855 | const char *sec_name; | |
2856 | Elf_Scn *scn; | |
2857 | ||
2858 | if (seg->sec_cnt == 0) | |
2859 | continue; | |
2860 | ||
2861 | seg->sec_idxs = calloc(seg->sec_cnt, sizeof(*seg->sec_idxs)); | |
2862 | if (!seg->sec_idxs) { | |
2863 | err = -ENOMEM; | |
2864 | goto out; | |
2865 | } | |
2866 | ||
2867 | sec_num = 0; | |
2868 | for_each_btf_ext_sec(seg, sec) { | |
2869 | /* preventively increment index to avoid doing | |
2870 | * this before every continue below | |
2871 | */ | |
2872 | sec_num++; | |
2873 | ||
2874 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); | |
2875 | if (str_is_empty(sec_name)) | |
2876 | continue; | |
2877 | scn = elf_sec_by_name(obj, sec_name); | |
2878 | if (!scn) | |
2879 | continue; | |
2880 | ||
2881 | seg->sec_idxs[sec_num - 1] = elf_ndxscn(scn); | |
2882 | } | |
2883 | } | |
9c6660d0 AN |
2884 | } |
2885 | out: | |
b35f14f4 | 2886 | if (err && libbpf_needs_btf(obj)) { |
be18010e | 2887 | pr_warn("BTF is required, but is missing or corrupted.\n"); |
b7d7f3e1 | 2888 | return err; |
abd29c93 | 2889 | } |
9c6660d0 AN |
2890 | return 0; |
2891 | } | |
2892 | ||
b96c07f3 AN |
2893 | static int compare_vsi_off(const void *_a, const void *_b) |
2894 | { | |
2895 | const struct btf_var_secinfo *a = _a; | |
2896 | const struct btf_var_secinfo *b = _b; | |
2897 | ||
2898 | return a->offset - b->offset; | |
2899 | } | |
2900 | ||
2901 | static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf, | |
2902 | struct btf_type *t) | |
2903 | { | |
f33f742d AN |
2904 | __u32 size = 0, i, vars = btf_vlen(t); |
2905 | const char *sec_name = btf__name_by_offset(btf, t->name_off); | |
b96c07f3 | 2906 | struct btf_var_secinfo *vsi; |
4fcac46c | 2907 | bool fixup_offsets = false; |
f33f742d | 2908 | int err; |
b96c07f3 | 2909 | |
f33f742d | 2910 | if (!sec_name) { |
b96c07f3 AN |
2911 | pr_debug("No name found in string section for DATASEC kind.\n"); |
2912 | return -ENOENT; | |
2913 | } | |
2914 | ||
4fcac46c AN |
2915 | /* Extern-backing datasecs (.ksyms, .kconfig) have their size and |
2916 | * variable offsets set at the previous step. Further, not every | |
2917 | * extern BTF VAR has corresponding ELF symbol preserved, so we skip | |
2918 | * all fixups altogether for such sections and go straight to sorting | |
2919 | * VARs within their DATASEC. | |
b96c07f3 | 2920 | */ |
4fcac46c | 2921 | if (strcmp(sec_name, KCONFIG_SEC) == 0 || strcmp(sec_name, KSYMS_SEC) == 0) |
b96c07f3 AN |
2922 | goto sort_vars; |
2923 | ||
4fcac46c AN |
2924 | /* Clang leaves DATASEC size and VAR offsets as zeroes, so we need to |
2925 | * fix this up. But BPF static linker already fixes this up and fills | |
2926 | * all the sizes and offsets during static linking. So this step has | |
2927 | * to be optional. But the STV_HIDDEN handling is non-optional for any | |
2928 | * non-extern DATASEC, so the variable fixup loop below handles both | |
2929 | * functions at the same time, paying the cost of BTF VAR <-> ELF | |
2930 | * symbol matching just once. | |
2931 | */ | |
2932 | if (t->size == 0) { | |
2933 | err = find_elf_sec_sz(obj, sec_name, &size); | |
2934 | if (err || !size) { | |
2935 | pr_debug("sec '%s': failed to determine size from ELF: size %u, err %d\n", | |
2936 | sec_name, size, err); | |
2937 | return -ENOENT; | |
2938 | } | |
b96c07f3 | 2939 | |
4fcac46c AN |
2940 | t->size = size; |
2941 | fixup_offsets = true; | |
2942 | } | |
b96c07f3 AN |
2943 | |
2944 | for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) { | |
f33f742d AN |
2945 | const struct btf_type *t_var; |
2946 | struct btf_var *var; | |
2947 | const char *var_name; | |
2948 | Elf64_Sym *sym; | |
2949 | ||
b96c07f3 | 2950 | t_var = btf__type_by_id(btf, vsi->type); |
88918dc1 | 2951 | if (!t_var || !btf_is_var(t_var)) { |
f33f742d | 2952 | pr_debug("sec '%s': unexpected non-VAR type found\n", sec_name); |
b96c07f3 AN |
2953 | return -EINVAL; |
2954 | } | |
2955 | ||
88918dc1 | 2956 | var = btf_var(t_var); |
f33f742d | 2957 | if (var->linkage == BTF_VAR_STATIC || var->linkage == BTF_VAR_GLOBAL_EXTERN) |
b96c07f3 AN |
2958 | continue; |
2959 | ||
f33f742d AN |
2960 | var_name = btf__name_by_offset(btf, t_var->name_off); |
2961 | if (!var_name) { | |
2962 | pr_debug("sec '%s': failed to find name of DATASEC's member #%d\n", | |
2963 | sec_name, i); | |
b96c07f3 AN |
2964 | return -ENOENT; |
2965 | } | |
2966 | ||
f33f742d AN |
2967 | sym = find_elf_var_sym(obj, var_name); |
2968 | if (IS_ERR(sym)) { | |
2969 | pr_debug("sec '%s': failed to find ELF symbol for VAR '%s'\n", | |
2970 | sec_name, var_name); | |
b96c07f3 AN |
2971 | return -ENOENT; |
2972 | } | |
2973 | ||
4fcac46c AN |
2974 | if (fixup_offsets) |
2975 | vsi->offset = sym->st_value; | |
2976 | ||
2977 | /* if variable is a global/weak symbol, but has restricted | |
2978 | * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF VAR | |
2979 | * as static. This follows similar logic for functions (BPF | |
2980 | * subprogs) and influences libbpf's further decisions about | |
2981 | * whether to make global data BPF array maps as | |
2982 | * BPF_F_MMAPABLE. | |
2983 | */ | |
2984 | if (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN | |
2985 | || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL) | |
2986 | var->linkage = BTF_VAR_STATIC; | |
b96c07f3 AN |
2987 | } |
2988 | ||
2989 | sort_vars: | |
2990 | qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off); | |
2991 | return 0; | |
2992 | } | |
2993 | ||
f33f742d | 2994 | static int bpf_object_fixup_btf(struct bpf_object *obj) |
b96c07f3 | 2995 | { |
f33f742d | 2996 | int i, n, err = 0; |
b96c07f3 | 2997 | |
f33f742d AN |
2998 | if (!obj->btf) |
2999 | return 0; | |
3000 | ||
3001 | n = btf__type_cnt(obj->btf); | |
6a886de0 | 3002 | for (i = 1; i < n; i++) { |
f33f742d | 3003 | struct btf_type *t = btf_type_by_id(obj->btf, i); |
b96c07f3 AN |
3004 | |
3005 | /* Loader needs to fix up some of the things compiler | |
3006 | * couldn't get its hands on while emitting BTF. This | |
3007 | * is section size and global variable offset. We use | |
3008 | * the info from the ELF itself for this purpose. | |
3009 | */ | |
3010 | if (btf_is_datasec(t)) { | |
f33f742d | 3011 | err = btf_fixup_datasec(obj, obj->btf, t); |
b96c07f3 | 3012 | if (err) |
f33f742d | 3013 | return err; |
b96c07f3 AN |
3014 | } |
3015 | } | |
3016 | ||
166750bc AN |
3017 | return 0; |
3018 | } | |
3019 | ||
fe62de31 | 3020 | static bool prog_needs_vmlinux_btf(struct bpf_program *prog) |
a6ed02ca | 3021 | { |
1e092a03 KS |
3022 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS || |
3023 | prog->type == BPF_PROG_TYPE_LSM) | |
a6ed02ca KS |
3024 | return true; |
3025 | ||
3026 | /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs | |
3027 | * also need vmlinux BTF | |
3028 | */ | |
3029 | if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd) | |
3030 | return true; | |
3031 | ||
3032 | return false; | |
3033 | } | |
3034 | ||
fe62de31 | 3035 | static bool obj_needs_vmlinux_btf(const struct bpf_object *obj) |
a6ed02ca KS |
3036 | { |
3037 | struct bpf_program *prog; | |
fe62de31 | 3038 | int i; |
a6ed02ca | 3039 | |
1373ff59 SC |
3040 | /* CO-RE relocations need kernel BTF, only when btf_custom_path |
3041 | * is not specified | |
3042 | */ | |
3043 | if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path) | |
fe62de31 | 3044 | return true; |
192b6638 | 3045 | |
d370bbe1 HL |
3046 | /* Support for typed ksyms needs kernel BTF */ |
3047 | for (i = 0; i < obj->nr_extern; i++) { | |
3048 | const struct extern_desc *ext; | |
3049 | ||
3050 | ext = &obj->externs[i]; | |
fe62de31 AN |
3051 | if (ext->type == EXT_KSYM && ext->ksym.type_id) |
3052 | return true; | |
d370bbe1 HL |
3053 | } |
3054 | ||
a6ed02ca | 3055 | bpf_object__for_each_program(prog, obj) { |
a3820c48 | 3056 | if (!prog->autoload) |
d9297581 | 3057 | continue; |
fe62de31 AN |
3058 | if (prog_needs_vmlinux_btf(prog)) |
3059 | return true; | |
a6ed02ca KS |
3060 | } |
3061 | ||
fe62de31 AN |
3062 | return false; |
3063 | } | |
3064 | ||
3065 | static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force) | |
3066 | { | |
3067 | int err; | |
3068 | ||
3069 | /* btf_vmlinux could be loaded earlier */ | |
67234743 | 3070 | if (obj->btf_vmlinux || obj->gen_loader) |
fe62de31 AN |
3071 | return 0; |
3072 | ||
3073 | if (!force && !obj_needs_vmlinux_btf(obj)) | |
192b6638 AN |
3074 | return 0; |
3075 | ||
a710eed3 | 3076 | obj->btf_vmlinux = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
3077 | err = libbpf_get_error(obj->btf_vmlinux); |
3078 | if (err) { | |
192b6638 AN |
3079 | pr_warn("Error loading vmlinux BTF: %d\n", err); |
3080 | obj->btf_vmlinux = NULL; | |
3081 | return err; | |
3082 | } | |
a6ed02ca KS |
3083 | return 0; |
3084 | } | |
3085 | ||
063183bf AN |
3086 | static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj) |
3087 | { | |
0f0e55d8 AN |
3088 | struct btf *kern_btf = obj->btf; |
3089 | bool btf_mandatory, sanitize; | |
aea28a60 | 3090 | int i, err = 0; |
063183bf AN |
3091 | |
3092 | if (!obj->btf) | |
3093 | return 0; | |
3094 | ||
9ca1f56a | 3095 | if (!kernel_supports(obj, FEAT_BTF)) { |
68b08647 AN |
3096 | if (kernel_needs_btf(obj)) { |
3097 | err = -EOPNOTSUPP; | |
3098 | goto report; | |
3099 | } | |
3100 | pr_debug("Kernel doesn't support BTF, skipping uploading it.\n"); | |
3101 | return 0; | |
3102 | } | |
3103 | ||
aea28a60 AN |
3104 | /* Even though some subprogs are global/weak, user might prefer more |
3105 | * permissive BPF verification process that BPF verifier performs for | |
3106 | * static functions, taking into account more context from the caller | |
3107 | * functions. In such case, they need to mark such subprogs with | |
3108 | * __attribute__((visibility("hidden"))) and libbpf will adjust | |
3109 | * corresponding FUNC BTF type to be marked as static and trigger more | |
3110 | * involved BPF verification process. | |
3111 | */ | |
3112 | for (i = 0; i < obj->nr_programs; i++) { | |
3113 | struct bpf_program *prog = &obj->programs[i]; | |
3114 | struct btf_type *t; | |
3115 | const char *name; | |
3116 | int j, n; | |
3117 | ||
3118 | if (!prog->mark_btf_static || !prog_is_subprog(obj, prog)) | |
3119 | continue; | |
3120 | ||
6a886de0 HC |
3121 | n = btf__type_cnt(obj->btf); |
3122 | for (j = 1; j < n; j++) { | |
aea28a60 AN |
3123 | t = btf_type_by_id(obj->btf, j); |
3124 | if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) | |
3125 | continue; | |
3126 | ||
3127 | name = btf__str_by_offset(obj->btf, t->name_off); | |
3128 | if (strcmp(name, prog->name) != 0) | |
3129 | continue; | |
3130 | ||
3131 | t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0); | |
3132 | break; | |
3133 | } | |
3134 | } | |
3135 | ||
0f0e55d8 AN |
3136 | sanitize = btf_needs_sanitization(obj); |
3137 | if (sanitize) { | |
5c3320d7 | 3138 | const void *raw_data; |
0f0e55d8 | 3139 | __u32 sz; |
063183bf | 3140 | |
0f0e55d8 | 3141 | /* clone BTF to sanitize a copy and leave the original intact */ |
6a886de0 | 3142 | raw_data = btf__raw_data(obj->btf, &sz); |
5c3320d7 | 3143 | kern_btf = btf__new(raw_data, sz); |
e9fc3ce9 AN |
3144 | err = libbpf_get_error(kern_btf); |
3145 | if (err) | |
3146 | return err; | |
04efe591 | 3147 | |
4c01925f AN |
3148 | /* enforce 8-byte pointers for BPF-targeted BTFs */ |
3149 | btf__set_pointer_size(obj->btf, 8); | |
f2a62588 YS |
3150 | err = bpf_object__sanitize_btf(obj, kern_btf); |
3151 | if (err) | |
3152 | return err; | |
063183bf | 3153 | } |
0f0e55d8 | 3154 | |
67234743 AS |
3155 | if (obj->gen_loader) { |
3156 | __u32 raw_size = 0; | |
6a886de0 | 3157 | const void *raw_data = btf__raw_data(kern_btf, &raw_size); |
67234743 AS |
3158 | |
3159 | if (!raw_data) | |
3160 | return -ENOMEM; | |
3161 | bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size); | |
3162 | /* Pretend to have valid FD to pass various fd >= 0 checks. | |
3163 | * This fd == 0 will not be used with any syscall and will be reset to -1 eventually. | |
3164 | */ | |
3165 | btf__set_fd(kern_btf, 0); | |
3166 | } else { | |
e0e3ea88 AN |
3167 | /* currently BPF_BTF_LOAD only supports log_level 1 */ |
3168 | err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size, | |
3169 | obj->log_level ? 1 : 0); | |
67234743 | 3170 | } |
0f0e55d8 AN |
3171 | if (sanitize) { |
3172 | if (!err) { | |
3173 | /* move fd to libbpf's BTF */ | |
3174 | btf__set_fd(obj->btf, btf__fd(kern_btf)); | |
3175 | btf__set_fd(kern_btf, -1); | |
3176 | } | |
3177 | btf__free(kern_btf); | |
3178 | } | |
68b08647 | 3179 | report: |
0f0e55d8 AN |
3180 | if (err) { |
3181 | btf_mandatory = kernel_needs_btf(obj); | |
3182 | pr_warn("Error loading .BTF into kernel: %d. %s\n", err, | |
3183 | btf_mandatory ? "BTF is mandatory, can't proceed." | |
3184 | : "BTF is optional, ignoring."); | |
3185 | if (!btf_mandatory) | |
3186 | err = 0; | |
3187 | } | |
3188 | return err; | |
063183bf AN |
3189 | } |
3190 | ||
88a82120 AN |
3191 | static const char *elf_sym_str(const struct bpf_object *obj, size_t off) |
3192 | { | |
3193 | const char *name; | |
3194 | ||
3195 | name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off); | |
3196 | if (!name) { | |
3197 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3198 | off, obj->path, elf_errmsg(-1)); | |
3199 | return NULL; | |
3200 | } | |
3201 | ||
3202 | return name; | |
3203 | } | |
3204 | ||
3205 | static const char *elf_sec_str(const struct bpf_object *obj, size_t off) | |
3206 | { | |
3207 | const char *name; | |
3208 | ||
3209 | name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off); | |
3210 | if (!name) { | |
3211 | pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n", | |
3212 | off, obj->path, elf_errmsg(-1)); | |
3213 | return NULL; | |
3214 | } | |
3215 | ||
3216 | return name; | |
3217 | } | |
3218 | ||
3219 | static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx) | |
3220 | { | |
3221 | Elf_Scn *scn; | |
3222 | ||
3223 | scn = elf_getscn(obj->efile.elf, idx); | |
3224 | if (!scn) { | |
3225 | pr_warn("elf: failed to get section(%zu) from %s: %s\n", | |
3226 | idx, obj->path, elf_errmsg(-1)); | |
3227 | return NULL; | |
3228 | } | |
3229 | return scn; | |
3230 | } | |
3231 | ||
3232 | static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name) | |
3233 | { | |
3234 | Elf_Scn *scn = NULL; | |
3235 | Elf *elf = obj->efile.elf; | |
3236 | const char *sec_name; | |
3237 | ||
3238 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
3239 | sec_name = elf_sec_name(obj, scn); | |
3240 | if (!sec_name) | |
3241 | return NULL; | |
3242 | ||
3243 | if (strcmp(sec_name, name) != 0) | |
3244 | continue; | |
3245 | ||
3246 | return scn; | |
3247 | } | |
3248 | return NULL; | |
3249 | } | |
3250 | ||
ad23b723 | 3251 | static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn) |
88a82120 | 3252 | { |
ad23b723 AN |
3253 | Elf64_Shdr *shdr; |
3254 | ||
88a82120 | 3255 | if (!scn) |
ad23b723 | 3256 | return NULL; |
88a82120 | 3257 | |
ad23b723 AN |
3258 | shdr = elf64_getshdr(scn); |
3259 | if (!shdr) { | |
88a82120 AN |
3260 | pr_warn("elf: failed to get section(%zu) header from %s: %s\n", |
3261 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
ad23b723 | 3262 | return NULL; |
88a82120 AN |
3263 | } |
3264 | ||
ad23b723 | 3265 | return shdr; |
88a82120 AN |
3266 | } |
3267 | ||
3268 | static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn) | |
3269 | { | |
3270 | const char *name; | |
ad23b723 | 3271 | Elf64_Shdr *sh; |
88a82120 AN |
3272 | |
3273 | if (!scn) | |
3274 | return NULL; | |
3275 | ||
ad23b723 AN |
3276 | sh = elf_sec_hdr(obj, scn); |
3277 | if (!sh) | |
88a82120 AN |
3278 | return NULL; |
3279 | ||
ad23b723 | 3280 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 AN |
3281 | if (!name) { |
3282 | pr_warn("elf: failed to get section(%zu) name from %s: %s\n", | |
3283 | elf_ndxscn(scn), obj->path, elf_errmsg(-1)); | |
3284 | return NULL; | |
3285 | } | |
3286 | ||
3287 | return name; | |
3288 | } | |
3289 | ||
3290 | static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn) | |
3291 | { | |
3292 | Elf_Data *data; | |
3293 | ||
3294 | if (!scn) | |
3295 | return NULL; | |
3296 | ||
3297 | data = elf_getdata(scn, 0); | |
3298 | if (!data) { | |
3299 | pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n", | |
3300 | elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>", | |
3301 | obj->path, elf_errmsg(-1)); | |
3302 | return NULL; | |
3303 | } | |
3304 | ||
3305 | return data; | |
3306 | } | |
3307 | ||
ad23b723 AN |
3308 | static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx) |
3309 | { | |
3310 | if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym)) | |
3311 | return NULL; | |
3312 | ||
3313 | return (Elf64_Sym *)obj->efile.symbols->d_buf + idx; | |
3314 | } | |
3315 | ||
3316 | static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx) | |
3317 | { | |
3318 | if (idx >= data->d_size / sizeof(Elf64_Rel)) | |
3319 | return NULL; | |
3320 | ||
3321 | return (Elf64_Rel *)data->d_buf + idx; | |
3322 | } | |
3323 | ||
50e09460 AN |
3324 | static bool is_sec_name_dwarf(const char *name) |
3325 | { | |
3326 | /* approximation, but the actual list is too long */ | |
13d35a0c | 3327 | return str_has_pfx(name, ".debug_"); |
50e09460 AN |
3328 | } |
3329 | ||
ad23b723 | 3330 | static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name) |
50e09460 AN |
3331 | { |
3332 | /* no special handling of .strtab */ | |
3333 | if (hdr->sh_type == SHT_STRTAB) | |
3334 | return true; | |
3335 | ||
3336 | /* ignore .llvm_addrsig section as well */ | |
faf6ed32 | 3337 | if (hdr->sh_type == SHT_LLVM_ADDRSIG) |
50e09460 AN |
3338 | return true; |
3339 | ||
3340 | /* no subprograms will lead to an empty .text section, ignore it */ | |
3341 | if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 && | |
3342 | strcmp(name, ".text") == 0) | |
3343 | return true; | |
3344 | ||
3345 | /* DWARF sections */ | |
3346 | if (is_sec_name_dwarf(name)) | |
3347 | return true; | |
3348 | ||
13d35a0c | 3349 | if (str_has_pfx(name, ".rel")) { |
50e09460 AN |
3350 | name += sizeof(".rel") - 1; |
3351 | /* DWARF section relocations */ | |
3352 | if (is_sec_name_dwarf(name)) | |
3353 | return true; | |
3354 | ||
3355 | /* .BTF and .BTF.ext don't need relocations */ | |
3356 | if (strcmp(name, BTF_ELF_SEC) == 0 || | |
3357 | strcmp(name, BTF_EXT_ELF_SEC) == 0) | |
3358 | return true; | |
3359 | } | |
3360 | ||
3361 | return false; | |
3362 | } | |
3363 | ||
db2b8b06 AN |
3364 | static int cmp_progs(const void *_a, const void *_b) |
3365 | { | |
3366 | const struct bpf_program *a = _a; | |
3367 | const struct bpf_program *b = _b; | |
3368 | ||
3369 | if (a->sec_idx != b->sec_idx) | |
3370 | return a->sec_idx < b->sec_idx ? -1 : 1; | |
3371 | ||
3372 | /* sec_insn_off can't be the same within the section */ | |
3373 | return a->sec_insn_off < b->sec_insn_off ? -1 : 1; | |
3374 | } | |
3375 | ||
0d13bfce | 3376 | static int bpf_object__elf_collect(struct bpf_object *obj) |
29603665 | 3377 | { |
25bbbd7a | 3378 | struct elf_sec_desc *sec_desc; |
29603665 | 3379 | Elf *elf = obj->efile.elf; |
f0187f0b | 3380 | Elf_Data *btf_ext_data = NULL; |
1713d68b | 3381 | Elf_Data *btf_data = NULL; |
666810e8 | 3382 | int idx = 0, err = 0; |
0201c575 AN |
3383 | const char *name; |
3384 | Elf_Data *data; | |
3385 | Elf_Scn *scn; | |
ad23b723 | 3386 | Elf64_Shdr *sh; |
29603665 | 3387 | |
0d6988e1 | 3388 | /* ELF section indices are 0-based, but sec #0 is special "invalid" |
51deedc9 SHY |
3389 | * section. Since section count retrieved by elf_getshdrnum() does |
3390 | * include sec #0, it is already the necessary size of an array to keep | |
3391 | * all the sections. | |
25bbbd7a | 3392 | */ |
51deedc9 SHY |
3393 | if (elf_getshdrnum(obj->efile.elf, &obj->efile.sec_cnt)) { |
3394 | pr_warn("elf: failed to get the number of sections for %s: %s\n", | |
3395 | obj->path, elf_errmsg(-1)); | |
3396 | return -LIBBPF_ERRNO__FORMAT; | |
3397 | } | |
25bbbd7a AN |
3398 | obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs)); |
3399 | if (!obj->efile.secs) | |
3400 | return -ENOMEM; | |
29603665 | 3401 | |
0201c575 AN |
3402 | /* a bunch of ELF parsing functionality depends on processing symbols, |
3403 | * so do the first pass and find the symbol table | |
3404 | */ | |
3405 | scn = NULL; | |
29603665 | 3406 | while ((scn = elf_nextscn(elf, scn)) != NULL) { |
ad23b723 AN |
3407 | sh = elf_sec_hdr(obj, scn); |
3408 | if (!sh) | |
0201c575 AN |
3409 | return -LIBBPF_ERRNO__FORMAT; |
3410 | ||
ad23b723 | 3411 | if (sh->sh_type == SHT_SYMTAB) { |
0201c575 AN |
3412 | if (obj->efile.symbols) { |
3413 | pr_warn("elf: multiple symbol tables in %s\n", obj->path); | |
3414 | return -LIBBPF_ERRNO__FORMAT; | |
3415 | } | |
29603665 | 3416 | |
0201c575 AN |
3417 | data = elf_sec_data(obj, scn); |
3418 | if (!data) | |
3419 | return -LIBBPF_ERRNO__FORMAT; | |
3420 | ||
25bbbd7a AN |
3421 | idx = elf_ndxscn(scn); |
3422 | ||
0201c575 | 3423 | obj->efile.symbols = data; |
25bbbd7a | 3424 | obj->efile.symbols_shndx = idx; |
ad23b723 | 3425 | obj->efile.strtabidx = sh->sh_link; |
0201c575 AN |
3426 | } |
3427 | } | |
3428 | ||
03e601f4 THJ |
3429 | if (!obj->efile.symbols) { |
3430 | pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n", | |
3431 | obj->path); | |
3432 | return -ENOENT; | |
3433 | } | |
3434 | ||
0201c575 AN |
3435 | scn = NULL; |
3436 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
25bbbd7a AN |
3437 | idx = elf_ndxscn(scn); |
3438 | sec_desc = &obj->efile.secs[idx]; | |
88a82120 | 3439 | |
ad23b723 AN |
3440 | sh = elf_sec_hdr(obj, scn); |
3441 | if (!sh) | |
01b29d1d | 3442 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3443 | |
ad23b723 | 3444 | name = elf_sec_str(obj, sh->sh_name); |
88a82120 | 3445 | if (!name) |
01b29d1d | 3446 | return -LIBBPF_ERRNO__FORMAT; |
29603665 | 3447 | |
ad23b723 | 3448 | if (ignore_elf_section(sh, name)) |
50e09460 AN |
3449 | continue; |
3450 | ||
88a82120 AN |
3451 | data = elf_sec_data(obj, scn); |
3452 | if (!data) | |
01b29d1d | 3453 | return -LIBBPF_ERRNO__FORMAT; |
88a82120 AN |
3454 | |
3455 | pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n", | |
077c066a | 3456 | idx, name, (unsigned long)data->d_size, |
ad23b723 AN |
3457 | (int)sh->sh_link, (unsigned long)sh->sh_flags, |
3458 | (int)sh->sh_type); | |
cb1e5e96 | 3459 | |
1713d68b | 3460 | if (strcmp(name, "license") == 0) { |
88a82120 | 3461 | err = bpf_object__init_license(obj, data->d_buf, data->d_size); |
01b29d1d AN |
3462 | if (err) |
3463 | return err; | |
1713d68b | 3464 | } else if (strcmp(name, "version") == 0) { |
88a82120 | 3465 | err = bpf_object__init_kversion(obj, data->d_buf, data->d_size); |
54b8625c JF |
3466 | if (err) |
3467 | return err; | |
1713d68b | 3468 | } else if (strcmp(name, "maps") == 0) { |
e19db676 AN |
3469 | pr_warn("elf: legacy map definitions in 'maps' section are not supported by libbpf v1.0+\n"); |
3470 | return -ENOTSUP; | |
abd29c93 AN |
3471 | } else if (strcmp(name, MAPS_ELF_SEC) == 0) { |
3472 | obj->efile.btf_maps_shndx = idx; | |
1713d68b | 3473 | } else if (strcmp(name, BTF_ELF_SEC) == 0) { |
62554d52 AN |
3474 | if (sh->sh_type != SHT_PROGBITS) |
3475 | return -LIBBPF_ERRNO__FORMAT; | |
1713d68b | 3476 | btf_data = data; |
2993e051 | 3477 | } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) { |
62554d52 AN |
3478 | if (sh->sh_type != SHT_PROGBITS) |
3479 | return -LIBBPF_ERRNO__FORMAT; | |
f0187f0b | 3480 | btf_ext_data = data; |
ad23b723 | 3481 | } else if (sh->sh_type == SHT_SYMTAB) { |
0201c575 | 3482 | /* already processed during the first pass above */ |
ad23b723 AN |
3483 | } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) { |
3484 | if (sh->sh_flags & SHF_EXECINSTR) { | |
f8c7a4d4 JS |
3485 | if (strcmp(name, ".text") == 0) |
3486 | obj->efile.text_shndx = idx; | |
c1122392 | 3487 | err = bpf_object__add_programs(obj, data, name, idx); |
88a82120 | 3488 | if (err) |
01b29d1d | 3489 | return err; |
aed65917 AN |
3490 | } else if (strcmp(name, DATA_SEC) == 0 || |
3491 | str_has_pfx(name, DATA_SEC ".")) { | |
25bbbd7a AN |
3492 | sec_desc->sec_type = SEC_DATA; |
3493 | sec_desc->shdr = sh; | |
3494 | sec_desc->data = data; | |
aed65917 AN |
3495 | } else if (strcmp(name, RODATA_SEC) == 0 || |
3496 | str_has_pfx(name, RODATA_SEC ".")) { | |
25bbbd7a AN |
3497 | sec_desc->sec_type = SEC_RODATA; |
3498 | sec_desc->shdr = sh; | |
3499 | sec_desc->data = data; | |
590a0088 MKL |
3500 | } else if (strcmp(name, STRUCT_OPS_SEC) == 0) { |
3501 | obj->efile.st_ops_data = data; | |
3502 | obj->efile.st_ops_shndx = idx; | |
809a69d6 KFL |
3503 | } else if (strcmp(name, STRUCT_OPS_LINK_SEC) == 0) { |
3504 | obj->efile.st_ops_link_data = data; | |
3505 | obj->efile.st_ops_link_shndx = idx; | |
d859900c | 3506 | } else { |
50e09460 AN |
3507 | pr_info("elf: skipping unrecognized data section(%d) %s\n", |
3508 | idx, name); | |
a5b8bd47 | 3509 | } |
ad23b723 | 3510 | } else if (sh->sh_type == SHT_REL) { |
25bbbd7a | 3511 | int targ_sec_idx = sh->sh_info; /* points to other section */ |
e3d91b0c | 3512 | |
b7332d28 AN |
3513 | if (sh->sh_entsize != sizeof(Elf64_Rel) || |
3514 | targ_sec_idx >= obj->efile.sec_cnt) | |
3515 | return -LIBBPF_ERRNO__FORMAT; | |
3516 | ||
e3d91b0c | 3517 | /* Only do relo for section with exec instructions */ |
25bbbd7a | 3518 | if (!section_have_execinstr(obj, targ_sec_idx) && |
646f02ff | 3519 | strcmp(name, ".rel" STRUCT_OPS_SEC) && |
809a69d6 | 3520 | strcmp(name, ".rel" STRUCT_OPS_LINK_SEC) && |
646f02ff | 3521 | strcmp(name, ".rel" MAPS_ELF_SEC)) { |
50e09460 | 3522 | pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n", |
25bbbd7a AN |
3523 | idx, name, targ_sec_idx, |
3524 | elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>"); | |
e3d91b0c JDB |
3525 | continue; |
3526 | } | |
b62f06e8 | 3527 | |
25bbbd7a AN |
3528 | sec_desc->sec_type = SEC_RELO; |
3529 | sec_desc->shdr = sh; | |
3530 | sec_desc->data = data; | |
dc79f035 AS |
3531 | } else if (sh->sh_type == SHT_NOBITS && (strcmp(name, BSS_SEC) == 0 || |
3532 | str_has_pfx(name, BSS_SEC "."))) { | |
25bbbd7a AN |
3533 | sec_desc->sec_type = SEC_BSS; |
3534 | sec_desc->shdr = sh; | |
3535 | sec_desc->data = data; | |
077c066a | 3536 | } else { |
2e80be60 | 3537 | pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name, |
ad23b723 | 3538 | (size_t)sh->sh_size); |
bec7d68c | 3539 | } |
29603665 | 3540 | } |
561bbcca | 3541 | |
d3a3aa0c | 3542 | if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) { |
88a82120 | 3543 | pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path); |
f102154d | 3544 | return -LIBBPF_ERRNO__FORMAT; |
77ba9a5b | 3545 | } |
db2b8b06 AN |
3546 | |
3547 | /* sort BPF programs by section name and in-section instruction offset | |
e3ba8e4e KM |
3548 | * for faster search |
3549 | */ | |
2a6a9bf2 AN |
3550 | if (obj->nr_programs) |
3551 | qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs); | |
db2b8b06 | 3552 | |
0d13bfce | 3553 | return bpf_object__init_btf(obj, btf_data, btf_ext_data); |
29603665 WN |
3554 | } |
3555 | ||
ad23b723 | 3556 | static bool sym_is_extern(const Elf64_Sym *sym) |
166750bc | 3557 | { |
ad23b723 | 3558 | int bind = ELF64_ST_BIND(sym->st_info); |
166750bc AN |
3559 | /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */ |
3560 | return sym->st_shndx == SHN_UNDEF && | |
3561 | (bind == STB_GLOBAL || bind == STB_WEAK) && | |
ad23b723 | 3562 | ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE; |
166750bc AN |
3563 | } |
3564 | ||
ad23b723 | 3565 | static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx) |
53eddb5e | 3566 | { |
ad23b723 AN |
3567 | int bind = ELF64_ST_BIND(sym->st_info); |
3568 | int type = ELF64_ST_TYPE(sym->st_info); | |
53eddb5e YS |
3569 | |
3570 | /* in .text section */ | |
3571 | if (sym->st_shndx != text_shndx) | |
3572 | return false; | |
3573 | ||
3574 | /* local function */ | |
3575 | if (bind == STB_LOCAL && type == STT_SECTION) | |
3576 | return true; | |
3577 | ||
3578 | /* global function */ | |
3579 | return bind == STB_GLOBAL && type == STT_FUNC; | |
3580 | } | |
3581 | ||
166750bc AN |
3582 | static int find_extern_btf_id(const struct btf *btf, const char *ext_name) |
3583 | { | |
3584 | const struct btf_type *t; | |
5bd022ec | 3585 | const char *tname; |
166750bc AN |
3586 | int i, n; |
3587 | ||
3588 | if (!btf) | |
3589 | return -ESRCH; | |
3590 | ||
6a886de0 HC |
3591 | n = btf__type_cnt(btf); |
3592 | for (i = 1; i < n; i++) { | |
166750bc AN |
3593 | t = btf__type_by_id(btf, i); |
3594 | ||
5bd022ec | 3595 | if (!btf_is_var(t) && !btf_is_func(t)) |
166750bc AN |
3596 | continue; |
3597 | ||
5bd022ec MKL |
3598 | tname = btf__name_by_offset(btf, t->name_off); |
3599 | if (strcmp(tname, ext_name)) | |
166750bc AN |
3600 | continue; |
3601 | ||
5bd022ec MKL |
3602 | if (btf_is_var(t) && |
3603 | btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN) | |
3604 | return -EINVAL; | |
3605 | ||
3606 | if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN) | |
166750bc AN |
3607 | return -EINVAL; |
3608 | ||
3609 | return i; | |
3610 | } | |
3611 | ||
3612 | return -ENOENT; | |
3613 | } | |
3614 | ||
2e33efe3 AN |
3615 | static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) { |
3616 | const struct btf_var_secinfo *vs; | |
3617 | const struct btf_type *t; | |
3618 | int i, j, n; | |
3619 | ||
3620 | if (!btf) | |
3621 | return -ESRCH; | |
3622 | ||
6a886de0 HC |
3623 | n = btf__type_cnt(btf); |
3624 | for (i = 1; i < n; i++) { | |
2e33efe3 AN |
3625 | t = btf__type_by_id(btf, i); |
3626 | ||
3627 | if (!btf_is_datasec(t)) | |
3628 | continue; | |
3629 | ||
3630 | vs = btf_var_secinfos(t); | |
3631 | for (j = 0; j < btf_vlen(t); j++, vs++) { | |
3632 | if (vs->type == ext_btf_id) | |
3633 | return i; | |
3634 | } | |
3635 | } | |
3636 | ||
3637 | return -ENOENT; | |
3638 | } | |
3639 | ||
3640 | static enum kcfg_type find_kcfg_type(const struct btf *btf, int id, | |
3641 | bool *is_signed) | |
166750bc AN |
3642 | { |
3643 | const struct btf_type *t; | |
3644 | const char *name; | |
3645 | ||
3646 | t = skip_mods_and_typedefs(btf, id, NULL); | |
3647 | name = btf__name_by_offset(btf, t->name_off); | |
3648 | ||
3649 | if (is_signed) | |
3650 | *is_signed = false; | |
3651 | switch (btf_kind(t)) { | |
3652 | case BTF_KIND_INT: { | |
3653 | int enc = btf_int_encoding(t); | |
3654 | ||
3655 | if (enc & BTF_INT_BOOL) | |
2e33efe3 | 3656 | return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN; |
166750bc AN |
3657 | if (is_signed) |
3658 | *is_signed = enc & BTF_INT_SIGNED; | |
3659 | if (t->size == 1) | |
2e33efe3 | 3660 | return KCFG_CHAR; |
166750bc | 3661 | if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1))) |
2e33efe3 AN |
3662 | return KCFG_UNKNOWN; |
3663 | return KCFG_INT; | |
166750bc AN |
3664 | } |
3665 | case BTF_KIND_ENUM: | |
3666 | if (t->size != 4) | |
2e33efe3 | 3667 | return KCFG_UNKNOWN; |
166750bc | 3668 | if (strcmp(name, "libbpf_tristate")) |
2e33efe3 AN |
3669 | return KCFG_UNKNOWN; |
3670 | return KCFG_TRISTATE; | |
f2a62588 YS |
3671 | case BTF_KIND_ENUM64: |
3672 | if (strcmp(name, "libbpf_tristate")) | |
3673 | return KCFG_UNKNOWN; | |
3674 | return KCFG_TRISTATE; | |
166750bc AN |
3675 | case BTF_KIND_ARRAY: |
3676 | if (btf_array(t)->nelems == 0) | |
2e33efe3 AN |
3677 | return KCFG_UNKNOWN; |
3678 | if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR) | |
3679 | return KCFG_UNKNOWN; | |
3680 | return KCFG_CHAR_ARR; | |
166750bc | 3681 | default: |
2e33efe3 | 3682 | return KCFG_UNKNOWN; |
166750bc AN |
3683 | } |
3684 | } | |
3685 | ||
3686 | static int cmp_externs(const void *_a, const void *_b) | |
3687 | { | |
3688 | const struct extern_desc *a = _a; | |
3689 | const struct extern_desc *b = _b; | |
3690 | ||
2e33efe3 AN |
3691 | if (a->type != b->type) |
3692 | return a->type < b->type ? -1 : 1; | |
3693 | ||
3694 | if (a->type == EXT_KCFG) { | |
3695 | /* descending order by alignment requirements */ | |
3696 | if (a->kcfg.align != b->kcfg.align) | |
3697 | return a->kcfg.align > b->kcfg.align ? -1 : 1; | |
3698 | /* ascending order by size, within same alignment class */ | |
3699 | if (a->kcfg.sz != b->kcfg.sz) | |
3700 | return a->kcfg.sz < b->kcfg.sz ? -1 : 1; | |
3701 | } | |
3702 | ||
166750bc AN |
3703 | /* resolve ties by name */ |
3704 | return strcmp(a->name, b->name); | |
3705 | } | |
3706 | ||
1c0c7074 AN |
3707 | static int find_int_btf_id(const struct btf *btf) |
3708 | { | |
3709 | const struct btf_type *t; | |
3710 | int i, n; | |
3711 | ||
6a886de0 HC |
3712 | n = btf__type_cnt(btf); |
3713 | for (i = 1; i < n; i++) { | |
1c0c7074 AN |
3714 | t = btf__type_by_id(btf, i); |
3715 | ||
3716 | if (btf_is_int(t) && btf_int_bits(t) == 32) | |
3717 | return i; | |
3718 | } | |
3719 | ||
3720 | return 0; | |
3721 | } | |
3722 | ||
5bd022ec MKL |
3723 | static int add_dummy_ksym_var(struct btf *btf) |
3724 | { | |
3725 | int i, int_btf_id, sec_btf_id, dummy_var_btf_id; | |
3726 | const struct btf_var_secinfo *vs; | |
3727 | const struct btf_type *sec; | |
3728 | ||
9683e577 IR |
3729 | if (!btf) |
3730 | return 0; | |
3731 | ||
5bd022ec MKL |
3732 | sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC, |
3733 | BTF_KIND_DATASEC); | |
3734 | if (sec_btf_id < 0) | |
3735 | return 0; | |
3736 | ||
3737 | sec = btf__type_by_id(btf, sec_btf_id); | |
3738 | vs = btf_var_secinfos(sec); | |
3739 | for (i = 0; i < btf_vlen(sec); i++, vs++) { | |
3740 | const struct btf_type *vt; | |
3741 | ||
3742 | vt = btf__type_by_id(btf, vs->type); | |
3743 | if (btf_is_func(vt)) | |
3744 | break; | |
3745 | } | |
3746 | ||
3747 | /* No func in ksyms sec. No need to add dummy var. */ | |
3748 | if (i == btf_vlen(sec)) | |
3749 | return 0; | |
3750 | ||
3751 | int_btf_id = find_int_btf_id(btf); | |
3752 | dummy_var_btf_id = btf__add_var(btf, | |
3753 | "dummy_ksym", | |
3754 | BTF_VAR_GLOBAL_ALLOCATED, | |
3755 | int_btf_id); | |
3756 | if (dummy_var_btf_id < 0) | |
3757 | pr_warn("cannot create a dummy_ksym var\n"); | |
3758 | ||
3759 | return dummy_var_btf_id; | |
3760 | } | |
3761 | ||
166750bc AN |
3762 | static int bpf_object__collect_externs(struct bpf_object *obj) |
3763 | { | |
1c0c7074 | 3764 | struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL; |
166750bc AN |
3765 | const struct btf_type *t; |
3766 | struct extern_desc *ext; | |
5bd022ec | 3767 | int i, n, off, dummy_var_btf_id; |
2e33efe3 | 3768 | const char *ext_name, *sec_name; |
166750bc | 3769 | Elf_Scn *scn; |
ad23b723 | 3770 | Elf64_Shdr *sh; |
166750bc AN |
3771 | |
3772 | if (!obj->efile.symbols) | |
3773 | return 0; | |
3774 | ||
88a82120 | 3775 | scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx); |
ad23b723 | 3776 | sh = elf_sec_hdr(obj, scn); |
83390787 | 3777 | if (!sh || sh->sh_entsize != sizeof(Elf64_Sym)) |
166750bc | 3778 | return -LIBBPF_ERRNO__FORMAT; |
166750bc | 3779 | |
5bd022ec MKL |
3780 | dummy_var_btf_id = add_dummy_ksym_var(obj->btf); |
3781 | if (dummy_var_btf_id < 0) | |
3782 | return dummy_var_btf_id; | |
3783 | ||
ad23b723 | 3784 | n = sh->sh_size / sh->sh_entsize; |
166750bc | 3785 | pr_debug("looking for externs among %d symbols...\n", n); |
88a82120 | 3786 | |
166750bc | 3787 | for (i = 0; i < n; i++) { |
ad23b723 | 3788 | Elf64_Sym *sym = elf_sym_by_idx(obj, i); |
166750bc | 3789 | |
ad23b723 | 3790 | if (!sym) |
166750bc | 3791 | return -LIBBPF_ERRNO__FORMAT; |
ad23b723 | 3792 | if (!sym_is_extern(sym)) |
166750bc | 3793 | continue; |
ad23b723 | 3794 | ext_name = elf_sym_str(obj, sym->st_name); |
166750bc AN |
3795 | if (!ext_name || !ext_name[0]) |
3796 | continue; | |
3797 | ||
3798 | ext = obj->externs; | |
029258d7 | 3799 | ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext)); |
166750bc AN |
3800 | if (!ext) |
3801 | return -ENOMEM; | |
3802 | obj->externs = ext; | |
3803 | ext = &ext[obj->nr_extern]; | |
3804 | memset(ext, 0, sizeof(*ext)); | |
3805 | obj->nr_extern++; | |
3806 | ||
3807 | ext->btf_id = find_extern_btf_id(obj->btf, ext_name); | |
3808 | if (ext->btf_id <= 0) { | |
3809 | pr_warn("failed to find BTF for extern '%s': %d\n", | |
3810 | ext_name, ext->btf_id); | |
3811 | return ext->btf_id; | |
3812 | } | |
3813 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
3814 | ext->name = btf__name_by_offset(obj->btf, t->name_off); | |
3815 | ext->sym_idx = i; | |
ad23b723 | 3816 | ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK; |
2e33efe3 AN |
3817 | |
3818 | ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id); | |
3819 | if (ext->sec_btf_id <= 0) { | |
3820 | pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n", | |
3821 | ext_name, ext->btf_id, ext->sec_btf_id); | |
3822 | return ext->sec_btf_id; | |
166750bc | 3823 | } |
2e33efe3 AN |
3824 | sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id); |
3825 | sec_name = btf__name_by_offset(obj->btf, sec->name_off); | |
3826 | ||
3827 | if (strcmp(sec_name, KCONFIG_SEC) == 0) { | |
5bd022ec MKL |
3828 | if (btf_is_func(t)) { |
3829 | pr_warn("extern function %s is unsupported under %s section\n", | |
3830 | ext->name, KCONFIG_SEC); | |
3831 | return -ENOTSUP; | |
3832 | } | |
2e33efe3 AN |
3833 | kcfg_sec = sec; |
3834 | ext->type = EXT_KCFG; | |
3835 | ext->kcfg.sz = btf__resolve_size(obj->btf, t->type); | |
3836 | if (ext->kcfg.sz <= 0) { | |
3837 | pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n", | |
3838 | ext_name, ext->kcfg.sz); | |
3839 | return ext->kcfg.sz; | |
3840 | } | |
3841 | ext->kcfg.align = btf__align_of(obj->btf, t->type); | |
3842 | if (ext->kcfg.align <= 0) { | |
3843 | pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n", | |
3844 | ext_name, ext->kcfg.align); | |
3845 | return -EINVAL; | |
3846 | } | |
3847 | ext->kcfg.type = find_kcfg_type(obj->btf, t->type, | |
e3ba8e4e | 3848 | &ext->kcfg.is_signed); |
2e33efe3 | 3849 | if (ext->kcfg.type == KCFG_UNKNOWN) { |
55d00c37 | 3850 | pr_warn("extern (kcfg) '%s': type is unsupported\n", ext_name); |
2e33efe3 AN |
3851 | return -ENOTSUP; |
3852 | } | |
1c0c7074 | 3853 | } else if (strcmp(sec_name, KSYMS_SEC) == 0) { |
1c0c7074 AN |
3854 | ksym_sec = sec; |
3855 | ext->type = EXT_KSYM; | |
d370bbe1 HL |
3856 | skip_mods_and_typedefs(obj->btf, t->type, |
3857 | &ext->ksym.type_id); | |
2e33efe3 AN |
3858 | } else { |
3859 | pr_warn("unrecognized extern section '%s'\n", sec_name); | |
166750bc AN |
3860 | return -ENOTSUP; |
3861 | } | |
3862 | } | |
3863 | pr_debug("collected %d externs total\n", obj->nr_extern); | |
3864 | ||
3865 | if (!obj->nr_extern) | |
3866 | return 0; | |
3867 | ||
2e33efe3 | 3868 | /* sort externs by type, for kcfg ones also by (align, size, name) */ |
166750bc | 3869 | qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs); |
166750bc | 3870 | |
1c0c7074 AN |
3871 | /* for .ksyms section, we need to turn all externs into allocated |
3872 | * variables in BTF to pass kernel verification; we do this by | |
3873 | * pretending that each extern is a 8-byte variable | |
3874 | */ | |
3875 | if (ksym_sec) { | |
3876 | /* find existing 4-byte integer type in BTF to use for fake | |
3877 | * extern variables in DATASEC | |
3878 | */ | |
3879 | int int_btf_id = find_int_btf_id(obj->btf); | |
5bd022ec MKL |
3880 | /* For extern function, a dummy_var added earlier |
3881 | * will be used to replace the vs->type and | |
3882 | * its name string will be used to refill | |
3883 | * the missing param's name. | |
3884 | */ | |
3885 | const struct btf_type *dummy_var; | |
1c0c7074 | 3886 | |
5bd022ec | 3887 | dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id); |
1c0c7074 AN |
3888 | for (i = 0; i < obj->nr_extern; i++) { |
3889 | ext = &obj->externs[i]; | |
3890 | if (ext->type != EXT_KSYM) | |
3891 | continue; | |
3892 | pr_debug("extern (ksym) #%d: symbol %d, name %s\n", | |
3893 | i, ext->sym_idx, ext->name); | |
3894 | } | |
3895 | ||
3896 | sec = ksym_sec; | |
3897 | n = btf_vlen(sec); | |
3898 | for (i = 0, off = 0; i < n; i++, off += sizeof(int)) { | |
3899 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
3900 | struct btf_type *vt; | |
3901 | ||
3902 | vt = (void *)btf__type_by_id(obj->btf, vs->type); | |
3903 | ext_name = btf__name_by_offset(obj->btf, vt->name_off); | |
3904 | ext = find_extern_by_name(obj, ext_name); | |
3905 | if (!ext) { | |
5bd022ec MKL |
3906 | pr_warn("failed to find extern definition for BTF %s '%s'\n", |
3907 | btf_kind_str(vt), ext_name); | |
1c0c7074 AN |
3908 | return -ESRCH; |
3909 | } | |
5bd022ec MKL |
3910 | if (btf_is_func(vt)) { |
3911 | const struct btf_type *func_proto; | |
3912 | struct btf_param *param; | |
3913 | int j; | |
3914 | ||
3915 | func_proto = btf__type_by_id(obj->btf, | |
3916 | vt->type); | |
3917 | param = btf_params(func_proto); | |
3918 | /* Reuse the dummy_var string if the | |
3919 | * func proto does not have param name. | |
3920 | */ | |
3921 | for (j = 0; j < btf_vlen(func_proto); j++) | |
3922 | if (param[j].type && !param[j].name_off) | |
3923 | param[j].name_off = | |
3924 | dummy_var->name_off; | |
3925 | vs->type = dummy_var_btf_id; | |
3926 | vt->info &= ~0xffff; | |
3927 | vt->info |= BTF_FUNC_GLOBAL; | |
3928 | } else { | |
3929 | btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
3930 | vt->type = int_btf_id; | |
3931 | } | |
1c0c7074 AN |
3932 | vs->offset = off; |
3933 | vs->size = sizeof(int); | |
3934 | } | |
3935 | sec->size = off; | |
3936 | } | |
3937 | ||
2e33efe3 AN |
3938 | if (kcfg_sec) { |
3939 | sec = kcfg_sec; | |
3940 | /* for kcfg externs calculate their offsets within a .kconfig map */ | |
3941 | off = 0; | |
3942 | for (i = 0; i < obj->nr_extern; i++) { | |
3943 | ext = &obj->externs[i]; | |
3944 | if (ext->type != EXT_KCFG) | |
3945 | continue; | |
166750bc | 3946 | |
2e33efe3 AN |
3947 | ext->kcfg.data_off = roundup(off, ext->kcfg.align); |
3948 | off = ext->kcfg.data_off + ext->kcfg.sz; | |
1c0c7074 | 3949 | pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n", |
2e33efe3 AN |
3950 | i, ext->sym_idx, ext->kcfg.data_off, ext->name); |
3951 | } | |
3952 | sec->size = off; | |
3953 | n = btf_vlen(sec); | |
3954 | for (i = 0; i < n; i++) { | |
3955 | struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i; | |
3956 | ||
3957 | t = btf__type_by_id(obj->btf, vs->type); | |
3958 | ext_name = btf__name_by_offset(obj->btf, t->name_off); | |
3959 | ext = find_extern_by_name(obj, ext_name); | |
3960 | if (!ext) { | |
3961 | pr_warn("failed to find extern definition for BTF var '%s'\n", | |
3962 | ext_name); | |
3963 | return -ESRCH; | |
3964 | } | |
3965 | btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED; | |
3966 | vs->offset = ext->kcfg.data_off; | |
166750bc | 3967 | } |
166750bc | 3968 | } |
166750bc AN |
3969 | return 0; |
3970 | } | |
3971 | ||
bd054102 AN |
3972 | static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog) |
3973 | { | |
197afc63 | 3974 | return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1; |
c3c55696 AN |
3975 | } |
3976 | ||
01af3bf0 AN |
3977 | struct bpf_program * |
3978 | bpf_object__find_program_by_name(const struct bpf_object *obj, | |
3979 | const char *name) | |
3980 | { | |
3981 | struct bpf_program *prog; | |
3982 | ||
3983 | bpf_object__for_each_program(prog, obj) { | |
c3c55696 AN |
3984 | if (prog_is_subprog(obj, prog)) |
3985 | continue; | |
01af3bf0 AN |
3986 | if (!strcmp(prog->name, name)) |
3987 | return prog; | |
3988 | } | |
e9fc3ce9 | 3989 | return errno = ENOENT, NULL; |
01af3bf0 AN |
3990 | } |
3991 | ||
d859900c DB |
3992 | static bool bpf_object__shndx_is_data(const struct bpf_object *obj, |
3993 | int shndx) | |
3994 | { | |
25bbbd7a AN |
3995 | switch (obj->efile.secs[shndx].sec_type) { |
3996 | case SEC_BSS: | |
3997 | case SEC_DATA: | |
3998 | case SEC_RODATA: | |
3999 | return true; | |
4000 | default: | |
4001 | return false; | |
4002 | } | |
d859900c DB |
4003 | } |
4004 | ||
4005 | static bool bpf_object__shndx_is_maps(const struct bpf_object *obj, | |
4006 | int shndx) | |
4007 | { | |
e19db676 | 4008 | return shndx == obj->efile.btf_maps_shndx; |
d859900c DB |
4009 | } |
4010 | ||
d859900c DB |
4011 | static enum libbpf_map_type |
4012 | bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx) | |
4013 | { | |
25bbbd7a AN |
4014 | if (shndx == obj->efile.symbols_shndx) |
4015 | return LIBBPF_MAP_KCONFIG; | |
4016 | ||
4017 | switch (obj->efile.secs[shndx].sec_type) { | |
4018 | case SEC_BSS: | |
d859900c | 4019 | return LIBBPF_MAP_BSS; |
25bbbd7a AN |
4020 | case SEC_DATA: |
4021 | return LIBBPF_MAP_DATA; | |
4022 | case SEC_RODATA: | |
d859900c | 4023 | return LIBBPF_MAP_RODATA; |
25bbbd7a | 4024 | default: |
d859900c | 4025 | return LIBBPF_MAP_UNSPEC; |
25bbbd7a | 4026 | } |
d859900c DB |
4027 | } |
4028 | ||
1f8e2bcb AN |
4029 | static int bpf_program__record_reloc(struct bpf_program *prog, |
4030 | struct reloc_desc *reloc_desc, | |
9c0f8cbd | 4031 | __u32 insn_idx, const char *sym_name, |
ad23b723 | 4032 | const Elf64_Sym *sym, const Elf64_Rel *rel) |
1f8e2bcb AN |
4033 | { |
4034 | struct bpf_insn *insn = &prog->insns[insn_idx]; | |
4035 | size_t map_idx, nr_maps = prog->obj->nr_maps; | |
4036 | struct bpf_object *obj = prog->obj; | |
4037 | __u32 shdr_idx = sym->st_shndx; | |
4038 | enum libbpf_map_type type; | |
9c0f8cbd | 4039 | const char *sym_sec_name; |
1f8e2bcb AN |
4040 | struct bpf_map *map; |
4041 | ||
aa0b8d43 | 4042 | if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) { |
9c0f8cbd AN |
4043 | pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n", |
4044 | prog->name, sym_name, insn_idx, insn->code); | |
1f8e2bcb AN |
4045 | return -LIBBPF_ERRNO__RELOC; |
4046 | } | |
166750bc AN |
4047 | |
4048 | if (sym_is_extern(sym)) { | |
ad23b723 | 4049 | int sym_idx = ELF64_R_SYM(rel->r_info); |
166750bc AN |
4050 | int i, n = obj->nr_extern; |
4051 | struct extern_desc *ext; | |
4052 | ||
4053 | for (i = 0; i < n; i++) { | |
4054 | ext = &obj->externs[i]; | |
4055 | if (ext->sym_idx == sym_idx) | |
4056 | break; | |
4057 | } | |
4058 | if (i >= n) { | |
9c0f8cbd AN |
4059 | pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n", |
4060 | prog->name, sym_name, sym_idx); | |
166750bc AN |
4061 | return -LIBBPF_ERRNO__RELOC; |
4062 | } | |
9c0f8cbd AN |
4063 | pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n", |
4064 | prog->name, i, ext->name, ext->sym_idx, insn_idx); | |
5bd022ec | 4065 | if (insn->code == (BPF_JMP | BPF_CALL)) |
a18f7214 | 4066 | reloc_desc->type = RELO_EXTERN_CALL; |
5bd022ec | 4067 | else |
a18f7214 | 4068 | reloc_desc->type = RELO_EXTERN_LD64; |
166750bc | 4069 | reloc_desc->insn_idx = insn_idx; |
3055ddd6 | 4070 | reloc_desc->ext_idx = i; |
166750bc AN |
4071 | return 0; |
4072 | } | |
4073 | ||
aa0b8d43 MKL |
4074 | /* sub-program call relocation */ |
4075 | if (is_call_insn(insn)) { | |
4076 | if (insn->src_reg != BPF_PSEUDO_CALL) { | |
4077 | pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name); | |
4078 | return -LIBBPF_ERRNO__RELOC; | |
4079 | } | |
4080 | /* text_shndx can be 0, if no default "main" program exists */ | |
4081 | if (!shdr_idx || shdr_idx != obj->efile.text_shndx) { | |
4082 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); | |
4083 | pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n", | |
4084 | prog->name, sym_name, sym_sec_name); | |
4085 | return -LIBBPF_ERRNO__RELOC; | |
4086 | } | |
4087 | if (sym->st_value % BPF_INSN_SZ) { | |
4088 | pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n", | |
4089 | prog->name, sym_name, (size_t)sym->st_value); | |
4090 | return -LIBBPF_ERRNO__RELOC; | |
4091 | } | |
4092 | reloc_desc->type = RELO_CALL; | |
4093 | reloc_desc->insn_idx = insn_idx; | |
4094 | reloc_desc->sym_off = sym->st_value; | |
4095 | return 0; | |
4096 | } | |
4097 | ||
1f8e2bcb | 4098 | if (!shdr_idx || shdr_idx >= SHN_LORESERVE) { |
9c0f8cbd AN |
4099 | pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n", |
4100 | prog->name, sym_name, shdr_idx); | |
1f8e2bcb AN |
4101 | return -LIBBPF_ERRNO__RELOC; |
4102 | } | |
4103 | ||
53eddb5e YS |
4104 | /* loading subprog addresses */ |
4105 | if (sym_is_subprog(sym, obj->efile.text_shndx)) { | |
4106 | /* global_func: sym->st_value = offset in the section, insn->imm = 0. | |
4107 | * local_func: sym->st_value = 0, insn->imm = offset in the section. | |
4108 | */ | |
4109 | if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) { | |
4110 | pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n", | |
4111 | prog->name, sym_name, (size_t)sym->st_value, insn->imm); | |
4112 | return -LIBBPF_ERRNO__RELOC; | |
4113 | } | |
4114 | ||
4115 | reloc_desc->type = RELO_SUBPROG_ADDR; | |
4116 | reloc_desc->insn_idx = insn_idx; | |
4117 | reloc_desc->sym_off = sym->st_value; | |
4118 | return 0; | |
4119 | } | |
4120 | ||
1f8e2bcb | 4121 | type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx); |
9c0f8cbd | 4122 | sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx)); |
1f8e2bcb AN |
4123 | |
4124 | /* generic map reference relocation */ | |
4125 | if (type == LIBBPF_MAP_UNSPEC) { | |
4126 | if (!bpf_object__shndx_is_maps(obj, shdr_idx)) { | |
9c0f8cbd AN |
4127 | pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n", |
4128 | prog->name, sym_name, sym_sec_name); | |
1f8e2bcb AN |
4129 | return -LIBBPF_ERRNO__RELOC; |
4130 | } | |
4131 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { | |
4132 | map = &obj->maps[map_idx]; | |
4133 | if (map->libbpf_type != type || | |
4134 | map->sec_idx != sym->st_shndx || | |
4135 | map->sec_offset != sym->st_value) | |
4136 | continue; | |
9c0f8cbd AN |
4137 | pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n", |
4138 | prog->name, map_idx, map->name, map->sec_idx, | |
1f8e2bcb AN |
4139 | map->sec_offset, insn_idx); |
4140 | break; | |
4141 | } | |
4142 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
4143 | pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n", |
4144 | prog->name, sym_sec_name, (size_t)sym->st_value); | |
1f8e2bcb AN |
4145 | return -LIBBPF_ERRNO__RELOC; |
4146 | } | |
4147 | reloc_desc->type = RELO_LD64; | |
4148 | reloc_desc->insn_idx = insn_idx; | |
4149 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 4150 | reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */ |
1f8e2bcb AN |
4151 | return 0; |
4152 | } | |
4153 | ||
4154 | /* global data map relocation */ | |
4155 | if (!bpf_object__shndx_is_data(obj, shdr_idx)) { | |
9c0f8cbd AN |
4156 | pr_warn("prog '%s': bad data relo against section '%s'\n", |
4157 | prog->name, sym_sec_name); | |
1f8e2bcb | 4158 | return -LIBBPF_ERRNO__RELOC; |
1f8e2bcb | 4159 | } |
1f8e2bcb AN |
4160 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { |
4161 | map = &obj->maps[map_idx]; | |
25bbbd7a | 4162 | if (map->libbpf_type != type || map->sec_idx != sym->st_shndx) |
1f8e2bcb | 4163 | continue; |
9c0f8cbd AN |
4164 | pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n", |
4165 | prog->name, map_idx, map->name, map->sec_idx, | |
4166 | map->sec_offset, insn_idx); | |
1f8e2bcb AN |
4167 | break; |
4168 | } | |
4169 | if (map_idx >= nr_maps) { | |
9c0f8cbd AN |
4170 | pr_warn("prog '%s': data relo failed to find map for section '%s'\n", |
4171 | prog->name, sym_sec_name); | |
1f8e2bcb AN |
4172 | return -LIBBPF_ERRNO__RELOC; |
4173 | } | |
4174 | ||
4175 | reloc_desc->type = RELO_DATA; | |
4176 | reloc_desc->insn_idx = insn_idx; | |
4177 | reloc_desc->map_idx = map_idx; | |
53f8dd43 | 4178 | reloc_desc->sym_off = sym->st_value; |
1f8e2bcb AN |
4179 | return 0; |
4180 | } | |
4181 | ||
db2b8b06 AN |
4182 | static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx) |
4183 | { | |
4184 | return insn_idx >= prog->sec_insn_off && | |
4185 | insn_idx < prog->sec_insn_off + prog->sec_insn_cnt; | |
4186 | } | |
4187 | ||
4188 | static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj, | |
4189 | size_t sec_idx, size_t insn_idx) | |
4190 | { | |
4191 | int l = 0, r = obj->nr_programs - 1, m; | |
4192 | struct bpf_program *prog; | |
4193 | ||
d0d382f9 SHY |
4194 | if (!obj->nr_programs) |
4195 | return NULL; | |
4196 | ||
db2b8b06 AN |
4197 | while (l < r) { |
4198 | m = l + (r - l + 1) / 2; | |
4199 | prog = &obj->programs[m]; | |
4200 | ||
4201 | if (prog->sec_idx < sec_idx || | |
4202 | (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx)) | |
4203 | l = m; | |
4204 | else | |
4205 | r = m - 1; | |
4206 | } | |
4207 | /* matching program could be at index l, but it still might be the | |
4208 | * wrong one, so we need to double check conditions for the last time | |
4209 | */ | |
4210 | prog = &obj->programs[l]; | |
4211 | if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx)) | |
4212 | return prog; | |
4213 | return NULL; | |
4214 | } | |
4215 | ||
34090915 | 4216 | static int |
ad23b723 | 4217 | bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data) |
34090915 | 4218 | { |
9c0f8cbd | 4219 | const char *relo_sec_name, *sec_name; |
b7332d28 | 4220 | size_t sec_idx = shdr->sh_info, sym_idx; |
c3c55696 AN |
4221 | struct bpf_program *prog; |
4222 | struct reloc_desc *relos; | |
1f8e2bcb | 4223 | int err, i, nrels; |
c3c55696 AN |
4224 | const char *sym_name; |
4225 | __u32 insn_idx; | |
6245947c AN |
4226 | Elf_Scn *scn; |
4227 | Elf_Data *scn_data; | |
ad23b723 AN |
4228 | Elf64_Sym *sym; |
4229 | Elf64_Rel *rel; | |
34090915 | 4230 | |
b7332d28 AN |
4231 | if (sec_idx >= obj->efile.sec_cnt) |
4232 | return -EINVAL; | |
4233 | ||
6245947c AN |
4234 | scn = elf_sec_by_idx(obj, sec_idx); |
4235 | scn_data = elf_sec_data(obj, scn); | |
4236 | ||
9c0f8cbd | 4237 | relo_sec_name = elf_sec_str(obj, shdr->sh_name); |
6245947c | 4238 | sec_name = elf_sec_name(obj, scn); |
9c0f8cbd AN |
4239 | if (!relo_sec_name || !sec_name) |
4240 | return -EINVAL; | |
4241 | ||
4242 | pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n", | |
4243 | relo_sec_name, sec_idx, sec_name); | |
34090915 WN |
4244 | nrels = shdr->sh_size / shdr->sh_entsize; |
4245 | ||
34090915 | 4246 | for (i = 0; i < nrels; i++) { |
ad23b723 AN |
4247 | rel = elf_rel_by_idx(data, i); |
4248 | if (!rel) { | |
9c0f8cbd | 4249 | pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i); |
6371ca3b | 4250 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4251 | } |
ad23b723 | 4252 | |
b7332d28 AN |
4253 | sym_idx = ELF64_R_SYM(rel->r_info); |
4254 | sym = elf_sym_by_idx(obj, sym_idx); | |
ad23b723 | 4255 | if (!sym) { |
b7332d28 AN |
4256 | pr_warn("sec '%s': symbol #%zu not found for relo #%d\n", |
4257 | relo_sec_name, sym_idx, i); | |
4258 | return -LIBBPF_ERRNO__FORMAT; | |
4259 | } | |
4260 | ||
4261 | if (sym->st_shndx >= obj->efile.sec_cnt) { | |
4262 | pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n", | |
4263 | relo_sec_name, sym_idx, (size_t)sym->st_shndx, i); | |
6371ca3b | 4264 | return -LIBBPF_ERRNO__FORMAT; |
34090915 | 4265 | } |
6245947c | 4266 | |
ad23b723 | 4267 | if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) { |
9c0f8cbd | 4268 | pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n", |
b7332d28 | 4269 | relo_sec_name, (size_t)rel->r_offset, i); |
1f8e2bcb | 4270 | return -LIBBPF_ERRNO__FORMAT; |
9c0f8cbd | 4271 | } |
d859900c | 4272 | |
ad23b723 | 4273 | insn_idx = rel->r_offset / BPF_INSN_SZ; |
c3c55696 AN |
4274 | /* relocations against static functions are recorded as |
4275 | * relocations against the section that contains a function; | |
4276 | * in such case, symbol will be STT_SECTION and sym.st_name | |
4277 | * will point to empty string (0), so fetch section name | |
4278 | * instead | |
4279 | */ | |
ad23b723 AN |
4280 | if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0) |
4281 | sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx)); | |
c3c55696 | 4282 | else |
ad23b723 | 4283 | sym_name = elf_sym_str(obj, sym->st_name); |
c3c55696 | 4284 | sym_name = sym_name ?: "<?"; |
d859900c | 4285 | |
9c0f8cbd AN |
4286 | pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n", |
4287 | relo_sec_name, i, insn_idx, sym_name); | |
666810e8 | 4288 | |
c3c55696 AN |
4289 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); |
4290 | if (!prog) { | |
6245947c | 4291 | pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n", |
c3c55696 | 4292 | relo_sec_name, i, sec_name, insn_idx); |
6245947c | 4293 | continue; |
c3c55696 AN |
4294 | } |
4295 | ||
4296 | relos = libbpf_reallocarray(prog->reloc_desc, | |
4297 | prog->nr_reloc + 1, sizeof(*relos)); | |
4298 | if (!relos) | |
4299 | return -ENOMEM; | |
4300 | prog->reloc_desc = relos; | |
4301 | ||
4302 | /* adjust insn_idx to local BPF program frame of reference */ | |
4303 | insn_idx -= prog->sec_insn_off; | |
4304 | err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc], | |
ad23b723 | 4305 | insn_idx, sym_name, sym, rel); |
1f8e2bcb AN |
4306 | if (err) |
4307 | return err; | |
c3c55696 AN |
4308 | |
4309 | prog->nr_reloc++; | |
34090915 WN |
4310 | } |
4311 | return 0; | |
4312 | } | |
4313 | ||
4fcac46c | 4314 | static int map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map) |
8a138aed | 4315 | { |
aaf6886d | 4316 | int id; |
8a138aed | 4317 | |
a8fee962 AN |
4318 | if (!obj->btf) |
4319 | return -ENOENT; | |
4320 | ||
590a0088 MKL |
4321 | /* if it's BTF-defined map, we don't need to search for type IDs. |
4322 | * For struct_ops map, it does not need btf_key_type_id and | |
4323 | * btf_value_type_id. | |
4324 | */ | |
aaf6886d | 4325 | if (map->sec_idx == obj->efile.btf_maps_shndx || bpf_map__is_struct_ops(map)) |
abd29c93 AN |
4326 | return 0; |
4327 | ||
aaf6886d AN |
4328 | /* |
4329 | * LLVM annotates global data differently in BTF, that is, | |
4330 | * only as '.data', '.bss' or '.rodata'. | |
4331 | */ | |
4332 | if (!bpf_map__is_internal(map)) | |
4333 | return -ENOENT; | |
4334 | ||
4335 | id = btf__find_by_name(obj->btf, map->real_name); | |
4336 | if (id < 0) | |
4337 | return id; | |
8a138aed | 4338 | |
aaf6886d AN |
4339 | map->btf_key_type_id = 0; |
4340 | map->btf_value_type_id = id; | |
8a138aed MKL |
4341 | return 0; |
4342 | } | |
4343 | ||
97eb3138 MP |
4344 | static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info) |
4345 | { | |
4346 | char file[PATH_MAX], buff[4096]; | |
4347 | FILE *fp; | |
4348 | __u32 val; | |
4349 | int err; | |
4350 | ||
4351 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
4352 | memset(info, 0, sizeof(*info)); | |
4353 | ||
59842c54 | 4354 | fp = fopen(file, "re"); |
97eb3138 MP |
4355 | if (!fp) { |
4356 | err = -errno; | |
4357 | pr_warn("failed to open %s: %d. No procfs support?\n", file, | |
4358 | err); | |
4359 | return err; | |
4360 | } | |
4361 | ||
4362 | while (fgets(buff, sizeof(buff), fp)) { | |
4363 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
4364 | info->type = val; | |
4365 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) | |
4366 | info->key_size = val; | |
4367 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) | |
4368 | info->value_size = val; | |
4369 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) | |
4370 | info->max_entries = val; | |
4371 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) | |
4372 | info->map_flags = val; | |
4373 | } | |
4374 | ||
4375 | fclose(fp); | |
4376 | ||
4377 | return 0; | |
4378 | } | |
4379 | ||
ec41817b AN |
4380 | bool bpf_map__autocreate(const struct bpf_map *map) |
4381 | { | |
4382 | return map->autocreate; | |
4383 | } | |
4384 | ||
4385 | int bpf_map__set_autocreate(struct bpf_map *map, bool autocreate) | |
4386 | { | |
4387 | if (map->obj->loaded) | |
4388 | return libbpf_err(-EBUSY); | |
4389 | ||
4390 | map->autocreate = autocreate; | |
4391 | return 0; | |
4392 | } | |
4393 | ||
26736eb9 JK |
4394 | int bpf_map__reuse_fd(struct bpf_map *map, int fd) |
4395 | { | |
813847a3 | 4396 | struct bpf_map_info info; |
bf3f0037 | 4397 | __u32 len = sizeof(info), name_len; |
26736eb9 JK |
4398 | int new_fd, err; |
4399 | char *new_name; | |
4400 | ||
813847a3 | 4401 | memset(&info, 0, len); |
629dfc66 | 4402 | err = bpf_map_get_info_by_fd(fd, &info, &len); |
97eb3138 MP |
4403 | if (err && errno == EINVAL) |
4404 | err = bpf_get_map_info_from_fdinfo(fd, &info); | |
26736eb9 | 4405 | if (err) |
e9fc3ce9 | 4406 | return libbpf_err(err); |
26736eb9 | 4407 | |
bf3f0037 AW |
4408 | name_len = strlen(info.name); |
4409 | if (name_len == BPF_OBJ_NAME_LEN - 1 && strncmp(map->name, info.name, name_len) == 0) | |
4410 | new_name = strdup(map->name); | |
4411 | else | |
4412 | new_name = strdup(info.name); | |
4413 | ||
26736eb9 | 4414 | if (!new_name) |
e9fc3ce9 | 4415 | return libbpf_err(-errno); |
26736eb9 JK |
4416 | |
4417 | new_fd = open("/", O_RDONLY | O_CLOEXEC); | |
d1b4574a THJ |
4418 | if (new_fd < 0) { |
4419 | err = -errno; | |
26736eb9 | 4420 | goto err_free_new_name; |
d1b4574a | 4421 | } |
26736eb9 JK |
4422 | |
4423 | new_fd = dup3(fd, new_fd, O_CLOEXEC); | |
d1b4574a THJ |
4424 | if (new_fd < 0) { |
4425 | err = -errno; | |
26736eb9 | 4426 | goto err_close_new_fd; |
d1b4574a | 4427 | } |
26736eb9 JK |
4428 | |
4429 | err = zclose(map->fd); | |
d1b4574a THJ |
4430 | if (err) { |
4431 | err = -errno; | |
26736eb9 | 4432 | goto err_close_new_fd; |
d1b4574a | 4433 | } |
26736eb9 JK |
4434 | free(map->name); |
4435 | ||
4436 | map->fd = new_fd; | |
4437 | map->name = new_name; | |
4438 | map->def.type = info.type; | |
4439 | map->def.key_size = info.key_size; | |
4440 | map->def.value_size = info.value_size; | |
4441 | map->def.max_entries = info.max_entries; | |
4442 | map->def.map_flags = info.map_flags; | |
4443 | map->btf_key_type_id = info.btf_key_type_id; | |
4444 | map->btf_value_type_id = info.btf_value_type_id; | |
ec6d5f47 | 4445 | map->reused = true; |
47512102 | 4446 | map->map_extra = info.map_extra; |
26736eb9 JK |
4447 | |
4448 | return 0; | |
4449 | ||
4450 | err_close_new_fd: | |
4451 | close(new_fd); | |
4452 | err_free_new_name: | |
4453 | free(new_name); | |
e9fc3ce9 | 4454 | return libbpf_err(err); |
26736eb9 JK |
4455 | } |
4456 | ||
1bdb6c9a | 4457 | __u32 bpf_map__max_entries(const struct bpf_map *map) |
1a11a4c7 | 4458 | { |
1bdb6c9a AN |
4459 | return map->def.max_entries; |
4460 | } | |
1a11a4c7 | 4461 | |
b3278099 AN |
4462 | struct bpf_map *bpf_map__inner_map(struct bpf_map *map) |
4463 | { | |
4464 | if (!bpf_map_type__is_map_in_map(map->def.type)) | |
e9fc3ce9 | 4465 | return errno = EINVAL, NULL; |
b3278099 AN |
4466 | |
4467 | return map->inner_map; | |
4468 | } | |
4469 | ||
1bdb6c9a AN |
4470 | int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries) |
4471 | { | |
597fbc46 | 4472 | if (map->obj->loaded) |
e9fc3ce9 | 4473 | return libbpf_err(-EBUSY); |
597fbc46 | 4474 | |
1a11a4c7 | 4475 | map->def.max_entries = max_entries; |
597fbc46 AN |
4476 | |
4477 | /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */ | |
b66ccae0 | 4478 | if (map_is_ringbuf(map)) |
597fbc46 AN |
4479 | map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries); |
4480 | ||
1a11a4c7 AI |
4481 | return 0; |
4482 | } | |
4483 | ||
47eff617 | 4484 | static int |
fd9eef1a | 4485 | bpf_object__probe_loading(struct bpf_object *obj) |
47eff617 | 4486 | { |
47eff617 SF |
4487 | char *cp, errmsg[STRERR_BUFSIZE]; |
4488 | struct bpf_insn insns[] = { | |
4489 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4490 | BPF_EXIT_INSN(), | |
4491 | }; | |
e32660ac | 4492 | int ret, insn_cnt = ARRAY_SIZE(insns); |
47eff617 | 4493 | |
f9bceaa5 SF |
4494 | if (obj->gen_loader) |
4495 | return 0; | |
4496 | ||
e542f2c4 AN |
4497 | ret = bump_rlimit_memlock(); |
4498 | if (ret) | |
4499 | pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret); | |
4500 | ||
47eff617 | 4501 | /* make sure basic loading works */ |
e32660ac AN |
4502 | ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL); |
4503 | if (ret < 0) | |
4504 | ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL); | |
47eff617 | 4505 | if (ret < 0) { |
fd9eef1a EC |
4506 | ret = errno; |
4507 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4508 | pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF " | |
4509 | "program. Make sure your kernel supports BPF " | |
4510 | "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is " | |
4511 | "set to big enough value.\n", __func__, cp, ret); | |
4512 | return -ret; | |
47eff617 SF |
4513 | } |
4514 | close(ret); | |
4515 | ||
fd9eef1a EC |
4516 | return 0; |
4517 | } | |
4518 | ||
bb180fb2 AN |
4519 | static int probe_fd(int fd) |
4520 | { | |
4521 | if (fd >= 0) | |
4522 | close(fd); | |
4523 | return fd >= 0; | |
4524 | } | |
4525 | ||
47b6cb4d | 4526 | static int probe_kern_prog_name(void) |
fd9eef1a | 4527 | { |
1f235777 | 4528 | const size_t attr_sz = offsetofend(union bpf_attr, prog_name); |
fd9eef1a EC |
4529 | struct bpf_insn insns[] = { |
4530 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4531 | BPF_EXIT_INSN(), | |
4532 | }; | |
1f235777 HL |
4533 | union bpf_attr attr; |
4534 | int ret; | |
4535 | ||
4536 | memset(&attr, 0, attr_sz); | |
4537 | attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; | |
4538 | attr.license = ptr_to_u64("GPL"); | |
4539 | attr.insns = ptr_to_u64(insns); | |
4540 | attr.insn_cnt = (__u32)ARRAY_SIZE(insns); | |
4541 | libbpf_strlcpy(attr.prog_name, "libbpf_nametest", sizeof(attr.prog_name)); | |
fd9eef1a EC |
4542 | |
4543 | /* make sure loading with name works */ | |
1f235777 | 4544 | ret = sys_bpf_prog_load(&attr, attr_sz, PROG_LOAD_ATTEMPTS); |
bb180fb2 | 4545 | return probe_fd(ret); |
47eff617 SF |
4546 | } |
4547 | ||
47b6cb4d | 4548 | static int probe_kern_global_data(void) |
8837fe5d | 4549 | { |
8837fe5d DB |
4550 | char *cp, errmsg[STRERR_BUFSIZE]; |
4551 | struct bpf_insn insns[] = { | |
4552 | BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16), | |
4553 | BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42), | |
4554 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4555 | BPF_EXIT_INSN(), | |
4556 | }; | |
e32660ac | 4557 | int ret, map, insn_cnt = ARRAY_SIZE(insns); |
8837fe5d | 4558 | |
10b62d6a | 4559 | map = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_global", sizeof(int), 32, 1, NULL); |
8837fe5d | 4560 | if (map < 0) { |
23ab656b THJ |
4561 | ret = -errno; |
4562 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
be18010e | 4563 | pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n", |
23ab656b THJ |
4564 | __func__, cp, -ret); |
4565 | return ret; | |
8837fe5d DB |
4566 | } |
4567 | ||
4568 | insns[0].imm = map; | |
4569 | ||
e32660ac | 4570 | ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL); |
47b6cb4d | 4571 | close(map); |
bb180fb2 | 4572 | return probe_fd(ret); |
8837fe5d DB |
4573 | } |
4574 | ||
68b08647 AN |
4575 | static int probe_kern_btf(void) |
4576 | { | |
4577 | static const char strs[] = "\0int"; | |
4578 | __u32 types[] = { | |
4579 | /* int */ | |
4580 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), | |
4581 | }; | |
4582 | ||
4583 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4584 | strs, sizeof(strs))); | |
4585 | } | |
4586 | ||
47b6cb4d | 4587 | static int probe_kern_btf_func(void) |
d7c4b398 | 4588 | { |
8983b731 | 4589 | static const char strs[] = "\0int\0x\0a"; |
d7c4b398 AN |
4590 | /* void x(int a) {} */ |
4591 | __u32 types[] = { | |
4592 | /* int */ | |
4593 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4594 | /* FUNC_PROTO */ /* [2] */ | |
4595 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0), | |
4596 | BTF_PARAM_ENC(7, 1), | |
4597 | /* FUNC x */ /* [3] */ | |
4598 | BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2), | |
4599 | }; | |
d7c4b398 | 4600 | |
bb180fb2 AN |
4601 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4602 | strs, sizeof(strs))); | |
d7c4b398 AN |
4603 | } |
4604 | ||
47b6cb4d | 4605 | static int probe_kern_btf_func_global(void) |
2d3eb67f AS |
4606 | { |
4607 | static const char strs[] = "\0int\0x\0a"; | |
4608 | /* static void x(int a) {} */ | |
4609 | __u32 types[] = { | |
4610 | /* int */ | |
4611 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4612 | /* FUNC_PROTO */ /* [2] */ | |
4613 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0), | |
4614 | BTF_PARAM_ENC(7, 1), | |
4615 | /* FUNC x BTF_FUNC_GLOBAL */ /* [3] */ | |
4616 | BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2), | |
4617 | }; | |
2d3eb67f | 4618 | |
bb180fb2 AN |
4619 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4620 | strs, sizeof(strs))); | |
2d3eb67f AS |
4621 | } |
4622 | ||
47b6cb4d | 4623 | static int probe_kern_btf_datasec(void) |
d7c4b398 | 4624 | { |
8983b731 | 4625 | static const char strs[] = "\0x\0.data"; |
d7c4b398 AN |
4626 | /* static int a; */ |
4627 | __u32 types[] = { | |
4628 | /* int */ | |
4629 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4630 | /* VAR x */ /* [2] */ | |
4631 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1), | |
4632 | BTF_VAR_STATIC, | |
4633 | /* DATASEC val */ /* [3] */ | |
4634 | BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4), | |
4635 | BTF_VAR_SECINFO_ENC(2, 0, 4), | |
4636 | }; | |
cfd49210 | 4637 | |
bb180fb2 AN |
4638 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), |
4639 | strs, sizeof(strs))); | |
d7c4b398 AN |
4640 | } |
4641 | ||
22541a9e IL |
4642 | static int probe_kern_btf_float(void) |
4643 | { | |
4644 | static const char strs[] = "\0float"; | |
4645 | __u32 types[] = { | |
4646 | /* float */ | |
4647 | BTF_TYPE_FLOAT_ENC(1, 4), | |
4648 | }; | |
4649 | ||
4650 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4651 | strs, sizeof(strs))); | |
4652 | } | |
4653 | ||
223f903e | 4654 | static int probe_kern_btf_decl_tag(void) |
5b84bd10 YS |
4655 | { |
4656 | static const char strs[] = "\0tag"; | |
4657 | __u32 types[] = { | |
4658 | /* int */ | |
4659 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4660 | /* VAR x */ /* [2] */ | |
4661 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1), | |
4662 | BTF_VAR_STATIC, | |
4663 | /* attr */ | |
223f903e | 4664 | BTF_TYPE_DECL_TAG_ENC(1, 2, -1), |
5b84bd10 YS |
4665 | }; |
4666 | ||
4667 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4668 | strs, sizeof(strs))); | |
4669 | } | |
4670 | ||
2dc1e488 YS |
4671 | static int probe_kern_btf_type_tag(void) |
4672 | { | |
4673 | static const char strs[] = "\0tag"; | |
4674 | __u32 types[] = { | |
4675 | /* int */ | |
4676 | BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */ | |
4677 | /* attr */ | |
4678 | BTF_TYPE_TYPE_TAG_ENC(1, 1), /* [2] */ | |
4679 | /* ptr */ | |
4680 | BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_PTR, 0, 0), 2), /* [3] */ | |
4681 | }; | |
4682 | ||
4683 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4684 | strs, sizeof(strs))); | |
4685 | } | |
4686 | ||
47b6cb4d | 4687 | static int probe_kern_array_mmap(void) |
7fe74b43 | 4688 | { |
a9606f40 AN |
4689 | LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); |
4690 | int fd; | |
7fe74b43 | 4691 | |
10b62d6a | 4692 | fd = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_mmap", sizeof(int), sizeof(int), 1, &opts); |
a9606f40 | 4693 | return probe_fd(fd); |
7fe74b43 AN |
4694 | } |
4695 | ||
47b6cb4d | 4696 | static int probe_kern_exp_attach_type(void) |
25498a19 | 4697 | { |
e32660ac | 4698 | LIBBPF_OPTS(bpf_prog_load_opts, opts, .expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE); |
25498a19 AN |
4699 | struct bpf_insn insns[] = { |
4700 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4701 | BPF_EXIT_INSN(), | |
4702 | }; | |
e32660ac | 4703 | int fd, insn_cnt = ARRAY_SIZE(insns); |
25498a19 | 4704 | |
25498a19 AN |
4705 | /* use any valid combination of program type and (optional) |
4706 | * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS) | |
4707 | * to see if kernel supports expected_attach_type field for | |
4708 | * BPF_PROG_LOAD command | |
4709 | */ | |
e32660ac AN |
4710 | fd = bpf_prog_load(BPF_PROG_TYPE_CGROUP_SOCK, NULL, "GPL", insns, insn_cnt, &opts); |
4711 | return probe_fd(fd); | |
25498a19 AN |
4712 | } |
4713 | ||
109cea5a AN |
4714 | static int probe_kern_probe_read_kernel(void) |
4715 | { | |
109cea5a AN |
4716 | struct bpf_insn insns[] = { |
4717 | BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), /* r1 = r10 (fp) */ | |
4718 | BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8), /* r1 += -8 */ | |
4719 | BPF_MOV64_IMM(BPF_REG_2, 8), /* r2 = 8 */ | |
4720 | BPF_MOV64_IMM(BPF_REG_3, 0), /* r3 = 0 */ | |
4721 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel), | |
4722 | BPF_EXIT_INSN(), | |
4723 | }; | |
e32660ac | 4724 | int fd, insn_cnt = ARRAY_SIZE(insns); |
109cea5a | 4725 | |
d252a4a4 | 4726 | fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL); |
e32660ac | 4727 | return probe_fd(fd); |
109cea5a AN |
4728 | } |
4729 | ||
5d23328d YZ |
4730 | static int probe_prog_bind_map(void) |
4731 | { | |
5d23328d YZ |
4732 | char *cp, errmsg[STRERR_BUFSIZE]; |
4733 | struct bpf_insn insns[] = { | |
4734 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4735 | BPF_EXIT_INSN(), | |
4736 | }; | |
e32660ac | 4737 | int ret, map, prog, insn_cnt = ARRAY_SIZE(insns); |
5d23328d | 4738 | |
10b62d6a | 4739 | map = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_det_bind", sizeof(int), 32, 1, NULL); |
5d23328d YZ |
4740 | if (map < 0) { |
4741 | ret = -errno; | |
4742 | cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg)); | |
4743 | pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n", | |
4744 | __func__, cp, -ret); | |
4745 | return ret; | |
4746 | } | |
4747 | ||
e32660ac | 4748 | prog = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL); |
5d23328d YZ |
4749 | if (prog < 0) { |
4750 | close(map); | |
4751 | return 0; | |
4752 | } | |
4753 | ||
4754 | ret = bpf_prog_bind_map(prog, map, NULL); | |
4755 | ||
4756 | close(map); | |
4757 | close(prog); | |
4758 | ||
4759 | return ret >= 0; | |
4760 | } | |
4761 | ||
4f33a53d AN |
4762 | static int probe_module_btf(void) |
4763 | { | |
4764 | static const char strs[] = "\0int"; | |
4765 | __u32 types[] = { | |
4766 | /* int */ | |
4767 | BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), | |
4768 | }; | |
4769 | struct bpf_btf_info info; | |
4770 | __u32 len = sizeof(info); | |
4771 | char name[16]; | |
4772 | int fd, err; | |
4773 | ||
4774 | fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs)); | |
4775 | if (fd < 0) | |
4776 | return 0; /* BTF not supported at all */ | |
4777 | ||
4778 | memset(&info, 0, sizeof(info)); | |
4779 | info.name = ptr_to_u64(name); | |
4780 | info.name_len = sizeof(name); | |
4781 | ||
4782 | /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer; | |
4783 | * kernel's module BTF support coincides with support for | |
4784 | * name/name_len fields in struct bpf_btf_info. | |
4785 | */ | |
629dfc66 | 4786 | err = bpf_btf_get_info_by_fd(fd, &info, &len); |
4f33a53d AN |
4787 | close(fd); |
4788 | return !err; | |
4789 | } | |
4790 | ||
668ace0e AN |
4791 | static int probe_perf_link(void) |
4792 | { | |
668ace0e AN |
4793 | struct bpf_insn insns[] = { |
4794 | BPF_MOV64_IMM(BPF_REG_0, 0), | |
4795 | BPF_EXIT_INSN(), | |
4796 | }; | |
4797 | int prog_fd, link_fd, err; | |
4798 | ||
e32660ac AN |
4799 | prog_fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", |
4800 | insns, ARRAY_SIZE(insns), NULL); | |
668ace0e AN |
4801 | if (prog_fd < 0) |
4802 | return -errno; | |
4803 | ||
4804 | /* use invalid perf_event FD to get EBADF, if link is supported; | |
4805 | * otherwise EINVAL should be returned | |
4806 | */ | |
4807 | link_fd = bpf_link_create(prog_fd, -1, BPF_PERF_EVENT, NULL); | |
4808 | err = -errno; /* close() can clobber errno */ | |
4809 | ||
4810 | if (link_fd >= 0) | |
4811 | close(link_fd); | |
4812 | close(prog_fd); | |
4813 | ||
4814 | return link_fd < 0 && err == -EBADF; | |
4815 | } | |
4816 | ||
2e4913e0 AN |
4817 | static int probe_kern_bpf_cookie(void) |
4818 | { | |
4819 | struct bpf_insn insns[] = { | |
4820 | BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_get_attach_cookie), | |
4821 | BPF_EXIT_INSN(), | |
4822 | }; | |
4823 | int ret, insn_cnt = ARRAY_SIZE(insns); | |
4824 | ||
4825 | ret = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL, "GPL", insns, insn_cnt, NULL); | |
4826 | return probe_fd(ret); | |
4827 | } | |
4828 | ||
f2a62588 YS |
4829 | static int probe_kern_btf_enum64(void) |
4830 | { | |
4831 | static const char strs[] = "\0enum64"; | |
4832 | __u32 types[] = { | |
4833 | BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_ENUM64, 0, 0), 8), | |
4834 | }; | |
4835 | ||
4836 | return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types), | |
4837 | strs, sizeof(strs))); | |
4838 | } | |
4839 | ||
708ac5be AN |
4840 | static int probe_kern_syscall_wrapper(void); |
4841 | ||
47b6cb4d AN |
4842 | enum kern_feature_result { |
4843 | FEAT_UNKNOWN = 0, | |
4844 | FEAT_SUPPORTED = 1, | |
4845 | FEAT_MISSING = 2, | |
4846 | }; | |
4847 | ||
4848 | typedef int (*feature_probe_fn)(void); | |
4849 | ||
4850 | static struct kern_feature_desc { | |
4851 | const char *desc; | |
4852 | feature_probe_fn probe; | |
4853 | enum kern_feature_result res; | |
4854 | } feature_probes[__FEAT_CNT] = { | |
4855 | [FEAT_PROG_NAME] = { | |
4856 | "BPF program name", probe_kern_prog_name, | |
4857 | }, | |
4858 | [FEAT_GLOBAL_DATA] = { | |
4859 | "global variables", probe_kern_global_data, | |
4860 | }, | |
68b08647 AN |
4861 | [FEAT_BTF] = { |
4862 | "minimal BTF", probe_kern_btf, | |
4863 | }, | |
47b6cb4d AN |
4864 | [FEAT_BTF_FUNC] = { |
4865 | "BTF functions", probe_kern_btf_func, | |
4866 | }, | |
4867 | [FEAT_BTF_GLOBAL_FUNC] = { | |
4868 | "BTF global function", probe_kern_btf_func_global, | |
4869 | }, | |
4870 | [FEAT_BTF_DATASEC] = { | |
4871 | "BTF data section and variable", probe_kern_btf_datasec, | |
4872 | }, | |
4873 | [FEAT_ARRAY_MMAP] = { | |
4874 | "ARRAY map mmap()", probe_kern_array_mmap, | |
4875 | }, | |
4876 | [FEAT_EXP_ATTACH_TYPE] = { | |
4877 | "BPF_PROG_LOAD expected_attach_type attribute", | |
4878 | probe_kern_exp_attach_type, | |
4879 | }, | |
109cea5a AN |
4880 | [FEAT_PROBE_READ_KERN] = { |
4881 | "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel, | |
5d23328d YZ |
4882 | }, |
4883 | [FEAT_PROG_BIND_MAP] = { | |
4884 | "BPF_PROG_BIND_MAP support", probe_prog_bind_map, | |
4f33a53d AN |
4885 | }, |
4886 | [FEAT_MODULE_BTF] = { | |
4887 | "module BTF support", probe_module_btf, | |
4888 | }, | |
22541a9e IL |
4889 | [FEAT_BTF_FLOAT] = { |
4890 | "BTF_KIND_FLOAT support", probe_kern_btf_float, | |
4891 | }, | |
668ace0e AN |
4892 | [FEAT_PERF_LINK] = { |
4893 | "BPF perf link support", probe_perf_link, | |
4894 | }, | |
223f903e YS |
4895 | [FEAT_BTF_DECL_TAG] = { |
4896 | "BTF_KIND_DECL_TAG support", probe_kern_btf_decl_tag, | |
5b84bd10 | 4897 | }, |
2dc1e488 YS |
4898 | [FEAT_BTF_TYPE_TAG] = { |
4899 | "BTF_KIND_TYPE_TAG support", probe_kern_btf_type_tag, | |
4900 | }, | |
e542f2c4 AN |
4901 | [FEAT_MEMCG_ACCOUNT] = { |
4902 | "memcg-based memory accounting", probe_memcg_account, | |
4903 | }, | |
2e4913e0 AN |
4904 | [FEAT_BPF_COOKIE] = { |
4905 | "BPF cookie support", probe_kern_bpf_cookie, | |
4906 | }, | |
f2a62588 YS |
4907 | [FEAT_BTF_ENUM64] = { |
4908 | "BTF_KIND_ENUM64 support", probe_kern_btf_enum64, | |
4909 | }, | |
708ac5be AN |
4910 | [FEAT_SYSCALL_WRAPPER] = { |
4911 | "Kernel using syscall wrapper", probe_kern_syscall_wrapper, | |
4912 | }, | |
47b6cb4d | 4913 | }; |
8837fe5d | 4914 | |
e542f2c4 | 4915 | bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id) |
47b6cb4d AN |
4916 | { |
4917 | struct kern_feature_desc *feat = &feature_probes[feat_id]; | |
4918 | int ret; | |
4919 | ||
e542f2c4 | 4920 | if (obj && obj->gen_loader) |
67234743 AS |
4921 | /* To generate loader program assume the latest kernel |
4922 | * to avoid doing extra prog_load, map_create syscalls. | |
4923 | */ | |
4924 | return true; | |
4925 | ||
47b6cb4d AN |
4926 | if (READ_ONCE(feat->res) == FEAT_UNKNOWN) { |
4927 | ret = feat->probe(); | |
4928 | if (ret > 0) { | |
4929 | WRITE_ONCE(feat->res, FEAT_SUPPORTED); | |
4930 | } else if (ret == 0) { | |
4931 | WRITE_ONCE(feat->res, FEAT_MISSING); | |
4932 | } else { | |
4933 | pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret); | |
4934 | WRITE_ONCE(feat->res, FEAT_MISSING); | |
4935 | } | |
8837fe5d DB |
4936 | } |
4937 | ||
47b6cb4d | 4938 | return READ_ONCE(feat->res) == FEAT_SUPPORTED; |
47eff617 SF |
4939 | } |
4940 | ||
57a00f41 THJ |
4941 | static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd) |
4942 | { | |
813847a3 | 4943 | struct bpf_map_info map_info; |
57a00f41 | 4944 | char msg[STRERR_BUFSIZE]; |
813847a3 | 4945 | __u32 map_info_len = sizeof(map_info); |
97eb3138 | 4946 | int err; |
57a00f41 | 4947 | |
813847a3 | 4948 | memset(&map_info, 0, map_info_len); |
629dfc66 | 4949 | err = bpf_map_get_info_by_fd(map_fd, &map_info, &map_info_len); |
97eb3138 MP |
4950 | if (err && errno == EINVAL) |
4951 | err = bpf_get_map_info_from_fdinfo(map_fd, &map_info); | |
4952 | if (err) { | |
4953 | pr_warn("failed to get map info for map FD %d: %s\n", map_fd, | |
4954 | libbpf_strerror_r(errno, msg, sizeof(msg))); | |
57a00f41 THJ |
4955 | return false; |
4956 | } | |
4957 | ||
4958 | return (map_info.type == map->def.type && | |
4959 | map_info.key_size == map->def.key_size && | |
4960 | map_info.value_size == map->def.value_size && | |
4961 | map_info.max_entries == map->def.max_entries && | |
47512102 JK |
4962 | map_info.map_flags == map->def.map_flags && |
4963 | map_info.map_extra == map->map_extra); | |
57a00f41 THJ |
4964 | } |
4965 | ||
4966 | static int | |
4967 | bpf_object__reuse_map(struct bpf_map *map) | |
4968 | { | |
4969 | char *cp, errmsg[STRERR_BUFSIZE]; | |
4970 | int err, pin_fd; | |
4971 | ||
4972 | pin_fd = bpf_obj_get(map->pin_path); | |
4973 | if (pin_fd < 0) { | |
4974 | err = -errno; | |
4975 | if (err == -ENOENT) { | |
4976 | pr_debug("found no pinned map to reuse at '%s'\n", | |
4977 | map->pin_path); | |
4978 | return 0; | |
4979 | } | |
4980 | ||
4981 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
4982 | pr_warn("couldn't retrieve pinned map '%s': %s\n", | |
4983 | map->pin_path, cp); | |
4984 | return err; | |
4985 | } | |
4986 | ||
4987 | if (!map_is_reuse_compat(map, pin_fd)) { | |
4988 | pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n", | |
4989 | map->pin_path); | |
4990 | close(pin_fd); | |
4991 | return -EINVAL; | |
4992 | } | |
4993 | ||
4994 | err = bpf_map__reuse_fd(map, pin_fd); | |
d0f325c3 | 4995 | close(pin_fd); |
e3ba8e4e | 4996 | if (err) |
57a00f41 | 4997 | return err; |
e3ba8e4e | 4998 | |
57a00f41 THJ |
4999 | map->pinned = true; |
5000 | pr_debug("reused pinned map at '%s'\n", map->pin_path); | |
5001 | ||
5002 | return 0; | |
5003 | } | |
5004 | ||
d859900c DB |
5005 | static int |
5006 | bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map) | |
5007 | { | |
166750bc | 5008 | enum libbpf_map_type map_type = map->libbpf_type; |
d859900c DB |
5009 | char *cp, errmsg[STRERR_BUFSIZE]; |
5010 | int err, zero = 0; | |
d859900c | 5011 | |
67234743 AS |
5012 | if (obj->gen_loader) { |
5013 | bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps, | |
5014 | map->mmaped, map->def.value_size); | |
5015 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) | |
5016 | bpf_gen__map_freeze(obj->gen_loader, map - obj->maps); | |
5017 | return 0; | |
5018 | } | |
eba9c5f4 AN |
5019 | err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0); |
5020 | if (err) { | |
5021 | err = -errno; | |
5022 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
5023 | pr_warn("Error setting initial map(%s) contents: %s\n", | |
5024 | map->name, cp); | |
5025 | return err; | |
5026 | } | |
d859900c | 5027 | |
81bfdd08 AN |
5028 | /* Freeze .rodata and .kconfig map as read-only from syscall side. */ |
5029 | if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) { | |
d859900c DB |
5030 | err = bpf_map_freeze(map->fd); |
5031 | if (err) { | |
eba9c5f4 AN |
5032 | err = -errno; |
5033 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
be18010e KW |
5034 | pr_warn("Error freezing map(%s) as read-only: %s\n", |
5035 | map->name, cp); | |
eba9c5f4 | 5036 | return err; |
d859900c DB |
5037 | } |
5038 | } | |
eba9c5f4 | 5039 | return 0; |
d859900c DB |
5040 | } |
5041 | ||
2d39d7c5 AN |
5042 | static void bpf_map__destroy(struct bpf_map *map); |
5043 | ||
67234743 | 5044 | static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner) |
2d39d7c5 | 5045 | { |
992c4225 | 5046 | LIBBPF_OPTS(bpf_map_create_opts, create_attr); |
2d39d7c5 | 5047 | struct bpf_map_def *def = &map->def; |
992c4225 | 5048 | const char *map_name = NULL; |
a21ab4c5 | 5049 | int err = 0; |
2d39d7c5 | 5050 | |
9ca1f56a | 5051 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
992c4225 | 5052 | map_name = map->name; |
2d39d7c5 | 5053 | create_attr.map_ifindex = map->map_ifindex; |
2d39d7c5 | 5054 | create_attr.map_flags = def->map_flags; |
1bdb6c9a | 5055 | create_attr.numa_node = map->numa_node; |
47512102 | 5056 | create_attr.map_extra = map->map_extra; |
2d39d7c5 | 5057 | |
2d39d7c5 | 5058 | if (bpf_map__is_struct_ops(map)) |
992c4225 | 5059 | create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id; |
2d39d7c5 | 5060 | |
262cfb74 | 5061 | if (obj->btf && btf__fd(obj->btf) >= 0) { |
2d39d7c5 AN |
5062 | create_attr.btf_fd = btf__fd(obj->btf); |
5063 | create_attr.btf_key_type_id = map->btf_key_type_id; | |
5064 | create_attr.btf_value_type_id = map->btf_value_type_id; | |
5065 | } | |
5066 | ||
646f02ff AN |
5067 | if (bpf_map_type__is_map_in_map(def->type)) { |
5068 | if (map->inner_map) { | |
67234743 | 5069 | err = bpf_object__create_map(obj, map->inner_map, true); |
646f02ff AN |
5070 | if (err) { |
5071 | pr_warn("map '%s': failed to create inner map: %d\n", | |
5072 | map->name, err); | |
5073 | return err; | |
5074 | } | |
5075 | map->inner_map_fd = bpf_map__fd(map->inner_map); | |
5076 | } | |
5077 | if (map->inner_map_fd >= 0) | |
5078 | create_attr.inner_map_fd = map->inner_map_fd; | |
5079 | } | |
5080 | ||
f7310523 HC |
5081 | switch (def->type) { |
5082 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
5083 | case BPF_MAP_TYPE_CGROUP_ARRAY: | |
5084 | case BPF_MAP_TYPE_STACK_TRACE: | |
5085 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
5086 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
5087 | case BPF_MAP_TYPE_DEVMAP: | |
5088 | case BPF_MAP_TYPE_DEVMAP_HASH: | |
5089 | case BPF_MAP_TYPE_CPUMAP: | |
5090 | case BPF_MAP_TYPE_XSKMAP: | |
5091 | case BPF_MAP_TYPE_SOCKMAP: | |
5092 | case BPF_MAP_TYPE_SOCKHASH: | |
5093 | case BPF_MAP_TYPE_QUEUE: | |
5094 | case BPF_MAP_TYPE_STACK: | |
f7310523 HC |
5095 | create_attr.btf_fd = 0; |
5096 | create_attr.btf_key_type_id = 0; | |
5097 | create_attr.btf_value_type_id = 0; | |
5098 | map->btf_key_type_id = 0; | |
5099 | map->btf_value_type_id = 0; | |
5100 | default: | |
5101 | break; | |
5102 | } | |
5103 | ||
67234743 | 5104 | if (obj->gen_loader) { |
992c4225 | 5105 | bpf_gen__map_create(obj->gen_loader, def->type, map_name, |
a4fbfdd7 | 5106 | def->key_size, def->value_size, def->max_entries, |
992c4225 | 5107 | &create_attr, is_inner ? -1 : map - obj->maps); |
67234743 AS |
5108 | /* Pretend to have valid FD to pass various fd >= 0 checks. |
5109 | * This fd == 0 will not be used with any syscall and will be reset to -1 eventually. | |
5110 | */ | |
5111 | map->fd = 0; | |
5112 | } else { | |
992c4225 AN |
5113 | map->fd = bpf_map_create(def->type, map_name, |
5114 | def->key_size, def->value_size, | |
a4fbfdd7 | 5115 | def->max_entries, &create_attr); |
67234743 | 5116 | } |
2d39d7c5 AN |
5117 | if (map->fd < 0 && (create_attr.btf_key_type_id || |
5118 | create_attr.btf_value_type_id)) { | |
5119 | char *cp, errmsg[STRERR_BUFSIZE]; | |
2d39d7c5 | 5120 | |
a21ab4c5 | 5121 | err = -errno; |
2d39d7c5 AN |
5122 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); |
5123 | pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n", | |
5124 | map->name, cp, err); | |
5125 | create_attr.btf_fd = 0; | |
5126 | create_attr.btf_key_type_id = 0; | |
5127 | create_attr.btf_value_type_id = 0; | |
5128 | map->btf_key_type_id = 0; | |
5129 | map->btf_value_type_id = 0; | |
992c4225 AN |
5130 | map->fd = bpf_map_create(def->type, map_name, |
5131 | def->key_size, def->value_size, | |
a4fbfdd7 | 5132 | def->max_entries, &create_attr); |
2d39d7c5 AN |
5133 | } |
5134 | ||
a21ab4c5 | 5135 | err = map->fd < 0 ? -errno : 0; |
2d39d7c5 | 5136 | |
646f02ff | 5137 | if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) { |
67234743 AS |
5138 | if (obj->gen_loader) |
5139 | map->inner_map->fd = -1; | |
646f02ff AN |
5140 | bpf_map__destroy(map->inner_map); |
5141 | zfree(&map->inner_map); | |
5142 | } | |
5143 | ||
a21ab4c5 | 5144 | return err; |
2d39d7c5 AN |
5145 | } |
5146 | ||
341ac5ff | 5147 | static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map) |
a0f2b7ac HL |
5148 | { |
5149 | const struct bpf_map *targ_map; | |
5150 | unsigned int i; | |
67234743 | 5151 | int fd, err = 0; |
a0f2b7ac HL |
5152 | |
5153 | for (i = 0; i < map->init_slots_sz; i++) { | |
5154 | if (!map->init_slots[i]) | |
5155 | continue; | |
5156 | ||
5157 | targ_map = map->init_slots[i]; | |
5158 | fd = bpf_map__fd(targ_map); | |
341ac5ff | 5159 | |
67234743 | 5160 | if (obj->gen_loader) { |
be05c944 AS |
5161 | bpf_gen__populate_outer_map(obj->gen_loader, |
5162 | map - obj->maps, i, | |
5163 | targ_map - obj->maps); | |
67234743 AS |
5164 | } else { |
5165 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
5166 | } | |
a0f2b7ac HL |
5167 | if (err) { |
5168 | err = -errno; | |
5169 | pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n", | |
341ac5ff | 5170 | map->name, i, targ_map->name, fd, err); |
a0f2b7ac HL |
5171 | return err; |
5172 | } | |
5173 | pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n", | |
5174 | map->name, i, targ_map->name, fd); | |
5175 | } | |
5176 | ||
5177 | zfree(&map->init_slots); | |
5178 | map->init_slots_sz = 0; | |
5179 | ||
5180 | return 0; | |
5181 | } | |
5182 | ||
341ac5ff HC |
5183 | static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map) |
5184 | { | |
5185 | const struct bpf_program *targ_prog; | |
5186 | unsigned int i; | |
5187 | int fd, err; | |
5188 | ||
5189 | if (obj->gen_loader) | |
5190 | return -ENOTSUP; | |
5191 | ||
5192 | for (i = 0; i < map->init_slots_sz; i++) { | |
5193 | if (!map->init_slots[i]) | |
5194 | continue; | |
5195 | ||
5196 | targ_prog = map->init_slots[i]; | |
5197 | fd = bpf_program__fd(targ_prog); | |
5198 | ||
5199 | err = bpf_map_update_elem(map->fd, &i, &fd, 0); | |
5200 | if (err) { | |
5201 | err = -errno; | |
5202 | pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %d\n", | |
5203 | map->name, i, targ_prog->name, fd, err); | |
5204 | return err; | |
5205 | } | |
5206 | pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n", | |
5207 | map->name, i, targ_prog->name, fd); | |
5208 | } | |
5209 | ||
5210 | zfree(&map->init_slots); | |
5211 | map->init_slots_sz = 0; | |
5212 | ||
5213 | return 0; | |
5214 | } | |
5215 | ||
5216 | static int bpf_object_init_prog_arrays(struct bpf_object *obj) | |
5217 | { | |
5218 | struct bpf_map *map; | |
5219 | int i, err; | |
5220 | ||
5221 | for (i = 0; i < obj->nr_maps; i++) { | |
5222 | map = &obj->maps[i]; | |
5223 | ||
5224 | if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY) | |
5225 | continue; | |
5226 | ||
5227 | err = init_prog_array_slots(obj, map); | |
5228 | if (err < 0) { | |
5229 | zclose(map->fd); | |
5230 | return err; | |
5231 | } | |
5232 | } | |
5233 | return 0; | |
5234 | } | |
5235 | ||
a4fbfdd7 ST |
5236 | static int map_set_def_max_entries(struct bpf_map *map) |
5237 | { | |
5238 | if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) { | |
5239 | int nr_cpus; | |
5240 | ||
5241 | nr_cpus = libbpf_num_possible_cpus(); | |
5242 | if (nr_cpus < 0) { | |
5243 | pr_warn("map '%s': failed to determine number of system CPUs: %d\n", | |
5244 | map->name, nr_cpus); | |
5245 | return nr_cpus; | |
5246 | } | |
5247 | pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus); | |
5248 | map->def.max_entries = nr_cpus; | |
5249 | } | |
5250 | ||
5251 | return 0; | |
5252 | } | |
5253 | ||
52d3352e WN |
5254 | static int |
5255 | bpf_object__create_maps(struct bpf_object *obj) | |
5256 | { | |
2d39d7c5 AN |
5257 | struct bpf_map *map; |
5258 | char *cp, errmsg[STRERR_BUFSIZE]; | |
5259 | unsigned int i, j; | |
8a138aed | 5260 | int err; |
043c5bb3 | 5261 | bool retried; |
52d3352e | 5262 | |
9d759a9b | 5263 | for (i = 0; i < obj->nr_maps; i++) { |
2d39d7c5 | 5264 | map = &obj->maps[i]; |
8a138aed | 5265 | |
16e0c35c AN |
5266 | /* To support old kernels, we skip creating global data maps |
5267 | * (.rodata, .data, .kconfig, etc); later on, during program | |
5268 | * loading, if we detect that at least one of the to-be-loaded | |
5269 | * programs is referencing any global data map, we'll error | |
5270 | * out with program name and relocation index logged. | |
5271 | * This approach allows to accommodate Clang emitting | |
5272 | * unnecessary .rodata.str1.1 sections for string literals, | |
5273 | * but also it allows to have CO-RE applications that use | |
5274 | * global variables in some of BPF programs, but not others. | |
5275 | * If those global variable-using programs are not loaded at | |
5276 | * runtime due to bpf_program__set_autoload(prog, false), | |
5277 | * bpf_object loading will succeed just fine even on old | |
5278 | * kernels. | |
5279 | */ | |
ec41817b AN |
5280 | if (bpf_map__is_internal(map) && !kernel_supports(obj, FEAT_GLOBAL_DATA)) |
5281 | map->autocreate = false; | |
5282 | ||
5283 | if (!map->autocreate) { | |
5284 | pr_debug("map '%s': skipped auto-creating...\n", map->name); | |
16e0c35c | 5285 | continue; |
229fae38 | 5286 | } |
16e0c35c | 5287 | |
a4fbfdd7 ST |
5288 | err = map_set_def_max_entries(map); |
5289 | if (err) | |
5290 | goto err_out; | |
5291 | ||
043c5bb3 MP |
5292 | retried = false; |
5293 | retry: | |
57a00f41 THJ |
5294 | if (map->pin_path) { |
5295 | err = bpf_object__reuse_map(map); | |
5296 | if (err) { | |
2d39d7c5 | 5297 | pr_warn("map '%s': error reusing pinned map\n", |
57a00f41 | 5298 | map->name); |
2d39d7c5 | 5299 | goto err_out; |
57a00f41 | 5300 | } |
043c5bb3 MP |
5301 | if (retried && map->fd < 0) { |
5302 | pr_warn("map '%s': cannot find pinned map\n", | |
5303 | map->name); | |
5304 | err = -ENOENT; | |
5305 | goto err_out; | |
5306 | } | |
57a00f41 THJ |
5307 | } |
5308 | ||
26736eb9 | 5309 | if (map->fd >= 0) { |
2d39d7c5 | 5310 | pr_debug("map '%s': skipping creation (preset fd=%d)\n", |
26736eb9 | 5311 | map->name, map->fd); |
2c193d32 | 5312 | } else { |
67234743 | 5313 | err = bpf_object__create_map(obj, map, false); |
2c193d32 | 5314 | if (err) |
d859900c | 5315 | goto err_out; |
d859900c | 5316 | |
2c193d32 HL |
5317 | pr_debug("map '%s': created successfully, fd=%d\n", |
5318 | map->name, map->fd); | |
646f02ff | 5319 | |
2c193d32 HL |
5320 | if (bpf_map__is_internal(map)) { |
5321 | err = bpf_object__populate_internal_map(obj, map); | |
5322 | if (err < 0) { | |
5323 | zclose(map->fd); | |
5324 | goto err_out; | |
5325 | } | |
d859900c | 5326 | } |
646f02ff | 5327 | |
341ac5ff HC |
5328 | if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) { |
5329 | err = init_map_in_map_slots(obj, map); | |
2c193d32 HL |
5330 | if (err < 0) { |
5331 | zclose(map->fd); | |
646f02ff AN |
5332 | goto err_out; |
5333 | } | |
646f02ff | 5334 | } |
646f02ff AN |
5335 | } |
5336 | ||
57a00f41 THJ |
5337 | if (map->pin_path && !map->pinned) { |
5338 | err = bpf_map__pin(map, NULL); | |
5339 | if (err) { | |
043c5bb3 MP |
5340 | zclose(map->fd); |
5341 | if (!retried && err == -EEXIST) { | |
5342 | retried = true; | |
5343 | goto retry; | |
5344 | } | |
2d39d7c5 AN |
5345 | pr_warn("map '%s': failed to auto-pin at '%s': %d\n", |
5346 | map->name, map->pin_path, err); | |
2d39d7c5 | 5347 | goto err_out; |
57a00f41 THJ |
5348 | } |
5349 | } | |
52d3352e WN |
5350 | } |
5351 | ||
52d3352e | 5352 | return 0; |
2d39d7c5 AN |
5353 | |
5354 | err_out: | |
5355 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
5356 | pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err); | |
5357 | pr_perm_msg(err); | |
5358 | for (j = 0; j < i; j++) | |
5359 | zclose(obj->maps[j].fd); | |
5360 | return err; | |
52d3352e WN |
5361 | } |
5362 | ||
ddc7c304 AN |
5363 | static bool bpf_core_is_flavor_sep(const char *s) |
5364 | { | |
5365 | /* check X___Y name pattern, where X and Y are not underscores */ | |
5366 | return s[0] != '_' && /* X */ | |
5367 | s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */ | |
5368 | s[4] != '_'; /* Y */ | |
5369 | } | |
5370 | ||
5371 | /* Given 'some_struct_name___with_flavor' return the length of a name prefix | |
5372 | * before last triple underscore. Struct name part after last triple | |
5373 | * underscore is ignored by BPF CO-RE relocation during relocation matching. | |
5374 | */ | |
b0588390 | 5375 | size_t bpf_core_essential_name_len(const char *name) |
ddc7c304 AN |
5376 | { |
5377 | size_t n = strlen(name); | |
5378 | int i; | |
5379 | ||
5380 | for (i = n - 5; i >= 0; i--) { | |
5381 | if (bpf_core_is_flavor_sep(name + i)) | |
5382 | return i + 1; | |
5383 | } | |
5384 | return n; | |
5385 | } | |
5386 | ||
8de6cae4 | 5387 | void bpf_core_free_cands(struct bpf_core_cand_list *cands) |
ddc7c304 | 5388 | { |
8de6cae4 MV |
5389 | if (!cands) |
5390 | return; | |
5391 | ||
0f7515ca AN |
5392 | free(cands->cands); |
5393 | free(cands); | |
ddc7c304 AN |
5394 | } |
5395 | ||
8de6cae4 MV |
5396 | int bpf_core_add_cands(struct bpf_core_cand *local_cand, |
5397 | size_t local_essent_len, | |
5398 | const struct btf *targ_btf, | |
5399 | const char *targ_btf_name, | |
5400 | int targ_start_id, | |
5401 | struct bpf_core_cand_list *cands) | |
ddc7c304 | 5402 | { |
301ba4d7 | 5403 | struct bpf_core_cand *new_cands, *cand; |
03d5b991 AN |
5404 | const struct btf_type *t, *local_t; |
5405 | const char *targ_name, *local_name; | |
0f7515ca AN |
5406 | size_t targ_essent_len; |
5407 | int n, i; | |
ddc7c304 | 5408 | |
03d5b991 AN |
5409 | local_t = btf__type_by_id(local_cand->btf, local_cand->id); |
5410 | local_name = btf__str_by_offset(local_cand->btf, local_t->name_off); | |
5411 | ||
6a886de0 HC |
5412 | n = btf__type_cnt(targ_btf); |
5413 | for (i = targ_start_id; i < n; i++) { | |
ddc7c304 | 5414 | t = btf__type_by_id(targ_btf, i); |
23b2a3a8 | 5415 | if (!btf_kind_core_compat(t, local_t)) |
ddc7c304 AN |
5416 | continue; |
5417 | ||
3fc32f40 AN |
5418 | targ_name = btf__name_by_offset(targ_btf, t->name_off); |
5419 | if (str_is_empty(targ_name)) | |
d121e1d3 AN |
5420 | continue; |
5421 | ||
ddc7c304 AN |
5422 | targ_essent_len = bpf_core_essential_name_len(targ_name); |
5423 | if (targ_essent_len != local_essent_len) | |
5424 | continue; | |
5425 | ||
03d5b991 | 5426 | if (strncmp(local_name, targ_name, local_essent_len) != 0) |
0f7515ca AN |
5427 | continue; |
5428 | ||
5429 | pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n", | |
03d5b991 AN |
5430 | local_cand->id, btf_kind_str(local_t), |
5431 | local_name, i, btf_kind_str(t), targ_name, | |
0f7515ca AN |
5432 | targ_btf_name); |
5433 | new_cands = libbpf_reallocarray(cands->cands, cands->len + 1, | |
5434 | sizeof(*cands->cands)); | |
5435 | if (!new_cands) | |
5436 | return -ENOMEM; | |
5437 | ||
5438 | cand = &new_cands[cands->len]; | |
5439 | cand->btf = targ_btf; | |
0f7515ca AN |
5440 | cand->id = i; |
5441 | ||
5442 | cands->cands = new_cands; | |
5443 | cands->len++; | |
ddc7c304 | 5444 | } |
0f7515ca AN |
5445 | return 0; |
5446 | } | |
5447 | ||
4f33a53d AN |
5448 | static int load_module_btfs(struct bpf_object *obj) |
5449 | { | |
5450 | struct bpf_btf_info info; | |
5451 | struct module_btf *mod_btf; | |
5452 | struct btf *btf; | |
5453 | char name[64]; | |
5454 | __u32 id = 0, len; | |
5455 | int err, fd; | |
5456 | ||
5457 | if (obj->btf_modules_loaded) | |
5458 | return 0; | |
5459 | ||
67234743 AS |
5460 | if (obj->gen_loader) |
5461 | return 0; | |
5462 | ||
4f33a53d AN |
5463 | /* don't do this again, even if we find no module BTFs */ |
5464 | obj->btf_modules_loaded = true; | |
5465 | ||
5466 | /* kernel too old to support module BTFs */ | |
9ca1f56a | 5467 | if (!kernel_supports(obj, FEAT_MODULE_BTF)) |
4f33a53d AN |
5468 | return 0; |
5469 | ||
5470 | while (true) { | |
5471 | err = bpf_btf_get_next_id(id, &id); | |
5472 | if (err && errno == ENOENT) | |
5473 | return 0; | |
5474 | if (err) { | |
5475 | err = -errno; | |
5476 | pr_warn("failed to iterate BTF objects: %d\n", err); | |
5477 | return err; | |
ddc7c304 | 5478 | } |
4f33a53d AN |
5479 | |
5480 | fd = bpf_btf_get_fd_by_id(id); | |
5481 | if (fd < 0) { | |
5482 | if (errno == ENOENT) | |
5483 | continue; /* expected race: BTF was unloaded */ | |
5484 | err = -errno; | |
5485 | pr_warn("failed to get BTF object #%d FD: %d\n", id, err); | |
5486 | return err; | |
5487 | } | |
5488 | ||
5489 | len = sizeof(info); | |
5490 | memset(&info, 0, sizeof(info)); | |
5491 | info.name = ptr_to_u64(name); | |
5492 | info.name_len = sizeof(name); | |
5493 | ||
629dfc66 | 5494 | err = bpf_btf_get_info_by_fd(fd, &info, &len); |
4f33a53d AN |
5495 | if (err) { |
5496 | err = -errno; | |
5497 | pr_warn("failed to get BTF object #%d info: %d\n", id, err); | |
91abb4a6 | 5498 | goto err_out; |
4f33a53d AN |
5499 | } |
5500 | ||
5501 | /* ignore non-module BTFs */ | |
5502 | if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) { | |
5503 | close(fd); | |
5504 | continue; | |
5505 | } | |
5506 | ||
5507 | btf = btf_get_from_fd(fd, obj->btf_vmlinux); | |
e9fc3ce9 AN |
5508 | err = libbpf_get_error(btf); |
5509 | if (err) { | |
5510 | pr_warn("failed to load module [%s]'s BTF object #%d: %d\n", | |
5511 | name, id, err); | |
91abb4a6 | 5512 | goto err_out; |
4f33a53d AN |
5513 | } |
5514 | ||
3b029e06 | 5515 | err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap, |
e3ba8e4e | 5516 | sizeof(*obj->btf_modules), obj->btf_module_cnt + 1); |
4f33a53d | 5517 | if (err) |
91abb4a6 | 5518 | goto err_out; |
4f33a53d AN |
5519 | |
5520 | mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; | |
5521 | ||
5522 | mod_btf->btf = btf; | |
5523 | mod_btf->id = id; | |
91abb4a6 | 5524 | mod_btf->fd = fd; |
4f33a53d | 5525 | mod_btf->name = strdup(name); |
91abb4a6 AN |
5526 | if (!mod_btf->name) { |
5527 | err = -ENOMEM; | |
5528 | goto err_out; | |
5529 | } | |
5530 | continue; | |
5531 | ||
5532 | err_out: | |
5533 | close(fd); | |
5534 | return err; | |
ddc7c304 | 5535 | } |
4f33a53d AN |
5536 | |
5537 | return 0; | |
5538 | } | |
5539 | ||
301ba4d7 | 5540 | static struct bpf_core_cand_list * |
0f7515ca AN |
5541 | bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id) |
5542 | { | |
301ba4d7 AS |
5543 | struct bpf_core_cand local_cand = {}; |
5544 | struct bpf_core_cand_list *cands; | |
4f33a53d | 5545 | const struct btf *main_btf; |
03d5b991 AN |
5546 | const struct btf_type *local_t; |
5547 | const char *local_name; | |
0f7515ca | 5548 | size_t local_essent_len; |
4f33a53d | 5549 | int err, i; |
0f7515ca AN |
5550 | |
5551 | local_cand.btf = local_btf; | |
03d5b991 AN |
5552 | local_cand.id = local_type_id; |
5553 | local_t = btf__type_by_id(local_btf, local_type_id); | |
5554 | if (!local_t) | |
0f7515ca AN |
5555 | return ERR_PTR(-EINVAL); |
5556 | ||
03d5b991 AN |
5557 | local_name = btf__name_by_offset(local_btf, local_t->name_off); |
5558 | if (str_is_empty(local_name)) | |
0f7515ca | 5559 | return ERR_PTR(-EINVAL); |
03d5b991 | 5560 | local_essent_len = bpf_core_essential_name_len(local_name); |
0f7515ca AN |
5561 | |
5562 | cands = calloc(1, sizeof(*cands)); | |
5563 | if (!cands) | |
5564 | return ERR_PTR(-ENOMEM); | |
5565 | ||
5566 | /* Attempt to find target candidates in vmlinux BTF first */ | |
4f33a53d AN |
5567 | main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux; |
5568 | err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands); | |
5569 | if (err) | |
5570 | goto err_out; | |
5571 | ||
5572 | /* if vmlinux BTF has any candidate, don't got for module BTFs */ | |
5573 | if (cands->len) | |
5574 | return cands; | |
5575 | ||
5576 | /* if vmlinux BTF was overridden, don't attempt to load module BTFs */ | |
5577 | if (obj->btf_vmlinux_override) | |
5578 | return cands; | |
5579 | ||
5580 | /* now look through module BTFs, trying to still find candidates */ | |
5581 | err = load_module_btfs(obj); | |
5582 | if (err) | |
5583 | goto err_out; | |
5584 | ||
5585 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
5586 | err = bpf_core_add_cands(&local_cand, local_essent_len, | |
5587 | obj->btf_modules[i].btf, | |
5588 | obj->btf_modules[i].name, | |
6a886de0 | 5589 | btf__type_cnt(obj->btf_vmlinux), |
4f33a53d AN |
5590 | cands); |
5591 | if (err) | |
5592 | goto err_out; | |
0f7515ca AN |
5593 | } |
5594 | ||
5595 | return cands; | |
ddc7c304 | 5596 | err_out: |
4f33a53d | 5597 | bpf_core_free_cands(cands); |
ddc7c304 AN |
5598 | return ERR_PTR(err); |
5599 | } | |
5600 | ||
3fc32f40 AN |
5601 | /* Check local and target types for compatibility. This check is used for |
5602 | * type-based CO-RE relocations and follow slightly different rules than | |
5603 | * field-based relocations. This function assumes that root types were already | |
5604 | * checked for name match. Beyond that initial root-level name check, names | |
5605 | * are completely ignored. Compatibility rules are as follows: | |
5606 | * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but | |
5607 | * kind should match for local and target types (i.e., STRUCT is not | |
5608 | * compatible with UNION); | |
5609 | * - for ENUMs, the size is ignored; | |
5610 | * - for INT, size and signedness are ignored; | |
5611 | * - for ARRAY, dimensionality is ignored, element types are checked for | |
5612 | * compatibility recursively; | |
5613 | * - CONST/VOLATILE/RESTRICT modifiers are ignored; | |
5614 | * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible; | |
5615 | * - FUNC_PROTOs are compatible if they have compatible signature: same | |
5616 | * number of input args and compatible return and argument types. | |
5617 | * These rules are not set in stone and probably will be adjusted as we get | |
5618 | * more experience with using BPF CO-RE relocations. | |
5619 | */ | |
b0588390 AS |
5620 | int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id, |
5621 | const struct btf *targ_btf, __u32 targ_id) | |
3fc32f40 | 5622 | { |
fd75733d | 5623 | return __bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id, 32); |
3fc32f40 AN |
5624 | } |
5625 | ||
ec6209c8 DM |
5626 | int bpf_core_types_match(const struct btf *local_btf, __u32 local_id, |
5627 | const struct btf *targ_btf, __u32 targ_id) | |
5628 | { | |
5629 | return __bpf_core_types_match(local_btf, local_id, targ_btf, targ_id, false, 32); | |
5630 | } | |
5631 | ||
c302378b | 5632 | static size_t bpf_core_hash_fn(const long key, void *ctx) |
ddc7c304 | 5633 | { |
c302378b | 5634 | return key; |
ddc7c304 AN |
5635 | } |
5636 | ||
c302378b | 5637 | static bool bpf_core_equal_fn(const long k1, const long k2, void *ctx) |
ddc7c304 AN |
5638 | { |
5639 | return k1 == k2; | |
5640 | } | |
5641 | ||
d0e92887 AS |
5642 | static int record_relo_core(struct bpf_program *prog, |
5643 | const struct bpf_core_relo *core_relo, int insn_idx) | |
5644 | { | |
5645 | struct reloc_desc *relos, *relo; | |
5646 | ||
5647 | relos = libbpf_reallocarray(prog->reloc_desc, | |
5648 | prog->nr_reloc + 1, sizeof(*relos)); | |
5649 | if (!relos) | |
5650 | return -ENOMEM; | |
5651 | relo = &relos[prog->nr_reloc]; | |
5652 | relo->type = RELO_CORE; | |
5653 | relo->insn_idx = insn_idx; | |
5654 | relo->core_relo = core_relo; | |
5655 | prog->reloc_desc = relos; | |
5656 | prog->nr_reloc++; | |
5657 | return 0; | |
5658 | } | |
5659 | ||
9fdc4273 AN |
5660 | static const struct bpf_core_relo *find_relo_core(struct bpf_program *prog, int insn_idx) |
5661 | { | |
5662 | struct reloc_desc *relo; | |
5663 | int i; | |
5664 | ||
5665 | for (i = 0; i < prog->nr_reloc; i++) { | |
5666 | relo = &prog->reloc_desc[i]; | |
5667 | if (relo->type != RELO_CORE || relo->insn_idx != insn_idx) | |
5668 | continue; | |
5669 | ||
5670 | return relo->core_relo; | |
5671 | } | |
5672 | ||
5673 | return NULL; | |
5674 | } | |
5675 | ||
adb8fa19 MV |
5676 | static int bpf_core_resolve_relo(struct bpf_program *prog, |
5677 | const struct bpf_core_relo *relo, | |
5678 | int relo_idx, | |
5679 | const struct btf *local_btf, | |
5680 | struct hashmap *cand_cache, | |
5681 | struct bpf_core_relo_res *targ_res) | |
3ee4f533 | 5682 | { |
78c1f8d0 | 5683 | struct bpf_core_spec specs_scratch[3] = {}; |
301ba4d7 | 5684 | struct bpf_core_cand_list *cands = NULL; |
3ee4f533 AS |
5685 | const char *prog_name = prog->name; |
5686 | const struct btf_type *local_type; | |
5687 | const char *local_name; | |
5688 | __u32 local_id = relo->type_id; | |
adb8fa19 | 5689 | int err; |
3ee4f533 AS |
5690 | |
5691 | local_type = btf__type_by_id(local_btf, local_id); | |
5692 | if (!local_type) | |
5693 | return -EINVAL; | |
5694 | ||
5695 | local_name = btf__name_by_offset(local_btf, local_type->name_off); | |
5696 | if (!local_name) | |
5697 | return -EINVAL; | |
5698 | ||
46334a0c | 5699 | if (relo->kind != BPF_CORE_TYPE_ID_LOCAL && |
c302378b | 5700 | !hashmap__find(cand_cache, local_id, &cands)) { |
3ee4f533 AS |
5701 | cands = bpf_core_find_cands(prog->obj, local_btf, local_id); |
5702 | if (IS_ERR(cands)) { | |
5703 | pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n", | |
5704 | prog_name, relo_idx, local_id, btf_kind_str(local_type), | |
5705 | local_name, PTR_ERR(cands)); | |
5706 | return PTR_ERR(cands); | |
5707 | } | |
c302378b | 5708 | err = hashmap__set(cand_cache, local_id, cands, NULL, NULL); |
3ee4f533 AS |
5709 | if (err) { |
5710 | bpf_core_free_cands(cands); | |
5711 | return err; | |
5712 | } | |
5713 | } | |
5714 | ||
adb8fa19 MV |
5715 | return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch, |
5716 | targ_res); | |
3ee4f533 AS |
5717 | } |
5718 | ||
ddc7c304 | 5719 | static int |
28b93c64 | 5720 | bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) |
ddc7c304 AN |
5721 | { |
5722 | const struct btf_ext_info_sec *sec; | |
adb8fa19 | 5723 | struct bpf_core_relo_res targ_res; |
28b93c64 | 5724 | const struct bpf_core_relo *rec; |
ddc7c304 AN |
5725 | const struct btf_ext_info *seg; |
5726 | struct hashmap_entry *entry; | |
5727 | struct hashmap *cand_cache = NULL; | |
5728 | struct bpf_program *prog; | |
adb8fa19 | 5729 | struct bpf_insn *insn; |
ddc7c304 | 5730 | const char *sec_name; |
11d5daa8 | 5731 | int i, err = 0, insn_idx, sec_idx, sec_num; |
ddc7c304 | 5732 | |
28b93c64 AN |
5733 | if (obj->btf_ext->core_relo_info.len == 0) |
5734 | return 0; | |
5735 | ||
0f7515ca AN |
5736 | if (targ_btf_path) { |
5737 | obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL); | |
e9fc3ce9 AN |
5738 | err = libbpf_get_error(obj->btf_vmlinux_override); |
5739 | if (err) { | |
0f7515ca AN |
5740 | pr_warn("failed to parse target BTF: %d\n", err); |
5741 | return err; | |
5742 | } | |
ddc7c304 AN |
5743 | } |
5744 | ||
5745 | cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL); | |
5746 | if (IS_ERR(cand_cache)) { | |
5747 | err = PTR_ERR(cand_cache); | |
5748 | goto out; | |
5749 | } | |
5750 | ||
28b93c64 | 5751 | seg = &obj->btf_ext->core_relo_info; |
11d5daa8 | 5752 | sec_num = 0; |
ddc7c304 | 5753 | for_each_btf_ext_sec(seg, sec) { |
11d5daa8 AN |
5754 | sec_idx = seg->sec_idxs[sec_num]; |
5755 | sec_num++; | |
5756 | ||
ddc7c304 AN |
5757 | sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off); |
5758 | if (str_is_empty(sec_name)) { | |
5759 | err = -EINVAL; | |
5760 | goto out; | |
5761 | } | |
ddc7c304 | 5762 | |
11d5daa8 | 5763 | pr_debug("sec '%s': found %d CO-RE relocations\n", sec_name, sec->num_info); |
ddc7c304 AN |
5764 | |
5765 | for_each_btf_ext_rec(seg, sec, i, rec) { | |
adb8fa19 MV |
5766 | if (rec->insn_off % BPF_INSN_SZ) |
5767 | return -EINVAL; | |
db2b8b06 AN |
5768 | insn_idx = rec->insn_off / BPF_INSN_SZ; |
5769 | prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx); | |
5770 | if (!prog) { | |
e89d57d9 AN |
5771 | /* When __weak subprog is "overridden" by another instance |
5772 | * of the subprog from a different object file, linker still | |
5773 | * appends all the .BTF.ext info that used to belong to that | |
5774 | * eliminated subprogram. | |
5775 | * This is similar to what x86-64 linker does for relocations. | |
5776 | * So just ignore such relocations just like we ignore | |
5777 | * subprog instructions when discovering subprograms. | |
5778 | */ | |
5779 | pr_debug("sec '%s': skipping CO-RE relocation #%d for insn #%d belonging to eliminated weak subprogram\n", | |
5780 | sec_name, i, insn_idx); | |
5781 | continue; | |
db2b8b06 | 5782 | } |
47f7cf63 AN |
5783 | /* no need to apply CO-RE relocation if the program is |
5784 | * not going to be loaded | |
5785 | */ | |
a3820c48 | 5786 | if (!prog->autoload) |
47f7cf63 | 5787 | continue; |
db2b8b06 | 5788 | |
adb8fa19 MV |
5789 | /* adjust insn_idx from section frame of reference to the local |
5790 | * program's frame of reference; (sub-)program code is not yet | |
5791 | * relocated, so it's enough to just subtract in-section offset | |
5792 | */ | |
5793 | insn_idx = insn_idx - prog->sec_insn_off; | |
5794 | if (insn_idx >= prog->insns_cnt) | |
5795 | return -EINVAL; | |
5796 | insn = &prog->insns[insn_idx]; | |
5797 | ||
185cfe83 AN |
5798 | err = record_relo_core(prog, rec, insn_idx); |
5799 | if (err) { | |
5800 | pr_warn("prog '%s': relo #%d: failed to record relocation: %d\n", | |
5801 | prog->name, i, err); | |
5802 | goto out; | |
adb8fa19 MV |
5803 | } |
5804 | ||
185cfe83 AN |
5805 | if (prog->obj->gen_loader) |
5806 | continue; | |
5807 | ||
adb8fa19 | 5808 | err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res); |
ddc7c304 | 5809 | if (err) { |
be18010e | 5810 | pr_warn("prog '%s': relo #%d: failed to relocate: %d\n", |
9c0f8cbd | 5811 | prog->name, i, err); |
ddc7c304 AN |
5812 | goto out; |
5813 | } | |
adb8fa19 MV |
5814 | |
5815 | err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res); | |
5816 | if (err) { | |
5817 | pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n", | |
5818 | prog->name, i, insn_idx, err); | |
5819 | goto out; | |
5820 | } | |
ddc7c304 AN |
5821 | } |
5822 | } | |
5823 | ||
5824 | out: | |
4f33a53d | 5825 | /* obj->btf_vmlinux and module BTFs are freed after object load */ |
0f7515ca AN |
5826 | btf__free(obj->btf_vmlinux_override); |
5827 | obj->btf_vmlinux_override = NULL; | |
5828 | ||
ddc7c304 AN |
5829 | if (!IS_ERR_OR_NULL(cand_cache)) { |
5830 | hashmap__for_each_entry(cand_cache, entry, i) { | |
c302378b | 5831 | bpf_core_free_cands(entry->pvalue); |
ddc7c304 AN |
5832 | } |
5833 | hashmap__free(cand_cache); | |
5834 | } | |
5835 | return err; | |
5836 | } | |
5837 | ||
ec41817b | 5838 | /* base map load ldimm64 special constant, used also for log fixup logic */ |
3055ddd6 AN |
5839 | #define POISON_LDIMM64_MAP_BASE 2001000000 |
5840 | #define POISON_LDIMM64_MAP_PFX "200100" | |
ec41817b AN |
5841 | |
5842 | static void poison_map_ldimm64(struct bpf_program *prog, int relo_idx, | |
5843 | int insn_idx, struct bpf_insn *insn, | |
5844 | int map_idx, const struct bpf_map *map) | |
5845 | { | |
5846 | int i; | |
5847 | ||
5848 | pr_debug("prog '%s': relo #%d: poisoning insn #%d that loads map #%d '%s'\n", | |
5849 | prog->name, relo_idx, insn_idx, map_idx, map->name); | |
5850 | ||
5851 | /* we turn single ldimm64 into two identical invalid calls */ | |
5852 | for (i = 0; i < 2; i++) { | |
5853 | insn->code = BPF_JMP | BPF_CALL; | |
5854 | insn->dst_reg = 0; | |
5855 | insn->src_reg = 0; | |
5856 | insn->off = 0; | |
5857 | /* if this instruction is reachable (not a dead code), | |
5858 | * verifier will complain with something like: | |
5859 | * invalid func unknown#2001000123 | |
5860 | * where lower 123 is map index into obj->maps[] array | |
5861 | */ | |
3055ddd6 | 5862 | insn->imm = POISON_LDIMM64_MAP_BASE + map_idx; |
ec41817b AN |
5863 | |
5864 | insn++; | |
5865 | } | |
5866 | } | |
5867 | ||
05b6f766 AN |
5868 | /* unresolved kfunc call special constant, used also for log fixup logic */ |
5869 | #define POISON_CALL_KFUNC_BASE 2002000000 | |
5870 | #define POISON_CALL_KFUNC_PFX "2002" | |
5871 | ||
5872 | static void poison_kfunc_call(struct bpf_program *prog, int relo_idx, | |
5873 | int insn_idx, struct bpf_insn *insn, | |
5874 | int ext_idx, const struct extern_desc *ext) | |
5875 | { | |
5876 | pr_debug("prog '%s': relo #%d: poisoning insn #%d that calls kfunc '%s'\n", | |
5877 | prog->name, relo_idx, insn_idx, ext->name); | |
5878 | ||
5879 | /* we turn kfunc call into invalid helper call with identifiable constant */ | |
5880 | insn->code = BPF_JMP | BPF_CALL; | |
5881 | insn->dst_reg = 0; | |
5882 | insn->src_reg = 0; | |
5883 | insn->off = 0; | |
5884 | /* if this instruction is reachable (not a dead code), | |
5885 | * verifier will complain with something like: | |
5886 | * invalid func unknown#2001000123 | |
5887 | * where lower 123 is extern index into obj->externs[] array | |
5888 | */ | |
5889 | insn->imm = POISON_CALL_KFUNC_BASE + ext_idx; | |
5890 | } | |
5891 | ||
c3c55696 AN |
5892 | /* Relocate data references within program code: |
5893 | * - map references; | |
5894 | * - global variable references; | |
5895 | * - extern references. | |
5896 | */ | |
48cca7e4 | 5897 | static int |
c3c55696 | 5898 | bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog) |
8a47a6c5 | 5899 | { |
c3c55696 | 5900 | int i; |
8a47a6c5 WN |
5901 | |
5902 | for (i = 0; i < prog->nr_reloc; i++) { | |
53f8dd43 | 5903 | struct reloc_desc *relo = &prog->reloc_desc[i]; |
166750bc | 5904 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; |
ec41817b | 5905 | const struct bpf_map *map; |
2e33efe3 | 5906 | struct extern_desc *ext; |
8a47a6c5 | 5907 | |
166750bc AN |
5908 | switch (relo->type) { |
5909 | case RELO_LD64: | |
ec41817b | 5910 | map = &obj->maps[relo->map_idx]; |
e2fa0156 AS |
5911 | if (obj->gen_loader) { |
5912 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX; | |
5913 | insn[0].imm = relo->map_idx; | |
ec41817b | 5914 | } else if (map->autocreate) { |
e2fa0156 | 5915 | insn[0].src_reg = BPF_PSEUDO_MAP_FD; |
ec41817b AN |
5916 | insn[0].imm = map->fd; |
5917 | } else { | |
5918 | poison_map_ldimm64(prog, i, relo->insn_idx, insn, | |
5919 | relo->map_idx, map); | |
e2fa0156 | 5920 | } |
166750bc AN |
5921 | break; |
5922 | case RELO_DATA: | |
ec41817b | 5923 | map = &obj->maps[relo->map_idx]; |
166750bc | 5924 | insn[1].imm = insn[0].imm + relo->sym_off; |
e2fa0156 AS |
5925 | if (obj->gen_loader) { |
5926 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
5927 | insn[0].imm = relo->map_idx; | |
ec41817b | 5928 | } else if (map->autocreate) { |
e2fa0156 | 5929 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; |
ec41817b AN |
5930 | insn[0].imm = map->fd; |
5931 | } else { | |
5932 | poison_map_ldimm64(prog, i, relo->insn_idx, insn, | |
5933 | relo->map_idx, map); | |
e2fa0156 | 5934 | } |
166750bc | 5935 | break; |
a18f7214 | 5936 | case RELO_EXTERN_LD64: |
3055ddd6 | 5937 | ext = &obj->externs[relo->ext_idx]; |
1c0c7074 | 5938 | if (ext->type == EXT_KCFG) { |
e2fa0156 AS |
5939 | if (obj->gen_loader) { |
5940 | insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE; | |
5941 | insn[0].imm = obj->kconfig_map_idx; | |
5942 | } else { | |
5943 | insn[0].src_reg = BPF_PSEUDO_MAP_VALUE; | |
5944 | insn[0].imm = obj->maps[obj->kconfig_map_idx].fd; | |
5945 | } | |
1c0c7074 AN |
5946 | insn[1].imm = ext->kcfg.data_off; |
5947 | } else /* EXT_KSYM */ { | |
2211c825 | 5948 | if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */ |
d370bbe1 | 5949 | insn[0].src_reg = BPF_PSEUDO_BTF_ID; |
284d2587 AN |
5950 | insn[0].imm = ext->ksym.kernel_btf_id; |
5951 | insn[1].imm = ext->ksym.kernel_btf_obj_fd; | |
2211c825 | 5952 | } else { /* typeless ksyms or unresolved typed ksyms */ |
d370bbe1 HL |
5953 | insn[0].imm = (__u32)ext->ksym.addr; |
5954 | insn[1].imm = ext->ksym.addr >> 32; | |
5955 | } | |
1c0c7074 | 5956 | } |
166750bc | 5957 | break; |
a18f7214 | 5958 | case RELO_EXTERN_CALL: |
3055ddd6 | 5959 | ext = &obj->externs[relo->ext_idx]; |
5bd022ec | 5960 | insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL; |
466b2e13 KKD |
5961 | if (ext->is_set) { |
5962 | insn[0].imm = ext->ksym.kernel_btf_id; | |
5963 | insn[0].off = ext->ksym.btf_fd_idx; | |
05b6f766 AN |
5964 | } else { /* unresolved weak kfunc call */ |
5965 | poison_kfunc_call(prog, i, relo->insn_idx, insn, | |
5966 | relo->ext_idx, ext); | |
466b2e13 | 5967 | } |
5bd022ec | 5968 | break; |
53eddb5e | 5969 | case RELO_SUBPROG_ADDR: |
b1268826 AS |
5970 | if (insn[0].src_reg != BPF_PSEUDO_FUNC) { |
5971 | pr_warn("prog '%s': relo #%d: bad insn\n", | |
5972 | prog->name, i); | |
5973 | return -EINVAL; | |
5974 | } | |
5975 | /* handled already */ | |
53eddb5e | 5976 | break; |
166750bc | 5977 | case RELO_CALL: |
b1268826 | 5978 | /* handled already */ |
166750bc | 5979 | break; |
d0e92887 AS |
5980 | case RELO_CORE: |
5981 | /* will be handled by bpf_program_record_relos() */ | |
5982 | break; | |
166750bc | 5983 | default: |
9c0f8cbd AN |
5984 | pr_warn("prog '%s': relo #%d: bad relo type %d\n", |
5985 | prog->name, i, relo->type); | |
166750bc | 5986 | return -EINVAL; |
8a47a6c5 | 5987 | } |
8a47a6c5 WN |
5988 | } |
5989 | ||
c3c55696 AN |
5990 | return 0; |
5991 | } | |
5992 | ||
8505e870 AN |
5993 | static int adjust_prog_btf_ext_info(const struct bpf_object *obj, |
5994 | const struct bpf_program *prog, | |
5995 | const struct btf_ext_info *ext_info, | |
5996 | void **prog_info, __u32 *prog_rec_cnt, | |
5997 | __u32 *prog_rec_sz) | |
5998 | { | |
5999 | void *copy_start = NULL, *copy_end = NULL; | |
6000 | void *rec, *rec_end, *new_prog_info; | |
6001 | const struct btf_ext_info_sec *sec; | |
6002 | size_t old_sz, new_sz; | |
11d5daa8 | 6003 | int i, sec_num, sec_idx, off_adj; |
8505e870 | 6004 | |
11d5daa8 | 6005 | sec_num = 0; |
8505e870 | 6006 | for_each_btf_ext_sec(ext_info, sec) { |
11d5daa8 AN |
6007 | sec_idx = ext_info->sec_idxs[sec_num]; |
6008 | sec_num++; | |
6009 | if (prog->sec_idx != sec_idx) | |
8505e870 AN |
6010 | continue; |
6011 | ||
6012 | for_each_btf_ext_rec(ext_info, sec, i, rec) { | |
6013 | __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ; | |
6014 | ||
6015 | if (insn_off < prog->sec_insn_off) | |
6016 | continue; | |
6017 | if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt) | |
6018 | break; | |
6019 | ||
6020 | if (!copy_start) | |
6021 | copy_start = rec; | |
6022 | copy_end = rec + ext_info->rec_size; | |
6023 | } | |
6024 | ||
6025 | if (!copy_start) | |
6026 | return -ENOENT; | |
6027 | ||
6028 | /* append func/line info of a given (sub-)program to the main | |
6029 | * program func/line info | |
6030 | */ | |
8eb62958 | 6031 | old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size; |
8505e870 AN |
6032 | new_sz = old_sz + (copy_end - copy_start); |
6033 | new_prog_info = realloc(*prog_info, new_sz); | |
6034 | if (!new_prog_info) | |
6035 | return -ENOMEM; | |
6036 | *prog_info = new_prog_info; | |
6037 | *prog_rec_cnt = new_sz / ext_info->rec_size; | |
6038 | memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start); | |
6039 | ||
6040 | /* Kernel instruction offsets are in units of 8-byte | |
6041 | * instructions, while .BTF.ext instruction offsets generated | |
6042 | * by Clang are in units of bytes. So convert Clang offsets | |
6043 | * into kernel offsets and adjust offset according to program | |
6044 | * relocated position. | |
6045 | */ | |
6046 | off_adj = prog->sub_insn_off - prog->sec_insn_off; | |
6047 | rec = new_prog_info + old_sz; | |
6048 | rec_end = new_prog_info + new_sz; | |
6049 | for (; rec < rec_end; rec += ext_info->rec_size) { | |
6050 | __u32 *insn_off = rec; | |
6051 | ||
6052 | *insn_off = *insn_off / BPF_INSN_SZ + off_adj; | |
6053 | } | |
6054 | *prog_rec_sz = ext_info->rec_size; | |
6055 | return 0; | |
6056 | } | |
6057 | ||
6058 | return -ENOENT; | |
6059 | } | |
6060 | ||
6061 | static int | |
6062 | reloc_prog_func_and_line_info(const struct bpf_object *obj, | |
6063 | struct bpf_program *main_prog, | |
6064 | const struct bpf_program *prog) | |
6065 | { | |
6066 | int err; | |
6067 | ||
6068 | /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't | |
6069 | * supprot func/line info | |
6070 | */ | |
9ca1f56a | 6071 | if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC)) |
8505e870 AN |
6072 | return 0; |
6073 | ||
6074 | /* only attempt func info relocation if main program's func_info | |
6075 | * relocation was successful | |
6076 | */ | |
6077 | if (main_prog != prog && !main_prog->func_info) | |
6078 | goto line_info; | |
6079 | ||
6080 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info, | |
6081 | &main_prog->func_info, | |
6082 | &main_prog->func_info_cnt, | |
6083 | &main_prog->func_info_rec_size); | |
6084 | if (err) { | |
6085 | if (err != -ENOENT) { | |
6086 | pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n", | |
6087 | prog->name, err); | |
6088 | return err; | |
6089 | } | |
6090 | if (main_prog->func_info) { | |
6091 | /* | |
6092 | * Some info has already been found but has problem | |
6093 | * in the last btf_ext reloc. Must have to error out. | |
6094 | */ | |
6095 | pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name); | |
6096 | return err; | |
6097 | } | |
6098 | /* Have problem loading the very first info. Ignore the rest. */ | |
6099 | pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n", | |
6100 | prog->name); | |
6101 | } | |
6102 | ||
6103 | line_info: | |
6104 | /* don't relocate line info if main program's relocation failed */ | |
6105 | if (main_prog != prog && !main_prog->line_info) | |
6106 | return 0; | |
6107 | ||
6108 | err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info, | |
6109 | &main_prog->line_info, | |
6110 | &main_prog->line_info_cnt, | |
6111 | &main_prog->line_info_rec_size); | |
6112 | if (err) { | |
6113 | if (err != -ENOENT) { | |
6114 | pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n", | |
6115 | prog->name, err); | |
6116 | return err; | |
6117 | } | |
6118 | if (main_prog->line_info) { | |
6119 | /* | |
6120 | * Some info has already been found but has problem | |
6121 | * in the last btf_ext reloc. Must have to error out. | |
6122 | */ | |
6123 | pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name); | |
6124 | return err; | |
6125 | } | |
6126 | /* Have problem loading the very first info. Ignore the rest. */ | |
6127 | pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n", | |
6128 | prog->name); | |
6129 | } | |
6130 | return 0; | |
6131 | } | |
6132 | ||
c3c55696 AN |
6133 | static int cmp_relo_by_insn_idx(const void *key, const void *elem) |
6134 | { | |
6135 | size_t insn_idx = *(const size_t *)key; | |
6136 | const struct reloc_desc *relo = elem; | |
6137 | ||
6138 | if (insn_idx == relo->insn_idx) | |
6139 | return 0; | |
6140 | return insn_idx < relo->insn_idx ? -1 : 1; | |
6141 | } | |
6142 | ||
6143 | static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx) | |
6144 | { | |
2a6a9bf2 AN |
6145 | if (!prog->nr_reloc) |
6146 | return NULL; | |
c3c55696 AN |
6147 | return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc, |
6148 | sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx); | |
6149 | } | |
6150 | ||
b1268826 AS |
6151 | static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog) |
6152 | { | |
6153 | int new_cnt = main_prog->nr_reloc + subprog->nr_reloc; | |
6154 | struct reloc_desc *relos; | |
6155 | int i; | |
6156 | ||
6157 | if (main_prog == subprog) | |
6158 | return 0; | |
6159 | relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos)); | |
6160 | if (!relos) | |
6161 | return -ENOMEM; | |
2a6a9bf2 AN |
6162 | if (subprog->nr_reloc) |
6163 | memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc, | |
6164 | sizeof(*relos) * subprog->nr_reloc); | |
b1268826 AS |
6165 | |
6166 | for (i = main_prog->nr_reloc; i < new_cnt; i++) | |
6167 | relos[i].insn_idx += subprog->sub_insn_off; | |
6168 | /* After insn_idx adjustment the 'relos' array is still sorted | |
6169 | * by insn_idx and doesn't break bsearch. | |
6170 | */ | |
6171 | main_prog->reloc_desc = relos; | |
6172 | main_prog->nr_reloc = new_cnt; | |
6173 | return 0; | |
6174 | } | |
6175 | ||
c3c55696 AN |
6176 | static int |
6177 | bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog, | |
6178 | struct bpf_program *prog) | |
6179 | { | |
6180 | size_t sub_insn_idx, insn_idx, new_cnt; | |
6181 | struct bpf_program *subprog; | |
6182 | struct bpf_insn *insns, *insn; | |
6183 | struct reloc_desc *relo; | |
6184 | int err; | |
6185 | ||
6186 | err = reloc_prog_func_and_line_info(obj, main_prog, prog); | |
6187 | if (err) | |
6188 | return err; | |
6189 | ||
6190 | for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) { | |
6191 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
53eddb5e | 6192 | if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn)) |
c3c55696 AN |
6193 | continue; |
6194 | ||
6195 | relo = find_prog_insn_relo(prog, insn_idx); | |
a18f7214 | 6196 | if (relo && relo->type == RELO_EXTERN_CALL) |
b1268826 AS |
6197 | /* kfunc relocations will be handled later |
6198 | * in bpf_object__relocate_data() | |
6199 | */ | |
6200 | continue; | |
53eddb5e | 6201 | if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) { |
c3c55696 AN |
6202 | pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n", |
6203 | prog->name, insn_idx, relo->type); | |
6204 | return -LIBBPF_ERRNO__RELOC; | |
6205 | } | |
6206 | if (relo) { | |
6207 | /* sub-program instruction index is a combination of | |
6208 | * an offset of a symbol pointed to by relocation and | |
6209 | * call instruction's imm field; for global functions, | |
6210 | * call always has imm = -1, but for static functions | |
6211 | * relocation is against STT_SECTION and insn->imm | |
6212 | * points to a start of a static function | |
53eddb5e YS |
6213 | * |
6214 | * for subprog addr relocation, the relo->sym_off + insn->imm is | |
6215 | * the byte offset in the corresponding section. | |
c3c55696 | 6216 | */ |
53eddb5e YS |
6217 | if (relo->type == RELO_CALL) |
6218 | sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1; | |
6219 | else | |
6220 | sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ; | |
6221 | } else if (insn_is_pseudo_func(insn)) { | |
6222 | /* | |
6223 | * RELO_SUBPROG_ADDR relo is always emitted even if both | |
6224 | * functions are in the same section, so it shouldn't reach here. | |
6225 | */ | |
6226 | pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n", | |
6227 | prog->name, insn_idx); | |
6228 | return -LIBBPF_ERRNO__RELOC; | |
c3c55696 AN |
6229 | } else { |
6230 | /* if subprogram call is to a static function within | |
6231 | * the same ELF section, there won't be any relocation | |
6232 | * emitted, but it also means there is no additional | |
6233 | * offset necessary, insns->imm is relative to | |
6234 | * instruction's original position within the section | |
6235 | */ | |
6236 | sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1; | |
6237 | } | |
6238 | ||
6239 | /* we enforce that sub-programs should be in .text section */ | |
6240 | subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx); | |
6241 | if (!subprog) { | |
6242 | pr_warn("prog '%s': no .text section found yet sub-program call exists\n", | |
6243 | prog->name); | |
6244 | return -LIBBPF_ERRNO__RELOC; | |
6245 | } | |
6246 | ||
6247 | /* if it's the first call instruction calling into this | |
6248 | * subprogram (meaning this subprog hasn't been processed | |
6249 | * yet) within the context of current main program: | |
6250 | * - append it at the end of main program's instructions blog; | |
6251 | * - process is recursively, while current program is put on hold; | |
6252 | * - if that subprogram calls some other not yet processes | |
6253 | * subprogram, same thing will happen recursively until | |
6254 | * there are no more unprocesses subprograms left to append | |
6255 | * and relocate. | |
6256 | */ | |
6257 | if (subprog->sub_insn_off == 0) { | |
6258 | subprog->sub_insn_off = main_prog->insns_cnt; | |
6259 | ||
6260 | new_cnt = main_prog->insns_cnt + subprog->insns_cnt; | |
6261 | insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns)); | |
6262 | if (!insns) { | |
6263 | pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name); | |
6264 | return -ENOMEM; | |
6265 | } | |
6266 | main_prog->insns = insns; | |
6267 | main_prog->insns_cnt = new_cnt; | |
6268 | ||
6269 | memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns, | |
6270 | subprog->insns_cnt * sizeof(*insns)); | |
6271 | ||
6272 | pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n", | |
6273 | main_prog->name, subprog->insns_cnt, subprog->name); | |
6274 | ||
b1268826 AS |
6275 | /* The subprog insns are now appended. Append its relos too. */ |
6276 | err = append_subprog_relos(main_prog, subprog); | |
6277 | if (err) | |
6278 | return err; | |
c3c55696 AN |
6279 | err = bpf_object__reloc_code(obj, main_prog, subprog); |
6280 | if (err) | |
6281 | return err; | |
6282 | } | |
6283 | ||
6284 | /* main_prog->insns memory could have been re-allocated, so | |
6285 | * calculate pointer again | |
6286 | */ | |
6287 | insn = &main_prog->insns[prog->sub_insn_off + insn_idx]; | |
6288 | /* calculate correct instruction position within current main | |
6289 | * prog; each main prog can have a different set of | |
6290 | * subprograms appended (potentially in different order as | |
6291 | * well), so position of any subprog can be different for | |
e3ba8e4e KM |
6292 | * different main programs |
6293 | */ | |
c3c55696 AN |
6294 | insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1; |
6295 | ||
c3c55696 AN |
6296 | pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n", |
6297 | prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off); | |
6298 | } | |
6299 | ||
6300 | return 0; | |
6301 | } | |
6302 | ||
6303 | /* | |
6304 | * Relocate sub-program calls. | |
6305 | * | |
6306 | * Algorithm operates as follows. Each entry-point BPF program (referred to as | |
6307 | * main prog) is processed separately. For each subprog (non-entry functions, | |
6308 | * that can be called from either entry progs or other subprogs) gets their | |
6309 | * sub_insn_off reset to zero. This serves as indicator that this subprogram | |
6310 | * hasn't been yet appended and relocated within current main prog. Once its | |
6311 | * relocated, sub_insn_off will point at the position within current main prog | |
6312 | * where given subprog was appended. This will further be used to relocate all | |
6313 | * the call instructions jumping into this subprog. | |
6314 | * | |
6315 | * We start with main program and process all call instructions. If the call | |
6316 | * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off | |
6317 | * is zero), subprog instructions are appended at the end of main program's | |
6318 | * instruction array. Then main program is "put on hold" while we recursively | |
6319 | * process newly appended subprogram. If that subprogram calls into another | |
6320 | * subprogram that hasn't been appended, new subprogram is appended again to | |
6321 | * the *main* prog's instructions (subprog's instructions are always left | |
6322 | * untouched, as they need to be in unmodified state for subsequent main progs | |
6323 | * and subprog instructions are always sent only as part of a main prog) and | |
6324 | * the process continues recursively. Once all the subprogs called from a main | |
6325 | * prog or any of its subprogs are appended (and relocated), all their | |
6326 | * positions within finalized instructions array are known, so it's easy to | |
6327 | * rewrite call instructions with correct relative offsets, corresponding to | |
6328 | * desired target subprog. | |
6329 | * | |
6330 | * Its important to realize that some subprogs might not be called from some | |
6331 | * main prog and any of its called/used subprogs. Those will keep their | |
6332 | * subprog->sub_insn_off as zero at all times and won't be appended to current | |
6333 | * main prog and won't be relocated within the context of current main prog. | |
6334 | * They might still be used from other main progs later. | |
6335 | * | |
6336 | * Visually this process can be shown as below. Suppose we have two main | |
6337 | * programs mainA and mainB and BPF object contains three subprogs: subA, | |
6338 | * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and | |
6339 | * subC both call subB: | |
6340 | * | |
6341 | * +--------+ +-------+ | |
6342 | * | v v | | |
6343 | * +--+---+ +--+-+-+ +---+--+ | |
6344 | * | subA | | subB | | subC | | |
6345 | * +--+---+ +------+ +---+--+ | |
6346 | * ^ ^ | |
6347 | * | | | |
6348 | * +---+-------+ +------+----+ | |
6349 | * | mainA | | mainB | | |
6350 | * +-----------+ +-----------+ | |
6351 | * | |
6352 | * We'll start relocating mainA, will find subA, append it and start | |
6353 | * processing sub A recursively: | |
6354 | * | |
6355 | * +-----------+------+ | |
6356 | * | mainA | subA | | |
6357 | * +-----------+------+ | |
6358 | * | |
6359 | * At this point we notice that subB is used from subA, so we append it and | |
6360 | * relocate (there are no further subcalls from subB): | |
6361 | * | |
6362 | * +-----------+------+------+ | |
6363 | * | mainA | subA | subB | | |
6364 | * +-----------+------+------+ | |
6365 | * | |
6366 | * At this point, we relocate subA calls, then go one level up and finish with | |
6367 | * relocatin mainA calls. mainA is done. | |
6368 | * | |
6369 | * For mainB process is similar but results in different order. We start with | |
6370 | * mainB and skip subA and subB, as mainB never calls them (at least | |
6371 | * directly), but we see subC is needed, so we append and start processing it: | |
6372 | * | |
6373 | * +-----------+------+ | |
6374 | * | mainB | subC | | |
6375 | * +-----------+------+ | |
6376 | * Now we see subC needs subB, so we go back to it, append and relocate it: | |
6377 | * | |
6378 | * +-----------+------+------+ | |
6379 | * | mainB | subC | subB | | |
6380 | * +-----------+------+------+ | |
6381 | * | |
6382 | * At this point we unwind recursion, relocate calls in subC, then in mainB. | |
6383 | */ | |
6384 | static int | |
6385 | bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog) | |
6386 | { | |
6387 | struct bpf_program *subprog; | |
d3d93e34 | 6388 | int i, err; |
c3c55696 | 6389 | |
c3c55696 AN |
6390 | /* mark all subprogs as not relocated (yet) within the context of |
6391 | * current main program | |
6392 | */ | |
6393 | for (i = 0; i < obj->nr_programs; i++) { | |
6394 | subprog = &obj->programs[i]; | |
6395 | if (!prog_is_subprog(obj, subprog)) | |
6396 | continue; | |
6397 | ||
6398 | subprog->sub_insn_off = 0; | |
c3c55696 AN |
6399 | } |
6400 | ||
6401 | err = bpf_object__reloc_code(obj, prog, prog); | |
6402 | if (err) | |
6403 | return err; | |
6404 | ||
8a47a6c5 WN |
6405 | return 0; |
6406 | } | |
6407 | ||
67234743 AS |
6408 | static void |
6409 | bpf_object__free_relocs(struct bpf_object *obj) | |
6410 | { | |
6411 | struct bpf_program *prog; | |
6412 | int i; | |
6413 | ||
6414 | /* free up relocation descriptors */ | |
6415 | for (i = 0; i < obj->nr_programs; i++) { | |
6416 | prog = &obj->programs[i]; | |
6417 | zfree(&prog->reloc_desc); | |
6418 | prog->nr_reloc = 0; | |
6419 | } | |
6420 | } | |
6421 | ||
d0e92887 AS |
6422 | static int cmp_relocs(const void *_a, const void *_b) |
6423 | { | |
6424 | const struct reloc_desc *a = _a; | |
6425 | const struct reloc_desc *b = _b; | |
6426 | ||
6427 | if (a->insn_idx != b->insn_idx) | |
6428 | return a->insn_idx < b->insn_idx ? -1 : 1; | |
6429 | ||
6430 | /* no two relocations should have the same insn_idx, but ... */ | |
6431 | if (a->type != b->type) | |
6432 | return a->type < b->type ? -1 : 1; | |
6433 | ||
6434 | return 0; | |
6435 | } | |
6436 | ||
6437 | static void bpf_object__sort_relos(struct bpf_object *obj) | |
6438 | { | |
6439 | int i; | |
6440 | ||
6441 | for (i = 0; i < obj->nr_programs; i++) { | |
6442 | struct bpf_program *p = &obj->programs[i]; | |
6443 | ||
6444 | if (!p->nr_reloc) | |
6445 | continue; | |
6446 | ||
6447 | qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs); | |
6448 | } | |
6449 | } | |
6450 | ||
8a47a6c5 | 6451 | static int |
ddc7c304 | 6452 | bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path) |
8a47a6c5 WN |
6453 | { |
6454 | struct bpf_program *prog; | |
b1268826 | 6455 | size_t i, j; |
8a47a6c5 WN |
6456 | int err; |
6457 | ||
ddc7c304 AN |
6458 | if (obj->btf_ext) { |
6459 | err = bpf_object__relocate_core(obj, targ_btf_path); | |
6460 | if (err) { | |
be18010e KW |
6461 | pr_warn("failed to perform CO-RE relocations: %d\n", |
6462 | err); | |
ddc7c304 AN |
6463 | return err; |
6464 | } | |
185cfe83 | 6465 | bpf_object__sort_relos(obj); |
ddc7c304 | 6466 | } |
b1268826 AS |
6467 | |
6468 | /* Before relocating calls pre-process relocations and mark | |
6469 | * few ld_imm64 instructions that points to subprogs. | |
6470 | * Otherwise bpf_object__reloc_code() later would have to consider | |
6471 | * all ld_imm64 insns as relocation candidates. That would | |
6472 | * reduce relocation speed, since amount of find_prog_insn_relo() | |
6473 | * would increase and most of them will fail to find a relo. | |
9173cac3 AN |
6474 | */ |
6475 | for (i = 0; i < obj->nr_programs; i++) { | |
6476 | prog = &obj->programs[i]; | |
b1268826 AS |
6477 | for (j = 0; j < prog->nr_reloc; j++) { |
6478 | struct reloc_desc *relo = &prog->reloc_desc[j]; | |
6479 | struct bpf_insn *insn = &prog->insns[relo->insn_idx]; | |
6480 | ||
6481 | /* mark the insn, so it's recognized by insn_is_pseudo_func() */ | |
6482 | if (relo->type == RELO_SUBPROG_ADDR) | |
6483 | insn[0].src_reg = BPF_PSEUDO_FUNC; | |
9173cac3 | 6484 | } |
9173cac3 | 6485 | } |
b1268826 AS |
6486 | |
6487 | /* relocate subprogram calls and append used subprograms to main | |
c3c55696 AN |
6488 | * programs; each copy of subprogram code needs to be relocated |
6489 | * differently for each main program, because its code location might | |
b1268826 AS |
6490 | * have changed. |
6491 | * Append subprog relos to main programs to allow data relos to be | |
6492 | * processed after text is completely relocated. | |
9173cac3 | 6493 | */ |
8a47a6c5 WN |
6494 | for (i = 0; i < obj->nr_programs; i++) { |
6495 | prog = &obj->programs[i]; | |
c3c55696 AN |
6496 | /* sub-program's sub-calls are relocated within the context of |
6497 | * its main program only | |
6498 | */ | |
6499 | if (prog_is_subprog(obj, prog)) | |
9173cac3 | 6500 | continue; |
a3820c48 | 6501 | if (!prog->autoload) |
16e0c35c | 6502 | continue; |
8a47a6c5 | 6503 | |
c3c55696 | 6504 | err = bpf_object__relocate_calls(obj, prog); |
8a47a6c5 | 6505 | if (err) { |
9c0f8cbd AN |
6506 | pr_warn("prog '%s': failed to relocate calls: %d\n", |
6507 | prog->name, err); | |
8a47a6c5 WN |
6508 | return err; |
6509 | } | |
6510 | } | |
b1268826 AS |
6511 | /* Process data relos for main programs */ |
6512 | for (i = 0; i < obj->nr_programs; i++) { | |
6513 | prog = &obj->programs[i]; | |
6514 | if (prog_is_subprog(obj, prog)) | |
6515 | continue; | |
a3820c48 | 6516 | if (!prog->autoload) |
16e0c35c | 6517 | continue; |
b1268826 AS |
6518 | err = bpf_object__relocate_data(obj, prog); |
6519 | if (err) { | |
6520 | pr_warn("prog '%s': failed to relocate data references: %d\n", | |
6521 | prog->name, err); | |
6522 | return err; | |
6523 | } | |
6524 | } | |
185cfe83 | 6525 | |
8a47a6c5 WN |
6526 | return 0; |
6527 | } | |
6528 | ||
646f02ff | 6529 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 6530 | Elf64_Shdr *shdr, Elf_Data *data); |
646f02ff AN |
6531 | |
6532 | static int bpf_object__collect_map_relos(struct bpf_object *obj, | |
ad23b723 | 6533 | Elf64_Shdr *shdr, Elf_Data *data) |
646f02ff | 6534 | { |
15728ad3 AN |
6535 | const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *); |
6536 | int i, j, nrels, new_sz; | |
063e6881 | 6537 | const struct btf_var_secinfo *vi = NULL; |
646f02ff | 6538 | const struct btf_type *sec, *var, *def; |
341ac5ff HC |
6539 | struct bpf_map *map = NULL, *targ_map = NULL; |
6540 | struct bpf_program *targ_prog = NULL; | |
6541 | bool is_prog_array, is_map_in_map; | |
646f02ff | 6542 | const struct btf_member *member; |
341ac5ff | 6543 | const char *name, *mname, *type; |
646f02ff | 6544 | unsigned int moff; |
ad23b723 AN |
6545 | Elf64_Sym *sym; |
6546 | Elf64_Rel *rel; | |
646f02ff AN |
6547 | void *tmp; |
6548 | ||
6549 | if (!obj->efile.btf_maps_sec_btf_id || !obj->btf) | |
6550 | return -EINVAL; | |
6551 | sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id); | |
6552 | if (!sec) | |
6553 | return -EINVAL; | |
6554 | ||
646f02ff AN |
6555 | nrels = shdr->sh_size / shdr->sh_entsize; |
6556 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
6557 | rel = elf_rel_by_idx(data, i); |
6558 | if (!rel) { | |
646f02ff AN |
6559 | pr_warn(".maps relo #%d: failed to get ELF relo\n", i); |
6560 | return -LIBBPF_ERRNO__FORMAT; | |
6561 | } | |
ad23b723 AN |
6562 | |
6563 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); | |
6564 | if (!sym) { | |
646f02ff | 6565 | pr_warn(".maps relo #%d: symbol %zx not found\n", |
ad23b723 | 6566 | i, (size_t)ELF64_R_SYM(rel->r_info)); |
646f02ff AN |
6567 | return -LIBBPF_ERRNO__FORMAT; |
6568 | } | |
ad23b723 | 6569 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
646f02ff | 6570 | |
ad23b723 AN |
6571 | pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n", |
6572 | i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value, | |
6573 | (size_t)rel->r_offset, sym->st_name, name); | |
646f02ff AN |
6574 | |
6575 | for (j = 0; j < obj->nr_maps; j++) { | |
6576 | map = &obj->maps[j]; | |
6577 | if (map->sec_idx != obj->efile.btf_maps_shndx) | |
6578 | continue; | |
6579 | ||
6580 | vi = btf_var_secinfos(sec) + map->btf_var_idx; | |
ad23b723 AN |
6581 | if (vi->offset <= rel->r_offset && |
6582 | rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size) | |
646f02ff AN |
6583 | break; |
6584 | } | |
6585 | if (j == obj->nr_maps) { | |
ad23b723 AN |
6586 | pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n", |
6587 | i, name, (size_t)rel->r_offset); | |
646f02ff AN |
6588 | return -EINVAL; |
6589 | } | |
6590 | ||
341ac5ff HC |
6591 | is_map_in_map = bpf_map_type__is_map_in_map(map->def.type); |
6592 | is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY; | |
6593 | type = is_map_in_map ? "map" : "prog"; | |
6594 | if (is_map_in_map) { | |
6595 | if (sym->st_shndx != obj->efile.btf_maps_shndx) { | |
6596 | pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n", | |
6597 | i, name); | |
6598 | return -LIBBPF_ERRNO__RELOC; | |
6599 | } | |
6600 | if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS && | |
6601 | map->def.key_size != sizeof(int)) { | |
6602 | pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n", | |
6603 | i, map->name, sizeof(int)); | |
6604 | return -EINVAL; | |
6605 | } | |
6606 | targ_map = bpf_object__find_map_by_name(obj, name); | |
6607 | if (!targ_map) { | |
6608 | pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n", | |
6609 | i, name); | |
6610 | return -ESRCH; | |
6611 | } | |
6612 | } else if (is_prog_array) { | |
6613 | targ_prog = bpf_object__find_program_by_name(obj, name); | |
6614 | if (!targ_prog) { | |
6615 | pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n", | |
6616 | i, name); | |
6617 | return -ESRCH; | |
6618 | } | |
6619 | if (targ_prog->sec_idx != sym->st_shndx || | |
6620 | targ_prog->sec_insn_off * 8 != sym->st_value || | |
6621 | prog_is_subprog(obj, targ_prog)) { | |
6622 | pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n", | |
6623 | i, name); | |
6624 | return -LIBBPF_ERRNO__RELOC; | |
6625 | } | |
6626 | } else { | |
646f02ff AN |
6627 | return -EINVAL; |
6628 | } | |
6629 | ||
646f02ff AN |
6630 | var = btf__type_by_id(obj->btf, vi->type); |
6631 | def = skip_mods_and_typedefs(obj->btf, var->type, NULL); | |
6632 | if (btf_vlen(def) == 0) | |
6633 | return -EINVAL; | |
6634 | member = btf_members(def) + btf_vlen(def) - 1; | |
6635 | mname = btf__name_by_offset(obj->btf, member->name_off); | |
6636 | if (strcmp(mname, "values")) | |
6637 | return -EINVAL; | |
6638 | ||
6639 | moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8; | |
ad23b723 | 6640 | if (rel->r_offset - vi->offset < moff) |
646f02ff AN |
6641 | return -EINVAL; |
6642 | ||
ad23b723 | 6643 | moff = rel->r_offset - vi->offset - moff; |
15728ad3 AN |
6644 | /* here we use BPF pointer size, which is always 64 bit, as we |
6645 | * are parsing ELF that was built for BPF target | |
6646 | */ | |
6647 | if (moff % bpf_ptr_sz) | |
646f02ff | 6648 | return -EINVAL; |
15728ad3 | 6649 | moff /= bpf_ptr_sz; |
646f02ff AN |
6650 | if (moff >= map->init_slots_sz) { |
6651 | new_sz = moff + 1; | |
029258d7 | 6652 | tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz); |
646f02ff AN |
6653 | if (!tmp) |
6654 | return -ENOMEM; | |
6655 | map->init_slots = tmp; | |
6656 | memset(map->init_slots + map->init_slots_sz, 0, | |
15728ad3 | 6657 | (new_sz - map->init_slots_sz) * host_ptr_sz); |
646f02ff AN |
6658 | map->init_slots_sz = new_sz; |
6659 | } | |
341ac5ff | 6660 | map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog; |
646f02ff | 6661 | |
341ac5ff HC |
6662 | pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n", |
6663 | i, map->name, moff, type, name); | |
646f02ff AN |
6664 | } |
6665 | ||
6666 | return 0; | |
6667 | } | |
590a0088 | 6668 | |
c3c55696 AN |
6669 | static int bpf_object__collect_relos(struct bpf_object *obj) |
6670 | { | |
6671 | int i, err; | |
34090915 | 6672 | |
25bbbd7a AN |
6673 | for (i = 0; i < obj->efile.sec_cnt; i++) { |
6674 | struct elf_sec_desc *sec_desc = &obj->efile.secs[i]; | |
6675 | Elf64_Shdr *shdr; | |
6676 | Elf_Data *data; | |
6677 | int idx; | |
6678 | ||
6679 | if (sec_desc->sec_type != SEC_RELO) | |
6680 | continue; | |
6681 | ||
6682 | shdr = sec_desc->shdr; | |
6683 | data = sec_desc->data; | |
6684 | idx = shdr->sh_info; | |
34090915 WN |
6685 | |
6686 | if (shdr->sh_type != SHT_REL) { | |
be18010e | 6687 | pr_warn("internal error at %d\n", __LINE__); |
6371ca3b | 6688 | return -LIBBPF_ERRNO__INTERNAL; |
34090915 WN |
6689 | } |
6690 | ||
809a69d6 | 6691 | if (idx == obj->efile.st_ops_shndx || idx == obj->efile.st_ops_link_shndx) |
646f02ff | 6692 | err = bpf_object__collect_st_ops_relos(obj, shdr, data); |
c3c55696 | 6693 | else if (idx == obj->efile.btf_maps_shndx) |
646f02ff | 6694 | err = bpf_object__collect_map_relos(obj, shdr, data); |
c3c55696 AN |
6695 | else |
6696 | err = bpf_object__collect_prog_relos(obj, shdr, data); | |
34090915 | 6697 | if (err) |
6371ca3b | 6698 | return err; |
34090915 | 6699 | } |
c3c55696 | 6700 | |
d0e92887 | 6701 | bpf_object__sort_relos(obj); |
34090915 WN |
6702 | return 0; |
6703 | } | |
6704 | ||
109cea5a AN |
6705 | static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id) |
6706 | { | |
9b2f6fec | 6707 | if (BPF_CLASS(insn->code) == BPF_JMP && |
109cea5a AN |
6708 | BPF_OP(insn->code) == BPF_CALL && |
6709 | BPF_SRC(insn->code) == BPF_K && | |
9b2f6fec AN |
6710 | insn->src_reg == 0 && |
6711 | insn->dst_reg == 0) { | |
6712 | *func_id = insn->imm; | |
109cea5a AN |
6713 | return true; |
6714 | } | |
6715 | return false; | |
6716 | } | |
6717 | ||
42869d28 | 6718 | static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog) |
109cea5a AN |
6719 | { |
6720 | struct bpf_insn *insn = prog->insns; | |
6721 | enum bpf_func_id func_id; | |
6722 | int i; | |
6723 | ||
67234743 AS |
6724 | if (obj->gen_loader) |
6725 | return 0; | |
6726 | ||
109cea5a AN |
6727 | for (i = 0; i < prog->insns_cnt; i++, insn++) { |
6728 | if (!insn_is_helper_call(insn, &func_id)) | |
6729 | continue; | |
6730 | ||
6731 | /* on kernels that don't yet support | |
6732 | * bpf_probe_read_{kernel,user}[_str] helpers, fall back | |
6733 | * to bpf_probe_read() which works well for old kernels | |
6734 | */ | |
6735 | switch (func_id) { | |
6736 | case BPF_FUNC_probe_read_kernel: | |
6737 | case BPF_FUNC_probe_read_user: | |
9ca1f56a | 6738 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
6739 | insn->imm = BPF_FUNC_probe_read; |
6740 | break; | |
6741 | case BPF_FUNC_probe_read_kernel_str: | |
6742 | case BPF_FUNC_probe_read_user_str: | |
9ca1f56a | 6743 | if (!kernel_supports(obj, FEAT_PROBE_READ_KERN)) |
109cea5a AN |
6744 | insn->imm = BPF_FUNC_probe_read_str; |
6745 | break; | |
6746 | default: | |
6747 | break; | |
6748 | } | |
6749 | } | |
6750 | return 0; | |
6751 | } | |
6752 | ||
15ea31fa AN |
6753 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
6754 | int *btf_obj_fd, int *btf_type_id); | |
12d9466d | 6755 | |
4fa5bcfe AN |
6756 | /* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */ |
6757 | static int libbpf_prepare_prog_load(struct bpf_program *prog, | |
6758 | struct bpf_prog_load_opts *opts, long cookie) | |
12d9466d | 6759 | { |
15ea31fa AN |
6760 | enum sec_def_flags def = cookie; |
6761 | ||
12d9466d | 6762 | /* old kernels might not support specifying expected_attach_type */ |
15ea31fa | 6763 | if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE)) |
d10ef2b8 | 6764 | opts->expected_attach_type = 0; |
12d9466d | 6765 | |
15ea31fa | 6766 | if (def & SEC_SLEEPABLE) |
d10ef2b8 | 6767 | opts->prog_flags |= BPF_F_SLEEPABLE; |
12d9466d | 6768 | |
082c4bfb LB |
6769 | if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS)) |
6770 | opts->prog_flags |= BPF_F_XDP_HAS_FRAGS; | |
6771 | ||
cc7d8f2c | 6772 | if ((def & SEC_ATTACH_BTF) && !prog->attach_btf_id) { |
12d9466d | 6773 | int btf_obj_fd = 0, btf_type_id = 0, err; |
15ea31fa | 6774 | const char *attach_name; |
12d9466d | 6775 | |
cc7d8f2c AN |
6776 | attach_name = strchr(prog->sec_name, '/'); |
6777 | if (!attach_name) { | |
6778 | /* if BPF program is annotated with just SEC("fentry") | |
6779 | * (or similar) without declaratively specifying | |
6780 | * target, then it is expected that target will be | |
6781 | * specified with bpf_program__set_attach_target() at | |
6782 | * runtime before BPF object load step. If not, then | |
6783 | * there is nothing to load into the kernel as BPF | |
6784 | * verifier won't be able to validate BPF program | |
6785 | * correctness anyways. | |
6786 | */ | |
6787 | pr_warn("prog '%s': no BTF-based attach target is specified, use bpf_program__set_attach_target()\n", | |
6788 | prog->name); | |
6789 | return -EINVAL; | |
6790 | } | |
6791 | attach_name++; /* skip over / */ | |
6792 | ||
15ea31fa | 6793 | err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id); |
12d9466d AN |
6794 | if (err) |
6795 | return err; | |
6796 | ||
6797 | /* cache resolved BTF FD and BTF type ID in the prog */ | |
6798 | prog->attach_btf_obj_fd = btf_obj_fd; | |
6799 | prog->attach_btf_id = btf_type_id; | |
6800 | ||
6801 | /* but by now libbpf common logic is not utilizing | |
6802 | * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because | |
d10ef2b8 AN |
6803 | * this callback is called after opts were populated by |
6804 | * libbpf, so this callback has to update opts explicitly here | |
12d9466d | 6805 | */ |
d10ef2b8 AN |
6806 | opts->attach_btf_obj_fd = btf_obj_fd; |
6807 | opts->attach_btf_id = btf_type_id; | |
12d9466d AN |
6808 | } |
6809 | return 0; | |
6810 | } | |
6811 | ||
9fdc4273 AN |
6812 | static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz); |
6813 | ||
cf90a20d AN |
6814 | static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog, |
6815 | struct bpf_insn *insns, int insns_cnt, | |
6816 | const char *license, __u32 kern_version, int *prog_fd) | |
55cffde2 | 6817 | { |
d10ef2b8 AN |
6818 | LIBBPF_OPTS(bpf_prog_load_opts, load_attr); |
6819 | const char *prog_name = NULL; | |
1ce6a9fc | 6820 | char *cp, errmsg[STRERR_BUFSIZE]; |
8395f320 | 6821 | size_t log_buf_size = 0; |
b3ce9079 | 6822 | char *log_buf = NULL, *tmp; |
12d9466d | 6823 | int btf_fd, ret, err; |
b3ce9079 AN |
6824 | bool own_log_buf = true; |
6825 | __u32 log_level = prog->log_level; | |
55cffde2 | 6826 | |
80b2b5c3 AM |
6827 | if (prog->type == BPF_PROG_TYPE_UNSPEC) { |
6828 | /* | |
6829 | * The program type must be set. Most likely we couldn't find a proper | |
6830 | * section definition at load time, and thus we didn't infer the type. | |
6831 | */ | |
6832 | pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n", | |
6833 | prog->name, prog->sec_name); | |
6834 | return -EINVAL; | |
6835 | } | |
6836 | ||
fba01a06 AN |
6837 | if (!insns || !insns_cnt) |
6838 | return -EINVAL; | |
6839 | ||
12d9466d | 6840 | load_attr.expected_attach_type = prog->expected_attach_type; |
25bbbd7a | 6841 | if (kernel_supports(obj, FEAT_PROG_NAME)) |
d10ef2b8 | 6842 | prog_name = prog->name; |
12d9466d AN |
6843 | load_attr.attach_prog_fd = prog->attach_prog_fd; |
6844 | load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd; | |
6aef10a4 AN |
6845 | load_attr.attach_btf_id = prog->attach_btf_id; |
6846 | load_attr.kern_version = kern_version; | |
6847 | load_attr.prog_ifindex = prog->prog_ifindex; | |
6848 | ||
0f0e55d8 | 6849 | /* specify func_info/line_info only if kernel supports them */ |
25bbbd7a AN |
6850 | btf_fd = bpf_object__btf_fd(obj); |
6851 | if (btf_fd >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) { | |
0f0e55d8 AN |
6852 | load_attr.prog_btf_fd = btf_fd; |
6853 | load_attr.func_info = prog->func_info; | |
6854 | load_attr.func_info_rec_size = prog->func_info_rec_size; | |
6855 | load_attr.func_info_cnt = prog->func_info_cnt; | |
6856 | load_attr.line_info = prog->line_info; | |
6857 | load_attr.line_info_rec_size = prog->line_info_rec_size; | |
6858 | load_attr.line_info_cnt = prog->line_info_cnt; | |
6859 | } | |
b3ce9079 | 6860 | load_attr.log_level = log_level; |
04656198 | 6861 | load_attr.prog_flags = prog->prog_flags; |
25bbbd7a | 6862 | load_attr.fd_array = obj->fd_array; |
55cffde2 | 6863 | |
12d9466d | 6864 | /* adjust load_attr if sec_def provides custom preload callback */ |
4fa5bcfe AN |
6865 | if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) { |
6866 | err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie); | |
12d9466d AN |
6867 | if (err < 0) { |
6868 | pr_warn("prog '%s': failed to prepare load attributes: %d\n", | |
6869 | prog->name, err); | |
6870 | return err; | |
6871 | } | |
b63b3c49 JO |
6872 | insns = prog->insns; |
6873 | insns_cnt = prog->insns_cnt; | |
12d9466d AN |
6874 | } |
6875 | ||
25bbbd7a | 6876 | if (obj->gen_loader) { |
d10ef2b8 AN |
6877 | bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name, |
6878 | license, insns, insns_cnt, &load_attr, | |
25bbbd7a | 6879 | prog - obj->programs); |
be2f2d16 | 6880 | *prog_fd = -1; |
67234743 AS |
6881 | return 0; |
6882 | } | |
8395f320 | 6883 | |
b3ce9079 | 6884 | retry_load: |
bb412cf1 | 6885 | /* if log_level is zero, we don't request logs initially even if |
b3ce9079 AN |
6886 | * custom log_buf is specified; if the program load fails, then we'll |
6887 | * bump log_level to 1 and use either custom log_buf or we'll allocate | |
6888 | * our own and retry the load to get details on what failed | |
6889 | */ | |
6890 | if (log_level) { | |
6891 | if (prog->log_buf) { | |
6892 | log_buf = prog->log_buf; | |
6893 | log_buf_size = prog->log_size; | |
6894 | own_log_buf = false; | |
6895 | } else if (obj->log_buf) { | |
6896 | log_buf = obj->log_buf; | |
6897 | log_buf_size = obj->log_size; | |
6898 | own_log_buf = false; | |
6899 | } else { | |
6900 | log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2); | |
6901 | tmp = realloc(log_buf, log_buf_size); | |
6902 | if (!tmp) { | |
6903 | ret = -ENOMEM; | |
6904 | goto out; | |
6905 | } | |
6906 | log_buf = tmp; | |
6907 | log_buf[0] = '\0'; | |
6908 | own_log_buf = true; | |
6909 | } | |
8395f320 | 6910 | } |
55cffde2 | 6911 | |
6aef10a4 | 6912 | load_attr.log_buf = log_buf; |
d10ef2b8 | 6913 | load_attr.log_size = log_buf_size; |
b3ce9079 | 6914 | load_attr.log_level = log_level; |
55cffde2 | 6915 | |
b3ce9079 | 6916 | ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr); |
55cffde2 | 6917 | if (ret >= 0) { |
b3ce9079 | 6918 | if (log_level && own_log_buf) { |
ad9a7f96 AN |
6919 | pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n", |
6920 | prog->name, log_buf); | |
6921 | } | |
5d23328d | 6922 | |
25bbbd7a AN |
6923 | if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) { |
6924 | struct bpf_map *map; | |
6925 | int i; | |
6926 | ||
6927 | for (i = 0; i < obj->nr_maps; i++) { | |
6928 | map = &prog->obj->maps[i]; | |
6929 | if (map->libbpf_type != LIBBPF_MAP_RODATA) | |
6930 | continue; | |
5d23328d | 6931 | |
25bbbd7a AN |
6932 | if (bpf_prog_bind_map(ret, bpf_map__fd(map), NULL)) { |
6933 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); | |
ad9a7f96 AN |
6934 | pr_warn("prog '%s': failed to bind map '%s': %s\n", |
6935 | prog->name, map->real_name, cp); | |
25bbbd7a AN |
6936 | /* Don't fail hard if can't bind rodata. */ |
6937 | } | |
5d23328d YZ |
6938 | } |
6939 | } | |
6940 | ||
be2f2d16 | 6941 | *prog_fd = ret; |
55cffde2 WN |
6942 | ret = 0; |
6943 | goto out; | |
6944 | } | |
6945 | ||
b3ce9079 AN |
6946 | if (log_level == 0) { |
6947 | log_level = 1; | |
da11b417 AS |
6948 | goto retry_load; |
6949 | } | |
b3ce9079 AN |
6950 | /* On ENOSPC, increase log buffer size and retry, unless custom |
6951 | * log_buf is specified. | |
6952 | * Be careful to not overflow u32, though. Kernel's log buf size limit | |
6953 | * isn't part of UAPI so it can always be bumped to full 4GB. So don't | |
6954 | * multiply by 2 unless we are sure we'll fit within 32 bits. | |
6955 | * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2). | |
6956 | */ | |
6957 | if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2) | |
6958 | goto retry_load; | |
2eda2145 AN |
6959 | |
6960 | ret = -errno; | |
9fdc4273 AN |
6961 | |
6962 | /* post-process verifier log to improve error descriptions */ | |
6963 | fixup_verifier_log(prog, log_buf, log_buf_size); | |
6964 | ||
24d6a808 | 6965 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
ad9a7f96 | 6966 | pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, cp); |
dc3a2d25 | 6967 | pr_perm_msg(ret); |
55cffde2 | 6968 | |
b3ce9079 | 6969 | if (own_log_buf && log_buf && log_buf[0] != '\0') { |
ad9a7f96 AN |
6970 | pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n", |
6971 | prog->name, log_buf); | |
6972 | } | |
55cffde2 WN |
6973 | |
6974 | out: | |
b3ce9079 AN |
6975 | if (own_log_buf) |
6976 | free(log_buf); | |
55cffde2 WN |
6977 | return ret; |
6978 | } | |
6979 | ||
9fdc4273 AN |
6980 | static char *find_prev_line(char *buf, char *cur) |
6981 | { | |
6982 | char *p; | |
6983 | ||
6984 | if (cur == buf) /* end of a log buf */ | |
6985 | return NULL; | |
6986 | ||
6987 | p = cur - 1; | |
6988 | while (p - 1 >= buf && *(p - 1) != '\n') | |
6989 | p--; | |
6990 | ||
6991 | return p; | |
6992 | } | |
6993 | ||
6994 | static void patch_log(char *buf, size_t buf_sz, size_t log_sz, | |
6995 | char *orig, size_t orig_sz, const char *patch) | |
6996 | { | |
6997 | /* size of the remaining log content to the right from the to-be-replaced part */ | |
6998 | size_t rem_sz = (buf + log_sz) - (orig + orig_sz); | |
6999 | size_t patch_sz = strlen(patch); | |
7000 | ||
7001 | if (patch_sz != orig_sz) { | |
7002 | /* If patch line(s) are longer than original piece of verifier log, | |
7003 | * shift log contents by (patch_sz - orig_sz) bytes to the right | |
7004 | * starting from after to-be-replaced part of the log. | |
7005 | * | |
7006 | * If patch line(s) are shorter than original piece of verifier log, | |
7007 | * shift log contents by (orig_sz - patch_sz) bytes to the left | |
7008 | * starting from after to-be-replaced part of the log | |
7009 | * | |
7010 | * We need to be careful about not overflowing available | |
7011 | * buf_sz capacity. If that's the case, we'll truncate the end | |
7012 | * of the original log, as necessary. | |
7013 | */ | |
7014 | if (patch_sz > orig_sz) { | |
7015 | if (orig + patch_sz >= buf + buf_sz) { | |
7016 | /* patch is big enough to cover remaining space completely */ | |
7017 | patch_sz -= (orig + patch_sz) - (buf + buf_sz) + 1; | |
7018 | rem_sz = 0; | |
7019 | } else if (patch_sz - orig_sz > buf_sz - log_sz) { | |
7020 | /* patch causes part of remaining log to be truncated */ | |
7021 | rem_sz -= (patch_sz - orig_sz) - (buf_sz - log_sz); | |
7022 | } | |
7023 | } | |
7024 | /* shift remaining log to the right by calculated amount */ | |
7025 | memmove(orig + patch_sz, orig + orig_sz, rem_sz); | |
7026 | } | |
7027 | ||
7028 | memcpy(orig, patch, patch_sz); | |
7029 | } | |
7030 | ||
7031 | static void fixup_log_failed_core_relo(struct bpf_program *prog, | |
7032 | char *buf, size_t buf_sz, size_t log_sz, | |
7033 | char *line1, char *line2, char *line3) | |
7034 | { | |
7035 | /* Expected log for failed and not properly guarded CO-RE relocation: | |
7036 | * line1 -> 123: (85) call unknown#195896080 | |
7037 | * line2 -> invalid func unknown#195896080 | |
7038 | * line3 -> <anything else or end of buffer> | |
7039 | * | |
7040 | * "123" is the index of the instruction that was poisoned. We extract | |
7041 | * instruction index to find corresponding CO-RE relocation and | |
7042 | * replace this part of the log with more relevant information about | |
7043 | * failed CO-RE relocation. | |
7044 | */ | |
7045 | const struct bpf_core_relo *relo; | |
7046 | struct bpf_core_spec spec; | |
7047 | char patch[512], spec_buf[256]; | |
b198881d | 7048 | int insn_idx, err, spec_len; |
9fdc4273 AN |
7049 | |
7050 | if (sscanf(line1, "%d: (%*d) call unknown#195896080\n", &insn_idx) != 1) | |
7051 | return; | |
7052 | ||
7053 | relo = find_relo_core(prog, insn_idx); | |
7054 | if (!relo) | |
7055 | return; | |
7056 | ||
7057 | err = bpf_core_parse_spec(prog->name, prog->obj->btf, relo, &spec); | |
7058 | if (err) | |
7059 | return; | |
7060 | ||
b198881d | 7061 | spec_len = bpf_core_format_spec(spec_buf, sizeof(spec_buf), &spec); |
9fdc4273 AN |
7062 | snprintf(patch, sizeof(patch), |
7063 | "%d: <invalid CO-RE relocation>\n" | |
b198881d AN |
7064 | "failed to resolve CO-RE relocation %s%s\n", |
7065 | insn_idx, spec_buf, spec_len >= sizeof(spec_buf) ? "..." : ""); | |
9fdc4273 AN |
7066 | |
7067 | patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch); | |
7068 | } | |
7069 | ||
ec41817b AN |
7070 | static void fixup_log_missing_map_load(struct bpf_program *prog, |
7071 | char *buf, size_t buf_sz, size_t log_sz, | |
7072 | char *line1, char *line2, char *line3) | |
7073 | { | |
3055ddd6 | 7074 | /* Expected log for failed and not properly guarded map reference: |
ec41817b AN |
7075 | * line1 -> 123: (85) call unknown#2001000345 |
7076 | * line2 -> invalid func unknown#2001000345 | |
7077 | * line3 -> <anything else or end of buffer> | |
7078 | * | |
7079 | * "123" is the index of the instruction that was poisoned. | |
3055ddd6 | 7080 | * "345" in "2001000345" is a map index in obj->maps to fetch map name. |
ec41817b AN |
7081 | */ |
7082 | struct bpf_object *obj = prog->obj; | |
7083 | const struct bpf_map *map; | |
7084 | int insn_idx, map_idx; | |
7085 | char patch[128]; | |
7086 | ||
7087 | if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &map_idx) != 2) | |
7088 | return; | |
7089 | ||
3055ddd6 | 7090 | map_idx -= POISON_LDIMM64_MAP_BASE; |
ec41817b AN |
7091 | if (map_idx < 0 || map_idx >= obj->nr_maps) |
7092 | return; | |
7093 | map = &obj->maps[map_idx]; | |
7094 | ||
7095 | snprintf(patch, sizeof(patch), | |
7096 | "%d: <invalid BPF map reference>\n" | |
7097 | "BPF map '%s' is referenced but wasn't created\n", | |
7098 | insn_idx, map->name); | |
7099 | ||
7100 | patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch); | |
7101 | } | |
7102 | ||
05b6f766 AN |
7103 | static void fixup_log_missing_kfunc_call(struct bpf_program *prog, |
7104 | char *buf, size_t buf_sz, size_t log_sz, | |
7105 | char *line1, char *line2, char *line3) | |
7106 | { | |
7107 | /* Expected log for failed and not properly guarded kfunc call: | |
7108 | * line1 -> 123: (85) call unknown#2002000345 | |
7109 | * line2 -> invalid func unknown#2002000345 | |
7110 | * line3 -> <anything else or end of buffer> | |
7111 | * | |
7112 | * "123" is the index of the instruction that was poisoned. | |
7113 | * "345" in "2002000345" is an extern index in obj->externs to fetch kfunc name. | |
7114 | */ | |
7115 | struct bpf_object *obj = prog->obj; | |
7116 | const struct extern_desc *ext; | |
7117 | int insn_idx, ext_idx; | |
7118 | char patch[128]; | |
7119 | ||
7120 | if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &ext_idx) != 2) | |
7121 | return; | |
7122 | ||
7123 | ext_idx -= POISON_CALL_KFUNC_BASE; | |
7124 | if (ext_idx < 0 || ext_idx >= obj->nr_extern) | |
7125 | return; | |
7126 | ext = &obj->externs[ext_idx]; | |
7127 | ||
7128 | snprintf(patch, sizeof(patch), | |
7129 | "%d: <invalid kfunc call>\n" | |
7130 | "kfunc '%s' is referenced but wasn't resolved\n", | |
7131 | insn_idx, ext->name); | |
7132 | ||
7133 | patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch); | |
7134 | } | |
7135 | ||
9fdc4273 AN |
7136 | static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz) |
7137 | { | |
7138 | /* look for familiar error patterns in last N lines of the log */ | |
7139 | const size_t max_last_line_cnt = 10; | |
7140 | char *prev_line, *cur_line, *next_line; | |
7141 | size_t log_sz; | |
7142 | int i; | |
7143 | ||
7144 | if (!buf) | |
7145 | return; | |
7146 | ||
7147 | log_sz = strlen(buf) + 1; | |
7148 | next_line = buf + log_sz - 1; | |
7149 | ||
7150 | for (i = 0; i < max_last_line_cnt; i++, next_line = cur_line) { | |
7151 | cur_line = find_prev_line(buf, next_line); | |
7152 | if (!cur_line) | |
7153 | return; | |
7154 | ||
9fdc4273 AN |
7155 | if (str_has_pfx(cur_line, "invalid func unknown#195896080\n")) { |
7156 | prev_line = find_prev_line(buf, cur_line); | |
7157 | if (!prev_line) | |
7158 | continue; | |
7159 | ||
3055ddd6 | 7160 | /* failed CO-RE relocation case */ |
9fdc4273 AN |
7161 | fixup_log_failed_core_relo(prog, buf, buf_sz, log_sz, |
7162 | prev_line, cur_line, next_line); | |
7163 | return; | |
3055ddd6 | 7164 | } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_LDIMM64_MAP_PFX)) { |
ec41817b AN |
7165 | prev_line = find_prev_line(buf, cur_line); |
7166 | if (!prev_line) | |
7167 | continue; | |
7168 | ||
3055ddd6 | 7169 | /* reference to uncreated BPF map */ |
ec41817b AN |
7170 | fixup_log_missing_map_load(prog, buf, buf_sz, log_sz, |
7171 | prev_line, cur_line, next_line); | |
7172 | return; | |
05b6f766 AN |
7173 | } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_CALL_KFUNC_PFX)) { |
7174 | prev_line = find_prev_line(buf, cur_line); | |
7175 | if (!prev_line) | |
7176 | continue; | |
7177 | ||
7178 | /* reference to unresolved kfunc */ | |
7179 | fixup_log_missing_kfunc_call(prog, buf, buf_sz, log_sz, | |
7180 | prev_line, cur_line, next_line); | |
7181 | return; | |
9fdc4273 AN |
7182 | } |
7183 | } | |
7184 | } | |
7185 | ||
d0e92887 | 7186 | static int bpf_program_record_relos(struct bpf_program *prog) |
67234743 AS |
7187 | { |
7188 | struct bpf_object *obj = prog->obj; | |
7189 | int i; | |
7190 | ||
7191 | for (i = 0; i < prog->nr_reloc; i++) { | |
7192 | struct reloc_desc *relo = &prog->reloc_desc[i]; | |
3055ddd6 | 7193 | struct extern_desc *ext = &obj->externs[relo->ext_idx]; |
708cdc57 | 7194 | int kind; |
67234743 AS |
7195 | |
7196 | switch (relo->type) { | |
a18f7214 | 7197 | case RELO_EXTERN_LD64: |
67234743 AS |
7198 | if (ext->type != EXT_KSYM) |
7199 | continue; | |
708cdc57 AS |
7200 | kind = btf_is_var(btf__type_by_id(obj->btf, ext->btf_id)) ? |
7201 | BTF_KIND_VAR : BTF_KIND_FUNC; | |
c24941cd KKD |
7202 | bpf_gen__record_extern(obj->gen_loader, ext->name, |
7203 | ext->is_weak, !ext->ksym.type_id, | |
708cdc57 | 7204 | true, kind, relo->insn_idx); |
67234743 | 7205 | break; |
a18f7214 | 7206 | case RELO_EXTERN_CALL: |
c24941cd | 7207 | bpf_gen__record_extern(obj->gen_loader, ext->name, |
708cdc57 | 7208 | ext->is_weak, false, false, BTF_KIND_FUNC, |
67234743 AS |
7209 | relo->insn_idx); |
7210 | break; | |
d0e92887 AS |
7211 | case RELO_CORE: { |
7212 | struct bpf_core_relo cr = { | |
7213 | .insn_off = relo->insn_idx * 8, | |
7214 | .type_id = relo->core_relo->type_id, | |
7215 | .access_str_off = relo->core_relo->access_str_off, | |
7216 | .kind = relo->core_relo->kind, | |
7217 | }; | |
7218 | ||
7219 | bpf_gen__record_relo_core(obj->gen_loader, &cr); | |
7220 | break; | |
7221 | } | |
67234743 AS |
7222 | default: |
7223 | continue; | |
7224 | } | |
7225 | } | |
7226 | return 0; | |
7227 | } | |
7228 | ||
55cffde2 | 7229 | static int |
60276f98 | 7230 | bpf_object__load_progs(struct bpf_object *obj, int log_level) |
55cffde2 | 7231 | { |
d9297581 | 7232 | struct bpf_program *prog; |
55cffde2 WN |
7233 | size_t i; |
7234 | int err; | |
7235 | ||
109cea5a AN |
7236 | for (i = 0; i < obj->nr_programs; i++) { |
7237 | prog = &obj->programs[i]; | |
7238 | err = bpf_object__sanitize_prog(obj, prog); | |
7239 | if (err) | |
7240 | return err; | |
7241 | } | |
7242 | ||
55cffde2 | 7243 | for (i = 0; i < obj->nr_programs; i++) { |
d9297581 | 7244 | prog = &obj->programs[i]; |
c3c55696 | 7245 | if (prog_is_subprog(obj, prog)) |
48cca7e4 | 7246 | continue; |
a3820c48 | 7247 | if (!prog->autoload) { |
9c0f8cbd | 7248 | pr_debug("prog '%s': skipped loading\n", prog->name); |
d9297581 AN |
7249 | continue; |
7250 | } | |
7251 | prog->log_level |= log_level; | |
cf90a20d AN |
7252 | |
7253 | if (obj->gen_loader) | |
7254 | bpf_program_record_relos(prog); | |
7255 | ||
7256 | err = bpf_object_load_prog(obj, prog, prog->insns, prog->insns_cnt, | |
7257 | obj->license, obj->kern_version, &prog->fd); | |
7258 | if (err) { | |
7259 | pr_warn("prog '%s': failed to load: %d\n", prog->name, err); | |
55cffde2 | 7260 | return err; |
cf90a20d | 7261 | } |
55cffde2 | 7262 | } |
185cfe83 AN |
7263 | |
7264 | bpf_object__free_relocs(obj); | |
55cffde2 WN |
7265 | return 0; |
7266 | } | |
7267 | ||
25498a19 AN |
7268 | static const struct bpf_sec_def *find_sec_def(const char *sec_name); |
7269 | ||
91b4d1d1 AN |
7270 | static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts) |
7271 | { | |
7272 | struct bpf_program *prog; | |
12d9466d | 7273 | int err; |
91b4d1d1 AN |
7274 | |
7275 | bpf_object__for_each_program(prog, obj) { | |
7276 | prog->sec_def = find_sec_def(prog->sec_name); | |
7277 | if (!prog->sec_def) { | |
7278 | /* couldn't guess, but user might manually specify */ | |
7279 | pr_debug("prog '%s': unrecognized ELF section name '%s'\n", | |
7280 | prog->name, prog->sec_name); | |
7281 | continue; | |
7282 | } | |
7283 | ||
df286716 GS |
7284 | prog->type = prog->sec_def->prog_type; |
7285 | prog->expected_attach_type = prog->sec_def->expected_attach_type; | |
91b4d1d1 | 7286 | |
12d9466d AN |
7287 | /* sec_def can have custom callback which should be called |
7288 | * after bpf_program is initialized to adjust its properties | |
7289 | */ | |
4fa5bcfe AN |
7290 | if (prog->sec_def->prog_setup_fn) { |
7291 | err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie); | |
12d9466d AN |
7292 | if (err < 0) { |
7293 | pr_warn("prog '%s': failed to initialize: %d\n", | |
7294 | prog->name, err); | |
7295 | return err; | |
7296 | } | |
7297 | } | |
91b4d1d1 AN |
7298 | } |
7299 | ||
7300 | return 0; | |
7301 | } | |
7302 | ||
ad9a7f96 AN |
7303 | static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz, |
7304 | const struct bpf_object_open_opts *opts) | |
1a5e3fb1 | 7305 | { |
1373ff59 | 7306 | const char *obj_name, *kconfig, *btf_tmp_path; |
1a5e3fb1 | 7307 | struct bpf_object *obj; |
291ee02b | 7308 | char tmp_name[64]; |
6371ca3b | 7309 | int err; |
e0e3ea88 AN |
7310 | char *log_buf; |
7311 | size_t log_size; | |
7312 | __u32 log_level; | |
1a5e3fb1 WN |
7313 | |
7314 | if (elf_version(EV_CURRENT) == EV_NONE) { | |
be18010e KW |
7315 | pr_warn("failed to init libelf for %s\n", |
7316 | path ? : "(mem buf)"); | |
6371ca3b | 7317 | return ERR_PTR(-LIBBPF_ERRNO__LIBELF); |
1a5e3fb1 WN |
7318 | } |
7319 | ||
291ee02b AN |
7320 | if (!OPTS_VALID(opts, bpf_object_open_opts)) |
7321 | return ERR_PTR(-EINVAL); | |
7322 | ||
1aace10f | 7323 | obj_name = OPTS_GET(opts, object_name, NULL); |
291ee02b AN |
7324 | if (obj_buf) { |
7325 | if (!obj_name) { | |
7326 | snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx", | |
7327 | (unsigned long)obj_buf, | |
7328 | (unsigned long)obj_buf_sz); | |
7329 | obj_name = tmp_name; | |
7330 | } | |
7331 | path = obj_name; | |
7332 | pr_debug("loading object '%s' from buffer\n", obj_name); | |
7333 | } | |
7334 | ||
e0e3ea88 AN |
7335 | log_buf = OPTS_GET(opts, kernel_log_buf, NULL); |
7336 | log_size = OPTS_GET(opts, kernel_log_size, 0); | |
7337 | log_level = OPTS_GET(opts, kernel_log_level, 0); | |
7338 | if (log_size > UINT_MAX) | |
7339 | return ERR_PTR(-EINVAL); | |
7340 | if (log_size && !log_buf) | |
7341 | return ERR_PTR(-EINVAL); | |
7342 | ||
2ce8450e | 7343 | obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name); |
6371ca3b WN |
7344 | if (IS_ERR(obj)) |
7345 | return obj; | |
1a5e3fb1 | 7346 | |
e0e3ea88 AN |
7347 | obj->log_buf = log_buf; |
7348 | obj->log_size = log_size; | |
7349 | obj->log_level = log_level; | |
7350 | ||
1373ff59 SC |
7351 | btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL); |
7352 | if (btf_tmp_path) { | |
7353 | if (strlen(btf_tmp_path) >= PATH_MAX) { | |
7354 | err = -ENAMETOOLONG; | |
7355 | goto out; | |
7356 | } | |
7357 | obj->btf_custom_path = strdup(btf_tmp_path); | |
7358 | if (!obj->btf_custom_path) { | |
7359 | err = -ENOMEM; | |
7360 | goto out; | |
7361 | } | |
7362 | } | |
7363 | ||
8601fd42 AN |
7364 | kconfig = OPTS_GET(opts, kconfig, NULL); |
7365 | if (kconfig) { | |
7366 | obj->kconfig = strdup(kconfig); | |
18353c87 SC |
7367 | if (!obj->kconfig) { |
7368 | err = -ENOMEM; | |
7369 | goto out; | |
7370 | } | |
166750bc | 7371 | } |
291ee02b | 7372 | |
0d13bfce AN |
7373 | err = bpf_object__elf_init(obj); |
7374 | err = err ? : bpf_object__check_endianness(obj); | |
7375 | err = err ? : bpf_object__elf_collect(obj); | |
166750bc | 7376 | err = err ? : bpf_object__collect_externs(obj); |
f33f742d | 7377 | err = err ? : bpf_object_fixup_btf(obj); |
0d13bfce | 7378 | err = err ? : bpf_object__init_maps(obj, opts); |
91b4d1d1 | 7379 | err = err ? : bpf_object_init_progs(obj, opts); |
c3c55696 | 7380 | err = err ? : bpf_object__collect_relos(obj); |
0d13bfce AN |
7381 | if (err) |
7382 | goto out; | |
dd4436bb | 7383 | |
91b4d1d1 | 7384 | bpf_object__elf_finish(obj); |
dd4436bb | 7385 | |
1a5e3fb1 WN |
7386 | return obj; |
7387 | out: | |
7388 | bpf_object__close(obj); | |
6371ca3b | 7389 | return ERR_PTR(err); |
1a5e3fb1 WN |
7390 | } |
7391 | ||
2ce8450e | 7392 | struct bpf_object * |
01af3bf0 | 7393 | bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts) |
2ce8450e | 7394 | { |
2ce8450e | 7395 | if (!path) |
e9fc3ce9 | 7396 | return libbpf_err_ptr(-EINVAL); |
2ce8450e AN |
7397 | |
7398 | pr_debug("loading %s\n", path); | |
7399 | ||
ad9a7f96 | 7400 | return libbpf_ptr(bpf_object_open(path, NULL, 0, opts)); |
2ce8450e AN |
7401 | } |
7402 | ||
146bf811 AN |
7403 | struct bpf_object *bpf_object__open(const char *path) |
7404 | { | |
7405 | return bpf_object__open_file(path, NULL); | |
7406 | } | |
7407 | ||
2ce8450e AN |
7408 | struct bpf_object * |
7409 | bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz, | |
01af3bf0 | 7410 | const struct bpf_object_open_opts *opts) |
6c956392 | 7411 | { |
2ce8450e | 7412 | if (!obj_buf || obj_buf_sz == 0) |
e9fc3ce9 | 7413 | return libbpf_err_ptr(-EINVAL); |
6c956392 | 7414 | |
ad9a7f96 | 7415 | return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, opts)); |
2ce8450e AN |
7416 | } |
7417 | ||
4a404a7e | 7418 | static int bpf_object_unload(struct bpf_object *obj) |
52d3352e WN |
7419 | { |
7420 | size_t i; | |
7421 | ||
7422 | if (!obj) | |
e9fc3ce9 | 7423 | return libbpf_err(-EINVAL); |
52d3352e | 7424 | |
590a0088 | 7425 | for (i = 0; i < obj->nr_maps; i++) { |
9d759a9b | 7426 | zclose(obj->maps[i].fd); |
590a0088 MKL |
7427 | if (obj->maps[i].st_ops) |
7428 | zfree(&obj->maps[i].st_ops->kern_vdata); | |
7429 | } | |
52d3352e | 7430 | |
55cffde2 WN |
7431 | for (i = 0; i < obj->nr_programs; i++) |
7432 | bpf_program__unload(&obj->programs[i]); | |
7433 | ||
52d3352e WN |
7434 | return 0; |
7435 | } | |
7436 | ||
0d13bfce AN |
7437 | static int bpf_object__sanitize_maps(struct bpf_object *obj) |
7438 | { | |
7439 | struct bpf_map *m; | |
7440 | ||
7441 | bpf_object__for_each_map(m, obj) { | |
7442 | if (!bpf_map__is_internal(m)) | |
7443 | continue; | |
9ca1f56a | 7444 | if (!kernel_supports(obj, FEAT_ARRAY_MMAP)) |
6920b086 | 7445 | m->def.map_flags &= ~BPF_F_MMAPABLE; |
0d13bfce AN |
7446 | } |
7447 | ||
7448 | return 0; | |
7449 | } | |
7450 | ||
85153ac0 | 7451 | int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *ctx) |
1c0c7074 AN |
7452 | { |
7453 | char sym_type, sym_name[500]; | |
7454 | unsigned long long sym_addr; | |
1c0c7074 AN |
7455 | int ret, err = 0; |
7456 | FILE *f; | |
7457 | ||
59842c54 | 7458 | f = fopen("/proc/kallsyms", "re"); |
1c0c7074 AN |
7459 | if (!f) { |
7460 | err = -errno; | |
7461 | pr_warn("failed to open /proc/kallsyms: %d\n", err); | |
7462 | return err; | |
7463 | } | |
7464 | ||
7465 | while (true) { | |
7466 | ret = fscanf(f, "%llx %c %499s%*[^\n]\n", | |
7467 | &sym_addr, &sym_type, sym_name); | |
7468 | if (ret == EOF && feof(f)) | |
7469 | break; | |
7470 | if (ret != 3) { | |
135c783f | 7471 | pr_warn("failed to read kallsyms entry: %d\n", ret); |
1c0c7074 | 7472 | err = -EINVAL; |
85153ac0 | 7473 | break; |
1c0c7074 AN |
7474 | } |
7475 | ||
85153ac0 JO |
7476 | err = cb(sym_addr, sym_type, sym_name, ctx); |
7477 | if (err) | |
7478 | break; | |
1c0c7074 AN |
7479 | } |
7480 | ||
1c0c7074 AN |
7481 | fclose(f); |
7482 | return err; | |
7483 | } | |
7484 | ||
85153ac0 JO |
7485 | static int kallsyms_cb(unsigned long long sym_addr, char sym_type, |
7486 | const char *sym_name, void *ctx) | |
7487 | { | |
7488 | struct bpf_object *obj = ctx; | |
7489 | const struct btf_type *t; | |
7490 | struct extern_desc *ext; | |
7491 | ||
7492 | ext = find_extern_by_name(obj, sym_name); | |
7493 | if (!ext || ext->type != EXT_KSYM) | |
7494 | return 0; | |
7495 | ||
7496 | t = btf__type_by_id(obj->btf, ext->btf_id); | |
7497 | if (!btf_is_var(t)) | |
7498 | return 0; | |
7499 | ||
7500 | if (ext->is_set && ext->ksym.addr != sym_addr) { | |
55d00c37 | 7501 | pr_warn("extern (ksym) '%s': resolution is ambiguous: 0x%llx or 0x%llx\n", |
85153ac0 JO |
7502 | sym_name, ext->ksym.addr, sym_addr); |
7503 | return -EINVAL; | |
7504 | } | |
7505 | if (!ext->is_set) { | |
7506 | ext->is_set = true; | |
7507 | ext->ksym.addr = sym_addr; | |
55d00c37 | 7508 | pr_debug("extern (ksym) '%s': set to 0x%llx\n", sym_name, sym_addr); |
85153ac0 JO |
7509 | } |
7510 | return 0; | |
7511 | } | |
7512 | ||
7513 | static int bpf_object__read_kallsyms_file(struct bpf_object *obj) | |
7514 | { | |
7515 | return libbpf_kallsyms_parse(kallsyms_cb, obj); | |
7516 | } | |
7517 | ||
774e132e MKL |
7518 | static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name, |
7519 | __u16 kind, struct btf **res_btf, | |
9dbe6015 | 7520 | struct module_btf **res_mod_btf) |
d370bbe1 | 7521 | { |
9dbe6015 | 7522 | struct module_btf *mod_btf; |
284d2587 | 7523 | struct btf *btf; |
9dbe6015 | 7524 | int i, id, err; |
d370bbe1 | 7525 | |
933d1aa3 | 7526 | btf = obj->btf_vmlinux; |
9dbe6015 | 7527 | mod_btf = NULL; |
774e132e MKL |
7528 | id = btf__find_by_name_kind(btf, ksym_name, kind); |
7529 | ||
933d1aa3 MKL |
7530 | if (id == -ENOENT) { |
7531 | err = load_module_btfs(obj); | |
7532 | if (err) | |
7533 | return err; | |
d370bbe1 | 7534 | |
933d1aa3 | 7535 | for (i = 0; i < obj->btf_module_cnt; i++) { |
9dbe6015 KKD |
7536 | /* we assume module_btf's BTF FD is always >0 */ |
7537 | mod_btf = &obj->btf_modules[i]; | |
7538 | btf = mod_btf->btf; | |
7539 | id = btf__find_by_name_kind_own(btf, ksym_name, kind); | |
933d1aa3 MKL |
7540 | if (id != -ENOENT) |
7541 | break; | |
7542 | } | |
7543 | } | |
2211c825 | 7544 | if (id <= 0) |
933d1aa3 | 7545 | return -ESRCH; |
d370bbe1 | 7546 | |
774e132e | 7547 | *res_btf = btf; |
9dbe6015 | 7548 | *res_mod_btf = mod_btf; |
774e132e MKL |
7549 | return id; |
7550 | } | |
7551 | ||
7552 | static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj, | |
7553 | struct extern_desc *ext) | |
7554 | { | |
7555 | const struct btf_type *targ_var, *targ_type; | |
7556 | __u32 targ_type_id, local_type_id; | |
9dbe6015 | 7557 | struct module_btf *mod_btf = NULL; |
774e132e | 7558 | const char *targ_var_name; |
774e132e | 7559 | struct btf *btf = NULL; |
9dbe6015 | 7560 | int id, err; |
774e132e | 7561 | |
9dbe6015 | 7562 | id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf); |
466b2e13 KKD |
7563 | if (id < 0) { |
7564 | if (id == -ESRCH && ext->is_weak) | |
7565 | return 0; | |
2211c825 HL |
7566 | pr_warn("extern (var ksym) '%s': not found in kernel BTF\n", |
7567 | ext->name); | |
774e132e | 7568 | return id; |
2211c825 | 7569 | } |
774e132e | 7570 | |
933d1aa3 MKL |
7571 | /* find local type_id */ |
7572 | local_type_id = ext->ksym.type_id; | |
284d2587 | 7573 | |
933d1aa3 MKL |
7574 | /* find target type_id */ |
7575 | targ_var = btf__type_by_id(btf, id); | |
7576 | targ_var_name = btf__name_by_offset(btf, targ_var->name_off); | |
7577 | targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id); | |
d370bbe1 | 7578 | |
933d1aa3 MKL |
7579 | err = bpf_core_types_are_compat(obj->btf, local_type_id, |
7580 | btf, targ_type_id); | |
7581 | if (err <= 0) { | |
7582 | const struct btf_type *local_type; | |
7583 | const char *targ_name, *local_name; | |
d370bbe1 | 7584 | |
933d1aa3 MKL |
7585 | local_type = btf__type_by_id(obj->btf, local_type_id); |
7586 | local_name = btf__name_by_offset(obj->btf, local_type->name_off); | |
7587 | targ_name = btf__name_by_offset(btf, targ_type->name_off); | |
d370bbe1 | 7588 | |
933d1aa3 MKL |
7589 | pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n", |
7590 | ext->name, local_type_id, | |
7591 | btf_kind_str(local_type), local_name, targ_type_id, | |
7592 | btf_kind_str(targ_type), targ_name); | |
7593 | return -EINVAL; | |
7594 | } | |
d370bbe1 | 7595 | |
933d1aa3 | 7596 | ext->is_set = true; |
9dbe6015 | 7597 | ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0; |
933d1aa3 MKL |
7598 | ext->ksym.kernel_btf_id = id; |
7599 | pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n", | |
7600 | ext->name, id, btf_kind_str(targ_var), targ_var_name); | |
d370bbe1 | 7601 | |
933d1aa3 MKL |
7602 | return 0; |
7603 | } | |
d370bbe1 | 7604 | |
5bd022ec MKL |
7605 | static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj, |
7606 | struct extern_desc *ext) | |
7607 | { | |
7608 | int local_func_proto_id, kfunc_proto_id, kfunc_id; | |
9dbe6015 | 7609 | struct module_btf *mod_btf = NULL; |
5bd022ec MKL |
7610 | const struct btf_type *kern_func; |
7611 | struct btf *kern_btf = NULL; | |
9dbe6015 | 7612 | int ret; |
5bd022ec MKL |
7613 | |
7614 | local_func_proto_id = ext->ksym.type_id; | |
7615 | ||
9dbe6015 | 7616 | kfunc_id = find_ksym_btf_id(obj, ext->name, BTF_KIND_FUNC, &kern_btf, &mod_btf); |
5bd022ec | 7617 | if (kfunc_id < 0) { |
466b2e13 KKD |
7618 | if (kfunc_id == -ESRCH && ext->is_weak) |
7619 | return 0; | |
7620 | pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n", | |
5bd022ec MKL |
7621 | ext->name); |
7622 | return kfunc_id; | |
7623 | } | |
7624 | ||
5bd022ec MKL |
7625 | kern_func = btf__type_by_id(kern_btf, kfunc_id); |
7626 | kfunc_proto_id = kern_func->type; | |
7627 | ||
7628 | ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id, | |
7629 | kern_btf, kfunc_proto_id); | |
7630 | if (ret <= 0) { | |
f709160d AN |
7631 | pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with %s [%d]\n", |
7632 | ext->name, local_func_proto_id, | |
7633 | mod_btf ? mod_btf->name : "vmlinux", kfunc_proto_id); | |
5bd022ec MKL |
7634 | return -EINVAL; |
7635 | } | |
7636 | ||
9dbe6015 KKD |
7637 | /* set index for module BTF fd in fd_array, if unset */ |
7638 | if (mod_btf && !mod_btf->fd_array_idx) { | |
7639 | /* insn->off is s16 */ | |
7640 | if (obj->fd_array_cnt == INT16_MAX) { | |
7641 | pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n", | |
7642 | ext->name, mod_btf->fd_array_idx); | |
7643 | return -E2BIG; | |
7644 | } | |
7645 | /* Cannot use index 0 for module BTF fd */ | |
7646 | if (!obj->fd_array_cnt) | |
7647 | obj->fd_array_cnt = 1; | |
7648 | ||
7649 | ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int), | |
7650 | obj->fd_array_cnt + 1); | |
7651 | if (ret) | |
7652 | return ret; | |
7653 | mod_btf->fd_array_idx = obj->fd_array_cnt; | |
7654 | /* we assume module BTF FD is always >0 */ | |
7655 | obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd; | |
7656 | } | |
7657 | ||
5bd022ec | 7658 | ext->is_set = true; |
5bd022ec | 7659 | ext->ksym.kernel_btf_id = kfunc_id; |
9dbe6015 | 7660 | ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0; |
5fc13ad5 AS |
7661 | /* Also set kernel_btf_obj_fd to make sure that bpf_object__relocate_data() |
7662 | * populates FD into ld_imm64 insn when it's used to point to kfunc. | |
7663 | * {kernel_btf_id, btf_fd_idx} -> fixup bpf_call. | |
7664 | * {kernel_btf_id, kernel_btf_obj_fd} -> fixup ld_imm64. | |
7665 | */ | |
7666 | ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0; | |
f709160d AN |
7667 | pr_debug("extern (func ksym) '%s': resolved to %s [%d]\n", |
7668 | ext->name, mod_btf ? mod_btf->name : "vmlinux", kfunc_id); | |
5bd022ec MKL |
7669 | |
7670 | return 0; | |
7671 | } | |
7672 | ||
933d1aa3 MKL |
7673 | static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj) |
7674 | { | |
5bd022ec | 7675 | const struct btf_type *t; |
933d1aa3 MKL |
7676 | struct extern_desc *ext; |
7677 | int i, err; | |
7678 | ||
7679 | for (i = 0; i < obj->nr_extern; i++) { | |
7680 | ext = &obj->externs[i]; | |
7681 | if (ext->type != EXT_KSYM || !ext->ksym.type_id) | |
7682 | continue; | |
7683 | ||
67234743 AS |
7684 | if (obj->gen_loader) { |
7685 | ext->is_set = true; | |
7686 | ext->ksym.kernel_btf_obj_fd = 0; | |
7687 | ext->ksym.kernel_btf_id = 0; | |
7688 | continue; | |
7689 | } | |
5bd022ec MKL |
7690 | t = btf__type_by_id(obj->btf, ext->btf_id); |
7691 | if (btf_is_var(t)) | |
7692 | err = bpf_object__resolve_ksym_var_btf_id(obj, ext); | |
7693 | else | |
7694 | err = bpf_object__resolve_ksym_func_btf_id(obj, ext); | |
933d1aa3 MKL |
7695 | if (err) |
7696 | return err; | |
d370bbe1 HL |
7697 | } |
7698 | return 0; | |
7699 | } | |
7700 | ||
166750bc | 7701 | static int bpf_object__resolve_externs(struct bpf_object *obj, |
8601fd42 | 7702 | const char *extra_kconfig) |
166750bc | 7703 | { |
1c0c7074 | 7704 | bool need_config = false, need_kallsyms = false; |
d370bbe1 | 7705 | bool need_vmlinux_btf = false; |
166750bc | 7706 | struct extern_desc *ext; |
2e33efe3 | 7707 | void *kcfg_data = NULL; |
166750bc | 7708 | int err, i; |
166750bc AN |
7709 | |
7710 | if (obj->nr_extern == 0) | |
7711 | return 0; | |
7712 | ||
2e33efe3 AN |
7713 | if (obj->kconfig_map_idx >= 0) |
7714 | kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped; | |
166750bc AN |
7715 | |
7716 | for (i = 0; i < obj->nr_extern; i++) { | |
7717 | ext = &obj->externs[i]; | |
7718 | ||
55d00c37 AN |
7719 | if (ext->type == EXT_KSYM) { |
7720 | if (ext->ksym.type_id) | |
7721 | need_vmlinux_btf = true; | |
7722 | else | |
7723 | need_kallsyms = true; | |
7724 | continue; | |
7725 | } else if (ext->type == EXT_KCFG) { | |
7726 | void *ext_ptr = kcfg_data + ext->kcfg.data_off; | |
7727 | __u64 value = 0; | |
166750bc | 7728 | |
55d00c37 AN |
7729 | /* Kconfig externs need actual /proc/config.gz */ |
7730 | if (str_has_pfx(ext->name, "CONFIG_")) { | |
7731 | need_config = true; | |
7732 | continue; | |
7733 | } | |
7734 | ||
7735 | /* Virtual kcfg externs are customly handled by libbpf */ | |
7736 | if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) { | |
7737 | value = get_kernel_version(); | |
7738 | if (!value) { | |
7739 | pr_warn("extern (kcfg) '%s': failed to get kernel version\n", ext->name); | |
7740 | return -EINVAL; | |
7741 | } | |
7742 | } else if (strcmp(ext->name, "LINUX_HAS_BPF_COOKIE") == 0) { | |
7743 | value = kernel_supports(obj, FEAT_BPF_COOKIE); | |
6f5d467d AN |
7744 | } else if (strcmp(ext->name, "LINUX_HAS_SYSCALL_WRAPPER") == 0) { |
7745 | value = kernel_supports(obj, FEAT_SYSCALL_WRAPPER); | |
55d00c37 AN |
7746 | } else if (!str_has_pfx(ext->name, "LINUX_") || !ext->is_weak) { |
7747 | /* Currently libbpf supports only CONFIG_ and LINUX_ prefixed | |
7748 | * __kconfig externs, where LINUX_ ones are virtual and filled out | |
7749 | * customly by libbpf (their values don't come from Kconfig). | |
7750 | * If LINUX_xxx variable is not recognized by libbpf, but is marked | |
7751 | * __weak, it defaults to zero value, just like for CONFIG_xxx | |
7752 | * externs. | |
7753 | */ | |
7754 | pr_warn("extern (kcfg) '%s': unrecognized virtual extern\n", ext->name); | |
166750bc AN |
7755 | return -EINVAL; |
7756 | } | |
55d00c37 AN |
7757 | |
7758 | err = set_kcfg_value_num(ext, ext_ptr, value); | |
166750bc AN |
7759 | if (err) |
7760 | return err; | |
55d00c37 AN |
7761 | pr_debug("extern (kcfg) '%s': set to 0x%llx\n", |
7762 | ext->name, (long long)value); | |
166750bc | 7763 | } else { |
55d00c37 | 7764 | pr_warn("extern '%s': unrecognized extern kind\n", ext->name); |
166750bc AN |
7765 | return -EINVAL; |
7766 | } | |
7767 | } | |
8601fd42 | 7768 | if (need_config && extra_kconfig) { |
2e33efe3 | 7769 | err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data); |
8601fd42 AN |
7770 | if (err) |
7771 | return -EINVAL; | |
7772 | need_config = false; | |
7773 | for (i = 0; i < obj->nr_extern; i++) { | |
7774 | ext = &obj->externs[i]; | |
2e33efe3 | 7775 | if (ext->type == EXT_KCFG && !ext->is_set) { |
8601fd42 AN |
7776 | need_config = true; |
7777 | break; | |
7778 | } | |
7779 | } | |
7780 | } | |
166750bc | 7781 | if (need_config) { |
2e33efe3 | 7782 | err = bpf_object__read_kconfig_file(obj, kcfg_data); |
166750bc AN |
7783 | if (err) |
7784 | return -EINVAL; | |
7785 | } | |
1c0c7074 AN |
7786 | if (need_kallsyms) { |
7787 | err = bpf_object__read_kallsyms_file(obj); | |
7788 | if (err) | |
7789 | return -EINVAL; | |
7790 | } | |
d370bbe1 HL |
7791 | if (need_vmlinux_btf) { |
7792 | err = bpf_object__resolve_ksyms_btf_id(obj); | |
7793 | if (err) | |
7794 | return -EINVAL; | |
7795 | } | |
166750bc AN |
7796 | for (i = 0; i < obj->nr_extern; i++) { |
7797 | ext = &obj->externs[i]; | |
7798 | ||
7799 | if (!ext->is_set && !ext->is_weak) { | |
55d00c37 | 7800 | pr_warn("extern '%s' (strong): not resolved\n", ext->name); |
166750bc AN |
7801 | return -ESRCH; |
7802 | } else if (!ext->is_set) { | |
55d00c37 | 7803 | pr_debug("extern '%s' (weak): not resolved, defaulting to zero\n", |
166750bc AN |
7804 | ext->name); |
7805 | } | |
7806 | } | |
7807 | ||
7808 | return 0; | |
7809 | } | |
7810 | ||
8d1608d7 KFL |
7811 | static void bpf_map_prepare_vdata(const struct bpf_map *map) |
7812 | { | |
7813 | struct bpf_struct_ops *st_ops; | |
7814 | __u32 i; | |
7815 | ||
7816 | st_ops = map->st_ops; | |
7817 | for (i = 0; i < btf_vlen(st_ops->type); i++) { | |
7818 | struct bpf_program *prog = st_ops->progs[i]; | |
7819 | void *kern_data; | |
7820 | int prog_fd; | |
7821 | ||
7822 | if (!prog) | |
7823 | continue; | |
7824 | ||
7825 | prog_fd = bpf_program__fd(prog); | |
7826 | kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i]; | |
7827 | *(unsigned long *)kern_data = prog_fd; | |
7828 | } | |
7829 | } | |
7830 | ||
7831 | static int bpf_object_prepare_struct_ops(struct bpf_object *obj) | |
7832 | { | |
7833 | int i; | |
7834 | ||
7835 | for (i = 0; i < obj->nr_maps; i++) | |
7836 | if (bpf_map__is_struct_ops(&obj->maps[i])) | |
7837 | bpf_map_prepare_vdata(&obj->maps[i]); | |
7838 | ||
7839 | return 0; | |
7840 | } | |
7841 | ||
e7b924ca | 7842 | static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path) |
52d3352e | 7843 | { |
ec6d5f47 | 7844 | int err, i; |
6371ca3b | 7845 | |
52d3352e | 7846 | if (!obj) |
e9fc3ce9 | 7847 | return libbpf_err(-EINVAL); |
52d3352e WN |
7848 | |
7849 | if (obj->loaded) { | |
d9297581 | 7850 | pr_warn("object '%s': load can't be attempted twice\n", obj->name); |
e9fc3ce9 | 7851 | return libbpf_err(-EINVAL); |
52d3352e WN |
7852 | } |
7853 | ||
67234743 | 7854 | if (obj->gen_loader) |
be315829 | 7855 | bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps); |
67234743 | 7856 | |
fd9eef1a | 7857 | err = bpf_object__probe_loading(obj); |
fe62de31 | 7858 | err = err ? : bpf_object__load_vmlinux_btf(obj, false); |
8601fd42 | 7859 | err = err ? : bpf_object__resolve_externs(obj, obj->kconfig); |
0d13bfce AN |
7860 | err = err ? : bpf_object__sanitize_and_load_btf(obj); |
7861 | err = err ? : bpf_object__sanitize_maps(obj); | |
590a0088 | 7862 | err = err ? : bpf_object__init_kern_struct_ops_maps(obj); |
0d13bfce | 7863 | err = err ? : bpf_object__create_maps(obj); |
e7b924ca AN |
7864 | err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path); |
7865 | err = err ? : bpf_object__load_progs(obj, extra_log_level); | |
341ac5ff | 7866 | err = err ? : bpf_object_init_prog_arrays(obj); |
8d1608d7 | 7867 | err = err ? : bpf_object_prepare_struct_ops(obj); |
a6ed02ca | 7868 | |
67234743 AS |
7869 | if (obj->gen_loader) { |
7870 | /* reset FDs */ | |
4729445b KKD |
7871 | if (obj->btf) |
7872 | btf__set_fd(obj->btf, -1); | |
67234743 AS |
7873 | for (i = 0; i < obj->nr_maps; i++) |
7874 | obj->maps[i].fd = -1; | |
7875 | if (!err) | |
ba05fd36 | 7876 | err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps); |
67234743 AS |
7877 | } |
7878 | ||
9dbe6015 KKD |
7879 | /* clean up fd_array */ |
7880 | zfree(&obj->fd_array); | |
7881 | ||
4f33a53d AN |
7882 | /* clean up module BTFs */ |
7883 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
91abb4a6 | 7884 | close(obj->btf_modules[i].fd); |
4f33a53d AN |
7885 | btf__free(obj->btf_modules[i].btf); |
7886 | free(obj->btf_modules[i].name); | |
7887 | } | |
7888 | free(obj->btf_modules); | |
7889 | ||
7890 | /* clean up vmlinux BTF */ | |
a6ed02ca KS |
7891 | btf__free(obj->btf_vmlinux); |
7892 | obj->btf_vmlinux = NULL; | |
7893 | ||
d9297581 AN |
7894 | obj->loaded = true; /* doesn't matter if successfully or not */ |
7895 | ||
0d13bfce AN |
7896 | if (err) |
7897 | goto out; | |
52d3352e WN |
7898 | |
7899 | return 0; | |
7900 | out: | |
ec6d5f47 THJ |
7901 | /* unpin any maps that were auto-pinned during load */ |
7902 | for (i = 0; i < obj->nr_maps; i++) | |
7903 | if (obj->maps[i].pinned && !obj->maps[i].reused) | |
7904 | bpf_map__unpin(&obj->maps[i], NULL); | |
7905 | ||
4a404a7e | 7906 | bpf_object_unload(obj); |
be18010e | 7907 | pr_warn("failed to load object '%s'\n", obj->path); |
e9fc3ce9 | 7908 | return libbpf_err(err); |
52d3352e WN |
7909 | } |
7910 | ||
e7b924ca AN |
7911 | int bpf_object__load(struct bpf_object *obj) |
7912 | { | |
7913 | return bpf_object_load(obj, 0, NULL); | |
60276f98 QM |
7914 | } |
7915 | ||
196f8487 THJ |
7916 | static int make_parent_dir(const char *path) |
7917 | { | |
7918 | char *cp, errmsg[STRERR_BUFSIZE]; | |
7919 | char *dname, *dir; | |
7920 | int err = 0; | |
7921 | ||
7922 | dname = strdup(path); | |
7923 | if (dname == NULL) | |
7924 | return -ENOMEM; | |
7925 | ||
7926 | dir = dirname(dname); | |
7927 | if (mkdir(dir, 0700) && errno != EEXIST) | |
7928 | err = -errno; | |
7929 | ||
7930 | free(dname); | |
7931 | if (err) { | |
7932 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
7933 | pr_warn("failed to mkdir %s: %s\n", path, cp); | |
7934 | } | |
7935 | return err; | |
7936 | } | |
7937 | ||
f367540c JS |
7938 | static int check_path(const char *path) |
7939 | { | |
1ce6a9fc | 7940 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
7941 | struct statfs st_fs; |
7942 | char *dname, *dir; | |
7943 | int err = 0; | |
7944 | ||
7945 | if (path == NULL) | |
7946 | return -EINVAL; | |
7947 | ||
7948 | dname = strdup(path); | |
7949 | if (dname == NULL) | |
7950 | return -ENOMEM; | |
7951 | ||
7952 | dir = dirname(dname); | |
7953 | if (statfs(dir, &st_fs)) { | |
24d6a808 | 7954 | cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg)); |
be18010e | 7955 | pr_warn("failed to statfs %s: %s\n", dir, cp); |
f367540c JS |
7956 | err = -errno; |
7957 | } | |
7958 | free(dname); | |
7959 | ||
7960 | if (!err && st_fs.f_type != BPF_FS_MAGIC) { | |
be18010e | 7961 | pr_warn("specified path %s is not on BPF FS\n", path); |
f367540c JS |
7962 | err = -EINVAL; |
7963 | } | |
7964 | ||
7965 | return err; | |
7966 | } | |
7967 | ||
cf90a20d | 7968 | int bpf_program__pin(struct bpf_program *prog, const char *path) |
f367540c | 7969 | { |
1ce6a9fc | 7970 | char *cp, errmsg[STRERR_BUFSIZE]; |
f367540c JS |
7971 | int err; |
7972 | ||
cf90a20d AN |
7973 | if (prog->fd < 0) { |
7974 | pr_warn("prog '%s': can't pin program that wasn't loaded\n", prog->name); | |
7975 | return libbpf_err(-EINVAL); | |
7976 | } | |
7977 | ||
196f8487 THJ |
7978 | err = make_parent_dir(path); |
7979 | if (err) | |
e9fc3ce9 | 7980 | return libbpf_err(err); |
196f8487 | 7981 | |
f367540c JS |
7982 | err = check_path(path); |
7983 | if (err) | |
e9fc3ce9 | 7984 | return libbpf_err(err); |
f367540c | 7985 | |
cf90a20d | 7986 | if (bpf_obj_pin(prog->fd, path)) { |
23ab656b THJ |
7987 | err = -errno; |
7988 | cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg)); | |
cf90a20d | 7989 | pr_warn("prog '%s': failed to pin at '%s': %s\n", prog->name, path, cp); |
e9fc3ce9 | 7990 | return libbpf_err(err); |
f367540c | 7991 | } |
f367540c | 7992 | |
cf90a20d | 7993 | pr_debug("prog '%s': pinned at '%s'\n", prog->name, path); |
f367540c JS |
7994 | return 0; |
7995 | } | |
7996 | ||
cf90a20d | 7997 | int bpf_program__unpin(struct bpf_program *prog, const char *path) |
0c19a9fb SF |
7998 | { |
7999 | int err; | |
8000 | ||
cf90a20d AN |
8001 | if (prog->fd < 0) { |
8002 | pr_warn("prog '%s': can't unpin program that wasn't loaded\n", prog->name); | |
e9fc3ce9 | 8003 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
8004 | } |
8005 | ||
f367540c JS |
8006 | err = check_path(path); |
8007 | if (err) | |
e9fc3ce9 | 8008 | return libbpf_err(err); |
f367540c | 8009 | |
cf90a20d | 8010 | err = unlink(path); |
0c19a9fb | 8011 | if (err) |
e9fc3ce9 | 8012 | return libbpf_err(-errno); |
0c19a9fb | 8013 | |
cf90a20d | 8014 | pr_debug("prog '%s': unpinned from '%s'\n", prog->name, path); |
f367540c JS |
8015 | return 0; |
8016 | } | |
8017 | ||
b6989f35 JS |
8018 | int bpf_map__pin(struct bpf_map *map, const char *path) |
8019 | { | |
1ce6a9fc | 8020 | char *cp, errmsg[STRERR_BUFSIZE]; |
b6989f35 JS |
8021 | int err; |
8022 | ||
b6989f35 | 8023 | if (map == NULL) { |
be18010e | 8024 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 8025 | return libbpf_err(-EINVAL); |
b6989f35 JS |
8026 | } |
8027 | ||
4580b25f THJ |
8028 | if (map->pin_path) { |
8029 | if (path && strcmp(path, map->pin_path)) { | |
8030 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
8031 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 8032 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8033 | } else if (map->pinned) { |
8034 | pr_debug("map '%s' already pinned at '%s'; not re-pinning\n", | |
8035 | bpf_map__name(map), map->pin_path); | |
8036 | return 0; | |
8037 | } | |
8038 | } else { | |
8039 | if (!path) { | |
8040 | pr_warn("missing a path to pin map '%s' at\n", | |
8041 | bpf_map__name(map)); | |
e9fc3ce9 | 8042 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8043 | } else if (map->pinned) { |
8044 | pr_warn("map '%s' already pinned\n", bpf_map__name(map)); | |
e9fc3ce9 | 8045 | return libbpf_err(-EEXIST); |
4580b25f THJ |
8046 | } |
8047 | ||
8048 | map->pin_path = strdup(path); | |
8049 | if (!map->pin_path) { | |
8050 | err = -errno; | |
8051 | goto out_err; | |
8052 | } | |
b6989f35 JS |
8053 | } |
8054 | ||
196f8487 THJ |
8055 | err = make_parent_dir(map->pin_path); |
8056 | if (err) | |
e9fc3ce9 | 8057 | return libbpf_err(err); |
196f8487 | 8058 | |
4580b25f THJ |
8059 | err = check_path(map->pin_path); |
8060 | if (err) | |
e9fc3ce9 | 8061 | return libbpf_err(err); |
4580b25f THJ |
8062 | |
8063 | if (bpf_obj_pin(map->fd, map->pin_path)) { | |
8064 | err = -errno; | |
8065 | goto out_err; | |
8066 | } | |
8067 | ||
8068 | map->pinned = true; | |
8069 | pr_debug("pinned map '%s'\n", map->pin_path); | |
0c19a9fb | 8070 | |
b6989f35 | 8071 | return 0; |
4580b25f THJ |
8072 | |
8073 | out_err: | |
8074 | cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg)); | |
8075 | pr_warn("failed to pin map: %s\n", cp); | |
e9fc3ce9 | 8076 | return libbpf_err(err); |
b6989f35 JS |
8077 | } |
8078 | ||
0c19a9fb SF |
8079 | int bpf_map__unpin(struct bpf_map *map, const char *path) |
8080 | { | |
8081 | int err; | |
8082 | ||
0c19a9fb | 8083 | if (map == NULL) { |
be18010e | 8084 | pr_warn("invalid map pointer\n"); |
e9fc3ce9 | 8085 | return libbpf_err(-EINVAL); |
0c19a9fb SF |
8086 | } |
8087 | ||
4580b25f THJ |
8088 | if (map->pin_path) { |
8089 | if (path && strcmp(path, map->pin_path)) { | |
8090 | pr_warn("map '%s' already has pin path '%s' different from '%s'\n", | |
8091 | bpf_map__name(map), map->pin_path, path); | |
e9fc3ce9 | 8092 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8093 | } |
8094 | path = map->pin_path; | |
8095 | } else if (!path) { | |
8096 | pr_warn("no path to unpin map '%s' from\n", | |
8097 | bpf_map__name(map)); | |
e9fc3ce9 | 8098 | return libbpf_err(-EINVAL); |
4580b25f THJ |
8099 | } |
8100 | ||
8101 | err = check_path(path); | |
8102 | if (err) | |
e9fc3ce9 | 8103 | return libbpf_err(err); |
4580b25f | 8104 | |
0c19a9fb SF |
8105 | err = unlink(path); |
8106 | if (err != 0) | |
e9fc3ce9 | 8107 | return libbpf_err(-errno); |
4580b25f THJ |
8108 | |
8109 | map->pinned = false; | |
8110 | pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path); | |
0c19a9fb SF |
8111 | |
8112 | return 0; | |
8113 | } | |
8114 | ||
4580b25f THJ |
8115 | int bpf_map__set_pin_path(struct bpf_map *map, const char *path) |
8116 | { | |
8117 | char *new = NULL; | |
8118 | ||
8119 | if (path) { | |
8120 | new = strdup(path); | |
8121 | if (!new) | |
e9fc3ce9 | 8122 | return libbpf_err(-errno); |
4580b25f THJ |
8123 | } |
8124 | ||
8125 | free(map->pin_path); | |
8126 | map->pin_path = new; | |
8127 | return 0; | |
8128 | } | |
8129 | ||
20eccf29 AN |
8130 | __alias(bpf_map__pin_path) |
8131 | const char *bpf_map__get_pin_path(const struct bpf_map *map); | |
4580b25f | 8132 | |
e244d34d EL |
8133 | const char *bpf_map__pin_path(const struct bpf_map *map) |
8134 | { | |
8135 | return map->pin_path; | |
8136 | } | |
8137 | ||
4580b25f THJ |
8138 | bool bpf_map__is_pinned(const struct bpf_map *map) |
8139 | { | |
8140 | return map->pinned; | |
8141 | } | |
8142 | ||
9cf309c5 THJ |
8143 | static void sanitize_pin_path(char *s) |
8144 | { | |
8145 | /* bpffs disallows periods in path names */ | |
8146 | while (*s) { | |
8147 | if (*s == '.') | |
8148 | *s = '_'; | |
8149 | s++; | |
8150 | } | |
8151 | } | |
8152 | ||
0c19a9fb | 8153 | int bpf_object__pin_maps(struct bpf_object *obj, const char *path) |
d5148d85 | 8154 | { |
d5148d85 JS |
8155 | struct bpf_map *map; |
8156 | int err; | |
8157 | ||
8158 | if (!obj) | |
e9fc3ce9 | 8159 | return libbpf_err(-ENOENT); |
d5148d85 JS |
8160 | |
8161 | if (!obj->loaded) { | |
be18010e | 8162 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 8163 | return libbpf_err(-ENOENT); |
d5148d85 JS |
8164 | } |
8165 | ||
f74a53d9 | 8166 | bpf_object__for_each_map(map, obj) { |
4580b25f | 8167 | char *pin_path = NULL; |
0c19a9fb | 8168 | char buf[PATH_MAX]; |
0c19a9fb | 8169 | |
ec41817b | 8170 | if (!map->autocreate) |
229fae38 SC |
8171 | continue; |
8172 | ||
4580b25f | 8173 | if (path) { |
e588c116 WY |
8174 | err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map)); |
8175 | if (err) | |
4580b25f | 8176 | goto err_unpin_maps; |
9cf309c5 | 8177 | sanitize_pin_path(buf); |
4580b25f THJ |
8178 | pin_path = buf; |
8179 | } else if (!map->pin_path) { | |
8180 | continue; | |
0c19a9fb SF |
8181 | } |
8182 | ||
4580b25f | 8183 | err = bpf_map__pin(map, pin_path); |
0c19a9fb SF |
8184 | if (err) |
8185 | goto err_unpin_maps; | |
8186 | } | |
8187 | ||
8188 | return 0; | |
8189 | ||
8190 | err_unpin_maps: | |
bcc40fc0 | 8191 | while ((map = bpf_object__prev_map(obj, map))) { |
4580b25f | 8192 | if (!map->pin_path) |
0c19a9fb SF |
8193 | continue; |
8194 | ||
4580b25f | 8195 | bpf_map__unpin(map, NULL); |
0c19a9fb SF |
8196 | } |
8197 | ||
e9fc3ce9 | 8198 | return libbpf_err(err); |
0c19a9fb SF |
8199 | } |
8200 | ||
8201 | int bpf_object__unpin_maps(struct bpf_object *obj, const char *path) | |
8202 | { | |
8203 | struct bpf_map *map; | |
8204 | int err; | |
8205 | ||
8206 | if (!obj) | |
e9fc3ce9 | 8207 | return libbpf_err(-ENOENT); |
0c19a9fb | 8208 | |
f74a53d9 | 8209 | bpf_object__for_each_map(map, obj) { |
4580b25f | 8210 | char *pin_path = NULL; |
d5148d85 | 8211 | char buf[PATH_MAX]; |
d5148d85 | 8212 | |
4580b25f | 8213 | if (path) { |
e588c116 WY |
8214 | err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map)); |
8215 | if (err) | |
8216 | return libbpf_err(err); | |
9cf309c5 | 8217 | sanitize_pin_path(buf); |
4580b25f THJ |
8218 | pin_path = buf; |
8219 | } else if (!map->pin_path) { | |
8220 | continue; | |
8221 | } | |
d5148d85 | 8222 | |
4580b25f | 8223 | err = bpf_map__unpin(map, pin_path); |
d5148d85 | 8224 | if (err) |
e9fc3ce9 | 8225 | return libbpf_err(err); |
d5148d85 JS |
8226 | } |
8227 | ||
0c19a9fb SF |
8228 | return 0; |
8229 | } | |
8230 | ||
8231 | int bpf_object__pin_programs(struct bpf_object *obj, const char *path) | |
8232 | { | |
8233 | struct bpf_program *prog; | |
e588c116 | 8234 | char buf[PATH_MAX]; |
0c19a9fb SF |
8235 | int err; |
8236 | ||
8237 | if (!obj) | |
e9fc3ce9 | 8238 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
8239 | |
8240 | if (!obj->loaded) { | |
be18010e | 8241 | pr_warn("object not yet loaded; load it first\n"); |
e9fc3ce9 | 8242 | return libbpf_err(-ENOENT); |
0c19a9fb SF |
8243 | } |
8244 | ||
0c19a9fb | 8245 | bpf_object__for_each_program(prog, obj) { |
e588c116 WY |
8246 | err = pathname_concat(buf, sizeof(buf), path, prog->name); |
8247 | if (err) | |
0c19a9fb | 8248 | goto err_unpin_programs; |
0c19a9fb SF |
8249 | |
8250 | err = bpf_program__pin(prog, buf); | |
8251 | if (err) | |
8252 | goto err_unpin_programs; | |
8253 | } | |
8254 | ||
8255 | return 0; | |
8256 | ||
8257 | err_unpin_programs: | |
bcc40fc0 | 8258 | while ((prog = bpf_object__prev_program(obj, prog))) { |
e588c116 | 8259 | if (pathname_concat(buf, sizeof(buf), path, prog->name)) |
0c19a9fb SF |
8260 | continue; |
8261 | ||
8262 | bpf_program__unpin(prog, buf); | |
8263 | } | |
8264 | ||
e9fc3ce9 | 8265 | return libbpf_err(err); |
0c19a9fb SF |
8266 | } |
8267 | ||
8268 | int bpf_object__unpin_programs(struct bpf_object *obj, const char *path) | |
8269 | { | |
8270 | struct bpf_program *prog; | |
8271 | int err; | |
8272 | ||
8273 | if (!obj) | |
e9fc3ce9 | 8274 | return libbpf_err(-ENOENT); |
0c19a9fb | 8275 | |
d5148d85 JS |
8276 | bpf_object__for_each_program(prog, obj) { |
8277 | char buf[PATH_MAX]; | |
d5148d85 | 8278 | |
e588c116 WY |
8279 | err = pathname_concat(buf, sizeof(buf), path, prog->name); |
8280 | if (err) | |
8281 | return libbpf_err(err); | |
d5148d85 | 8282 | |
0c19a9fb | 8283 | err = bpf_program__unpin(prog, buf); |
d5148d85 | 8284 | if (err) |
e9fc3ce9 | 8285 | return libbpf_err(err); |
d5148d85 JS |
8286 | } |
8287 | ||
8288 | return 0; | |
8289 | } | |
8290 | ||
0c19a9fb SF |
8291 | int bpf_object__pin(struct bpf_object *obj, const char *path) |
8292 | { | |
8293 | int err; | |
8294 | ||
8295 | err = bpf_object__pin_maps(obj, path); | |
8296 | if (err) | |
e9fc3ce9 | 8297 | return libbpf_err(err); |
0c19a9fb SF |
8298 | |
8299 | err = bpf_object__pin_programs(obj, path); | |
8300 | if (err) { | |
8301 | bpf_object__unpin_maps(obj, path); | |
e9fc3ce9 | 8302 | return libbpf_err(err); |
0c19a9fb SF |
8303 | } |
8304 | ||
8305 | return 0; | |
8306 | } | |
8307 | ||
2d39d7c5 AN |
8308 | static void bpf_map__destroy(struct bpf_map *map) |
8309 | { | |
646f02ff AN |
8310 | if (map->inner_map) { |
8311 | bpf_map__destroy(map->inner_map); | |
8312 | zfree(&map->inner_map); | |
8313 | } | |
8314 | ||
8315 | zfree(&map->init_slots); | |
8316 | map->init_slots_sz = 0; | |
8317 | ||
2d39d7c5 | 8318 | if (map->mmaped) { |
9d0a2331 JK |
8319 | size_t mmap_sz; |
8320 | ||
8321 | mmap_sz = bpf_map_mmap_sz(map->def.value_size, map->def.max_entries); | |
8322 | munmap(map->mmaped, mmap_sz); | |
2d39d7c5 AN |
8323 | map->mmaped = NULL; |
8324 | } | |
8325 | ||
8326 | if (map->st_ops) { | |
8327 | zfree(&map->st_ops->data); | |
8328 | zfree(&map->st_ops->progs); | |
8329 | zfree(&map->st_ops->kern_func_off); | |
8330 | zfree(&map->st_ops); | |
8331 | } | |
8332 | ||
8333 | zfree(&map->name); | |
aed65917 | 8334 | zfree(&map->real_name); |
2d39d7c5 AN |
8335 | zfree(&map->pin_path); |
8336 | ||
8337 | if (map->fd >= 0) | |
8338 | zclose(map->fd); | |
8339 | } | |
8340 | ||
1a5e3fb1 WN |
8341 | void bpf_object__close(struct bpf_object *obj) |
8342 | { | |
a5b8bd47 WN |
8343 | size_t i; |
8344 | ||
50450fc7 | 8345 | if (IS_ERR_OR_NULL(obj)) |
1a5e3fb1 WN |
8346 | return; |
8347 | ||
2e4913e0 AN |
8348 | usdt_manager_free(obj->usdt_man); |
8349 | obj->usdt_man = NULL; | |
8350 | ||
67234743 | 8351 | bpf_gen__free(obj->gen_loader); |
1a5e3fb1 | 8352 | bpf_object__elf_finish(obj); |
4a404a7e | 8353 | bpf_object_unload(obj); |
8a138aed | 8354 | btf__free(obj->btf); |
2993e051 | 8355 | btf_ext__free(obj->btf_ext); |
1a5e3fb1 | 8356 | |
2d39d7c5 AN |
8357 | for (i = 0; i < obj->nr_maps; i++) |
8358 | bpf_map__destroy(&obj->maps[i]); | |
d859900c | 8359 | |
1373ff59 | 8360 | zfree(&obj->btf_custom_path); |
8601fd42 | 8361 | zfree(&obj->kconfig); |
166750bc AN |
8362 | zfree(&obj->externs); |
8363 | obj->nr_extern = 0; | |
8364 | ||
9d759a9b WN |
8365 | zfree(&obj->maps); |
8366 | obj->nr_maps = 0; | |
a5b8bd47 WN |
8367 | |
8368 | if (obj->programs && obj->nr_programs) { | |
8369 | for (i = 0; i < obj->nr_programs; i++) | |
8370 | bpf_program__exit(&obj->programs[i]); | |
8371 | } | |
8372 | zfree(&obj->programs); | |
8373 | ||
1a5e3fb1 WN |
8374 | free(obj); |
8375 | } | |
aa9b1ac3 | 8376 | |
a324aae3 | 8377 | const char *bpf_object__name(const struct bpf_object *obj) |
acf860ae | 8378 | { |
e9fc3ce9 | 8379 | return obj ? obj->name : libbpf_err_ptr(-EINVAL); |
acf860ae WN |
8380 | } |
8381 | ||
a324aae3 | 8382 | unsigned int bpf_object__kversion(const struct bpf_object *obj) |
45825d8a | 8383 | { |
a7fe0450 | 8384 | return obj ? obj->kern_version : 0; |
45825d8a WN |
8385 | } |
8386 | ||
a324aae3 | 8387 | struct btf *bpf_object__btf(const struct bpf_object *obj) |
789f6bab AI |
8388 | { |
8389 | return obj ? obj->btf : NULL; | |
8390 | } | |
8391 | ||
8a138aed MKL |
8392 | int bpf_object__btf_fd(const struct bpf_object *obj) |
8393 | { | |
8394 | return obj->btf ? btf__fd(obj->btf) : -1; | |
8395 | } | |
8396 | ||
155f556d RDT |
8397 | int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version) |
8398 | { | |
8399 | if (obj->loaded) | |
e9fc3ce9 | 8400 | return libbpf_err(-EINVAL); |
155f556d RDT |
8401 | |
8402 | obj->kern_version = kern_version; | |
8403 | ||
8404 | return 0; | |
8405 | } | |
8406 | ||
67234743 AS |
8407 | int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts) |
8408 | { | |
8409 | struct bpf_gen *gen; | |
8410 | ||
8411 | if (!opts) | |
8412 | return -EFAULT; | |
8413 | if (!OPTS_VALID(opts, gen_loader_opts)) | |
8414 | return -EINVAL; | |
8415 | gen = calloc(sizeof(*gen), 1); | |
8416 | if (!gen) | |
8417 | return -ENOMEM; | |
8418 | gen->opts = opts; | |
8419 | obj->gen_loader = gen; | |
8420 | return 0; | |
8421 | } | |
8422 | ||
eac7d845 | 8423 | static struct bpf_program * |
a324aae3 AN |
8424 | __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj, |
8425 | bool forward) | |
aa9b1ac3 | 8426 | { |
a83d6e76 | 8427 | size_t nr_programs = obj->nr_programs; |
0c19a9fb | 8428 | ssize_t idx; |
aa9b1ac3 | 8429 | |
a83d6e76 | 8430 | if (!nr_programs) |
aa9b1ac3 | 8431 | return NULL; |
aa9b1ac3 | 8432 | |
a83d6e76 MKL |
8433 | if (!p) |
8434 | /* Iter from the beginning */ | |
8435 | return forward ? &obj->programs[0] : | |
8436 | &obj->programs[nr_programs - 1]; | |
8437 | ||
0c19a9fb | 8438 | if (p->obj != obj) { |
be18010e | 8439 | pr_warn("error: program handler doesn't match object\n"); |
e9fc3ce9 | 8440 | return errno = EINVAL, NULL; |
aa9b1ac3 WN |
8441 | } |
8442 | ||
a83d6e76 | 8443 | idx = (p - obj->programs) + (forward ? 1 : -1); |
0c19a9fb | 8444 | if (idx >= obj->nr_programs || idx < 0) |
aa9b1ac3 WN |
8445 | return NULL; |
8446 | return &obj->programs[idx]; | |
8447 | } | |
8448 | ||
2088a3a7 HC |
8449 | struct bpf_program * |
8450 | bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev) | |
eac7d845 JK |
8451 | { |
8452 | struct bpf_program *prog = prev; | |
8453 | ||
8454 | do { | |
a83d6e76 | 8455 | prog = __bpf_program__iter(prog, obj, true); |
c3c55696 | 8456 | } while (prog && prog_is_subprog(obj, prog)); |
0c19a9fb SF |
8457 | |
8458 | return prog; | |
8459 | } | |
8460 | ||
2088a3a7 HC |
8461 | struct bpf_program * |
8462 | bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next) | |
0c19a9fb SF |
8463 | { |
8464 | struct bpf_program *prog = next; | |
8465 | ||
0c19a9fb | 8466 | do { |
a83d6e76 | 8467 | prog = __bpf_program__iter(prog, obj, false); |
c3c55696 | 8468 | } while (prog && prog_is_subprog(obj, prog)); |
eac7d845 JK |
8469 | |
8470 | return prog; | |
8471 | } | |
8472 | ||
9aba3613 JK |
8473 | void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex) |
8474 | { | |
8475 | prog->prog_ifindex = ifindex; | |
8476 | } | |
8477 | ||
01af3bf0 AN |
8478 | const char *bpf_program__name(const struct bpf_program *prog) |
8479 | { | |
8480 | return prog->name; | |
8481 | } | |
8482 | ||
52109584 AN |
8483 | const char *bpf_program__section_name(const struct bpf_program *prog) |
8484 | { | |
8485 | return prog->sec_name; | |
8486 | } | |
8487 | ||
d9297581 AN |
8488 | bool bpf_program__autoload(const struct bpf_program *prog) |
8489 | { | |
a3820c48 | 8490 | return prog->autoload; |
d9297581 AN |
8491 | } |
8492 | ||
8493 | int bpf_program__set_autoload(struct bpf_program *prog, bool autoload) | |
8494 | { | |
8495 | if (prog->obj->loaded) | |
e9fc3ce9 | 8496 | return libbpf_err(-EINVAL); |
d9297581 | 8497 | |
a3820c48 | 8498 | prog->autoload = autoload; |
d9297581 AN |
8499 | return 0; |
8500 | } | |
8501 | ||
43cb8cba HL |
8502 | bool bpf_program__autoattach(const struct bpf_program *prog) |
8503 | { | |
8504 | return prog->autoattach; | |
8505 | } | |
8506 | ||
8507 | void bpf_program__set_autoattach(struct bpf_program *prog, bool autoattach) | |
8508 | { | |
8509 | prog->autoattach = autoattach; | |
8510 | } | |
8511 | ||
65a7fa2e AN |
8512 | const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog) |
8513 | { | |
8514 | return prog->insns; | |
8515 | } | |
8516 | ||
8517 | size_t bpf_program__insn_cnt(const struct bpf_program *prog) | |
8518 | { | |
8519 | return prog->insns_cnt; | |
8520 | } | |
8521 | ||
b63b3c49 JO |
8522 | int bpf_program__set_insns(struct bpf_program *prog, |
8523 | struct bpf_insn *new_insns, size_t new_insn_cnt) | |
8524 | { | |
8525 | struct bpf_insn *insns; | |
8526 | ||
8527 | if (prog->obj->loaded) | |
8528 | return -EBUSY; | |
8529 | ||
8530 | insns = libbpf_reallocarray(prog->insns, new_insn_cnt, sizeof(*insns)); | |
8531 | if (!insns) { | |
8532 | pr_warn("prog '%s': failed to realloc prog code\n", prog->name); | |
8533 | return -ENOMEM; | |
8534 | } | |
8535 | memcpy(insns, new_insns, new_insn_cnt * sizeof(*insns)); | |
8536 | ||
8537 | prog->insns = insns; | |
8538 | prog->insns_cnt = new_insn_cnt; | |
8539 | return 0; | |
8540 | } | |
8541 | ||
cf90a20d | 8542 | int bpf_program__fd(const struct bpf_program *prog) |
b580563e | 8543 | { |
1e960043 | 8544 | if (!prog) |
e9fc3ce9 | 8545 | return libbpf_err(-EINVAL); |
1e960043 | 8546 | |
cf90a20d | 8547 | if (prog->fd < 0) |
e9fc3ce9 | 8548 | return libbpf_err(-ENOENT); |
b580563e | 8549 | |
cf90a20d | 8550 | return prog->fd; |
aa9b1ac3 | 8551 | } |
9d759a9b | 8552 | |
20eccf29 AN |
8553 | __alias(bpf_program__type) |
8554 | enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog); | |
8555 | ||
8556 | enum bpf_prog_type bpf_program__type(const struct bpf_program *prog) | |
f1eead9e AN |
8557 | { |
8558 | return prog->type; | |
8559 | } | |
8560 | ||
93442f13 | 8561 | int bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type) |
5f44e4c8 | 8562 | { |
93442f13 GS |
8563 | if (prog->obj->loaded) |
8564 | return libbpf_err(-EBUSY); | |
8565 | ||
5f44e4c8 | 8566 | prog->type = type; |
d6e6286a | 8567 | prog->sec_def = NULL; |
93442f13 | 8568 | return 0; |
5f44e4c8 WN |
8569 | } |
8570 | ||
20eccf29 AN |
8571 | __alias(bpf_program__expected_attach_type) |
8572 | enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog); | |
8573 | ||
8574 | enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog) | |
f1eead9e AN |
8575 | { |
8576 | return prog->expected_attach_type; | |
8577 | } | |
8578 | ||
93442f13 | 8579 | int bpf_program__set_expected_attach_type(struct bpf_program *prog, |
16962b24 | 8580 | enum bpf_attach_type type) |
d7be143b | 8581 | { |
93442f13 GS |
8582 | if (prog->obj->loaded) |
8583 | return libbpf_err(-EBUSY); | |
8584 | ||
d7be143b | 8585 | prog->expected_attach_type = type; |
93442f13 | 8586 | return 0; |
d7be143b AI |
8587 | } |
8588 | ||
a6ca7158 AN |
8589 | __u32 bpf_program__flags(const struct bpf_program *prog) |
8590 | { | |
8591 | return prog->prog_flags; | |
8592 | } | |
8593 | ||
8cccee9e | 8594 | int bpf_program__set_flags(struct bpf_program *prog, __u32 flags) |
a6ca7158 AN |
8595 | { |
8596 | if (prog->obj->loaded) | |
8597 | return libbpf_err(-EBUSY); | |
8598 | ||
8cccee9e | 8599 | prog->prog_flags = flags; |
a6ca7158 AN |
8600 | return 0; |
8601 | } | |
8602 | ||
dbdd2c7f AN |
8603 | __u32 bpf_program__log_level(const struct bpf_program *prog) |
8604 | { | |
8605 | return prog->log_level; | |
8606 | } | |
8607 | ||
8608 | int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level) | |
8609 | { | |
8610 | if (prog->obj->loaded) | |
8611 | return libbpf_err(-EBUSY); | |
8612 | ||
8613 | prog->log_level = log_level; | |
8614 | return 0; | |
8615 | } | |
8616 | ||
b3ce9079 AN |
8617 | const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size) |
8618 | { | |
8619 | *log_size = prog->log_size; | |
8620 | return prog->log_buf; | |
8621 | } | |
8622 | ||
8623 | int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size) | |
8624 | { | |
8625 | if (log_size && !log_buf) | |
8626 | return -EINVAL; | |
8627 | if (prog->log_size > UINT_MAX) | |
8628 | return -EINVAL; | |
8629 | if (prog->obj->loaded) | |
8630 | return -EBUSY; | |
8631 | ||
8632 | prog->log_buf = log_buf; | |
8633 | prog->log_size = log_size; | |
a6ca7158 AN |
8634 | return 0; |
8635 | } | |
8636 | ||
15ea31fa | 8637 | #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \ |
697f104d | 8638 | .sec = (char *)sec_pfx, \ |
d7a18ea7 | 8639 | .prog_type = BPF_PROG_TYPE_##ptype, \ |
15ea31fa AN |
8640 | .expected_attach_type = atype, \ |
8641 | .cookie = (long)(flags), \ | |
4fa5bcfe | 8642 | .prog_prepare_load_fn = libbpf_prepare_prog_load, \ |
d7a18ea7 AN |
8643 | __VA_ARGS__ \ |
8644 | } | |
8645 | ||
4fa5bcfe | 8646 | static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
39f8dc43 | 8647 | static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
708ac5be | 8648 | static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
2e4913e0 | 8649 | static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
4fa5bcfe AN |
8650 | static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
8651 | static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
8652 | static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
ddc6b049 | 8653 | static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
4fa5bcfe AN |
8654 | static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link); |
8655 | static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link); | |
d7a18ea7 | 8656 | |
d7a18ea7 | 8657 | static const struct bpf_sec_def section_defs[] = { |
450b167f AN |
8658 | SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE), |
8659 | SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE), | |
8660 | SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE), | |
9af8efc4 | 8661 | SEC_DEF("kprobe+", KPROBE, 0, SEC_NONE, attach_kprobe), |
39f8dc43 | 8662 | SEC_DEF("uprobe+", KPROBE, 0, SEC_NONE, attach_uprobe), |
c4cac71f | 8663 | SEC_DEF("uprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe), |
9af8efc4 | 8664 | SEC_DEF("kretprobe+", KPROBE, 0, SEC_NONE, attach_kprobe), |
39f8dc43 | 8665 | SEC_DEF("uretprobe+", KPROBE, 0, SEC_NONE, attach_uprobe), |
c4cac71f | 8666 | SEC_DEF("uretprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe), |
9af8efc4 AN |
8667 | SEC_DEF("kprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi), |
8668 | SEC_DEF("kretprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi), | |
708ac5be AN |
8669 | SEC_DEF("ksyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall), |
8670 | SEC_DEF("kretsyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall), | |
2e4913e0 | 8671 | SEC_DEF("usdt+", KPROBE, 0, SEC_NONE, attach_usdt), |
15ea31fa | 8672 | SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE), |
450b167f AN |
8673 | SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE), |
8674 | SEC_DEF("action", SCHED_ACT, 0, SEC_NONE), | |
9af8efc4 AN |
8675 | SEC_DEF("tracepoint+", TRACEPOINT, 0, SEC_NONE, attach_tp), |
8676 | SEC_DEF("tp+", TRACEPOINT, 0, SEC_NONE, attach_tp), | |
8677 | SEC_DEF("raw_tracepoint+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
8678 | SEC_DEF("raw_tp+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp), | |
8679 | SEC_DEF("raw_tracepoint.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), | |
8680 | SEC_DEF("raw_tp.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp), | |
cc7d8f2c AN |
8681 | SEC_DEF("tp_btf+", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace), |
8682 | SEC_DEF("fentry+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace), | |
8683 | SEC_DEF("fmod_ret+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace), | |
8684 | SEC_DEF("fexit+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace), | |
8685 | SEC_DEF("fentry.s+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8686 | SEC_DEF("fmod_ret.s+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8687 | SEC_DEF("fexit.s+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace), | |
8688 | SEC_DEF("freplace+", EXT, 0, SEC_ATTACH_BTF, attach_trace), | |
8689 | SEC_DEF("lsm+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm), | |
8690 | SEC_DEF("lsm.s+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm), | |
bffcf348 | 8691 | SEC_DEF("lsm_cgroup+", LSM, BPF_LSM_CGROUP, SEC_ATTACH_BTF), |
cc7d8f2c AN |
8692 | SEC_DEF("iter+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter), |
8693 | SEC_DEF("iter.s+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter), | |
15ea31fa | 8694 | SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE), |
082c4bfb | 8695 | SEC_DEF("xdp.frags/devmap", XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS), |
4a4d4cee | 8696 | SEC_DEF("xdp/devmap", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE), |
082c4bfb | 8697 | SEC_DEF("xdp.frags/cpumap", XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS), |
4a4d4cee | 8698 | SEC_DEF("xdp/cpumap", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE), |
082c4bfb | 8699 | SEC_DEF("xdp.frags", XDP, BPF_XDP, SEC_XDP_FRAGS), |
450b167f AN |
8700 | SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT), |
8701 | SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE), | |
8702 | SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE), | |
8703 | SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE), | |
8704 | SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE), | |
8705 | SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE), | |
8706 | SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT), | |
8707 | SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT), | |
8708 | SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT), | |
8709 | SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE), | |
8710 | SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT), | |
8711 | SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT), | |
8712 | SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT), | |
8713 | SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT), | |
8714 | SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT), | |
8715 | SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE), | |
8716 | SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE), | |
8717 | SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE), | |
8718 | SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT), | |
8719 | SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE), | |
8720 | SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE), | |
8721 | SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE), | |
8722 | SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE), | |
8723 | SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE), | |
8724 | SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE), | |
8725 | SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE), | |
8726 | SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE), | |
8727 | SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE), | |
8728 | SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE), | |
8729 | SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE), | |
8730 | SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE), | |
8731 | SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE), | |
8732 | SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE), | |
8733 | SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE), | |
8734 | SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE), | |
8735 | SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE), | |
8736 | SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT), | |
dd94d45c | 8737 | SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), |
913b2255 | 8738 | SEC_DEF("struct_ops.s+", STRUCT_OPS, 0, SEC_SLEEPABLE), |
450b167f | 8739 | SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE), |
d0fe92fb | 8740 | SEC_DEF("netfilter", NETFILTER, 0, SEC_NONE), |
583c9009 | 8741 | }; |
d7be143b | 8742 | |
697f104d AN |
8743 | static size_t custom_sec_def_cnt; |
8744 | static struct bpf_sec_def *custom_sec_defs; | |
8745 | static struct bpf_sec_def custom_fallback_def; | |
8746 | static bool has_custom_fallback_def; | |
c76e4c22 | 8747 | |
697f104d AN |
8748 | static int last_custom_sec_def_handler_id; |
8749 | ||
8750 | int libbpf_register_prog_handler(const char *sec, | |
8751 | enum bpf_prog_type prog_type, | |
8752 | enum bpf_attach_type exp_attach_type, | |
8753 | const struct libbpf_prog_handler_opts *opts) | |
d7a18ea7 | 8754 | { |
697f104d | 8755 | struct bpf_sec_def *sec_def; |
d7a18ea7 | 8756 | |
697f104d AN |
8757 | if (!OPTS_VALID(opts, libbpf_prog_handler_opts)) |
8758 | return libbpf_err(-EINVAL); | |
dd94d45c | 8759 | |
697f104d AN |
8760 | if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */ |
8761 | return libbpf_err(-E2BIG); | |
dd94d45c | 8762 | |
697f104d AN |
8763 | if (sec) { |
8764 | sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1, | |
8765 | sizeof(*sec_def)); | |
8766 | if (!sec_def) | |
8767 | return libbpf_err(-ENOMEM); | |
dd94d45c | 8768 | |
697f104d AN |
8769 | custom_sec_defs = sec_def; |
8770 | sec_def = &custom_sec_defs[custom_sec_def_cnt]; | |
8771 | } else { | |
8772 | if (has_custom_fallback_def) | |
8773 | return libbpf_err(-EBUSY); | |
dd94d45c | 8774 | |
697f104d AN |
8775 | sec_def = &custom_fallback_def; |
8776 | } | |
8777 | ||
8778 | sec_def->sec = sec ? strdup(sec) : NULL; | |
8779 | if (sec && !sec_def->sec) | |
8780 | return libbpf_err(-ENOMEM); | |
8781 | ||
8782 | sec_def->prog_type = prog_type; | |
8783 | sec_def->expected_attach_type = exp_attach_type; | |
8784 | sec_def->cookie = OPTS_GET(opts, cookie, 0); | |
8785 | ||
8786 | sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL); | |
8787 | sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL); | |
8788 | sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL); | |
8789 | ||
8790 | sec_def->handler_id = ++last_custom_sec_def_handler_id; | |
8791 | ||
8792 | if (sec) | |
8793 | custom_sec_def_cnt++; | |
8794 | else | |
8795 | has_custom_fallback_def = true; | |
8796 | ||
8797 | return sec_def->handler_id; | |
8798 | } | |
8799 | ||
8800 | int libbpf_unregister_prog_handler(int handler_id) | |
8801 | { | |
8802 | struct bpf_sec_def *sec_defs; | |
8803 | int i; | |
8804 | ||
8805 | if (handler_id <= 0) | |
8806 | return libbpf_err(-EINVAL); | |
8807 | ||
8808 | if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) { | |
8809 | memset(&custom_fallback_def, 0, sizeof(custom_fallback_def)); | |
8810 | has_custom_fallback_def = false; | |
8811 | return 0; | |
8812 | } | |
8813 | ||
8814 | for (i = 0; i < custom_sec_def_cnt; i++) { | |
8815 | if (custom_sec_defs[i].handler_id == handler_id) | |
8816 | break; | |
8817 | } | |
8818 | ||
8819 | if (i == custom_sec_def_cnt) | |
8820 | return libbpf_err(-ENOENT); | |
8821 | ||
8822 | free(custom_sec_defs[i].sec); | |
8823 | for (i = i + 1; i < custom_sec_def_cnt; i++) | |
8824 | custom_sec_defs[i - 1] = custom_sec_defs[i]; | |
8825 | custom_sec_def_cnt--; | |
8826 | ||
8827 | /* try to shrink the array, but it's ok if we couldn't */ | |
8828 | sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs)); | |
8829 | if (sec_defs) | |
8830 | custom_sec_defs = sec_defs; | |
8831 | ||
8832 | return 0; | |
8833 | } | |
8834 | ||
450b167f | 8835 | static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name) |
697f104d AN |
8836 | { |
8837 | size_t len = strlen(sec_def->sec); | |
8838 | ||
8839 | /* "type/" always has to have proper SEC("type/extras") form */ | |
8840 | if (sec_def->sec[len - 1] == '/') { | |
8841 | if (str_has_pfx(sec_name, sec_def->sec)) | |
8842 | return true; | |
8843 | return false; | |
8844 | } | |
8845 | ||
8846 | /* "type+" means it can be either exact SEC("type") or | |
8847 | * well-formed SEC("type/extras") with proper '/' separator | |
8848 | */ | |
8849 | if (sec_def->sec[len - 1] == '+') { | |
8850 | len--; | |
8851 | /* not even a prefix */ | |
8852 | if (strncmp(sec_name, sec_def->sec, len) != 0) | |
8853 | return false; | |
8854 | /* exact match or has '/' separator */ | |
8855 | if (sec_name[len] == '\0' || sec_name[len] == '/') | |
8856 | return true; | |
8857 | return false; | |
8858 | } | |
8859 | ||
697f104d AN |
8860 | return strcmp(sec_name, sec_def->sec) == 0; |
8861 | } | |
8862 | ||
8863 | static const struct bpf_sec_def *find_sec_def(const char *sec_name) | |
8864 | { | |
8865 | const struct bpf_sec_def *sec_def; | |
8866 | int i, n; | |
697f104d AN |
8867 | |
8868 | n = custom_sec_def_cnt; | |
8869 | for (i = 0; i < n; i++) { | |
8870 | sec_def = &custom_sec_defs[i]; | |
450b167f | 8871 | if (sec_def_matches(sec_def, sec_name)) |
697f104d AN |
8872 | return sec_def; |
8873 | } | |
8874 | ||
8875 | n = ARRAY_SIZE(section_defs); | |
8876 | for (i = 0; i < n; i++) { | |
8877 | sec_def = §ion_defs[i]; | |
450b167f | 8878 | if (sec_def_matches(sec_def, sec_name)) |
dd94d45c | 8879 | return sec_def; |
d7a18ea7 | 8880 | } |
697f104d AN |
8881 | |
8882 | if (has_custom_fallback_def) | |
8883 | return &custom_fallback_def; | |
8884 | ||
d7a18ea7 AN |
8885 | return NULL; |
8886 | } | |
8887 | ||
697f104d AN |
8888 | #define MAX_TYPE_NAME_SIZE 32 |
8889 | ||
c76e4c22 TS |
8890 | static char *libbpf_get_type_names(bool attach_type) |
8891 | { | |
d7a18ea7 | 8892 | int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE; |
c76e4c22 TS |
8893 | char *buf; |
8894 | ||
8895 | buf = malloc(len); | |
8896 | if (!buf) | |
8897 | return NULL; | |
8898 | ||
8899 | buf[0] = '\0'; | |
8900 | /* Forge string buf with all available names */ | |
d7a18ea7 | 8901 | for (i = 0; i < ARRAY_SIZE(section_defs); i++) { |
15ea31fa AN |
8902 | const struct bpf_sec_def *sec_def = §ion_defs[i]; |
8903 | ||
8904 | if (attach_type) { | |
4fa5bcfe | 8905 | if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load) |
15ea31fa AN |
8906 | continue; |
8907 | ||
8908 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
8909 | continue; | |
8910 | } | |
c76e4c22 | 8911 | |
d7a18ea7 | 8912 | if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) { |
c76e4c22 TS |
8913 | free(buf); |
8914 | return NULL; | |
8915 | } | |
8916 | strcat(buf, " "); | |
d7a18ea7 | 8917 | strcat(buf, section_defs[i].sec); |
c76e4c22 TS |
8918 | } |
8919 | ||
8920 | return buf; | |
8921 | } | |
8922 | ||
b60df2a0 JK |
8923 | int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type, |
8924 | enum bpf_attach_type *expected_attach_type) | |
583c9009 | 8925 | { |
d7a18ea7 | 8926 | const struct bpf_sec_def *sec_def; |
c76e4c22 | 8927 | char *type_names; |
583c9009 | 8928 | |
b60df2a0 | 8929 | if (!name) |
e9fc3ce9 | 8930 | return libbpf_err(-EINVAL); |
583c9009 | 8931 | |
d7a18ea7 AN |
8932 | sec_def = find_sec_def(name); |
8933 | if (sec_def) { | |
8934 | *prog_type = sec_def->prog_type; | |
8935 | *expected_attach_type = sec_def->expected_attach_type; | |
b60df2a0 JK |
8936 | return 0; |
8937 | } | |
d7a18ea7 | 8938 | |
4a3d6c6a | 8939 | pr_debug("failed to guess program type from ELF section '%s'\n", name); |
c76e4c22 TS |
8940 | type_names = libbpf_get_type_names(false); |
8941 | if (type_names != NULL) { | |
3f519353 | 8942 | pr_debug("supported section(type) names are:%s\n", type_names); |
c76e4c22 TS |
8943 | free(type_names); |
8944 | } | |
8945 | ||
e9fc3ce9 | 8946 | return libbpf_err(-ESRCH); |
b60df2a0 | 8947 | } |
583c9009 | 8948 | |
ccde5760 DM |
8949 | const char *libbpf_bpf_attach_type_str(enum bpf_attach_type t) |
8950 | { | |
8951 | if (t < 0 || t >= ARRAY_SIZE(attach_type_name)) | |
8952 | return NULL; | |
8953 | ||
8954 | return attach_type_name[t]; | |
8955 | } | |
8956 | ||
ba5d1b58 DM |
8957 | const char *libbpf_bpf_link_type_str(enum bpf_link_type t) |
8958 | { | |
8959 | if (t < 0 || t >= ARRAY_SIZE(link_type_name)) | |
8960 | return NULL; | |
8961 | ||
8962 | return link_type_name[t]; | |
8963 | } | |
8964 | ||
3e6dc020 DM |
8965 | const char *libbpf_bpf_map_type_str(enum bpf_map_type t) |
8966 | { | |
8967 | if (t < 0 || t >= ARRAY_SIZE(map_type_name)) | |
8968 | return NULL; | |
8969 | ||
8970 | return map_type_name[t]; | |
8971 | } | |
8972 | ||
d18616e7 DM |
8973 | const char *libbpf_bpf_prog_type_str(enum bpf_prog_type t) |
8974 | { | |
8975 | if (t < 0 || t >= ARRAY_SIZE(prog_type_name)) | |
8976 | return NULL; | |
8977 | ||
8978 | return prog_type_name[t]; | |
8979 | } | |
8980 | ||
590a0088 | 8981 | static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj, |
809a69d6 | 8982 | int sec_idx, |
590a0088 MKL |
8983 | size_t offset) |
8984 | { | |
8985 | struct bpf_map *map; | |
8986 | size_t i; | |
8987 | ||
8988 | for (i = 0; i < obj->nr_maps; i++) { | |
8989 | map = &obj->maps[i]; | |
8990 | if (!bpf_map__is_struct_ops(map)) | |
8991 | continue; | |
809a69d6 KFL |
8992 | if (map->sec_idx == sec_idx && |
8993 | map->sec_offset <= offset && | |
590a0088 MKL |
8994 | offset - map->sec_offset < map->def.value_size) |
8995 | return map; | |
8996 | } | |
8997 | ||
8998 | return NULL; | |
8999 | } | |
9000 | ||
9001 | /* Collect the reloc from ELF and populate the st_ops->progs[] */ | |
646f02ff | 9002 | static int bpf_object__collect_st_ops_relos(struct bpf_object *obj, |
ad23b723 | 9003 | Elf64_Shdr *shdr, Elf_Data *data) |
590a0088 MKL |
9004 | { |
9005 | const struct btf_member *member; | |
9006 | struct bpf_struct_ops *st_ops; | |
9007 | struct bpf_program *prog; | |
9008 | unsigned int shdr_idx; | |
9009 | const struct btf *btf; | |
9010 | struct bpf_map *map; | |
7e06aad5 | 9011 | unsigned int moff, insn_idx; |
590a0088 | 9012 | const char *name; |
1d1a3bcf | 9013 | __u32 member_idx; |
ad23b723 AN |
9014 | Elf64_Sym *sym; |
9015 | Elf64_Rel *rel; | |
590a0088 MKL |
9016 | int i, nrels; |
9017 | ||
590a0088 MKL |
9018 | btf = obj->btf; |
9019 | nrels = shdr->sh_size / shdr->sh_entsize; | |
9020 | for (i = 0; i < nrels; i++) { | |
ad23b723 AN |
9021 | rel = elf_rel_by_idx(data, i); |
9022 | if (!rel) { | |
590a0088 MKL |
9023 | pr_warn("struct_ops reloc: failed to get %d reloc\n", i); |
9024 | return -LIBBPF_ERRNO__FORMAT; | |
9025 | } | |
9026 | ||
ad23b723 AN |
9027 | sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info)); |
9028 | if (!sym) { | |
590a0088 | 9029 | pr_warn("struct_ops reloc: symbol %zx not found\n", |
ad23b723 | 9030 | (size_t)ELF64_R_SYM(rel->r_info)); |
590a0088 MKL |
9031 | return -LIBBPF_ERRNO__FORMAT; |
9032 | } | |
9033 | ||
ad23b723 | 9034 | name = elf_sym_str(obj, sym->st_name) ?: "<?>"; |
809a69d6 | 9035 | map = find_struct_ops_map_by_offset(obj, shdr->sh_info, rel->r_offset); |
590a0088 | 9036 | if (!map) { |
ad23b723 AN |
9037 | pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n", |
9038 | (size_t)rel->r_offset); | |
590a0088 MKL |
9039 | return -EINVAL; |
9040 | } | |
9041 | ||
ad23b723 AN |
9042 | moff = rel->r_offset - map->sec_offset; |
9043 | shdr_idx = sym->st_shndx; | |
590a0088 | 9044 | st_ops = map->st_ops; |
ad23b723 | 9045 | 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 | 9046 | map->name, |
ad23b723 AN |
9047 | (long long)(rel->r_info >> 32), |
9048 | (long long)sym->st_value, | |
9049 | shdr_idx, (size_t)rel->r_offset, | |
9050 | map->sec_offset, sym->st_name, name); | |
590a0088 MKL |
9051 | |
9052 | if (shdr_idx >= SHN_LORESERVE) { | |
ad23b723 AN |
9053 | pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n", |
9054 | map->name, (size_t)rel->r_offset, shdr_idx); | |
590a0088 MKL |
9055 | return -LIBBPF_ERRNO__RELOC; |
9056 | } | |
ad23b723 | 9057 | if (sym->st_value % BPF_INSN_SZ) { |
7e06aad5 | 9058 | pr_warn("struct_ops reloc %s: invalid target program offset %llu\n", |
ad23b723 | 9059 | map->name, (unsigned long long)sym->st_value); |
7e06aad5 AN |
9060 | return -LIBBPF_ERRNO__FORMAT; |
9061 | } | |
ad23b723 | 9062 | insn_idx = sym->st_value / BPF_INSN_SZ; |
590a0088 MKL |
9063 | |
9064 | member = find_member_by_offset(st_ops->type, moff * 8); | |
9065 | if (!member) { | |
9066 | pr_warn("struct_ops reloc %s: cannot find member at moff %u\n", | |
9067 | map->name, moff); | |
9068 | return -EINVAL; | |
9069 | } | |
9070 | member_idx = member - btf_members(st_ops->type); | |
9071 | name = btf__name_by_offset(btf, member->name_off); | |
9072 | ||
9073 | if (!resolve_func_ptr(btf, member->type, NULL)) { | |
9074 | pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n", | |
9075 | map->name, name); | |
9076 | return -EINVAL; | |
9077 | } | |
9078 | ||
7e06aad5 | 9079 | prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx); |
590a0088 MKL |
9080 | if (!prog) { |
9081 | pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n", | |
9082 | map->name, shdr_idx, name); | |
9083 | return -EINVAL; | |
9084 | } | |
9085 | ||
91b4d1d1 AN |
9086 | /* prevent the use of BPF prog with invalid type */ |
9087 | if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) { | |
9088 | pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n", | |
9089 | map->name, prog->name); | |
9090 | return -EINVAL; | |
9091 | } | |
590a0088 | 9092 | |
91b4d1d1 AN |
9093 | /* if we haven't yet processed this BPF program, record proper |
9094 | * attach_btf_id and member_idx | |
9095 | */ | |
9096 | if (!prog->attach_btf_id) { | |
590a0088 MKL |
9097 | prog->attach_btf_id = st_ops->type_id; |
9098 | prog->expected_attach_type = member_idx; | |
590a0088 | 9099 | } |
91b4d1d1 AN |
9100 | |
9101 | /* struct_ops BPF prog can be re-used between multiple | |
809a69d6 KFL |
9102 | * .struct_ops & .struct_ops.link as long as it's the |
9103 | * same struct_ops struct definition and the same | |
9104 | * function pointer field | |
91b4d1d1 AN |
9105 | */ |
9106 | if (prog->attach_btf_id != st_ops->type_id || | |
9107 | prog->expected_attach_type != member_idx) { | |
9108 | pr_warn("struct_ops reloc %s: cannot use prog %s in sec %s with type %u attach_btf_id %u expected_attach_type %u for func ptr %s\n", | |
9109 | map->name, prog->name, prog->sec_name, prog->type, | |
9110 | prog->attach_btf_id, prog->expected_attach_type, name); | |
9111 | return -EINVAL; | |
9112 | } | |
9113 | ||
590a0088 MKL |
9114 | st_ops->progs[member_idx] = prog; |
9115 | } | |
9116 | ||
9117 | return 0; | |
590a0088 MKL |
9118 | } |
9119 | ||
a6ed02ca | 9120 | #define BTF_TRACE_PREFIX "btf_trace_" |
1e092a03 | 9121 | #define BTF_LSM_PREFIX "bpf_lsm_" |
21aef70e | 9122 | #define BTF_ITER_PREFIX "bpf_iter_" |
a6ed02ca KS |
9123 | #define BTF_MAX_NAME_SIZE 128 |
9124 | ||
67234743 AS |
9125 | void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type, |
9126 | const char **prefix, int *kind) | |
9127 | { | |
9128 | switch (attach_type) { | |
9129 | case BPF_TRACE_RAW_TP: | |
9130 | *prefix = BTF_TRACE_PREFIX; | |
9131 | *kind = BTF_KIND_TYPEDEF; | |
9132 | break; | |
9133 | case BPF_LSM_MAC: | |
bffcf348 | 9134 | case BPF_LSM_CGROUP: |
67234743 AS |
9135 | *prefix = BTF_LSM_PREFIX; |
9136 | *kind = BTF_KIND_FUNC; | |
9137 | break; | |
9138 | case BPF_TRACE_ITER: | |
9139 | *prefix = BTF_ITER_PREFIX; | |
9140 | *kind = BTF_KIND_FUNC; | |
9141 | break; | |
9142 | default: | |
9143 | *prefix = ""; | |
9144 | *kind = BTF_KIND_FUNC; | |
9145 | } | |
9146 | } | |
9147 | ||
a6ed02ca KS |
9148 | static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix, |
9149 | const char *name, __u32 kind) | |
9150 | { | |
9151 | char btf_type_name[BTF_MAX_NAME_SIZE]; | |
9152 | int ret; | |
9153 | ||
9154 | ret = snprintf(btf_type_name, sizeof(btf_type_name), | |
9155 | "%s%s", prefix, name); | |
9156 | /* snprintf returns the number of characters written excluding the | |
c139e40a | 9157 | * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it |
a6ed02ca KS |
9158 | * indicates truncation. |
9159 | */ | |
9160 | if (ret < 0 || ret >= sizeof(btf_type_name)) | |
9161 | return -ENAMETOOLONG; | |
9162 | return btf__find_by_name_kind(btf, btf_type_name, kind); | |
9163 | } | |
9164 | ||
91abb4a6 AN |
9165 | static inline int find_attach_btf_id(struct btf *btf, const char *name, |
9166 | enum bpf_attach_type attach_type) | |
a6ed02ca | 9167 | { |
67234743 AS |
9168 | const char *prefix; |
9169 | int kind; | |
a6ed02ca | 9170 | |
67234743 AS |
9171 | btf_get_kernel_prefix_kind(attach_type, &prefix, &kind); |
9172 | return find_btf_by_prefix_kind(btf, prefix, name, kind); | |
a6ed02ca KS |
9173 | } |
9174 | ||
b8c54ea4 AS |
9175 | int libbpf_find_vmlinux_btf_id(const char *name, |
9176 | enum bpf_attach_type attach_type) | |
12a8654b | 9177 | { |
a6ed02ca | 9178 | struct btf *btf; |
3521ffa2 | 9179 | int err; |
12a8654b | 9180 | |
a710eed3 | 9181 | btf = btf__load_vmlinux_btf(); |
e9fc3ce9 AN |
9182 | err = libbpf_get_error(btf); |
9183 | if (err) { | |
12a8654b | 9184 | pr_warn("vmlinux BTF is not found\n"); |
e9fc3ce9 | 9185 | return libbpf_err(err); |
12a8654b AS |
9186 | } |
9187 | ||
91abb4a6 AN |
9188 | err = find_attach_btf_id(btf, name, attach_type); |
9189 | if (err <= 0) | |
9190 | pr_warn("%s is not found in vmlinux BTF\n", name); | |
9191 | ||
3521ffa2 | 9192 | btf__free(btf); |
e9fc3ce9 | 9193 | return libbpf_err(err); |
b8c54ea4 AS |
9194 | } |
9195 | ||
e7bf94db AS |
9196 | static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd) |
9197 | { | |
813847a3 | 9198 | struct bpf_prog_info info; |
ebc7b50a | 9199 | __u32 info_len = sizeof(info); |
6cc93e2f | 9200 | struct btf *btf; |
6d2d73cd | 9201 | int err; |
e7bf94db | 9202 | |
813847a3 | 9203 | memset(&info, 0, info_len); |
629dfc66 | 9204 | err = bpf_prog_get_info_by_fd(attach_prog_fd, &info, &info_len); |
e9fc3ce9 | 9205 | if (err) { |
629dfc66 | 9206 | pr_warn("failed bpf_prog_get_info_by_fd for FD %d: %d\n", |
ebc7b50a | 9207 | attach_prog_fd, err); |
e9fc3ce9 | 9208 | return err; |
e7bf94db | 9209 | } |
6d2d73cd QM |
9210 | |
9211 | err = -EINVAL; | |
ebc7b50a | 9212 | if (!info.btf_id) { |
e7bf94db AS |
9213 | pr_warn("The target program doesn't have BTF\n"); |
9214 | goto out; | |
9215 | } | |
ebc7b50a DM |
9216 | btf = btf__load_from_kernel_by_id(info.btf_id); |
9217 | err = libbpf_get_error(btf); | |
9218 | if (err) { | |
9219 | pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err); | |
e7bf94db AS |
9220 | goto out; |
9221 | } | |
9222 | err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC); | |
9223 | btf__free(btf); | |
9224 | if (err <= 0) { | |
9225 | pr_warn("%s is not found in prog's BTF\n", name); | |
9226 | goto out; | |
9227 | } | |
9228 | out: | |
e7bf94db AS |
9229 | return err; |
9230 | } | |
9231 | ||
91abb4a6 AN |
9232 | static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name, |
9233 | enum bpf_attach_type attach_type, | |
9234 | int *btf_obj_fd, int *btf_type_id) | |
9235 | { | |
9236 | int ret, i; | |
9237 | ||
9238 | ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type); | |
9239 | if (ret > 0) { | |
9240 | *btf_obj_fd = 0; /* vmlinux BTF */ | |
9241 | *btf_type_id = ret; | |
9242 | return 0; | |
9243 | } | |
9244 | if (ret != -ENOENT) | |
9245 | return ret; | |
9246 | ||
9247 | ret = load_module_btfs(obj); | |
9248 | if (ret) | |
9249 | return ret; | |
9250 | ||
9251 | for (i = 0; i < obj->btf_module_cnt; i++) { | |
9252 | const struct module_btf *mod = &obj->btf_modules[i]; | |
9253 | ||
9254 | ret = find_attach_btf_id(mod->btf, attach_name, attach_type); | |
9255 | if (ret > 0) { | |
9256 | *btf_obj_fd = mod->fd; | |
9257 | *btf_type_id = ret; | |
9258 | return 0; | |
9259 | } | |
9260 | if (ret == -ENOENT) | |
9261 | continue; | |
9262 | ||
9263 | return ret; | |
9264 | } | |
9265 | ||
9266 | return -ESRCH; | |
9267 | } | |
9268 | ||
15ea31fa AN |
9269 | static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name, |
9270 | int *btf_obj_fd, int *btf_type_id) | |
b8c54ea4 | 9271 | { |
a6ed02ca KS |
9272 | enum bpf_attach_type attach_type = prog->expected_attach_type; |
9273 | __u32 attach_prog_fd = prog->attach_prog_fd; | |
b6291a6f | 9274 | int err = 0; |
b8c54ea4 | 9275 | |
91abb4a6 | 9276 | /* BPF program's BTF ID */ |
749c202c AN |
9277 | if (prog->type == BPF_PROG_TYPE_EXT || attach_prog_fd) { |
9278 | if (!attach_prog_fd) { | |
9279 | pr_warn("prog '%s': attach program FD is not set\n", prog->name); | |
9280 | return -EINVAL; | |
9281 | } | |
91abb4a6 AN |
9282 | err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd); |
9283 | if (err < 0) { | |
749c202c AN |
9284 | pr_warn("prog '%s': failed to find BPF program (FD %d) BTF ID for '%s': %d\n", |
9285 | prog->name, attach_prog_fd, attach_name, err); | |
91abb4a6 AN |
9286 | return err; |
9287 | } | |
9288 | *btf_obj_fd = 0; | |
9289 | *btf_type_id = err; | |
9290 | return 0; | |
9291 | } | |
9292 | ||
9293 | /* kernel/module BTF ID */ | |
67234743 AS |
9294 | if (prog->obj->gen_loader) { |
9295 | bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type); | |
9296 | *btf_obj_fd = 0; | |
9297 | *btf_type_id = 1; | |
9298 | } else { | |
9299 | err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id); | |
9300 | } | |
91abb4a6 | 9301 | if (err) { |
749c202c AN |
9302 | pr_warn("prog '%s': failed to find kernel BTF type ID of '%s': %d\n", |
9303 | prog->name, attach_name, err); | |
b8c54ea4 | 9304 | return err; |
12a8654b | 9305 | } |
91abb4a6 | 9306 | return 0; |
12a8654b AS |
9307 | } |
9308 | ||
956b620f AI |
9309 | int libbpf_attach_type_by_name(const char *name, |
9310 | enum bpf_attach_type *attach_type) | |
9311 | { | |
c76e4c22 | 9312 | char *type_names; |
b6291a6f | 9313 | const struct bpf_sec_def *sec_def; |
956b620f AI |
9314 | |
9315 | if (!name) | |
e9fc3ce9 | 9316 | return libbpf_err(-EINVAL); |
956b620f | 9317 | |
b6291a6f AN |
9318 | sec_def = find_sec_def(name); |
9319 | if (!sec_def) { | |
9320 | pr_debug("failed to guess attach type based on ELF section name '%s'\n", name); | |
9321 | type_names = libbpf_get_type_names(true); | |
9322 | if (type_names != NULL) { | |
9323 | pr_debug("attachable section(type) names are:%s\n", type_names); | |
9324 | free(type_names); | |
9325 | } | |
9326 | ||
9327 | return libbpf_err(-EINVAL); | |
c76e4c22 TS |
9328 | } |
9329 | ||
4fa5bcfe | 9330 | if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load) |
15ea31fa AN |
9331 | return libbpf_err(-EINVAL); |
9332 | if (!(sec_def->cookie & SEC_ATTACHABLE)) | |
b6291a6f AN |
9333 | return libbpf_err(-EINVAL); |
9334 | ||
9335 | *attach_type = sec_def->expected_attach_type; | |
9336 | return 0; | |
956b620f AI |
9337 | } |
9338 | ||
a324aae3 | 9339 | int bpf_map__fd(const struct bpf_map *map) |
9d759a9b | 9340 | { |
e9fc3ce9 | 9341 | return map ? map->fd : libbpf_err(-EINVAL); |
9d759a9b WN |
9342 | } |
9343 | ||
aed65917 AN |
9344 | static bool map_uses_real_name(const struct bpf_map *map) |
9345 | { | |
9346 | /* Since libbpf started to support custom .data.* and .rodata.* maps, | |
9347 | * their user-visible name differs from kernel-visible name. Users see | |
9348 | * such map's corresponding ELF section name as a map name. | |
9349 | * This check distinguishes .data/.rodata from .data.* and .rodata.* | |
9350 | * maps to know which name has to be returned to the user. | |
9351 | */ | |
9352 | if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0) | |
9353 | return true; | |
9354 | if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0) | |
9355 | return true; | |
9356 | return false; | |
9357 | } | |
9358 | ||
a324aae3 | 9359 | const char *bpf_map__name(const struct bpf_map *map) |
561bbcca | 9360 | { |
aed65917 AN |
9361 | if (!map) |
9362 | return NULL; | |
9363 | ||
9364 | if (map_uses_real_name(map)) | |
9365 | return map->real_name; | |
9366 | ||
9367 | return map->name; | |
561bbcca WN |
9368 | } |
9369 | ||
1bdb6c9a AN |
9370 | enum bpf_map_type bpf_map__type(const struct bpf_map *map) |
9371 | { | |
9372 | return map->def.type; | |
9373 | } | |
9374 | ||
9375 | int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type) | |
9376 | { | |
9377 | if (map->fd >= 0) | |
e9fc3ce9 | 9378 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9379 | map->def.type = type; |
9380 | return 0; | |
9381 | } | |
9382 | ||
9383 | __u32 bpf_map__map_flags(const struct bpf_map *map) | |
9384 | { | |
9385 | return map->def.map_flags; | |
9386 | } | |
9387 | ||
9388 | int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags) | |
9389 | { | |
9390 | if (map->fd >= 0) | |
e9fc3ce9 | 9391 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9392 | map->def.map_flags = flags; |
9393 | return 0; | |
9394 | } | |
9395 | ||
47512102 JK |
9396 | __u64 bpf_map__map_extra(const struct bpf_map *map) |
9397 | { | |
9398 | return map->map_extra; | |
9399 | } | |
9400 | ||
9401 | int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra) | |
9402 | { | |
9403 | if (map->fd >= 0) | |
9404 | return libbpf_err(-EBUSY); | |
9405 | map->map_extra = map_extra; | |
9406 | return 0; | |
9407 | } | |
9408 | ||
1bdb6c9a AN |
9409 | __u32 bpf_map__numa_node(const struct bpf_map *map) |
9410 | { | |
9411 | return map->numa_node; | |
9412 | } | |
9413 | ||
9414 | int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node) | |
9415 | { | |
9416 | if (map->fd >= 0) | |
e9fc3ce9 | 9417 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9418 | map->numa_node = numa_node; |
9419 | return 0; | |
9420 | } | |
9421 | ||
9422 | __u32 bpf_map__key_size(const struct bpf_map *map) | |
9423 | { | |
9424 | return map->def.key_size; | |
9425 | } | |
9426 | ||
9427 | int bpf_map__set_key_size(struct bpf_map *map, __u32 size) | |
9428 | { | |
9429 | if (map->fd >= 0) | |
e9fc3ce9 | 9430 | return libbpf_err(-EBUSY); |
1bdb6c9a AN |
9431 | map->def.key_size = size; |
9432 | return 0; | |
9433 | } | |
9434 | ||
9435 | __u32 bpf_map__value_size(const struct bpf_map *map) | |
9436 | { | |
9437 | return map->def.value_size; | |
9438 | } | |
9439 | ||
9d0a2331 JK |
9440 | static int map_btf_datasec_resize(struct bpf_map *map, __u32 size) |
9441 | { | |
9442 | struct btf *btf; | |
9443 | struct btf_type *datasec_type, *var_type; | |
9444 | struct btf_var_secinfo *var; | |
9445 | const struct btf_type *array_type; | |
9446 | const struct btf_array *array; | |
4c857a71 JK |
9447 | int vlen, element_sz, new_array_id; |
9448 | __u32 nr_elements; | |
9d0a2331 JK |
9449 | |
9450 | /* check btf existence */ | |
9451 | btf = bpf_object__btf(map->obj); | |
9452 | if (!btf) | |
9453 | return -ENOENT; | |
9454 | ||
9455 | /* verify map is datasec */ | |
9456 | datasec_type = btf_type_by_id(btf, bpf_map__btf_value_type_id(map)); | |
9457 | if (!btf_is_datasec(datasec_type)) { | |
9458 | pr_warn("map '%s': cannot be resized, map value type is not a datasec\n", | |
9459 | bpf_map__name(map)); | |
9460 | return -EINVAL; | |
9461 | } | |
9462 | ||
9463 | /* verify datasec has at least one var */ | |
9464 | vlen = btf_vlen(datasec_type); | |
9465 | if (vlen == 0) { | |
9466 | pr_warn("map '%s': cannot be resized, map value datasec is empty\n", | |
9467 | bpf_map__name(map)); | |
9468 | return -EINVAL; | |
9469 | } | |
9470 | ||
9471 | /* verify last var in the datasec is an array */ | |
9472 | var = &btf_var_secinfos(datasec_type)[vlen - 1]; | |
9473 | var_type = btf_type_by_id(btf, var->type); | |
9474 | array_type = skip_mods_and_typedefs(btf, var_type->type, NULL); | |
9475 | if (!btf_is_array(array_type)) { | |
9476 | pr_warn("map '%s': cannot be resized, last var must be an array\n", | |
9477 | bpf_map__name(map)); | |
9478 | return -EINVAL; | |
9479 | } | |
9480 | ||
9481 | /* verify request size aligns with array */ | |
9482 | array = btf_array(array_type); | |
9483 | element_sz = btf__resolve_size(btf, array->type); | |
9484 | if (element_sz <= 0 || (size - var->offset) % element_sz != 0) { | |
9485 | pr_warn("map '%s': cannot be resized, element size (%d) doesn't align with new total size (%u)\n", | |
9486 | bpf_map__name(map), element_sz, size); | |
9487 | return -EINVAL; | |
9488 | } | |
9489 | ||
9490 | /* create a new array based on the existing array, but with new length */ | |
9491 | nr_elements = (size - var->offset) / element_sz; | |
9492 | new_array_id = btf__add_array(btf, array->index_type, array->type, nr_elements); | |
9493 | if (new_array_id < 0) | |
9494 | return new_array_id; | |
9495 | ||
9496 | /* adding a new btf type invalidates existing pointers to btf objects, | |
9497 | * so refresh pointers before proceeding | |
9498 | */ | |
9499 | datasec_type = btf_type_by_id(btf, map->btf_value_type_id); | |
9500 | var = &btf_var_secinfos(datasec_type)[vlen - 1]; | |
9501 | var_type = btf_type_by_id(btf, var->type); | |
9502 | ||
9503 | /* finally update btf info */ | |
9504 | datasec_type->size = size; | |
9505 | var->size = size - var->offset; | |
9506 | var_type->type = new_array_id; | |
9507 | ||
9508 | return 0; | |
9509 | } | |
9510 | ||
1bdb6c9a AN |
9511 | int bpf_map__set_value_size(struct bpf_map *map, __u32 size) |
9512 | { | |
9513 | if (map->fd >= 0) | |
e9fc3ce9 | 9514 | return libbpf_err(-EBUSY); |
9d0a2331 JK |
9515 | |
9516 | if (map->mmaped) { | |
9517 | int err; | |
9518 | size_t mmap_old_sz, mmap_new_sz; | |
9519 | ||
9520 | mmap_old_sz = bpf_map_mmap_sz(map->def.value_size, map->def.max_entries); | |
9521 | mmap_new_sz = bpf_map_mmap_sz(size, map->def.max_entries); | |
9522 | err = bpf_map_mmap_resize(map, mmap_old_sz, mmap_new_sz); | |
9523 | if (err) { | |
9524 | pr_warn("map '%s': failed to resize memory-mapped region: %d\n", | |
9525 | bpf_map__name(map), err); | |
9526 | return err; | |
9527 | } | |
9528 | err = map_btf_datasec_resize(map, size); | |
9529 | if (err && err != -ENOENT) { | |
9530 | pr_warn("map '%s': failed to adjust resized BTF, clearing BTF key/value info: %d\n", | |
9531 | bpf_map__name(map), err); | |
9532 | map->btf_value_type_id = 0; | |
9533 | map->btf_key_type_id = 0; | |
9534 | } | |
9535 | } | |
9536 | ||
1bdb6c9a AN |
9537 | map->def.value_size = size; |
9538 | return 0; | |
9539 | } | |
9540 | ||
5b891af7 | 9541 | __u32 bpf_map__btf_key_type_id(const struct bpf_map *map) |
8a138aed | 9542 | { |
61746dbe | 9543 | return map ? map->btf_key_type_id : 0; |
8a138aed MKL |
9544 | } |
9545 | ||
5b891af7 | 9546 | __u32 bpf_map__btf_value_type_id(const struct bpf_map *map) |
8a138aed | 9547 | { |
61746dbe | 9548 | return map ? map->btf_value_type_id : 0; |
8a138aed MKL |
9549 | } |
9550 | ||
e2842be5 THJ |
9551 | int bpf_map__set_initial_value(struct bpf_map *map, |
9552 | const void *data, size_t size) | |
9553 | { | |
9554 | if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG || | |
9555 | size != map->def.value_size || map->fd >= 0) | |
e9fc3ce9 | 9556 | return libbpf_err(-EINVAL); |
e2842be5 THJ |
9557 | |
9558 | memcpy(map->mmaped, data, size); | |
9559 | return 0; | |
9560 | } | |
9561 | ||
9d0a2331 | 9562 | void *bpf_map__initial_value(struct bpf_map *map, size_t *psize) |
7723256b AS |
9563 | { |
9564 | if (!map->mmaped) | |
9565 | return NULL; | |
9566 | *psize = map->def.value_size; | |
9567 | return map->mmaped; | |
9568 | } | |
9569 | ||
a324aae3 | 9570 | bool bpf_map__is_internal(const struct bpf_map *map) |
d859900c DB |
9571 | { |
9572 | return map->libbpf_type != LIBBPF_MAP_UNSPEC; | |
9573 | } | |
9574 | ||
1bdb6c9a AN |
9575 | __u32 bpf_map__ifindex(const struct bpf_map *map) |
9576 | { | |
9577 | return map->map_ifindex; | |
9578 | } | |
9579 | ||
9580 | int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex) | |
9aba3613 | 9581 | { |
1bdb6c9a | 9582 | if (map->fd >= 0) |
e9fc3ce9 | 9583 | return libbpf_err(-EBUSY); |
9aba3613 | 9584 | map->map_ifindex = ifindex; |
1bdb6c9a | 9585 | return 0; |
9aba3613 JK |
9586 | } |
9587 | ||
addb9fc9 NS |
9588 | int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd) |
9589 | { | |
9590 | if (!bpf_map_type__is_map_in_map(map->def.type)) { | |
be18010e | 9591 | pr_warn("error: unsupported map type\n"); |
e9fc3ce9 | 9592 | return libbpf_err(-EINVAL); |
addb9fc9 NS |
9593 | } |
9594 | if (map->inner_map_fd != -1) { | |
be18010e | 9595 | pr_warn("error: inner_map_fd already specified\n"); |
e9fc3ce9 | 9596 | return libbpf_err(-EINVAL); |
addb9fc9 | 9597 | } |
8f7b239e AN |
9598 | if (map->inner_map) { |
9599 | bpf_map__destroy(map->inner_map); | |
9600 | zfree(&map->inner_map); | |
9601 | } | |
addb9fc9 NS |
9602 | map->inner_map_fd = fd; |
9603 | return 0; | |
9604 | } | |
9605 | ||
0c19a9fb | 9606 | static struct bpf_map * |
a324aae3 | 9607 | __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i) |
9d759a9b | 9608 | { |
0c19a9fb | 9609 | ssize_t idx; |
9d759a9b WN |
9610 | struct bpf_map *s, *e; |
9611 | ||
9612 | if (!obj || !obj->maps) | |
e9fc3ce9 | 9613 | return errno = EINVAL, NULL; |
9d759a9b WN |
9614 | |
9615 | s = obj->maps; | |
9616 | e = obj->maps + obj->nr_maps; | |
9617 | ||
0c19a9fb | 9618 | if ((m < s) || (m >= e)) { |
be18010e KW |
9619 | pr_warn("error in %s: map handler doesn't belong to object\n", |
9620 | __func__); | |
e9fc3ce9 | 9621 | return errno = EINVAL, NULL; |
9d759a9b WN |
9622 | } |
9623 | ||
0c19a9fb SF |
9624 | idx = (m - obj->maps) + i; |
9625 | if (idx >= obj->nr_maps || idx < 0) | |
9d759a9b WN |
9626 | return NULL; |
9627 | return &obj->maps[idx]; | |
9628 | } | |
561bbcca | 9629 | |
2088a3a7 HC |
9630 | struct bpf_map * |
9631 | bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev) | |
0c19a9fb SF |
9632 | { |
9633 | if (prev == NULL) | |
9634 | return obj->maps; | |
9635 | ||
9636 | return __bpf_map__iter(prev, obj, 1); | |
9637 | } | |
9638 | ||
2088a3a7 HC |
9639 | struct bpf_map * |
9640 | bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next) | |
0c19a9fb SF |
9641 | { |
9642 | if (next == NULL) { | |
9643 | if (!obj->nr_maps) | |
9644 | return NULL; | |
9645 | return obj->maps + obj->nr_maps - 1; | |
9646 | } | |
9647 | ||
9648 | return __bpf_map__iter(next, obj, -1); | |
9649 | } | |
9650 | ||
561bbcca | 9651 | struct bpf_map * |
a324aae3 | 9652 | bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name) |
561bbcca WN |
9653 | { |
9654 | struct bpf_map *pos; | |
9655 | ||
f74a53d9 | 9656 | bpf_object__for_each_map(pos, obj) { |
26071635 AN |
9657 | /* if it's a special internal map name (which always starts |
9658 | * with dot) then check if that special name matches the | |
9659 | * real map name (ELF section name) | |
9660 | */ | |
9661 | if (name[0] == '.') { | |
9662 | if (pos->real_name && strcmp(pos->real_name, name) == 0) | |
9663 | return pos; | |
9664 | continue; | |
9665 | } | |
9666 | /* otherwise map name has to be an exact match */ | |
aed65917 AN |
9667 | if (map_uses_real_name(pos)) { |
9668 | if (strcmp(pos->real_name, name) == 0) | |
9669 | return pos; | |
9670 | continue; | |
9671 | } | |
9672 | if (strcmp(pos->name, name) == 0) | |
561bbcca WN |
9673 | return pos; |
9674 | } | |
e9fc3ce9 | 9675 | return errno = ENOENT, NULL; |
561bbcca | 9676 | } |
5a6acad1 | 9677 | |
f3cea32d | 9678 | int |
a324aae3 | 9679 | bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name) |
f3cea32d MF |
9680 | { |
9681 | return bpf_map__fd(bpf_object__find_map_by_name(obj, name)); | |
9682 | } | |
9683 | ||
737d0646 AN |
9684 | static int validate_map_op(const struct bpf_map *map, size_t key_sz, |
9685 | size_t value_sz, bool check_value_sz) | |
9686 | { | |
9687 | if (map->fd <= 0) | |
9688 | return -ENOENT; | |
9689 | ||
9690 | if (map->def.key_size != key_sz) { | |
9691 | pr_warn("map '%s': unexpected key size %zu provided, expected %u\n", | |
9692 | map->name, key_sz, map->def.key_size); | |
9693 | return -EINVAL; | |
9694 | } | |
9695 | ||
9696 | if (!check_value_sz) | |
9697 | return 0; | |
9698 | ||
9699 | switch (map->def.type) { | |
9700 | case BPF_MAP_TYPE_PERCPU_ARRAY: | |
9701 | case BPF_MAP_TYPE_PERCPU_HASH: | |
9702 | case BPF_MAP_TYPE_LRU_PERCPU_HASH: | |
9703 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: { | |
9704 | int num_cpu = libbpf_num_possible_cpus(); | |
9705 | size_t elem_sz = roundup(map->def.value_size, 8); | |
9706 | ||
9707 | if (value_sz != num_cpu * elem_sz) { | |
9708 | pr_warn("map '%s': unexpected value size %zu provided for per-CPU map, expected %d * %zu = %zd\n", | |
9709 | map->name, value_sz, num_cpu, elem_sz, num_cpu * elem_sz); | |
9710 | return -EINVAL; | |
9711 | } | |
9712 | break; | |
9713 | } | |
9714 | default: | |
9715 | if (map->def.value_size != value_sz) { | |
9716 | pr_warn("map '%s': unexpected value size %zu provided, expected %u\n", | |
9717 | map->name, value_sz, map->def.value_size); | |
9718 | return -EINVAL; | |
9719 | } | |
9720 | break; | |
9721 | } | |
9722 | return 0; | |
9723 | } | |
9724 | ||
9725 | int bpf_map__lookup_elem(const struct bpf_map *map, | |
9726 | const void *key, size_t key_sz, | |
9727 | void *value, size_t value_sz, __u64 flags) | |
9728 | { | |
9729 | int err; | |
9730 | ||
9731 | err = validate_map_op(map, key_sz, value_sz, true); | |
9732 | if (err) | |
9733 | return libbpf_err(err); | |
9734 | ||
9735 | return bpf_map_lookup_elem_flags(map->fd, key, value, flags); | |
9736 | } | |
9737 | ||
9738 | int bpf_map__update_elem(const struct bpf_map *map, | |
9739 | const void *key, size_t key_sz, | |
9740 | const void *value, size_t value_sz, __u64 flags) | |
9741 | { | |
9742 | int err; | |
9743 | ||
9744 | err = validate_map_op(map, key_sz, value_sz, true); | |
9745 | if (err) | |
9746 | return libbpf_err(err); | |
9747 | ||
9748 | return bpf_map_update_elem(map->fd, key, value, flags); | |
9749 | } | |
9750 | ||
9751 | int bpf_map__delete_elem(const struct bpf_map *map, | |
9752 | const void *key, size_t key_sz, __u64 flags) | |
9753 | { | |
9754 | int err; | |
9755 | ||
9756 | err = validate_map_op(map, key_sz, 0, false /* check_value_sz */); | |
9757 | if (err) | |
9758 | return libbpf_err(err); | |
9759 | ||
9760 | return bpf_map_delete_elem_flags(map->fd, key, flags); | |
9761 | } | |
9762 | ||
9763 | int bpf_map__lookup_and_delete_elem(const struct bpf_map *map, | |
9764 | const void *key, size_t key_sz, | |
9765 | void *value, size_t value_sz, __u64 flags) | |
9766 | { | |
9767 | int err; | |
9768 | ||
9769 | err = validate_map_op(map, key_sz, value_sz, true); | |
9770 | if (err) | |
9771 | return libbpf_err(err); | |
9772 | ||
9773 | return bpf_map_lookup_and_delete_elem_flags(map->fd, key, value, flags); | |
9774 | } | |
9775 | ||
9776 | int bpf_map__get_next_key(const struct bpf_map *map, | |
9777 | const void *cur_key, void *next_key, size_t key_sz) | |
9778 | { | |
9779 | int err; | |
9780 | ||
9781 | err = validate_map_op(map, key_sz, 0, false /* check_value_sz */); | |
9782 | if (err) | |
9783 | return libbpf_err(err); | |
9784 | ||
9785 | return bpf_map_get_next_key(map->fd, cur_key, next_key); | |
9786 | } | |
9787 | ||
e28ff1a8 JS |
9788 | long libbpf_get_error(const void *ptr) |
9789 | { | |
e9fc3ce9 AN |
9790 | if (!IS_ERR_OR_NULL(ptr)) |
9791 | return 0; | |
9792 | ||
9793 | if (IS_ERR(ptr)) | |
9794 | errno = -PTR_ERR(ptr); | |
9795 | ||
9796 | /* If ptr == NULL, then errno should be already set by the failing | |
9797 | * API, because libbpf never returns NULL on success and it now always | |
9798 | * sets errno on error. So no extra errno handling for ptr == NULL | |
9799 | * case. | |
9800 | */ | |
9801 | return -errno; | |
e28ff1a8 | 9802 | } |
6f6d33f3 | 9803 | |
cc4f864b AN |
9804 | /* Replace link's underlying BPF program with the new one */ |
9805 | int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog) | |
9806 | { | |
e9fc3ce9 | 9807 | int ret; |
c139e40a | 9808 | |
e9fc3ce9 AN |
9809 | ret = bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL); |
9810 | return libbpf_err_errno(ret); | |
cc4f864b AN |
9811 | } |
9812 | ||
d6958706 AN |
9813 | /* Release "ownership" of underlying BPF resource (typically, BPF program |
9814 | * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected | |
9815 | * link, when destructed through bpf_link__destroy() call won't attempt to | |
9816 | * detach/unregisted that BPF resource. This is useful in situations where, | |
9817 | * say, attached BPF program has to outlive userspace program that attached it | |
9818 | * in the system. Depending on type of BPF program, though, there might be | |
9819 | * additional steps (like pinning BPF program in BPF FS) necessary to ensure | |
9820 | * exit of userspace program doesn't trigger automatic detachment and clean up | |
9821 | * inside the kernel. | |
9822 | */ | |
9823 | void bpf_link__disconnect(struct bpf_link *link) | |
9824 | { | |
9825 | link->disconnected = true; | |
9826 | } | |
9827 | ||
1c2e9efc AN |
9828 | int bpf_link__destroy(struct bpf_link *link) |
9829 | { | |
d6958706 | 9830 | int err = 0; |
1c2e9efc | 9831 | |
50450fc7 | 9832 | if (IS_ERR_OR_NULL(link)) |
1c2e9efc AN |
9833 | return 0; |
9834 | ||
d6958706 AN |
9835 | if (!link->disconnected && link->detach) |
9836 | err = link->detach(link); | |
c016b68e AN |
9837 | if (link->pin_path) |
9838 | free(link->pin_path); | |
d88b71d4 AN |
9839 | if (link->dealloc) |
9840 | link->dealloc(link); | |
9841 | else | |
9842 | free(link); | |
1c2e9efc | 9843 | |
e9fc3ce9 | 9844 | return libbpf_err(err); |
1c2e9efc AN |
9845 | } |
9846 | ||
c016b68e AN |
9847 | int bpf_link__fd(const struct bpf_link *link) |
9848 | { | |
9849 | return link->fd; | |
9850 | } | |
9851 | ||
9852 | const char *bpf_link__pin_path(const struct bpf_link *link) | |
9853 | { | |
9854 | return link->pin_path; | |
9855 | } | |
9856 | ||
9857 | static int bpf_link__detach_fd(struct bpf_link *link) | |
9858 | { | |
e9fc3ce9 | 9859 | return libbpf_err_errno(close(link->fd)); |
c016b68e AN |
9860 | } |
9861 | ||
9862 | struct bpf_link *bpf_link__open(const char *path) | |
9863 | { | |
9864 | struct bpf_link *link; | |
9865 | int fd; | |
9866 | ||
9867 | fd = bpf_obj_get(path); | |
9868 | if (fd < 0) { | |
9869 | fd = -errno; | |
9870 | pr_warn("failed to open link at %s: %d\n", path, fd); | |
e9fc3ce9 | 9871 | return libbpf_err_ptr(fd); |
c016b68e AN |
9872 | } |
9873 | ||
9874 | link = calloc(1, sizeof(*link)); | |
9875 | if (!link) { | |
9876 | close(fd); | |
e9fc3ce9 | 9877 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
9878 | } |
9879 | link->detach = &bpf_link__detach_fd; | |
9880 | link->fd = fd; | |
9881 | ||
9882 | link->pin_path = strdup(path); | |
9883 | if (!link->pin_path) { | |
9884 | bpf_link__destroy(link); | |
e9fc3ce9 | 9885 | return libbpf_err_ptr(-ENOMEM); |
c016b68e AN |
9886 | } |
9887 | ||
9888 | return link; | |
9889 | } | |
9890 | ||
2e49527e AN |
9891 | int bpf_link__detach(struct bpf_link *link) |
9892 | { | |
9893 | return bpf_link_detach(link->fd) ? -errno : 0; | |
9894 | } | |
9895 | ||
c016b68e AN |
9896 | int bpf_link__pin(struct bpf_link *link, const char *path) |
9897 | { | |
9898 | int err; | |
9899 | ||
9900 | if (link->pin_path) | |
e9fc3ce9 | 9901 | return libbpf_err(-EBUSY); |
c016b68e AN |
9902 | err = make_parent_dir(path); |
9903 | if (err) | |
e9fc3ce9 | 9904 | return libbpf_err(err); |
c016b68e AN |
9905 | err = check_path(path); |
9906 | if (err) | |
e9fc3ce9 | 9907 | return libbpf_err(err); |
c016b68e AN |
9908 | |
9909 | link->pin_path = strdup(path); | |
9910 | if (!link->pin_path) | |
e9fc3ce9 | 9911 | return libbpf_err(-ENOMEM); |
c016b68e AN |
9912 | |
9913 | if (bpf_obj_pin(link->fd, link->pin_path)) { | |
9914 | err = -errno; | |
9915 | zfree(&link->pin_path); | |
e9fc3ce9 | 9916 | return libbpf_err(err); |
c016b68e AN |
9917 | } |
9918 | ||
9919 | pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path); | |
9920 | return 0; | |
9921 | } | |
9922 | ||
9923 | int bpf_link__unpin(struct bpf_link *link) | |
9924 | { | |
9925 | int err; | |
9926 | ||
9927 | if (!link->pin_path) | |
e9fc3ce9 | 9928 | return libbpf_err(-EINVAL); |
c016b68e AN |
9929 | |
9930 | err = unlink(link->pin_path); | |
9931 | if (err != 0) | |
af0efa05 | 9932 | return -errno; |
c016b68e AN |
9933 | |
9934 | pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path); | |
9935 | zfree(&link->pin_path); | |
9936 | return 0; | |
9937 | } | |
63f2f5ee | 9938 | |
668ace0e AN |
9939 | struct bpf_link_perf { |
9940 | struct bpf_link link; | |
9941 | int perf_event_fd; | |
ca304b40 RDT |
9942 | /* legacy kprobe support: keep track of probe identifier and type */ |
9943 | char *legacy_probe_name; | |
46ed5fc3 | 9944 | bool legacy_is_kprobe; |
ca304b40 | 9945 | bool legacy_is_retprobe; |
668ace0e AN |
9946 | }; |
9947 | ||
46ed5fc3 | 9948 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe); |
cc10623c | 9949 | static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe); |
46ed5fc3 | 9950 | |
668ace0e | 9951 | static int bpf_link_perf_detach(struct bpf_link *link) |
63f2f5ee | 9952 | { |
668ace0e AN |
9953 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); |
9954 | int err = 0; | |
63f2f5ee | 9955 | |
668ace0e | 9956 | if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0) |
63f2f5ee AN |
9957 | err = -errno; |
9958 | ||
668ace0e AN |
9959 | if (perf_link->perf_event_fd != link->fd) |
9960 | close(perf_link->perf_event_fd); | |
c016b68e | 9961 | close(link->fd); |
668ace0e | 9962 | |
cc10623c | 9963 | /* legacy uprobe/kprobe needs to be removed after perf event fd closure */ |
46ed5fc3 AN |
9964 | if (perf_link->legacy_probe_name) { |
9965 | if (perf_link->legacy_is_kprobe) { | |
9966 | err = remove_kprobe_event_legacy(perf_link->legacy_probe_name, | |
9967 | perf_link->legacy_is_retprobe); | |
cc10623c AN |
9968 | } else { |
9969 | err = remove_uprobe_event_legacy(perf_link->legacy_probe_name, | |
9970 | perf_link->legacy_is_retprobe); | |
46ed5fc3 AN |
9971 | } |
9972 | } | |
ca304b40 RDT |
9973 | |
9974 | return err; | |
63f2f5ee AN |
9975 | } |
9976 | ||
668ace0e AN |
9977 | static void bpf_link_perf_dealloc(struct bpf_link *link) |
9978 | { | |
9979 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
9980 | ||
ca304b40 | 9981 | free(perf_link->legacy_probe_name); |
668ace0e AN |
9982 | free(perf_link); |
9983 | } | |
9984 | ||
942025c9 | 9985 | struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd, |
47faff37 | 9986 | const struct bpf_perf_event_opts *opts) |
63f2f5ee AN |
9987 | { |
9988 | char errmsg[STRERR_BUFSIZE]; | |
668ace0e AN |
9989 | struct bpf_link_perf *link; |
9990 | int prog_fd, link_fd = -1, err; | |
f8b299bc | 9991 | bool force_ioctl_attach; |
63f2f5ee | 9992 | |
47faff37 AN |
9993 | if (!OPTS_VALID(opts, bpf_perf_event_opts)) |
9994 | return libbpf_err_ptr(-EINVAL); | |
9995 | ||
63f2f5ee | 9996 | if (pfd < 0) { |
52109584 AN |
9997 | pr_warn("prog '%s': invalid perf event FD %d\n", |
9998 | prog->name, pfd); | |
e9fc3ce9 | 9999 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
10000 | } |
10001 | prog_fd = bpf_program__fd(prog); | |
10002 | if (prog_fd < 0) { | |
52109584 AN |
10003 | pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n", |
10004 | prog->name); | |
e9fc3ce9 | 10005 | return libbpf_err_ptr(-EINVAL); |
63f2f5ee AN |
10006 | } |
10007 | ||
d6958706 | 10008 | link = calloc(1, sizeof(*link)); |
63f2f5ee | 10009 | if (!link) |
e9fc3ce9 | 10010 | return libbpf_err_ptr(-ENOMEM); |
668ace0e AN |
10011 | link->link.detach = &bpf_link_perf_detach; |
10012 | link->link.dealloc = &bpf_link_perf_dealloc; | |
10013 | link->perf_event_fd = pfd; | |
63f2f5ee | 10014 | |
f8b299bc MD |
10015 | force_ioctl_attach = OPTS_GET(opts, force_ioctl_attach, false); |
10016 | if (kernel_supports(prog->obj, FEAT_PERF_LINK) && !force_ioctl_attach) { | |
47faff37 AN |
10017 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts, |
10018 | .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0)); | |
10019 | ||
10020 | link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts); | |
668ace0e AN |
10021 | if (link_fd < 0) { |
10022 | err = -errno; | |
10023 | pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n", | |
10024 | prog->name, pfd, | |
10025 | err, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10026 | goto err_out; | |
10027 | } | |
10028 | link->link.fd = link_fd; | |
10029 | } else { | |
47faff37 AN |
10030 | if (OPTS_GET(opts, bpf_cookie, 0)) { |
10031 | pr_warn("prog '%s': user context value is not supported\n", prog->name); | |
10032 | err = -EOPNOTSUPP; | |
10033 | goto err_out; | |
10034 | } | |
10035 | ||
668ace0e AN |
10036 | if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) { |
10037 | err = -errno; | |
10038 | pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n", | |
10039 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10040 | if (err == -EPROTO) | |
10041 | pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n", | |
10042 | prog->name, pfd); | |
10043 | goto err_out; | |
10044 | } | |
10045 | link->link.fd = pfd; | |
63f2f5ee AN |
10046 | } |
10047 | if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
10048 | err = -errno; | |
668ace0e | 10049 | pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n", |
52109584 | 10050 | prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
668ace0e | 10051 | goto err_out; |
63f2f5ee | 10052 | } |
668ace0e AN |
10053 | |
10054 | return &link->link; | |
10055 | err_out: | |
10056 | if (link_fd >= 0) | |
10057 | close(link_fd); | |
10058 | free(link); | |
10059 | return libbpf_err_ptr(err); | |
63f2f5ee AN |
10060 | } |
10061 | ||
942025c9 | 10062 | struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd) |
47faff37 AN |
10063 | { |
10064 | return bpf_program__attach_perf_event_opts(prog, pfd, NULL); | |
10065 | } | |
10066 | ||
b2650027 AN |
10067 | /* |
10068 | * this function is expected to parse integer in the range of [0, 2^31-1] from | |
10069 | * given file using scanf format string fmt. If actual parsed value is | |
10070 | * negative, the result might be indistinguishable from error | |
10071 | */ | |
10072 | static int parse_uint_from_file(const char *file, const char *fmt) | |
10073 | { | |
10074 | char buf[STRERR_BUFSIZE]; | |
10075 | int err, ret; | |
10076 | FILE *f; | |
10077 | ||
59842c54 | 10078 | f = fopen(file, "re"); |
b2650027 AN |
10079 | if (!f) { |
10080 | err = -errno; | |
10081 | pr_debug("failed to open '%s': %s\n", file, | |
10082 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
10083 | return err; | |
10084 | } | |
10085 | err = fscanf(f, fmt, &ret); | |
10086 | if (err != 1) { | |
10087 | err = err == EOF ? -EIO : -errno; | |
10088 | pr_debug("failed to parse '%s': %s\n", file, | |
10089 | libbpf_strerror_r(err, buf, sizeof(buf))); | |
10090 | fclose(f); | |
10091 | return err; | |
10092 | } | |
10093 | fclose(f); | |
10094 | return ret; | |
10095 | } | |
10096 | ||
10097 | static int determine_kprobe_perf_type(void) | |
10098 | { | |
10099 | const char *file = "/sys/bus/event_source/devices/kprobe/type"; | |
10100 | ||
10101 | return parse_uint_from_file(file, "%d\n"); | |
10102 | } | |
10103 | ||
10104 | static int determine_uprobe_perf_type(void) | |
10105 | { | |
10106 | const char *file = "/sys/bus/event_source/devices/uprobe/type"; | |
10107 | ||
10108 | return parse_uint_from_file(file, "%d\n"); | |
10109 | } | |
10110 | ||
10111 | static int determine_kprobe_retprobe_bit(void) | |
10112 | { | |
10113 | const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe"; | |
10114 | ||
10115 | return parse_uint_from_file(file, "config:%d\n"); | |
10116 | } | |
10117 | ||
10118 | static int determine_uprobe_retprobe_bit(void) | |
10119 | { | |
10120 | const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe"; | |
10121 | ||
10122 | return parse_uint_from_file(file, "config:%d\n"); | |
10123 | } | |
10124 | ||
5e3b8356 AN |
10125 | #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32 |
10126 | #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32 | |
10127 | ||
b2650027 | 10128 | static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name, |
5e3b8356 | 10129 | uint64_t offset, int pid, size_t ref_ctr_off) |
b2650027 | 10130 | { |
813847a3 AN |
10131 | const size_t attr_sz = sizeof(struct perf_event_attr); |
10132 | struct perf_event_attr attr; | |
b2650027 | 10133 | char errmsg[STRERR_BUFSIZE]; |
708ac5be | 10134 | int type, pfd; |
b2650027 | 10135 | |
1520e846 | 10136 | if ((__u64)ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS)) |
5e3b8356 AN |
10137 | return -EINVAL; |
10138 | ||
813847a3 AN |
10139 | memset(&attr, 0, attr_sz); |
10140 | ||
b2650027 AN |
10141 | type = uprobe ? determine_uprobe_perf_type() |
10142 | : determine_kprobe_perf_type(); | |
10143 | if (type < 0) { | |
be18010e KW |
10144 | pr_warn("failed to determine %s perf type: %s\n", |
10145 | uprobe ? "uprobe" : "kprobe", | |
10146 | libbpf_strerror_r(type, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10147 | return type; |
10148 | } | |
10149 | if (retprobe) { | |
10150 | int bit = uprobe ? determine_uprobe_retprobe_bit() | |
10151 | : determine_kprobe_retprobe_bit(); | |
10152 | ||
10153 | if (bit < 0) { | |
be18010e KW |
10154 | pr_warn("failed to determine %s retprobe bit: %s\n", |
10155 | uprobe ? "uprobe" : "kprobe", | |
10156 | libbpf_strerror_r(bit, errmsg, sizeof(errmsg))); | |
b2650027 AN |
10157 | return bit; |
10158 | } | |
10159 | attr.config |= 1 << bit; | |
10160 | } | |
813847a3 | 10161 | attr.size = attr_sz; |
b2650027 | 10162 | attr.type = type; |
5e3b8356 | 10163 | attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT; |
36db2a94 AN |
10164 | attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */ |
10165 | attr.config2 = offset; /* kprobe_addr or probe_offset */ | |
b2650027 AN |
10166 | |
10167 | /* pid filter is meaningful only for uprobes */ | |
10168 | pfd = syscall(__NR_perf_event_open, &attr, | |
10169 | pid < 0 ? -1 : pid /* pid */, | |
10170 | pid == -1 ? 0 : -1 /* cpu */, | |
10171 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
708ac5be | 10172 | return pfd >= 0 ? pfd : -errno; |
b2650027 AN |
10173 | } |
10174 | ||
46ed5fc3 AN |
10175 | static int append_to_file(const char *file, const char *fmt, ...) |
10176 | { | |
10177 | int fd, n, err = 0; | |
10178 | va_list ap; | |
01dc26c9 LP |
10179 | char buf[1024]; |
10180 | ||
10181 | va_start(ap, fmt); | |
10182 | n = vsnprintf(buf, sizeof(buf), fmt, ap); | |
10183 | va_end(ap); | |
10184 | ||
10185 | if (n < 0 || n >= sizeof(buf)) | |
10186 | return -EINVAL; | |
46ed5fc3 | 10187 | |
92274e24 | 10188 | fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0); |
46ed5fc3 AN |
10189 | if (fd < 0) |
10190 | return -errno; | |
10191 | ||
01dc26c9 | 10192 | if (write(fd, buf, n) < 0) |
46ed5fc3 AN |
10193 | err = -errno; |
10194 | ||
10195 | close(fd); | |
10196 | return err; | |
10197 | } | |
10198 | ||
a1ac9fd6 AN |
10199 | #define DEBUGFS "/sys/kernel/debug/tracing" |
10200 | #define TRACEFS "/sys/kernel/tracing" | |
10201 | ||
10202 | static bool use_debugfs(void) | |
10203 | { | |
10204 | static int has_debugfs = -1; | |
10205 | ||
10206 | if (has_debugfs < 0) | |
6a4ab886 | 10207 | has_debugfs = faccessat(AT_FDCWD, DEBUGFS, F_OK, AT_EACCESS) == 0; |
a1ac9fd6 AN |
10208 | |
10209 | return has_debugfs == 1; | |
10210 | } | |
10211 | ||
10212 | static const char *tracefs_path(void) | |
10213 | { | |
10214 | return use_debugfs() ? DEBUGFS : TRACEFS; | |
10215 | } | |
10216 | ||
10217 | static const char *tracefs_kprobe_events(void) | |
10218 | { | |
10219 | return use_debugfs() ? DEBUGFS"/kprobe_events" : TRACEFS"/kprobe_events"; | |
10220 | } | |
10221 | ||
10222 | static const char *tracefs_uprobe_events(void) | |
10223 | { | |
10224 | return use_debugfs() ? DEBUGFS"/uprobe_events" : TRACEFS"/uprobe_events"; | |
10225 | } | |
10226 | ||
46ed5fc3 AN |
10227 | static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz, |
10228 | const char *kfunc_name, size_t offset) | |
10229 | { | |
51a33c60 | 10230 | static int index = 0; |
2fa07453 | 10231 | int i; |
51a33c60 QW |
10232 | |
10233 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset, | |
10234 | __sync_fetch_and_add(&index, 1)); | |
2fa07453 MD |
10235 | |
10236 | /* sanitize binary_path in the probe name */ | |
10237 | for (i = 0; buf[i]; i++) { | |
10238 | if (!isalnum(buf[i])) | |
10239 | buf[i] = '_'; | |
10240 | } | |
46ed5fc3 AN |
10241 | } |
10242 | ||
10243 | static int add_kprobe_event_legacy(const char *probe_name, bool retprobe, | |
10244 | const char *kfunc_name, size_t offset) | |
10245 | { | |
a1ac9fd6 | 10246 | return append_to_file(tracefs_kprobe_events(), "%c:%s/%s %s+0x%zx", |
46ed5fc3 AN |
10247 | retprobe ? 'r' : 'p', |
10248 | retprobe ? "kretprobes" : "kprobes", | |
10249 | probe_name, kfunc_name, offset); | |
10250 | } | |
10251 | ||
10252 | static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe) | |
10253 | { | |
a1ac9fd6 AN |
10254 | return append_to_file(tracefs_kprobe_events(), "-:%s/%s", |
10255 | retprobe ? "kretprobes" : "kprobes", probe_name); | |
46ed5fc3 AN |
10256 | } |
10257 | ||
10258 | static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
10259 | { | |
10260 | char file[256]; | |
10261 | ||
a1ac9fd6 AN |
10262 | snprintf(file, sizeof(file), "%s/events/%s/%s/id", |
10263 | tracefs_path(), retprobe ? "kretprobes" : "kprobes", probe_name); | |
46ed5fc3 AN |
10264 | |
10265 | return parse_uint_from_file(file, "%d\n"); | |
10266 | } | |
10267 | ||
10268 | static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe, | |
10269 | const char *kfunc_name, size_t offset, int pid) | |
ca304b40 | 10270 | { |
813847a3 AN |
10271 | const size_t attr_sz = sizeof(struct perf_event_attr); |
10272 | struct perf_event_attr attr; | |
ca304b40 RDT |
10273 | char errmsg[STRERR_BUFSIZE]; |
10274 | int type, pfd, err; | |
10275 | ||
46ed5fc3 | 10276 | err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset); |
ca304b40 | 10277 | if (err < 0) { |
46ed5fc3 AN |
10278 | pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n", |
10279 | kfunc_name, offset, | |
ca304b40 RDT |
10280 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10281 | return err; | |
10282 | } | |
46ed5fc3 | 10283 | type = determine_kprobe_perf_type_legacy(probe_name, retprobe); |
ca304b40 | 10284 | if (type < 0) { |
80940293 | 10285 | err = type; |
46ed5fc3 AN |
10286 | pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n", |
10287 | kfunc_name, offset, | |
80940293 CW |
10288 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10289 | goto err_clean_legacy; | |
ca304b40 | 10290 | } |
813847a3 AN |
10291 | |
10292 | memset(&attr, 0, attr_sz); | |
10293 | attr.size = attr_sz; | |
ca304b40 RDT |
10294 | attr.config = type; |
10295 | attr.type = PERF_TYPE_TRACEPOINT; | |
10296 | ||
10297 | pfd = syscall(__NR_perf_event_open, &attr, | |
10298 | pid < 0 ? -1 : pid, /* pid */ | |
10299 | pid == -1 ? 0 : -1, /* cpu */ | |
10300 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
10301 | if (pfd < 0) { | |
10302 | err = -errno; | |
10303 | pr_warn("legacy kprobe perf_event_open() failed: %s\n", | |
10304 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
80940293 | 10305 | goto err_clean_legacy; |
ca304b40 RDT |
10306 | } |
10307 | return pfd; | |
80940293 CW |
10308 | |
10309 | err_clean_legacy: | |
10310 | /* Clear the newly added legacy kprobe_event */ | |
10311 | remove_kprobe_event_legacy(probe_name, retprobe); | |
10312 | return err; | |
ca304b40 RDT |
10313 | } |
10314 | ||
708ac5be AN |
10315 | static const char *arch_specific_syscall_pfx(void) |
10316 | { | |
10317 | #if defined(__x86_64__) | |
10318 | return "x64"; | |
10319 | #elif defined(__i386__) | |
10320 | return "ia32"; | |
10321 | #elif defined(__s390x__) | |
10322 | return "s390x"; | |
10323 | #elif defined(__s390__) | |
10324 | return "s390"; | |
10325 | #elif defined(__arm__) | |
10326 | return "arm"; | |
10327 | #elif defined(__aarch64__) | |
10328 | return "arm64"; | |
10329 | #elif defined(__mips__) | |
10330 | return "mips"; | |
10331 | #elif defined(__riscv) | |
10332 | return "riscv"; | |
64893e83 DM |
10333 | #elif defined(__powerpc__) |
10334 | return "powerpc"; | |
10335 | #elif defined(__powerpc64__) | |
10336 | return "powerpc64"; | |
708ac5be AN |
10337 | #else |
10338 | return NULL; | |
10339 | #endif | |
10340 | } | |
10341 | ||
10342 | static int probe_kern_syscall_wrapper(void) | |
10343 | { | |
10344 | char syscall_name[64]; | |
10345 | const char *ksys_pfx; | |
10346 | ||
10347 | ksys_pfx = arch_specific_syscall_pfx(); | |
10348 | if (!ksys_pfx) | |
10349 | return 0; | |
10350 | ||
10351 | snprintf(syscall_name, sizeof(syscall_name), "__%s_sys_bpf", ksys_pfx); | |
10352 | ||
10353 | if (determine_kprobe_perf_type() >= 0) { | |
10354 | int pfd; | |
10355 | ||
10356 | pfd = perf_event_open_probe(false, false, syscall_name, 0, getpid(), 0); | |
10357 | if (pfd >= 0) | |
10358 | close(pfd); | |
10359 | ||
10360 | return pfd >= 0 ? 1 : 0; | |
10361 | } else { /* legacy mode */ | |
10362 | char probe_name[128]; | |
10363 | ||
10364 | gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), syscall_name, 0); | |
10365 | if (add_kprobe_event_legacy(probe_name, false, syscall_name, 0) < 0) | |
10366 | return 0; | |
10367 | ||
10368 | (void)remove_kprobe_event_legacy(probe_name, false); | |
10369 | return 1; | |
10370 | } | |
10371 | } | |
10372 | ||
da97553e | 10373 | struct bpf_link * |
942025c9 | 10374 | bpf_program__attach_kprobe_opts(const struct bpf_program *prog, |
ac0ed488 | 10375 | const char *func_name, |
47faff37 | 10376 | const struct bpf_kprobe_opts *opts) |
b2650027 | 10377 | { |
47faff37 | 10378 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f8b299bc | 10379 | enum probe_attach_mode attach_mode; |
b2650027 | 10380 | char errmsg[STRERR_BUFSIZE]; |
ca304b40 | 10381 | char *legacy_probe = NULL; |
b2650027 | 10382 | struct bpf_link *link; |
46ed5fc3 | 10383 | size_t offset; |
ca304b40 | 10384 | bool retprobe, legacy; |
b2650027 AN |
10385 | int pfd, err; |
10386 | ||
da97553e JO |
10387 | if (!OPTS_VALID(opts, bpf_kprobe_opts)) |
10388 | return libbpf_err_ptr(-EINVAL); | |
10389 | ||
f8b299bc | 10390 | attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT); |
da97553e JO |
10391 | retprobe = OPTS_GET(opts, retprobe, false); |
10392 | offset = OPTS_GET(opts, offset, 0); | |
47faff37 | 10393 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
da97553e | 10394 | |
ca304b40 | 10395 | legacy = determine_kprobe_perf_type() < 0; |
f8b299bc MD |
10396 | switch (attach_mode) { |
10397 | case PROBE_ATTACH_MODE_LEGACY: | |
10398 | legacy = true; | |
10399 | pe_opts.force_ioctl_attach = true; | |
10400 | break; | |
10401 | case PROBE_ATTACH_MODE_PERF: | |
10402 | if (legacy) | |
10403 | return libbpf_err_ptr(-ENOTSUP); | |
10404 | pe_opts.force_ioctl_attach = true; | |
10405 | break; | |
10406 | case PROBE_ATTACH_MODE_LINK: | |
10407 | if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK)) | |
10408 | return libbpf_err_ptr(-ENOTSUP); | |
10409 | break; | |
10410 | case PROBE_ATTACH_MODE_DEFAULT: | |
10411 | break; | |
10412 | default: | |
10413 | return libbpf_err_ptr(-EINVAL); | |
10414 | } | |
10415 | ||
ca304b40 RDT |
10416 | if (!legacy) { |
10417 | pfd = perf_event_open_probe(false /* uprobe */, retprobe, | |
10418 | func_name, offset, | |
10419 | -1 /* pid */, 0 /* ref_ctr_off */); | |
10420 | } else { | |
46ed5fc3 AN |
10421 | char probe_name[256]; |
10422 | ||
10423 | gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
10424 | func_name, offset); | |
10425 | ||
71cff670 | 10426 | legacy_probe = strdup(probe_name); |
ca304b40 RDT |
10427 | if (!legacy_probe) |
10428 | return libbpf_err_ptr(-ENOMEM); | |
10429 | ||
46ed5fc3 | 10430 | pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name, |
ca304b40 RDT |
10431 | offset, -1 /* pid */); |
10432 | } | |
b2650027 | 10433 | if (pfd < 0) { |
46ed5fc3 AN |
10434 | err = -errno; |
10435 | pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n", | |
10436 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
10437 | func_name, offset, | |
303a2572 AN |
10438 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
10439 | goto err_out; | |
b2650027 | 10440 | } |
47faff37 | 10441 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
10442 | err = libbpf_get_error(link); |
10443 | if (err) { | |
b2650027 | 10444 | close(pfd); |
46ed5fc3 AN |
10445 | pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n", |
10446 | prog->name, retprobe ? "kretprobe" : "kprobe", | |
10447 | func_name, offset, | |
be18010e | 10448 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
80940293 | 10449 | goto err_clean_legacy; |
b2650027 | 10450 | } |
ca304b40 RDT |
10451 | if (legacy) { |
10452 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
10453 | ||
10454 | perf_link->legacy_probe_name = legacy_probe; | |
46ed5fc3 | 10455 | perf_link->legacy_is_kprobe = true; |
ca304b40 RDT |
10456 | perf_link->legacy_is_retprobe = retprobe; |
10457 | } | |
10458 | ||
b2650027 | 10459 | return link; |
80940293 CW |
10460 | |
10461 | err_clean_legacy: | |
10462 | if (legacy) | |
10463 | remove_kprobe_event_legacy(legacy_probe, retprobe); | |
303a2572 AN |
10464 | err_out: |
10465 | free(legacy_probe); | |
10466 | return libbpf_err_ptr(err); | |
b2650027 AN |
10467 | } |
10468 | ||
942025c9 | 10469 | struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog, |
ac0ed488 JO |
10470 | bool retprobe, |
10471 | const char *func_name) | |
10472 | { | |
da97553e | 10473 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts, |
ac0ed488 | 10474 | .retprobe = retprobe, |
da97553e | 10475 | ); |
ac0ed488 JO |
10476 | |
10477 | return bpf_program__attach_kprobe_opts(prog, func_name, &opts); | |
10478 | } | |
10479 | ||
708ac5be AN |
10480 | struct bpf_link *bpf_program__attach_ksyscall(const struct bpf_program *prog, |
10481 | const char *syscall_name, | |
10482 | const struct bpf_ksyscall_opts *opts) | |
10483 | { | |
10484 | LIBBPF_OPTS(bpf_kprobe_opts, kprobe_opts); | |
10485 | char func_name[128]; | |
10486 | ||
10487 | if (!OPTS_VALID(opts, bpf_ksyscall_opts)) | |
10488 | return libbpf_err_ptr(-EINVAL); | |
10489 | ||
10490 | if (kernel_supports(prog->obj, FEAT_SYSCALL_WRAPPER)) { | |
64893e83 DM |
10491 | /* arch_specific_syscall_pfx() should never return NULL here |
10492 | * because it is guarded by kernel_supports(). However, since | |
10493 | * compiler does not know that we have an explicit conditional | |
10494 | * as well. | |
10495 | */ | |
708ac5be | 10496 | snprintf(func_name, sizeof(func_name), "__%s_sys_%s", |
64893e83 | 10497 | arch_specific_syscall_pfx() ? : "", syscall_name); |
708ac5be AN |
10498 | } else { |
10499 | snprintf(func_name, sizeof(func_name), "__se_sys_%s", syscall_name); | |
10500 | } | |
10501 | ||
10502 | kprobe_opts.retprobe = OPTS_GET(opts, retprobe, false); | |
10503 | kprobe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); | |
10504 | ||
10505 | return bpf_program__attach_kprobe_opts(prog, func_name, &kprobe_opts); | |
10506 | } | |
10507 | ||
ddc6b049 JO |
10508 | /* Adapted from perf/util/string.c */ |
10509 | static bool glob_match(const char *str, const char *pat) | |
10510 | { | |
10511 | while (*str && *pat && *pat != '*') { | |
10512 | if (*pat == '?') { /* Matches any single character */ | |
10513 | str++; | |
10514 | pat++; | |
10515 | continue; | |
10516 | } | |
10517 | if (*str != *pat) | |
10518 | return false; | |
10519 | str++; | |
10520 | pat++; | |
10521 | } | |
10522 | /* Check wild card */ | |
10523 | if (*pat == '*') { | |
10524 | while (*pat == '*') | |
10525 | pat++; | |
10526 | if (!*pat) /* Tail wild card matches all */ | |
10527 | return true; | |
10528 | while (*str) | |
10529 | if (glob_match(str++, pat)) | |
10530 | return true; | |
10531 | } | |
10532 | return !*str && !*pat; | |
10533 | } | |
10534 | ||
10535 | struct kprobe_multi_resolve { | |
10536 | const char *pattern; | |
10537 | unsigned long *addrs; | |
10538 | size_t cap; | |
10539 | size_t cnt; | |
10540 | }; | |
10541 | ||
10542 | static int | |
10543 | resolve_kprobe_multi_cb(unsigned long long sym_addr, char sym_type, | |
10544 | const char *sym_name, void *ctx) | |
10545 | { | |
10546 | struct kprobe_multi_resolve *res = ctx; | |
10547 | int err; | |
10548 | ||
10549 | if (!glob_match(sym_name, res->pattern)) | |
10550 | return 0; | |
10551 | ||
10552 | err = libbpf_ensure_mem((void **) &res->addrs, &res->cap, sizeof(unsigned long), | |
10553 | res->cnt + 1); | |
10554 | if (err) | |
10555 | return err; | |
10556 | ||
10557 | res->addrs[res->cnt++] = (unsigned long) sym_addr; | |
10558 | return 0; | |
10559 | } | |
10560 | ||
10561 | struct bpf_link * | |
10562 | bpf_program__attach_kprobe_multi_opts(const struct bpf_program *prog, | |
10563 | const char *pattern, | |
10564 | const struct bpf_kprobe_multi_opts *opts) | |
10565 | { | |
10566 | LIBBPF_OPTS(bpf_link_create_opts, lopts); | |
10567 | struct kprobe_multi_resolve res = { | |
10568 | .pattern = pattern, | |
10569 | }; | |
10570 | struct bpf_link *link = NULL; | |
10571 | char errmsg[STRERR_BUFSIZE]; | |
10572 | const unsigned long *addrs; | |
10573 | int err, link_fd, prog_fd; | |
10574 | const __u64 *cookies; | |
10575 | const char **syms; | |
10576 | bool retprobe; | |
10577 | size_t cnt; | |
10578 | ||
10579 | if (!OPTS_VALID(opts, bpf_kprobe_multi_opts)) | |
10580 | return libbpf_err_ptr(-EINVAL); | |
10581 | ||
10582 | syms = OPTS_GET(opts, syms, false); | |
10583 | addrs = OPTS_GET(opts, addrs, false); | |
10584 | cnt = OPTS_GET(opts, cnt, false); | |
10585 | cookies = OPTS_GET(opts, cookies, false); | |
10586 | ||
10587 | if (!pattern && !addrs && !syms) | |
10588 | return libbpf_err_ptr(-EINVAL); | |
10589 | if (pattern && (addrs || syms || cookies || cnt)) | |
10590 | return libbpf_err_ptr(-EINVAL); | |
10591 | if (!pattern && !cnt) | |
10592 | return libbpf_err_ptr(-EINVAL); | |
10593 | if (addrs && syms) | |
10594 | return libbpf_err_ptr(-EINVAL); | |
10595 | ||
10596 | if (pattern) { | |
10597 | err = libbpf_kallsyms_parse(resolve_kprobe_multi_cb, &res); | |
10598 | if (err) | |
10599 | goto error; | |
10600 | if (!res.cnt) { | |
10601 | err = -ENOENT; | |
10602 | goto error; | |
10603 | } | |
10604 | addrs = res.addrs; | |
10605 | cnt = res.cnt; | |
10606 | } | |
10607 | ||
10608 | retprobe = OPTS_GET(opts, retprobe, false); | |
10609 | ||
10610 | lopts.kprobe_multi.syms = syms; | |
10611 | lopts.kprobe_multi.addrs = addrs; | |
10612 | lopts.kprobe_multi.cookies = cookies; | |
10613 | lopts.kprobe_multi.cnt = cnt; | |
10614 | lopts.kprobe_multi.flags = retprobe ? BPF_F_KPROBE_MULTI_RETURN : 0; | |
10615 | ||
10616 | link = calloc(1, sizeof(*link)); | |
10617 | if (!link) { | |
10618 | err = -ENOMEM; | |
10619 | goto error; | |
10620 | } | |
10621 | link->detach = &bpf_link__detach_fd; | |
10622 | ||
10623 | prog_fd = bpf_program__fd(prog); | |
10624 | link_fd = bpf_link_create(prog_fd, 0, BPF_TRACE_KPROBE_MULTI, &lopts); | |
10625 | if (link_fd < 0) { | |
10626 | err = -errno; | |
10627 | pr_warn("prog '%s': failed to attach: %s\n", | |
10628 | prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
10629 | goto error; | |
10630 | } | |
10631 | link->fd = link_fd; | |
10632 | free(res.addrs); | |
10633 | return link; | |
10634 | ||
10635 | error: | |
10636 | free(link); | |
10637 | free(res.addrs); | |
10638 | return libbpf_err_ptr(err); | |
10639 | } | |
10640 | ||
4fa5bcfe | 10641 | static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 10642 | { |
da97553e | 10643 | DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts); |
a2488b5f | 10644 | unsigned long offset = 0; |
d7a18ea7 | 10645 | const char *func_name; |
a2488b5f | 10646 | char *func; |
4fa5bcfe | 10647 | int n; |
d7a18ea7 | 10648 | |
9af8efc4 AN |
10649 | *link = NULL; |
10650 | ||
10651 | /* no auto-attach for SEC("kprobe") and SEC("kretprobe") */ | |
10652 | if (strcmp(prog->sec_name, "kprobe") == 0 || strcmp(prog->sec_name, "kretprobe") == 0) | |
10653 | return 0; | |
10654 | ||
13d35a0c AN |
10655 | opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/"); |
10656 | if (opts.retprobe) | |
10657 | func_name = prog->sec_name + sizeof("kretprobe/") - 1; | |
10658 | else | |
10659 | func_name = prog->sec_name + sizeof("kprobe/") - 1; | |
d7a18ea7 | 10660 | |
e3f9bc35 | 10661 | n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset); |
a2488b5f | 10662 | if (n < 1) { |
a2488b5f | 10663 | pr_warn("kprobe name is invalid: %s\n", func_name); |
4fa5bcfe | 10664 | return -EINVAL; |
a2488b5f AM |
10665 | } |
10666 | if (opts.retprobe && offset != 0) { | |
1f71a468 | 10667 | free(func); |
a2488b5f | 10668 | pr_warn("kretprobes do not support offset specification\n"); |
4fa5bcfe | 10669 | return -EINVAL; |
a2488b5f | 10670 | } |
d7a18ea7 | 10671 | |
a2488b5f | 10672 | opts.offset = offset; |
4fa5bcfe | 10673 | *link = bpf_program__attach_kprobe_opts(prog, func, &opts); |
a2488b5f | 10674 | free(func); |
4fa5bcfe | 10675 | return libbpf_get_error(*link); |
d7a18ea7 AN |
10676 | } |
10677 | ||
708ac5be AN |
10678 | static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
10679 | { | |
10680 | LIBBPF_OPTS(bpf_ksyscall_opts, opts); | |
10681 | const char *syscall_name; | |
10682 | ||
10683 | *link = NULL; | |
10684 | ||
10685 | /* no auto-attach for SEC("ksyscall") and SEC("kretsyscall") */ | |
10686 | if (strcmp(prog->sec_name, "ksyscall") == 0 || strcmp(prog->sec_name, "kretsyscall") == 0) | |
10687 | return 0; | |
10688 | ||
10689 | opts.retprobe = str_has_pfx(prog->sec_name, "kretsyscall/"); | |
10690 | if (opts.retprobe) | |
10691 | syscall_name = prog->sec_name + sizeof("kretsyscall/") - 1; | |
10692 | else | |
10693 | syscall_name = prog->sec_name + sizeof("ksyscall/") - 1; | |
10694 | ||
10695 | *link = bpf_program__attach_ksyscall(prog, syscall_name, &opts); | |
10696 | return *link ? 0 : -errno; | |
10697 | } | |
10698 | ||
ddc6b049 JO |
10699 | static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
10700 | { | |
10701 | LIBBPF_OPTS(bpf_kprobe_multi_opts, opts); | |
10702 | const char *spec; | |
10703 | char *pattern; | |
10704 | int n; | |
10705 | ||
9af8efc4 AN |
10706 | *link = NULL; |
10707 | ||
10708 | /* no auto-attach for SEC("kprobe.multi") and SEC("kretprobe.multi") */ | |
10709 | if (strcmp(prog->sec_name, "kprobe.multi") == 0 || | |
10710 | strcmp(prog->sec_name, "kretprobe.multi") == 0) | |
10711 | return 0; | |
10712 | ||
ddc6b049 JO |
10713 | opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe.multi/"); |
10714 | if (opts.retprobe) | |
10715 | spec = prog->sec_name + sizeof("kretprobe.multi/") - 1; | |
10716 | else | |
10717 | spec = prog->sec_name + sizeof("kprobe.multi/") - 1; | |
10718 | ||
10719 | n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern); | |
10720 | if (n < 1) { | |
10721 | pr_warn("kprobe multi pattern is invalid: %s\n", pattern); | |
10722 | return -EINVAL; | |
10723 | } | |
10724 | ||
10725 | *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts); | |
10726 | free(pattern); | |
10727 | return libbpf_get_error(*link); | |
10728 | } | |
10729 | ||
cc10623c AN |
10730 | static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz, |
10731 | const char *binary_path, uint64_t offset) | |
10732 | { | |
10733 | int i; | |
10734 | ||
10735 | snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset); | |
10736 | ||
10737 | /* sanitize binary_path in the probe name */ | |
10738 | for (i = 0; buf[i]; i++) { | |
10739 | if (!isalnum(buf[i])) | |
10740 | buf[i] = '_'; | |
10741 | } | |
10742 | } | |
10743 | ||
10744 | static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe, | |
10745 | const char *binary_path, size_t offset) | |
10746 | { | |
a1ac9fd6 | 10747 | return append_to_file(tracefs_uprobe_events(), "%c:%s/%s %s:0x%zx", |
cc10623c AN |
10748 | retprobe ? 'r' : 'p', |
10749 | retprobe ? "uretprobes" : "uprobes", | |
10750 | probe_name, binary_path, offset); | |
10751 | } | |
10752 | ||
10753 | static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe) | |
10754 | { | |
a1ac9fd6 AN |
10755 | return append_to_file(tracefs_uprobe_events(), "-:%s/%s", |
10756 | retprobe ? "uretprobes" : "uprobes", probe_name); | |
cc10623c AN |
10757 | } |
10758 | ||
10759 | static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe) | |
10760 | { | |
10761 | char file[512]; | |
10762 | ||
a1ac9fd6 AN |
10763 | snprintf(file, sizeof(file), "%s/events/%s/%s/id", |
10764 | tracefs_path(), retprobe ? "uretprobes" : "uprobes", probe_name); | |
cc10623c AN |
10765 | |
10766 | return parse_uint_from_file(file, "%d\n"); | |
10767 | } | |
10768 | ||
10769 | static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe, | |
10770 | const char *binary_path, size_t offset, int pid) | |
10771 | { | |
813847a3 | 10772 | const size_t attr_sz = sizeof(struct perf_event_attr); |
cc10623c AN |
10773 | struct perf_event_attr attr; |
10774 | int type, pfd, err; | |
10775 | ||
10776 | err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset); | |
10777 | if (err < 0) { | |
10778 | pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n", | |
10779 | binary_path, (size_t)offset, err); | |
10780 | return err; | |
10781 | } | |
10782 | type = determine_uprobe_perf_type_legacy(probe_name, retprobe); | |
10783 | if (type < 0) { | |
2655144f | 10784 | err = type; |
cc10623c | 10785 | pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n", |
2655144f CW |
10786 | binary_path, offset, err); |
10787 | goto err_clean_legacy; | |
cc10623c AN |
10788 | } |
10789 | ||
813847a3 AN |
10790 | memset(&attr, 0, attr_sz); |
10791 | attr.size = attr_sz; | |
cc10623c AN |
10792 | attr.config = type; |
10793 | attr.type = PERF_TYPE_TRACEPOINT; | |
10794 | ||
10795 | pfd = syscall(__NR_perf_event_open, &attr, | |
10796 | pid < 0 ? -1 : pid, /* pid */ | |
10797 | pid == -1 ? 0 : -1, /* cpu */ | |
10798 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
10799 | if (pfd < 0) { | |
10800 | err = -errno; | |
10801 | pr_warn("legacy uprobe perf_event_open() failed: %d\n", err); | |
2655144f | 10802 | goto err_clean_legacy; |
cc10623c AN |
10803 | } |
10804 | return pfd; | |
2655144f CW |
10805 | |
10806 | err_clean_legacy: | |
10807 | /* Clear the newly added legacy uprobe_event */ | |
10808 | remove_uprobe_event_legacy(probe_name, retprobe); | |
10809 | return err; | |
cc10623c AN |
10810 | } |
10811 | ||
433966e3 AM |
10812 | /* Return next ELF section of sh_type after scn, or first of that type if scn is NULL. */ |
10813 | static Elf_Scn *elf_find_next_scn_by_type(Elf *elf, int sh_type, Elf_Scn *scn) | |
10814 | { | |
10815 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
10816 | GElf_Shdr sh; | |
10817 | ||
10818 | if (!gelf_getshdr(scn, &sh)) | |
10819 | continue; | |
10820 | if (sh.sh_type == sh_type) | |
10821 | return scn; | |
10822 | } | |
10823 | return NULL; | |
10824 | } | |
10825 | ||
434fdcea DM |
10826 | /* Find offset of function name in the provided ELF object. "binary_path" is |
10827 | * the path to the ELF binary represented by "elf", and only used for error | |
10828 | * reporting matters. "name" matches symbol name or name@@LIB for library | |
10829 | * functions. | |
433966e3 | 10830 | */ |
434fdcea | 10831 | static long elf_find_func_offset(Elf *elf, const char *binary_path, const char *name) |
433966e3 | 10832 | { |
434fdcea | 10833 | int i, sh_types[2] = { SHT_DYNSYM, SHT_SYMTAB }; |
433966e3 | 10834 | bool is_shared_lib, is_name_qualified; |
433966e3 AM |
10835 | long ret = -ENOENT; |
10836 | size_t name_len; | |
10837 | GElf_Ehdr ehdr; | |
433966e3 | 10838 | |
433966e3 AM |
10839 | if (!gelf_getehdr(elf, &ehdr)) { |
10840 | pr_warn("elf: failed to get ehdr from %s: %s\n", binary_path, elf_errmsg(-1)); | |
10841 | ret = -LIBBPF_ERRNO__FORMAT; | |
10842 | goto out; | |
10843 | } | |
10844 | /* for shared lib case, we do not need to calculate relative offset */ | |
10845 | is_shared_lib = ehdr.e_type == ET_DYN; | |
10846 | ||
10847 | name_len = strlen(name); | |
10848 | /* Does name specify "@@LIB"? */ | |
10849 | is_name_qualified = strstr(name, "@@") != NULL; | |
10850 | ||
434fdcea | 10851 | /* Search SHT_DYNSYM, SHT_SYMTAB for symbol. This search order is used because if |
433966e3 AM |
10852 | * a binary is stripped, it may only have SHT_DYNSYM, and a fully-statically |
10853 | * linked binary may not have SHT_DYMSYM, so absence of a section should not be | |
10854 | * reported as a warning/error. | |
10855 | */ | |
10856 | for (i = 0; i < ARRAY_SIZE(sh_types); i++) { | |
10857 | size_t nr_syms, strtabidx, idx; | |
10858 | Elf_Data *symbols = NULL; | |
10859 | Elf_Scn *scn = NULL; | |
10860 | int last_bind = -1; | |
10861 | const char *sname; | |
10862 | GElf_Shdr sh; | |
10863 | ||
10864 | scn = elf_find_next_scn_by_type(elf, sh_types[i], NULL); | |
10865 | if (!scn) { | |
10866 | pr_debug("elf: failed to find symbol table ELF sections in '%s'\n", | |
10867 | binary_path); | |
10868 | continue; | |
10869 | } | |
10870 | if (!gelf_getshdr(scn, &sh)) | |
10871 | continue; | |
10872 | strtabidx = sh.sh_link; | |
10873 | symbols = elf_getdata(scn, 0); | |
10874 | if (!symbols) { | |
10875 | pr_warn("elf: failed to get symbols for symtab section in '%s': %s\n", | |
10876 | binary_path, elf_errmsg(-1)); | |
10877 | ret = -LIBBPF_ERRNO__FORMAT; | |
10878 | goto out; | |
10879 | } | |
10880 | nr_syms = symbols->d_size / sh.sh_entsize; | |
10881 | ||
10882 | for (idx = 0; idx < nr_syms; idx++) { | |
10883 | int curr_bind; | |
10884 | GElf_Sym sym; | |
fe928335 AN |
10885 | Elf_Scn *sym_scn; |
10886 | GElf_Shdr sym_sh; | |
433966e3 AM |
10887 | |
10888 | if (!gelf_getsym(symbols, idx, &sym)) | |
10889 | continue; | |
10890 | ||
10891 | if (GELF_ST_TYPE(sym.st_info) != STT_FUNC) | |
10892 | continue; | |
10893 | ||
10894 | sname = elf_strptr(elf, strtabidx, sym.st_name); | |
10895 | if (!sname) | |
10896 | continue; | |
10897 | ||
10898 | curr_bind = GELF_ST_BIND(sym.st_info); | |
10899 | ||
10900 | /* User can specify func, func@@LIB or func@@LIB_VERSION. */ | |
10901 | if (strncmp(sname, name, name_len) != 0) | |
10902 | continue; | |
10903 | /* ...but we don't want a search for "foo" to match 'foo2" also, so any | |
10904 | * additional characters in sname should be of the form "@@LIB". | |
10905 | */ | |
10906 | if (!is_name_qualified && sname[name_len] != '\0' && sname[name_len] != '@') | |
10907 | continue; | |
10908 | ||
10909 | if (ret >= 0) { | |
10910 | /* handle multiple matches */ | |
10911 | if (last_bind != STB_WEAK && curr_bind != STB_WEAK) { | |
10912 | /* Only accept one non-weak bind. */ | |
10913 | pr_warn("elf: ambiguous match for '%s', '%s' in '%s'\n", | |
10914 | sname, name, binary_path); | |
10915 | ret = -LIBBPF_ERRNO__FORMAT; | |
10916 | goto out; | |
10917 | } else if (curr_bind == STB_WEAK) { | |
10918 | /* already have a non-weak bind, and | |
10919 | * this is a weak bind, so ignore. | |
10920 | */ | |
10921 | continue; | |
10922 | } | |
10923 | } | |
fe928335 AN |
10924 | |
10925 | /* Transform symbol's virtual address (absolute for | |
10926 | * binaries and relative for shared libs) into file | |
10927 | * offset, which is what kernel is expecting for | |
10928 | * uprobe/uretprobe attachment. | |
10929 | * See Documentation/trace/uprobetracer.rst for more | |
10930 | * details. | |
10931 | * This is done by looking up symbol's containing | |
10932 | * section's header and using it's virtual address | |
10933 | * (sh_addr) and corresponding file offset (sh_offset) | |
10934 | * to transform sym.st_value (virtual address) into | |
10935 | * desired final file offset. | |
10936 | */ | |
10937 | sym_scn = elf_getscn(elf, sym.st_shndx); | |
10938 | if (!sym_scn) | |
10939 | continue; | |
10940 | if (!gelf_getshdr(sym_scn, &sym_sh)) | |
10941 | continue; | |
10942 | ||
10943 | ret = sym.st_value - sym_sh.sh_addr + sym_sh.sh_offset; | |
433966e3 AM |
10944 | last_bind = curr_bind; |
10945 | } | |
433966e3 AM |
10946 | if (ret > 0) |
10947 | break; | |
10948 | } | |
10949 | ||
10950 | if (ret > 0) { | |
10951 | pr_debug("elf: symbol address match for '%s' in '%s': 0x%lx\n", name, binary_path, | |
10952 | ret); | |
10953 | } else { | |
10954 | if (ret == 0) { | |
10955 | pr_warn("elf: '%s' is 0 in symtab for '%s': %s\n", name, binary_path, | |
10956 | is_shared_lib ? "should not be 0 in a shared library" : | |
10957 | "try using shared library path instead"); | |
10958 | ret = -ENOENT; | |
10959 | } else { | |
10960 | pr_warn("elf: failed to find symbol '%s' in '%s'\n", name, binary_path); | |
10961 | } | |
10962 | } | |
10963 | out: | |
434fdcea DM |
10964 | return ret; |
10965 | } | |
10966 | ||
10967 | /* Find offset of function name in ELF object specified by path. "name" matches | |
10968 | * symbol name or name@@LIB for library functions. | |
10969 | */ | |
10970 | static long elf_find_func_offset_from_file(const char *binary_path, const char *name) | |
10971 | { | |
10972 | char errmsg[STRERR_BUFSIZE]; | |
10973 | long ret = -ENOENT; | |
10974 | Elf *elf; | |
10975 | int fd; | |
10976 | ||
10977 | fd = open(binary_path, O_RDONLY | O_CLOEXEC); | |
10978 | if (fd < 0) { | |
10979 | ret = -errno; | |
10980 | pr_warn("failed to open %s: %s\n", binary_path, | |
10981 | libbpf_strerror_r(ret, errmsg, sizeof(errmsg))); | |
10982 | return ret; | |
10983 | } | |
10984 | elf = elf_begin(fd, ELF_C_READ_MMAP, NULL); | |
10985 | if (!elf) { | |
10986 | pr_warn("elf: could not read elf from %s: %s\n", binary_path, elf_errmsg(-1)); | |
10987 | close(fd); | |
10988 | return -LIBBPF_ERRNO__FORMAT; | |
10989 | } | |
10990 | ||
10991 | ret = elf_find_func_offset(elf, binary_path, name); | |
433966e3 AM |
10992 | elf_end(elf); |
10993 | close(fd); | |
10994 | return ret; | |
10995 | } | |
10996 | ||
c44fd845 DM |
10997 | /* Find offset of function name in archive specified by path. Currently |
10998 | * supported are .zip files that do not compress their contents, as used on | |
10999 | * Android in the form of APKs, for example. "file_name" is the name of the ELF | |
11000 | * file inside the archive. "func_name" matches symbol name or name@@LIB for | |
11001 | * library functions. | |
11002 | * | |
11003 | * An overview of the APK format specifically provided here: | |
11004 | * https://en.wikipedia.org/w/index.php?title=Apk_(file_format)&oldid=1139099120#Package_contents | |
11005 | */ | |
11006 | static long elf_find_func_offset_from_archive(const char *archive_path, const char *file_name, | |
11007 | const char *func_name) | |
11008 | { | |
11009 | struct zip_archive *archive; | |
11010 | struct zip_entry entry; | |
11011 | long ret; | |
11012 | Elf *elf; | |
11013 | ||
11014 | archive = zip_archive_open(archive_path); | |
11015 | if (IS_ERR(archive)) { | |
11016 | ret = PTR_ERR(archive); | |
11017 | pr_warn("zip: failed to open %s: %ld\n", archive_path, ret); | |
11018 | return ret; | |
11019 | } | |
11020 | ||
11021 | ret = zip_archive_find_entry(archive, file_name, &entry); | |
11022 | if (ret) { | |
11023 | pr_warn("zip: could not find archive member %s in %s: %ld\n", file_name, | |
11024 | archive_path, ret); | |
11025 | goto out; | |
11026 | } | |
11027 | pr_debug("zip: found entry for %s in %s at 0x%lx\n", file_name, archive_path, | |
11028 | (unsigned long)entry.data_offset); | |
11029 | ||
11030 | if (entry.compression) { | |
11031 | pr_warn("zip: entry %s of %s is compressed and cannot be handled\n", file_name, | |
11032 | archive_path); | |
11033 | ret = -LIBBPF_ERRNO__FORMAT; | |
11034 | goto out; | |
11035 | } | |
11036 | ||
11037 | elf = elf_memory((void *)entry.data, entry.data_length); | |
11038 | if (!elf) { | |
11039 | pr_warn("elf: could not read elf file %s from %s: %s\n", file_name, archive_path, | |
11040 | elf_errmsg(-1)); | |
11041 | ret = -LIBBPF_ERRNO__LIBELF; | |
11042 | goto out; | |
11043 | } | |
11044 | ||
11045 | ret = elf_find_func_offset(elf, file_name, func_name); | |
11046 | if (ret > 0) { | |
11047 | pr_debug("elf: symbol address match for %s of %s in %s: 0x%x + 0x%lx = 0x%lx\n", | |
11048 | func_name, file_name, archive_path, entry.data_offset, ret, | |
11049 | ret + entry.data_offset); | |
11050 | ret += entry.data_offset; | |
11051 | } | |
11052 | elf_end(elf); | |
11053 | ||
11054 | out: | |
11055 | zip_archive_close(archive); | |
11056 | return ret; | |
11057 | } | |
11058 | ||
56818931 IL |
11059 | static const char *arch_specific_lib_paths(void) |
11060 | { | |
11061 | /* | |
11062 | * Based on https://packages.debian.org/sid/libc6. | |
11063 | * | |
11064 | * Assume that the traced program is built for the same architecture | |
11065 | * as libbpf, which should cover the vast majority of cases. | |
11066 | */ | |
11067 | #if defined(__x86_64__) | |
11068 | return "/lib/x86_64-linux-gnu"; | |
11069 | #elif defined(__i386__) | |
11070 | return "/lib/i386-linux-gnu"; | |
11071 | #elif defined(__s390x__) | |
11072 | return "/lib/s390x-linux-gnu"; | |
11073 | #elif defined(__s390__) | |
11074 | return "/lib/s390-linux-gnu"; | |
11075 | #elif defined(__arm__) && defined(__SOFTFP__) | |
11076 | return "/lib/arm-linux-gnueabi"; | |
11077 | #elif defined(__arm__) && !defined(__SOFTFP__) | |
11078 | return "/lib/arm-linux-gnueabihf"; | |
11079 | #elif defined(__aarch64__) | |
11080 | return "/lib/aarch64-linux-gnu"; | |
11081 | #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 64 | |
11082 | return "/lib/mips64el-linux-gnuabi64"; | |
11083 | #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 32 | |
11084 | return "/lib/mipsel-linux-gnu"; | |
11085 | #elif defined(__powerpc64__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ | |
11086 | return "/lib/powerpc64le-linux-gnu"; | |
11087 | #elif defined(__sparc__) && defined(__arch64__) | |
11088 | return "/lib/sparc64-linux-gnu"; | |
11089 | #elif defined(__riscv) && __riscv_xlen == 64 | |
11090 | return "/lib/riscv64-linux-gnu"; | |
11091 | #else | |
11092 | return NULL; | |
11093 | #endif | |
11094 | } | |
11095 | ||
1ce3a60e AM |
11096 | /* Get full path to program/shared library. */ |
11097 | static int resolve_full_path(const char *file, char *result, size_t result_sz) | |
11098 | { | |
56818931 | 11099 | const char *search_paths[3] = {}; |
9e32084e | 11100 | int i, perm; |
1ce3a60e | 11101 | |
a1c9d61b | 11102 | if (str_has_sfx(file, ".so") || strstr(file, ".so.")) { |
1ce3a60e AM |
11103 | search_paths[0] = getenv("LD_LIBRARY_PATH"); |
11104 | search_paths[1] = "/usr/lib64:/usr/lib"; | |
56818931 | 11105 | search_paths[2] = arch_specific_lib_paths(); |
9e32084e | 11106 | perm = R_OK; |
1ce3a60e AM |
11107 | } else { |
11108 | search_paths[0] = getenv("PATH"); | |
11109 | search_paths[1] = "/usr/bin:/usr/sbin"; | |
9e32084e | 11110 | perm = R_OK | X_OK; |
1ce3a60e AM |
11111 | } |
11112 | ||
11113 | for (i = 0; i < ARRAY_SIZE(search_paths); i++) { | |
11114 | const char *s; | |
11115 | ||
11116 | if (!search_paths[i]) | |
11117 | continue; | |
11118 | for (s = search_paths[i]; s != NULL; s = strchr(s, ':')) { | |
11119 | char *next_path; | |
11120 | int seg_len; | |
11121 | ||
11122 | if (s[0] == ':') | |
11123 | s++; | |
11124 | next_path = strchr(s, ':'); | |
11125 | seg_len = next_path ? next_path - s : strlen(s); | |
11126 | if (!seg_len) | |
11127 | continue; | |
11128 | snprintf(result, result_sz, "%.*s/%s", seg_len, s, file); | |
9e32084e | 11129 | /* ensure it has required permissions */ |
6a4ab886 | 11130 | if (faccessat(AT_FDCWD, result, perm, AT_EACCESS) < 0) |
1ce3a60e AM |
11131 | continue; |
11132 | pr_debug("resolved '%s' to '%s'\n", file, result); | |
11133 | return 0; | |
11134 | } | |
11135 | } | |
11136 | return -ENOENT; | |
11137 | } | |
11138 | ||
47faff37 | 11139 | LIBBPF_API struct bpf_link * |
942025c9 | 11140 | bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid, |
47faff37 AN |
11141 | const char *binary_path, size_t func_offset, |
11142 | const struct bpf_uprobe_opts *opts) | |
b2650027 | 11143 | { |
c44fd845 | 11144 | const char *archive_path = NULL, *archive_sep = NULL; |
cc10623c | 11145 | char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL; |
c44fd845 | 11146 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f8b299bc | 11147 | enum probe_attach_mode attach_mode; |
c44fd845 | 11148 | char full_path[PATH_MAX]; |
b2650027 | 11149 | struct bpf_link *link; |
5e3b8356 | 11150 | size_t ref_ctr_off; |
b2650027 | 11151 | int pfd, err; |
cc10623c | 11152 | bool retprobe, legacy; |
433966e3 | 11153 | const char *func_name; |
47faff37 AN |
11154 | |
11155 | if (!OPTS_VALID(opts, bpf_uprobe_opts)) | |
11156 | return libbpf_err_ptr(-EINVAL); | |
11157 | ||
f8b299bc | 11158 | attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT); |
47faff37 | 11159 | retprobe = OPTS_GET(opts, retprobe, false); |
5e3b8356 | 11160 | ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0); |
47faff37 | 11161 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); |
b2650027 | 11162 | |
8ed2f5a6 HC |
11163 | if (!binary_path) |
11164 | return libbpf_err_ptr(-EINVAL); | |
11165 | ||
c44fd845 DM |
11166 | /* Check if "binary_path" refers to an archive. */ |
11167 | archive_sep = strstr(binary_path, "!/"); | |
11168 | if (archive_sep) { | |
11169 | full_path[0] = '\0'; | |
11170 | libbpf_strlcpy(full_path, binary_path, | |
11171 | min(sizeof(full_path), (size_t)(archive_sep - binary_path + 1))); | |
11172 | archive_path = full_path; | |
11173 | binary_path = archive_sep + 2; | |
11174 | } else if (!strchr(binary_path, '/')) { | |
11175 | err = resolve_full_path(binary_path, full_path, sizeof(full_path)); | |
1ce3a60e | 11176 | if (err) { |
2e4913e0 AN |
11177 | pr_warn("prog '%s': failed to resolve full path for '%s': %d\n", |
11178 | prog->name, binary_path, err); | |
1ce3a60e AM |
11179 | return libbpf_err_ptr(err); |
11180 | } | |
c44fd845 | 11181 | binary_path = full_path; |
1ce3a60e | 11182 | } |
433966e3 AM |
11183 | func_name = OPTS_GET(opts, func_name, NULL); |
11184 | if (func_name) { | |
11185 | long sym_off; | |
11186 | ||
c44fd845 DM |
11187 | if (archive_path) { |
11188 | sym_off = elf_find_func_offset_from_archive(archive_path, binary_path, | |
11189 | func_name); | |
11190 | binary_path = archive_path; | |
11191 | } else { | |
11192 | sym_off = elf_find_func_offset_from_file(binary_path, func_name); | |
11193 | } | |
433966e3 AM |
11194 | if (sym_off < 0) |
11195 | return libbpf_err_ptr(sym_off); | |
11196 | func_offset += sym_off; | |
11197 | } | |
1ce3a60e | 11198 | |
cc10623c | 11199 | legacy = determine_uprobe_perf_type() < 0; |
f8b299bc MD |
11200 | switch (attach_mode) { |
11201 | case PROBE_ATTACH_MODE_LEGACY: | |
11202 | legacy = true; | |
11203 | pe_opts.force_ioctl_attach = true; | |
11204 | break; | |
11205 | case PROBE_ATTACH_MODE_PERF: | |
11206 | if (legacy) | |
11207 | return libbpf_err_ptr(-ENOTSUP); | |
11208 | pe_opts.force_ioctl_attach = true; | |
11209 | break; | |
11210 | case PROBE_ATTACH_MODE_LINK: | |
11211 | if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK)) | |
11212 | return libbpf_err_ptr(-ENOTSUP); | |
11213 | break; | |
11214 | case PROBE_ATTACH_MODE_DEFAULT: | |
11215 | break; | |
11216 | default: | |
11217 | return libbpf_err_ptr(-EINVAL); | |
11218 | } | |
11219 | ||
cc10623c AN |
11220 | if (!legacy) { |
11221 | pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path, | |
11222 | func_offset, pid, ref_ctr_off); | |
11223 | } else { | |
1ce3a60e | 11224 | char probe_name[PATH_MAX + 64]; |
cc10623c AN |
11225 | |
11226 | if (ref_ctr_off) | |
11227 | return libbpf_err_ptr(-EINVAL); | |
11228 | ||
11229 | gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name), | |
11230 | binary_path, func_offset); | |
11231 | ||
11232 | legacy_probe = strdup(probe_name); | |
11233 | if (!legacy_probe) | |
11234 | return libbpf_err_ptr(-ENOMEM); | |
11235 | ||
11236 | pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe, | |
11237 | binary_path, func_offset, pid); | |
11238 | } | |
b2650027 | 11239 | if (pfd < 0) { |
cc10623c | 11240 | err = -errno; |
52109584 AN |
11241 | pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n", |
11242 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e | 11243 | binary_path, func_offset, |
cc10623c AN |
11244 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
11245 | goto err_out; | |
b2650027 | 11246 | } |
cc10623c | 11247 | |
47faff37 | 11248 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
11249 | err = libbpf_get_error(link); |
11250 | if (err) { | |
b2650027 | 11251 | close(pfd); |
52109584 AN |
11252 | pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n", |
11253 | prog->name, retprobe ? "uretprobe" : "uprobe", | |
be18010e KW |
11254 | binary_path, func_offset, |
11255 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
2655144f | 11256 | goto err_clean_legacy; |
cc10623c AN |
11257 | } |
11258 | if (legacy) { | |
11259 | struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link); | |
11260 | ||
11261 | perf_link->legacy_probe_name = legacy_probe; | |
11262 | perf_link->legacy_is_kprobe = false; | |
11263 | perf_link->legacy_is_retprobe = retprobe; | |
b2650027 AN |
11264 | } |
11265 | return link; | |
2655144f CW |
11266 | |
11267 | err_clean_legacy: | |
11268 | if (legacy) | |
11269 | remove_uprobe_event_legacy(legacy_probe, retprobe); | |
cc10623c AN |
11270 | err_out: |
11271 | free(legacy_probe); | |
11272 | return libbpf_err_ptr(err); | |
b2650027 AN |
11273 | } |
11274 | ||
39f8dc43 AM |
11275 | /* Format of u[ret]probe section definition supporting auto-attach: |
11276 | * u[ret]probe/binary:function[+offset] | |
11277 | * | |
11278 | * binary can be an absolute/relative path or a filename; the latter is resolved to a | |
11279 | * full binary path via bpf_program__attach_uprobe_opts. | |
11280 | * | |
11281 | * Specifying uprobe+ ensures we carry out strict matching; either "uprobe" must be | |
11282 | * specified (and auto-attach is not possible) or the above format is specified for | |
11283 | * auto-attach. | |
11284 | */ | |
11285 | static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link) | |
11286 | { | |
11287 | DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts); | |
90db26e6 AM |
11288 | char *probe_type = NULL, *binary_path = NULL, *func_name = NULL; |
11289 | int n, ret = -EINVAL; | |
11290 | long offset = 0; | |
39f8dc43 AM |
11291 | |
11292 | *link = NULL; | |
11293 | ||
90db26e6 AM |
11294 | n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[a-zA-Z0-9_.]+%li", |
11295 | &probe_type, &binary_path, &func_name, &offset); | |
11296 | switch (n) { | |
11297 | case 1: | |
11298 | /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */ | |
11299 | ret = 0; | |
11300 | break; | |
11301 | case 2: | |
11302 | pr_warn("prog '%s': section '%s' missing ':function[+offset]' specification\n", | |
11303 | prog->name, prog->sec_name); | |
11304 | break; | |
11305 | case 3: | |
11306 | case 4: | |
c4cac71f DK |
11307 | opts.retprobe = strcmp(probe_type, "uretprobe") == 0 || |
11308 | strcmp(probe_type, "uretprobe.s") == 0; | |
90db26e6 AM |
11309 | if (opts.retprobe && offset != 0) { |
11310 | pr_warn("prog '%s': uretprobes do not support offset specification\n", | |
11311 | prog->name); | |
11312 | break; | |
11313 | } | |
11314 | opts.func_name = func_name; | |
11315 | *link = bpf_program__attach_uprobe_opts(prog, -1, binary_path, offset, &opts); | |
11316 | ret = libbpf_get_error(*link); | |
11317 | break; | |
11318 | default: | |
11319 | pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name, | |
39f8dc43 | 11320 | prog->sec_name); |
90db26e6 | 11321 | break; |
39f8dc43 | 11322 | } |
90db26e6 AM |
11323 | free(probe_type); |
11324 | free(binary_path); | |
11325 | free(func_name); | |
39f8dc43 | 11326 | |
90db26e6 | 11327 | return ret; |
39f8dc43 AM |
11328 | } |
11329 | ||
942025c9 | 11330 | struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog, |
47faff37 AN |
11331 | bool retprobe, pid_t pid, |
11332 | const char *binary_path, | |
11333 | size_t func_offset) | |
11334 | { | |
11335 | DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe); | |
11336 | ||
11337 | return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts); | |
11338 | } | |
11339 | ||
2e4913e0 AN |
11340 | struct bpf_link *bpf_program__attach_usdt(const struct bpf_program *prog, |
11341 | pid_t pid, const char *binary_path, | |
11342 | const char *usdt_provider, const char *usdt_name, | |
11343 | const struct bpf_usdt_opts *opts) | |
11344 | { | |
11345 | char resolved_path[512]; | |
11346 | struct bpf_object *obj = prog->obj; | |
11347 | struct bpf_link *link; | |
5af25a41 | 11348 | __u64 usdt_cookie; |
2e4913e0 AN |
11349 | int err; |
11350 | ||
11351 | if (!OPTS_VALID(opts, bpf_uprobe_opts)) | |
11352 | return libbpf_err_ptr(-EINVAL); | |
11353 | ||
11354 | if (bpf_program__fd(prog) < 0) { | |
11355 | pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n", | |
11356 | prog->name); | |
11357 | return libbpf_err_ptr(-EINVAL); | |
11358 | } | |
11359 | ||
8ed2f5a6 HC |
11360 | if (!binary_path) |
11361 | return libbpf_err_ptr(-EINVAL); | |
11362 | ||
2e4913e0 AN |
11363 | if (!strchr(binary_path, '/')) { |
11364 | err = resolve_full_path(binary_path, resolved_path, sizeof(resolved_path)); | |
11365 | if (err) { | |
11366 | pr_warn("prog '%s': failed to resolve full path for '%s': %d\n", | |
11367 | prog->name, binary_path, err); | |
11368 | return libbpf_err_ptr(err); | |
11369 | } | |
11370 | binary_path = resolved_path; | |
11371 | } | |
11372 | ||
11373 | /* USDT manager is instantiated lazily on first USDT attach. It will | |
11374 | * be destroyed together with BPF object in bpf_object__close(). | |
11375 | */ | |
11376 | if (IS_ERR(obj->usdt_man)) | |
11377 | return libbpf_ptr(obj->usdt_man); | |
11378 | if (!obj->usdt_man) { | |
11379 | obj->usdt_man = usdt_manager_new(obj); | |
11380 | if (IS_ERR(obj->usdt_man)) | |
11381 | return libbpf_ptr(obj->usdt_man); | |
11382 | } | |
11383 | ||
11384 | usdt_cookie = OPTS_GET(opts, usdt_cookie, 0); | |
11385 | link = usdt_manager_attach_usdt(obj->usdt_man, prog, pid, binary_path, | |
e3ba8e4e | 11386 | usdt_provider, usdt_name, usdt_cookie); |
2e4913e0 AN |
11387 | err = libbpf_get_error(link); |
11388 | if (err) | |
11389 | return libbpf_err_ptr(err); | |
11390 | return link; | |
11391 | } | |
11392 | ||
11393 | static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link) | |
11394 | { | |
11395 | char *path = NULL, *provider = NULL, *name = NULL; | |
11396 | const char *sec_name; | |
11397 | int n, err; | |
11398 | ||
11399 | sec_name = bpf_program__section_name(prog); | |
11400 | if (strcmp(sec_name, "usdt") == 0) { | |
11401 | /* no auto-attach for just SEC("usdt") */ | |
11402 | *link = NULL; | |
11403 | return 0; | |
11404 | } | |
11405 | ||
11406 | n = sscanf(sec_name, "usdt/%m[^:]:%m[^:]:%m[^:]", &path, &provider, &name); | |
11407 | if (n != 3) { | |
11408 | pr_warn("invalid section '%s', expected SEC(\"usdt/<path>:<provider>:<name>\")\n", | |
11409 | sec_name); | |
11410 | err = -EINVAL; | |
11411 | } else { | |
11412 | *link = bpf_program__attach_usdt(prog, -1 /* any process */, path, | |
11413 | provider, name, NULL); | |
11414 | err = libbpf_get_error(*link); | |
11415 | } | |
11416 | free(path); | |
11417 | free(provider); | |
11418 | free(name); | |
11419 | return err; | |
11420 | } | |
11421 | ||
f6de59c1 AN |
11422 | static int determine_tracepoint_id(const char *tp_category, |
11423 | const char *tp_name) | |
11424 | { | |
11425 | char file[PATH_MAX]; | |
11426 | int ret; | |
11427 | ||
a1ac9fd6 AN |
11428 | ret = snprintf(file, sizeof(file), "%s/events/%s/%s/id", |
11429 | tracefs_path(), tp_category, tp_name); | |
f6de59c1 AN |
11430 | if (ret < 0) |
11431 | return -errno; | |
11432 | if (ret >= sizeof(file)) { | |
11433 | pr_debug("tracepoint %s/%s path is too long\n", | |
11434 | tp_category, tp_name); | |
11435 | return -E2BIG; | |
11436 | } | |
11437 | return parse_uint_from_file(file, "%d\n"); | |
11438 | } | |
11439 | ||
11440 | static int perf_event_open_tracepoint(const char *tp_category, | |
11441 | const char *tp_name) | |
11442 | { | |
813847a3 AN |
11443 | const size_t attr_sz = sizeof(struct perf_event_attr); |
11444 | struct perf_event_attr attr; | |
f6de59c1 AN |
11445 | char errmsg[STRERR_BUFSIZE]; |
11446 | int tp_id, pfd, err; | |
11447 | ||
11448 | tp_id = determine_tracepoint_id(tp_category, tp_name); | |
11449 | if (tp_id < 0) { | |
be18010e KW |
11450 | pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n", |
11451 | tp_category, tp_name, | |
11452 | libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
11453 | return tp_id; |
11454 | } | |
11455 | ||
813847a3 | 11456 | memset(&attr, 0, attr_sz); |
f6de59c1 | 11457 | attr.type = PERF_TYPE_TRACEPOINT; |
813847a3 | 11458 | attr.size = attr_sz; |
f6de59c1 AN |
11459 | attr.config = tp_id; |
11460 | ||
11461 | pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */, | |
11462 | -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC); | |
11463 | if (pfd < 0) { | |
11464 | err = -errno; | |
be18010e KW |
11465 | pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n", |
11466 | tp_category, tp_name, | |
11467 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); | |
f6de59c1 AN |
11468 | return err; |
11469 | } | |
11470 | return pfd; | |
11471 | } | |
11472 | ||
942025c9 | 11473 | struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog, |
47faff37 AN |
11474 | const char *tp_category, |
11475 | const char *tp_name, | |
11476 | const struct bpf_tracepoint_opts *opts) | |
f6de59c1 | 11477 | { |
47faff37 | 11478 | DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts); |
f6de59c1 AN |
11479 | char errmsg[STRERR_BUFSIZE]; |
11480 | struct bpf_link *link; | |
11481 | int pfd, err; | |
11482 | ||
47faff37 AN |
11483 | if (!OPTS_VALID(opts, bpf_tracepoint_opts)) |
11484 | return libbpf_err_ptr(-EINVAL); | |
11485 | ||
11486 | pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0); | |
11487 | ||
f6de59c1 AN |
11488 | pfd = perf_event_open_tracepoint(tp_category, tp_name); |
11489 | if (pfd < 0) { | |
52109584 AN |
11490 | pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n", |
11491 | prog->name, tp_category, tp_name, | |
be18010e | 11492 | libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 11493 | return libbpf_err_ptr(pfd); |
f6de59c1 | 11494 | } |
47faff37 | 11495 | link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts); |
e9fc3ce9 AN |
11496 | err = libbpf_get_error(link); |
11497 | if (err) { | |
f6de59c1 | 11498 | close(pfd); |
52109584 AN |
11499 | pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n", |
11500 | prog->name, tp_category, tp_name, | |
be18010e | 11501 | libbpf_strerror_r(err, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 11502 | return libbpf_err_ptr(err); |
f6de59c1 AN |
11503 | } |
11504 | return link; | |
11505 | } | |
11506 | ||
942025c9 | 11507 | struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog, |
47faff37 AN |
11508 | const char *tp_category, |
11509 | const char *tp_name) | |
11510 | { | |
11511 | return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL); | |
11512 | } | |
11513 | ||
4fa5bcfe | 11514 | static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 AN |
11515 | { |
11516 | char *sec_name, *tp_cat, *tp_name; | |
d7a18ea7 | 11517 | |
9af8efc4 AN |
11518 | *link = NULL; |
11519 | ||
11520 | /* no auto-attach for SEC("tp") or SEC("tracepoint") */ | |
11521 | if (strcmp(prog->sec_name, "tp") == 0 || strcmp(prog->sec_name, "tracepoint") == 0) | |
11522 | return 0; | |
11523 | ||
ac6a6586 AN |
11524 | sec_name = strdup(prog->sec_name); |
11525 | if (!sec_name) | |
11526 | return -ENOMEM; | |
11527 | ||
13d35a0c AN |
11528 | /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */ |
11529 | if (str_has_pfx(prog->sec_name, "tp/")) | |
11530 | tp_cat = sec_name + sizeof("tp/") - 1; | |
11531 | else | |
11532 | tp_cat = sec_name + sizeof("tracepoint/") - 1; | |
d7a18ea7 AN |
11533 | tp_name = strchr(tp_cat, '/'); |
11534 | if (!tp_name) { | |
e9fc3ce9 | 11535 | free(sec_name); |
4fa5bcfe | 11536 | return -EINVAL; |
d7a18ea7 AN |
11537 | } |
11538 | *tp_name = '\0'; | |
11539 | tp_name++; | |
11540 | ||
4fa5bcfe | 11541 | *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name); |
d7a18ea7 | 11542 | free(sec_name); |
4fa5bcfe | 11543 | return libbpf_get_error(*link); |
d7a18ea7 AN |
11544 | } |
11545 | ||
942025c9 | 11546 | struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog, |
84bf5e1f AN |
11547 | const char *tp_name) |
11548 | { | |
11549 | char errmsg[STRERR_BUFSIZE]; | |
c016b68e | 11550 | struct bpf_link *link; |
84bf5e1f AN |
11551 | int prog_fd, pfd; |
11552 | ||
11553 | prog_fd = bpf_program__fd(prog); | |
11554 | if (prog_fd < 0) { | |
52109584 | 11555 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 11556 | return libbpf_err_ptr(-EINVAL); |
84bf5e1f AN |
11557 | } |
11558 | ||
d6958706 | 11559 | link = calloc(1, sizeof(*link)); |
84bf5e1f | 11560 | if (!link) |
e9fc3ce9 | 11561 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 11562 | link->detach = &bpf_link__detach_fd; |
84bf5e1f AN |
11563 | |
11564 | pfd = bpf_raw_tracepoint_open(tp_name, prog_fd); | |
11565 | if (pfd < 0) { | |
11566 | pfd = -errno; | |
11567 | free(link); | |
52109584 AN |
11568 | pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n", |
11569 | prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 11570 | return libbpf_err_ptr(pfd); |
84bf5e1f AN |
11571 | } |
11572 | link->fd = pfd; | |
c016b68e | 11573 | return link; |
84bf5e1f AN |
11574 | } |
11575 | ||
4fa5bcfe | 11576 | static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 11577 | { |
ccaf12d6 | 11578 | static const char *const prefixes[] = { |
9af8efc4 AN |
11579 | "raw_tp", |
11580 | "raw_tracepoint", | |
11581 | "raw_tp.w", | |
11582 | "raw_tracepoint.w", | |
ccaf12d6 HT |
11583 | }; |
11584 | size_t i; | |
11585 | const char *tp_name = NULL; | |
13d35a0c | 11586 | |
9af8efc4 AN |
11587 | *link = NULL; |
11588 | ||
ccaf12d6 | 11589 | for (i = 0; i < ARRAY_SIZE(prefixes); i++) { |
9af8efc4 AN |
11590 | size_t pfx_len; |
11591 | ||
11592 | if (!str_has_pfx(prog->sec_name, prefixes[i])) | |
11593 | continue; | |
11594 | ||
11595 | pfx_len = strlen(prefixes[i]); | |
11596 | /* no auto-attach case of, e.g., SEC("raw_tp") */ | |
11597 | if (prog->sec_name[pfx_len] == '\0') | |
11598 | return 0; | |
11599 | ||
11600 | if (prog->sec_name[pfx_len] != '/') | |
11601 | continue; | |
11602 | ||
11603 | tp_name = prog->sec_name + pfx_len + 1; | |
11604 | break; | |
ccaf12d6 | 11605 | } |
9af8efc4 | 11606 | |
ccaf12d6 HT |
11607 | if (!tp_name) { |
11608 | pr_warn("prog '%s': invalid section name '%s'\n", | |
11609 | prog->name, prog->sec_name); | |
4fa5bcfe | 11610 | return -EINVAL; |
ccaf12d6 | 11611 | } |
d7a18ea7 | 11612 | |
4fa5bcfe | 11613 | *link = bpf_program__attach_raw_tracepoint(prog, tp_name); |
5fd2a60a | 11614 | return libbpf_get_error(*link); |
d7a18ea7 AN |
11615 | } |
11616 | ||
1e092a03 | 11617 | /* Common logic for all BPF program types that attach to a btf_id */ |
129b9c5e KFL |
11618 | static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog, |
11619 | const struct bpf_trace_opts *opts) | |
b8c54ea4 | 11620 | { |
129b9c5e | 11621 | LIBBPF_OPTS(bpf_link_create_opts, link_opts); |
b8c54ea4 | 11622 | char errmsg[STRERR_BUFSIZE]; |
c016b68e | 11623 | struct bpf_link *link; |
b8c54ea4 AS |
11624 | int prog_fd, pfd; |
11625 | ||
129b9c5e KFL |
11626 | if (!OPTS_VALID(opts, bpf_trace_opts)) |
11627 | return libbpf_err_ptr(-EINVAL); | |
11628 | ||
b8c54ea4 AS |
11629 | prog_fd = bpf_program__fd(prog); |
11630 | if (prog_fd < 0) { | |
52109584 | 11631 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 11632 | return libbpf_err_ptr(-EINVAL); |
b8c54ea4 AS |
11633 | } |
11634 | ||
d6958706 | 11635 | link = calloc(1, sizeof(*link)); |
b8c54ea4 | 11636 | if (!link) |
e9fc3ce9 | 11637 | return libbpf_err_ptr(-ENOMEM); |
c016b68e | 11638 | link->detach = &bpf_link__detach_fd; |
b8c54ea4 | 11639 | |
8462e0b4 | 11640 | /* libbpf is smart enough to redirect to BPF_RAW_TRACEPOINT_OPEN on old kernels */ |
129b9c5e KFL |
11641 | link_opts.tracing.cookie = OPTS_GET(opts, cookie, 0); |
11642 | pfd = bpf_link_create(prog_fd, 0, bpf_program__expected_attach_type(prog), &link_opts); | |
b8c54ea4 AS |
11643 | if (pfd < 0) { |
11644 | pfd = -errno; | |
11645 | free(link); | |
52109584 AN |
11646 | pr_warn("prog '%s': failed to attach: %s\n", |
11647 | prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 11648 | return libbpf_err_ptr(pfd); |
b8c54ea4 AS |
11649 | } |
11650 | link->fd = pfd; | |
003fed59 | 11651 | return link; |
b8c54ea4 AS |
11652 | } |
11653 | ||
942025c9 | 11654 | struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog) |
1e092a03 | 11655 | { |
129b9c5e KFL |
11656 | return bpf_program__attach_btf_id(prog, NULL); |
11657 | } | |
11658 | ||
11659 | struct bpf_link *bpf_program__attach_trace_opts(const struct bpf_program *prog, | |
11660 | const struct bpf_trace_opts *opts) | |
11661 | { | |
11662 | return bpf_program__attach_btf_id(prog, opts); | |
1e092a03 KS |
11663 | } |
11664 | ||
942025c9 | 11665 | struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog) |
1e092a03 | 11666 | { |
129b9c5e | 11667 | return bpf_program__attach_btf_id(prog, NULL); |
1e092a03 KS |
11668 | } |
11669 | ||
4fa5bcfe | 11670 | static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
d7a18ea7 | 11671 | { |
4fa5bcfe AN |
11672 | *link = bpf_program__attach_trace(prog); |
11673 | return libbpf_get_error(*link); | |
d7a18ea7 AN |
11674 | } |
11675 | ||
4fa5bcfe | 11676 | static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
1e092a03 | 11677 | { |
4fa5bcfe AN |
11678 | *link = bpf_program__attach_lsm(prog); |
11679 | return libbpf_get_error(*link); | |
1e092a03 KS |
11680 | } |
11681 | ||
d60d81ac | 11682 | static struct bpf_link * |
942025c9 | 11683 | bpf_program__attach_fd(const struct bpf_program *prog, int target_fd, int btf_id, |
d60d81ac | 11684 | const char *target_name) |
cc4f864b | 11685 | { |
a5359091 THJ |
11686 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts, |
11687 | .target_btf_id = btf_id); | |
cc4f864b AN |
11688 | enum bpf_attach_type attach_type; |
11689 | char errmsg[STRERR_BUFSIZE]; | |
11690 | struct bpf_link *link; | |
11691 | int prog_fd, link_fd; | |
11692 | ||
11693 | prog_fd = bpf_program__fd(prog); | |
11694 | if (prog_fd < 0) { | |
52109584 | 11695 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 11696 | return libbpf_err_ptr(-EINVAL); |
cc4f864b AN |
11697 | } |
11698 | ||
11699 | link = calloc(1, sizeof(*link)); | |
11700 | if (!link) | |
e9fc3ce9 | 11701 | return libbpf_err_ptr(-ENOMEM); |
cc4f864b AN |
11702 | link->detach = &bpf_link__detach_fd; |
11703 | ||
20eccf29 | 11704 | attach_type = bpf_program__expected_attach_type(prog); |
a5359091 | 11705 | link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts); |
cc4f864b AN |
11706 | if (link_fd < 0) { |
11707 | link_fd = -errno; | |
11708 | free(link); | |
52109584 AN |
11709 | pr_warn("prog '%s': failed to attach to %s: %s\n", |
11710 | prog->name, target_name, | |
cc4f864b | 11711 | libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); |
e9fc3ce9 | 11712 | return libbpf_err_ptr(link_fd); |
cc4f864b AN |
11713 | } |
11714 | link->fd = link_fd; | |
11715 | return link; | |
11716 | } | |
11717 | ||
d60d81ac | 11718 | struct bpf_link * |
942025c9 | 11719 | bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd) |
d60d81ac | 11720 | { |
a5359091 | 11721 | return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup"); |
d60d81ac JS |
11722 | } |
11723 | ||
11724 | struct bpf_link * | |
942025c9 | 11725 | bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd) |
d60d81ac | 11726 | { |
a5359091 | 11727 | return bpf_program__attach_fd(prog, netns_fd, 0, "netns"); |
d60d81ac JS |
11728 | } |
11729 | ||
942025c9 | 11730 | struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex) |
dc8698ca AN |
11731 | { |
11732 | /* target_fd/target_ifindex use the same field in LINK_CREATE */ | |
a5359091 THJ |
11733 | return bpf_program__attach_fd(prog, ifindex, 0, "xdp"); |
11734 | } | |
11735 | ||
942025c9 | 11736 | struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog, |
a5359091 THJ |
11737 | int target_fd, |
11738 | const char *attach_func_name) | |
11739 | { | |
11740 | int btf_id; | |
11741 | ||
11742 | if (!!target_fd != !!attach_func_name) { | |
11743 | pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n", | |
11744 | prog->name); | |
e9fc3ce9 | 11745 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
11746 | } |
11747 | ||
11748 | if (prog->type != BPF_PROG_TYPE_EXT) { | |
11749 | pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace", | |
11750 | prog->name); | |
e9fc3ce9 | 11751 | return libbpf_err_ptr(-EINVAL); |
a5359091 THJ |
11752 | } |
11753 | ||
11754 | if (target_fd) { | |
11755 | btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd); | |
11756 | if (btf_id < 0) | |
e9fc3ce9 | 11757 | return libbpf_err_ptr(btf_id); |
a5359091 THJ |
11758 | |
11759 | return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace"); | |
11760 | } else { | |
11761 | /* no target, so use raw_tracepoint_open for compatibility | |
11762 | * with old kernels | |
11763 | */ | |
11764 | return bpf_program__attach_trace(prog); | |
11765 | } | |
dc8698ca AN |
11766 | } |
11767 | ||
c09add2f | 11768 | struct bpf_link * |
942025c9 | 11769 | bpf_program__attach_iter(const struct bpf_program *prog, |
c09add2f YS |
11770 | const struct bpf_iter_attach_opts *opts) |
11771 | { | |
cd31039a | 11772 | DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts); |
c09add2f YS |
11773 | char errmsg[STRERR_BUFSIZE]; |
11774 | struct bpf_link *link; | |
11775 | int prog_fd, link_fd; | |
cd31039a | 11776 | __u32 target_fd = 0; |
c09add2f YS |
11777 | |
11778 | if (!OPTS_VALID(opts, bpf_iter_attach_opts)) | |
e9fc3ce9 | 11779 | return libbpf_err_ptr(-EINVAL); |
c09add2f | 11780 | |
74fc097d YS |
11781 | link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0); |
11782 | link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0); | |
cd31039a | 11783 | |
c09add2f YS |
11784 | prog_fd = bpf_program__fd(prog); |
11785 | if (prog_fd < 0) { | |
52109584 | 11786 | pr_warn("prog '%s': can't attach before loaded\n", prog->name); |
e9fc3ce9 | 11787 | return libbpf_err_ptr(-EINVAL); |
c09add2f YS |
11788 | } |
11789 | ||
11790 | link = calloc(1, sizeof(*link)); | |
11791 | if (!link) | |
e9fc3ce9 | 11792 | return libbpf_err_ptr(-ENOMEM); |
c09add2f YS |
11793 | link->detach = &bpf_link__detach_fd; |
11794 | ||
cd31039a YS |
11795 | link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER, |
11796 | &link_create_opts); | |
c09add2f YS |
11797 | if (link_fd < 0) { |
11798 | link_fd = -errno; | |
11799 | free(link); | |
52109584 AN |
11800 | pr_warn("prog '%s': failed to attach to iterator: %s\n", |
11801 | prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg))); | |
e9fc3ce9 | 11802 | return libbpf_err_ptr(link_fd); |
c09add2f YS |
11803 | } |
11804 | link->fd = link_fd; | |
11805 | return link; | |
11806 | } | |
11807 | ||
4fa5bcfe | 11808 | static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link) |
e9fc3ce9 | 11809 | { |
4fa5bcfe AN |
11810 | *link = bpf_program__attach_iter(prog, NULL); |
11811 | return libbpf_get_error(*link); | |
e9fc3ce9 AN |
11812 | } |
11813 | ||
942025c9 | 11814 | struct bpf_link *bpf_program__attach(const struct bpf_program *prog) |
d7a18ea7 | 11815 | { |
4fa5bcfe AN |
11816 | struct bpf_link *link = NULL; |
11817 | int err; | |
11818 | ||
11819 | if (!prog->sec_def || !prog->sec_def->prog_attach_fn) | |
11820 | return libbpf_err_ptr(-EOPNOTSUPP); | |
d7a18ea7 | 11821 | |
4fa5bcfe AN |
11822 | err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link); |
11823 | if (err) | |
11824 | return libbpf_err_ptr(err); | |
11825 | ||
11826 | /* When calling bpf_program__attach() explicitly, auto-attach support | |
11827 | * is expected to work, so NULL returned link is considered an error. | |
11828 | * This is different for skeleton's attach, see comment in | |
11829 | * bpf_object__attach_skeleton(). | |
11830 | */ | |
11831 | if (!link) | |
11832 | return libbpf_err_ptr(-EOPNOTSUPP); | |
11833 | ||
11834 | return link; | |
d7a18ea7 AN |
11835 | } |
11836 | ||
8d1608d7 KFL |
11837 | struct bpf_link_struct_ops { |
11838 | struct bpf_link link; | |
11839 | int map_fd; | |
11840 | }; | |
11841 | ||
590a0088 MKL |
11842 | static int bpf_link__detach_struct_ops(struct bpf_link *link) |
11843 | { | |
8d1608d7 | 11844 | struct bpf_link_struct_ops *st_link; |
590a0088 MKL |
11845 | __u32 zero = 0; |
11846 | ||
8d1608d7 | 11847 | st_link = container_of(link, struct bpf_link_struct_ops, link); |
590a0088 | 11848 | |
8d1608d7 KFL |
11849 | if (st_link->map_fd < 0) |
11850 | /* w/o a real link */ | |
11851 | return bpf_map_delete_elem(link->fd, &zero); | |
11852 | ||
11853 | return close(link->fd); | |
590a0088 MKL |
11854 | } |
11855 | ||
942025c9 | 11856 | struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map) |
590a0088 | 11857 | { |
8d1608d7 KFL |
11858 | struct bpf_link_struct_ops *link; |
11859 | __u32 zero = 0; | |
11860 | int err, fd; | |
590a0088 MKL |
11861 | |
11862 | if (!bpf_map__is_struct_ops(map) || map->fd == -1) | |
e9fc3ce9 | 11863 | return libbpf_err_ptr(-EINVAL); |
590a0088 MKL |
11864 | |
11865 | link = calloc(1, sizeof(*link)); | |
11866 | if (!link) | |
e9fc3ce9 | 11867 | return libbpf_err_ptr(-EINVAL); |
590a0088 | 11868 | |
8d1608d7 KFL |
11869 | /* kern_vdata should be prepared during the loading phase. */ |
11870 | err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0); | |
11871 | /* It can be EBUSY if the map has been used to create or | |
11872 | * update a link before. We don't allow updating the value of | |
11873 | * a struct_ops once it is set. That ensures that the value | |
11874 | * never changed. So, it is safe to skip EBUSY. | |
11875 | */ | |
11876 | if (err && (!(map->def.map_flags & BPF_F_LINK) || err != -EBUSY)) { | |
11877 | free(link); | |
11878 | return libbpf_err_ptr(err); | |
11879 | } | |
590a0088 | 11880 | |
8d1608d7 | 11881 | link->link.detach = bpf_link__detach_struct_ops; |
590a0088 | 11882 | |
8d1608d7 KFL |
11883 | if (!(map->def.map_flags & BPF_F_LINK)) { |
11884 | /* w/o a real link */ | |
11885 | link->link.fd = map->fd; | |
11886 | link->map_fd = -1; | |
11887 | return &link->link; | |
590a0088 MKL |
11888 | } |
11889 | ||
8d1608d7 KFL |
11890 | fd = bpf_link_create(map->fd, 0, BPF_STRUCT_OPS, NULL); |
11891 | if (fd < 0) { | |
590a0088 | 11892 | free(link); |
8d1608d7 | 11893 | return libbpf_err_ptr(fd); |
590a0088 MKL |
11894 | } |
11895 | ||
8d1608d7 KFL |
11896 | link->link.fd = fd; |
11897 | link->map_fd = map->fd; | |
590a0088 | 11898 | |
8d1608d7 | 11899 | return &link->link; |
590a0088 MKL |
11900 | } |
11901 | ||
912dd4b0 KFL |
11902 | /* |
11903 | * Swap the back struct_ops of a link with a new struct_ops map. | |
11904 | */ | |
11905 | int bpf_link__update_map(struct bpf_link *link, const struct bpf_map *map) | |
11906 | { | |
11907 | struct bpf_link_struct_ops *st_ops_link; | |
11908 | __u32 zero = 0; | |
11909 | int err; | |
11910 | ||
11911 | if (!bpf_map__is_struct_ops(map) || map->fd < 0) | |
11912 | return -EINVAL; | |
11913 | ||
11914 | st_ops_link = container_of(link, struct bpf_link_struct_ops, link); | |
11915 | /* Ensure the type of a link is correct */ | |
11916 | if (st_ops_link->map_fd < 0) | |
11917 | return -EINVAL; | |
11918 | ||
11919 | err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0); | |
11920 | /* It can be EBUSY if the map has been used to create or | |
11921 | * update a link before. We don't allow updating the value of | |
11922 | * a struct_ops once it is set. That ensures that the value | |
11923 | * never changed. So, it is safe to skip EBUSY. | |
11924 | */ | |
11925 | if (err && err != -EBUSY) | |
11926 | return err; | |
11927 | ||
11928 | err = bpf_link_update(link->fd, map->fd, NULL); | |
11929 | if (err < 0) | |
11930 | return err; | |
11931 | ||
11932 | st_ops_link->map_fd = map->fd; | |
11933 | ||
11934 | return 0; | |
11935 | } | |
11936 | ||
22dd7a58 AN |
11937 | typedef enum bpf_perf_event_ret (*bpf_perf_event_print_t)(struct perf_event_header *hdr, |
11938 | void *private_data); | |
11939 | ||
7218c28c CL |
11940 | static enum bpf_perf_event_ret |
11941 | perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size, | |
11942 | void **copy_mem, size_t *copy_size, | |
11943 | bpf_perf_event_print_t fn, void *private_data) | |
d0cabbb0 | 11944 | { |
3dca2115 | 11945 | struct perf_event_mmap_page *header = mmap_mem; |
a64af0ef | 11946 | __u64 data_head = ring_buffer_read_head(header); |
d0cabbb0 | 11947 | __u64 data_tail = header->data_tail; |
3dca2115 DB |
11948 | void *base = ((__u8 *)header) + page_size; |
11949 | int ret = LIBBPF_PERF_EVENT_CONT; | |
11950 | struct perf_event_header *ehdr; | |
11951 | size_t ehdr_size; | |
11952 | ||
11953 | while (data_head != data_tail) { | |
11954 | ehdr = base + (data_tail & (mmap_size - 1)); | |
11955 | ehdr_size = ehdr->size; | |
11956 | ||
11957 | if (((void *)ehdr) + ehdr_size > base + mmap_size) { | |
11958 | void *copy_start = ehdr; | |
11959 | size_t len_first = base + mmap_size - copy_start; | |
11960 | size_t len_secnd = ehdr_size - len_first; | |
11961 | ||
11962 | if (*copy_size < ehdr_size) { | |
11963 | free(*copy_mem); | |
11964 | *copy_mem = malloc(ehdr_size); | |
11965 | if (!*copy_mem) { | |
11966 | *copy_size = 0; | |
d0cabbb0 JK |
11967 | ret = LIBBPF_PERF_EVENT_ERROR; |
11968 | break; | |
11969 | } | |
3dca2115 | 11970 | *copy_size = ehdr_size; |
d0cabbb0 JK |
11971 | } |
11972 | ||
3dca2115 DB |
11973 | memcpy(*copy_mem, copy_start, len_first); |
11974 | memcpy(*copy_mem + len_first, base, len_secnd); | |
11975 | ehdr = *copy_mem; | |
d0cabbb0 JK |
11976 | } |
11977 | ||
3dca2115 DB |
11978 | ret = fn(ehdr, private_data); |
11979 | data_tail += ehdr_size; | |
d0cabbb0 JK |
11980 | if (ret != LIBBPF_PERF_EVENT_CONT) |
11981 | break; | |
d0cabbb0 JK |
11982 | } |
11983 | ||
a64af0ef | 11984 | ring_buffer_write_tail(header, data_tail); |
e9fc3ce9 | 11985 | return libbpf_err(ret); |
d0cabbb0 | 11986 | } |
34be1646 | 11987 | |
fb84b822 AN |
11988 | struct perf_buffer; |
11989 | ||
11990 | struct perf_buffer_params { | |
11991 | struct perf_event_attr *attr; | |
11992 | /* if event_cb is specified, it takes precendence */ | |
11993 | perf_buffer_event_fn event_cb; | |
11994 | /* sample_cb and lost_cb are higher-level common-case callbacks */ | |
11995 | perf_buffer_sample_fn sample_cb; | |
11996 | perf_buffer_lost_fn lost_cb; | |
11997 | void *ctx; | |
11998 | int cpu_cnt; | |
11999 | int *cpus; | |
12000 | int *map_keys; | |
12001 | }; | |
12002 | ||
12003 | struct perf_cpu_buf { | |
12004 | struct perf_buffer *pb; | |
12005 | void *base; /* mmap()'ed memory */ | |
12006 | void *buf; /* for reconstructing segmented data */ | |
12007 | size_t buf_size; | |
12008 | int fd; | |
12009 | int cpu; | |
12010 | int map_key; | |
12011 | }; | |
12012 | ||
12013 | struct perf_buffer { | |
12014 | perf_buffer_event_fn event_cb; | |
12015 | perf_buffer_sample_fn sample_cb; | |
12016 | perf_buffer_lost_fn lost_cb; | |
12017 | void *ctx; /* passed into callbacks */ | |
12018 | ||
12019 | size_t page_size; | |
12020 | size_t mmap_size; | |
12021 | struct perf_cpu_buf **cpu_bufs; | |
12022 | struct epoll_event *events; | |
783b8f01 | 12023 | int cpu_cnt; /* number of allocated CPU buffers */ |
fb84b822 AN |
12024 | int epoll_fd; /* perf event FD */ |
12025 | int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */ | |
12026 | }; | |
12027 | ||
12028 | static void perf_buffer__free_cpu_buf(struct perf_buffer *pb, | |
12029 | struct perf_cpu_buf *cpu_buf) | |
12030 | { | |
12031 | if (!cpu_buf) | |
12032 | return; | |
12033 | if (cpu_buf->base && | |
12034 | munmap(cpu_buf->base, pb->mmap_size + pb->page_size)) | |
be18010e | 12035 | pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu); |
fb84b822 AN |
12036 | if (cpu_buf->fd >= 0) { |
12037 | ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0); | |
12038 | close(cpu_buf->fd); | |
12039 | } | |
12040 | free(cpu_buf->buf); | |
12041 | free(cpu_buf); | |
12042 | } | |
12043 | ||
12044 | void perf_buffer__free(struct perf_buffer *pb) | |
12045 | { | |
12046 | int i; | |
12047 | ||
50450fc7 | 12048 | if (IS_ERR_OR_NULL(pb)) |
fb84b822 AN |
12049 | return; |
12050 | if (pb->cpu_bufs) { | |
601b05ca | 12051 | for (i = 0; i < pb->cpu_cnt; i++) { |
fb84b822 AN |
12052 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; |
12053 | ||
601b05ca EC |
12054 | if (!cpu_buf) |
12055 | continue; | |
12056 | ||
fb84b822 AN |
12057 | bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key); |
12058 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
12059 | } | |
12060 | free(pb->cpu_bufs); | |
12061 | } | |
12062 | if (pb->epoll_fd >= 0) | |
12063 | close(pb->epoll_fd); | |
12064 | free(pb->events); | |
12065 | free(pb); | |
12066 | } | |
12067 | ||
12068 | static struct perf_cpu_buf * | |
12069 | perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr, | |
12070 | int cpu, int map_key) | |
12071 | { | |
12072 | struct perf_cpu_buf *cpu_buf; | |
12073 | char msg[STRERR_BUFSIZE]; | |
12074 | int err; | |
12075 | ||
12076 | cpu_buf = calloc(1, sizeof(*cpu_buf)); | |
12077 | if (!cpu_buf) | |
12078 | return ERR_PTR(-ENOMEM); | |
12079 | ||
12080 | cpu_buf->pb = pb; | |
12081 | cpu_buf->cpu = cpu; | |
12082 | cpu_buf->map_key = map_key; | |
12083 | ||
12084 | cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu, | |
12085 | -1, PERF_FLAG_FD_CLOEXEC); | |
12086 | if (cpu_buf->fd < 0) { | |
12087 | err = -errno; | |
be18010e KW |
12088 | pr_warn("failed to open perf buffer event on cpu #%d: %s\n", |
12089 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
12090 | goto error; |
12091 | } | |
12092 | ||
12093 | cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size, | |
12094 | PROT_READ | PROT_WRITE, MAP_SHARED, | |
12095 | cpu_buf->fd, 0); | |
12096 | if (cpu_buf->base == MAP_FAILED) { | |
12097 | cpu_buf->base = NULL; | |
12098 | err = -errno; | |
be18010e KW |
12099 | pr_warn("failed to mmap perf buffer on cpu #%d: %s\n", |
12100 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
12101 | goto error; |
12102 | } | |
12103 | ||
12104 | if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) { | |
12105 | err = -errno; | |
be18010e KW |
12106 | pr_warn("failed to enable perf buffer event on cpu #%d: %s\n", |
12107 | cpu, libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
12108 | goto error; |
12109 | } | |
12110 | ||
12111 | return cpu_buf; | |
12112 | ||
12113 | error: | |
12114 | perf_buffer__free_cpu_buf(pb, cpu_buf); | |
12115 | return (struct perf_cpu_buf *)ERR_PTR(err); | |
12116 | } | |
12117 | ||
12118 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
12119 | struct perf_buffer_params *p); | |
12120 | ||
22dd7a58 AN |
12121 | struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt, |
12122 | perf_buffer_sample_fn sample_cb, | |
12123 | perf_buffer_lost_fn lost_cb, | |
12124 | void *ctx, | |
12125 | const struct perf_buffer_opts *opts) | |
fb84b822 | 12126 | { |
813847a3 | 12127 | const size_t attr_sz = sizeof(struct perf_event_attr); |
fb84b822 | 12128 | struct perf_buffer_params p = {}; |
813847a3 | 12129 | struct perf_event_attr attr; |
ab8684b8 | 12130 | __u32 sample_period; |
41788934 AN |
12131 | |
12132 | if (!OPTS_VALID(opts, perf_buffer_opts)) | |
12133 | return libbpf_err_ptr(-EINVAL); | |
4be6e05c | 12134 | |
ab8684b8 JD |
12135 | sample_period = OPTS_GET(opts, sample_period, 1); |
12136 | if (!sample_period) | |
12137 | sample_period = 1; | |
12138 | ||
813847a3 AN |
12139 | memset(&attr, 0, attr_sz); |
12140 | attr.size = attr_sz; | |
65bb2e0f | 12141 | attr.config = PERF_COUNT_SW_BPF_OUTPUT; |
4be6e05c ACM |
12142 | attr.type = PERF_TYPE_SOFTWARE; |
12143 | attr.sample_type = PERF_SAMPLE_RAW; | |
ab8684b8 JD |
12144 | attr.sample_period = sample_period; |
12145 | attr.wakeup_events = sample_period; | |
fb84b822 AN |
12146 | |
12147 | p.attr = &attr; | |
41788934 AN |
12148 | p.sample_cb = sample_cb; |
12149 | p.lost_cb = lost_cb; | |
12150 | p.ctx = ctx; | |
fb84b822 | 12151 | |
e9fc3ce9 | 12152 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
12153 | } |
12154 | ||
22dd7a58 AN |
12155 | struct perf_buffer *perf_buffer__new_raw(int map_fd, size_t page_cnt, |
12156 | struct perf_event_attr *attr, | |
12157 | perf_buffer_event_fn event_cb, void *ctx, | |
12158 | const struct perf_buffer_raw_opts *opts) | |
fb84b822 AN |
12159 | { |
12160 | struct perf_buffer_params p = {}; | |
12161 | ||
41332d6e | 12162 | if (!attr) |
41788934 AN |
12163 | return libbpf_err_ptr(-EINVAL); |
12164 | ||
12165 | if (!OPTS_VALID(opts, perf_buffer_raw_opts)) | |
12166 | return libbpf_err_ptr(-EINVAL); | |
12167 | ||
12168 | p.attr = attr; | |
12169 | p.event_cb = event_cb; | |
12170 | p.ctx = ctx; | |
12171 | p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0); | |
12172 | p.cpus = OPTS_GET(opts, cpus, NULL); | |
12173 | p.map_keys = OPTS_GET(opts, map_keys, NULL); | |
fb84b822 | 12174 | |
e9fc3ce9 | 12175 | return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p)); |
fb84b822 AN |
12176 | } |
12177 | ||
12178 | static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt, | |
12179 | struct perf_buffer_params *p) | |
12180 | { | |
783b8f01 | 12181 | const char *online_cpus_file = "/sys/devices/system/cpu/online"; |
0e289487 | 12182 | struct bpf_map_info map; |
fb84b822 AN |
12183 | char msg[STRERR_BUFSIZE]; |
12184 | struct perf_buffer *pb; | |
783b8f01 | 12185 | bool *online = NULL; |
fb84b822 | 12186 | __u32 map_info_len; |
783b8f01 | 12187 | int err, i, j, n; |
fb84b822 | 12188 | |
41332d6e | 12189 | if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) { |
be18010e KW |
12190 | pr_warn("page count should be power of two, but is %zu\n", |
12191 | page_cnt); | |
fb84b822 AN |
12192 | return ERR_PTR(-EINVAL); |
12193 | } | |
12194 | ||
0e289487 AN |
12195 | /* best-effort sanity checks */ |
12196 | memset(&map, 0, sizeof(map)); | |
fb84b822 | 12197 | map_info_len = sizeof(map); |
629dfc66 | 12198 | err = bpf_map_get_info_by_fd(map_fd, &map, &map_info_len); |
fb84b822 AN |
12199 | if (err) { |
12200 | err = -errno; | |
0e289487 AN |
12201 | /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return |
12202 | * -EBADFD, -EFAULT, or -E2BIG on real error | |
12203 | */ | |
12204 | if (err != -EINVAL) { | |
12205 | pr_warn("failed to get map info for map FD %d: %s\n", | |
12206 | map_fd, libbpf_strerror_r(err, msg, sizeof(msg))); | |
12207 | return ERR_PTR(err); | |
12208 | } | |
12209 | pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n", | |
12210 | map_fd); | |
12211 | } else { | |
12212 | if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) { | |
12213 | pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n", | |
12214 | map.name); | |
12215 | return ERR_PTR(-EINVAL); | |
12216 | } | |
fb84b822 AN |
12217 | } |
12218 | ||
12219 | pb = calloc(1, sizeof(*pb)); | |
12220 | if (!pb) | |
12221 | return ERR_PTR(-ENOMEM); | |
12222 | ||
12223 | pb->event_cb = p->event_cb; | |
12224 | pb->sample_cb = p->sample_cb; | |
12225 | pb->lost_cb = p->lost_cb; | |
12226 | pb->ctx = p->ctx; | |
12227 | ||
12228 | pb->page_size = getpagesize(); | |
12229 | pb->mmap_size = pb->page_size * page_cnt; | |
12230 | pb->map_fd = map_fd; | |
12231 | ||
12232 | pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
12233 | if (pb->epoll_fd < 0) { | |
12234 | err = -errno; | |
be18010e KW |
12235 | pr_warn("failed to create epoll instance: %s\n", |
12236 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
12237 | goto error; |
12238 | } | |
12239 | ||
12240 | if (p->cpu_cnt > 0) { | |
12241 | pb->cpu_cnt = p->cpu_cnt; | |
12242 | } else { | |
12243 | pb->cpu_cnt = libbpf_num_possible_cpus(); | |
12244 | if (pb->cpu_cnt < 0) { | |
12245 | err = pb->cpu_cnt; | |
12246 | goto error; | |
12247 | } | |
0e289487 | 12248 | if (map.max_entries && map.max_entries < pb->cpu_cnt) |
fb84b822 AN |
12249 | pb->cpu_cnt = map.max_entries; |
12250 | } | |
12251 | ||
12252 | pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events)); | |
12253 | if (!pb->events) { | |
12254 | err = -ENOMEM; | |
be18010e | 12255 | pr_warn("failed to allocate events: out of memory\n"); |
fb84b822 AN |
12256 | goto error; |
12257 | } | |
12258 | pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs)); | |
12259 | if (!pb->cpu_bufs) { | |
12260 | err = -ENOMEM; | |
be18010e | 12261 | pr_warn("failed to allocate buffers: out of memory\n"); |
fb84b822 AN |
12262 | goto error; |
12263 | } | |
12264 | ||
783b8f01 AN |
12265 | err = parse_cpu_mask_file(online_cpus_file, &online, &n); |
12266 | if (err) { | |
12267 | pr_warn("failed to get online CPU mask: %d\n", err); | |
12268 | goto error; | |
12269 | } | |
12270 | ||
12271 | for (i = 0, j = 0; i < pb->cpu_cnt; i++) { | |
fb84b822 AN |
12272 | struct perf_cpu_buf *cpu_buf; |
12273 | int cpu, map_key; | |
12274 | ||
12275 | cpu = p->cpu_cnt > 0 ? p->cpus[i] : i; | |
12276 | map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i; | |
12277 | ||
783b8f01 AN |
12278 | /* in case user didn't explicitly requested particular CPUs to |
12279 | * be attached to, skip offline/not present CPUs | |
12280 | */ | |
12281 | if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu])) | |
12282 | continue; | |
12283 | ||
fb84b822 AN |
12284 | cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key); |
12285 | if (IS_ERR(cpu_buf)) { | |
12286 | err = PTR_ERR(cpu_buf); | |
12287 | goto error; | |
12288 | } | |
12289 | ||
783b8f01 | 12290 | pb->cpu_bufs[j] = cpu_buf; |
fb84b822 AN |
12291 | |
12292 | err = bpf_map_update_elem(pb->map_fd, &map_key, | |
12293 | &cpu_buf->fd, 0); | |
12294 | if (err) { | |
12295 | err = -errno; | |
be18010e KW |
12296 | pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n", |
12297 | cpu, map_key, cpu_buf->fd, | |
12298 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
12299 | goto error; |
12300 | } | |
12301 | ||
783b8f01 AN |
12302 | pb->events[j].events = EPOLLIN; |
12303 | pb->events[j].data.ptr = cpu_buf; | |
fb84b822 | 12304 | if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd, |
783b8f01 | 12305 | &pb->events[j]) < 0) { |
fb84b822 | 12306 | err = -errno; |
be18010e KW |
12307 | pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n", |
12308 | cpu, cpu_buf->fd, | |
12309 | libbpf_strerror_r(err, msg, sizeof(msg))); | |
fb84b822 AN |
12310 | goto error; |
12311 | } | |
783b8f01 | 12312 | j++; |
fb84b822 | 12313 | } |
783b8f01 AN |
12314 | pb->cpu_cnt = j; |
12315 | free(online); | |
fb84b822 AN |
12316 | |
12317 | return pb; | |
12318 | ||
12319 | error: | |
783b8f01 | 12320 | free(online); |
fb84b822 AN |
12321 | if (pb) |
12322 | perf_buffer__free(pb); | |
12323 | return ERR_PTR(err); | |
12324 | } | |
12325 | ||
12326 | struct perf_sample_raw { | |
12327 | struct perf_event_header header; | |
12328 | uint32_t size; | |
385bbf7b | 12329 | char data[]; |
fb84b822 AN |
12330 | }; |
12331 | ||
12332 | struct perf_sample_lost { | |
12333 | struct perf_event_header header; | |
12334 | uint64_t id; | |
12335 | uint64_t lost; | |
12336 | uint64_t sample_id; | |
12337 | }; | |
12338 | ||
12339 | static enum bpf_perf_event_ret | |
12340 | perf_buffer__process_record(struct perf_event_header *e, void *ctx) | |
12341 | { | |
12342 | struct perf_cpu_buf *cpu_buf = ctx; | |
12343 | struct perf_buffer *pb = cpu_buf->pb; | |
12344 | void *data = e; | |
12345 | ||
12346 | /* user wants full control over parsing perf event */ | |
12347 | if (pb->event_cb) | |
12348 | return pb->event_cb(pb->ctx, cpu_buf->cpu, e); | |
12349 | ||
12350 | switch (e->type) { | |
12351 | case PERF_RECORD_SAMPLE: { | |
12352 | struct perf_sample_raw *s = data; | |
12353 | ||
12354 | if (pb->sample_cb) | |
12355 | pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size); | |
12356 | break; | |
12357 | } | |
12358 | case PERF_RECORD_LOST: { | |
12359 | struct perf_sample_lost *s = data; | |
12360 | ||
12361 | if (pb->lost_cb) | |
12362 | pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost); | |
12363 | break; | |
12364 | } | |
12365 | default: | |
be18010e | 12366 | pr_warn("unknown perf sample type %d\n", e->type); |
fb84b822 AN |
12367 | return LIBBPF_PERF_EVENT_ERROR; |
12368 | } | |
12369 | return LIBBPF_PERF_EVENT_CONT; | |
12370 | } | |
12371 | ||
12372 | static int perf_buffer__process_records(struct perf_buffer *pb, | |
12373 | struct perf_cpu_buf *cpu_buf) | |
12374 | { | |
12375 | enum bpf_perf_event_ret ret; | |
12376 | ||
7218c28c CL |
12377 | ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size, |
12378 | pb->page_size, &cpu_buf->buf, | |
12379 | &cpu_buf->buf_size, | |
12380 | perf_buffer__process_record, cpu_buf); | |
fb84b822 AN |
12381 | if (ret != LIBBPF_PERF_EVENT_CONT) |
12382 | return ret; | |
12383 | return 0; | |
12384 | } | |
12385 | ||
dca5612f AN |
12386 | int perf_buffer__epoll_fd(const struct perf_buffer *pb) |
12387 | { | |
12388 | return pb->epoll_fd; | |
12389 | } | |
12390 | ||
fb84b822 AN |
12391 | int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms) |
12392 | { | |
12393 | int i, cnt, err; | |
12394 | ||
12395 | cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms); | |
e9fc3ce9 | 12396 | if (cnt < 0) |
af0efa05 | 12397 | return -errno; |
e9fc3ce9 | 12398 | |
fb84b822 AN |
12399 | for (i = 0; i < cnt; i++) { |
12400 | struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr; | |
12401 | ||
12402 | err = perf_buffer__process_records(pb, cpu_buf); | |
12403 | if (err) { | |
be18010e | 12404 | pr_warn("error while processing records: %d\n", err); |
e9fc3ce9 | 12405 | return libbpf_err(err); |
fb84b822 AN |
12406 | } |
12407 | } | |
e9fc3ce9 | 12408 | return cnt; |
fb84b822 AN |
12409 | } |
12410 | ||
dca5612f AN |
12411 | /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer |
12412 | * manager. | |
12413 | */ | |
12414 | size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb) | |
12415 | { | |
12416 | return pb->cpu_cnt; | |
12417 | } | |
12418 | ||
12419 | /* | |
12420 | * Return perf_event FD of a ring buffer in *buf_idx* slot of | |
12421 | * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using | |
12422 | * select()/poll()/epoll() Linux syscalls. | |
12423 | */ | |
12424 | int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx) | |
12425 | { | |
12426 | struct perf_cpu_buf *cpu_buf; | |
12427 | ||
12428 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 12429 | return libbpf_err(-EINVAL); |
dca5612f AN |
12430 | |
12431 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
12432 | if (!cpu_buf) | |
e9fc3ce9 | 12433 | return libbpf_err(-ENOENT); |
dca5612f AN |
12434 | |
12435 | return cpu_buf->fd; | |
12436 | } | |
12437 | ||
9ff5efde JD |
12438 | int perf_buffer__buffer(struct perf_buffer *pb, int buf_idx, void **buf, size_t *buf_size) |
12439 | { | |
12440 | struct perf_cpu_buf *cpu_buf; | |
12441 | ||
12442 | if (buf_idx >= pb->cpu_cnt) | |
12443 | return libbpf_err(-EINVAL); | |
12444 | ||
12445 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
12446 | if (!cpu_buf) | |
12447 | return libbpf_err(-ENOENT); | |
12448 | ||
12449 | *buf = cpu_buf->base; | |
12450 | *buf_size = pb->mmap_size; | |
12451 | return 0; | |
12452 | } | |
12453 | ||
dca5612f AN |
12454 | /* |
12455 | * Consume data from perf ring buffer corresponding to slot *buf_idx* in | |
12456 | * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to | |
12457 | * consume, do nothing and return success. | |
12458 | * Returns: | |
12459 | * - 0 on success; | |
12460 | * - <0 on failure. | |
12461 | */ | |
12462 | int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx) | |
12463 | { | |
12464 | struct perf_cpu_buf *cpu_buf; | |
12465 | ||
12466 | if (buf_idx >= pb->cpu_cnt) | |
e9fc3ce9 | 12467 | return libbpf_err(-EINVAL); |
dca5612f AN |
12468 | |
12469 | cpu_buf = pb->cpu_bufs[buf_idx]; | |
12470 | if (!cpu_buf) | |
e9fc3ce9 | 12471 | return libbpf_err(-ENOENT); |
dca5612f AN |
12472 | |
12473 | return perf_buffer__process_records(pb, cpu_buf); | |
12474 | } | |
12475 | ||
272d51af EC |
12476 | int perf_buffer__consume(struct perf_buffer *pb) |
12477 | { | |
12478 | int i, err; | |
12479 | ||
12480 | for (i = 0; i < pb->cpu_cnt; i++) { | |
12481 | struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i]; | |
12482 | ||
12483 | if (!cpu_buf) | |
12484 | continue; | |
12485 | ||
12486 | err = perf_buffer__process_records(pb, cpu_buf); | |
12487 | if (err) { | |
dca5612f | 12488 | pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err); |
e9fc3ce9 | 12489 | return libbpf_err(err); |
272d51af EC |
12490 | } |
12491 | } | |
12492 | return 0; | |
12493 | } | |
12494 | ||
ff26ce5c EC |
12495 | int bpf_program__set_attach_target(struct bpf_program *prog, |
12496 | int attach_prog_fd, | |
12497 | const char *attach_func_name) | |
12498 | { | |
fe62de31 | 12499 | int btf_obj_fd = 0, btf_id = 0, err; |
ff26ce5c | 12500 | |
2d5ec1c6 | 12501 | if (!prog || attach_prog_fd < 0) |
e9fc3ce9 | 12502 | return libbpf_err(-EINVAL); |
ff26ce5c | 12503 | |
fe62de31 | 12504 | if (prog->obj->loaded) |
e9fc3ce9 | 12505 | return libbpf_err(-EINVAL); |
fe62de31 | 12506 | |
2d5ec1c6 AN |
12507 | if (attach_prog_fd && !attach_func_name) { |
12508 | /* remember attach_prog_fd and let bpf_program__load() find | |
12509 | * BTF ID during the program load | |
12510 | */ | |
12511 | prog->attach_prog_fd = attach_prog_fd; | |
12512 | return 0; | |
12513 | } | |
12514 | ||
fe62de31 | 12515 | if (attach_prog_fd) { |
ff26ce5c EC |
12516 | btf_id = libbpf_find_prog_btf_id(attach_func_name, |
12517 | attach_prog_fd); | |
fe62de31 | 12518 | if (btf_id < 0) |
e9fc3ce9 | 12519 | return libbpf_err(btf_id); |
fe62de31 | 12520 | } else { |
2d5ec1c6 AN |
12521 | if (!attach_func_name) |
12522 | return libbpf_err(-EINVAL); | |
12523 | ||
fe62de31 AN |
12524 | /* load btf_vmlinux, if not yet */ |
12525 | err = bpf_object__load_vmlinux_btf(prog->obj, true); | |
12526 | if (err) | |
e9fc3ce9 | 12527 | return libbpf_err(err); |
fe62de31 AN |
12528 | err = find_kernel_btf_id(prog->obj, attach_func_name, |
12529 | prog->expected_attach_type, | |
12530 | &btf_obj_fd, &btf_id); | |
12531 | if (err) | |
e9fc3ce9 | 12532 | return libbpf_err(err); |
fe62de31 | 12533 | } |
ff26ce5c EC |
12534 | |
12535 | prog->attach_btf_id = btf_id; | |
fe62de31 | 12536 | prog->attach_btf_obj_fd = btf_obj_fd; |
ff26ce5c EC |
12537 | prog->attach_prog_fd = attach_prog_fd; |
12538 | return 0; | |
12539 | } | |
12540 | ||
6803ee25 | 12541 | int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz) |
6446b315 | 12542 | { |
6803ee25 AN |
12543 | int err = 0, n, len, start, end = -1; |
12544 | bool *tmp; | |
6446b315 | 12545 | |
6803ee25 AN |
12546 | *mask = NULL; |
12547 | *mask_sz = 0; | |
12548 | ||
12549 | /* Each sub string separated by ',' has format \d+-\d+ or \d+ */ | |
12550 | while (*s) { | |
12551 | if (*s == ',' || *s == '\n') { | |
12552 | s++; | |
12553 | continue; | |
12554 | } | |
12555 | n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len); | |
12556 | if (n <= 0 || n > 2) { | |
12557 | pr_warn("Failed to get CPU range %s: %d\n", s, n); | |
12558 | err = -EINVAL; | |
12559 | goto cleanup; | |
12560 | } else if (n == 1) { | |
12561 | end = start; | |
12562 | } | |
12563 | if (start < 0 || start > end) { | |
12564 | pr_warn("Invalid CPU range [%d,%d] in %s\n", | |
12565 | start, end, s); | |
12566 | err = -EINVAL; | |
12567 | goto cleanup; | |
12568 | } | |
12569 | tmp = realloc(*mask, end + 1); | |
12570 | if (!tmp) { | |
12571 | err = -ENOMEM; | |
12572 | goto cleanup; | |
12573 | } | |
12574 | *mask = tmp; | |
12575 | memset(tmp + *mask_sz, 0, start - *mask_sz); | |
12576 | memset(tmp + start, 1, end - start + 1); | |
12577 | *mask_sz = end + 1; | |
12578 | s += len; | |
12579 | } | |
12580 | if (!*mask_sz) { | |
12581 | pr_warn("Empty CPU range\n"); | |
12582 | return -EINVAL; | |
12583 | } | |
12584 | return 0; | |
12585 | cleanup: | |
12586 | free(*mask); | |
12587 | *mask = NULL; | |
12588 | return err; | |
12589 | } | |
12590 | ||
12591 | int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz) | |
12592 | { | |
12593 | int fd, err = 0, len; | |
12594 | char buf[128]; | |
6446b315 | 12595 | |
92274e24 | 12596 | fd = open(fcpu, O_RDONLY | O_CLOEXEC); |
6446b315 | 12597 | if (fd < 0) { |
6803ee25 AN |
12598 | err = -errno; |
12599 | pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err); | |
12600 | return err; | |
6446b315 HL |
12601 | } |
12602 | len = read(fd, buf, sizeof(buf)); | |
12603 | close(fd); | |
12604 | if (len <= 0) { | |
6803ee25 AN |
12605 | err = len ? -errno : -EINVAL; |
12606 | pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err); | |
12607 | return err; | |
6446b315 | 12608 | } |
6803ee25 AN |
12609 | if (len >= sizeof(buf)) { |
12610 | pr_warn("CPU mask is too big in file %s\n", fcpu); | |
12611 | return -E2BIG; | |
6446b315 HL |
12612 | } |
12613 | buf[len] = '\0'; | |
12614 | ||
6803ee25 AN |
12615 | return parse_cpu_mask_str(buf, mask, mask_sz); |
12616 | } | |
12617 | ||
12618 | int libbpf_num_possible_cpus(void) | |
12619 | { | |
12620 | static const char *fcpu = "/sys/devices/system/cpu/possible"; | |
12621 | static int cpus; | |
12622 | int err, n, i, tmp_cpus; | |
12623 | bool *mask; | |
12624 | ||
12625 | tmp_cpus = READ_ONCE(cpus); | |
12626 | if (tmp_cpus > 0) | |
12627 | return tmp_cpus; | |
12628 | ||
12629 | err = parse_cpu_mask_file(fcpu, &mask, &n); | |
12630 | if (err) | |
e9fc3ce9 | 12631 | return libbpf_err(err); |
6803ee25 AN |
12632 | |
12633 | tmp_cpus = 0; | |
12634 | for (i = 0; i < n; i++) { | |
12635 | if (mask[i]) | |
12636 | tmp_cpus++; | |
6446b315 | 12637 | } |
6803ee25 | 12638 | free(mask); |
56fbc241 TC |
12639 | |
12640 | WRITE_ONCE(cpus, tmp_cpus); | |
12641 | return tmp_cpus; | |
6446b315 | 12642 | } |
d66562fb | 12643 | |
430025e5 DK |
12644 | static int populate_skeleton_maps(const struct bpf_object *obj, |
12645 | struct bpf_map_skeleton *maps, | |
12646 | size_t map_cnt) | |
12647 | { | |
12648 | int i; | |
12649 | ||
12650 | for (i = 0; i < map_cnt; i++) { | |
12651 | struct bpf_map **map = maps[i].map; | |
12652 | const char *name = maps[i].name; | |
12653 | void **mmaped = maps[i].mmaped; | |
12654 | ||
12655 | *map = bpf_object__find_map_by_name(obj, name); | |
12656 | if (!*map) { | |
12657 | pr_warn("failed to find skeleton map '%s'\n", name); | |
12658 | return -ESRCH; | |
12659 | } | |
12660 | ||
12661 | /* externs shouldn't be pre-setup from user code */ | |
12662 | if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG) | |
12663 | *mmaped = (*map)->mmaped; | |
12664 | } | |
12665 | return 0; | |
12666 | } | |
12667 | ||
12668 | static int populate_skeleton_progs(const struct bpf_object *obj, | |
12669 | struct bpf_prog_skeleton *progs, | |
12670 | size_t prog_cnt) | |
12671 | { | |
12672 | int i; | |
12673 | ||
12674 | for (i = 0; i < prog_cnt; i++) { | |
12675 | struct bpf_program **prog = progs[i].prog; | |
12676 | const char *name = progs[i].name; | |
12677 | ||
12678 | *prog = bpf_object__find_program_by_name(obj, name); | |
12679 | if (!*prog) { | |
12680 | pr_warn("failed to find skeleton program '%s'\n", name); | |
12681 | return -ESRCH; | |
12682 | } | |
12683 | } | |
12684 | return 0; | |
12685 | } | |
12686 | ||
d66562fb AN |
12687 | int bpf_object__open_skeleton(struct bpf_object_skeleton *s, |
12688 | const struct bpf_object_open_opts *opts) | |
12689 | { | |
12690 | DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts, | |
12691 | .object_name = s->name, | |
12692 | ); | |
12693 | struct bpf_object *obj; | |
430025e5 | 12694 | int err; |
d66562fb AN |
12695 | |
12696 | /* Attempt to preserve opts->object_name, unless overriden by user | |
12697 | * explicitly. Overwriting object name for skeletons is discouraged, | |
12698 | * as it breaks global data maps, because they contain object name | |
12699 | * prefix as their own map name prefix. When skeleton is generated, | |
12700 | * bpftool is making an assumption that this name will stay the same. | |
12701 | */ | |
12702 | if (opts) { | |
12703 | memcpy(&skel_opts, opts, sizeof(*opts)); | |
12704 | if (!opts->object_name) | |
12705 | skel_opts.object_name = s->name; | |
12706 | } | |
12707 | ||
12708 | obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts); | |
e9fc3ce9 AN |
12709 | err = libbpf_get_error(obj); |
12710 | if (err) { | |
12711 | pr_warn("failed to initialize skeleton BPF object '%s': %d\n", | |
12712 | s->name, err); | |
12713 | return libbpf_err(err); | |
d66562fb AN |
12714 | } |
12715 | ||
12716 | *s->obj = obj; | |
430025e5 DK |
12717 | err = populate_skeleton_maps(obj, s->maps, s->map_cnt); |
12718 | if (err) { | |
12719 | pr_warn("failed to populate skeleton maps for '%s': %d\n", s->name, err); | |
12720 | return libbpf_err(err); | |
12721 | } | |
12722 | ||
12723 | err = populate_skeleton_progs(obj, s->progs, s->prog_cnt); | |
12724 | if (err) { | |
12725 | pr_warn("failed to populate skeleton progs for '%s': %d\n", s->name, err); | |
12726 | return libbpf_err(err); | |
12727 | } | |
d66562fb | 12728 | |
430025e5 DK |
12729 | return 0; |
12730 | } | |
d66562fb | 12731 | |
430025e5 DK |
12732 | int bpf_object__open_subskeleton(struct bpf_object_subskeleton *s) |
12733 | { | |
12734 | int err, len, var_idx, i; | |
12735 | const char *var_name; | |
12736 | const struct bpf_map *map; | |
12737 | struct btf *btf; | |
12738 | __u32 map_type_id; | |
12739 | const struct btf_type *map_type, *var_type; | |
12740 | const struct bpf_var_skeleton *var_skel; | |
12741 | struct btf_var_secinfo *var; | |
d66562fb | 12742 | |
430025e5 DK |
12743 | if (!s->obj) |
12744 | return libbpf_err(-EINVAL); | |
12745 | ||
12746 | btf = bpf_object__btf(s->obj); | |
12747 | if (!btf) { | |
12748 | pr_warn("subskeletons require BTF at runtime (object %s)\n", | |
e3ba8e4e | 12749 | bpf_object__name(s->obj)); |
430025e5 | 12750 | return libbpf_err(-errno); |
d66562fb AN |
12751 | } |
12752 | ||
430025e5 DK |
12753 | err = populate_skeleton_maps(s->obj, s->maps, s->map_cnt); |
12754 | if (err) { | |
12755 | pr_warn("failed to populate subskeleton maps: %d\n", err); | |
12756 | return libbpf_err(err); | |
12757 | } | |
d66562fb | 12758 | |
430025e5 DK |
12759 | err = populate_skeleton_progs(s->obj, s->progs, s->prog_cnt); |
12760 | if (err) { | |
12761 | pr_warn("failed to populate subskeleton maps: %d\n", err); | |
12762 | return libbpf_err(err); | |
d66562fb AN |
12763 | } |
12764 | ||
430025e5 DK |
12765 | for (var_idx = 0; var_idx < s->var_cnt; var_idx++) { |
12766 | var_skel = &s->vars[var_idx]; | |
12767 | map = *var_skel->map; | |
12768 | map_type_id = bpf_map__btf_value_type_id(map); | |
12769 | map_type = btf__type_by_id(btf, map_type_id); | |
12770 | ||
12771 | if (!btf_is_datasec(map_type)) { | |
12772 | pr_warn("type for map '%1$s' is not a datasec: %2$s", | |
12773 | bpf_map__name(map), | |
12774 | __btf_kind_str(btf_kind(map_type))); | |
12775 | return libbpf_err(-EINVAL); | |
12776 | } | |
12777 | ||
12778 | len = btf_vlen(map_type); | |
12779 | var = btf_var_secinfos(map_type); | |
12780 | for (i = 0; i < len; i++, var++) { | |
12781 | var_type = btf__type_by_id(btf, var->type); | |
12782 | var_name = btf__name_by_offset(btf, var_type->name_off); | |
12783 | if (strcmp(var_name, var_skel->name) == 0) { | |
12784 | *var_skel->addr = map->mmaped + var->offset; | |
12785 | break; | |
12786 | } | |
12787 | } | |
12788 | } | |
d66562fb AN |
12789 | return 0; |
12790 | } | |
12791 | ||
430025e5 DK |
12792 | void bpf_object__destroy_subskeleton(struct bpf_object_subskeleton *s) |
12793 | { | |
12794 | if (!s) | |
12795 | return; | |
12796 | free(s->maps); | |
12797 | free(s->progs); | |
12798 | free(s->vars); | |
12799 | free(s); | |
12800 | } | |
12801 | ||
d66562fb AN |
12802 | int bpf_object__load_skeleton(struct bpf_object_skeleton *s) |
12803 | { | |
12804 | int i, err; | |
12805 | ||
12806 | err = bpf_object__load(*s->obj); | |
12807 | if (err) { | |
12808 | pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err); | |
e9fc3ce9 | 12809 | return libbpf_err(err); |
d66562fb AN |
12810 | } |
12811 | ||
12812 | for (i = 0; i < s->map_cnt; i++) { | |
12813 | struct bpf_map *map = *s->maps[i].map; | |
9d0a2331 | 12814 | size_t mmap_sz = bpf_map_mmap_sz(map->def.value_size, map->def.max_entries); |
d66562fb AN |
12815 | int prot, map_fd = bpf_map__fd(map); |
12816 | void **mmaped = s->maps[i].mmaped; | |
d66562fb AN |
12817 | |
12818 | if (!mmaped) | |
12819 | continue; | |
12820 | ||
12821 | if (!(map->def.map_flags & BPF_F_MMAPABLE)) { | |
12822 | *mmaped = NULL; | |
12823 | continue; | |
12824 | } | |
12825 | ||
12826 | if (map->def.map_flags & BPF_F_RDONLY_PROG) | |
12827 | prot = PROT_READ; | |
12828 | else | |
12829 | prot = PROT_READ | PROT_WRITE; | |
12830 | ||
12831 | /* Remap anonymous mmap()-ed "map initialization image" as | |
12832 | * a BPF map-backed mmap()-ed memory, but preserving the same | |
12833 | * memory address. This will cause kernel to change process' | |
12834 | * page table to point to a different piece of kernel memory, | |
12835 | * but from userspace point of view memory address (and its | |
12836 | * contents, being identical at this point) will stay the | |
12837 | * same. This mapping will be released by bpf_object__close() | |
12838 | * as per normal clean up procedure, so we don't need to worry | |
12839 | * about it from skeleton's clean up perspective. | |
12840 | */ | |
9d0a2331 | 12841 | *mmaped = mmap(map->mmaped, mmap_sz, prot, MAP_SHARED | MAP_FIXED, map_fd, 0); |
2ad97d47 | 12842 | if (*mmaped == MAP_FAILED) { |
d66562fb AN |
12843 | err = -errno; |
12844 | *mmaped = NULL; | |
12845 | pr_warn("failed to re-mmap() map '%s': %d\n", | |
12846 | bpf_map__name(map), err); | |
e9fc3ce9 | 12847 | return libbpf_err(err); |
d66562fb AN |
12848 | } |
12849 | } | |
12850 | ||
12851 | return 0; | |
12852 | } | |
12853 | ||
12854 | int bpf_object__attach_skeleton(struct bpf_object_skeleton *s) | |
12855 | { | |
e9fc3ce9 | 12856 | int i, err; |
d66562fb AN |
12857 | |
12858 | for (i = 0; i < s->prog_cnt; i++) { | |
12859 | struct bpf_program *prog = *s->progs[i].prog; | |
12860 | struct bpf_link **link = s->progs[i].link; | |
d66562fb | 12861 | |
43cb8cba | 12862 | if (!prog->autoload || !prog->autoattach) |
d9297581 AN |
12863 | continue; |
12864 | ||
5532dfd4 | 12865 | /* auto-attaching not supported for this program */ |
4fa5bcfe | 12866 | if (!prog->sec_def || !prog->sec_def->prog_attach_fn) |
d66562fb AN |
12867 | continue; |
12868 | ||
4fa5bcfe AN |
12869 | /* if user already set the link manually, don't attempt auto-attach */ |
12870 | if (*link) | |
12871 | continue; | |
12872 | ||
12873 | err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link); | |
e9fc3ce9 | 12874 | if (err) { |
4fa5bcfe | 12875 | pr_warn("prog '%s': failed to auto-attach: %d\n", |
e9fc3ce9 AN |
12876 | bpf_program__name(prog), err); |
12877 | return libbpf_err(err); | |
d66562fb | 12878 | } |
4fa5bcfe AN |
12879 | |
12880 | /* It's possible that for some SEC() definitions auto-attach | |
12881 | * is supported in some cases (e.g., if definition completely | |
12882 | * specifies target information), but is not in other cases. | |
12883 | * SEC("uprobe") is one such case. If user specified target | |
12884 | * binary and function name, such BPF program can be | |
12885 | * auto-attached. But if not, it shouldn't trigger skeleton's | |
12886 | * attach to fail. It should just be skipped. | |
12887 | * attach_fn signals such case with returning 0 (no error) and | |
12888 | * setting link to NULL. | |
12889 | */ | |
d66562fb AN |
12890 | } |
12891 | ||
12892 | return 0; | |
12893 | } | |
12894 | ||
12895 | void bpf_object__detach_skeleton(struct bpf_object_skeleton *s) | |
12896 | { | |
12897 | int i; | |
12898 | ||
12899 | for (i = 0; i < s->prog_cnt; i++) { | |
12900 | struct bpf_link **link = s->progs[i].link; | |
12901 | ||
50450fc7 | 12902 | bpf_link__destroy(*link); |
d66562fb AN |
12903 | *link = NULL; |
12904 | } | |
12905 | } | |
12906 | ||
12907 | void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s) | |
12908 | { | |
a32ea51a YS |
12909 | if (!s) |
12910 | return; | |
12911 | ||
d66562fb AN |
12912 | if (s->progs) |
12913 | bpf_object__detach_skeleton(s); | |
12914 | if (s->obj) | |
12915 | bpf_object__close(*s->obj); | |
12916 | free(s->maps); | |
12917 | free(s->progs); | |
12918 | free(s); | |
12919 | } |