Debian Codegraph @

tools/lib/bpf/bpf.c

raw
1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2
3 /*
4 * common eBPF ELF operations.
5 *
6 * Copyright (C) 2013-2015 Alexei Starovoitov <[email protected]>
7 * Copyright (C) 2015 Wang Nan <[email protected]>
8 * Copyright (C) 2015 Huawei Inc.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation;
13 * version 2.1 of the License (not later!)
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU Lesser General Public License for more details.
19 *
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with this program; if not, see <http://www.gnu.org/licenses>
22 */
23
24 #include <stdlib.h>
25 #include <string.h>
26 #include <memory.h>
27 #include <unistd.h>
28 #include <asm/unistd.h>
29 #include <errno.h>
30 #include <linux/bpf.h>
31 #include <linux/filter.h>
32 #include <linux/kernel.h>
33 #include <limits.h>
34 #include <sys/resource.h>
35 #include "bpf.h"
36 #include "libbpf.h"
37 #include "libbpf_internal.h"
38
39 /*
40 * When building perf, unistd.h is overridden. __NR_bpf is
41 * required to be defined explicitly.
42 */
43 #ifndef __NR_bpf
44 # if defined(__i386__)
45 # define __NR_bpf 357
46 # elif defined(__x86_64__)
47 # define __NR_bpf 321
48 # elif defined(__aarch64__)
49 # define __NR_bpf 280
50 # elif defined(__sparc__)
51 # define __NR_bpf 349
52 # elif defined(__s390__)
53 # define __NR_bpf 351
54 # elif defined(__arc__)
55 # define __NR_bpf 280
56 # elif defined(__mips__) && defined(_ABIO32)
57 # define __NR_bpf 4355
58 # elif defined(__mips__) && defined(_ABIN32)
59 # define __NR_bpf 6319
60 # elif defined(__mips__) && defined(_ABI64)
61 # define __NR_bpf 5315
62 # else
63 # error __NR_bpf not defined. libbpf does not support your arch.
64 # endif
65 #endif
66
67 static inline __u64 ptr_to_u64(const void *ptr)
68 {
69 return (__u64) (unsigned long) ptr;
70 }
71
72 static inline int sys_bpf(enum bpf_cmd cmd, union bpf_attr *attr,
73 unsigned int size)
74 {
75 return syscall(__NR_bpf, cmd, attr, size);
76 }
77
78 static inline int sys_bpf_fd(enum bpf_cmd cmd, union bpf_attr *attr,
79 unsigned int size)
80 {
81 int fd;
82
83 fd = sys_bpf(cmd, attr, size);
84 return ensure_good_fd(fd);
85 }
86
87 int sys_bpf_prog_load(union bpf_attr *attr, unsigned int size, int attempts)
88 {
89 int fd;
90
91 do {
92 fd = sys_bpf_fd(BPF_PROG_LOAD, attr, size);
93 } while (fd < 0 && errno == EAGAIN && --attempts > 0);
94
95 return fd;
96 }
97
98 /* Probe whether kernel switched from memlock-based (RLIMIT_MEMLOCK) to
99 * memcg-based memory accounting for BPF maps and progs. This was done in [0].
100 * We use the support for bpf_ktime_get_coarse_ns() helper, which was added in
101 * the same 5.11 Linux release ([1]), to detect memcg-based accounting for BPF.
102 *
103 * [0] https://lore.kernel.org/bpf/[email protected]/
104 * [1] d05512618056 ("bpf: Add bpf_ktime_get_coarse_ns helper")
105 */
106 int probe_memcg_account(int token_fd)
107 {
108 const size_t attr_sz = offsetofend(union bpf_attr, prog_token_fd);
109 struct bpf_insn insns[] = {
110 BPF_EMIT_CALL(BPF_FUNC_ktime_get_coarse_ns),
111 BPF_EXIT_INSN(),
112 };
113 size_t insn_cnt = ARRAY_SIZE(insns);
114 union bpf_attr attr;
115 int prog_fd;
116
117 /* attempt loading freplace trying to use custom BTF */
118 memset(&attr, 0, attr_sz);
119 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
120 attr.insns = ptr_to_u64(insns);
121 attr.insn_cnt = insn_cnt;
122 attr.license = ptr_to_u64("GPL");
123 attr.prog_token_fd = token_fd;
124 if (token_fd)
125 attr.prog_flags |= BPF_F_TOKEN_FD;
126
127 prog_fd = sys_bpf_fd(BPF_PROG_LOAD, &attr, attr_sz);
128 if (prog_fd >= 0) {
129 close(prog_fd);
130 return 1;
131 }
132 return 0;
133 }
134
135 static bool memlock_bumped;
136 static rlim_t memlock_rlim = RLIM_INFINITY;
137
138 int libbpf_set_memlock_rlim(size_t memlock_bytes)
139 {
140 if (memlock_bumped)
141 return libbpf_err(-EBUSY);
142
143 memlock_rlim = memlock_bytes;
144 return 0;
145 }
146
147 int bump_rlimit_memlock(void)
148 {
149 struct rlimit rlim;
150
151 /* if kernel supports memcg-based accounting, skip bumping RLIMIT_MEMLOCK */
152 if (memlock_bumped || feat_supported(NULL, FEAT_MEMCG_ACCOUNT))
153 return 0;
154
155 memlock_bumped = true;
156
157 /* zero memlock_rlim disables auto-bumping RLIMIT_MEMLOCK */
158 if (memlock_rlim == 0)
159 return 0;
160
161 rlim.rlim_cur = rlim.rlim_max = memlock_rlim;
162 if (setrlimit(RLIMIT_MEMLOCK, &rlim))
163 return -errno;
164
165 return 0;
166 }
167
168 int bpf_map_create(enum bpf_map_type map_type,
169 const char *map_name,
170 __u32 key_size,
171 __u32 value_size,
172 __u32 max_entries,
173 const struct bpf_map_create_opts *opts)
174 {
175 const size_t attr_sz = offsetofend(union bpf_attr, excl_prog_hash_size);
176 union bpf_attr attr;
177 int fd;
178
179 bump_rlimit_memlock();
180
181 memset(&attr, 0, attr_sz);
182
183 if (!OPTS_VALID(opts, bpf_map_create_opts))
184 return libbpf_err(-EINVAL);
185
186 attr.map_type = map_type;
187 if (map_name && feat_supported(NULL, FEAT_PROG_NAME))
188 libbpf_strlcpy(attr.map_name, map_name, sizeof(attr.map_name));
189 attr.key_size = key_size;
190 attr.value_size = value_size;
191 attr.max_entries = max_entries;
192
193 attr.btf_fd = OPTS_GET(opts, btf_fd, 0);
194 attr.btf_key_type_id = OPTS_GET(opts, btf_key_type_id, 0);
195 attr.btf_value_type_id = OPTS_GET(opts, btf_value_type_id, 0);
196 attr.btf_vmlinux_value_type_id = OPTS_GET(opts, btf_vmlinux_value_type_id, 0);
197 attr.value_type_btf_obj_fd = OPTS_GET(opts, value_type_btf_obj_fd, 0);
198
199 attr.inner_map_fd = OPTS_GET(opts, inner_map_fd, 0);
200 attr.map_flags = OPTS_GET(opts, map_flags, 0);
201 attr.map_extra = OPTS_GET(opts, map_extra, 0);
202 attr.numa_node = OPTS_GET(opts, numa_node, 0);
203 attr.map_ifindex = OPTS_GET(opts, map_ifindex, 0);
204
205 attr.map_token_fd = OPTS_GET(opts, token_fd, 0);
206 attr.excl_prog_hash = ptr_to_u64(OPTS_GET(opts, excl_prog_hash, NULL));
207 attr.excl_prog_hash_size = OPTS_GET(opts, excl_prog_hash_size, 0);
208
209 fd = sys_bpf_fd(BPF_MAP_CREATE, &attr, attr_sz);
210 return libbpf_err_errno(fd);
211 }
212
213 static void *
214 alloc_zero_tailing_info(const void *orecord, __u32 cnt,
215 __u32 actual_rec_size, __u32 expected_rec_size)
216 {
217 __u64 info_len = (__u64)actual_rec_size * cnt;
218 void *info, *nrecord;
219 int i;
220
221 info = malloc(info_len);
222 if (!info)
223 return NULL;
224
225 /* zero out bytes kernel does not understand */
226 nrecord = info;
227 for (i = 0; i < cnt; i++) {
228 memcpy(nrecord, orecord, expected_rec_size);
229 memset(nrecord + expected_rec_size, 0,
230 actual_rec_size - expected_rec_size);
231 orecord += actual_rec_size;
232 nrecord += actual_rec_size;
233 }
234
235 return info;
236 }
237
238 int bpf_prog_load(enum bpf_prog_type prog_type,
239 const char *prog_name, const char *license,
240 const struct bpf_insn *insns, size_t insn_cnt,
241 struct bpf_prog_load_opts *opts)
242 {
243 const size_t attr_sz = offsetofend(union bpf_attr, keyring_id);
244 void *finfo = NULL, *linfo = NULL;
245 const char *func_info, *line_info;
246 __u32 log_size, log_level, attach_prog_fd, attach_btf_obj_fd;
247 __u32 func_info_rec_size, line_info_rec_size;
248 int fd, attempts;
249 union bpf_attr attr;
250 char *log_buf;
251
252 bump_rlimit_memlock();
253
254 if (!OPTS_VALID(opts, bpf_prog_load_opts))
255 return libbpf_err(-EINVAL);
256
257 attempts = OPTS_GET(opts, attempts, 0);
258 if (attempts < 0)
259 return libbpf_err(-EINVAL);
260 if (attempts == 0)
261 attempts = PROG_LOAD_ATTEMPTS;
262
263 memset(&attr, 0, attr_sz);
264
265 attr.prog_type = prog_type;
266 attr.expected_attach_type = OPTS_GET(opts, expected_attach_type, 0);
267
268 attr.prog_btf_fd = OPTS_GET(opts, prog_btf_fd, 0);
269 attr.prog_flags = OPTS_GET(opts, prog_flags, 0);
270 attr.prog_ifindex = OPTS_GET(opts, prog_ifindex, 0);
271 attr.kern_version = OPTS_GET(opts, kern_version, 0);
272 attr.prog_token_fd = OPTS_GET(opts, token_fd, 0);
273
274 if (prog_name && feat_supported(NULL, FEAT_PROG_NAME))
275 libbpf_strlcpy(attr.prog_name, prog_name, sizeof(attr.prog_name));
276 attr.license = ptr_to_u64(license);
277
278 if (insn_cnt > UINT_MAX)
279 return libbpf_err(-E2BIG);
280
281 attr.insns = ptr_to_u64(insns);
282 attr.insn_cnt = (__u32)insn_cnt;
283
284 attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
285 attach_btf_obj_fd = OPTS_GET(opts, attach_btf_obj_fd, 0);
286
287 if (attach_prog_fd && attach_btf_obj_fd)
288 return libbpf_err(-EINVAL);
289
290 attr.attach_btf_id = OPTS_GET(opts, attach_btf_id, 0);
291 if (attach_prog_fd)
292 attr.attach_prog_fd = attach_prog_fd;
293 else
294 attr.attach_btf_obj_fd = attach_btf_obj_fd;
295
296 log_buf = OPTS_GET(opts, log_buf, NULL);
297 log_size = OPTS_GET(opts, log_size, 0);
298 log_level = OPTS_GET(opts, log_level, 0);
299
300 if (!!log_buf != !!log_size)
301 return libbpf_err(-EINVAL);
302
303 func_info_rec_size = OPTS_GET(opts, func_info_rec_size, 0);
304 func_info = OPTS_GET(opts, func_info, NULL);
305 attr.func_info_rec_size = func_info_rec_size;
306 attr.func_info = ptr_to_u64(func_info);
307 attr.func_info_cnt = OPTS_GET(opts, func_info_cnt, 0);
308
309 line_info_rec_size = OPTS_GET(opts, line_info_rec_size, 0);
310 line_info = OPTS_GET(opts, line_info, NULL);
311 attr.line_info_rec_size = line_info_rec_size;
312 attr.line_info = ptr_to_u64(line_info);
313 attr.line_info_cnt = OPTS_GET(opts, line_info_cnt, 0);
314
315 attr.fd_array = ptr_to_u64(OPTS_GET(opts, fd_array, NULL));
316 attr.fd_array_cnt = OPTS_GET(opts, fd_array_cnt, 0);
317
318 if (log_level) {
319 attr.log_buf = ptr_to_u64(log_buf);
320 attr.log_size = log_size;
321 attr.log_level = log_level;
322 }
323
324 fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
325 OPTS_SET(opts, log_true_size, attr.log_true_size);
326 if (fd >= 0)
327 return fd;
328
329 /* After bpf_prog_load, the kernel may modify certain attributes
330 * to give user space a hint how to deal with loading failure.
331 * Check to see whether we can make some changes and load again.
332 */
333 while (errno == E2BIG && (!finfo || !linfo)) {
334 if (!finfo && attr.func_info_cnt &&
335 attr.func_info_rec_size < func_info_rec_size) {
336 /* try with corrected func info records */
337 finfo = alloc_zero_tailing_info(func_info,
338 attr.func_info_cnt,
339 func_info_rec_size,
340 attr.func_info_rec_size);
341 if (!finfo) {
342 errno = E2BIG;
343 goto done;
344 }
345
346 attr.func_info = ptr_to_u64(finfo);
347 attr.func_info_rec_size = func_info_rec_size;
348 } else if (!linfo && attr.line_info_cnt &&
349 attr.line_info_rec_size < line_info_rec_size) {
350 linfo = alloc_zero_tailing_info(line_info,
351 attr.line_info_cnt,
352 line_info_rec_size,
353 attr.line_info_rec_size);
354 if (!linfo) {
355 errno = E2BIG;
356 goto done;
357 }
358
359 attr.line_info = ptr_to_u64(linfo);
360 attr.line_info_rec_size = line_info_rec_size;
361 } else {
362 break;
363 }
364
365 fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
366 OPTS_SET(opts, log_true_size, attr.log_true_size);
367 if (fd >= 0)
368 goto done;
369 }
370
371 if (log_level == 0 && log_buf) {
372 /* log_level == 0 with non-NULL log_buf requires retrying on error
373 * with log_level == 1 and log_buf/log_buf_size set, to get details of
374 * failure
375 */
376 attr.log_buf = ptr_to_u64(log_buf);
377 attr.log_size = log_size;
378 attr.log_level = 1;
379
380 fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
381 OPTS_SET(opts, log_true_size, attr.log_true_size);
382 }
383 done:
384 /* free() doesn't affect errno, so we don't need to restore it */
385 free(finfo);
386 free(linfo);
387 return libbpf_err_errno(fd);
388 }
389
390 int bpf_map_update_elem(int fd, const void *key, const void *value,
391 __u64 flags)
392 {
393 const size_t attr_sz = offsetofend(union bpf_attr, flags);
394 union bpf_attr attr;
395 int ret;
396
397 memset(&attr, 0, attr_sz);
398 attr.map_fd = fd;
399 attr.key = ptr_to_u64(key);
400 attr.value = ptr_to_u64(value);
401 attr.flags = flags;
402
403 ret = sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, attr_sz);
404 return libbpf_err_errno(ret);
405 }
406
407 int bpf_map_lookup_elem(int fd, const void *key, void *value)
408 {
409 const size_t attr_sz = offsetofend(union bpf_attr, flags);
410 union bpf_attr attr;
411 int ret;
412
413 memset(&attr, 0, attr_sz);
414 attr.map_fd = fd;
415 attr.key = ptr_to_u64(key);
416 attr.value = ptr_to_u64(value);
417
418 ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, attr_sz);
419 return libbpf_err_errno(ret);
420 }
421
422 int bpf_map_lookup_elem_flags(int fd, const void *key, void *value, __u64 flags)
423 {
424 const size_t attr_sz = offsetofend(union bpf_attr, flags);
425 union bpf_attr attr;
426 int ret;
427
428 memset(&attr, 0, attr_sz);
429 attr.map_fd = fd;
430 attr.key = ptr_to_u64(key);
431 attr.value = ptr_to_u64(value);
432 attr.flags = flags;
433
434 ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, attr_sz);
435 return libbpf_err_errno(ret);
436 }
437
438 int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value)
439 {
440 const size_t attr_sz = offsetofend(union bpf_attr, flags);
441 union bpf_attr attr;
442 int ret;
443
444 memset(&attr, 0, attr_sz);
445 attr.map_fd = fd;
446 attr.key = ptr_to_u64(key);
447 attr.value = ptr_to_u64(value);
448
449 ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, attr_sz);
450 return libbpf_err_errno(ret);
451 }
452
453 int bpf_map_lookup_and_delete_elem_flags(int fd, const void *key, void *value, __u64 flags)
454 {
455 const size_t attr_sz = offsetofend(union bpf_attr, flags);
456 union bpf_attr attr;
457 int ret;
458
459 memset(&attr, 0, attr_sz);
460 attr.map_fd = fd;
461 attr.key = ptr_to_u64(key);
462 attr.value = ptr_to_u64(value);
463 attr.flags = flags;
464
465 ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, attr_sz);
466 return libbpf_err_errno(ret);
467 }
468
469 int bpf_map_delete_elem(int fd, const void *key)
470 {
471 const size_t attr_sz = offsetofend(union bpf_attr, flags);
472 union bpf_attr attr;
473 int ret;
474
475 memset(&attr, 0, attr_sz);
476 attr.map_fd = fd;
477 attr.key = ptr_to_u64(key);
478
479 ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, attr_sz);
480 return libbpf_err_errno(ret);
481 }
482
483 int bpf_map_delete_elem_flags(int fd, const void *key, __u64 flags)
484 {
485 const size_t attr_sz = offsetofend(union bpf_attr, flags);
486 union bpf_attr attr;
487 int ret;
488
489 memset(&attr, 0, attr_sz);
490 attr.map_fd = fd;
491 attr.key = ptr_to_u64(key);
492 attr.flags = flags;
493
494 ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, attr_sz);
495 return libbpf_err_errno(ret);
496 }
497
498 int bpf_map_get_next_key(int fd, const void *key, void *next_key)
499 {
500 const size_t attr_sz = offsetofend(union bpf_attr, next_key);
501 union bpf_attr attr;
502 int ret;
503
504 memset(&attr, 0, attr_sz);
505 attr.map_fd = fd;
506 attr.key = ptr_to_u64(key);
507 attr.next_key = ptr_to_u64(next_key);
508
509 ret = sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, attr_sz);
510 return libbpf_err_errno(ret);
511 }
512
513 int bpf_map_freeze(int fd)
514 {
515 const size_t attr_sz = offsetofend(union bpf_attr, map_fd);
516 union bpf_attr attr;
517 int ret;
518
519 memset(&attr, 0, attr_sz);
520 attr.map_fd = fd;
521
522 ret = sys_bpf(BPF_MAP_FREEZE, &attr, attr_sz);
523 return libbpf_err_errno(ret);
524 }
525
526 static int bpf_map_batch_common(int cmd, int fd, void *in_batch,
527 void *out_batch, void *keys, void *values,
528 __u32 *count,
529 const struct bpf_map_batch_opts *opts)
530 {
531 const size_t attr_sz = offsetofend(union bpf_attr, batch);
532 union bpf_attr attr;
533 int ret;
534
535 if (!OPTS_VALID(opts, bpf_map_batch_opts))
536 return libbpf_err(-EINVAL);
537
538 memset(&attr, 0, attr_sz);
539 attr.batch.map_fd = fd;
540 attr.batch.in_batch = ptr_to_u64(in_batch);
541 attr.batch.out_batch = ptr_to_u64(out_batch);
542 attr.batch.keys = ptr_to_u64(keys);
543 attr.batch.values = ptr_to_u64(values);
544 attr.batch.count = *count;
545 attr.batch.elem_flags = OPTS_GET(opts, elem_flags, 0);
546 attr.batch.flags = OPTS_GET(opts, flags, 0);
547
548 ret = sys_bpf(cmd, &attr, attr_sz);
549 *count = attr.batch.count;
550
551 return libbpf_err_errno(ret);
552 }
553
554 int bpf_map_delete_batch(int fd, const void *keys, __u32 *count,
555 const struct bpf_map_batch_opts *opts)
556 {
557 return bpf_map_batch_common(BPF_MAP_DELETE_BATCH, fd, NULL,
558 NULL, (void *)keys, NULL, count, opts);
559 }
560
561 int bpf_map_lookup_batch(int fd, void *in_batch, void *out_batch, void *keys,
562 void *values, __u32 *count,
563 const struct bpf_map_batch_opts *opts)
564 {
565 return bpf_map_batch_common(BPF_MAP_LOOKUP_BATCH, fd, in_batch,
566 out_batch, keys, values, count, opts);
567 }
568
569 int bpf_map_lookup_and_delete_batch(int fd, void *in_batch, void *out_batch,
570 void *keys, void *values, __u32 *count,
571 const struct bpf_map_batch_opts *opts)
572 {
573 return bpf_map_batch_common(BPF_MAP_LOOKUP_AND_DELETE_BATCH,
574 fd, in_batch, out_batch, keys, values,
575 count, opts);
576 }
577
578 int bpf_map_update_batch(int fd, const void *keys, const void *values, __u32 *count,
579 const struct bpf_map_batch_opts *opts)
580 {
581 return bpf_map_batch_common(BPF_MAP_UPDATE_BATCH, fd, NULL, NULL,
582 (void *)keys, (void *)values, count, opts);
583 }
584
585 int bpf_obj_pin_opts(int fd, const char *pathname, const struct bpf_obj_pin_opts *opts)
586 {
587 const size_t attr_sz = offsetofend(union bpf_attr, path_fd);
588 union bpf_attr attr;
589 int ret;
590
591 if (!OPTS_VALID(opts, bpf_obj_pin_opts))
592 return libbpf_err(-EINVAL);
593
594 memset(&attr, 0, attr_sz);
595 attr.path_fd = OPTS_GET(opts, path_fd, 0);
596 attr.pathname = ptr_to_u64((void *)pathname);
597 attr.file_flags = OPTS_GET(opts, file_flags, 0);
598 attr.bpf_fd = fd;
599
600 ret = sys_bpf(BPF_OBJ_PIN, &attr, attr_sz);
601 return libbpf_err_errno(ret);
602 }
603
604 int bpf_obj_pin(int fd, const char *pathname)
605 {
606 return bpf_obj_pin_opts(fd, pathname, NULL);
607 }
608
609 int bpf_obj_get(const char *pathname)
610 {
611 return bpf_obj_get_opts(pathname, NULL);
612 }
613
614 int bpf_obj_get_opts(const char *pathname, const struct bpf_obj_get_opts *opts)
615 {
616 const size_t attr_sz = offsetofend(union bpf_attr, path_fd);
617 union bpf_attr attr;
618 int fd;
619
620 if (!OPTS_VALID(opts, bpf_obj_get_opts))
621 return libbpf_err(-EINVAL);
622
623 memset(&attr, 0, attr_sz);
624 attr.path_fd = OPTS_GET(opts, path_fd, 0);
625 attr.pathname = ptr_to_u64((void *)pathname);
626 attr.file_flags = OPTS_GET(opts, file_flags, 0);
627
628 fd = sys_bpf_fd(BPF_OBJ_GET, &attr, attr_sz);
629 return libbpf_err_errno(fd);
630 }
631
632 int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type,
633 unsigned int flags)
634 {
635 DECLARE_LIBBPF_OPTS(bpf_prog_attach_opts, opts,
636 .flags = flags,
637 );
638
639 return bpf_prog_attach_opts(prog_fd, target_fd, type, &opts);
640 }
641
642 int bpf_prog_attach_opts(int prog_fd, int target, enum bpf_attach_type type,
643 const struct bpf_prog_attach_opts *opts)
644 {
645 const size_t attr_sz = offsetofend(union bpf_attr, expected_revision);
646 __u32 relative_id, flags;
647 int ret, relative_fd;
648 union bpf_attr attr;
649
650 if (!OPTS_VALID(opts, bpf_prog_attach_opts))
651 return libbpf_err(-EINVAL);
652
653 relative_id = OPTS_GET(opts, relative_id, 0);
654 relative_fd = OPTS_GET(opts, relative_fd, 0);
655 flags = OPTS_GET(opts, flags, 0);
656
657 /* validate we don't have unexpected combinations of non-zero fields */
658 if (relative_fd && relative_id)
659 return libbpf_err(-EINVAL);
660
661 memset(&attr, 0, attr_sz);
662 attr.target_fd = target;
663 attr.attach_bpf_fd = prog_fd;
664 attr.attach_type = type;
665 attr.replace_bpf_fd = OPTS_GET(opts, replace_fd, 0);
666 attr.expected_revision = OPTS_GET(opts, expected_revision, 0);
667
668 if (relative_id) {
669 attr.attach_flags = flags | BPF_F_ID;
670 attr.relative_id = relative_id;
671 } else {
672 attr.attach_flags = flags;
673 attr.relative_fd = relative_fd;
674 }
675
676 ret = sys_bpf(BPF_PROG_ATTACH, &attr, attr_sz);
677 return libbpf_err_errno(ret);
678 }
679
680 int bpf_prog_detach_opts(int prog_fd, int target, enum bpf_attach_type type,
681 const struct bpf_prog_detach_opts *opts)
682 {
683 const size_t attr_sz = offsetofend(union bpf_attr, expected_revision);
684 __u32 relative_id, flags;
685 int ret, relative_fd;
686 union bpf_attr attr;
687
688 if (!OPTS_VALID(opts, bpf_prog_detach_opts))
689 return libbpf_err(-EINVAL);
690
691 relative_id = OPTS_GET(opts, relative_id, 0);
692 relative_fd = OPTS_GET(opts, relative_fd, 0);
693 flags = OPTS_GET(opts, flags, 0);
694
695 /* validate we don't have unexpected combinations of non-zero fields */
696 if (relative_fd && relative_id)
697 return libbpf_err(-EINVAL);
698
699 memset(&attr, 0, attr_sz);
700 attr.target_fd = target;
701 attr.attach_bpf_fd = prog_fd;
702 attr.attach_type = type;
703 attr.expected_revision = OPTS_GET(opts, expected_revision, 0);
704
705 if (relative_id) {
706 attr.attach_flags = flags | BPF_F_ID;
707 attr.relative_id = relative_id;
708 } else {
709 attr.attach_flags = flags;
710 attr.relative_fd = relative_fd;
711 }
712
713 ret = sys_bpf(BPF_PROG_DETACH, &attr, attr_sz);
714 return libbpf_err_errno(ret);
715 }
716
717 int bpf_prog_detach(int target_fd, enum bpf_attach_type type)
718 {
719 return bpf_prog_detach_opts(0, target_fd, type, NULL);
720 }
721
722 int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type)
723 {
724 return bpf_prog_detach_opts(prog_fd, target_fd, type, NULL);
725 }
726
727 int bpf_link_create(int prog_fd, int target_fd,
728 enum bpf_attach_type attach_type,
729 const struct bpf_link_create_opts *opts)
730 {
731 const size_t attr_sz = offsetofend(union bpf_attr, link_create);
732 __u32 target_btf_id, iter_info_len, relative_id;
733 int fd, err, relative_fd;
734 union bpf_attr attr;
735
736 if (!OPTS_VALID(opts, bpf_link_create_opts))
737 return libbpf_err(-EINVAL);
738
739 iter_info_len = OPTS_GET(opts, iter_info_len, 0);
740 target_btf_id = OPTS_GET(opts, target_btf_id, 0);
741
742 /* validate we don't have unexpected combinations of non-zero fields */
743 if (iter_info_len || target_btf_id) {
744 if (iter_info_len && target_btf_id)
745 return libbpf_err(-EINVAL);
746 if (!OPTS_ZEROED(opts, target_btf_id))
747 return libbpf_err(-EINVAL);
748 }
749
750 memset(&attr, 0, attr_sz);
751 attr.link_create.prog_fd = prog_fd;
752 attr.link_create.target_fd = target_fd;
753 attr.link_create.attach_type = attach_type;
754 attr.link_create.flags = OPTS_GET(opts, flags, 0);
755
756 if (target_btf_id) {
757 attr.link_create.target_btf_id = target_btf_id;
758 goto proceed;
759 }
760
761 switch (attach_type) {
762 case BPF_TRACE_ITER:
763 attr.link_create.iter_info = ptr_to_u64(OPTS_GET(opts, iter_info, (void *)0));
764 attr.link_create.iter_info_len = iter_info_len;
765 break;
766 case BPF_PERF_EVENT:
767 attr.link_create.perf_event.bpf_cookie = OPTS_GET(opts, perf_event.bpf_cookie, 0);
768 if (!OPTS_ZEROED(opts, perf_event))
769 return libbpf_err(-EINVAL);
770 break;
771 case BPF_TRACE_KPROBE_MULTI:
772 case BPF_TRACE_KPROBE_SESSION:
773 attr.link_create.kprobe_multi.flags = OPTS_GET(opts, kprobe_multi.flags, 0);
774 attr.link_create.kprobe_multi.cnt = OPTS_GET(opts, kprobe_multi.cnt, 0);
775 attr.link_create.kprobe_multi.syms = ptr_to_u64(OPTS_GET(opts, kprobe_multi.syms, 0));
776 attr.link_create.kprobe_multi.addrs = ptr_to_u64(OPTS_GET(opts, kprobe_multi.addrs, 0));
777 attr.link_create.kprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, kprobe_multi.cookies, 0));
778 if (!OPTS_ZEROED(opts, kprobe_multi))
779 return libbpf_err(-EINVAL);
780 break;
781 case BPF_TRACE_UPROBE_MULTI:
782 case BPF_TRACE_UPROBE_SESSION:
783 attr.link_create.uprobe_multi.flags = OPTS_GET(opts, uprobe_multi.flags, 0);
784 attr.link_create.uprobe_multi.cnt = OPTS_GET(opts, uprobe_multi.cnt, 0);
785 attr.link_create.uprobe_multi.path = ptr_to_u64(OPTS_GET(opts, uprobe_multi.path, 0));
786 attr.link_create.uprobe_multi.offsets = ptr_to_u64(OPTS_GET(opts, uprobe_multi.offsets, 0));
787 attr.link_create.uprobe_multi.ref_ctr_offsets = ptr_to_u64(OPTS_GET(opts, uprobe_multi.ref_ctr_offsets, 0));
788 attr.link_create.uprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, uprobe_multi.cookies, 0));
789 attr.link_create.uprobe_multi.pid = OPTS_GET(opts, uprobe_multi.pid, 0);
790 if (!OPTS_ZEROED(opts, uprobe_multi))
791 return libbpf_err(-EINVAL);
792 break;
793 case BPF_TRACE_RAW_TP:
794 case BPF_TRACE_FENTRY:
795 case BPF_TRACE_FEXIT:
796 case BPF_MODIFY_RETURN:
797 case BPF_TRACE_FSESSION:
798 case BPF_LSM_MAC:
799 attr.link_create.tracing.cookie = OPTS_GET(opts, tracing.cookie, 0);
800 if (!OPTS_ZEROED(opts, tracing))
801 return libbpf_err(-EINVAL);
802 break;
803 case BPF_NETFILTER:
804 attr.link_create.netfilter.pf = OPTS_GET(opts, netfilter.pf, 0);
805 attr.link_create.netfilter.hooknum = OPTS_GET(opts, netfilter.hooknum, 0);
806 attr.link_create.netfilter.priority = OPTS_GET(opts, netfilter.priority, 0);
807 attr.link_create.netfilter.flags = OPTS_GET(opts, netfilter.flags, 0);
808 if (!OPTS_ZEROED(opts, netfilter))
809 return libbpf_err(-EINVAL);
810 break;
811 case BPF_TCX_INGRESS:
812 case BPF_TCX_EGRESS:
813 relative_fd = OPTS_GET(opts, tcx.relative_fd, 0);
814 relative_id = OPTS_GET(opts, tcx.relative_id, 0);
815 if (relative_fd && relative_id)
816 return libbpf_err(-EINVAL);
817 if (relative_id) {
818 attr.link_create.tcx.relative_id = relative_id;
819 attr.link_create.flags |= BPF_F_ID;
820 } else {
821 attr.link_create.tcx.relative_fd = relative_fd;
822 }
823 attr.link_create.tcx.expected_revision = OPTS_GET(opts, tcx.expected_revision, 0);
824 if (!OPTS_ZEROED(opts, tcx))
825 return libbpf_err(-EINVAL);
826 break;
827 case BPF_NETKIT_PRIMARY:
828 case BPF_NETKIT_PEER:
829 relative_fd = OPTS_GET(opts, netkit.relative_fd, 0);
830 relative_id = OPTS_GET(opts, netkit.relative_id, 0);
831 if (relative_fd && relative_id)
832 return libbpf_err(-EINVAL);
833 if (relative_id) {
834 attr.link_create.netkit.relative_id = relative_id;
835 attr.link_create.flags |= BPF_F_ID;
836 } else {
837 attr.link_create.netkit.relative_fd = relative_fd;
838 }
839 attr.link_create.netkit.expected_revision = OPTS_GET(opts, netkit.expected_revision, 0);
840 if (!OPTS_ZEROED(opts, netkit))
841 return libbpf_err(-EINVAL);
842 break;
843 case BPF_CGROUP_INET_INGRESS:
844 case BPF_CGROUP_INET_EGRESS:
845 case BPF_CGROUP_INET_SOCK_CREATE:
846 case BPF_CGROUP_INET_SOCK_RELEASE:
847 case BPF_CGROUP_INET4_BIND:
848 case BPF_CGROUP_INET6_BIND:
849 case BPF_CGROUP_INET4_POST_BIND:
850 case BPF_CGROUP_INET6_POST_BIND:
851 case BPF_CGROUP_INET4_CONNECT:
852 case BPF_CGROUP_INET6_CONNECT:
853 case BPF_CGROUP_UNIX_CONNECT:
854 case BPF_CGROUP_INET4_GETPEERNAME:
855 case BPF_CGROUP_INET6_GETPEERNAME:
856 case BPF_CGROUP_UNIX_GETPEERNAME:
857 case BPF_CGROUP_INET4_GETSOCKNAME:
858 case BPF_CGROUP_INET6_GETSOCKNAME:
859 case BPF_CGROUP_UNIX_GETSOCKNAME:
860 case BPF_CGROUP_UDP4_SENDMSG:
861 case BPF_CGROUP_UDP6_SENDMSG:
862 case BPF_CGROUP_UNIX_SENDMSG:
863 case BPF_CGROUP_UDP4_RECVMSG:
864 case BPF_CGROUP_UDP6_RECVMSG:
865 case BPF_CGROUP_UNIX_RECVMSG:
866 case BPF_CGROUP_SOCK_OPS:
867 case BPF_CGROUP_DEVICE:
868 case BPF_CGROUP_SYSCTL:
869 case BPF_CGROUP_GETSOCKOPT:
870 case BPF_CGROUP_SETSOCKOPT:
871 case BPF_LSM_CGROUP:
872 relative_fd = OPTS_GET(opts, cgroup.relative_fd, 0);
873 relative_id = OPTS_GET(opts, cgroup.relative_id, 0);
874 if (relative_fd && relative_id)
875 return libbpf_err(-EINVAL);
876 if (relative_id) {
877 attr.link_create.cgroup.relative_id = relative_id;
878 attr.link_create.flags |= BPF_F_ID;
879 } else {
880 attr.link_create.cgroup.relative_fd = relative_fd;
881 }
882 attr.link_create.cgroup.expected_revision =
883 OPTS_GET(opts, cgroup.expected_revision, 0);
884 if (!OPTS_ZEROED(opts, cgroup))
885 return libbpf_err(-EINVAL);
886 break;
887 default:
888 if (!OPTS_ZEROED(opts, flags))
889 return libbpf_err(-EINVAL);
890 break;
891 }
892 proceed:
893 fd = sys_bpf_fd(BPF_LINK_CREATE, &attr, attr_sz);
894 if (fd >= 0)
895 return fd;
896 /* we'll get EINVAL if LINK_CREATE doesn't support attaching fentry
897 * and other similar programs
898 */
899 err = -errno;
900 if (err != -EINVAL)
901 return libbpf_err(err);
902
903 /* if user used features not supported by
904 * BPF_RAW_TRACEPOINT_OPEN command, then just give up immediately
905 */
906 if (attr.link_create.target_fd || attr.link_create.target_btf_id)
907 return libbpf_err(err);
908 if (!OPTS_ZEROED(opts, sz))
909 return libbpf_err(err);
910
911 /* otherwise, for few select kinds of programs that can be
912 * attached using BPF_RAW_TRACEPOINT_OPEN command, try that as
913 * a fallback for older kernels
914 */
915 switch (attach_type) {
916 case BPF_TRACE_RAW_TP:
917 case BPF_LSM_MAC:
918 case BPF_TRACE_FENTRY:
919 case BPF_TRACE_FEXIT:
920 case BPF_MODIFY_RETURN:
921 return bpf_raw_tracepoint_open(NULL, prog_fd);
922 default:
923 return libbpf_err(err);
924 }
925 }
926
927 int bpf_link_detach(int link_fd)
928 {
929 const size_t attr_sz = offsetofend(union bpf_attr, link_detach);
930 union bpf_attr attr;
931 int ret;
932
933 memset(&attr, 0, attr_sz);
934 attr.link_detach.link_fd = link_fd;
935
936 ret = sys_bpf(BPF_LINK_DETACH, &attr, attr_sz);
937 return libbpf_err_errno(ret);
938 }
939
940 int bpf_link_update(int link_fd, int new_prog_fd,
941 const struct bpf_link_update_opts *opts)
942 {
943 const size_t attr_sz = offsetofend(union bpf_attr, link_update);
944 union bpf_attr attr;
945 int ret;
946
947 if (!OPTS_VALID(opts, bpf_link_update_opts))
948 return libbpf_err(-EINVAL);
949
950 if (OPTS_GET(opts, old_prog_fd, 0) && OPTS_GET(opts, old_map_fd, 0))
951 return libbpf_err(-EINVAL);
952
953 memset(&attr, 0, attr_sz);
954 attr.link_update.link_fd = link_fd;
955 attr.link_update.new_prog_fd = new_prog_fd;
956 attr.link_update.flags = OPTS_GET(opts, flags, 0);
957 if (OPTS_GET(opts, old_prog_fd, 0))
958 attr.link_update.old_prog_fd = OPTS_GET(opts, old_prog_fd, 0);
959 else if (OPTS_GET(opts, old_map_fd, 0))
960 attr.link_update.old_map_fd = OPTS_GET(opts, old_map_fd, 0);
961
962 ret = sys_bpf(BPF_LINK_UPDATE, &attr, attr_sz);
963 return libbpf_err_errno(ret);
964 }
965
966 int bpf_iter_create(int link_fd)
967 {
968 const size_t attr_sz = offsetofend(union bpf_attr, iter_create);
969 union bpf_attr attr;
970 int fd;
971
972 memset(&attr, 0, attr_sz);
973 attr.iter_create.link_fd = link_fd;
974
975 fd = sys_bpf_fd(BPF_ITER_CREATE, &attr, attr_sz);
976 return libbpf_err_errno(fd);
977 }
978
979 int bpf_prog_query_opts(int target, enum bpf_attach_type type,
980 struct bpf_prog_query_opts *opts)
981 {
982 const size_t attr_sz = offsetofend(union bpf_attr, query);
983 union bpf_attr attr;
984 int ret;
985
986 if (!OPTS_VALID(opts, bpf_prog_query_opts))
987 return libbpf_err(-EINVAL);
988
989 memset(&attr, 0, attr_sz);
990 attr.query.target_fd = target;
991 attr.query.attach_type = type;
992 attr.query.query_flags = OPTS_GET(opts, query_flags, 0);
993 attr.query.count = OPTS_GET(opts, count, 0);
994 attr.query.prog_ids = ptr_to_u64(OPTS_GET(opts, prog_ids, NULL));
995 attr.query.link_ids = ptr_to_u64(OPTS_GET(opts, link_ids, NULL));
996 attr.query.prog_attach_flags = ptr_to_u64(OPTS_GET(opts, prog_attach_flags, NULL));
997 attr.query.link_attach_flags = ptr_to_u64(OPTS_GET(opts, link_attach_flags, NULL));
998
999 ret = sys_bpf(BPF_PROG_QUERY, &attr, attr_sz);
1000
1001 OPTS_SET(opts, attach_flags, attr.query.attach_flags);
1002 OPTS_SET(opts, revision, attr.query.revision);
1003 OPTS_SET(opts, count, attr.query.count);
1004
1005 return libbpf_err_errno(ret);
1006 }
1007
1008 int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags,
1009 __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt)
1010 {
1011 LIBBPF_OPTS(bpf_prog_query_opts, opts);
1012 int ret;
1013
1014 opts.query_flags = query_flags;
1015 opts.prog_ids = prog_ids;
1016 opts.prog_cnt = *prog_cnt;
1017
1018 ret = bpf_prog_query_opts(target_fd, type, &opts);
1019
1020 if (attach_flags)
1021 *attach_flags = opts.attach_flags;
1022 *prog_cnt = opts.prog_cnt;
1023
1024 return libbpf_err_errno(ret);
1025 }
1026
1027 int bpf_prog_test_run_opts(int prog_fd, struct bpf_test_run_opts *opts)
1028 {
1029 const size_t attr_sz = offsetofend(union bpf_attr, test);
1030 union bpf_attr attr;
1031 int ret;
1032
1033 if (!OPTS_VALID(opts, bpf_test_run_opts))
1034 return libbpf_err(-EINVAL);
1035
1036 memset(&attr, 0, attr_sz);
1037 attr.test.prog_fd = prog_fd;
1038 attr.test.batch_size = OPTS_GET(opts, batch_size, 0);
1039 attr.test.cpu = OPTS_GET(opts, cpu, 0);
1040 attr.test.flags = OPTS_GET(opts, flags, 0);
1041 attr.test.repeat = OPTS_GET(opts, repeat, 0);
1042 attr.test.duration = OPTS_GET(opts, duration, 0);
1043 attr.test.ctx_size_in = OPTS_GET(opts, ctx_size_in, 0);
1044 attr.test.ctx_size_out = OPTS_GET(opts, ctx_size_out, 0);
1045 attr.test.data_size_in = OPTS_GET(opts, data_size_in, 0);
1046 attr.test.data_size_out = OPTS_GET(opts, data_size_out, 0);
1047 attr.test.ctx_in = ptr_to_u64(OPTS_GET(opts, ctx_in, NULL));
1048 attr.test.ctx_out = ptr_to_u64(OPTS_GET(opts, ctx_out, NULL));
1049 attr.test.data_in = ptr_to_u64(OPTS_GET(opts, data_in, NULL));
1050 attr.test.data_out = ptr_to_u64(OPTS_GET(opts, data_out, NULL));
1051
1052 ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, attr_sz);
1053
1054 OPTS_SET(opts, data_size_out, attr.test.data_size_out);
1055 OPTS_SET(opts, ctx_size_out, attr.test.ctx_size_out);
1056 OPTS_SET(opts, duration, attr.test.duration);
1057 OPTS_SET(opts, retval, attr.test.retval);
1058
1059 return libbpf_err_errno(ret);
1060 }
1061
1062 static int bpf_obj_get_next_id(__u32 start_id, __u32 *next_id, int cmd)
1063 {
1064 const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1065 union bpf_attr attr;
1066 int err;
1067
1068 memset(&attr, 0, attr_sz);
1069 attr.start_id = start_id;
1070
1071 err = sys_bpf(cmd, &attr, attr_sz);
1072 if (!err)
1073 *next_id = attr.next_id;
1074
1075 return libbpf_err_errno(err);
1076 }
1077
1078 int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id)
1079 {
1080 return bpf_obj_get_next_id(start_id, next_id, BPF_PROG_GET_NEXT_ID);
1081 }
1082
1083 int bpf_map_get_next_id(__u32 start_id, __u32 *next_id)
1084 {
1085 return bpf_obj_get_next_id(start_id, next_id, BPF_MAP_GET_NEXT_ID);
1086 }
1087
1088 int bpf_btf_get_next_id(__u32 start_id, __u32 *next_id)
1089 {
1090 return bpf_obj_get_next_id(start_id, next_id, BPF_BTF_GET_NEXT_ID);
1091 }
1092
1093 int bpf_link_get_next_id(__u32 start_id, __u32 *next_id)
1094 {
1095 return bpf_obj_get_next_id(start_id, next_id, BPF_LINK_GET_NEXT_ID);
1096 }
1097
1098 int bpf_prog_get_fd_by_id_opts(__u32 id,
1099 const struct bpf_get_fd_by_id_opts *opts)
1100 {
1101 const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1102 union bpf_attr attr;
1103 int fd;
1104
1105 if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1106 return libbpf_err(-EINVAL);
1107
1108 memset(&attr, 0, attr_sz);
1109 attr.prog_id = id;
1110 attr.open_flags = OPTS_GET(opts, open_flags, 0);
1111
1112 fd = sys_bpf_fd(BPF_PROG_GET_FD_BY_ID, &attr, attr_sz);
1113 return libbpf_err_errno(fd);
1114 }
1115
1116 int bpf_prog_get_fd_by_id(__u32 id)
1117 {
1118 return bpf_prog_get_fd_by_id_opts(id, NULL);
1119 }
1120
1121 int bpf_map_get_fd_by_id_opts(__u32 id,
1122 const struct bpf_get_fd_by_id_opts *opts)
1123 {
1124 const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1125 union bpf_attr attr;
1126 int fd;
1127
1128 if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1129 return libbpf_err(-EINVAL);
1130
1131 memset(&attr, 0, attr_sz);
1132 attr.map_id = id;
1133 attr.open_flags = OPTS_GET(opts, open_flags, 0);
1134
1135 fd = sys_bpf_fd(BPF_MAP_GET_FD_BY_ID, &attr, attr_sz);
1136 return libbpf_err_errno(fd);
1137 }
1138
1139 int bpf_map_get_fd_by_id(__u32 id)
1140 {
1141 return bpf_map_get_fd_by_id_opts(id, NULL);
1142 }
1143
1144 int bpf_btf_get_fd_by_id_opts(__u32 id,
1145 const struct bpf_get_fd_by_id_opts *opts)
1146 {
1147 const size_t attr_sz = offsetofend(union bpf_attr, fd_by_id_token_fd);
1148 union bpf_attr attr;
1149 int fd;
1150
1151 if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1152 return libbpf_err(-EINVAL);
1153
1154 memset(&attr, 0, attr_sz);
1155 attr.btf_id = id;
1156 attr.open_flags = OPTS_GET(opts, open_flags, 0);
1157 attr.fd_by_id_token_fd = OPTS_GET(opts, token_fd, 0);
1158
1159 fd = sys_bpf_fd(BPF_BTF_GET_FD_BY_ID, &attr, attr_sz);
1160 return libbpf_err_errno(fd);
1161 }
1162
1163 int bpf_btf_get_fd_by_id(__u32 id)
1164 {
1165 return bpf_btf_get_fd_by_id_opts(id, NULL);
1166 }
1167
1168 int bpf_link_get_fd_by_id_opts(__u32 id,
1169 const struct bpf_get_fd_by_id_opts *opts)
1170 {
1171 const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1172 union bpf_attr attr;
1173 int fd;
1174
1175 if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1176 return libbpf_err(-EINVAL);
1177
1178 memset(&attr, 0, attr_sz);
1179 attr.link_id = id;
1180 attr.open_flags = OPTS_GET(opts, open_flags, 0);
1181
1182 fd = sys_bpf_fd(BPF_LINK_GET_FD_BY_ID, &attr, attr_sz);
1183 return libbpf_err_errno(fd);
1184 }
1185
1186 int bpf_link_get_fd_by_id(__u32 id)
1187 {
1188 return bpf_link_get_fd_by_id_opts(id, NULL);
1189 }
1190
1191 int bpf_obj_get_info_by_fd(int bpf_fd, void *info, __u32 *info_len)
1192 {
1193 const size_t attr_sz = offsetofend(union bpf_attr, info);
1194 union bpf_attr attr;
1195 int err;
1196
1197 memset(&attr, 0, attr_sz);
1198 attr.info.bpf_fd = bpf_fd;
1199 attr.info.info_len = *info_len;
1200 attr.info.info = ptr_to_u64(info);
1201
1202 err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, attr_sz);
1203 if (!err)
1204 *info_len = attr.info.info_len;
1205 return libbpf_err_errno(err);
1206 }
1207
1208 int bpf_prog_get_info_by_fd(int prog_fd, struct bpf_prog_info *info, __u32 *info_len)
1209 {
1210 return bpf_obj_get_info_by_fd(prog_fd, info, info_len);
1211 }
1212
1213 int bpf_map_get_info_by_fd(int map_fd, struct bpf_map_info *info, __u32 *info_len)
1214 {
1215 return bpf_obj_get_info_by_fd(map_fd, info, info_len);
1216 }
1217
1218 int bpf_btf_get_info_by_fd(int btf_fd, struct bpf_btf_info *info, __u32 *info_len)
1219 {
1220 return bpf_obj_get_info_by_fd(btf_fd, info, info_len);
1221 }
1222
1223 int bpf_link_get_info_by_fd(int link_fd, struct bpf_link_info *info, __u32 *info_len)
1224 {
1225 return bpf_obj_get_info_by_fd(link_fd, info, info_len);
1226 }
1227
1228 int bpf_raw_tracepoint_open_opts(int prog_fd, struct bpf_raw_tp_opts *opts)
1229 {
1230 const size_t attr_sz = offsetofend(union bpf_attr, raw_tracepoint);
1231 union bpf_attr attr;
1232 int fd;
1233
1234 if (!OPTS_VALID(opts, bpf_raw_tp_opts))
1235 return libbpf_err(-EINVAL);
1236
1237 memset(&attr, 0, attr_sz);
1238 attr.raw_tracepoint.prog_fd = prog_fd;
1239 attr.raw_tracepoint.name = ptr_to_u64(OPTS_GET(opts, tp_name, NULL));
1240 attr.raw_tracepoint.cookie = OPTS_GET(opts, cookie, 0);
1241
1242 fd = sys_bpf_fd(BPF_RAW_TRACEPOINT_OPEN, &attr, attr_sz);
1243 return libbpf_err_errno(fd);
1244 }
1245
1246 int bpf_raw_tracepoint_open(const char *name, int prog_fd)
1247 {
1248 LIBBPF_OPTS(bpf_raw_tp_opts, opts, .tp_name = name);
1249
1250 return bpf_raw_tracepoint_open_opts(prog_fd, &opts);
1251 }
1252
1253 int bpf_btf_load(const void *btf_data, size_t btf_size, struct bpf_btf_load_opts *opts)
1254 {
1255 const size_t attr_sz = offsetofend(union bpf_attr, btf_token_fd);
1256 union bpf_attr attr;
1257 char *log_buf;
1258 size_t log_size;
1259 __u32 log_level;
1260 int fd;
1261
1262 bump_rlimit_memlock();
1263
1264 memset(&attr, 0, attr_sz);
1265
1266 if (!OPTS_VALID(opts, bpf_btf_load_opts))
1267 return libbpf_err(-EINVAL);
1268
1269 log_buf = OPTS_GET(opts, log_buf, NULL);
1270 log_size = OPTS_GET(opts, log_size, 0);
1271 log_level = OPTS_GET(opts, log_level, 0);
1272
1273 if (log_size > UINT_MAX)
1274 return libbpf_err(-EINVAL);
1275 if (log_size && !log_buf)
1276 return libbpf_err(-EINVAL);
1277
1278 attr.btf = ptr_to_u64(btf_data);
1279 attr.btf_size = btf_size;
1280
1281 attr.btf_flags = OPTS_GET(opts, btf_flags, 0);
1282 attr.btf_token_fd = OPTS_GET(opts, token_fd, 0);
1283
1284 /* log_level == 0 and log_buf != NULL means "try loading without
1285 * log_buf, but retry with log_buf and log_level=1 on error", which is
1286 * consistent across low-level and high-level BTF and program loading
1287 * APIs within libbpf and provides a sensible behavior in practice
1288 */
1289 if (log_level) {
1290 attr.btf_log_buf = ptr_to_u64(log_buf);
1291 attr.btf_log_size = (__u32)log_size;
1292 attr.btf_log_level = log_level;
1293 }
1294
1295 fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1296 if (fd < 0 && log_buf && log_level == 0) {
1297 attr.btf_log_buf = ptr_to_u64(log_buf);
1298 attr.btf_log_size = (__u32)log_size;
1299 attr.btf_log_level = 1;
1300 fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1301 }
1302
1303 OPTS_SET(opts, log_true_size, attr.btf_log_true_size);
1304 return libbpf_err_errno(fd);
1305 }
1306
1307 int bpf_task_fd_query(int pid, int fd, __u32 flags, char *buf, __u32 *buf_len,
1308 __u32 *prog_id, __u32 *fd_type, __u64 *probe_offset,
1309 __u64 *probe_addr)
1310 {
1311 const size_t attr_sz = offsetofend(union bpf_attr, task_fd_query);
1312 union bpf_attr attr;
1313 int err;
1314
1315 memset(&attr, 0, attr_sz);
1316 attr.task_fd_query.pid = pid;
1317 attr.task_fd_query.fd = fd;
1318 attr.task_fd_query.flags = flags;
1319 attr.task_fd_query.buf = ptr_to_u64(buf);
1320 attr.task_fd_query.buf_len = *buf_len;
1321
1322 err = sys_bpf(BPF_TASK_FD_QUERY, &attr, attr_sz);
1323
1324 *buf_len = attr.task_fd_query.buf_len;
1325 *prog_id = attr.task_fd_query.prog_id;
1326 *fd_type = attr.task_fd_query.fd_type;
1327 *probe_offset = attr.task_fd_query.probe_offset;
1328 *probe_addr = attr.task_fd_query.probe_addr;
1329
1330 return libbpf_err_errno(err);
1331 }
1332
1333 int bpf_enable_stats(enum bpf_stats_type type)
1334 {
1335 const size_t attr_sz = offsetofend(union bpf_attr, enable_stats);
1336 union bpf_attr attr;
1337 int fd;
1338
1339 memset(&attr, 0, attr_sz);
1340 attr.enable_stats.type = type;
1341
1342 fd = sys_bpf_fd(BPF_ENABLE_STATS, &attr, attr_sz);
1343 return libbpf_err_errno(fd);
1344 }
1345
1346 int bpf_prog_bind_map(int prog_fd, int map_fd,
1347 const struct bpf_prog_bind_opts *opts)
1348 {
1349 const size_t attr_sz = offsetofend(union bpf_attr, prog_bind_map);
1350 union bpf_attr attr;
1351 int ret;
1352
1353 if (!OPTS_VALID(opts, bpf_prog_bind_opts))
1354 return libbpf_err(-EINVAL);
1355
1356 memset(&attr, 0, attr_sz);
1357 attr.prog_bind_map.prog_fd = prog_fd;
1358 attr.prog_bind_map.map_fd = map_fd;
1359 attr.prog_bind_map.flags = OPTS_GET(opts, flags, 0);
1360
1361 ret = sys_bpf(BPF_PROG_BIND_MAP, &attr, attr_sz);
1362 return libbpf_err_errno(ret);
1363 }
1364
1365 int bpf_token_create(int bpffs_fd, struct bpf_token_create_opts *opts)
1366 {
1367 const size_t attr_sz = offsetofend(union bpf_attr, token_create);
1368 union bpf_attr attr;
1369 int fd;
1370
1371 if (!OPTS_VALID(opts, bpf_token_create_opts))
1372 return libbpf_err(-EINVAL);
1373
1374 memset(&attr, 0, attr_sz);
1375 attr.token_create.bpffs_fd = bpffs_fd;
1376 attr.token_create.flags = OPTS_GET(opts, flags, 0);
1377
1378 fd = sys_bpf_fd(BPF_TOKEN_CREATE, &attr, attr_sz);
1379 return libbpf_err_errno(fd);
1380 }
1381
1382 int bpf_prog_stream_read(int prog_fd, __u32 stream_id, void *buf, __u32 buf_len,
1383 struct bpf_prog_stream_read_opts *opts)
1384 {
1385 const size_t attr_sz = offsetofend(union bpf_attr, prog_stream_read);
1386 union bpf_attr attr;
1387 int err;
1388
1389 if (!OPTS_VALID(opts, bpf_prog_stream_read_opts))
1390 return libbpf_err(-EINVAL);
1391
1392 memset(&attr, 0, attr_sz);
1393 attr.prog_stream_read.stream_buf = ptr_to_u64(buf);
1394 attr.prog_stream_read.stream_buf_len = buf_len;
1395 attr.prog_stream_read.stream_id = stream_id;
1396 attr.prog_stream_read.prog_fd = prog_fd;
1397
1398 err = sys_bpf(BPF_PROG_STREAM_READ_BY_FD, &attr, attr_sz);
1399 return libbpf_err_errno(err);
1400 }
1401
1402 int bpf_prog_assoc_struct_ops(int prog_fd, int map_fd,
1403 struct bpf_prog_assoc_struct_ops_opts *opts)
1404 {
1405 const size_t attr_sz = offsetofend(union bpf_attr, prog_assoc_struct_ops);
1406 union bpf_attr attr;
1407 int err;
1408
1409 if (!OPTS_VALID(opts, bpf_prog_assoc_struct_ops_opts))
1410 return libbpf_err(-EINVAL);
1411
1412 memset(&attr, 0, attr_sz);
1413 attr.prog_assoc_struct_ops.map_fd = map_fd;
1414 attr.prog_assoc_struct_ops.prog_fd = prog_fd;
1415 attr.prog_assoc_struct_ops.flags = OPTS_GET(opts, flags, 0);
1416
1417 err = sys_bpf(BPF_PROG_ASSOC_STRUCT_OPS, &attr, attr_sz);
1418 return libbpf_err_errno(err);
1419 }