Loading...
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 <ast@kernel.org>
7 * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
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
67static inline __u64 ptr_to_u64(const void *ptr)
68{
69 return (__u64) (unsigned long) ptr;
70}
71
72static 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
78static 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
87int 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/20201201215900.3569844-1-guro@fb.com/
104 * [1] d05512618056 ("bpf: Add bpf_ktime_get_coarse_ns helper")
105 */
106int probe_memcg_account(void)
107{
108 const size_t attr_sz = offsetofend(union bpf_attr, attach_btf_obj_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
124 prog_fd = sys_bpf_fd(BPF_PROG_LOAD, &attr, attr_sz);
125 if (prog_fd >= 0) {
126 close(prog_fd);
127 return 1;
128 }
129 return 0;
130}
131
132static bool memlock_bumped;
133static rlim_t memlock_rlim = RLIM_INFINITY;
134
135int libbpf_set_memlock_rlim(size_t memlock_bytes)
136{
137 if (memlock_bumped)
138 return libbpf_err(-EBUSY);
139
140 memlock_rlim = memlock_bytes;
141 return 0;
142}
143
144int bump_rlimit_memlock(void)
145{
146 struct rlimit rlim;
147
148 /* if kernel supports memcg-based accounting, skip bumping RLIMIT_MEMLOCK */
149 if (memlock_bumped || kernel_supports(NULL, FEAT_MEMCG_ACCOUNT))
150 return 0;
151
152 memlock_bumped = true;
153
154 /* zero memlock_rlim_max disables auto-bumping RLIMIT_MEMLOCK */
155 if (memlock_rlim == 0)
156 return 0;
157
158 rlim.rlim_cur = rlim.rlim_max = memlock_rlim;
159 if (setrlimit(RLIMIT_MEMLOCK, &rlim))
160 return -errno;
161
162 return 0;
163}
164
165int bpf_map_create(enum bpf_map_type map_type,
166 const char *map_name,
167 __u32 key_size,
168 __u32 value_size,
169 __u32 max_entries,
170 const struct bpf_map_create_opts *opts)
171{
172 const size_t attr_sz = offsetofend(union bpf_attr, map_extra);
173 union bpf_attr attr;
174 int fd;
175
176 bump_rlimit_memlock();
177
178 memset(&attr, 0, attr_sz);
179
180 if (!OPTS_VALID(opts, bpf_map_create_opts))
181 return libbpf_err(-EINVAL);
182
183 attr.map_type = map_type;
184 if (map_name && kernel_supports(NULL, FEAT_PROG_NAME))
185 libbpf_strlcpy(attr.map_name, map_name, sizeof(attr.map_name));
186 attr.key_size = key_size;
187 attr.value_size = value_size;
188 attr.max_entries = max_entries;
189
190 attr.btf_fd = OPTS_GET(opts, btf_fd, 0);
191 attr.btf_key_type_id = OPTS_GET(opts, btf_key_type_id, 0);
192 attr.btf_value_type_id = OPTS_GET(opts, btf_value_type_id, 0);
193 attr.btf_vmlinux_value_type_id = OPTS_GET(opts, btf_vmlinux_value_type_id, 0);
194
195 attr.inner_map_fd = OPTS_GET(opts, inner_map_fd, 0);
196 attr.map_flags = OPTS_GET(opts, map_flags, 0);
197 attr.map_extra = OPTS_GET(opts, map_extra, 0);
198 attr.numa_node = OPTS_GET(opts, numa_node, 0);
199 attr.map_ifindex = OPTS_GET(opts, map_ifindex, 0);
200
201 fd = sys_bpf_fd(BPF_MAP_CREATE, &attr, attr_sz);
202 return libbpf_err_errno(fd);
203}
204
205static void *
206alloc_zero_tailing_info(const void *orecord, __u32 cnt,
207 __u32 actual_rec_size, __u32 expected_rec_size)
208{
209 __u64 info_len = (__u64)actual_rec_size * cnt;
210 void *info, *nrecord;
211 int i;
212
213 info = malloc(info_len);
214 if (!info)
215 return NULL;
216
217 /* zero out bytes kernel does not understand */
218 nrecord = info;
219 for (i = 0; i < cnt; i++) {
220 memcpy(nrecord, orecord, expected_rec_size);
221 memset(nrecord + expected_rec_size, 0,
222 actual_rec_size - expected_rec_size);
223 orecord += actual_rec_size;
224 nrecord += actual_rec_size;
225 }
226
227 return info;
228}
229
230int bpf_prog_load(enum bpf_prog_type prog_type,
231 const char *prog_name, const char *license,
232 const struct bpf_insn *insns, size_t insn_cnt,
233 struct bpf_prog_load_opts *opts)
234{
235 const size_t attr_sz = offsetofend(union bpf_attr, log_true_size);
236 void *finfo = NULL, *linfo = NULL;
237 const char *func_info, *line_info;
238 __u32 log_size, log_level, attach_prog_fd, attach_btf_obj_fd;
239 __u32 func_info_rec_size, line_info_rec_size;
240 int fd, attempts;
241 union bpf_attr attr;
242 char *log_buf;
243
244 bump_rlimit_memlock();
245
246 if (!OPTS_VALID(opts, bpf_prog_load_opts))
247 return libbpf_err(-EINVAL);
248
249 attempts = OPTS_GET(opts, attempts, 0);
250 if (attempts < 0)
251 return libbpf_err(-EINVAL);
252 if (attempts == 0)
253 attempts = PROG_LOAD_ATTEMPTS;
254
255 memset(&attr, 0, attr_sz);
256
257 attr.prog_type = prog_type;
258 attr.expected_attach_type = OPTS_GET(opts, expected_attach_type, 0);
259
260 attr.prog_btf_fd = OPTS_GET(opts, prog_btf_fd, 0);
261 attr.prog_flags = OPTS_GET(opts, prog_flags, 0);
262 attr.prog_ifindex = OPTS_GET(opts, prog_ifindex, 0);
263 attr.kern_version = OPTS_GET(opts, kern_version, 0);
264
265 if (prog_name && kernel_supports(NULL, FEAT_PROG_NAME))
266 libbpf_strlcpy(attr.prog_name, prog_name, sizeof(attr.prog_name));
267 attr.license = ptr_to_u64(license);
268
269 if (insn_cnt > UINT_MAX)
270 return libbpf_err(-E2BIG);
271
272 attr.insns = ptr_to_u64(insns);
273 attr.insn_cnt = (__u32)insn_cnt;
274
275 attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
276 attach_btf_obj_fd = OPTS_GET(opts, attach_btf_obj_fd, 0);
277
278 if (attach_prog_fd && attach_btf_obj_fd)
279 return libbpf_err(-EINVAL);
280
281 attr.attach_btf_id = OPTS_GET(opts, attach_btf_id, 0);
282 if (attach_prog_fd)
283 attr.attach_prog_fd = attach_prog_fd;
284 else
285 attr.attach_btf_obj_fd = attach_btf_obj_fd;
286
287 log_buf = OPTS_GET(opts, log_buf, NULL);
288 log_size = OPTS_GET(opts, log_size, 0);
289 log_level = OPTS_GET(opts, log_level, 0);
290
291 if (!!log_buf != !!log_size)
292 return libbpf_err(-EINVAL);
293
294 func_info_rec_size = OPTS_GET(opts, func_info_rec_size, 0);
295 func_info = OPTS_GET(opts, func_info, NULL);
296 attr.func_info_rec_size = func_info_rec_size;
297 attr.func_info = ptr_to_u64(func_info);
298 attr.func_info_cnt = OPTS_GET(opts, func_info_cnt, 0);
299
300 line_info_rec_size = OPTS_GET(opts, line_info_rec_size, 0);
301 line_info = OPTS_GET(opts, line_info, NULL);
302 attr.line_info_rec_size = line_info_rec_size;
303 attr.line_info = ptr_to_u64(line_info);
304 attr.line_info_cnt = OPTS_GET(opts, line_info_cnt, 0);
305
306 attr.fd_array = ptr_to_u64(OPTS_GET(opts, fd_array, NULL));
307
308 if (log_level) {
309 attr.log_buf = ptr_to_u64(log_buf);
310 attr.log_size = log_size;
311 attr.log_level = log_level;
312 }
313
314 fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
315 OPTS_SET(opts, log_true_size, attr.log_true_size);
316 if (fd >= 0)
317 return fd;
318
319 /* After bpf_prog_load, the kernel may modify certain attributes
320 * to give user space a hint how to deal with loading failure.
321 * Check to see whether we can make some changes and load again.
322 */
323 while (errno == E2BIG && (!finfo || !linfo)) {
324 if (!finfo && attr.func_info_cnt &&
325 attr.func_info_rec_size < func_info_rec_size) {
326 /* try with corrected func info records */
327 finfo = alloc_zero_tailing_info(func_info,
328 attr.func_info_cnt,
329 func_info_rec_size,
330 attr.func_info_rec_size);
331 if (!finfo) {
332 errno = E2BIG;
333 goto done;
334 }
335
336 attr.func_info = ptr_to_u64(finfo);
337 attr.func_info_rec_size = func_info_rec_size;
338 } else if (!linfo && attr.line_info_cnt &&
339 attr.line_info_rec_size < line_info_rec_size) {
340 linfo = alloc_zero_tailing_info(line_info,
341 attr.line_info_cnt,
342 line_info_rec_size,
343 attr.line_info_rec_size);
344 if (!linfo) {
345 errno = E2BIG;
346 goto done;
347 }
348
349 attr.line_info = ptr_to_u64(linfo);
350 attr.line_info_rec_size = line_info_rec_size;
351 } else {
352 break;
353 }
354
355 fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
356 OPTS_SET(opts, log_true_size, attr.log_true_size);
357 if (fd >= 0)
358 goto done;
359 }
360
361 if (log_level == 0 && log_buf) {
362 /* log_level == 0 with non-NULL log_buf requires retrying on error
363 * with log_level == 1 and log_buf/log_buf_size set, to get details of
364 * failure
365 */
366 attr.log_buf = ptr_to_u64(log_buf);
367 attr.log_size = log_size;
368 attr.log_level = 1;
369
370 fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
371 OPTS_SET(opts, log_true_size, attr.log_true_size);
372 }
373done:
374 /* free() doesn't affect errno, so we don't need to restore it */
375 free(finfo);
376 free(linfo);
377 return libbpf_err_errno(fd);
378}
379
380int bpf_map_update_elem(int fd, const void *key, const void *value,
381 __u64 flags)
382{
383 const size_t attr_sz = offsetofend(union bpf_attr, flags);
384 union bpf_attr attr;
385 int ret;
386
387 memset(&attr, 0, attr_sz);
388 attr.map_fd = fd;
389 attr.key = ptr_to_u64(key);
390 attr.value = ptr_to_u64(value);
391 attr.flags = flags;
392
393 ret = sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, attr_sz);
394 return libbpf_err_errno(ret);
395}
396
397int bpf_map_lookup_elem(int fd, const void *key, void *value)
398{
399 const size_t attr_sz = offsetofend(union bpf_attr, flags);
400 union bpf_attr attr;
401 int ret;
402
403 memset(&attr, 0, attr_sz);
404 attr.map_fd = fd;
405 attr.key = ptr_to_u64(key);
406 attr.value = ptr_to_u64(value);
407
408 ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, attr_sz);
409 return libbpf_err_errno(ret);
410}
411
412int bpf_map_lookup_elem_flags(int fd, const void *key, void *value, __u64 flags)
413{
414 const size_t attr_sz = offsetofend(union bpf_attr, flags);
415 union bpf_attr attr;
416 int ret;
417
418 memset(&attr, 0, attr_sz);
419 attr.map_fd = fd;
420 attr.key = ptr_to_u64(key);
421 attr.value = ptr_to_u64(value);
422 attr.flags = flags;
423
424 ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, attr_sz);
425 return libbpf_err_errno(ret);
426}
427
428int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value)
429{
430 const size_t attr_sz = offsetofend(union bpf_attr, flags);
431 union bpf_attr attr;
432 int ret;
433
434 memset(&attr, 0, attr_sz);
435 attr.map_fd = fd;
436 attr.key = ptr_to_u64(key);
437 attr.value = ptr_to_u64(value);
438
439 ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, attr_sz);
440 return libbpf_err_errno(ret);
441}
442
443int bpf_map_lookup_and_delete_elem_flags(int fd, const void *key, void *value, __u64 flags)
444{
445 const size_t attr_sz = offsetofend(union bpf_attr, flags);
446 union bpf_attr attr;
447 int ret;
448
449 memset(&attr, 0, attr_sz);
450 attr.map_fd = fd;
451 attr.key = ptr_to_u64(key);
452 attr.value = ptr_to_u64(value);
453 attr.flags = flags;
454
455 ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, attr_sz);
456 return libbpf_err_errno(ret);
457}
458
459int bpf_map_delete_elem(int fd, const void *key)
460{
461 const size_t attr_sz = offsetofend(union bpf_attr, flags);
462 union bpf_attr attr;
463 int ret;
464
465 memset(&attr, 0, attr_sz);
466 attr.map_fd = fd;
467 attr.key = ptr_to_u64(key);
468
469 ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, attr_sz);
470 return libbpf_err_errno(ret);
471}
472
473int bpf_map_delete_elem_flags(int fd, const void *key, __u64 flags)
474{
475 const size_t attr_sz = offsetofend(union bpf_attr, flags);
476 union bpf_attr attr;
477 int ret;
478
479 memset(&attr, 0, attr_sz);
480 attr.map_fd = fd;
481 attr.key = ptr_to_u64(key);
482 attr.flags = flags;
483
484 ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, attr_sz);
485 return libbpf_err_errno(ret);
486}
487
488int bpf_map_get_next_key(int fd, const void *key, void *next_key)
489{
490 const size_t attr_sz = offsetofend(union bpf_attr, next_key);
491 union bpf_attr attr;
492 int ret;
493
494 memset(&attr, 0, attr_sz);
495 attr.map_fd = fd;
496 attr.key = ptr_to_u64(key);
497 attr.next_key = ptr_to_u64(next_key);
498
499 ret = sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, attr_sz);
500 return libbpf_err_errno(ret);
501}
502
503int bpf_map_freeze(int fd)
504{
505 const size_t attr_sz = offsetofend(union bpf_attr, map_fd);
506 union bpf_attr attr;
507 int ret;
508
509 memset(&attr, 0, attr_sz);
510 attr.map_fd = fd;
511
512 ret = sys_bpf(BPF_MAP_FREEZE, &attr, attr_sz);
513 return libbpf_err_errno(ret);
514}
515
516static int bpf_map_batch_common(int cmd, int fd, void *in_batch,
517 void *out_batch, void *keys, void *values,
518 __u32 *count,
519 const struct bpf_map_batch_opts *opts)
520{
521 const size_t attr_sz = offsetofend(union bpf_attr, batch);
522 union bpf_attr attr;
523 int ret;
524
525 if (!OPTS_VALID(opts, bpf_map_batch_opts))
526 return libbpf_err(-EINVAL);
527
528 memset(&attr, 0, attr_sz);
529 attr.batch.map_fd = fd;
530 attr.batch.in_batch = ptr_to_u64(in_batch);
531 attr.batch.out_batch = ptr_to_u64(out_batch);
532 attr.batch.keys = ptr_to_u64(keys);
533 attr.batch.values = ptr_to_u64(values);
534 attr.batch.count = *count;
535 attr.batch.elem_flags = OPTS_GET(opts, elem_flags, 0);
536 attr.batch.flags = OPTS_GET(opts, flags, 0);
537
538 ret = sys_bpf(cmd, &attr, attr_sz);
539 *count = attr.batch.count;
540
541 return libbpf_err_errno(ret);
542}
543
544int bpf_map_delete_batch(int fd, const void *keys, __u32 *count,
545 const struct bpf_map_batch_opts *opts)
546{
547 return bpf_map_batch_common(BPF_MAP_DELETE_BATCH, fd, NULL,
548 NULL, (void *)keys, NULL, count, opts);
549}
550
551int bpf_map_lookup_batch(int fd, void *in_batch, void *out_batch, void *keys,
552 void *values, __u32 *count,
553 const struct bpf_map_batch_opts *opts)
554{
555 return bpf_map_batch_common(BPF_MAP_LOOKUP_BATCH, fd, in_batch,
556 out_batch, keys, values, count, opts);
557}
558
559int bpf_map_lookup_and_delete_batch(int fd, void *in_batch, void *out_batch,
560 void *keys, void *values, __u32 *count,
561 const struct bpf_map_batch_opts *opts)
562{
563 return bpf_map_batch_common(BPF_MAP_LOOKUP_AND_DELETE_BATCH,
564 fd, in_batch, out_batch, keys, values,
565 count, opts);
566}
567
568int bpf_map_update_batch(int fd, const void *keys, const void *values, __u32 *count,
569 const struct bpf_map_batch_opts *opts)
570{
571 return bpf_map_batch_common(BPF_MAP_UPDATE_BATCH, fd, NULL, NULL,
572 (void *)keys, (void *)values, count, opts);
573}
574
575int bpf_obj_pin_opts(int fd, const char *pathname, const struct bpf_obj_pin_opts *opts)
576{
577 const size_t attr_sz = offsetofend(union bpf_attr, path_fd);
578 union bpf_attr attr;
579 int ret;
580
581 if (!OPTS_VALID(opts, bpf_obj_pin_opts))
582 return libbpf_err(-EINVAL);
583
584 memset(&attr, 0, attr_sz);
585 attr.path_fd = OPTS_GET(opts, path_fd, 0);
586 attr.pathname = ptr_to_u64((void *)pathname);
587 attr.file_flags = OPTS_GET(opts, file_flags, 0);
588 attr.bpf_fd = fd;
589
590 ret = sys_bpf(BPF_OBJ_PIN, &attr, attr_sz);
591 return libbpf_err_errno(ret);
592}
593
594int bpf_obj_pin(int fd, const char *pathname)
595{
596 return bpf_obj_pin_opts(fd, pathname, NULL);
597}
598
599int bpf_obj_get(const char *pathname)
600{
601 return bpf_obj_get_opts(pathname, NULL);
602}
603
604int bpf_obj_get_opts(const char *pathname, const struct bpf_obj_get_opts *opts)
605{
606 const size_t attr_sz = offsetofend(union bpf_attr, path_fd);
607 union bpf_attr attr;
608 int fd;
609
610 if (!OPTS_VALID(opts, bpf_obj_get_opts))
611 return libbpf_err(-EINVAL);
612
613 memset(&attr, 0, attr_sz);
614 attr.path_fd = OPTS_GET(opts, path_fd, 0);
615 attr.pathname = ptr_to_u64((void *)pathname);
616 attr.file_flags = OPTS_GET(opts, file_flags, 0);
617
618 fd = sys_bpf_fd(BPF_OBJ_GET, &attr, attr_sz);
619 return libbpf_err_errno(fd);
620}
621
622int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type,
623 unsigned int flags)
624{
625 DECLARE_LIBBPF_OPTS(bpf_prog_attach_opts, opts,
626 .flags = flags,
627 );
628
629 return bpf_prog_attach_opts(prog_fd, target_fd, type, &opts);
630}
631
632int bpf_prog_attach_opts(int prog_fd, int target, enum bpf_attach_type type,
633 const struct bpf_prog_attach_opts *opts)
634{
635 const size_t attr_sz = offsetofend(union bpf_attr, expected_revision);
636 __u32 relative_id, flags;
637 int ret, relative_fd;
638 union bpf_attr attr;
639
640 if (!OPTS_VALID(opts, bpf_prog_attach_opts))
641 return libbpf_err(-EINVAL);
642
643 relative_id = OPTS_GET(opts, relative_id, 0);
644 relative_fd = OPTS_GET(opts, relative_fd, 0);
645 flags = OPTS_GET(opts, flags, 0);
646
647 /* validate we don't have unexpected combinations of non-zero fields */
648 if (relative_fd && relative_id)
649 return libbpf_err(-EINVAL);
650
651 memset(&attr, 0, attr_sz);
652 attr.target_fd = target;
653 attr.attach_bpf_fd = prog_fd;
654 attr.attach_type = type;
655 attr.replace_bpf_fd = OPTS_GET(opts, replace_fd, 0);
656 attr.expected_revision = OPTS_GET(opts, expected_revision, 0);
657
658 if (relative_id) {
659 attr.attach_flags = flags | BPF_F_ID;
660 attr.relative_id = relative_id;
661 } else {
662 attr.attach_flags = flags;
663 attr.relative_fd = relative_fd;
664 }
665
666 ret = sys_bpf(BPF_PROG_ATTACH, &attr, attr_sz);
667 return libbpf_err_errno(ret);
668}
669
670int bpf_prog_detach_opts(int prog_fd, int target, enum bpf_attach_type type,
671 const struct bpf_prog_detach_opts *opts)
672{
673 const size_t attr_sz = offsetofend(union bpf_attr, expected_revision);
674 __u32 relative_id, flags;
675 int ret, relative_fd;
676 union bpf_attr attr;
677
678 if (!OPTS_VALID(opts, bpf_prog_detach_opts))
679 return libbpf_err(-EINVAL);
680
681 relative_id = OPTS_GET(opts, relative_id, 0);
682 relative_fd = OPTS_GET(opts, relative_fd, 0);
683 flags = OPTS_GET(opts, flags, 0);
684
685 /* validate we don't have unexpected combinations of non-zero fields */
686 if (relative_fd && relative_id)
687 return libbpf_err(-EINVAL);
688
689 memset(&attr, 0, attr_sz);
690 attr.target_fd = target;
691 attr.attach_bpf_fd = prog_fd;
692 attr.attach_type = type;
693 attr.expected_revision = OPTS_GET(opts, expected_revision, 0);
694
695 if (relative_id) {
696 attr.attach_flags = flags | BPF_F_ID;
697 attr.relative_id = relative_id;
698 } else {
699 attr.attach_flags = flags;
700 attr.relative_fd = relative_fd;
701 }
702
703 ret = sys_bpf(BPF_PROG_DETACH, &attr, attr_sz);
704 return libbpf_err_errno(ret);
705}
706
707int bpf_prog_detach(int target_fd, enum bpf_attach_type type)
708{
709 return bpf_prog_detach_opts(0, target_fd, type, NULL);
710}
711
712int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type)
713{
714 return bpf_prog_detach_opts(prog_fd, target_fd, type, NULL);
715}
716
717int bpf_link_create(int prog_fd, int target_fd,
718 enum bpf_attach_type attach_type,
719 const struct bpf_link_create_opts *opts)
720{
721 const size_t attr_sz = offsetofend(union bpf_attr, link_create);
722 __u32 target_btf_id, iter_info_len, relative_id;
723 int fd, err, relative_fd;
724 union bpf_attr attr;
725
726 if (!OPTS_VALID(opts, bpf_link_create_opts))
727 return libbpf_err(-EINVAL);
728
729 iter_info_len = OPTS_GET(opts, iter_info_len, 0);
730 target_btf_id = OPTS_GET(opts, target_btf_id, 0);
731
732 /* validate we don't have unexpected combinations of non-zero fields */
733 if (iter_info_len || target_btf_id) {
734 if (iter_info_len && target_btf_id)
735 return libbpf_err(-EINVAL);
736 if (!OPTS_ZEROED(opts, target_btf_id))
737 return libbpf_err(-EINVAL);
738 }
739
740 memset(&attr, 0, attr_sz);
741 attr.link_create.prog_fd = prog_fd;
742 attr.link_create.target_fd = target_fd;
743 attr.link_create.attach_type = attach_type;
744 attr.link_create.flags = OPTS_GET(opts, flags, 0);
745
746 if (target_btf_id) {
747 attr.link_create.target_btf_id = target_btf_id;
748 goto proceed;
749 }
750
751 switch (attach_type) {
752 case BPF_TRACE_ITER:
753 attr.link_create.iter_info = ptr_to_u64(OPTS_GET(opts, iter_info, (void *)0));
754 attr.link_create.iter_info_len = iter_info_len;
755 break;
756 case BPF_PERF_EVENT:
757 attr.link_create.perf_event.bpf_cookie = OPTS_GET(opts, perf_event.bpf_cookie, 0);
758 if (!OPTS_ZEROED(opts, perf_event))
759 return libbpf_err(-EINVAL);
760 break;
761 case BPF_TRACE_KPROBE_MULTI:
762 attr.link_create.kprobe_multi.flags = OPTS_GET(opts, kprobe_multi.flags, 0);
763 attr.link_create.kprobe_multi.cnt = OPTS_GET(opts, kprobe_multi.cnt, 0);
764 attr.link_create.kprobe_multi.syms = ptr_to_u64(OPTS_GET(opts, kprobe_multi.syms, 0));
765 attr.link_create.kprobe_multi.addrs = ptr_to_u64(OPTS_GET(opts, kprobe_multi.addrs, 0));
766 attr.link_create.kprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, kprobe_multi.cookies, 0));
767 if (!OPTS_ZEROED(opts, kprobe_multi))
768 return libbpf_err(-EINVAL);
769 break;
770 case BPF_TRACE_UPROBE_MULTI:
771 attr.link_create.uprobe_multi.flags = OPTS_GET(opts, uprobe_multi.flags, 0);
772 attr.link_create.uprobe_multi.cnt = OPTS_GET(opts, uprobe_multi.cnt, 0);
773 attr.link_create.uprobe_multi.path = ptr_to_u64(OPTS_GET(opts, uprobe_multi.path, 0));
774 attr.link_create.uprobe_multi.offsets = ptr_to_u64(OPTS_GET(opts, uprobe_multi.offsets, 0));
775 attr.link_create.uprobe_multi.ref_ctr_offsets = ptr_to_u64(OPTS_GET(opts, uprobe_multi.ref_ctr_offsets, 0));
776 attr.link_create.uprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, uprobe_multi.cookies, 0));
777 attr.link_create.uprobe_multi.pid = OPTS_GET(opts, uprobe_multi.pid, 0);
778 if (!OPTS_ZEROED(opts, uprobe_multi))
779 return libbpf_err(-EINVAL);
780 break;
781 case BPF_TRACE_FENTRY:
782 case BPF_TRACE_FEXIT:
783 case BPF_MODIFY_RETURN:
784 case BPF_LSM_MAC:
785 attr.link_create.tracing.cookie = OPTS_GET(opts, tracing.cookie, 0);
786 if (!OPTS_ZEROED(opts, tracing))
787 return libbpf_err(-EINVAL);
788 break;
789 case BPF_NETFILTER:
790 attr.link_create.netfilter.pf = OPTS_GET(opts, netfilter.pf, 0);
791 attr.link_create.netfilter.hooknum = OPTS_GET(opts, netfilter.hooknum, 0);
792 attr.link_create.netfilter.priority = OPTS_GET(opts, netfilter.priority, 0);
793 attr.link_create.netfilter.flags = OPTS_GET(opts, netfilter.flags, 0);
794 if (!OPTS_ZEROED(opts, netfilter))
795 return libbpf_err(-EINVAL);
796 break;
797 case BPF_TCX_INGRESS:
798 case BPF_TCX_EGRESS:
799 relative_fd = OPTS_GET(opts, tcx.relative_fd, 0);
800 relative_id = OPTS_GET(opts, tcx.relative_id, 0);
801 if (relative_fd && relative_id)
802 return libbpf_err(-EINVAL);
803 if (relative_id) {
804 attr.link_create.tcx.relative_id = relative_id;
805 attr.link_create.flags |= BPF_F_ID;
806 } else {
807 attr.link_create.tcx.relative_fd = relative_fd;
808 }
809 attr.link_create.tcx.expected_revision = OPTS_GET(opts, tcx.expected_revision, 0);
810 if (!OPTS_ZEROED(opts, tcx))
811 return libbpf_err(-EINVAL);
812 break;
813 case BPF_NETKIT_PRIMARY:
814 case BPF_NETKIT_PEER:
815 relative_fd = OPTS_GET(opts, netkit.relative_fd, 0);
816 relative_id = OPTS_GET(opts, netkit.relative_id, 0);
817 if (relative_fd && relative_id)
818 return libbpf_err(-EINVAL);
819 if (relative_id) {
820 attr.link_create.netkit.relative_id = relative_id;
821 attr.link_create.flags |= BPF_F_ID;
822 } else {
823 attr.link_create.netkit.relative_fd = relative_fd;
824 }
825 attr.link_create.netkit.expected_revision = OPTS_GET(opts, netkit.expected_revision, 0);
826 if (!OPTS_ZEROED(opts, netkit))
827 return libbpf_err(-EINVAL);
828 break;
829 default:
830 if (!OPTS_ZEROED(opts, flags))
831 return libbpf_err(-EINVAL);
832 break;
833 }
834proceed:
835 fd = sys_bpf_fd(BPF_LINK_CREATE, &attr, attr_sz);
836 if (fd >= 0)
837 return fd;
838 /* we'll get EINVAL if LINK_CREATE doesn't support attaching fentry
839 * and other similar programs
840 */
841 err = -errno;
842 if (err != -EINVAL)
843 return libbpf_err(err);
844
845 /* if user used features not supported by
846 * BPF_RAW_TRACEPOINT_OPEN command, then just give up immediately
847 */
848 if (attr.link_create.target_fd || attr.link_create.target_btf_id)
849 return libbpf_err(err);
850 if (!OPTS_ZEROED(opts, sz))
851 return libbpf_err(err);
852
853 /* otherwise, for few select kinds of programs that can be
854 * attached using BPF_RAW_TRACEPOINT_OPEN command, try that as
855 * a fallback for older kernels
856 */
857 switch (attach_type) {
858 case BPF_TRACE_RAW_TP:
859 case BPF_LSM_MAC:
860 case BPF_TRACE_FENTRY:
861 case BPF_TRACE_FEXIT:
862 case BPF_MODIFY_RETURN:
863 return bpf_raw_tracepoint_open(NULL, prog_fd);
864 default:
865 return libbpf_err(err);
866 }
867}
868
869int bpf_link_detach(int link_fd)
870{
871 const size_t attr_sz = offsetofend(union bpf_attr, link_detach);
872 union bpf_attr attr;
873 int ret;
874
875 memset(&attr, 0, attr_sz);
876 attr.link_detach.link_fd = link_fd;
877
878 ret = sys_bpf(BPF_LINK_DETACH, &attr, attr_sz);
879 return libbpf_err_errno(ret);
880}
881
882int bpf_link_update(int link_fd, int new_prog_fd,
883 const struct bpf_link_update_opts *opts)
884{
885 const size_t attr_sz = offsetofend(union bpf_attr, link_update);
886 union bpf_attr attr;
887 int ret;
888
889 if (!OPTS_VALID(opts, bpf_link_update_opts))
890 return libbpf_err(-EINVAL);
891
892 if (OPTS_GET(opts, old_prog_fd, 0) && OPTS_GET(opts, old_map_fd, 0))
893 return libbpf_err(-EINVAL);
894
895 memset(&attr, 0, attr_sz);
896 attr.link_update.link_fd = link_fd;
897 attr.link_update.new_prog_fd = new_prog_fd;
898 attr.link_update.flags = OPTS_GET(opts, flags, 0);
899 if (OPTS_GET(opts, old_prog_fd, 0))
900 attr.link_update.old_prog_fd = OPTS_GET(opts, old_prog_fd, 0);
901 else if (OPTS_GET(opts, old_map_fd, 0))
902 attr.link_update.old_map_fd = OPTS_GET(opts, old_map_fd, 0);
903
904 ret = sys_bpf(BPF_LINK_UPDATE, &attr, attr_sz);
905 return libbpf_err_errno(ret);
906}
907
908int bpf_iter_create(int link_fd)
909{
910 const size_t attr_sz = offsetofend(union bpf_attr, iter_create);
911 union bpf_attr attr;
912 int fd;
913
914 memset(&attr, 0, attr_sz);
915 attr.iter_create.link_fd = link_fd;
916
917 fd = sys_bpf_fd(BPF_ITER_CREATE, &attr, attr_sz);
918 return libbpf_err_errno(fd);
919}
920
921int bpf_prog_query_opts(int target, enum bpf_attach_type type,
922 struct bpf_prog_query_opts *opts)
923{
924 const size_t attr_sz = offsetofend(union bpf_attr, query);
925 union bpf_attr attr;
926 int ret;
927
928 if (!OPTS_VALID(opts, bpf_prog_query_opts))
929 return libbpf_err(-EINVAL);
930
931 memset(&attr, 0, attr_sz);
932 attr.query.target_fd = target;
933 attr.query.attach_type = type;
934 attr.query.query_flags = OPTS_GET(opts, query_flags, 0);
935 attr.query.count = OPTS_GET(opts, count, 0);
936 attr.query.prog_ids = ptr_to_u64(OPTS_GET(opts, prog_ids, NULL));
937 attr.query.link_ids = ptr_to_u64(OPTS_GET(opts, link_ids, NULL));
938 attr.query.prog_attach_flags = ptr_to_u64(OPTS_GET(opts, prog_attach_flags, NULL));
939 attr.query.link_attach_flags = ptr_to_u64(OPTS_GET(opts, link_attach_flags, NULL));
940
941 ret = sys_bpf(BPF_PROG_QUERY, &attr, attr_sz);
942
943 OPTS_SET(opts, attach_flags, attr.query.attach_flags);
944 OPTS_SET(opts, revision, attr.query.revision);
945 OPTS_SET(opts, count, attr.query.count);
946
947 return libbpf_err_errno(ret);
948}
949
950int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags,
951 __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt)
952{
953 LIBBPF_OPTS(bpf_prog_query_opts, opts);
954 int ret;
955
956 opts.query_flags = query_flags;
957 opts.prog_ids = prog_ids;
958 opts.prog_cnt = *prog_cnt;
959
960 ret = bpf_prog_query_opts(target_fd, type, &opts);
961
962 if (attach_flags)
963 *attach_flags = opts.attach_flags;
964 *prog_cnt = opts.prog_cnt;
965
966 return libbpf_err_errno(ret);
967}
968
969int bpf_prog_test_run_opts(int prog_fd, struct bpf_test_run_opts *opts)
970{
971 const size_t attr_sz = offsetofend(union bpf_attr, test);
972 union bpf_attr attr;
973 int ret;
974
975 if (!OPTS_VALID(opts, bpf_test_run_opts))
976 return libbpf_err(-EINVAL);
977
978 memset(&attr, 0, attr_sz);
979 attr.test.prog_fd = prog_fd;
980 attr.test.batch_size = OPTS_GET(opts, batch_size, 0);
981 attr.test.cpu = OPTS_GET(opts, cpu, 0);
982 attr.test.flags = OPTS_GET(opts, flags, 0);
983 attr.test.repeat = OPTS_GET(opts, repeat, 0);
984 attr.test.duration = OPTS_GET(opts, duration, 0);
985 attr.test.ctx_size_in = OPTS_GET(opts, ctx_size_in, 0);
986 attr.test.ctx_size_out = OPTS_GET(opts, ctx_size_out, 0);
987 attr.test.data_size_in = OPTS_GET(opts, data_size_in, 0);
988 attr.test.data_size_out = OPTS_GET(opts, data_size_out, 0);
989 attr.test.ctx_in = ptr_to_u64(OPTS_GET(opts, ctx_in, NULL));
990 attr.test.ctx_out = ptr_to_u64(OPTS_GET(opts, ctx_out, NULL));
991 attr.test.data_in = ptr_to_u64(OPTS_GET(opts, data_in, NULL));
992 attr.test.data_out = ptr_to_u64(OPTS_GET(opts, data_out, NULL));
993
994 ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, attr_sz);
995
996 OPTS_SET(opts, data_size_out, attr.test.data_size_out);
997 OPTS_SET(opts, ctx_size_out, attr.test.ctx_size_out);
998 OPTS_SET(opts, duration, attr.test.duration);
999 OPTS_SET(opts, retval, attr.test.retval);
1000
1001 return libbpf_err_errno(ret);
1002}
1003
1004static int bpf_obj_get_next_id(__u32 start_id, __u32 *next_id, int cmd)
1005{
1006 const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1007 union bpf_attr attr;
1008 int err;
1009
1010 memset(&attr, 0, attr_sz);
1011 attr.start_id = start_id;
1012
1013 err = sys_bpf(cmd, &attr, attr_sz);
1014 if (!err)
1015 *next_id = attr.next_id;
1016
1017 return libbpf_err_errno(err);
1018}
1019
1020int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id)
1021{
1022 return bpf_obj_get_next_id(start_id, next_id, BPF_PROG_GET_NEXT_ID);
1023}
1024
1025int bpf_map_get_next_id(__u32 start_id, __u32 *next_id)
1026{
1027 return bpf_obj_get_next_id(start_id, next_id, BPF_MAP_GET_NEXT_ID);
1028}
1029
1030int bpf_btf_get_next_id(__u32 start_id, __u32 *next_id)
1031{
1032 return bpf_obj_get_next_id(start_id, next_id, BPF_BTF_GET_NEXT_ID);
1033}
1034
1035int bpf_link_get_next_id(__u32 start_id, __u32 *next_id)
1036{
1037 return bpf_obj_get_next_id(start_id, next_id, BPF_LINK_GET_NEXT_ID);
1038}
1039
1040int bpf_prog_get_fd_by_id_opts(__u32 id,
1041 const struct bpf_get_fd_by_id_opts *opts)
1042{
1043 const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1044 union bpf_attr attr;
1045 int fd;
1046
1047 if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1048 return libbpf_err(-EINVAL);
1049
1050 memset(&attr, 0, attr_sz);
1051 attr.prog_id = id;
1052 attr.open_flags = OPTS_GET(opts, open_flags, 0);
1053
1054 fd = sys_bpf_fd(BPF_PROG_GET_FD_BY_ID, &attr, attr_sz);
1055 return libbpf_err_errno(fd);
1056}
1057
1058int bpf_prog_get_fd_by_id(__u32 id)
1059{
1060 return bpf_prog_get_fd_by_id_opts(id, NULL);
1061}
1062
1063int bpf_map_get_fd_by_id_opts(__u32 id,
1064 const struct bpf_get_fd_by_id_opts *opts)
1065{
1066 const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1067 union bpf_attr attr;
1068 int fd;
1069
1070 if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1071 return libbpf_err(-EINVAL);
1072
1073 memset(&attr, 0, attr_sz);
1074 attr.map_id = id;
1075 attr.open_flags = OPTS_GET(opts, open_flags, 0);
1076
1077 fd = sys_bpf_fd(BPF_MAP_GET_FD_BY_ID, &attr, attr_sz);
1078 return libbpf_err_errno(fd);
1079}
1080
1081int bpf_map_get_fd_by_id(__u32 id)
1082{
1083 return bpf_map_get_fd_by_id_opts(id, NULL);
1084}
1085
1086int bpf_btf_get_fd_by_id_opts(__u32 id,
1087 const struct bpf_get_fd_by_id_opts *opts)
1088{
1089 const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1090 union bpf_attr attr;
1091 int fd;
1092
1093 if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1094 return libbpf_err(-EINVAL);
1095
1096 memset(&attr, 0, attr_sz);
1097 attr.btf_id = id;
1098 attr.open_flags = OPTS_GET(opts, open_flags, 0);
1099
1100 fd = sys_bpf_fd(BPF_BTF_GET_FD_BY_ID, &attr, attr_sz);
1101 return libbpf_err_errno(fd);
1102}
1103
1104int bpf_btf_get_fd_by_id(__u32 id)
1105{
1106 return bpf_btf_get_fd_by_id_opts(id, NULL);
1107}
1108
1109int bpf_link_get_fd_by_id_opts(__u32 id,
1110 const struct bpf_get_fd_by_id_opts *opts)
1111{
1112 const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1113 union bpf_attr attr;
1114 int fd;
1115
1116 if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1117 return libbpf_err(-EINVAL);
1118
1119 memset(&attr, 0, attr_sz);
1120 attr.link_id = id;
1121 attr.open_flags = OPTS_GET(opts, open_flags, 0);
1122
1123 fd = sys_bpf_fd(BPF_LINK_GET_FD_BY_ID, &attr, attr_sz);
1124 return libbpf_err_errno(fd);
1125}
1126
1127int bpf_link_get_fd_by_id(__u32 id)
1128{
1129 return bpf_link_get_fd_by_id_opts(id, NULL);
1130}
1131
1132int bpf_obj_get_info_by_fd(int bpf_fd, void *info, __u32 *info_len)
1133{
1134 const size_t attr_sz = offsetofend(union bpf_attr, info);
1135 union bpf_attr attr;
1136 int err;
1137
1138 memset(&attr, 0, attr_sz);
1139 attr.info.bpf_fd = bpf_fd;
1140 attr.info.info_len = *info_len;
1141 attr.info.info = ptr_to_u64(info);
1142
1143 err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, attr_sz);
1144 if (!err)
1145 *info_len = attr.info.info_len;
1146 return libbpf_err_errno(err);
1147}
1148
1149int bpf_prog_get_info_by_fd(int prog_fd, struct bpf_prog_info *info, __u32 *info_len)
1150{
1151 return bpf_obj_get_info_by_fd(prog_fd, info, info_len);
1152}
1153
1154int bpf_map_get_info_by_fd(int map_fd, struct bpf_map_info *info, __u32 *info_len)
1155{
1156 return bpf_obj_get_info_by_fd(map_fd, info, info_len);
1157}
1158
1159int bpf_btf_get_info_by_fd(int btf_fd, struct bpf_btf_info *info, __u32 *info_len)
1160{
1161 return bpf_obj_get_info_by_fd(btf_fd, info, info_len);
1162}
1163
1164int bpf_link_get_info_by_fd(int link_fd, struct bpf_link_info *info, __u32 *info_len)
1165{
1166 return bpf_obj_get_info_by_fd(link_fd, info, info_len);
1167}
1168
1169int bpf_raw_tracepoint_open(const char *name, int prog_fd)
1170{
1171 const size_t attr_sz = offsetofend(union bpf_attr, raw_tracepoint);
1172 union bpf_attr attr;
1173 int fd;
1174
1175 memset(&attr, 0, attr_sz);
1176 attr.raw_tracepoint.name = ptr_to_u64(name);
1177 attr.raw_tracepoint.prog_fd = prog_fd;
1178
1179 fd = sys_bpf_fd(BPF_RAW_TRACEPOINT_OPEN, &attr, attr_sz);
1180 return libbpf_err_errno(fd);
1181}
1182
1183int bpf_btf_load(const void *btf_data, size_t btf_size, struct bpf_btf_load_opts *opts)
1184{
1185 const size_t attr_sz = offsetofend(union bpf_attr, btf_log_true_size);
1186 union bpf_attr attr;
1187 char *log_buf;
1188 size_t log_size;
1189 __u32 log_level;
1190 int fd;
1191
1192 bump_rlimit_memlock();
1193
1194 memset(&attr, 0, attr_sz);
1195
1196 if (!OPTS_VALID(opts, bpf_btf_load_opts))
1197 return libbpf_err(-EINVAL);
1198
1199 log_buf = OPTS_GET(opts, log_buf, NULL);
1200 log_size = OPTS_GET(opts, log_size, 0);
1201 log_level = OPTS_GET(opts, log_level, 0);
1202
1203 if (log_size > UINT_MAX)
1204 return libbpf_err(-EINVAL);
1205 if (log_size && !log_buf)
1206 return libbpf_err(-EINVAL);
1207
1208 attr.btf = ptr_to_u64(btf_data);
1209 attr.btf_size = btf_size;
1210 /* log_level == 0 and log_buf != NULL means "try loading without
1211 * log_buf, but retry with log_buf and log_level=1 on error", which is
1212 * consistent across low-level and high-level BTF and program loading
1213 * APIs within libbpf and provides a sensible behavior in practice
1214 */
1215 if (log_level) {
1216 attr.btf_log_buf = ptr_to_u64(log_buf);
1217 attr.btf_log_size = (__u32)log_size;
1218 attr.btf_log_level = log_level;
1219 }
1220
1221 fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1222 if (fd < 0 && log_buf && log_level == 0) {
1223 attr.btf_log_buf = ptr_to_u64(log_buf);
1224 attr.btf_log_size = (__u32)log_size;
1225 attr.btf_log_level = 1;
1226 fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1227 }
1228
1229 OPTS_SET(opts, log_true_size, attr.btf_log_true_size);
1230 return libbpf_err_errno(fd);
1231}
1232
1233int bpf_task_fd_query(int pid, int fd, __u32 flags, char *buf, __u32 *buf_len,
1234 __u32 *prog_id, __u32 *fd_type, __u64 *probe_offset,
1235 __u64 *probe_addr)
1236{
1237 const size_t attr_sz = offsetofend(union bpf_attr, task_fd_query);
1238 union bpf_attr attr;
1239 int err;
1240
1241 memset(&attr, 0, attr_sz);
1242 attr.task_fd_query.pid = pid;
1243 attr.task_fd_query.fd = fd;
1244 attr.task_fd_query.flags = flags;
1245 attr.task_fd_query.buf = ptr_to_u64(buf);
1246 attr.task_fd_query.buf_len = *buf_len;
1247
1248 err = sys_bpf(BPF_TASK_FD_QUERY, &attr, attr_sz);
1249
1250 *buf_len = attr.task_fd_query.buf_len;
1251 *prog_id = attr.task_fd_query.prog_id;
1252 *fd_type = attr.task_fd_query.fd_type;
1253 *probe_offset = attr.task_fd_query.probe_offset;
1254 *probe_addr = attr.task_fd_query.probe_addr;
1255
1256 return libbpf_err_errno(err);
1257}
1258
1259int bpf_enable_stats(enum bpf_stats_type type)
1260{
1261 const size_t attr_sz = offsetofend(union bpf_attr, enable_stats);
1262 union bpf_attr attr;
1263 int fd;
1264
1265 memset(&attr, 0, attr_sz);
1266 attr.enable_stats.type = type;
1267
1268 fd = sys_bpf_fd(BPF_ENABLE_STATS, &attr, attr_sz);
1269 return libbpf_err_errno(fd);
1270}
1271
1272int bpf_prog_bind_map(int prog_fd, int map_fd,
1273 const struct bpf_prog_bind_opts *opts)
1274{
1275 const size_t attr_sz = offsetofend(union bpf_attr, prog_bind_map);
1276 union bpf_attr attr;
1277 int ret;
1278
1279 if (!OPTS_VALID(opts, bpf_prog_bind_opts))
1280 return libbpf_err(-EINVAL);
1281
1282 memset(&attr, 0, attr_sz);
1283 attr.prog_bind_map.prog_fd = prog_fd;
1284 attr.prog_bind_map.map_fd = map_fd;
1285 attr.prog_bind_map.flags = OPTS_GET(opts, flags, 0);
1286
1287 ret = sys_bpf(BPF_PROG_BIND_MAP, &attr, attr_sz);
1288 return libbpf_err_errno(ret);
1289}
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 <ast@kernel.org>
7 * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
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 "bpf.h"
32#include "libbpf.h"
33#include "libbpf_internal.h"
34
35/* make sure libbpf doesn't use kernel-only integer typedefs */
36#pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64
37
38/*
39 * When building perf, unistd.h is overridden. __NR_bpf is
40 * required to be defined explicitly.
41 */
42#ifndef __NR_bpf
43# if defined(__i386__)
44# define __NR_bpf 357
45# elif defined(__x86_64__)
46# define __NR_bpf 321
47# elif defined(__aarch64__)
48# define __NR_bpf 280
49# elif defined(__sparc__)
50# define __NR_bpf 349
51# elif defined(__s390__)
52# define __NR_bpf 351
53# elif defined(__arc__)
54# define __NR_bpf 280
55# else
56# error __NR_bpf not defined. libbpf does not support your arch.
57# endif
58#endif
59
60static inline __u64 ptr_to_u64(const void *ptr)
61{
62 return (__u64) (unsigned long) ptr;
63}
64
65static inline int sys_bpf(enum bpf_cmd cmd, union bpf_attr *attr,
66 unsigned int size)
67{
68 return syscall(__NR_bpf, cmd, attr, size);
69}
70
71static inline int sys_bpf_prog_load(union bpf_attr *attr, unsigned int size)
72{
73 int fd;
74
75 do {
76 fd = sys_bpf(BPF_PROG_LOAD, attr, size);
77 } while (fd < 0 && errno == EAGAIN);
78
79 return fd;
80}
81
82int bpf_create_map_xattr(const struct bpf_create_map_attr *create_attr)
83{
84 union bpf_attr attr;
85
86 memset(&attr, '\0', sizeof(attr));
87
88 attr.map_type = create_attr->map_type;
89 attr.key_size = create_attr->key_size;
90 attr.value_size = create_attr->value_size;
91 attr.max_entries = create_attr->max_entries;
92 attr.map_flags = create_attr->map_flags;
93 if (create_attr->name)
94 memcpy(attr.map_name, create_attr->name,
95 min(strlen(create_attr->name), BPF_OBJ_NAME_LEN - 1));
96 attr.numa_node = create_attr->numa_node;
97 attr.btf_fd = create_attr->btf_fd;
98 attr.btf_key_type_id = create_attr->btf_key_type_id;
99 attr.btf_value_type_id = create_attr->btf_value_type_id;
100 attr.map_ifindex = create_attr->map_ifindex;
101 if (attr.map_type == BPF_MAP_TYPE_STRUCT_OPS)
102 attr.btf_vmlinux_value_type_id =
103 create_attr->btf_vmlinux_value_type_id;
104 else
105 attr.inner_map_fd = create_attr->inner_map_fd;
106
107 return sys_bpf(BPF_MAP_CREATE, &attr, sizeof(attr));
108}
109
110int bpf_create_map_node(enum bpf_map_type map_type, const char *name,
111 int key_size, int value_size, int max_entries,
112 __u32 map_flags, int node)
113{
114 struct bpf_create_map_attr map_attr = {};
115
116 map_attr.name = name;
117 map_attr.map_type = map_type;
118 map_attr.map_flags = map_flags;
119 map_attr.key_size = key_size;
120 map_attr.value_size = value_size;
121 map_attr.max_entries = max_entries;
122 if (node >= 0) {
123 map_attr.numa_node = node;
124 map_attr.map_flags |= BPF_F_NUMA_NODE;
125 }
126
127 return bpf_create_map_xattr(&map_attr);
128}
129
130int bpf_create_map(enum bpf_map_type map_type, int key_size,
131 int value_size, int max_entries, __u32 map_flags)
132{
133 struct bpf_create_map_attr map_attr = {};
134
135 map_attr.map_type = map_type;
136 map_attr.map_flags = map_flags;
137 map_attr.key_size = key_size;
138 map_attr.value_size = value_size;
139 map_attr.max_entries = max_entries;
140
141 return bpf_create_map_xattr(&map_attr);
142}
143
144int bpf_create_map_name(enum bpf_map_type map_type, const char *name,
145 int key_size, int value_size, int max_entries,
146 __u32 map_flags)
147{
148 struct bpf_create_map_attr map_attr = {};
149
150 map_attr.name = name;
151 map_attr.map_type = map_type;
152 map_attr.map_flags = map_flags;
153 map_attr.key_size = key_size;
154 map_attr.value_size = value_size;
155 map_attr.max_entries = max_entries;
156
157 return bpf_create_map_xattr(&map_attr);
158}
159
160int bpf_create_map_in_map_node(enum bpf_map_type map_type, const char *name,
161 int key_size, int inner_map_fd, int max_entries,
162 __u32 map_flags, int node)
163{
164 union bpf_attr attr;
165
166 memset(&attr, '\0', sizeof(attr));
167
168 attr.map_type = map_type;
169 attr.key_size = key_size;
170 attr.value_size = 4;
171 attr.inner_map_fd = inner_map_fd;
172 attr.max_entries = max_entries;
173 attr.map_flags = map_flags;
174 if (name)
175 memcpy(attr.map_name, name,
176 min(strlen(name), BPF_OBJ_NAME_LEN - 1));
177
178 if (node >= 0) {
179 attr.map_flags |= BPF_F_NUMA_NODE;
180 attr.numa_node = node;
181 }
182
183 return sys_bpf(BPF_MAP_CREATE, &attr, sizeof(attr));
184}
185
186int bpf_create_map_in_map(enum bpf_map_type map_type, const char *name,
187 int key_size, int inner_map_fd, int max_entries,
188 __u32 map_flags)
189{
190 return bpf_create_map_in_map_node(map_type, name, key_size,
191 inner_map_fd, max_entries, map_flags,
192 -1);
193}
194
195static void *
196alloc_zero_tailing_info(const void *orecord, __u32 cnt,
197 __u32 actual_rec_size, __u32 expected_rec_size)
198{
199 __u64 info_len = (__u64)actual_rec_size * cnt;
200 void *info, *nrecord;
201 int i;
202
203 info = malloc(info_len);
204 if (!info)
205 return NULL;
206
207 /* zero out bytes kernel does not understand */
208 nrecord = info;
209 for (i = 0; i < cnt; i++) {
210 memcpy(nrecord, orecord, expected_rec_size);
211 memset(nrecord + expected_rec_size, 0,
212 actual_rec_size - expected_rec_size);
213 orecord += actual_rec_size;
214 nrecord += actual_rec_size;
215 }
216
217 return info;
218}
219
220int bpf_load_program_xattr(const struct bpf_load_program_attr *load_attr,
221 char *log_buf, size_t log_buf_sz)
222{
223 void *finfo = NULL, *linfo = NULL;
224 union bpf_attr attr;
225 __u32 log_level;
226 int fd;
227
228 if (!load_attr || !log_buf != !log_buf_sz)
229 return -EINVAL;
230
231 log_level = load_attr->log_level;
232 if (log_level > (4 | 2 | 1) || (log_level && !log_buf))
233 return -EINVAL;
234
235 memset(&attr, 0, sizeof(attr));
236 attr.prog_type = load_attr->prog_type;
237 attr.expected_attach_type = load_attr->expected_attach_type;
238 if (attr.prog_type == BPF_PROG_TYPE_STRUCT_OPS ||
239 attr.prog_type == BPF_PROG_TYPE_LSM) {
240 attr.attach_btf_id = load_attr->attach_btf_id;
241 } else if (attr.prog_type == BPF_PROG_TYPE_TRACING ||
242 attr.prog_type == BPF_PROG_TYPE_EXT) {
243 attr.attach_btf_id = load_attr->attach_btf_id;
244 attr.attach_prog_fd = load_attr->attach_prog_fd;
245 } else {
246 attr.prog_ifindex = load_attr->prog_ifindex;
247 attr.kern_version = load_attr->kern_version;
248 }
249 attr.insn_cnt = (__u32)load_attr->insns_cnt;
250 attr.insns = ptr_to_u64(load_attr->insns);
251 attr.license = ptr_to_u64(load_attr->license);
252
253 attr.log_level = log_level;
254 if (log_level) {
255 attr.log_buf = ptr_to_u64(log_buf);
256 attr.log_size = log_buf_sz;
257 } else {
258 attr.log_buf = ptr_to_u64(NULL);
259 attr.log_size = 0;
260 }
261
262 attr.prog_btf_fd = load_attr->prog_btf_fd;
263 attr.func_info_rec_size = load_attr->func_info_rec_size;
264 attr.func_info_cnt = load_attr->func_info_cnt;
265 attr.func_info = ptr_to_u64(load_attr->func_info);
266 attr.line_info_rec_size = load_attr->line_info_rec_size;
267 attr.line_info_cnt = load_attr->line_info_cnt;
268 attr.line_info = ptr_to_u64(load_attr->line_info);
269 if (load_attr->name)
270 memcpy(attr.prog_name, load_attr->name,
271 min(strlen(load_attr->name), BPF_OBJ_NAME_LEN - 1));
272 attr.prog_flags = load_attr->prog_flags;
273
274 fd = sys_bpf_prog_load(&attr, sizeof(attr));
275 if (fd >= 0)
276 return fd;
277
278 /* After bpf_prog_load, the kernel may modify certain attributes
279 * to give user space a hint how to deal with loading failure.
280 * Check to see whether we can make some changes and load again.
281 */
282 while (errno == E2BIG && (!finfo || !linfo)) {
283 if (!finfo && attr.func_info_cnt &&
284 attr.func_info_rec_size < load_attr->func_info_rec_size) {
285 /* try with corrected func info records */
286 finfo = alloc_zero_tailing_info(load_attr->func_info,
287 load_attr->func_info_cnt,
288 load_attr->func_info_rec_size,
289 attr.func_info_rec_size);
290 if (!finfo)
291 goto done;
292
293 attr.func_info = ptr_to_u64(finfo);
294 attr.func_info_rec_size = load_attr->func_info_rec_size;
295 } else if (!linfo && attr.line_info_cnt &&
296 attr.line_info_rec_size <
297 load_attr->line_info_rec_size) {
298 linfo = alloc_zero_tailing_info(load_attr->line_info,
299 load_attr->line_info_cnt,
300 load_attr->line_info_rec_size,
301 attr.line_info_rec_size);
302 if (!linfo)
303 goto done;
304
305 attr.line_info = ptr_to_u64(linfo);
306 attr.line_info_rec_size = load_attr->line_info_rec_size;
307 } else {
308 break;
309 }
310
311 fd = sys_bpf_prog_load(&attr, sizeof(attr));
312
313 if (fd >= 0)
314 goto done;
315 }
316
317 if (log_level || !log_buf)
318 goto done;
319
320 /* Try again with log */
321 attr.log_buf = ptr_to_u64(log_buf);
322 attr.log_size = log_buf_sz;
323 attr.log_level = 1;
324 log_buf[0] = 0;
325 fd = sys_bpf_prog_load(&attr, sizeof(attr));
326done:
327 free(finfo);
328 free(linfo);
329 return fd;
330}
331
332int bpf_load_program(enum bpf_prog_type type, const struct bpf_insn *insns,
333 size_t insns_cnt, const char *license,
334 __u32 kern_version, char *log_buf,
335 size_t log_buf_sz)
336{
337 struct bpf_load_program_attr load_attr;
338
339 memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
340 load_attr.prog_type = type;
341 load_attr.expected_attach_type = 0;
342 load_attr.name = NULL;
343 load_attr.insns = insns;
344 load_attr.insns_cnt = insns_cnt;
345 load_attr.license = license;
346 load_attr.kern_version = kern_version;
347
348 return bpf_load_program_xattr(&load_attr, log_buf, log_buf_sz);
349}
350
351int bpf_verify_program(enum bpf_prog_type type, const struct bpf_insn *insns,
352 size_t insns_cnt, __u32 prog_flags, const char *license,
353 __u32 kern_version, char *log_buf, size_t log_buf_sz,
354 int log_level)
355{
356 union bpf_attr attr;
357
358 memset(&attr, 0, sizeof(attr));
359 attr.prog_type = type;
360 attr.insn_cnt = (__u32)insns_cnt;
361 attr.insns = ptr_to_u64(insns);
362 attr.license = ptr_to_u64(license);
363 attr.log_buf = ptr_to_u64(log_buf);
364 attr.log_size = log_buf_sz;
365 attr.log_level = log_level;
366 log_buf[0] = 0;
367 attr.kern_version = kern_version;
368 attr.prog_flags = prog_flags;
369
370 return sys_bpf_prog_load(&attr, sizeof(attr));
371}
372
373int bpf_map_update_elem(int fd, const void *key, const void *value,
374 __u64 flags)
375{
376 union bpf_attr attr;
377
378 memset(&attr, 0, sizeof(attr));
379 attr.map_fd = fd;
380 attr.key = ptr_to_u64(key);
381 attr.value = ptr_to_u64(value);
382 attr.flags = flags;
383
384 return sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr));
385}
386
387int bpf_map_lookup_elem(int fd, const void *key, void *value)
388{
389 union bpf_attr attr;
390
391 memset(&attr, 0, sizeof(attr));
392 attr.map_fd = fd;
393 attr.key = ptr_to_u64(key);
394 attr.value = ptr_to_u64(value);
395
396 return sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr));
397}
398
399int bpf_map_lookup_elem_flags(int fd, const void *key, void *value, __u64 flags)
400{
401 union bpf_attr attr;
402
403 memset(&attr, 0, sizeof(attr));
404 attr.map_fd = fd;
405 attr.key = ptr_to_u64(key);
406 attr.value = ptr_to_u64(value);
407 attr.flags = flags;
408
409 return sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr));
410}
411
412int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value)
413{
414 union bpf_attr attr;
415
416 memset(&attr, 0, sizeof(attr));
417 attr.map_fd = fd;
418 attr.key = ptr_to_u64(key);
419 attr.value = ptr_to_u64(value);
420
421 return sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, sizeof(attr));
422}
423
424int bpf_map_delete_elem(int fd, const void *key)
425{
426 union bpf_attr attr;
427
428 memset(&attr, 0, sizeof(attr));
429 attr.map_fd = fd;
430 attr.key = ptr_to_u64(key);
431
432 return sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr));
433}
434
435int bpf_map_get_next_key(int fd, const void *key, void *next_key)
436{
437 union bpf_attr attr;
438
439 memset(&attr, 0, sizeof(attr));
440 attr.map_fd = fd;
441 attr.key = ptr_to_u64(key);
442 attr.next_key = ptr_to_u64(next_key);
443
444 return sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, sizeof(attr));
445}
446
447int bpf_map_freeze(int fd)
448{
449 union bpf_attr attr;
450
451 memset(&attr, 0, sizeof(attr));
452 attr.map_fd = fd;
453
454 return sys_bpf(BPF_MAP_FREEZE, &attr, sizeof(attr));
455}
456
457static int bpf_map_batch_common(int cmd, int fd, void *in_batch,
458 void *out_batch, void *keys, void *values,
459 __u32 *count,
460 const struct bpf_map_batch_opts *opts)
461{
462 union bpf_attr attr;
463 int ret;
464
465 if (!OPTS_VALID(opts, bpf_map_batch_opts))
466 return -EINVAL;
467
468 memset(&attr, 0, sizeof(attr));
469 attr.batch.map_fd = fd;
470 attr.batch.in_batch = ptr_to_u64(in_batch);
471 attr.batch.out_batch = ptr_to_u64(out_batch);
472 attr.batch.keys = ptr_to_u64(keys);
473 attr.batch.values = ptr_to_u64(values);
474 attr.batch.count = *count;
475 attr.batch.elem_flags = OPTS_GET(opts, elem_flags, 0);
476 attr.batch.flags = OPTS_GET(opts, flags, 0);
477
478 ret = sys_bpf(cmd, &attr, sizeof(attr));
479 *count = attr.batch.count;
480
481 return ret;
482}
483
484int bpf_map_delete_batch(int fd, void *keys, __u32 *count,
485 const struct bpf_map_batch_opts *opts)
486{
487 return bpf_map_batch_common(BPF_MAP_DELETE_BATCH, fd, NULL,
488 NULL, keys, NULL, count, opts);
489}
490
491int bpf_map_lookup_batch(int fd, void *in_batch, void *out_batch, void *keys,
492 void *values, __u32 *count,
493 const struct bpf_map_batch_opts *opts)
494{
495 return bpf_map_batch_common(BPF_MAP_LOOKUP_BATCH, fd, in_batch,
496 out_batch, keys, values, count, opts);
497}
498
499int bpf_map_lookup_and_delete_batch(int fd, void *in_batch, void *out_batch,
500 void *keys, void *values, __u32 *count,
501 const struct bpf_map_batch_opts *opts)
502{
503 return bpf_map_batch_common(BPF_MAP_LOOKUP_AND_DELETE_BATCH,
504 fd, in_batch, out_batch, keys, values,
505 count, opts);
506}
507
508int bpf_map_update_batch(int fd, void *keys, void *values, __u32 *count,
509 const struct bpf_map_batch_opts *opts)
510{
511 return bpf_map_batch_common(BPF_MAP_UPDATE_BATCH, fd, NULL, NULL,
512 keys, values, count, opts);
513}
514
515int bpf_obj_pin(int fd, const char *pathname)
516{
517 union bpf_attr attr;
518
519 memset(&attr, 0, sizeof(attr));
520 attr.pathname = ptr_to_u64((void *)pathname);
521 attr.bpf_fd = fd;
522
523 return sys_bpf(BPF_OBJ_PIN, &attr, sizeof(attr));
524}
525
526int bpf_obj_get(const char *pathname)
527{
528 union bpf_attr attr;
529
530 memset(&attr, 0, sizeof(attr));
531 attr.pathname = ptr_to_u64((void *)pathname);
532
533 return sys_bpf(BPF_OBJ_GET, &attr, sizeof(attr));
534}
535
536int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type,
537 unsigned int flags)
538{
539 DECLARE_LIBBPF_OPTS(bpf_prog_attach_opts, opts,
540 .flags = flags,
541 );
542
543 return bpf_prog_attach_xattr(prog_fd, target_fd, type, &opts);
544}
545
546int bpf_prog_attach_xattr(int prog_fd, int target_fd,
547 enum bpf_attach_type type,
548 const struct bpf_prog_attach_opts *opts)
549{
550 union bpf_attr attr;
551
552 if (!OPTS_VALID(opts, bpf_prog_attach_opts))
553 return -EINVAL;
554
555 memset(&attr, 0, sizeof(attr));
556 attr.target_fd = target_fd;
557 attr.attach_bpf_fd = prog_fd;
558 attr.attach_type = type;
559 attr.attach_flags = OPTS_GET(opts, flags, 0);
560 attr.replace_bpf_fd = OPTS_GET(opts, replace_prog_fd, 0);
561
562 return sys_bpf(BPF_PROG_ATTACH, &attr, sizeof(attr));
563}
564
565int bpf_prog_detach(int target_fd, enum bpf_attach_type type)
566{
567 union bpf_attr attr;
568
569 memset(&attr, 0, sizeof(attr));
570 attr.target_fd = target_fd;
571 attr.attach_type = type;
572
573 return sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr));
574}
575
576int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type)
577{
578 union bpf_attr attr;
579
580 memset(&attr, 0, sizeof(attr));
581 attr.target_fd = target_fd;
582 attr.attach_bpf_fd = prog_fd;
583 attr.attach_type = type;
584
585 return sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr));
586}
587
588int bpf_link_create(int prog_fd, int target_fd,
589 enum bpf_attach_type attach_type,
590 const struct bpf_link_create_opts *opts)
591{
592 union bpf_attr attr;
593
594 if (!OPTS_VALID(opts, bpf_link_create_opts))
595 return -EINVAL;
596
597 memset(&attr, 0, sizeof(attr));
598 attr.link_create.prog_fd = prog_fd;
599 attr.link_create.target_fd = target_fd;
600 attr.link_create.attach_type = attach_type;
601 attr.link_create.flags = OPTS_GET(opts, flags, 0);
602 attr.link_create.iter_info =
603 ptr_to_u64(OPTS_GET(opts, iter_info, (void *)0));
604 attr.link_create.iter_info_len = OPTS_GET(opts, iter_info_len, 0);
605
606 return sys_bpf(BPF_LINK_CREATE, &attr, sizeof(attr));
607}
608
609int bpf_link_detach(int link_fd)
610{
611 union bpf_attr attr;
612
613 memset(&attr, 0, sizeof(attr));
614 attr.link_detach.link_fd = link_fd;
615
616 return sys_bpf(BPF_LINK_DETACH, &attr, sizeof(attr));
617}
618
619int bpf_link_update(int link_fd, int new_prog_fd,
620 const struct bpf_link_update_opts *opts)
621{
622 union bpf_attr attr;
623
624 if (!OPTS_VALID(opts, bpf_link_update_opts))
625 return -EINVAL;
626
627 memset(&attr, 0, sizeof(attr));
628 attr.link_update.link_fd = link_fd;
629 attr.link_update.new_prog_fd = new_prog_fd;
630 attr.link_update.flags = OPTS_GET(opts, flags, 0);
631 attr.link_update.old_prog_fd = OPTS_GET(opts, old_prog_fd, 0);
632
633 return sys_bpf(BPF_LINK_UPDATE, &attr, sizeof(attr));
634}
635
636int bpf_iter_create(int link_fd)
637{
638 union bpf_attr attr;
639
640 memset(&attr, 0, sizeof(attr));
641 attr.iter_create.link_fd = link_fd;
642
643 return sys_bpf(BPF_ITER_CREATE, &attr, sizeof(attr));
644}
645
646int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags,
647 __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt)
648{
649 union bpf_attr attr;
650 int ret;
651
652 memset(&attr, 0, sizeof(attr));
653 attr.query.target_fd = target_fd;
654 attr.query.attach_type = type;
655 attr.query.query_flags = query_flags;
656 attr.query.prog_cnt = *prog_cnt;
657 attr.query.prog_ids = ptr_to_u64(prog_ids);
658
659 ret = sys_bpf(BPF_PROG_QUERY, &attr, sizeof(attr));
660 if (attach_flags)
661 *attach_flags = attr.query.attach_flags;
662 *prog_cnt = attr.query.prog_cnt;
663 return ret;
664}
665
666int bpf_prog_test_run(int prog_fd, int repeat, void *data, __u32 size,
667 void *data_out, __u32 *size_out, __u32 *retval,
668 __u32 *duration)
669{
670 union bpf_attr attr;
671 int ret;
672
673 memset(&attr, 0, sizeof(attr));
674 attr.test.prog_fd = prog_fd;
675 attr.test.data_in = ptr_to_u64(data);
676 attr.test.data_out = ptr_to_u64(data_out);
677 attr.test.data_size_in = size;
678 attr.test.repeat = repeat;
679
680 ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr));
681 if (size_out)
682 *size_out = attr.test.data_size_out;
683 if (retval)
684 *retval = attr.test.retval;
685 if (duration)
686 *duration = attr.test.duration;
687 return ret;
688}
689
690int bpf_prog_test_run_xattr(struct bpf_prog_test_run_attr *test_attr)
691{
692 union bpf_attr attr;
693 int ret;
694
695 if (!test_attr->data_out && test_attr->data_size_out > 0)
696 return -EINVAL;
697
698 memset(&attr, 0, sizeof(attr));
699 attr.test.prog_fd = test_attr->prog_fd;
700 attr.test.data_in = ptr_to_u64(test_attr->data_in);
701 attr.test.data_out = ptr_to_u64(test_attr->data_out);
702 attr.test.data_size_in = test_attr->data_size_in;
703 attr.test.data_size_out = test_attr->data_size_out;
704 attr.test.ctx_in = ptr_to_u64(test_attr->ctx_in);
705 attr.test.ctx_out = ptr_to_u64(test_attr->ctx_out);
706 attr.test.ctx_size_in = test_attr->ctx_size_in;
707 attr.test.ctx_size_out = test_attr->ctx_size_out;
708 attr.test.repeat = test_attr->repeat;
709
710 ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr));
711 test_attr->data_size_out = attr.test.data_size_out;
712 test_attr->ctx_size_out = attr.test.ctx_size_out;
713 test_attr->retval = attr.test.retval;
714 test_attr->duration = attr.test.duration;
715 return ret;
716}
717
718static int bpf_obj_get_next_id(__u32 start_id, __u32 *next_id, int cmd)
719{
720 union bpf_attr attr;
721 int err;
722
723 memset(&attr, 0, sizeof(attr));
724 attr.start_id = start_id;
725
726 err = sys_bpf(cmd, &attr, sizeof(attr));
727 if (!err)
728 *next_id = attr.next_id;
729
730 return err;
731}
732
733int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id)
734{
735 return bpf_obj_get_next_id(start_id, next_id, BPF_PROG_GET_NEXT_ID);
736}
737
738int bpf_map_get_next_id(__u32 start_id, __u32 *next_id)
739{
740 return bpf_obj_get_next_id(start_id, next_id, BPF_MAP_GET_NEXT_ID);
741}
742
743int bpf_btf_get_next_id(__u32 start_id, __u32 *next_id)
744{
745 return bpf_obj_get_next_id(start_id, next_id, BPF_BTF_GET_NEXT_ID);
746}
747
748int bpf_link_get_next_id(__u32 start_id, __u32 *next_id)
749{
750 return bpf_obj_get_next_id(start_id, next_id, BPF_LINK_GET_NEXT_ID);
751}
752
753int bpf_prog_get_fd_by_id(__u32 id)
754{
755 union bpf_attr attr;
756
757 memset(&attr, 0, sizeof(attr));
758 attr.prog_id = id;
759
760 return sys_bpf(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr));
761}
762
763int bpf_map_get_fd_by_id(__u32 id)
764{
765 union bpf_attr attr;
766
767 memset(&attr, 0, sizeof(attr));
768 attr.map_id = id;
769
770 return sys_bpf(BPF_MAP_GET_FD_BY_ID, &attr, sizeof(attr));
771}
772
773int bpf_btf_get_fd_by_id(__u32 id)
774{
775 union bpf_attr attr;
776
777 memset(&attr, 0, sizeof(attr));
778 attr.btf_id = id;
779
780 return sys_bpf(BPF_BTF_GET_FD_BY_ID, &attr, sizeof(attr));
781}
782
783int bpf_link_get_fd_by_id(__u32 id)
784{
785 union bpf_attr attr;
786
787 memset(&attr, 0, sizeof(attr));
788 attr.link_id = id;
789
790 return sys_bpf(BPF_LINK_GET_FD_BY_ID, &attr, sizeof(attr));
791}
792
793int bpf_obj_get_info_by_fd(int bpf_fd, void *info, __u32 *info_len)
794{
795 union bpf_attr attr;
796 int err;
797
798 memset(&attr, 0, sizeof(attr));
799 attr.info.bpf_fd = bpf_fd;
800 attr.info.info_len = *info_len;
801 attr.info.info = ptr_to_u64(info);
802
803 err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr));
804 if (!err)
805 *info_len = attr.info.info_len;
806
807 return err;
808}
809
810int bpf_raw_tracepoint_open(const char *name, int prog_fd)
811{
812 union bpf_attr attr;
813
814 memset(&attr, 0, sizeof(attr));
815 attr.raw_tracepoint.name = ptr_to_u64(name);
816 attr.raw_tracepoint.prog_fd = prog_fd;
817
818 return sys_bpf(BPF_RAW_TRACEPOINT_OPEN, &attr, sizeof(attr));
819}
820
821int bpf_load_btf(void *btf, __u32 btf_size, char *log_buf, __u32 log_buf_size,
822 bool do_log)
823{
824 union bpf_attr attr = {};
825 int fd;
826
827 attr.btf = ptr_to_u64(btf);
828 attr.btf_size = btf_size;
829
830retry:
831 if (do_log && log_buf && log_buf_size) {
832 attr.btf_log_level = 1;
833 attr.btf_log_size = log_buf_size;
834 attr.btf_log_buf = ptr_to_u64(log_buf);
835 }
836
837 fd = sys_bpf(BPF_BTF_LOAD, &attr, sizeof(attr));
838 if (fd == -1 && !do_log && log_buf && log_buf_size) {
839 do_log = true;
840 goto retry;
841 }
842
843 return fd;
844}
845
846int bpf_task_fd_query(int pid, int fd, __u32 flags, char *buf, __u32 *buf_len,
847 __u32 *prog_id, __u32 *fd_type, __u64 *probe_offset,
848 __u64 *probe_addr)
849{
850 union bpf_attr attr = {};
851 int err;
852
853 attr.task_fd_query.pid = pid;
854 attr.task_fd_query.fd = fd;
855 attr.task_fd_query.flags = flags;
856 attr.task_fd_query.buf = ptr_to_u64(buf);
857 attr.task_fd_query.buf_len = *buf_len;
858
859 err = sys_bpf(BPF_TASK_FD_QUERY, &attr, sizeof(attr));
860 *buf_len = attr.task_fd_query.buf_len;
861 *prog_id = attr.task_fd_query.prog_id;
862 *fd_type = attr.task_fd_query.fd_type;
863 *probe_offset = attr.task_fd_query.probe_offset;
864 *probe_addr = attr.task_fd_query.probe_addr;
865
866 return err;
867}
868
869int bpf_enable_stats(enum bpf_stats_type type)
870{
871 union bpf_attr attr;
872
873 memset(&attr, 0, sizeof(attr));
874 attr.enable_stats.type = type;
875
876 return sys_bpf(BPF_ENABLE_STATS, &attr, sizeof(attr));
877}