1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
4 * common eBPF ELF operations.
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.
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!)
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.
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>
28 #include <asm/unistd.h>
30 #include <linux/bpf.h>
31 #include <linux/filter.h>
32 #include <linux/kernel.h>
34 #include <sys/resource.h>
37 #include "libbpf_internal.h"
40 * When building perf, unistd.h is overridden. __NR_bpf is
41 * required to be defined explicitly.
44 # if defined(__i386__)
46 # elif defined(__x86_64__)
48 # elif defined(__aarch64__)
50 # elif defined(__sparc__)
52 # elif defined(__s390__)
54 # elif defined(__arc__)
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
63 # error __NR_bpf not defined. libbpf does not support your arch.
67 static inline __u64 ptr_to_u64(const void *ptr)
69 return (__u64) (unsigned long) ptr;
72 static inline int sys_bpf(enum bpf_cmd cmd, union bpf_attr *attr,
75 return syscall(__NR_bpf, cmd, attr, size);
78 static inline int sys_bpf_fd(enum bpf_cmd cmd, union bpf_attr *attr,
83 fd = sys_bpf(cmd, attr, size);
84 return ensure_good_fd(fd);
87 int sys_bpf_prog_load(union bpf_attr *attr, unsigned int size, int attempts)
92 fd = sys_bpf_fd(BPF_PROG_LOAD, attr, size);
93 } while (fd < 0 && errno == EAGAIN && --attempts > 0);
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.
103 * [0] https://lore.kernel.org/bpf/20201201215900.3569844-1-guro@fb.com/
104 * [1] d05512618056 ("bpf: Add bpf_ktime_get_coarse_ns helper")
106 int probe_memcg_account(void)
108 const size_t prog_load_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),
113 size_t insn_cnt = ARRAY_SIZE(insns);
117 /* attempt loading freplace trying to use custom BTF */
118 memset(&attr, 0, prog_load_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");
124 prog_fd = sys_bpf_fd(BPF_PROG_LOAD, &attr, prog_load_attr_sz);
132 static bool memlock_bumped;
133 static rlim_t memlock_rlim = RLIM_INFINITY;
135 int libbpf_set_memlock_rlim(size_t memlock_bytes)
138 return libbpf_err(-EBUSY);
140 memlock_rlim = memlock_bytes;
144 int bump_rlimit_memlock(void)
148 /* if kernel supports memcg-based accounting, skip bumping RLIMIT_MEMLOCK */
149 if (memlock_bumped || kernel_supports(NULL, FEAT_MEMCG_ACCOUNT))
152 memlock_bumped = true;
154 /* zero memlock_rlim_max disables auto-bumping RLIMIT_MEMLOCK */
155 if (memlock_rlim == 0)
158 rlim.rlim_cur = rlim.rlim_max = memlock_rlim;
159 if (setrlimit(RLIMIT_MEMLOCK, &rlim))
165 int bpf_map_create(enum bpf_map_type map_type,
166 const char *map_name,
170 const struct bpf_map_create_opts *opts)
172 const size_t attr_sz = offsetofend(union bpf_attr, map_extra);
176 bump_rlimit_memlock();
178 memset(&attr, 0, attr_sz);
180 if (!OPTS_VALID(opts, bpf_map_create_opts))
181 return libbpf_err(-EINVAL);
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;
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);
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);
201 fd = sys_bpf_fd(BPF_MAP_CREATE, &attr, attr_sz);
202 return libbpf_err_errno(fd);
206 alloc_zero_tailing_info(const void *orecord, __u32 cnt,
207 __u32 actual_rec_size, __u32 expected_rec_size)
209 __u64 info_len = (__u64)actual_rec_size * cnt;
210 void *info, *nrecord;
213 info = malloc(info_len);
217 /* zero out bytes kernel does not understand */
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;
230 int 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 const struct bpf_prog_load_opts *opts)
235 void *finfo = NULL, *linfo = NULL;
236 const char *func_info, *line_info;
237 __u32 log_size, log_level, attach_prog_fd, attach_btf_obj_fd;
238 __u32 func_info_rec_size, line_info_rec_size;
243 bump_rlimit_memlock();
245 if (!OPTS_VALID(opts, bpf_prog_load_opts))
246 return libbpf_err(-EINVAL);
248 attempts = OPTS_GET(opts, attempts, 0);
250 return libbpf_err(-EINVAL);
252 attempts = PROG_LOAD_ATTEMPTS;
254 memset(&attr, 0, sizeof(attr));
256 attr.prog_type = prog_type;
257 attr.expected_attach_type = OPTS_GET(opts, expected_attach_type, 0);
259 attr.prog_btf_fd = OPTS_GET(opts, prog_btf_fd, 0);
260 attr.prog_flags = OPTS_GET(opts, prog_flags, 0);
261 attr.prog_ifindex = OPTS_GET(opts, prog_ifindex, 0);
262 attr.kern_version = OPTS_GET(opts, kern_version, 0);
264 if (prog_name && kernel_supports(NULL, FEAT_PROG_NAME))
265 libbpf_strlcpy(attr.prog_name, prog_name, sizeof(attr.prog_name));
266 attr.license = ptr_to_u64(license);
268 if (insn_cnt > UINT_MAX)
269 return libbpf_err(-E2BIG);
271 attr.insns = ptr_to_u64(insns);
272 attr.insn_cnt = (__u32)insn_cnt;
274 attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
275 attach_btf_obj_fd = OPTS_GET(opts, attach_btf_obj_fd, 0);
277 if (attach_prog_fd && attach_btf_obj_fd)
278 return libbpf_err(-EINVAL);
280 attr.attach_btf_id = OPTS_GET(opts, attach_btf_id, 0);
282 attr.attach_prog_fd = attach_prog_fd;
284 attr.attach_btf_obj_fd = attach_btf_obj_fd;
286 log_buf = OPTS_GET(opts, log_buf, NULL);
287 log_size = OPTS_GET(opts, log_size, 0);
288 log_level = OPTS_GET(opts, log_level, 0);
290 if (!!log_buf != !!log_size)
291 return libbpf_err(-EINVAL);
292 if (log_level > (4 | 2 | 1))
293 return libbpf_err(-EINVAL);
294 if (log_level && !log_buf)
295 return libbpf_err(-EINVAL);
297 func_info_rec_size = OPTS_GET(opts, func_info_rec_size, 0);
298 func_info = OPTS_GET(opts, func_info, NULL);
299 attr.func_info_rec_size = func_info_rec_size;
300 attr.func_info = ptr_to_u64(func_info);
301 attr.func_info_cnt = OPTS_GET(opts, func_info_cnt, 0);
303 line_info_rec_size = OPTS_GET(opts, line_info_rec_size, 0);
304 line_info = OPTS_GET(opts, line_info, NULL);
305 attr.line_info_rec_size = line_info_rec_size;
306 attr.line_info = ptr_to_u64(line_info);
307 attr.line_info_cnt = OPTS_GET(opts, line_info_cnt, 0);
309 attr.fd_array = ptr_to_u64(OPTS_GET(opts, fd_array, NULL));
312 attr.log_buf = ptr_to_u64(log_buf);
313 attr.log_size = log_size;
314 attr.log_level = log_level;
317 fd = sys_bpf_prog_load(&attr, sizeof(attr), attempts);
321 /* After bpf_prog_load, the kernel may modify certain attributes
322 * to give user space a hint how to deal with loading failure.
323 * Check to see whether we can make some changes and load again.
325 while (errno == E2BIG && (!finfo || !linfo)) {
326 if (!finfo && attr.func_info_cnt &&
327 attr.func_info_rec_size < func_info_rec_size) {
328 /* try with corrected func info records */
329 finfo = alloc_zero_tailing_info(func_info,
332 attr.func_info_rec_size);
338 attr.func_info = ptr_to_u64(finfo);
339 attr.func_info_rec_size = func_info_rec_size;
340 } else if (!linfo && attr.line_info_cnt &&
341 attr.line_info_rec_size < line_info_rec_size) {
342 linfo = alloc_zero_tailing_info(line_info,
345 attr.line_info_rec_size);
351 attr.line_info = ptr_to_u64(linfo);
352 attr.line_info_rec_size = line_info_rec_size;
357 fd = sys_bpf_prog_load(&attr, sizeof(attr), attempts);
362 if (log_level == 0 && log_buf) {
363 /* log_level == 0 with non-NULL log_buf requires retrying on error
364 * with log_level == 1 and log_buf/log_buf_size set, to get details of
367 attr.log_buf = ptr_to_u64(log_buf);
368 attr.log_size = log_size;
371 fd = sys_bpf_prog_load(&attr, sizeof(attr), attempts);
374 /* free() doesn't affect errno, so we don't need to restore it */
377 return libbpf_err_errno(fd);
380 int bpf_map_update_elem(int fd, const void *key, const void *value,
386 memset(&attr, 0, sizeof(attr));
388 attr.key = ptr_to_u64(key);
389 attr.value = ptr_to_u64(value);
392 ret = sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr));
393 return libbpf_err_errno(ret);
396 int bpf_map_lookup_elem(int fd, const void *key, void *value)
401 memset(&attr, 0, sizeof(attr));
403 attr.key = ptr_to_u64(key);
404 attr.value = ptr_to_u64(value);
406 ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr));
407 return libbpf_err_errno(ret);
410 int bpf_map_lookup_elem_flags(int fd, const void *key, void *value, __u64 flags)
415 memset(&attr, 0, sizeof(attr));
417 attr.key = ptr_to_u64(key);
418 attr.value = ptr_to_u64(value);
421 ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr));
422 return libbpf_err_errno(ret);
425 int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value)
430 memset(&attr, 0, sizeof(attr));
432 attr.key = ptr_to_u64(key);
433 attr.value = ptr_to_u64(value);
435 ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, sizeof(attr));
436 return libbpf_err_errno(ret);
439 int bpf_map_lookup_and_delete_elem_flags(int fd, const void *key, void *value, __u64 flags)
444 memset(&attr, 0, sizeof(attr));
446 attr.key = ptr_to_u64(key);
447 attr.value = ptr_to_u64(value);
450 ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, sizeof(attr));
451 return libbpf_err_errno(ret);
454 int bpf_map_delete_elem(int fd, const void *key)
459 memset(&attr, 0, sizeof(attr));
461 attr.key = ptr_to_u64(key);
463 ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr));
464 return libbpf_err_errno(ret);
467 int bpf_map_delete_elem_flags(int fd, const void *key, __u64 flags)
472 memset(&attr, 0, sizeof(attr));
474 attr.key = ptr_to_u64(key);
477 ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr));
478 return libbpf_err_errno(ret);
481 int bpf_map_get_next_key(int fd, const void *key, void *next_key)
486 memset(&attr, 0, sizeof(attr));
488 attr.key = ptr_to_u64(key);
489 attr.next_key = ptr_to_u64(next_key);
491 ret = sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, sizeof(attr));
492 return libbpf_err_errno(ret);
495 int bpf_map_freeze(int fd)
500 memset(&attr, 0, sizeof(attr));
503 ret = sys_bpf(BPF_MAP_FREEZE, &attr, sizeof(attr));
504 return libbpf_err_errno(ret);
507 static int bpf_map_batch_common(int cmd, int fd, void *in_batch,
508 void *out_batch, void *keys, void *values,
510 const struct bpf_map_batch_opts *opts)
515 if (!OPTS_VALID(opts, bpf_map_batch_opts))
516 return libbpf_err(-EINVAL);
518 memset(&attr, 0, sizeof(attr));
519 attr.batch.map_fd = fd;
520 attr.batch.in_batch = ptr_to_u64(in_batch);
521 attr.batch.out_batch = ptr_to_u64(out_batch);
522 attr.batch.keys = ptr_to_u64(keys);
523 attr.batch.values = ptr_to_u64(values);
524 attr.batch.count = *count;
525 attr.batch.elem_flags = OPTS_GET(opts, elem_flags, 0);
526 attr.batch.flags = OPTS_GET(opts, flags, 0);
528 ret = sys_bpf(cmd, &attr, sizeof(attr));
529 *count = attr.batch.count;
531 return libbpf_err_errno(ret);
534 int bpf_map_delete_batch(int fd, const void *keys, __u32 *count,
535 const struct bpf_map_batch_opts *opts)
537 return bpf_map_batch_common(BPF_MAP_DELETE_BATCH, fd, NULL,
538 NULL, (void *)keys, NULL, count, opts);
541 int bpf_map_lookup_batch(int fd, void *in_batch, void *out_batch, void *keys,
542 void *values, __u32 *count,
543 const struct bpf_map_batch_opts *opts)
545 return bpf_map_batch_common(BPF_MAP_LOOKUP_BATCH, fd, in_batch,
546 out_batch, keys, values, count, opts);
549 int bpf_map_lookup_and_delete_batch(int fd, void *in_batch, void *out_batch,
550 void *keys, void *values, __u32 *count,
551 const struct bpf_map_batch_opts *opts)
553 return bpf_map_batch_common(BPF_MAP_LOOKUP_AND_DELETE_BATCH,
554 fd, in_batch, out_batch, keys, values,
558 int bpf_map_update_batch(int fd, const void *keys, const void *values, __u32 *count,
559 const struct bpf_map_batch_opts *opts)
561 return bpf_map_batch_common(BPF_MAP_UPDATE_BATCH, fd, NULL, NULL,
562 (void *)keys, (void *)values, count, opts);
565 int bpf_obj_pin(int fd, const char *pathname)
570 memset(&attr, 0, sizeof(attr));
571 attr.pathname = ptr_to_u64((void *)pathname);
574 ret = sys_bpf(BPF_OBJ_PIN, &attr, sizeof(attr));
575 return libbpf_err_errno(ret);
578 int bpf_obj_get(const char *pathname)
580 return bpf_obj_get_opts(pathname, NULL);
583 int bpf_obj_get_opts(const char *pathname, const struct bpf_obj_get_opts *opts)
588 if (!OPTS_VALID(opts, bpf_obj_get_opts))
589 return libbpf_err(-EINVAL);
591 memset(&attr, 0, sizeof(attr));
592 attr.pathname = ptr_to_u64((void *)pathname);
593 attr.file_flags = OPTS_GET(opts, file_flags, 0);
595 fd = sys_bpf_fd(BPF_OBJ_GET, &attr, sizeof(attr));
596 return libbpf_err_errno(fd);
599 int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type,
602 DECLARE_LIBBPF_OPTS(bpf_prog_attach_opts, opts,
606 return bpf_prog_attach_opts(prog_fd, target_fd, type, &opts);
609 int bpf_prog_attach_opts(int prog_fd, int target_fd,
610 enum bpf_attach_type type,
611 const struct bpf_prog_attach_opts *opts)
616 if (!OPTS_VALID(opts, bpf_prog_attach_opts))
617 return libbpf_err(-EINVAL);
619 memset(&attr, 0, sizeof(attr));
620 attr.target_fd = target_fd;
621 attr.attach_bpf_fd = prog_fd;
622 attr.attach_type = type;
623 attr.attach_flags = OPTS_GET(opts, flags, 0);
624 attr.replace_bpf_fd = OPTS_GET(opts, replace_prog_fd, 0);
626 ret = sys_bpf(BPF_PROG_ATTACH, &attr, sizeof(attr));
627 return libbpf_err_errno(ret);
630 __attribute__((alias("bpf_prog_attach_opts")))
631 int bpf_prog_attach_xattr(int prog_fd, int target_fd,
632 enum bpf_attach_type type,
633 const struct bpf_prog_attach_opts *opts);
635 int bpf_prog_detach(int target_fd, enum bpf_attach_type type)
640 memset(&attr, 0, sizeof(attr));
641 attr.target_fd = target_fd;
642 attr.attach_type = type;
644 ret = sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr));
645 return libbpf_err_errno(ret);
648 int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type)
653 memset(&attr, 0, sizeof(attr));
654 attr.target_fd = target_fd;
655 attr.attach_bpf_fd = prog_fd;
656 attr.attach_type = type;
658 ret = sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr));
659 return libbpf_err_errno(ret);
662 int bpf_link_create(int prog_fd, int target_fd,
663 enum bpf_attach_type attach_type,
664 const struct bpf_link_create_opts *opts)
666 __u32 target_btf_id, iter_info_len;
670 if (!OPTS_VALID(opts, bpf_link_create_opts))
671 return libbpf_err(-EINVAL);
673 iter_info_len = OPTS_GET(opts, iter_info_len, 0);
674 target_btf_id = OPTS_GET(opts, target_btf_id, 0);
676 /* validate we don't have unexpected combinations of non-zero fields */
677 if (iter_info_len || target_btf_id) {
678 if (iter_info_len && target_btf_id)
679 return libbpf_err(-EINVAL);
680 if (!OPTS_ZEROED(opts, target_btf_id))
681 return libbpf_err(-EINVAL);
684 memset(&attr, 0, sizeof(attr));
685 attr.link_create.prog_fd = prog_fd;
686 attr.link_create.target_fd = target_fd;
687 attr.link_create.attach_type = attach_type;
688 attr.link_create.flags = OPTS_GET(opts, flags, 0);
691 attr.link_create.target_btf_id = target_btf_id;
695 switch (attach_type) {
697 attr.link_create.iter_info = ptr_to_u64(OPTS_GET(opts, iter_info, (void *)0));
698 attr.link_create.iter_info_len = iter_info_len;
701 attr.link_create.perf_event.bpf_cookie = OPTS_GET(opts, perf_event.bpf_cookie, 0);
702 if (!OPTS_ZEROED(opts, perf_event))
703 return libbpf_err(-EINVAL);
705 case BPF_TRACE_KPROBE_MULTI:
706 attr.link_create.kprobe_multi.flags = OPTS_GET(opts, kprobe_multi.flags, 0);
707 attr.link_create.kprobe_multi.cnt = OPTS_GET(opts, kprobe_multi.cnt, 0);
708 attr.link_create.kprobe_multi.syms = ptr_to_u64(OPTS_GET(opts, kprobe_multi.syms, 0));
709 attr.link_create.kprobe_multi.addrs = ptr_to_u64(OPTS_GET(opts, kprobe_multi.addrs, 0));
710 attr.link_create.kprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, kprobe_multi.cookies, 0));
711 if (!OPTS_ZEROED(opts, kprobe_multi))
712 return libbpf_err(-EINVAL);
714 case BPF_TRACE_FENTRY:
715 case BPF_TRACE_FEXIT:
716 case BPF_MODIFY_RETURN:
718 attr.link_create.tracing.cookie = OPTS_GET(opts, tracing.cookie, 0);
719 if (!OPTS_ZEROED(opts, tracing))
720 return libbpf_err(-EINVAL);
723 if (!OPTS_ZEROED(opts, flags))
724 return libbpf_err(-EINVAL);
728 fd = sys_bpf_fd(BPF_LINK_CREATE, &attr, sizeof(attr));
731 /* we'll get EINVAL if LINK_CREATE doesn't support attaching fentry
732 * and other similar programs
736 return libbpf_err(err);
738 /* if user used features not supported by
739 * BPF_RAW_TRACEPOINT_OPEN command, then just give up immediately
741 if (attr.link_create.target_fd || attr.link_create.target_btf_id)
742 return libbpf_err(err);
743 if (!OPTS_ZEROED(opts, sz))
744 return libbpf_err(err);
746 /* otherwise, for few select kinds of programs that can be
747 * attached using BPF_RAW_TRACEPOINT_OPEN command, try that as
748 * a fallback for older kernels
750 switch (attach_type) {
751 case BPF_TRACE_RAW_TP:
753 case BPF_TRACE_FENTRY:
754 case BPF_TRACE_FEXIT:
755 case BPF_MODIFY_RETURN:
756 return bpf_raw_tracepoint_open(NULL, prog_fd);
758 return libbpf_err(err);
762 int bpf_link_detach(int link_fd)
767 memset(&attr, 0, sizeof(attr));
768 attr.link_detach.link_fd = link_fd;
770 ret = sys_bpf(BPF_LINK_DETACH, &attr, sizeof(attr));
771 return libbpf_err_errno(ret);
774 int bpf_link_update(int link_fd, int new_prog_fd,
775 const struct bpf_link_update_opts *opts)
780 if (!OPTS_VALID(opts, bpf_link_update_opts))
781 return libbpf_err(-EINVAL);
783 memset(&attr, 0, sizeof(attr));
784 attr.link_update.link_fd = link_fd;
785 attr.link_update.new_prog_fd = new_prog_fd;
786 attr.link_update.flags = OPTS_GET(opts, flags, 0);
787 attr.link_update.old_prog_fd = OPTS_GET(opts, old_prog_fd, 0);
789 ret = sys_bpf(BPF_LINK_UPDATE, &attr, sizeof(attr));
790 return libbpf_err_errno(ret);
793 int bpf_iter_create(int link_fd)
798 memset(&attr, 0, sizeof(attr));
799 attr.iter_create.link_fd = link_fd;
801 fd = sys_bpf_fd(BPF_ITER_CREATE, &attr, sizeof(attr));
802 return libbpf_err_errno(fd);
805 int bpf_prog_query_opts(int target_fd,
806 enum bpf_attach_type type,
807 struct bpf_prog_query_opts *opts)
812 if (!OPTS_VALID(opts, bpf_prog_query_opts))
813 return libbpf_err(-EINVAL);
815 memset(&attr, 0, sizeof(attr));
817 attr.query.target_fd = target_fd;
818 attr.query.attach_type = type;
819 attr.query.query_flags = OPTS_GET(opts, query_flags, 0);
820 attr.query.prog_cnt = OPTS_GET(opts, prog_cnt, 0);
821 attr.query.prog_ids = ptr_to_u64(OPTS_GET(opts, prog_ids, NULL));
822 attr.query.prog_attach_flags = ptr_to_u64(OPTS_GET(opts, prog_attach_flags, NULL));
824 ret = sys_bpf(BPF_PROG_QUERY, &attr, sizeof(attr));
826 OPTS_SET(opts, attach_flags, attr.query.attach_flags);
827 OPTS_SET(opts, prog_cnt, attr.query.prog_cnt);
829 return libbpf_err_errno(ret);
832 int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags,
833 __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt)
835 LIBBPF_OPTS(bpf_prog_query_opts, opts);
838 opts.query_flags = query_flags;
839 opts.prog_ids = prog_ids;
840 opts.prog_cnt = *prog_cnt;
842 ret = bpf_prog_query_opts(target_fd, type, &opts);
845 *attach_flags = opts.attach_flags;
846 *prog_cnt = opts.prog_cnt;
848 return libbpf_err_errno(ret);
851 int bpf_prog_test_run_opts(int prog_fd, struct bpf_test_run_opts *opts)
856 if (!OPTS_VALID(opts, bpf_test_run_opts))
857 return libbpf_err(-EINVAL);
859 memset(&attr, 0, sizeof(attr));
860 attr.test.prog_fd = prog_fd;
861 attr.test.batch_size = OPTS_GET(opts, batch_size, 0);
862 attr.test.cpu = OPTS_GET(opts, cpu, 0);
863 attr.test.flags = OPTS_GET(opts, flags, 0);
864 attr.test.repeat = OPTS_GET(opts, repeat, 0);
865 attr.test.duration = OPTS_GET(opts, duration, 0);
866 attr.test.ctx_size_in = OPTS_GET(opts, ctx_size_in, 0);
867 attr.test.ctx_size_out = OPTS_GET(opts, ctx_size_out, 0);
868 attr.test.data_size_in = OPTS_GET(opts, data_size_in, 0);
869 attr.test.data_size_out = OPTS_GET(opts, data_size_out, 0);
870 attr.test.ctx_in = ptr_to_u64(OPTS_GET(opts, ctx_in, NULL));
871 attr.test.ctx_out = ptr_to_u64(OPTS_GET(opts, ctx_out, NULL));
872 attr.test.data_in = ptr_to_u64(OPTS_GET(opts, data_in, NULL));
873 attr.test.data_out = ptr_to_u64(OPTS_GET(opts, data_out, NULL));
875 ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr));
877 OPTS_SET(opts, data_size_out, attr.test.data_size_out);
878 OPTS_SET(opts, ctx_size_out, attr.test.ctx_size_out);
879 OPTS_SET(opts, duration, attr.test.duration);
880 OPTS_SET(opts, retval, attr.test.retval);
882 return libbpf_err_errno(ret);
885 static int bpf_obj_get_next_id(__u32 start_id, __u32 *next_id, int cmd)
890 memset(&attr, 0, sizeof(attr));
891 attr.start_id = start_id;
893 err = sys_bpf(cmd, &attr, sizeof(attr));
895 *next_id = attr.next_id;
897 return libbpf_err_errno(err);
900 int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id)
902 return bpf_obj_get_next_id(start_id, next_id, BPF_PROG_GET_NEXT_ID);
905 int bpf_map_get_next_id(__u32 start_id, __u32 *next_id)
907 return bpf_obj_get_next_id(start_id, next_id, BPF_MAP_GET_NEXT_ID);
910 int bpf_btf_get_next_id(__u32 start_id, __u32 *next_id)
912 return bpf_obj_get_next_id(start_id, next_id, BPF_BTF_GET_NEXT_ID);
915 int bpf_link_get_next_id(__u32 start_id, __u32 *next_id)
917 return bpf_obj_get_next_id(start_id, next_id, BPF_LINK_GET_NEXT_ID);
920 int bpf_prog_get_fd_by_id(__u32 id)
925 memset(&attr, 0, sizeof(attr));
928 fd = sys_bpf_fd(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr));
929 return libbpf_err_errno(fd);
932 int bpf_map_get_fd_by_id(__u32 id)
937 memset(&attr, 0, sizeof(attr));
940 fd = sys_bpf_fd(BPF_MAP_GET_FD_BY_ID, &attr, sizeof(attr));
941 return libbpf_err_errno(fd);
944 int bpf_btf_get_fd_by_id(__u32 id)
949 memset(&attr, 0, sizeof(attr));
952 fd = sys_bpf_fd(BPF_BTF_GET_FD_BY_ID, &attr, sizeof(attr));
953 return libbpf_err_errno(fd);
956 int bpf_link_get_fd_by_id(__u32 id)
961 memset(&attr, 0, sizeof(attr));
964 fd = sys_bpf_fd(BPF_LINK_GET_FD_BY_ID, &attr, sizeof(attr));
965 return libbpf_err_errno(fd);
968 int bpf_obj_get_info_by_fd(int bpf_fd, void *info, __u32 *info_len)
973 memset(&attr, 0, sizeof(attr));
974 attr.info.bpf_fd = bpf_fd;
975 attr.info.info_len = *info_len;
976 attr.info.info = ptr_to_u64(info);
978 err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr));
981 *info_len = attr.info.info_len;
983 return libbpf_err_errno(err);
986 int bpf_raw_tracepoint_open(const char *name, int prog_fd)
991 memset(&attr, 0, sizeof(attr));
992 attr.raw_tracepoint.name = ptr_to_u64(name);
993 attr.raw_tracepoint.prog_fd = prog_fd;
995 fd = sys_bpf_fd(BPF_RAW_TRACEPOINT_OPEN, &attr, sizeof(attr));
996 return libbpf_err_errno(fd);
999 int bpf_btf_load(const void *btf_data, size_t btf_size, const struct bpf_btf_load_opts *opts)
1001 const size_t attr_sz = offsetofend(union bpf_attr, btf_log_level);
1002 union bpf_attr attr;
1008 bump_rlimit_memlock();
1010 memset(&attr, 0, attr_sz);
1012 if (!OPTS_VALID(opts, bpf_btf_load_opts))
1013 return libbpf_err(-EINVAL);
1015 log_buf = OPTS_GET(opts, log_buf, NULL);
1016 log_size = OPTS_GET(opts, log_size, 0);
1017 log_level = OPTS_GET(opts, log_level, 0);
1019 if (log_size > UINT_MAX)
1020 return libbpf_err(-EINVAL);
1021 if (log_size && !log_buf)
1022 return libbpf_err(-EINVAL);
1024 attr.btf = ptr_to_u64(btf_data);
1025 attr.btf_size = btf_size;
1026 /* log_level == 0 and log_buf != NULL means "try loading without
1027 * log_buf, but retry with log_buf and log_level=1 on error", which is
1028 * consistent across low-level and high-level BTF and program loading
1029 * APIs within libbpf and provides a sensible behavior in practice
1032 attr.btf_log_buf = ptr_to_u64(log_buf);
1033 attr.btf_log_size = (__u32)log_size;
1034 attr.btf_log_level = log_level;
1037 fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1038 if (fd < 0 && log_buf && log_level == 0) {
1039 attr.btf_log_buf = ptr_to_u64(log_buf);
1040 attr.btf_log_size = (__u32)log_size;
1041 attr.btf_log_level = 1;
1042 fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1044 return libbpf_err_errno(fd);
1047 int bpf_task_fd_query(int pid, int fd, __u32 flags, char *buf, __u32 *buf_len,
1048 __u32 *prog_id, __u32 *fd_type, __u64 *probe_offset,
1051 union bpf_attr attr = {};
1054 attr.task_fd_query.pid = pid;
1055 attr.task_fd_query.fd = fd;
1056 attr.task_fd_query.flags = flags;
1057 attr.task_fd_query.buf = ptr_to_u64(buf);
1058 attr.task_fd_query.buf_len = *buf_len;
1060 err = sys_bpf(BPF_TASK_FD_QUERY, &attr, sizeof(attr));
1062 *buf_len = attr.task_fd_query.buf_len;
1063 *prog_id = attr.task_fd_query.prog_id;
1064 *fd_type = attr.task_fd_query.fd_type;
1065 *probe_offset = attr.task_fd_query.probe_offset;
1066 *probe_addr = attr.task_fd_query.probe_addr;
1068 return libbpf_err_errno(err);
1071 int bpf_enable_stats(enum bpf_stats_type type)
1073 union bpf_attr attr;
1076 memset(&attr, 0, sizeof(attr));
1077 attr.enable_stats.type = type;
1079 fd = sys_bpf_fd(BPF_ENABLE_STATS, &attr, sizeof(attr));
1080 return libbpf_err_errno(fd);
1083 int bpf_prog_bind_map(int prog_fd, int map_fd,
1084 const struct bpf_prog_bind_opts *opts)
1086 union bpf_attr attr;
1089 if (!OPTS_VALID(opts, bpf_prog_bind_opts))
1090 return libbpf_err(-EINVAL);
1092 memset(&attr, 0, sizeof(attr));
1093 attr.prog_bind_map.prog_fd = prog_fd;
1094 attr.prog_bind_map.map_fd = map_fd;
1095 attr.prog_bind_map.flags = OPTS_GET(opts, flags, 0);
1097 ret = sys_bpf(BPF_PROG_BIND_MAP, &attr, sizeof(attr));
1098 return libbpf_err_errno(ret);