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
87#define PROG_LOAD_ATTEMPTS 5
88
89static inline int sys_bpf_prog_load(union bpf_attr *attr, unsigned int size, int attempts)
90{
91	int fd;
92
93	do {
94		fd = sys_bpf_fd(BPF_PROG_LOAD, attr, size);
95	} while (fd < 0 && errno == EAGAIN && --attempts > 0);
96
97	return fd;
98}
99
100/* Probe whether kernel switched from memlock-based (RLIMIT_MEMLOCK) to
101 * memcg-based memory accounting for BPF maps and progs. This was done in [0].
102 * We use the support for bpf_ktime_get_coarse_ns() helper, which was added in
103 * the same 5.11 Linux release ([1]), to detect memcg-based accounting for BPF.
104 *
105 *   [0] https://lore.kernel.org/bpf/20201201215900.3569844-1-guro@fb.com/
106 *   [1] d05512618056 ("bpf: Add bpf_ktime_get_coarse_ns helper")
107 */
108int probe_memcg_account(void)
109{
110	const size_t prog_load_attr_sz = offsetofend(union bpf_attr, attach_btf_obj_fd);
111	struct bpf_insn insns[] = {
112		BPF_EMIT_CALL(BPF_FUNC_ktime_get_coarse_ns),
113		BPF_EXIT_INSN(),
114	};
115	size_t insn_cnt = ARRAY_SIZE(insns);
116	union bpf_attr attr;
117	int prog_fd;
118
119	/* attempt loading freplace trying to use custom BTF */
120	memset(&attr, 0, prog_load_attr_sz);
121	attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
122	attr.insns = ptr_to_u64(insns);
123	attr.insn_cnt = insn_cnt;
124	attr.license = ptr_to_u64("GPL");
125
126	prog_fd = sys_bpf_fd(BPF_PROG_LOAD, &attr, prog_load_attr_sz);
127	if (prog_fd >= 0) {
128		close(prog_fd);
129		return 1;
130	}
131	return 0;
132}
133
134static bool memlock_bumped;
135static rlim_t memlock_rlim = RLIM_INFINITY;
136
137int libbpf_set_memlock_rlim(size_t memlock_bytes)
138{
139	if (memlock_bumped)
140		return libbpf_err(-EBUSY);
141
142	memlock_rlim = memlock_bytes;
143	return 0;
144}
145
146int bump_rlimit_memlock(void)
147{
148	struct rlimit rlim;
149
150	/* if kernel supports memcg-based accounting, skip bumping RLIMIT_MEMLOCK */
151	if (memlock_bumped || kernel_supports(NULL, FEAT_MEMCG_ACCOUNT))
152		return 0;
153
154	memlock_bumped = true;
155
156	/* zero memlock_rlim_max disables auto-bumping RLIMIT_MEMLOCK */
157	if (memlock_rlim == 0)
158		return 0;
159
160	rlim.rlim_cur = rlim.rlim_max = memlock_rlim;
161	if (setrlimit(RLIMIT_MEMLOCK, &rlim))
162		return -errno;
163
164	return 0;
165}
166
167int bpf_map_create(enum bpf_map_type map_type,
168		   const char *map_name,
169		   __u32 key_size,
170		   __u32 value_size,
171		   __u32 max_entries,
172		   const struct bpf_map_create_opts *opts)
173{
174	const size_t attr_sz = offsetofend(union bpf_attr, map_extra);
175	union bpf_attr attr;
176	int fd;
177
178	bump_rlimit_memlock();
179
180	memset(&attr, 0, attr_sz);
181
182	if (!OPTS_VALID(opts, bpf_map_create_opts))
183		return libbpf_err(-EINVAL);
184
185	attr.map_type = map_type;
186	if (map_name)
187		libbpf_strlcpy(attr.map_name, map_name, sizeof(attr.map_name));
188	attr.key_size = key_size;
189	attr.value_size = value_size;
190	attr.max_entries = max_entries;
191
192	attr.btf_fd = OPTS_GET(opts, btf_fd, 0);
193	attr.btf_key_type_id = OPTS_GET(opts, btf_key_type_id, 0);
194	attr.btf_value_type_id = OPTS_GET(opts, btf_value_type_id, 0);
195	attr.btf_vmlinux_value_type_id = OPTS_GET(opts, btf_vmlinux_value_type_id, 0);
196
197	attr.inner_map_fd = OPTS_GET(opts, inner_map_fd, 0);
198	attr.map_flags = OPTS_GET(opts, map_flags, 0);
199	attr.map_extra = OPTS_GET(opts, map_extra, 0);
200	attr.numa_node = OPTS_GET(opts, numa_node, 0);
201	attr.map_ifindex = OPTS_GET(opts, map_ifindex, 0);
202
203	fd = sys_bpf_fd(BPF_MAP_CREATE, &attr, attr_sz);
204	return libbpf_err_errno(fd);
205}
206
207static void *
208alloc_zero_tailing_info(const void *orecord, __u32 cnt,
209			__u32 actual_rec_size, __u32 expected_rec_size)
210{
211	__u64 info_len = (__u64)actual_rec_size * cnt;
212	void *info, *nrecord;
213	int i;
214
215	info = malloc(info_len);
216	if (!info)
217		return NULL;
218
219	/* zero out bytes kernel does not understand */
220	nrecord = info;
221	for (i = 0; i < cnt; i++) {
222		memcpy(nrecord, orecord, expected_rec_size);
223		memset(nrecord + expected_rec_size, 0,
224		       actual_rec_size - expected_rec_size);
225		orecord += actual_rec_size;
226		nrecord += actual_rec_size;
227	}
228
229	return info;
230}
231
232int bpf_prog_load(enum bpf_prog_type prog_type,
233		  const char *prog_name, const char *license,
234		  const struct bpf_insn *insns, size_t insn_cnt,
235		  const struct bpf_prog_load_opts *opts)
236{
237	void *finfo = NULL, *linfo = NULL;
238	const char *func_info, *line_info;
239	__u32 log_size, log_level, attach_prog_fd, attach_btf_obj_fd;
240	__u32 func_info_rec_size, line_info_rec_size;
241	int fd, attempts;
242	union bpf_attr attr;
243	char *log_buf;
244
245	bump_rlimit_memlock();
246
247	if (!OPTS_VALID(opts, bpf_prog_load_opts))
248		return libbpf_err(-EINVAL);
249
250	attempts = OPTS_GET(opts, attempts, 0);
251	if (attempts < 0)
252		return libbpf_err(-EINVAL);
253	if (attempts == 0)
254		attempts = PROG_LOAD_ATTEMPTS;
255
256	memset(&attr, 0, sizeof(attr));
257
258	attr.prog_type = prog_type;
259	attr.expected_attach_type = OPTS_GET(opts, expected_attach_type, 0);
260
261	attr.prog_btf_fd = OPTS_GET(opts, prog_btf_fd, 0);
262	attr.prog_flags = OPTS_GET(opts, prog_flags, 0);
263	attr.prog_ifindex = OPTS_GET(opts, prog_ifindex, 0);
264	attr.kern_version = OPTS_GET(opts, kern_version, 0);
265
266	if (prog_name)
267		libbpf_strlcpy(attr.prog_name, prog_name, sizeof(attr.prog_name));
268	attr.license = ptr_to_u64(license);
269
270	if (insn_cnt > UINT_MAX)
271		return libbpf_err(-E2BIG);
272
273	attr.insns = ptr_to_u64(insns);
274	attr.insn_cnt = (__u32)insn_cnt;
275
276	attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
277	attach_btf_obj_fd = OPTS_GET(opts, attach_btf_obj_fd, 0);
278
279	if (attach_prog_fd && attach_btf_obj_fd)
280		return libbpf_err(-EINVAL);
281
282	attr.attach_btf_id = OPTS_GET(opts, attach_btf_id, 0);
283	if (attach_prog_fd)
284		attr.attach_prog_fd = attach_prog_fd;
285	else
286		attr.attach_btf_obj_fd = attach_btf_obj_fd;
287
288	log_buf = OPTS_GET(opts, log_buf, NULL);
289	log_size = OPTS_GET(opts, log_size, 0);
290	log_level = OPTS_GET(opts, log_level, 0);
291
292	if (!!log_buf != !!log_size)
293		return libbpf_err(-EINVAL);
294	if (log_level > (4 | 2 | 1))
295		return libbpf_err(-EINVAL);
296	if (log_level && !log_buf)
297		return libbpf_err(-EINVAL);
298
299	func_info_rec_size = OPTS_GET(opts, func_info_rec_size, 0);
300	func_info = OPTS_GET(opts, func_info, NULL);
301	attr.func_info_rec_size = func_info_rec_size;
302	attr.func_info = ptr_to_u64(func_info);
303	attr.func_info_cnt = OPTS_GET(opts, func_info_cnt, 0);
304
305	line_info_rec_size = OPTS_GET(opts, line_info_rec_size, 0);
306	line_info = OPTS_GET(opts, line_info, NULL);
307	attr.line_info_rec_size = line_info_rec_size;
308	attr.line_info = ptr_to_u64(line_info);
309	attr.line_info_cnt = OPTS_GET(opts, line_info_cnt, 0);
310
311	attr.fd_array = ptr_to_u64(OPTS_GET(opts, fd_array, NULL));
312
313	if (log_level) {
314		attr.log_buf = ptr_to_u64(log_buf);
315		attr.log_size = log_size;
316		attr.log_level = log_level;
317	}
318
319	fd = sys_bpf_prog_load(&attr, sizeof(attr), attempts);
320	if (fd >= 0)
321		return fd;
322
323	/* After bpf_prog_load, the kernel may modify certain attributes
324	 * to give user space a hint how to deal with loading failure.
325	 * Check to see whether we can make some changes and load again.
326	 */
327	while (errno == E2BIG && (!finfo || !linfo)) {
328		if (!finfo && attr.func_info_cnt &&
329		    attr.func_info_rec_size < func_info_rec_size) {
330			/* try with corrected func info records */
331			finfo = alloc_zero_tailing_info(func_info,
332							attr.func_info_cnt,
333							func_info_rec_size,
334							attr.func_info_rec_size);
335			if (!finfo) {
336				errno = E2BIG;
337				goto done;
338			}
339
340			attr.func_info = ptr_to_u64(finfo);
341			attr.func_info_rec_size = func_info_rec_size;
342		} else if (!linfo && attr.line_info_cnt &&
343			   attr.line_info_rec_size < line_info_rec_size) {
344			linfo = alloc_zero_tailing_info(line_info,
345							attr.line_info_cnt,
346							line_info_rec_size,
347							attr.line_info_rec_size);
348			if (!linfo) {
349				errno = E2BIG;
350				goto done;
351			}
352
353			attr.line_info = ptr_to_u64(linfo);
354			attr.line_info_rec_size = line_info_rec_size;
355		} else {
356			break;
357		}
358
359		fd = sys_bpf_prog_load(&attr, sizeof(attr), attempts);
360		if (fd >= 0)
361			goto done;
362	}
363
364	if (log_level == 0 && log_buf) {
365		/* log_level == 0 with non-NULL log_buf requires retrying on error
366		 * with log_level == 1 and log_buf/log_buf_size set, to get details of
367		 * failure
368		 */
369		attr.log_buf = ptr_to_u64(log_buf);
370		attr.log_size = log_size;
371		attr.log_level = 1;
372
373		fd = sys_bpf_prog_load(&attr, sizeof(attr), attempts);
374	}
375done:
376	/* free() doesn't affect errno, so we don't need to restore it */
377	free(finfo);
378	free(linfo);
379	return libbpf_err_errno(fd);
380}
381
382int bpf_map_update_elem(int fd, const void *key, const void *value,
383			__u64 flags)
384{
385	union bpf_attr attr;
386	int ret;
387
388	memset(&attr, 0, sizeof(attr));
389	attr.map_fd = fd;
390	attr.key = ptr_to_u64(key);
391	attr.value = ptr_to_u64(value);
392	attr.flags = flags;
393
394	ret = sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr));
395	return libbpf_err_errno(ret);
396}
397
398int bpf_map_lookup_elem(int fd, const void *key, void *value)
399{
400	union bpf_attr attr;
401	int ret;
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
408	ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr));
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	union bpf_attr attr;
415	int ret;
416
417	memset(&attr, 0, sizeof(attr));
418	attr.map_fd = fd;
419	attr.key = ptr_to_u64(key);
420	attr.value = ptr_to_u64(value);
421	attr.flags = flags;
422
423	ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr));
424	return libbpf_err_errno(ret);
425}
426
427int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value)
428{
429	union bpf_attr attr;
430	int ret;
431
432	memset(&attr, 0, sizeof(attr));
433	attr.map_fd = fd;
434	attr.key = ptr_to_u64(key);
435	attr.value = ptr_to_u64(value);
436
437	ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, sizeof(attr));
438	return libbpf_err_errno(ret);
439}
440
441int bpf_map_lookup_and_delete_elem_flags(int fd, const void *key, void *value, __u64 flags)
442{
443	union bpf_attr attr;
444	int ret;
445
446	memset(&attr, 0, sizeof(attr));
447	attr.map_fd = fd;
448	attr.key = ptr_to_u64(key);
449	attr.value = ptr_to_u64(value);
450	attr.flags = flags;
451
452	ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, sizeof(attr));
453	return libbpf_err_errno(ret);
454}
455
456int bpf_map_delete_elem(int fd, const void *key)
457{
458	union bpf_attr attr;
459	int ret;
460
461	memset(&attr, 0, sizeof(attr));
462	attr.map_fd = fd;
463	attr.key = ptr_to_u64(key);
464
465	ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr));
466	return libbpf_err_errno(ret);
467}
468
469int bpf_map_delete_elem_flags(int fd, const void *key, __u64 flags)
470{
471	union bpf_attr attr;
472	int ret;
473
474	memset(&attr, 0, sizeof(attr));
475	attr.map_fd = fd;
476	attr.key = ptr_to_u64(key);
477	attr.flags = flags;
478
479	ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr));
480	return libbpf_err_errno(ret);
481}
482
483int bpf_map_get_next_key(int fd, const void *key, void *next_key)
484{
485	union bpf_attr attr;
486	int ret;
487
488	memset(&attr, 0, sizeof(attr));
489	attr.map_fd = fd;
490	attr.key = ptr_to_u64(key);
491	attr.next_key = ptr_to_u64(next_key);
492
493	ret = sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, sizeof(attr));
494	return libbpf_err_errno(ret);
495}
496
497int bpf_map_freeze(int fd)
498{
499	union bpf_attr attr;
500	int ret;
501
502	memset(&attr, 0, sizeof(attr));
503	attr.map_fd = fd;
504
505	ret = sys_bpf(BPF_MAP_FREEZE, &attr, sizeof(attr));
506	return libbpf_err_errno(ret);
507}
508
509static int bpf_map_batch_common(int cmd, int fd, void  *in_batch,
510				void *out_batch, void *keys, void *values,
511				__u32 *count,
512				const struct bpf_map_batch_opts *opts)
513{
514	union bpf_attr attr;
515	int ret;
516
517	if (!OPTS_VALID(opts, bpf_map_batch_opts))
518		return libbpf_err(-EINVAL);
519
520	memset(&attr, 0, sizeof(attr));
521	attr.batch.map_fd = fd;
522	attr.batch.in_batch = ptr_to_u64(in_batch);
523	attr.batch.out_batch = ptr_to_u64(out_batch);
524	attr.batch.keys = ptr_to_u64(keys);
525	attr.batch.values = ptr_to_u64(values);
526	attr.batch.count = *count;
527	attr.batch.elem_flags  = OPTS_GET(opts, elem_flags, 0);
528	attr.batch.flags = OPTS_GET(opts, flags, 0);
529
530	ret = sys_bpf(cmd, &attr, sizeof(attr));
531	*count = attr.batch.count;
532
533	return libbpf_err_errno(ret);
534}
535
536int bpf_map_delete_batch(int fd, const void *keys, __u32 *count,
537			 const struct bpf_map_batch_opts *opts)
538{
539	return bpf_map_batch_common(BPF_MAP_DELETE_BATCH, fd, NULL,
540				    NULL, (void *)keys, NULL, count, opts);
541}
542
543int bpf_map_lookup_batch(int fd, void *in_batch, void *out_batch, void *keys,
544			 void *values, __u32 *count,
545			 const struct bpf_map_batch_opts *opts)
546{
547	return bpf_map_batch_common(BPF_MAP_LOOKUP_BATCH, fd, in_batch,
548				    out_batch, keys, values, count, opts);
549}
550
551int bpf_map_lookup_and_delete_batch(int fd, void *in_batch, void *out_batch,
552				    void *keys, void *values, __u32 *count,
553				    const struct bpf_map_batch_opts *opts)
554{
555	return bpf_map_batch_common(BPF_MAP_LOOKUP_AND_DELETE_BATCH,
556				    fd, in_batch, out_batch, keys, values,
557				    count, opts);
558}
559
560int bpf_map_update_batch(int fd, const void *keys, const void *values, __u32 *count,
561			 const struct bpf_map_batch_opts *opts)
562{
563	return bpf_map_batch_common(BPF_MAP_UPDATE_BATCH, fd, NULL, NULL,
564				    (void *)keys, (void *)values, count, opts);
565}
566
567int bpf_obj_pin(int fd, const char *pathname)
568{
569	union bpf_attr attr;
570	int ret;
571
572	memset(&attr, 0, sizeof(attr));
573	attr.pathname = ptr_to_u64((void *)pathname);
574	attr.bpf_fd = fd;
575
576	ret = sys_bpf(BPF_OBJ_PIN, &attr, sizeof(attr));
577	return libbpf_err_errno(ret);
578}
579
580int bpf_obj_get(const char *pathname)
581{
582	return bpf_obj_get_opts(pathname, NULL);
583}
584
585int bpf_obj_get_opts(const char *pathname, const struct bpf_obj_get_opts *opts)
586{
587	union bpf_attr attr;
588	int fd;
589
590	if (!OPTS_VALID(opts, bpf_obj_get_opts))
591		return libbpf_err(-EINVAL);
592
593	memset(&attr, 0, sizeof(attr));
594	attr.pathname = ptr_to_u64((void *)pathname);
595	attr.file_flags = OPTS_GET(opts, file_flags, 0);
596
597	fd = sys_bpf_fd(BPF_OBJ_GET, &attr, sizeof(attr));
598	return libbpf_err_errno(fd);
599}
600
601int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type,
602		    unsigned int flags)
603{
604	DECLARE_LIBBPF_OPTS(bpf_prog_attach_opts, opts,
605		.flags = flags,
606	);
607
608	return bpf_prog_attach_opts(prog_fd, target_fd, type, &opts);
609}
610
611int bpf_prog_attach_opts(int prog_fd, int target_fd,
612			  enum bpf_attach_type type,
613			  const struct bpf_prog_attach_opts *opts)
614{
615	union bpf_attr attr;
616	int ret;
617
618	if (!OPTS_VALID(opts, bpf_prog_attach_opts))
619		return libbpf_err(-EINVAL);
620
621	memset(&attr, 0, sizeof(attr));
622	attr.target_fd	   = target_fd;
623	attr.attach_bpf_fd = prog_fd;
624	attr.attach_type   = type;
625	attr.attach_flags  = OPTS_GET(opts, flags, 0);
626	attr.replace_bpf_fd = OPTS_GET(opts, replace_prog_fd, 0);
627
628	ret = sys_bpf(BPF_PROG_ATTACH, &attr, sizeof(attr));
629	return libbpf_err_errno(ret);
630}
631
632__attribute__((alias("bpf_prog_attach_opts")))
633int bpf_prog_attach_xattr(int prog_fd, int target_fd,
634			  enum bpf_attach_type type,
635			  const struct bpf_prog_attach_opts *opts);
636
637int bpf_prog_detach(int target_fd, enum bpf_attach_type type)
638{
639	union bpf_attr attr;
640	int ret;
641
642	memset(&attr, 0, sizeof(attr));
643	attr.target_fd	 = target_fd;
644	attr.attach_type = type;
645
646	ret = sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr));
647	return libbpf_err_errno(ret);
648}
649
650int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type)
651{
652	union bpf_attr attr;
653	int ret;
654
655	memset(&attr, 0, sizeof(attr));
656	attr.target_fd	 = target_fd;
657	attr.attach_bpf_fd = prog_fd;
658	attr.attach_type = type;
659
660	ret = sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr));
661	return libbpf_err_errno(ret);
662}
663
664int bpf_link_create(int prog_fd, int target_fd,
665		    enum bpf_attach_type attach_type,
666		    const struct bpf_link_create_opts *opts)
667{
668	__u32 target_btf_id, iter_info_len;
669	union bpf_attr attr;
670	int fd, err;
671
672	if (!OPTS_VALID(opts, bpf_link_create_opts))
673		return libbpf_err(-EINVAL);
674
675	iter_info_len = OPTS_GET(opts, iter_info_len, 0);
676	target_btf_id = OPTS_GET(opts, target_btf_id, 0);
677
678	/* validate we don't have unexpected combinations of non-zero fields */
679	if (iter_info_len || target_btf_id) {
680		if (iter_info_len && target_btf_id)
681			return libbpf_err(-EINVAL);
682		if (!OPTS_ZEROED(opts, target_btf_id))
683			return libbpf_err(-EINVAL);
684	}
685
686	memset(&attr, 0, sizeof(attr));
687	attr.link_create.prog_fd = prog_fd;
688	attr.link_create.target_fd = target_fd;
689	attr.link_create.attach_type = attach_type;
690	attr.link_create.flags = OPTS_GET(opts, flags, 0);
691
692	if (target_btf_id) {
693		attr.link_create.target_btf_id = target_btf_id;
694		goto proceed;
695	}
696
697	switch (attach_type) {
698	case BPF_TRACE_ITER:
699		attr.link_create.iter_info = ptr_to_u64(OPTS_GET(opts, iter_info, (void *)0));
700		attr.link_create.iter_info_len = iter_info_len;
701		break;
702	case BPF_PERF_EVENT:
703		attr.link_create.perf_event.bpf_cookie = OPTS_GET(opts, perf_event.bpf_cookie, 0);
704		if (!OPTS_ZEROED(opts, perf_event))
705			return libbpf_err(-EINVAL);
706		break;
707	case BPF_TRACE_KPROBE_MULTI:
708		attr.link_create.kprobe_multi.flags = OPTS_GET(opts, kprobe_multi.flags, 0);
709		attr.link_create.kprobe_multi.cnt = OPTS_GET(opts, kprobe_multi.cnt, 0);
710		attr.link_create.kprobe_multi.syms = ptr_to_u64(OPTS_GET(opts, kprobe_multi.syms, 0));
711		attr.link_create.kprobe_multi.addrs = ptr_to_u64(OPTS_GET(opts, kprobe_multi.addrs, 0));
712		attr.link_create.kprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, kprobe_multi.cookies, 0));
713		if (!OPTS_ZEROED(opts, kprobe_multi))
714			return libbpf_err(-EINVAL);
715		break;
716	case BPF_TRACE_FENTRY:
717	case BPF_TRACE_FEXIT:
718	case BPF_MODIFY_RETURN:
719	case BPF_LSM_MAC:
720		attr.link_create.tracing.cookie = OPTS_GET(opts, tracing.cookie, 0);
721		if (!OPTS_ZEROED(opts, tracing))
722			return libbpf_err(-EINVAL);
723		break;
724	default:
725		if (!OPTS_ZEROED(opts, flags))
726			return libbpf_err(-EINVAL);
727		break;
728	}
729proceed:
730	fd = sys_bpf_fd(BPF_LINK_CREATE, &attr, sizeof(attr));
731	if (fd >= 0)
732		return fd;
733	/* we'll get EINVAL if LINK_CREATE doesn't support attaching fentry
734	 * and other similar programs
735	 */
736	err = -errno;
737	if (err != -EINVAL)
738		return libbpf_err(err);
739
740	/* if user used features not supported by
741	 * BPF_RAW_TRACEPOINT_OPEN command, then just give up immediately
742	 */
743	if (attr.link_create.target_fd || attr.link_create.target_btf_id)
744		return libbpf_err(err);
745	if (!OPTS_ZEROED(opts, sz))
746		return libbpf_err(err);
747
748	/* otherwise, for few select kinds of programs that can be
749	 * attached using BPF_RAW_TRACEPOINT_OPEN command, try that as
750	 * a fallback for older kernels
751	 */
752	switch (attach_type) {
753	case BPF_TRACE_RAW_TP:
754	case BPF_LSM_MAC:
755	case BPF_TRACE_FENTRY:
756	case BPF_TRACE_FEXIT:
757	case BPF_MODIFY_RETURN:
758		return bpf_raw_tracepoint_open(NULL, prog_fd);
759	default:
760		return libbpf_err(err);
761	}
762}
763
764int bpf_link_detach(int link_fd)
765{
766	union bpf_attr attr;
767	int ret;
768
769	memset(&attr, 0, sizeof(attr));
770	attr.link_detach.link_fd = link_fd;
771
772	ret = sys_bpf(BPF_LINK_DETACH, &attr, sizeof(attr));
773	return libbpf_err_errno(ret);
774}
775
776int bpf_link_update(int link_fd, int new_prog_fd,
777		    const struct bpf_link_update_opts *opts)
778{
779	union bpf_attr attr;
780	int ret;
781
782	if (!OPTS_VALID(opts, bpf_link_update_opts))
783		return libbpf_err(-EINVAL);
784
785	memset(&attr, 0, sizeof(attr));
786	attr.link_update.link_fd = link_fd;
787	attr.link_update.new_prog_fd = new_prog_fd;
788	attr.link_update.flags = OPTS_GET(opts, flags, 0);
789	attr.link_update.old_prog_fd = OPTS_GET(opts, old_prog_fd, 0);
790
791	ret = sys_bpf(BPF_LINK_UPDATE, &attr, sizeof(attr));
792	return libbpf_err_errno(ret);
793}
794
795int bpf_iter_create(int link_fd)
796{
797	union bpf_attr attr;
798	int fd;
799
800	memset(&attr, 0, sizeof(attr));
801	attr.iter_create.link_fd = link_fd;
802
803	fd = sys_bpf_fd(BPF_ITER_CREATE, &attr, sizeof(attr));
804	return libbpf_err_errno(fd);
805}
806
807int bpf_prog_query_opts(int target_fd,
808			enum bpf_attach_type type,
809			struct bpf_prog_query_opts *opts)
810{
811	union bpf_attr attr;
812	int ret;
813
814	if (!OPTS_VALID(opts, bpf_prog_query_opts))
815		return libbpf_err(-EINVAL);
816
817	memset(&attr, 0, sizeof(attr));
818
819	attr.query.target_fd	= target_fd;
820	attr.query.attach_type	= type;
821	attr.query.query_flags	= OPTS_GET(opts, query_flags, 0);
822	attr.query.prog_cnt	= OPTS_GET(opts, prog_cnt, 0);
823	attr.query.prog_ids	= ptr_to_u64(OPTS_GET(opts, prog_ids, NULL));
824	attr.query.prog_attach_flags = ptr_to_u64(OPTS_GET(opts, prog_attach_flags, NULL));
825
826	ret = sys_bpf(BPF_PROG_QUERY, &attr, sizeof(attr));
827
828	OPTS_SET(opts, attach_flags, attr.query.attach_flags);
829	OPTS_SET(opts, prog_cnt, attr.query.prog_cnt);
830
831	return libbpf_err_errno(ret);
832}
833
834int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags,
835		   __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt)
836{
837	LIBBPF_OPTS(bpf_prog_query_opts, opts);
838	int ret;
839
840	opts.query_flags = query_flags;
841	opts.prog_ids = prog_ids;
842	opts.prog_cnt = *prog_cnt;
843
844	ret = bpf_prog_query_opts(target_fd, type, &opts);
845
846	if (attach_flags)
847		*attach_flags = opts.attach_flags;
848	*prog_cnt = opts.prog_cnt;
849
850	return libbpf_err_errno(ret);
851}
852
853int bpf_prog_test_run_opts(int prog_fd, struct bpf_test_run_opts *opts)
854{
855	union bpf_attr attr;
856	int ret;
857
858	if (!OPTS_VALID(opts, bpf_test_run_opts))
859		return libbpf_err(-EINVAL);
860
861	memset(&attr, 0, sizeof(attr));
862	attr.test.prog_fd = prog_fd;
863	attr.test.batch_size = OPTS_GET(opts, batch_size, 0);
864	attr.test.cpu = OPTS_GET(opts, cpu, 0);
865	attr.test.flags = OPTS_GET(opts, flags, 0);
866	attr.test.repeat = OPTS_GET(opts, repeat, 0);
867	attr.test.duration = OPTS_GET(opts, duration, 0);
868	attr.test.ctx_size_in = OPTS_GET(opts, ctx_size_in, 0);
869	attr.test.ctx_size_out = OPTS_GET(opts, ctx_size_out, 0);
870	attr.test.data_size_in = OPTS_GET(opts, data_size_in, 0);
871	attr.test.data_size_out = OPTS_GET(opts, data_size_out, 0);
872	attr.test.ctx_in = ptr_to_u64(OPTS_GET(opts, ctx_in, NULL));
873	attr.test.ctx_out = ptr_to_u64(OPTS_GET(opts, ctx_out, NULL));
874	attr.test.data_in = ptr_to_u64(OPTS_GET(opts, data_in, NULL));
875	attr.test.data_out = ptr_to_u64(OPTS_GET(opts, data_out, NULL));
876
877	ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr));
878
879	OPTS_SET(opts, data_size_out, attr.test.data_size_out);
880	OPTS_SET(opts, ctx_size_out, attr.test.ctx_size_out);
881	OPTS_SET(opts, duration, attr.test.duration);
882	OPTS_SET(opts, retval, attr.test.retval);
883
884	return libbpf_err_errno(ret);
885}
886
887static int bpf_obj_get_next_id(__u32 start_id, __u32 *next_id, int cmd)
888{
889	union bpf_attr attr;
890	int err;
891
892	memset(&attr, 0, sizeof(attr));
893	attr.start_id = start_id;
894
895	err = sys_bpf(cmd, &attr, sizeof(attr));
896	if (!err)
897		*next_id = attr.next_id;
898
899	return libbpf_err_errno(err);
900}
901
902int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id)
903{
904	return bpf_obj_get_next_id(start_id, next_id, BPF_PROG_GET_NEXT_ID);
905}
906
907int bpf_map_get_next_id(__u32 start_id, __u32 *next_id)
908{
909	return bpf_obj_get_next_id(start_id, next_id, BPF_MAP_GET_NEXT_ID);
910}
911
912int bpf_btf_get_next_id(__u32 start_id, __u32 *next_id)
913{
914	return bpf_obj_get_next_id(start_id, next_id, BPF_BTF_GET_NEXT_ID);
915}
916
917int bpf_link_get_next_id(__u32 start_id, __u32 *next_id)
918{
919	return bpf_obj_get_next_id(start_id, next_id, BPF_LINK_GET_NEXT_ID);
920}
921
922int bpf_prog_get_fd_by_id(__u32 id)
923{
924	union bpf_attr attr;
925	int fd;
926
927	memset(&attr, 0, sizeof(attr));
928	attr.prog_id = id;
929
930	fd = sys_bpf_fd(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr));
931	return libbpf_err_errno(fd);
932}
933
934int bpf_map_get_fd_by_id(__u32 id)
935{
936	union bpf_attr attr;
937	int fd;
938
939	memset(&attr, 0, sizeof(attr));
940	attr.map_id = id;
941
942	fd = sys_bpf_fd(BPF_MAP_GET_FD_BY_ID, &attr, sizeof(attr));
943	return libbpf_err_errno(fd);
944}
945
946int bpf_btf_get_fd_by_id(__u32 id)
947{
948	union bpf_attr attr;
949	int fd;
950
951	memset(&attr, 0, sizeof(attr));
952	attr.btf_id = id;
953
954	fd = sys_bpf_fd(BPF_BTF_GET_FD_BY_ID, &attr, sizeof(attr));
955	return libbpf_err_errno(fd);
956}
957
958int bpf_link_get_fd_by_id(__u32 id)
959{
960	union bpf_attr attr;
961	int fd;
962
963	memset(&attr, 0, sizeof(attr));
964	attr.link_id = id;
965
966	fd = sys_bpf_fd(BPF_LINK_GET_FD_BY_ID, &attr, sizeof(attr));
967	return libbpf_err_errno(fd);
968}
969
970int bpf_obj_get_info_by_fd(int bpf_fd, void *info, __u32 *info_len)
971{
972	union bpf_attr attr;
973	int err;
974
975	memset(&attr, 0, sizeof(attr));
976	attr.info.bpf_fd = bpf_fd;
977	attr.info.info_len = *info_len;
978	attr.info.info = ptr_to_u64(info);
979
980	err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr));
981
982	if (!err)
983		*info_len = attr.info.info_len;
984
985	return libbpf_err_errno(err);
986}
987
988int bpf_raw_tracepoint_open(const char *name, int prog_fd)
989{
990	union bpf_attr attr;
991	int fd;
992
993	memset(&attr, 0, sizeof(attr));
994	attr.raw_tracepoint.name = ptr_to_u64(name);
995	attr.raw_tracepoint.prog_fd = prog_fd;
996
997	fd = sys_bpf_fd(BPF_RAW_TRACEPOINT_OPEN, &attr, sizeof(attr));
998	return libbpf_err_errno(fd);
999}
1000
1001int bpf_btf_load(const void *btf_data, size_t btf_size, const struct bpf_btf_load_opts *opts)
1002{
1003	const size_t attr_sz = offsetofend(union bpf_attr, btf_log_level);
1004	union bpf_attr attr;
1005	char *log_buf;
1006	size_t log_size;
1007	__u32 log_level;
1008	int fd;
1009
1010	bump_rlimit_memlock();
1011
1012	memset(&attr, 0, attr_sz);
1013
1014	if (!OPTS_VALID(opts, bpf_btf_load_opts))
1015		return libbpf_err(-EINVAL);
1016
1017	log_buf = OPTS_GET(opts, log_buf, NULL);
1018	log_size = OPTS_GET(opts, log_size, 0);
1019	log_level = OPTS_GET(opts, log_level, 0);
1020
1021	if (log_size > UINT_MAX)
1022		return libbpf_err(-EINVAL);
1023	if (log_size && !log_buf)
1024		return libbpf_err(-EINVAL);
1025
1026	attr.btf = ptr_to_u64(btf_data);
1027	attr.btf_size = btf_size;
1028	/* log_level == 0 and log_buf != NULL means "try loading without
1029	 * log_buf, but retry with log_buf and log_level=1 on error", which is
1030	 * consistent across low-level and high-level BTF and program loading
1031	 * APIs within libbpf and provides a sensible behavior in practice
1032	 */
1033	if (log_level) {
1034		attr.btf_log_buf = ptr_to_u64(log_buf);
1035		attr.btf_log_size = (__u32)log_size;
1036		attr.btf_log_level = log_level;
1037	}
1038
1039	fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1040	if (fd < 0 && log_buf && log_level == 0) {
1041		attr.btf_log_buf = ptr_to_u64(log_buf);
1042		attr.btf_log_size = (__u32)log_size;
1043		attr.btf_log_level = 1;
1044		fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1045	}
1046	return libbpf_err_errno(fd);
1047}
1048
1049int bpf_task_fd_query(int pid, int fd, __u32 flags, char *buf, __u32 *buf_len,
1050		      __u32 *prog_id, __u32 *fd_type, __u64 *probe_offset,
1051		      __u64 *probe_addr)
1052{
1053	union bpf_attr attr = {};
1054	int err;
1055
1056	attr.task_fd_query.pid = pid;
1057	attr.task_fd_query.fd = fd;
1058	attr.task_fd_query.flags = flags;
1059	attr.task_fd_query.buf = ptr_to_u64(buf);
1060	attr.task_fd_query.buf_len = *buf_len;
1061
1062	err = sys_bpf(BPF_TASK_FD_QUERY, &attr, sizeof(attr));
1063
1064	*buf_len = attr.task_fd_query.buf_len;
1065	*prog_id = attr.task_fd_query.prog_id;
1066	*fd_type = attr.task_fd_query.fd_type;
1067	*probe_offset = attr.task_fd_query.probe_offset;
1068	*probe_addr = attr.task_fd_query.probe_addr;
1069
1070	return libbpf_err_errno(err);
1071}
1072
1073int bpf_enable_stats(enum bpf_stats_type type)
1074{
1075	union bpf_attr attr;
1076	int fd;
1077
1078	memset(&attr, 0, sizeof(attr));
1079	attr.enable_stats.type = type;
1080
1081	fd = sys_bpf_fd(BPF_ENABLE_STATS, &attr, sizeof(attr));
1082	return libbpf_err_errno(fd);
1083}
1084
1085int bpf_prog_bind_map(int prog_fd, int map_fd,
1086		      const struct bpf_prog_bind_opts *opts)
1087{
1088	union bpf_attr attr;
1089	int ret;
1090
1091	if (!OPTS_VALID(opts, bpf_prog_bind_opts))
1092		return libbpf_err(-EINVAL);
1093
1094	memset(&attr, 0, sizeof(attr));
1095	attr.prog_bind_map.prog_fd = prog_fd;
1096	attr.prog_bind_map.map_fd = map_fd;
1097	attr.prog_bind_map.flags = OPTS_GET(opts, flags, 0);
1098
1099	ret = sys_bpf(BPF_PROG_BIND_MAP, &attr, sizeof(attr));
1100	return libbpf_err_errno(ret);
1101}
1102