Linux Audio

Check our new training course

Loading...
v6.9.4
   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(int token_fd)
 107{
 108	const size_t attr_sz = offsetofend(union bpf_attr, prog_token_fd);
 109	struct bpf_insn insns[] = {
 110		BPF_EMIT_CALL(BPF_FUNC_ktime_get_coarse_ns),
 111		BPF_EXIT_INSN(),
 112	};
 113	size_t insn_cnt = ARRAY_SIZE(insns);
 114	union bpf_attr attr;
 115	int prog_fd;
 116
 117	/* attempt loading freplace trying to use custom BTF */
 118	memset(&attr, 0, attr_sz);
 119	attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
 120	attr.insns = ptr_to_u64(insns);
 121	attr.insn_cnt = insn_cnt;
 122	attr.license = ptr_to_u64("GPL");
 123	attr.prog_token_fd = token_fd;
 124	if (token_fd)
 125		attr.prog_flags |= BPF_F_TOKEN_FD;
 126
 127	prog_fd = sys_bpf_fd(BPF_PROG_LOAD, &attr, attr_sz);
 128	if (prog_fd >= 0) {
 129		close(prog_fd);
 130		return 1;
 131	}
 132	return 0;
 133}
 134
 135static bool memlock_bumped;
 136static rlim_t memlock_rlim = RLIM_INFINITY;
 137
 138int libbpf_set_memlock_rlim(size_t memlock_bytes)
 139{
 140	if (memlock_bumped)
 141		return libbpf_err(-EBUSY);
 142
 143	memlock_rlim = memlock_bytes;
 144	return 0;
 145}
 146
 147int bump_rlimit_memlock(void)
 148{
 149	struct rlimit rlim;
 150
 151	/* if kernel supports memcg-based accounting, skip bumping RLIMIT_MEMLOCK */
 152	if (memlock_bumped || feat_supported(NULL, FEAT_MEMCG_ACCOUNT))
 153		return 0;
 154
 155	memlock_bumped = true;
 156
 157	/* zero memlock_rlim_max disables auto-bumping RLIMIT_MEMLOCK */
 158	if (memlock_rlim == 0)
 159		return 0;
 160
 161	rlim.rlim_cur = rlim.rlim_max = memlock_rlim;
 162	if (setrlimit(RLIMIT_MEMLOCK, &rlim))
 163		return -errno;
 164
 165	return 0;
 166}
 167
 168int bpf_map_create(enum bpf_map_type map_type,
 169		   const char *map_name,
 170		   __u32 key_size,
 171		   __u32 value_size,
 172		   __u32 max_entries,
 173		   const struct bpf_map_create_opts *opts)
 174{
 175	const size_t attr_sz = offsetofend(union bpf_attr, map_token_fd);
 176	union bpf_attr attr;
 177	int fd;
 178
 179	bump_rlimit_memlock();
 180
 181	memset(&attr, 0, attr_sz);
 182
 183	if (!OPTS_VALID(opts, bpf_map_create_opts))
 184		return libbpf_err(-EINVAL);
 185
 186	attr.map_type = map_type;
 187	if (map_name && feat_supported(NULL, FEAT_PROG_NAME))
 188		libbpf_strlcpy(attr.map_name, map_name, sizeof(attr.map_name));
 189	attr.key_size = key_size;
 190	attr.value_size = value_size;
 191	attr.max_entries = max_entries;
 192
 193	attr.btf_fd = OPTS_GET(opts, btf_fd, 0);
 194	attr.btf_key_type_id = OPTS_GET(opts, btf_key_type_id, 0);
 195	attr.btf_value_type_id = OPTS_GET(opts, btf_value_type_id, 0);
 196	attr.btf_vmlinux_value_type_id = OPTS_GET(opts, btf_vmlinux_value_type_id, 0);
 197	attr.value_type_btf_obj_fd = OPTS_GET(opts, value_type_btf_obj_fd, 0);
 198
 199	attr.inner_map_fd = OPTS_GET(opts, inner_map_fd, 0);
 200	attr.map_flags = OPTS_GET(opts, map_flags, 0);
 201	attr.map_extra = OPTS_GET(opts, map_extra, 0);
 202	attr.numa_node = OPTS_GET(opts, numa_node, 0);
 203	attr.map_ifindex = OPTS_GET(opts, map_ifindex, 0);
 204
 205	attr.map_token_fd = OPTS_GET(opts, token_fd, 0);
 206
 207	fd = sys_bpf_fd(BPF_MAP_CREATE, &attr, attr_sz);
 208	return libbpf_err_errno(fd);
 209}
 210
 211static void *
 212alloc_zero_tailing_info(const void *orecord, __u32 cnt,
 213			__u32 actual_rec_size, __u32 expected_rec_size)
 214{
 215	__u64 info_len = (__u64)actual_rec_size * cnt;
 216	void *info, *nrecord;
 217	int i;
 218
 219	info = malloc(info_len);
 220	if (!info)
 221		return NULL;
 222
 223	/* zero out bytes kernel does not understand */
 224	nrecord = info;
 225	for (i = 0; i < cnt; i++) {
 226		memcpy(nrecord, orecord, expected_rec_size);
 227		memset(nrecord + expected_rec_size, 0,
 228		       actual_rec_size - expected_rec_size);
 229		orecord += actual_rec_size;
 230		nrecord += actual_rec_size;
 231	}
 232
 233	return info;
 234}
 235
 236int bpf_prog_load(enum bpf_prog_type prog_type,
 237		  const char *prog_name, const char *license,
 238		  const struct bpf_insn *insns, size_t insn_cnt,
 239		  struct bpf_prog_load_opts *opts)
 240{
 241	const size_t attr_sz = offsetofend(union bpf_attr, prog_token_fd);
 242	void *finfo = NULL, *linfo = NULL;
 243	const char *func_info, *line_info;
 244	__u32 log_size, log_level, attach_prog_fd, attach_btf_obj_fd;
 245	__u32 func_info_rec_size, line_info_rec_size;
 246	int fd, attempts;
 247	union bpf_attr attr;
 248	char *log_buf;
 249
 250	bump_rlimit_memlock();
 251
 252	if (!OPTS_VALID(opts, bpf_prog_load_opts))
 253		return libbpf_err(-EINVAL);
 254
 255	attempts = OPTS_GET(opts, attempts, 0);
 256	if (attempts < 0)
 257		return libbpf_err(-EINVAL);
 258	if (attempts == 0)
 259		attempts = PROG_LOAD_ATTEMPTS;
 260
 261	memset(&attr, 0, attr_sz);
 262
 263	attr.prog_type = prog_type;
 264	attr.expected_attach_type = OPTS_GET(opts, expected_attach_type, 0);
 265
 266	attr.prog_btf_fd = OPTS_GET(opts, prog_btf_fd, 0);
 267	attr.prog_flags = OPTS_GET(opts, prog_flags, 0);
 268	attr.prog_ifindex = OPTS_GET(opts, prog_ifindex, 0);
 269	attr.kern_version = OPTS_GET(opts, kern_version, 0);
 270	attr.prog_token_fd = OPTS_GET(opts, token_fd, 0);
 271
 272	if (prog_name && feat_supported(NULL, FEAT_PROG_NAME))
 273		libbpf_strlcpy(attr.prog_name, prog_name, sizeof(attr.prog_name));
 274	attr.license = ptr_to_u64(license);
 275
 276	if (insn_cnt > UINT_MAX)
 277		return libbpf_err(-E2BIG);
 278
 279	attr.insns = ptr_to_u64(insns);
 280	attr.insn_cnt = (__u32)insn_cnt;
 281
 282	attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
 283	attach_btf_obj_fd = OPTS_GET(opts, attach_btf_obj_fd, 0);
 284
 285	if (attach_prog_fd && attach_btf_obj_fd)
 286		return libbpf_err(-EINVAL);
 287
 288	attr.attach_btf_id = OPTS_GET(opts, attach_btf_id, 0);
 289	if (attach_prog_fd)
 290		attr.attach_prog_fd = attach_prog_fd;
 291	else
 292		attr.attach_btf_obj_fd = attach_btf_obj_fd;
 293
 294	log_buf = OPTS_GET(opts, log_buf, NULL);
 295	log_size = OPTS_GET(opts, log_size, 0);
 296	log_level = OPTS_GET(opts, log_level, 0);
 297
 298	if (!!log_buf != !!log_size)
 299		return libbpf_err(-EINVAL);
 300
 301	func_info_rec_size = OPTS_GET(opts, func_info_rec_size, 0);
 302	func_info = OPTS_GET(opts, func_info, NULL);
 303	attr.func_info_rec_size = func_info_rec_size;
 304	attr.func_info = ptr_to_u64(func_info);
 305	attr.func_info_cnt = OPTS_GET(opts, func_info_cnt, 0);
 306
 307	line_info_rec_size = OPTS_GET(opts, line_info_rec_size, 0);
 308	line_info = OPTS_GET(opts, line_info, NULL);
 309	attr.line_info_rec_size = line_info_rec_size;
 310	attr.line_info = ptr_to_u64(line_info);
 311	attr.line_info_cnt = OPTS_GET(opts, line_info_cnt, 0);
 312
 313	attr.fd_array = ptr_to_u64(OPTS_GET(opts, fd_array, NULL));
 314
 315	if (log_level) {
 316		attr.log_buf = ptr_to_u64(log_buf);
 317		attr.log_size = log_size;
 318		attr.log_level = log_level;
 319	}
 320
 321	fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
 322	OPTS_SET(opts, log_true_size, attr.log_true_size);
 323	if (fd >= 0)
 324		return fd;
 325
 326	/* After bpf_prog_load, the kernel may modify certain attributes
 327	 * to give user space a hint how to deal with loading failure.
 328	 * Check to see whether we can make some changes and load again.
 329	 */
 330	while (errno == E2BIG && (!finfo || !linfo)) {
 331		if (!finfo && attr.func_info_cnt &&
 332		    attr.func_info_rec_size < func_info_rec_size) {
 333			/* try with corrected func info records */
 334			finfo = alloc_zero_tailing_info(func_info,
 335							attr.func_info_cnt,
 336							func_info_rec_size,
 337							attr.func_info_rec_size);
 338			if (!finfo) {
 339				errno = E2BIG;
 340				goto done;
 341			}
 342
 343			attr.func_info = ptr_to_u64(finfo);
 344			attr.func_info_rec_size = func_info_rec_size;
 345		} else if (!linfo && attr.line_info_cnt &&
 346			   attr.line_info_rec_size < line_info_rec_size) {
 347			linfo = alloc_zero_tailing_info(line_info,
 348							attr.line_info_cnt,
 349							line_info_rec_size,
 350							attr.line_info_rec_size);
 351			if (!linfo) {
 352				errno = E2BIG;
 353				goto done;
 354			}
 355
 356			attr.line_info = ptr_to_u64(linfo);
 357			attr.line_info_rec_size = line_info_rec_size;
 358		} else {
 359			break;
 360		}
 361
 362		fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
 363		OPTS_SET(opts, log_true_size, attr.log_true_size);
 364		if (fd >= 0)
 365			goto done;
 366	}
 367
 368	if (log_level == 0 && log_buf) {
 369		/* log_level == 0 with non-NULL log_buf requires retrying on error
 370		 * with log_level == 1 and log_buf/log_buf_size set, to get details of
 371		 * failure
 372		 */
 373		attr.log_buf = ptr_to_u64(log_buf);
 374		attr.log_size = log_size;
 375		attr.log_level = 1;
 376
 377		fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
 378		OPTS_SET(opts, log_true_size, attr.log_true_size);
 379	}
 380done:
 381	/* free() doesn't affect errno, so we don't need to restore it */
 382	free(finfo);
 383	free(linfo);
 384	return libbpf_err_errno(fd);
 385}
 386
 387int bpf_map_update_elem(int fd, const void *key, const void *value,
 388			__u64 flags)
 389{
 390	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 391	union bpf_attr attr;
 392	int ret;
 393
 394	memset(&attr, 0, attr_sz);
 395	attr.map_fd = fd;
 396	attr.key = ptr_to_u64(key);
 397	attr.value = ptr_to_u64(value);
 398	attr.flags = flags;
 399
 400	ret = sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, attr_sz);
 401	return libbpf_err_errno(ret);
 402}
 403
 404int bpf_map_lookup_elem(int fd, const void *key, void *value)
 405{
 406	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 407	union bpf_attr attr;
 408	int ret;
 409
 410	memset(&attr, 0, attr_sz);
 411	attr.map_fd = fd;
 412	attr.key = ptr_to_u64(key);
 413	attr.value = ptr_to_u64(value);
 414
 415	ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, attr_sz);
 416	return libbpf_err_errno(ret);
 417}
 418
 419int bpf_map_lookup_elem_flags(int fd, const void *key, void *value, __u64 flags)
 420{
 421	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 422	union bpf_attr attr;
 423	int ret;
 424
 425	memset(&attr, 0, attr_sz);
 426	attr.map_fd = fd;
 427	attr.key = ptr_to_u64(key);
 428	attr.value = ptr_to_u64(value);
 429	attr.flags = flags;
 430
 431	ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, attr_sz);
 432	return libbpf_err_errno(ret);
 433}
 434
 435int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value)
 436{
 437	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 438	union bpf_attr attr;
 439	int ret;
 440
 441	memset(&attr, 0, attr_sz);
 442	attr.map_fd = fd;
 443	attr.key = ptr_to_u64(key);
 444	attr.value = ptr_to_u64(value);
 445
 446	ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, attr_sz);
 447	return libbpf_err_errno(ret);
 448}
 449
 450int bpf_map_lookup_and_delete_elem_flags(int fd, const void *key, void *value, __u64 flags)
 451{
 452	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 453	union bpf_attr attr;
 454	int ret;
 455
 456	memset(&attr, 0, attr_sz);
 457	attr.map_fd = fd;
 458	attr.key = ptr_to_u64(key);
 459	attr.value = ptr_to_u64(value);
 460	attr.flags = flags;
 461
 462	ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, attr_sz);
 463	return libbpf_err_errno(ret);
 464}
 465
 466int bpf_map_delete_elem(int fd, const void *key)
 467{
 468	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 469	union bpf_attr attr;
 470	int ret;
 471
 472	memset(&attr, 0, attr_sz);
 473	attr.map_fd = fd;
 474	attr.key = ptr_to_u64(key);
 475
 476	ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, attr_sz);
 477	return libbpf_err_errno(ret);
 478}
 479
 480int bpf_map_delete_elem_flags(int fd, const void *key, __u64 flags)
 481{
 482	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 483	union bpf_attr attr;
 484	int ret;
 485
 486	memset(&attr, 0, attr_sz);
 487	attr.map_fd = fd;
 488	attr.key = ptr_to_u64(key);
 489	attr.flags = flags;
 490
 491	ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, attr_sz);
 492	return libbpf_err_errno(ret);
 493}
 494
 495int bpf_map_get_next_key(int fd, const void *key, void *next_key)
 496{
 497	const size_t attr_sz = offsetofend(union bpf_attr, next_key);
 498	union bpf_attr attr;
 499	int ret;
 500
 501	memset(&attr, 0, attr_sz);
 502	attr.map_fd = fd;
 503	attr.key = ptr_to_u64(key);
 504	attr.next_key = ptr_to_u64(next_key);
 505
 506	ret = sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, attr_sz);
 507	return libbpf_err_errno(ret);
 508}
 509
 510int bpf_map_freeze(int fd)
 511{
 512	const size_t attr_sz = offsetofend(union bpf_attr, map_fd);
 513	union bpf_attr attr;
 514	int ret;
 515
 516	memset(&attr, 0, attr_sz);
 517	attr.map_fd = fd;
 518
 519	ret = sys_bpf(BPF_MAP_FREEZE, &attr, attr_sz);
 520	return libbpf_err_errno(ret);
 521}
 522
 523static int bpf_map_batch_common(int cmd, int fd, void  *in_batch,
 524				void *out_batch, void *keys, void *values,
 525				__u32 *count,
 526				const struct bpf_map_batch_opts *opts)
 527{
 528	const size_t attr_sz = offsetofend(union bpf_attr, batch);
 529	union bpf_attr attr;
 530	int ret;
 531
 532	if (!OPTS_VALID(opts, bpf_map_batch_opts))
 533		return libbpf_err(-EINVAL);
 534
 535	memset(&attr, 0, attr_sz);
 536	attr.batch.map_fd = fd;
 537	attr.batch.in_batch = ptr_to_u64(in_batch);
 538	attr.batch.out_batch = ptr_to_u64(out_batch);
 539	attr.batch.keys = ptr_to_u64(keys);
 540	attr.batch.values = ptr_to_u64(values);
 541	attr.batch.count = *count;
 542	attr.batch.elem_flags  = OPTS_GET(opts, elem_flags, 0);
 543	attr.batch.flags = OPTS_GET(opts, flags, 0);
 544
 545	ret = sys_bpf(cmd, &attr, attr_sz);
 546	*count = attr.batch.count;
 547
 548	return libbpf_err_errno(ret);
 549}
 550
 551int bpf_map_delete_batch(int fd, const void *keys, __u32 *count,
 552			 const struct bpf_map_batch_opts *opts)
 553{
 554	return bpf_map_batch_common(BPF_MAP_DELETE_BATCH, fd, NULL,
 555				    NULL, (void *)keys, NULL, count, opts);
 556}
 557
 558int bpf_map_lookup_batch(int fd, void *in_batch, void *out_batch, void *keys,
 559			 void *values, __u32 *count,
 560			 const struct bpf_map_batch_opts *opts)
 561{
 562	return bpf_map_batch_common(BPF_MAP_LOOKUP_BATCH, fd, in_batch,
 563				    out_batch, keys, values, count, opts);
 564}
 565
 566int bpf_map_lookup_and_delete_batch(int fd, void *in_batch, void *out_batch,
 567				    void *keys, void *values, __u32 *count,
 568				    const struct bpf_map_batch_opts *opts)
 569{
 570	return bpf_map_batch_common(BPF_MAP_LOOKUP_AND_DELETE_BATCH,
 571				    fd, in_batch, out_batch, keys, values,
 572				    count, opts);
 573}
 574
 575int bpf_map_update_batch(int fd, const void *keys, const void *values, __u32 *count,
 576			 const struct bpf_map_batch_opts *opts)
 577{
 578	return bpf_map_batch_common(BPF_MAP_UPDATE_BATCH, fd, NULL, NULL,
 579				    (void *)keys, (void *)values, count, opts);
 580}
 581
 582int bpf_obj_pin_opts(int fd, const char *pathname, const struct bpf_obj_pin_opts *opts)
 583{
 584	const size_t attr_sz = offsetofend(union bpf_attr, path_fd);
 585	union bpf_attr attr;
 586	int ret;
 587
 588	if (!OPTS_VALID(opts, bpf_obj_pin_opts))
 589		return libbpf_err(-EINVAL);
 590
 591	memset(&attr, 0, attr_sz);
 592	attr.path_fd = OPTS_GET(opts, path_fd, 0);
 593	attr.pathname = ptr_to_u64((void *)pathname);
 594	attr.file_flags = OPTS_GET(opts, file_flags, 0);
 595	attr.bpf_fd = fd;
 596
 597	ret = sys_bpf(BPF_OBJ_PIN, &attr, attr_sz);
 598	return libbpf_err_errno(ret);
 599}
 600
 601int bpf_obj_pin(int fd, const char *pathname)
 602{
 603	return bpf_obj_pin_opts(fd, pathname, NULL);
 604}
 605
 606int bpf_obj_get(const char *pathname)
 607{
 608	return bpf_obj_get_opts(pathname, NULL);
 609}
 610
 611int bpf_obj_get_opts(const char *pathname, const struct bpf_obj_get_opts *opts)
 612{
 613	const size_t attr_sz = offsetofend(union bpf_attr, path_fd);
 614	union bpf_attr attr;
 615	int fd;
 616
 617	if (!OPTS_VALID(opts, bpf_obj_get_opts))
 618		return libbpf_err(-EINVAL);
 619
 620	memset(&attr, 0, attr_sz);
 621	attr.path_fd = OPTS_GET(opts, path_fd, 0);
 622	attr.pathname = ptr_to_u64((void *)pathname);
 623	attr.file_flags = OPTS_GET(opts, file_flags, 0);
 624
 625	fd = sys_bpf_fd(BPF_OBJ_GET, &attr, attr_sz);
 626	return libbpf_err_errno(fd);
 627}
 628
 629int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type,
 630		    unsigned int flags)
 631{
 632	DECLARE_LIBBPF_OPTS(bpf_prog_attach_opts, opts,
 633		.flags = flags,
 634	);
 635
 636	return bpf_prog_attach_opts(prog_fd, target_fd, type, &opts);
 637}
 638
 639int bpf_prog_attach_opts(int prog_fd, int target, enum bpf_attach_type type,
 640			 const struct bpf_prog_attach_opts *opts)
 641{
 642	const size_t attr_sz = offsetofend(union bpf_attr, expected_revision);
 643	__u32 relative_id, flags;
 644	int ret, relative_fd;
 645	union bpf_attr attr;
 646
 647	if (!OPTS_VALID(opts, bpf_prog_attach_opts))
 648		return libbpf_err(-EINVAL);
 649
 650	relative_id = OPTS_GET(opts, relative_id, 0);
 651	relative_fd = OPTS_GET(opts, relative_fd, 0);
 652	flags = OPTS_GET(opts, flags, 0);
 653
 654	/* validate we don't have unexpected combinations of non-zero fields */
 655	if (relative_fd && relative_id)
 656		return libbpf_err(-EINVAL);
 657
 658	memset(&attr, 0, attr_sz);
 659	attr.target_fd		= target;
 660	attr.attach_bpf_fd	= prog_fd;
 661	attr.attach_type	= type;
 662	attr.replace_bpf_fd	= OPTS_GET(opts, replace_fd, 0);
 663	attr.expected_revision	= OPTS_GET(opts, expected_revision, 0);
 664
 665	if (relative_id) {
 666		attr.attach_flags = flags | BPF_F_ID;
 667		attr.relative_id  = relative_id;
 668	} else {
 669		attr.attach_flags = flags;
 670		attr.relative_fd  = relative_fd;
 671	}
 672
 673	ret = sys_bpf(BPF_PROG_ATTACH, &attr, attr_sz);
 674	return libbpf_err_errno(ret);
 675}
 676
 677int bpf_prog_detach_opts(int prog_fd, int target, enum bpf_attach_type type,
 678			 const struct bpf_prog_detach_opts *opts)
 679{
 680	const size_t attr_sz = offsetofend(union bpf_attr, expected_revision);
 681	__u32 relative_id, flags;
 682	int ret, relative_fd;
 683	union bpf_attr attr;
 684
 685	if (!OPTS_VALID(opts, bpf_prog_detach_opts))
 686		return libbpf_err(-EINVAL);
 687
 688	relative_id = OPTS_GET(opts, relative_id, 0);
 689	relative_fd = OPTS_GET(opts, relative_fd, 0);
 690	flags = OPTS_GET(opts, flags, 0);
 691
 692	/* validate we don't have unexpected combinations of non-zero fields */
 693	if (relative_fd && relative_id)
 694		return libbpf_err(-EINVAL);
 695
 696	memset(&attr, 0, attr_sz);
 697	attr.target_fd		= target;
 698	attr.attach_bpf_fd	= prog_fd;
 699	attr.attach_type	= type;
 700	attr.expected_revision	= OPTS_GET(opts, expected_revision, 0);
 701
 702	if (relative_id) {
 703		attr.attach_flags = flags | BPF_F_ID;
 704		attr.relative_id  = relative_id;
 705	} else {
 706		attr.attach_flags = flags;
 707		attr.relative_fd  = relative_fd;
 708	}
 709
 710	ret = sys_bpf(BPF_PROG_DETACH, &attr, attr_sz);
 711	return libbpf_err_errno(ret);
 712}
 713
 714int bpf_prog_detach(int target_fd, enum bpf_attach_type type)
 715{
 716	return bpf_prog_detach_opts(0, target_fd, type, NULL);
 717}
 718
 719int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type)
 720{
 721	return bpf_prog_detach_opts(prog_fd, target_fd, type, NULL);
 722}
 723
 724int bpf_link_create(int prog_fd, int target_fd,
 725		    enum bpf_attach_type attach_type,
 726		    const struct bpf_link_create_opts *opts)
 727{
 728	const size_t attr_sz = offsetofend(union bpf_attr, link_create);
 729	__u32 target_btf_id, iter_info_len, relative_id;
 730	int fd, err, relative_fd;
 731	union bpf_attr attr;
 732
 733	if (!OPTS_VALID(opts, bpf_link_create_opts))
 734		return libbpf_err(-EINVAL);
 735
 736	iter_info_len = OPTS_GET(opts, iter_info_len, 0);
 737	target_btf_id = OPTS_GET(opts, target_btf_id, 0);
 738
 739	/* validate we don't have unexpected combinations of non-zero fields */
 740	if (iter_info_len || target_btf_id) {
 741		if (iter_info_len && target_btf_id)
 742			return libbpf_err(-EINVAL);
 743		if (!OPTS_ZEROED(opts, target_btf_id))
 744			return libbpf_err(-EINVAL);
 745	}
 746
 747	memset(&attr, 0, attr_sz);
 748	attr.link_create.prog_fd = prog_fd;
 749	attr.link_create.target_fd = target_fd;
 750	attr.link_create.attach_type = attach_type;
 751	attr.link_create.flags = OPTS_GET(opts, flags, 0);
 752
 753	if (target_btf_id) {
 754		attr.link_create.target_btf_id = target_btf_id;
 755		goto proceed;
 756	}
 757
 758	switch (attach_type) {
 759	case BPF_TRACE_ITER:
 760		attr.link_create.iter_info = ptr_to_u64(OPTS_GET(opts, iter_info, (void *)0));
 761		attr.link_create.iter_info_len = iter_info_len;
 762		break;
 763	case BPF_PERF_EVENT:
 764		attr.link_create.perf_event.bpf_cookie = OPTS_GET(opts, perf_event.bpf_cookie, 0);
 765		if (!OPTS_ZEROED(opts, perf_event))
 766			return libbpf_err(-EINVAL);
 767		break;
 768	case BPF_TRACE_KPROBE_MULTI:
 
 769		attr.link_create.kprobe_multi.flags = OPTS_GET(opts, kprobe_multi.flags, 0);
 770		attr.link_create.kprobe_multi.cnt = OPTS_GET(opts, kprobe_multi.cnt, 0);
 771		attr.link_create.kprobe_multi.syms = ptr_to_u64(OPTS_GET(opts, kprobe_multi.syms, 0));
 772		attr.link_create.kprobe_multi.addrs = ptr_to_u64(OPTS_GET(opts, kprobe_multi.addrs, 0));
 773		attr.link_create.kprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, kprobe_multi.cookies, 0));
 774		if (!OPTS_ZEROED(opts, kprobe_multi))
 775			return libbpf_err(-EINVAL);
 776		break;
 777	case BPF_TRACE_UPROBE_MULTI:
 
 778		attr.link_create.uprobe_multi.flags = OPTS_GET(opts, uprobe_multi.flags, 0);
 779		attr.link_create.uprobe_multi.cnt = OPTS_GET(opts, uprobe_multi.cnt, 0);
 780		attr.link_create.uprobe_multi.path = ptr_to_u64(OPTS_GET(opts, uprobe_multi.path, 0));
 781		attr.link_create.uprobe_multi.offsets = ptr_to_u64(OPTS_GET(opts, uprobe_multi.offsets, 0));
 782		attr.link_create.uprobe_multi.ref_ctr_offsets = ptr_to_u64(OPTS_GET(opts, uprobe_multi.ref_ctr_offsets, 0));
 783		attr.link_create.uprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, uprobe_multi.cookies, 0));
 784		attr.link_create.uprobe_multi.pid = OPTS_GET(opts, uprobe_multi.pid, 0);
 785		if (!OPTS_ZEROED(opts, uprobe_multi))
 786			return libbpf_err(-EINVAL);
 787		break;
 
 788	case BPF_TRACE_FENTRY:
 789	case BPF_TRACE_FEXIT:
 790	case BPF_MODIFY_RETURN:
 791	case BPF_LSM_MAC:
 792		attr.link_create.tracing.cookie = OPTS_GET(opts, tracing.cookie, 0);
 793		if (!OPTS_ZEROED(opts, tracing))
 794			return libbpf_err(-EINVAL);
 795		break;
 796	case BPF_NETFILTER:
 797		attr.link_create.netfilter.pf = OPTS_GET(opts, netfilter.pf, 0);
 798		attr.link_create.netfilter.hooknum = OPTS_GET(opts, netfilter.hooknum, 0);
 799		attr.link_create.netfilter.priority = OPTS_GET(opts, netfilter.priority, 0);
 800		attr.link_create.netfilter.flags = OPTS_GET(opts, netfilter.flags, 0);
 801		if (!OPTS_ZEROED(opts, netfilter))
 802			return libbpf_err(-EINVAL);
 803		break;
 804	case BPF_TCX_INGRESS:
 805	case BPF_TCX_EGRESS:
 806		relative_fd = OPTS_GET(opts, tcx.relative_fd, 0);
 807		relative_id = OPTS_GET(opts, tcx.relative_id, 0);
 808		if (relative_fd && relative_id)
 809			return libbpf_err(-EINVAL);
 810		if (relative_id) {
 811			attr.link_create.tcx.relative_id = relative_id;
 812			attr.link_create.flags |= BPF_F_ID;
 813		} else {
 814			attr.link_create.tcx.relative_fd = relative_fd;
 815		}
 816		attr.link_create.tcx.expected_revision = OPTS_GET(opts, tcx.expected_revision, 0);
 817		if (!OPTS_ZEROED(opts, tcx))
 818			return libbpf_err(-EINVAL);
 819		break;
 820	case BPF_NETKIT_PRIMARY:
 821	case BPF_NETKIT_PEER:
 822		relative_fd = OPTS_GET(opts, netkit.relative_fd, 0);
 823		relative_id = OPTS_GET(opts, netkit.relative_id, 0);
 824		if (relative_fd && relative_id)
 825			return libbpf_err(-EINVAL);
 826		if (relative_id) {
 827			attr.link_create.netkit.relative_id = relative_id;
 828			attr.link_create.flags |= BPF_F_ID;
 829		} else {
 830			attr.link_create.netkit.relative_fd = relative_fd;
 831		}
 832		attr.link_create.netkit.expected_revision = OPTS_GET(opts, netkit.expected_revision, 0);
 833		if (!OPTS_ZEROED(opts, netkit))
 834			return libbpf_err(-EINVAL);
 835		break;
 836	default:
 837		if (!OPTS_ZEROED(opts, flags))
 838			return libbpf_err(-EINVAL);
 839		break;
 840	}
 841proceed:
 842	fd = sys_bpf_fd(BPF_LINK_CREATE, &attr, attr_sz);
 843	if (fd >= 0)
 844		return fd;
 845	/* we'll get EINVAL if LINK_CREATE doesn't support attaching fentry
 846	 * and other similar programs
 847	 */
 848	err = -errno;
 849	if (err != -EINVAL)
 850		return libbpf_err(err);
 851
 852	/* if user used features not supported by
 853	 * BPF_RAW_TRACEPOINT_OPEN command, then just give up immediately
 854	 */
 855	if (attr.link_create.target_fd || attr.link_create.target_btf_id)
 856		return libbpf_err(err);
 857	if (!OPTS_ZEROED(opts, sz))
 858		return libbpf_err(err);
 859
 860	/* otherwise, for few select kinds of programs that can be
 861	 * attached using BPF_RAW_TRACEPOINT_OPEN command, try that as
 862	 * a fallback for older kernels
 863	 */
 864	switch (attach_type) {
 865	case BPF_TRACE_RAW_TP:
 866	case BPF_LSM_MAC:
 867	case BPF_TRACE_FENTRY:
 868	case BPF_TRACE_FEXIT:
 869	case BPF_MODIFY_RETURN:
 870		return bpf_raw_tracepoint_open(NULL, prog_fd);
 871	default:
 872		return libbpf_err(err);
 873	}
 874}
 875
 876int bpf_link_detach(int link_fd)
 877{
 878	const size_t attr_sz = offsetofend(union bpf_attr, link_detach);
 879	union bpf_attr attr;
 880	int ret;
 881
 882	memset(&attr, 0, attr_sz);
 883	attr.link_detach.link_fd = link_fd;
 884
 885	ret = sys_bpf(BPF_LINK_DETACH, &attr, attr_sz);
 886	return libbpf_err_errno(ret);
 887}
 888
 889int bpf_link_update(int link_fd, int new_prog_fd,
 890		    const struct bpf_link_update_opts *opts)
 891{
 892	const size_t attr_sz = offsetofend(union bpf_attr, link_update);
 893	union bpf_attr attr;
 894	int ret;
 895
 896	if (!OPTS_VALID(opts, bpf_link_update_opts))
 897		return libbpf_err(-EINVAL);
 898
 899	if (OPTS_GET(opts, old_prog_fd, 0) && OPTS_GET(opts, old_map_fd, 0))
 900		return libbpf_err(-EINVAL);
 901
 902	memset(&attr, 0, attr_sz);
 903	attr.link_update.link_fd = link_fd;
 904	attr.link_update.new_prog_fd = new_prog_fd;
 905	attr.link_update.flags = OPTS_GET(opts, flags, 0);
 906	if (OPTS_GET(opts, old_prog_fd, 0))
 907		attr.link_update.old_prog_fd = OPTS_GET(opts, old_prog_fd, 0);
 908	else if (OPTS_GET(opts, old_map_fd, 0))
 909		attr.link_update.old_map_fd = OPTS_GET(opts, old_map_fd, 0);
 910
 911	ret = sys_bpf(BPF_LINK_UPDATE, &attr, attr_sz);
 912	return libbpf_err_errno(ret);
 913}
 914
 915int bpf_iter_create(int link_fd)
 916{
 917	const size_t attr_sz = offsetofend(union bpf_attr, iter_create);
 918	union bpf_attr attr;
 919	int fd;
 920
 921	memset(&attr, 0, attr_sz);
 922	attr.iter_create.link_fd = link_fd;
 923
 924	fd = sys_bpf_fd(BPF_ITER_CREATE, &attr, attr_sz);
 925	return libbpf_err_errno(fd);
 926}
 927
 928int bpf_prog_query_opts(int target, enum bpf_attach_type type,
 929			struct bpf_prog_query_opts *opts)
 930{
 931	const size_t attr_sz = offsetofend(union bpf_attr, query);
 932	union bpf_attr attr;
 933	int ret;
 934
 935	if (!OPTS_VALID(opts, bpf_prog_query_opts))
 936		return libbpf_err(-EINVAL);
 937
 938	memset(&attr, 0, attr_sz);
 939	attr.query.target_fd		= target;
 940	attr.query.attach_type		= type;
 941	attr.query.query_flags		= OPTS_GET(opts, query_flags, 0);
 942	attr.query.count		= OPTS_GET(opts, count, 0);
 943	attr.query.prog_ids		= ptr_to_u64(OPTS_GET(opts, prog_ids, NULL));
 944	attr.query.link_ids		= ptr_to_u64(OPTS_GET(opts, link_ids, NULL));
 945	attr.query.prog_attach_flags	= ptr_to_u64(OPTS_GET(opts, prog_attach_flags, NULL));
 946	attr.query.link_attach_flags	= ptr_to_u64(OPTS_GET(opts, link_attach_flags, NULL));
 947
 948	ret = sys_bpf(BPF_PROG_QUERY, &attr, attr_sz);
 949
 950	OPTS_SET(opts, attach_flags, attr.query.attach_flags);
 951	OPTS_SET(opts, revision, attr.query.revision);
 952	OPTS_SET(opts, count, attr.query.count);
 953
 954	return libbpf_err_errno(ret);
 955}
 956
 957int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags,
 958		   __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt)
 959{
 960	LIBBPF_OPTS(bpf_prog_query_opts, opts);
 961	int ret;
 962
 963	opts.query_flags = query_flags;
 964	opts.prog_ids = prog_ids;
 965	opts.prog_cnt = *prog_cnt;
 966
 967	ret = bpf_prog_query_opts(target_fd, type, &opts);
 968
 969	if (attach_flags)
 970		*attach_flags = opts.attach_flags;
 971	*prog_cnt = opts.prog_cnt;
 972
 973	return libbpf_err_errno(ret);
 974}
 975
 976int bpf_prog_test_run_opts(int prog_fd, struct bpf_test_run_opts *opts)
 977{
 978	const size_t attr_sz = offsetofend(union bpf_attr, test);
 979	union bpf_attr attr;
 980	int ret;
 981
 982	if (!OPTS_VALID(opts, bpf_test_run_opts))
 983		return libbpf_err(-EINVAL);
 984
 985	memset(&attr, 0, attr_sz);
 986	attr.test.prog_fd = prog_fd;
 987	attr.test.batch_size = OPTS_GET(opts, batch_size, 0);
 988	attr.test.cpu = OPTS_GET(opts, cpu, 0);
 989	attr.test.flags = OPTS_GET(opts, flags, 0);
 990	attr.test.repeat = OPTS_GET(opts, repeat, 0);
 991	attr.test.duration = OPTS_GET(opts, duration, 0);
 992	attr.test.ctx_size_in = OPTS_GET(opts, ctx_size_in, 0);
 993	attr.test.ctx_size_out = OPTS_GET(opts, ctx_size_out, 0);
 994	attr.test.data_size_in = OPTS_GET(opts, data_size_in, 0);
 995	attr.test.data_size_out = OPTS_GET(opts, data_size_out, 0);
 996	attr.test.ctx_in = ptr_to_u64(OPTS_GET(opts, ctx_in, NULL));
 997	attr.test.ctx_out = ptr_to_u64(OPTS_GET(opts, ctx_out, NULL));
 998	attr.test.data_in = ptr_to_u64(OPTS_GET(opts, data_in, NULL));
 999	attr.test.data_out = ptr_to_u64(OPTS_GET(opts, data_out, NULL));
1000
1001	ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, attr_sz);
1002
1003	OPTS_SET(opts, data_size_out, attr.test.data_size_out);
1004	OPTS_SET(opts, ctx_size_out, attr.test.ctx_size_out);
1005	OPTS_SET(opts, duration, attr.test.duration);
1006	OPTS_SET(opts, retval, attr.test.retval);
1007
1008	return libbpf_err_errno(ret);
1009}
1010
1011static int bpf_obj_get_next_id(__u32 start_id, __u32 *next_id, int cmd)
1012{
1013	const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1014	union bpf_attr attr;
1015	int err;
1016
1017	memset(&attr, 0, attr_sz);
1018	attr.start_id = start_id;
1019
1020	err = sys_bpf(cmd, &attr, attr_sz);
1021	if (!err)
1022		*next_id = attr.next_id;
1023
1024	return libbpf_err_errno(err);
1025}
1026
1027int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id)
1028{
1029	return bpf_obj_get_next_id(start_id, next_id, BPF_PROG_GET_NEXT_ID);
1030}
1031
1032int bpf_map_get_next_id(__u32 start_id, __u32 *next_id)
1033{
1034	return bpf_obj_get_next_id(start_id, next_id, BPF_MAP_GET_NEXT_ID);
1035}
1036
1037int bpf_btf_get_next_id(__u32 start_id, __u32 *next_id)
1038{
1039	return bpf_obj_get_next_id(start_id, next_id, BPF_BTF_GET_NEXT_ID);
1040}
1041
1042int bpf_link_get_next_id(__u32 start_id, __u32 *next_id)
1043{
1044	return bpf_obj_get_next_id(start_id, next_id, BPF_LINK_GET_NEXT_ID);
1045}
1046
1047int bpf_prog_get_fd_by_id_opts(__u32 id,
1048			       const struct bpf_get_fd_by_id_opts *opts)
1049{
1050	const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1051	union bpf_attr attr;
1052	int fd;
1053
1054	if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1055		return libbpf_err(-EINVAL);
1056
1057	memset(&attr, 0, attr_sz);
1058	attr.prog_id = id;
1059	attr.open_flags = OPTS_GET(opts, open_flags, 0);
1060
1061	fd = sys_bpf_fd(BPF_PROG_GET_FD_BY_ID, &attr, attr_sz);
1062	return libbpf_err_errno(fd);
1063}
1064
1065int bpf_prog_get_fd_by_id(__u32 id)
1066{
1067	return bpf_prog_get_fd_by_id_opts(id, NULL);
1068}
1069
1070int bpf_map_get_fd_by_id_opts(__u32 id,
1071			      const struct bpf_get_fd_by_id_opts *opts)
1072{
1073	const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1074	union bpf_attr attr;
1075	int fd;
1076
1077	if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1078		return libbpf_err(-EINVAL);
1079
1080	memset(&attr, 0, attr_sz);
1081	attr.map_id = id;
1082	attr.open_flags = OPTS_GET(opts, open_flags, 0);
1083
1084	fd = sys_bpf_fd(BPF_MAP_GET_FD_BY_ID, &attr, attr_sz);
1085	return libbpf_err_errno(fd);
1086}
1087
1088int bpf_map_get_fd_by_id(__u32 id)
1089{
1090	return bpf_map_get_fd_by_id_opts(id, NULL);
1091}
1092
1093int bpf_btf_get_fd_by_id_opts(__u32 id,
1094			      const struct bpf_get_fd_by_id_opts *opts)
1095{
1096	const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1097	union bpf_attr attr;
1098	int fd;
1099
1100	if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1101		return libbpf_err(-EINVAL);
1102
1103	memset(&attr, 0, attr_sz);
1104	attr.btf_id = id;
1105	attr.open_flags = OPTS_GET(opts, open_flags, 0);
1106
1107	fd = sys_bpf_fd(BPF_BTF_GET_FD_BY_ID, &attr, attr_sz);
1108	return libbpf_err_errno(fd);
1109}
1110
1111int bpf_btf_get_fd_by_id(__u32 id)
1112{
1113	return bpf_btf_get_fd_by_id_opts(id, NULL);
1114}
1115
1116int bpf_link_get_fd_by_id_opts(__u32 id,
1117			       const struct bpf_get_fd_by_id_opts *opts)
1118{
1119	const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1120	union bpf_attr attr;
1121	int fd;
1122
1123	if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1124		return libbpf_err(-EINVAL);
1125
1126	memset(&attr, 0, attr_sz);
1127	attr.link_id = id;
1128	attr.open_flags = OPTS_GET(opts, open_flags, 0);
1129
1130	fd = sys_bpf_fd(BPF_LINK_GET_FD_BY_ID, &attr, attr_sz);
1131	return libbpf_err_errno(fd);
1132}
1133
1134int bpf_link_get_fd_by_id(__u32 id)
1135{
1136	return bpf_link_get_fd_by_id_opts(id, NULL);
1137}
1138
1139int bpf_obj_get_info_by_fd(int bpf_fd, void *info, __u32 *info_len)
1140{
1141	const size_t attr_sz = offsetofend(union bpf_attr, info);
1142	union bpf_attr attr;
1143	int err;
1144
1145	memset(&attr, 0, attr_sz);
1146	attr.info.bpf_fd = bpf_fd;
1147	attr.info.info_len = *info_len;
1148	attr.info.info = ptr_to_u64(info);
1149
1150	err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, attr_sz);
1151	if (!err)
1152		*info_len = attr.info.info_len;
1153	return libbpf_err_errno(err);
1154}
1155
1156int bpf_prog_get_info_by_fd(int prog_fd, struct bpf_prog_info *info, __u32 *info_len)
1157{
1158	return bpf_obj_get_info_by_fd(prog_fd, info, info_len);
1159}
1160
1161int bpf_map_get_info_by_fd(int map_fd, struct bpf_map_info *info, __u32 *info_len)
1162{
1163	return bpf_obj_get_info_by_fd(map_fd, info, info_len);
1164}
1165
1166int bpf_btf_get_info_by_fd(int btf_fd, struct bpf_btf_info *info, __u32 *info_len)
1167{
1168	return bpf_obj_get_info_by_fd(btf_fd, info, info_len);
1169}
1170
1171int bpf_link_get_info_by_fd(int link_fd, struct bpf_link_info *info, __u32 *info_len)
1172{
1173	return bpf_obj_get_info_by_fd(link_fd, info, info_len);
1174}
1175
1176int bpf_raw_tracepoint_open(const char *name, int prog_fd)
1177{
1178	const size_t attr_sz = offsetofend(union bpf_attr, raw_tracepoint);
1179	union bpf_attr attr;
1180	int fd;
1181
 
 
 
1182	memset(&attr, 0, attr_sz);
1183	attr.raw_tracepoint.name = ptr_to_u64(name);
1184	attr.raw_tracepoint.prog_fd = prog_fd;
 
 
1185
1186	fd = sys_bpf_fd(BPF_RAW_TRACEPOINT_OPEN, &attr, attr_sz);
1187	return libbpf_err_errno(fd);
 
 
 
 
 
 
 
1188}
1189
1190int bpf_btf_load(const void *btf_data, size_t btf_size, struct bpf_btf_load_opts *opts)
1191{
1192	const size_t attr_sz = offsetofend(union bpf_attr, btf_token_fd);
1193	union bpf_attr attr;
1194	char *log_buf;
1195	size_t log_size;
1196	__u32 log_level;
1197	int fd;
1198
1199	bump_rlimit_memlock();
1200
1201	memset(&attr, 0, attr_sz);
1202
1203	if (!OPTS_VALID(opts, bpf_btf_load_opts))
1204		return libbpf_err(-EINVAL);
1205
1206	log_buf = OPTS_GET(opts, log_buf, NULL);
1207	log_size = OPTS_GET(opts, log_size, 0);
1208	log_level = OPTS_GET(opts, log_level, 0);
1209
1210	if (log_size > UINT_MAX)
1211		return libbpf_err(-EINVAL);
1212	if (log_size && !log_buf)
1213		return libbpf_err(-EINVAL);
1214
1215	attr.btf = ptr_to_u64(btf_data);
1216	attr.btf_size = btf_size;
1217
1218	attr.btf_flags = OPTS_GET(opts, btf_flags, 0);
1219	attr.btf_token_fd = OPTS_GET(opts, token_fd, 0);
1220
1221	/* log_level == 0 and log_buf != NULL means "try loading without
1222	 * log_buf, but retry with log_buf and log_level=1 on error", which is
1223	 * consistent across low-level and high-level BTF and program loading
1224	 * APIs within libbpf and provides a sensible behavior in practice
1225	 */
1226	if (log_level) {
1227		attr.btf_log_buf = ptr_to_u64(log_buf);
1228		attr.btf_log_size = (__u32)log_size;
1229		attr.btf_log_level = log_level;
1230	}
1231
1232	fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1233	if (fd < 0 && log_buf && log_level == 0) {
1234		attr.btf_log_buf = ptr_to_u64(log_buf);
1235		attr.btf_log_size = (__u32)log_size;
1236		attr.btf_log_level = 1;
1237		fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1238	}
1239
1240	OPTS_SET(opts, log_true_size, attr.btf_log_true_size);
1241	return libbpf_err_errno(fd);
1242}
1243
1244int bpf_task_fd_query(int pid, int fd, __u32 flags, char *buf, __u32 *buf_len,
1245		      __u32 *prog_id, __u32 *fd_type, __u64 *probe_offset,
1246		      __u64 *probe_addr)
1247{
1248	const size_t attr_sz = offsetofend(union bpf_attr, task_fd_query);
1249	union bpf_attr attr;
1250	int err;
1251
1252	memset(&attr, 0, attr_sz);
1253	attr.task_fd_query.pid = pid;
1254	attr.task_fd_query.fd = fd;
1255	attr.task_fd_query.flags = flags;
1256	attr.task_fd_query.buf = ptr_to_u64(buf);
1257	attr.task_fd_query.buf_len = *buf_len;
1258
1259	err = sys_bpf(BPF_TASK_FD_QUERY, &attr, attr_sz);
1260
1261	*buf_len = attr.task_fd_query.buf_len;
1262	*prog_id = attr.task_fd_query.prog_id;
1263	*fd_type = attr.task_fd_query.fd_type;
1264	*probe_offset = attr.task_fd_query.probe_offset;
1265	*probe_addr = attr.task_fd_query.probe_addr;
1266
1267	return libbpf_err_errno(err);
1268}
1269
1270int bpf_enable_stats(enum bpf_stats_type type)
1271{
1272	const size_t attr_sz = offsetofend(union bpf_attr, enable_stats);
1273	union bpf_attr attr;
1274	int fd;
1275
1276	memset(&attr, 0, attr_sz);
1277	attr.enable_stats.type = type;
1278
1279	fd = sys_bpf_fd(BPF_ENABLE_STATS, &attr, attr_sz);
1280	return libbpf_err_errno(fd);
1281}
1282
1283int bpf_prog_bind_map(int prog_fd, int map_fd,
1284		      const struct bpf_prog_bind_opts *opts)
1285{
1286	const size_t attr_sz = offsetofend(union bpf_attr, prog_bind_map);
1287	union bpf_attr attr;
1288	int ret;
1289
1290	if (!OPTS_VALID(opts, bpf_prog_bind_opts))
1291		return libbpf_err(-EINVAL);
1292
1293	memset(&attr, 0, attr_sz);
1294	attr.prog_bind_map.prog_fd = prog_fd;
1295	attr.prog_bind_map.map_fd = map_fd;
1296	attr.prog_bind_map.flags = OPTS_GET(opts, flags, 0);
1297
1298	ret = sys_bpf(BPF_PROG_BIND_MAP, &attr, attr_sz);
1299	return libbpf_err_errno(ret);
1300}
1301
1302int bpf_token_create(int bpffs_fd, struct bpf_token_create_opts *opts)
1303{
1304	const size_t attr_sz = offsetofend(union bpf_attr, token_create);
1305	union bpf_attr attr;
1306	int fd;
1307
1308	if (!OPTS_VALID(opts, bpf_token_create_opts))
1309		return libbpf_err(-EINVAL);
1310
1311	memset(&attr, 0, attr_sz);
1312	attr.token_create.bpffs_fd = bpffs_fd;
1313	attr.token_create.flags = OPTS_GET(opts, flags, 0);
1314
1315	fd = sys_bpf_fd(BPF_TOKEN_CREATE, &attr, attr_sz);
1316	return libbpf_err_errno(fd);
1317}
v6.13.7
   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(int token_fd)
 107{
 108	const size_t attr_sz = offsetofend(union bpf_attr, prog_token_fd);
 109	struct bpf_insn insns[] = {
 110		BPF_EMIT_CALL(BPF_FUNC_ktime_get_coarse_ns),
 111		BPF_EXIT_INSN(),
 112	};
 113	size_t insn_cnt = ARRAY_SIZE(insns);
 114	union bpf_attr attr;
 115	int prog_fd;
 116
 117	/* attempt loading freplace trying to use custom BTF */
 118	memset(&attr, 0, attr_sz);
 119	attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
 120	attr.insns = ptr_to_u64(insns);
 121	attr.insn_cnt = insn_cnt;
 122	attr.license = ptr_to_u64("GPL");
 123	attr.prog_token_fd = token_fd;
 124	if (token_fd)
 125		attr.prog_flags |= BPF_F_TOKEN_FD;
 126
 127	prog_fd = sys_bpf_fd(BPF_PROG_LOAD, &attr, attr_sz);
 128	if (prog_fd >= 0) {
 129		close(prog_fd);
 130		return 1;
 131	}
 132	return 0;
 133}
 134
 135static bool memlock_bumped;
 136static rlim_t memlock_rlim = RLIM_INFINITY;
 137
 138int libbpf_set_memlock_rlim(size_t memlock_bytes)
 139{
 140	if (memlock_bumped)
 141		return libbpf_err(-EBUSY);
 142
 143	memlock_rlim = memlock_bytes;
 144	return 0;
 145}
 146
 147int bump_rlimit_memlock(void)
 148{
 149	struct rlimit rlim;
 150
 151	/* if kernel supports memcg-based accounting, skip bumping RLIMIT_MEMLOCK */
 152	if (memlock_bumped || feat_supported(NULL, FEAT_MEMCG_ACCOUNT))
 153		return 0;
 154
 155	memlock_bumped = true;
 156
 157	/* zero memlock_rlim_max disables auto-bumping RLIMIT_MEMLOCK */
 158	if (memlock_rlim == 0)
 159		return 0;
 160
 161	rlim.rlim_cur = rlim.rlim_max = memlock_rlim;
 162	if (setrlimit(RLIMIT_MEMLOCK, &rlim))
 163		return -errno;
 164
 165	return 0;
 166}
 167
 168int bpf_map_create(enum bpf_map_type map_type,
 169		   const char *map_name,
 170		   __u32 key_size,
 171		   __u32 value_size,
 172		   __u32 max_entries,
 173		   const struct bpf_map_create_opts *opts)
 174{
 175	const size_t attr_sz = offsetofend(union bpf_attr, map_token_fd);
 176	union bpf_attr attr;
 177	int fd;
 178
 179	bump_rlimit_memlock();
 180
 181	memset(&attr, 0, attr_sz);
 182
 183	if (!OPTS_VALID(opts, bpf_map_create_opts))
 184		return libbpf_err(-EINVAL);
 185
 186	attr.map_type = map_type;
 187	if (map_name && feat_supported(NULL, FEAT_PROG_NAME))
 188		libbpf_strlcpy(attr.map_name, map_name, sizeof(attr.map_name));
 189	attr.key_size = key_size;
 190	attr.value_size = value_size;
 191	attr.max_entries = max_entries;
 192
 193	attr.btf_fd = OPTS_GET(opts, btf_fd, 0);
 194	attr.btf_key_type_id = OPTS_GET(opts, btf_key_type_id, 0);
 195	attr.btf_value_type_id = OPTS_GET(opts, btf_value_type_id, 0);
 196	attr.btf_vmlinux_value_type_id = OPTS_GET(opts, btf_vmlinux_value_type_id, 0);
 197	attr.value_type_btf_obj_fd = OPTS_GET(opts, value_type_btf_obj_fd, 0);
 198
 199	attr.inner_map_fd = OPTS_GET(opts, inner_map_fd, 0);
 200	attr.map_flags = OPTS_GET(opts, map_flags, 0);
 201	attr.map_extra = OPTS_GET(opts, map_extra, 0);
 202	attr.numa_node = OPTS_GET(opts, numa_node, 0);
 203	attr.map_ifindex = OPTS_GET(opts, map_ifindex, 0);
 204
 205	attr.map_token_fd = OPTS_GET(opts, token_fd, 0);
 206
 207	fd = sys_bpf_fd(BPF_MAP_CREATE, &attr, attr_sz);
 208	return libbpf_err_errno(fd);
 209}
 210
 211static void *
 212alloc_zero_tailing_info(const void *orecord, __u32 cnt,
 213			__u32 actual_rec_size, __u32 expected_rec_size)
 214{
 215	__u64 info_len = (__u64)actual_rec_size * cnt;
 216	void *info, *nrecord;
 217	int i;
 218
 219	info = malloc(info_len);
 220	if (!info)
 221		return NULL;
 222
 223	/* zero out bytes kernel does not understand */
 224	nrecord = info;
 225	for (i = 0; i < cnt; i++) {
 226		memcpy(nrecord, orecord, expected_rec_size);
 227		memset(nrecord + expected_rec_size, 0,
 228		       actual_rec_size - expected_rec_size);
 229		orecord += actual_rec_size;
 230		nrecord += actual_rec_size;
 231	}
 232
 233	return info;
 234}
 235
 236int bpf_prog_load(enum bpf_prog_type prog_type,
 237		  const char *prog_name, const char *license,
 238		  const struct bpf_insn *insns, size_t insn_cnt,
 239		  struct bpf_prog_load_opts *opts)
 240{
 241	const size_t attr_sz = offsetofend(union bpf_attr, prog_token_fd);
 242	void *finfo = NULL, *linfo = NULL;
 243	const char *func_info, *line_info;
 244	__u32 log_size, log_level, attach_prog_fd, attach_btf_obj_fd;
 245	__u32 func_info_rec_size, line_info_rec_size;
 246	int fd, attempts;
 247	union bpf_attr attr;
 248	char *log_buf;
 249
 250	bump_rlimit_memlock();
 251
 252	if (!OPTS_VALID(opts, bpf_prog_load_opts))
 253		return libbpf_err(-EINVAL);
 254
 255	attempts = OPTS_GET(opts, attempts, 0);
 256	if (attempts < 0)
 257		return libbpf_err(-EINVAL);
 258	if (attempts == 0)
 259		attempts = PROG_LOAD_ATTEMPTS;
 260
 261	memset(&attr, 0, attr_sz);
 262
 263	attr.prog_type = prog_type;
 264	attr.expected_attach_type = OPTS_GET(opts, expected_attach_type, 0);
 265
 266	attr.prog_btf_fd = OPTS_GET(opts, prog_btf_fd, 0);
 267	attr.prog_flags = OPTS_GET(opts, prog_flags, 0);
 268	attr.prog_ifindex = OPTS_GET(opts, prog_ifindex, 0);
 269	attr.kern_version = OPTS_GET(opts, kern_version, 0);
 270	attr.prog_token_fd = OPTS_GET(opts, token_fd, 0);
 271
 272	if (prog_name && feat_supported(NULL, FEAT_PROG_NAME))
 273		libbpf_strlcpy(attr.prog_name, prog_name, sizeof(attr.prog_name));
 274	attr.license = ptr_to_u64(license);
 275
 276	if (insn_cnt > UINT_MAX)
 277		return libbpf_err(-E2BIG);
 278
 279	attr.insns = ptr_to_u64(insns);
 280	attr.insn_cnt = (__u32)insn_cnt;
 281
 282	attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
 283	attach_btf_obj_fd = OPTS_GET(opts, attach_btf_obj_fd, 0);
 284
 285	if (attach_prog_fd && attach_btf_obj_fd)
 286		return libbpf_err(-EINVAL);
 287
 288	attr.attach_btf_id = OPTS_GET(opts, attach_btf_id, 0);
 289	if (attach_prog_fd)
 290		attr.attach_prog_fd = attach_prog_fd;
 291	else
 292		attr.attach_btf_obj_fd = attach_btf_obj_fd;
 293
 294	log_buf = OPTS_GET(opts, log_buf, NULL);
 295	log_size = OPTS_GET(opts, log_size, 0);
 296	log_level = OPTS_GET(opts, log_level, 0);
 297
 298	if (!!log_buf != !!log_size)
 299		return libbpf_err(-EINVAL);
 300
 301	func_info_rec_size = OPTS_GET(opts, func_info_rec_size, 0);
 302	func_info = OPTS_GET(opts, func_info, NULL);
 303	attr.func_info_rec_size = func_info_rec_size;
 304	attr.func_info = ptr_to_u64(func_info);
 305	attr.func_info_cnt = OPTS_GET(opts, func_info_cnt, 0);
 306
 307	line_info_rec_size = OPTS_GET(opts, line_info_rec_size, 0);
 308	line_info = OPTS_GET(opts, line_info, NULL);
 309	attr.line_info_rec_size = line_info_rec_size;
 310	attr.line_info = ptr_to_u64(line_info);
 311	attr.line_info_cnt = OPTS_GET(opts, line_info_cnt, 0);
 312
 313	attr.fd_array = ptr_to_u64(OPTS_GET(opts, fd_array, NULL));
 314
 315	if (log_level) {
 316		attr.log_buf = ptr_to_u64(log_buf);
 317		attr.log_size = log_size;
 318		attr.log_level = log_level;
 319	}
 320
 321	fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
 322	OPTS_SET(opts, log_true_size, attr.log_true_size);
 323	if (fd >= 0)
 324		return fd;
 325
 326	/* After bpf_prog_load, the kernel may modify certain attributes
 327	 * to give user space a hint how to deal with loading failure.
 328	 * Check to see whether we can make some changes and load again.
 329	 */
 330	while (errno == E2BIG && (!finfo || !linfo)) {
 331		if (!finfo && attr.func_info_cnt &&
 332		    attr.func_info_rec_size < func_info_rec_size) {
 333			/* try with corrected func info records */
 334			finfo = alloc_zero_tailing_info(func_info,
 335							attr.func_info_cnt,
 336							func_info_rec_size,
 337							attr.func_info_rec_size);
 338			if (!finfo) {
 339				errno = E2BIG;
 340				goto done;
 341			}
 342
 343			attr.func_info = ptr_to_u64(finfo);
 344			attr.func_info_rec_size = func_info_rec_size;
 345		} else if (!linfo && attr.line_info_cnt &&
 346			   attr.line_info_rec_size < line_info_rec_size) {
 347			linfo = alloc_zero_tailing_info(line_info,
 348							attr.line_info_cnt,
 349							line_info_rec_size,
 350							attr.line_info_rec_size);
 351			if (!linfo) {
 352				errno = E2BIG;
 353				goto done;
 354			}
 355
 356			attr.line_info = ptr_to_u64(linfo);
 357			attr.line_info_rec_size = line_info_rec_size;
 358		} else {
 359			break;
 360		}
 361
 362		fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
 363		OPTS_SET(opts, log_true_size, attr.log_true_size);
 364		if (fd >= 0)
 365			goto done;
 366	}
 367
 368	if (log_level == 0 && log_buf) {
 369		/* log_level == 0 with non-NULL log_buf requires retrying on error
 370		 * with log_level == 1 and log_buf/log_buf_size set, to get details of
 371		 * failure
 372		 */
 373		attr.log_buf = ptr_to_u64(log_buf);
 374		attr.log_size = log_size;
 375		attr.log_level = 1;
 376
 377		fd = sys_bpf_prog_load(&attr, attr_sz, attempts);
 378		OPTS_SET(opts, log_true_size, attr.log_true_size);
 379	}
 380done:
 381	/* free() doesn't affect errno, so we don't need to restore it */
 382	free(finfo);
 383	free(linfo);
 384	return libbpf_err_errno(fd);
 385}
 386
 387int bpf_map_update_elem(int fd, const void *key, const void *value,
 388			__u64 flags)
 389{
 390	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 391	union bpf_attr attr;
 392	int ret;
 393
 394	memset(&attr, 0, attr_sz);
 395	attr.map_fd = fd;
 396	attr.key = ptr_to_u64(key);
 397	attr.value = ptr_to_u64(value);
 398	attr.flags = flags;
 399
 400	ret = sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, attr_sz);
 401	return libbpf_err_errno(ret);
 402}
 403
 404int bpf_map_lookup_elem(int fd, const void *key, void *value)
 405{
 406	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 407	union bpf_attr attr;
 408	int ret;
 409
 410	memset(&attr, 0, attr_sz);
 411	attr.map_fd = fd;
 412	attr.key = ptr_to_u64(key);
 413	attr.value = ptr_to_u64(value);
 414
 415	ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, attr_sz);
 416	return libbpf_err_errno(ret);
 417}
 418
 419int bpf_map_lookup_elem_flags(int fd, const void *key, void *value, __u64 flags)
 420{
 421	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 422	union bpf_attr attr;
 423	int ret;
 424
 425	memset(&attr, 0, attr_sz);
 426	attr.map_fd = fd;
 427	attr.key = ptr_to_u64(key);
 428	attr.value = ptr_to_u64(value);
 429	attr.flags = flags;
 430
 431	ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, attr_sz);
 432	return libbpf_err_errno(ret);
 433}
 434
 435int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value)
 436{
 437	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 438	union bpf_attr attr;
 439	int ret;
 440
 441	memset(&attr, 0, attr_sz);
 442	attr.map_fd = fd;
 443	attr.key = ptr_to_u64(key);
 444	attr.value = ptr_to_u64(value);
 445
 446	ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, attr_sz);
 447	return libbpf_err_errno(ret);
 448}
 449
 450int bpf_map_lookup_and_delete_elem_flags(int fd, const void *key, void *value, __u64 flags)
 451{
 452	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 453	union bpf_attr attr;
 454	int ret;
 455
 456	memset(&attr, 0, attr_sz);
 457	attr.map_fd = fd;
 458	attr.key = ptr_to_u64(key);
 459	attr.value = ptr_to_u64(value);
 460	attr.flags = flags;
 461
 462	ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, attr_sz);
 463	return libbpf_err_errno(ret);
 464}
 465
 466int bpf_map_delete_elem(int fd, const void *key)
 467{
 468	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 469	union bpf_attr attr;
 470	int ret;
 471
 472	memset(&attr, 0, attr_sz);
 473	attr.map_fd = fd;
 474	attr.key = ptr_to_u64(key);
 475
 476	ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, attr_sz);
 477	return libbpf_err_errno(ret);
 478}
 479
 480int bpf_map_delete_elem_flags(int fd, const void *key, __u64 flags)
 481{
 482	const size_t attr_sz = offsetofend(union bpf_attr, flags);
 483	union bpf_attr attr;
 484	int ret;
 485
 486	memset(&attr, 0, attr_sz);
 487	attr.map_fd = fd;
 488	attr.key = ptr_to_u64(key);
 489	attr.flags = flags;
 490
 491	ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, attr_sz);
 492	return libbpf_err_errno(ret);
 493}
 494
 495int bpf_map_get_next_key(int fd, const void *key, void *next_key)
 496{
 497	const size_t attr_sz = offsetofend(union bpf_attr, next_key);
 498	union bpf_attr attr;
 499	int ret;
 500
 501	memset(&attr, 0, attr_sz);
 502	attr.map_fd = fd;
 503	attr.key = ptr_to_u64(key);
 504	attr.next_key = ptr_to_u64(next_key);
 505
 506	ret = sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, attr_sz);
 507	return libbpf_err_errno(ret);
 508}
 509
 510int bpf_map_freeze(int fd)
 511{
 512	const size_t attr_sz = offsetofend(union bpf_attr, map_fd);
 513	union bpf_attr attr;
 514	int ret;
 515
 516	memset(&attr, 0, attr_sz);
 517	attr.map_fd = fd;
 518
 519	ret = sys_bpf(BPF_MAP_FREEZE, &attr, attr_sz);
 520	return libbpf_err_errno(ret);
 521}
 522
 523static int bpf_map_batch_common(int cmd, int fd, void  *in_batch,
 524				void *out_batch, void *keys, void *values,
 525				__u32 *count,
 526				const struct bpf_map_batch_opts *opts)
 527{
 528	const size_t attr_sz = offsetofend(union bpf_attr, batch);
 529	union bpf_attr attr;
 530	int ret;
 531
 532	if (!OPTS_VALID(opts, bpf_map_batch_opts))
 533		return libbpf_err(-EINVAL);
 534
 535	memset(&attr, 0, attr_sz);
 536	attr.batch.map_fd = fd;
 537	attr.batch.in_batch = ptr_to_u64(in_batch);
 538	attr.batch.out_batch = ptr_to_u64(out_batch);
 539	attr.batch.keys = ptr_to_u64(keys);
 540	attr.batch.values = ptr_to_u64(values);
 541	attr.batch.count = *count;
 542	attr.batch.elem_flags  = OPTS_GET(opts, elem_flags, 0);
 543	attr.batch.flags = OPTS_GET(opts, flags, 0);
 544
 545	ret = sys_bpf(cmd, &attr, attr_sz);
 546	*count = attr.batch.count;
 547
 548	return libbpf_err_errno(ret);
 549}
 550
 551int bpf_map_delete_batch(int fd, const void *keys, __u32 *count,
 552			 const struct bpf_map_batch_opts *opts)
 553{
 554	return bpf_map_batch_common(BPF_MAP_DELETE_BATCH, fd, NULL,
 555				    NULL, (void *)keys, NULL, count, opts);
 556}
 557
 558int bpf_map_lookup_batch(int fd, void *in_batch, void *out_batch, void *keys,
 559			 void *values, __u32 *count,
 560			 const struct bpf_map_batch_opts *opts)
 561{
 562	return bpf_map_batch_common(BPF_MAP_LOOKUP_BATCH, fd, in_batch,
 563				    out_batch, keys, values, count, opts);
 564}
 565
 566int bpf_map_lookup_and_delete_batch(int fd, void *in_batch, void *out_batch,
 567				    void *keys, void *values, __u32 *count,
 568				    const struct bpf_map_batch_opts *opts)
 569{
 570	return bpf_map_batch_common(BPF_MAP_LOOKUP_AND_DELETE_BATCH,
 571				    fd, in_batch, out_batch, keys, values,
 572				    count, opts);
 573}
 574
 575int bpf_map_update_batch(int fd, const void *keys, const void *values, __u32 *count,
 576			 const struct bpf_map_batch_opts *opts)
 577{
 578	return bpf_map_batch_common(BPF_MAP_UPDATE_BATCH, fd, NULL, NULL,
 579				    (void *)keys, (void *)values, count, opts);
 580}
 581
 582int bpf_obj_pin_opts(int fd, const char *pathname, const struct bpf_obj_pin_opts *opts)
 583{
 584	const size_t attr_sz = offsetofend(union bpf_attr, path_fd);
 585	union bpf_attr attr;
 586	int ret;
 587
 588	if (!OPTS_VALID(opts, bpf_obj_pin_opts))
 589		return libbpf_err(-EINVAL);
 590
 591	memset(&attr, 0, attr_sz);
 592	attr.path_fd = OPTS_GET(opts, path_fd, 0);
 593	attr.pathname = ptr_to_u64((void *)pathname);
 594	attr.file_flags = OPTS_GET(opts, file_flags, 0);
 595	attr.bpf_fd = fd;
 596
 597	ret = sys_bpf(BPF_OBJ_PIN, &attr, attr_sz);
 598	return libbpf_err_errno(ret);
 599}
 600
 601int bpf_obj_pin(int fd, const char *pathname)
 602{
 603	return bpf_obj_pin_opts(fd, pathname, NULL);
 604}
 605
 606int bpf_obj_get(const char *pathname)
 607{
 608	return bpf_obj_get_opts(pathname, NULL);
 609}
 610
 611int bpf_obj_get_opts(const char *pathname, const struct bpf_obj_get_opts *opts)
 612{
 613	const size_t attr_sz = offsetofend(union bpf_attr, path_fd);
 614	union bpf_attr attr;
 615	int fd;
 616
 617	if (!OPTS_VALID(opts, bpf_obj_get_opts))
 618		return libbpf_err(-EINVAL);
 619
 620	memset(&attr, 0, attr_sz);
 621	attr.path_fd = OPTS_GET(opts, path_fd, 0);
 622	attr.pathname = ptr_to_u64((void *)pathname);
 623	attr.file_flags = OPTS_GET(opts, file_flags, 0);
 624
 625	fd = sys_bpf_fd(BPF_OBJ_GET, &attr, attr_sz);
 626	return libbpf_err_errno(fd);
 627}
 628
 629int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type,
 630		    unsigned int flags)
 631{
 632	DECLARE_LIBBPF_OPTS(bpf_prog_attach_opts, opts,
 633		.flags = flags,
 634	);
 635
 636	return bpf_prog_attach_opts(prog_fd, target_fd, type, &opts);
 637}
 638
 639int bpf_prog_attach_opts(int prog_fd, int target, enum bpf_attach_type type,
 640			 const struct bpf_prog_attach_opts *opts)
 641{
 642	const size_t attr_sz = offsetofend(union bpf_attr, expected_revision);
 643	__u32 relative_id, flags;
 644	int ret, relative_fd;
 645	union bpf_attr attr;
 646
 647	if (!OPTS_VALID(opts, bpf_prog_attach_opts))
 648		return libbpf_err(-EINVAL);
 649
 650	relative_id = OPTS_GET(opts, relative_id, 0);
 651	relative_fd = OPTS_GET(opts, relative_fd, 0);
 652	flags = OPTS_GET(opts, flags, 0);
 653
 654	/* validate we don't have unexpected combinations of non-zero fields */
 655	if (relative_fd && relative_id)
 656		return libbpf_err(-EINVAL);
 657
 658	memset(&attr, 0, attr_sz);
 659	attr.target_fd		= target;
 660	attr.attach_bpf_fd	= prog_fd;
 661	attr.attach_type	= type;
 662	attr.replace_bpf_fd	= OPTS_GET(opts, replace_fd, 0);
 663	attr.expected_revision	= OPTS_GET(opts, expected_revision, 0);
 664
 665	if (relative_id) {
 666		attr.attach_flags = flags | BPF_F_ID;
 667		attr.relative_id  = relative_id;
 668	} else {
 669		attr.attach_flags = flags;
 670		attr.relative_fd  = relative_fd;
 671	}
 672
 673	ret = sys_bpf(BPF_PROG_ATTACH, &attr, attr_sz);
 674	return libbpf_err_errno(ret);
 675}
 676
 677int bpf_prog_detach_opts(int prog_fd, int target, enum bpf_attach_type type,
 678			 const struct bpf_prog_detach_opts *opts)
 679{
 680	const size_t attr_sz = offsetofend(union bpf_attr, expected_revision);
 681	__u32 relative_id, flags;
 682	int ret, relative_fd;
 683	union bpf_attr attr;
 684
 685	if (!OPTS_VALID(opts, bpf_prog_detach_opts))
 686		return libbpf_err(-EINVAL);
 687
 688	relative_id = OPTS_GET(opts, relative_id, 0);
 689	relative_fd = OPTS_GET(opts, relative_fd, 0);
 690	flags = OPTS_GET(opts, flags, 0);
 691
 692	/* validate we don't have unexpected combinations of non-zero fields */
 693	if (relative_fd && relative_id)
 694		return libbpf_err(-EINVAL);
 695
 696	memset(&attr, 0, attr_sz);
 697	attr.target_fd		= target;
 698	attr.attach_bpf_fd	= prog_fd;
 699	attr.attach_type	= type;
 700	attr.expected_revision	= OPTS_GET(opts, expected_revision, 0);
 701
 702	if (relative_id) {
 703		attr.attach_flags = flags | BPF_F_ID;
 704		attr.relative_id  = relative_id;
 705	} else {
 706		attr.attach_flags = flags;
 707		attr.relative_fd  = relative_fd;
 708	}
 709
 710	ret = sys_bpf(BPF_PROG_DETACH, &attr, attr_sz);
 711	return libbpf_err_errno(ret);
 712}
 713
 714int bpf_prog_detach(int target_fd, enum bpf_attach_type type)
 715{
 716	return bpf_prog_detach_opts(0, target_fd, type, NULL);
 717}
 718
 719int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type)
 720{
 721	return bpf_prog_detach_opts(prog_fd, target_fd, type, NULL);
 722}
 723
 724int bpf_link_create(int prog_fd, int target_fd,
 725		    enum bpf_attach_type attach_type,
 726		    const struct bpf_link_create_opts *opts)
 727{
 728	const size_t attr_sz = offsetofend(union bpf_attr, link_create);
 729	__u32 target_btf_id, iter_info_len, relative_id;
 730	int fd, err, relative_fd;
 731	union bpf_attr attr;
 732
 733	if (!OPTS_VALID(opts, bpf_link_create_opts))
 734		return libbpf_err(-EINVAL);
 735
 736	iter_info_len = OPTS_GET(opts, iter_info_len, 0);
 737	target_btf_id = OPTS_GET(opts, target_btf_id, 0);
 738
 739	/* validate we don't have unexpected combinations of non-zero fields */
 740	if (iter_info_len || target_btf_id) {
 741		if (iter_info_len && target_btf_id)
 742			return libbpf_err(-EINVAL);
 743		if (!OPTS_ZEROED(opts, target_btf_id))
 744			return libbpf_err(-EINVAL);
 745	}
 746
 747	memset(&attr, 0, attr_sz);
 748	attr.link_create.prog_fd = prog_fd;
 749	attr.link_create.target_fd = target_fd;
 750	attr.link_create.attach_type = attach_type;
 751	attr.link_create.flags = OPTS_GET(opts, flags, 0);
 752
 753	if (target_btf_id) {
 754		attr.link_create.target_btf_id = target_btf_id;
 755		goto proceed;
 756	}
 757
 758	switch (attach_type) {
 759	case BPF_TRACE_ITER:
 760		attr.link_create.iter_info = ptr_to_u64(OPTS_GET(opts, iter_info, (void *)0));
 761		attr.link_create.iter_info_len = iter_info_len;
 762		break;
 763	case BPF_PERF_EVENT:
 764		attr.link_create.perf_event.bpf_cookie = OPTS_GET(opts, perf_event.bpf_cookie, 0);
 765		if (!OPTS_ZEROED(opts, perf_event))
 766			return libbpf_err(-EINVAL);
 767		break;
 768	case BPF_TRACE_KPROBE_MULTI:
 769	case BPF_TRACE_KPROBE_SESSION:
 770		attr.link_create.kprobe_multi.flags = OPTS_GET(opts, kprobe_multi.flags, 0);
 771		attr.link_create.kprobe_multi.cnt = OPTS_GET(opts, kprobe_multi.cnt, 0);
 772		attr.link_create.kprobe_multi.syms = ptr_to_u64(OPTS_GET(opts, kprobe_multi.syms, 0));
 773		attr.link_create.kprobe_multi.addrs = ptr_to_u64(OPTS_GET(opts, kprobe_multi.addrs, 0));
 774		attr.link_create.kprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, kprobe_multi.cookies, 0));
 775		if (!OPTS_ZEROED(opts, kprobe_multi))
 776			return libbpf_err(-EINVAL);
 777		break;
 778	case BPF_TRACE_UPROBE_MULTI:
 779	case BPF_TRACE_UPROBE_SESSION:
 780		attr.link_create.uprobe_multi.flags = OPTS_GET(opts, uprobe_multi.flags, 0);
 781		attr.link_create.uprobe_multi.cnt = OPTS_GET(opts, uprobe_multi.cnt, 0);
 782		attr.link_create.uprobe_multi.path = ptr_to_u64(OPTS_GET(opts, uprobe_multi.path, 0));
 783		attr.link_create.uprobe_multi.offsets = ptr_to_u64(OPTS_GET(opts, uprobe_multi.offsets, 0));
 784		attr.link_create.uprobe_multi.ref_ctr_offsets = ptr_to_u64(OPTS_GET(opts, uprobe_multi.ref_ctr_offsets, 0));
 785		attr.link_create.uprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, uprobe_multi.cookies, 0));
 786		attr.link_create.uprobe_multi.pid = OPTS_GET(opts, uprobe_multi.pid, 0);
 787		if (!OPTS_ZEROED(opts, uprobe_multi))
 788			return libbpf_err(-EINVAL);
 789		break;
 790	case BPF_TRACE_RAW_TP:
 791	case BPF_TRACE_FENTRY:
 792	case BPF_TRACE_FEXIT:
 793	case BPF_MODIFY_RETURN:
 794	case BPF_LSM_MAC:
 795		attr.link_create.tracing.cookie = OPTS_GET(opts, tracing.cookie, 0);
 796		if (!OPTS_ZEROED(opts, tracing))
 797			return libbpf_err(-EINVAL);
 798		break;
 799	case BPF_NETFILTER:
 800		attr.link_create.netfilter.pf = OPTS_GET(opts, netfilter.pf, 0);
 801		attr.link_create.netfilter.hooknum = OPTS_GET(opts, netfilter.hooknum, 0);
 802		attr.link_create.netfilter.priority = OPTS_GET(opts, netfilter.priority, 0);
 803		attr.link_create.netfilter.flags = OPTS_GET(opts, netfilter.flags, 0);
 804		if (!OPTS_ZEROED(opts, netfilter))
 805			return libbpf_err(-EINVAL);
 806		break;
 807	case BPF_TCX_INGRESS:
 808	case BPF_TCX_EGRESS:
 809		relative_fd = OPTS_GET(opts, tcx.relative_fd, 0);
 810		relative_id = OPTS_GET(opts, tcx.relative_id, 0);
 811		if (relative_fd && relative_id)
 812			return libbpf_err(-EINVAL);
 813		if (relative_id) {
 814			attr.link_create.tcx.relative_id = relative_id;
 815			attr.link_create.flags |= BPF_F_ID;
 816		} else {
 817			attr.link_create.tcx.relative_fd = relative_fd;
 818		}
 819		attr.link_create.tcx.expected_revision = OPTS_GET(opts, tcx.expected_revision, 0);
 820		if (!OPTS_ZEROED(opts, tcx))
 821			return libbpf_err(-EINVAL);
 822		break;
 823	case BPF_NETKIT_PRIMARY:
 824	case BPF_NETKIT_PEER:
 825		relative_fd = OPTS_GET(opts, netkit.relative_fd, 0);
 826		relative_id = OPTS_GET(opts, netkit.relative_id, 0);
 827		if (relative_fd && relative_id)
 828			return libbpf_err(-EINVAL);
 829		if (relative_id) {
 830			attr.link_create.netkit.relative_id = relative_id;
 831			attr.link_create.flags |= BPF_F_ID;
 832		} else {
 833			attr.link_create.netkit.relative_fd = relative_fd;
 834		}
 835		attr.link_create.netkit.expected_revision = OPTS_GET(opts, netkit.expected_revision, 0);
 836		if (!OPTS_ZEROED(opts, netkit))
 837			return libbpf_err(-EINVAL);
 838		break;
 839	default:
 840		if (!OPTS_ZEROED(opts, flags))
 841			return libbpf_err(-EINVAL);
 842		break;
 843	}
 844proceed:
 845	fd = sys_bpf_fd(BPF_LINK_CREATE, &attr, attr_sz);
 846	if (fd >= 0)
 847		return fd;
 848	/* we'll get EINVAL if LINK_CREATE doesn't support attaching fentry
 849	 * and other similar programs
 850	 */
 851	err = -errno;
 852	if (err != -EINVAL)
 853		return libbpf_err(err);
 854
 855	/* if user used features not supported by
 856	 * BPF_RAW_TRACEPOINT_OPEN command, then just give up immediately
 857	 */
 858	if (attr.link_create.target_fd || attr.link_create.target_btf_id)
 859		return libbpf_err(err);
 860	if (!OPTS_ZEROED(opts, sz))
 861		return libbpf_err(err);
 862
 863	/* otherwise, for few select kinds of programs that can be
 864	 * attached using BPF_RAW_TRACEPOINT_OPEN command, try that as
 865	 * a fallback for older kernels
 866	 */
 867	switch (attach_type) {
 868	case BPF_TRACE_RAW_TP:
 869	case BPF_LSM_MAC:
 870	case BPF_TRACE_FENTRY:
 871	case BPF_TRACE_FEXIT:
 872	case BPF_MODIFY_RETURN:
 873		return bpf_raw_tracepoint_open(NULL, prog_fd);
 874	default:
 875		return libbpf_err(err);
 876	}
 877}
 878
 879int bpf_link_detach(int link_fd)
 880{
 881	const size_t attr_sz = offsetofend(union bpf_attr, link_detach);
 882	union bpf_attr attr;
 883	int ret;
 884
 885	memset(&attr, 0, attr_sz);
 886	attr.link_detach.link_fd = link_fd;
 887
 888	ret = sys_bpf(BPF_LINK_DETACH, &attr, attr_sz);
 889	return libbpf_err_errno(ret);
 890}
 891
 892int bpf_link_update(int link_fd, int new_prog_fd,
 893		    const struct bpf_link_update_opts *opts)
 894{
 895	const size_t attr_sz = offsetofend(union bpf_attr, link_update);
 896	union bpf_attr attr;
 897	int ret;
 898
 899	if (!OPTS_VALID(opts, bpf_link_update_opts))
 900		return libbpf_err(-EINVAL);
 901
 902	if (OPTS_GET(opts, old_prog_fd, 0) && OPTS_GET(opts, old_map_fd, 0))
 903		return libbpf_err(-EINVAL);
 904
 905	memset(&attr, 0, attr_sz);
 906	attr.link_update.link_fd = link_fd;
 907	attr.link_update.new_prog_fd = new_prog_fd;
 908	attr.link_update.flags = OPTS_GET(opts, flags, 0);
 909	if (OPTS_GET(opts, old_prog_fd, 0))
 910		attr.link_update.old_prog_fd = OPTS_GET(opts, old_prog_fd, 0);
 911	else if (OPTS_GET(opts, old_map_fd, 0))
 912		attr.link_update.old_map_fd = OPTS_GET(opts, old_map_fd, 0);
 913
 914	ret = sys_bpf(BPF_LINK_UPDATE, &attr, attr_sz);
 915	return libbpf_err_errno(ret);
 916}
 917
 918int bpf_iter_create(int link_fd)
 919{
 920	const size_t attr_sz = offsetofend(union bpf_attr, iter_create);
 921	union bpf_attr attr;
 922	int fd;
 923
 924	memset(&attr, 0, attr_sz);
 925	attr.iter_create.link_fd = link_fd;
 926
 927	fd = sys_bpf_fd(BPF_ITER_CREATE, &attr, attr_sz);
 928	return libbpf_err_errno(fd);
 929}
 930
 931int bpf_prog_query_opts(int target, enum bpf_attach_type type,
 932			struct bpf_prog_query_opts *opts)
 933{
 934	const size_t attr_sz = offsetofend(union bpf_attr, query);
 935	union bpf_attr attr;
 936	int ret;
 937
 938	if (!OPTS_VALID(opts, bpf_prog_query_opts))
 939		return libbpf_err(-EINVAL);
 940
 941	memset(&attr, 0, attr_sz);
 942	attr.query.target_fd		= target;
 943	attr.query.attach_type		= type;
 944	attr.query.query_flags		= OPTS_GET(opts, query_flags, 0);
 945	attr.query.count		= OPTS_GET(opts, count, 0);
 946	attr.query.prog_ids		= ptr_to_u64(OPTS_GET(opts, prog_ids, NULL));
 947	attr.query.link_ids		= ptr_to_u64(OPTS_GET(opts, link_ids, NULL));
 948	attr.query.prog_attach_flags	= ptr_to_u64(OPTS_GET(opts, prog_attach_flags, NULL));
 949	attr.query.link_attach_flags	= ptr_to_u64(OPTS_GET(opts, link_attach_flags, NULL));
 950
 951	ret = sys_bpf(BPF_PROG_QUERY, &attr, attr_sz);
 952
 953	OPTS_SET(opts, attach_flags, attr.query.attach_flags);
 954	OPTS_SET(opts, revision, attr.query.revision);
 955	OPTS_SET(opts, count, attr.query.count);
 956
 957	return libbpf_err_errno(ret);
 958}
 959
 960int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags,
 961		   __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt)
 962{
 963	LIBBPF_OPTS(bpf_prog_query_opts, opts);
 964	int ret;
 965
 966	opts.query_flags = query_flags;
 967	opts.prog_ids = prog_ids;
 968	opts.prog_cnt = *prog_cnt;
 969
 970	ret = bpf_prog_query_opts(target_fd, type, &opts);
 971
 972	if (attach_flags)
 973		*attach_flags = opts.attach_flags;
 974	*prog_cnt = opts.prog_cnt;
 975
 976	return libbpf_err_errno(ret);
 977}
 978
 979int bpf_prog_test_run_opts(int prog_fd, struct bpf_test_run_opts *opts)
 980{
 981	const size_t attr_sz = offsetofend(union bpf_attr, test);
 982	union bpf_attr attr;
 983	int ret;
 984
 985	if (!OPTS_VALID(opts, bpf_test_run_opts))
 986		return libbpf_err(-EINVAL);
 987
 988	memset(&attr, 0, attr_sz);
 989	attr.test.prog_fd = prog_fd;
 990	attr.test.batch_size = OPTS_GET(opts, batch_size, 0);
 991	attr.test.cpu = OPTS_GET(opts, cpu, 0);
 992	attr.test.flags = OPTS_GET(opts, flags, 0);
 993	attr.test.repeat = OPTS_GET(opts, repeat, 0);
 994	attr.test.duration = OPTS_GET(opts, duration, 0);
 995	attr.test.ctx_size_in = OPTS_GET(opts, ctx_size_in, 0);
 996	attr.test.ctx_size_out = OPTS_GET(opts, ctx_size_out, 0);
 997	attr.test.data_size_in = OPTS_GET(opts, data_size_in, 0);
 998	attr.test.data_size_out = OPTS_GET(opts, data_size_out, 0);
 999	attr.test.ctx_in = ptr_to_u64(OPTS_GET(opts, ctx_in, NULL));
1000	attr.test.ctx_out = ptr_to_u64(OPTS_GET(opts, ctx_out, NULL));
1001	attr.test.data_in = ptr_to_u64(OPTS_GET(opts, data_in, NULL));
1002	attr.test.data_out = ptr_to_u64(OPTS_GET(opts, data_out, NULL));
1003
1004	ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, attr_sz);
1005
1006	OPTS_SET(opts, data_size_out, attr.test.data_size_out);
1007	OPTS_SET(opts, ctx_size_out, attr.test.ctx_size_out);
1008	OPTS_SET(opts, duration, attr.test.duration);
1009	OPTS_SET(opts, retval, attr.test.retval);
1010
1011	return libbpf_err_errno(ret);
1012}
1013
1014static int bpf_obj_get_next_id(__u32 start_id, __u32 *next_id, int cmd)
1015{
1016	const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1017	union bpf_attr attr;
1018	int err;
1019
1020	memset(&attr, 0, attr_sz);
1021	attr.start_id = start_id;
1022
1023	err = sys_bpf(cmd, &attr, attr_sz);
1024	if (!err)
1025		*next_id = attr.next_id;
1026
1027	return libbpf_err_errno(err);
1028}
1029
1030int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id)
1031{
1032	return bpf_obj_get_next_id(start_id, next_id, BPF_PROG_GET_NEXT_ID);
1033}
1034
1035int bpf_map_get_next_id(__u32 start_id, __u32 *next_id)
1036{
1037	return bpf_obj_get_next_id(start_id, next_id, BPF_MAP_GET_NEXT_ID);
1038}
1039
1040int bpf_btf_get_next_id(__u32 start_id, __u32 *next_id)
1041{
1042	return bpf_obj_get_next_id(start_id, next_id, BPF_BTF_GET_NEXT_ID);
1043}
1044
1045int bpf_link_get_next_id(__u32 start_id, __u32 *next_id)
1046{
1047	return bpf_obj_get_next_id(start_id, next_id, BPF_LINK_GET_NEXT_ID);
1048}
1049
1050int bpf_prog_get_fd_by_id_opts(__u32 id,
1051			       const struct bpf_get_fd_by_id_opts *opts)
1052{
1053	const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1054	union bpf_attr attr;
1055	int fd;
1056
1057	if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1058		return libbpf_err(-EINVAL);
1059
1060	memset(&attr, 0, attr_sz);
1061	attr.prog_id = id;
1062	attr.open_flags = OPTS_GET(opts, open_flags, 0);
1063
1064	fd = sys_bpf_fd(BPF_PROG_GET_FD_BY_ID, &attr, attr_sz);
1065	return libbpf_err_errno(fd);
1066}
1067
1068int bpf_prog_get_fd_by_id(__u32 id)
1069{
1070	return bpf_prog_get_fd_by_id_opts(id, NULL);
1071}
1072
1073int bpf_map_get_fd_by_id_opts(__u32 id,
1074			      const struct bpf_get_fd_by_id_opts *opts)
1075{
1076	const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1077	union bpf_attr attr;
1078	int fd;
1079
1080	if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1081		return libbpf_err(-EINVAL);
1082
1083	memset(&attr, 0, attr_sz);
1084	attr.map_id = id;
1085	attr.open_flags = OPTS_GET(opts, open_flags, 0);
1086
1087	fd = sys_bpf_fd(BPF_MAP_GET_FD_BY_ID, &attr, attr_sz);
1088	return libbpf_err_errno(fd);
1089}
1090
1091int bpf_map_get_fd_by_id(__u32 id)
1092{
1093	return bpf_map_get_fd_by_id_opts(id, NULL);
1094}
1095
1096int bpf_btf_get_fd_by_id_opts(__u32 id,
1097			      const struct bpf_get_fd_by_id_opts *opts)
1098{
1099	const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1100	union bpf_attr attr;
1101	int fd;
1102
1103	if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1104		return libbpf_err(-EINVAL);
1105
1106	memset(&attr, 0, attr_sz);
1107	attr.btf_id = id;
1108	attr.open_flags = OPTS_GET(opts, open_flags, 0);
1109
1110	fd = sys_bpf_fd(BPF_BTF_GET_FD_BY_ID, &attr, attr_sz);
1111	return libbpf_err_errno(fd);
1112}
1113
1114int bpf_btf_get_fd_by_id(__u32 id)
1115{
1116	return bpf_btf_get_fd_by_id_opts(id, NULL);
1117}
1118
1119int bpf_link_get_fd_by_id_opts(__u32 id,
1120			       const struct bpf_get_fd_by_id_opts *opts)
1121{
1122	const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
1123	union bpf_attr attr;
1124	int fd;
1125
1126	if (!OPTS_VALID(opts, bpf_get_fd_by_id_opts))
1127		return libbpf_err(-EINVAL);
1128
1129	memset(&attr, 0, attr_sz);
1130	attr.link_id = id;
1131	attr.open_flags = OPTS_GET(opts, open_flags, 0);
1132
1133	fd = sys_bpf_fd(BPF_LINK_GET_FD_BY_ID, &attr, attr_sz);
1134	return libbpf_err_errno(fd);
1135}
1136
1137int bpf_link_get_fd_by_id(__u32 id)
1138{
1139	return bpf_link_get_fd_by_id_opts(id, NULL);
1140}
1141
1142int bpf_obj_get_info_by_fd(int bpf_fd, void *info, __u32 *info_len)
1143{
1144	const size_t attr_sz = offsetofend(union bpf_attr, info);
1145	union bpf_attr attr;
1146	int err;
1147
1148	memset(&attr, 0, attr_sz);
1149	attr.info.bpf_fd = bpf_fd;
1150	attr.info.info_len = *info_len;
1151	attr.info.info = ptr_to_u64(info);
1152
1153	err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, attr_sz);
1154	if (!err)
1155		*info_len = attr.info.info_len;
1156	return libbpf_err_errno(err);
1157}
1158
1159int bpf_prog_get_info_by_fd(int prog_fd, struct bpf_prog_info *info, __u32 *info_len)
1160{
1161	return bpf_obj_get_info_by_fd(prog_fd, info, info_len);
1162}
1163
1164int bpf_map_get_info_by_fd(int map_fd, struct bpf_map_info *info, __u32 *info_len)
1165{
1166	return bpf_obj_get_info_by_fd(map_fd, info, info_len);
1167}
1168
1169int bpf_btf_get_info_by_fd(int btf_fd, struct bpf_btf_info *info, __u32 *info_len)
1170{
1171	return bpf_obj_get_info_by_fd(btf_fd, info, info_len);
1172}
1173
1174int bpf_link_get_info_by_fd(int link_fd, struct bpf_link_info *info, __u32 *info_len)
1175{
1176	return bpf_obj_get_info_by_fd(link_fd, info, info_len);
1177}
1178
1179int bpf_raw_tracepoint_open_opts(int prog_fd, struct bpf_raw_tp_opts *opts)
1180{
1181	const size_t attr_sz = offsetofend(union bpf_attr, raw_tracepoint);
1182	union bpf_attr attr;
1183	int fd;
1184
1185	if (!OPTS_VALID(opts, bpf_raw_tp_opts))
1186		return libbpf_err(-EINVAL);
1187
1188	memset(&attr, 0, attr_sz);
 
1189	attr.raw_tracepoint.prog_fd = prog_fd;
1190	attr.raw_tracepoint.name = ptr_to_u64(OPTS_GET(opts, tp_name, NULL));
1191	attr.raw_tracepoint.cookie = OPTS_GET(opts, cookie, 0);
1192
1193	fd = sys_bpf_fd(BPF_RAW_TRACEPOINT_OPEN, &attr, attr_sz);
1194	return libbpf_err_errno(fd);
1195}
1196
1197int bpf_raw_tracepoint_open(const char *name, int prog_fd)
1198{
1199	LIBBPF_OPTS(bpf_raw_tp_opts, opts, .tp_name = name);
1200
1201	return bpf_raw_tracepoint_open_opts(prog_fd, &opts);
1202}
1203
1204int bpf_btf_load(const void *btf_data, size_t btf_size, struct bpf_btf_load_opts *opts)
1205{
1206	const size_t attr_sz = offsetofend(union bpf_attr, btf_token_fd);
1207	union bpf_attr attr;
1208	char *log_buf;
1209	size_t log_size;
1210	__u32 log_level;
1211	int fd;
1212
1213	bump_rlimit_memlock();
1214
1215	memset(&attr, 0, attr_sz);
1216
1217	if (!OPTS_VALID(opts, bpf_btf_load_opts))
1218		return libbpf_err(-EINVAL);
1219
1220	log_buf = OPTS_GET(opts, log_buf, NULL);
1221	log_size = OPTS_GET(opts, log_size, 0);
1222	log_level = OPTS_GET(opts, log_level, 0);
1223
1224	if (log_size > UINT_MAX)
1225		return libbpf_err(-EINVAL);
1226	if (log_size && !log_buf)
1227		return libbpf_err(-EINVAL);
1228
1229	attr.btf = ptr_to_u64(btf_data);
1230	attr.btf_size = btf_size;
1231
1232	attr.btf_flags = OPTS_GET(opts, btf_flags, 0);
1233	attr.btf_token_fd = OPTS_GET(opts, token_fd, 0);
1234
1235	/* log_level == 0 and log_buf != NULL means "try loading without
1236	 * log_buf, but retry with log_buf and log_level=1 on error", which is
1237	 * consistent across low-level and high-level BTF and program loading
1238	 * APIs within libbpf and provides a sensible behavior in practice
1239	 */
1240	if (log_level) {
1241		attr.btf_log_buf = ptr_to_u64(log_buf);
1242		attr.btf_log_size = (__u32)log_size;
1243		attr.btf_log_level = log_level;
1244	}
1245
1246	fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1247	if (fd < 0 && log_buf && log_level == 0) {
1248		attr.btf_log_buf = ptr_to_u64(log_buf);
1249		attr.btf_log_size = (__u32)log_size;
1250		attr.btf_log_level = 1;
1251		fd = sys_bpf_fd(BPF_BTF_LOAD, &attr, attr_sz);
1252	}
1253
1254	OPTS_SET(opts, log_true_size, attr.btf_log_true_size);
1255	return libbpf_err_errno(fd);
1256}
1257
1258int bpf_task_fd_query(int pid, int fd, __u32 flags, char *buf, __u32 *buf_len,
1259		      __u32 *prog_id, __u32 *fd_type, __u64 *probe_offset,
1260		      __u64 *probe_addr)
1261{
1262	const size_t attr_sz = offsetofend(union bpf_attr, task_fd_query);
1263	union bpf_attr attr;
1264	int err;
1265
1266	memset(&attr, 0, attr_sz);
1267	attr.task_fd_query.pid = pid;
1268	attr.task_fd_query.fd = fd;
1269	attr.task_fd_query.flags = flags;
1270	attr.task_fd_query.buf = ptr_to_u64(buf);
1271	attr.task_fd_query.buf_len = *buf_len;
1272
1273	err = sys_bpf(BPF_TASK_FD_QUERY, &attr, attr_sz);
1274
1275	*buf_len = attr.task_fd_query.buf_len;
1276	*prog_id = attr.task_fd_query.prog_id;
1277	*fd_type = attr.task_fd_query.fd_type;
1278	*probe_offset = attr.task_fd_query.probe_offset;
1279	*probe_addr = attr.task_fd_query.probe_addr;
1280
1281	return libbpf_err_errno(err);
1282}
1283
1284int bpf_enable_stats(enum bpf_stats_type type)
1285{
1286	const size_t attr_sz = offsetofend(union bpf_attr, enable_stats);
1287	union bpf_attr attr;
1288	int fd;
1289
1290	memset(&attr, 0, attr_sz);
1291	attr.enable_stats.type = type;
1292
1293	fd = sys_bpf_fd(BPF_ENABLE_STATS, &attr, attr_sz);
1294	return libbpf_err_errno(fd);
1295}
1296
1297int bpf_prog_bind_map(int prog_fd, int map_fd,
1298		      const struct bpf_prog_bind_opts *opts)
1299{
1300	const size_t attr_sz = offsetofend(union bpf_attr, prog_bind_map);
1301	union bpf_attr attr;
1302	int ret;
1303
1304	if (!OPTS_VALID(opts, bpf_prog_bind_opts))
1305		return libbpf_err(-EINVAL);
1306
1307	memset(&attr, 0, attr_sz);
1308	attr.prog_bind_map.prog_fd = prog_fd;
1309	attr.prog_bind_map.map_fd = map_fd;
1310	attr.prog_bind_map.flags = OPTS_GET(opts, flags, 0);
1311
1312	ret = sys_bpf(BPF_PROG_BIND_MAP, &attr, attr_sz);
1313	return libbpf_err_errno(ret);
1314}
1315
1316int bpf_token_create(int bpffs_fd, struct bpf_token_create_opts *opts)
1317{
1318	const size_t attr_sz = offsetofend(union bpf_attr, token_create);
1319	union bpf_attr attr;
1320	int fd;
1321
1322	if (!OPTS_VALID(opts, bpf_token_create_opts))
1323		return libbpf_err(-EINVAL);
1324
1325	memset(&attr, 0, attr_sz);
1326	attr.token_create.bpffs_fd = bpffs_fd;
1327	attr.token_create.flags = OPTS_GET(opts, flags, 0);
1328
1329	fd = sys_bpf_fd(BPF_TOKEN_CREATE, &attr, attr_sz);
1330	return libbpf_err_errno(fd);
1331}