Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
0002 
0003 /*
0004  * common eBPF ELF operations.
0005  *
0006  * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
0007  * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
0008  * Copyright (C) 2015 Huawei Inc.
0009  *
0010  * This program is free software; you can redistribute it and/or
0011  * modify it under the terms of the GNU Lesser General Public
0012  * License as published by the Free Software Foundation;
0013  * version 2.1 of the License (not later!)
0014  *
0015  * This program is distributed in the hope that it will be useful,
0016  * but WITHOUT ANY WARRANTY; without even the implied warranty of
0017  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
0018  * GNU Lesser General Public License for more details.
0019  *
0020  * You should have received a copy of the GNU Lesser General Public
0021  * License along with this program; if not,  see <http://www.gnu.org/licenses>
0022  */
0023 
0024 #include <stdlib.h>
0025 #include <string.h>
0026 #include <memory.h>
0027 #include <unistd.h>
0028 #include <asm/unistd.h>
0029 #include <errno.h>
0030 #include <linux/bpf.h>
0031 #include <linux/filter.h>
0032 #include <linux/kernel.h>
0033 #include <limits.h>
0034 #include <sys/resource.h>
0035 #include "bpf.h"
0036 #include "libbpf.h"
0037 #include "libbpf_internal.h"
0038 
0039 /*
0040  * When building perf, unistd.h is overridden. __NR_bpf is
0041  * required to be defined explicitly.
0042  */
0043 #ifndef __NR_bpf
0044 # if defined(__i386__)
0045 #  define __NR_bpf 357
0046 # elif defined(__x86_64__)
0047 #  define __NR_bpf 321
0048 # elif defined(__aarch64__)
0049 #  define __NR_bpf 280
0050 # elif defined(__sparc__)
0051 #  define __NR_bpf 349
0052 # elif defined(__s390__)
0053 #  define __NR_bpf 351
0054 # elif defined(__arc__)
0055 #  define __NR_bpf 280
0056 # elif defined(__mips__) && defined(_ABIO32)
0057 #  define __NR_bpf 4355
0058 # elif defined(__mips__) && defined(_ABIN32)
0059 #  define __NR_bpf 6319
0060 # elif defined(__mips__) && defined(_ABI64)
0061 #  define __NR_bpf 5315
0062 # else
0063 #  error __NR_bpf not defined. libbpf does not support your arch.
0064 # endif
0065 #endif
0066 
0067 static inline __u64 ptr_to_u64(const void *ptr)
0068 {
0069     return (__u64) (unsigned long) ptr;
0070 }
0071 
0072 static inline int sys_bpf(enum bpf_cmd cmd, union bpf_attr *attr,
0073               unsigned int size)
0074 {
0075     return syscall(__NR_bpf, cmd, attr, size);
0076 }
0077 
0078 static inline int sys_bpf_fd(enum bpf_cmd cmd, union bpf_attr *attr,
0079                  unsigned int size)
0080 {
0081     int fd;
0082 
0083     fd = sys_bpf(cmd, attr, size);
0084     return ensure_good_fd(fd);
0085 }
0086 
0087 #define PROG_LOAD_ATTEMPTS 5
0088 
0089 static inline int sys_bpf_prog_load(union bpf_attr *attr, unsigned int size, int attempts)
0090 {
0091     int fd;
0092 
0093     do {
0094         fd = sys_bpf_fd(BPF_PROG_LOAD, attr, size);
0095     } while (fd < 0 && errno == EAGAIN && --attempts > 0);
0096 
0097     return fd;
0098 }
0099 
0100 /* Probe whether kernel switched from memlock-based (RLIMIT_MEMLOCK) to
0101  * memcg-based memory accounting for BPF maps and progs. This was done in [0].
0102  * We use the support for bpf_ktime_get_coarse_ns() helper, which was added in
0103  * the same 5.11 Linux release ([1]), to detect memcg-based accounting for BPF.
0104  *
0105  *   [0] https://lore.kernel.org/bpf/20201201215900.3569844-1-guro@fb.com/
0106  *   [1] d05512618056 ("bpf: Add bpf_ktime_get_coarse_ns helper")
0107  */
0108 int probe_memcg_account(void)
0109 {
0110     const size_t prog_load_attr_sz = offsetofend(union bpf_attr, attach_btf_obj_fd);
0111     struct bpf_insn insns[] = {
0112         BPF_EMIT_CALL(BPF_FUNC_ktime_get_coarse_ns),
0113         BPF_EXIT_INSN(),
0114     };
0115     size_t insn_cnt = ARRAY_SIZE(insns);
0116     union bpf_attr attr;
0117     int prog_fd;
0118 
0119     /* attempt loading freplace trying to use custom BTF */
0120     memset(&attr, 0, prog_load_attr_sz);
0121     attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
0122     attr.insns = ptr_to_u64(insns);
0123     attr.insn_cnt = insn_cnt;
0124     attr.license = ptr_to_u64("GPL");
0125 
0126     prog_fd = sys_bpf_fd(BPF_PROG_LOAD, &attr, prog_load_attr_sz);
0127     if (prog_fd >= 0) {
0128         close(prog_fd);
0129         return 1;
0130     }
0131     return 0;
0132 }
0133 
0134 static bool memlock_bumped;
0135 static rlim_t memlock_rlim = RLIM_INFINITY;
0136 
0137 int libbpf_set_memlock_rlim(size_t memlock_bytes)
0138 {
0139     if (memlock_bumped)
0140         return libbpf_err(-EBUSY);
0141 
0142     memlock_rlim = memlock_bytes;
0143     return 0;
0144 }
0145 
0146 int bump_rlimit_memlock(void)
0147 {
0148     struct rlimit rlim;
0149 
0150     /* if kernel supports memcg-based accounting, skip bumping RLIMIT_MEMLOCK */
0151     if (memlock_bumped || kernel_supports(NULL, FEAT_MEMCG_ACCOUNT))
0152         return 0;
0153 
0154     memlock_bumped = true;
0155 
0156     /* zero memlock_rlim_max disables auto-bumping RLIMIT_MEMLOCK */
0157     if (memlock_rlim == 0)
0158         return 0;
0159 
0160     rlim.rlim_cur = rlim.rlim_max = memlock_rlim;
0161     if (setrlimit(RLIMIT_MEMLOCK, &rlim))
0162         return -errno;
0163 
0164     return 0;
0165 }
0166 
0167 int bpf_map_create(enum bpf_map_type map_type,
0168            const char *map_name,
0169            __u32 key_size,
0170            __u32 value_size,
0171            __u32 max_entries,
0172            const struct bpf_map_create_opts *opts)
0173 {
0174     const size_t attr_sz = offsetofend(union bpf_attr, map_extra);
0175     union bpf_attr attr;
0176     int fd;
0177 
0178     bump_rlimit_memlock();
0179 
0180     memset(&attr, 0, attr_sz);
0181 
0182     if (!OPTS_VALID(opts, bpf_map_create_opts))
0183         return libbpf_err(-EINVAL);
0184 
0185     attr.map_type = map_type;
0186     if (map_name)
0187         libbpf_strlcpy(attr.map_name, map_name, sizeof(attr.map_name));
0188     attr.key_size = key_size;
0189     attr.value_size = value_size;
0190     attr.max_entries = max_entries;
0191 
0192     attr.btf_fd = OPTS_GET(opts, btf_fd, 0);
0193     attr.btf_key_type_id = OPTS_GET(opts, btf_key_type_id, 0);
0194     attr.btf_value_type_id = OPTS_GET(opts, btf_value_type_id, 0);
0195     attr.btf_vmlinux_value_type_id = OPTS_GET(opts, btf_vmlinux_value_type_id, 0);
0196 
0197     attr.inner_map_fd = OPTS_GET(opts, inner_map_fd, 0);
0198     attr.map_flags = OPTS_GET(opts, map_flags, 0);
0199     attr.map_extra = OPTS_GET(opts, map_extra, 0);
0200     attr.numa_node = OPTS_GET(opts, numa_node, 0);
0201     attr.map_ifindex = OPTS_GET(opts, map_ifindex, 0);
0202 
0203     fd = sys_bpf_fd(BPF_MAP_CREATE, &attr, attr_sz);
0204     return libbpf_err_errno(fd);
0205 }
0206 
0207 static void *
0208 alloc_zero_tailing_info(const void *orecord, __u32 cnt,
0209             __u32 actual_rec_size, __u32 expected_rec_size)
0210 {
0211     __u64 info_len = (__u64)actual_rec_size * cnt;
0212     void *info, *nrecord;
0213     int i;
0214 
0215     info = malloc(info_len);
0216     if (!info)
0217         return NULL;
0218 
0219     /* zero out bytes kernel does not understand */
0220     nrecord = info;
0221     for (i = 0; i < cnt; i++) {
0222         memcpy(nrecord, orecord, expected_rec_size);
0223         memset(nrecord + expected_rec_size, 0,
0224                actual_rec_size - expected_rec_size);
0225         orecord += actual_rec_size;
0226         nrecord += actual_rec_size;
0227     }
0228 
0229     return info;
0230 }
0231 
0232 int bpf_prog_load(enum bpf_prog_type prog_type,
0233           const char *prog_name, const char *license,
0234           const struct bpf_insn *insns, size_t insn_cnt,
0235           const struct bpf_prog_load_opts *opts)
0236 {
0237     void *finfo = NULL, *linfo = NULL;
0238     const char *func_info, *line_info;
0239     __u32 log_size, log_level, attach_prog_fd, attach_btf_obj_fd;
0240     __u32 func_info_rec_size, line_info_rec_size;
0241     int fd, attempts;
0242     union bpf_attr attr;
0243     char *log_buf;
0244 
0245     bump_rlimit_memlock();
0246 
0247     if (!OPTS_VALID(opts, bpf_prog_load_opts))
0248         return libbpf_err(-EINVAL);
0249 
0250     attempts = OPTS_GET(opts, attempts, 0);
0251     if (attempts < 0)
0252         return libbpf_err(-EINVAL);
0253     if (attempts == 0)
0254         attempts = PROG_LOAD_ATTEMPTS;
0255 
0256     memset(&attr, 0, sizeof(attr));
0257 
0258     attr.prog_type = prog_type;
0259     attr.expected_attach_type = OPTS_GET(opts, expected_attach_type, 0);
0260 
0261     attr.prog_btf_fd = OPTS_GET(opts, prog_btf_fd, 0);
0262     attr.prog_flags = OPTS_GET(opts, prog_flags, 0);
0263     attr.prog_ifindex = OPTS_GET(opts, prog_ifindex, 0);
0264     attr.kern_version = OPTS_GET(opts, kern_version, 0);
0265 
0266     if (prog_name)
0267         libbpf_strlcpy(attr.prog_name, prog_name, sizeof(attr.prog_name));
0268     attr.license = ptr_to_u64(license);
0269 
0270     if (insn_cnt > UINT_MAX)
0271         return libbpf_err(-E2BIG);
0272 
0273     attr.insns = ptr_to_u64(insns);
0274     attr.insn_cnt = (__u32)insn_cnt;
0275 
0276     attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
0277     attach_btf_obj_fd = OPTS_GET(opts, attach_btf_obj_fd, 0);
0278 
0279     if (attach_prog_fd && attach_btf_obj_fd)
0280         return libbpf_err(-EINVAL);
0281 
0282     attr.attach_btf_id = OPTS_GET(opts, attach_btf_id, 0);
0283     if (attach_prog_fd)
0284         attr.attach_prog_fd = attach_prog_fd;
0285     else
0286         attr.attach_btf_obj_fd = attach_btf_obj_fd;
0287 
0288     log_buf = OPTS_GET(opts, log_buf, NULL);
0289     log_size = OPTS_GET(opts, log_size, 0);
0290     log_level = OPTS_GET(opts, log_level, 0);
0291 
0292     if (!!log_buf != !!log_size)
0293         return libbpf_err(-EINVAL);
0294     if (log_level > (4 | 2 | 1))
0295         return libbpf_err(-EINVAL);
0296     if (log_level && !log_buf)
0297         return libbpf_err(-EINVAL);
0298 
0299     func_info_rec_size = OPTS_GET(opts, func_info_rec_size, 0);
0300     func_info = OPTS_GET(opts, func_info, NULL);
0301     attr.func_info_rec_size = func_info_rec_size;
0302     attr.func_info = ptr_to_u64(func_info);
0303     attr.func_info_cnt = OPTS_GET(opts, func_info_cnt, 0);
0304 
0305     line_info_rec_size = OPTS_GET(opts, line_info_rec_size, 0);
0306     line_info = OPTS_GET(opts, line_info, NULL);
0307     attr.line_info_rec_size = line_info_rec_size;
0308     attr.line_info = ptr_to_u64(line_info);
0309     attr.line_info_cnt = OPTS_GET(opts, line_info_cnt, 0);
0310 
0311     attr.fd_array = ptr_to_u64(OPTS_GET(opts, fd_array, NULL));
0312 
0313     if (log_level) {
0314         attr.log_buf = ptr_to_u64(log_buf);
0315         attr.log_size = log_size;
0316         attr.log_level = log_level;
0317     }
0318 
0319     fd = sys_bpf_prog_load(&attr, sizeof(attr), attempts);
0320     if (fd >= 0)
0321         return fd;
0322 
0323     /* After bpf_prog_load, the kernel may modify certain attributes
0324      * to give user space a hint how to deal with loading failure.
0325      * Check to see whether we can make some changes and load again.
0326      */
0327     while (errno == E2BIG && (!finfo || !linfo)) {
0328         if (!finfo && attr.func_info_cnt &&
0329             attr.func_info_rec_size < func_info_rec_size) {
0330             /* try with corrected func info records */
0331             finfo = alloc_zero_tailing_info(func_info,
0332                             attr.func_info_cnt,
0333                             func_info_rec_size,
0334                             attr.func_info_rec_size);
0335             if (!finfo) {
0336                 errno = E2BIG;
0337                 goto done;
0338             }
0339 
0340             attr.func_info = ptr_to_u64(finfo);
0341             attr.func_info_rec_size = func_info_rec_size;
0342         } else if (!linfo && attr.line_info_cnt &&
0343                attr.line_info_rec_size < line_info_rec_size) {
0344             linfo = alloc_zero_tailing_info(line_info,
0345                             attr.line_info_cnt,
0346                             line_info_rec_size,
0347                             attr.line_info_rec_size);
0348             if (!linfo) {
0349                 errno = E2BIG;
0350                 goto done;
0351             }
0352 
0353             attr.line_info = ptr_to_u64(linfo);
0354             attr.line_info_rec_size = line_info_rec_size;
0355         } else {
0356             break;
0357         }
0358 
0359         fd = sys_bpf_prog_load(&attr, sizeof(attr), attempts);
0360         if (fd >= 0)
0361             goto done;
0362     }
0363 
0364     if (log_level == 0 && log_buf) {
0365         /* log_level == 0 with non-NULL log_buf requires retrying on error
0366          * with log_level == 1 and log_buf/log_buf_size set, to get details of
0367          * failure
0368          */
0369         attr.log_buf = ptr_to_u64(log_buf);
0370         attr.log_size = log_size;
0371         attr.log_level = 1;
0372 
0373         fd = sys_bpf_prog_load(&attr, sizeof(attr), attempts);
0374     }
0375 done:
0376     /* free() doesn't affect errno, so we don't need to restore it */
0377     free(finfo);
0378     free(linfo);
0379     return libbpf_err_errno(fd);
0380 }
0381 
0382 int bpf_map_update_elem(int fd, const void *key, const void *value,
0383             __u64 flags)
0384 {
0385     union bpf_attr attr;
0386     int ret;
0387 
0388     memset(&attr, 0, sizeof(attr));
0389     attr.map_fd = fd;
0390     attr.key = ptr_to_u64(key);
0391     attr.value = ptr_to_u64(value);
0392     attr.flags = flags;
0393 
0394     ret = sys_bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr));
0395     return libbpf_err_errno(ret);
0396 }
0397 
0398 int bpf_map_lookup_elem(int fd, const void *key, void *value)
0399 {
0400     union bpf_attr attr;
0401     int ret;
0402 
0403     memset(&attr, 0, sizeof(attr));
0404     attr.map_fd = fd;
0405     attr.key = ptr_to_u64(key);
0406     attr.value = ptr_to_u64(value);
0407 
0408     ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr));
0409     return libbpf_err_errno(ret);
0410 }
0411 
0412 int bpf_map_lookup_elem_flags(int fd, const void *key, void *value, __u64 flags)
0413 {
0414     union bpf_attr attr;
0415     int ret;
0416 
0417     memset(&attr, 0, sizeof(attr));
0418     attr.map_fd = fd;
0419     attr.key = ptr_to_u64(key);
0420     attr.value = ptr_to_u64(value);
0421     attr.flags = flags;
0422 
0423     ret = sys_bpf(BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr));
0424     return libbpf_err_errno(ret);
0425 }
0426 
0427 int bpf_map_lookup_and_delete_elem(int fd, const void *key, void *value)
0428 {
0429     union bpf_attr attr;
0430     int ret;
0431 
0432     memset(&attr, 0, sizeof(attr));
0433     attr.map_fd = fd;
0434     attr.key = ptr_to_u64(key);
0435     attr.value = ptr_to_u64(value);
0436 
0437     ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, sizeof(attr));
0438     return libbpf_err_errno(ret);
0439 }
0440 
0441 int bpf_map_lookup_and_delete_elem_flags(int fd, const void *key, void *value, __u64 flags)
0442 {
0443     union bpf_attr attr;
0444     int ret;
0445 
0446     memset(&attr, 0, sizeof(attr));
0447     attr.map_fd = fd;
0448     attr.key = ptr_to_u64(key);
0449     attr.value = ptr_to_u64(value);
0450     attr.flags = flags;
0451 
0452     ret = sys_bpf(BPF_MAP_LOOKUP_AND_DELETE_ELEM, &attr, sizeof(attr));
0453     return libbpf_err_errno(ret);
0454 }
0455 
0456 int bpf_map_delete_elem(int fd, const void *key)
0457 {
0458     union bpf_attr attr;
0459     int ret;
0460 
0461     memset(&attr, 0, sizeof(attr));
0462     attr.map_fd = fd;
0463     attr.key = ptr_to_u64(key);
0464 
0465     ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr));
0466     return libbpf_err_errno(ret);
0467 }
0468 
0469 int bpf_map_delete_elem_flags(int fd, const void *key, __u64 flags)
0470 {
0471     union bpf_attr attr;
0472     int ret;
0473 
0474     memset(&attr, 0, sizeof(attr));
0475     attr.map_fd = fd;
0476     attr.key = ptr_to_u64(key);
0477     attr.flags = flags;
0478 
0479     ret = sys_bpf(BPF_MAP_DELETE_ELEM, &attr, sizeof(attr));
0480     return libbpf_err_errno(ret);
0481 }
0482 
0483 int bpf_map_get_next_key(int fd, const void *key, void *next_key)
0484 {
0485     union bpf_attr attr;
0486     int ret;
0487 
0488     memset(&attr, 0, sizeof(attr));
0489     attr.map_fd = fd;
0490     attr.key = ptr_to_u64(key);
0491     attr.next_key = ptr_to_u64(next_key);
0492 
0493     ret = sys_bpf(BPF_MAP_GET_NEXT_KEY, &attr, sizeof(attr));
0494     return libbpf_err_errno(ret);
0495 }
0496 
0497 int bpf_map_freeze(int fd)
0498 {
0499     union bpf_attr attr;
0500     int ret;
0501 
0502     memset(&attr, 0, sizeof(attr));
0503     attr.map_fd = fd;
0504 
0505     ret = sys_bpf(BPF_MAP_FREEZE, &attr, sizeof(attr));
0506     return libbpf_err_errno(ret);
0507 }
0508 
0509 static int bpf_map_batch_common(int cmd, int fd, void  *in_batch,
0510                 void *out_batch, void *keys, void *values,
0511                 __u32 *count,
0512                 const struct bpf_map_batch_opts *opts)
0513 {
0514     union bpf_attr attr;
0515     int ret;
0516 
0517     if (!OPTS_VALID(opts, bpf_map_batch_opts))
0518         return libbpf_err(-EINVAL);
0519 
0520     memset(&attr, 0, sizeof(attr));
0521     attr.batch.map_fd = fd;
0522     attr.batch.in_batch = ptr_to_u64(in_batch);
0523     attr.batch.out_batch = ptr_to_u64(out_batch);
0524     attr.batch.keys = ptr_to_u64(keys);
0525     attr.batch.values = ptr_to_u64(values);
0526     attr.batch.count = *count;
0527     attr.batch.elem_flags  = OPTS_GET(opts, elem_flags, 0);
0528     attr.batch.flags = OPTS_GET(opts, flags, 0);
0529 
0530     ret = sys_bpf(cmd, &attr, sizeof(attr));
0531     *count = attr.batch.count;
0532 
0533     return libbpf_err_errno(ret);
0534 }
0535 
0536 int bpf_map_delete_batch(int fd, const void *keys, __u32 *count,
0537              const struct bpf_map_batch_opts *opts)
0538 {
0539     return bpf_map_batch_common(BPF_MAP_DELETE_BATCH, fd, NULL,
0540                     NULL, (void *)keys, NULL, count, opts);
0541 }
0542 
0543 int bpf_map_lookup_batch(int fd, void *in_batch, void *out_batch, void *keys,
0544              void *values, __u32 *count,
0545              const struct bpf_map_batch_opts *opts)
0546 {
0547     return bpf_map_batch_common(BPF_MAP_LOOKUP_BATCH, fd, in_batch,
0548                     out_batch, keys, values, count, opts);
0549 }
0550 
0551 int bpf_map_lookup_and_delete_batch(int fd, void *in_batch, void *out_batch,
0552                     void *keys, void *values, __u32 *count,
0553                     const struct bpf_map_batch_opts *opts)
0554 {
0555     return bpf_map_batch_common(BPF_MAP_LOOKUP_AND_DELETE_BATCH,
0556                     fd, in_batch, out_batch, keys, values,
0557                     count, opts);
0558 }
0559 
0560 int bpf_map_update_batch(int fd, const void *keys, const void *values, __u32 *count,
0561              const struct bpf_map_batch_opts *opts)
0562 {
0563     return bpf_map_batch_common(BPF_MAP_UPDATE_BATCH, fd, NULL, NULL,
0564                     (void *)keys, (void *)values, count, opts);
0565 }
0566 
0567 int bpf_obj_pin(int fd, const char *pathname)
0568 {
0569     union bpf_attr attr;
0570     int ret;
0571 
0572     memset(&attr, 0, sizeof(attr));
0573     attr.pathname = ptr_to_u64((void *)pathname);
0574     attr.bpf_fd = fd;
0575 
0576     ret = sys_bpf(BPF_OBJ_PIN, &attr, sizeof(attr));
0577     return libbpf_err_errno(ret);
0578 }
0579 
0580 int bpf_obj_get(const char *pathname)
0581 {
0582     return bpf_obj_get_opts(pathname, NULL);
0583 }
0584 
0585 int bpf_obj_get_opts(const char *pathname, const struct bpf_obj_get_opts *opts)
0586 {
0587     union bpf_attr attr;
0588     int fd;
0589 
0590     if (!OPTS_VALID(opts, bpf_obj_get_opts))
0591         return libbpf_err(-EINVAL);
0592 
0593     memset(&attr, 0, sizeof(attr));
0594     attr.pathname = ptr_to_u64((void *)pathname);
0595     attr.file_flags = OPTS_GET(opts, file_flags, 0);
0596 
0597     fd = sys_bpf_fd(BPF_OBJ_GET, &attr, sizeof(attr));
0598     return libbpf_err_errno(fd);
0599 }
0600 
0601 int bpf_prog_attach(int prog_fd, int target_fd, enum bpf_attach_type type,
0602             unsigned int flags)
0603 {
0604     DECLARE_LIBBPF_OPTS(bpf_prog_attach_opts, opts,
0605         .flags = flags,
0606     );
0607 
0608     return bpf_prog_attach_opts(prog_fd, target_fd, type, &opts);
0609 }
0610 
0611 int bpf_prog_attach_opts(int prog_fd, int target_fd,
0612               enum bpf_attach_type type,
0613               const struct bpf_prog_attach_opts *opts)
0614 {
0615     union bpf_attr attr;
0616     int ret;
0617 
0618     if (!OPTS_VALID(opts, bpf_prog_attach_opts))
0619         return libbpf_err(-EINVAL);
0620 
0621     memset(&attr, 0, sizeof(attr));
0622     attr.target_fd     = target_fd;
0623     attr.attach_bpf_fd = prog_fd;
0624     attr.attach_type   = type;
0625     attr.attach_flags  = OPTS_GET(opts, flags, 0);
0626     attr.replace_bpf_fd = OPTS_GET(opts, replace_prog_fd, 0);
0627 
0628     ret = sys_bpf(BPF_PROG_ATTACH, &attr, sizeof(attr));
0629     return libbpf_err_errno(ret);
0630 }
0631 
0632 __attribute__((alias("bpf_prog_attach_opts")))
0633 int bpf_prog_attach_xattr(int prog_fd, int target_fd,
0634               enum bpf_attach_type type,
0635               const struct bpf_prog_attach_opts *opts);
0636 
0637 int bpf_prog_detach(int target_fd, enum bpf_attach_type type)
0638 {
0639     union bpf_attr attr;
0640     int ret;
0641 
0642     memset(&attr, 0, sizeof(attr));
0643     attr.target_fd   = target_fd;
0644     attr.attach_type = type;
0645 
0646     ret = sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr));
0647     return libbpf_err_errno(ret);
0648 }
0649 
0650 int bpf_prog_detach2(int prog_fd, int target_fd, enum bpf_attach_type type)
0651 {
0652     union bpf_attr attr;
0653     int ret;
0654 
0655     memset(&attr, 0, sizeof(attr));
0656     attr.target_fd   = target_fd;
0657     attr.attach_bpf_fd = prog_fd;
0658     attr.attach_type = type;
0659 
0660     ret = sys_bpf(BPF_PROG_DETACH, &attr, sizeof(attr));
0661     return libbpf_err_errno(ret);
0662 }
0663 
0664 int bpf_link_create(int prog_fd, int target_fd,
0665             enum bpf_attach_type attach_type,
0666             const struct bpf_link_create_opts *opts)
0667 {
0668     __u32 target_btf_id, iter_info_len;
0669     union bpf_attr attr;
0670     int fd, err;
0671 
0672     if (!OPTS_VALID(opts, bpf_link_create_opts))
0673         return libbpf_err(-EINVAL);
0674 
0675     iter_info_len = OPTS_GET(opts, iter_info_len, 0);
0676     target_btf_id = OPTS_GET(opts, target_btf_id, 0);
0677 
0678     /* validate we don't have unexpected combinations of non-zero fields */
0679     if (iter_info_len || target_btf_id) {
0680         if (iter_info_len && target_btf_id)
0681             return libbpf_err(-EINVAL);
0682         if (!OPTS_ZEROED(opts, target_btf_id))
0683             return libbpf_err(-EINVAL);
0684     }
0685 
0686     memset(&attr, 0, sizeof(attr));
0687     attr.link_create.prog_fd = prog_fd;
0688     attr.link_create.target_fd = target_fd;
0689     attr.link_create.attach_type = attach_type;
0690     attr.link_create.flags = OPTS_GET(opts, flags, 0);
0691 
0692     if (target_btf_id) {
0693         attr.link_create.target_btf_id = target_btf_id;
0694         goto proceed;
0695     }
0696 
0697     switch (attach_type) {
0698     case BPF_TRACE_ITER:
0699         attr.link_create.iter_info = ptr_to_u64(OPTS_GET(opts, iter_info, (void *)0));
0700         attr.link_create.iter_info_len = iter_info_len;
0701         break;
0702     case BPF_PERF_EVENT:
0703         attr.link_create.perf_event.bpf_cookie = OPTS_GET(opts, perf_event.bpf_cookie, 0);
0704         if (!OPTS_ZEROED(opts, perf_event))
0705             return libbpf_err(-EINVAL);
0706         break;
0707     case BPF_TRACE_KPROBE_MULTI:
0708         attr.link_create.kprobe_multi.flags = OPTS_GET(opts, kprobe_multi.flags, 0);
0709         attr.link_create.kprobe_multi.cnt = OPTS_GET(opts, kprobe_multi.cnt, 0);
0710         attr.link_create.kprobe_multi.syms = ptr_to_u64(OPTS_GET(opts, kprobe_multi.syms, 0));
0711         attr.link_create.kprobe_multi.addrs = ptr_to_u64(OPTS_GET(opts, kprobe_multi.addrs, 0));
0712         attr.link_create.kprobe_multi.cookies = ptr_to_u64(OPTS_GET(opts, kprobe_multi.cookies, 0));
0713         if (!OPTS_ZEROED(opts, kprobe_multi))
0714             return libbpf_err(-EINVAL);
0715         break;
0716     case BPF_TRACE_FENTRY:
0717     case BPF_TRACE_FEXIT:
0718     case BPF_MODIFY_RETURN:
0719     case BPF_LSM_MAC:
0720         attr.link_create.tracing.cookie = OPTS_GET(opts, tracing.cookie, 0);
0721         if (!OPTS_ZEROED(opts, tracing))
0722             return libbpf_err(-EINVAL);
0723         break;
0724     default:
0725         if (!OPTS_ZEROED(opts, flags))
0726             return libbpf_err(-EINVAL);
0727         break;
0728     }
0729 proceed:
0730     fd = sys_bpf_fd(BPF_LINK_CREATE, &attr, sizeof(attr));
0731     if (fd >= 0)
0732         return fd;
0733     /* we'll get EINVAL if LINK_CREATE doesn't support attaching fentry
0734      * and other similar programs
0735      */
0736     err = -errno;
0737     if (err != -EINVAL)
0738         return libbpf_err(err);
0739 
0740     /* if user used features not supported by
0741      * BPF_RAW_TRACEPOINT_OPEN command, then just give up immediately
0742      */
0743     if (attr.link_create.target_fd || attr.link_create.target_btf_id)
0744         return libbpf_err(err);
0745     if (!OPTS_ZEROED(opts, sz))
0746         return libbpf_err(err);
0747 
0748     /* otherwise, for few select kinds of programs that can be
0749      * attached using BPF_RAW_TRACEPOINT_OPEN command, try that as
0750      * a fallback for older kernels
0751      */
0752     switch (attach_type) {
0753     case BPF_TRACE_RAW_TP:
0754     case BPF_LSM_MAC:
0755     case BPF_TRACE_FENTRY:
0756     case BPF_TRACE_FEXIT:
0757     case BPF_MODIFY_RETURN:
0758         return bpf_raw_tracepoint_open(NULL, prog_fd);
0759     default:
0760         return libbpf_err(err);
0761     }
0762 }
0763 
0764 int bpf_link_detach(int link_fd)
0765 {
0766     union bpf_attr attr;
0767     int ret;
0768 
0769     memset(&attr, 0, sizeof(attr));
0770     attr.link_detach.link_fd = link_fd;
0771 
0772     ret = sys_bpf(BPF_LINK_DETACH, &attr, sizeof(attr));
0773     return libbpf_err_errno(ret);
0774 }
0775 
0776 int bpf_link_update(int link_fd, int new_prog_fd,
0777             const struct bpf_link_update_opts *opts)
0778 {
0779     union bpf_attr attr;
0780     int ret;
0781 
0782     if (!OPTS_VALID(opts, bpf_link_update_opts))
0783         return libbpf_err(-EINVAL);
0784 
0785     memset(&attr, 0, sizeof(attr));
0786     attr.link_update.link_fd = link_fd;
0787     attr.link_update.new_prog_fd = new_prog_fd;
0788     attr.link_update.flags = OPTS_GET(opts, flags, 0);
0789     attr.link_update.old_prog_fd = OPTS_GET(opts, old_prog_fd, 0);
0790 
0791     ret = sys_bpf(BPF_LINK_UPDATE, &attr, sizeof(attr));
0792     return libbpf_err_errno(ret);
0793 }
0794 
0795 int bpf_iter_create(int link_fd)
0796 {
0797     union bpf_attr attr;
0798     int fd;
0799 
0800     memset(&attr, 0, sizeof(attr));
0801     attr.iter_create.link_fd = link_fd;
0802 
0803     fd = sys_bpf_fd(BPF_ITER_CREATE, &attr, sizeof(attr));
0804     return libbpf_err_errno(fd);
0805 }
0806 
0807 int bpf_prog_query_opts(int target_fd,
0808             enum bpf_attach_type type,
0809             struct bpf_prog_query_opts *opts)
0810 {
0811     union bpf_attr attr;
0812     int ret;
0813 
0814     if (!OPTS_VALID(opts, bpf_prog_query_opts))
0815         return libbpf_err(-EINVAL);
0816 
0817     memset(&attr, 0, sizeof(attr));
0818 
0819     attr.query.target_fd    = target_fd;
0820     attr.query.attach_type  = type;
0821     attr.query.query_flags  = OPTS_GET(opts, query_flags, 0);
0822     attr.query.prog_cnt = OPTS_GET(opts, prog_cnt, 0);
0823     attr.query.prog_ids = ptr_to_u64(OPTS_GET(opts, prog_ids, NULL));
0824     attr.query.prog_attach_flags = ptr_to_u64(OPTS_GET(opts, prog_attach_flags, NULL));
0825 
0826     ret = sys_bpf(BPF_PROG_QUERY, &attr, sizeof(attr));
0827 
0828     OPTS_SET(opts, attach_flags, attr.query.attach_flags);
0829     OPTS_SET(opts, prog_cnt, attr.query.prog_cnt);
0830 
0831     return libbpf_err_errno(ret);
0832 }
0833 
0834 int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags,
0835            __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt)
0836 {
0837     LIBBPF_OPTS(bpf_prog_query_opts, opts);
0838     int ret;
0839 
0840     opts.query_flags = query_flags;
0841     opts.prog_ids = prog_ids;
0842     opts.prog_cnt = *prog_cnt;
0843 
0844     ret = bpf_prog_query_opts(target_fd, type, &opts);
0845 
0846     if (attach_flags)
0847         *attach_flags = opts.attach_flags;
0848     *prog_cnt = opts.prog_cnt;
0849 
0850     return libbpf_err_errno(ret);
0851 }
0852 
0853 int bpf_prog_test_run_opts(int prog_fd, struct bpf_test_run_opts *opts)
0854 {
0855     union bpf_attr attr;
0856     int ret;
0857 
0858     if (!OPTS_VALID(opts, bpf_test_run_opts))
0859         return libbpf_err(-EINVAL);
0860 
0861     memset(&attr, 0, sizeof(attr));
0862     attr.test.prog_fd = prog_fd;
0863     attr.test.batch_size = OPTS_GET(opts, batch_size, 0);
0864     attr.test.cpu = OPTS_GET(opts, cpu, 0);
0865     attr.test.flags = OPTS_GET(opts, flags, 0);
0866     attr.test.repeat = OPTS_GET(opts, repeat, 0);
0867     attr.test.duration = OPTS_GET(opts, duration, 0);
0868     attr.test.ctx_size_in = OPTS_GET(opts, ctx_size_in, 0);
0869     attr.test.ctx_size_out = OPTS_GET(opts, ctx_size_out, 0);
0870     attr.test.data_size_in = OPTS_GET(opts, data_size_in, 0);
0871     attr.test.data_size_out = OPTS_GET(opts, data_size_out, 0);
0872     attr.test.ctx_in = ptr_to_u64(OPTS_GET(opts, ctx_in, NULL));
0873     attr.test.ctx_out = ptr_to_u64(OPTS_GET(opts, ctx_out, NULL));
0874     attr.test.data_in = ptr_to_u64(OPTS_GET(opts, data_in, NULL));
0875     attr.test.data_out = ptr_to_u64(OPTS_GET(opts, data_out, NULL));
0876 
0877     ret = sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr));
0878 
0879     OPTS_SET(opts, data_size_out, attr.test.data_size_out);
0880     OPTS_SET(opts, ctx_size_out, attr.test.ctx_size_out);
0881     OPTS_SET(opts, duration, attr.test.duration);
0882     OPTS_SET(opts, retval, attr.test.retval);
0883 
0884     return libbpf_err_errno(ret);
0885 }
0886 
0887 static int bpf_obj_get_next_id(__u32 start_id, __u32 *next_id, int cmd)
0888 {
0889     union bpf_attr attr;
0890     int err;
0891 
0892     memset(&attr, 0, sizeof(attr));
0893     attr.start_id = start_id;
0894 
0895     err = sys_bpf(cmd, &attr, sizeof(attr));
0896     if (!err)
0897         *next_id = attr.next_id;
0898 
0899     return libbpf_err_errno(err);
0900 }
0901 
0902 int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id)
0903 {
0904     return bpf_obj_get_next_id(start_id, next_id, BPF_PROG_GET_NEXT_ID);
0905 }
0906 
0907 int bpf_map_get_next_id(__u32 start_id, __u32 *next_id)
0908 {
0909     return bpf_obj_get_next_id(start_id, next_id, BPF_MAP_GET_NEXT_ID);
0910 }
0911 
0912 int bpf_btf_get_next_id(__u32 start_id, __u32 *next_id)
0913 {
0914     return bpf_obj_get_next_id(start_id, next_id, BPF_BTF_GET_NEXT_ID);
0915 }
0916 
0917 int bpf_link_get_next_id(__u32 start_id, __u32 *next_id)
0918 {
0919     return bpf_obj_get_next_id(start_id, next_id, BPF_LINK_GET_NEXT_ID);
0920 }
0921 
0922 int bpf_prog_get_fd_by_id(__u32 id)
0923 {
0924     union bpf_attr attr;
0925     int fd;
0926 
0927     memset(&attr, 0, sizeof(attr));
0928     attr.prog_id = id;
0929 
0930     fd = sys_bpf_fd(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr));
0931     return libbpf_err_errno(fd);
0932 }
0933 
0934 int bpf_map_get_fd_by_id(__u32 id)
0935 {
0936     union bpf_attr attr;
0937     int fd;
0938 
0939     memset(&attr, 0, sizeof(attr));
0940     attr.map_id = id;
0941 
0942     fd = sys_bpf_fd(BPF_MAP_GET_FD_BY_ID, &attr, sizeof(attr));
0943     return libbpf_err_errno(fd);
0944 }
0945 
0946 int bpf_btf_get_fd_by_id(__u32 id)
0947 {
0948     union bpf_attr attr;
0949     int fd;
0950 
0951     memset(&attr, 0, sizeof(attr));
0952     attr.btf_id = id;
0953 
0954     fd = sys_bpf_fd(BPF_BTF_GET_FD_BY_ID, &attr, sizeof(attr));
0955     return libbpf_err_errno(fd);
0956 }
0957 
0958 int bpf_link_get_fd_by_id(__u32 id)
0959 {
0960     union bpf_attr attr;
0961     int fd;
0962 
0963     memset(&attr, 0, sizeof(attr));
0964     attr.link_id = id;
0965 
0966     fd = sys_bpf_fd(BPF_LINK_GET_FD_BY_ID, &attr, sizeof(attr));
0967     return libbpf_err_errno(fd);
0968 }
0969 
0970 int bpf_obj_get_info_by_fd(int bpf_fd, void *info, __u32 *info_len)
0971 {
0972     union bpf_attr attr;
0973     int err;
0974 
0975     memset(&attr, 0, sizeof(attr));
0976     attr.info.bpf_fd = bpf_fd;
0977     attr.info.info_len = *info_len;
0978     attr.info.info = ptr_to_u64(info);
0979 
0980     err = sys_bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr));
0981 
0982     if (!err)
0983         *info_len = attr.info.info_len;
0984 
0985     return libbpf_err_errno(err);
0986 }
0987 
0988 int bpf_raw_tracepoint_open(const char *name, int prog_fd)
0989 {
0990     union bpf_attr attr;
0991     int fd;
0992 
0993     memset(&attr, 0, sizeof(attr));
0994     attr.raw_tracepoint.name = ptr_to_u64(name);
0995     attr.raw_tracepoint.prog_fd = prog_fd;
0996 
0997     fd = sys_bpf_fd(BPF_RAW_TRACEPOINT_OPEN, &attr, sizeof(attr));
0998     return libbpf_err_errno(fd);
0999 }
1000 
1001 int 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 
1049 int 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 
1073 int 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 
1085 int 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 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!