1 /* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
4 * Internal libbpf helpers.
6 * Copyright (c) 2019 Facebook
9 #ifndef __LIBBPF_LIBBPF_INTERNAL_H
10 #define __LIBBPF_LIBBPF_INTERNAL_H
16 #include <linux/err.h>
19 #include <sys/syscall.h>
21 #include "relo_core.h"
23 /* Android's libc doesn't support AT_EACCESS in faccessat() implementation
24 * ([0]), and just returns -EINVAL even if file exists and is accessible.
25 * See [1] for issues caused by this.
27 * So just redefine it to 0 on Android.
29 * [0] https://android.googlesource.com/platform/bionic/+/refs/heads/android13-release/libc/bionic/faccessat.cpp#50
30 * [1] https://github.com/libbpf/libbpf-bootstrap/issues/250#issuecomment-1911324250
37 /* make sure libbpf doesn't use kernel-only integer typedefs */
38 #pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64
40 /* prevent accidental re-addition of reallocarray() */
41 #pragma GCC poison reallocarray
53 #ifndef R_BPF_64_ABS64
54 #define R_BPF_64_ABS64 2
56 #ifndef R_BPF_64_ABS32
57 #define R_BPF_64_ABS32 3
60 #define R_BPF_64_32 10
63 #ifndef SHT_LLVM_ADDRSIG
64 #define SHT_LLVM_ADDRSIG 0x6FFF4C03
67 /* if libelf is old and doesn't support mmap(), fall back to read() */
68 #ifndef ELF_C_READ_MMAP
69 #define ELF_C_READ_MMAP ELF_C_READ
72 /* Older libelf all end up in this expression, for both 32 and 64 bit */
73 #ifndef ELF64_ST_VISIBILITY
74 #define ELF64_ST_VISIBILITY(o) ((o) & 0x03)
77 #define BTF_INFO_ENC(kind, kind_flag, vlen) \
78 ((!!(kind_flag) << 31) | ((kind) << 24) | ((vlen) & BTF_MAX_VLEN))
79 #define BTF_TYPE_ENC(name, info, size_or_type) (name), (info), (size_or_type)
80 #define BTF_INT_ENC(encoding, bits_offset, nr_bits) \
81 ((encoding) << 24 | (bits_offset) << 16 | (nr_bits))
82 #define BTF_TYPE_INT_ENC(name, encoding, bits_offset, bits, sz) \
83 BTF_TYPE_ENC(name, BTF_INFO_ENC(BTF_KIND_INT, 0, 0), sz), \
84 BTF_INT_ENC(encoding, bits_offset, bits)
85 #define BTF_MEMBER_ENC(name, type, bits_offset) (name), (type), (bits_offset)
86 #define BTF_PARAM_ENC(name, type) (name), (type)
87 #define BTF_VAR_SECINFO_ENC(type, offset, size) (type), (offset), (size)
88 #define BTF_TYPE_FLOAT_ENC(name, sz) \
89 BTF_TYPE_ENC(name, BTF_INFO_ENC(BTF_KIND_FLOAT, 0, 0), sz)
90 #define BTF_TYPE_DECL_TAG_ENC(value, type, component_idx) \
91 BTF_TYPE_ENC(value, BTF_INFO_ENC(BTF_KIND_DECL_TAG, 0, 0), type), (component_idx)
92 #define BTF_TYPE_TYPE_TAG_ENC(value, type) \
93 BTF_TYPE_ENC(value, BTF_INFO_ENC(BTF_KIND_TYPE_TAG, 0, 0), type)
96 #define likely(x) __builtin_expect(!!(x), 1)
99 #define unlikely(x) __builtin_expect(!!(x), 0)
102 # define min(x, y) ((x) < (y) ? (x) : (y))
105 # define max(x, y) ((x) < (y) ? (y) : (x))
108 # define offsetofend(TYPE, FIELD) \
109 (offsetof(TYPE, FIELD) + sizeof(((TYPE *)0)->FIELD))
112 #define __alias(symbol) __attribute__((alias(#symbol)))
115 /* Check whether a string `str` has prefix `pfx`, regardless if `pfx` is
116 * a string literal known at compilation time or char * pointer known only at
119 #define str_has_pfx(str, pfx) \
120 (strncmp(str, pfx, __builtin_constant_p(pfx) ? sizeof(pfx) - 1 : strlen(pfx)) == 0)
123 static inline bool str_has_sfx(const char *str
, const char *sfx
)
125 size_t str_len
= strlen(str
);
126 size_t sfx_len
= strlen(sfx
);
128 if (sfx_len
> str_len
)
130 return strcmp(str
+ str_len
- sfx_len
, sfx
) == 0;
133 /* Symbol versioning is different between static and shared library.
134 * Properly versioned symbols are needed for shared library, but
135 * only the symbol of the new version is needed for static library.
136 * Starting with GNU C 10, use symver attribute instead of .symver assembler
137 * directive, which works better with GCC LTO builds.
139 #if defined(SHARED) && defined(__GNUC__) && __GNUC__ >= 10
141 #define DEFAULT_VERSION(internal_name, api_name, version) \
142 __attribute__((symver(#api_name "@@" #version)))
143 #define COMPAT_VERSION(internal_name, api_name, version) \
144 __attribute__((symver(#api_name "@" #version)))
146 #elif defined(SHARED)
148 #define COMPAT_VERSION(internal_name, api_name, version) \
149 asm(".symver " #internal_name "," #api_name "@" #version);
150 #define DEFAULT_VERSION(internal_name, api_name, version) \
151 asm(".symver " #internal_name "," #api_name "@@" #version);
155 #define COMPAT_VERSION(internal_name, api_name, version)
156 #define DEFAULT_VERSION(internal_name, api_name, version) \
157 extern typeof(internal_name) api_name \
158 __attribute__((alias(#internal_name)));
162 extern void libbpf_print(enum libbpf_print_level level
,
163 const char *format
, ...)
164 __attribute__((format(printf
, 2, 3)));
166 #define __pr(level, fmt, ...) \
168 libbpf_print(level, "libbpf: " fmt, ##__VA_ARGS__); \
171 #define pr_warn(fmt, ...) __pr(LIBBPF_WARN, fmt, ##__VA_ARGS__)
172 #define pr_info(fmt, ...) __pr(LIBBPF_INFO, fmt, ##__VA_ARGS__)
173 #define pr_debug(fmt, ...) __pr(LIBBPF_DEBUG, fmt, ##__VA_ARGS__)
175 #ifndef __has_builtin
176 #define __has_builtin(x) 0
180 int (*detach
)(struct bpf_link
*link
);
181 void (*dealloc
)(struct bpf_link
*link
);
182 char *pin_path
; /* NULL, if not pinned */
183 int fd
; /* hook FD, -1 if not applicable */
188 * Re-implement glibc's reallocarray() for libbpf internal-only use.
189 * reallocarray(), unfortunately, is not available in all versions of glibc,
190 * so requires extra feature detection and using reallocarray() stub from
191 * <tools/libc_compat.h> and COMPAT_NEED_REALLOCARRAY. All this complicates
192 * build of libbpf unnecessarily and is just a maintenance burden. Instead,
193 * it's trivial to implement libbpf-specific internal version and use it
196 static inline void *libbpf_reallocarray(void *ptr
, size_t nmemb
, size_t size
)
200 #if __has_builtin(__builtin_mul_overflow)
201 if (unlikely(__builtin_mul_overflow(nmemb
, size
, &total
)))
204 if (size
== 0 || nmemb
> ULONG_MAX
/ size
)
206 total
= nmemb
* size
;
208 return realloc(ptr
, total
);
211 /* Copy up to sz - 1 bytes from zero-terminated src string and ensure that dst
212 * is zero-terminated string no matter what (unless sz == 0, in which case
213 * it's a no-op). It's conceptually close to FreeBSD's strlcpy(), but differs
214 * in what is returned. Given this is internal helper, it's trivial to extend
215 * this, when necessary. Use this instead of strncpy inside libbpf source code.
217 static inline void libbpf_strlcpy(char *dst
, const char *src
, size_t sz
)
225 for (i
= 0; i
< sz
&& src
[i
]; i
++)
230 __u32
get_kernel_version(void);
235 struct btf_type
*btf_type_by_id(const struct btf
*btf
, __u32 type_id
);
236 const char *btf_kind_str(const struct btf_type
*t
);
237 const struct btf_type
*skip_mods_and_typedefs(const struct btf
*btf
, __u32 id
, __u32
*res_id
);
238 const struct btf_header
*btf_header(const struct btf
*btf
);
239 void btf_set_base_btf(struct btf
*btf
, const struct btf
*base_btf
);
240 int btf_relocate(struct btf
*btf
, const struct btf
*base_btf
, __u32
**id_map
);
242 static inline enum btf_func_linkage
btf_func_linkage(const struct btf_type
*t
)
244 return (enum btf_func_linkage
)(int)btf_vlen(t
);
247 static inline __u32
btf_type_info(int kind
, int vlen
, int kflag
)
249 return (kflag
<< 31) | (kind
<< 24) | vlen
;
253 MAP_DEF_MAP_TYPE
= 0x001,
254 MAP_DEF_KEY_TYPE
= 0x002,
255 MAP_DEF_KEY_SIZE
= 0x004,
256 MAP_DEF_VALUE_TYPE
= 0x008,
257 MAP_DEF_VALUE_SIZE
= 0x010,
258 MAP_DEF_MAX_ENTRIES
= 0x020,
259 MAP_DEF_MAP_FLAGS
= 0x040,
260 MAP_DEF_NUMA_NODE
= 0x080,
261 MAP_DEF_PINNING
= 0x100,
262 MAP_DEF_INNER_MAP
= 0x200,
263 MAP_DEF_MAP_EXTRA
= 0x400,
265 MAP_DEF_ALL
= 0x7ff, /* combination of all above */
269 enum map_def_parts parts
;
282 int parse_btf_map_def(const char *map_name
, struct btf
*btf
,
283 const struct btf_type
*def_t
, bool strict
,
284 struct btf_map_def
*map_def
, struct btf_map_def
*inner_def
);
286 void *libbpf_add_mem(void **data
, size_t *cap_cnt
, size_t elem_sz
,
287 size_t cur_cnt
, size_t max_cnt
, size_t add_cnt
);
288 int libbpf_ensure_mem(void **data
, size_t *cap_cnt
, size_t elem_sz
, size_t need_cnt
);
290 static inline bool libbpf_is_mem_zeroed(const char *p
, ssize_t len
)
301 static inline bool libbpf_validate_opts(const char *opts
,
302 size_t opts_sz
, size_t user_sz
,
303 const char *type_name
)
305 if (user_sz
< sizeof(size_t)) {
306 pr_warn("%s size (%zu) is too small\n", type_name
, user_sz
);
309 if (!libbpf_is_mem_zeroed(opts
+ opts_sz
, (ssize_t
)user_sz
- opts_sz
)) {
310 pr_warn("%s has non-zero extra bytes\n", type_name
);
316 #define OPTS_VALID(opts, type) \
317 (!(opts) || libbpf_validate_opts((const char *)opts, \
318 offsetofend(struct type, \
319 type##__last_field), \
321 #define OPTS_HAS(opts, field) \
322 ((opts) && opts->sz >= offsetofend(typeof(*(opts)), field))
323 #define OPTS_GET(opts, field, fallback_value) \
324 (OPTS_HAS(opts, field) ? (opts)->field : fallback_value)
325 #define OPTS_SET(opts, field, value) \
327 if (OPTS_HAS(opts, field)) \
328 (opts)->field = value; \
331 #define OPTS_ZEROED(opts, last_nonzero_field) \
333 ssize_t __off = offsetofend(typeof(*(opts)), last_nonzero_field); \
334 !(opts) || libbpf_is_mem_zeroed((const void *)opts + __off, \
335 (opts)->sz - __off); \
338 enum kern_feature_id
{
339 /* v4.14: kernel support for program & map names. */
341 /* v5.2: kernel support for global data sections. */
345 /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
347 /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
349 /* BTF_FUNC_GLOBAL is supported */
350 FEAT_BTF_GLOBAL_FUNC
,
351 /* BPF_F_MMAPABLE is supported for arrays */
353 /* kernel support for expected_attach_type in BPF_PROG_LOAD */
354 FEAT_EXP_ATTACH_TYPE
,
355 /* bpf_probe_read_{kernel,user}[_str] helpers */
356 FEAT_PROBE_READ_KERN
,
357 /* BPF_PROG_BIND_MAP is supported */
359 /* Kernel support for module BTFs */
361 /* BTF_KIND_FLOAT support */
363 /* BPF perf link support */
365 /* BTF_KIND_DECL_TAG support */
367 /* BTF_KIND_TYPE_TAG support */
369 /* memcg-based accounting for BPF maps and progs */
371 /* BPF cookie (bpf_get_attach_cookie() BPF helper) support */
373 /* BTF_KIND_ENUM64 support and BTF_KIND_ENUM kflag support */
375 /* Kernel uses syscall wrapper (CONFIG_ARCH_HAS_SYSCALL_WRAPPER) */
376 FEAT_SYSCALL_WRAPPER
,
377 /* BPF multi-uprobe link support */
378 FEAT_UPROBE_MULTI_LINK
,
379 /* Kernel supports arg:ctx tag (__arg_ctx) for global subprogs natively */
381 /* Kernel supports '?' at the front of datasec names */
382 FEAT_BTF_QMARK_DATASEC
,
386 enum kern_feature_result
{
392 struct kern_feature_cache
{
393 enum kern_feature_result res
[__FEAT_CNT
];
397 bool feat_supported(struct kern_feature_cache
*cache
, enum kern_feature_id feat_id
);
398 bool kernel_supports(const struct bpf_object
*obj
, enum kern_feature_id feat_id
);
400 int probe_kern_syscall_wrapper(int token_fd
);
401 int probe_memcg_account(int token_fd
);
402 int bump_rlimit_memlock(void);
404 int parse_cpu_mask_str(const char *s
, bool **mask
, int *mask_sz
);
405 int parse_cpu_mask_file(const char *fcpu
, bool **mask
, int *mask_sz
);
406 int libbpf__load_raw_btf(const char *raw_types
, size_t types_len
,
407 const char *str_sec
, size_t str_len
,
409 int btf_load_into_kernel(struct btf
*btf
,
410 char *log_buf
, size_t log_sz
, __u32 log_level
,
413 struct btf
*btf_get_from_fd(int btf_fd
, struct btf
*base_btf
);
414 void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type
,
415 const char **prefix
, int *kind
);
417 struct btf_ext_info
{
419 * info points to the individual info section (e.g. func_info and
420 * line_info) from the .BTF.ext. It does not include the __u32 rec_size.
425 /* optional (maintained internally by libbpf) mapping between .BTF.ext
426 * section and corresponding ELF section. This is used to join
427 * information like CO-RE relocation records with corresponding BPF
428 * programs defined in ELF sections
434 #define for_each_btf_ext_sec(seg, sec) \
435 for (sec = (seg)->info; \
436 (void *)sec < (seg)->info + (seg)->len; \
437 sec = (void *)sec + sizeof(struct btf_ext_info_sec) + \
438 (seg)->rec_size * sec->num_info)
440 #define for_each_btf_ext_rec(seg, sec, i, rec) \
441 for (i = 0, rec = (void *)&(sec)->data; \
442 i < (sec)->num_info; \
443 i++, rec = (void *)rec + (seg)->rec_size)
446 * The .BTF.ext ELF section layout defined as
447 * struct btf_ext_header
448 * func_info subsection
450 * The func_info subsection layout:
451 * record size for struct bpf_func_info in the func_info subsection
452 * struct btf_ext_info_sec for section #1
453 * a list of bpf_func_info records for section #1
454 * where struct bpf_func_info mimics one in include/uapi/linux/bpf.h
455 * but may not be identical
456 * struct btf_ext_info_sec for section #2
457 * a list of bpf_func_info records for section #2
460 * Note that the bpf_func_info record size in .BTF.ext may not
461 * be the same as the one defined in include/uapi/linux/bpf.h.
462 * The loader should ensure that record_size meets minimum
463 * requirement and pass the record as is to the kernel. The
464 * kernel will handle the func_info properly based on its contents.
466 struct btf_ext_header
{
472 /* All offsets are in bytes relative to the end of this header */
478 /* optional part of .BTF.ext header */
485 struct btf_ext_header
*hdr
;
490 struct btf_ext_info func_info
;
491 struct btf_ext_info line_info
;
492 struct btf_ext_info core_relo_info
;
496 struct btf_ext_info_sec
{
499 /* Followed by num_info * record_size number of bytes */
503 /* The minimum bpf_func_info checked by the loader */
504 struct bpf_func_info_min
{
509 /* The minimum bpf_line_info checked by the loader */
510 struct bpf_line_info_min
{
517 /* Functions to byte-swap info records */
519 typedef void (*info_rec_bswap_fn
)(void *);
521 static inline void bpf_func_info_bswap(struct bpf_func_info
*i
)
523 i
->insn_off
= bswap_32(i
->insn_off
);
524 i
->type_id
= bswap_32(i
->type_id
);
527 static inline void bpf_line_info_bswap(struct bpf_line_info
*i
)
529 i
->insn_off
= bswap_32(i
->insn_off
);
530 i
->file_name_off
= bswap_32(i
->file_name_off
);
531 i
->line_off
= bswap_32(i
->line_off
);
532 i
->line_col
= bswap_32(i
->line_col
);
535 static inline void bpf_core_relo_bswap(struct bpf_core_relo
*i
)
537 i
->insn_off
= bswap_32(i
->insn_off
);
538 i
->type_id
= bswap_32(i
->type_id
);
539 i
->access_str_off
= bswap_32(i
->access_str_off
);
540 i
->kind
= bswap_32(i
->kind
);
543 enum btf_field_iter_kind
{
548 struct btf_field_desc
{
549 /* once-per-type offsets */
550 int t_off_cnt
, t_offs
[2];
551 /* member struct size, or zero, if no members */
553 /* repeated per-member offsets */
554 int m_off_cnt
, m_offs
[1];
557 struct btf_field_iter
{
558 struct btf_field_desc desc
;
565 int btf_field_iter_init(struct btf_field_iter
*it
, struct btf_type
*t
, enum btf_field_iter_kind iter_kind
);
566 __u32
*btf_field_iter_next(struct btf_field_iter
*it
);
568 typedef int (*type_id_visit_fn
)(__u32
*type_id
, void *ctx
);
569 typedef int (*str_off_visit_fn
)(__u32
*str_off
, void *ctx
);
570 int btf_ext_visit_type_ids(struct btf_ext
*btf_ext
, type_id_visit_fn visit
, void *ctx
);
571 int btf_ext_visit_str_offs(struct btf_ext
*btf_ext
, str_off_visit_fn visit
, void *ctx
);
572 __s32
btf__find_by_name_kind_own(const struct btf
*btf
, const char *type_name
,
575 /* handle direct returned errors */
576 static inline int libbpf_err(int ret
)
583 /* handle errno-based (e.g., syscall or libc) errors according to libbpf's
584 * strict mode settings
586 static inline int libbpf_err_errno(int ret
)
588 /* errno is already assumed to be set on error */
589 return ret
< 0 ? -errno
: ret
;
592 /* handle error for pointer-returning APIs, err is assumed to be < 0 always */
593 static inline void *libbpf_err_ptr(int err
)
595 /* set errno on error, this doesn't break anything */
600 /* handle pointer-returning APIs' error handling */
601 static inline void *libbpf_ptr(void *ret
)
603 /* set errno on error, this doesn't break anything */
605 errno
= -PTR_ERR(ret
);
607 return IS_ERR(ret
) ? NULL
: ret
;
610 static inline bool str_is_empty(const char *s
)
615 static inline bool is_ldimm64_insn(struct bpf_insn
*insn
)
617 return insn
->code
== (BPF_LD
| BPF_IMM
| BPF_DW
);
620 static inline void bpf_insn_bswap(struct bpf_insn
*insn
)
622 __u8 tmp_reg
= insn
->dst_reg
;
624 insn
->dst_reg
= insn
->src_reg
;
625 insn
->src_reg
= tmp_reg
;
626 insn
->off
= bswap_16(insn
->off
);
627 insn
->imm
= bswap_32(insn
->imm
);
630 /* Unconditionally dup FD, ensuring it doesn't use [0, 2] range.
631 * Original FD is not closed or altered in any other way.
632 * Preserves original FD value, if it's invalid (negative).
634 static inline int dup_good_fd(int fd
)
638 return fcntl(fd
, F_DUPFD_CLOEXEC
, 3);
641 /* if fd is stdin, stdout, or stderr, dup to a fd greater than 2
642 * Takes ownership of the fd passed in, and closes it if calling
643 * fcntl(fd, F_DUPFD_CLOEXEC, 3).
645 static inline int ensure_good_fd(int fd
)
647 int old_fd
= fd
, saved_errno
;
652 fd
= dup_good_fd(fd
);
657 pr_warn("failed to dup FD %d to FD > 2: %d\n", old_fd
, -saved_errno
);
664 static inline int sys_dup3(int oldfd
, int newfd
, int flags
)
666 return syscall(__NR_dup3
, oldfd
, newfd
, flags
);
669 /* Point *fixed_fd* to the same file that *tmp_fd* points to.
670 * Regardless of success, *tmp_fd* is closed.
671 * Whatever *fixed_fd* pointed to is closed silently.
673 static inline int reuse_fd(int fixed_fd
, int tmp_fd
)
677 err
= sys_dup3(tmp_fd
, fixed_fd
, O_CLOEXEC
);
678 err
= err
< 0 ? -errno
: 0;
679 close(tmp_fd
); /* clean up temporary FD */
683 /* The following two functions are exposed to bpftool */
684 int bpf_core_add_cands(struct bpf_core_cand
*local_cand
,
685 size_t local_essent_len
,
686 const struct btf
*targ_btf
,
687 const char *targ_btf_name
,
689 struct bpf_core_cand_list
*cands
);
690 void bpf_core_free_cands(struct bpf_core_cand_list
*cands
);
692 struct usdt_manager
*usdt_manager_new(struct bpf_object
*obj
);
693 void usdt_manager_free(struct usdt_manager
*man
);
694 struct bpf_link
* usdt_manager_attach_usdt(struct usdt_manager
*man
,
695 const struct bpf_program
*prog
,
696 pid_t pid
, const char *path
,
697 const char *usdt_provider
, const char *usdt_name
,
700 static inline bool is_pow_of_2(size_t x
)
702 return x
&& (x
& (x
- 1)) == 0;
705 #define PROG_LOAD_ATTEMPTS 5
706 int sys_bpf_prog_load(union bpf_attr
*attr
, unsigned int size
, int attempts
);
708 bool glob_match(const char *str
, const char *pat
);
710 long elf_find_func_offset(Elf
*elf
, const char *binary_path
, const char *name
);
711 long elf_find_func_offset_from_file(const char *binary_path
, const char *name
);
718 int elf_open(const char *binary_path
, struct elf_fd
*elf_fd
);
719 void elf_close(struct elf_fd
*elf_fd
);
721 int elf_resolve_syms_offsets(const char *binary_path
, int cnt
,
722 const char **syms
, unsigned long **poffsets
,
724 int elf_resolve_pattern_offsets(const char *binary_path
, const char *pattern
,
725 unsigned long **poffsets
, size_t *pcnt
);
727 int probe_fd(int fd
);
729 #endif /* __LIBBPF_LIBBPF_INTERNAL_H */