1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
4 * Common eBPF ELF object loading operations.
6 * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
7 * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
8 * Copyright (C) 2015 Huawei Inc.
9 * Copyright (C) 2017 Nicira, Inc.
10 * Copyright (C) 2019 Isovalent, Inc.
26 #include <asm/unistd.h>
27 #include <linux/err.h>
28 #include <linux/kernel.h>
29 #include <linux/bpf.h>
30 #include <linux/btf.h>
31 #include <linux/filter.h>
32 #include <linux/list.h>
33 #include <linux/limits.h>
34 #include <linux/perf_event.h>
35 #include <linux/ring_buffer.h>
36 #include <sys/epoll.h>
37 #include <sys/ioctl.h>
40 #include <sys/types.h>
42 #include <sys/utsname.h>
43 #include <tools/libc_compat.h>
50 #include "str_error.h"
51 #include "libbpf_internal.h"
59 #define BPF_FS_MAGIC 0xcafe4a11
62 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
63 * compilation if user enables corresponding warning. Disable it explicitly.
65 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
67 #define __printf(a, b) __attribute__((format(printf, a, b)))
69 static int __base_pr(enum libbpf_print_level level, const char *format,
72 if (level == LIBBPF_DEBUG)
75 return vfprintf(stderr, format, args);
78 static libbpf_print_fn_t __libbpf_pr = __base_pr;
80 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
82 libbpf_print_fn_t old_print_fn = __libbpf_pr;
89 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
96 va_start(args, format);
97 __libbpf_pr(level, format, args);
101 #define STRERR_BUFSIZE 128
103 #define CHECK_ERR(action, err, out) do { \
110 /* Copied from tools/perf/util/util.h */
112 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
116 # define zclose(fd) ({ \
119 ___err = close((fd)); \
124 #ifdef HAVE_LIBELF_MMAP_SUPPORT
125 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
127 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
130 static inline __u64 ptr_to_u64(const void *ptr)
132 return (__u64) (unsigned long) ptr;
135 struct bpf_capabilities {
136 /* v4.14: kernel support for program & map names. */
138 /* v5.2: kernel support for global data sections. */
140 /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
142 /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
147 * bpf_prog should be a better name but it has been used in
151 /* Index in elf obj file, for relocation use. */
156 /* section_name with / replaced by _; makes recursive pinning
157 * in bpf_object__pin_programs easier
160 struct bpf_insn *insns;
161 size_t insns_cnt, main_prog_cnt;
162 enum bpf_prog_type type;
183 bpf_program_prep_t preprocessor;
185 struct bpf_object *obj;
187 bpf_program_clear_priv_t clear_priv;
189 enum bpf_attach_type expected_attach_type;
191 __u32 func_info_rec_size;
194 struct bpf_capabilities *caps;
197 __u32 line_info_rec_size;
202 enum libbpf_map_type {
209 static const char * const libbpf_type_to_btf_name[] = {
210 [LIBBPF_MAP_DATA] = ".data",
211 [LIBBPF_MAP_BSS] = ".bss",
212 [LIBBPF_MAP_RODATA] = ".rodata",
222 struct bpf_map_def def;
223 __u32 btf_key_type_id;
224 __u32 btf_value_type_id;
226 bpf_map_clear_priv_t clear_priv;
227 enum libbpf_map_type libbpf_type;
235 static LIST_HEAD(bpf_objects_list);
238 char name[BPF_OBJ_NAME_LEN];
242 struct bpf_program *programs;
244 struct bpf_map *maps;
247 struct bpf_secdata sections;
250 bool has_pseudo_calls;
253 * Information when doing elf related work. Only valid if fd
280 * All loaded bpf_object is linked in a list, which is
281 * hidden to caller. bpf_objects__<func> handlers deal with
284 struct list_head list;
287 struct btf_ext *btf_ext;
290 bpf_object_clear_priv_t clear_priv;
292 struct bpf_capabilities caps;
296 #define obj_elf_valid(o) ((o)->efile.elf)
298 void bpf_program__unload(struct bpf_program *prog)
306 * If the object is opened but the program was never loaded,
307 * it is possible that prog->instances.nr == -1.
309 if (prog->instances.nr > 0) {
310 for (i = 0; i < prog->instances.nr; i++)
311 zclose(prog->instances.fds[i]);
312 } else if (prog->instances.nr != -1) {
313 pr_warning("Internal error: instances.nr is %d\n",
317 prog->instances.nr = -1;
318 zfree(&prog->instances.fds);
320 zfree(&prog->func_info);
321 zfree(&prog->line_info);
324 static void bpf_program__exit(struct bpf_program *prog)
329 if (prog->clear_priv)
330 prog->clear_priv(prog, prog->priv);
333 prog->clear_priv = NULL;
335 bpf_program__unload(prog);
337 zfree(&prog->section_name);
338 zfree(&prog->pin_name);
340 zfree(&prog->reloc_desc);
347 static char *__bpf_program__pin_name(struct bpf_program *prog)
351 name = p = strdup(prog->section_name);
352 while ((p = strchr(p, '/')))
359 bpf_program__init(void *data, size_t size, char *section_name, int idx,
360 struct bpf_program *prog)
362 const size_t bpf_insn_sz = sizeof(struct bpf_insn);
364 if (size == 0 || size % bpf_insn_sz) {
365 pr_warning("corrupted section '%s', size: %zu\n",
370 memset(prog, 0, sizeof(*prog));
372 prog->section_name = strdup(section_name);
373 if (!prog->section_name) {
374 pr_warning("failed to alloc name for prog under section(%d) %s\n",
379 prog->pin_name = __bpf_program__pin_name(prog);
380 if (!prog->pin_name) {
381 pr_warning("failed to alloc pin name for prog under section(%d) %s\n",
386 prog->insns = malloc(size);
388 pr_warning("failed to alloc insns for prog under section %s\n",
392 prog->insns_cnt = size / bpf_insn_sz;
393 memcpy(prog->insns, data, size);
395 prog->instances.fds = NULL;
396 prog->instances.nr = -1;
397 prog->type = BPF_PROG_TYPE_UNSPEC;
401 bpf_program__exit(prog);
406 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
407 char *section_name, int idx)
409 struct bpf_program prog, *progs;
412 err = bpf_program__init(data, size, section_name, idx, &prog);
416 prog.caps = &obj->caps;
417 progs = obj->programs;
418 nr_progs = obj->nr_programs;
420 progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
423 * In this case the original obj->programs
424 * is still valid, so don't need special treat for
425 * bpf_close_object().
427 pr_warning("failed to alloc a new program under section '%s'\n",
429 bpf_program__exit(&prog);
433 pr_debug("found program %s\n", prog.section_name);
434 obj->programs = progs;
435 obj->nr_programs = nr_progs + 1;
437 progs[nr_progs] = prog;
442 bpf_object__init_prog_names(struct bpf_object *obj)
444 Elf_Data *symbols = obj->efile.symbols;
445 struct bpf_program *prog;
448 for (pi = 0; pi < obj->nr_programs; pi++) {
449 const char *name = NULL;
451 prog = &obj->programs[pi];
453 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
457 if (!gelf_getsym(symbols, si, &sym))
459 if (sym.st_shndx != prog->idx)
461 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
464 name = elf_strptr(obj->efile.elf,
465 obj->efile.strtabidx,
468 pr_warning("failed to get sym name string for prog %s\n",
470 return -LIBBPF_ERRNO__LIBELF;
474 if (!name && prog->idx == obj->efile.text_shndx)
478 pr_warning("failed to find sym for prog %s\n",
483 prog->name = strdup(name);
485 pr_warning("failed to allocate memory for prog sym %s\n",
494 static struct bpf_object *bpf_object__new(const char *path,
498 struct bpf_object *obj;
501 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
503 pr_warning("alloc memory failed for %s\n", path);
504 return ERR_PTR(-ENOMEM);
507 strcpy(obj->path, path);
508 /* Using basename() GNU version which doesn't modify arg. */
509 strncpy(obj->name, basename((void *)path), sizeof(obj->name) - 1);
510 end = strchr(obj->name, '.');
516 * Caller of this function should also call
517 * bpf_object__elf_finish() after data collection to return
518 * obj_buf to user. If not, we should duplicate the buffer to
519 * avoid user freeing them before elf finish.
521 obj->efile.obj_buf = obj_buf;
522 obj->efile.obj_buf_sz = obj_buf_sz;
523 obj->efile.maps_shndx = -1;
524 obj->efile.btf_maps_shndx = -1;
525 obj->efile.data_shndx = -1;
526 obj->efile.rodata_shndx = -1;
527 obj->efile.bss_shndx = -1;
531 INIT_LIST_HEAD(&obj->list);
532 list_add(&obj->list, &bpf_objects_list);
536 static void bpf_object__elf_finish(struct bpf_object *obj)
538 if (!obj_elf_valid(obj))
541 if (obj->efile.elf) {
542 elf_end(obj->efile.elf);
543 obj->efile.elf = NULL;
545 obj->efile.symbols = NULL;
546 obj->efile.data = NULL;
547 obj->efile.rodata = NULL;
548 obj->efile.bss = NULL;
550 zfree(&obj->efile.reloc);
551 obj->efile.nr_reloc = 0;
552 zclose(obj->efile.fd);
553 obj->efile.obj_buf = NULL;
554 obj->efile.obj_buf_sz = 0;
557 static int bpf_object__elf_init(struct bpf_object *obj)
562 if (obj_elf_valid(obj)) {
563 pr_warning("elf init: internal error\n");
564 return -LIBBPF_ERRNO__LIBELF;
567 if (obj->efile.obj_buf_sz > 0) {
569 * obj_buf should have been validated by
570 * bpf_object__open_buffer().
572 obj->efile.elf = elf_memory(obj->efile.obj_buf,
573 obj->efile.obj_buf_sz);
575 obj->efile.fd = open(obj->path, O_RDONLY);
576 if (obj->efile.fd < 0) {
577 char errmsg[STRERR_BUFSIZE], *cp;
580 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
581 pr_warning("failed to open %s: %s\n", obj->path, cp);
585 obj->efile.elf = elf_begin(obj->efile.fd,
586 LIBBPF_ELF_C_READ_MMAP, NULL);
589 if (!obj->efile.elf) {
590 pr_warning("failed to open %s as ELF file\n", obj->path);
591 err = -LIBBPF_ERRNO__LIBELF;
595 if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
596 pr_warning("failed to get EHDR from %s\n", obj->path);
597 err = -LIBBPF_ERRNO__FORMAT;
600 ep = &obj->efile.ehdr;
602 /* Old LLVM set e_machine to EM_NONE */
603 if (ep->e_type != ET_REL ||
604 (ep->e_machine && ep->e_machine != EM_BPF)) {
605 pr_warning("%s is not an eBPF object file\n", obj->path);
606 err = -LIBBPF_ERRNO__FORMAT;
612 bpf_object__elf_finish(obj);
616 static int bpf_object__check_endianness(struct bpf_object *obj)
618 #if __BYTE_ORDER == __LITTLE_ENDIAN
619 if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
621 #elif __BYTE_ORDER == __BIG_ENDIAN
622 if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
625 # error "Unrecognized __BYTE_ORDER__"
627 pr_warning("endianness mismatch.\n");
628 return -LIBBPF_ERRNO__ENDIAN;
632 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
634 memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
635 pr_debug("license of %s is %s\n", obj->path, obj->license);
640 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
644 if (size != sizeof(kver)) {
645 pr_warning("invalid kver section in %s\n", obj->path);
646 return -LIBBPF_ERRNO__FORMAT;
648 memcpy(&kver, data, sizeof(kver));
649 obj->kern_version = kver;
650 pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
654 static int compare_bpf_map(const void *_a, const void *_b)
656 const struct bpf_map *a = _a;
657 const struct bpf_map *b = _b;
659 if (a->sec_idx != b->sec_idx)
660 return a->sec_idx - b->sec_idx;
661 return a->sec_offset - b->sec_offset;
664 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
666 if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
667 type == BPF_MAP_TYPE_HASH_OF_MAPS)
672 static int bpf_object_search_section_size(const struct bpf_object *obj,
673 const char *name, size_t *d_size)
675 const GElf_Ehdr *ep = &obj->efile.ehdr;
676 Elf *elf = obj->efile.elf;
680 while ((scn = elf_nextscn(elf, scn)) != NULL) {
681 const char *sec_name;
686 if (gelf_getshdr(scn, &sh) != &sh) {
687 pr_warning("failed to get section(%d) header from %s\n",
692 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
694 pr_warning("failed to get section(%d) name from %s\n",
699 if (strcmp(name, sec_name))
702 data = elf_getdata(scn, 0);
704 pr_warning("failed to get section(%d) data from %s(%s)\n",
705 idx, name, obj->path);
709 *d_size = data->d_size;
716 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
725 } else if (!strcmp(name, ".data")) {
727 *size = obj->efile.data->d_size;
728 } else if (!strcmp(name, ".bss")) {
730 *size = obj->efile.bss->d_size;
731 } else if (!strcmp(name, ".rodata")) {
732 if (obj->efile.rodata)
733 *size = obj->efile.rodata->d_size;
735 ret = bpf_object_search_section_size(obj, name, &d_size);
740 return *size ? 0 : ret;
743 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
746 Elf_Data *symbols = obj->efile.symbols;
753 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
756 if (!gelf_getsym(symbols, si, &sym))
758 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
759 GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
762 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
765 pr_warning("failed to get sym name string for var %s\n",
769 if (strcmp(name, sname) == 0) {
778 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
780 struct bpf_map *new_maps;
784 if (obj->nr_maps < obj->maps_cap)
785 return &obj->maps[obj->nr_maps++];
787 new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
788 new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
790 pr_warning("alloc maps for object failed\n");
791 return ERR_PTR(-ENOMEM);
794 obj->maps_cap = new_cap;
795 obj->maps = new_maps;
797 /* zero out new maps */
798 memset(obj->maps + obj->nr_maps, 0,
799 (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
801 * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
802 * when failure (zclose won't close negative fd)).
804 for (i = obj->nr_maps; i < obj->maps_cap; i++) {
805 obj->maps[i].fd = -1;
806 obj->maps[i].inner_map_fd = -1;
809 return &obj->maps[obj->nr_maps++];
813 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
814 int sec_idx, Elf_Data *data, void **data_buff)
816 char map_name[BPF_OBJ_NAME_LEN];
817 struct bpf_map_def *def;
820 map = bpf_object__add_map(obj);
824 map->libbpf_type = type;
825 map->sec_idx = sec_idx;
827 snprintf(map_name, sizeof(map_name), "%.8s%.7s", obj->name,
828 libbpf_type_to_btf_name[type]);
829 map->name = strdup(map_name);
831 pr_warning("failed to alloc map name\n");
834 pr_debug("map '%s' (global data): at sec_idx %d, offset %zu.\n",
835 map_name, map->sec_idx, map->sec_offset);
838 def->type = BPF_MAP_TYPE_ARRAY;
839 def->key_size = sizeof(int);
840 def->value_size = data->d_size;
841 def->max_entries = 1;
842 def->map_flags = type == LIBBPF_MAP_RODATA ? BPF_F_RDONLY_PROG : 0;
844 *data_buff = malloc(data->d_size);
847 pr_warning("failed to alloc map content buffer\n");
850 memcpy(*data_buff, data->d_buf, data->d_size);
853 pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
857 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
861 if (!obj->caps.global_data)
864 * Populate obj->maps with libbpf internal maps.
866 if (obj->efile.data_shndx >= 0) {
867 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
868 obj->efile.data_shndx,
870 &obj->sections.data);
874 if (obj->efile.rodata_shndx >= 0) {
875 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
876 obj->efile.rodata_shndx,
878 &obj->sections.rodata);
882 if (obj->efile.bss_shndx >= 0) {
883 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
884 obj->efile.bss_shndx,
885 obj->efile.bss, NULL);
892 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
894 Elf_Data *symbols = obj->efile.symbols;
895 int i, map_def_sz = 0, nr_maps = 0, nr_syms;
896 Elf_Data *data = NULL;
899 if (obj->efile.maps_shndx < 0)
905 scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
907 data = elf_getdata(scn, NULL);
909 pr_warning("failed to get Elf_Data from map section %d\n",
910 obj->efile.maps_shndx);
915 * Count number of maps. Each map has a name.
916 * Array of maps is not supported: only the first element is
919 * TODO: Detect array of map and report error.
921 nr_syms = symbols->d_size / sizeof(GElf_Sym);
922 for (i = 0; i < nr_syms; i++) {
925 if (!gelf_getsym(symbols, i, &sym))
927 if (sym.st_shndx != obj->efile.maps_shndx)
931 /* Assume equally sized map definitions */
932 pr_debug("maps in %s: %d maps in %zd bytes\n",
933 obj->path, nr_maps, data->d_size);
935 map_def_sz = data->d_size / nr_maps;
936 if (!data->d_size || (data->d_size % nr_maps) != 0) {
937 pr_warning("unable to determine map definition size "
938 "section %s, %d maps in %zd bytes\n",
939 obj->path, nr_maps, data->d_size);
943 /* Fill obj->maps using data in "maps" section. */
944 for (i = 0; i < nr_syms; i++) {
946 const char *map_name;
947 struct bpf_map_def *def;
950 if (!gelf_getsym(symbols, i, &sym))
952 if (sym.st_shndx != obj->efile.maps_shndx)
955 map = bpf_object__add_map(obj);
959 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
962 pr_warning("failed to get map #%d name sym string for obj %s\n",
964 return -LIBBPF_ERRNO__FORMAT;
967 map->libbpf_type = LIBBPF_MAP_UNSPEC;
968 map->sec_idx = sym.st_shndx;
969 map->sec_offset = sym.st_value;
970 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
971 map_name, map->sec_idx, map->sec_offset);
972 if (sym.st_value + map_def_sz > data->d_size) {
973 pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
974 obj->path, map_name);
978 map->name = strdup(map_name);
980 pr_warning("failed to alloc map name\n");
983 pr_debug("map %d is \"%s\"\n", i, map->name);
984 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
986 * If the definition of the map in the object file fits in
987 * bpf_map_def, copy it. Any extra fields in our version
988 * of bpf_map_def will default to zero as a result of the
991 if (map_def_sz <= sizeof(struct bpf_map_def)) {
992 memcpy(&map->def, def, map_def_sz);
995 * Here the map structure being read is bigger than what
996 * we expect, truncate if the excess bits are all zero.
997 * If they are not zero, reject this map as
1001 for (b = ((char *)def) + sizeof(struct bpf_map_def);
1002 b < ((char *)def) + map_def_sz; b++) {
1004 pr_warning("maps section in %s: \"%s\" "
1005 "has unrecognized, non-zero "
1007 obj->path, map_name);
1012 memcpy(&map->def, def, sizeof(struct bpf_map_def));
1018 static const struct btf_type *
1019 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1021 const struct btf_type *t = btf__type_by_id(btf, id);
1026 while (btf_is_mod(t) || btf_is_typedef(t)) {
1029 t = btf__type_by_id(btf, t->type);
1036 * Fetch integer attribute of BTF map definition. Such attributes are
1037 * represented using a pointer to an array, in which dimensionality of array
1038 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1039 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1040 * type definition, while using only sizeof(void *) space in ELF data section.
1042 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1043 const struct btf_type *def,
1044 const struct btf_member *m, __u32 *res) {
1045 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
1046 const char *name = btf__name_by_offset(btf, m->name_off);
1047 const struct btf_array *arr_info;
1048 const struct btf_type *arr_t;
1050 if (!btf_is_ptr(t)) {
1051 pr_warning("map '%s': attr '%s': expected PTR, got %u.\n",
1052 map_name, name, btf_kind(t));
1056 arr_t = btf__type_by_id(btf, t->type);
1058 pr_warning("map '%s': attr '%s': type [%u] not found.\n",
1059 map_name, name, t->type);
1062 if (!btf_is_array(arr_t)) {
1063 pr_warning("map '%s': attr '%s': expected ARRAY, got %u.\n",
1064 map_name, name, btf_kind(arr_t));
1067 arr_info = btf_array(arr_t);
1068 *res = arr_info->nelems;
1072 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1073 const struct btf_type *sec,
1074 int var_idx, int sec_idx,
1075 const Elf_Data *data, bool strict)
1077 const struct btf_type *var, *def, *t;
1078 const struct btf_var_secinfo *vi;
1079 const struct btf_var *var_extra;
1080 const struct btf_member *m;
1081 const char *map_name;
1082 struct bpf_map *map;
1085 vi = btf_var_secinfos(sec) + var_idx;
1086 var = btf__type_by_id(obj->btf, vi->type);
1087 var_extra = btf_var(var);
1088 map_name = btf__name_by_offset(obj->btf, var->name_off);
1089 vlen = btf_vlen(var);
1091 if (map_name == NULL || map_name[0] == '\0') {
1092 pr_warning("map #%d: empty name.\n", var_idx);
1095 if ((__u64)vi->offset + vi->size > data->d_size) {
1096 pr_warning("map '%s' BTF data is corrupted.\n", map_name);
1099 if (!btf_is_var(var)) {
1100 pr_warning("map '%s': unexpected var kind %u.\n",
1101 map_name, btf_kind(var));
1104 if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1105 var_extra->linkage != BTF_VAR_STATIC) {
1106 pr_warning("map '%s': unsupported var linkage %u.\n",
1107 map_name, var_extra->linkage);
1111 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
1112 if (!btf_is_struct(def)) {
1113 pr_warning("map '%s': unexpected def kind %u.\n",
1114 map_name, btf_kind(var));
1117 if (def->size > vi->size) {
1118 pr_warning("map '%s': invalid def size.\n", map_name);
1122 map = bpf_object__add_map(obj);
1124 return PTR_ERR(map);
1125 map->name = strdup(map_name);
1127 pr_warning("map '%s': failed to alloc map name.\n", map_name);
1130 map->libbpf_type = LIBBPF_MAP_UNSPEC;
1131 map->def.type = BPF_MAP_TYPE_UNSPEC;
1132 map->sec_idx = sec_idx;
1133 map->sec_offset = vi->offset;
1134 pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1135 map_name, map->sec_idx, map->sec_offset);
1137 vlen = btf_vlen(def);
1138 m = btf_members(def);
1139 for (i = 0; i < vlen; i++, m++) {
1140 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1143 pr_warning("map '%s': invalid field #%d.\n",
1147 if (strcmp(name, "type") == 0) {
1148 if (!get_map_field_int(map_name, obj->btf, def, m,
1151 pr_debug("map '%s': found type = %u.\n",
1152 map_name, map->def.type);
1153 } else if (strcmp(name, "max_entries") == 0) {
1154 if (!get_map_field_int(map_name, obj->btf, def, m,
1155 &map->def.max_entries))
1157 pr_debug("map '%s': found max_entries = %u.\n",
1158 map_name, map->def.max_entries);
1159 } else if (strcmp(name, "map_flags") == 0) {
1160 if (!get_map_field_int(map_name, obj->btf, def, m,
1161 &map->def.map_flags))
1163 pr_debug("map '%s': found map_flags = %u.\n",
1164 map_name, map->def.map_flags);
1165 } else if (strcmp(name, "key_size") == 0) {
1168 if (!get_map_field_int(map_name, obj->btf, def, m,
1171 pr_debug("map '%s': found key_size = %u.\n",
1173 if (map->def.key_size && map->def.key_size != sz) {
1174 pr_warning("map '%s': conflicting key size %u != %u.\n",
1175 map_name, map->def.key_size, sz);
1178 map->def.key_size = sz;
1179 } else if (strcmp(name, "key") == 0) {
1182 t = btf__type_by_id(obj->btf, m->type);
1184 pr_warning("map '%s': key type [%d] not found.\n",
1188 if (!btf_is_ptr(t)) {
1189 pr_warning("map '%s': key spec is not PTR: %u.\n",
1190 map_name, btf_kind(t));
1193 sz = btf__resolve_size(obj->btf, t->type);
1195 pr_warning("map '%s': can't determine key size for type [%u]: %lld.\n",
1196 map_name, t->type, sz);
1199 pr_debug("map '%s': found key [%u], sz = %lld.\n",
1200 map_name, t->type, sz);
1201 if (map->def.key_size && map->def.key_size != sz) {
1202 pr_warning("map '%s': conflicting key size %u != %lld.\n",
1203 map_name, map->def.key_size, sz);
1206 map->def.key_size = sz;
1207 map->btf_key_type_id = t->type;
1208 } else if (strcmp(name, "value_size") == 0) {
1211 if (!get_map_field_int(map_name, obj->btf, def, m,
1214 pr_debug("map '%s': found value_size = %u.\n",
1216 if (map->def.value_size && map->def.value_size != sz) {
1217 pr_warning("map '%s': conflicting value size %u != %u.\n",
1218 map_name, map->def.value_size, sz);
1221 map->def.value_size = sz;
1222 } else if (strcmp(name, "value") == 0) {
1225 t = btf__type_by_id(obj->btf, m->type);
1227 pr_warning("map '%s': value type [%d] not found.\n",
1231 if (!btf_is_ptr(t)) {
1232 pr_warning("map '%s': value spec is not PTR: %u.\n",
1233 map_name, btf_kind(t));
1236 sz = btf__resolve_size(obj->btf, t->type);
1238 pr_warning("map '%s': can't determine value size for type [%u]: %lld.\n",
1239 map_name, t->type, sz);
1242 pr_debug("map '%s': found value [%u], sz = %lld.\n",
1243 map_name, t->type, sz);
1244 if (map->def.value_size && map->def.value_size != sz) {
1245 pr_warning("map '%s': conflicting value size %u != %lld.\n",
1246 map_name, map->def.value_size, sz);
1249 map->def.value_size = sz;
1250 map->btf_value_type_id = t->type;
1253 pr_warning("map '%s': unknown field '%s'.\n",
1257 pr_debug("map '%s': ignoring unknown field '%s'.\n",
1262 if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
1263 pr_warning("map '%s': map type isn't specified.\n", map_name);
1270 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict)
1272 const struct btf_type *sec = NULL;
1273 int nr_types, i, vlen, err;
1274 const struct btf_type *t;
1279 if (obj->efile.btf_maps_shndx < 0)
1282 scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
1284 data = elf_getdata(scn, NULL);
1285 if (!scn || !data) {
1286 pr_warning("failed to get Elf_Data from map section %d (%s)\n",
1287 obj->efile.maps_shndx, MAPS_ELF_SEC);
1291 nr_types = btf__get_nr_types(obj->btf);
1292 for (i = 1; i <= nr_types; i++) {
1293 t = btf__type_by_id(obj->btf, i);
1294 if (!btf_is_datasec(t))
1296 name = btf__name_by_offset(obj->btf, t->name_off);
1297 if (strcmp(name, MAPS_ELF_SEC) == 0) {
1304 pr_warning("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
1308 vlen = btf_vlen(sec);
1309 for (i = 0; i < vlen; i++) {
1310 err = bpf_object__init_user_btf_map(obj, sec, i,
1311 obj->efile.btf_maps_shndx,
1320 static int bpf_object__init_maps(struct bpf_object *obj, int flags)
1322 bool strict = !(flags & MAPS_RELAX_COMPAT);
1325 err = bpf_object__init_user_maps(obj, strict);
1329 err = bpf_object__init_user_btf_maps(obj, strict);
1333 err = bpf_object__init_global_data_maps(obj);
1338 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]),
1344 static bool section_have_execinstr(struct bpf_object *obj, int idx)
1349 scn = elf_getscn(obj->efile.elf, idx);
1353 if (gelf_getshdr(scn, &sh) != &sh)
1356 if (sh.sh_flags & SHF_EXECINSTR)
1362 static void bpf_object__sanitize_btf(struct bpf_object *obj)
1364 bool has_datasec = obj->caps.btf_datasec;
1365 bool has_func = obj->caps.btf_func;
1366 struct btf *btf = obj->btf;
1370 if (!obj->btf || (has_func && has_datasec))
1373 for (i = 1; i <= btf__get_nr_types(btf); i++) {
1374 t = (struct btf_type *)btf__type_by_id(btf, i);
1376 if (!has_datasec && btf_is_var(t)) {
1377 /* replace VAR with INT */
1378 t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1380 * using size = 1 is the safest choice, 4 will be too
1381 * big and cause kernel BTF validation failure if
1382 * original variable took less than 4 bytes
1385 *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
1386 } else if (!has_datasec && btf_is_datasec(t)) {
1387 /* replace DATASEC with STRUCT */
1388 const struct btf_var_secinfo *v = btf_var_secinfos(t);
1389 struct btf_member *m = btf_members(t);
1390 struct btf_type *vt;
1393 name = (char *)btf__name_by_offset(btf, t->name_off);
1401 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
1402 for (j = 0; j < vlen; j++, v++, m++) {
1403 /* order of field assignments is important */
1404 m->offset = v->offset * 8;
1406 /* preserve variable name as member name */
1407 vt = (void *)btf__type_by_id(btf, v->type);
1408 m->name_off = vt->name_off;
1410 } else if (!has_func && btf_is_func_proto(t)) {
1411 /* replace FUNC_PROTO with ENUM */
1413 t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
1414 t->size = sizeof(__u32); /* kernel enforced */
1415 } else if (!has_func && btf_is_func(t)) {
1416 /* replace FUNC with TYPEDEF */
1417 t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
1422 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
1427 if (!obj->caps.btf_func) {
1428 btf_ext__free(obj->btf_ext);
1429 obj->btf_ext = NULL;
1433 static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
1435 return obj->efile.btf_maps_shndx >= 0;
1438 static int bpf_object__init_btf(struct bpf_object *obj,
1440 Elf_Data *btf_ext_data)
1442 bool btf_required = bpf_object__is_btf_mandatory(obj);
1446 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
1447 if (IS_ERR(obj->btf)) {
1448 pr_warning("Error loading ELF section %s: %d.\n",
1452 err = btf__finalize_data(obj, obj->btf);
1454 pr_warning("Error finalizing %s: %d.\n",
1461 pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
1462 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
1465 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
1466 btf_ext_data->d_size);
1467 if (IS_ERR(obj->btf_ext)) {
1468 pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
1469 BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
1470 obj->btf_ext = NULL;
1475 if (err || IS_ERR(obj->btf)) {
1477 err = err ? : PTR_ERR(obj->btf);
1480 if (!IS_ERR_OR_NULL(obj->btf))
1481 btf__free(obj->btf);
1484 if (btf_required && !obj->btf) {
1485 pr_warning("BTF is required, but is missing or corrupted.\n");
1486 return err == 0 ? -ENOENT : err;
1491 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
1498 bpf_object__sanitize_btf(obj);
1499 bpf_object__sanitize_btf_ext(obj);
1501 err = btf__load(obj->btf);
1503 pr_warning("Error loading %s into kernel: %d.\n",
1505 btf__free(obj->btf);
1507 /* btf_ext can't exist without btf, so free it as well */
1509 btf_ext__free(obj->btf_ext);
1510 obj->btf_ext = NULL;
1513 if (bpf_object__is_btf_mandatory(obj))
1519 static int bpf_object__elf_collect(struct bpf_object *obj, int flags)
1521 Elf *elf = obj->efile.elf;
1522 GElf_Ehdr *ep = &obj->efile.ehdr;
1523 Elf_Data *btf_ext_data = NULL;
1524 Elf_Data *btf_data = NULL;
1525 Elf_Scn *scn = NULL;
1526 int idx = 0, err = 0;
1528 /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1529 if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
1530 pr_warning("failed to get e_shstrndx from %s\n", obj->path);
1531 return -LIBBPF_ERRNO__FORMAT;
1534 while ((scn = elf_nextscn(elf, scn)) != NULL) {
1540 if (gelf_getshdr(scn, &sh) != &sh) {
1541 pr_warning("failed to get section(%d) header from %s\n",
1543 return -LIBBPF_ERRNO__FORMAT;
1546 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1548 pr_warning("failed to get section(%d) name from %s\n",
1550 return -LIBBPF_ERRNO__FORMAT;
1553 data = elf_getdata(scn, 0);
1555 pr_warning("failed to get section(%d) data from %s(%s)\n",
1556 idx, name, obj->path);
1557 return -LIBBPF_ERRNO__FORMAT;
1559 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
1560 idx, name, (unsigned long)data->d_size,
1561 (int)sh.sh_link, (unsigned long)sh.sh_flags,
1564 if (strcmp(name, "license") == 0) {
1565 err = bpf_object__init_license(obj,
1570 } else if (strcmp(name, "version") == 0) {
1571 err = bpf_object__init_kversion(obj,
1576 } else if (strcmp(name, "maps") == 0) {
1577 obj->efile.maps_shndx = idx;
1578 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
1579 obj->efile.btf_maps_shndx = idx;
1580 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
1582 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
1583 btf_ext_data = data;
1584 } else if (sh.sh_type == SHT_SYMTAB) {
1585 if (obj->efile.symbols) {
1586 pr_warning("bpf: multiple SYMTAB in %s\n",
1588 return -LIBBPF_ERRNO__FORMAT;
1590 obj->efile.symbols = data;
1591 obj->efile.strtabidx = sh.sh_link;
1592 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
1593 if (sh.sh_flags & SHF_EXECINSTR) {
1594 if (strcmp(name, ".text") == 0)
1595 obj->efile.text_shndx = idx;
1596 err = bpf_object__add_program(obj, data->d_buf,
1597 data->d_size, name, idx);
1599 char errmsg[STRERR_BUFSIZE];
1600 char *cp = libbpf_strerror_r(-err, errmsg,
1603 pr_warning("failed to alloc program %s (%s): %s",
1604 name, obj->path, cp);
1607 } else if (strcmp(name, ".data") == 0) {
1608 obj->efile.data = data;
1609 obj->efile.data_shndx = idx;
1610 } else if (strcmp(name, ".rodata") == 0) {
1611 obj->efile.rodata = data;
1612 obj->efile.rodata_shndx = idx;
1614 pr_debug("skip section(%d) %s\n", idx, name);
1616 } else if (sh.sh_type == SHT_REL) {
1617 int nr_reloc = obj->efile.nr_reloc;
1618 void *reloc = obj->efile.reloc;
1619 int sec = sh.sh_info; /* points to other section */
1621 /* Only do relo for section with exec instructions */
1622 if (!section_have_execinstr(obj, sec)) {
1623 pr_debug("skip relo %s(%d) for section(%d)\n",
1628 reloc = reallocarray(reloc, nr_reloc + 1,
1629 sizeof(*obj->efile.reloc));
1631 pr_warning("realloc failed\n");
1635 obj->efile.reloc = reloc;
1636 obj->efile.nr_reloc++;
1638 obj->efile.reloc[nr_reloc].shdr = sh;
1639 obj->efile.reloc[nr_reloc].data = data;
1640 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, ".bss") == 0) {
1641 obj->efile.bss = data;
1642 obj->efile.bss_shndx = idx;
1644 pr_debug("skip section(%d) %s\n", idx, name);
1648 if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
1649 pr_warning("Corrupted ELF file: index of strtab invalid\n");
1650 return -LIBBPF_ERRNO__FORMAT;
1652 err = bpf_object__init_btf(obj, btf_data, btf_ext_data);
1654 err = bpf_object__init_maps(obj, flags);
1656 err = bpf_object__sanitize_and_load_btf(obj);
1658 err = bpf_object__init_prog_names(obj);
1662 static struct bpf_program *
1663 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
1665 struct bpf_program *prog;
1668 for (i = 0; i < obj->nr_programs; i++) {
1669 prog = &obj->programs[i];
1670 if (prog->idx == idx)
1676 struct bpf_program *
1677 bpf_object__find_program_by_title(const struct bpf_object *obj,
1680 struct bpf_program *pos;
1682 bpf_object__for_each_program(pos, obj) {
1683 if (pos->section_name && !strcmp(pos->section_name, title))
1689 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
1692 return shndx == obj->efile.data_shndx ||
1693 shndx == obj->efile.bss_shndx ||
1694 shndx == obj->efile.rodata_shndx;
1697 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
1700 return shndx == obj->efile.maps_shndx ||
1701 shndx == obj->efile.btf_maps_shndx;
1704 static bool bpf_object__relo_in_known_section(const struct bpf_object *obj,
1707 return shndx == obj->efile.text_shndx ||
1708 bpf_object__shndx_is_maps(obj, shndx) ||
1709 bpf_object__shndx_is_data(obj, shndx);
1712 static enum libbpf_map_type
1713 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
1715 if (shndx == obj->efile.data_shndx)
1716 return LIBBPF_MAP_DATA;
1717 else if (shndx == obj->efile.bss_shndx)
1718 return LIBBPF_MAP_BSS;
1719 else if (shndx == obj->efile.rodata_shndx)
1720 return LIBBPF_MAP_RODATA;
1722 return LIBBPF_MAP_UNSPEC;
1726 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
1727 Elf_Data *data, struct bpf_object *obj)
1729 Elf_Data *symbols = obj->efile.symbols;
1730 struct bpf_map *maps = obj->maps;
1731 size_t nr_maps = obj->nr_maps;
1734 pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
1735 nrels = shdr->sh_size / shdr->sh_entsize;
1737 prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
1738 if (!prog->reloc_desc) {
1739 pr_warning("failed to alloc memory in relocation\n");
1742 prog->nr_reloc = nrels;
1744 for (i = 0; i < nrels; i++) {
1745 struct bpf_insn *insns = prog->insns;
1746 enum libbpf_map_type type;
1747 unsigned int insn_idx;
1748 unsigned int shdr_idx;
1754 if (!gelf_getrel(data, i, &rel)) {
1755 pr_warning("relocation: failed to get %d reloc\n", i);
1756 return -LIBBPF_ERRNO__FORMAT;
1759 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
1760 pr_warning("relocation: symbol %"PRIx64" not found\n",
1761 GELF_R_SYM(rel.r_info));
1762 return -LIBBPF_ERRNO__FORMAT;
1765 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1766 sym.st_name) ? : "<?>";
1768 pr_debug("relo for %lld value %lld name %d (\'%s\')\n",
1769 (long long) (rel.r_info >> 32),
1770 (long long) sym.st_value, sym.st_name, name);
1772 shdr_idx = sym.st_shndx;
1773 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
1774 pr_debug("relocation: insn_idx=%u, shdr_idx=%u\n",
1775 insn_idx, shdr_idx);
1777 if (shdr_idx >= SHN_LORESERVE) {
1778 pr_warning("relocation: not yet supported relo for non-static global \'%s\' variable in special section (0x%x) found in insns[%d].code 0x%x\n",
1779 name, shdr_idx, insn_idx,
1780 insns[insn_idx].code);
1781 return -LIBBPF_ERRNO__RELOC;
1783 if (!bpf_object__relo_in_known_section(obj, shdr_idx)) {
1784 pr_warning("Program '%s' contains unrecognized relo data pointing to section %u\n",
1785 prog->section_name, shdr_idx);
1786 return -LIBBPF_ERRNO__RELOC;
1789 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
1790 if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
1791 pr_warning("incorrect bpf_call opcode\n");
1792 return -LIBBPF_ERRNO__RELOC;
1794 prog->reloc_desc[i].type = RELO_CALL;
1795 prog->reloc_desc[i].insn_idx = insn_idx;
1796 prog->reloc_desc[i].text_off = sym.st_value;
1797 obj->has_pseudo_calls = true;
1801 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
1802 pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
1803 insn_idx, insns[insn_idx].code);
1804 return -LIBBPF_ERRNO__RELOC;
1807 if (bpf_object__shndx_is_maps(obj, shdr_idx) ||
1808 bpf_object__shndx_is_data(obj, shdr_idx)) {
1809 type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
1810 if (type != LIBBPF_MAP_UNSPEC) {
1811 if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL) {
1812 pr_warning("bpf: relocation: not yet supported relo for non-static global \'%s\' variable found in insns[%d].code 0x%x\n",
1813 name, insn_idx, insns[insn_idx].code);
1814 return -LIBBPF_ERRNO__RELOC;
1816 if (!obj->caps.global_data) {
1817 pr_warning("bpf: relocation: kernel does not support global \'%s\' variable access in insns[%d]\n",
1819 return -LIBBPF_ERRNO__RELOC;
1823 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
1824 if (maps[map_idx].libbpf_type != type)
1826 if (type != LIBBPF_MAP_UNSPEC ||
1827 (maps[map_idx].sec_idx == sym.st_shndx &&
1828 maps[map_idx].sec_offset == sym.st_value)) {
1829 pr_debug("relocation: found map %zd (%s, sec_idx %d, offset %zu) for insn %u\n",
1830 map_idx, maps[map_idx].name,
1831 maps[map_idx].sec_idx,
1832 maps[map_idx].sec_offset,
1838 if (map_idx >= nr_maps) {
1839 pr_warning("bpf relocation: map_idx %d larger than %d\n",
1840 (int)map_idx, (int)nr_maps - 1);
1841 return -LIBBPF_ERRNO__RELOC;
1844 prog->reloc_desc[i].type = type != LIBBPF_MAP_UNSPEC ?
1845 RELO_DATA : RELO_LD64;
1846 prog->reloc_desc[i].insn_idx = insn_idx;
1847 prog->reloc_desc[i].map_idx = map_idx;
1853 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
1855 struct bpf_map_def *def = &map->def;
1856 __u32 key_type_id = 0, value_type_id = 0;
1859 /* if it's BTF-defined map, we don't need to search for type IDs */
1860 if (map->sec_idx == obj->efile.btf_maps_shndx)
1863 if (!bpf_map__is_internal(map)) {
1864 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
1865 def->value_size, &key_type_id,
1869 * LLVM annotates global data differently in BTF, that is,
1870 * only as '.data', '.bss' or '.rodata'.
1872 ret = btf__find_by_name(obj->btf,
1873 libbpf_type_to_btf_name[map->libbpf_type]);
1878 map->btf_key_type_id = key_type_id;
1879 map->btf_value_type_id = bpf_map__is_internal(map) ?
1880 ret : value_type_id;
1884 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1886 struct bpf_map_info info = {};
1887 __u32 len = sizeof(info);
1891 err = bpf_obj_get_info_by_fd(fd, &info, &len);
1895 new_name = strdup(info.name);
1899 new_fd = open("/", O_RDONLY | O_CLOEXEC);
1901 goto err_free_new_name;
1903 new_fd = dup3(fd, new_fd, O_CLOEXEC);
1905 goto err_close_new_fd;
1907 err = zclose(map->fd);
1909 goto err_close_new_fd;
1913 map->name = new_name;
1914 map->def.type = info.type;
1915 map->def.key_size = info.key_size;
1916 map->def.value_size = info.value_size;
1917 map->def.max_entries = info.max_entries;
1918 map->def.map_flags = info.map_flags;
1919 map->btf_key_type_id = info.btf_key_type_id;
1920 map->btf_value_type_id = info.btf_value_type_id;
1931 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
1933 if (!map || !max_entries)
1936 /* If map already created, its attributes can't be changed. */
1940 map->def.max_entries = max_entries;
1946 bpf_object__probe_name(struct bpf_object *obj)
1948 struct bpf_load_program_attr attr;
1949 char *cp, errmsg[STRERR_BUFSIZE];
1950 struct bpf_insn insns[] = {
1951 BPF_MOV64_IMM(BPF_REG_0, 0),
1956 /* make sure basic loading works */
1958 memset(&attr, 0, sizeof(attr));
1959 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
1961 attr.insns_cnt = ARRAY_SIZE(insns);
1962 attr.license = "GPL";
1964 ret = bpf_load_program_xattr(&attr, NULL, 0);
1966 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1967 pr_warning("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
1968 __func__, cp, errno);
1973 /* now try the same program, but with the name */
1976 ret = bpf_load_program_xattr(&attr, NULL, 0);
1986 bpf_object__probe_global_data(struct bpf_object *obj)
1988 struct bpf_load_program_attr prg_attr;
1989 struct bpf_create_map_attr map_attr;
1990 char *cp, errmsg[STRERR_BUFSIZE];
1991 struct bpf_insn insns[] = {
1992 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
1993 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
1994 BPF_MOV64_IMM(BPF_REG_0, 0),
1999 memset(&map_attr, 0, sizeof(map_attr));
2000 map_attr.map_type = BPF_MAP_TYPE_ARRAY;
2001 map_attr.key_size = sizeof(int);
2002 map_attr.value_size = 32;
2003 map_attr.max_entries = 1;
2005 map = bpf_create_map_xattr(&map_attr);
2007 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2008 pr_warning("Error in %s():%s(%d). Couldn't create simple array map.\n",
2009 __func__, cp, errno);
2015 memset(&prg_attr, 0, sizeof(prg_attr));
2016 prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
2017 prg_attr.insns = insns;
2018 prg_attr.insns_cnt = ARRAY_SIZE(insns);
2019 prg_attr.license = "GPL";
2021 ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
2023 obj->caps.global_data = 1;
2031 static int bpf_object__probe_btf_func(struct bpf_object *obj)
2033 const char strs[] = "\0int\0x\0a";
2034 /* void x(int a) {} */
2037 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
2038 /* FUNC_PROTO */ /* [2] */
2039 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
2040 BTF_PARAM_ENC(7, 1),
2041 /* FUNC x */ /* [3] */
2042 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
2046 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2047 strs, sizeof(strs));
2049 obj->caps.btf_func = 1;
2057 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
2059 const char strs[] = "\0x\0.data";
2063 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
2064 /* VAR x */ /* [2] */
2065 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
2067 /* DATASEC val */ /* [3] */
2068 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
2069 BTF_VAR_SECINFO_ENC(2, 0, 4),
2073 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2074 strs, sizeof(strs));
2076 obj->caps.btf_datasec = 1;
2085 bpf_object__probe_caps(struct bpf_object *obj)
2087 int (*probe_fn[])(struct bpf_object *obj) = {
2088 bpf_object__probe_name,
2089 bpf_object__probe_global_data,
2090 bpf_object__probe_btf_func,
2091 bpf_object__probe_btf_datasec,
2095 for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
2096 ret = probe_fn[i](obj);
2098 pr_debug("Probe #%d failed with %d.\n", i, ret);
2105 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
2107 char *cp, errmsg[STRERR_BUFSIZE];
2111 /* Nothing to do here since kernel already zero-initializes .bss map. */
2112 if (map->libbpf_type == LIBBPF_MAP_BSS)
2115 data = map->libbpf_type == LIBBPF_MAP_DATA ?
2116 obj->sections.data : obj->sections.rodata;
2118 err = bpf_map_update_elem(map->fd, &zero, data, 0);
2119 /* Freeze .rodata map as read-only from syscall side. */
2120 if (!err && map->libbpf_type == LIBBPF_MAP_RODATA) {
2121 err = bpf_map_freeze(map->fd);
2123 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2124 pr_warning("Error freezing map(%s) as read-only: %s\n",
2133 bpf_object__create_maps(struct bpf_object *obj)
2135 struct bpf_create_map_attr create_attr = {};
2140 for (i = 0; i < obj->nr_maps; i++) {
2141 struct bpf_map *map = &obj->maps[i];
2142 struct bpf_map_def *def = &map->def;
2143 char *cp, errmsg[STRERR_BUFSIZE];
2144 int *pfd = &map->fd;
2147 pr_debug("skip map create (preset) %s: fd=%d\n",
2148 map->name, map->fd);
2153 create_attr.name = map->name;
2154 create_attr.map_ifindex = map->map_ifindex;
2155 create_attr.map_type = def->type;
2156 create_attr.map_flags = def->map_flags;
2157 create_attr.key_size = def->key_size;
2158 create_attr.value_size = def->value_size;
2159 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
2160 !def->max_entries) {
2162 nr_cpus = libbpf_num_possible_cpus();
2164 pr_warning("failed to determine number of system CPUs: %d\n",
2169 pr_debug("map '%s': setting size to %d\n",
2170 map->name, nr_cpus);
2171 create_attr.max_entries = nr_cpus;
2173 create_attr.max_entries = def->max_entries;
2175 create_attr.btf_fd = 0;
2176 create_attr.btf_key_type_id = 0;
2177 create_attr.btf_value_type_id = 0;
2178 if (bpf_map_type__is_map_in_map(def->type) &&
2179 map->inner_map_fd >= 0)
2180 create_attr.inner_map_fd = map->inner_map_fd;
2182 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
2183 create_attr.btf_fd = btf__fd(obj->btf);
2184 create_attr.btf_key_type_id = map->btf_key_type_id;
2185 create_attr.btf_value_type_id = map->btf_value_type_id;
2188 *pfd = bpf_create_map_xattr(&create_attr);
2189 if (*pfd < 0 && (create_attr.btf_key_type_id ||
2190 create_attr.btf_value_type_id)) {
2192 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2193 pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
2194 map->name, cp, err);
2195 create_attr.btf_fd = 0;
2196 create_attr.btf_key_type_id = 0;
2197 create_attr.btf_value_type_id = 0;
2198 map->btf_key_type_id = 0;
2199 map->btf_value_type_id = 0;
2200 *pfd = bpf_create_map_xattr(&create_attr);
2208 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2209 pr_warning("failed to create map (name: '%s'): %s(%d)\n",
2210 map->name, cp, err);
2211 for (j = 0; j < i; j++)
2212 zclose(obj->maps[j].fd);
2216 if (bpf_map__is_internal(map)) {
2217 err = bpf_object__populate_internal_map(obj, map);
2224 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
2231 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
2232 void *btf_prog_info, const char *info_name)
2234 if (err != -ENOENT) {
2235 pr_warning("Error in loading %s for sec %s.\n",
2236 info_name, prog->section_name);
2240 /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
2242 if (btf_prog_info) {
2244 * Some info has already been found but has problem
2245 * in the last btf_ext reloc. Must have to error out.
2247 pr_warning("Error in relocating %s for sec %s.\n",
2248 info_name, prog->section_name);
2252 /* Have problem loading the very first info. Ignore the rest. */
2253 pr_warning("Cannot find %s for main program sec %s. Ignore all %s.\n",
2254 info_name, prog->section_name, info_name);
2259 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
2260 const char *section_name, __u32 insn_offset)
2264 if (!insn_offset || prog->func_info) {
2266 * !insn_offset => main program
2268 * For sub prog, the main program's func_info has to
2269 * be loaded first (i.e. prog->func_info != NULL)
2271 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
2272 section_name, insn_offset,
2274 &prog->func_info_cnt);
2276 return check_btf_ext_reloc_err(prog, err,
2280 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
2283 if (!insn_offset || prog->line_info) {
2284 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
2285 section_name, insn_offset,
2287 &prog->line_info_cnt);
2289 return check_btf_ext_reloc_err(prog, err,
2293 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
2299 #define BPF_CORE_SPEC_MAX_LEN 64
2301 /* represents BPF CO-RE field or array element accessor */
2302 struct bpf_core_accessor {
2303 __u32 type_id; /* struct/union type or array element type */
2304 __u32 idx; /* field index or array index */
2305 const char *name; /* field name or NULL for array accessor */
2308 struct bpf_core_spec {
2309 const struct btf *btf;
2310 /* high-level spec: named fields and array indices only */
2311 struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
2312 /* high-level spec length */
2314 /* raw, low-level spec: 1-to-1 with accessor spec string */
2315 int raw_spec[BPF_CORE_SPEC_MAX_LEN];
2316 /* raw spec length */
2318 /* field byte offset represented by spec */
2322 static bool str_is_empty(const char *s)
2328 * Turn bpf_offset_reloc into a low- and high-level spec representation,
2329 * validating correctness along the way, as well as calculating resulting
2330 * field offset (in bytes), specified by accessor string. Low-level spec
2331 * captures every single level of nestedness, including traversing anonymous
2332 * struct/union members. High-level one only captures semantically meaningful
2333 * "turning points": named fields and array indicies.
2334 * E.g., for this case:
2337 * int __unimportant;
2345 * struct sample *s = ...;
2347 * int x = &s->a[3]; // access string = '0:1:2:3'
2349 * Low-level spec has 1:1 mapping with each element of access string (it's
2350 * just a parsed access string representation): [0, 1, 2, 3].
2352 * High-level spec will capture only 3 points:
2353 * - intial zero-index access by pointer (&s->... is the same as &s[0]...);
2354 * - field 'a' access (corresponds to '2' in low-level spec);
2355 * - array element #3 access (corresponds to '3' in low-level spec).
2358 static int bpf_core_spec_parse(const struct btf *btf,
2360 const char *spec_str,
2361 struct bpf_core_spec *spec)
2363 int access_idx, parsed_len, i;
2364 const struct btf_type *t;
2369 if (str_is_empty(spec_str) || *spec_str == ':')
2372 memset(spec, 0, sizeof(*spec));
2375 /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
2377 if (*spec_str == ':')
2379 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
2381 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2383 spec_str += parsed_len;
2384 spec->raw_spec[spec->raw_len++] = access_idx;
2387 if (spec->raw_len == 0)
2390 /* first spec value is always reloc type array index */
2391 t = skip_mods_and_typedefs(btf, type_id, &id);
2395 access_idx = spec->raw_spec[0];
2396 spec->spec[0].type_id = id;
2397 spec->spec[0].idx = access_idx;
2400 sz = btf__resolve_size(btf, id);
2403 spec->offset = access_idx * sz;
2405 for (i = 1; i < spec->raw_len; i++) {
2406 t = skip_mods_and_typedefs(btf, id, &id);
2410 access_idx = spec->raw_spec[i];
2412 if (btf_is_composite(t)) {
2413 const struct btf_member *m;
2416 if (access_idx >= btf_vlen(t))
2418 if (btf_member_bitfield_size(t, access_idx))
2421 offset = btf_member_bit_offset(t, access_idx);
2424 spec->offset += offset / 8;
2426 m = btf_members(t) + access_idx;
2428 name = btf__name_by_offset(btf, m->name_off);
2429 if (str_is_empty(name))
2432 spec->spec[spec->len].type_id = id;
2433 spec->spec[spec->len].idx = access_idx;
2434 spec->spec[spec->len].name = name;
2439 } else if (btf_is_array(t)) {
2440 const struct btf_array *a = btf_array(t);
2442 t = skip_mods_and_typedefs(btf, a->type, &id);
2443 if (!t || access_idx >= a->nelems)
2446 spec->spec[spec->len].type_id = id;
2447 spec->spec[spec->len].idx = access_idx;
2450 sz = btf__resolve_size(btf, id);
2453 spec->offset += access_idx * sz;
2455 pr_warning("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
2456 type_id, spec_str, i, id, btf_kind(t));
2464 static bool bpf_core_is_flavor_sep(const char *s)
2466 /* check X___Y name pattern, where X and Y are not underscores */
2467 return s[0] != '_' && /* X */
2468 s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */
2469 s[4] != '_'; /* Y */
2472 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
2473 * before last triple underscore. Struct name part after last triple
2474 * underscore is ignored by BPF CO-RE relocation during relocation matching.
2476 static size_t bpf_core_essential_name_len(const char *name)
2478 size_t n = strlen(name);
2481 for (i = n - 5; i >= 0; i--) {
2482 if (bpf_core_is_flavor_sep(name + i))
2488 /* dynamically sized list of type IDs */
2494 static void bpf_core_free_cands(struct ids_vec *cand_ids)
2496 free(cand_ids->data);
2500 static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
2501 __u32 local_type_id,
2502 const struct btf *targ_btf)
2504 size_t local_essent_len, targ_essent_len;
2505 const char *local_name, *targ_name;
2506 const struct btf_type *t;
2507 struct ids_vec *cand_ids;
2511 t = btf__type_by_id(local_btf, local_type_id);
2513 return ERR_PTR(-EINVAL);
2515 local_name = btf__name_by_offset(local_btf, t->name_off);
2516 if (str_is_empty(local_name))
2517 return ERR_PTR(-EINVAL);
2518 local_essent_len = bpf_core_essential_name_len(local_name);
2520 cand_ids = calloc(1, sizeof(*cand_ids));
2522 return ERR_PTR(-ENOMEM);
2524 n = btf__get_nr_types(targ_btf);
2525 for (i = 1; i <= n; i++) {
2526 t = btf__type_by_id(targ_btf, i);
2527 targ_name = btf__name_by_offset(targ_btf, t->name_off);
2528 if (str_is_empty(targ_name))
2531 targ_essent_len = bpf_core_essential_name_len(targ_name);
2532 if (targ_essent_len != local_essent_len)
2535 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
2536 pr_debug("[%d] %s: found candidate [%d] %s\n",
2537 local_type_id, local_name, i, targ_name);
2538 new_ids = realloc(cand_ids->data, cand_ids->len + 1);
2543 cand_ids->data = new_ids;
2544 cand_ids->data[cand_ids->len++] = i;
2549 bpf_core_free_cands(cand_ids);
2550 return ERR_PTR(err);
2553 /* Check two types for compatibility, skipping const/volatile/restrict and
2554 * typedefs, to ensure we are relocating offset to the compatible entities:
2555 * - any two STRUCTs/UNIONs are compatible and can be mixed;
2556 * - any two FWDs are compatible;
2557 * - any two PTRs are always compatible;
2558 * - for ENUMs, check sizes, names are ignored;
2559 * - for INT, size and bitness should match, signedness is ignored;
2560 * - for ARRAY, dimensionality is ignored, element types are checked for
2561 * compatibility recursively;
2562 * - everything else shouldn't be ever a target of relocation.
2563 * These rules are not set in stone and probably will be adjusted as we get
2564 * more experience with using BPF CO-RE relocations.
2566 static int bpf_core_fields_are_compat(const struct btf *local_btf,
2568 const struct btf *targ_btf,
2571 const struct btf_type *local_type, *targ_type;
2574 local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
2575 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2576 if (!local_type || !targ_type)
2579 if (btf_is_composite(local_type) && btf_is_composite(targ_type))
2581 if (btf_kind(local_type) != btf_kind(targ_type))
2584 switch (btf_kind(local_type)) {
2589 return local_type->size == targ_type->size;
2591 return btf_int_offset(local_type) == 0 &&
2592 btf_int_offset(targ_type) == 0 &&
2593 local_type->size == targ_type->size &&
2594 btf_int_bits(local_type) == btf_int_bits(targ_type);
2595 case BTF_KIND_ARRAY:
2596 local_id = btf_array(local_type)->type;
2597 targ_id = btf_array(targ_type)->type;
2600 pr_warning("unexpected kind %d relocated, local [%d], target [%d]\n",
2601 btf_kind(local_type), local_id, targ_id);
2607 * Given single high-level named field accessor in local type, find
2608 * corresponding high-level accessor for a target type. Along the way,
2609 * maintain low-level spec for target as well. Also keep updating target
2612 * Searching is performed through recursive exhaustive enumeration of all
2613 * fields of a struct/union. If there are any anonymous (embedded)
2614 * structs/unions, they are recursively searched as well. If field with
2615 * desired name is found, check compatibility between local and target types,
2616 * before returning result.
2618 * 1 is returned, if field is found.
2619 * 0 is returned if no compatible field is found.
2620 * <0 is returned on error.
2622 static int bpf_core_match_member(const struct btf *local_btf,
2623 const struct bpf_core_accessor *local_acc,
2624 const struct btf *targ_btf,
2626 struct bpf_core_spec *spec,
2627 __u32 *next_targ_id)
2629 const struct btf_type *local_type, *targ_type;
2630 const struct btf_member *local_member, *m;
2631 const char *local_name, *targ_name;
2635 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2638 if (!btf_is_composite(targ_type))
2641 local_id = local_acc->type_id;
2642 local_type = btf__type_by_id(local_btf, local_id);
2643 local_member = btf_members(local_type) + local_acc->idx;
2644 local_name = btf__name_by_offset(local_btf, local_member->name_off);
2646 n = btf_vlen(targ_type);
2647 m = btf_members(targ_type);
2648 for (i = 0; i < n; i++, m++) {
2651 /* bitfield relocations not supported */
2652 if (btf_member_bitfield_size(targ_type, i))
2654 offset = btf_member_bit_offset(targ_type, i);
2658 /* too deep struct/union/array nesting */
2659 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2662 /* speculate this member will be the good one */
2663 spec->offset += offset / 8;
2664 spec->raw_spec[spec->raw_len++] = i;
2666 targ_name = btf__name_by_offset(targ_btf, m->name_off);
2667 if (str_is_empty(targ_name)) {
2668 /* embedded struct/union, we need to go deeper */
2669 found = bpf_core_match_member(local_btf, local_acc,
2671 spec, next_targ_id);
2672 if (found) /* either found or error */
2674 } else if (strcmp(local_name, targ_name) == 0) {
2675 /* matching named field */
2676 struct bpf_core_accessor *targ_acc;
2678 targ_acc = &spec->spec[spec->len++];
2679 targ_acc->type_id = targ_id;
2681 targ_acc->name = targ_name;
2683 *next_targ_id = m->type;
2684 found = bpf_core_fields_are_compat(local_btf,
2688 spec->len--; /* pop accessor */
2691 /* member turned out not to be what we looked for */
2692 spec->offset -= offset / 8;
2700 * Try to match local spec to a target type and, if successful, produce full
2701 * target spec (high-level, low-level + offset).
2703 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
2704 const struct btf *targ_btf, __u32 targ_id,
2705 struct bpf_core_spec *targ_spec)
2707 const struct btf_type *targ_type;
2708 const struct bpf_core_accessor *local_acc;
2709 struct bpf_core_accessor *targ_acc;
2712 memset(targ_spec, 0, sizeof(*targ_spec));
2713 targ_spec->btf = targ_btf;
2715 local_acc = &local_spec->spec[0];
2716 targ_acc = &targ_spec->spec[0];
2718 for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
2719 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
2724 if (local_acc->name) {
2725 matched = bpf_core_match_member(local_spec->btf,
2728 targ_spec, &targ_id);
2732 /* for i=0, targ_id is already treated as array element
2733 * type (because it's the original struct), for others
2734 * we should find array element type first
2737 const struct btf_array *a;
2739 if (!btf_is_array(targ_type))
2742 a = btf_array(targ_type);
2743 if (local_acc->idx >= a->nelems)
2745 if (!skip_mods_and_typedefs(targ_btf, a->type,
2750 /* too deep struct/union/array nesting */
2751 if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2754 targ_acc->type_id = targ_id;
2755 targ_acc->idx = local_acc->idx;
2756 targ_acc->name = NULL;
2758 targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
2759 targ_spec->raw_len++;
2761 sz = btf__resolve_size(targ_btf, targ_id);
2764 targ_spec->offset += local_acc->idx * sz;
2772 * Patch relocatable BPF instruction.
2773 * Expected insn->imm value is provided for validation, as well as the new
2776 * Currently three kinds of BPF instructions are supported:
2777 * 1. rX = <imm> (assignment with immediate operand);
2778 * 2. rX += <imm> (arithmetic operations with immediate operand);
2779 * 3. *(rX) = <imm> (indirect memory assignment with immediate operand).
2781 * If actual insn->imm value is wrong, bail out.
2783 static int bpf_core_reloc_insn(struct bpf_program *prog, int insn_off,
2784 __u32 orig_off, __u32 new_off)
2786 struct bpf_insn *insn;
2790 if (insn_off % sizeof(struct bpf_insn))
2792 insn_idx = insn_off / sizeof(struct bpf_insn);
2794 insn = &prog->insns[insn_idx];
2795 class = BPF_CLASS(insn->code);
2797 if (class == BPF_ALU || class == BPF_ALU64) {
2798 if (BPF_SRC(insn->code) != BPF_K)
2800 if (insn->imm != orig_off)
2802 insn->imm = new_off;
2803 pr_debug("prog '%s': patched insn #%d (ALU/ALU64) imm %d -> %d\n",
2804 bpf_program__title(prog, false),
2805 insn_idx, orig_off, new_off);
2807 pr_warning("prog '%s': trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
2808 bpf_program__title(prog, false),
2809 insn_idx, insn->code, insn->src_reg, insn->dst_reg,
2810 insn->off, insn->imm);
2816 static struct btf *btf_load_raw(const char *path)
2824 if (stat(path, &st))
2825 return ERR_PTR(-errno);
2827 data = malloc(st.st_size);
2829 return ERR_PTR(-ENOMEM);
2831 f = fopen(path, "rb");
2833 btf = ERR_PTR(-errno);
2837 read_cnt = fread(data, 1, st.st_size, f);
2839 if (read_cnt < st.st_size) {
2840 btf = ERR_PTR(-EBADF);
2844 btf = btf__new(data, read_cnt);
2852 * Probe few well-known locations for vmlinux kernel image and try to load BTF
2853 * data out of it to use for target BTF.
2855 static struct btf *bpf_core_find_kernel_btf(void)
2858 const char *path_fmt;
2861 /* try canonical vmlinux BTF through sysfs first */
2862 { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
2863 /* fall back to trying to find vmlinux ELF on disk otherwise */
2864 { "/boot/vmlinux-%1$s" },
2865 { "/lib/modules/%1$s/vmlinux-%1$s" },
2866 { "/lib/modules/%1$s/build/vmlinux" },
2867 { "/usr/lib/modules/%1$s/kernel/vmlinux" },
2868 { "/usr/lib/debug/boot/vmlinux-%1$s" },
2869 { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
2870 { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
2872 char path[PATH_MAX + 1];
2879 for (i = 0; i < ARRAY_SIZE(locations); i++) {
2880 snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
2882 if (access(path, R_OK))
2885 if (locations[i].raw_btf)
2886 btf = btf_load_raw(path);
2888 btf = btf__parse_elf(path, NULL);
2890 pr_debug("loading kernel BTF '%s': %ld\n",
2891 path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
2898 pr_warning("failed to find valid kernel BTF\n");
2899 return ERR_PTR(-ESRCH);
2902 /* Output spec definition in the format:
2903 * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
2904 * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
2906 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
2908 const struct btf_type *t;
2913 type_id = spec->spec[0].type_id;
2914 t = btf__type_by_id(spec->btf, type_id);
2915 s = btf__name_by_offset(spec->btf, t->name_off);
2916 libbpf_print(level, "[%u] %s + ", type_id, s);
2918 for (i = 0; i < spec->raw_len; i++)
2919 libbpf_print(level, "%d%s", spec->raw_spec[i],
2920 i == spec->raw_len - 1 ? " => " : ":");
2922 libbpf_print(level, "%u @ &x", spec->offset);
2924 for (i = 0; i < spec->len; i++) {
2925 if (spec->spec[i].name)
2926 libbpf_print(level, ".%s", spec->spec[i].name);
2928 libbpf_print(level, "[%u]", spec->spec[i].idx);
2933 static size_t bpf_core_hash_fn(const void *key, void *ctx)
2938 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
2943 static void *u32_as_hash_key(__u32 x)
2945 return (void *)(uintptr_t)x;
2949 * CO-RE relocate single instruction.
2951 * The outline and important points of the algorithm:
2952 * 1. For given local type, find corresponding candidate target types.
2953 * Candidate type is a type with the same "essential" name, ignoring
2954 * everything after last triple underscore (___). E.g., `sample`,
2955 * `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
2956 * for each other. Names with triple underscore are referred to as
2957 * "flavors" and are useful, among other things, to allow to
2958 * specify/support incompatible variations of the same kernel struct, which
2959 * might differ between different kernel versions and/or build
2962 * N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
2963 * converter, when deduplicated BTF of a kernel still contains more than
2964 * one different types with the same name. In that case, ___2, ___3, etc
2965 * are appended starting from second name conflict. But start flavors are
2966 * also useful to be defined "locally", in BPF program, to extract same
2967 * data from incompatible changes between different kernel
2968 * versions/configurations. For instance, to handle field renames between
2969 * kernel versions, one can use two flavors of the struct name with the
2970 * same common name and use conditional relocations to extract that field,
2971 * depending on target kernel version.
2972 * 2. For each candidate type, try to match local specification to this
2973 * candidate target type. Matching involves finding corresponding
2974 * high-level spec accessors, meaning that all named fields should match,
2975 * as well as all array accesses should be within the actual bounds. Also,
2976 * types should be compatible (see bpf_core_fields_are_compat for details).
2977 * 3. It is supported and expected that there might be multiple flavors
2978 * matching the spec. As long as all the specs resolve to the same set of
2979 * offsets across all candidates, there is not error. If there is any
2980 * ambiguity, CO-RE relocation will fail. This is necessary to accomodate
2981 * imprefection of BTF deduplication, which can cause slight duplication of
2982 * the same BTF type, if some directly or indirectly referenced (by
2983 * pointer) type gets resolved to different actual types in different
2984 * object files. If such situation occurs, deduplicated BTF will end up
2985 * with two (or more) structurally identical types, which differ only in
2986 * types they refer to through pointer. This should be OK in most cases and
2988 * 4. Candidate types search is performed by linearly scanning through all
2989 * types in target BTF. It is anticipated that this is overall more
2990 * efficient memory-wise and not significantly worse (if not better)
2991 * CPU-wise compared to prebuilding a map from all local type names to
2992 * a list of candidate type names. It's also sped up by caching resolved
2993 * list of matching candidates per each local "root" type ID, that has at
2994 * least one bpf_offset_reloc associated with it. This list is shared
2995 * between multiple relocations for the same type ID and is updated as some
2996 * of the candidates are pruned due to structural incompatibility.
2998 static int bpf_core_reloc_offset(struct bpf_program *prog,
2999 const struct bpf_offset_reloc *relo,
3001 const struct btf *local_btf,
3002 const struct btf *targ_btf,
3003 struct hashmap *cand_cache)
3005 const char *prog_name = bpf_program__title(prog, false);
3006 struct bpf_core_spec local_spec, cand_spec, targ_spec;
3007 const void *type_key = u32_as_hash_key(relo->type_id);
3008 const struct btf_type *local_type, *cand_type;
3009 const char *local_name, *cand_name;
3010 struct ids_vec *cand_ids;
3011 __u32 local_id, cand_id;
3012 const char *spec_str;
3015 local_id = relo->type_id;
3016 local_type = btf__type_by_id(local_btf, local_id);
3020 local_name = btf__name_by_offset(local_btf, local_type->name_off);
3021 if (str_is_empty(local_name))
3024 spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
3025 if (str_is_empty(spec_str))
3028 err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
3030 pr_warning("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
3031 prog_name, relo_idx, local_id, local_name, spec_str,
3036 pr_debug("prog '%s': relo #%d: spec is ", prog_name, relo_idx);
3037 bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
3038 libbpf_print(LIBBPF_DEBUG, "\n");
3040 if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
3041 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
3042 if (IS_ERR(cand_ids)) {
3043 pr_warning("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
3044 prog_name, relo_idx, local_id, local_name,
3046 return PTR_ERR(cand_ids);
3048 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
3050 bpf_core_free_cands(cand_ids);
3055 for (i = 0, j = 0; i < cand_ids->len; i++) {
3056 cand_id = cand_ids->data[i];
3057 cand_type = btf__type_by_id(targ_btf, cand_id);
3058 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
3060 err = bpf_core_spec_match(&local_spec, targ_btf,
3061 cand_id, &cand_spec);
3062 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
3063 prog_name, relo_idx, i, cand_name);
3064 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
3065 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
3067 pr_warning("prog '%s': relo #%d: matching error: %d\n",
3068 prog_name, relo_idx, err);
3075 targ_spec = cand_spec;
3076 } else if (cand_spec.offset != targ_spec.offset) {
3077 /* if there are many candidates, they should all
3078 * resolve to the same offset
3080 pr_warning("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
3081 prog_name, relo_idx, cand_spec.offset,
3086 cand_ids->data[j++] = cand_spec.spec[0].type_id;
3090 if (cand_ids->len == 0) {
3091 pr_warning("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
3092 prog_name, relo_idx, local_id, local_name, spec_str);
3096 err = bpf_core_reloc_insn(prog, relo->insn_off,
3097 local_spec.offset, targ_spec.offset);
3099 pr_warning("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
3100 prog_name, relo_idx, relo->insn_off, err);
3108 bpf_core_reloc_offsets(struct bpf_object *obj, const char *targ_btf_path)
3110 const struct btf_ext_info_sec *sec;
3111 const struct bpf_offset_reloc *rec;
3112 const struct btf_ext_info *seg;
3113 struct hashmap_entry *entry;
3114 struct hashmap *cand_cache = NULL;
3115 struct bpf_program *prog;
3116 struct btf *targ_btf;
3117 const char *sec_name;
3121 targ_btf = btf__parse_elf(targ_btf_path, NULL);
3123 targ_btf = bpf_core_find_kernel_btf();
3124 if (IS_ERR(targ_btf)) {
3125 pr_warning("failed to get target BTF: %ld\n",
3127 return PTR_ERR(targ_btf);
3130 cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
3131 if (IS_ERR(cand_cache)) {
3132 err = PTR_ERR(cand_cache);
3136 seg = &obj->btf_ext->offset_reloc_info;
3137 for_each_btf_ext_sec(seg, sec) {
3138 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
3139 if (str_is_empty(sec_name)) {
3143 prog = bpf_object__find_program_by_title(obj, sec_name);
3145 pr_warning("failed to find program '%s' for CO-RE offset relocation\n",
3151 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
3152 sec_name, sec->num_info);
3154 for_each_btf_ext_rec(seg, sec, i, rec) {
3155 err = bpf_core_reloc_offset(prog, rec, i, obj->btf,
3156 targ_btf, cand_cache);
3158 pr_warning("prog '%s': relo #%d: failed to relocate: %d\n",
3166 btf__free(targ_btf);
3167 if (!IS_ERR_OR_NULL(cand_cache)) {
3168 hashmap__for_each_entry(cand_cache, entry, i) {
3169 bpf_core_free_cands(entry->value);
3171 hashmap__free(cand_cache);
3177 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
3181 if (obj->btf_ext->offset_reloc_info.len)
3182 err = bpf_core_reloc_offsets(obj, targ_btf_path);
3188 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
3189 struct reloc_desc *relo)
3191 struct bpf_insn *insn, *new_insn;
3192 struct bpf_program *text;
3196 if (relo->type != RELO_CALL)
3197 return -LIBBPF_ERRNO__RELOC;
3199 if (prog->idx == obj->efile.text_shndx) {
3200 pr_warning("relo in .text insn %d into off %d\n",
3201 relo->insn_idx, relo->text_off);
3202 return -LIBBPF_ERRNO__RELOC;
3205 if (prog->main_prog_cnt == 0) {
3206 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
3208 pr_warning("no .text section found yet relo into text exist\n");
3209 return -LIBBPF_ERRNO__RELOC;
3211 new_cnt = prog->insns_cnt + text->insns_cnt;
3212 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
3214 pr_warning("oom in prog realloc\n");
3219 err = bpf_program_reloc_btf_ext(prog, obj,
3226 memcpy(new_insn + prog->insns_cnt, text->insns,
3227 text->insns_cnt * sizeof(*insn));
3228 prog->insns = new_insn;
3229 prog->main_prog_cnt = prog->insns_cnt;
3230 prog->insns_cnt = new_cnt;
3231 pr_debug("added %zd insn from %s to prog %s\n",
3232 text->insns_cnt, text->section_name,
3233 prog->section_name);
3235 insn = &prog->insns[relo->insn_idx];
3236 insn->imm += prog->main_prog_cnt - relo->insn_idx;
3241 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
3249 err = bpf_program_reloc_btf_ext(prog, obj,
3250 prog->section_name, 0);
3255 if (!prog->reloc_desc)
3258 for (i = 0; i < prog->nr_reloc; i++) {
3259 if (prog->reloc_desc[i].type == RELO_LD64 ||
3260 prog->reloc_desc[i].type == RELO_DATA) {
3261 bool relo_data = prog->reloc_desc[i].type == RELO_DATA;
3262 struct bpf_insn *insns = prog->insns;
3263 int insn_idx, map_idx;
3265 insn_idx = prog->reloc_desc[i].insn_idx;
3266 map_idx = prog->reloc_desc[i].map_idx;
3268 if (insn_idx + 1 >= (int)prog->insns_cnt) {
3269 pr_warning("relocation out of range: '%s'\n",
3270 prog->section_name);
3271 return -LIBBPF_ERRNO__RELOC;
3275 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
3277 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_VALUE;
3278 insns[insn_idx + 1].imm = insns[insn_idx].imm;
3280 insns[insn_idx].imm = obj->maps[map_idx].fd;
3281 } else if (prog->reloc_desc[i].type == RELO_CALL) {
3282 err = bpf_program__reloc_text(prog, obj,
3283 &prog->reloc_desc[i]);
3289 zfree(&prog->reloc_desc);
3295 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
3297 struct bpf_program *prog;
3302 err = bpf_object__relocate_core(obj, targ_btf_path);
3304 pr_warning("failed to perform CO-RE relocations: %d\n",
3309 for (i = 0; i < obj->nr_programs; i++) {
3310 prog = &obj->programs[i];
3312 err = bpf_program__relocate(prog, obj);
3314 pr_warning("failed to relocate '%s'\n",
3315 prog->section_name);
3322 static int bpf_object__collect_reloc(struct bpf_object *obj)
3326 if (!obj_elf_valid(obj)) {
3327 pr_warning("Internal error: elf object is closed\n");
3328 return -LIBBPF_ERRNO__INTERNAL;
3331 for (i = 0; i < obj->efile.nr_reloc; i++) {
3332 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
3333 Elf_Data *data = obj->efile.reloc[i].data;
3334 int idx = shdr->sh_info;
3335 struct bpf_program *prog;
3337 if (shdr->sh_type != SHT_REL) {
3338 pr_warning("internal error at %d\n", __LINE__);
3339 return -LIBBPF_ERRNO__INTERNAL;
3342 prog = bpf_object__find_prog_by_idx(obj, idx);
3344 pr_warning("relocation failed: no section(%d)\n", idx);
3345 return -LIBBPF_ERRNO__RELOC;
3348 err = bpf_program__collect_reloc(prog, shdr, data, obj);
3356 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
3357 char *license, __u32 kern_version, int *pfd)
3359 struct bpf_load_program_attr load_attr;
3360 char *cp, errmsg[STRERR_BUFSIZE];
3361 int log_buf_size = BPF_LOG_BUF_SIZE;
3365 if (!insns || !insns_cnt)
3368 memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
3369 load_attr.prog_type = prog->type;
3370 load_attr.expected_attach_type = prog->expected_attach_type;
3371 if (prog->caps->name)
3372 load_attr.name = prog->name;
3373 load_attr.insns = insns;
3374 load_attr.insns_cnt = insns_cnt;
3375 load_attr.license = license;
3376 load_attr.kern_version = kern_version;
3377 load_attr.prog_ifindex = prog->prog_ifindex;
3378 /* if .BTF.ext was loaded, kernel supports associated BTF for prog */
3379 if (prog->obj->btf_ext)
3380 btf_fd = bpf_object__btf_fd(prog->obj);
3383 load_attr.prog_btf_fd = btf_fd >= 0 ? btf_fd : 0;
3384 load_attr.func_info = prog->func_info;
3385 load_attr.func_info_rec_size = prog->func_info_rec_size;
3386 load_attr.func_info_cnt = prog->func_info_cnt;
3387 load_attr.line_info = prog->line_info;
3388 load_attr.line_info_rec_size = prog->line_info_rec_size;
3389 load_attr.line_info_cnt = prog->line_info_cnt;
3390 load_attr.log_level = prog->log_level;
3391 load_attr.prog_flags = prog->prog_flags;
3394 log_buf = malloc(log_buf_size);
3396 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
3398 ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
3401 if (load_attr.log_level)
3402 pr_debug("verifier log:\n%s", log_buf);
3408 if (errno == ENOSPC) {
3413 ret = -LIBBPF_ERRNO__LOAD;
3414 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3415 pr_warning("load bpf program failed: %s\n", cp);
3417 if (log_buf && log_buf[0] != '\0') {
3418 ret = -LIBBPF_ERRNO__VERIFY;
3419 pr_warning("-- BEGIN DUMP LOG ---\n");
3420 pr_warning("\n%s\n", log_buf);
3421 pr_warning("-- END LOG --\n");
3422 } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
3423 pr_warning("Program too large (%zu insns), at most %d insns\n",
3424 load_attr.insns_cnt, BPF_MAXINSNS);
3425 ret = -LIBBPF_ERRNO__PROG2BIG;
3427 /* Wrong program type? */
3428 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
3431 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
3432 load_attr.expected_attach_type = 0;
3433 fd = bpf_load_program_xattr(&load_attr, NULL, 0);
3436 ret = -LIBBPF_ERRNO__PROGTYPE;
3442 ret = -LIBBPF_ERRNO__KVER;
3451 bpf_program__load(struct bpf_program *prog,
3452 char *license, __u32 kern_version)
3456 if (prog->instances.nr < 0 || !prog->instances.fds) {
3457 if (prog->preprocessor) {
3458 pr_warning("Internal error: can't load program '%s'\n",
3459 prog->section_name);
3460 return -LIBBPF_ERRNO__INTERNAL;
3463 prog->instances.fds = malloc(sizeof(int));
3464 if (!prog->instances.fds) {
3465 pr_warning("Not enough memory for BPF fds\n");
3468 prog->instances.nr = 1;
3469 prog->instances.fds[0] = -1;
3472 if (!prog->preprocessor) {
3473 if (prog->instances.nr != 1) {
3474 pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
3475 prog->section_name, prog->instances.nr);
3477 err = load_program(prog, prog->insns, prog->insns_cnt,
3478 license, kern_version, &fd);
3480 prog->instances.fds[0] = fd;
3484 for (i = 0; i < prog->instances.nr; i++) {
3485 struct bpf_prog_prep_result result;
3486 bpf_program_prep_t preprocessor = prog->preprocessor;
3488 memset(&result, 0, sizeof(result));
3489 err = preprocessor(prog, i, prog->insns,
3490 prog->insns_cnt, &result);
3492 pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
3493 i, prog->section_name);
3497 if (!result.new_insn_ptr || !result.new_insn_cnt) {
3498 pr_debug("Skip loading the %dth instance of program '%s'\n",
3499 i, prog->section_name);
3500 prog->instances.fds[i] = -1;
3506 err = load_program(prog, result.new_insn_ptr,
3507 result.new_insn_cnt,
3508 license, kern_version, &fd);
3511 pr_warning("Loading the %dth instance of program '%s' failed\n",
3512 i, prog->section_name);
3518 prog->instances.fds[i] = fd;
3522 pr_warning("failed to load program '%s'\n",
3523 prog->section_name);
3524 zfree(&prog->insns);
3525 prog->insns_cnt = 0;
3529 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
3530 const struct bpf_object *obj)
3532 return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
3536 bpf_object__load_progs(struct bpf_object *obj, int log_level)
3541 for (i = 0; i < obj->nr_programs; i++) {
3542 if (bpf_program__is_function_storage(&obj->programs[i], obj))
3544 obj->programs[i].log_level |= log_level;
3545 err = bpf_program__load(&obj->programs[i],
3554 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
3557 case BPF_PROG_TYPE_SOCKET_FILTER:
3558 case BPF_PROG_TYPE_SCHED_CLS:
3559 case BPF_PROG_TYPE_SCHED_ACT:
3560 case BPF_PROG_TYPE_XDP:
3561 case BPF_PROG_TYPE_CGROUP_SKB:
3562 case BPF_PROG_TYPE_CGROUP_SOCK:
3563 case BPF_PROG_TYPE_LWT_IN:
3564 case BPF_PROG_TYPE_LWT_OUT:
3565 case BPF_PROG_TYPE_LWT_XMIT:
3566 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
3567 case BPF_PROG_TYPE_SOCK_OPS:
3568 case BPF_PROG_TYPE_SK_SKB:
3569 case BPF_PROG_TYPE_CGROUP_DEVICE:
3570 case BPF_PROG_TYPE_SK_MSG:
3571 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3572 case BPF_PROG_TYPE_LIRC_MODE2:
3573 case BPF_PROG_TYPE_SK_REUSEPORT:
3574 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3575 case BPF_PROG_TYPE_UNSPEC:
3576 case BPF_PROG_TYPE_TRACEPOINT:
3577 case BPF_PROG_TYPE_RAW_TRACEPOINT:
3578 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3579 case BPF_PROG_TYPE_PERF_EVENT:
3580 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3581 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3583 case BPF_PROG_TYPE_KPROBE:
3589 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
3591 if (needs_kver && obj->kern_version == 0) {
3592 pr_warning("%s doesn't provide kernel version\n",
3594 return -LIBBPF_ERRNO__KVERSION;
3599 static struct bpf_object *
3600 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
3601 bool needs_kver, int flags)
3603 struct bpf_object *obj;
3606 if (elf_version(EV_CURRENT) == EV_NONE) {
3607 pr_warning("failed to init libelf for %s\n", path);
3608 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
3611 obj = bpf_object__new(path, obj_buf, obj_buf_sz);
3615 CHECK_ERR(bpf_object__elf_init(obj), err, out);
3616 CHECK_ERR(bpf_object__check_endianness(obj), err, out);
3617 CHECK_ERR(bpf_object__probe_caps(obj), err, out);
3618 CHECK_ERR(bpf_object__elf_collect(obj, flags), err, out);
3619 CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
3620 CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
3622 bpf_object__elf_finish(obj);
3625 bpf_object__close(obj);
3626 return ERR_PTR(err);
3629 struct bpf_object *__bpf_object__open_xattr(struct bpf_object_open_attr *attr,
3632 /* param validation */
3636 pr_debug("loading %s\n", attr->file);
3638 return __bpf_object__open(attr->file, NULL, 0,
3639 bpf_prog_type__needs_kver(attr->prog_type),
3643 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
3645 return __bpf_object__open_xattr(attr, 0);
3648 struct bpf_object *bpf_object__open(const char *path)
3650 struct bpf_object_open_attr attr = {
3652 .prog_type = BPF_PROG_TYPE_UNSPEC,
3655 return bpf_object__open_xattr(&attr);
3658 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
3664 /* param validation */
3665 if (!obj_buf || obj_buf_sz <= 0)
3669 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
3670 (unsigned long)obj_buf,
3671 (unsigned long)obj_buf_sz);
3674 pr_debug("loading object '%s' from buffer\n", name);
3676 return __bpf_object__open(name, obj_buf, obj_buf_sz, true, true);
3679 int bpf_object__unload(struct bpf_object *obj)
3686 for (i = 0; i < obj->nr_maps; i++)
3687 zclose(obj->maps[i].fd);
3689 for (i = 0; i < obj->nr_programs; i++)
3690 bpf_program__unload(&obj->programs[i]);
3695 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
3697 struct bpf_object *obj;
3707 pr_warning("object should not be loaded twice\n");
3713 CHECK_ERR(bpf_object__create_maps(obj), err, out);
3714 CHECK_ERR(bpf_object__relocate(obj, attr->target_btf_path), err, out);
3715 CHECK_ERR(bpf_object__load_progs(obj, attr->log_level), err, out);
3719 bpf_object__unload(obj);
3720 pr_warning("failed to load object '%s'\n", obj->path);
3724 int bpf_object__load(struct bpf_object *obj)
3726 struct bpf_object_load_attr attr = {
3730 return bpf_object__load_xattr(&attr);
3733 static int check_path(const char *path)
3735 char *cp, errmsg[STRERR_BUFSIZE];
3736 struct statfs st_fs;
3743 dname = strdup(path);
3747 dir = dirname(dname);
3748 if (statfs(dir, &st_fs)) {
3749 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3750 pr_warning("failed to statfs %s: %s\n", dir, cp);
3755 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
3756 pr_warning("specified path %s is not on BPF FS\n", path);
3763 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
3766 char *cp, errmsg[STRERR_BUFSIZE];
3769 err = check_path(path);
3774 pr_warning("invalid program pointer\n");
3778 if (instance < 0 || instance >= prog->instances.nr) {
3779 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
3780 instance, prog->section_name, prog->instances.nr);
3784 if (bpf_obj_pin(prog->instances.fds[instance], path)) {
3785 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3786 pr_warning("failed to pin program: %s\n", cp);
3789 pr_debug("pinned program '%s'\n", path);
3794 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
3799 err = check_path(path);
3804 pr_warning("invalid program pointer\n");
3808 if (instance < 0 || instance >= prog->instances.nr) {
3809 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
3810 instance, prog->section_name, prog->instances.nr);
3817 pr_debug("unpinned program '%s'\n", path);
3822 static int make_dir(const char *path)
3824 char *cp, errmsg[STRERR_BUFSIZE];
3827 if (mkdir(path, 0700) && errno != EEXIST)
3831 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
3832 pr_warning("failed to mkdir %s: %s\n", path, cp);
3837 int bpf_program__pin(struct bpf_program *prog, const char *path)
3841 err = check_path(path);
3846 pr_warning("invalid program pointer\n");
3850 if (prog->instances.nr <= 0) {
3851 pr_warning("no instances of prog %s to pin\n",
3852 prog->section_name);
3856 if (prog->instances.nr == 1) {
3857 /* don't create subdirs when pinning single instance */
3858 return bpf_program__pin_instance(prog, path, 0);
3861 err = make_dir(path);
3865 for (i = 0; i < prog->instances.nr; i++) {
3869 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3873 } else if (len >= PATH_MAX) {
3874 err = -ENAMETOOLONG;
3878 err = bpf_program__pin_instance(prog, buf, i);
3886 for (i = i - 1; i >= 0; i--) {
3890 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3893 else if (len >= PATH_MAX)
3896 bpf_program__unpin_instance(prog, buf, i);
3904 int bpf_program__unpin(struct bpf_program *prog, const char *path)
3908 err = check_path(path);
3913 pr_warning("invalid program pointer\n");
3917 if (prog->instances.nr <= 0) {
3918 pr_warning("no instances of prog %s to pin\n",
3919 prog->section_name);
3923 if (prog->instances.nr == 1) {
3924 /* don't create subdirs when pinning single instance */
3925 return bpf_program__unpin_instance(prog, path, 0);
3928 for (i = 0; i < prog->instances.nr; i++) {
3932 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3935 else if (len >= PATH_MAX)
3936 return -ENAMETOOLONG;
3938 err = bpf_program__unpin_instance(prog, buf, i);
3950 int bpf_map__pin(struct bpf_map *map, const char *path)
3952 char *cp, errmsg[STRERR_BUFSIZE];
3955 err = check_path(path);
3960 pr_warning("invalid map pointer\n");
3964 if (bpf_obj_pin(map->fd, path)) {
3965 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3966 pr_warning("failed to pin map: %s\n", cp);
3970 pr_debug("pinned map '%s'\n", path);
3975 int bpf_map__unpin(struct bpf_map *map, const char *path)
3979 err = check_path(path);
3984 pr_warning("invalid map pointer\n");
3991 pr_debug("unpinned map '%s'\n", path);
3996 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
3998 struct bpf_map *map;
4005 pr_warning("object not yet loaded; load it first\n");
4009 err = make_dir(path);
4013 bpf_object__for_each_map(map, obj) {
4017 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4018 bpf_map__name(map));
4021 goto err_unpin_maps;
4022 } else if (len >= PATH_MAX) {
4023 err = -ENAMETOOLONG;
4024 goto err_unpin_maps;
4027 err = bpf_map__pin(map, buf);
4029 goto err_unpin_maps;
4035 while ((map = bpf_map__prev(map, obj))) {
4039 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4040 bpf_map__name(map));
4043 else if (len >= PATH_MAX)
4046 bpf_map__unpin(map, buf);
4052 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
4054 struct bpf_map *map;
4060 bpf_object__for_each_map(map, obj) {
4064 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4065 bpf_map__name(map));
4068 else if (len >= PATH_MAX)
4069 return -ENAMETOOLONG;
4071 err = bpf_map__unpin(map, buf);
4079 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
4081 struct bpf_program *prog;
4088 pr_warning("object not yet loaded; load it first\n");
4092 err = make_dir(path);
4096 bpf_object__for_each_program(prog, obj) {
4100 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4104 goto err_unpin_programs;
4105 } else if (len >= PATH_MAX) {
4106 err = -ENAMETOOLONG;
4107 goto err_unpin_programs;
4110 err = bpf_program__pin(prog, buf);
4112 goto err_unpin_programs;
4118 while ((prog = bpf_program__prev(prog, obj))) {
4122 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4126 else if (len >= PATH_MAX)
4129 bpf_program__unpin(prog, buf);
4135 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
4137 struct bpf_program *prog;
4143 bpf_object__for_each_program(prog, obj) {
4147 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4151 else if (len >= PATH_MAX)
4152 return -ENAMETOOLONG;
4154 err = bpf_program__unpin(prog, buf);
4162 int bpf_object__pin(struct bpf_object *obj, const char *path)
4166 err = bpf_object__pin_maps(obj, path);
4170 err = bpf_object__pin_programs(obj, path);
4172 bpf_object__unpin_maps(obj, path);
4179 void bpf_object__close(struct bpf_object *obj)
4186 if (obj->clear_priv)
4187 obj->clear_priv(obj, obj->priv);
4189 bpf_object__elf_finish(obj);
4190 bpf_object__unload(obj);
4191 btf__free(obj->btf);
4192 btf_ext__free(obj->btf_ext);
4194 for (i = 0; i < obj->nr_maps; i++) {
4195 zfree(&obj->maps[i].name);
4196 if (obj->maps[i].clear_priv)
4197 obj->maps[i].clear_priv(&obj->maps[i],
4199 obj->maps[i].priv = NULL;
4200 obj->maps[i].clear_priv = NULL;
4203 zfree(&obj->sections.rodata);
4204 zfree(&obj->sections.data);
4208 if (obj->programs && obj->nr_programs) {
4209 for (i = 0; i < obj->nr_programs; i++)
4210 bpf_program__exit(&obj->programs[i]);
4212 zfree(&obj->programs);
4214 list_del(&obj->list);
4219 bpf_object__next(struct bpf_object *prev)
4221 struct bpf_object *next;
4224 next = list_first_entry(&bpf_objects_list,
4228 next = list_next_entry(prev, list);
4230 /* Empty list is noticed here so don't need checking on entry. */
4231 if (&next->list == &bpf_objects_list)
4237 const char *bpf_object__name(const struct bpf_object *obj)
4239 return obj ? obj->path : ERR_PTR(-EINVAL);
4242 unsigned int bpf_object__kversion(const struct bpf_object *obj)
4244 return obj ? obj->kern_version : 0;
4247 struct btf *bpf_object__btf(const struct bpf_object *obj)
4249 return obj ? obj->btf : NULL;
4252 int bpf_object__btf_fd(const struct bpf_object *obj)
4254 return obj->btf ? btf__fd(obj->btf) : -1;
4257 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
4258 bpf_object_clear_priv_t clear_priv)
4260 if (obj->priv && obj->clear_priv)
4261 obj->clear_priv(obj, obj->priv);
4264 obj->clear_priv = clear_priv;
4268 void *bpf_object__priv(const struct bpf_object *obj)
4270 return obj ? obj->priv : ERR_PTR(-EINVAL);
4273 static struct bpf_program *
4274 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
4277 size_t nr_programs = obj->nr_programs;
4284 /* Iter from the beginning */
4285 return forward ? &obj->programs[0] :
4286 &obj->programs[nr_programs - 1];
4288 if (p->obj != obj) {
4289 pr_warning("error: program handler doesn't match object\n");
4293 idx = (p - obj->programs) + (forward ? 1 : -1);
4294 if (idx >= obj->nr_programs || idx < 0)
4296 return &obj->programs[idx];
4299 struct bpf_program *
4300 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
4302 struct bpf_program *prog = prev;
4305 prog = __bpf_program__iter(prog, obj, true);
4306 } while (prog && bpf_program__is_function_storage(prog, obj));
4311 struct bpf_program *
4312 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
4314 struct bpf_program *prog = next;
4317 prog = __bpf_program__iter(prog, obj, false);
4318 } while (prog && bpf_program__is_function_storage(prog, obj));
4323 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
4324 bpf_program_clear_priv_t clear_priv)
4326 if (prog->priv && prog->clear_priv)
4327 prog->clear_priv(prog, prog->priv);
4330 prog->clear_priv = clear_priv;
4334 void *bpf_program__priv(const struct bpf_program *prog)
4336 return prog ? prog->priv : ERR_PTR(-EINVAL);
4339 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
4341 prog->prog_ifindex = ifindex;
4344 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
4348 title = prog->section_name;
4350 title = strdup(title);
4352 pr_warning("failed to strdup program title\n");
4353 return ERR_PTR(-ENOMEM);
4360 int bpf_program__fd(const struct bpf_program *prog)
4362 return bpf_program__nth_fd(prog, 0);
4365 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
4366 bpf_program_prep_t prep)
4370 if (nr_instances <= 0 || !prep)
4373 if (prog->instances.nr > 0 || prog->instances.fds) {
4374 pr_warning("Can't set pre-processor after loading\n");
4378 instances_fds = malloc(sizeof(int) * nr_instances);
4379 if (!instances_fds) {
4380 pr_warning("alloc memory failed for fds\n");
4384 /* fill all fd with -1 */
4385 memset(instances_fds, -1, sizeof(int) * nr_instances);
4387 prog->instances.nr = nr_instances;
4388 prog->instances.fds = instances_fds;
4389 prog->preprocessor = prep;
4393 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
4400 if (n >= prog->instances.nr || n < 0) {
4401 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
4402 n, prog->section_name, prog->instances.nr);
4406 fd = prog->instances.fds[n];
4408 pr_warning("%dth instance of program '%s' is invalid\n",
4409 n, prog->section_name);
4416 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
4421 static bool bpf_program__is_type(const struct bpf_program *prog,
4422 enum bpf_prog_type type)
4424 return prog ? (prog->type == type) : false;
4427 #define BPF_PROG_TYPE_FNS(NAME, TYPE) \
4428 int bpf_program__set_##NAME(struct bpf_program *prog) \
4432 bpf_program__set_type(prog, TYPE); \
4436 bool bpf_program__is_##NAME(const struct bpf_program *prog) \
4438 return bpf_program__is_type(prog, TYPE); \
4441 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
4442 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
4443 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
4444 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
4445 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
4446 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
4447 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
4448 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
4450 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
4451 enum bpf_attach_type type)
4453 prog->expected_attach_type = type;
4456 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, atype) \
4457 { string, sizeof(string) - 1, ptype, eatype, is_attachable, atype }
4459 /* Programs that can NOT be attached. */
4460 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0)
4462 /* Programs that can be attached. */
4463 #define BPF_APROG_SEC(string, ptype, atype) \
4464 BPF_PROG_SEC_IMPL(string, ptype, 0, 1, atype)
4466 /* Programs that must specify expected attach type at load time. */
4467 #define BPF_EAPROG_SEC(string, ptype, eatype) \
4468 BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, eatype)
4470 /* Programs that can be attached but attach type can't be identified by section
4471 * name. Kept for backward compatibility.
4473 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
4475 static const struct {
4478 enum bpf_prog_type prog_type;
4479 enum bpf_attach_type expected_attach_type;
4481 enum bpf_attach_type attach_type;
4482 } section_names[] = {
4483 BPF_PROG_SEC("socket", BPF_PROG_TYPE_SOCKET_FILTER),
4484 BPF_PROG_SEC("kprobe/", BPF_PROG_TYPE_KPROBE),
4485 BPF_PROG_SEC("kretprobe/", BPF_PROG_TYPE_KPROBE),
4486 BPF_PROG_SEC("classifier", BPF_PROG_TYPE_SCHED_CLS),
4487 BPF_PROG_SEC("action", BPF_PROG_TYPE_SCHED_ACT),
4488 BPF_PROG_SEC("tracepoint/", BPF_PROG_TYPE_TRACEPOINT),
4489 BPF_PROG_SEC("raw_tracepoint/", BPF_PROG_TYPE_RAW_TRACEPOINT),
4490 BPF_PROG_SEC("xdp", BPF_PROG_TYPE_XDP),
4491 BPF_PROG_SEC("perf_event", BPF_PROG_TYPE_PERF_EVENT),
4492 BPF_PROG_SEC("lwt_in", BPF_PROG_TYPE_LWT_IN),
4493 BPF_PROG_SEC("lwt_out", BPF_PROG_TYPE_LWT_OUT),
4494 BPF_PROG_SEC("lwt_xmit", BPF_PROG_TYPE_LWT_XMIT),
4495 BPF_PROG_SEC("lwt_seg6local", BPF_PROG_TYPE_LWT_SEG6LOCAL),
4496 BPF_APROG_SEC("cgroup_skb/ingress", BPF_PROG_TYPE_CGROUP_SKB,
4497 BPF_CGROUP_INET_INGRESS),
4498 BPF_APROG_SEC("cgroup_skb/egress", BPF_PROG_TYPE_CGROUP_SKB,
4499 BPF_CGROUP_INET_EGRESS),
4500 BPF_APROG_COMPAT("cgroup/skb", BPF_PROG_TYPE_CGROUP_SKB),
4501 BPF_APROG_SEC("cgroup/sock", BPF_PROG_TYPE_CGROUP_SOCK,
4502 BPF_CGROUP_INET_SOCK_CREATE),
4503 BPF_EAPROG_SEC("cgroup/post_bind4", BPF_PROG_TYPE_CGROUP_SOCK,
4504 BPF_CGROUP_INET4_POST_BIND),
4505 BPF_EAPROG_SEC("cgroup/post_bind6", BPF_PROG_TYPE_CGROUP_SOCK,
4506 BPF_CGROUP_INET6_POST_BIND),
4507 BPF_APROG_SEC("cgroup/dev", BPF_PROG_TYPE_CGROUP_DEVICE,
4509 BPF_APROG_SEC("sockops", BPF_PROG_TYPE_SOCK_OPS,
4510 BPF_CGROUP_SOCK_OPS),
4511 BPF_APROG_SEC("sk_skb/stream_parser", BPF_PROG_TYPE_SK_SKB,
4512 BPF_SK_SKB_STREAM_PARSER),
4513 BPF_APROG_SEC("sk_skb/stream_verdict", BPF_PROG_TYPE_SK_SKB,
4514 BPF_SK_SKB_STREAM_VERDICT),
4515 BPF_APROG_COMPAT("sk_skb", BPF_PROG_TYPE_SK_SKB),
4516 BPF_APROG_SEC("sk_msg", BPF_PROG_TYPE_SK_MSG,
4517 BPF_SK_MSG_VERDICT),
4518 BPF_APROG_SEC("lirc_mode2", BPF_PROG_TYPE_LIRC_MODE2,
4520 BPF_APROG_SEC("flow_dissector", BPF_PROG_TYPE_FLOW_DISSECTOR,
4521 BPF_FLOW_DISSECTOR),
4522 BPF_EAPROG_SEC("cgroup/bind4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4523 BPF_CGROUP_INET4_BIND),
4524 BPF_EAPROG_SEC("cgroup/bind6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4525 BPF_CGROUP_INET6_BIND),
4526 BPF_EAPROG_SEC("cgroup/connect4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4527 BPF_CGROUP_INET4_CONNECT),
4528 BPF_EAPROG_SEC("cgroup/connect6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4529 BPF_CGROUP_INET6_CONNECT),
4530 BPF_EAPROG_SEC("cgroup/sendmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4531 BPF_CGROUP_UDP4_SENDMSG),
4532 BPF_EAPROG_SEC("cgroup/sendmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4533 BPF_CGROUP_UDP6_SENDMSG),
4534 BPF_EAPROG_SEC("cgroup/recvmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4535 BPF_CGROUP_UDP4_RECVMSG),
4536 BPF_EAPROG_SEC("cgroup/recvmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4537 BPF_CGROUP_UDP6_RECVMSG),
4538 BPF_EAPROG_SEC("cgroup/sysctl", BPF_PROG_TYPE_CGROUP_SYSCTL,
4540 BPF_EAPROG_SEC("cgroup/getsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT,
4541 BPF_CGROUP_GETSOCKOPT),
4542 BPF_EAPROG_SEC("cgroup/setsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT,
4543 BPF_CGROUP_SETSOCKOPT),
4546 #undef BPF_PROG_SEC_IMPL
4548 #undef BPF_APROG_SEC
4549 #undef BPF_EAPROG_SEC
4550 #undef BPF_APROG_COMPAT
4552 #define MAX_TYPE_NAME_SIZE 32
4554 static char *libbpf_get_type_names(bool attach_type)
4556 int i, len = ARRAY_SIZE(section_names) * MAX_TYPE_NAME_SIZE;
4564 /* Forge string buf with all available names */
4565 for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4566 if (attach_type && !section_names[i].is_attachable)
4569 if (strlen(buf) + strlen(section_names[i].sec) + 2 > len) {
4574 strcat(buf, section_names[i].sec);
4580 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
4581 enum bpf_attach_type *expected_attach_type)
4589 for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4590 if (strncmp(name, section_names[i].sec, section_names[i].len))
4592 *prog_type = section_names[i].prog_type;
4593 *expected_attach_type = section_names[i].expected_attach_type;
4596 pr_warning("failed to guess program type based on ELF section name '%s'\n", name);
4597 type_names = libbpf_get_type_names(false);
4598 if (type_names != NULL) {
4599 pr_info("supported section(type) names are:%s\n", type_names);
4606 int libbpf_attach_type_by_name(const char *name,
4607 enum bpf_attach_type *attach_type)
4615 for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4616 if (strncmp(name, section_names[i].sec, section_names[i].len))
4618 if (!section_names[i].is_attachable)
4620 *attach_type = section_names[i].attach_type;
4623 pr_warning("failed to guess attach type based on ELF section name '%s'\n", name);
4624 type_names = libbpf_get_type_names(true);
4625 if (type_names != NULL) {
4626 pr_info("attachable section(type) names are:%s\n", type_names);
4634 bpf_program__identify_section(struct bpf_program *prog,
4635 enum bpf_prog_type *prog_type,
4636 enum bpf_attach_type *expected_attach_type)
4638 return libbpf_prog_type_by_name(prog->section_name, prog_type,
4639 expected_attach_type);
4642 int bpf_map__fd(const struct bpf_map *map)
4644 return map ? map->fd : -EINVAL;
4647 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
4649 return map ? &map->def : ERR_PTR(-EINVAL);
4652 const char *bpf_map__name(const struct bpf_map *map)
4654 return map ? map->name : NULL;
4657 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
4659 return map ? map->btf_key_type_id : 0;
4662 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
4664 return map ? map->btf_value_type_id : 0;
4667 int bpf_map__set_priv(struct bpf_map *map, void *priv,
4668 bpf_map_clear_priv_t clear_priv)
4674 if (map->clear_priv)
4675 map->clear_priv(map, map->priv);
4679 map->clear_priv = clear_priv;
4683 void *bpf_map__priv(const struct bpf_map *map)
4685 return map ? map->priv : ERR_PTR(-EINVAL);
4688 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
4690 return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
4693 bool bpf_map__is_internal(const struct bpf_map *map)
4695 return map->libbpf_type != LIBBPF_MAP_UNSPEC;
4698 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
4700 map->map_ifindex = ifindex;
4703 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
4705 if (!bpf_map_type__is_map_in_map(map->def.type)) {
4706 pr_warning("error: unsupported map type\n");
4709 if (map->inner_map_fd != -1) {
4710 pr_warning("error: inner_map_fd already specified\n");
4713 map->inner_map_fd = fd;
4717 static struct bpf_map *
4718 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
4721 struct bpf_map *s, *e;
4723 if (!obj || !obj->maps)
4727 e = obj->maps + obj->nr_maps;
4729 if ((m < s) || (m >= e)) {
4730 pr_warning("error in %s: map handler doesn't belong to object\n",
4735 idx = (m - obj->maps) + i;
4736 if (idx >= obj->nr_maps || idx < 0)
4738 return &obj->maps[idx];
4742 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
4747 return __bpf_map__iter(prev, obj, 1);
4751 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
4756 return obj->maps + obj->nr_maps - 1;
4759 return __bpf_map__iter(next, obj, -1);
4763 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
4765 struct bpf_map *pos;
4767 bpf_object__for_each_map(pos, obj) {
4768 if (pos->name && !strcmp(pos->name, name))
4775 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
4777 return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
4781 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
4783 return ERR_PTR(-ENOTSUP);
4786 long libbpf_get_error(const void *ptr)
4788 return PTR_ERR_OR_ZERO(ptr);
4791 int bpf_prog_load(const char *file, enum bpf_prog_type type,
4792 struct bpf_object **pobj, int *prog_fd)
4794 struct bpf_prog_load_attr attr;
4796 memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
4798 attr.prog_type = type;
4799 attr.expected_attach_type = 0;
4801 return bpf_prog_load_xattr(&attr, pobj, prog_fd);
4804 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
4805 struct bpf_object **pobj, int *prog_fd)
4807 struct bpf_object_open_attr open_attr = {};
4808 struct bpf_program *prog, *first_prog = NULL;
4809 enum bpf_attach_type expected_attach_type;
4810 enum bpf_prog_type prog_type;
4811 struct bpf_object *obj;
4812 struct bpf_map *map;
4820 open_attr.file = attr->file;
4821 open_attr.prog_type = attr->prog_type;
4823 obj = bpf_object__open_xattr(&open_attr);
4824 if (IS_ERR_OR_NULL(obj))
4827 bpf_object__for_each_program(prog, obj) {
4829 * If type is not specified, try to guess it based on
4832 prog_type = attr->prog_type;
4833 prog->prog_ifindex = attr->ifindex;
4834 expected_attach_type = attr->expected_attach_type;
4835 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
4836 err = bpf_program__identify_section(prog, &prog_type,
4837 &expected_attach_type);
4839 bpf_object__close(obj);
4844 bpf_program__set_type(prog, prog_type);
4845 bpf_program__set_expected_attach_type(prog,
4846 expected_attach_type);
4848 prog->log_level = attr->log_level;
4849 prog->prog_flags = attr->prog_flags;
4854 bpf_object__for_each_map(map, obj) {
4855 if (!bpf_map__is_offload_neutral(map))
4856 map->map_ifindex = attr->ifindex;
4860 pr_warning("object file doesn't contain bpf program\n");
4861 bpf_object__close(obj);
4865 err = bpf_object__load(obj);
4867 bpf_object__close(obj);
4872 *prog_fd = bpf_program__fd(first_prog);
4877 int (*destroy)(struct bpf_link *link);
4880 int bpf_link__destroy(struct bpf_link *link)
4887 err = link->destroy(link);
4893 struct bpf_link_fd {
4894 struct bpf_link link; /* has to be at the top of struct */
4895 int fd; /* hook FD */
4898 static int bpf_link__destroy_perf_event(struct bpf_link *link)
4900 struct bpf_link_fd *l = (void *)link;
4903 err = ioctl(l->fd, PERF_EVENT_IOC_DISABLE, 0);
4911 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
4914 char errmsg[STRERR_BUFSIZE];
4915 struct bpf_link_fd *link;
4919 pr_warning("program '%s': invalid perf event FD %d\n",
4920 bpf_program__title(prog, false), pfd);
4921 return ERR_PTR(-EINVAL);
4923 prog_fd = bpf_program__fd(prog);
4925 pr_warning("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
4926 bpf_program__title(prog, false));
4927 return ERR_PTR(-EINVAL);
4930 link = malloc(sizeof(*link));
4932 return ERR_PTR(-ENOMEM);
4933 link->link.destroy = &bpf_link__destroy_perf_event;
4936 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
4939 pr_warning("program '%s': failed to attach to pfd %d: %s\n",
4940 bpf_program__title(prog, false), pfd,
4941 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4942 return ERR_PTR(err);
4944 if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
4947 pr_warning("program '%s': failed to enable pfd %d: %s\n",
4948 bpf_program__title(prog, false), pfd,
4949 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4950 return ERR_PTR(err);
4952 return (struct bpf_link *)link;
4956 * this function is expected to parse integer in the range of [0, 2^31-1] from
4957 * given file using scanf format string fmt. If actual parsed value is
4958 * negative, the result might be indistinguishable from error
4960 static int parse_uint_from_file(const char *file, const char *fmt)
4962 char buf[STRERR_BUFSIZE];
4966 f = fopen(file, "r");
4969 pr_debug("failed to open '%s': %s\n", file,
4970 libbpf_strerror_r(err, buf, sizeof(buf)));
4973 err = fscanf(f, fmt, &ret);
4975 err = err == EOF ? -EIO : -errno;
4976 pr_debug("failed to parse '%s': %s\n", file,
4977 libbpf_strerror_r(err, buf, sizeof(buf)));
4985 static int determine_kprobe_perf_type(void)
4987 const char *file = "/sys/bus/event_source/devices/kprobe/type";
4989 return parse_uint_from_file(file, "%d\n");
4992 static int determine_uprobe_perf_type(void)
4994 const char *file = "/sys/bus/event_source/devices/uprobe/type";
4996 return parse_uint_from_file(file, "%d\n");
4999 static int determine_kprobe_retprobe_bit(void)
5001 const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
5003 return parse_uint_from_file(file, "config:%d\n");
5006 static int determine_uprobe_retprobe_bit(void)
5008 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
5010 return parse_uint_from_file(file, "config:%d\n");
5013 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
5014 uint64_t offset, int pid)
5016 struct perf_event_attr attr = {};
5017 char errmsg[STRERR_BUFSIZE];
5020 type = uprobe ? determine_uprobe_perf_type()
5021 : determine_kprobe_perf_type();
5023 pr_warning("failed to determine %s perf type: %s\n",
5024 uprobe ? "uprobe" : "kprobe",
5025 libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
5029 int bit = uprobe ? determine_uprobe_retprobe_bit()
5030 : determine_kprobe_retprobe_bit();
5033 pr_warning("failed to determine %s retprobe bit: %s\n",
5034 uprobe ? "uprobe" : "kprobe",
5035 libbpf_strerror_r(bit, errmsg,
5039 attr.config |= 1 << bit;
5041 attr.size = sizeof(attr);
5043 attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
5044 attr.config2 = offset; /* kprobe_addr or probe_offset */
5046 /* pid filter is meaningful only for uprobes */
5047 pfd = syscall(__NR_perf_event_open, &attr,
5048 pid < 0 ? -1 : pid /* pid */,
5049 pid == -1 ? 0 : -1 /* cpu */,
5050 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5053 pr_warning("%s perf_event_open() failed: %s\n",
5054 uprobe ? "uprobe" : "kprobe",
5055 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5061 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
5063 const char *func_name)
5065 char errmsg[STRERR_BUFSIZE];
5066 struct bpf_link *link;
5069 pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
5070 0 /* offset */, -1 /* pid */);
5072 pr_warning("program '%s': failed to create %s '%s' perf event: %s\n",
5073 bpf_program__title(prog, false),
5074 retprobe ? "kretprobe" : "kprobe", func_name,
5075 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5076 return ERR_PTR(pfd);
5078 link = bpf_program__attach_perf_event(prog, pfd);
5081 err = PTR_ERR(link);
5082 pr_warning("program '%s': failed to attach to %s '%s': %s\n",
5083 bpf_program__title(prog, false),
5084 retprobe ? "kretprobe" : "kprobe", func_name,
5085 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5091 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
5092 bool retprobe, pid_t pid,
5093 const char *binary_path,
5096 char errmsg[STRERR_BUFSIZE];
5097 struct bpf_link *link;
5100 pfd = perf_event_open_probe(true /* uprobe */, retprobe,
5101 binary_path, func_offset, pid);
5103 pr_warning("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
5104 bpf_program__title(prog, false),
5105 retprobe ? "uretprobe" : "uprobe",
5106 binary_path, func_offset,
5107 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5108 return ERR_PTR(pfd);
5110 link = bpf_program__attach_perf_event(prog, pfd);
5113 err = PTR_ERR(link);
5114 pr_warning("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
5115 bpf_program__title(prog, false),
5116 retprobe ? "uretprobe" : "uprobe",
5117 binary_path, func_offset,
5118 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5124 static int determine_tracepoint_id(const char *tp_category,
5125 const char *tp_name)
5127 char file[PATH_MAX];
5130 ret = snprintf(file, sizeof(file),
5131 "/sys/kernel/debug/tracing/events/%s/%s/id",
5132 tp_category, tp_name);
5135 if (ret >= sizeof(file)) {
5136 pr_debug("tracepoint %s/%s path is too long\n",
5137 tp_category, tp_name);
5140 return parse_uint_from_file(file, "%d\n");
5143 static int perf_event_open_tracepoint(const char *tp_category,
5144 const char *tp_name)
5146 struct perf_event_attr attr = {};
5147 char errmsg[STRERR_BUFSIZE];
5148 int tp_id, pfd, err;
5150 tp_id = determine_tracepoint_id(tp_category, tp_name);
5152 pr_warning("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
5153 tp_category, tp_name,
5154 libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
5158 attr.type = PERF_TYPE_TRACEPOINT;
5159 attr.size = sizeof(attr);
5160 attr.config = tp_id;
5162 pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
5163 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5166 pr_warning("tracepoint '%s/%s' perf_event_open() failed: %s\n",
5167 tp_category, tp_name,
5168 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5174 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
5175 const char *tp_category,
5176 const char *tp_name)
5178 char errmsg[STRERR_BUFSIZE];
5179 struct bpf_link *link;
5182 pfd = perf_event_open_tracepoint(tp_category, tp_name);
5184 pr_warning("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
5185 bpf_program__title(prog, false),
5186 tp_category, tp_name,
5187 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5188 return ERR_PTR(pfd);
5190 link = bpf_program__attach_perf_event(prog, pfd);
5193 err = PTR_ERR(link);
5194 pr_warning("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
5195 bpf_program__title(prog, false),
5196 tp_category, tp_name,
5197 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5203 static int bpf_link__destroy_fd(struct bpf_link *link)
5205 struct bpf_link_fd *l = (void *)link;
5207 return close(l->fd);
5210 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
5211 const char *tp_name)
5213 char errmsg[STRERR_BUFSIZE];
5214 struct bpf_link_fd *link;
5217 prog_fd = bpf_program__fd(prog);
5219 pr_warning("program '%s': can't attach before loaded\n",
5220 bpf_program__title(prog, false));
5221 return ERR_PTR(-EINVAL);
5224 link = malloc(sizeof(*link));
5226 return ERR_PTR(-ENOMEM);
5227 link->link.destroy = &bpf_link__destroy_fd;
5229 pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
5233 pr_warning("program '%s': failed to attach to raw tracepoint '%s': %s\n",
5234 bpf_program__title(prog, false), tp_name,
5235 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5236 return ERR_PTR(pfd);
5239 return (struct bpf_link *)link;
5242 enum bpf_perf_event_ret
5243 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
5244 void **copy_mem, size_t *copy_size,
5245 bpf_perf_event_print_t fn, void *private_data)
5247 struct perf_event_mmap_page *header = mmap_mem;
5248 __u64 data_head = ring_buffer_read_head(header);
5249 __u64 data_tail = header->data_tail;
5250 void *base = ((__u8 *)header) + page_size;
5251 int ret = LIBBPF_PERF_EVENT_CONT;
5252 struct perf_event_header *ehdr;
5255 while (data_head != data_tail) {
5256 ehdr = base + (data_tail & (mmap_size - 1));
5257 ehdr_size = ehdr->size;
5259 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
5260 void *copy_start = ehdr;
5261 size_t len_first = base + mmap_size - copy_start;
5262 size_t len_secnd = ehdr_size - len_first;
5264 if (*copy_size < ehdr_size) {
5266 *copy_mem = malloc(ehdr_size);
5269 ret = LIBBPF_PERF_EVENT_ERROR;
5272 *copy_size = ehdr_size;
5275 memcpy(*copy_mem, copy_start, len_first);
5276 memcpy(*copy_mem + len_first, base, len_secnd);
5280 ret = fn(ehdr, private_data);
5281 data_tail += ehdr_size;
5282 if (ret != LIBBPF_PERF_EVENT_CONT)
5286 ring_buffer_write_tail(header, data_tail);
5292 struct perf_buffer_params {
5293 struct perf_event_attr *attr;
5294 /* if event_cb is specified, it takes precendence */
5295 perf_buffer_event_fn event_cb;
5296 /* sample_cb and lost_cb are higher-level common-case callbacks */
5297 perf_buffer_sample_fn sample_cb;
5298 perf_buffer_lost_fn lost_cb;
5305 struct perf_cpu_buf {
5306 struct perf_buffer *pb;
5307 void *base; /* mmap()'ed memory */
5308 void *buf; /* for reconstructing segmented data */
5315 struct perf_buffer {
5316 perf_buffer_event_fn event_cb;
5317 perf_buffer_sample_fn sample_cb;
5318 perf_buffer_lost_fn lost_cb;
5319 void *ctx; /* passed into callbacks */
5323 struct perf_cpu_buf **cpu_bufs;
5324 struct epoll_event *events;
5326 int epoll_fd; /* perf event FD */
5327 int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
5330 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
5331 struct perf_cpu_buf *cpu_buf)
5335 if (cpu_buf->base &&
5336 munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
5337 pr_warning("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
5338 if (cpu_buf->fd >= 0) {
5339 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
5346 void perf_buffer__free(struct perf_buffer *pb)
5353 for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
5354 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
5356 bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
5357 perf_buffer__free_cpu_buf(pb, cpu_buf);
5361 if (pb->epoll_fd >= 0)
5362 close(pb->epoll_fd);
5367 static struct perf_cpu_buf *
5368 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
5369 int cpu, int map_key)
5371 struct perf_cpu_buf *cpu_buf;
5372 char msg[STRERR_BUFSIZE];
5375 cpu_buf = calloc(1, sizeof(*cpu_buf));
5377 return ERR_PTR(-ENOMEM);
5381 cpu_buf->map_key = map_key;
5383 cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
5384 -1, PERF_FLAG_FD_CLOEXEC);
5385 if (cpu_buf->fd < 0) {
5387 pr_warning("failed to open perf buffer event on cpu #%d: %s\n",
5388 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5392 cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
5393 PROT_READ | PROT_WRITE, MAP_SHARED,
5395 if (cpu_buf->base == MAP_FAILED) {
5396 cpu_buf->base = NULL;
5398 pr_warning("failed to mmap perf buffer on cpu #%d: %s\n",
5399 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5403 if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
5405 pr_warning("failed to enable perf buffer event on cpu #%d: %s\n",
5406 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5413 perf_buffer__free_cpu_buf(pb, cpu_buf);
5414 return (struct perf_cpu_buf *)ERR_PTR(err);
5417 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
5418 struct perf_buffer_params *p);
5420 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
5421 const struct perf_buffer_opts *opts)
5423 struct perf_buffer_params p = {};
5424 struct perf_event_attr attr = { 0, };
5426 attr.config = PERF_COUNT_SW_BPF_OUTPUT,
5427 attr.type = PERF_TYPE_SOFTWARE;
5428 attr.sample_type = PERF_SAMPLE_RAW;
5429 attr.sample_period = 1;
5430 attr.wakeup_events = 1;
5433 p.sample_cb = opts ? opts->sample_cb : NULL;
5434 p.lost_cb = opts ? opts->lost_cb : NULL;
5435 p.ctx = opts ? opts->ctx : NULL;
5437 return __perf_buffer__new(map_fd, page_cnt, &p);
5440 struct perf_buffer *
5441 perf_buffer__new_raw(int map_fd, size_t page_cnt,
5442 const struct perf_buffer_raw_opts *opts)
5444 struct perf_buffer_params p = {};
5446 p.attr = opts->attr;
5447 p.event_cb = opts->event_cb;
5449 p.cpu_cnt = opts->cpu_cnt;
5450 p.cpus = opts->cpus;
5451 p.map_keys = opts->map_keys;
5453 return __perf_buffer__new(map_fd, page_cnt, &p);
5456 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
5457 struct perf_buffer_params *p)
5459 struct bpf_map_info map = {};
5460 char msg[STRERR_BUFSIZE];
5461 struct perf_buffer *pb;
5465 if (page_cnt & (page_cnt - 1)) {
5466 pr_warning("page count should be power of two, but is %zu\n",
5468 return ERR_PTR(-EINVAL);
5471 map_info_len = sizeof(map);
5472 err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
5475 pr_warning("failed to get map info for map FD %d: %s\n",
5476 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
5477 return ERR_PTR(err);
5480 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
5481 pr_warning("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
5483 return ERR_PTR(-EINVAL);
5486 pb = calloc(1, sizeof(*pb));
5488 return ERR_PTR(-ENOMEM);
5490 pb->event_cb = p->event_cb;
5491 pb->sample_cb = p->sample_cb;
5492 pb->lost_cb = p->lost_cb;
5495 pb->page_size = getpagesize();
5496 pb->mmap_size = pb->page_size * page_cnt;
5497 pb->map_fd = map_fd;
5499 pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
5500 if (pb->epoll_fd < 0) {
5502 pr_warning("failed to create epoll instance: %s\n",
5503 libbpf_strerror_r(err, msg, sizeof(msg)));
5507 if (p->cpu_cnt > 0) {
5508 pb->cpu_cnt = p->cpu_cnt;
5510 pb->cpu_cnt = libbpf_num_possible_cpus();
5511 if (pb->cpu_cnt < 0) {
5515 if (map.max_entries < pb->cpu_cnt)
5516 pb->cpu_cnt = map.max_entries;
5519 pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
5522 pr_warning("failed to allocate events: out of memory\n");
5525 pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
5526 if (!pb->cpu_bufs) {
5528 pr_warning("failed to allocate buffers: out of memory\n");
5532 for (i = 0; i < pb->cpu_cnt; i++) {
5533 struct perf_cpu_buf *cpu_buf;
5536 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
5537 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
5539 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
5540 if (IS_ERR(cpu_buf)) {
5541 err = PTR_ERR(cpu_buf);
5545 pb->cpu_bufs[i] = cpu_buf;
5547 err = bpf_map_update_elem(pb->map_fd, &map_key,
5551 pr_warning("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
5552 cpu, map_key, cpu_buf->fd,
5553 libbpf_strerror_r(err, msg, sizeof(msg)));
5557 pb->events[i].events = EPOLLIN;
5558 pb->events[i].data.ptr = cpu_buf;
5559 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
5560 &pb->events[i]) < 0) {
5562 pr_warning("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
5564 libbpf_strerror_r(err, msg, sizeof(msg)));
5573 perf_buffer__free(pb);
5574 return ERR_PTR(err);
5577 struct perf_sample_raw {
5578 struct perf_event_header header;
5583 struct perf_sample_lost {
5584 struct perf_event_header header;
5590 static enum bpf_perf_event_ret
5591 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
5593 struct perf_cpu_buf *cpu_buf = ctx;
5594 struct perf_buffer *pb = cpu_buf->pb;
5597 /* user wants full control over parsing perf event */
5599 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
5602 case PERF_RECORD_SAMPLE: {
5603 struct perf_sample_raw *s = data;
5606 pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
5609 case PERF_RECORD_LOST: {
5610 struct perf_sample_lost *s = data;
5613 pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
5617 pr_warning("unknown perf sample type %d\n", e->type);
5618 return LIBBPF_PERF_EVENT_ERROR;
5620 return LIBBPF_PERF_EVENT_CONT;
5623 static int perf_buffer__process_records(struct perf_buffer *pb,
5624 struct perf_cpu_buf *cpu_buf)
5626 enum bpf_perf_event_ret ret;
5628 ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
5629 pb->page_size, &cpu_buf->buf,
5631 perf_buffer__process_record, cpu_buf);
5632 if (ret != LIBBPF_PERF_EVENT_CONT)
5637 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
5641 cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
5642 for (i = 0; i < cnt; i++) {
5643 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
5645 err = perf_buffer__process_records(pb, cpu_buf);
5647 pr_warning("error while processing records: %d\n", err);
5651 return cnt < 0 ? -errno : cnt;
5654 struct bpf_prog_info_array_desc {
5655 int array_offset; /* e.g. offset of jited_prog_insns */
5656 int count_offset; /* e.g. offset of jited_prog_len */
5657 int size_offset; /* > 0: offset of rec size,
5658 * < 0: fix size of -size_offset
5662 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
5663 [BPF_PROG_INFO_JITED_INSNS] = {
5664 offsetof(struct bpf_prog_info, jited_prog_insns),
5665 offsetof(struct bpf_prog_info, jited_prog_len),
5668 [BPF_PROG_INFO_XLATED_INSNS] = {
5669 offsetof(struct bpf_prog_info, xlated_prog_insns),
5670 offsetof(struct bpf_prog_info, xlated_prog_len),
5673 [BPF_PROG_INFO_MAP_IDS] = {
5674 offsetof(struct bpf_prog_info, map_ids),
5675 offsetof(struct bpf_prog_info, nr_map_ids),
5676 -(int)sizeof(__u32),
5678 [BPF_PROG_INFO_JITED_KSYMS] = {
5679 offsetof(struct bpf_prog_info, jited_ksyms),
5680 offsetof(struct bpf_prog_info, nr_jited_ksyms),
5681 -(int)sizeof(__u64),
5683 [BPF_PROG_INFO_JITED_FUNC_LENS] = {
5684 offsetof(struct bpf_prog_info, jited_func_lens),
5685 offsetof(struct bpf_prog_info, nr_jited_func_lens),
5686 -(int)sizeof(__u32),
5688 [BPF_PROG_INFO_FUNC_INFO] = {
5689 offsetof(struct bpf_prog_info, func_info),
5690 offsetof(struct bpf_prog_info, nr_func_info),
5691 offsetof(struct bpf_prog_info, func_info_rec_size),
5693 [BPF_PROG_INFO_LINE_INFO] = {
5694 offsetof(struct bpf_prog_info, line_info),
5695 offsetof(struct bpf_prog_info, nr_line_info),
5696 offsetof(struct bpf_prog_info, line_info_rec_size),
5698 [BPF_PROG_INFO_JITED_LINE_INFO] = {
5699 offsetof(struct bpf_prog_info, jited_line_info),
5700 offsetof(struct bpf_prog_info, nr_jited_line_info),
5701 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
5703 [BPF_PROG_INFO_PROG_TAGS] = {
5704 offsetof(struct bpf_prog_info, prog_tags),
5705 offsetof(struct bpf_prog_info, nr_prog_tags),
5706 -(int)sizeof(__u8) * BPF_TAG_SIZE,
5711 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, int offset)
5713 __u32 *array = (__u32 *)info;
5716 return array[offset / sizeof(__u32)];
5717 return -(int)offset;
5720 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, int offset)
5722 __u64 *array = (__u64 *)info;
5725 return array[offset / sizeof(__u64)];
5726 return -(int)offset;
5729 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
5732 __u32 *array = (__u32 *)info;
5735 array[offset / sizeof(__u32)] = val;
5738 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
5741 __u64 *array = (__u64 *)info;
5744 array[offset / sizeof(__u64)] = val;
5747 struct bpf_prog_info_linear *
5748 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
5750 struct bpf_prog_info_linear *info_linear;
5751 struct bpf_prog_info info = {};
5752 __u32 info_len = sizeof(info);
5757 if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
5758 return ERR_PTR(-EINVAL);
5760 /* step 1: get array dimensions */
5761 err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
5763 pr_debug("can't get prog info: %s", strerror(errno));
5764 return ERR_PTR(-EFAULT);
5767 /* step 2: calculate total size of all arrays */
5768 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5769 bool include_array = (arrays & (1UL << i)) > 0;
5770 struct bpf_prog_info_array_desc *desc;
5773 desc = bpf_prog_info_array_desc + i;
5775 /* kernel is too old to support this field */
5776 if (info_len < desc->array_offset + sizeof(__u32) ||
5777 info_len < desc->count_offset + sizeof(__u32) ||
5778 (desc->size_offset > 0 && info_len < desc->size_offset))
5779 include_array = false;
5781 if (!include_array) {
5782 arrays &= ~(1UL << i); /* clear the bit */
5786 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5787 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5789 data_len += count * size;
5792 /* step 3: allocate continuous memory */
5793 data_len = roundup(data_len, sizeof(__u64));
5794 info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
5796 return ERR_PTR(-ENOMEM);
5798 /* step 4: fill data to info_linear->info */
5799 info_linear->arrays = arrays;
5800 memset(&info_linear->info, 0, sizeof(info));
5801 ptr = info_linear->data;
5803 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5804 struct bpf_prog_info_array_desc *desc;
5807 if ((arrays & (1UL << i)) == 0)
5810 desc = bpf_prog_info_array_desc + i;
5811 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5812 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5813 bpf_prog_info_set_offset_u32(&info_linear->info,
5814 desc->count_offset, count);
5815 bpf_prog_info_set_offset_u32(&info_linear->info,
5816 desc->size_offset, size);
5817 bpf_prog_info_set_offset_u64(&info_linear->info,
5820 ptr += count * size;
5823 /* step 5: call syscall again to get required arrays */
5824 err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
5826 pr_debug("can't get prog info: %s", strerror(errno));
5828 return ERR_PTR(-EFAULT);
5831 /* step 6: verify the data */
5832 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5833 struct bpf_prog_info_array_desc *desc;
5836 if ((arrays & (1UL << i)) == 0)
5839 desc = bpf_prog_info_array_desc + i;
5840 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5841 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
5842 desc->count_offset);
5844 pr_warning("%s: mismatch in element count\n", __func__);
5846 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5847 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
5850 pr_warning("%s: mismatch in rec size\n", __func__);
5853 /* step 7: update info_len and data_len */
5854 info_linear->info_len = sizeof(struct bpf_prog_info);
5855 info_linear->data_len = data_len;
5860 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
5864 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5865 struct bpf_prog_info_array_desc *desc;
5868 if ((info_linear->arrays & (1UL << i)) == 0)
5871 desc = bpf_prog_info_array_desc + i;
5872 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
5873 desc->array_offset);
5874 offs = addr - ptr_to_u64(info_linear->data);
5875 bpf_prog_info_set_offset_u64(&info_linear->info,
5876 desc->array_offset, offs);
5880 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
5884 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5885 struct bpf_prog_info_array_desc *desc;
5888 if ((info_linear->arrays & (1UL << i)) == 0)
5891 desc = bpf_prog_info_array_desc + i;
5892 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
5893 desc->array_offset);
5894 addr = offs + ptr_to_u64(info_linear->data);
5895 bpf_prog_info_set_offset_u64(&info_linear->info,
5896 desc->array_offset, addr);
5900 int libbpf_num_possible_cpus(void)
5902 static const char *fcpu = "/sys/devices/system/cpu/possible";
5903 int len = 0, n = 0, il = 0, ir = 0;
5904 unsigned int start = 0, end = 0;
5911 tmp_cpus = READ_ONCE(cpus);
5915 fd = open(fcpu, O_RDONLY);
5918 pr_warning("Failed to open file %s: %s\n",
5919 fcpu, strerror(error));
5922 len = read(fd, buf, sizeof(buf));
5925 error = len ? errno : EINVAL;
5926 pr_warning("Failed to read # of possible cpus from %s: %s\n",
5927 fcpu, strerror(error));
5930 if (len == sizeof(buf)) {
5931 pr_warning("File %s size overflow\n", fcpu);
5936 for (ir = 0, tmp_cpus = 0; ir <= len; ir++) {
5937 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
5938 if (buf[ir] == ',' || buf[ir] == '\0') {
5940 n = sscanf(&buf[il], "%u-%u", &start, &end);
5942 pr_warning("Failed to get # CPUs from %s\n",
5945 } else if (n == 1) {
5948 tmp_cpus += end - start + 1;
5952 if (tmp_cpus <= 0) {
5953 pr_warning("Invalid #CPUs %d from %s\n", tmp_cpus, fcpu);
5957 WRITE_ONCE(cpus, tmp_cpus);