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;
192 __u32 func_info_rec_size;
195 struct bpf_capabilities *caps;
198 __u32 line_info_rec_size;
203 enum libbpf_map_type {
210 static const char * const libbpf_type_to_btf_name[] = {
211 [LIBBPF_MAP_DATA] = ".data",
212 [LIBBPF_MAP_BSS] = ".bss",
213 [LIBBPF_MAP_RODATA] = ".rodata",
223 struct bpf_map_def def;
224 __u32 btf_key_type_id;
225 __u32 btf_value_type_id;
227 bpf_map_clear_priv_t clear_priv;
228 enum libbpf_map_type libbpf_type;
236 static LIST_HEAD(bpf_objects_list);
239 char name[BPF_OBJ_NAME_LEN];
243 struct bpf_program *programs;
245 struct bpf_map *maps;
248 struct bpf_secdata sections;
251 bool has_pseudo_calls;
254 * Information when doing elf related work. Only valid if fd
281 * All loaded bpf_object is linked in a list, which is
282 * hidden to caller. bpf_objects__<func> handlers deal with
285 struct list_head list;
288 struct btf_ext *btf_ext;
291 bpf_object_clear_priv_t clear_priv;
293 struct bpf_capabilities caps;
297 #define obj_elf_valid(o) ((o)->efile.elf)
299 void bpf_program__unload(struct bpf_program *prog)
307 * If the object is opened but the program was never loaded,
308 * it is possible that prog->instances.nr == -1.
310 if (prog->instances.nr > 0) {
311 for (i = 0; i < prog->instances.nr; i++)
312 zclose(prog->instances.fds[i]);
313 } else if (prog->instances.nr != -1) {
314 pr_warning("Internal error: instances.nr is %d\n",
318 prog->instances.nr = -1;
319 zfree(&prog->instances.fds);
321 zclose(prog->btf_fd);
322 zfree(&prog->func_info);
323 zfree(&prog->line_info);
326 static void bpf_program__exit(struct bpf_program *prog)
331 if (prog->clear_priv)
332 prog->clear_priv(prog, prog->priv);
335 prog->clear_priv = NULL;
337 bpf_program__unload(prog);
339 zfree(&prog->section_name);
340 zfree(&prog->pin_name);
342 zfree(&prog->reloc_desc);
349 static char *__bpf_program__pin_name(struct bpf_program *prog)
353 name = p = strdup(prog->section_name);
354 while ((p = strchr(p, '/')))
361 bpf_program__init(void *data, size_t size, char *section_name, int idx,
362 struct bpf_program *prog)
364 const size_t bpf_insn_sz = sizeof(struct bpf_insn);
366 if (size == 0 || size % bpf_insn_sz) {
367 pr_warning("corrupted section '%s', size: %zu\n",
372 memset(prog, 0, sizeof(*prog));
374 prog->section_name = strdup(section_name);
375 if (!prog->section_name) {
376 pr_warning("failed to alloc name for prog under section(%d) %s\n",
381 prog->pin_name = __bpf_program__pin_name(prog);
382 if (!prog->pin_name) {
383 pr_warning("failed to alloc pin name for prog under section(%d) %s\n",
388 prog->insns = malloc(size);
390 pr_warning("failed to alloc insns for prog under section %s\n",
394 prog->insns_cnt = size / bpf_insn_sz;
395 memcpy(prog->insns, data, size);
397 prog->instances.fds = NULL;
398 prog->instances.nr = -1;
399 prog->type = BPF_PROG_TYPE_UNSPEC;
404 bpf_program__exit(prog);
409 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
410 char *section_name, int idx)
412 struct bpf_program prog, *progs;
415 err = bpf_program__init(data, size, section_name, idx, &prog);
419 prog.caps = &obj->caps;
420 progs = obj->programs;
421 nr_progs = obj->nr_programs;
423 progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
426 * In this case the original obj->programs
427 * is still valid, so don't need special treat for
428 * bpf_close_object().
430 pr_warning("failed to alloc a new program under section '%s'\n",
432 bpf_program__exit(&prog);
436 pr_debug("found program %s\n", prog.section_name);
437 obj->programs = progs;
438 obj->nr_programs = nr_progs + 1;
440 progs[nr_progs] = prog;
445 bpf_object__init_prog_names(struct bpf_object *obj)
447 Elf_Data *symbols = obj->efile.symbols;
448 struct bpf_program *prog;
451 for (pi = 0; pi < obj->nr_programs; pi++) {
452 const char *name = NULL;
454 prog = &obj->programs[pi];
456 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
460 if (!gelf_getsym(symbols, si, &sym))
462 if (sym.st_shndx != prog->idx)
464 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
467 name = elf_strptr(obj->efile.elf,
468 obj->efile.strtabidx,
471 pr_warning("failed to get sym name string for prog %s\n",
473 return -LIBBPF_ERRNO__LIBELF;
477 if (!name && prog->idx == obj->efile.text_shndx)
481 pr_warning("failed to find sym for prog %s\n",
486 prog->name = strdup(name);
488 pr_warning("failed to allocate memory for prog sym %s\n",
497 static struct bpf_object *bpf_object__new(const char *path,
501 struct bpf_object *obj;
504 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
506 pr_warning("alloc memory failed for %s\n", path);
507 return ERR_PTR(-ENOMEM);
510 strcpy(obj->path, path);
511 /* Using basename() GNU version which doesn't modify arg. */
512 strncpy(obj->name, basename((void *)path), sizeof(obj->name) - 1);
513 end = strchr(obj->name, '.');
519 * Caller of this function should also call
520 * bpf_object__elf_finish() after data collection to return
521 * obj_buf to user. If not, we should duplicate the buffer to
522 * avoid user freeing them before elf finish.
524 obj->efile.obj_buf = obj_buf;
525 obj->efile.obj_buf_sz = obj_buf_sz;
526 obj->efile.maps_shndx = -1;
527 obj->efile.btf_maps_shndx = -1;
528 obj->efile.data_shndx = -1;
529 obj->efile.rodata_shndx = -1;
530 obj->efile.bss_shndx = -1;
534 INIT_LIST_HEAD(&obj->list);
535 list_add(&obj->list, &bpf_objects_list);
539 static void bpf_object__elf_finish(struct bpf_object *obj)
541 if (!obj_elf_valid(obj))
544 if (obj->efile.elf) {
545 elf_end(obj->efile.elf);
546 obj->efile.elf = NULL;
548 obj->efile.symbols = NULL;
549 obj->efile.data = NULL;
550 obj->efile.rodata = NULL;
551 obj->efile.bss = NULL;
553 zfree(&obj->efile.reloc);
554 obj->efile.nr_reloc = 0;
555 zclose(obj->efile.fd);
556 obj->efile.obj_buf = NULL;
557 obj->efile.obj_buf_sz = 0;
560 static int bpf_object__elf_init(struct bpf_object *obj)
565 if (obj_elf_valid(obj)) {
566 pr_warning("elf init: internal error\n");
567 return -LIBBPF_ERRNO__LIBELF;
570 if (obj->efile.obj_buf_sz > 0) {
572 * obj_buf should have been validated by
573 * bpf_object__open_buffer().
575 obj->efile.elf = elf_memory(obj->efile.obj_buf,
576 obj->efile.obj_buf_sz);
578 obj->efile.fd = open(obj->path, O_RDONLY);
579 if (obj->efile.fd < 0) {
580 char errmsg[STRERR_BUFSIZE], *cp;
583 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
584 pr_warning("failed to open %s: %s\n", obj->path, cp);
588 obj->efile.elf = elf_begin(obj->efile.fd,
589 LIBBPF_ELF_C_READ_MMAP, NULL);
592 if (!obj->efile.elf) {
593 pr_warning("failed to open %s as ELF file\n", obj->path);
594 err = -LIBBPF_ERRNO__LIBELF;
598 if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
599 pr_warning("failed to get EHDR from %s\n", obj->path);
600 err = -LIBBPF_ERRNO__FORMAT;
603 ep = &obj->efile.ehdr;
605 /* Old LLVM set e_machine to EM_NONE */
606 if (ep->e_type != ET_REL ||
607 (ep->e_machine && ep->e_machine != EM_BPF)) {
608 pr_warning("%s is not an eBPF object file\n", obj->path);
609 err = -LIBBPF_ERRNO__FORMAT;
615 bpf_object__elf_finish(obj);
619 static int bpf_object__check_endianness(struct bpf_object *obj)
621 #if __BYTE_ORDER == __LITTLE_ENDIAN
622 if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
624 #elif __BYTE_ORDER == __BIG_ENDIAN
625 if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
628 # error "Unrecognized __BYTE_ORDER__"
630 pr_warning("endianness mismatch.\n");
631 return -LIBBPF_ERRNO__ENDIAN;
635 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
637 memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
638 pr_debug("license of %s is %s\n", obj->path, obj->license);
643 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
647 if (size != sizeof(kver)) {
648 pr_warning("invalid kver section in %s\n", obj->path);
649 return -LIBBPF_ERRNO__FORMAT;
651 memcpy(&kver, data, sizeof(kver));
652 obj->kern_version = kver;
653 pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
657 static int compare_bpf_map(const void *_a, const void *_b)
659 const struct bpf_map *a = _a;
660 const struct bpf_map *b = _b;
662 if (a->sec_idx != b->sec_idx)
663 return a->sec_idx - b->sec_idx;
664 return a->sec_offset - b->sec_offset;
667 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
669 if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
670 type == BPF_MAP_TYPE_HASH_OF_MAPS)
675 static int bpf_object_search_section_size(const struct bpf_object *obj,
676 const char *name, size_t *d_size)
678 const GElf_Ehdr *ep = &obj->efile.ehdr;
679 Elf *elf = obj->efile.elf;
683 while ((scn = elf_nextscn(elf, scn)) != NULL) {
684 const char *sec_name;
689 if (gelf_getshdr(scn, &sh) != &sh) {
690 pr_warning("failed to get section(%d) header from %s\n",
695 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
697 pr_warning("failed to get section(%d) name from %s\n",
702 if (strcmp(name, sec_name))
705 data = elf_getdata(scn, 0);
707 pr_warning("failed to get section(%d) data from %s(%s)\n",
708 idx, name, obj->path);
712 *d_size = data->d_size;
719 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
728 } else if (!strcmp(name, ".data")) {
730 *size = obj->efile.data->d_size;
731 } else if (!strcmp(name, ".bss")) {
733 *size = obj->efile.bss->d_size;
734 } else if (!strcmp(name, ".rodata")) {
735 if (obj->efile.rodata)
736 *size = obj->efile.rodata->d_size;
738 ret = bpf_object_search_section_size(obj, name, &d_size);
743 return *size ? 0 : ret;
746 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
749 Elf_Data *symbols = obj->efile.symbols;
756 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
759 if (!gelf_getsym(symbols, si, &sym))
761 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
762 GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
765 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
768 pr_warning("failed to get sym name string for var %s\n",
772 if (strcmp(name, sname) == 0) {
781 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
783 struct bpf_map *new_maps;
787 if (obj->nr_maps < obj->maps_cap)
788 return &obj->maps[obj->nr_maps++];
790 new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
791 new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
793 pr_warning("alloc maps for object failed\n");
794 return ERR_PTR(-ENOMEM);
797 obj->maps_cap = new_cap;
798 obj->maps = new_maps;
800 /* zero out new maps */
801 memset(obj->maps + obj->nr_maps, 0,
802 (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
804 * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
805 * when failure (zclose won't close negative fd)).
807 for (i = obj->nr_maps; i < obj->maps_cap; i++) {
808 obj->maps[i].fd = -1;
809 obj->maps[i].inner_map_fd = -1;
812 return &obj->maps[obj->nr_maps++];
816 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
817 int sec_idx, Elf_Data *data, void **data_buff)
819 char map_name[BPF_OBJ_NAME_LEN];
820 struct bpf_map_def *def;
823 map = bpf_object__add_map(obj);
827 map->libbpf_type = type;
828 map->sec_idx = sec_idx;
830 snprintf(map_name, sizeof(map_name), "%.8s%.7s", obj->name,
831 libbpf_type_to_btf_name[type]);
832 map->name = strdup(map_name);
834 pr_warning("failed to alloc map name\n");
837 pr_debug("map '%s' (global data): at sec_idx %d, offset %zu.\n",
838 map_name, map->sec_idx, map->sec_offset);
841 def->type = BPF_MAP_TYPE_ARRAY;
842 def->key_size = sizeof(int);
843 def->value_size = data->d_size;
844 def->max_entries = 1;
845 def->map_flags = type == LIBBPF_MAP_RODATA ? BPF_F_RDONLY_PROG : 0;
847 *data_buff = malloc(data->d_size);
850 pr_warning("failed to alloc map content buffer\n");
853 memcpy(*data_buff, data->d_buf, data->d_size);
856 pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
860 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
864 if (!obj->caps.global_data)
867 * Populate obj->maps with libbpf internal maps.
869 if (obj->efile.data_shndx >= 0) {
870 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
871 obj->efile.data_shndx,
873 &obj->sections.data);
877 if (obj->efile.rodata_shndx >= 0) {
878 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
879 obj->efile.rodata_shndx,
881 &obj->sections.rodata);
885 if (obj->efile.bss_shndx >= 0) {
886 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
887 obj->efile.bss_shndx,
888 obj->efile.bss, NULL);
895 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
897 Elf_Data *symbols = obj->efile.symbols;
898 int i, map_def_sz = 0, nr_maps = 0, nr_syms;
899 Elf_Data *data = NULL;
902 if (obj->efile.maps_shndx < 0)
908 scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
910 data = elf_getdata(scn, NULL);
912 pr_warning("failed to get Elf_Data from map section %d\n",
913 obj->efile.maps_shndx);
918 * Count number of maps. Each map has a name.
919 * Array of maps is not supported: only the first element is
922 * TODO: Detect array of map and report error.
924 nr_syms = symbols->d_size / sizeof(GElf_Sym);
925 for (i = 0; i < nr_syms; i++) {
928 if (!gelf_getsym(symbols, i, &sym))
930 if (sym.st_shndx != obj->efile.maps_shndx)
934 /* Assume equally sized map definitions */
935 pr_debug("maps in %s: %d maps in %zd bytes\n",
936 obj->path, nr_maps, data->d_size);
938 map_def_sz = data->d_size / nr_maps;
939 if (!data->d_size || (data->d_size % nr_maps) != 0) {
940 pr_warning("unable to determine map definition size "
941 "section %s, %d maps in %zd bytes\n",
942 obj->path, nr_maps, data->d_size);
946 /* Fill obj->maps using data in "maps" section. */
947 for (i = 0; i < nr_syms; i++) {
949 const char *map_name;
950 struct bpf_map_def *def;
953 if (!gelf_getsym(symbols, i, &sym))
955 if (sym.st_shndx != obj->efile.maps_shndx)
958 map = bpf_object__add_map(obj);
962 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
965 pr_warning("failed to get map #%d name sym string for obj %s\n",
967 return -LIBBPF_ERRNO__FORMAT;
970 map->libbpf_type = LIBBPF_MAP_UNSPEC;
971 map->sec_idx = sym.st_shndx;
972 map->sec_offset = sym.st_value;
973 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
974 map_name, map->sec_idx, map->sec_offset);
975 if (sym.st_value + map_def_sz > data->d_size) {
976 pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
977 obj->path, map_name);
981 map->name = strdup(map_name);
983 pr_warning("failed to alloc map name\n");
986 pr_debug("map %d is \"%s\"\n", i, map->name);
987 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
989 * If the definition of the map in the object file fits in
990 * bpf_map_def, copy it. Any extra fields in our version
991 * of bpf_map_def will default to zero as a result of the
994 if (map_def_sz <= sizeof(struct bpf_map_def)) {
995 memcpy(&map->def, def, map_def_sz);
998 * Here the map structure being read is bigger than what
999 * we expect, truncate if the excess bits are all zero.
1000 * If they are not zero, reject this map as
1004 for (b = ((char *)def) + sizeof(struct bpf_map_def);
1005 b < ((char *)def) + map_def_sz; b++) {
1007 pr_warning("maps section in %s: \"%s\" "
1008 "has unrecognized, non-zero "
1010 obj->path, map_name);
1015 memcpy(&map->def, def, sizeof(struct bpf_map_def));
1021 static const struct btf_type *
1022 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1024 const struct btf_type *t = btf__type_by_id(btf, id);
1029 while (btf_is_mod(t) || btf_is_typedef(t)) {
1032 t = btf__type_by_id(btf, t->type);
1039 * Fetch integer attribute of BTF map definition. Such attributes are
1040 * represented using a pointer to an array, in which dimensionality of array
1041 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1042 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1043 * type definition, while using only sizeof(void *) space in ELF data section.
1045 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1046 const struct btf_type *def,
1047 const struct btf_member *m, __u32 *res) {
1048 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
1049 const char *name = btf__name_by_offset(btf, m->name_off);
1050 const struct btf_array *arr_info;
1051 const struct btf_type *arr_t;
1053 if (!btf_is_ptr(t)) {
1054 pr_warning("map '%s': attr '%s': expected PTR, got %u.\n",
1055 map_name, name, btf_kind(t));
1059 arr_t = btf__type_by_id(btf, t->type);
1061 pr_warning("map '%s': attr '%s': type [%u] not found.\n",
1062 map_name, name, t->type);
1065 if (!btf_is_array(arr_t)) {
1066 pr_warning("map '%s': attr '%s': expected ARRAY, got %u.\n",
1067 map_name, name, btf_kind(arr_t));
1070 arr_info = btf_array(arr_t);
1071 *res = arr_info->nelems;
1075 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1076 const struct btf_type *sec,
1077 int var_idx, int sec_idx,
1078 const Elf_Data *data, bool strict)
1080 const struct btf_type *var, *def, *t;
1081 const struct btf_var_secinfo *vi;
1082 const struct btf_var *var_extra;
1083 const struct btf_member *m;
1084 const char *map_name;
1085 struct bpf_map *map;
1088 vi = btf_var_secinfos(sec) + var_idx;
1089 var = btf__type_by_id(obj->btf, vi->type);
1090 var_extra = btf_var(var);
1091 map_name = btf__name_by_offset(obj->btf, var->name_off);
1092 vlen = btf_vlen(var);
1094 if (map_name == NULL || map_name[0] == '\0') {
1095 pr_warning("map #%d: empty name.\n", var_idx);
1098 if ((__u64)vi->offset + vi->size > data->d_size) {
1099 pr_warning("map '%s' BTF data is corrupted.\n", map_name);
1102 if (!btf_is_var(var)) {
1103 pr_warning("map '%s': unexpected var kind %u.\n",
1104 map_name, btf_kind(var));
1107 if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1108 var_extra->linkage != BTF_VAR_STATIC) {
1109 pr_warning("map '%s': unsupported var linkage %u.\n",
1110 map_name, var_extra->linkage);
1114 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
1115 if (!btf_is_struct(def)) {
1116 pr_warning("map '%s': unexpected def kind %u.\n",
1117 map_name, btf_kind(var));
1120 if (def->size > vi->size) {
1121 pr_warning("map '%s': invalid def size.\n", map_name);
1125 map = bpf_object__add_map(obj);
1127 return PTR_ERR(map);
1128 map->name = strdup(map_name);
1130 pr_warning("map '%s': failed to alloc map name.\n", map_name);
1133 map->libbpf_type = LIBBPF_MAP_UNSPEC;
1134 map->def.type = BPF_MAP_TYPE_UNSPEC;
1135 map->sec_idx = sec_idx;
1136 map->sec_offset = vi->offset;
1137 pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1138 map_name, map->sec_idx, map->sec_offset);
1140 vlen = btf_vlen(def);
1141 m = btf_members(def);
1142 for (i = 0; i < vlen; i++, m++) {
1143 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1146 pr_warning("map '%s': invalid field #%d.\n",
1150 if (strcmp(name, "type") == 0) {
1151 if (!get_map_field_int(map_name, obj->btf, def, m,
1154 pr_debug("map '%s': found type = %u.\n",
1155 map_name, map->def.type);
1156 } else if (strcmp(name, "max_entries") == 0) {
1157 if (!get_map_field_int(map_name, obj->btf, def, m,
1158 &map->def.max_entries))
1160 pr_debug("map '%s': found max_entries = %u.\n",
1161 map_name, map->def.max_entries);
1162 } else if (strcmp(name, "map_flags") == 0) {
1163 if (!get_map_field_int(map_name, obj->btf, def, m,
1164 &map->def.map_flags))
1166 pr_debug("map '%s': found map_flags = %u.\n",
1167 map_name, map->def.map_flags);
1168 } else if (strcmp(name, "key_size") == 0) {
1171 if (!get_map_field_int(map_name, obj->btf, def, m,
1174 pr_debug("map '%s': found key_size = %u.\n",
1176 if (map->def.key_size && map->def.key_size != sz) {
1177 pr_warning("map '%s': conflicting key size %u != %u.\n",
1178 map_name, map->def.key_size, sz);
1181 map->def.key_size = sz;
1182 } else if (strcmp(name, "key") == 0) {
1185 t = btf__type_by_id(obj->btf, m->type);
1187 pr_warning("map '%s': key type [%d] not found.\n",
1191 if (!btf_is_ptr(t)) {
1192 pr_warning("map '%s': key spec is not PTR: %u.\n",
1193 map_name, btf_kind(t));
1196 sz = btf__resolve_size(obj->btf, t->type);
1198 pr_warning("map '%s': can't determine key size for type [%u]: %lld.\n",
1199 map_name, t->type, sz);
1202 pr_debug("map '%s': found key [%u], sz = %lld.\n",
1203 map_name, t->type, sz);
1204 if (map->def.key_size && map->def.key_size != sz) {
1205 pr_warning("map '%s': conflicting key size %u != %lld.\n",
1206 map_name, map->def.key_size, sz);
1209 map->def.key_size = sz;
1210 map->btf_key_type_id = t->type;
1211 } else if (strcmp(name, "value_size") == 0) {
1214 if (!get_map_field_int(map_name, obj->btf, def, m,
1217 pr_debug("map '%s': found value_size = %u.\n",
1219 if (map->def.value_size && map->def.value_size != sz) {
1220 pr_warning("map '%s': conflicting value size %u != %u.\n",
1221 map_name, map->def.value_size, sz);
1224 map->def.value_size = sz;
1225 } else if (strcmp(name, "value") == 0) {
1228 t = btf__type_by_id(obj->btf, m->type);
1230 pr_warning("map '%s': value type [%d] not found.\n",
1234 if (!btf_is_ptr(t)) {
1235 pr_warning("map '%s': value spec is not PTR: %u.\n",
1236 map_name, btf_kind(t));
1239 sz = btf__resolve_size(obj->btf, t->type);
1241 pr_warning("map '%s': can't determine value size for type [%u]: %lld.\n",
1242 map_name, t->type, sz);
1245 pr_debug("map '%s': found value [%u], sz = %lld.\n",
1246 map_name, t->type, sz);
1247 if (map->def.value_size && map->def.value_size != sz) {
1248 pr_warning("map '%s': conflicting value size %u != %lld.\n",
1249 map_name, map->def.value_size, sz);
1252 map->def.value_size = sz;
1253 map->btf_value_type_id = t->type;
1256 pr_warning("map '%s': unknown field '%s'.\n",
1260 pr_debug("map '%s': ignoring unknown field '%s'.\n",
1265 if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
1266 pr_warning("map '%s': map type isn't specified.\n", map_name);
1273 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict)
1275 const struct btf_type *sec = NULL;
1276 int nr_types, i, vlen, err;
1277 const struct btf_type *t;
1282 if (obj->efile.btf_maps_shndx < 0)
1285 scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
1287 data = elf_getdata(scn, NULL);
1288 if (!scn || !data) {
1289 pr_warning("failed to get Elf_Data from map section %d (%s)\n",
1290 obj->efile.maps_shndx, MAPS_ELF_SEC);
1294 nr_types = btf__get_nr_types(obj->btf);
1295 for (i = 1; i <= nr_types; i++) {
1296 t = btf__type_by_id(obj->btf, i);
1297 if (!btf_is_datasec(t))
1299 name = btf__name_by_offset(obj->btf, t->name_off);
1300 if (strcmp(name, MAPS_ELF_SEC) == 0) {
1307 pr_warning("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
1311 vlen = btf_vlen(sec);
1312 for (i = 0; i < vlen; i++) {
1313 err = bpf_object__init_user_btf_map(obj, sec, i,
1314 obj->efile.btf_maps_shndx,
1323 static int bpf_object__init_maps(struct bpf_object *obj, int flags)
1325 bool strict = !(flags & MAPS_RELAX_COMPAT);
1328 err = bpf_object__init_user_maps(obj, strict);
1332 err = bpf_object__init_user_btf_maps(obj, strict);
1336 err = bpf_object__init_global_data_maps(obj);
1341 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]),
1347 static bool section_have_execinstr(struct bpf_object *obj, int idx)
1352 scn = elf_getscn(obj->efile.elf, idx);
1356 if (gelf_getshdr(scn, &sh) != &sh)
1359 if (sh.sh_flags & SHF_EXECINSTR)
1365 static void bpf_object__sanitize_btf(struct bpf_object *obj)
1367 bool has_datasec = obj->caps.btf_datasec;
1368 bool has_func = obj->caps.btf_func;
1369 struct btf *btf = obj->btf;
1373 if (!obj->btf || (has_func && has_datasec))
1376 for (i = 1; i <= btf__get_nr_types(btf); i++) {
1377 t = (struct btf_type *)btf__type_by_id(btf, i);
1379 if (!has_datasec && btf_is_var(t)) {
1380 /* replace VAR with INT */
1381 t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1383 * using size = 1 is the safest choice, 4 will be too
1384 * big and cause kernel BTF validation failure if
1385 * original variable took less than 4 bytes
1388 *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
1389 } else if (!has_datasec && btf_is_datasec(t)) {
1390 /* replace DATASEC with STRUCT */
1391 const struct btf_var_secinfo *v = btf_var_secinfos(t);
1392 struct btf_member *m = btf_members(t);
1393 struct btf_type *vt;
1396 name = (char *)btf__name_by_offset(btf, t->name_off);
1404 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
1405 for (j = 0; j < vlen; j++, v++, m++) {
1406 /* order of field assignments is important */
1407 m->offset = v->offset * 8;
1409 /* preserve variable name as member name */
1410 vt = (void *)btf__type_by_id(btf, v->type);
1411 m->name_off = vt->name_off;
1413 } else if (!has_func && btf_is_func_proto(t)) {
1414 /* replace FUNC_PROTO with ENUM */
1416 t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
1417 t->size = sizeof(__u32); /* kernel enforced */
1418 } else if (!has_func && btf_is_func(t)) {
1419 /* replace FUNC with TYPEDEF */
1420 t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
1425 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
1430 if (!obj->caps.btf_func) {
1431 btf_ext__free(obj->btf_ext);
1432 obj->btf_ext = NULL;
1436 static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
1438 return obj->efile.btf_maps_shndx >= 0;
1441 static int bpf_object__init_btf(struct bpf_object *obj,
1443 Elf_Data *btf_ext_data)
1445 bool btf_required = bpf_object__is_btf_mandatory(obj);
1449 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
1450 if (IS_ERR(obj->btf)) {
1451 pr_warning("Error loading ELF section %s: %d.\n",
1455 err = btf__finalize_data(obj, obj->btf);
1457 pr_warning("Error finalizing %s: %d.\n",
1464 pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
1465 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
1468 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
1469 btf_ext_data->d_size);
1470 if (IS_ERR(obj->btf_ext)) {
1471 pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
1472 BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
1473 obj->btf_ext = NULL;
1478 if (err || IS_ERR(obj->btf)) {
1480 err = err ? : PTR_ERR(obj->btf);
1483 if (!IS_ERR_OR_NULL(obj->btf))
1484 btf__free(obj->btf);
1487 if (btf_required && !obj->btf) {
1488 pr_warning("BTF is required, but is missing or corrupted.\n");
1489 return err == 0 ? -ENOENT : err;
1494 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
1501 bpf_object__sanitize_btf(obj);
1502 bpf_object__sanitize_btf_ext(obj);
1504 err = btf__load(obj->btf);
1506 pr_warning("Error loading %s into kernel: %d.\n",
1508 btf__free(obj->btf);
1510 /* btf_ext can't exist without btf, so free it as well */
1512 btf_ext__free(obj->btf_ext);
1513 obj->btf_ext = NULL;
1516 if (bpf_object__is_btf_mandatory(obj))
1522 static int bpf_object__elf_collect(struct bpf_object *obj, int flags)
1524 Elf *elf = obj->efile.elf;
1525 GElf_Ehdr *ep = &obj->efile.ehdr;
1526 Elf_Data *btf_ext_data = NULL;
1527 Elf_Data *btf_data = NULL;
1528 Elf_Scn *scn = NULL;
1529 int idx = 0, err = 0;
1531 /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1532 if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
1533 pr_warning("failed to get e_shstrndx from %s\n", obj->path);
1534 return -LIBBPF_ERRNO__FORMAT;
1537 while ((scn = elf_nextscn(elf, scn)) != NULL) {
1543 if (gelf_getshdr(scn, &sh) != &sh) {
1544 pr_warning("failed to get section(%d) header from %s\n",
1546 return -LIBBPF_ERRNO__FORMAT;
1549 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1551 pr_warning("failed to get section(%d) name from %s\n",
1553 return -LIBBPF_ERRNO__FORMAT;
1556 data = elf_getdata(scn, 0);
1558 pr_warning("failed to get section(%d) data from %s(%s)\n",
1559 idx, name, obj->path);
1560 return -LIBBPF_ERRNO__FORMAT;
1562 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
1563 idx, name, (unsigned long)data->d_size,
1564 (int)sh.sh_link, (unsigned long)sh.sh_flags,
1567 if (strcmp(name, "license") == 0) {
1568 err = bpf_object__init_license(obj,
1573 } else if (strcmp(name, "version") == 0) {
1574 err = bpf_object__init_kversion(obj,
1579 } else if (strcmp(name, "maps") == 0) {
1580 obj->efile.maps_shndx = idx;
1581 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
1582 obj->efile.btf_maps_shndx = idx;
1583 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
1585 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
1586 btf_ext_data = data;
1587 } else if (sh.sh_type == SHT_SYMTAB) {
1588 if (obj->efile.symbols) {
1589 pr_warning("bpf: multiple SYMTAB in %s\n",
1591 return -LIBBPF_ERRNO__FORMAT;
1593 obj->efile.symbols = data;
1594 obj->efile.strtabidx = sh.sh_link;
1595 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
1596 if (sh.sh_flags & SHF_EXECINSTR) {
1597 if (strcmp(name, ".text") == 0)
1598 obj->efile.text_shndx = idx;
1599 err = bpf_object__add_program(obj, data->d_buf,
1600 data->d_size, name, idx);
1602 char errmsg[STRERR_BUFSIZE];
1603 char *cp = libbpf_strerror_r(-err, errmsg,
1606 pr_warning("failed to alloc program %s (%s): %s",
1607 name, obj->path, cp);
1610 } else if (strcmp(name, ".data") == 0) {
1611 obj->efile.data = data;
1612 obj->efile.data_shndx = idx;
1613 } else if (strcmp(name, ".rodata") == 0) {
1614 obj->efile.rodata = data;
1615 obj->efile.rodata_shndx = idx;
1617 pr_debug("skip section(%d) %s\n", idx, name);
1619 } else if (sh.sh_type == SHT_REL) {
1620 int nr_reloc = obj->efile.nr_reloc;
1621 void *reloc = obj->efile.reloc;
1622 int sec = sh.sh_info; /* points to other section */
1624 /* Only do relo for section with exec instructions */
1625 if (!section_have_execinstr(obj, sec)) {
1626 pr_debug("skip relo %s(%d) for section(%d)\n",
1631 reloc = reallocarray(reloc, nr_reloc + 1,
1632 sizeof(*obj->efile.reloc));
1634 pr_warning("realloc failed\n");
1638 obj->efile.reloc = reloc;
1639 obj->efile.nr_reloc++;
1641 obj->efile.reloc[nr_reloc].shdr = sh;
1642 obj->efile.reloc[nr_reloc].data = data;
1643 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, ".bss") == 0) {
1644 obj->efile.bss = data;
1645 obj->efile.bss_shndx = idx;
1647 pr_debug("skip section(%d) %s\n", idx, name);
1651 if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
1652 pr_warning("Corrupted ELF file: index of strtab invalid\n");
1653 return -LIBBPF_ERRNO__FORMAT;
1655 err = bpf_object__init_btf(obj, btf_data, btf_ext_data);
1657 err = bpf_object__init_maps(obj, flags);
1659 err = bpf_object__sanitize_and_load_btf(obj);
1661 err = bpf_object__init_prog_names(obj);
1665 static struct bpf_program *
1666 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
1668 struct bpf_program *prog;
1671 for (i = 0; i < obj->nr_programs; i++) {
1672 prog = &obj->programs[i];
1673 if (prog->idx == idx)
1679 struct bpf_program *
1680 bpf_object__find_program_by_title(const struct bpf_object *obj,
1683 struct bpf_program *pos;
1685 bpf_object__for_each_program(pos, obj) {
1686 if (pos->section_name && !strcmp(pos->section_name, title))
1692 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
1695 return shndx == obj->efile.data_shndx ||
1696 shndx == obj->efile.bss_shndx ||
1697 shndx == obj->efile.rodata_shndx;
1700 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
1703 return shndx == obj->efile.maps_shndx ||
1704 shndx == obj->efile.btf_maps_shndx;
1707 static bool bpf_object__relo_in_known_section(const struct bpf_object *obj,
1710 return shndx == obj->efile.text_shndx ||
1711 bpf_object__shndx_is_maps(obj, shndx) ||
1712 bpf_object__shndx_is_data(obj, shndx);
1715 static enum libbpf_map_type
1716 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
1718 if (shndx == obj->efile.data_shndx)
1719 return LIBBPF_MAP_DATA;
1720 else if (shndx == obj->efile.bss_shndx)
1721 return LIBBPF_MAP_BSS;
1722 else if (shndx == obj->efile.rodata_shndx)
1723 return LIBBPF_MAP_RODATA;
1725 return LIBBPF_MAP_UNSPEC;
1729 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
1730 Elf_Data *data, struct bpf_object *obj)
1732 Elf_Data *symbols = obj->efile.symbols;
1733 struct bpf_map *maps = obj->maps;
1734 size_t nr_maps = obj->nr_maps;
1737 pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
1738 nrels = shdr->sh_size / shdr->sh_entsize;
1740 prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
1741 if (!prog->reloc_desc) {
1742 pr_warning("failed to alloc memory in relocation\n");
1745 prog->nr_reloc = nrels;
1747 for (i = 0; i < nrels; i++) {
1748 struct bpf_insn *insns = prog->insns;
1749 enum libbpf_map_type type;
1750 unsigned int insn_idx;
1751 unsigned int shdr_idx;
1757 if (!gelf_getrel(data, i, &rel)) {
1758 pr_warning("relocation: failed to get %d reloc\n", i);
1759 return -LIBBPF_ERRNO__FORMAT;
1762 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
1763 pr_warning("relocation: symbol %"PRIx64" not found\n",
1764 GELF_R_SYM(rel.r_info));
1765 return -LIBBPF_ERRNO__FORMAT;
1768 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1769 sym.st_name) ? : "<?>";
1771 pr_debug("relo for %lld value %lld name %d (\'%s\')\n",
1772 (long long) (rel.r_info >> 32),
1773 (long long) sym.st_value, sym.st_name, name);
1775 shdr_idx = sym.st_shndx;
1776 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
1777 pr_debug("relocation: insn_idx=%u, shdr_idx=%u\n",
1778 insn_idx, shdr_idx);
1780 if (shdr_idx >= SHN_LORESERVE) {
1781 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",
1782 name, shdr_idx, insn_idx,
1783 insns[insn_idx].code);
1784 return -LIBBPF_ERRNO__RELOC;
1786 if (!bpf_object__relo_in_known_section(obj, shdr_idx)) {
1787 pr_warning("Program '%s' contains unrecognized relo data pointing to section %u\n",
1788 prog->section_name, shdr_idx);
1789 return -LIBBPF_ERRNO__RELOC;
1792 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
1793 if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
1794 pr_warning("incorrect bpf_call opcode\n");
1795 return -LIBBPF_ERRNO__RELOC;
1797 prog->reloc_desc[i].type = RELO_CALL;
1798 prog->reloc_desc[i].insn_idx = insn_idx;
1799 prog->reloc_desc[i].text_off = sym.st_value;
1800 obj->has_pseudo_calls = true;
1804 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
1805 pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
1806 insn_idx, insns[insn_idx].code);
1807 return -LIBBPF_ERRNO__RELOC;
1810 if (bpf_object__shndx_is_maps(obj, shdr_idx) ||
1811 bpf_object__shndx_is_data(obj, shdr_idx)) {
1812 type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
1813 if (type != LIBBPF_MAP_UNSPEC) {
1814 if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL) {
1815 pr_warning("bpf: relocation: not yet supported relo for non-static global \'%s\' variable found in insns[%d].code 0x%x\n",
1816 name, insn_idx, insns[insn_idx].code);
1817 return -LIBBPF_ERRNO__RELOC;
1819 if (!obj->caps.global_data) {
1820 pr_warning("bpf: relocation: kernel does not support global \'%s\' variable access in insns[%d]\n",
1822 return -LIBBPF_ERRNO__RELOC;
1826 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
1827 if (maps[map_idx].libbpf_type != type)
1829 if (type != LIBBPF_MAP_UNSPEC ||
1830 (maps[map_idx].sec_idx == sym.st_shndx &&
1831 maps[map_idx].sec_offset == sym.st_value)) {
1832 pr_debug("relocation: found map %zd (%s, sec_idx %d, offset %zu) for insn %u\n",
1833 map_idx, maps[map_idx].name,
1834 maps[map_idx].sec_idx,
1835 maps[map_idx].sec_offset,
1841 if (map_idx >= nr_maps) {
1842 pr_warning("bpf relocation: map_idx %d larger than %d\n",
1843 (int)map_idx, (int)nr_maps - 1);
1844 return -LIBBPF_ERRNO__RELOC;
1847 prog->reloc_desc[i].type = type != LIBBPF_MAP_UNSPEC ?
1848 RELO_DATA : RELO_LD64;
1849 prog->reloc_desc[i].insn_idx = insn_idx;
1850 prog->reloc_desc[i].map_idx = map_idx;
1856 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
1858 struct bpf_map_def *def = &map->def;
1859 __u32 key_type_id = 0, value_type_id = 0;
1862 /* if it's BTF-defined map, we don't need to search for type IDs */
1863 if (map->sec_idx == obj->efile.btf_maps_shndx)
1866 if (!bpf_map__is_internal(map)) {
1867 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
1868 def->value_size, &key_type_id,
1872 * LLVM annotates global data differently in BTF, that is,
1873 * only as '.data', '.bss' or '.rodata'.
1875 ret = btf__find_by_name(obj->btf,
1876 libbpf_type_to_btf_name[map->libbpf_type]);
1881 map->btf_key_type_id = key_type_id;
1882 map->btf_value_type_id = bpf_map__is_internal(map) ?
1883 ret : value_type_id;
1887 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1889 struct bpf_map_info info = {};
1890 __u32 len = sizeof(info);
1894 err = bpf_obj_get_info_by_fd(fd, &info, &len);
1898 new_name = strdup(info.name);
1902 new_fd = open("/", O_RDONLY | O_CLOEXEC);
1904 goto err_free_new_name;
1906 new_fd = dup3(fd, new_fd, O_CLOEXEC);
1908 goto err_close_new_fd;
1910 err = zclose(map->fd);
1912 goto err_close_new_fd;
1916 map->name = new_name;
1917 map->def.type = info.type;
1918 map->def.key_size = info.key_size;
1919 map->def.value_size = info.value_size;
1920 map->def.max_entries = info.max_entries;
1921 map->def.map_flags = info.map_flags;
1922 map->btf_key_type_id = info.btf_key_type_id;
1923 map->btf_value_type_id = info.btf_value_type_id;
1934 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
1936 if (!map || !max_entries)
1939 /* If map already created, its attributes can't be changed. */
1943 map->def.max_entries = max_entries;
1949 bpf_object__probe_name(struct bpf_object *obj)
1951 struct bpf_load_program_attr attr;
1952 char *cp, errmsg[STRERR_BUFSIZE];
1953 struct bpf_insn insns[] = {
1954 BPF_MOV64_IMM(BPF_REG_0, 0),
1959 /* make sure basic loading works */
1961 memset(&attr, 0, sizeof(attr));
1962 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
1964 attr.insns_cnt = ARRAY_SIZE(insns);
1965 attr.license = "GPL";
1967 ret = bpf_load_program_xattr(&attr, NULL, 0);
1969 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1970 pr_warning("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
1971 __func__, cp, errno);
1976 /* now try the same program, but with the name */
1979 ret = bpf_load_program_xattr(&attr, NULL, 0);
1989 bpf_object__probe_global_data(struct bpf_object *obj)
1991 struct bpf_load_program_attr prg_attr;
1992 struct bpf_create_map_attr map_attr;
1993 char *cp, errmsg[STRERR_BUFSIZE];
1994 struct bpf_insn insns[] = {
1995 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
1996 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
1997 BPF_MOV64_IMM(BPF_REG_0, 0),
2002 memset(&map_attr, 0, sizeof(map_attr));
2003 map_attr.map_type = BPF_MAP_TYPE_ARRAY;
2004 map_attr.key_size = sizeof(int);
2005 map_attr.value_size = 32;
2006 map_attr.max_entries = 1;
2008 map = bpf_create_map_xattr(&map_attr);
2010 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2011 pr_warning("Error in %s():%s(%d). Couldn't create simple array map.\n",
2012 __func__, cp, errno);
2018 memset(&prg_attr, 0, sizeof(prg_attr));
2019 prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
2020 prg_attr.insns = insns;
2021 prg_attr.insns_cnt = ARRAY_SIZE(insns);
2022 prg_attr.license = "GPL";
2024 ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
2026 obj->caps.global_data = 1;
2034 static int bpf_object__probe_btf_func(struct bpf_object *obj)
2036 const char strs[] = "\0int\0x\0a";
2037 /* void x(int a) {} */
2040 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
2041 /* FUNC_PROTO */ /* [2] */
2042 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
2043 BTF_PARAM_ENC(7, 1),
2044 /* FUNC x */ /* [3] */
2045 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
2049 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2050 strs, sizeof(strs));
2052 obj->caps.btf_func = 1;
2060 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
2062 const char strs[] = "\0x\0.data";
2066 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4), /* [1] */
2067 /* VAR x */ /* [2] */
2068 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
2070 /* DATASEC val */ /* [3] */
2071 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
2072 BTF_VAR_SECINFO_ENC(2, 0, 4),
2076 btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2077 strs, sizeof(strs));
2079 obj->caps.btf_datasec = 1;
2088 bpf_object__probe_caps(struct bpf_object *obj)
2090 int (*probe_fn[])(struct bpf_object *obj) = {
2091 bpf_object__probe_name,
2092 bpf_object__probe_global_data,
2093 bpf_object__probe_btf_func,
2094 bpf_object__probe_btf_datasec,
2098 for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
2099 ret = probe_fn[i](obj);
2101 pr_debug("Probe #%d failed with %d.\n", i, ret);
2108 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
2110 char *cp, errmsg[STRERR_BUFSIZE];
2114 /* Nothing to do here since kernel already zero-initializes .bss map. */
2115 if (map->libbpf_type == LIBBPF_MAP_BSS)
2118 data = map->libbpf_type == LIBBPF_MAP_DATA ?
2119 obj->sections.data : obj->sections.rodata;
2121 err = bpf_map_update_elem(map->fd, &zero, data, 0);
2122 /* Freeze .rodata map as read-only from syscall side. */
2123 if (!err && map->libbpf_type == LIBBPF_MAP_RODATA) {
2124 err = bpf_map_freeze(map->fd);
2126 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2127 pr_warning("Error freezing map(%s) as read-only: %s\n",
2136 bpf_object__create_maps(struct bpf_object *obj)
2138 struct bpf_create_map_attr create_attr = {};
2143 for (i = 0; i < obj->nr_maps; i++) {
2144 struct bpf_map *map = &obj->maps[i];
2145 struct bpf_map_def *def = &map->def;
2146 char *cp, errmsg[STRERR_BUFSIZE];
2147 int *pfd = &map->fd;
2150 pr_debug("skip map create (preset) %s: fd=%d\n",
2151 map->name, map->fd);
2156 create_attr.name = map->name;
2157 create_attr.map_ifindex = map->map_ifindex;
2158 create_attr.map_type = def->type;
2159 create_attr.map_flags = def->map_flags;
2160 create_attr.key_size = def->key_size;
2161 create_attr.value_size = def->value_size;
2162 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
2163 !def->max_entries) {
2165 nr_cpus = libbpf_num_possible_cpus();
2167 pr_warning("failed to determine number of system CPUs: %d\n",
2172 pr_debug("map '%s': setting size to %d\n",
2173 map->name, nr_cpus);
2174 create_attr.max_entries = nr_cpus;
2176 create_attr.max_entries = def->max_entries;
2178 create_attr.btf_fd = 0;
2179 create_attr.btf_key_type_id = 0;
2180 create_attr.btf_value_type_id = 0;
2181 if (bpf_map_type__is_map_in_map(def->type) &&
2182 map->inner_map_fd >= 0)
2183 create_attr.inner_map_fd = map->inner_map_fd;
2185 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
2186 create_attr.btf_fd = btf__fd(obj->btf);
2187 create_attr.btf_key_type_id = map->btf_key_type_id;
2188 create_attr.btf_value_type_id = map->btf_value_type_id;
2191 *pfd = bpf_create_map_xattr(&create_attr);
2192 if (*pfd < 0 && (create_attr.btf_key_type_id ||
2193 create_attr.btf_value_type_id)) {
2195 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2196 pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
2197 map->name, cp, err);
2198 create_attr.btf_fd = 0;
2199 create_attr.btf_key_type_id = 0;
2200 create_attr.btf_value_type_id = 0;
2201 map->btf_key_type_id = 0;
2202 map->btf_value_type_id = 0;
2203 *pfd = bpf_create_map_xattr(&create_attr);
2211 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2212 pr_warning("failed to create map (name: '%s'): %s(%d)\n",
2213 map->name, cp, err);
2214 for (j = 0; j < i; j++)
2215 zclose(obj->maps[j].fd);
2219 if (bpf_map__is_internal(map)) {
2220 err = bpf_object__populate_internal_map(obj, map);
2227 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
2234 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
2235 void *btf_prog_info, const char *info_name)
2237 if (err != -ENOENT) {
2238 pr_warning("Error in loading %s for sec %s.\n",
2239 info_name, prog->section_name);
2243 /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
2245 if (btf_prog_info) {
2247 * Some info has already been found but has problem
2248 * in the last btf_ext reloc. Must have to error out.
2250 pr_warning("Error in relocating %s for sec %s.\n",
2251 info_name, prog->section_name);
2255 /* Have problem loading the very first info. Ignore the rest. */
2256 pr_warning("Cannot find %s for main program sec %s. Ignore all %s.\n",
2257 info_name, prog->section_name, info_name);
2262 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
2263 const char *section_name, __u32 insn_offset)
2267 if (!insn_offset || prog->func_info) {
2269 * !insn_offset => main program
2271 * For sub prog, the main program's func_info has to
2272 * be loaded first (i.e. prog->func_info != NULL)
2274 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
2275 section_name, insn_offset,
2277 &prog->func_info_cnt);
2279 return check_btf_ext_reloc_err(prog, err,
2283 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
2286 if (!insn_offset || prog->line_info) {
2287 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
2288 section_name, insn_offset,
2290 &prog->line_info_cnt);
2292 return check_btf_ext_reloc_err(prog, err,
2296 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
2300 prog->btf_fd = btf__fd(obj->btf);
2305 #define BPF_CORE_SPEC_MAX_LEN 64
2307 /* represents BPF CO-RE field or array element accessor */
2308 struct bpf_core_accessor {
2309 __u32 type_id; /* struct/union type or array element type */
2310 __u32 idx; /* field index or array index */
2311 const char *name; /* field name or NULL for array accessor */
2314 struct bpf_core_spec {
2315 const struct btf *btf;
2316 /* high-level spec: named fields and array indices only */
2317 struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
2318 /* high-level spec length */
2320 /* raw, low-level spec: 1-to-1 with accessor spec string */
2321 int raw_spec[BPF_CORE_SPEC_MAX_LEN];
2322 /* raw spec length */
2324 /* field byte offset represented by spec */
2328 static bool str_is_empty(const char *s)
2334 * Turn bpf_offset_reloc into a low- and high-level spec representation,
2335 * validating correctness along the way, as well as calculating resulting
2336 * field offset (in bytes), specified by accessor string. Low-level spec
2337 * captures every single level of nestedness, including traversing anonymous
2338 * struct/union members. High-level one only captures semantically meaningful
2339 * "turning points": named fields and array indicies.
2340 * E.g., for this case:
2343 * int __unimportant;
2351 * struct sample *s = ...;
2353 * int x = &s->a[3]; // access string = '0:1:2:3'
2355 * Low-level spec has 1:1 mapping with each element of access string (it's
2356 * just a parsed access string representation): [0, 1, 2, 3].
2358 * High-level spec will capture only 3 points:
2359 * - intial zero-index access by pointer (&s->... is the same as &s[0]...);
2360 * - field 'a' access (corresponds to '2' in low-level spec);
2361 * - array element #3 access (corresponds to '3' in low-level spec).
2364 static int bpf_core_spec_parse(const struct btf *btf,
2366 const char *spec_str,
2367 struct bpf_core_spec *spec)
2369 int access_idx, parsed_len, i;
2370 const struct btf_type *t;
2375 if (str_is_empty(spec_str) || *spec_str == ':')
2378 memset(spec, 0, sizeof(*spec));
2381 /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
2383 if (*spec_str == ':')
2385 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
2387 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2389 spec_str += parsed_len;
2390 spec->raw_spec[spec->raw_len++] = access_idx;
2393 if (spec->raw_len == 0)
2396 /* first spec value is always reloc type array index */
2397 t = skip_mods_and_typedefs(btf, type_id, &id);
2401 access_idx = spec->raw_spec[0];
2402 spec->spec[0].type_id = id;
2403 spec->spec[0].idx = access_idx;
2406 sz = btf__resolve_size(btf, id);
2409 spec->offset = access_idx * sz;
2411 for (i = 1; i < spec->raw_len; i++) {
2412 t = skip_mods_and_typedefs(btf, id, &id);
2416 access_idx = spec->raw_spec[i];
2418 if (btf_is_composite(t)) {
2419 const struct btf_member *m;
2422 if (access_idx >= btf_vlen(t))
2424 if (btf_member_bitfield_size(t, access_idx))
2427 offset = btf_member_bit_offset(t, access_idx);
2430 spec->offset += offset / 8;
2432 m = btf_members(t) + access_idx;
2434 name = btf__name_by_offset(btf, m->name_off);
2435 if (str_is_empty(name))
2438 spec->spec[spec->len].type_id = id;
2439 spec->spec[spec->len].idx = access_idx;
2440 spec->spec[spec->len].name = name;
2445 } else if (btf_is_array(t)) {
2446 const struct btf_array *a = btf_array(t);
2448 t = skip_mods_and_typedefs(btf, a->type, &id);
2449 if (!t || access_idx >= a->nelems)
2452 spec->spec[spec->len].type_id = id;
2453 spec->spec[spec->len].idx = access_idx;
2456 sz = btf__resolve_size(btf, id);
2459 spec->offset += access_idx * sz;
2461 pr_warning("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
2462 type_id, spec_str, i, id, btf_kind(t));
2470 static bool bpf_core_is_flavor_sep(const char *s)
2472 /* check X___Y name pattern, where X and Y are not underscores */
2473 return s[0] != '_' && /* X */
2474 s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */
2475 s[4] != '_'; /* Y */
2478 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
2479 * before last triple underscore. Struct name part after last triple
2480 * underscore is ignored by BPF CO-RE relocation during relocation matching.
2482 static size_t bpf_core_essential_name_len(const char *name)
2484 size_t n = strlen(name);
2487 for (i = n - 5; i >= 0; i--) {
2488 if (bpf_core_is_flavor_sep(name + i))
2494 /* dynamically sized list of type IDs */
2500 static void bpf_core_free_cands(struct ids_vec *cand_ids)
2502 free(cand_ids->data);
2506 static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
2507 __u32 local_type_id,
2508 const struct btf *targ_btf)
2510 size_t local_essent_len, targ_essent_len;
2511 const char *local_name, *targ_name;
2512 const struct btf_type *t;
2513 struct ids_vec *cand_ids;
2517 t = btf__type_by_id(local_btf, local_type_id);
2519 return ERR_PTR(-EINVAL);
2521 local_name = btf__name_by_offset(local_btf, t->name_off);
2522 if (str_is_empty(local_name))
2523 return ERR_PTR(-EINVAL);
2524 local_essent_len = bpf_core_essential_name_len(local_name);
2526 cand_ids = calloc(1, sizeof(*cand_ids));
2528 return ERR_PTR(-ENOMEM);
2530 n = btf__get_nr_types(targ_btf);
2531 for (i = 1; i <= n; i++) {
2532 t = btf__type_by_id(targ_btf, i);
2533 targ_name = btf__name_by_offset(targ_btf, t->name_off);
2534 if (str_is_empty(targ_name))
2537 targ_essent_len = bpf_core_essential_name_len(targ_name);
2538 if (targ_essent_len != local_essent_len)
2541 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
2542 pr_debug("[%d] %s: found candidate [%d] %s\n",
2543 local_type_id, local_name, i, targ_name);
2544 new_ids = realloc(cand_ids->data, cand_ids->len + 1);
2549 cand_ids->data = new_ids;
2550 cand_ids->data[cand_ids->len++] = i;
2555 bpf_core_free_cands(cand_ids);
2556 return ERR_PTR(err);
2559 /* Check two types for compatibility, skipping const/volatile/restrict and
2560 * typedefs, to ensure we are relocating offset to the compatible entities:
2561 * - any two STRUCTs/UNIONs are compatible and can be mixed;
2562 * - any two FWDs are compatible;
2563 * - any two PTRs are always compatible;
2564 * - for ENUMs, check sizes, names are ignored;
2565 * - for INT, size and bitness should match, signedness is ignored;
2566 * - for ARRAY, dimensionality is ignored, element types are checked for
2567 * compatibility recursively;
2568 * - everything else shouldn't be ever a target of relocation.
2569 * These rules are not set in stone and probably will be adjusted as we get
2570 * more experience with using BPF CO-RE relocations.
2572 static int bpf_core_fields_are_compat(const struct btf *local_btf,
2574 const struct btf *targ_btf,
2577 const struct btf_type *local_type, *targ_type;
2580 local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
2581 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2582 if (!local_type || !targ_type)
2585 if (btf_is_composite(local_type) && btf_is_composite(targ_type))
2587 if (btf_kind(local_type) != btf_kind(targ_type))
2590 switch (btf_kind(local_type)) {
2595 return local_type->size == targ_type->size;
2597 return btf_int_offset(local_type) == 0 &&
2598 btf_int_offset(targ_type) == 0 &&
2599 local_type->size == targ_type->size &&
2600 btf_int_bits(local_type) == btf_int_bits(targ_type);
2601 case BTF_KIND_ARRAY:
2602 local_id = btf_array(local_type)->type;
2603 targ_id = btf_array(targ_type)->type;
2606 pr_warning("unexpected kind %d relocated, local [%d], target [%d]\n",
2607 btf_kind(local_type), local_id, targ_id);
2613 * Given single high-level named field accessor in local type, find
2614 * corresponding high-level accessor for a target type. Along the way,
2615 * maintain low-level spec for target as well. Also keep updating target
2618 * Searching is performed through recursive exhaustive enumeration of all
2619 * fields of a struct/union. If there are any anonymous (embedded)
2620 * structs/unions, they are recursively searched as well. If field with
2621 * desired name is found, check compatibility between local and target types,
2622 * before returning result.
2624 * 1 is returned, if field is found.
2625 * 0 is returned if no compatible field is found.
2626 * <0 is returned on error.
2628 static int bpf_core_match_member(const struct btf *local_btf,
2629 const struct bpf_core_accessor *local_acc,
2630 const struct btf *targ_btf,
2632 struct bpf_core_spec *spec,
2633 __u32 *next_targ_id)
2635 const struct btf_type *local_type, *targ_type;
2636 const struct btf_member *local_member, *m;
2637 const char *local_name, *targ_name;
2641 targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2644 if (!btf_is_composite(targ_type))
2647 local_id = local_acc->type_id;
2648 local_type = btf__type_by_id(local_btf, local_id);
2649 local_member = btf_members(local_type) + local_acc->idx;
2650 local_name = btf__name_by_offset(local_btf, local_member->name_off);
2652 n = btf_vlen(targ_type);
2653 m = btf_members(targ_type);
2654 for (i = 0; i < n; i++, m++) {
2657 /* bitfield relocations not supported */
2658 if (btf_member_bitfield_size(targ_type, i))
2660 offset = btf_member_bit_offset(targ_type, i);
2664 /* too deep struct/union/array nesting */
2665 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2668 /* speculate this member will be the good one */
2669 spec->offset += offset / 8;
2670 spec->raw_spec[spec->raw_len++] = i;
2672 targ_name = btf__name_by_offset(targ_btf, m->name_off);
2673 if (str_is_empty(targ_name)) {
2674 /* embedded struct/union, we need to go deeper */
2675 found = bpf_core_match_member(local_btf, local_acc,
2677 spec, next_targ_id);
2678 if (found) /* either found or error */
2680 } else if (strcmp(local_name, targ_name) == 0) {
2681 /* matching named field */
2682 struct bpf_core_accessor *targ_acc;
2684 targ_acc = &spec->spec[spec->len++];
2685 targ_acc->type_id = targ_id;
2687 targ_acc->name = targ_name;
2689 *next_targ_id = m->type;
2690 found = bpf_core_fields_are_compat(local_btf,
2694 spec->len--; /* pop accessor */
2697 /* member turned out not to be what we looked for */
2698 spec->offset -= offset / 8;
2706 * Try to match local spec to a target type and, if successful, produce full
2707 * target spec (high-level, low-level + offset).
2709 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
2710 const struct btf *targ_btf, __u32 targ_id,
2711 struct bpf_core_spec *targ_spec)
2713 const struct btf_type *targ_type;
2714 const struct bpf_core_accessor *local_acc;
2715 struct bpf_core_accessor *targ_acc;
2718 memset(targ_spec, 0, sizeof(*targ_spec));
2719 targ_spec->btf = targ_btf;
2721 local_acc = &local_spec->spec[0];
2722 targ_acc = &targ_spec->spec[0];
2724 for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
2725 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
2730 if (local_acc->name) {
2731 matched = bpf_core_match_member(local_spec->btf,
2734 targ_spec, &targ_id);
2738 /* for i=0, targ_id is already treated as array element
2739 * type (because it's the original struct), for others
2740 * we should find array element type first
2743 const struct btf_array *a;
2745 if (!btf_is_array(targ_type))
2748 a = btf_array(targ_type);
2749 if (local_acc->idx >= a->nelems)
2751 if (!skip_mods_and_typedefs(targ_btf, a->type,
2756 /* too deep struct/union/array nesting */
2757 if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2760 targ_acc->type_id = targ_id;
2761 targ_acc->idx = local_acc->idx;
2762 targ_acc->name = NULL;
2764 targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
2765 targ_spec->raw_len++;
2767 sz = btf__resolve_size(targ_btf, targ_id);
2770 targ_spec->offset += local_acc->idx * sz;
2778 * Patch relocatable BPF instruction.
2779 * Expected insn->imm value is provided for validation, as well as the new
2782 * Currently three kinds of BPF instructions are supported:
2783 * 1. rX = <imm> (assignment with immediate operand);
2784 * 2. rX += <imm> (arithmetic operations with immediate operand);
2785 * 3. *(rX) = <imm> (indirect memory assignment with immediate operand).
2787 * If actual insn->imm value is wrong, bail out.
2789 static int bpf_core_reloc_insn(struct bpf_program *prog, int insn_off,
2790 __u32 orig_off, __u32 new_off)
2792 struct bpf_insn *insn;
2796 if (insn_off % sizeof(struct bpf_insn))
2798 insn_idx = insn_off / sizeof(struct bpf_insn);
2800 insn = &prog->insns[insn_idx];
2801 class = BPF_CLASS(insn->code);
2803 if (class == BPF_ALU || class == BPF_ALU64) {
2804 if (BPF_SRC(insn->code) != BPF_K)
2806 if (insn->imm != orig_off)
2808 insn->imm = new_off;
2809 pr_debug("prog '%s': patched insn #%d (ALU/ALU64) imm %d -> %d\n",
2810 bpf_program__title(prog, false),
2811 insn_idx, orig_off, new_off);
2813 pr_warning("prog '%s': trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
2814 bpf_program__title(prog, false),
2815 insn_idx, insn->code, insn->src_reg, insn->dst_reg,
2816 insn->off, insn->imm);
2822 static struct btf *btf_load_raw(const char *path)
2830 if (stat(path, &st))
2831 return ERR_PTR(-errno);
2833 data = malloc(st.st_size);
2835 return ERR_PTR(-ENOMEM);
2837 f = fopen(path, "rb");
2839 btf = ERR_PTR(-errno);
2843 read_cnt = fread(data, 1, st.st_size, f);
2845 if (read_cnt < st.st_size) {
2846 btf = ERR_PTR(-EBADF);
2850 btf = btf__new(data, read_cnt);
2858 * Probe few well-known locations for vmlinux kernel image and try to load BTF
2859 * data out of it to use for target BTF.
2861 static struct btf *bpf_core_find_kernel_btf(void)
2864 const char *path_fmt;
2867 /* try canonical vmlinux BTF through sysfs first */
2868 { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
2869 /* fall back to trying to find vmlinux ELF on disk otherwise */
2870 { "/boot/vmlinux-%1$s" },
2871 { "/lib/modules/%1$s/vmlinux-%1$s" },
2872 { "/lib/modules/%1$s/build/vmlinux" },
2873 { "/usr/lib/modules/%1$s/kernel/vmlinux" },
2874 { "/usr/lib/debug/boot/vmlinux-%1$s" },
2875 { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
2876 { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
2878 char path[PATH_MAX + 1];
2885 for (i = 0; i < ARRAY_SIZE(locations); i++) {
2886 snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
2888 if (access(path, R_OK))
2891 if (locations[i].raw_btf)
2892 btf = btf_load_raw(path);
2894 btf = btf__parse_elf(path, NULL);
2896 pr_debug("loading kernel BTF '%s': %ld\n",
2897 path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
2904 pr_warning("failed to find valid kernel BTF\n");
2905 return ERR_PTR(-ESRCH);
2908 /* Output spec definition in the format:
2909 * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
2910 * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
2912 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
2914 const struct btf_type *t;
2919 type_id = spec->spec[0].type_id;
2920 t = btf__type_by_id(spec->btf, type_id);
2921 s = btf__name_by_offset(spec->btf, t->name_off);
2922 libbpf_print(level, "[%u] %s + ", type_id, s);
2924 for (i = 0; i < spec->raw_len; i++)
2925 libbpf_print(level, "%d%s", spec->raw_spec[i],
2926 i == spec->raw_len - 1 ? " => " : ":");
2928 libbpf_print(level, "%u @ &x", spec->offset);
2930 for (i = 0; i < spec->len; i++) {
2931 if (spec->spec[i].name)
2932 libbpf_print(level, ".%s", spec->spec[i].name);
2934 libbpf_print(level, "[%u]", spec->spec[i].idx);
2939 static size_t bpf_core_hash_fn(const void *key, void *ctx)
2944 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
2949 static void *u32_as_hash_key(__u32 x)
2951 return (void *)(uintptr_t)x;
2955 * CO-RE relocate single instruction.
2957 * The outline and important points of the algorithm:
2958 * 1. For given local type, find corresponding candidate target types.
2959 * Candidate type is a type with the same "essential" name, ignoring
2960 * everything after last triple underscore (___). E.g., `sample`,
2961 * `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
2962 * for each other. Names with triple underscore are referred to as
2963 * "flavors" and are useful, among other things, to allow to
2964 * specify/support incompatible variations of the same kernel struct, which
2965 * might differ between different kernel versions and/or build
2968 * N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
2969 * converter, when deduplicated BTF of a kernel still contains more than
2970 * one different types with the same name. In that case, ___2, ___3, etc
2971 * are appended starting from second name conflict. But start flavors are
2972 * also useful to be defined "locally", in BPF program, to extract same
2973 * data from incompatible changes between different kernel
2974 * versions/configurations. For instance, to handle field renames between
2975 * kernel versions, one can use two flavors of the struct name with the
2976 * same common name and use conditional relocations to extract that field,
2977 * depending on target kernel version.
2978 * 2. For each candidate type, try to match local specification to this
2979 * candidate target type. Matching involves finding corresponding
2980 * high-level spec accessors, meaning that all named fields should match,
2981 * as well as all array accesses should be within the actual bounds. Also,
2982 * types should be compatible (see bpf_core_fields_are_compat for details).
2983 * 3. It is supported and expected that there might be multiple flavors
2984 * matching the spec. As long as all the specs resolve to the same set of
2985 * offsets across all candidates, there is not error. If there is any
2986 * ambiguity, CO-RE relocation will fail. This is necessary to accomodate
2987 * imprefection of BTF deduplication, which can cause slight duplication of
2988 * the same BTF type, if some directly or indirectly referenced (by
2989 * pointer) type gets resolved to different actual types in different
2990 * object files. If such situation occurs, deduplicated BTF will end up
2991 * with two (or more) structurally identical types, which differ only in
2992 * types they refer to through pointer. This should be OK in most cases and
2994 * 4. Candidate types search is performed by linearly scanning through all
2995 * types in target BTF. It is anticipated that this is overall more
2996 * efficient memory-wise and not significantly worse (if not better)
2997 * CPU-wise compared to prebuilding a map from all local type names to
2998 * a list of candidate type names. It's also sped up by caching resolved
2999 * list of matching candidates per each local "root" type ID, that has at
3000 * least one bpf_offset_reloc associated with it. This list is shared
3001 * between multiple relocations for the same type ID and is updated as some
3002 * of the candidates are pruned due to structural incompatibility.
3004 static int bpf_core_reloc_offset(struct bpf_program *prog,
3005 const struct bpf_offset_reloc *relo,
3007 const struct btf *local_btf,
3008 const struct btf *targ_btf,
3009 struct hashmap *cand_cache)
3011 const char *prog_name = bpf_program__title(prog, false);
3012 struct bpf_core_spec local_spec, cand_spec, targ_spec;
3013 const void *type_key = u32_as_hash_key(relo->type_id);
3014 const struct btf_type *local_type, *cand_type;
3015 const char *local_name, *cand_name;
3016 struct ids_vec *cand_ids;
3017 __u32 local_id, cand_id;
3018 const char *spec_str;
3021 local_id = relo->type_id;
3022 local_type = btf__type_by_id(local_btf, local_id);
3026 local_name = btf__name_by_offset(local_btf, local_type->name_off);
3027 if (str_is_empty(local_name))
3030 spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
3031 if (str_is_empty(spec_str))
3034 err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
3036 pr_warning("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
3037 prog_name, relo_idx, local_id, local_name, spec_str,
3042 pr_debug("prog '%s': relo #%d: spec is ", prog_name, relo_idx);
3043 bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
3044 libbpf_print(LIBBPF_DEBUG, "\n");
3046 if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
3047 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
3048 if (IS_ERR(cand_ids)) {
3049 pr_warning("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
3050 prog_name, relo_idx, local_id, local_name,
3052 return PTR_ERR(cand_ids);
3054 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
3056 bpf_core_free_cands(cand_ids);
3061 for (i = 0, j = 0; i < cand_ids->len; i++) {
3062 cand_id = cand_ids->data[i];
3063 cand_type = btf__type_by_id(targ_btf, cand_id);
3064 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
3066 err = bpf_core_spec_match(&local_spec, targ_btf,
3067 cand_id, &cand_spec);
3068 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
3069 prog_name, relo_idx, i, cand_name);
3070 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
3071 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
3073 pr_warning("prog '%s': relo #%d: matching error: %d\n",
3074 prog_name, relo_idx, err);
3081 targ_spec = cand_spec;
3082 } else if (cand_spec.offset != targ_spec.offset) {
3083 /* if there are many candidates, they should all
3084 * resolve to the same offset
3086 pr_warning("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
3087 prog_name, relo_idx, cand_spec.offset,
3092 cand_ids->data[j++] = cand_spec.spec[0].type_id;
3096 if (cand_ids->len == 0) {
3097 pr_warning("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
3098 prog_name, relo_idx, local_id, local_name, spec_str);
3102 err = bpf_core_reloc_insn(prog, relo->insn_off,
3103 local_spec.offset, targ_spec.offset);
3105 pr_warning("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
3106 prog_name, relo_idx, relo->insn_off, err);
3114 bpf_core_reloc_offsets(struct bpf_object *obj, const char *targ_btf_path)
3116 const struct btf_ext_info_sec *sec;
3117 const struct bpf_offset_reloc *rec;
3118 const struct btf_ext_info *seg;
3119 struct hashmap_entry *entry;
3120 struct hashmap *cand_cache = NULL;
3121 struct bpf_program *prog;
3122 struct btf *targ_btf;
3123 const char *sec_name;
3127 targ_btf = btf__parse_elf(targ_btf_path, NULL);
3129 targ_btf = bpf_core_find_kernel_btf();
3130 if (IS_ERR(targ_btf)) {
3131 pr_warning("failed to get target BTF: %ld\n",
3133 return PTR_ERR(targ_btf);
3136 cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
3137 if (IS_ERR(cand_cache)) {
3138 err = PTR_ERR(cand_cache);
3142 seg = &obj->btf_ext->offset_reloc_info;
3143 for_each_btf_ext_sec(seg, sec) {
3144 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
3145 if (str_is_empty(sec_name)) {
3149 prog = bpf_object__find_program_by_title(obj, sec_name);
3151 pr_warning("failed to find program '%s' for CO-RE offset relocation\n",
3157 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
3158 sec_name, sec->num_info);
3160 for_each_btf_ext_rec(seg, sec, i, rec) {
3161 err = bpf_core_reloc_offset(prog, rec, i, obj->btf,
3162 targ_btf, cand_cache);
3164 pr_warning("prog '%s': relo #%d: failed to relocate: %d\n",
3172 btf__free(targ_btf);
3173 if (!IS_ERR_OR_NULL(cand_cache)) {
3174 hashmap__for_each_entry(cand_cache, entry, i) {
3175 bpf_core_free_cands(entry->value);
3177 hashmap__free(cand_cache);
3183 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
3187 if (obj->btf_ext->offset_reloc_info.len)
3188 err = bpf_core_reloc_offsets(obj, targ_btf_path);
3194 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
3195 struct reloc_desc *relo)
3197 struct bpf_insn *insn, *new_insn;
3198 struct bpf_program *text;
3202 if (relo->type != RELO_CALL)
3203 return -LIBBPF_ERRNO__RELOC;
3205 if (prog->idx == obj->efile.text_shndx) {
3206 pr_warning("relo in .text insn %d into off %d\n",
3207 relo->insn_idx, relo->text_off);
3208 return -LIBBPF_ERRNO__RELOC;
3211 if (prog->main_prog_cnt == 0) {
3212 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
3214 pr_warning("no .text section found yet relo into text exist\n");
3215 return -LIBBPF_ERRNO__RELOC;
3217 new_cnt = prog->insns_cnt + text->insns_cnt;
3218 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
3220 pr_warning("oom in prog realloc\n");
3225 err = bpf_program_reloc_btf_ext(prog, obj,
3232 memcpy(new_insn + prog->insns_cnt, text->insns,
3233 text->insns_cnt * sizeof(*insn));
3234 prog->insns = new_insn;
3235 prog->main_prog_cnt = prog->insns_cnt;
3236 prog->insns_cnt = new_cnt;
3237 pr_debug("added %zd insn from %s to prog %s\n",
3238 text->insns_cnt, text->section_name,
3239 prog->section_name);
3241 insn = &prog->insns[relo->insn_idx];
3242 insn->imm += prog->main_prog_cnt - relo->insn_idx;
3247 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
3255 err = bpf_program_reloc_btf_ext(prog, obj,
3256 prog->section_name, 0);
3261 if (!prog->reloc_desc)
3264 for (i = 0; i < prog->nr_reloc; i++) {
3265 if (prog->reloc_desc[i].type == RELO_LD64 ||
3266 prog->reloc_desc[i].type == RELO_DATA) {
3267 bool relo_data = prog->reloc_desc[i].type == RELO_DATA;
3268 struct bpf_insn *insns = prog->insns;
3269 int insn_idx, map_idx;
3271 insn_idx = prog->reloc_desc[i].insn_idx;
3272 map_idx = prog->reloc_desc[i].map_idx;
3274 if (insn_idx + 1 >= (int)prog->insns_cnt) {
3275 pr_warning("relocation out of range: '%s'\n",
3276 prog->section_name);
3277 return -LIBBPF_ERRNO__RELOC;
3281 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
3283 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_VALUE;
3284 insns[insn_idx + 1].imm = insns[insn_idx].imm;
3286 insns[insn_idx].imm = obj->maps[map_idx].fd;
3287 } else if (prog->reloc_desc[i].type == RELO_CALL) {
3288 err = bpf_program__reloc_text(prog, obj,
3289 &prog->reloc_desc[i]);
3295 zfree(&prog->reloc_desc);
3301 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
3303 struct bpf_program *prog;
3308 err = bpf_object__relocate_core(obj, targ_btf_path);
3310 pr_warning("failed to perform CO-RE relocations: %d\n",
3315 for (i = 0; i < obj->nr_programs; i++) {
3316 prog = &obj->programs[i];
3318 err = bpf_program__relocate(prog, obj);
3320 pr_warning("failed to relocate '%s'\n",
3321 prog->section_name);
3328 static int bpf_object__collect_reloc(struct bpf_object *obj)
3332 if (!obj_elf_valid(obj)) {
3333 pr_warning("Internal error: elf object is closed\n");
3334 return -LIBBPF_ERRNO__INTERNAL;
3337 for (i = 0; i < obj->efile.nr_reloc; i++) {
3338 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
3339 Elf_Data *data = obj->efile.reloc[i].data;
3340 int idx = shdr->sh_info;
3341 struct bpf_program *prog;
3343 if (shdr->sh_type != SHT_REL) {
3344 pr_warning("internal error at %d\n", __LINE__);
3345 return -LIBBPF_ERRNO__INTERNAL;
3348 prog = bpf_object__find_prog_by_idx(obj, idx);
3350 pr_warning("relocation failed: no section(%d)\n", idx);
3351 return -LIBBPF_ERRNO__RELOC;
3354 err = bpf_program__collect_reloc(prog, shdr, data, obj);
3362 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
3363 char *license, __u32 kern_version, int *pfd)
3365 struct bpf_load_program_attr load_attr;
3366 char *cp, errmsg[STRERR_BUFSIZE];
3367 int log_buf_size = BPF_LOG_BUF_SIZE;
3371 if (!insns || !insns_cnt)
3374 memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
3375 load_attr.prog_type = prog->type;
3376 load_attr.expected_attach_type = prog->expected_attach_type;
3377 if (prog->caps->name)
3378 load_attr.name = prog->name;
3379 load_attr.insns = insns;
3380 load_attr.insns_cnt = insns_cnt;
3381 load_attr.license = license;
3382 load_attr.kern_version = kern_version;
3383 load_attr.prog_ifindex = prog->prog_ifindex;
3384 load_attr.prog_btf_fd = prog->btf_fd >= 0 ? prog->btf_fd : 0;
3385 load_attr.func_info = prog->func_info;
3386 load_attr.func_info_rec_size = prog->func_info_rec_size;
3387 load_attr.func_info_cnt = prog->func_info_cnt;
3388 load_attr.line_info = prog->line_info;
3389 load_attr.line_info_rec_size = prog->line_info_rec_size;
3390 load_attr.line_info_cnt = prog->line_info_cnt;
3391 load_attr.log_level = prog->log_level;
3392 load_attr.prog_flags = prog->prog_flags;
3395 log_buf = malloc(log_buf_size);
3397 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
3399 ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
3402 if (load_attr.log_level)
3403 pr_debug("verifier log:\n%s", log_buf);
3409 if (errno == ENOSPC) {
3414 ret = -LIBBPF_ERRNO__LOAD;
3415 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3416 pr_warning("load bpf program failed: %s\n", cp);
3418 if (log_buf && log_buf[0] != '\0') {
3419 ret = -LIBBPF_ERRNO__VERIFY;
3420 pr_warning("-- BEGIN DUMP LOG ---\n");
3421 pr_warning("\n%s\n", log_buf);
3422 pr_warning("-- END LOG --\n");
3423 } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
3424 pr_warning("Program too large (%zu insns), at most %d insns\n",
3425 load_attr.insns_cnt, BPF_MAXINSNS);
3426 ret = -LIBBPF_ERRNO__PROG2BIG;
3428 /* Wrong program type? */
3429 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
3432 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
3433 load_attr.expected_attach_type = 0;
3434 fd = bpf_load_program_xattr(&load_attr, NULL, 0);
3437 ret = -LIBBPF_ERRNO__PROGTYPE;
3443 ret = -LIBBPF_ERRNO__KVER;
3452 bpf_program__load(struct bpf_program *prog,
3453 char *license, __u32 kern_version)
3457 if (prog->instances.nr < 0 || !prog->instances.fds) {
3458 if (prog->preprocessor) {
3459 pr_warning("Internal error: can't load program '%s'\n",
3460 prog->section_name);
3461 return -LIBBPF_ERRNO__INTERNAL;
3464 prog->instances.fds = malloc(sizeof(int));
3465 if (!prog->instances.fds) {
3466 pr_warning("Not enough memory for BPF fds\n");
3469 prog->instances.nr = 1;
3470 prog->instances.fds[0] = -1;
3473 if (!prog->preprocessor) {
3474 if (prog->instances.nr != 1) {
3475 pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
3476 prog->section_name, prog->instances.nr);
3478 err = load_program(prog, prog->insns, prog->insns_cnt,
3479 license, kern_version, &fd);
3481 prog->instances.fds[0] = fd;
3485 for (i = 0; i < prog->instances.nr; i++) {
3486 struct bpf_prog_prep_result result;
3487 bpf_program_prep_t preprocessor = prog->preprocessor;
3489 memset(&result, 0, sizeof(result));
3490 err = preprocessor(prog, i, prog->insns,
3491 prog->insns_cnt, &result);
3493 pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
3494 i, prog->section_name);
3498 if (!result.new_insn_ptr || !result.new_insn_cnt) {
3499 pr_debug("Skip loading the %dth instance of program '%s'\n",
3500 i, prog->section_name);
3501 prog->instances.fds[i] = -1;
3507 err = load_program(prog, result.new_insn_ptr,
3508 result.new_insn_cnt,
3509 license, kern_version, &fd);
3512 pr_warning("Loading the %dth instance of program '%s' failed\n",
3513 i, prog->section_name);
3519 prog->instances.fds[i] = fd;
3523 pr_warning("failed to load program '%s'\n",
3524 prog->section_name);
3525 zfree(&prog->insns);
3526 prog->insns_cnt = 0;
3530 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
3531 const struct bpf_object *obj)
3533 return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
3537 bpf_object__load_progs(struct bpf_object *obj, int log_level)
3542 for (i = 0; i < obj->nr_programs; i++) {
3543 if (bpf_program__is_function_storage(&obj->programs[i], obj))
3545 obj->programs[i].log_level |= log_level;
3546 err = bpf_program__load(&obj->programs[i],
3555 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
3558 case BPF_PROG_TYPE_SOCKET_FILTER:
3559 case BPF_PROG_TYPE_SCHED_CLS:
3560 case BPF_PROG_TYPE_SCHED_ACT:
3561 case BPF_PROG_TYPE_XDP:
3562 case BPF_PROG_TYPE_CGROUP_SKB:
3563 case BPF_PROG_TYPE_CGROUP_SOCK:
3564 case BPF_PROG_TYPE_LWT_IN:
3565 case BPF_PROG_TYPE_LWT_OUT:
3566 case BPF_PROG_TYPE_LWT_XMIT:
3567 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
3568 case BPF_PROG_TYPE_SOCK_OPS:
3569 case BPF_PROG_TYPE_SK_SKB:
3570 case BPF_PROG_TYPE_CGROUP_DEVICE:
3571 case BPF_PROG_TYPE_SK_MSG:
3572 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3573 case BPF_PROG_TYPE_LIRC_MODE2:
3574 case BPF_PROG_TYPE_SK_REUSEPORT:
3575 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3576 case BPF_PROG_TYPE_UNSPEC:
3577 case BPF_PROG_TYPE_TRACEPOINT:
3578 case BPF_PROG_TYPE_RAW_TRACEPOINT:
3579 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3580 case BPF_PROG_TYPE_PERF_EVENT:
3581 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3582 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3584 case BPF_PROG_TYPE_KPROBE:
3590 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
3592 if (needs_kver && obj->kern_version == 0) {
3593 pr_warning("%s doesn't provide kernel version\n",
3595 return -LIBBPF_ERRNO__KVERSION;
3600 static struct bpf_object *
3601 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
3602 bool needs_kver, int flags)
3604 struct bpf_object *obj;
3607 if (elf_version(EV_CURRENT) == EV_NONE) {
3608 pr_warning("failed to init libelf for %s\n", path);
3609 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
3612 obj = bpf_object__new(path, obj_buf, obj_buf_sz);
3616 CHECK_ERR(bpf_object__elf_init(obj), err, out);
3617 CHECK_ERR(bpf_object__check_endianness(obj), err, out);
3618 CHECK_ERR(bpf_object__probe_caps(obj), err, out);
3619 CHECK_ERR(bpf_object__elf_collect(obj, flags), err, out);
3620 CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
3621 CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
3623 bpf_object__elf_finish(obj);
3626 bpf_object__close(obj);
3627 return ERR_PTR(err);
3630 struct bpf_object *__bpf_object__open_xattr(struct bpf_object_open_attr *attr,
3633 /* param validation */
3637 pr_debug("loading %s\n", attr->file);
3639 return __bpf_object__open(attr->file, NULL, 0,
3640 bpf_prog_type__needs_kver(attr->prog_type),
3644 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
3646 return __bpf_object__open_xattr(attr, 0);
3649 struct bpf_object *bpf_object__open(const char *path)
3651 struct bpf_object_open_attr attr = {
3653 .prog_type = BPF_PROG_TYPE_UNSPEC,
3656 return bpf_object__open_xattr(&attr);
3659 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
3665 /* param validation */
3666 if (!obj_buf || obj_buf_sz <= 0)
3670 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
3671 (unsigned long)obj_buf,
3672 (unsigned long)obj_buf_sz);
3675 pr_debug("loading object '%s' from buffer\n", name);
3677 return __bpf_object__open(name, obj_buf, obj_buf_sz, true, true);
3680 int bpf_object__unload(struct bpf_object *obj)
3687 for (i = 0; i < obj->nr_maps; i++)
3688 zclose(obj->maps[i].fd);
3690 for (i = 0; i < obj->nr_programs; i++)
3691 bpf_program__unload(&obj->programs[i]);
3696 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
3698 struct bpf_object *obj;
3708 pr_warning("object should not be loaded twice\n");
3714 CHECK_ERR(bpf_object__create_maps(obj), err, out);
3715 CHECK_ERR(bpf_object__relocate(obj, attr->target_btf_path), err, out);
3716 CHECK_ERR(bpf_object__load_progs(obj, attr->log_level), err, out);
3720 bpf_object__unload(obj);
3721 pr_warning("failed to load object '%s'\n", obj->path);
3725 int bpf_object__load(struct bpf_object *obj)
3727 struct bpf_object_load_attr attr = {
3731 return bpf_object__load_xattr(&attr);
3734 static int check_path(const char *path)
3736 char *cp, errmsg[STRERR_BUFSIZE];
3737 struct statfs st_fs;
3744 dname = strdup(path);
3748 dir = dirname(dname);
3749 if (statfs(dir, &st_fs)) {
3750 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3751 pr_warning("failed to statfs %s: %s\n", dir, cp);
3756 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
3757 pr_warning("specified path %s is not on BPF FS\n", path);
3764 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
3767 char *cp, errmsg[STRERR_BUFSIZE];
3770 err = check_path(path);
3775 pr_warning("invalid program pointer\n");
3779 if (instance < 0 || instance >= prog->instances.nr) {
3780 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
3781 instance, prog->section_name, prog->instances.nr);
3785 if (bpf_obj_pin(prog->instances.fds[instance], path)) {
3786 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3787 pr_warning("failed to pin program: %s\n", cp);
3790 pr_debug("pinned program '%s'\n", path);
3795 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
3800 err = check_path(path);
3805 pr_warning("invalid program pointer\n");
3809 if (instance < 0 || instance >= prog->instances.nr) {
3810 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
3811 instance, prog->section_name, prog->instances.nr);
3818 pr_debug("unpinned program '%s'\n", path);
3823 static int make_dir(const char *path)
3825 char *cp, errmsg[STRERR_BUFSIZE];
3828 if (mkdir(path, 0700) && errno != EEXIST)
3832 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
3833 pr_warning("failed to mkdir %s: %s\n", path, cp);
3838 int bpf_program__pin(struct bpf_program *prog, const char *path)
3842 err = check_path(path);
3847 pr_warning("invalid program pointer\n");
3851 if (prog->instances.nr <= 0) {
3852 pr_warning("no instances of prog %s to pin\n",
3853 prog->section_name);
3857 if (prog->instances.nr == 1) {
3858 /* don't create subdirs when pinning single instance */
3859 return bpf_program__pin_instance(prog, path, 0);
3862 err = make_dir(path);
3866 for (i = 0; i < prog->instances.nr; i++) {
3870 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3874 } else if (len >= PATH_MAX) {
3875 err = -ENAMETOOLONG;
3879 err = bpf_program__pin_instance(prog, buf, i);
3887 for (i = i - 1; i >= 0; i--) {
3891 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3894 else if (len >= PATH_MAX)
3897 bpf_program__unpin_instance(prog, buf, i);
3905 int bpf_program__unpin(struct bpf_program *prog, const char *path)
3909 err = check_path(path);
3914 pr_warning("invalid program pointer\n");
3918 if (prog->instances.nr <= 0) {
3919 pr_warning("no instances of prog %s to pin\n",
3920 prog->section_name);
3924 if (prog->instances.nr == 1) {
3925 /* don't create subdirs when pinning single instance */
3926 return bpf_program__unpin_instance(prog, path, 0);
3929 for (i = 0; i < prog->instances.nr; i++) {
3933 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3936 else if (len >= PATH_MAX)
3937 return -ENAMETOOLONG;
3939 err = bpf_program__unpin_instance(prog, buf, i);
3951 int bpf_map__pin(struct bpf_map *map, const char *path)
3953 char *cp, errmsg[STRERR_BUFSIZE];
3956 err = check_path(path);
3961 pr_warning("invalid map pointer\n");
3965 if (bpf_obj_pin(map->fd, path)) {
3966 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3967 pr_warning("failed to pin map: %s\n", cp);
3971 pr_debug("pinned map '%s'\n", path);
3976 int bpf_map__unpin(struct bpf_map *map, const char *path)
3980 err = check_path(path);
3985 pr_warning("invalid map pointer\n");
3992 pr_debug("unpinned map '%s'\n", path);
3997 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
3999 struct bpf_map *map;
4006 pr_warning("object not yet loaded; load it first\n");
4010 err = make_dir(path);
4014 bpf_object__for_each_map(map, obj) {
4018 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4019 bpf_map__name(map));
4022 goto err_unpin_maps;
4023 } else if (len >= PATH_MAX) {
4024 err = -ENAMETOOLONG;
4025 goto err_unpin_maps;
4028 err = bpf_map__pin(map, buf);
4030 goto err_unpin_maps;
4036 while ((map = bpf_map__prev(map, obj))) {
4040 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4041 bpf_map__name(map));
4044 else if (len >= PATH_MAX)
4047 bpf_map__unpin(map, buf);
4053 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
4055 struct bpf_map *map;
4061 bpf_object__for_each_map(map, obj) {
4065 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4066 bpf_map__name(map));
4069 else if (len >= PATH_MAX)
4070 return -ENAMETOOLONG;
4072 err = bpf_map__unpin(map, buf);
4080 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
4082 struct bpf_program *prog;
4089 pr_warning("object not yet loaded; load it first\n");
4093 err = make_dir(path);
4097 bpf_object__for_each_program(prog, obj) {
4101 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4105 goto err_unpin_programs;
4106 } else if (len >= PATH_MAX) {
4107 err = -ENAMETOOLONG;
4108 goto err_unpin_programs;
4111 err = bpf_program__pin(prog, buf);
4113 goto err_unpin_programs;
4119 while ((prog = bpf_program__prev(prog, obj))) {
4123 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4127 else if (len >= PATH_MAX)
4130 bpf_program__unpin(prog, buf);
4136 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
4138 struct bpf_program *prog;
4144 bpf_object__for_each_program(prog, obj) {
4148 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4152 else if (len >= PATH_MAX)
4153 return -ENAMETOOLONG;
4155 err = bpf_program__unpin(prog, buf);
4163 int bpf_object__pin(struct bpf_object *obj, const char *path)
4167 err = bpf_object__pin_maps(obj, path);
4171 err = bpf_object__pin_programs(obj, path);
4173 bpf_object__unpin_maps(obj, path);
4180 void bpf_object__close(struct bpf_object *obj)
4187 if (obj->clear_priv)
4188 obj->clear_priv(obj, obj->priv);
4190 bpf_object__elf_finish(obj);
4191 bpf_object__unload(obj);
4192 btf__free(obj->btf);
4193 btf_ext__free(obj->btf_ext);
4195 for (i = 0; i < obj->nr_maps; i++) {
4196 zfree(&obj->maps[i].name);
4197 if (obj->maps[i].clear_priv)
4198 obj->maps[i].clear_priv(&obj->maps[i],
4200 obj->maps[i].priv = NULL;
4201 obj->maps[i].clear_priv = NULL;
4204 zfree(&obj->sections.rodata);
4205 zfree(&obj->sections.data);
4209 if (obj->programs && obj->nr_programs) {
4210 for (i = 0; i < obj->nr_programs; i++)
4211 bpf_program__exit(&obj->programs[i]);
4213 zfree(&obj->programs);
4215 list_del(&obj->list);
4220 bpf_object__next(struct bpf_object *prev)
4222 struct bpf_object *next;
4225 next = list_first_entry(&bpf_objects_list,
4229 next = list_next_entry(prev, list);
4231 /* Empty list is noticed here so don't need checking on entry. */
4232 if (&next->list == &bpf_objects_list)
4238 const char *bpf_object__name(const struct bpf_object *obj)
4240 return obj ? obj->path : ERR_PTR(-EINVAL);
4243 unsigned int bpf_object__kversion(const struct bpf_object *obj)
4245 return obj ? obj->kern_version : 0;
4248 struct btf *bpf_object__btf(const struct bpf_object *obj)
4250 return obj ? obj->btf : NULL;
4253 int bpf_object__btf_fd(const struct bpf_object *obj)
4255 return obj->btf ? btf__fd(obj->btf) : -1;
4258 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
4259 bpf_object_clear_priv_t clear_priv)
4261 if (obj->priv && obj->clear_priv)
4262 obj->clear_priv(obj, obj->priv);
4265 obj->clear_priv = clear_priv;
4269 void *bpf_object__priv(const struct bpf_object *obj)
4271 return obj ? obj->priv : ERR_PTR(-EINVAL);
4274 static struct bpf_program *
4275 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
4278 size_t nr_programs = obj->nr_programs;
4285 /* Iter from the beginning */
4286 return forward ? &obj->programs[0] :
4287 &obj->programs[nr_programs - 1];
4289 if (p->obj != obj) {
4290 pr_warning("error: program handler doesn't match object\n");
4294 idx = (p - obj->programs) + (forward ? 1 : -1);
4295 if (idx >= obj->nr_programs || idx < 0)
4297 return &obj->programs[idx];
4300 struct bpf_program *
4301 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
4303 struct bpf_program *prog = prev;
4306 prog = __bpf_program__iter(prog, obj, true);
4307 } while (prog && bpf_program__is_function_storage(prog, obj));
4312 struct bpf_program *
4313 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
4315 struct bpf_program *prog = next;
4318 prog = __bpf_program__iter(prog, obj, false);
4319 } while (prog && bpf_program__is_function_storage(prog, obj));
4324 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
4325 bpf_program_clear_priv_t clear_priv)
4327 if (prog->priv && prog->clear_priv)
4328 prog->clear_priv(prog, prog->priv);
4331 prog->clear_priv = clear_priv;
4335 void *bpf_program__priv(const struct bpf_program *prog)
4337 return prog ? prog->priv : ERR_PTR(-EINVAL);
4340 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
4342 prog->prog_ifindex = ifindex;
4345 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
4349 title = prog->section_name;
4351 title = strdup(title);
4353 pr_warning("failed to strdup program title\n");
4354 return ERR_PTR(-ENOMEM);
4361 int bpf_program__fd(const struct bpf_program *prog)
4363 return bpf_program__nth_fd(prog, 0);
4366 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
4367 bpf_program_prep_t prep)
4371 if (nr_instances <= 0 || !prep)
4374 if (prog->instances.nr > 0 || prog->instances.fds) {
4375 pr_warning("Can't set pre-processor after loading\n");
4379 instances_fds = malloc(sizeof(int) * nr_instances);
4380 if (!instances_fds) {
4381 pr_warning("alloc memory failed for fds\n");
4385 /* fill all fd with -1 */
4386 memset(instances_fds, -1, sizeof(int) * nr_instances);
4388 prog->instances.nr = nr_instances;
4389 prog->instances.fds = instances_fds;
4390 prog->preprocessor = prep;
4394 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
4401 if (n >= prog->instances.nr || n < 0) {
4402 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
4403 n, prog->section_name, prog->instances.nr);
4407 fd = prog->instances.fds[n];
4409 pr_warning("%dth instance of program '%s' is invalid\n",
4410 n, prog->section_name);
4417 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
4422 static bool bpf_program__is_type(const struct bpf_program *prog,
4423 enum bpf_prog_type type)
4425 return prog ? (prog->type == type) : false;
4428 #define BPF_PROG_TYPE_FNS(NAME, TYPE) \
4429 int bpf_program__set_##NAME(struct bpf_program *prog) \
4433 bpf_program__set_type(prog, TYPE); \
4437 bool bpf_program__is_##NAME(const struct bpf_program *prog) \
4439 return bpf_program__is_type(prog, TYPE); \
4442 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
4443 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
4444 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
4445 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
4446 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
4447 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
4448 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
4449 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
4451 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
4452 enum bpf_attach_type type)
4454 prog->expected_attach_type = type;
4457 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, atype) \
4458 { string, sizeof(string) - 1, ptype, eatype, is_attachable, atype }
4460 /* Programs that can NOT be attached. */
4461 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0)
4463 /* Programs that can be attached. */
4464 #define BPF_APROG_SEC(string, ptype, atype) \
4465 BPF_PROG_SEC_IMPL(string, ptype, 0, 1, atype)
4467 /* Programs that must specify expected attach type at load time. */
4468 #define BPF_EAPROG_SEC(string, ptype, eatype) \
4469 BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, eatype)
4471 /* Programs that can be attached but attach type can't be identified by section
4472 * name. Kept for backward compatibility.
4474 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
4476 static const struct {
4479 enum bpf_prog_type prog_type;
4480 enum bpf_attach_type expected_attach_type;
4482 enum bpf_attach_type attach_type;
4483 } section_names[] = {
4484 BPF_PROG_SEC("socket", BPF_PROG_TYPE_SOCKET_FILTER),
4485 BPF_PROG_SEC("kprobe/", BPF_PROG_TYPE_KPROBE),
4486 BPF_PROG_SEC("kretprobe/", BPF_PROG_TYPE_KPROBE),
4487 BPF_PROG_SEC("classifier", BPF_PROG_TYPE_SCHED_CLS),
4488 BPF_PROG_SEC("action", BPF_PROG_TYPE_SCHED_ACT),
4489 BPF_PROG_SEC("tracepoint/", BPF_PROG_TYPE_TRACEPOINT),
4490 BPF_PROG_SEC("raw_tracepoint/", BPF_PROG_TYPE_RAW_TRACEPOINT),
4491 BPF_PROG_SEC("xdp", BPF_PROG_TYPE_XDP),
4492 BPF_PROG_SEC("perf_event", BPF_PROG_TYPE_PERF_EVENT),
4493 BPF_PROG_SEC("lwt_in", BPF_PROG_TYPE_LWT_IN),
4494 BPF_PROG_SEC("lwt_out", BPF_PROG_TYPE_LWT_OUT),
4495 BPF_PROG_SEC("lwt_xmit", BPF_PROG_TYPE_LWT_XMIT),
4496 BPF_PROG_SEC("lwt_seg6local", BPF_PROG_TYPE_LWT_SEG6LOCAL),
4497 BPF_APROG_SEC("cgroup_skb/ingress", BPF_PROG_TYPE_CGROUP_SKB,
4498 BPF_CGROUP_INET_INGRESS),
4499 BPF_APROG_SEC("cgroup_skb/egress", BPF_PROG_TYPE_CGROUP_SKB,
4500 BPF_CGROUP_INET_EGRESS),
4501 BPF_APROG_COMPAT("cgroup/skb", BPF_PROG_TYPE_CGROUP_SKB),
4502 BPF_APROG_SEC("cgroup/sock", BPF_PROG_TYPE_CGROUP_SOCK,
4503 BPF_CGROUP_INET_SOCK_CREATE),
4504 BPF_EAPROG_SEC("cgroup/post_bind4", BPF_PROG_TYPE_CGROUP_SOCK,
4505 BPF_CGROUP_INET4_POST_BIND),
4506 BPF_EAPROG_SEC("cgroup/post_bind6", BPF_PROG_TYPE_CGROUP_SOCK,
4507 BPF_CGROUP_INET6_POST_BIND),
4508 BPF_APROG_SEC("cgroup/dev", BPF_PROG_TYPE_CGROUP_DEVICE,
4510 BPF_APROG_SEC("sockops", BPF_PROG_TYPE_SOCK_OPS,
4511 BPF_CGROUP_SOCK_OPS),
4512 BPF_APROG_SEC("sk_skb/stream_parser", BPF_PROG_TYPE_SK_SKB,
4513 BPF_SK_SKB_STREAM_PARSER),
4514 BPF_APROG_SEC("sk_skb/stream_verdict", BPF_PROG_TYPE_SK_SKB,
4515 BPF_SK_SKB_STREAM_VERDICT),
4516 BPF_APROG_COMPAT("sk_skb", BPF_PROG_TYPE_SK_SKB),
4517 BPF_APROG_SEC("sk_msg", BPF_PROG_TYPE_SK_MSG,
4518 BPF_SK_MSG_VERDICT),
4519 BPF_APROG_SEC("lirc_mode2", BPF_PROG_TYPE_LIRC_MODE2,
4521 BPF_APROG_SEC("flow_dissector", BPF_PROG_TYPE_FLOW_DISSECTOR,
4522 BPF_FLOW_DISSECTOR),
4523 BPF_EAPROG_SEC("cgroup/bind4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4524 BPF_CGROUP_INET4_BIND),
4525 BPF_EAPROG_SEC("cgroup/bind6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4526 BPF_CGROUP_INET6_BIND),
4527 BPF_EAPROG_SEC("cgroup/connect4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4528 BPF_CGROUP_INET4_CONNECT),
4529 BPF_EAPROG_SEC("cgroup/connect6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4530 BPF_CGROUP_INET6_CONNECT),
4531 BPF_EAPROG_SEC("cgroup/sendmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4532 BPF_CGROUP_UDP4_SENDMSG),
4533 BPF_EAPROG_SEC("cgroup/sendmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4534 BPF_CGROUP_UDP6_SENDMSG),
4535 BPF_EAPROG_SEC("cgroup/recvmsg4", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4536 BPF_CGROUP_UDP4_RECVMSG),
4537 BPF_EAPROG_SEC("cgroup/recvmsg6", BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4538 BPF_CGROUP_UDP6_RECVMSG),
4539 BPF_EAPROG_SEC("cgroup/sysctl", BPF_PROG_TYPE_CGROUP_SYSCTL,
4541 BPF_EAPROG_SEC("cgroup/getsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT,
4542 BPF_CGROUP_GETSOCKOPT),
4543 BPF_EAPROG_SEC("cgroup/setsockopt", BPF_PROG_TYPE_CGROUP_SOCKOPT,
4544 BPF_CGROUP_SETSOCKOPT),
4547 #undef BPF_PROG_SEC_IMPL
4549 #undef BPF_APROG_SEC
4550 #undef BPF_EAPROG_SEC
4551 #undef BPF_APROG_COMPAT
4553 #define MAX_TYPE_NAME_SIZE 32
4555 static char *libbpf_get_type_names(bool attach_type)
4557 int i, len = ARRAY_SIZE(section_names) * MAX_TYPE_NAME_SIZE;
4565 /* Forge string buf with all available names */
4566 for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4567 if (attach_type && !section_names[i].is_attachable)
4570 if (strlen(buf) + strlen(section_names[i].sec) + 2 > len) {
4575 strcat(buf, section_names[i].sec);
4581 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
4582 enum bpf_attach_type *expected_attach_type)
4590 for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4591 if (strncmp(name, section_names[i].sec, section_names[i].len))
4593 *prog_type = section_names[i].prog_type;
4594 *expected_attach_type = section_names[i].expected_attach_type;
4597 pr_warning("failed to guess program type based on ELF section name '%s'\n", name);
4598 type_names = libbpf_get_type_names(false);
4599 if (type_names != NULL) {
4600 pr_info("supported section(type) names are:%s\n", type_names);
4607 int libbpf_attach_type_by_name(const char *name,
4608 enum bpf_attach_type *attach_type)
4616 for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4617 if (strncmp(name, section_names[i].sec, section_names[i].len))
4619 if (!section_names[i].is_attachable)
4621 *attach_type = section_names[i].attach_type;
4624 pr_warning("failed to guess attach type based on ELF section name '%s'\n", name);
4625 type_names = libbpf_get_type_names(true);
4626 if (type_names != NULL) {
4627 pr_info("attachable section(type) names are:%s\n", type_names);
4635 bpf_program__identify_section(struct bpf_program *prog,
4636 enum bpf_prog_type *prog_type,
4637 enum bpf_attach_type *expected_attach_type)
4639 return libbpf_prog_type_by_name(prog->section_name, prog_type,
4640 expected_attach_type);
4643 int bpf_map__fd(const struct bpf_map *map)
4645 return map ? map->fd : -EINVAL;
4648 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
4650 return map ? &map->def : ERR_PTR(-EINVAL);
4653 const char *bpf_map__name(const struct bpf_map *map)
4655 return map ? map->name : NULL;
4658 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
4660 return map ? map->btf_key_type_id : 0;
4663 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
4665 return map ? map->btf_value_type_id : 0;
4668 int bpf_map__set_priv(struct bpf_map *map, void *priv,
4669 bpf_map_clear_priv_t clear_priv)
4675 if (map->clear_priv)
4676 map->clear_priv(map, map->priv);
4680 map->clear_priv = clear_priv;
4684 void *bpf_map__priv(const struct bpf_map *map)
4686 return map ? map->priv : ERR_PTR(-EINVAL);
4689 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
4691 return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
4694 bool bpf_map__is_internal(const struct bpf_map *map)
4696 return map->libbpf_type != LIBBPF_MAP_UNSPEC;
4699 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
4701 map->map_ifindex = ifindex;
4704 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
4706 if (!bpf_map_type__is_map_in_map(map->def.type)) {
4707 pr_warning("error: unsupported map type\n");
4710 if (map->inner_map_fd != -1) {
4711 pr_warning("error: inner_map_fd already specified\n");
4714 map->inner_map_fd = fd;
4718 static struct bpf_map *
4719 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
4722 struct bpf_map *s, *e;
4724 if (!obj || !obj->maps)
4728 e = obj->maps + obj->nr_maps;
4730 if ((m < s) || (m >= e)) {
4731 pr_warning("error in %s: map handler doesn't belong to object\n",
4736 idx = (m - obj->maps) + i;
4737 if (idx >= obj->nr_maps || idx < 0)
4739 return &obj->maps[idx];
4743 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
4748 return __bpf_map__iter(prev, obj, 1);
4752 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
4757 return obj->maps + obj->nr_maps - 1;
4760 return __bpf_map__iter(next, obj, -1);
4764 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
4766 struct bpf_map *pos;
4768 bpf_object__for_each_map(pos, obj) {
4769 if (pos->name && !strcmp(pos->name, name))
4776 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
4778 return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
4782 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
4784 return ERR_PTR(-ENOTSUP);
4787 long libbpf_get_error(const void *ptr)
4789 return PTR_ERR_OR_ZERO(ptr);
4792 int bpf_prog_load(const char *file, enum bpf_prog_type type,
4793 struct bpf_object **pobj, int *prog_fd)
4795 struct bpf_prog_load_attr attr;
4797 memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
4799 attr.prog_type = type;
4800 attr.expected_attach_type = 0;
4802 return bpf_prog_load_xattr(&attr, pobj, prog_fd);
4805 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
4806 struct bpf_object **pobj, int *prog_fd)
4808 struct bpf_object_open_attr open_attr = {};
4809 struct bpf_program *prog, *first_prog = NULL;
4810 enum bpf_attach_type expected_attach_type;
4811 enum bpf_prog_type prog_type;
4812 struct bpf_object *obj;
4813 struct bpf_map *map;
4821 open_attr.file = attr->file;
4822 open_attr.prog_type = attr->prog_type;
4824 obj = bpf_object__open_xattr(&open_attr);
4825 if (IS_ERR_OR_NULL(obj))
4828 bpf_object__for_each_program(prog, obj) {
4830 * If type is not specified, try to guess it based on
4833 prog_type = attr->prog_type;
4834 prog->prog_ifindex = attr->ifindex;
4835 expected_attach_type = attr->expected_attach_type;
4836 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
4837 err = bpf_program__identify_section(prog, &prog_type,
4838 &expected_attach_type);
4840 bpf_object__close(obj);
4845 bpf_program__set_type(prog, prog_type);
4846 bpf_program__set_expected_attach_type(prog,
4847 expected_attach_type);
4849 prog->log_level = attr->log_level;
4850 prog->prog_flags = attr->prog_flags;
4855 bpf_object__for_each_map(map, obj) {
4856 if (!bpf_map__is_offload_neutral(map))
4857 map->map_ifindex = attr->ifindex;
4861 pr_warning("object file doesn't contain bpf program\n");
4862 bpf_object__close(obj);
4866 err = bpf_object__load(obj);
4868 bpf_object__close(obj);
4873 *prog_fd = bpf_program__fd(first_prog);
4878 int (*destroy)(struct bpf_link *link);
4881 int bpf_link__destroy(struct bpf_link *link)
4888 err = link->destroy(link);
4894 struct bpf_link_fd {
4895 struct bpf_link link; /* has to be at the top of struct */
4896 int fd; /* hook FD */
4899 static int bpf_link__destroy_perf_event(struct bpf_link *link)
4901 struct bpf_link_fd *l = (void *)link;
4904 err = ioctl(l->fd, PERF_EVENT_IOC_DISABLE, 0);
4912 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
4915 char errmsg[STRERR_BUFSIZE];
4916 struct bpf_link_fd *link;
4920 pr_warning("program '%s': invalid perf event FD %d\n",
4921 bpf_program__title(prog, false), pfd);
4922 return ERR_PTR(-EINVAL);
4924 prog_fd = bpf_program__fd(prog);
4926 pr_warning("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
4927 bpf_program__title(prog, false));
4928 return ERR_PTR(-EINVAL);
4931 link = malloc(sizeof(*link));
4933 return ERR_PTR(-ENOMEM);
4934 link->link.destroy = &bpf_link__destroy_perf_event;
4937 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
4940 pr_warning("program '%s': failed to attach to pfd %d: %s\n",
4941 bpf_program__title(prog, false), pfd,
4942 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4943 return ERR_PTR(err);
4945 if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
4948 pr_warning("program '%s': failed to enable pfd %d: %s\n",
4949 bpf_program__title(prog, false), pfd,
4950 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4951 return ERR_PTR(err);
4953 return (struct bpf_link *)link;
4957 * this function is expected to parse integer in the range of [0, 2^31-1] from
4958 * given file using scanf format string fmt. If actual parsed value is
4959 * negative, the result might be indistinguishable from error
4961 static int parse_uint_from_file(const char *file, const char *fmt)
4963 char buf[STRERR_BUFSIZE];
4967 f = fopen(file, "r");
4970 pr_debug("failed to open '%s': %s\n", file,
4971 libbpf_strerror_r(err, buf, sizeof(buf)));
4974 err = fscanf(f, fmt, &ret);
4976 err = err == EOF ? -EIO : -errno;
4977 pr_debug("failed to parse '%s': %s\n", file,
4978 libbpf_strerror_r(err, buf, sizeof(buf)));
4986 static int determine_kprobe_perf_type(void)
4988 const char *file = "/sys/bus/event_source/devices/kprobe/type";
4990 return parse_uint_from_file(file, "%d\n");
4993 static int determine_uprobe_perf_type(void)
4995 const char *file = "/sys/bus/event_source/devices/uprobe/type";
4997 return parse_uint_from_file(file, "%d\n");
5000 static int determine_kprobe_retprobe_bit(void)
5002 const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
5004 return parse_uint_from_file(file, "config:%d\n");
5007 static int determine_uprobe_retprobe_bit(void)
5009 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
5011 return parse_uint_from_file(file, "config:%d\n");
5014 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
5015 uint64_t offset, int pid)
5017 struct perf_event_attr attr = {};
5018 char errmsg[STRERR_BUFSIZE];
5021 type = uprobe ? determine_uprobe_perf_type()
5022 : determine_kprobe_perf_type();
5024 pr_warning("failed to determine %s perf type: %s\n",
5025 uprobe ? "uprobe" : "kprobe",
5026 libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
5030 int bit = uprobe ? determine_uprobe_retprobe_bit()
5031 : determine_kprobe_retprobe_bit();
5034 pr_warning("failed to determine %s retprobe bit: %s\n",
5035 uprobe ? "uprobe" : "kprobe",
5036 libbpf_strerror_r(bit, errmsg,
5040 attr.config |= 1 << bit;
5042 attr.size = sizeof(attr);
5044 attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
5045 attr.config2 = offset; /* kprobe_addr or probe_offset */
5047 /* pid filter is meaningful only for uprobes */
5048 pfd = syscall(__NR_perf_event_open, &attr,
5049 pid < 0 ? -1 : pid /* pid */,
5050 pid == -1 ? 0 : -1 /* cpu */,
5051 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5054 pr_warning("%s perf_event_open() failed: %s\n",
5055 uprobe ? "uprobe" : "kprobe",
5056 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5062 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
5064 const char *func_name)
5066 char errmsg[STRERR_BUFSIZE];
5067 struct bpf_link *link;
5070 pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
5071 0 /* offset */, -1 /* pid */);
5073 pr_warning("program '%s': failed to create %s '%s' perf event: %s\n",
5074 bpf_program__title(prog, false),
5075 retprobe ? "kretprobe" : "kprobe", func_name,
5076 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5077 return ERR_PTR(pfd);
5079 link = bpf_program__attach_perf_event(prog, pfd);
5082 err = PTR_ERR(link);
5083 pr_warning("program '%s': failed to attach to %s '%s': %s\n",
5084 bpf_program__title(prog, false),
5085 retprobe ? "kretprobe" : "kprobe", func_name,
5086 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5092 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
5093 bool retprobe, pid_t pid,
5094 const char *binary_path,
5097 char errmsg[STRERR_BUFSIZE];
5098 struct bpf_link *link;
5101 pfd = perf_event_open_probe(true /* uprobe */, retprobe,
5102 binary_path, func_offset, pid);
5104 pr_warning("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
5105 bpf_program__title(prog, false),
5106 retprobe ? "uretprobe" : "uprobe",
5107 binary_path, func_offset,
5108 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5109 return ERR_PTR(pfd);
5111 link = bpf_program__attach_perf_event(prog, pfd);
5114 err = PTR_ERR(link);
5115 pr_warning("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
5116 bpf_program__title(prog, false),
5117 retprobe ? "uretprobe" : "uprobe",
5118 binary_path, func_offset,
5119 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5125 static int determine_tracepoint_id(const char *tp_category,
5126 const char *tp_name)
5128 char file[PATH_MAX];
5131 ret = snprintf(file, sizeof(file),
5132 "/sys/kernel/debug/tracing/events/%s/%s/id",
5133 tp_category, tp_name);
5136 if (ret >= sizeof(file)) {
5137 pr_debug("tracepoint %s/%s path is too long\n",
5138 tp_category, tp_name);
5141 return parse_uint_from_file(file, "%d\n");
5144 static int perf_event_open_tracepoint(const char *tp_category,
5145 const char *tp_name)
5147 struct perf_event_attr attr = {};
5148 char errmsg[STRERR_BUFSIZE];
5149 int tp_id, pfd, err;
5151 tp_id = determine_tracepoint_id(tp_category, tp_name);
5153 pr_warning("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
5154 tp_category, tp_name,
5155 libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
5159 attr.type = PERF_TYPE_TRACEPOINT;
5160 attr.size = sizeof(attr);
5161 attr.config = tp_id;
5163 pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
5164 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5167 pr_warning("tracepoint '%s/%s' perf_event_open() failed: %s\n",
5168 tp_category, tp_name,
5169 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5175 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
5176 const char *tp_category,
5177 const char *tp_name)
5179 char errmsg[STRERR_BUFSIZE];
5180 struct bpf_link *link;
5183 pfd = perf_event_open_tracepoint(tp_category, tp_name);
5185 pr_warning("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
5186 bpf_program__title(prog, false),
5187 tp_category, tp_name,
5188 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5189 return ERR_PTR(pfd);
5191 link = bpf_program__attach_perf_event(prog, pfd);
5194 err = PTR_ERR(link);
5195 pr_warning("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
5196 bpf_program__title(prog, false),
5197 tp_category, tp_name,
5198 libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5204 static int bpf_link__destroy_fd(struct bpf_link *link)
5206 struct bpf_link_fd *l = (void *)link;
5208 return close(l->fd);
5211 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
5212 const char *tp_name)
5214 char errmsg[STRERR_BUFSIZE];
5215 struct bpf_link_fd *link;
5218 prog_fd = bpf_program__fd(prog);
5220 pr_warning("program '%s': can't attach before loaded\n",
5221 bpf_program__title(prog, false));
5222 return ERR_PTR(-EINVAL);
5225 link = malloc(sizeof(*link));
5227 return ERR_PTR(-ENOMEM);
5228 link->link.destroy = &bpf_link__destroy_fd;
5230 pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
5234 pr_warning("program '%s': failed to attach to raw tracepoint '%s': %s\n",
5235 bpf_program__title(prog, false), tp_name,
5236 libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5237 return ERR_PTR(pfd);
5240 return (struct bpf_link *)link;
5243 enum bpf_perf_event_ret
5244 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
5245 void **copy_mem, size_t *copy_size,
5246 bpf_perf_event_print_t fn, void *private_data)
5248 struct perf_event_mmap_page *header = mmap_mem;
5249 __u64 data_head = ring_buffer_read_head(header);
5250 __u64 data_tail = header->data_tail;
5251 void *base = ((__u8 *)header) + page_size;
5252 int ret = LIBBPF_PERF_EVENT_CONT;
5253 struct perf_event_header *ehdr;
5256 while (data_head != data_tail) {
5257 ehdr = base + (data_tail & (mmap_size - 1));
5258 ehdr_size = ehdr->size;
5260 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
5261 void *copy_start = ehdr;
5262 size_t len_first = base + mmap_size - copy_start;
5263 size_t len_secnd = ehdr_size - len_first;
5265 if (*copy_size < ehdr_size) {
5267 *copy_mem = malloc(ehdr_size);
5270 ret = LIBBPF_PERF_EVENT_ERROR;
5273 *copy_size = ehdr_size;
5276 memcpy(*copy_mem, copy_start, len_first);
5277 memcpy(*copy_mem + len_first, base, len_secnd);
5281 ret = fn(ehdr, private_data);
5282 data_tail += ehdr_size;
5283 if (ret != LIBBPF_PERF_EVENT_CONT)
5287 ring_buffer_write_tail(header, data_tail);
5293 struct perf_buffer_params {
5294 struct perf_event_attr *attr;
5295 /* if event_cb is specified, it takes precendence */
5296 perf_buffer_event_fn event_cb;
5297 /* sample_cb and lost_cb are higher-level common-case callbacks */
5298 perf_buffer_sample_fn sample_cb;
5299 perf_buffer_lost_fn lost_cb;
5306 struct perf_cpu_buf {
5307 struct perf_buffer *pb;
5308 void *base; /* mmap()'ed memory */
5309 void *buf; /* for reconstructing segmented data */
5316 struct perf_buffer {
5317 perf_buffer_event_fn event_cb;
5318 perf_buffer_sample_fn sample_cb;
5319 perf_buffer_lost_fn lost_cb;
5320 void *ctx; /* passed into callbacks */
5324 struct perf_cpu_buf **cpu_bufs;
5325 struct epoll_event *events;
5327 int epoll_fd; /* perf event FD */
5328 int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
5331 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
5332 struct perf_cpu_buf *cpu_buf)
5336 if (cpu_buf->base &&
5337 munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
5338 pr_warning("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
5339 if (cpu_buf->fd >= 0) {
5340 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
5347 void perf_buffer__free(struct perf_buffer *pb)
5354 for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
5355 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
5357 bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
5358 perf_buffer__free_cpu_buf(pb, cpu_buf);
5362 if (pb->epoll_fd >= 0)
5363 close(pb->epoll_fd);
5368 static struct perf_cpu_buf *
5369 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
5370 int cpu, int map_key)
5372 struct perf_cpu_buf *cpu_buf;
5373 char msg[STRERR_BUFSIZE];
5376 cpu_buf = calloc(1, sizeof(*cpu_buf));
5378 return ERR_PTR(-ENOMEM);
5382 cpu_buf->map_key = map_key;
5384 cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
5385 -1, PERF_FLAG_FD_CLOEXEC);
5386 if (cpu_buf->fd < 0) {
5388 pr_warning("failed to open perf buffer event on cpu #%d: %s\n",
5389 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5393 cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
5394 PROT_READ | PROT_WRITE, MAP_SHARED,
5396 if (cpu_buf->base == MAP_FAILED) {
5397 cpu_buf->base = NULL;
5399 pr_warning("failed to mmap perf buffer on cpu #%d: %s\n",
5400 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5404 if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
5406 pr_warning("failed to enable perf buffer event on cpu #%d: %s\n",
5407 cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5414 perf_buffer__free_cpu_buf(pb, cpu_buf);
5415 return (struct perf_cpu_buf *)ERR_PTR(err);
5418 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
5419 struct perf_buffer_params *p);
5421 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
5422 const struct perf_buffer_opts *opts)
5424 struct perf_buffer_params p = {};
5425 struct perf_event_attr attr = { 0, };
5427 attr.config = PERF_COUNT_SW_BPF_OUTPUT,
5428 attr.type = PERF_TYPE_SOFTWARE;
5429 attr.sample_type = PERF_SAMPLE_RAW;
5430 attr.sample_period = 1;
5431 attr.wakeup_events = 1;
5434 p.sample_cb = opts ? opts->sample_cb : NULL;
5435 p.lost_cb = opts ? opts->lost_cb : NULL;
5436 p.ctx = opts ? opts->ctx : NULL;
5438 return __perf_buffer__new(map_fd, page_cnt, &p);
5441 struct perf_buffer *
5442 perf_buffer__new_raw(int map_fd, size_t page_cnt,
5443 const struct perf_buffer_raw_opts *opts)
5445 struct perf_buffer_params p = {};
5447 p.attr = opts->attr;
5448 p.event_cb = opts->event_cb;
5450 p.cpu_cnt = opts->cpu_cnt;
5451 p.cpus = opts->cpus;
5452 p.map_keys = opts->map_keys;
5454 return __perf_buffer__new(map_fd, page_cnt, &p);
5457 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
5458 struct perf_buffer_params *p)
5460 struct bpf_map_info map = {};
5461 char msg[STRERR_BUFSIZE];
5462 struct perf_buffer *pb;
5466 if (page_cnt & (page_cnt - 1)) {
5467 pr_warning("page count should be power of two, but is %zu\n",
5469 return ERR_PTR(-EINVAL);
5472 map_info_len = sizeof(map);
5473 err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
5476 pr_warning("failed to get map info for map FD %d: %s\n",
5477 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
5478 return ERR_PTR(err);
5481 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
5482 pr_warning("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
5484 return ERR_PTR(-EINVAL);
5487 pb = calloc(1, sizeof(*pb));
5489 return ERR_PTR(-ENOMEM);
5491 pb->event_cb = p->event_cb;
5492 pb->sample_cb = p->sample_cb;
5493 pb->lost_cb = p->lost_cb;
5496 pb->page_size = getpagesize();
5497 pb->mmap_size = pb->page_size * page_cnt;
5498 pb->map_fd = map_fd;
5500 pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
5501 if (pb->epoll_fd < 0) {
5503 pr_warning("failed to create epoll instance: %s\n",
5504 libbpf_strerror_r(err, msg, sizeof(msg)));
5508 if (p->cpu_cnt > 0) {
5509 pb->cpu_cnt = p->cpu_cnt;
5511 pb->cpu_cnt = libbpf_num_possible_cpus();
5512 if (pb->cpu_cnt < 0) {
5516 if (map.max_entries < pb->cpu_cnt)
5517 pb->cpu_cnt = map.max_entries;
5520 pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
5523 pr_warning("failed to allocate events: out of memory\n");
5526 pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
5527 if (!pb->cpu_bufs) {
5529 pr_warning("failed to allocate buffers: out of memory\n");
5533 for (i = 0; i < pb->cpu_cnt; i++) {
5534 struct perf_cpu_buf *cpu_buf;
5537 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
5538 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
5540 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
5541 if (IS_ERR(cpu_buf)) {
5542 err = PTR_ERR(cpu_buf);
5546 pb->cpu_bufs[i] = cpu_buf;
5548 err = bpf_map_update_elem(pb->map_fd, &map_key,
5552 pr_warning("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
5553 cpu, map_key, cpu_buf->fd,
5554 libbpf_strerror_r(err, msg, sizeof(msg)));
5558 pb->events[i].events = EPOLLIN;
5559 pb->events[i].data.ptr = cpu_buf;
5560 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
5561 &pb->events[i]) < 0) {
5563 pr_warning("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
5565 libbpf_strerror_r(err, msg, sizeof(msg)));
5574 perf_buffer__free(pb);
5575 return ERR_PTR(err);
5578 struct perf_sample_raw {
5579 struct perf_event_header header;
5584 struct perf_sample_lost {
5585 struct perf_event_header header;
5591 static enum bpf_perf_event_ret
5592 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
5594 struct perf_cpu_buf *cpu_buf = ctx;
5595 struct perf_buffer *pb = cpu_buf->pb;
5598 /* user wants full control over parsing perf event */
5600 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
5603 case PERF_RECORD_SAMPLE: {
5604 struct perf_sample_raw *s = data;
5607 pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
5610 case PERF_RECORD_LOST: {
5611 struct perf_sample_lost *s = data;
5614 pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
5618 pr_warning("unknown perf sample type %d\n", e->type);
5619 return LIBBPF_PERF_EVENT_ERROR;
5621 return LIBBPF_PERF_EVENT_CONT;
5624 static int perf_buffer__process_records(struct perf_buffer *pb,
5625 struct perf_cpu_buf *cpu_buf)
5627 enum bpf_perf_event_ret ret;
5629 ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
5630 pb->page_size, &cpu_buf->buf,
5632 perf_buffer__process_record, cpu_buf);
5633 if (ret != LIBBPF_PERF_EVENT_CONT)
5638 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
5642 cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
5643 for (i = 0; i < cnt; i++) {
5644 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
5646 err = perf_buffer__process_records(pb, cpu_buf);
5648 pr_warning("error while processing records: %d\n", err);
5652 return cnt < 0 ? -errno : cnt;
5655 struct bpf_prog_info_array_desc {
5656 int array_offset; /* e.g. offset of jited_prog_insns */
5657 int count_offset; /* e.g. offset of jited_prog_len */
5658 int size_offset; /* > 0: offset of rec size,
5659 * < 0: fix size of -size_offset
5663 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
5664 [BPF_PROG_INFO_JITED_INSNS] = {
5665 offsetof(struct bpf_prog_info, jited_prog_insns),
5666 offsetof(struct bpf_prog_info, jited_prog_len),
5669 [BPF_PROG_INFO_XLATED_INSNS] = {
5670 offsetof(struct bpf_prog_info, xlated_prog_insns),
5671 offsetof(struct bpf_prog_info, xlated_prog_len),
5674 [BPF_PROG_INFO_MAP_IDS] = {
5675 offsetof(struct bpf_prog_info, map_ids),
5676 offsetof(struct bpf_prog_info, nr_map_ids),
5677 -(int)sizeof(__u32),
5679 [BPF_PROG_INFO_JITED_KSYMS] = {
5680 offsetof(struct bpf_prog_info, jited_ksyms),
5681 offsetof(struct bpf_prog_info, nr_jited_ksyms),
5682 -(int)sizeof(__u64),
5684 [BPF_PROG_INFO_JITED_FUNC_LENS] = {
5685 offsetof(struct bpf_prog_info, jited_func_lens),
5686 offsetof(struct bpf_prog_info, nr_jited_func_lens),
5687 -(int)sizeof(__u32),
5689 [BPF_PROG_INFO_FUNC_INFO] = {
5690 offsetof(struct bpf_prog_info, func_info),
5691 offsetof(struct bpf_prog_info, nr_func_info),
5692 offsetof(struct bpf_prog_info, func_info_rec_size),
5694 [BPF_PROG_INFO_LINE_INFO] = {
5695 offsetof(struct bpf_prog_info, line_info),
5696 offsetof(struct bpf_prog_info, nr_line_info),
5697 offsetof(struct bpf_prog_info, line_info_rec_size),
5699 [BPF_PROG_INFO_JITED_LINE_INFO] = {
5700 offsetof(struct bpf_prog_info, jited_line_info),
5701 offsetof(struct bpf_prog_info, nr_jited_line_info),
5702 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
5704 [BPF_PROG_INFO_PROG_TAGS] = {
5705 offsetof(struct bpf_prog_info, prog_tags),
5706 offsetof(struct bpf_prog_info, nr_prog_tags),
5707 -(int)sizeof(__u8) * BPF_TAG_SIZE,
5712 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, int offset)
5714 __u32 *array = (__u32 *)info;
5717 return array[offset / sizeof(__u32)];
5718 return -(int)offset;
5721 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, int offset)
5723 __u64 *array = (__u64 *)info;
5726 return array[offset / sizeof(__u64)];
5727 return -(int)offset;
5730 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
5733 __u32 *array = (__u32 *)info;
5736 array[offset / sizeof(__u32)] = val;
5739 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
5742 __u64 *array = (__u64 *)info;
5745 array[offset / sizeof(__u64)] = val;
5748 struct bpf_prog_info_linear *
5749 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
5751 struct bpf_prog_info_linear *info_linear;
5752 struct bpf_prog_info info = {};
5753 __u32 info_len = sizeof(info);
5758 if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
5759 return ERR_PTR(-EINVAL);
5761 /* step 1: get array dimensions */
5762 err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
5764 pr_debug("can't get prog info: %s", strerror(errno));
5765 return ERR_PTR(-EFAULT);
5768 /* step 2: calculate total size of all arrays */
5769 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5770 bool include_array = (arrays & (1UL << i)) > 0;
5771 struct bpf_prog_info_array_desc *desc;
5774 desc = bpf_prog_info_array_desc + i;
5776 /* kernel is too old to support this field */
5777 if (info_len < desc->array_offset + sizeof(__u32) ||
5778 info_len < desc->count_offset + sizeof(__u32) ||
5779 (desc->size_offset > 0 && info_len < desc->size_offset))
5780 include_array = false;
5782 if (!include_array) {
5783 arrays &= ~(1UL << i); /* clear the bit */
5787 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5788 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5790 data_len += count * size;
5793 /* step 3: allocate continuous memory */
5794 data_len = roundup(data_len, sizeof(__u64));
5795 info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
5797 return ERR_PTR(-ENOMEM);
5799 /* step 4: fill data to info_linear->info */
5800 info_linear->arrays = arrays;
5801 memset(&info_linear->info, 0, sizeof(info));
5802 ptr = info_linear->data;
5804 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5805 struct bpf_prog_info_array_desc *desc;
5808 if ((arrays & (1UL << i)) == 0)
5811 desc = bpf_prog_info_array_desc + i;
5812 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5813 size = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5814 bpf_prog_info_set_offset_u32(&info_linear->info,
5815 desc->count_offset, count);
5816 bpf_prog_info_set_offset_u32(&info_linear->info,
5817 desc->size_offset, size);
5818 bpf_prog_info_set_offset_u64(&info_linear->info,
5821 ptr += count * size;
5824 /* step 5: call syscall again to get required arrays */
5825 err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
5827 pr_debug("can't get prog info: %s", strerror(errno));
5829 return ERR_PTR(-EFAULT);
5832 /* step 6: verify the data */
5833 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5834 struct bpf_prog_info_array_desc *desc;
5837 if ((arrays & (1UL << i)) == 0)
5840 desc = bpf_prog_info_array_desc + i;
5841 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5842 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
5843 desc->count_offset);
5845 pr_warning("%s: mismatch in element count\n", __func__);
5847 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5848 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
5851 pr_warning("%s: mismatch in rec size\n", __func__);
5854 /* step 7: update info_len and data_len */
5855 info_linear->info_len = sizeof(struct bpf_prog_info);
5856 info_linear->data_len = data_len;
5861 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
5865 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5866 struct bpf_prog_info_array_desc *desc;
5869 if ((info_linear->arrays & (1UL << i)) == 0)
5872 desc = bpf_prog_info_array_desc + i;
5873 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
5874 desc->array_offset);
5875 offs = addr - ptr_to_u64(info_linear->data);
5876 bpf_prog_info_set_offset_u64(&info_linear->info,
5877 desc->array_offset, offs);
5881 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
5885 for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5886 struct bpf_prog_info_array_desc *desc;
5889 if ((info_linear->arrays & (1UL << i)) == 0)
5892 desc = bpf_prog_info_array_desc + i;
5893 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
5894 desc->array_offset);
5895 addr = offs + ptr_to_u64(info_linear->data);
5896 bpf_prog_info_set_offset_u64(&info_linear->info,
5897 desc->array_offset, addr);
5901 int libbpf_num_possible_cpus(void)
5903 static const char *fcpu = "/sys/devices/system/cpu/possible";
5904 int len = 0, n = 0, il = 0, ir = 0;
5905 unsigned int start = 0, end = 0;
5914 fd = open(fcpu, O_RDONLY);
5917 pr_warning("Failed to open file %s: %s\n",
5918 fcpu, strerror(error));
5921 len = read(fd, buf, sizeof(buf));
5924 error = len ? errno : EINVAL;
5925 pr_warning("Failed to read # of possible cpus from %s: %s\n",
5926 fcpu, strerror(error));
5929 if (len == sizeof(buf)) {
5930 pr_warning("File %s size overflow\n", fcpu);
5935 for (ir = 0, cpus = 0; ir <= len; ir++) {
5936 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
5937 if (buf[ir] == ',' || buf[ir] == '\0') {
5939 n = sscanf(&buf[il], "%u-%u", &start, &end);
5941 pr_warning("Failed to get # CPUs from %s\n",
5944 } else if (n == 1) {
5947 cpus += end - start + 1;
5952 pr_warning("Invalid #CPUs %d from %s\n", cpus, fcpu);