1 // SPDX-License-Identifier: LGPL-2.1
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.
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public
13 * License as published by the Free Software Foundation;
14 * version 2.1 of the License (not later!)
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with this program; if not, see <http://www.gnu.org/licenses>
34 #include <asm/unistd.h>
35 #include <linux/err.h>
36 #include <linux/kernel.h>
37 #include <linux/bpf.h>
38 #include <linux/list.h>
39 #include <linux/limits.h>
41 #include <sys/types.h>
54 #define BPF_FS_MAGIC 0xcafe4a11
57 #define __printf(a, b) __attribute__((format(printf, a, b)))
60 static int __base_pr(const char *format, ...)
65 va_start(args, format);
66 err = vfprintf(stderr, format, args);
71 static __printf(1, 2) libbpf_print_fn_t __pr_warning = __base_pr;
72 static __printf(1, 2) libbpf_print_fn_t __pr_info = __base_pr;
73 static __printf(1, 2) libbpf_print_fn_t __pr_debug;
75 #define __pr(func, fmt, ...) \
78 (func)("libbpf: " fmt, ##__VA_ARGS__); \
81 #define pr_warning(fmt, ...) __pr(__pr_warning, fmt, ##__VA_ARGS__)
82 #define pr_info(fmt, ...) __pr(__pr_info, fmt, ##__VA_ARGS__)
83 #define pr_debug(fmt, ...) __pr(__pr_debug, fmt, ##__VA_ARGS__)
85 void libbpf_set_print(libbpf_print_fn_t warn,
86 libbpf_print_fn_t info,
87 libbpf_print_fn_t debug)
94 #define STRERR_BUFSIZE 128
96 #define ERRNO_OFFSET(e) ((e) - __LIBBPF_ERRNO__START)
97 #define ERRCODE_OFFSET(c) ERRNO_OFFSET(LIBBPF_ERRNO__##c)
98 #define NR_ERRNO (__LIBBPF_ERRNO__END - __LIBBPF_ERRNO__START)
100 static const char *libbpf_strerror_table[NR_ERRNO] = {
101 [ERRCODE_OFFSET(LIBELF)] = "Something wrong in libelf",
102 [ERRCODE_OFFSET(FORMAT)] = "BPF object format invalid",
103 [ERRCODE_OFFSET(KVERSION)] = "'version' section incorrect or lost",
104 [ERRCODE_OFFSET(ENDIAN)] = "Endian mismatch",
105 [ERRCODE_OFFSET(INTERNAL)] = "Internal error in libbpf",
106 [ERRCODE_OFFSET(RELOC)] = "Relocation failed",
107 [ERRCODE_OFFSET(VERIFY)] = "Kernel verifier blocks program loading",
108 [ERRCODE_OFFSET(PROG2BIG)] = "Program too big",
109 [ERRCODE_OFFSET(KVER)] = "Incorrect kernel version",
110 [ERRCODE_OFFSET(PROGTYPE)] = "Kernel doesn't support this program type",
111 [ERRCODE_OFFSET(WRNGPID)] = "Wrong pid in netlink message",
112 [ERRCODE_OFFSET(INVSEQ)] = "Invalid netlink sequence",
115 int libbpf_strerror(int err, char *buf, size_t size)
120 err = err > 0 ? err : -err;
122 if (err < __LIBBPF_ERRNO__START) {
125 ret = strerror_r(err, buf, size);
126 buf[size - 1] = '\0';
130 if (err < __LIBBPF_ERRNO__END) {
133 msg = libbpf_strerror_table[ERRNO_OFFSET(err)];
134 snprintf(buf, size, "%s", msg);
135 buf[size - 1] = '\0';
139 snprintf(buf, size, "Unknown libbpf error %d", err);
140 buf[size - 1] = '\0';
144 #define CHECK_ERR(action, err, out) do { \
151 /* Copied from tools/perf/util/util.h */
153 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
157 # define zclose(fd) ({ \
160 ___err = close((fd)); \
165 #ifdef HAVE_LIBELF_MMAP_SUPPORT
166 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
168 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
172 * bpf_prog should be a better name but it has been used in
176 /* Index in elf obj file, for relocation use. */
180 struct bpf_insn *insns;
181 size_t insns_cnt, main_prog_cnt;
182 enum bpf_prog_type type;
201 bpf_program_prep_t preprocessor;
203 struct bpf_object *obj;
205 bpf_program_clear_priv_t clear_priv;
212 struct bpf_map_def def;
214 bpf_map_clear_priv_t clear_priv;
217 static LIST_HEAD(bpf_objects_list);
223 struct bpf_program *programs;
225 struct bpf_map *maps;
231 * Information when doing elf related work. Only valid if fd
251 * All loaded bpf_object is linked in a list, which is
252 * hidden to caller. bpf_objects__<func> handlers deal with
255 struct list_head list;
258 bpf_object_clear_priv_t clear_priv;
262 #define obj_elf_valid(o) ((o)->efile.elf)
264 static void bpf_program__unload(struct bpf_program *prog)
272 * If the object is opened but the program was never loaded,
273 * it is possible that prog->instances.nr == -1.
275 if (prog->instances.nr > 0) {
276 for (i = 0; i < prog->instances.nr; i++)
277 zclose(prog->instances.fds[i]);
278 } else if (prog->instances.nr != -1) {
279 pr_warning("Internal error: instances.nr is %d\n",
283 prog->instances.nr = -1;
284 zfree(&prog->instances.fds);
287 static void bpf_program__exit(struct bpf_program *prog)
292 if (prog->clear_priv)
293 prog->clear_priv(prog, prog->priv);
296 prog->clear_priv = NULL;
298 bpf_program__unload(prog);
300 zfree(&prog->section_name);
302 zfree(&prog->reloc_desc);
310 bpf_program__init(void *data, size_t size, char *section_name, int idx,
311 struct bpf_program *prog)
313 if (size < sizeof(struct bpf_insn)) {
314 pr_warning("corrupted section '%s'\n", section_name);
318 bzero(prog, sizeof(*prog));
320 prog->section_name = strdup(section_name);
321 if (!prog->section_name) {
322 pr_warning("failed to alloc name for prog under section(%d) %s\n",
327 prog->insns = malloc(size);
329 pr_warning("failed to alloc insns for prog under section %s\n",
333 prog->insns_cnt = size / sizeof(struct bpf_insn);
334 memcpy(prog->insns, data,
335 prog->insns_cnt * sizeof(struct bpf_insn));
337 prog->instances.fds = NULL;
338 prog->instances.nr = -1;
339 prog->type = BPF_PROG_TYPE_KPROBE;
343 bpf_program__exit(prog);
348 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
349 char *section_name, int idx)
351 struct bpf_program prog, *progs;
354 err = bpf_program__init(data, size, section_name, idx, &prog);
358 progs = obj->programs;
359 nr_progs = obj->nr_programs;
361 progs = realloc(progs, sizeof(progs[0]) * (nr_progs + 1));
364 * In this case the original obj->programs
365 * is still valid, so don't need special treat for
366 * bpf_close_object().
368 pr_warning("failed to alloc a new program under section '%s'\n",
370 bpf_program__exit(&prog);
374 pr_debug("found program %s\n", prog.section_name);
375 obj->programs = progs;
376 obj->nr_programs = nr_progs + 1;
378 progs[nr_progs] = prog;
383 bpf_object__init_prog_names(struct bpf_object *obj)
385 Elf_Data *symbols = obj->efile.symbols;
386 struct bpf_program *prog;
389 for (pi = 0; pi < obj->nr_programs; pi++) {
390 const char *name = NULL;
392 prog = &obj->programs[pi];
393 if (prog->idx == obj->efile.text_shndx) {
398 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
402 if (!gelf_getsym(symbols, si, &sym))
404 if (sym.st_shndx != prog->idx)
406 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
409 name = elf_strptr(obj->efile.elf,
410 obj->efile.strtabidx,
413 pr_warning("failed to get sym name string for prog %s\n",
415 return -LIBBPF_ERRNO__LIBELF;
420 pr_warning("failed to find sym for prog %s\n",
425 prog->name = strdup(name);
427 pr_warning("failed to allocate memory for prog sym %s\n",
436 static struct bpf_object *bpf_object__new(const char *path,
440 struct bpf_object *obj;
442 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
444 pr_warning("alloc memory failed for %s\n", path);
445 return ERR_PTR(-ENOMEM);
448 strcpy(obj->path, path);
452 * Caller of this function should also calls
453 * bpf_object__elf_finish() after data collection to return
454 * obj_buf to user. If not, we should duplicate the buffer to
455 * avoid user freeing them before elf finish.
457 obj->efile.obj_buf = obj_buf;
458 obj->efile.obj_buf_sz = obj_buf_sz;
459 obj->efile.maps_shndx = -1;
463 INIT_LIST_HEAD(&obj->list);
464 list_add(&obj->list, &bpf_objects_list);
468 static void bpf_object__elf_finish(struct bpf_object *obj)
470 if (!obj_elf_valid(obj))
473 if (obj->efile.elf) {
474 elf_end(obj->efile.elf);
475 obj->efile.elf = NULL;
477 obj->efile.symbols = NULL;
479 zfree(&obj->efile.reloc);
480 obj->efile.nr_reloc = 0;
481 zclose(obj->efile.fd);
482 obj->efile.obj_buf = NULL;
483 obj->efile.obj_buf_sz = 0;
486 static int bpf_object__elf_init(struct bpf_object *obj)
491 if (obj_elf_valid(obj)) {
492 pr_warning("elf init: internal error\n");
493 return -LIBBPF_ERRNO__LIBELF;
496 if (obj->efile.obj_buf_sz > 0) {
498 * obj_buf should have been validated by
499 * bpf_object__open_buffer().
501 obj->efile.elf = elf_memory(obj->efile.obj_buf,
502 obj->efile.obj_buf_sz);
504 obj->efile.fd = open(obj->path, O_RDONLY);
505 if (obj->efile.fd < 0) {
506 pr_warning("failed to open %s: %s\n", obj->path,
511 obj->efile.elf = elf_begin(obj->efile.fd,
512 LIBBPF_ELF_C_READ_MMAP,
516 if (!obj->efile.elf) {
517 pr_warning("failed to open %s as ELF file\n",
519 err = -LIBBPF_ERRNO__LIBELF;
523 if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
524 pr_warning("failed to get EHDR from %s\n",
526 err = -LIBBPF_ERRNO__FORMAT;
529 ep = &obj->efile.ehdr;
531 /* Old LLVM set e_machine to EM_NONE */
532 if ((ep->e_type != ET_REL) || (ep->e_machine && (ep->e_machine != EM_BPF))) {
533 pr_warning("%s is not an eBPF object file\n",
535 err = -LIBBPF_ERRNO__FORMAT;
541 bpf_object__elf_finish(obj);
546 bpf_object__check_endianness(struct bpf_object *obj)
548 static unsigned int const endian = 1;
550 switch (obj->efile.ehdr.e_ident[EI_DATA]) {
552 /* We are big endian, BPF obj is little endian. */
553 if (*(unsigned char const *)&endian != 1)
558 /* We are little endian, BPF obj is big endian. */
559 if (*(unsigned char const *)&endian != 0)
563 return -LIBBPF_ERRNO__ENDIAN;
569 pr_warning("Error: endianness mismatch.\n");
570 return -LIBBPF_ERRNO__ENDIAN;
574 bpf_object__init_license(struct bpf_object *obj,
575 void *data, size_t size)
577 memcpy(obj->license, data,
578 min(size, sizeof(obj->license) - 1));
579 pr_debug("license of %s is %s\n", obj->path, obj->license);
584 bpf_object__init_kversion(struct bpf_object *obj,
585 void *data, size_t size)
589 if (size != sizeof(kver)) {
590 pr_warning("invalid kver section in %s\n", obj->path);
591 return -LIBBPF_ERRNO__FORMAT;
593 memcpy(&kver, data, sizeof(kver));
594 obj->kern_version = kver;
595 pr_debug("kernel version of %s is %x\n", obj->path,
600 static int compare_bpf_map(const void *_a, const void *_b)
602 const struct bpf_map *a = _a;
603 const struct bpf_map *b = _b;
605 return a->offset - b->offset;
609 bpf_object__init_maps(struct bpf_object *obj)
611 int i, map_idx, map_def_sz, nr_maps = 0;
614 Elf_Data *symbols = obj->efile.symbols;
616 if (obj->efile.maps_shndx < 0)
621 scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
623 data = elf_getdata(scn, NULL);
625 pr_warning("failed to get Elf_Data from map section %d\n",
626 obj->efile.maps_shndx);
631 * Count number of maps. Each map has a name.
632 * Array of maps is not supported: only the first element is
635 * TODO: Detect array of map and report error.
637 for (i = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
640 if (!gelf_getsym(symbols, i, &sym))
642 if (sym.st_shndx != obj->efile.maps_shndx)
647 /* Alloc obj->maps and fill nr_maps. */
648 pr_debug("maps in %s: %d maps in %zd bytes\n", obj->path,
649 nr_maps, data->d_size);
654 /* Assume equally sized map definitions */
655 map_def_sz = data->d_size / nr_maps;
656 if (!data->d_size || (data->d_size % nr_maps) != 0) {
657 pr_warning("unable to determine map definition size "
658 "section %s, %d maps in %zd bytes\n",
659 obj->path, nr_maps, data->d_size);
663 obj->maps = calloc(nr_maps, sizeof(obj->maps[0]));
665 pr_warning("alloc maps for object failed\n");
668 obj->nr_maps = nr_maps;
671 * fill all fd with -1 so won't close incorrect
672 * fd (fd=0 is stdin) when failure (zclose won't close
675 for (i = 0; i < nr_maps; i++)
676 obj->maps[i].fd = -1;
679 * Fill obj->maps using data in "maps" section.
681 for (i = 0, map_idx = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
683 const char *map_name;
684 struct bpf_map_def *def;
686 if (!gelf_getsym(symbols, i, &sym))
688 if (sym.st_shndx != obj->efile.maps_shndx)
691 map_name = elf_strptr(obj->efile.elf,
692 obj->efile.strtabidx,
694 obj->maps[map_idx].offset = sym.st_value;
695 if (sym.st_value + map_def_sz > data->d_size) {
696 pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
697 obj->path, map_name);
701 obj->maps[map_idx].name = strdup(map_name);
702 if (!obj->maps[map_idx].name) {
703 pr_warning("failed to alloc map name\n");
706 pr_debug("map %d is \"%s\"\n", map_idx,
707 obj->maps[map_idx].name);
708 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
710 * If the definition of the map in the object file fits in
711 * bpf_map_def, copy it. Any extra fields in our version
712 * of bpf_map_def will default to zero as a result of the
715 if (map_def_sz <= sizeof(struct bpf_map_def)) {
716 memcpy(&obj->maps[map_idx].def, def, map_def_sz);
719 * Here the map structure being read is bigger than what
720 * we expect, truncate if the excess bits are all zero.
721 * If they are not zero, reject this map as
725 for (b = ((char *)def) + sizeof(struct bpf_map_def);
726 b < ((char *)def) + map_def_sz; b++) {
728 pr_warning("maps section in %s: \"%s\" "
729 "has unrecognized, non-zero "
731 obj->path, map_name);
735 memcpy(&obj->maps[map_idx].def, def,
736 sizeof(struct bpf_map_def));
741 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]), compare_bpf_map);
745 static bool section_have_execinstr(struct bpf_object *obj, int idx)
750 scn = elf_getscn(obj->efile.elf, idx);
754 if (gelf_getshdr(scn, &sh) != &sh)
757 if (sh.sh_flags & SHF_EXECINSTR)
763 static int bpf_object__elf_collect(struct bpf_object *obj)
765 Elf *elf = obj->efile.elf;
766 GElf_Ehdr *ep = &obj->efile.ehdr;
768 int idx = 0, err = 0;
770 /* Elf is corrupted/truncated, avoid calling elf_strptr. */
771 if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
772 pr_warning("failed to get e_shstrndx from %s\n",
774 return -LIBBPF_ERRNO__FORMAT;
777 while ((scn = elf_nextscn(elf, scn)) != NULL) {
783 if (gelf_getshdr(scn, &sh) != &sh) {
784 pr_warning("failed to get section(%d) header from %s\n",
786 err = -LIBBPF_ERRNO__FORMAT;
790 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
792 pr_warning("failed to get section(%d) name from %s\n",
794 err = -LIBBPF_ERRNO__FORMAT;
798 data = elf_getdata(scn, 0);
800 pr_warning("failed to get section(%d) data from %s(%s)\n",
801 idx, name, obj->path);
802 err = -LIBBPF_ERRNO__FORMAT;
805 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
806 idx, name, (unsigned long)data->d_size,
807 (int)sh.sh_link, (unsigned long)sh.sh_flags,
810 if (strcmp(name, "license") == 0)
811 err = bpf_object__init_license(obj,
814 else if (strcmp(name, "version") == 0)
815 err = bpf_object__init_kversion(obj,
818 else if (strcmp(name, "maps") == 0)
819 obj->efile.maps_shndx = idx;
820 else if (sh.sh_type == SHT_SYMTAB) {
821 if (obj->efile.symbols) {
822 pr_warning("bpf: multiple SYMTAB in %s\n",
824 err = -LIBBPF_ERRNO__FORMAT;
826 obj->efile.symbols = data;
827 obj->efile.strtabidx = sh.sh_link;
829 } else if ((sh.sh_type == SHT_PROGBITS) &&
830 (sh.sh_flags & SHF_EXECINSTR) &&
831 (data->d_size > 0)) {
832 if (strcmp(name, ".text") == 0)
833 obj->efile.text_shndx = idx;
834 err = bpf_object__add_program(obj, data->d_buf,
835 data->d_size, name, idx);
837 char errmsg[STRERR_BUFSIZE];
839 strerror_r(-err, errmsg, sizeof(errmsg));
840 pr_warning("failed to alloc program %s (%s): %s",
841 name, obj->path, errmsg);
843 } else if (sh.sh_type == SHT_REL) {
844 void *reloc = obj->efile.reloc;
845 int nr_reloc = obj->efile.nr_reloc + 1;
846 int sec = sh.sh_info; /* points to other section */
848 /* Only do relo for section with exec instructions */
849 if (!section_have_execinstr(obj, sec)) {
850 pr_debug("skip relo %s(%d) for section(%d)\n",
855 reloc = realloc(reloc,
856 sizeof(*obj->efile.reloc) * nr_reloc);
858 pr_warning("realloc failed\n");
861 int n = nr_reloc - 1;
863 obj->efile.reloc = reloc;
864 obj->efile.nr_reloc = nr_reloc;
866 obj->efile.reloc[n].shdr = sh;
867 obj->efile.reloc[n].data = data;
870 pr_debug("skip section(%d) %s\n", idx, name);
876 if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
877 pr_warning("Corrupted ELF file: index of strtab invalid\n");
878 return LIBBPF_ERRNO__FORMAT;
880 if (obj->efile.maps_shndx >= 0) {
881 err = bpf_object__init_maps(obj);
885 err = bpf_object__init_prog_names(obj);
890 static struct bpf_program *
891 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
893 struct bpf_program *prog;
896 for (i = 0; i < obj->nr_programs; i++) {
897 prog = &obj->programs[i];
898 if (prog->idx == idx)
905 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
906 Elf_Data *data, struct bpf_object *obj)
908 Elf_Data *symbols = obj->efile.symbols;
909 int text_shndx = obj->efile.text_shndx;
910 int maps_shndx = obj->efile.maps_shndx;
911 struct bpf_map *maps = obj->maps;
912 size_t nr_maps = obj->nr_maps;
915 pr_debug("collecting relocating info for: '%s'\n",
917 nrels = shdr->sh_size / shdr->sh_entsize;
919 prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
920 if (!prog->reloc_desc) {
921 pr_warning("failed to alloc memory in relocation\n");
924 prog->nr_reloc = nrels;
926 for (i = 0; i < nrels; i++) {
929 unsigned int insn_idx;
930 struct bpf_insn *insns = prog->insns;
933 if (!gelf_getrel(data, i, &rel)) {
934 pr_warning("relocation: failed to get %d reloc\n", i);
935 return -LIBBPF_ERRNO__FORMAT;
938 if (!gelf_getsym(symbols,
939 GELF_R_SYM(rel.r_info),
941 pr_warning("relocation: symbol %"PRIx64" not found\n",
942 GELF_R_SYM(rel.r_info));
943 return -LIBBPF_ERRNO__FORMAT;
945 pr_debug("relo for %lld value %lld name %d\n",
946 (long long) (rel.r_info >> 32),
947 (long long) sym.st_value, sym.st_name);
949 if (sym.st_shndx != maps_shndx && sym.st_shndx != text_shndx) {
950 pr_warning("Program '%s' contains non-map related relo data pointing to section %u\n",
951 prog->section_name, sym.st_shndx);
952 return -LIBBPF_ERRNO__RELOC;
955 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
956 pr_debug("relocation: insn_idx=%u\n", insn_idx);
958 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
959 if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
960 pr_warning("incorrect bpf_call opcode\n");
961 return -LIBBPF_ERRNO__RELOC;
963 prog->reloc_desc[i].type = RELO_CALL;
964 prog->reloc_desc[i].insn_idx = insn_idx;
965 prog->reloc_desc[i].text_off = sym.st_value;
969 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
970 pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
971 insn_idx, insns[insn_idx].code);
972 return -LIBBPF_ERRNO__RELOC;
975 /* TODO: 'maps' is sorted. We can use bsearch to make it faster. */
976 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
977 if (maps[map_idx].offset == sym.st_value) {
978 pr_debug("relocation: find map %zd (%s) for insn %u\n",
979 map_idx, maps[map_idx].name, insn_idx);
984 if (map_idx >= nr_maps) {
985 pr_warning("bpf relocation: map_idx %d large than %d\n",
986 (int)map_idx, (int)nr_maps - 1);
987 return -LIBBPF_ERRNO__RELOC;
990 prog->reloc_desc[i].type = RELO_LD64;
991 prog->reloc_desc[i].insn_idx = insn_idx;
992 prog->reloc_desc[i].map_idx = map_idx;
998 bpf_object__create_maps(struct bpf_object *obj)
1002 for (i = 0; i < obj->nr_maps; i++) {
1003 struct bpf_map_def *def = &obj->maps[i].def;
1004 int *pfd = &obj->maps[i].fd;
1006 *pfd = bpf_create_map_name(def->type,
1016 pr_warning("failed to create map (name: '%s'): %s\n",
1019 for (j = 0; j < i; j++)
1020 zclose(obj->maps[j].fd);
1023 pr_debug("create map %s: fd=%d\n", obj->maps[i].name, *pfd);
1030 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
1031 struct reloc_desc *relo)
1033 struct bpf_insn *insn, *new_insn;
1034 struct bpf_program *text;
1037 if (relo->type != RELO_CALL)
1038 return -LIBBPF_ERRNO__RELOC;
1040 if (prog->idx == obj->efile.text_shndx) {
1041 pr_warning("relo in .text insn %d into off %d\n",
1042 relo->insn_idx, relo->text_off);
1043 return -LIBBPF_ERRNO__RELOC;
1046 if (prog->main_prog_cnt == 0) {
1047 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
1049 pr_warning("no .text section found yet relo into text exist\n");
1050 return -LIBBPF_ERRNO__RELOC;
1052 new_cnt = prog->insns_cnt + text->insns_cnt;
1053 new_insn = realloc(prog->insns, new_cnt * sizeof(*insn));
1055 pr_warning("oom in prog realloc\n");
1058 memcpy(new_insn + prog->insns_cnt, text->insns,
1059 text->insns_cnt * sizeof(*insn));
1060 prog->insns = new_insn;
1061 prog->main_prog_cnt = prog->insns_cnt;
1062 prog->insns_cnt = new_cnt;
1064 insn = &prog->insns[relo->insn_idx];
1065 insn->imm += prog->main_prog_cnt - relo->insn_idx;
1066 pr_debug("added %zd insn from %s to prog %s\n",
1067 text->insns_cnt, text->section_name, prog->section_name);
1072 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
1076 if (!prog || !prog->reloc_desc)
1079 for (i = 0; i < prog->nr_reloc; i++) {
1080 if (prog->reloc_desc[i].type == RELO_LD64) {
1081 struct bpf_insn *insns = prog->insns;
1082 int insn_idx, map_idx;
1084 insn_idx = prog->reloc_desc[i].insn_idx;
1085 map_idx = prog->reloc_desc[i].map_idx;
1087 if (insn_idx >= (int)prog->insns_cnt) {
1088 pr_warning("relocation out of range: '%s'\n",
1089 prog->section_name);
1090 return -LIBBPF_ERRNO__RELOC;
1092 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
1093 insns[insn_idx].imm = obj->maps[map_idx].fd;
1095 err = bpf_program__reloc_text(prog, obj,
1096 &prog->reloc_desc[i]);
1102 zfree(&prog->reloc_desc);
1109 bpf_object__relocate(struct bpf_object *obj)
1111 struct bpf_program *prog;
1115 for (i = 0; i < obj->nr_programs; i++) {
1116 prog = &obj->programs[i];
1118 err = bpf_program__relocate(prog, obj);
1120 pr_warning("failed to relocate '%s'\n",
1121 prog->section_name);
1128 static int bpf_object__collect_reloc(struct bpf_object *obj)
1132 if (!obj_elf_valid(obj)) {
1133 pr_warning("Internal error: elf object is closed\n");
1134 return -LIBBPF_ERRNO__INTERNAL;
1137 for (i = 0; i < obj->efile.nr_reloc; i++) {
1138 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
1139 Elf_Data *data = obj->efile.reloc[i].data;
1140 int idx = shdr->sh_info;
1141 struct bpf_program *prog;
1143 if (shdr->sh_type != SHT_REL) {
1144 pr_warning("internal error at %d\n", __LINE__);
1145 return -LIBBPF_ERRNO__INTERNAL;
1148 prog = bpf_object__find_prog_by_idx(obj, idx);
1150 pr_warning("relocation failed: no section(%d)\n", idx);
1151 return -LIBBPF_ERRNO__RELOC;
1154 err = bpf_program__collect_reloc(prog,
1164 load_program(enum bpf_prog_type type, const char *name, struct bpf_insn *insns,
1165 int insns_cnt, char *license, u32 kern_version, int *pfd)
1170 if (!insns || !insns_cnt)
1173 log_buf = malloc(BPF_LOG_BUF_SIZE);
1175 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
1177 ret = bpf_load_program_name(type, name, insns, insns_cnt, license,
1178 kern_version, log_buf, BPF_LOG_BUF_SIZE);
1186 ret = -LIBBPF_ERRNO__LOAD;
1187 pr_warning("load bpf program failed: %s\n", strerror(errno));
1189 if (log_buf && log_buf[0] != '\0') {
1190 ret = -LIBBPF_ERRNO__VERIFY;
1191 pr_warning("-- BEGIN DUMP LOG ---\n");
1192 pr_warning("\n%s\n", log_buf);
1193 pr_warning("-- END LOG --\n");
1194 } else if (insns_cnt >= BPF_MAXINSNS) {
1195 pr_warning("Program too large (%d insns), at most %d insns\n",
1196 insns_cnt, BPF_MAXINSNS);
1197 ret = -LIBBPF_ERRNO__PROG2BIG;
1199 /* Wrong program type? */
1200 if (type != BPF_PROG_TYPE_KPROBE) {
1203 fd = bpf_load_program_name(BPF_PROG_TYPE_KPROBE, name,
1204 insns, insns_cnt, license,
1205 kern_version, NULL, 0);
1208 ret = -LIBBPF_ERRNO__PROGTYPE;
1214 ret = -LIBBPF_ERRNO__KVER;
1223 bpf_program__load(struct bpf_program *prog,
1224 char *license, u32 kern_version)
1228 if (prog->instances.nr < 0 || !prog->instances.fds) {
1229 if (prog->preprocessor) {
1230 pr_warning("Internal error: can't load program '%s'\n",
1231 prog->section_name);
1232 return -LIBBPF_ERRNO__INTERNAL;
1235 prog->instances.fds = malloc(sizeof(int));
1236 if (!prog->instances.fds) {
1237 pr_warning("Not enough memory for BPF fds\n");
1240 prog->instances.nr = 1;
1241 prog->instances.fds[0] = -1;
1244 if (!prog->preprocessor) {
1245 if (prog->instances.nr != 1) {
1246 pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
1247 prog->section_name, prog->instances.nr);
1249 err = load_program(prog->type, prog->name, prog->insns,
1250 prog->insns_cnt, license, kern_version, &fd);
1252 prog->instances.fds[0] = fd;
1256 for (i = 0; i < prog->instances.nr; i++) {
1257 struct bpf_prog_prep_result result;
1258 bpf_program_prep_t preprocessor = prog->preprocessor;
1260 bzero(&result, sizeof(result));
1261 err = preprocessor(prog, i, prog->insns,
1262 prog->insns_cnt, &result);
1264 pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
1265 i, prog->section_name);
1269 if (!result.new_insn_ptr || !result.new_insn_cnt) {
1270 pr_debug("Skip loading the %dth instance of program '%s'\n",
1271 i, prog->section_name);
1272 prog->instances.fds[i] = -1;
1278 err = load_program(prog->type, prog->name,
1279 result.new_insn_ptr,
1280 result.new_insn_cnt,
1281 license, kern_version, &fd);
1284 pr_warning("Loading the %dth instance of program '%s' failed\n",
1285 i, prog->section_name);
1291 prog->instances.fds[i] = fd;
1295 pr_warning("failed to load program '%s'\n",
1296 prog->section_name);
1297 zfree(&prog->insns);
1298 prog->insns_cnt = 0;
1303 bpf_object__load_progs(struct bpf_object *obj)
1308 for (i = 0; i < obj->nr_programs; i++) {
1309 if (obj->programs[i].idx == obj->efile.text_shndx)
1311 err = bpf_program__load(&obj->programs[i],
1320 static int bpf_object__validate(struct bpf_object *obj)
1322 if (obj->kern_version == 0) {
1323 pr_warning("%s doesn't provide kernel version\n",
1325 return -LIBBPF_ERRNO__KVERSION;
1330 static struct bpf_object *
1331 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz)
1333 struct bpf_object *obj;
1336 if (elf_version(EV_CURRENT) == EV_NONE) {
1337 pr_warning("failed to init libelf for %s\n", path);
1338 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1341 obj = bpf_object__new(path, obj_buf, obj_buf_sz);
1345 CHECK_ERR(bpf_object__elf_init(obj), err, out);
1346 CHECK_ERR(bpf_object__check_endianness(obj), err, out);
1347 CHECK_ERR(bpf_object__elf_collect(obj), err, out);
1348 CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
1349 CHECK_ERR(bpf_object__validate(obj), err, out);
1351 bpf_object__elf_finish(obj);
1354 bpf_object__close(obj);
1355 return ERR_PTR(err);
1358 struct bpf_object *bpf_object__open(const char *path)
1360 /* param validation */
1364 pr_debug("loading %s\n", path);
1366 return __bpf_object__open(path, NULL, 0);
1369 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
1375 /* param validation */
1376 if (!obj_buf || obj_buf_sz <= 0)
1380 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
1381 (unsigned long)obj_buf,
1382 (unsigned long)obj_buf_sz);
1383 tmp_name[sizeof(tmp_name) - 1] = '\0';
1386 pr_debug("loading object '%s' from buffer\n",
1389 return __bpf_object__open(name, obj_buf, obj_buf_sz);
1392 int bpf_object__unload(struct bpf_object *obj)
1399 for (i = 0; i < obj->nr_maps; i++)
1400 zclose(obj->maps[i].fd);
1402 for (i = 0; i < obj->nr_programs; i++)
1403 bpf_program__unload(&obj->programs[i]);
1408 int bpf_object__load(struct bpf_object *obj)
1416 pr_warning("object should not be loaded twice\n");
1422 CHECK_ERR(bpf_object__create_maps(obj), err, out);
1423 CHECK_ERR(bpf_object__relocate(obj), err, out);
1424 CHECK_ERR(bpf_object__load_progs(obj), err, out);
1428 bpf_object__unload(obj);
1429 pr_warning("failed to load object '%s'\n", obj->path);
1433 static int check_path(const char *path)
1435 struct statfs st_fs;
1442 dname = strdup(path);
1446 dir = dirname(dname);
1447 if (statfs(dir, &st_fs)) {
1448 pr_warning("failed to statfs %s: %s\n", dir, strerror(errno));
1453 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
1454 pr_warning("specified path %s is not on BPF FS\n", path);
1461 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
1466 err = check_path(path);
1471 pr_warning("invalid program pointer\n");
1475 if (instance < 0 || instance >= prog->instances.nr) {
1476 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
1477 instance, prog->section_name, prog->instances.nr);
1481 if (bpf_obj_pin(prog->instances.fds[instance], path)) {
1482 pr_warning("failed to pin program: %s\n", strerror(errno));
1485 pr_debug("pinned program '%s'\n", path);
1490 static int make_dir(const char *path)
1494 if (mkdir(path, 0700) && errno != EEXIST)
1498 pr_warning("failed to mkdir %s: %s\n", path, strerror(-err));
1502 int bpf_program__pin(struct bpf_program *prog, const char *path)
1506 err = check_path(path);
1511 pr_warning("invalid program pointer\n");
1515 if (prog->instances.nr <= 0) {
1516 pr_warning("no instances of prog %s to pin\n",
1517 prog->section_name);
1521 err = make_dir(path);
1525 for (i = 0; i < prog->instances.nr; i++) {
1529 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
1532 else if (len >= PATH_MAX)
1533 return -ENAMETOOLONG;
1535 err = bpf_program__pin_instance(prog, buf, i);
1543 int bpf_map__pin(struct bpf_map *map, const char *path)
1547 err = check_path(path);
1552 pr_warning("invalid map pointer\n");
1556 if (bpf_obj_pin(map->fd, path)) {
1557 pr_warning("failed to pin map: %s\n", strerror(errno));
1561 pr_debug("pinned map '%s'\n", path);
1565 int bpf_object__pin(struct bpf_object *obj, const char *path)
1567 struct bpf_program *prog;
1568 struct bpf_map *map;
1575 pr_warning("object not yet loaded; load it first\n");
1579 err = make_dir(path);
1583 bpf_map__for_each(map, obj) {
1587 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1588 bpf_map__name(map));
1591 else if (len >= PATH_MAX)
1592 return -ENAMETOOLONG;
1594 err = bpf_map__pin(map, buf);
1599 bpf_object__for_each_program(prog, obj) {
1603 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1604 prog->section_name);
1607 else if (len >= PATH_MAX)
1608 return -ENAMETOOLONG;
1610 err = bpf_program__pin(prog, buf);
1618 void bpf_object__close(struct bpf_object *obj)
1625 if (obj->clear_priv)
1626 obj->clear_priv(obj, obj->priv);
1628 bpf_object__elf_finish(obj);
1629 bpf_object__unload(obj);
1631 for (i = 0; i < obj->nr_maps; i++) {
1632 zfree(&obj->maps[i].name);
1633 if (obj->maps[i].clear_priv)
1634 obj->maps[i].clear_priv(&obj->maps[i],
1636 obj->maps[i].priv = NULL;
1637 obj->maps[i].clear_priv = NULL;
1642 if (obj->programs && obj->nr_programs) {
1643 for (i = 0; i < obj->nr_programs; i++)
1644 bpf_program__exit(&obj->programs[i]);
1646 zfree(&obj->programs);
1648 list_del(&obj->list);
1653 bpf_object__next(struct bpf_object *prev)
1655 struct bpf_object *next;
1658 next = list_first_entry(&bpf_objects_list,
1662 next = list_next_entry(prev, list);
1664 /* Empty list is noticed here so don't need checking on entry. */
1665 if (&next->list == &bpf_objects_list)
1671 const char *bpf_object__name(struct bpf_object *obj)
1673 return obj ? obj->path : ERR_PTR(-EINVAL);
1676 unsigned int bpf_object__kversion(struct bpf_object *obj)
1678 return obj ? obj->kern_version : 0;
1681 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
1682 bpf_object_clear_priv_t clear_priv)
1684 if (obj->priv && obj->clear_priv)
1685 obj->clear_priv(obj, obj->priv);
1688 obj->clear_priv = clear_priv;
1692 void *bpf_object__priv(struct bpf_object *obj)
1694 return obj ? obj->priv : ERR_PTR(-EINVAL);
1697 struct bpf_program *
1698 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1706 return &obj->programs[0];
1708 if (prev->obj != obj) {
1709 pr_warning("error: program handler doesn't match object\n");
1713 idx = (prev - obj->programs) + 1;
1714 if (idx >= obj->nr_programs)
1716 return &obj->programs[idx];
1719 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
1720 bpf_program_clear_priv_t clear_priv)
1722 if (prog->priv && prog->clear_priv)
1723 prog->clear_priv(prog, prog->priv);
1726 prog->clear_priv = clear_priv;
1730 void *bpf_program__priv(struct bpf_program *prog)
1732 return prog ? prog->priv : ERR_PTR(-EINVAL);
1735 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
1739 title = prog->section_name;
1741 title = strdup(title);
1743 pr_warning("failed to strdup program title\n");
1744 return ERR_PTR(-ENOMEM);
1751 int bpf_program__fd(struct bpf_program *prog)
1753 return bpf_program__nth_fd(prog, 0);
1756 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
1757 bpf_program_prep_t prep)
1761 if (nr_instances <= 0 || !prep)
1764 if (prog->instances.nr > 0 || prog->instances.fds) {
1765 pr_warning("Can't set pre-processor after loading\n");
1769 instances_fds = malloc(sizeof(int) * nr_instances);
1770 if (!instances_fds) {
1771 pr_warning("alloc memory failed for fds\n");
1775 /* fill all fd with -1 */
1776 memset(instances_fds, -1, sizeof(int) * nr_instances);
1778 prog->instances.nr = nr_instances;
1779 prog->instances.fds = instances_fds;
1780 prog->preprocessor = prep;
1784 int bpf_program__nth_fd(struct bpf_program *prog, int n)
1788 if (n >= prog->instances.nr || n < 0) {
1789 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
1790 n, prog->section_name, prog->instances.nr);
1794 fd = prog->instances.fds[n];
1796 pr_warning("%dth instance of program '%s' is invalid\n",
1797 n, prog->section_name);
1804 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
1809 static bool bpf_program__is_type(struct bpf_program *prog,
1810 enum bpf_prog_type type)
1812 return prog ? (prog->type == type) : false;
1815 #define BPF_PROG_TYPE_FNS(NAME, TYPE) \
1816 int bpf_program__set_##NAME(struct bpf_program *prog) \
1820 bpf_program__set_type(prog, TYPE); \
1824 bool bpf_program__is_##NAME(struct bpf_program *prog) \
1826 return bpf_program__is_type(prog, TYPE); \
1829 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
1830 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
1831 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
1832 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
1833 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
1834 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
1835 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
1837 #define BPF_PROG_SEC(string, type) { string, sizeof(string) - 1, type }
1838 static const struct {
1841 enum bpf_prog_type prog_type;
1842 } section_names[] = {
1843 BPF_PROG_SEC("socket", BPF_PROG_TYPE_SOCKET_FILTER),
1844 BPF_PROG_SEC("kprobe/", BPF_PROG_TYPE_KPROBE),
1845 BPF_PROG_SEC("kretprobe/", BPF_PROG_TYPE_KPROBE),
1846 BPF_PROG_SEC("classifier", BPF_PROG_TYPE_SCHED_CLS),
1847 BPF_PROG_SEC("action", BPF_PROG_TYPE_SCHED_ACT),
1848 BPF_PROG_SEC("tracepoint/", BPF_PROG_TYPE_TRACEPOINT),
1849 BPF_PROG_SEC("xdp", BPF_PROG_TYPE_XDP),
1850 BPF_PROG_SEC("perf_event", BPF_PROG_TYPE_PERF_EVENT),
1851 BPF_PROG_SEC("cgroup/skb", BPF_PROG_TYPE_CGROUP_SKB),
1852 BPF_PROG_SEC("cgroup/sock", BPF_PROG_TYPE_CGROUP_SOCK),
1853 BPF_PROG_SEC("cgroup/dev", BPF_PROG_TYPE_CGROUP_DEVICE),
1854 BPF_PROG_SEC("lwt_in", BPF_PROG_TYPE_LWT_IN),
1855 BPF_PROG_SEC("lwt_out", BPF_PROG_TYPE_LWT_OUT),
1856 BPF_PROG_SEC("lwt_xmit", BPF_PROG_TYPE_LWT_XMIT),
1857 BPF_PROG_SEC("sockops", BPF_PROG_TYPE_SOCK_OPS),
1858 BPF_PROG_SEC("sk_skb", BPF_PROG_TYPE_SK_SKB),
1862 static enum bpf_prog_type bpf_program__guess_type(struct bpf_program *prog)
1866 if (!prog->section_name)
1869 for (i = 0; i < ARRAY_SIZE(section_names); i++)
1870 if (strncmp(prog->section_name, section_names[i].sec,
1871 section_names[i].len) == 0)
1872 return section_names[i].prog_type;
1875 pr_warning("failed to guess program type based on section name %s\n",
1876 prog->section_name);
1878 return BPF_PROG_TYPE_UNSPEC;
1881 int bpf_map__fd(struct bpf_map *map)
1883 return map ? map->fd : -EINVAL;
1886 const struct bpf_map_def *bpf_map__def(struct bpf_map *map)
1888 return map ? &map->def : ERR_PTR(-EINVAL);
1891 const char *bpf_map__name(struct bpf_map *map)
1893 return map ? map->name : NULL;
1896 int bpf_map__set_priv(struct bpf_map *map, void *priv,
1897 bpf_map_clear_priv_t clear_priv)
1903 if (map->clear_priv)
1904 map->clear_priv(map, map->priv);
1908 map->clear_priv = clear_priv;
1912 void *bpf_map__priv(struct bpf_map *map)
1914 return map ? map->priv : ERR_PTR(-EINVAL);
1918 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
1921 struct bpf_map *s, *e;
1923 if (!obj || !obj->maps)
1927 e = obj->maps + obj->nr_maps;
1932 if ((prev < s) || (prev >= e)) {
1933 pr_warning("error in %s: map handler doesn't belong to object\n",
1938 idx = (prev - obj->maps) + 1;
1939 if (idx >= obj->nr_maps)
1941 return &obj->maps[idx];
1945 bpf_object__find_map_by_name(struct bpf_object *obj, const char *name)
1947 struct bpf_map *pos;
1949 bpf_map__for_each(pos, obj) {
1950 if (pos->name && !strcmp(pos->name, name))
1957 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
1961 for (i = 0; i < obj->nr_maps; i++) {
1962 if (obj->maps[i].offset == offset)
1963 return &obj->maps[i];
1965 return ERR_PTR(-ENOENT);
1968 long libbpf_get_error(const void *ptr)
1971 return PTR_ERR(ptr);
1975 int bpf_prog_load(const char *file, enum bpf_prog_type type,
1976 struct bpf_object **pobj, int *prog_fd)
1978 struct bpf_program *prog, *first_prog = NULL;
1979 struct bpf_object *obj;
1982 obj = bpf_object__open(file);
1986 bpf_object__for_each_program(prog, obj) {
1988 * If type is not specified, try to guess it based on
1991 if (type == BPF_PROG_TYPE_UNSPEC) {
1992 type = bpf_program__guess_type(prog);
1993 if (type == BPF_PROG_TYPE_UNSPEC) {
1994 bpf_object__close(obj);
1999 bpf_program__set_type(prog, type);
2000 if (prog->idx != obj->efile.text_shndx && !first_prog)
2005 pr_warning("object file doesn't contain bpf program\n");
2006 bpf_object__close(obj);
2010 err = bpf_object__load(obj);
2012 bpf_object__close(obj);
2017 *prog_fd = bpf_program__fd(first_prog);