1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
4 * resolve_btfids scans ELF object for .BTF_ids section and resolves
5 * its symbols with BTF ID values.
7 * Each symbol points to 4 bytes data and is expected to have
8 * following name syntax:
10 * __BTF_ID__<type>__<symbol>[__<id>]
14 * func - lookup BTF_KIND_FUNC symbol with <symbol> name
15 * and store its ID into the data:
17 * __BTF_ID__func__vfs_close__1:
20 * struct - lookup BTF_KIND_STRUCT symbol with <symbol> name
21 * and store its ID into the data:
23 * __BTF_ID__struct__sk_buff__1:
26 * union - lookup BTF_KIND_UNION symbol with <symbol> name
27 * and store its ID into the data:
29 * __BTF_ID__union__thread_union__1:
32 * typedef - lookup BTF_KIND_TYPEDEF symbol with <symbol> name
33 * and store its ID into the data:
35 * __BTF_ID__typedef__pid_t__1:
38 * set - store symbol size into first 4 bytes and sort following
41 * __BTF_ID__set__list:
44 * __BTF_ID__func__vfs_getattr__3:
46 * __BTF_ID__func__vfs_fallocate__4:
49 * set8 - store symbol size into first 4 bytes and sort following
52 * __BTF_ID__set8__list:
55 * __BTF_ID__func__vfs_getattr__3:
57 * .word (1 << 0) | (1 << 2)
58 * __BTF_ID__func__vfs_fallocate__5:
60 * .word (1 << 3) | (1 << 1) | (1 << 2)
73 #include <linux/btf_ids.h>
74 #include <linux/rbtree.h>
75 #include <linux/zalloc.h>
76 #include <linux/err.h>
78 #include <bpf/libbpf.h>
79 #include <subcmd/parse-options.h>
81 #define BTF_IDS_SECTION ".BTF_ids"
82 #define BTF_ID_PREFIX "__BTF_ID__"
84 #define BTF_STRUCT "struct"
85 #define BTF_UNION "union"
86 #define BTF_TYPEDEF "typedef"
87 #define BTF_FUNC "func"
89 #define BTF_SET8 "set8"
93 #if __BYTE_ORDER == __LITTLE_ENDIAN
94 # define ELFDATANATIVE ELFDATA2LSB
95 #elif __BYTE_ORDER == __BIG_ENDIAN
96 # define ELFDATANATIVE ELFDATA2MSB
98 # error "Unknown machine endianness!"
102 struct rb_node rb_node;
111 Elf64_Addr addr[ADDR_CNT];
117 const char *base_btf_path;
127 unsigned long idlist_addr;
132 struct rb_root structs;
133 struct rb_root unions;
134 struct rb_root typedefs;
135 struct rb_root funcs;
145 static int eprintf(int level, int var, const char *fmt, ...)
152 ret = vfprintf(stderr, fmt, args);
159 #define pr_fmt(fmt) fmt
162 #define pr_debug(fmt, ...) \
163 eprintf(1, verbose, pr_fmt(fmt), ##__VA_ARGS__)
164 #define pr_debugN(n, fmt, ...) \
165 eprintf(n, verbose, pr_fmt(fmt), ##__VA_ARGS__)
166 #define pr_debug2(fmt, ...) pr_debugN(2, pr_fmt(fmt), ##__VA_ARGS__)
167 #define pr_err(fmt, ...) \
168 eprintf(0, verbose, pr_fmt(fmt), ##__VA_ARGS__)
169 #define pr_info(fmt, ...) \
170 eprintf(0, verbose, pr_fmt(fmt), ##__VA_ARGS__)
172 static bool is_btf_id(const char *name)
174 return name && !strncmp(name, BTF_ID_PREFIX, sizeof(BTF_ID_PREFIX) - 1);
177 static struct btf_id *btf_id__find(struct rb_root *root, const char *name)
179 struct rb_node *p = root->rb_node;
184 id = rb_entry(p, struct btf_id, rb_node);
185 cmp = strcmp(id->name, name);
196 static struct btf_id *
197 btf_id__add(struct rb_root *root, char *name, bool unique)
199 struct rb_node **p = &root->rb_node;
200 struct rb_node *parent = NULL;
206 id = rb_entry(parent, struct btf_id, rb_node);
207 cmp = strcmp(id->name, name);
213 return unique ? NULL : id;
216 id = zalloc(sizeof(*id));
218 pr_debug("adding symbol %s\n", name);
220 rb_link_node(&id->rb_node, parent, p);
221 rb_insert_color(&id->rb_node, root);
226 static char *get_id(const char *prefix_end)
229 * __BTF_ID__func__vfs_truncate__0
233 int len = strlen(prefix_end);
234 int pos = sizeof("__") - 1;
240 id = strdup(prefix_end + pos);
243 * __BTF_ID__func__vfs_truncate__0
246 * cut the unique id part
248 p = strrchr(id, '_');
259 static struct btf_id *add_set(struct object *obj, char *name, bool is_set8)
262 * __BTF_ID__set__name
266 char *id = name + (is_set8 ? sizeof(BTF_SET8 "__") : sizeof(BTF_SET "__")) - 1;
267 int len = strlen(name);
269 if (id >= name + len) {
270 pr_err("FAILED to parse set name: %s\n", name);
274 return btf_id__add(&obj->sets, id, true);
277 static struct btf_id *add_symbol(struct rb_root *root, char *name, size_t size)
281 id = get_id(name + size);
283 pr_err("FAILED to parse symbol name: %s\n", name);
287 return btf_id__add(root, id, false);
290 /* Older libelf.h and glibc elf.h might not yet define the ELF compression types. */
291 #ifndef SHF_COMPRESSED
292 #define SHF_COMPRESSED (1 << 11) /* Section with compressed data. */
296 * The data of compressed section should be aligned to 4
297 * (for 32bit) or 8 (for 64 bit) bytes. The binutils ld
298 * sets sh_addralign to 1, which makes libelf fail with
299 * misaligned section error during the update:
300 * FAILED elf_update(WRITE): invalid section alignment
302 * While waiting for ld fix, we fix the compressed sections
303 * sh_addralign value manualy.
305 static int compressed_section_fix(Elf *elf, Elf_Scn *scn, GElf_Shdr *sh)
307 int expected = gelf_getclass(elf) == ELFCLASS32 ? 4 : 8;
309 if (!(sh->sh_flags & SHF_COMPRESSED))
312 if (sh->sh_addralign == expected)
315 pr_debug2(" - fixing wrong alignment sh_addralign %u, expected %u\n",
316 sh->sh_addralign, expected);
318 sh->sh_addralign = expected;
320 if (gelf_update_shdr(scn, sh) == 0) {
321 pr_err("FAILED cannot update section header: %s\n",
328 static int elf_collect(struct object *obj)
337 fd = open(obj->path, O_RDWR, 0666);
339 pr_err("FAILED cannot open %s: %s\n",
340 obj->path, strerror(errno));
344 elf_version(EV_CURRENT);
346 elf = elf_begin(fd, ELF_C_RDWR_MMAP, NULL);
349 pr_err("FAILED cannot create ELF descriptor: %s\n",
355 obj->efile.elf = elf;
357 elf_flagelf(elf, ELF_C_SET, ELF_F_LAYOUT);
359 if (elf_getshdrstrndx(elf, &shdrstrndx) != 0) {
360 pr_err("FAILED cannot get shdr str ndx\n");
364 if (gelf_getehdr(obj->efile.elf, &ehdr) == NULL) {
365 pr_err("FAILED cannot get ELF header: %s\n",
369 obj->efile.encoding = ehdr.e_ident[EI_DATA];
372 * Scan all the elf sections and look for save data
373 * from .BTF_ids section and symbols.
375 while ((scn = elf_nextscn(elf, scn)) != NULL) {
381 if (gelf_getshdr(scn, &sh) != &sh) {
382 pr_err("FAILED get section(%d) header\n", idx);
386 name = elf_strptr(elf, shdrstrndx, sh.sh_name);
388 pr_err("FAILED get section(%d) name\n", idx);
392 data = elf_getdata(scn, 0);
394 pr_err("FAILED to get section(%d) data from %s\n",
399 pr_debug2("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
400 idx, name, (unsigned long) data->d_size,
401 (int) sh.sh_link, (unsigned long) sh.sh_flags,
404 if (sh.sh_type == SHT_SYMTAB) {
405 obj->efile.symbols = data;
406 obj->efile.symbols_shndx = idx;
407 obj->efile.strtabidx = sh.sh_link;
408 } else if (!strcmp(name, BTF_IDS_SECTION)) {
409 obj->efile.idlist = data;
410 obj->efile.idlist_shndx = idx;
411 obj->efile.idlist_addr = sh.sh_addr;
414 if (compressed_section_fix(elf, scn, &sh))
421 static int symbols_collect(struct object *obj)
428 scn = elf_getscn(obj->efile.elf, obj->efile.symbols_shndx);
432 if (gelf_getshdr(scn, &sh) != &sh)
435 n = sh.sh_size / sh.sh_entsize;
438 * Scan symbols and look for the ones starting with
439 * __BTF_ID__* over .BTF_ids section.
441 for (i = 0; i < n; i++) {
446 if (!gelf_getsym(obj->efile.symbols, i, &sym))
449 if (sym.st_shndx != obj->efile.idlist_shndx)
452 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
455 if (!is_btf_id(name))
459 * __BTF_ID__TYPE__vfs_truncate__0
462 prefix = name + sizeof(BTF_ID_PREFIX) - 1;
465 if (!strncmp(prefix, BTF_STRUCT, sizeof(BTF_STRUCT) - 1)) {
467 id = add_symbol(&obj->structs, prefix, sizeof(BTF_STRUCT) - 1);
469 } else if (!strncmp(prefix, BTF_UNION, sizeof(BTF_UNION) - 1)) {
471 id = add_symbol(&obj->unions, prefix, sizeof(BTF_UNION) - 1);
473 } else if (!strncmp(prefix, BTF_TYPEDEF, sizeof(BTF_TYPEDEF) - 1)) {
475 id = add_symbol(&obj->typedefs, prefix, sizeof(BTF_TYPEDEF) - 1);
477 } else if (!strncmp(prefix, BTF_FUNC, sizeof(BTF_FUNC) - 1)) {
479 id = add_symbol(&obj->funcs, prefix, sizeof(BTF_FUNC) - 1);
481 } else if (!strncmp(prefix, BTF_SET8, sizeof(BTF_SET8) - 1)) {
482 id = add_set(obj, prefix, true);
484 * SET8 objects store list's count, which is encoded
485 * in symbol's size, together with 'cnt' field hence
489 id->cnt = sym.st_size / sizeof(uint64_t) - 1;
493 } else if (!strncmp(prefix, BTF_SET, sizeof(BTF_SET) - 1)) {
494 id = add_set(obj, prefix, false);
496 * SET objects store list's count, which is encoded
497 * in symbol's size, together with 'cnt' field hence
501 id->cnt = sym.st_size / sizeof(int) - 1;
505 pr_err("FAILED unsupported prefix %s\n", prefix);
512 if (id->addr_cnt >= ADDR_CNT) {
513 pr_err("FAILED symbol %s crossed the number of allowed lists\n",
517 id->addr[id->addr_cnt++] = sym.st_value;
523 static int symbols_resolve(struct object *obj)
525 int nr_typedefs = obj->nr_typedefs;
526 int nr_structs = obj->nr_structs;
527 int nr_unions = obj->nr_unions;
528 int nr_funcs = obj->nr_funcs;
529 struct btf *base_btf = NULL;
534 if (obj->base_btf_path) {
535 base_btf = btf__parse(obj->base_btf_path, NULL);
536 err = libbpf_get_error(base_btf);
538 pr_err("FAILED: load base BTF from %s: %s\n",
539 obj->base_btf_path, strerror(-err));
544 btf = btf__parse_split(obj->btf ?: obj->path, base_btf);
545 err = libbpf_get_error(btf);
547 pr_err("FAILED: load BTF from %s: %s\n",
548 obj->btf ?: obj->path, strerror(-err));
553 nr_types = btf__type_cnt(btf);
556 * Iterate all the BTF types and search for collected symbol IDs.
558 for (type_id = 1; type_id < nr_types; type_id++) {
559 const struct btf_type *type;
560 struct rb_root *root;
565 type = btf__type_by_id(btf, type_id);
567 pr_err("FAILED: malformed BTF, can't resolve type for ID %d\n",
572 if (btf_is_func(type) && nr_funcs) {
575 } else if (btf_is_struct(type) && nr_structs) {
577 root = &obj->structs;
578 } else if (btf_is_union(type) && nr_unions) {
581 } else if (btf_is_typedef(type) && nr_typedefs) {
583 root = &obj->typedefs;
587 str = btf__name_by_offset(btf, type->name_off);
589 pr_err("FAILED: malformed BTF, can't resolve name for ID %d\n",
594 id = btf_id__find(root, str);
597 pr_info("WARN: multiple IDs found for '%s': %d, %d - using %d\n",
598 str, id->id, type_id, id->id);
613 static int id_patch(struct object *obj, struct btf_id *id)
615 Elf_Data *data = obj->efile.idlist;
616 int *ptr = data->d_buf;
619 /* For set, set8, id->id may be 0 */
620 if (!id->id && !id->is_set && !id->is_set8)
621 pr_err("WARN: resolve_btfids: unresolved symbol %s\n", id->name);
623 for (i = 0; i < id->addr_cnt; i++) {
624 unsigned long addr = id->addr[i];
625 unsigned long idx = addr - obj->efile.idlist_addr;
627 pr_debug("patching addr %5lu: ID %7d [%s]\n",
628 idx, id->id, id->name);
630 if (idx >= data->d_size) {
631 pr_err("FAILED patching index %lu out of bounds %lu\n",
636 idx = idx / sizeof(int);
643 static int __symbols_patch(struct object *obj, struct rb_root *root)
645 struct rb_node *next;
648 next = rb_first(root);
650 id = rb_entry(next, struct btf_id, rb_node);
652 if (id_patch(obj, id))
655 next = rb_next(next);
660 static int cmp_id(const void *pa, const void *pb)
662 const int *a = pa, *b = pb;
667 static int sets_patch(struct object *obj)
669 Elf_Data *data = obj->efile.idlist;
670 struct rb_node *next;
672 next = rb_first(&obj->sets);
674 struct btf_id_set8 *set8;
675 struct btf_id_set *set;
676 unsigned long addr, off;
679 id = rb_entry(next, struct btf_id, rb_node);
681 off = addr - obj->efile.idlist_addr;
683 /* sets are unique */
684 if (id->addr_cnt != 1) {
685 pr_err("FAILED malformed data for set '%s'\n",
691 set = data->d_buf + off;
692 qsort(set->ids, set->cnt, sizeof(set->ids[0]), cmp_id);
694 set8 = data->d_buf + off;
696 * Make sure id is at the beginning of the pairs
697 * struct, otherwise the below qsort would not work.
699 BUILD_BUG_ON(set8->pairs != &set8->pairs[0].id);
700 qsort(set8->pairs, set8->cnt, sizeof(set8->pairs[0]), cmp_id);
703 * When ELF endianness does not match endianness of the
704 * host, libelf will do the translation when updating
705 * the ELF. This, however, corrupts SET8 flags which are
706 * already in the target endianness. So, let's bswap
707 * them to the host endianness and libelf will then
708 * correctly translate everything.
710 if (obj->efile.encoding != ELFDATANATIVE) {
713 set8->flags = bswap_32(set8->flags);
714 for (i = 0; i < set8->cnt; i++) {
715 set8->pairs[i].flags =
716 bswap_32(set8->pairs[i].flags);
721 pr_debug("sorting addr %5lu: cnt %6d [%s]\n",
722 off, id->is_set ? set->cnt : set8->cnt, id->name);
724 next = rb_next(next);
729 static int symbols_patch(struct object *obj)
733 if (__symbols_patch(obj, &obj->structs) ||
734 __symbols_patch(obj, &obj->unions) ||
735 __symbols_patch(obj, &obj->typedefs) ||
736 __symbols_patch(obj, &obj->funcs) ||
737 __symbols_patch(obj, &obj->sets))
743 /* Set type to ensure endian translation occurs. */
744 obj->efile.idlist->d_type = ELF_T_WORD;
746 elf_flagdata(obj->efile.idlist, ELF_C_SET, ELF_F_DIRTY);
748 err = elf_update(obj->efile.elf, ELF_C_WRITE);
750 pr_err("FAILED elf_update(WRITE): %s\n",
754 pr_debug("update %s for %s\n",
755 err >= 0 ? "ok" : "failed", obj->path);
756 return err < 0 ? -1 : 0;
759 static const char * const resolve_btfids_usage[] = {
760 "resolve_btfids [<options>] <ELF object>",
764 int main(int argc, const char **argv)
766 struct object obj = {
777 struct option btfid_options[] = {
778 OPT_INCR('v', "verbose", &verbose,
779 "be more verbose (show errors, etc)"),
780 OPT_STRING(0, "btf", &obj.btf, "BTF data",
782 OPT_STRING('b', "btf_base", &obj.base_btf_path, "file",
783 "path of file providing base BTF"),
788 argc = parse_options(argc, argv, btfid_options, resolve_btfids_usage,
789 PARSE_OPT_STOP_AT_NON_OPTION);
791 usage_with_options(resolve_btfids_usage, btfid_options);
795 if (elf_collect(&obj))
799 * We did not find .BTF_ids section or symbols section,
802 if (obj.efile.idlist_shndx == -1 ||
803 obj.efile.symbols_shndx == -1) {
804 pr_debug("Cannot find .BTF_ids or symbols sections, nothing to do\n");
809 if (symbols_collect(&obj))
812 if (symbols_resolve(&obj))
815 if (symbols_patch(&obj))
821 elf_end(obj.efile.elf);