1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 /* Copyright (c) 2018 Facebook */
12 #include <sys/utsname.h>
13 #include <sys/param.h>
15 #include <linux/kernel.h>
16 #include <linux/err.h>
17 #include <linux/btf.h>
22 #include "libbpf_internal.h"
26 #define BTF_MAX_NR_TYPES 0x7fffffffU
27 #define BTF_MAX_STR_OFFSET 0x7fffffffU
29 static struct btf_type btf_void;
32 /* raw BTF data in native endianness */
34 /* raw BTF data in non-native endianness */
35 void *raw_data_swapped;
37 /* whether target endianness differs from the native one */
41 * When BTF is loaded from an ELF or raw memory it is stored
42 * in a contiguous memory block. The hdr, type_data, and, strs_data
43 * point inside that memory region to their respective parts of BTF
46 * +--------------------------------+
47 * | Header | Types | Strings |
48 * +--------------------------------+
53 * strs_data------------+
55 * If BTF data is later modified, e.g., due to types added or
56 * removed, BTF deduplication performed, etc, this contiguous
57 * representation is broken up into three independently allocated
58 * memory regions to be able to modify them independently.
59 * raw_data is nulled out at that point, but can be later allocated
60 * and cached again if user calls btf__get_raw_data(), at which point
61 * raw_data will contain a contiguous copy of header, types, and
64 * +----------+ +---------+ +-----------+
65 * | Header | | Types | | Strings |
66 * +----------+ +---------+ +-----------+
71 * strset__data(strs_set)-----+
73 * +----------+---------+-----------+
74 * | Header | Types | Strings |
75 * raw_data----->+----------+---------+-----------+
77 struct btf_header *hdr;
80 size_t types_data_cap; /* used size stored in hdr->type_len */
82 /* type ID to `struct btf_type *` lookup index
83 * type_offs[0] corresponds to the first non-VOID type:
84 * - for base BTF it's type [1];
85 * - for split BTF it's the first non-base BTF type.
89 /* number of types in this BTF instance:
90 * - doesn't include special [0] void type;
91 * - for split BTF counts number of types added on top of base BTF.
94 /* if not NULL, points to the base BTF on top of which the current
98 /* BTF type ID of the first type in this BTF instance:
99 * - for base BTF it's equal to 1;
100 * - for split BTF it's equal to biggest type ID of base BTF plus 1.
103 /* logical string offset of this BTF instance:
104 * - for base BTF it's equal to 0;
105 * - for split BTF it's equal to total size of base BTF's string section size.
109 /* only one of strs_data or strs_set can be non-NULL, depending on
110 * whether BTF is in a modifiable state (strs_set is used) or not
111 * (strs_data points inside raw_data)
114 /* a set of unique strings */
115 struct strset *strs_set;
116 /* whether strings are already deduplicated */
119 /* BTF object FD, if loaded into kernel */
122 /* Pointer size (in bytes) for a target architecture of this BTF */
126 static inline __u64 ptr_to_u64(const void *ptr)
128 return (__u64) (unsigned long) ptr;
131 /* Ensure given dynamically allocated memory region pointed to by *data* with
132 * capacity of *cap_cnt* elements each taking *elem_sz* bytes has enough
133 * memory to accomodate *add_cnt* new elements, assuming *cur_cnt* elements
134 * are already used. At most *max_cnt* elements can be ever allocated.
135 * If necessary, memory is reallocated and all existing data is copied over,
136 * new pointer to the memory region is stored at *data, new memory region
137 * capacity (in number of elements) is stored in *cap.
138 * On success, memory pointer to the beginning of unused memory is returned.
139 * On error, NULL is returned.
141 void *libbpf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz,
142 size_t cur_cnt, size_t max_cnt, size_t add_cnt)
147 if (cur_cnt + add_cnt <= *cap_cnt)
148 return *data + cur_cnt * elem_sz;
150 /* requested more than the set limit */
151 if (cur_cnt + add_cnt > max_cnt)
155 new_cnt += new_cnt / 4; /* expand by 25% */
156 if (new_cnt < 16) /* but at least 16 elements */
158 if (new_cnt > max_cnt) /* but not exceeding a set limit */
160 if (new_cnt < cur_cnt + add_cnt) /* also ensure we have enough memory */
161 new_cnt = cur_cnt + add_cnt;
163 new_data = libbpf_reallocarray(*data, new_cnt, elem_sz);
167 /* zero out newly allocated portion of memory */
168 memset(new_data + (*cap_cnt) * elem_sz, 0, (new_cnt - *cap_cnt) * elem_sz);
172 return new_data + cur_cnt * elem_sz;
175 /* Ensure given dynamically allocated memory region has enough allocated space
176 * to accommodate *need_cnt* elements of size *elem_sz* bytes each
178 int libbpf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt)
182 if (need_cnt <= *cap_cnt)
185 p = libbpf_add_mem(data, cap_cnt, elem_sz, *cap_cnt, SIZE_MAX, need_cnt - *cap_cnt);
192 static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
196 p = libbpf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
197 btf->nr_types, BTF_MAX_NR_TYPES, 1);
205 static void btf_bswap_hdr(struct btf_header *h)
207 h->magic = bswap_16(h->magic);
208 h->hdr_len = bswap_32(h->hdr_len);
209 h->type_off = bswap_32(h->type_off);
210 h->type_len = bswap_32(h->type_len);
211 h->str_off = bswap_32(h->str_off);
212 h->str_len = bswap_32(h->str_len);
215 static int btf_parse_hdr(struct btf *btf)
217 struct btf_header *hdr = btf->hdr;
220 if (btf->raw_size < sizeof(struct btf_header)) {
221 pr_debug("BTF header not found\n");
225 if (hdr->magic == bswap_16(BTF_MAGIC)) {
226 btf->swapped_endian = true;
227 if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
228 pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
229 bswap_32(hdr->hdr_len));
233 } else if (hdr->magic != BTF_MAGIC) {
234 pr_debug("Invalid BTF magic:%x\n", hdr->magic);
238 meta_left = btf->raw_size - sizeof(*hdr);
239 if (meta_left < hdr->str_off + hdr->str_len) {
240 pr_debug("Invalid BTF total size:%u\n", btf->raw_size);
244 if (hdr->type_off + hdr->type_len > hdr->str_off) {
245 pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
246 hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
250 if (hdr->type_off % 4) {
251 pr_debug("BTF type section is not aligned to 4 bytes\n");
258 static int btf_parse_str_sec(struct btf *btf)
260 const struct btf_header *hdr = btf->hdr;
261 const char *start = btf->strs_data;
262 const char *end = start + btf->hdr->str_len;
264 if (btf->base_btf && hdr->str_len == 0)
266 if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET || end[-1]) {
267 pr_debug("Invalid BTF string section\n");
270 if (!btf->base_btf && start[0]) {
271 pr_debug("Invalid BTF string section\n");
277 static int btf_type_size(const struct btf_type *t)
279 const int base_size = sizeof(struct btf_type);
280 __u16 vlen = btf_vlen(t);
282 switch (btf_kind(t)) {
285 case BTF_KIND_VOLATILE:
286 case BTF_KIND_RESTRICT:
288 case BTF_KIND_TYPEDEF:
293 return base_size + sizeof(__u32);
295 return base_size + vlen * sizeof(struct btf_enum);
297 return base_size + sizeof(struct btf_array);
298 case BTF_KIND_STRUCT:
300 return base_size + vlen * sizeof(struct btf_member);
301 case BTF_KIND_FUNC_PROTO:
302 return base_size + vlen * sizeof(struct btf_param);
304 return base_size + sizeof(struct btf_var);
305 case BTF_KIND_DATASEC:
306 return base_size + vlen * sizeof(struct btf_var_secinfo);
308 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
313 static void btf_bswap_type_base(struct btf_type *t)
315 t->name_off = bswap_32(t->name_off);
316 t->info = bswap_32(t->info);
317 t->type = bswap_32(t->type);
320 static int btf_bswap_type_rest(struct btf_type *t)
322 struct btf_var_secinfo *v;
323 struct btf_member *m;
327 __u16 vlen = btf_vlen(t);
330 switch (btf_kind(t)) {
333 case BTF_KIND_VOLATILE:
334 case BTF_KIND_RESTRICT:
336 case BTF_KIND_TYPEDEF:
341 *(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
344 for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
345 e->name_off = bswap_32(e->name_off);
346 e->val = bswap_32(e->val);
351 a->type = bswap_32(a->type);
352 a->index_type = bswap_32(a->index_type);
353 a->nelems = bswap_32(a->nelems);
355 case BTF_KIND_STRUCT:
357 for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
358 m->name_off = bswap_32(m->name_off);
359 m->type = bswap_32(m->type);
360 m->offset = bswap_32(m->offset);
363 case BTF_KIND_FUNC_PROTO:
364 for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
365 p->name_off = bswap_32(p->name_off);
366 p->type = bswap_32(p->type);
370 btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
372 case BTF_KIND_DATASEC:
373 for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
374 v->type = bswap_32(v->type);
375 v->offset = bswap_32(v->offset);
376 v->size = bswap_32(v->size);
380 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
385 static int btf_parse_type_sec(struct btf *btf)
387 struct btf_header *hdr = btf->hdr;
388 void *next_type = btf->types_data;
389 void *end_type = next_type + hdr->type_len;
392 while (next_type + sizeof(struct btf_type) <= end_type) {
393 if (btf->swapped_endian)
394 btf_bswap_type_base(next_type);
396 type_size = btf_type_size(next_type);
399 if (next_type + type_size > end_type) {
400 pr_warn("BTF type [%d] is malformed\n", btf->start_id + btf->nr_types);
404 if (btf->swapped_endian && btf_bswap_type_rest(next_type))
407 err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
411 next_type += type_size;
415 if (next_type != end_type) {
416 pr_warn("BTF types data is malformed\n");
423 __u32 btf__get_nr_types(const struct btf *btf)
425 return btf->start_id + btf->nr_types - 1;
428 const struct btf *btf__base_btf(const struct btf *btf)
430 return btf->base_btf;
433 /* internal helper returning non-const pointer to a type */
434 struct btf_type *btf_type_by_id(struct btf *btf, __u32 type_id)
438 if (type_id < btf->start_id)
439 return btf_type_by_id(btf->base_btf, type_id);
440 return btf->types_data + btf->type_offs[type_id - btf->start_id];
443 const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
445 if (type_id >= btf->start_id + btf->nr_types)
446 return errno = EINVAL, NULL;
447 return btf_type_by_id((struct btf *)btf, type_id);
450 static int determine_ptr_size(const struct btf *btf)
452 const struct btf_type *t;
456 if (btf->base_btf && btf->base_btf->ptr_sz > 0)
457 return btf->base_btf->ptr_sz;
459 n = btf__get_nr_types(btf);
460 for (i = 1; i <= n; i++) {
461 t = btf__type_by_id(btf, i);
465 name = btf__name_by_offset(btf, t->name_off);
469 if (strcmp(name, "long int") == 0 ||
470 strcmp(name, "long unsigned int") == 0) {
471 if (t->size != 4 && t->size != 8)
480 static size_t btf_ptr_sz(const struct btf *btf)
483 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
484 return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
487 /* Return pointer size this BTF instance assumes. The size is heuristically
488 * determined by looking for 'long' or 'unsigned long' integer type and
489 * recording its size in bytes. If BTF type information doesn't have any such
490 * type, this function returns 0. In the latter case, native architecture's
491 * pointer size is assumed, so will be either 4 or 8, depending on
492 * architecture that libbpf was compiled for. It's possible to override
493 * guessed value by using btf__set_pointer_size() API.
495 size_t btf__pointer_size(const struct btf *btf)
498 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
501 /* not enough BTF type info to guess */
507 /* Override or set pointer size in bytes. Only values of 4 and 8 are
510 int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
512 if (ptr_sz != 4 && ptr_sz != 8)
513 return libbpf_err(-EINVAL);
514 btf->ptr_sz = ptr_sz;
518 static bool is_host_big_endian(void)
520 #if __BYTE_ORDER == __LITTLE_ENDIAN
522 #elif __BYTE_ORDER == __BIG_ENDIAN
525 # error "Unrecognized __BYTE_ORDER__"
529 enum btf_endianness btf__endianness(const struct btf *btf)
531 if (is_host_big_endian())
532 return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
534 return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
537 int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
539 if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
540 return libbpf_err(-EINVAL);
542 btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
543 if (!btf->swapped_endian) {
544 free(btf->raw_data_swapped);
545 btf->raw_data_swapped = NULL;
550 static bool btf_type_is_void(const struct btf_type *t)
552 return t == &btf_void || btf_is_fwd(t);
555 static bool btf_type_is_void_or_null(const struct btf_type *t)
557 return !t || btf_type_is_void(t);
560 #define MAX_RESOLVE_DEPTH 32
562 __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
564 const struct btf_array *array;
565 const struct btf_type *t;
570 t = btf__type_by_id(btf, type_id);
571 for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t); i++) {
572 switch (btf_kind(t)) {
574 case BTF_KIND_STRUCT:
577 case BTF_KIND_DATASEC:
582 size = btf_ptr_sz(btf);
584 case BTF_KIND_TYPEDEF:
585 case BTF_KIND_VOLATILE:
587 case BTF_KIND_RESTRICT:
592 array = btf_array(t);
593 if (nelems && array->nelems > UINT32_MAX / nelems)
594 return libbpf_err(-E2BIG);
595 nelems *= array->nelems;
596 type_id = array->type;
599 return libbpf_err(-EINVAL);
602 t = btf__type_by_id(btf, type_id);
607 return libbpf_err(-EINVAL);
608 if (nelems && size > UINT32_MAX / nelems)
609 return libbpf_err(-E2BIG);
611 return nelems * size;
614 int btf__align_of(const struct btf *btf, __u32 id)
616 const struct btf_type *t = btf__type_by_id(btf, id);
617 __u16 kind = btf_kind(t);
623 return min(btf_ptr_sz(btf), (size_t)t->size);
625 return btf_ptr_sz(btf);
626 case BTF_KIND_TYPEDEF:
627 case BTF_KIND_VOLATILE:
629 case BTF_KIND_RESTRICT:
630 return btf__align_of(btf, t->type);
632 return btf__align_of(btf, btf_array(t)->type);
633 case BTF_KIND_STRUCT:
634 case BTF_KIND_UNION: {
635 const struct btf_member *m = btf_members(t);
636 __u16 vlen = btf_vlen(t);
637 int i, max_align = 1, align;
639 for (i = 0; i < vlen; i++, m++) {
640 align = btf__align_of(btf, m->type);
642 return libbpf_err(align);
643 max_align = max(max_align, align);
649 pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
650 return errno = EINVAL, 0;
654 int btf__resolve_type(const struct btf *btf, __u32 type_id)
656 const struct btf_type *t;
659 t = btf__type_by_id(btf, type_id);
660 while (depth < MAX_RESOLVE_DEPTH &&
661 !btf_type_is_void_or_null(t) &&
662 (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
664 t = btf__type_by_id(btf, type_id);
668 if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
669 return libbpf_err(-EINVAL);
674 __s32 btf__find_by_name(const struct btf *btf, const char *type_name)
676 __u32 i, nr_types = btf__get_nr_types(btf);
678 if (!strcmp(type_name, "void"))
681 for (i = 1; i <= nr_types; i++) {
682 const struct btf_type *t = btf__type_by_id(btf, i);
683 const char *name = btf__name_by_offset(btf, t->name_off);
685 if (name && !strcmp(type_name, name))
689 return libbpf_err(-ENOENT);
692 __s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
695 __u32 i, nr_types = btf__get_nr_types(btf);
697 if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
700 for (i = 1; i <= nr_types; i++) {
701 const struct btf_type *t = btf__type_by_id(btf, i);
704 if (btf_kind(t) != kind)
706 name = btf__name_by_offset(btf, t->name_off);
707 if (name && !strcmp(type_name, name))
711 return libbpf_err(-ENOENT);
714 static bool btf_is_modifiable(const struct btf *btf)
716 return (void *)btf->hdr != btf->raw_data;
719 void btf__free(struct btf *btf)
721 if (IS_ERR_OR_NULL(btf))
727 if (btf_is_modifiable(btf)) {
728 /* if BTF was modified after loading, it will have a split
729 * in-memory representation for header, types, and strings
730 * sections, so we need to free all of them individually. It
731 * might still have a cached contiguous raw data present,
732 * which will be unconditionally freed below.
735 free(btf->types_data);
736 strset__free(btf->strs_set);
739 free(btf->raw_data_swapped);
740 free(btf->type_offs);
744 static struct btf *btf_new_empty(struct btf *base_btf)
748 btf = calloc(1, sizeof(*btf));
750 return ERR_PTR(-ENOMEM);
754 btf->start_str_off = 0;
756 btf->ptr_sz = sizeof(void *);
757 btf->swapped_endian = false;
760 btf->base_btf = base_btf;
761 btf->start_id = btf__get_nr_types(base_btf) + 1;
762 btf->start_str_off = base_btf->hdr->str_len;
765 /* +1 for empty string at offset 0 */
766 btf->raw_size = sizeof(struct btf_header) + (base_btf ? 0 : 1);
767 btf->raw_data = calloc(1, btf->raw_size);
768 if (!btf->raw_data) {
770 return ERR_PTR(-ENOMEM);
773 btf->hdr = btf->raw_data;
774 btf->hdr->hdr_len = sizeof(struct btf_header);
775 btf->hdr->magic = BTF_MAGIC;
776 btf->hdr->version = BTF_VERSION;
778 btf->types_data = btf->raw_data + btf->hdr->hdr_len;
779 btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
780 btf->hdr->str_len = base_btf ? 0 : 1; /* empty string at offset 0 */
785 struct btf *btf__new_empty(void)
787 return libbpf_ptr(btf_new_empty(NULL));
790 struct btf *btf__new_empty_split(struct btf *base_btf)
792 return libbpf_ptr(btf_new_empty(base_btf));
795 static struct btf *btf_new(const void *data, __u32 size, struct btf *base_btf)
800 btf = calloc(1, sizeof(struct btf));
802 return ERR_PTR(-ENOMEM);
806 btf->start_str_off = 0;
809 btf->base_btf = base_btf;
810 btf->start_id = btf__get_nr_types(base_btf) + 1;
811 btf->start_str_off = base_btf->hdr->str_len;
814 btf->raw_data = malloc(size);
815 if (!btf->raw_data) {
819 memcpy(btf->raw_data, data, size);
820 btf->raw_size = size;
822 btf->hdr = btf->raw_data;
823 err = btf_parse_hdr(btf);
827 btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
828 btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
830 err = btf_parse_str_sec(btf);
831 err = err ?: btf_parse_type_sec(btf);
846 struct btf *btf__new(const void *data, __u32 size)
848 return libbpf_ptr(btf_new(data, size, NULL));
851 static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,
852 struct btf_ext **btf_ext)
854 Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
855 int err = 0, fd = -1, idx = 0;
856 struct btf *btf = NULL;
862 if (elf_version(EV_CURRENT) == EV_NONE) {
863 pr_warn("failed to init libelf for %s\n", path);
864 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
867 fd = open(path, O_RDONLY);
870 pr_warn("failed to open %s: %s\n", path, strerror(errno));
874 err = -LIBBPF_ERRNO__FORMAT;
876 elf = elf_begin(fd, ELF_C_READ, NULL);
878 pr_warn("failed to open %s as ELF file\n", path);
881 if (!gelf_getehdr(elf, &ehdr)) {
882 pr_warn("failed to get EHDR from %s\n", path);
886 if (elf_getshdrstrndx(elf, &shstrndx)) {
887 pr_warn("failed to get section names section index for %s\n",
892 if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
893 pr_warn("failed to get e_shstrndx from %s\n", path);
897 while ((scn = elf_nextscn(elf, scn)) != NULL) {
902 if (gelf_getshdr(scn, &sh) != &sh) {
903 pr_warn("failed to get section(%d) header from %s\n",
907 name = elf_strptr(elf, shstrndx, sh.sh_name);
909 pr_warn("failed to get section(%d) name from %s\n",
913 if (strcmp(name, BTF_ELF_SEC) == 0) {
914 btf_data = elf_getdata(scn, 0);
916 pr_warn("failed to get section(%d, %s) data from %s\n",
921 } else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
922 btf_ext_data = elf_getdata(scn, 0);
924 pr_warn("failed to get section(%d, %s) data from %s\n",
938 btf = btf_new(btf_data->d_buf, btf_data->d_size, base_btf);
939 err = libbpf_get_error(btf);
943 switch (gelf_getclass(elf)) {
945 btf__set_pointer_size(btf, 4);
948 btf__set_pointer_size(btf, 8);
951 pr_warn("failed to get ELF class (bitness) for %s\n", path);
955 if (btf_ext && btf_ext_data) {
956 *btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
957 err = libbpf_get_error(*btf_ext);
960 } else if (btf_ext) {
972 btf_ext__free(*btf_ext);
978 struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
980 return libbpf_ptr(btf_parse_elf(path, NULL, btf_ext));
983 struct btf *btf__parse_elf_split(const char *path, struct btf *base_btf)
985 return libbpf_ptr(btf_parse_elf(path, base_btf, NULL));
988 static struct btf *btf_parse_raw(const char *path, struct btf *base_btf)
990 struct btf *btf = NULL;
997 f = fopen(path, "rb");
1003 /* check BTF magic */
1004 if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
1008 if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
1009 /* definitely not a raw BTF */
1015 if (fseek(f, 0, SEEK_END)) {
1024 /* rewind to the start */
1025 if (fseek(f, 0, SEEK_SET)) {
1030 /* pre-alloc memory and read all of BTF data */
1036 if (fread(data, 1, sz, f) < sz) {
1041 /* finally parse BTF data */
1042 btf = btf_new(data, sz, base_btf);
1048 return err ? ERR_PTR(err) : btf;
1051 struct btf *btf__parse_raw(const char *path)
1053 return libbpf_ptr(btf_parse_raw(path, NULL));
1056 struct btf *btf__parse_raw_split(const char *path, struct btf *base_btf)
1058 return libbpf_ptr(btf_parse_raw(path, base_btf));
1061 static struct btf *btf_parse(const char *path, struct btf *base_btf, struct btf_ext **btf_ext)
1069 btf = btf_parse_raw(path, base_btf);
1070 err = libbpf_get_error(btf);
1074 return ERR_PTR(err);
1075 return btf_parse_elf(path, base_btf, btf_ext);
1078 struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
1080 return libbpf_ptr(btf_parse(path, NULL, btf_ext));
1083 struct btf *btf__parse_split(const char *path, struct btf *base_btf)
1085 return libbpf_ptr(btf_parse(path, base_btf, NULL));
1088 static int compare_vsi_off(const void *_a, const void *_b)
1090 const struct btf_var_secinfo *a = _a;
1091 const struct btf_var_secinfo *b = _b;
1093 return a->offset - b->offset;
1096 static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
1099 __u32 size = 0, off = 0, i, vars = btf_vlen(t);
1100 const char *name = btf__name_by_offset(btf, t->name_off);
1101 const struct btf_type *t_var;
1102 struct btf_var_secinfo *vsi;
1103 const struct btf_var *var;
1107 pr_debug("No name found in string section for DATASEC kind.\n");
1111 /* .extern datasec size and var offsets were set correctly during
1112 * extern collection step, so just skip straight to sorting variables
1117 ret = bpf_object__section_size(obj, name, &size);
1118 if (ret || !size || (t->size && t->size != size)) {
1119 pr_debug("Invalid size for section %s: %u bytes\n", name, size);
1125 for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
1126 t_var = btf__type_by_id(btf, vsi->type);
1127 var = btf_var(t_var);
1129 if (!btf_is_var(t_var)) {
1130 pr_debug("Non-VAR type seen in section %s\n", name);
1134 if (var->linkage == BTF_VAR_STATIC)
1137 name = btf__name_by_offset(btf, t_var->name_off);
1139 pr_debug("No name found in string section for VAR kind\n");
1143 ret = bpf_object__variable_offset(obj, name, &off);
1145 pr_debug("No offset found in symbol table for VAR %s\n",
1154 qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
1158 int btf__finalize_data(struct bpf_object *obj, struct btf *btf)
1163 for (i = 1; i <= btf->nr_types; i++) {
1164 struct btf_type *t = btf_type_by_id(btf, i);
1166 /* Loader needs to fix up some of the things compiler
1167 * couldn't get its hands on while emitting BTF. This
1168 * is section size and global variable offset. We use
1169 * the info from the ELF itself for this purpose.
1171 if (btf_is_datasec(t)) {
1172 err = btf_fixup_datasec(obj, btf, t);
1178 return libbpf_err(err);
1181 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1183 int btf__load(struct btf *btf)
1185 __u32 log_buf_size = 0, raw_size;
1186 char *log_buf = NULL;
1191 return libbpf_err(-EEXIST);
1195 log_buf = malloc(log_buf_size);
1197 return libbpf_err(-ENOMEM);
1202 raw_data = btf_get_raw_data(btf, &raw_size, false);
1207 /* cache native raw data representation */
1208 btf->raw_size = raw_size;
1209 btf->raw_data = raw_data;
1211 btf->fd = bpf_load_btf(raw_data, raw_size, log_buf, log_buf_size, false);
1213 if (!log_buf || errno == ENOSPC) {
1214 log_buf_size = max((__u32)BPF_LOG_BUF_SIZE,
1221 pr_warn("Error loading BTF: %s(%d)\n", strerror(errno), errno);
1223 pr_warn("%s\n", log_buf);
1229 return libbpf_err(err);
1232 int btf__fd(const struct btf *btf)
1237 void btf__set_fd(struct btf *btf, int fd)
1242 static const void *btf_strs_data(const struct btf *btf)
1244 return btf->strs_data ? btf->strs_data : strset__data(btf->strs_set);
1247 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1249 struct btf_header *hdr = btf->hdr;
1255 data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1257 *size = btf->raw_size;
1261 data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1262 data = calloc(1, data_sz);
1267 memcpy(p, hdr, hdr->hdr_len);
1272 memcpy(p, btf->types_data, hdr->type_len);
1274 for (i = 0; i < btf->nr_types; i++) {
1275 t = p + btf->type_offs[i];
1276 /* btf_bswap_type_rest() relies on native t->info, so
1277 * we swap base type info after we swapped all the
1278 * additional information
1280 if (btf_bswap_type_rest(t))
1282 btf_bswap_type_base(t);
1287 memcpy(p, btf_strs_data(btf), hdr->str_len);
1297 const void *btf__get_raw_data(const struct btf *btf_ro, __u32 *size)
1299 struct btf *btf = (struct btf *)btf_ro;
1303 data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1305 return errno = -ENOMEM, NULL;
1307 btf->raw_size = data_sz;
1308 if (btf->swapped_endian)
1309 btf->raw_data_swapped = data;
1311 btf->raw_data = data;
1316 const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
1318 if (offset < btf->start_str_off)
1319 return btf__str_by_offset(btf->base_btf, offset);
1320 else if (offset - btf->start_str_off < btf->hdr->str_len)
1321 return btf_strs_data(btf) + (offset - btf->start_str_off);
1323 return errno = EINVAL, NULL;
1326 const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1328 return btf__str_by_offset(btf, offset);
1331 struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf)
1333 struct bpf_btf_info btf_info;
1334 __u32 len = sizeof(btf_info);
1340 /* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
1341 * let's start with a sane default - 4KiB here - and resize it only if
1342 * bpf_obj_get_info_by_fd() needs a bigger buffer.
1345 ptr = malloc(last_size);
1347 return ERR_PTR(-ENOMEM);
1349 memset(&btf_info, 0, sizeof(btf_info));
1350 btf_info.btf = ptr_to_u64(ptr);
1351 btf_info.btf_size = last_size;
1352 err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1354 if (!err && btf_info.btf_size > last_size) {
1357 last_size = btf_info.btf_size;
1358 temp_ptr = realloc(ptr, last_size);
1360 btf = ERR_PTR(-ENOMEM);
1365 len = sizeof(btf_info);
1366 memset(&btf_info, 0, sizeof(btf_info));
1367 btf_info.btf = ptr_to_u64(ptr);
1368 btf_info.btf_size = last_size;
1370 err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1373 if (err || btf_info.btf_size > last_size) {
1374 btf = err ? ERR_PTR(-errno) : ERR_PTR(-E2BIG);
1378 btf = btf_new(ptr, btf_info.btf_size, base_btf);
1385 int btf__get_from_id(__u32 id, struct btf **btf)
1391 btf_fd = bpf_btf_get_fd_by_id(id);
1393 return libbpf_err(-errno);
1395 res = btf_get_from_fd(btf_fd, NULL);
1396 err = libbpf_get_error(res);
1401 return libbpf_err(err);
1407 int btf__get_map_kv_tids(const struct btf *btf, const char *map_name,
1408 __u32 expected_key_size, __u32 expected_value_size,
1409 __u32 *key_type_id, __u32 *value_type_id)
1411 const struct btf_type *container_type;
1412 const struct btf_member *key, *value;
1413 const size_t max_name = 256;
1414 char container_name[max_name];
1415 __s64 key_size, value_size;
1418 if (snprintf(container_name, max_name, "____btf_map_%s", map_name) == max_name) {
1419 pr_warn("map:%s length of '____btf_map_%s' is too long\n",
1420 map_name, map_name);
1421 return libbpf_err(-EINVAL);
1424 container_id = btf__find_by_name(btf, container_name);
1425 if (container_id < 0) {
1426 pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
1427 map_name, container_name);
1428 return libbpf_err(container_id);
1431 container_type = btf__type_by_id(btf, container_id);
1432 if (!container_type) {
1433 pr_warn("map:%s cannot find BTF type for container_id:%u\n",
1434 map_name, container_id);
1435 return libbpf_err(-EINVAL);
1438 if (!btf_is_struct(container_type) || btf_vlen(container_type) < 2) {
1439 pr_warn("map:%s container_name:%s is an invalid container struct\n",
1440 map_name, container_name);
1441 return libbpf_err(-EINVAL);
1444 key = btf_members(container_type);
1447 key_size = btf__resolve_size(btf, key->type);
1449 pr_warn("map:%s invalid BTF key_type_size\n", map_name);
1450 return libbpf_err(key_size);
1453 if (expected_key_size != key_size) {
1454 pr_warn("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
1455 map_name, (__u32)key_size, expected_key_size);
1456 return libbpf_err(-EINVAL);
1459 value_size = btf__resolve_size(btf, value->type);
1460 if (value_size < 0) {
1461 pr_warn("map:%s invalid BTF value_type_size\n", map_name);
1462 return libbpf_err(value_size);
1465 if (expected_value_size != value_size) {
1466 pr_warn("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1467 map_name, (__u32)value_size, expected_value_size);
1468 return libbpf_err(-EINVAL);
1471 *key_type_id = key->type;
1472 *value_type_id = value->type;
1477 static void btf_invalidate_raw_data(struct btf *btf)
1479 if (btf->raw_data) {
1480 free(btf->raw_data);
1481 btf->raw_data = NULL;
1483 if (btf->raw_data_swapped) {
1484 free(btf->raw_data_swapped);
1485 btf->raw_data_swapped = NULL;
1489 /* Ensure BTF is ready to be modified (by splitting into a three memory
1490 * regions for header, types, and strings). Also invalidate cached
1493 static int btf_ensure_modifiable(struct btf *btf)
1496 struct strset *set = NULL;
1499 if (btf_is_modifiable(btf)) {
1500 /* any BTF modification invalidates raw_data */
1501 btf_invalidate_raw_data(btf);
1505 /* split raw data into three memory regions */
1506 hdr = malloc(btf->hdr->hdr_len);
1507 types = malloc(btf->hdr->type_len);
1511 memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1512 memcpy(types, btf->types_data, btf->hdr->type_len);
1514 /* build lookup index for all strings */
1515 set = strset__new(BTF_MAX_STR_OFFSET, btf->strs_data, btf->hdr->str_len);
1521 /* only when everything was successful, update internal state */
1523 btf->types_data = types;
1524 btf->types_data_cap = btf->hdr->type_len;
1525 btf->strs_data = NULL;
1526 btf->strs_set = set;
1527 /* if BTF was created from scratch, all strings are guaranteed to be
1528 * unique and deduplicated
1530 if (btf->hdr->str_len == 0)
1531 btf->strs_deduped = true;
1532 if (!btf->base_btf && btf->hdr->str_len == 1)
1533 btf->strs_deduped = true;
1535 /* invalidate raw_data representation */
1536 btf_invalidate_raw_data(btf);
1547 /* Find an offset in BTF string section that corresponds to a given string *s*.
1549 * - >0 offset into string section, if string is found;
1550 * - -ENOENT, if string is not in the string section;
1551 * - <0, on any other error.
1553 int btf__find_str(struct btf *btf, const char *s)
1557 if (btf->base_btf) {
1558 off = btf__find_str(btf->base_btf, s);
1563 /* BTF needs to be in a modifiable state to build string lookup index */
1564 if (btf_ensure_modifiable(btf))
1565 return libbpf_err(-ENOMEM);
1567 off = strset__find_str(btf->strs_set, s);
1569 return libbpf_err(off);
1571 return btf->start_str_off + off;
1574 /* Add a string s to the BTF string section.
1576 * - > 0 offset into string section, on success;
1579 int btf__add_str(struct btf *btf, const char *s)
1583 if (btf->base_btf) {
1584 off = btf__find_str(btf->base_btf, s);
1589 if (btf_ensure_modifiable(btf))
1590 return libbpf_err(-ENOMEM);
1592 off = strset__add_str(btf->strs_set, s);
1594 return libbpf_err(off);
1596 btf->hdr->str_len = strset__data_size(btf->strs_set);
1598 return btf->start_str_off + off;
1601 static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1603 return libbpf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1604 btf->hdr->type_len, UINT_MAX, add_sz);
1607 static void btf_type_inc_vlen(struct btf_type *t)
1609 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1612 static int btf_commit_type(struct btf *btf, int data_sz)
1616 err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1618 return libbpf_err(err);
1620 btf->hdr->type_len += data_sz;
1621 btf->hdr->str_off += data_sz;
1623 return btf->start_id + btf->nr_types - 1;
1627 const struct btf *src;
1631 static int btf_rewrite_str(__u32 *str_off, void *ctx)
1633 struct btf_pipe *p = ctx;
1636 if (!*str_off) /* nothing to do for empty strings */
1639 off = btf__add_str(p->dst, btf__str_by_offset(p->src, *str_off));
1647 int btf__add_type(struct btf *btf, const struct btf *src_btf, const struct btf_type *src_type)
1649 struct btf_pipe p = { .src = src_btf, .dst = btf };
1653 sz = btf_type_size(src_type);
1655 return libbpf_err(sz);
1657 /* deconstruct BTF, if necessary, and invalidate raw_data */
1658 if (btf_ensure_modifiable(btf))
1659 return libbpf_err(-ENOMEM);
1661 t = btf_add_type_mem(btf, sz);
1663 return libbpf_err(-ENOMEM);
1665 memcpy(t, src_type, sz);
1667 err = btf_type_visit_str_offs(t, btf_rewrite_str, &p);
1669 return libbpf_err(err);
1671 return btf_commit_type(btf, sz);
1675 * Append new BTF_KIND_INT type with:
1676 * - *name* - non-empty, non-NULL type name;
1677 * - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1678 * - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1680 * - >0, type ID of newly added BTF type;
1683 int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
1688 /* non-empty name */
1689 if (!name || !name[0])
1690 return libbpf_err(-EINVAL);
1691 /* byte_sz must be power of 2 */
1692 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
1693 return libbpf_err(-EINVAL);
1694 if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
1695 return libbpf_err(-EINVAL);
1697 /* deconstruct BTF, if necessary, and invalidate raw_data */
1698 if (btf_ensure_modifiable(btf))
1699 return libbpf_err(-ENOMEM);
1701 sz = sizeof(struct btf_type) + sizeof(int);
1702 t = btf_add_type_mem(btf, sz);
1704 return libbpf_err(-ENOMEM);
1706 /* if something goes wrong later, we might end up with an extra string,
1707 * but that shouldn't be a problem, because BTF can't be constructed
1708 * completely anyway and will most probably be just discarded
1710 name_off = btf__add_str(btf, name);
1714 t->name_off = name_off;
1715 t->info = btf_type_info(BTF_KIND_INT, 0, 0);
1717 /* set INT info, we don't allow setting legacy bit offset/size */
1718 *(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
1720 return btf_commit_type(btf, sz);
1724 * Append new BTF_KIND_FLOAT type with:
1725 * - *name* - non-empty, non-NULL type name;
1726 * - *sz* - size of the type, in bytes;
1728 * - >0, type ID of newly added BTF type;
1731 int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
1736 /* non-empty name */
1737 if (!name || !name[0])
1738 return libbpf_err(-EINVAL);
1740 /* byte_sz must be one of the explicitly allowed values */
1741 if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
1743 return libbpf_err(-EINVAL);
1745 if (btf_ensure_modifiable(btf))
1746 return libbpf_err(-ENOMEM);
1748 sz = sizeof(struct btf_type);
1749 t = btf_add_type_mem(btf, sz);
1751 return libbpf_err(-ENOMEM);
1753 name_off = btf__add_str(btf, name);
1757 t->name_off = name_off;
1758 t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
1761 return btf_commit_type(btf, sz);
1764 /* it's completely legal to append BTF types with type IDs pointing forward to
1765 * types that haven't been appended yet, so we only make sure that id looks
1766 * sane, we can't guarantee that ID will always be valid
1768 static int validate_type_id(int id)
1770 if (id < 0 || id > BTF_MAX_NR_TYPES)
1775 /* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
1776 static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
1779 int sz, name_off = 0;
1781 if (validate_type_id(ref_type_id))
1782 return libbpf_err(-EINVAL);
1784 if (btf_ensure_modifiable(btf))
1785 return libbpf_err(-ENOMEM);
1787 sz = sizeof(struct btf_type);
1788 t = btf_add_type_mem(btf, sz);
1790 return libbpf_err(-ENOMEM);
1792 if (name && name[0]) {
1793 name_off = btf__add_str(btf, name);
1798 t->name_off = name_off;
1799 t->info = btf_type_info(kind, 0, 0);
1800 t->type = ref_type_id;
1802 return btf_commit_type(btf, sz);
1806 * Append new BTF_KIND_PTR type with:
1807 * - *ref_type_id* - referenced type ID, it might not exist yet;
1809 * - >0, type ID of newly added BTF type;
1812 int btf__add_ptr(struct btf *btf, int ref_type_id)
1814 return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
1818 * Append new BTF_KIND_ARRAY type with:
1819 * - *index_type_id* - type ID of the type describing array index;
1820 * - *elem_type_id* - type ID of the type describing array element;
1821 * - *nr_elems* - the size of the array;
1823 * - >0, type ID of newly added BTF type;
1826 int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
1829 struct btf_array *a;
1832 if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
1833 return libbpf_err(-EINVAL);
1835 if (btf_ensure_modifiable(btf))
1836 return libbpf_err(-ENOMEM);
1838 sz = sizeof(struct btf_type) + sizeof(struct btf_array);
1839 t = btf_add_type_mem(btf, sz);
1841 return libbpf_err(-ENOMEM);
1844 t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
1848 a->type = elem_type_id;
1849 a->index_type = index_type_id;
1850 a->nelems = nr_elems;
1852 return btf_commit_type(btf, sz);
1855 /* generic STRUCT/UNION append function */
1856 static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
1859 int sz, name_off = 0;
1861 if (btf_ensure_modifiable(btf))
1862 return libbpf_err(-ENOMEM);
1864 sz = sizeof(struct btf_type);
1865 t = btf_add_type_mem(btf, sz);
1867 return libbpf_err(-ENOMEM);
1869 if (name && name[0]) {
1870 name_off = btf__add_str(btf, name);
1875 /* start out with vlen=0 and no kflag; this will be adjusted when
1876 * adding each member
1878 t->name_off = name_off;
1879 t->info = btf_type_info(kind, 0, 0);
1882 return btf_commit_type(btf, sz);
1886 * Append new BTF_KIND_STRUCT type with:
1887 * - *name* - name of the struct, can be NULL or empty for anonymous structs;
1888 * - *byte_sz* - size of the struct, in bytes;
1890 * Struct initially has no fields in it. Fields can be added by
1891 * btf__add_field() right after btf__add_struct() succeeds.
1894 * - >0, type ID of newly added BTF type;
1897 int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
1899 return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
1903 * Append new BTF_KIND_UNION type with:
1904 * - *name* - name of the union, can be NULL or empty for anonymous union;
1905 * - *byte_sz* - size of the union, in bytes;
1907 * Union initially has no fields in it. Fields can be added by
1908 * btf__add_field() right after btf__add_union() succeeds. All fields
1909 * should have *bit_offset* of 0.
1912 * - >0, type ID of newly added BTF type;
1915 int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
1917 return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
1920 static struct btf_type *btf_last_type(struct btf *btf)
1922 return btf_type_by_id(btf, btf__get_nr_types(btf));
1926 * Append new field for the current STRUCT/UNION type with:
1927 * - *name* - name of the field, can be NULL or empty for anonymous field;
1928 * - *type_id* - type ID for the type describing field type;
1929 * - *bit_offset* - bit offset of the start of the field within struct/union;
1930 * - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
1935 int btf__add_field(struct btf *btf, const char *name, int type_id,
1936 __u32 bit_offset, __u32 bit_size)
1939 struct btf_member *m;
1941 int sz, name_off = 0;
1943 /* last type should be union/struct */
1944 if (btf->nr_types == 0)
1945 return libbpf_err(-EINVAL);
1946 t = btf_last_type(btf);
1947 if (!btf_is_composite(t))
1948 return libbpf_err(-EINVAL);
1950 if (validate_type_id(type_id))
1951 return libbpf_err(-EINVAL);
1952 /* best-effort bit field offset/size enforcement */
1953 is_bitfield = bit_size || (bit_offset % 8 != 0);
1954 if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
1955 return libbpf_err(-EINVAL);
1957 /* only offset 0 is allowed for unions */
1958 if (btf_is_union(t) && bit_offset)
1959 return libbpf_err(-EINVAL);
1961 /* decompose and invalidate raw data */
1962 if (btf_ensure_modifiable(btf))
1963 return libbpf_err(-ENOMEM);
1965 sz = sizeof(struct btf_member);
1966 m = btf_add_type_mem(btf, sz);
1968 return libbpf_err(-ENOMEM);
1970 if (name && name[0]) {
1971 name_off = btf__add_str(btf, name);
1976 m->name_off = name_off;
1978 m->offset = bit_offset | (bit_size << 24);
1980 /* btf_add_type_mem can invalidate t pointer */
1981 t = btf_last_type(btf);
1982 /* update parent type's vlen and kflag */
1983 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
1985 btf->hdr->type_len += sz;
1986 btf->hdr->str_off += sz;
1991 * Append new BTF_KIND_ENUM type with:
1992 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
1993 * - *byte_sz* - size of the enum, in bytes.
1995 * Enum initially has no enum values in it (and corresponds to enum forward
1996 * declaration). Enumerator values can be added by btf__add_enum_value()
1997 * immediately after btf__add_enum() succeeds.
2000 * - >0, type ID of newly added BTF type;
2003 int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2006 int sz, name_off = 0;
2008 /* byte_sz must be power of 2 */
2009 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2010 return libbpf_err(-EINVAL);
2012 if (btf_ensure_modifiable(btf))
2013 return libbpf_err(-ENOMEM);
2015 sz = sizeof(struct btf_type);
2016 t = btf_add_type_mem(btf, sz);
2018 return libbpf_err(-ENOMEM);
2020 if (name && name[0]) {
2021 name_off = btf__add_str(btf, name);
2026 /* start out with vlen=0; it will be adjusted when adding enum values */
2027 t->name_off = name_off;
2028 t->info = btf_type_info(BTF_KIND_ENUM, 0, 0);
2031 return btf_commit_type(btf, sz);
2035 * Append new enum value for the current ENUM type with:
2036 * - *name* - name of the enumerator value, can't be NULL or empty;
2037 * - *value* - integer value corresponding to enum value *name*;
2042 int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2048 /* last type should be BTF_KIND_ENUM */
2049 if (btf->nr_types == 0)
2050 return libbpf_err(-EINVAL);
2051 t = btf_last_type(btf);
2052 if (!btf_is_enum(t))
2053 return libbpf_err(-EINVAL);
2055 /* non-empty name */
2056 if (!name || !name[0])
2057 return libbpf_err(-EINVAL);
2058 if (value < INT_MIN || value > UINT_MAX)
2059 return libbpf_err(-E2BIG);
2061 /* decompose and invalidate raw data */
2062 if (btf_ensure_modifiable(btf))
2063 return libbpf_err(-ENOMEM);
2065 sz = sizeof(struct btf_enum);
2066 v = btf_add_type_mem(btf, sz);
2068 return libbpf_err(-ENOMEM);
2070 name_off = btf__add_str(btf, name);
2074 v->name_off = name_off;
2077 /* update parent type's vlen */
2078 t = btf_last_type(btf);
2079 btf_type_inc_vlen(t);
2081 btf->hdr->type_len += sz;
2082 btf->hdr->str_off += sz;
2087 * Append new BTF_KIND_FWD type with:
2088 * - *name*, non-empty/non-NULL name;
2089 * - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2090 * BTF_FWD_UNION, or BTF_FWD_ENUM;
2092 * - >0, type ID of newly added BTF type;
2095 int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2097 if (!name || !name[0])
2098 return libbpf_err(-EINVAL);
2101 case BTF_FWD_STRUCT:
2102 case BTF_FWD_UNION: {
2106 id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
2109 t = btf_type_by_id(btf, id);
2110 t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2114 /* enum forward in BTF currently is just an enum with no enum
2115 * values; we also assume a standard 4-byte size for it
2117 return btf__add_enum(btf, name, sizeof(int));
2119 return libbpf_err(-EINVAL);
2124 * Append new BTF_KING_TYPEDEF type with:
2125 * - *name*, non-empty/non-NULL name;
2126 * - *ref_type_id* - referenced type ID, it might not exist yet;
2128 * - >0, type ID of newly added BTF type;
2131 int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2133 if (!name || !name[0])
2134 return libbpf_err(-EINVAL);
2136 return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2140 * Append new BTF_KIND_VOLATILE type with:
2141 * - *ref_type_id* - referenced type ID, it might not exist yet;
2143 * - >0, type ID of newly added BTF type;
2146 int btf__add_volatile(struct btf *btf, int ref_type_id)
2148 return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2152 * Append new BTF_KIND_CONST type with:
2153 * - *ref_type_id* - referenced type ID, it might not exist yet;
2155 * - >0, type ID of newly added BTF type;
2158 int btf__add_const(struct btf *btf, int ref_type_id)
2160 return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2164 * Append new BTF_KIND_RESTRICT type with:
2165 * - *ref_type_id* - referenced type ID, it might not exist yet;
2167 * - >0, type ID of newly added BTF type;
2170 int btf__add_restrict(struct btf *btf, int ref_type_id)
2172 return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2176 * Append new BTF_KIND_FUNC type with:
2177 * - *name*, non-empty/non-NULL name;
2178 * - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2180 * - >0, type ID of newly added BTF type;
2183 int btf__add_func(struct btf *btf, const char *name,
2184 enum btf_func_linkage linkage, int proto_type_id)
2188 if (!name || !name[0])
2189 return libbpf_err(-EINVAL);
2190 if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2191 linkage != BTF_FUNC_EXTERN)
2192 return libbpf_err(-EINVAL);
2194 id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2196 struct btf_type *t = btf_type_by_id(btf, id);
2198 t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2200 return libbpf_err(id);
2204 * Append new BTF_KIND_FUNC_PROTO with:
2205 * - *ret_type_id* - type ID for return result of a function.
2207 * Function prototype initially has no arguments, but they can be added by
2208 * btf__add_func_param() one by one, immediately after
2209 * btf__add_func_proto() succeeded.
2212 * - >0, type ID of newly added BTF type;
2215 int btf__add_func_proto(struct btf *btf, int ret_type_id)
2220 if (validate_type_id(ret_type_id))
2221 return libbpf_err(-EINVAL);
2223 if (btf_ensure_modifiable(btf))
2224 return libbpf_err(-ENOMEM);
2226 sz = sizeof(struct btf_type);
2227 t = btf_add_type_mem(btf, sz);
2229 return libbpf_err(-ENOMEM);
2231 /* start out with vlen=0; this will be adjusted when adding enum
2232 * values, if necessary
2235 t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2236 t->type = ret_type_id;
2238 return btf_commit_type(btf, sz);
2242 * Append new function parameter for current FUNC_PROTO type with:
2243 * - *name* - parameter name, can be NULL or empty;
2244 * - *type_id* - type ID describing the type of the parameter.
2249 int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2252 struct btf_param *p;
2253 int sz, name_off = 0;
2255 if (validate_type_id(type_id))
2256 return libbpf_err(-EINVAL);
2258 /* last type should be BTF_KIND_FUNC_PROTO */
2259 if (btf->nr_types == 0)
2260 return libbpf_err(-EINVAL);
2261 t = btf_last_type(btf);
2262 if (!btf_is_func_proto(t))
2263 return libbpf_err(-EINVAL);
2265 /* decompose and invalidate raw data */
2266 if (btf_ensure_modifiable(btf))
2267 return libbpf_err(-ENOMEM);
2269 sz = sizeof(struct btf_param);
2270 p = btf_add_type_mem(btf, sz);
2272 return libbpf_err(-ENOMEM);
2274 if (name && name[0]) {
2275 name_off = btf__add_str(btf, name);
2280 p->name_off = name_off;
2283 /* update parent type's vlen */
2284 t = btf_last_type(btf);
2285 btf_type_inc_vlen(t);
2287 btf->hdr->type_len += sz;
2288 btf->hdr->str_off += sz;
2293 * Append new BTF_KIND_VAR type with:
2294 * - *name* - non-empty/non-NULL name;
2295 * - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2296 * BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2297 * - *type_id* - type ID of the type describing the type of the variable.
2299 * - >0, type ID of newly added BTF type;
2302 int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2308 /* non-empty name */
2309 if (!name || !name[0])
2310 return libbpf_err(-EINVAL);
2311 if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2312 linkage != BTF_VAR_GLOBAL_EXTERN)
2313 return libbpf_err(-EINVAL);
2314 if (validate_type_id(type_id))
2315 return libbpf_err(-EINVAL);
2317 /* deconstruct BTF, if necessary, and invalidate raw_data */
2318 if (btf_ensure_modifiable(btf))
2319 return libbpf_err(-ENOMEM);
2321 sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2322 t = btf_add_type_mem(btf, sz);
2324 return libbpf_err(-ENOMEM);
2326 name_off = btf__add_str(btf, name);
2330 t->name_off = name_off;
2331 t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2335 v->linkage = linkage;
2337 return btf_commit_type(btf, sz);
2341 * Append new BTF_KIND_DATASEC type with:
2342 * - *name* - non-empty/non-NULL name;
2343 * - *byte_sz* - data section size, in bytes.
2345 * Data section is initially empty. Variables info can be added with
2346 * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2349 * - >0, type ID of newly added BTF type;
2352 int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2357 /* non-empty name */
2358 if (!name || !name[0])
2359 return libbpf_err(-EINVAL);
2361 if (btf_ensure_modifiable(btf))
2362 return libbpf_err(-ENOMEM);
2364 sz = sizeof(struct btf_type);
2365 t = btf_add_type_mem(btf, sz);
2367 return libbpf_err(-ENOMEM);
2369 name_off = btf__add_str(btf, name);
2373 /* start with vlen=0, which will be update as var_secinfos are added */
2374 t->name_off = name_off;
2375 t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2378 return btf_commit_type(btf, sz);
2382 * Append new data section variable information entry for current DATASEC type:
2383 * - *var_type_id* - type ID, describing type of the variable;
2384 * - *offset* - variable offset within data section, in bytes;
2385 * - *byte_sz* - variable size, in bytes.
2391 int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2394 struct btf_var_secinfo *v;
2397 /* last type should be BTF_KIND_DATASEC */
2398 if (btf->nr_types == 0)
2399 return libbpf_err(-EINVAL);
2400 t = btf_last_type(btf);
2401 if (!btf_is_datasec(t))
2402 return libbpf_err(-EINVAL);
2404 if (validate_type_id(var_type_id))
2405 return libbpf_err(-EINVAL);
2407 /* decompose and invalidate raw data */
2408 if (btf_ensure_modifiable(btf))
2409 return libbpf_err(-ENOMEM);
2411 sz = sizeof(struct btf_var_secinfo);
2412 v = btf_add_type_mem(btf, sz);
2414 return libbpf_err(-ENOMEM);
2416 v->type = var_type_id;
2420 /* update parent type's vlen */
2421 t = btf_last_type(btf);
2422 btf_type_inc_vlen(t);
2424 btf->hdr->type_len += sz;
2425 btf->hdr->str_off += sz;
2429 struct btf_ext_sec_setup_param {
2433 struct btf_ext_info *ext_info;
2437 static int btf_ext_setup_info(struct btf_ext *btf_ext,
2438 struct btf_ext_sec_setup_param *ext_sec)
2440 const struct btf_ext_info_sec *sinfo;
2441 struct btf_ext_info *ext_info;
2442 __u32 info_left, record_size;
2443 /* The start of the info sec (including the __u32 record_size). */
2446 if (ext_sec->len == 0)
2449 if (ext_sec->off & 0x03) {
2450 pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2455 info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2456 info_left = ext_sec->len;
2458 if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2459 pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2460 ext_sec->desc, ext_sec->off, ext_sec->len);
2464 /* At least a record size */
2465 if (info_left < sizeof(__u32)) {
2466 pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2470 /* The record size needs to meet the minimum standard */
2471 record_size = *(__u32 *)info;
2472 if (record_size < ext_sec->min_rec_size ||
2473 record_size & 0x03) {
2474 pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2475 ext_sec->desc, record_size);
2479 sinfo = info + sizeof(__u32);
2480 info_left -= sizeof(__u32);
2482 /* If no records, return failure now so .BTF.ext won't be used. */
2484 pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
2489 unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2490 __u64 total_record_size;
2493 if (info_left < sec_hdrlen) {
2494 pr_debug("%s section header is not found in .BTF.ext\n",
2499 num_records = sinfo->num_info;
2500 if (num_records == 0) {
2501 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2506 total_record_size = sec_hdrlen +
2507 (__u64)num_records * record_size;
2508 if (info_left < total_record_size) {
2509 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2514 info_left -= total_record_size;
2515 sinfo = (void *)sinfo + total_record_size;
2518 ext_info = ext_sec->ext_info;
2519 ext_info->len = ext_sec->len - sizeof(__u32);
2520 ext_info->rec_size = record_size;
2521 ext_info->info = info + sizeof(__u32);
2526 static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
2528 struct btf_ext_sec_setup_param param = {
2529 .off = btf_ext->hdr->func_info_off,
2530 .len = btf_ext->hdr->func_info_len,
2531 .min_rec_size = sizeof(struct bpf_func_info_min),
2532 .ext_info = &btf_ext->func_info,
2536 return btf_ext_setup_info(btf_ext, ¶m);
2539 static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
2541 struct btf_ext_sec_setup_param param = {
2542 .off = btf_ext->hdr->line_info_off,
2543 .len = btf_ext->hdr->line_info_len,
2544 .min_rec_size = sizeof(struct bpf_line_info_min),
2545 .ext_info = &btf_ext->line_info,
2546 .desc = "line_info",
2549 return btf_ext_setup_info(btf_ext, ¶m);
2552 static int btf_ext_setup_core_relos(struct btf_ext *btf_ext)
2554 struct btf_ext_sec_setup_param param = {
2555 .off = btf_ext->hdr->core_relo_off,
2556 .len = btf_ext->hdr->core_relo_len,
2557 .min_rec_size = sizeof(struct bpf_core_relo),
2558 .ext_info = &btf_ext->core_relo_info,
2559 .desc = "core_relo",
2562 return btf_ext_setup_info(btf_ext, ¶m);
2565 static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
2567 const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
2569 if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
2570 data_size < hdr->hdr_len) {
2571 pr_debug("BTF.ext header not found");
2575 if (hdr->magic == bswap_16(BTF_MAGIC)) {
2576 pr_warn("BTF.ext in non-native endianness is not supported\n");
2578 } else if (hdr->magic != BTF_MAGIC) {
2579 pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
2583 if (hdr->version != BTF_VERSION) {
2584 pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
2589 pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
2593 if (data_size == hdr->hdr_len) {
2594 pr_debug("BTF.ext has no data\n");
2601 void btf_ext__free(struct btf_ext *btf_ext)
2603 if (IS_ERR_OR_NULL(btf_ext))
2605 free(btf_ext->data);
2609 struct btf_ext *btf_ext__new(__u8 *data, __u32 size)
2611 struct btf_ext *btf_ext;
2614 err = btf_ext_parse_hdr(data, size);
2616 return libbpf_err_ptr(err);
2618 btf_ext = calloc(1, sizeof(struct btf_ext));
2620 return libbpf_err_ptr(-ENOMEM);
2622 btf_ext->data_size = size;
2623 btf_ext->data = malloc(size);
2624 if (!btf_ext->data) {
2628 memcpy(btf_ext->data, data, size);
2630 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
2635 err = btf_ext_setup_func_info(btf_ext);
2639 err = btf_ext_setup_line_info(btf_ext);
2643 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len)) {
2648 err = btf_ext_setup_core_relos(btf_ext);
2654 btf_ext__free(btf_ext);
2655 return libbpf_err_ptr(err);
2661 const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
2663 *size = btf_ext->data_size;
2664 return btf_ext->data;
2667 static int btf_ext_reloc_info(const struct btf *btf,
2668 const struct btf_ext_info *ext_info,
2669 const char *sec_name, __u32 insns_cnt,
2670 void **info, __u32 *cnt)
2672 __u32 sec_hdrlen = sizeof(struct btf_ext_info_sec);
2673 __u32 i, record_size, existing_len, records_len;
2674 struct btf_ext_info_sec *sinfo;
2675 const char *info_sec_name;
2679 record_size = ext_info->rec_size;
2680 sinfo = ext_info->info;
2681 remain_len = ext_info->len;
2682 while (remain_len > 0) {
2683 records_len = sinfo->num_info * record_size;
2684 info_sec_name = btf__name_by_offset(btf, sinfo->sec_name_off);
2685 if (strcmp(info_sec_name, sec_name)) {
2686 remain_len -= sec_hdrlen + records_len;
2687 sinfo = (void *)sinfo + sec_hdrlen + records_len;
2691 existing_len = (*cnt) * record_size;
2692 data = realloc(*info, existing_len + records_len);
2694 return libbpf_err(-ENOMEM);
2696 memcpy(data + existing_len, sinfo->data, records_len);
2697 /* adjust insn_off only, the rest data will be passed
2700 for (i = 0; i < sinfo->num_info; i++) {
2703 insn_off = data + existing_len + (i * record_size);
2704 *insn_off = *insn_off / sizeof(struct bpf_insn) + insns_cnt;
2707 *cnt += sinfo->num_info;
2711 return libbpf_err(-ENOENT);
2714 int btf_ext__reloc_func_info(const struct btf *btf,
2715 const struct btf_ext *btf_ext,
2716 const char *sec_name, __u32 insns_cnt,
2717 void **func_info, __u32 *cnt)
2719 return btf_ext_reloc_info(btf, &btf_ext->func_info, sec_name,
2720 insns_cnt, func_info, cnt);
2723 int btf_ext__reloc_line_info(const struct btf *btf,
2724 const struct btf_ext *btf_ext,
2725 const char *sec_name, __u32 insns_cnt,
2726 void **line_info, __u32 *cnt)
2728 return btf_ext_reloc_info(btf, &btf_ext->line_info, sec_name,
2729 insns_cnt, line_info, cnt);
2732 __u32 btf_ext__func_info_rec_size(const struct btf_ext *btf_ext)
2734 return btf_ext->func_info.rec_size;
2737 __u32 btf_ext__line_info_rec_size(const struct btf_ext *btf_ext)
2739 return btf_ext->line_info.rec_size;
2744 static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
2745 const struct btf_dedup_opts *opts);
2746 static void btf_dedup_free(struct btf_dedup *d);
2747 static int btf_dedup_prep(struct btf_dedup *d);
2748 static int btf_dedup_strings(struct btf_dedup *d);
2749 static int btf_dedup_prim_types(struct btf_dedup *d);
2750 static int btf_dedup_struct_types(struct btf_dedup *d);
2751 static int btf_dedup_ref_types(struct btf_dedup *d);
2752 static int btf_dedup_compact_types(struct btf_dedup *d);
2753 static int btf_dedup_remap_types(struct btf_dedup *d);
2756 * Deduplicate BTF types and strings.
2758 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
2759 * section with all BTF type descriptors and string data. It overwrites that
2760 * memory in-place with deduplicated types and strings without any loss of
2761 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
2762 * is provided, all the strings referenced from .BTF.ext section are honored
2763 * and updated to point to the right offsets after deduplication.
2765 * If function returns with error, type/string data might be garbled and should
2768 * More verbose and detailed description of both problem btf_dedup is solving,
2769 * as well as solution could be found at:
2770 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
2772 * Problem description and justification
2773 * =====================================
2775 * BTF type information is typically emitted either as a result of conversion
2776 * from DWARF to BTF or directly by compiler. In both cases, each compilation
2777 * unit contains information about a subset of all the types that are used
2778 * in an application. These subsets are frequently overlapping and contain a lot
2779 * of duplicated information when later concatenated together into a single
2780 * binary. This algorithm ensures that each unique type is represented by single
2781 * BTF type descriptor, greatly reducing resulting size of BTF data.
2783 * Compilation unit isolation and subsequent duplication of data is not the only
2784 * problem. The same type hierarchy (e.g., struct and all the type that struct
2785 * references) in different compilation units can be represented in BTF to
2786 * various degrees of completeness (or, rather, incompleteness) due to
2787 * struct/union forward declarations.
2789 * Let's take a look at an example, that we'll use to better understand the
2790 * problem (and solution). Suppose we have two compilation units, each using
2791 * same `struct S`, but each of them having incomplete type information about
2820 * In case of CU #1, BTF data will know only that `struct B` exist (but no
2821 * more), but will know the complete type information about `struct A`. While
2822 * for CU #2, it will know full type information about `struct B`, but will
2823 * only know about forward declaration of `struct A` (in BTF terms, it will
2824 * have `BTF_KIND_FWD` type descriptor with name `B`).
2826 * This compilation unit isolation means that it's possible that there is no
2827 * single CU with complete type information describing structs `S`, `A`, and
2828 * `B`. Also, we might get tons of duplicated and redundant type information.
2830 * Additional complication we need to keep in mind comes from the fact that
2831 * types, in general, can form graphs containing cycles, not just DAGs.
2833 * While algorithm does deduplication, it also merges and resolves type
2834 * information (unless disabled throught `struct btf_opts`), whenever possible.
2835 * E.g., in the example above with two compilation units having partial type
2836 * information for structs `A` and `B`, the output of algorithm will emit
2837 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
2838 * (as well as type information for `int` and pointers), as if they were defined
2839 * in a single compilation unit as:
2859 * Algorithm completes its work in 6 separate passes:
2861 * 1. Strings deduplication.
2862 * 2. Primitive types deduplication (int, enum, fwd).
2863 * 3. Struct/union types deduplication.
2864 * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
2865 * protos, and const/volatile/restrict modifiers).
2866 * 5. Types compaction.
2867 * 6. Types remapping.
2869 * Algorithm determines canonical type descriptor, which is a single
2870 * representative type for each truly unique type. This canonical type is the
2871 * one that will go into final deduplicated BTF type information. For
2872 * struct/unions, it is also the type that algorithm will merge additional type
2873 * information into (while resolving FWDs), as it discovers it from data in
2874 * other CUs. Each input BTF type eventually gets either mapped to itself, if
2875 * that type is canonical, or to some other type, if that type is equivalent
2876 * and was chosen as canonical representative. This mapping is stored in
2877 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
2878 * FWD type got resolved to.
2880 * To facilitate fast discovery of canonical types, we also maintain canonical
2881 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
2882 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
2883 * that match that signature. With sufficiently good choice of type signature
2884 * hashing function, we can limit number of canonical types for each unique type
2885 * signature to a very small number, allowing to find canonical type for any
2886 * duplicated type very quickly.
2888 * Struct/union deduplication is the most critical part and algorithm for
2889 * deduplicating structs/unions is described in greater details in comments for
2890 * `btf_dedup_is_equiv` function.
2892 int btf__dedup(struct btf *btf, struct btf_ext *btf_ext,
2893 const struct btf_dedup_opts *opts)
2895 struct btf_dedup *d = btf_dedup_new(btf, btf_ext, opts);
2899 pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
2900 return libbpf_err(-EINVAL);
2903 if (btf_ensure_modifiable(btf))
2904 return libbpf_err(-ENOMEM);
2906 err = btf_dedup_prep(d);
2908 pr_debug("btf_dedup_prep failed:%d\n", err);
2911 err = btf_dedup_strings(d);
2913 pr_debug("btf_dedup_strings failed:%d\n", err);
2916 err = btf_dedup_prim_types(d);
2918 pr_debug("btf_dedup_prim_types failed:%d\n", err);
2921 err = btf_dedup_struct_types(d);
2923 pr_debug("btf_dedup_struct_types failed:%d\n", err);
2926 err = btf_dedup_ref_types(d);
2928 pr_debug("btf_dedup_ref_types failed:%d\n", err);
2931 err = btf_dedup_compact_types(d);
2933 pr_debug("btf_dedup_compact_types failed:%d\n", err);
2936 err = btf_dedup_remap_types(d);
2938 pr_debug("btf_dedup_remap_types failed:%d\n", err);
2944 return libbpf_err(err);
2947 #define BTF_UNPROCESSED_ID ((__u32)-1)
2948 #define BTF_IN_PROGRESS_ID ((__u32)-2)
2951 /* .BTF section to be deduped in-place */
2954 * Optional .BTF.ext section. When provided, any strings referenced
2955 * from it will be taken into account when deduping strings
2957 struct btf_ext *btf_ext;
2959 * This is a map from any type's signature hash to a list of possible
2960 * canonical representative type candidates. Hash collisions are
2961 * ignored, so even types of various kinds can share same list of
2962 * candidates, which is fine because we rely on subsequent
2963 * btf_xxx_equal() checks to authoritatively verify type equality.
2965 struct hashmap *dedup_table;
2966 /* Canonical types map */
2968 /* Hypothetical mapping, used during type graph equivalence checks */
2973 /* Whether hypothetical mapping, if successful, would need to adjust
2974 * already canonicalized types (due to a new forward declaration to
2975 * concrete type resolution). In such case, during split BTF dedup
2976 * candidate type would still be considered as different, because base
2977 * BTF is considered to be immutable.
2979 bool hypot_adjust_canon;
2980 /* Various option modifying behavior of algorithm */
2981 struct btf_dedup_opts opts;
2982 /* temporary strings deduplication state */
2983 struct strset *strs_set;
2986 static long hash_combine(long h, long value)
2988 return h * 31 + value;
2991 #define for_each_dedup_cand(d, node, hash) \
2992 hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
2994 static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
2996 return hashmap__append(d->dedup_table,
2997 (void *)hash, (void *)(long)type_id);
3000 static int btf_dedup_hypot_map_add(struct btf_dedup *d,
3001 __u32 from_id, __u32 to_id)
3003 if (d->hypot_cnt == d->hypot_cap) {
3006 d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3007 new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3010 d->hypot_list = new_list;
3012 d->hypot_list[d->hypot_cnt++] = from_id;
3013 d->hypot_map[from_id] = to_id;
3017 static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3021 for (i = 0; i < d->hypot_cnt; i++)
3022 d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3024 d->hypot_adjust_canon = false;
3027 static void btf_dedup_free(struct btf_dedup *d)
3029 hashmap__free(d->dedup_table);
3030 d->dedup_table = NULL;
3036 d->hypot_map = NULL;
3038 free(d->hypot_list);
3039 d->hypot_list = NULL;
3044 static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
3049 static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
3054 static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
3059 static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
3060 const struct btf_dedup_opts *opts)
3062 struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3063 hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3064 int i, err = 0, type_cnt;
3067 return ERR_PTR(-ENOMEM);
3069 d->opts.dont_resolve_fwds = opts && opts->dont_resolve_fwds;
3070 /* dedup_table_size is now used only to force collisions in tests */
3071 if (opts && opts->dedup_table_size == 1)
3072 hash_fn = btf_dedup_collision_hash_fn;
3075 d->btf_ext = btf_ext;
3077 d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3078 if (IS_ERR(d->dedup_table)) {
3079 err = PTR_ERR(d->dedup_table);
3080 d->dedup_table = NULL;
3084 type_cnt = btf__get_nr_types(btf) + 1;
3085 d->map = malloc(sizeof(__u32) * type_cnt);
3090 /* special BTF "void" type is made canonical immediately */
3092 for (i = 1; i < type_cnt; i++) {
3093 struct btf_type *t = btf_type_by_id(d->btf, i);
3095 /* VAR and DATASEC are never deduped and are self-canonical */
3096 if (btf_is_var(t) || btf_is_datasec(t))
3099 d->map[i] = BTF_UNPROCESSED_ID;
3102 d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3103 if (!d->hypot_map) {
3107 for (i = 0; i < type_cnt; i++)
3108 d->hypot_map[i] = BTF_UNPROCESSED_ID;
3113 return ERR_PTR(err);
3120 * Iterate over all possible places in .BTF and .BTF.ext that can reference
3121 * string and pass pointer to it to a provided callback `fn`.
3123 static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3127 for (i = 0; i < d->btf->nr_types; i++) {
3128 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3130 r = btf_type_visit_str_offs(t, fn, ctx);
3138 r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3145 static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3147 struct btf_dedup *d = ctx;
3148 __u32 str_off = *str_off_ptr;
3152 /* don't touch empty string or string in main BTF */
3153 if (str_off == 0 || str_off < d->btf->start_str_off)
3156 s = btf__str_by_offset(d->btf, str_off);
3157 if (d->btf->base_btf) {
3158 err = btf__find_str(d->btf->base_btf, s);
3167 off = strset__add_str(d->strs_set, s);
3171 *str_off_ptr = d->btf->start_str_off + off;
3176 * Dedup string and filter out those that are not referenced from either .BTF
3177 * or .BTF.ext (if provided) sections.
3179 * This is done by building index of all strings in BTF's string section,
3180 * then iterating over all entities that can reference strings (e.g., type
3181 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3182 * strings as used. After that all used strings are deduped and compacted into
3183 * sequential blob of memory and new offsets are calculated. Then all the string
3184 * references are iterated again and rewritten using new offsets.
3186 static int btf_dedup_strings(struct btf_dedup *d)
3190 if (d->btf->strs_deduped)
3193 d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3194 if (IS_ERR(d->strs_set)) {
3195 err = PTR_ERR(d->strs_set);
3199 if (!d->btf->base_btf) {
3200 /* insert empty string; we won't be looking it up during strings
3201 * dedup, but it's good to have it for generic BTF string lookups
3203 err = strset__add_str(d->strs_set, "");
3208 /* remap string offsets */
3209 err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3213 /* replace BTF string data and hash with deduped ones */
3214 strset__free(d->btf->strs_set);
3215 d->btf->hdr->str_len = strset__data_size(d->strs_set);
3216 d->btf->strs_set = d->strs_set;
3218 d->btf->strs_deduped = true;
3222 strset__free(d->strs_set);
3228 static long btf_hash_common(struct btf_type *t)
3232 h = hash_combine(0, t->name_off);
3233 h = hash_combine(h, t->info);
3234 h = hash_combine(h, t->size);
3238 static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3240 return t1->name_off == t2->name_off &&
3241 t1->info == t2->info &&
3242 t1->size == t2->size;
3245 /* Calculate type signature hash of INT. */
3246 static long btf_hash_int(struct btf_type *t)
3248 __u32 info = *(__u32 *)(t + 1);
3251 h = btf_hash_common(t);
3252 h = hash_combine(h, info);
3256 /* Check structural equality of two INTs. */
3257 static bool btf_equal_int(struct btf_type *t1, struct btf_type *t2)
3261 if (!btf_equal_common(t1, t2))
3263 info1 = *(__u32 *)(t1 + 1);
3264 info2 = *(__u32 *)(t2 + 1);
3265 return info1 == info2;
3268 /* Calculate type signature hash of ENUM. */
3269 static long btf_hash_enum(struct btf_type *t)
3273 /* don't hash vlen and enum members to support enum fwd resolving */
3274 h = hash_combine(0, t->name_off);
3275 h = hash_combine(h, t->info & ~0xffff);
3276 h = hash_combine(h, t->size);
3280 /* Check structural equality of two ENUMs. */
3281 static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3283 const struct btf_enum *m1, *m2;
3287 if (!btf_equal_common(t1, t2))
3290 vlen = btf_vlen(t1);
3293 for (i = 0; i < vlen; i++) {
3294 if (m1->name_off != m2->name_off || m1->val != m2->val)
3302 static inline bool btf_is_enum_fwd(struct btf_type *t)
3304 return btf_is_enum(t) && btf_vlen(t) == 0;
3307 static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3309 if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3310 return btf_equal_enum(t1, t2);
3311 /* ignore vlen when comparing */
3312 return t1->name_off == t2->name_off &&
3313 (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
3314 t1->size == t2->size;
3318 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3319 * as referenced type IDs equivalence is established separately during type
3320 * graph equivalence check algorithm.
3322 static long btf_hash_struct(struct btf_type *t)
3324 const struct btf_member *member = btf_members(t);
3325 __u32 vlen = btf_vlen(t);
3326 long h = btf_hash_common(t);
3329 for (i = 0; i < vlen; i++) {
3330 h = hash_combine(h, member->name_off);
3331 h = hash_combine(h, member->offset);
3332 /* no hashing of referenced type ID, it can be unresolved yet */
3339 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3340 * IDs. This check is performed during type graph equivalence check and
3341 * referenced types equivalence is checked separately.
3343 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3345 const struct btf_member *m1, *m2;
3349 if (!btf_equal_common(t1, t2))
3352 vlen = btf_vlen(t1);
3353 m1 = btf_members(t1);
3354 m2 = btf_members(t2);
3355 for (i = 0; i < vlen; i++) {
3356 if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3365 * Calculate type signature hash of ARRAY, including referenced type IDs,
3366 * under assumption that they were already resolved to canonical type IDs and
3367 * are not going to change.
3369 static long btf_hash_array(struct btf_type *t)
3371 const struct btf_array *info = btf_array(t);
3372 long h = btf_hash_common(t);
3374 h = hash_combine(h, info->type);
3375 h = hash_combine(h, info->index_type);
3376 h = hash_combine(h, info->nelems);
3381 * Check exact equality of two ARRAYs, taking into account referenced
3382 * type IDs, under assumption that they were already resolved to canonical
3383 * type IDs and are not going to change.
3384 * This function is called during reference types deduplication to compare
3385 * ARRAY to potential canonical representative.
3387 static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3389 const struct btf_array *info1, *info2;
3391 if (!btf_equal_common(t1, t2))
3394 info1 = btf_array(t1);
3395 info2 = btf_array(t2);
3396 return info1->type == info2->type &&
3397 info1->index_type == info2->index_type &&
3398 info1->nelems == info2->nelems;
3402 * Check structural compatibility of two ARRAYs, ignoring referenced type
3403 * IDs. This check is performed during type graph equivalence check and
3404 * referenced types equivalence is checked separately.
3406 static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
3408 if (!btf_equal_common(t1, t2))
3411 return btf_array(t1)->nelems == btf_array(t2)->nelems;
3415 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
3416 * under assumption that they were already resolved to canonical type IDs and
3417 * are not going to change.
3419 static long btf_hash_fnproto(struct btf_type *t)
3421 const struct btf_param *member = btf_params(t);
3422 __u16 vlen = btf_vlen(t);
3423 long h = btf_hash_common(t);
3426 for (i = 0; i < vlen; i++) {
3427 h = hash_combine(h, member->name_off);
3428 h = hash_combine(h, member->type);
3435 * Check exact equality of two FUNC_PROTOs, taking into account referenced
3436 * type IDs, under assumption that they were already resolved to canonical
3437 * type IDs and are not going to change.
3438 * This function is called during reference types deduplication to compare
3439 * FUNC_PROTO to potential canonical representative.
3441 static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
3443 const struct btf_param *m1, *m2;
3447 if (!btf_equal_common(t1, t2))
3450 vlen = btf_vlen(t1);
3451 m1 = btf_params(t1);
3452 m2 = btf_params(t2);
3453 for (i = 0; i < vlen; i++) {
3454 if (m1->name_off != m2->name_off || m1->type != m2->type)
3463 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3464 * IDs. This check is performed during type graph equivalence check and
3465 * referenced types equivalence is checked separately.
3467 static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
3469 const struct btf_param *m1, *m2;
3473 /* skip return type ID */
3474 if (t1->name_off != t2->name_off || t1->info != t2->info)
3477 vlen = btf_vlen(t1);
3478 m1 = btf_params(t1);
3479 m2 = btf_params(t2);
3480 for (i = 0; i < vlen; i++) {
3481 if (m1->name_off != m2->name_off)
3489 /* Prepare split BTF for deduplication by calculating hashes of base BTF's
3490 * types and initializing the rest of the state (canonical type mapping) for
3491 * the fixed base BTF part.
3493 static int btf_dedup_prep(struct btf_dedup *d)
3499 if (!d->btf->base_btf)
3502 for (type_id = 1; type_id < d->btf->start_id; type_id++) {
3503 t = btf_type_by_id(d->btf, type_id);
3505 /* all base BTF types are self-canonical by definition */
3506 d->map[type_id] = type_id;
3508 switch (btf_kind(t)) {
3510 case BTF_KIND_DATASEC:
3511 /* VAR and DATASEC are never hash/deduplicated */
3513 case BTF_KIND_CONST:
3514 case BTF_KIND_VOLATILE:
3515 case BTF_KIND_RESTRICT:
3518 case BTF_KIND_TYPEDEF:
3520 case BTF_KIND_FLOAT:
3521 h = btf_hash_common(t);
3524 h = btf_hash_int(t);
3527 h = btf_hash_enum(t);
3529 case BTF_KIND_STRUCT:
3530 case BTF_KIND_UNION:
3531 h = btf_hash_struct(t);
3533 case BTF_KIND_ARRAY:
3534 h = btf_hash_array(t);
3536 case BTF_KIND_FUNC_PROTO:
3537 h = btf_hash_fnproto(t);
3540 pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
3543 if (btf_dedup_table_add(d, h, type_id))
3551 * Deduplicate primitive types, that can't reference other types, by calculating
3552 * their type signature hash and comparing them with any possible canonical
3553 * candidate. If no canonical candidate matches, type itself is marked as
3554 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
3556 static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
3558 struct btf_type *t = btf_type_by_id(d->btf, type_id);
3559 struct hashmap_entry *hash_entry;
3560 struct btf_type *cand;
3561 /* if we don't find equivalent type, then we are canonical */
3562 __u32 new_id = type_id;
3566 switch (btf_kind(t)) {
3567 case BTF_KIND_CONST:
3568 case BTF_KIND_VOLATILE:
3569 case BTF_KIND_RESTRICT:
3571 case BTF_KIND_TYPEDEF:
3572 case BTF_KIND_ARRAY:
3573 case BTF_KIND_STRUCT:
3574 case BTF_KIND_UNION:
3576 case BTF_KIND_FUNC_PROTO:
3578 case BTF_KIND_DATASEC:
3582 h = btf_hash_int(t);
3583 for_each_dedup_cand(d, hash_entry, h) {
3584 cand_id = (__u32)(long)hash_entry->value;
3585 cand = btf_type_by_id(d->btf, cand_id);
3586 if (btf_equal_int(t, cand)) {
3594 h = btf_hash_enum(t);
3595 for_each_dedup_cand(d, hash_entry, h) {
3596 cand_id = (__u32)(long)hash_entry->value;
3597 cand = btf_type_by_id(d->btf, cand_id);
3598 if (btf_equal_enum(t, cand)) {
3602 if (d->opts.dont_resolve_fwds)
3604 if (btf_compat_enum(t, cand)) {
3605 if (btf_is_enum_fwd(t)) {
3606 /* resolve fwd to full enum */
3610 /* resolve canonical enum fwd to full enum */
3611 d->map[cand_id] = type_id;
3617 case BTF_KIND_FLOAT:
3618 h = btf_hash_common(t);
3619 for_each_dedup_cand(d, hash_entry, h) {
3620 cand_id = (__u32)(long)hash_entry->value;
3621 cand = btf_type_by_id(d->btf, cand_id);
3622 if (btf_equal_common(t, cand)) {
3633 d->map[type_id] = new_id;
3634 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
3640 static int btf_dedup_prim_types(struct btf_dedup *d)
3644 for (i = 0; i < d->btf->nr_types; i++) {
3645 err = btf_dedup_prim_type(d, d->btf->start_id + i);
3653 * Check whether type is already mapped into canonical one (could be to itself).
3655 static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
3657 return d->map[type_id] <= BTF_MAX_NR_TYPES;
3661 * Resolve type ID into its canonical type ID, if any; otherwise return original
3662 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
3663 * STRUCT/UNION link and resolve it into canonical type ID as well.
3665 static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
3667 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3668 type_id = d->map[type_id];
3673 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
3676 static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
3678 __u32 orig_type_id = type_id;
3680 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3683 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3684 type_id = d->map[type_id];
3686 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3689 return orig_type_id;
3693 static inline __u16 btf_fwd_kind(struct btf_type *t)
3695 return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
3698 /* Check if given two types are identical ARRAY definitions */
3699 static int btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
3701 struct btf_type *t1, *t2;
3703 t1 = btf_type_by_id(d->btf, id1);
3704 t2 = btf_type_by_id(d->btf, id2);
3705 if (!btf_is_array(t1) || !btf_is_array(t2))
3708 return btf_equal_array(t1, t2);
3712 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
3713 * call it "candidate graph" in this description for brevity) to a type graph
3714 * formed by (potential) canonical struct/union ("canonical graph" for brevity
3715 * here, though keep in mind that not all types in canonical graph are
3716 * necessarily canonical representatives themselves, some of them might be
3717 * duplicates or its uniqueness might not have been established yet).
3719 * - >0, if type graphs are equivalent;
3720 * - 0, if not equivalent;
3723 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
3724 * equivalence of BTF types at each step. If at any point BTF types in candidate
3725 * and canonical graphs are not compatible structurally, whole graphs are
3726 * incompatible. If types are structurally equivalent (i.e., all information
3727 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
3728 * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
3729 * If a type references other types, then those referenced types are checked
3730 * for equivalence recursively.
3732 * During DFS traversal, if we find that for current `canon_id` type we
3733 * already have some mapping in hypothetical map, we check for two possible
3735 * - `canon_id` is mapped to exactly the same type as `cand_id`. This will
3736 * happen when type graphs have cycles. In this case we assume those two
3737 * types are equivalent.
3738 * - `canon_id` is mapped to different type. This is contradiction in our
3739 * hypothetical mapping, because same graph in canonical graph corresponds
3740 * to two different types in candidate graph, which for equivalent type
3741 * graphs shouldn't happen. This condition terminates equivalence check
3742 * with negative result.
3744 * If type graphs traversal exhausts types to check and find no contradiction,
3745 * then type graphs are equivalent.
3747 * When checking types for equivalence, there is one special case: FWD types.
3748 * If FWD type resolution is allowed and one of the types (either from canonical
3749 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
3750 * flag) and their names match, hypothetical mapping is updated to point from
3751 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
3752 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
3754 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
3755 * if there are two exactly named (or anonymous) structs/unions that are
3756 * compatible structurally, one of which has FWD field, while other is concrete
3757 * STRUCT/UNION, but according to C sources they are different structs/unions
3758 * that are referencing different types with the same name. This is extremely
3759 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
3760 * this logic is causing problems.
3762 * Doing FWD resolution means that both candidate and/or canonical graphs can
3763 * consists of portions of the graph that come from multiple compilation units.
3764 * This is due to the fact that types within single compilation unit are always
3765 * deduplicated and FWDs are already resolved, if referenced struct/union
3766 * definiton is available. So, if we had unresolved FWD and found corresponding
3767 * STRUCT/UNION, they will be from different compilation units. This
3768 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
3769 * type graph will likely have at least two different BTF types that describe
3770 * same type (e.g., most probably there will be two different BTF types for the
3771 * same 'int' primitive type) and could even have "overlapping" parts of type
3772 * graph that describe same subset of types.
3774 * This in turn means that our assumption that each type in canonical graph
3775 * must correspond to exactly one type in candidate graph might not hold
3776 * anymore and will make it harder to detect contradictions using hypothetical
3777 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
3778 * resolution only in canonical graph. FWDs in candidate graphs are never
3779 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
3781 * - Both types in canonical and candidate graphs are FWDs. If they are
3782 * structurally equivalent, then they can either be both resolved to the
3783 * same STRUCT/UNION or not resolved at all. In both cases they are
3784 * equivalent and there is no need to resolve FWD on candidate side.
3785 * - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
3786 * so nothing to resolve as well, algorithm will check equivalence anyway.
3787 * - Type in canonical graph is FWD, while type in candidate is concrete
3788 * STRUCT/UNION. In this case candidate graph comes from single compilation
3789 * unit, so there is exactly one BTF type for each unique C type. After
3790 * resolving FWD into STRUCT/UNION, there might be more than one BTF type
3791 * in canonical graph mapping to single BTF type in candidate graph, but
3792 * because hypothetical mapping maps from canonical to candidate types, it's
3793 * alright, and we still maintain the property of having single `canon_id`
3794 * mapping to single `cand_id` (there could be two different `canon_id`
3795 * mapped to the same `cand_id`, but it's not contradictory).
3796 * - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
3797 * graph is FWD. In this case we are just going to check compatibility of
3798 * STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
3799 * assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
3800 * a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
3801 * turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
3804 static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
3807 struct btf_type *cand_type;
3808 struct btf_type *canon_type;
3809 __u32 hypot_type_id;
3814 /* if both resolve to the same canonical, they must be equivalent */
3815 if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
3818 canon_id = resolve_fwd_id(d, canon_id);
3820 hypot_type_id = d->hypot_map[canon_id];
3821 if (hypot_type_id <= BTF_MAX_NR_TYPES) {
3822 /* In some cases compiler will generate different DWARF types
3823 * for *identical* array type definitions and use them for
3824 * different fields within the *same* struct. This breaks type
3825 * equivalence check, which makes an assumption that candidate
3826 * types sub-graph has a consistent and deduped-by-compiler
3827 * types within a single CU. So work around that by explicitly
3828 * allowing identical array types here.
3830 return hypot_type_id == cand_id ||
3831 btf_dedup_identical_arrays(d, hypot_type_id, cand_id);
3834 if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
3837 cand_type = btf_type_by_id(d->btf, cand_id);
3838 canon_type = btf_type_by_id(d->btf, canon_id);
3839 cand_kind = btf_kind(cand_type);
3840 canon_kind = btf_kind(canon_type);
3842 if (cand_type->name_off != canon_type->name_off)
3845 /* FWD <--> STRUCT/UNION equivalence check, if enabled */
3846 if (!d->opts.dont_resolve_fwds
3847 && (cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
3848 && cand_kind != canon_kind) {
3852 if (cand_kind == BTF_KIND_FWD) {
3853 real_kind = canon_kind;
3854 fwd_kind = btf_fwd_kind(cand_type);
3856 real_kind = cand_kind;
3857 fwd_kind = btf_fwd_kind(canon_type);
3858 /* we'd need to resolve base FWD to STRUCT/UNION */
3859 if (fwd_kind == real_kind && canon_id < d->btf->start_id)
3860 d->hypot_adjust_canon = true;
3862 return fwd_kind == real_kind;
3865 if (cand_kind != canon_kind)
3868 switch (cand_kind) {
3870 return btf_equal_int(cand_type, canon_type);
3873 if (d->opts.dont_resolve_fwds)
3874 return btf_equal_enum(cand_type, canon_type);
3876 return btf_compat_enum(cand_type, canon_type);
3879 case BTF_KIND_FLOAT:
3880 return btf_equal_common(cand_type, canon_type);
3882 case BTF_KIND_CONST:
3883 case BTF_KIND_VOLATILE:
3884 case BTF_KIND_RESTRICT:
3886 case BTF_KIND_TYPEDEF:
3888 if (cand_type->info != canon_type->info)
3890 return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
3892 case BTF_KIND_ARRAY: {
3893 const struct btf_array *cand_arr, *canon_arr;
3895 if (!btf_compat_array(cand_type, canon_type))
3897 cand_arr = btf_array(cand_type);
3898 canon_arr = btf_array(canon_type);
3899 eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
3902 return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
3905 case BTF_KIND_STRUCT:
3906 case BTF_KIND_UNION: {
3907 const struct btf_member *cand_m, *canon_m;
3910 if (!btf_shallow_equal_struct(cand_type, canon_type))
3912 vlen = btf_vlen(cand_type);
3913 cand_m = btf_members(cand_type);
3914 canon_m = btf_members(canon_type);
3915 for (i = 0; i < vlen; i++) {
3916 eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
3926 case BTF_KIND_FUNC_PROTO: {
3927 const struct btf_param *cand_p, *canon_p;
3930 if (!btf_compat_fnproto(cand_type, canon_type))
3932 eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
3935 vlen = btf_vlen(cand_type);
3936 cand_p = btf_params(cand_type);
3937 canon_p = btf_params(canon_type);
3938 for (i = 0; i < vlen; i++) {
3939 eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
3955 * Use hypothetical mapping, produced by successful type graph equivalence
3956 * check, to augment existing struct/union canonical mapping, where possible.
3958 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
3959 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
3960 * it doesn't matter if FWD type was part of canonical graph or candidate one,
3961 * we are recording the mapping anyway. As opposed to carefulness required
3962 * for struct/union correspondence mapping (described below), for FWD resolution
3963 * it's not important, as by the time that FWD type (reference type) will be
3964 * deduplicated all structs/unions will be deduped already anyway.
3966 * Recording STRUCT/UNION mapping is purely a performance optimization and is
3967 * not required for correctness. It needs to be done carefully to ensure that
3968 * struct/union from candidate's type graph is not mapped into corresponding
3969 * struct/union from canonical type graph that itself hasn't been resolved into
3970 * canonical representative. The only guarantee we have is that canonical
3971 * struct/union was determined as canonical and that won't change. But any
3972 * types referenced through that struct/union fields could have been not yet
3973 * resolved, so in case like that it's too early to establish any kind of
3974 * correspondence between structs/unions.
3976 * No canonical correspondence is derived for primitive types (they are already
3977 * deduplicated completely already anyway) or reference types (they rely on
3978 * stability of struct/union canonical relationship for equivalence checks).
3980 static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
3982 __u32 canon_type_id, targ_type_id;
3983 __u16 t_kind, c_kind;
3987 for (i = 0; i < d->hypot_cnt; i++) {
3988 canon_type_id = d->hypot_list[i];
3989 targ_type_id = d->hypot_map[canon_type_id];
3990 t_id = resolve_type_id(d, targ_type_id);
3991 c_id = resolve_type_id(d, canon_type_id);
3992 t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
3993 c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
3995 * Resolve FWD into STRUCT/UNION.
3996 * It's ok to resolve FWD into STRUCT/UNION that's not yet
3997 * mapped to canonical representative (as opposed to
3998 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
3999 * eventually that struct is going to be mapped and all resolved
4000 * FWDs will automatically resolve to correct canonical
4001 * representative. This will happen before ref type deduping,
4002 * which critically depends on stability of these mapping. This
4003 * stability is not a requirement for STRUCT/UNION equivalence
4007 /* if it's the split BTF case, we still need to point base FWD
4008 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4009 * will be resolved against base FWD. If we don't point base
4010 * canonical FWD to the resolved STRUCT/UNION, then all the
4011 * FWDs in split BTF won't be correctly resolved to a proper
4014 if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4015 d->map[c_id] = t_id;
4017 /* if graph equivalence determined that we'd need to adjust
4018 * base canonical types, then we need to only point base FWDs
4019 * to STRUCTs/UNIONs and do no more modifications. For all
4020 * other purposes the type graphs were not equivalent.
4022 if (d->hypot_adjust_canon)
4025 if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4026 d->map[t_id] = c_id;
4028 if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4029 c_kind != BTF_KIND_FWD &&
4030 is_type_mapped(d, c_id) &&
4031 !is_type_mapped(d, t_id)) {
4033 * as a perf optimization, we can map struct/union
4034 * that's part of type graph we just verified for
4035 * equivalence. We can do that for struct/union that has
4036 * canonical representative only, though.
4038 d->map[t_id] = c_id;
4044 * Deduplicate struct/union types.
4046 * For each struct/union type its type signature hash is calculated, taking
4047 * into account type's name, size, number, order and names of fields, but
4048 * ignoring type ID's referenced from fields, because they might not be deduped
4049 * completely until after reference types deduplication phase. This type hash
4050 * is used to iterate over all potential canonical types, sharing same hash.
4051 * For each canonical candidate we check whether type graphs that they form
4052 * (through referenced types in fields and so on) are equivalent using algorithm
4053 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4054 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4055 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4056 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4057 * potentially map other structs/unions to their canonical representatives,
4058 * if such relationship hasn't yet been established. This speeds up algorithm
4059 * by eliminating some of the duplicate work.
4061 * If no matching canonical representative was found, struct/union is marked
4062 * as canonical for itself and is added into btf_dedup->dedup_table hash map
4063 * for further look ups.
4065 static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4067 struct btf_type *cand_type, *t;
4068 struct hashmap_entry *hash_entry;
4069 /* if we don't find equivalent type, then we are canonical */
4070 __u32 new_id = type_id;
4074 /* already deduped or is in process of deduping (loop detected) */
4075 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4078 t = btf_type_by_id(d->btf, type_id);
4081 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4084 h = btf_hash_struct(t);
4085 for_each_dedup_cand(d, hash_entry, h) {
4086 __u32 cand_id = (__u32)(long)hash_entry->value;
4090 * Even though btf_dedup_is_equiv() checks for
4091 * btf_shallow_equal_struct() internally when checking two
4092 * structs (unions) for equivalence, we need to guard here
4093 * from picking matching FWD type as a dedup candidate.
4094 * This can happen due to hash collision. In such case just
4095 * relying on btf_dedup_is_equiv() would lead to potentially
4096 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4097 * FWD and compatible STRUCT/UNION are considered equivalent.
4099 cand_type = btf_type_by_id(d->btf, cand_id);
4100 if (!btf_shallow_equal_struct(t, cand_type))
4103 btf_dedup_clear_hypot_map(d);
4104 eq = btf_dedup_is_equiv(d, type_id, cand_id);
4109 btf_dedup_merge_hypot_map(d);
4110 if (d->hypot_adjust_canon) /* not really equivalent */
4116 d->map[type_id] = new_id;
4117 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4123 static int btf_dedup_struct_types(struct btf_dedup *d)
4127 for (i = 0; i < d->btf->nr_types; i++) {
4128 err = btf_dedup_struct_type(d, d->btf->start_id + i);
4136 * Deduplicate reference type.
4138 * Once all primitive and struct/union types got deduplicated, we can easily
4139 * deduplicate all other (reference) BTF types. This is done in two steps:
4141 * 1. Resolve all referenced type IDs into their canonical type IDs. This
4142 * resolution can be done either immediately for primitive or struct/union types
4143 * (because they were deduped in previous two phases) or recursively for
4144 * reference types. Recursion will always terminate at either primitive or
4145 * struct/union type, at which point we can "unwind" chain of reference types
4146 * one by one. There is no danger of encountering cycles because in C type
4147 * system the only way to form type cycle is through struct/union, so any chain
4148 * of reference types, even those taking part in a type cycle, will inevitably
4149 * reach struct/union at some point.
4151 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4152 * becomes "stable", in the sense that no further deduplication will cause
4153 * any changes to it. With that, it's now possible to calculate type's signature
4154 * hash (this time taking into account referenced type IDs) and loop over all
4155 * potential canonical representatives. If no match was found, current type
4156 * will become canonical representative of itself and will be added into
4157 * btf_dedup->dedup_table as another possible canonical representative.
4159 static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4161 struct hashmap_entry *hash_entry;
4162 __u32 new_id = type_id, cand_id;
4163 struct btf_type *t, *cand;
4164 /* if we don't find equivalent type, then we are representative type */
4168 if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4170 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4171 return resolve_type_id(d, type_id);
4173 t = btf_type_by_id(d->btf, type_id);
4174 d->map[type_id] = BTF_IN_PROGRESS_ID;
4176 switch (btf_kind(t)) {
4177 case BTF_KIND_CONST:
4178 case BTF_KIND_VOLATILE:
4179 case BTF_KIND_RESTRICT:
4181 case BTF_KIND_TYPEDEF:
4183 ref_type_id = btf_dedup_ref_type(d, t->type);
4184 if (ref_type_id < 0)
4186 t->type = ref_type_id;
4188 h = btf_hash_common(t);
4189 for_each_dedup_cand(d, hash_entry, h) {
4190 cand_id = (__u32)(long)hash_entry->value;
4191 cand = btf_type_by_id(d->btf, cand_id);
4192 if (btf_equal_common(t, cand)) {
4199 case BTF_KIND_ARRAY: {
4200 struct btf_array *info = btf_array(t);
4202 ref_type_id = btf_dedup_ref_type(d, info->type);
4203 if (ref_type_id < 0)
4205 info->type = ref_type_id;
4207 ref_type_id = btf_dedup_ref_type(d, info->index_type);
4208 if (ref_type_id < 0)
4210 info->index_type = ref_type_id;
4212 h = btf_hash_array(t);
4213 for_each_dedup_cand(d, hash_entry, h) {
4214 cand_id = (__u32)(long)hash_entry->value;
4215 cand = btf_type_by_id(d->btf, cand_id);
4216 if (btf_equal_array(t, cand)) {
4224 case BTF_KIND_FUNC_PROTO: {
4225 struct btf_param *param;
4229 ref_type_id = btf_dedup_ref_type(d, t->type);
4230 if (ref_type_id < 0)
4232 t->type = ref_type_id;
4235 param = btf_params(t);
4236 for (i = 0; i < vlen; i++) {
4237 ref_type_id = btf_dedup_ref_type(d, param->type);
4238 if (ref_type_id < 0)
4240 param->type = ref_type_id;
4244 h = btf_hash_fnproto(t);
4245 for_each_dedup_cand(d, hash_entry, h) {
4246 cand_id = (__u32)(long)hash_entry->value;
4247 cand = btf_type_by_id(d->btf, cand_id);
4248 if (btf_equal_fnproto(t, cand)) {
4260 d->map[type_id] = new_id;
4261 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4267 static int btf_dedup_ref_types(struct btf_dedup *d)
4271 for (i = 0; i < d->btf->nr_types; i++) {
4272 err = btf_dedup_ref_type(d, d->btf->start_id + i);
4276 /* we won't need d->dedup_table anymore */
4277 hashmap__free(d->dedup_table);
4278 d->dedup_table = NULL;
4285 * After we established for each type its corresponding canonical representative
4286 * type, we now can eliminate types that are not canonical and leave only
4287 * canonical ones layed out sequentially in memory by copying them over
4288 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
4289 * a map from original type ID to a new compacted type ID, which will be used
4290 * during next phase to "fix up" type IDs, referenced from struct/union and
4293 static int btf_dedup_compact_types(struct btf_dedup *d)
4296 __u32 next_type_id = d->btf->start_id;
4297 const struct btf_type *t;
4301 /* we are going to reuse hypot_map to store compaction remapping */
4302 d->hypot_map[0] = 0;
4303 /* base BTF types are not renumbered */
4304 for (id = 1; id < d->btf->start_id; id++)
4305 d->hypot_map[id] = id;
4306 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
4307 d->hypot_map[id] = BTF_UNPROCESSED_ID;
4309 p = d->btf->types_data;
4311 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
4312 if (d->map[id] != id)
4315 t = btf__type_by_id(d->btf, id);
4316 len = btf_type_size(t);
4321 d->hypot_map[id] = next_type_id;
4322 d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
4327 /* shrink struct btf's internal types index and update btf_header */
4328 d->btf->nr_types = next_type_id - d->btf->start_id;
4329 d->btf->type_offs_cap = d->btf->nr_types;
4330 d->btf->hdr->type_len = p - d->btf->types_data;
4331 new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
4333 if (d->btf->type_offs_cap && !new_offs)
4335 d->btf->type_offs = new_offs;
4336 d->btf->hdr->str_off = d->btf->hdr->type_len;
4337 d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
4342 * Figure out final (deduplicated and compacted) type ID for provided original
4343 * `type_id` by first resolving it into corresponding canonical type ID and
4344 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
4345 * which is populated during compaction phase.
4347 static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
4349 struct btf_dedup *d = ctx;
4350 __u32 resolved_type_id, new_type_id;
4352 resolved_type_id = resolve_type_id(d, *type_id);
4353 new_type_id = d->hypot_map[resolved_type_id];
4354 if (new_type_id > BTF_MAX_NR_TYPES)
4357 *type_id = new_type_id;
4362 * Remap referenced type IDs into deduped type IDs.
4364 * After BTF types are deduplicated and compacted, their final type IDs may
4365 * differ from original ones. The map from original to a corresponding
4366 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
4367 * compaction phase. During remapping phase we are rewriting all type IDs
4368 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
4369 * their final deduped type IDs.
4371 static int btf_dedup_remap_types(struct btf_dedup *d)
4375 for (i = 0; i < d->btf->nr_types; i++) {
4376 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
4378 r = btf_type_visit_type_ids(t, btf_dedup_remap_type_id, d);
4386 r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
4394 * Probe few well-known locations for vmlinux kernel image and try to load BTF
4395 * data out of it to use for target BTF.
4397 struct btf *libbpf_find_kernel_btf(void)
4400 const char *path_fmt;
4403 /* try canonical vmlinux BTF through sysfs first */
4404 { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
4405 /* fall back to trying to find vmlinux ELF on disk otherwise */
4406 { "/boot/vmlinux-%1$s" },
4407 { "/lib/modules/%1$s/vmlinux-%1$s" },
4408 { "/lib/modules/%1$s/build/vmlinux" },
4409 { "/usr/lib/modules/%1$s/kernel/vmlinux" },
4410 { "/usr/lib/debug/boot/vmlinux-%1$s" },
4411 { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
4412 { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
4414 char path[PATH_MAX + 1];
4421 for (i = 0; i < ARRAY_SIZE(locations); i++) {
4422 snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
4424 if (access(path, R_OK))
4427 if (locations[i].raw_btf)
4428 btf = btf__parse_raw(path);
4430 btf = btf__parse_elf(path, NULL);
4431 err = libbpf_get_error(btf);
4432 pr_debug("loading kernel BTF '%s': %d\n", path, err);
4439 pr_warn("failed to find valid kernel BTF\n");
4440 return libbpf_err_ptr(-ESRCH);
4443 int btf_type_visit_type_ids(struct btf_type *t, type_id_visit_fn visit, void *ctx)
4447 switch (btf_kind(t)) {
4449 case BTF_KIND_FLOAT:
4454 case BTF_KIND_CONST:
4455 case BTF_KIND_VOLATILE:
4456 case BTF_KIND_RESTRICT:
4458 case BTF_KIND_TYPEDEF:
4461 return visit(&t->type, ctx);
4463 case BTF_KIND_ARRAY: {
4464 struct btf_array *a = btf_array(t);
4466 err = visit(&a->type, ctx);
4467 err = err ?: visit(&a->index_type, ctx);
4471 case BTF_KIND_STRUCT:
4472 case BTF_KIND_UNION: {
4473 struct btf_member *m = btf_members(t);
4475 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4476 err = visit(&m->type, ctx);
4483 case BTF_KIND_FUNC_PROTO: {
4484 struct btf_param *m = btf_params(t);
4486 err = visit(&t->type, ctx);
4489 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4490 err = visit(&m->type, ctx);
4497 case BTF_KIND_DATASEC: {
4498 struct btf_var_secinfo *m = btf_var_secinfos(t);
4500 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4501 err = visit(&m->type, ctx);
4513 int btf_type_visit_str_offs(struct btf_type *t, str_off_visit_fn visit, void *ctx)
4517 err = visit(&t->name_off, ctx);
4521 switch (btf_kind(t)) {
4522 case BTF_KIND_STRUCT:
4523 case BTF_KIND_UNION: {
4524 struct btf_member *m = btf_members(t);
4526 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4527 err = visit(&m->name_off, ctx);
4533 case BTF_KIND_ENUM: {
4534 struct btf_enum *m = btf_enum(t);
4536 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4537 err = visit(&m->name_off, ctx);
4543 case BTF_KIND_FUNC_PROTO: {
4544 struct btf_param *m = btf_params(t);
4546 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4547 err = visit(&m->name_off, ctx);
4560 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
4562 const struct btf_ext_info *seg;
4563 struct btf_ext_info_sec *sec;
4566 seg = &btf_ext->func_info;
4567 for_each_btf_ext_sec(seg, sec) {
4568 struct bpf_func_info_min *rec;
4570 for_each_btf_ext_rec(seg, sec, i, rec) {
4571 err = visit(&rec->type_id, ctx);
4577 seg = &btf_ext->core_relo_info;
4578 for_each_btf_ext_sec(seg, sec) {
4579 struct bpf_core_relo *rec;
4581 for_each_btf_ext_rec(seg, sec, i, rec) {
4582 err = visit(&rec->type_id, ctx);
4591 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
4593 const struct btf_ext_info *seg;
4594 struct btf_ext_info_sec *sec;
4597 seg = &btf_ext->func_info;
4598 for_each_btf_ext_sec(seg, sec) {
4599 err = visit(&sec->sec_name_off, ctx);
4604 seg = &btf_ext->line_info;
4605 for_each_btf_ext_sec(seg, sec) {
4606 struct bpf_line_info_min *rec;
4608 err = visit(&sec->sec_name_off, ctx);
4612 for_each_btf_ext_rec(seg, sec, i, rec) {
4613 err = visit(&rec->file_name_off, ctx);
4616 err = visit(&rec->line_off, ctx);
4622 seg = &btf_ext->core_relo_info;
4623 for_each_btf_ext_sec(seg, sec) {
4624 struct bpf_core_relo *rec;
4626 err = visit(&sec->sec_name_off, ctx);
4630 for_each_btf_ext_rec(seg, sec, i, rec) {
4631 err = visit(&rec->access_str_off, ctx);