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__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 accommodate *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 void *btf_add_type_offs_mem(struct btf *btf, size_t add_cnt)
194 return libbpf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
195 btf->nr_types, BTF_MAX_NR_TYPES, add_cnt);
198 static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
202 p = btf_add_type_offs_mem(btf, 1);
210 static void btf_bswap_hdr(struct btf_header *h)
212 h->magic = bswap_16(h->magic);
213 h->hdr_len = bswap_32(h->hdr_len);
214 h->type_off = bswap_32(h->type_off);
215 h->type_len = bswap_32(h->type_len);
216 h->str_off = bswap_32(h->str_off);
217 h->str_len = bswap_32(h->str_len);
220 static int btf_parse_hdr(struct btf *btf)
222 struct btf_header *hdr = btf->hdr;
225 if (btf->raw_size < sizeof(struct btf_header)) {
226 pr_debug("BTF header not found\n");
230 if (hdr->magic == bswap_16(BTF_MAGIC)) {
231 btf->swapped_endian = true;
232 if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
233 pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
234 bswap_32(hdr->hdr_len));
238 } else if (hdr->magic != BTF_MAGIC) {
239 pr_debug("Invalid BTF magic: %x\n", hdr->magic);
243 if (btf->raw_size < hdr->hdr_len) {
244 pr_debug("BTF header len %u larger than data size %u\n",
245 hdr->hdr_len, btf->raw_size);
249 meta_left = btf->raw_size - hdr->hdr_len;
250 if (meta_left < (long long)hdr->str_off + hdr->str_len) {
251 pr_debug("Invalid BTF total size: %u\n", btf->raw_size);
255 if ((long long)hdr->type_off + hdr->type_len > hdr->str_off) {
256 pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
257 hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
261 if (hdr->type_off % 4) {
262 pr_debug("BTF type section is not aligned to 4 bytes\n");
269 static int btf_parse_str_sec(struct btf *btf)
271 const struct btf_header *hdr = btf->hdr;
272 const char *start = btf->strs_data;
273 const char *end = start + btf->hdr->str_len;
275 if (btf->base_btf && hdr->str_len == 0)
277 if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET || end[-1]) {
278 pr_debug("Invalid BTF string section\n");
281 if (!btf->base_btf && start[0]) {
282 pr_debug("Invalid BTF string section\n");
288 static int btf_type_size(const struct btf_type *t)
290 const int base_size = sizeof(struct btf_type);
291 __u16 vlen = btf_vlen(t);
293 switch (btf_kind(t)) {
296 case BTF_KIND_VOLATILE:
297 case BTF_KIND_RESTRICT:
299 case BTF_KIND_TYPEDEF:
302 case BTF_KIND_TYPE_TAG:
305 return base_size + sizeof(__u32);
307 return base_size + vlen * sizeof(struct btf_enum);
308 case BTF_KIND_ENUM64:
309 return base_size + vlen * sizeof(struct btf_enum64);
311 return base_size + sizeof(struct btf_array);
312 case BTF_KIND_STRUCT:
314 return base_size + vlen * sizeof(struct btf_member);
315 case BTF_KIND_FUNC_PROTO:
316 return base_size + vlen * sizeof(struct btf_param);
318 return base_size + sizeof(struct btf_var);
319 case BTF_KIND_DATASEC:
320 return base_size + vlen * sizeof(struct btf_var_secinfo);
321 case BTF_KIND_DECL_TAG:
322 return base_size + sizeof(struct btf_decl_tag);
324 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
329 static void btf_bswap_type_base(struct btf_type *t)
331 t->name_off = bswap_32(t->name_off);
332 t->info = bswap_32(t->info);
333 t->type = bswap_32(t->type);
336 static int btf_bswap_type_rest(struct btf_type *t)
338 struct btf_var_secinfo *v;
339 struct btf_enum64 *e64;
340 struct btf_member *m;
344 __u16 vlen = btf_vlen(t);
347 switch (btf_kind(t)) {
350 case BTF_KIND_VOLATILE:
351 case BTF_KIND_RESTRICT:
353 case BTF_KIND_TYPEDEF:
356 case BTF_KIND_TYPE_TAG:
359 *(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
362 for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
363 e->name_off = bswap_32(e->name_off);
364 e->val = bswap_32(e->val);
367 case BTF_KIND_ENUM64:
368 for (i = 0, e64 = btf_enum64(t); i < vlen; i++, e64++) {
369 e64->name_off = bswap_32(e64->name_off);
370 e64->val_lo32 = bswap_32(e64->val_lo32);
371 e64->val_hi32 = bswap_32(e64->val_hi32);
376 a->type = bswap_32(a->type);
377 a->index_type = bswap_32(a->index_type);
378 a->nelems = bswap_32(a->nelems);
380 case BTF_KIND_STRUCT:
382 for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
383 m->name_off = bswap_32(m->name_off);
384 m->type = bswap_32(m->type);
385 m->offset = bswap_32(m->offset);
388 case BTF_KIND_FUNC_PROTO:
389 for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
390 p->name_off = bswap_32(p->name_off);
391 p->type = bswap_32(p->type);
395 btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
397 case BTF_KIND_DATASEC:
398 for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
399 v->type = bswap_32(v->type);
400 v->offset = bswap_32(v->offset);
401 v->size = bswap_32(v->size);
404 case BTF_KIND_DECL_TAG:
405 btf_decl_tag(t)->component_idx = bswap_32(btf_decl_tag(t)->component_idx);
408 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
413 static int btf_parse_type_sec(struct btf *btf)
415 struct btf_header *hdr = btf->hdr;
416 void *next_type = btf->types_data;
417 void *end_type = next_type + hdr->type_len;
420 while (next_type + sizeof(struct btf_type) <= end_type) {
421 if (btf->swapped_endian)
422 btf_bswap_type_base(next_type);
424 type_size = btf_type_size(next_type);
427 if (next_type + type_size > end_type) {
428 pr_warn("BTF type [%d] is malformed\n", btf->start_id + btf->nr_types);
432 if (btf->swapped_endian && btf_bswap_type_rest(next_type))
435 err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
439 next_type += type_size;
443 if (next_type != end_type) {
444 pr_warn("BTF types data is malformed\n");
451 __u32 btf__type_cnt(const struct btf *btf)
453 return btf->start_id + btf->nr_types;
456 const struct btf *btf__base_btf(const struct btf *btf)
458 return btf->base_btf;
461 /* internal helper returning non-const pointer to a type */
462 struct btf_type *btf_type_by_id(const struct btf *btf, __u32 type_id)
466 if (type_id < btf->start_id)
467 return btf_type_by_id(btf->base_btf, type_id);
468 return btf->types_data + btf->type_offs[type_id - btf->start_id];
471 const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
473 if (type_id >= btf->start_id + btf->nr_types)
474 return errno = EINVAL, NULL;
475 return btf_type_by_id((struct btf *)btf, type_id);
478 static int determine_ptr_size(const struct btf *btf)
480 static const char * const long_aliases[] = {
493 const struct btf_type *t;
497 if (btf->base_btf && btf->base_btf->ptr_sz > 0)
498 return btf->base_btf->ptr_sz;
500 n = btf__type_cnt(btf);
501 for (i = 1; i < n; i++) {
502 t = btf__type_by_id(btf, i);
506 if (t->size != 4 && t->size != 8)
509 name = btf__name_by_offset(btf, t->name_off);
513 for (j = 0; j < ARRAY_SIZE(long_aliases); j++) {
514 if (strcmp(name, long_aliases[j]) == 0)
522 static size_t btf_ptr_sz(const struct btf *btf)
525 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
526 return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
529 /* Return pointer size this BTF instance assumes. The size is heuristically
530 * determined by looking for 'long' or 'unsigned long' integer type and
531 * recording its size in bytes. If BTF type information doesn't have any such
532 * type, this function returns 0. In the latter case, native architecture's
533 * pointer size is assumed, so will be either 4 or 8, depending on
534 * architecture that libbpf was compiled for. It's possible to override
535 * guessed value by using btf__set_pointer_size() API.
537 size_t btf__pointer_size(const struct btf *btf)
540 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
543 /* not enough BTF type info to guess */
549 /* Override or set pointer size in bytes. Only values of 4 and 8 are
552 int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
554 if (ptr_sz != 4 && ptr_sz != 8)
555 return libbpf_err(-EINVAL);
556 btf->ptr_sz = ptr_sz;
560 static bool is_host_big_endian(void)
562 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
564 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
567 # error "Unrecognized __BYTE_ORDER__"
571 enum btf_endianness btf__endianness(const struct btf *btf)
573 if (is_host_big_endian())
574 return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
576 return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
579 int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
581 if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
582 return libbpf_err(-EINVAL);
584 btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
585 if (!btf->swapped_endian) {
586 free(btf->raw_data_swapped);
587 btf->raw_data_swapped = NULL;
592 static bool btf_type_is_void(const struct btf_type *t)
594 return t == &btf_void || btf_is_fwd(t);
597 static bool btf_type_is_void_or_null(const struct btf_type *t)
599 return !t || btf_type_is_void(t);
602 #define MAX_RESOLVE_DEPTH 32
604 __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
606 const struct btf_array *array;
607 const struct btf_type *t;
612 t = btf__type_by_id(btf, type_id);
613 for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t); i++) {
614 switch (btf_kind(t)) {
616 case BTF_KIND_STRUCT:
619 case BTF_KIND_ENUM64:
620 case BTF_KIND_DATASEC:
625 size = btf_ptr_sz(btf);
627 case BTF_KIND_TYPEDEF:
628 case BTF_KIND_VOLATILE:
630 case BTF_KIND_RESTRICT:
632 case BTF_KIND_DECL_TAG:
633 case BTF_KIND_TYPE_TAG:
637 array = btf_array(t);
638 if (nelems && array->nelems > UINT32_MAX / nelems)
639 return libbpf_err(-E2BIG);
640 nelems *= array->nelems;
641 type_id = array->type;
644 return libbpf_err(-EINVAL);
647 t = btf__type_by_id(btf, type_id);
652 return libbpf_err(-EINVAL);
653 if (nelems && size > UINT32_MAX / nelems)
654 return libbpf_err(-E2BIG);
656 return nelems * size;
659 int btf__align_of(const struct btf *btf, __u32 id)
661 const struct btf_type *t = btf__type_by_id(btf, id);
662 __u16 kind = btf_kind(t);
667 case BTF_KIND_ENUM64:
669 return min(btf_ptr_sz(btf), (size_t)t->size);
671 return btf_ptr_sz(btf);
672 case BTF_KIND_TYPEDEF:
673 case BTF_KIND_VOLATILE:
675 case BTF_KIND_RESTRICT:
676 case BTF_KIND_TYPE_TAG:
677 return btf__align_of(btf, t->type);
679 return btf__align_of(btf, btf_array(t)->type);
680 case BTF_KIND_STRUCT:
681 case BTF_KIND_UNION: {
682 const struct btf_member *m = btf_members(t);
683 __u16 vlen = btf_vlen(t);
684 int i, max_align = 1, align;
686 for (i = 0; i < vlen; i++, m++) {
687 align = btf__align_of(btf, m->type);
689 return libbpf_err(align);
690 max_align = max(max_align, align);
696 pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
697 return errno = EINVAL, 0;
701 int btf__resolve_type(const struct btf *btf, __u32 type_id)
703 const struct btf_type *t;
706 t = btf__type_by_id(btf, type_id);
707 while (depth < MAX_RESOLVE_DEPTH &&
708 !btf_type_is_void_or_null(t) &&
709 (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
711 t = btf__type_by_id(btf, type_id);
715 if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
716 return libbpf_err(-EINVAL);
721 __s32 btf__find_by_name(const struct btf *btf, const char *type_name)
723 __u32 i, nr_types = btf__type_cnt(btf);
725 if (!strcmp(type_name, "void"))
728 for (i = 1; i < nr_types; i++) {
729 const struct btf_type *t = btf__type_by_id(btf, i);
730 const char *name = btf__name_by_offset(btf, t->name_off);
732 if (name && !strcmp(type_name, name))
736 return libbpf_err(-ENOENT);
739 static __s32 btf_find_by_name_kind(const struct btf *btf, int start_id,
740 const char *type_name, __u32 kind)
742 __u32 i, nr_types = btf__type_cnt(btf);
744 if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
747 for (i = start_id; i < nr_types; i++) {
748 const struct btf_type *t = btf__type_by_id(btf, i);
751 if (btf_kind(t) != kind)
753 name = btf__name_by_offset(btf, t->name_off);
754 if (name && !strcmp(type_name, name))
758 return libbpf_err(-ENOENT);
761 __s32 btf__find_by_name_kind_own(const struct btf *btf, const char *type_name,
764 return btf_find_by_name_kind(btf, btf->start_id, type_name, kind);
767 __s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
770 return btf_find_by_name_kind(btf, 1, type_name, kind);
773 static bool btf_is_modifiable(const struct btf *btf)
775 return (void *)btf->hdr != btf->raw_data;
778 void btf__free(struct btf *btf)
780 if (IS_ERR_OR_NULL(btf))
786 if (btf_is_modifiable(btf)) {
787 /* if BTF was modified after loading, it will have a split
788 * in-memory representation for header, types, and strings
789 * sections, so we need to free all of them individually. It
790 * might still have a cached contiguous raw data present,
791 * which will be unconditionally freed below.
794 free(btf->types_data);
795 strset__free(btf->strs_set);
798 free(btf->raw_data_swapped);
799 free(btf->type_offs);
803 static struct btf *btf_new_empty(struct btf *base_btf)
807 btf = calloc(1, sizeof(*btf));
809 return ERR_PTR(-ENOMEM);
813 btf->start_str_off = 0;
815 btf->ptr_sz = sizeof(void *);
816 btf->swapped_endian = false;
819 btf->base_btf = base_btf;
820 btf->start_id = btf__type_cnt(base_btf);
821 btf->start_str_off = base_btf->hdr->str_len;
824 /* +1 for empty string at offset 0 */
825 btf->raw_size = sizeof(struct btf_header) + (base_btf ? 0 : 1);
826 btf->raw_data = calloc(1, btf->raw_size);
827 if (!btf->raw_data) {
829 return ERR_PTR(-ENOMEM);
832 btf->hdr = btf->raw_data;
833 btf->hdr->hdr_len = sizeof(struct btf_header);
834 btf->hdr->magic = BTF_MAGIC;
835 btf->hdr->version = BTF_VERSION;
837 btf->types_data = btf->raw_data + btf->hdr->hdr_len;
838 btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
839 btf->hdr->str_len = base_btf ? 0 : 1; /* empty string at offset 0 */
844 struct btf *btf__new_empty(void)
846 return libbpf_ptr(btf_new_empty(NULL));
849 struct btf *btf__new_empty_split(struct btf *base_btf)
851 return libbpf_ptr(btf_new_empty(base_btf));
854 static struct btf *btf_new(const void *data, __u32 size, struct btf *base_btf)
859 btf = calloc(1, sizeof(struct btf));
861 return ERR_PTR(-ENOMEM);
865 btf->start_str_off = 0;
869 btf->base_btf = base_btf;
870 btf->start_id = btf__type_cnt(base_btf);
871 btf->start_str_off = base_btf->hdr->str_len;
874 btf->raw_data = malloc(size);
875 if (!btf->raw_data) {
879 memcpy(btf->raw_data, data, size);
880 btf->raw_size = size;
882 btf->hdr = btf->raw_data;
883 err = btf_parse_hdr(btf);
887 btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
888 btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
890 err = btf_parse_str_sec(btf);
891 err = err ?: btf_parse_type_sec(btf);
904 struct btf *btf__new(const void *data, __u32 size)
906 return libbpf_ptr(btf_new(data, size, NULL));
909 static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,
910 struct btf_ext **btf_ext)
912 Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
913 int err = 0, fd = -1, idx = 0;
914 struct btf *btf = NULL;
920 if (elf_version(EV_CURRENT) == EV_NONE) {
921 pr_warn("failed to init libelf for %s\n", path);
922 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
925 fd = open(path, O_RDONLY | O_CLOEXEC);
928 pr_warn("failed to open %s: %s\n", path, strerror(errno));
932 err = -LIBBPF_ERRNO__FORMAT;
934 elf = elf_begin(fd, ELF_C_READ, NULL);
936 pr_warn("failed to open %s as ELF file\n", path);
939 if (!gelf_getehdr(elf, &ehdr)) {
940 pr_warn("failed to get EHDR from %s\n", path);
944 if (elf_getshdrstrndx(elf, &shstrndx)) {
945 pr_warn("failed to get section names section index for %s\n",
950 if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
951 pr_warn("failed to get e_shstrndx from %s\n", path);
955 while ((scn = elf_nextscn(elf, scn)) != NULL) {
960 if (gelf_getshdr(scn, &sh) != &sh) {
961 pr_warn("failed to get section(%d) header from %s\n",
965 name = elf_strptr(elf, shstrndx, sh.sh_name);
967 pr_warn("failed to get section(%d) name from %s\n",
971 if (strcmp(name, BTF_ELF_SEC) == 0) {
972 btf_data = elf_getdata(scn, 0);
974 pr_warn("failed to get section(%d, %s) data from %s\n",
979 } else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
980 btf_ext_data = elf_getdata(scn, 0);
982 pr_warn("failed to get section(%d, %s) data from %s\n",
996 btf = btf_new(btf_data->d_buf, btf_data->d_size, base_btf);
997 err = libbpf_get_error(btf);
1001 switch (gelf_getclass(elf)) {
1003 btf__set_pointer_size(btf, 4);
1006 btf__set_pointer_size(btf, 8);
1009 pr_warn("failed to get ELF class (bitness) for %s\n", path);
1013 if (btf_ext && btf_ext_data) {
1014 *btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
1015 err = libbpf_get_error(*btf_ext);
1018 } else if (btf_ext) {
1030 btf_ext__free(*btf_ext);
1033 return ERR_PTR(err);
1036 struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
1038 return libbpf_ptr(btf_parse_elf(path, NULL, btf_ext));
1041 struct btf *btf__parse_elf_split(const char *path, struct btf *base_btf)
1043 return libbpf_ptr(btf_parse_elf(path, base_btf, NULL));
1046 static struct btf *btf_parse_raw(const char *path, struct btf *base_btf)
1048 struct btf *btf = NULL;
1055 f = fopen(path, "rb");
1061 /* check BTF magic */
1062 if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
1066 if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
1067 /* definitely not a raw BTF */
1073 if (fseek(f, 0, SEEK_END)) {
1082 /* rewind to the start */
1083 if (fseek(f, 0, SEEK_SET)) {
1088 /* pre-alloc memory and read all of BTF data */
1094 if (fread(data, 1, sz, f) < sz) {
1099 /* finally parse BTF data */
1100 btf = btf_new(data, sz, base_btf);
1106 return err ? ERR_PTR(err) : btf;
1109 struct btf *btf__parse_raw(const char *path)
1111 return libbpf_ptr(btf_parse_raw(path, NULL));
1114 struct btf *btf__parse_raw_split(const char *path, struct btf *base_btf)
1116 return libbpf_ptr(btf_parse_raw(path, base_btf));
1119 static struct btf *btf_parse(const char *path, struct btf *base_btf, struct btf_ext **btf_ext)
1127 btf = btf_parse_raw(path, base_btf);
1128 err = libbpf_get_error(btf);
1132 return ERR_PTR(err);
1133 return btf_parse_elf(path, base_btf, btf_ext);
1136 struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
1138 return libbpf_ptr(btf_parse(path, NULL, btf_ext));
1141 struct btf *btf__parse_split(const char *path, struct btf *base_btf)
1143 return libbpf_ptr(btf_parse(path, base_btf, NULL));
1146 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1148 int btf_load_into_kernel(struct btf *btf, char *log_buf, size_t log_sz, __u32 log_level)
1150 LIBBPF_OPTS(bpf_btf_load_opts, opts);
1151 __u32 buf_sz = 0, raw_size;
1152 char *buf = NULL, *tmp;
1157 return libbpf_err(-EEXIST);
1158 if (log_sz && !log_buf)
1159 return libbpf_err(-EINVAL);
1161 /* cache native raw data representation */
1162 raw_data = btf_get_raw_data(btf, &raw_size, false);
1167 btf->raw_size = raw_size;
1168 btf->raw_data = raw_data;
1171 /* if log_level is 0, we won't provide log_buf/log_size to the kernel,
1172 * initially. Only if BTF loading fails, we bump log_level to 1 and
1173 * retry, using either auto-allocated or custom log_buf. This way
1174 * non-NULL custom log_buf provides a buffer just in case, but hopes
1175 * for successful load and no need for log_buf.
1178 /* if caller didn't provide custom log_buf, we'll keep
1179 * allocating our own progressively bigger buffers for BTF
1183 buf_sz = max((__u32)BPF_LOG_BUF_SIZE, buf_sz * 2);
1184 tmp = realloc(buf, buf_sz);
1193 opts.log_buf = log_buf ? log_buf : buf;
1194 opts.log_size = log_buf ? log_sz : buf_sz;
1195 opts.log_level = log_level;
1198 btf->fd = bpf_btf_load(raw_data, raw_size, &opts);
1200 /* time to turn on verbose mode and try again */
1201 if (log_level == 0) {
1205 /* only retry if caller didn't provide custom log_buf, but
1206 * make sure we can never overflow buf_sz
1208 if (!log_buf && errno == ENOSPC && buf_sz <= UINT_MAX / 2)
1212 pr_warn("BTF loading error: %d\n", err);
1213 /* don't print out contents of custom log_buf */
1214 if (!log_buf && buf[0])
1215 pr_warn("-- BEGIN BTF LOAD LOG ---\n%s\n-- END BTF LOAD LOG --\n", buf);
1220 return libbpf_err(err);
1223 int btf__load_into_kernel(struct btf *btf)
1225 return btf_load_into_kernel(btf, NULL, 0, 0);
1228 int btf__fd(const struct btf *btf)
1233 void btf__set_fd(struct btf *btf, int fd)
1238 static const void *btf_strs_data(const struct btf *btf)
1240 return btf->strs_data ? btf->strs_data : strset__data(btf->strs_set);
1243 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1245 struct btf_header *hdr = btf->hdr;
1251 data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1253 *size = btf->raw_size;
1257 data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1258 data = calloc(1, data_sz);
1263 memcpy(p, hdr, hdr->hdr_len);
1268 memcpy(p, btf->types_data, hdr->type_len);
1270 for (i = 0; i < btf->nr_types; i++) {
1271 t = p + btf->type_offs[i];
1272 /* btf_bswap_type_rest() relies on native t->info, so
1273 * we swap base type info after we swapped all the
1274 * additional information
1276 if (btf_bswap_type_rest(t))
1278 btf_bswap_type_base(t);
1283 memcpy(p, btf_strs_data(btf), hdr->str_len);
1293 const void *btf__raw_data(const struct btf *btf_ro, __u32 *size)
1295 struct btf *btf = (struct btf *)btf_ro;
1299 data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1301 return errno = ENOMEM, NULL;
1303 btf->raw_size = data_sz;
1304 if (btf->swapped_endian)
1305 btf->raw_data_swapped = data;
1307 btf->raw_data = data;
1312 __attribute__((alias("btf__raw_data")))
1313 const void *btf__get_raw_data(const struct btf *btf, __u32 *size);
1315 const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
1317 if (offset < btf->start_str_off)
1318 return btf__str_by_offset(btf->base_btf, offset);
1319 else if (offset - btf->start_str_off < btf->hdr->str_len)
1320 return btf_strs_data(btf) + (offset - btf->start_str_off);
1322 return errno = EINVAL, NULL;
1325 const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1327 return btf__str_by_offset(btf, offset);
1330 struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf)
1332 struct bpf_btf_info btf_info;
1333 __u32 len = sizeof(btf_info);
1339 /* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
1340 * let's start with a sane default - 4KiB here - and resize it only if
1341 * bpf_obj_get_info_by_fd() needs a bigger buffer.
1344 ptr = malloc(last_size);
1346 return ERR_PTR(-ENOMEM);
1348 memset(&btf_info, 0, sizeof(btf_info));
1349 btf_info.btf = ptr_to_u64(ptr);
1350 btf_info.btf_size = last_size;
1351 err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1353 if (!err && btf_info.btf_size > last_size) {
1356 last_size = btf_info.btf_size;
1357 temp_ptr = realloc(ptr, last_size);
1359 btf = ERR_PTR(-ENOMEM);
1364 len = sizeof(btf_info);
1365 memset(&btf_info, 0, sizeof(btf_info));
1366 btf_info.btf = ptr_to_u64(ptr);
1367 btf_info.btf_size = last_size;
1369 err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1372 if (err || btf_info.btf_size > last_size) {
1373 btf = err ? ERR_PTR(-errno) : ERR_PTR(-E2BIG);
1377 btf = btf_new(ptr, btf_info.btf_size, base_btf);
1384 struct btf *btf__load_from_kernel_by_id_split(__u32 id, struct btf *base_btf)
1389 btf_fd = bpf_btf_get_fd_by_id(id);
1391 return libbpf_err_ptr(-errno);
1393 btf = btf_get_from_fd(btf_fd, base_btf);
1396 return libbpf_ptr(btf);
1399 struct btf *btf__load_from_kernel_by_id(__u32 id)
1401 return btf__load_from_kernel_by_id_split(id, NULL);
1404 static void btf_invalidate_raw_data(struct btf *btf)
1406 if (btf->raw_data) {
1407 free(btf->raw_data);
1408 btf->raw_data = NULL;
1410 if (btf->raw_data_swapped) {
1411 free(btf->raw_data_swapped);
1412 btf->raw_data_swapped = NULL;
1416 /* Ensure BTF is ready to be modified (by splitting into a three memory
1417 * regions for header, types, and strings). Also invalidate cached
1420 static int btf_ensure_modifiable(struct btf *btf)
1423 struct strset *set = NULL;
1426 if (btf_is_modifiable(btf)) {
1427 /* any BTF modification invalidates raw_data */
1428 btf_invalidate_raw_data(btf);
1432 /* split raw data into three memory regions */
1433 hdr = malloc(btf->hdr->hdr_len);
1434 types = malloc(btf->hdr->type_len);
1438 memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1439 memcpy(types, btf->types_data, btf->hdr->type_len);
1441 /* build lookup index for all strings */
1442 set = strset__new(BTF_MAX_STR_OFFSET, btf->strs_data, btf->hdr->str_len);
1448 /* only when everything was successful, update internal state */
1450 btf->types_data = types;
1451 btf->types_data_cap = btf->hdr->type_len;
1452 btf->strs_data = NULL;
1453 btf->strs_set = set;
1454 /* if BTF was created from scratch, all strings are guaranteed to be
1455 * unique and deduplicated
1457 if (btf->hdr->str_len == 0)
1458 btf->strs_deduped = true;
1459 if (!btf->base_btf && btf->hdr->str_len == 1)
1460 btf->strs_deduped = true;
1462 /* invalidate raw_data representation */
1463 btf_invalidate_raw_data(btf);
1474 /* Find an offset in BTF string section that corresponds to a given string *s*.
1476 * - >0 offset into string section, if string is found;
1477 * - -ENOENT, if string is not in the string section;
1478 * - <0, on any other error.
1480 int btf__find_str(struct btf *btf, const char *s)
1484 if (btf->base_btf) {
1485 off = btf__find_str(btf->base_btf, s);
1490 /* BTF needs to be in a modifiable state to build string lookup index */
1491 if (btf_ensure_modifiable(btf))
1492 return libbpf_err(-ENOMEM);
1494 off = strset__find_str(btf->strs_set, s);
1496 return libbpf_err(off);
1498 return btf->start_str_off + off;
1501 /* Add a string s to the BTF string section.
1503 * - > 0 offset into string section, on success;
1506 int btf__add_str(struct btf *btf, const char *s)
1510 if (btf->base_btf) {
1511 off = btf__find_str(btf->base_btf, s);
1516 if (btf_ensure_modifiable(btf))
1517 return libbpf_err(-ENOMEM);
1519 off = strset__add_str(btf->strs_set, s);
1521 return libbpf_err(off);
1523 btf->hdr->str_len = strset__data_size(btf->strs_set);
1525 return btf->start_str_off + off;
1528 static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1530 return libbpf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1531 btf->hdr->type_len, UINT_MAX, add_sz);
1534 static void btf_type_inc_vlen(struct btf_type *t)
1536 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1539 static int btf_commit_type(struct btf *btf, int data_sz)
1543 err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1545 return libbpf_err(err);
1547 btf->hdr->type_len += data_sz;
1548 btf->hdr->str_off += data_sz;
1550 return btf->start_id + btf->nr_types - 1;
1554 const struct btf *src;
1556 struct hashmap *str_off_map; /* map string offsets from src to dst */
1559 static int btf_rewrite_str(__u32 *str_off, void *ctx)
1561 struct btf_pipe *p = ctx;
1565 if (!*str_off) /* nothing to do for empty strings */
1568 if (p->str_off_map &&
1569 hashmap__find(p->str_off_map, *str_off, &mapped_off)) {
1570 *str_off = mapped_off;
1574 off = btf__add_str(p->dst, btf__str_by_offset(p->src, *str_off));
1578 /* Remember string mapping from src to dst. It avoids
1579 * performing expensive string comparisons.
1581 if (p->str_off_map) {
1582 err = hashmap__append(p->str_off_map, *str_off, off);
1591 int btf__add_type(struct btf *btf, const struct btf *src_btf, const struct btf_type *src_type)
1593 struct btf_pipe p = { .src = src_btf, .dst = btf };
1597 sz = btf_type_size(src_type);
1599 return libbpf_err(sz);
1601 /* deconstruct BTF, if necessary, and invalidate raw_data */
1602 if (btf_ensure_modifiable(btf))
1603 return libbpf_err(-ENOMEM);
1605 t = btf_add_type_mem(btf, sz);
1607 return libbpf_err(-ENOMEM);
1609 memcpy(t, src_type, sz);
1611 err = btf_type_visit_str_offs(t, btf_rewrite_str, &p);
1613 return libbpf_err(err);
1615 return btf_commit_type(btf, sz);
1618 static int btf_rewrite_type_ids(__u32 *type_id, void *ctx)
1620 struct btf *btf = ctx;
1622 if (!*type_id) /* nothing to do for VOID references */
1625 /* we haven't updated btf's type count yet, so
1626 * btf->start_id + btf->nr_types - 1 is the type ID offset we should
1627 * add to all newly added BTF types
1629 *type_id += btf->start_id + btf->nr_types - 1;
1633 static size_t btf_dedup_identity_hash_fn(long key, void *ctx);
1634 static bool btf_dedup_equal_fn(long k1, long k2, void *ctx);
1636 int btf__add_btf(struct btf *btf, const struct btf *src_btf)
1638 struct btf_pipe p = { .src = src_btf, .dst = btf };
1639 int data_sz, sz, cnt, i, err, old_strs_len;
1643 /* appending split BTF isn't supported yet */
1644 if (src_btf->base_btf)
1645 return libbpf_err(-ENOTSUP);
1647 /* deconstruct BTF, if necessary, and invalidate raw_data */
1648 if (btf_ensure_modifiable(btf))
1649 return libbpf_err(-ENOMEM);
1651 /* remember original strings section size if we have to roll back
1652 * partial strings section changes
1654 old_strs_len = btf->hdr->str_len;
1656 data_sz = src_btf->hdr->type_len;
1657 cnt = btf__type_cnt(src_btf) - 1;
1659 /* pre-allocate enough memory for new types */
1660 t = btf_add_type_mem(btf, data_sz);
1662 return libbpf_err(-ENOMEM);
1664 /* pre-allocate enough memory for type offset index for new types */
1665 off = btf_add_type_offs_mem(btf, cnt);
1667 return libbpf_err(-ENOMEM);
1669 /* Map the string offsets from src_btf to the offsets from btf to improve performance */
1670 p.str_off_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
1671 if (IS_ERR(p.str_off_map))
1672 return libbpf_err(-ENOMEM);
1674 /* bulk copy types data for all types from src_btf */
1675 memcpy(t, src_btf->types_data, data_sz);
1677 for (i = 0; i < cnt; i++) {
1678 sz = btf_type_size(t);
1680 /* unlikely, has to be corrupted src_btf */
1685 /* fill out type ID to type offset mapping for lookups by type ID */
1686 *off = t - btf->types_data;
1688 /* add, dedup, and remap strings referenced by this BTF type */
1689 err = btf_type_visit_str_offs(t, btf_rewrite_str, &p);
1693 /* remap all type IDs referenced from this BTF type */
1694 err = btf_type_visit_type_ids(t, btf_rewrite_type_ids, btf);
1698 /* go to next type data and type offset index entry */
1703 /* Up until now any of the copied type data was effectively invisible,
1704 * so if we exited early before this point due to error, BTF would be
1705 * effectively unmodified. There would be extra internal memory
1706 * pre-allocated, but it would not be available for querying. But now
1707 * that we've copied and rewritten all the data successfully, we can
1708 * update type count and various internal offsets and sizes to
1709 * "commit" the changes and made them visible to the outside world.
1711 btf->hdr->type_len += data_sz;
1712 btf->hdr->str_off += data_sz;
1713 btf->nr_types += cnt;
1715 hashmap__free(p.str_off_map);
1717 /* return type ID of the first added BTF type */
1718 return btf->start_id + btf->nr_types - cnt;
1720 /* zero out preallocated memory as if it was just allocated with
1723 memset(btf->types_data + btf->hdr->type_len, 0, data_sz);
1724 memset(btf->strs_data + old_strs_len, 0, btf->hdr->str_len - old_strs_len);
1726 /* and now restore original strings section size; types data size
1727 * wasn't modified, so doesn't need restoring, see big comment above
1729 btf->hdr->str_len = old_strs_len;
1731 hashmap__free(p.str_off_map);
1733 return libbpf_err(err);
1737 * Append new BTF_KIND_INT type with:
1738 * - *name* - non-empty, non-NULL type name;
1739 * - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1740 * - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1742 * - >0, type ID of newly added BTF type;
1745 int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
1750 /* non-empty name */
1751 if (!name || !name[0])
1752 return libbpf_err(-EINVAL);
1753 /* byte_sz must be power of 2 */
1754 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
1755 return libbpf_err(-EINVAL);
1756 if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
1757 return libbpf_err(-EINVAL);
1759 /* deconstruct BTF, if necessary, and invalidate raw_data */
1760 if (btf_ensure_modifiable(btf))
1761 return libbpf_err(-ENOMEM);
1763 sz = sizeof(struct btf_type) + sizeof(int);
1764 t = btf_add_type_mem(btf, sz);
1766 return libbpf_err(-ENOMEM);
1768 /* if something goes wrong later, we might end up with an extra string,
1769 * but that shouldn't be a problem, because BTF can't be constructed
1770 * completely anyway and will most probably be just discarded
1772 name_off = btf__add_str(btf, name);
1776 t->name_off = name_off;
1777 t->info = btf_type_info(BTF_KIND_INT, 0, 0);
1779 /* set INT info, we don't allow setting legacy bit offset/size */
1780 *(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
1782 return btf_commit_type(btf, sz);
1786 * Append new BTF_KIND_FLOAT type with:
1787 * - *name* - non-empty, non-NULL type name;
1788 * - *sz* - size of the type, in bytes;
1790 * - >0, type ID of newly added BTF type;
1793 int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
1798 /* non-empty name */
1799 if (!name || !name[0])
1800 return libbpf_err(-EINVAL);
1802 /* byte_sz must be one of the explicitly allowed values */
1803 if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
1805 return libbpf_err(-EINVAL);
1807 if (btf_ensure_modifiable(btf))
1808 return libbpf_err(-ENOMEM);
1810 sz = sizeof(struct btf_type);
1811 t = btf_add_type_mem(btf, sz);
1813 return libbpf_err(-ENOMEM);
1815 name_off = btf__add_str(btf, name);
1819 t->name_off = name_off;
1820 t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
1823 return btf_commit_type(btf, sz);
1826 /* it's completely legal to append BTF types with type IDs pointing forward to
1827 * types that haven't been appended yet, so we only make sure that id looks
1828 * sane, we can't guarantee that ID will always be valid
1830 static int validate_type_id(int id)
1832 if (id < 0 || id > BTF_MAX_NR_TYPES)
1837 /* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
1838 static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
1841 int sz, name_off = 0;
1843 if (validate_type_id(ref_type_id))
1844 return libbpf_err(-EINVAL);
1846 if (btf_ensure_modifiable(btf))
1847 return libbpf_err(-ENOMEM);
1849 sz = sizeof(struct btf_type);
1850 t = btf_add_type_mem(btf, sz);
1852 return libbpf_err(-ENOMEM);
1854 if (name && name[0]) {
1855 name_off = btf__add_str(btf, name);
1860 t->name_off = name_off;
1861 t->info = btf_type_info(kind, 0, 0);
1862 t->type = ref_type_id;
1864 return btf_commit_type(btf, sz);
1868 * Append new BTF_KIND_PTR type with:
1869 * - *ref_type_id* - referenced type ID, it might not exist yet;
1871 * - >0, type ID of newly added BTF type;
1874 int btf__add_ptr(struct btf *btf, int ref_type_id)
1876 return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
1880 * Append new BTF_KIND_ARRAY type with:
1881 * - *index_type_id* - type ID of the type describing array index;
1882 * - *elem_type_id* - type ID of the type describing array element;
1883 * - *nr_elems* - the size of the array;
1885 * - >0, type ID of newly added BTF type;
1888 int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
1891 struct btf_array *a;
1894 if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
1895 return libbpf_err(-EINVAL);
1897 if (btf_ensure_modifiable(btf))
1898 return libbpf_err(-ENOMEM);
1900 sz = sizeof(struct btf_type) + sizeof(struct btf_array);
1901 t = btf_add_type_mem(btf, sz);
1903 return libbpf_err(-ENOMEM);
1906 t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
1910 a->type = elem_type_id;
1911 a->index_type = index_type_id;
1912 a->nelems = nr_elems;
1914 return btf_commit_type(btf, sz);
1917 /* generic STRUCT/UNION append function */
1918 static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
1921 int sz, name_off = 0;
1923 if (btf_ensure_modifiable(btf))
1924 return libbpf_err(-ENOMEM);
1926 sz = sizeof(struct btf_type);
1927 t = btf_add_type_mem(btf, sz);
1929 return libbpf_err(-ENOMEM);
1931 if (name && name[0]) {
1932 name_off = btf__add_str(btf, name);
1937 /* start out with vlen=0 and no kflag; this will be adjusted when
1938 * adding each member
1940 t->name_off = name_off;
1941 t->info = btf_type_info(kind, 0, 0);
1944 return btf_commit_type(btf, sz);
1948 * Append new BTF_KIND_STRUCT type with:
1949 * - *name* - name of the struct, can be NULL or empty for anonymous structs;
1950 * - *byte_sz* - size of the struct, in bytes;
1952 * Struct initially has no fields in it. Fields can be added by
1953 * btf__add_field() right after btf__add_struct() succeeds.
1956 * - >0, type ID of newly added BTF type;
1959 int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
1961 return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
1965 * Append new BTF_KIND_UNION type with:
1966 * - *name* - name of the union, can be NULL or empty for anonymous union;
1967 * - *byte_sz* - size of the union, in bytes;
1969 * Union initially has no fields in it. Fields can be added by
1970 * btf__add_field() right after btf__add_union() succeeds. All fields
1971 * should have *bit_offset* of 0.
1974 * - >0, type ID of newly added BTF type;
1977 int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
1979 return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
1982 static struct btf_type *btf_last_type(struct btf *btf)
1984 return btf_type_by_id(btf, btf__type_cnt(btf) - 1);
1988 * Append new field for the current STRUCT/UNION type with:
1989 * - *name* - name of the field, can be NULL or empty for anonymous field;
1990 * - *type_id* - type ID for the type describing field type;
1991 * - *bit_offset* - bit offset of the start of the field within struct/union;
1992 * - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
1997 int btf__add_field(struct btf *btf, const char *name, int type_id,
1998 __u32 bit_offset, __u32 bit_size)
2001 struct btf_member *m;
2003 int sz, name_off = 0;
2005 /* last type should be union/struct */
2006 if (btf->nr_types == 0)
2007 return libbpf_err(-EINVAL);
2008 t = btf_last_type(btf);
2009 if (!btf_is_composite(t))
2010 return libbpf_err(-EINVAL);
2012 if (validate_type_id(type_id))
2013 return libbpf_err(-EINVAL);
2014 /* best-effort bit field offset/size enforcement */
2015 is_bitfield = bit_size || (bit_offset % 8 != 0);
2016 if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
2017 return libbpf_err(-EINVAL);
2019 /* only offset 0 is allowed for unions */
2020 if (btf_is_union(t) && bit_offset)
2021 return libbpf_err(-EINVAL);
2023 /* decompose and invalidate raw data */
2024 if (btf_ensure_modifiable(btf))
2025 return libbpf_err(-ENOMEM);
2027 sz = sizeof(struct btf_member);
2028 m = btf_add_type_mem(btf, sz);
2030 return libbpf_err(-ENOMEM);
2032 if (name && name[0]) {
2033 name_off = btf__add_str(btf, name);
2038 m->name_off = name_off;
2040 m->offset = bit_offset | (bit_size << 24);
2042 /* btf_add_type_mem can invalidate t pointer */
2043 t = btf_last_type(btf);
2044 /* update parent type's vlen and kflag */
2045 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
2047 btf->hdr->type_len += sz;
2048 btf->hdr->str_off += sz;
2052 static int btf_add_enum_common(struct btf *btf, const char *name, __u32 byte_sz,
2053 bool is_signed, __u8 kind)
2056 int sz, name_off = 0;
2058 /* byte_sz must be power of 2 */
2059 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2060 return libbpf_err(-EINVAL);
2062 if (btf_ensure_modifiable(btf))
2063 return libbpf_err(-ENOMEM);
2065 sz = sizeof(struct btf_type);
2066 t = btf_add_type_mem(btf, sz);
2068 return libbpf_err(-ENOMEM);
2070 if (name && name[0]) {
2071 name_off = btf__add_str(btf, name);
2076 /* start out with vlen=0; it will be adjusted when adding enum values */
2077 t->name_off = name_off;
2078 t->info = btf_type_info(kind, 0, is_signed);
2081 return btf_commit_type(btf, sz);
2085 * Append new BTF_KIND_ENUM type with:
2086 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
2087 * - *byte_sz* - size of the enum, in bytes.
2089 * Enum initially has no enum values in it (and corresponds to enum forward
2090 * declaration). Enumerator values can be added by btf__add_enum_value()
2091 * immediately after btf__add_enum() succeeds.
2094 * - >0, type ID of newly added BTF type;
2097 int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2100 * set the signedness to be unsigned, it will change to signed
2101 * if any later enumerator is negative.
2103 return btf_add_enum_common(btf, name, byte_sz, false, BTF_KIND_ENUM);
2107 * Append new enum value for the current ENUM type with:
2108 * - *name* - name of the enumerator value, can't be NULL or empty;
2109 * - *value* - integer value corresponding to enum value *name*;
2114 int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2120 /* last type should be BTF_KIND_ENUM */
2121 if (btf->nr_types == 0)
2122 return libbpf_err(-EINVAL);
2123 t = btf_last_type(btf);
2124 if (!btf_is_enum(t))
2125 return libbpf_err(-EINVAL);
2127 /* non-empty name */
2128 if (!name || !name[0])
2129 return libbpf_err(-EINVAL);
2130 if (value < INT_MIN || value > UINT_MAX)
2131 return libbpf_err(-E2BIG);
2133 /* decompose and invalidate raw data */
2134 if (btf_ensure_modifiable(btf))
2135 return libbpf_err(-ENOMEM);
2137 sz = sizeof(struct btf_enum);
2138 v = btf_add_type_mem(btf, sz);
2140 return libbpf_err(-ENOMEM);
2142 name_off = btf__add_str(btf, name);
2146 v->name_off = name_off;
2149 /* update parent type's vlen */
2150 t = btf_last_type(btf);
2151 btf_type_inc_vlen(t);
2153 /* if negative value, set signedness to signed */
2155 t->info = btf_type_info(btf_kind(t), btf_vlen(t), true);
2157 btf->hdr->type_len += sz;
2158 btf->hdr->str_off += sz;
2163 * Append new BTF_KIND_ENUM64 type with:
2164 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
2165 * - *byte_sz* - size of the enum, in bytes.
2166 * - *is_signed* - whether the enum values are signed or not;
2168 * Enum initially has no enum values in it (and corresponds to enum forward
2169 * declaration). Enumerator values can be added by btf__add_enum64_value()
2170 * immediately after btf__add_enum64() succeeds.
2173 * - >0, type ID of newly added BTF type;
2176 int btf__add_enum64(struct btf *btf, const char *name, __u32 byte_sz,
2179 return btf_add_enum_common(btf, name, byte_sz, is_signed,
2184 * Append new enum value for the current ENUM64 type with:
2185 * - *name* - name of the enumerator value, can't be NULL or empty;
2186 * - *value* - integer value corresponding to enum value *name*;
2191 int btf__add_enum64_value(struct btf *btf, const char *name, __u64 value)
2193 struct btf_enum64 *v;
2197 /* last type should be BTF_KIND_ENUM64 */
2198 if (btf->nr_types == 0)
2199 return libbpf_err(-EINVAL);
2200 t = btf_last_type(btf);
2201 if (!btf_is_enum64(t))
2202 return libbpf_err(-EINVAL);
2204 /* non-empty name */
2205 if (!name || !name[0])
2206 return libbpf_err(-EINVAL);
2208 /* decompose and invalidate raw data */
2209 if (btf_ensure_modifiable(btf))
2210 return libbpf_err(-ENOMEM);
2212 sz = sizeof(struct btf_enum64);
2213 v = btf_add_type_mem(btf, sz);
2215 return libbpf_err(-ENOMEM);
2217 name_off = btf__add_str(btf, name);
2221 v->name_off = name_off;
2222 v->val_lo32 = (__u32)value;
2223 v->val_hi32 = value >> 32;
2225 /* update parent type's vlen */
2226 t = btf_last_type(btf);
2227 btf_type_inc_vlen(t);
2229 btf->hdr->type_len += sz;
2230 btf->hdr->str_off += sz;
2235 * Append new BTF_KIND_FWD type with:
2236 * - *name*, non-empty/non-NULL name;
2237 * - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2238 * BTF_FWD_UNION, or BTF_FWD_ENUM;
2240 * - >0, type ID of newly added BTF type;
2243 int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2245 if (!name || !name[0])
2246 return libbpf_err(-EINVAL);
2249 case BTF_FWD_STRUCT:
2250 case BTF_FWD_UNION: {
2254 id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
2257 t = btf_type_by_id(btf, id);
2258 t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2262 /* enum forward in BTF currently is just an enum with no enum
2263 * values; we also assume a standard 4-byte size for it
2265 return btf__add_enum(btf, name, sizeof(int));
2267 return libbpf_err(-EINVAL);
2272 * Append new BTF_KING_TYPEDEF type with:
2273 * - *name*, non-empty/non-NULL name;
2274 * - *ref_type_id* - referenced type ID, it might not exist yet;
2276 * - >0, type ID of newly added BTF type;
2279 int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2281 if (!name || !name[0])
2282 return libbpf_err(-EINVAL);
2284 return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2288 * Append new BTF_KIND_VOLATILE type with:
2289 * - *ref_type_id* - referenced type ID, it might not exist yet;
2291 * - >0, type ID of newly added BTF type;
2294 int btf__add_volatile(struct btf *btf, int ref_type_id)
2296 return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2300 * Append new BTF_KIND_CONST type with:
2301 * - *ref_type_id* - referenced type ID, it might not exist yet;
2303 * - >0, type ID of newly added BTF type;
2306 int btf__add_const(struct btf *btf, int ref_type_id)
2308 return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2312 * Append new BTF_KIND_RESTRICT type with:
2313 * - *ref_type_id* - referenced type ID, it might not exist yet;
2315 * - >0, type ID of newly added BTF type;
2318 int btf__add_restrict(struct btf *btf, int ref_type_id)
2320 return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2324 * Append new BTF_KIND_TYPE_TAG type with:
2325 * - *value*, non-empty/non-NULL tag value;
2326 * - *ref_type_id* - referenced type ID, it might not exist yet;
2328 * - >0, type ID of newly added BTF type;
2331 int btf__add_type_tag(struct btf *btf, const char *value, int ref_type_id)
2333 if (!value || !value[0])
2334 return libbpf_err(-EINVAL);
2336 return btf_add_ref_kind(btf, BTF_KIND_TYPE_TAG, value, ref_type_id);
2340 * Append new BTF_KIND_FUNC type with:
2341 * - *name*, non-empty/non-NULL name;
2342 * - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2344 * - >0, type ID of newly added BTF type;
2347 int btf__add_func(struct btf *btf, const char *name,
2348 enum btf_func_linkage linkage, int proto_type_id)
2352 if (!name || !name[0])
2353 return libbpf_err(-EINVAL);
2354 if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2355 linkage != BTF_FUNC_EXTERN)
2356 return libbpf_err(-EINVAL);
2358 id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2360 struct btf_type *t = btf_type_by_id(btf, id);
2362 t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2364 return libbpf_err(id);
2368 * Append new BTF_KIND_FUNC_PROTO with:
2369 * - *ret_type_id* - type ID for return result of a function.
2371 * Function prototype initially has no arguments, but they can be added by
2372 * btf__add_func_param() one by one, immediately after
2373 * btf__add_func_proto() succeeded.
2376 * - >0, type ID of newly added BTF type;
2379 int btf__add_func_proto(struct btf *btf, int ret_type_id)
2384 if (validate_type_id(ret_type_id))
2385 return libbpf_err(-EINVAL);
2387 if (btf_ensure_modifiable(btf))
2388 return libbpf_err(-ENOMEM);
2390 sz = sizeof(struct btf_type);
2391 t = btf_add_type_mem(btf, sz);
2393 return libbpf_err(-ENOMEM);
2395 /* start out with vlen=0; this will be adjusted when adding enum
2396 * values, if necessary
2399 t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2400 t->type = ret_type_id;
2402 return btf_commit_type(btf, sz);
2406 * Append new function parameter for current FUNC_PROTO type with:
2407 * - *name* - parameter name, can be NULL or empty;
2408 * - *type_id* - type ID describing the type of the parameter.
2413 int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2416 struct btf_param *p;
2417 int sz, name_off = 0;
2419 if (validate_type_id(type_id))
2420 return libbpf_err(-EINVAL);
2422 /* last type should be BTF_KIND_FUNC_PROTO */
2423 if (btf->nr_types == 0)
2424 return libbpf_err(-EINVAL);
2425 t = btf_last_type(btf);
2426 if (!btf_is_func_proto(t))
2427 return libbpf_err(-EINVAL);
2429 /* decompose and invalidate raw data */
2430 if (btf_ensure_modifiable(btf))
2431 return libbpf_err(-ENOMEM);
2433 sz = sizeof(struct btf_param);
2434 p = btf_add_type_mem(btf, sz);
2436 return libbpf_err(-ENOMEM);
2438 if (name && name[0]) {
2439 name_off = btf__add_str(btf, name);
2444 p->name_off = name_off;
2447 /* update parent type's vlen */
2448 t = btf_last_type(btf);
2449 btf_type_inc_vlen(t);
2451 btf->hdr->type_len += sz;
2452 btf->hdr->str_off += sz;
2457 * Append new BTF_KIND_VAR type with:
2458 * - *name* - non-empty/non-NULL name;
2459 * - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2460 * BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2461 * - *type_id* - type ID of the type describing the type of the variable.
2463 * - >0, type ID of newly added BTF type;
2466 int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2472 /* non-empty name */
2473 if (!name || !name[0])
2474 return libbpf_err(-EINVAL);
2475 if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2476 linkage != BTF_VAR_GLOBAL_EXTERN)
2477 return libbpf_err(-EINVAL);
2478 if (validate_type_id(type_id))
2479 return libbpf_err(-EINVAL);
2481 /* deconstruct BTF, if necessary, and invalidate raw_data */
2482 if (btf_ensure_modifiable(btf))
2483 return libbpf_err(-ENOMEM);
2485 sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2486 t = btf_add_type_mem(btf, sz);
2488 return libbpf_err(-ENOMEM);
2490 name_off = btf__add_str(btf, name);
2494 t->name_off = name_off;
2495 t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2499 v->linkage = linkage;
2501 return btf_commit_type(btf, sz);
2505 * Append new BTF_KIND_DATASEC type with:
2506 * - *name* - non-empty/non-NULL name;
2507 * - *byte_sz* - data section size, in bytes.
2509 * Data section is initially empty. Variables info can be added with
2510 * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2513 * - >0, type ID of newly added BTF type;
2516 int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2521 /* non-empty name */
2522 if (!name || !name[0])
2523 return libbpf_err(-EINVAL);
2525 if (btf_ensure_modifiable(btf))
2526 return libbpf_err(-ENOMEM);
2528 sz = sizeof(struct btf_type);
2529 t = btf_add_type_mem(btf, sz);
2531 return libbpf_err(-ENOMEM);
2533 name_off = btf__add_str(btf, name);
2537 /* start with vlen=0, which will be update as var_secinfos are added */
2538 t->name_off = name_off;
2539 t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2542 return btf_commit_type(btf, sz);
2546 * Append new data section variable information entry for current DATASEC type:
2547 * - *var_type_id* - type ID, describing type of the variable;
2548 * - *offset* - variable offset within data section, in bytes;
2549 * - *byte_sz* - variable size, in bytes.
2555 int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2558 struct btf_var_secinfo *v;
2561 /* last type should be BTF_KIND_DATASEC */
2562 if (btf->nr_types == 0)
2563 return libbpf_err(-EINVAL);
2564 t = btf_last_type(btf);
2565 if (!btf_is_datasec(t))
2566 return libbpf_err(-EINVAL);
2568 if (validate_type_id(var_type_id))
2569 return libbpf_err(-EINVAL);
2571 /* decompose and invalidate raw data */
2572 if (btf_ensure_modifiable(btf))
2573 return libbpf_err(-ENOMEM);
2575 sz = sizeof(struct btf_var_secinfo);
2576 v = btf_add_type_mem(btf, sz);
2578 return libbpf_err(-ENOMEM);
2580 v->type = var_type_id;
2584 /* update parent type's vlen */
2585 t = btf_last_type(btf);
2586 btf_type_inc_vlen(t);
2588 btf->hdr->type_len += sz;
2589 btf->hdr->str_off += sz;
2594 * Append new BTF_KIND_DECL_TAG type with:
2595 * - *value* - non-empty/non-NULL string;
2596 * - *ref_type_id* - referenced type ID, it might not exist yet;
2597 * - *component_idx* - -1 for tagging reference type, otherwise struct/union
2598 * member or function argument index;
2600 * - >0, type ID of newly added BTF type;
2603 int btf__add_decl_tag(struct btf *btf, const char *value, int ref_type_id,
2609 if (!value || !value[0] || component_idx < -1)
2610 return libbpf_err(-EINVAL);
2612 if (validate_type_id(ref_type_id))
2613 return libbpf_err(-EINVAL);
2615 if (btf_ensure_modifiable(btf))
2616 return libbpf_err(-ENOMEM);
2618 sz = sizeof(struct btf_type) + sizeof(struct btf_decl_tag);
2619 t = btf_add_type_mem(btf, sz);
2621 return libbpf_err(-ENOMEM);
2623 value_off = btf__add_str(btf, value);
2627 t->name_off = value_off;
2628 t->info = btf_type_info(BTF_KIND_DECL_TAG, 0, false);
2629 t->type = ref_type_id;
2630 btf_decl_tag(t)->component_idx = component_idx;
2632 return btf_commit_type(btf, sz);
2635 struct btf_ext_sec_setup_param {
2639 struct btf_ext_info *ext_info;
2643 static int btf_ext_setup_info(struct btf_ext *btf_ext,
2644 struct btf_ext_sec_setup_param *ext_sec)
2646 const struct btf_ext_info_sec *sinfo;
2647 struct btf_ext_info *ext_info;
2648 __u32 info_left, record_size;
2650 /* The start of the info sec (including the __u32 record_size). */
2653 if (ext_sec->len == 0)
2656 if (ext_sec->off & 0x03) {
2657 pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2662 info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2663 info_left = ext_sec->len;
2665 if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2666 pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2667 ext_sec->desc, ext_sec->off, ext_sec->len);
2671 /* At least a record size */
2672 if (info_left < sizeof(__u32)) {
2673 pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2677 /* The record size needs to meet the minimum standard */
2678 record_size = *(__u32 *)info;
2679 if (record_size < ext_sec->min_rec_size ||
2680 record_size & 0x03) {
2681 pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2682 ext_sec->desc, record_size);
2686 sinfo = info + sizeof(__u32);
2687 info_left -= sizeof(__u32);
2689 /* If no records, return failure now so .BTF.ext won't be used. */
2691 pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
2696 unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2697 __u64 total_record_size;
2700 if (info_left < sec_hdrlen) {
2701 pr_debug("%s section header is not found in .BTF.ext\n",
2706 num_records = sinfo->num_info;
2707 if (num_records == 0) {
2708 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2713 total_record_size = sec_hdrlen + (__u64)num_records * record_size;
2714 if (info_left < total_record_size) {
2715 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2720 info_left -= total_record_size;
2721 sinfo = (void *)sinfo + total_record_size;
2725 ext_info = ext_sec->ext_info;
2726 ext_info->len = ext_sec->len - sizeof(__u32);
2727 ext_info->rec_size = record_size;
2728 ext_info->info = info + sizeof(__u32);
2729 ext_info->sec_cnt = sec_cnt;
2734 static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
2736 struct btf_ext_sec_setup_param param = {
2737 .off = btf_ext->hdr->func_info_off,
2738 .len = btf_ext->hdr->func_info_len,
2739 .min_rec_size = sizeof(struct bpf_func_info_min),
2740 .ext_info = &btf_ext->func_info,
2744 return btf_ext_setup_info(btf_ext, ¶m);
2747 static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
2749 struct btf_ext_sec_setup_param param = {
2750 .off = btf_ext->hdr->line_info_off,
2751 .len = btf_ext->hdr->line_info_len,
2752 .min_rec_size = sizeof(struct bpf_line_info_min),
2753 .ext_info = &btf_ext->line_info,
2754 .desc = "line_info",
2757 return btf_ext_setup_info(btf_ext, ¶m);
2760 static int btf_ext_setup_core_relos(struct btf_ext *btf_ext)
2762 struct btf_ext_sec_setup_param param = {
2763 .off = btf_ext->hdr->core_relo_off,
2764 .len = btf_ext->hdr->core_relo_len,
2765 .min_rec_size = sizeof(struct bpf_core_relo),
2766 .ext_info = &btf_ext->core_relo_info,
2767 .desc = "core_relo",
2770 return btf_ext_setup_info(btf_ext, ¶m);
2773 static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
2775 const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
2777 if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
2778 data_size < hdr->hdr_len) {
2779 pr_debug("BTF.ext header not found");
2783 if (hdr->magic == bswap_16(BTF_MAGIC)) {
2784 pr_warn("BTF.ext in non-native endianness is not supported\n");
2786 } else if (hdr->magic != BTF_MAGIC) {
2787 pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
2791 if (hdr->version != BTF_VERSION) {
2792 pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
2797 pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
2801 if (data_size == hdr->hdr_len) {
2802 pr_debug("BTF.ext has no data\n");
2809 void btf_ext__free(struct btf_ext *btf_ext)
2811 if (IS_ERR_OR_NULL(btf_ext))
2813 free(btf_ext->func_info.sec_idxs);
2814 free(btf_ext->line_info.sec_idxs);
2815 free(btf_ext->core_relo_info.sec_idxs);
2816 free(btf_ext->data);
2820 struct btf_ext *btf_ext__new(const __u8 *data, __u32 size)
2822 struct btf_ext *btf_ext;
2825 btf_ext = calloc(1, sizeof(struct btf_ext));
2827 return libbpf_err_ptr(-ENOMEM);
2829 btf_ext->data_size = size;
2830 btf_ext->data = malloc(size);
2831 if (!btf_ext->data) {
2835 memcpy(btf_ext->data, data, size);
2837 err = btf_ext_parse_hdr(btf_ext->data, size);
2841 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
2846 err = btf_ext_setup_func_info(btf_ext);
2850 err = btf_ext_setup_line_info(btf_ext);
2854 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
2855 goto done; /* skip core relos parsing */
2857 err = btf_ext_setup_core_relos(btf_ext);
2863 btf_ext__free(btf_ext);
2864 return libbpf_err_ptr(err);
2870 const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
2872 *size = btf_ext->data_size;
2873 return btf_ext->data;
2878 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts);
2879 static void btf_dedup_free(struct btf_dedup *d);
2880 static int btf_dedup_prep(struct btf_dedup *d);
2881 static int btf_dedup_strings(struct btf_dedup *d);
2882 static int btf_dedup_prim_types(struct btf_dedup *d);
2883 static int btf_dedup_struct_types(struct btf_dedup *d);
2884 static int btf_dedup_ref_types(struct btf_dedup *d);
2885 static int btf_dedup_resolve_fwds(struct btf_dedup *d);
2886 static int btf_dedup_compact_types(struct btf_dedup *d);
2887 static int btf_dedup_remap_types(struct btf_dedup *d);
2890 * Deduplicate BTF types and strings.
2892 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
2893 * section with all BTF type descriptors and string data. It overwrites that
2894 * memory in-place with deduplicated types and strings without any loss of
2895 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
2896 * is provided, all the strings referenced from .BTF.ext section are honored
2897 * and updated to point to the right offsets after deduplication.
2899 * If function returns with error, type/string data might be garbled and should
2902 * More verbose and detailed description of both problem btf_dedup is solving,
2903 * as well as solution could be found at:
2904 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
2906 * Problem description and justification
2907 * =====================================
2909 * BTF type information is typically emitted either as a result of conversion
2910 * from DWARF to BTF or directly by compiler. In both cases, each compilation
2911 * unit contains information about a subset of all the types that are used
2912 * in an application. These subsets are frequently overlapping and contain a lot
2913 * of duplicated information when later concatenated together into a single
2914 * binary. This algorithm ensures that each unique type is represented by single
2915 * BTF type descriptor, greatly reducing resulting size of BTF data.
2917 * Compilation unit isolation and subsequent duplication of data is not the only
2918 * problem. The same type hierarchy (e.g., struct and all the type that struct
2919 * references) in different compilation units can be represented in BTF to
2920 * various degrees of completeness (or, rather, incompleteness) due to
2921 * struct/union forward declarations.
2923 * Let's take a look at an example, that we'll use to better understand the
2924 * problem (and solution). Suppose we have two compilation units, each using
2925 * same `struct S`, but each of them having incomplete type information about
2954 * In case of CU #1, BTF data will know only that `struct B` exist (but no
2955 * more), but will know the complete type information about `struct A`. While
2956 * for CU #2, it will know full type information about `struct B`, but will
2957 * only know about forward declaration of `struct A` (in BTF terms, it will
2958 * have `BTF_KIND_FWD` type descriptor with name `B`).
2960 * This compilation unit isolation means that it's possible that there is no
2961 * single CU with complete type information describing structs `S`, `A`, and
2962 * `B`. Also, we might get tons of duplicated and redundant type information.
2964 * Additional complication we need to keep in mind comes from the fact that
2965 * types, in general, can form graphs containing cycles, not just DAGs.
2967 * While algorithm does deduplication, it also merges and resolves type
2968 * information (unless disabled throught `struct btf_opts`), whenever possible.
2969 * E.g., in the example above with two compilation units having partial type
2970 * information for structs `A` and `B`, the output of algorithm will emit
2971 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
2972 * (as well as type information for `int` and pointers), as if they were defined
2973 * in a single compilation unit as:
2993 * Algorithm completes its work in 7 separate passes:
2995 * 1. Strings deduplication.
2996 * 2. Primitive types deduplication (int, enum, fwd).
2997 * 3. Struct/union types deduplication.
2998 * 4. Resolve unambiguous forward declarations.
2999 * 5. Reference types deduplication (pointers, typedefs, arrays, funcs, func
3000 * protos, and const/volatile/restrict modifiers).
3001 * 6. Types compaction.
3002 * 7. Types remapping.
3004 * Algorithm determines canonical type descriptor, which is a single
3005 * representative type for each truly unique type. This canonical type is the
3006 * one that will go into final deduplicated BTF type information. For
3007 * struct/unions, it is also the type that algorithm will merge additional type
3008 * information into (while resolving FWDs), as it discovers it from data in
3009 * other CUs. Each input BTF type eventually gets either mapped to itself, if
3010 * that type is canonical, or to some other type, if that type is equivalent
3011 * and was chosen as canonical representative. This mapping is stored in
3012 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
3013 * FWD type got resolved to.
3015 * To facilitate fast discovery of canonical types, we also maintain canonical
3016 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
3017 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
3018 * that match that signature. With sufficiently good choice of type signature
3019 * hashing function, we can limit number of canonical types for each unique type
3020 * signature to a very small number, allowing to find canonical type for any
3021 * duplicated type very quickly.
3023 * Struct/union deduplication is the most critical part and algorithm for
3024 * deduplicating structs/unions is described in greater details in comments for
3025 * `btf_dedup_is_equiv` function.
3027 int btf__dedup(struct btf *btf, const struct btf_dedup_opts *opts)
3029 struct btf_dedup *d;
3032 if (!OPTS_VALID(opts, btf_dedup_opts))
3033 return libbpf_err(-EINVAL);
3035 d = btf_dedup_new(btf, opts);
3037 pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
3038 return libbpf_err(-EINVAL);
3041 if (btf_ensure_modifiable(btf)) {
3046 err = btf_dedup_prep(d);
3048 pr_debug("btf_dedup_prep failed:%d\n", err);
3051 err = btf_dedup_strings(d);
3053 pr_debug("btf_dedup_strings failed:%d\n", err);
3056 err = btf_dedup_prim_types(d);
3058 pr_debug("btf_dedup_prim_types failed:%d\n", err);
3061 err = btf_dedup_struct_types(d);
3063 pr_debug("btf_dedup_struct_types failed:%d\n", err);
3066 err = btf_dedup_resolve_fwds(d);
3068 pr_debug("btf_dedup_resolve_fwds failed:%d\n", err);
3071 err = btf_dedup_ref_types(d);
3073 pr_debug("btf_dedup_ref_types failed:%d\n", err);
3076 err = btf_dedup_compact_types(d);
3078 pr_debug("btf_dedup_compact_types failed:%d\n", err);
3081 err = btf_dedup_remap_types(d);
3083 pr_debug("btf_dedup_remap_types failed:%d\n", err);
3089 return libbpf_err(err);
3092 #define BTF_UNPROCESSED_ID ((__u32)-1)
3093 #define BTF_IN_PROGRESS_ID ((__u32)-2)
3096 /* .BTF section to be deduped in-place */
3099 * Optional .BTF.ext section. When provided, any strings referenced
3100 * from it will be taken into account when deduping strings
3102 struct btf_ext *btf_ext;
3104 * This is a map from any type's signature hash to a list of possible
3105 * canonical representative type candidates. Hash collisions are
3106 * ignored, so even types of various kinds can share same list of
3107 * candidates, which is fine because we rely on subsequent
3108 * btf_xxx_equal() checks to authoritatively verify type equality.
3110 struct hashmap *dedup_table;
3111 /* Canonical types map */
3113 /* Hypothetical mapping, used during type graph equivalence checks */
3118 /* Whether hypothetical mapping, if successful, would need to adjust
3119 * already canonicalized types (due to a new forward declaration to
3120 * concrete type resolution). In such case, during split BTF dedup
3121 * candidate type would still be considered as different, because base
3122 * BTF is considered to be immutable.
3124 bool hypot_adjust_canon;
3125 /* Various option modifying behavior of algorithm */
3126 struct btf_dedup_opts opts;
3127 /* temporary strings deduplication state */
3128 struct strset *strs_set;
3131 static long hash_combine(long h, long value)
3133 return h * 31 + value;
3136 #define for_each_dedup_cand(d, node, hash) \
3137 hashmap__for_each_key_entry(d->dedup_table, node, hash)
3139 static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
3141 return hashmap__append(d->dedup_table, hash, type_id);
3144 static int btf_dedup_hypot_map_add(struct btf_dedup *d,
3145 __u32 from_id, __u32 to_id)
3147 if (d->hypot_cnt == d->hypot_cap) {
3150 d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3151 new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3154 d->hypot_list = new_list;
3156 d->hypot_list[d->hypot_cnt++] = from_id;
3157 d->hypot_map[from_id] = to_id;
3161 static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3165 for (i = 0; i < d->hypot_cnt; i++)
3166 d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3168 d->hypot_adjust_canon = false;
3171 static void btf_dedup_free(struct btf_dedup *d)
3173 hashmap__free(d->dedup_table);
3174 d->dedup_table = NULL;
3180 d->hypot_map = NULL;
3182 free(d->hypot_list);
3183 d->hypot_list = NULL;
3188 static size_t btf_dedup_identity_hash_fn(long key, void *ctx)
3193 static size_t btf_dedup_collision_hash_fn(long key, void *ctx)
3198 static bool btf_dedup_equal_fn(long k1, long k2, void *ctx)
3203 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts)
3205 struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3206 hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3207 int i, err = 0, type_cnt;
3210 return ERR_PTR(-ENOMEM);
3212 if (OPTS_GET(opts, force_collisions, false))
3213 hash_fn = btf_dedup_collision_hash_fn;
3216 d->btf_ext = OPTS_GET(opts, btf_ext, NULL);
3218 d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3219 if (IS_ERR(d->dedup_table)) {
3220 err = PTR_ERR(d->dedup_table);
3221 d->dedup_table = NULL;
3225 type_cnt = btf__type_cnt(btf);
3226 d->map = malloc(sizeof(__u32) * type_cnt);
3231 /* special BTF "void" type is made canonical immediately */
3233 for (i = 1; i < type_cnt; i++) {
3234 struct btf_type *t = btf_type_by_id(d->btf, i);
3236 /* VAR and DATASEC are never deduped and are self-canonical */
3237 if (btf_is_var(t) || btf_is_datasec(t))
3240 d->map[i] = BTF_UNPROCESSED_ID;
3243 d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3244 if (!d->hypot_map) {
3248 for (i = 0; i < type_cnt; i++)
3249 d->hypot_map[i] = BTF_UNPROCESSED_ID;
3254 return ERR_PTR(err);
3261 * Iterate over all possible places in .BTF and .BTF.ext that can reference
3262 * string and pass pointer to it to a provided callback `fn`.
3264 static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3268 for (i = 0; i < d->btf->nr_types; i++) {
3269 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3271 r = btf_type_visit_str_offs(t, fn, ctx);
3279 r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3286 static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3288 struct btf_dedup *d = ctx;
3289 __u32 str_off = *str_off_ptr;
3293 /* don't touch empty string or string in main BTF */
3294 if (str_off == 0 || str_off < d->btf->start_str_off)
3297 s = btf__str_by_offset(d->btf, str_off);
3298 if (d->btf->base_btf) {
3299 err = btf__find_str(d->btf->base_btf, s);
3308 off = strset__add_str(d->strs_set, s);
3312 *str_off_ptr = d->btf->start_str_off + off;
3317 * Dedup string and filter out those that are not referenced from either .BTF
3318 * or .BTF.ext (if provided) sections.
3320 * This is done by building index of all strings in BTF's string section,
3321 * then iterating over all entities that can reference strings (e.g., type
3322 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3323 * strings as used. After that all used strings are deduped and compacted into
3324 * sequential blob of memory and new offsets are calculated. Then all the string
3325 * references are iterated again and rewritten using new offsets.
3327 static int btf_dedup_strings(struct btf_dedup *d)
3331 if (d->btf->strs_deduped)
3334 d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3335 if (IS_ERR(d->strs_set)) {
3336 err = PTR_ERR(d->strs_set);
3340 if (!d->btf->base_btf) {
3341 /* insert empty string; we won't be looking it up during strings
3342 * dedup, but it's good to have it for generic BTF string lookups
3344 err = strset__add_str(d->strs_set, "");
3349 /* remap string offsets */
3350 err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3354 /* replace BTF string data and hash with deduped ones */
3355 strset__free(d->btf->strs_set);
3356 d->btf->hdr->str_len = strset__data_size(d->strs_set);
3357 d->btf->strs_set = d->strs_set;
3359 d->btf->strs_deduped = true;
3363 strset__free(d->strs_set);
3369 static long btf_hash_common(struct btf_type *t)
3373 h = hash_combine(0, t->name_off);
3374 h = hash_combine(h, t->info);
3375 h = hash_combine(h, t->size);
3379 static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3381 return t1->name_off == t2->name_off &&
3382 t1->info == t2->info &&
3383 t1->size == t2->size;
3386 /* Calculate type signature hash of INT or TAG. */
3387 static long btf_hash_int_decl_tag(struct btf_type *t)
3389 __u32 info = *(__u32 *)(t + 1);
3392 h = btf_hash_common(t);
3393 h = hash_combine(h, info);
3397 /* Check structural equality of two INTs or TAGs. */
3398 static bool btf_equal_int_tag(struct btf_type *t1, struct btf_type *t2)
3402 if (!btf_equal_common(t1, t2))
3404 info1 = *(__u32 *)(t1 + 1);
3405 info2 = *(__u32 *)(t2 + 1);
3406 return info1 == info2;
3409 /* Calculate type signature hash of ENUM/ENUM64. */
3410 static long btf_hash_enum(struct btf_type *t)
3414 /* don't hash vlen, enum members and size to support enum fwd resolving */
3415 h = hash_combine(0, t->name_off);
3419 static bool btf_equal_enum_members(struct btf_type *t1, struct btf_type *t2)
3421 const struct btf_enum *m1, *m2;
3425 vlen = btf_vlen(t1);
3428 for (i = 0; i < vlen; i++) {
3429 if (m1->name_off != m2->name_off || m1->val != m2->val)
3437 static bool btf_equal_enum64_members(struct btf_type *t1, struct btf_type *t2)
3439 const struct btf_enum64 *m1, *m2;
3443 vlen = btf_vlen(t1);
3444 m1 = btf_enum64(t1);
3445 m2 = btf_enum64(t2);
3446 for (i = 0; i < vlen; i++) {
3447 if (m1->name_off != m2->name_off || m1->val_lo32 != m2->val_lo32 ||
3448 m1->val_hi32 != m2->val_hi32)
3456 /* Check structural equality of two ENUMs or ENUM64s. */
3457 static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3459 if (!btf_equal_common(t1, t2))
3462 /* t1 & t2 kinds are identical because of btf_equal_common */
3463 if (btf_kind(t1) == BTF_KIND_ENUM)
3464 return btf_equal_enum_members(t1, t2);
3466 return btf_equal_enum64_members(t1, t2);
3469 static inline bool btf_is_enum_fwd(struct btf_type *t)
3471 return btf_is_any_enum(t) && btf_vlen(t) == 0;
3474 static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3476 if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3477 return btf_equal_enum(t1, t2);
3478 /* At this point either t1 or t2 or both are forward declarations, thus:
3479 * - skip comparing vlen because it is zero for forward declarations;
3480 * - skip comparing size to allow enum forward declarations
3481 * to be compatible with enum64 full declarations;
3482 * - skip comparing kind for the same reason.
3484 return t1->name_off == t2->name_off &&
3485 btf_is_any_enum(t1) && btf_is_any_enum(t2);
3489 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3490 * as referenced type IDs equivalence is established separately during type
3491 * graph equivalence check algorithm.
3493 static long btf_hash_struct(struct btf_type *t)
3495 const struct btf_member *member = btf_members(t);
3496 __u32 vlen = btf_vlen(t);
3497 long h = btf_hash_common(t);
3500 for (i = 0; i < vlen; i++) {
3501 h = hash_combine(h, member->name_off);
3502 h = hash_combine(h, member->offset);
3503 /* no hashing of referenced type ID, it can be unresolved yet */
3510 * Check structural compatibility of two STRUCTs/UNIONs, ignoring referenced
3511 * type IDs. This check is performed during type graph equivalence check and
3512 * referenced types equivalence is checked separately.
3514 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3516 const struct btf_member *m1, *m2;
3520 if (!btf_equal_common(t1, t2))
3523 vlen = btf_vlen(t1);
3524 m1 = btf_members(t1);
3525 m2 = btf_members(t2);
3526 for (i = 0; i < vlen; i++) {
3527 if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3536 * Calculate type signature hash of ARRAY, including referenced type IDs,
3537 * under assumption that they were already resolved to canonical type IDs and
3538 * are not going to change.
3540 static long btf_hash_array(struct btf_type *t)
3542 const struct btf_array *info = btf_array(t);
3543 long h = btf_hash_common(t);
3545 h = hash_combine(h, info->type);
3546 h = hash_combine(h, info->index_type);
3547 h = hash_combine(h, info->nelems);
3552 * Check exact equality of two ARRAYs, taking into account referenced
3553 * type IDs, under assumption that they were already resolved to canonical
3554 * type IDs and are not going to change.
3555 * This function is called during reference types deduplication to compare
3556 * ARRAY to potential canonical representative.
3558 static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3560 const struct btf_array *info1, *info2;
3562 if (!btf_equal_common(t1, t2))
3565 info1 = btf_array(t1);
3566 info2 = btf_array(t2);
3567 return info1->type == info2->type &&
3568 info1->index_type == info2->index_type &&
3569 info1->nelems == info2->nelems;
3573 * Check structural compatibility of two ARRAYs, ignoring referenced type
3574 * IDs. This check is performed during type graph equivalence check and
3575 * referenced types equivalence is checked separately.
3577 static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
3579 if (!btf_equal_common(t1, t2))
3582 return btf_array(t1)->nelems == btf_array(t2)->nelems;
3586 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
3587 * under assumption that they were already resolved to canonical type IDs and
3588 * are not going to change.
3590 static long btf_hash_fnproto(struct btf_type *t)
3592 const struct btf_param *member = btf_params(t);
3593 __u16 vlen = btf_vlen(t);
3594 long h = btf_hash_common(t);
3597 for (i = 0; i < vlen; i++) {
3598 h = hash_combine(h, member->name_off);
3599 h = hash_combine(h, member->type);
3606 * Check exact equality of two FUNC_PROTOs, taking into account referenced
3607 * type IDs, under assumption that they were already resolved to canonical
3608 * type IDs and are not going to change.
3609 * This function is called during reference types deduplication to compare
3610 * FUNC_PROTO to potential canonical representative.
3612 static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
3614 const struct btf_param *m1, *m2;
3618 if (!btf_equal_common(t1, t2))
3621 vlen = btf_vlen(t1);
3622 m1 = btf_params(t1);
3623 m2 = btf_params(t2);
3624 for (i = 0; i < vlen; i++) {
3625 if (m1->name_off != m2->name_off || m1->type != m2->type)
3634 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3635 * IDs. This check is performed during type graph equivalence check and
3636 * referenced types equivalence is checked separately.
3638 static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
3640 const struct btf_param *m1, *m2;
3644 /* skip return type ID */
3645 if (t1->name_off != t2->name_off || t1->info != t2->info)
3648 vlen = btf_vlen(t1);
3649 m1 = btf_params(t1);
3650 m2 = btf_params(t2);
3651 for (i = 0; i < vlen; i++) {
3652 if (m1->name_off != m2->name_off)
3660 /* Prepare split BTF for deduplication by calculating hashes of base BTF's
3661 * types and initializing the rest of the state (canonical type mapping) for
3662 * the fixed base BTF part.
3664 static int btf_dedup_prep(struct btf_dedup *d)
3670 if (!d->btf->base_btf)
3673 for (type_id = 1; type_id < d->btf->start_id; type_id++) {
3674 t = btf_type_by_id(d->btf, type_id);
3676 /* all base BTF types are self-canonical by definition */
3677 d->map[type_id] = type_id;
3679 switch (btf_kind(t)) {
3681 case BTF_KIND_DATASEC:
3682 /* VAR and DATASEC are never hash/deduplicated */
3684 case BTF_KIND_CONST:
3685 case BTF_KIND_VOLATILE:
3686 case BTF_KIND_RESTRICT:
3689 case BTF_KIND_TYPEDEF:
3691 case BTF_KIND_FLOAT:
3692 case BTF_KIND_TYPE_TAG:
3693 h = btf_hash_common(t);
3696 case BTF_KIND_DECL_TAG:
3697 h = btf_hash_int_decl_tag(t);
3700 case BTF_KIND_ENUM64:
3701 h = btf_hash_enum(t);
3703 case BTF_KIND_STRUCT:
3704 case BTF_KIND_UNION:
3705 h = btf_hash_struct(t);
3707 case BTF_KIND_ARRAY:
3708 h = btf_hash_array(t);
3710 case BTF_KIND_FUNC_PROTO:
3711 h = btf_hash_fnproto(t);
3714 pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
3717 if (btf_dedup_table_add(d, h, type_id))
3725 * Deduplicate primitive types, that can't reference other types, by calculating
3726 * their type signature hash and comparing them with any possible canonical
3727 * candidate. If no canonical candidate matches, type itself is marked as
3728 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
3730 static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
3732 struct btf_type *t = btf_type_by_id(d->btf, type_id);
3733 struct hashmap_entry *hash_entry;
3734 struct btf_type *cand;
3735 /* if we don't find equivalent type, then we are canonical */
3736 __u32 new_id = type_id;
3740 switch (btf_kind(t)) {
3741 case BTF_KIND_CONST:
3742 case BTF_KIND_VOLATILE:
3743 case BTF_KIND_RESTRICT:
3745 case BTF_KIND_TYPEDEF:
3746 case BTF_KIND_ARRAY:
3747 case BTF_KIND_STRUCT:
3748 case BTF_KIND_UNION:
3750 case BTF_KIND_FUNC_PROTO:
3752 case BTF_KIND_DATASEC:
3753 case BTF_KIND_DECL_TAG:
3754 case BTF_KIND_TYPE_TAG:
3758 h = btf_hash_int_decl_tag(t);
3759 for_each_dedup_cand(d, hash_entry, h) {
3760 cand_id = hash_entry->value;
3761 cand = btf_type_by_id(d->btf, cand_id);
3762 if (btf_equal_int_tag(t, cand)) {
3770 case BTF_KIND_ENUM64:
3771 h = btf_hash_enum(t);
3772 for_each_dedup_cand(d, hash_entry, h) {
3773 cand_id = hash_entry->value;
3774 cand = btf_type_by_id(d->btf, cand_id);
3775 if (btf_equal_enum(t, cand)) {
3779 if (btf_compat_enum(t, cand)) {
3780 if (btf_is_enum_fwd(t)) {
3781 /* resolve fwd to full enum */
3785 /* resolve canonical enum fwd to full enum */
3786 d->map[cand_id] = type_id;
3792 case BTF_KIND_FLOAT:
3793 h = btf_hash_common(t);
3794 for_each_dedup_cand(d, hash_entry, h) {
3795 cand_id = hash_entry->value;
3796 cand = btf_type_by_id(d->btf, cand_id);
3797 if (btf_equal_common(t, cand)) {
3808 d->map[type_id] = new_id;
3809 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
3815 static int btf_dedup_prim_types(struct btf_dedup *d)
3819 for (i = 0; i < d->btf->nr_types; i++) {
3820 err = btf_dedup_prim_type(d, d->btf->start_id + i);
3828 * Check whether type is already mapped into canonical one (could be to itself).
3830 static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
3832 return d->map[type_id] <= BTF_MAX_NR_TYPES;
3836 * Resolve type ID into its canonical type ID, if any; otherwise return original
3837 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
3838 * STRUCT/UNION link and resolve it into canonical type ID as well.
3840 static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
3842 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3843 type_id = d->map[type_id];
3848 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
3851 static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
3853 __u32 orig_type_id = type_id;
3855 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3858 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3859 type_id = d->map[type_id];
3861 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3864 return orig_type_id;
3868 static inline __u16 btf_fwd_kind(struct btf_type *t)
3870 return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
3873 /* Check if given two types are identical ARRAY definitions */
3874 static bool btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
3876 struct btf_type *t1, *t2;
3878 t1 = btf_type_by_id(d->btf, id1);
3879 t2 = btf_type_by_id(d->btf, id2);
3880 if (!btf_is_array(t1) || !btf_is_array(t2))
3883 return btf_equal_array(t1, t2);
3886 /* Check if given two types are identical STRUCT/UNION definitions */
3887 static bool btf_dedup_identical_structs(struct btf_dedup *d, __u32 id1, __u32 id2)
3889 const struct btf_member *m1, *m2;
3890 struct btf_type *t1, *t2;
3893 t1 = btf_type_by_id(d->btf, id1);
3894 t2 = btf_type_by_id(d->btf, id2);
3896 if (!btf_is_composite(t1) || btf_kind(t1) != btf_kind(t2))
3899 if (!btf_shallow_equal_struct(t1, t2))
3902 m1 = btf_members(t1);
3903 m2 = btf_members(t2);
3904 for (i = 0, n = btf_vlen(t1); i < n; i++, m1++, m2++) {
3905 if (m1->type != m2->type &&
3906 !btf_dedup_identical_arrays(d, m1->type, m2->type) &&
3907 !btf_dedup_identical_structs(d, m1->type, m2->type))
3914 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
3915 * call it "candidate graph" in this description for brevity) to a type graph
3916 * formed by (potential) canonical struct/union ("canonical graph" for brevity
3917 * here, though keep in mind that not all types in canonical graph are
3918 * necessarily canonical representatives themselves, some of them might be
3919 * duplicates or its uniqueness might not have been established yet).
3921 * - >0, if type graphs are equivalent;
3922 * - 0, if not equivalent;
3925 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
3926 * equivalence of BTF types at each step. If at any point BTF types in candidate
3927 * and canonical graphs are not compatible structurally, whole graphs are
3928 * incompatible. If types are structurally equivalent (i.e., all information
3929 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
3930 * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
3931 * If a type references other types, then those referenced types are checked
3932 * for equivalence recursively.
3934 * During DFS traversal, if we find that for current `canon_id` type we
3935 * already have some mapping in hypothetical map, we check for two possible
3937 * - `canon_id` is mapped to exactly the same type as `cand_id`. This will
3938 * happen when type graphs have cycles. In this case we assume those two
3939 * types are equivalent.
3940 * - `canon_id` is mapped to different type. This is contradiction in our
3941 * hypothetical mapping, because same graph in canonical graph corresponds
3942 * to two different types in candidate graph, which for equivalent type
3943 * graphs shouldn't happen. This condition terminates equivalence check
3944 * with negative result.
3946 * If type graphs traversal exhausts types to check and find no contradiction,
3947 * then type graphs are equivalent.
3949 * When checking types for equivalence, there is one special case: FWD types.
3950 * If FWD type resolution is allowed and one of the types (either from canonical
3951 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
3952 * flag) and their names match, hypothetical mapping is updated to point from
3953 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
3954 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
3956 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
3957 * if there are two exactly named (or anonymous) structs/unions that are
3958 * compatible structurally, one of which has FWD field, while other is concrete
3959 * STRUCT/UNION, but according to C sources they are different structs/unions
3960 * that are referencing different types with the same name. This is extremely
3961 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
3962 * this logic is causing problems.
3964 * Doing FWD resolution means that both candidate and/or canonical graphs can
3965 * consists of portions of the graph that come from multiple compilation units.
3966 * This is due to the fact that types within single compilation unit are always
3967 * deduplicated and FWDs are already resolved, if referenced struct/union
3968 * definiton is available. So, if we had unresolved FWD and found corresponding
3969 * STRUCT/UNION, they will be from different compilation units. This
3970 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
3971 * type graph will likely have at least two different BTF types that describe
3972 * same type (e.g., most probably there will be two different BTF types for the
3973 * same 'int' primitive type) and could even have "overlapping" parts of type
3974 * graph that describe same subset of types.
3976 * This in turn means that our assumption that each type in canonical graph
3977 * must correspond to exactly one type in candidate graph might not hold
3978 * anymore and will make it harder to detect contradictions using hypothetical
3979 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
3980 * resolution only in canonical graph. FWDs in candidate graphs are never
3981 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
3983 * - Both types in canonical and candidate graphs are FWDs. If they are
3984 * structurally equivalent, then they can either be both resolved to the
3985 * same STRUCT/UNION or not resolved at all. In both cases they are
3986 * equivalent and there is no need to resolve FWD on candidate side.
3987 * - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
3988 * so nothing to resolve as well, algorithm will check equivalence anyway.
3989 * - Type in canonical graph is FWD, while type in candidate is concrete
3990 * STRUCT/UNION. In this case candidate graph comes from single compilation
3991 * unit, so there is exactly one BTF type for each unique C type. After
3992 * resolving FWD into STRUCT/UNION, there might be more than one BTF type
3993 * in canonical graph mapping to single BTF type in candidate graph, but
3994 * because hypothetical mapping maps from canonical to candidate types, it's
3995 * alright, and we still maintain the property of having single `canon_id`
3996 * mapping to single `cand_id` (there could be two different `canon_id`
3997 * mapped to the same `cand_id`, but it's not contradictory).
3998 * - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
3999 * graph is FWD. In this case we are just going to check compatibility of
4000 * STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
4001 * assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
4002 * a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
4003 * turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
4006 static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
4009 struct btf_type *cand_type;
4010 struct btf_type *canon_type;
4011 __u32 hypot_type_id;
4016 /* if both resolve to the same canonical, they must be equivalent */
4017 if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
4020 canon_id = resolve_fwd_id(d, canon_id);
4022 hypot_type_id = d->hypot_map[canon_id];
4023 if (hypot_type_id <= BTF_MAX_NR_TYPES) {
4024 if (hypot_type_id == cand_id)
4026 /* In some cases compiler will generate different DWARF types
4027 * for *identical* array type definitions and use them for
4028 * different fields within the *same* struct. This breaks type
4029 * equivalence check, which makes an assumption that candidate
4030 * types sub-graph has a consistent and deduped-by-compiler
4031 * types within a single CU. So work around that by explicitly
4032 * allowing identical array types here.
4034 if (btf_dedup_identical_arrays(d, hypot_type_id, cand_id))
4036 /* It turns out that similar situation can happen with
4037 * struct/union sometimes, sigh... Handle the case where
4038 * structs/unions are exactly the same, down to the referenced
4039 * type IDs. Anything more complicated (e.g., if referenced
4040 * types are different, but equivalent) is *way more*
4041 * complicated and requires a many-to-many equivalence mapping.
4043 if (btf_dedup_identical_structs(d, hypot_type_id, cand_id))
4048 if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
4051 cand_type = btf_type_by_id(d->btf, cand_id);
4052 canon_type = btf_type_by_id(d->btf, canon_id);
4053 cand_kind = btf_kind(cand_type);
4054 canon_kind = btf_kind(canon_type);
4056 if (cand_type->name_off != canon_type->name_off)
4059 /* FWD <--> STRUCT/UNION equivalence check, if enabled */
4060 if ((cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
4061 && cand_kind != canon_kind) {
4065 if (cand_kind == BTF_KIND_FWD) {
4066 real_kind = canon_kind;
4067 fwd_kind = btf_fwd_kind(cand_type);
4069 real_kind = cand_kind;
4070 fwd_kind = btf_fwd_kind(canon_type);
4071 /* we'd need to resolve base FWD to STRUCT/UNION */
4072 if (fwd_kind == real_kind && canon_id < d->btf->start_id)
4073 d->hypot_adjust_canon = true;
4075 return fwd_kind == real_kind;
4078 if (cand_kind != canon_kind)
4081 switch (cand_kind) {
4083 return btf_equal_int_tag(cand_type, canon_type);
4086 case BTF_KIND_ENUM64:
4087 return btf_compat_enum(cand_type, canon_type);
4090 case BTF_KIND_FLOAT:
4091 return btf_equal_common(cand_type, canon_type);
4093 case BTF_KIND_CONST:
4094 case BTF_KIND_VOLATILE:
4095 case BTF_KIND_RESTRICT:
4097 case BTF_KIND_TYPEDEF:
4099 case BTF_KIND_TYPE_TAG:
4100 if (cand_type->info != canon_type->info)
4102 return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4104 case BTF_KIND_ARRAY: {
4105 const struct btf_array *cand_arr, *canon_arr;
4107 if (!btf_compat_array(cand_type, canon_type))
4109 cand_arr = btf_array(cand_type);
4110 canon_arr = btf_array(canon_type);
4111 eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
4114 return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
4117 case BTF_KIND_STRUCT:
4118 case BTF_KIND_UNION: {
4119 const struct btf_member *cand_m, *canon_m;
4122 if (!btf_shallow_equal_struct(cand_type, canon_type))
4124 vlen = btf_vlen(cand_type);
4125 cand_m = btf_members(cand_type);
4126 canon_m = btf_members(canon_type);
4127 for (i = 0; i < vlen; i++) {
4128 eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
4138 case BTF_KIND_FUNC_PROTO: {
4139 const struct btf_param *cand_p, *canon_p;
4142 if (!btf_compat_fnproto(cand_type, canon_type))
4144 eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4147 vlen = btf_vlen(cand_type);
4148 cand_p = btf_params(cand_type);
4149 canon_p = btf_params(canon_type);
4150 for (i = 0; i < vlen; i++) {
4151 eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
4167 * Use hypothetical mapping, produced by successful type graph equivalence
4168 * check, to augment existing struct/union canonical mapping, where possible.
4170 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
4171 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
4172 * it doesn't matter if FWD type was part of canonical graph or candidate one,
4173 * we are recording the mapping anyway. As opposed to carefulness required
4174 * for struct/union correspondence mapping (described below), for FWD resolution
4175 * it's not important, as by the time that FWD type (reference type) will be
4176 * deduplicated all structs/unions will be deduped already anyway.
4178 * Recording STRUCT/UNION mapping is purely a performance optimization and is
4179 * not required for correctness. It needs to be done carefully to ensure that
4180 * struct/union from candidate's type graph is not mapped into corresponding
4181 * struct/union from canonical type graph that itself hasn't been resolved into
4182 * canonical representative. The only guarantee we have is that canonical
4183 * struct/union was determined as canonical and that won't change. But any
4184 * types referenced through that struct/union fields could have been not yet
4185 * resolved, so in case like that it's too early to establish any kind of
4186 * correspondence between structs/unions.
4188 * No canonical correspondence is derived for primitive types (they are already
4189 * deduplicated completely already anyway) or reference types (they rely on
4190 * stability of struct/union canonical relationship for equivalence checks).
4192 static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
4194 __u32 canon_type_id, targ_type_id;
4195 __u16 t_kind, c_kind;
4199 for (i = 0; i < d->hypot_cnt; i++) {
4200 canon_type_id = d->hypot_list[i];
4201 targ_type_id = d->hypot_map[canon_type_id];
4202 t_id = resolve_type_id(d, targ_type_id);
4203 c_id = resolve_type_id(d, canon_type_id);
4204 t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
4205 c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
4207 * Resolve FWD into STRUCT/UNION.
4208 * It's ok to resolve FWD into STRUCT/UNION that's not yet
4209 * mapped to canonical representative (as opposed to
4210 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
4211 * eventually that struct is going to be mapped and all resolved
4212 * FWDs will automatically resolve to correct canonical
4213 * representative. This will happen before ref type deduping,
4214 * which critically depends on stability of these mapping. This
4215 * stability is not a requirement for STRUCT/UNION equivalence
4219 /* if it's the split BTF case, we still need to point base FWD
4220 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4221 * will be resolved against base FWD. If we don't point base
4222 * canonical FWD to the resolved STRUCT/UNION, then all the
4223 * FWDs in split BTF won't be correctly resolved to a proper
4226 if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4227 d->map[c_id] = t_id;
4229 /* if graph equivalence determined that we'd need to adjust
4230 * base canonical types, then we need to only point base FWDs
4231 * to STRUCTs/UNIONs and do no more modifications. For all
4232 * other purposes the type graphs were not equivalent.
4234 if (d->hypot_adjust_canon)
4237 if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4238 d->map[t_id] = c_id;
4240 if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4241 c_kind != BTF_KIND_FWD &&
4242 is_type_mapped(d, c_id) &&
4243 !is_type_mapped(d, t_id)) {
4245 * as a perf optimization, we can map struct/union
4246 * that's part of type graph we just verified for
4247 * equivalence. We can do that for struct/union that has
4248 * canonical representative only, though.
4250 d->map[t_id] = c_id;
4256 * Deduplicate struct/union types.
4258 * For each struct/union type its type signature hash is calculated, taking
4259 * into account type's name, size, number, order and names of fields, but
4260 * ignoring type ID's referenced from fields, because they might not be deduped
4261 * completely until after reference types deduplication phase. This type hash
4262 * is used to iterate over all potential canonical types, sharing same hash.
4263 * For each canonical candidate we check whether type graphs that they form
4264 * (through referenced types in fields and so on) are equivalent using algorithm
4265 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4266 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4267 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4268 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4269 * potentially map other structs/unions to their canonical representatives,
4270 * if such relationship hasn't yet been established. This speeds up algorithm
4271 * by eliminating some of the duplicate work.
4273 * If no matching canonical representative was found, struct/union is marked
4274 * as canonical for itself and is added into btf_dedup->dedup_table hash map
4275 * for further look ups.
4277 static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4279 struct btf_type *cand_type, *t;
4280 struct hashmap_entry *hash_entry;
4281 /* if we don't find equivalent type, then we are canonical */
4282 __u32 new_id = type_id;
4286 /* already deduped or is in process of deduping (loop detected) */
4287 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4290 t = btf_type_by_id(d->btf, type_id);
4293 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4296 h = btf_hash_struct(t);
4297 for_each_dedup_cand(d, hash_entry, h) {
4298 __u32 cand_id = hash_entry->value;
4302 * Even though btf_dedup_is_equiv() checks for
4303 * btf_shallow_equal_struct() internally when checking two
4304 * structs (unions) for equivalence, we need to guard here
4305 * from picking matching FWD type as a dedup candidate.
4306 * This can happen due to hash collision. In such case just
4307 * relying on btf_dedup_is_equiv() would lead to potentially
4308 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4309 * FWD and compatible STRUCT/UNION are considered equivalent.
4311 cand_type = btf_type_by_id(d->btf, cand_id);
4312 if (!btf_shallow_equal_struct(t, cand_type))
4315 btf_dedup_clear_hypot_map(d);
4316 eq = btf_dedup_is_equiv(d, type_id, cand_id);
4321 btf_dedup_merge_hypot_map(d);
4322 if (d->hypot_adjust_canon) /* not really equivalent */
4328 d->map[type_id] = new_id;
4329 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4335 static int btf_dedup_struct_types(struct btf_dedup *d)
4339 for (i = 0; i < d->btf->nr_types; i++) {
4340 err = btf_dedup_struct_type(d, d->btf->start_id + i);
4348 * Deduplicate reference type.
4350 * Once all primitive and struct/union types got deduplicated, we can easily
4351 * deduplicate all other (reference) BTF types. This is done in two steps:
4353 * 1. Resolve all referenced type IDs into their canonical type IDs. This
4354 * resolution can be done either immediately for primitive or struct/union types
4355 * (because they were deduped in previous two phases) or recursively for
4356 * reference types. Recursion will always terminate at either primitive or
4357 * struct/union type, at which point we can "unwind" chain of reference types
4358 * one by one. There is no danger of encountering cycles because in C type
4359 * system the only way to form type cycle is through struct/union, so any chain
4360 * of reference types, even those taking part in a type cycle, will inevitably
4361 * reach struct/union at some point.
4363 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4364 * becomes "stable", in the sense that no further deduplication will cause
4365 * any changes to it. With that, it's now possible to calculate type's signature
4366 * hash (this time taking into account referenced type IDs) and loop over all
4367 * potential canonical representatives. If no match was found, current type
4368 * will become canonical representative of itself and will be added into
4369 * btf_dedup->dedup_table as another possible canonical representative.
4371 static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4373 struct hashmap_entry *hash_entry;
4374 __u32 new_id = type_id, cand_id;
4375 struct btf_type *t, *cand;
4376 /* if we don't find equivalent type, then we are representative type */
4380 if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4382 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4383 return resolve_type_id(d, type_id);
4385 t = btf_type_by_id(d->btf, type_id);
4386 d->map[type_id] = BTF_IN_PROGRESS_ID;
4388 switch (btf_kind(t)) {
4389 case BTF_KIND_CONST:
4390 case BTF_KIND_VOLATILE:
4391 case BTF_KIND_RESTRICT:
4393 case BTF_KIND_TYPEDEF:
4395 case BTF_KIND_TYPE_TAG:
4396 ref_type_id = btf_dedup_ref_type(d, t->type);
4397 if (ref_type_id < 0)
4399 t->type = ref_type_id;
4401 h = btf_hash_common(t);
4402 for_each_dedup_cand(d, hash_entry, h) {
4403 cand_id = hash_entry->value;
4404 cand = btf_type_by_id(d->btf, cand_id);
4405 if (btf_equal_common(t, cand)) {
4412 case BTF_KIND_DECL_TAG:
4413 ref_type_id = btf_dedup_ref_type(d, t->type);
4414 if (ref_type_id < 0)
4416 t->type = ref_type_id;
4418 h = btf_hash_int_decl_tag(t);
4419 for_each_dedup_cand(d, hash_entry, h) {
4420 cand_id = hash_entry->value;
4421 cand = btf_type_by_id(d->btf, cand_id);
4422 if (btf_equal_int_tag(t, cand)) {
4429 case BTF_KIND_ARRAY: {
4430 struct btf_array *info = btf_array(t);
4432 ref_type_id = btf_dedup_ref_type(d, info->type);
4433 if (ref_type_id < 0)
4435 info->type = ref_type_id;
4437 ref_type_id = btf_dedup_ref_type(d, info->index_type);
4438 if (ref_type_id < 0)
4440 info->index_type = ref_type_id;
4442 h = btf_hash_array(t);
4443 for_each_dedup_cand(d, hash_entry, h) {
4444 cand_id = hash_entry->value;
4445 cand = btf_type_by_id(d->btf, cand_id);
4446 if (btf_equal_array(t, cand)) {
4454 case BTF_KIND_FUNC_PROTO: {
4455 struct btf_param *param;
4459 ref_type_id = btf_dedup_ref_type(d, t->type);
4460 if (ref_type_id < 0)
4462 t->type = ref_type_id;
4465 param = btf_params(t);
4466 for (i = 0; i < vlen; i++) {
4467 ref_type_id = btf_dedup_ref_type(d, param->type);
4468 if (ref_type_id < 0)
4470 param->type = ref_type_id;
4474 h = btf_hash_fnproto(t);
4475 for_each_dedup_cand(d, hash_entry, h) {
4476 cand_id = hash_entry->value;
4477 cand = btf_type_by_id(d->btf, cand_id);
4478 if (btf_equal_fnproto(t, cand)) {
4490 d->map[type_id] = new_id;
4491 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4497 static int btf_dedup_ref_types(struct btf_dedup *d)
4501 for (i = 0; i < d->btf->nr_types; i++) {
4502 err = btf_dedup_ref_type(d, d->btf->start_id + i);
4506 /* we won't need d->dedup_table anymore */
4507 hashmap__free(d->dedup_table);
4508 d->dedup_table = NULL;
4513 * Collect a map from type names to type ids for all canonical structs
4514 * and unions. If the same name is shared by several canonical types
4515 * use a special value 0 to indicate this fact.
4517 static int btf_dedup_fill_unique_names_map(struct btf_dedup *d, struct hashmap *names_map)
4519 __u32 nr_types = btf__type_cnt(d->btf);
4526 * Iterate over base and split module ids in order to get all
4527 * available structs in the map.
4529 for (type_id = 1; type_id < nr_types; ++type_id) {
4530 t = btf_type_by_id(d->btf, type_id);
4533 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4536 /* Skip non-canonical types */
4537 if (type_id != d->map[type_id])
4540 err = hashmap__add(names_map, t->name_off, type_id);
4542 err = hashmap__set(names_map, t->name_off, 0, NULL, NULL);
4551 static int btf_dedup_resolve_fwd(struct btf_dedup *d, struct hashmap *names_map, __u32 type_id)
4553 struct btf_type *t = btf_type_by_id(d->btf, type_id);
4554 enum btf_fwd_kind fwd_kind = btf_kflag(t);
4555 __u16 cand_kind, kind = btf_kind(t);
4556 struct btf_type *cand_t;
4559 if (kind != BTF_KIND_FWD)
4562 /* Skip if this FWD already has a mapping */
4563 if (type_id != d->map[type_id])
4566 if (!hashmap__find(names_map, t->name_off, &cand_id))
4569 /* Zero is a special value indicating that name is not unique */
4573 cand_t = btf_type_by_id(d->btf, cand_id);
4574 cand_kind = btf_kind(cand_t);
4575 if ((cand_kind == BTF_KIND_STRUCT && fwd_kind != BTF_FWD_STRUCT) ||
4576 (cand_kind == BTF_KIND_UNION && fwd_kind != BTF_FWD_UNION))
4579 d->map[type_id] = cand_id;
4585 * Resolve unambiguous forward declarations.
4587 * The lion's share of all FWD declarations is resolved during
4588 * `btf_dedup_struct_types` phase when different type graphs are
4589 * compared against each other. However, if in some compilation unit a
4590 * FWD declaration is not a part of a type graph compared against
4591 * another type graph that declaration's canonical type would not be
4597 * struct foo *some_global;
4601 * struct foo { int u; };
4602 * struct foo *another_global;
4604 * After `btf_dedup_struct_types` the BTF looks as follows:
4606 * [1] STRUCT 'foo' size=4 vlen=1 ...
4607 * [2] INT 'int' size=4 ...
4608 * [3] PTR '(anon)' type_id=1
4609 * [4] FWD 'foo' fwd_kind=struct
4610 * [5] PTR '(anon)' type_id=4
4612 * This pass assumes that such FWD declarations should be mapped to
4613 * structs or unions with identical name in case if the name is not
4616 static int btf_dedup_resolve_fwds(struct btf_dedup *d)
4619 struct hashmap *names_map;
4621 names_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
4622 if (IS_ERR(names_map))
4623 return PTR_ERR(names_map);
4625 err = btf_dedup_fill_unique_names_map(d, names_map);
4629 for (i = 0; i < d->btf->nr_types; i++) {
4630 err = btf_dedup_resolve_fwd(d, names_map, d->btf->start_id + i);
4636 hashmap__free(names_map);
4643 * After we established for each type its corresponding canonical representative
4644 * type, we now can eliminate types that are not canonical and leave only
4645 * canonical ones layed out sequentially in memory by copying them over
4646 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
4647 * a map from original type ID to a new compacted type ID, which will be used
4648 * during next phase to "fix up" type IDs, referenced from struct/union and
4651 static int btf_dedup_compact_types(struct btf_dedup *d)
4654 __u32 next_type_id = d->btf->start_id;
4655 const struct btf_type *t;
4659 /* we are going to reuse hypot_map to store compaction remapping */
4660 d->hypot_map[0] = 0;
4661 /* base BTF types are not renumbered */
4662 for (id = 1; id < d->btf->start_id; id++)
4663 d->hypot_map[id] = id;
4664 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
4665 d->hypot_map[id] = BTF_UNPROCESSED_ID;
4667 p = d->btf->types_data;
4669 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
4670 if (d->map[id] != id)
4673 t = btf__type_by_id(d->btf, id);
4674 len = btf_type_size(t);
4679 d->hypot_map[id] = next_type_id;
4680 d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
4685 /* shrink struct btf's internal types index and update btf_header */
4686 d->btf->nr_types = next_type_id - d->btf->start_id;
4687 d->btf->type_offs_cap = d->btf->nr_types;
4688 d->btf->hdr->type_len = p - d->btf->types_data;
4689 new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
4691 if (d->btf->type_offs_cap && !new_offs)
4693 d->btf->type_offs = new_offs;
4694 d->btf->hdr->str_off = d->btf->hdr->type_len;
4695 d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
4700 * Figure out final (deduplicated and compacted) type ID for provided original
4701 * `type_id` by first resolving it into corresponding canonical type ID and
4702 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
4703 * which is populated during compaction phase.
4705 static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
4707 struct btf_dedup *d = ctx;
4708 __u32 resolved_type_id, new_type_id;
4710 resolved_type_id = resolve_type_id(d, *type_id);
4711 new_type_id = d->hypot_map[resolved_type_id];
4712 if (new_type_id > BTF_MAX_NR_TYPES)
4715 *type_id = new_type_id;
4720 * Remap referenced type IDs into deduped type IDs.
4722 * After BTF types are deduplicated and compacted, their final type IDs may
4723 * differ from original ones. The map from original to a corresponding
4724 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
4725 * compaction phase. During remapping phase we are rewriting all type IDs
4726 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
4727 * their final deduped type IDs.
4729 static int btf_dedup_remap_types(struct btf_dedup *d)
4733 for (i = 0; i < d->btf->nr_types; i++) {
4734 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
4736 r = btf_type_visit_type_ids(t, btf_dedup_remap_type_id, d);
4744 r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
4752 * Probe few well-known locations for vmlinux kernel image and try to load BTF
4753 * data out of it to use for target BTF.
4755 struct btf *btf__load_vmlinux_btf(void)
4757 const char *locations[] = {
4758 /* try canonical vmlinux BTF through sysfs first */
4759 "/sys/kernel/btf/vmlinux",
4760 /* fall back to trying to find vmlinux on disk otherwise */
4761 "/boot/vmlinux-%1$s",
4762 "/lib/modules/%1$s/vmlinux-%1$s",
4763 "/lib/modules/%1$s/build/vmlinux",
4764 "/usr/lib/modules/%1$s/kernel/vmlinux",
4765 "/usr/lib/debug/boot/vmlinux-%1$s",
4766 "/usr/lib/debug/boot/vmlinux-%1$s.debug",
4767 "/usr/lib/debug/lib/modules/%1$s/vmlinux",
4769 char path[PATH_MAX + 1];
4776 for (i = 0; i < ARRAY_SIZE(locations); i++) {
4777 snprintf(path, PATH_MAX, locations[i], buf.release);
4779 if (faccessat(AT_FDCWD, path, R_OK, AT_EACCESS))
4782 btf = btf__parse(path, NULL);
4783 err = libbpf_get_error(btf);
4784 pr_debug("loading kernel BTF '%s': %d\n", path, err);
4791 pr_warn("failed to find valid kernel BTF\n");
4792 return libbpf_err_ptr(-ESRCH);
4795 struct btf *libbpf_find_kernel_btf(void) __attribute__((alias("btf__load_vmlinux_btf")));
4797 struct btf *btf__load_module_btf(const char *module_name, struct btf *vmlinux_btf)
4801 snprintf(path, sizeof(path), "/sys/kernel/btf/%s", module_name);
4802 return btf__parse_split(path, vmlinux_btf);
4805 int btf_type_visit_type_ids(struct btf_type *t, type_id_visit_fn visit, void *ctx)
4809 switch (btf_kind(t)) {
4811 case BTF_KIND_FLOAT:
4813 case BTF_KIND_ENUM64:
4817 case BTF_KIND_CONST:
4818 case BTF_KIND_VOLATILE:
4819 case BTF_KIND_RESTRICT:
4821 case BTF_KIND_TYPEDEF:
4824 case BTF_KIND_DECL_TAG:
4825 case BTF_KIND_TYPE_TAG:
4826 return visit(&t->type, ctx);
4828 case BTF_KIND_ARRAY: {
4829 struct btf_array *a = btf_array(t);
4831 err = visit(&a->type, ctx);
4832 err = err ?: visit(&a->index_type, ctx);
4836 case BTF_KIND_STRUCT:
4837 case BTF_KIND_UNION: {
4838 struct btf_member *m = btf_members(t);
4840 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4841 err = visit(&m->type, ctx);
4848 case BTF_KIND_FUNC_PROTO: {
4849 struct btf_param *m = btf_params(t);
4851 err = visit(&t->type, ctx);
4854 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4855 err = visit(&m->type, ctx);
4862 case BTF_KIND_DATASEC: {
4863 struct btf_var_secinfo *m = btf_var_secinfos(t);
4865 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4866 err = visit(&m->type, ctx);
4878 int btf_type_visit_str_offs(struct btf_type *t, str_off_visit_fn visit, void *ctx)
4882 err = visit(&t->name_off, ctx);
4886 switch (btf_kind(t)) {
4887 case BTF_KIND_STRUCT:
4888 case BTF_KIND_UNION: {
4889 struct btf_member *m = btf_members(t);
4891 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4892 err = visit(&m->name_off, ctx);
4898 case BTF_KIND_ENUM: {
4899 struct btf_enum *m = btf_enum(t);
4901 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4902 err = visit(&m->name_off, ctx);
4908 case BTF_KIND_ENUM64: {
4909 struct btf_enum64 *m = btf_enum64(t);
4911 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4912 err = visit(&m->name_off, ctx);
4918 case BTF_KIND_FUNC_PROTO: {
4919 struct btf_param *m = btf_params(t);
4921 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4922 err = visit(&m->name_off, ctx);
4935 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
4937 const struct btf_ext_info *seg;
4938 struct btf_ext_info_sec *sec;
4941 seg = &btf_ext->func_info;
4942 for_each_btf_ext_sec(seg, sec) {
4943 struct bpf_func_info_min *rec;
4945 for_each_btf_ext_rec(seg, sec, i, rec) {
4946 err = visit(&rec->type_id, ctx);
4952 seg = &btf_ext->core_relo_info;
4953 for_each_btf_ext_sec(seg, sec) {
4954 struct bpf_core_relo *rec;
4956 for_each_btf_ext_rec(seg, sec, i, rec) {
4957 err = visit(&rec->type_id, ctx);
4966 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
4968 const struct btf_ext_info *seg;
4969 struct btf_ext_info_sec *sec;
4972 seg = &btf_ext->func_info;
4973 for_each_btf_ext_sec(seg, sec) {
4974 err = visit(&sec->sec_name_off, ctx);
4979 seg = &btf_ext->line_info;
4980 for_each_btf_ext_sec(seg, sec) {
4981 struct bpf_line_info_min *rec;
4983 err = visit(&sec->sec_name_off, ctx);
4987 for_each_btf_ext_rec(seg, sec, i, rec) {
4988 err = visit(&rec->file_name_off, ctx);
4991 err = visit(&rec->line_off, ctx);
4997 seg = &btf_ext->core_relo_info;
4998 for_each_btf_ext_sec(seg, sec) {
4999 struct bpf_core_relo *rec;
5001 err = visit(&sec->sec_name_off, ctx);
5005 for_each_btf_ext_rec(seg, sec, i, rec) {
5006 err = visit(&rec->access_str_off, ctx);