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, (void *)(long)*str_off, &mapped_off)) {
1570 *str_off = (__u32)(long)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, (void *)(long)*str_off, (void *)(long)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(const void *key, void *ctx);
1634 static bool btf_dedup_equal_fn(const void *k1, const void *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 */
1728 btf->hdr->str_len = old_strs_len;
1730 hashmap__free(p.str_off_map);
1732 return libbpf_err(err);
1736 * Append new BTF_KIND_INT type with:
1737 * - *name* - non-empty, non-NULL type name;
1738 * - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1739 * - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1741 * - >0, type ID of newly added BTF type;
1744 int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
1749 /* non-empty name */
1750 if (!name || !name[0])
1751 return libbpf_err(-EINVAL);
1752 /* byte_sz must be power of 2 */
1753 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
1754 return libbpf_err(-EINVAL);
1755 if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
1756 return libbpf_err(-EINVAL);
1758 /* deconstruct BTF, if necessary, and invalidate raw_data */
1759 if (btf_ensure_modifiable(btf))
1760 return libbpf_err(-ENOMEM);
1762 sz = sizeof(struct btf_type) + sizeof(int);
1763 t = btf_add_type_mem(btf, sz);
1765 return libbpf_err(-ENOMEM);
1767 /* if something goes wrong later, we might end up with an extra string,
1768 * but that shouldn't be a problem, because BTF can't be constructed
1769 * completely anyway and will most probably be just discarded
1771 name_off = btf__add_str(btf, name);
1775 t->name_off = name_off;
1776 t->info = btf_type_info(BTF_KIND_INT, 0, 0);
1778 /* set INT info, we don't allow setting legacy bit offset/size */
1779 *(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
1781 return btf_commit_type(btf, sz);
1785 * Append new BTF_KIND_FLOAT type with:
1786 * - *name* - non-empty, non-NULL type name;
1787 * - *sz* - size of the type, in bytes;
1789 * - >0, type ID of newly added BTF type;
1792 int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
1797 /* non-empty name */
1798 if (!name || !name[0])
1799 return libbpf_err(-EINVAL);
1801 /* byte_sz must be one of the explicitly allowed values */
1802 if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
1804 return libbpf_err(-EINVAL);
1806 if (btf_ensure_modifiable(btf))
1807 return libbpf_err(-ENOMEM);
1809 sz = sizeof(struct btf_type);
1810 t = btf_add_type_mem(btf, sz);
1812 return libbpf_err(-ENOMEM);
1814 name_off = btf__add_str(btf, name);
1818 t->name_off = name_off;
1819 t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
1822 return btf_commit_type(btf, sz);
1825 /* it's completely legal to append BTF types with type IDs pointing forward to
1826 * types that haven't been appended yet, so we only make sure that id looks
1827 * sane, we can't guarantee that ID will always be valid
1829 static int validate_type_id(int id)
1831 if (id < 0 || id > BTF_MAX_NR_TYPES)
1836 /* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
1837 static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
1840 int sz, name_off = 0;
1842 if (validate_type_id(ref_type_id))
1843 return libbpf_err(-EINVAL);
1845 if (btf_ensure_modifiable(btf))
1846 return libbpf_err(-ENOMEM);
1848 sz = sizeof(struct btf_type);
1849 t = btf_add_type_mem(btf, sz);
1851 return libbpf_err(-ENOMEM);
1853 if (name && name[0]) {
1854 name_off = btf__add_str(btf, name);
1859 t->name_off = name_off;
1860 t->info = btf_type_info(kind, 0, 0);
1861 t->type = ref_type_id;
1863 return btf_commit_type(btf, sz);
1867 * Append new BTF_KIND_PTR type with:
1868 * - *ref_type_id* - referenced type ID, it might not exist yet;
1870 * - >0, type ID of newly added BTF type;
1873 int btf__add_ptr(struct btf *btf, int ref_type_id)
1875 return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
1879 * Append new BTF_KIND_ARRAY type with:
1880 * - *index_type_id* - type ID of the type describing array index;
1881 * - *elem_type_id* - type ID of the type describing array element;
1882 * - *nr_elems* - the size of the array;
1884 * - >0, type ID of newly added BTF type;
1887 int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
1890 struct btf_array *a;
1893 if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
1894 return libbpf_err(-EINVAL);
1896 if (btf_ensure_modifiable(btf))
1897 return libbpf_err(-ENOMEM);
1899 sz = sizeof(struct btf_type) + sizeof(struct btf_array);
1900 t = btf_add_type_mem(btf, sz);
1902 return libbpf_err(-ENOMEM);
1905 t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
1909 a->type = elem_type_id;
1910 a->index_type = index_type_id;
1911 a->nelems = nr_elems;
1913 return btf_commit_type(btf, sz);
1916 /* generic STRUCT/UNION append function */
1917 static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
1920 int sz, name_off = 0;
1922 if (btf_ensure_modifiable(btf))
1923 return libbpf_err(-ENOMEM);
1925 sz = sizeof(struct btf_type);
1926 t = btf_add_type_mem(btf, sz);
1928 return libbpf_err(-ENOMEM);
1930 if (name && name[0]) {
1931 name_off = btf__add_str(btf, name);
1936 /* start out with vlen=0 and no kflag; this will be adjusted when
1937 * adding each member
1939 t->name_off = name_off;
1940 t->info = btf_type_info(kind, 0, 0);
1943 return btf_commit_type(btf, sz);
1947 * Append new BTF_KIND_STRUCT type with:
1948 * - *name* - name of the struct, can be NULL or empty for anonymous structs;
1949 * - *byte_sz* - size of the struct, in bytes;
1951 * Struct initially has no fields in it. Fields can be added by
1952 * btf__add_field() right after btf__add_struct() succeeds.
1955 * - >0, type ID of newly added BTF type;
1958 int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
1960 return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
1964 * Append new BTF_KIND_UNION type with:
1965 * - *name* - name of the union, can be NULL or empty for anonymous union;
1966 * - *byte_sz* - size of the union, in bytes;
1968 * Union initially has no fields in it. Fields can be added by
1969 * btf__add_field() right after btf__add_union() succeeds. All fields
1970 * should have *bit_offset* of 0.
1973 * - >0, type ID of newly added BTF type;
1976 int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
1978 return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
1981 static struct btf_type *btf_last_type(struct btf *btf)
1983 return btf_type_by_id(btf, btf__type_cnt(btf) - 1);
1987 * Append new field for the current STRUCT/UNION type with:
1988 * - *name* - name of the field, can be NULL or empty for anonymous field;
1989 * - *type_id* - type ID for the type describing field type;
1990 * - *bit_offset* - bit offset of the start of the field within struct/union;
1991 * - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
1996 int btf__add_field(struct btf *btf, const char *name, int type_id,
1997 __u32 bit_offset, __u32 bit_size)
2000 struct btf_member *m;
2002 int sz, name_off = 0;
2004 /* last type should be union/struct */
2005 if (btf->nr_types == 0)
2006 return libbpf_err(-EINVAL);
2007 t = btf_last_type(btf);
2008 if (!btf_is_composite(t))
2009 return libbpf_err(-EINVAL);
2011 if (validate_type_id(type_id))
2012 return libbpf_err(-EINVAL);
2013 /* best-effort bit field offset/size enforcement */
2014 is_bitfield = bit_size || (bit_offset % 8 != 0);
2015 if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
2016 return libbpf_err(-EINVAL);
2018 /* only offset 0 is allowed for unions */
2019 if (btf_is_union(t) && bit_offset)
2020 return libbpf_err(-EINVAL);
2022 /* decompose and invalidate raw data */
2023 if (btf_ensure_modifiable(btf))
2024 return libbpf_err(-ENOMEM);
2026 sz = sizeof(struct btf_member);
2027 m = btf_add_type_mem(btf, sz);
2029 return libbpf_err(-ENOMEM);
2031 if (name && name[0]) {
2032 name_off = btf__add_str(btf, name);
2037 m->name_off = name_off;
2039 m->offset = bit_offset | (bit_size << 24);
2041 /* btf_add_type_mem can invalidate t pointer */
2042 t = btf_last_type(btf);
2043 /* update parent type's vlen and kflag */
2044 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
2046 btf->hdr->type_len += sz;
2047 btf->hdr->str_off += sz;
2051 static int btf_add_enum_common(struct btf *btf, const char *name, __u32 byte_sz,
2052 bool is_signed, __u8 kind)
2055 int sz, name_off = 0;
2057 /* byte_sz must be power of 2 */
2058 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2059 return libbpf_err(-EINVAL);
2061 if (btf_ensure_modifiable(btf))
2062 return libbpf_err(-ENOMEM);
2064 sz = sizeof(struct btf_type);
2065 t = btf_add_type_mem(btf, sz);
2067 return libbpf_err(-ENOMEM);
2069 if (name && name[0]) {
2070 name_off = btf__add_str(btf, name);
2075 /* start out with vlen=0; it will be adjusted when adding enum values */
2076 t->name_off = name_off;
2077 t->info = btf_type_info(kind, 0, is_signed);
2080 return btf_commit_type(btf, sz);
2084 * Append new BTF_KIND_ENUM type with:
2085 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
2086 * - *byte_sz* - size of the enum, in bytes.
2088 * Enum initially has no enum values in it (and corresponds to enum forward
2089 * declaration). Enumerator values can be added by btf__add_enum_value()
2090 * immediately after btf__add_enum() succeeds.
2093 * - >0, type ID of newly added BTF type;
2096 int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2099 * set the signedness to be unsigned, it will change to signed
2100 * if any later enumerator is negative.
2102 return btf_add_enum_common(btf, name, byte_sz, false, BTF_KIND_ENUM);
2106 * Append new enum value for the current ENUM type with:
2107 * - *name* - name of the enumerator value, can't be NULL or empty;
2108 * - *value* - integer value corresponding to enum value *name*;
2113 int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2119 /* last type should be BTF_KIND_ENUM */
2120 if (btf->nr_types == 0)
2121 return libbpf_err(-EINVAL);
2122 t = btf_last_type(btf);
2123 if (!btf_is_enum(t))
2124 return libbpf_err(-EINVAL);
2126 /* non-empty name */
2127 if (!name || !name[0])
2128 return libbpf_err(-EINVAL);
2129 if (value < INT_MIN || value > UINT_MAX)
2130 return libbpf_err(-E2BIG);
2132 /* decompose and invalidate raw data */
2133 if (btf_ensure_modifiable(btf))
2134 return libbpf_err(-ENOMEM);
2136 sz = sizeof(struct btf_enum);
2137 v = btf_add_type_mem(btf, sz);
2139 return libbpf_err(-ENOMEM);
2141 name_off = btf__add_str(btf, name);
2145 v->name_off = name_off;
2148 /* update parent type's vlen */
2149 t = btf_last_type(btf);
2150 btf_type_inc_vlen(t);
2152 /* if negative value, set signedness to signed */
2154 t->info = btf_type_info(btf_kind(t), btf_vlen(t), true);
2156 btf->hdr->type_len += sz;
2157 btf->hdr->str_off += sz;
2162 * Append new BTF_KIND_ENUM64 type with:
2163 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
2164 * - *byte_sz* - size of the enum, in bytes.
2165 * - *is_signed* - whether the enum values are signed or not;
2167 * Enum initially has no enum values in it (and corresponds to enum forward
2168 * declaration). Enumerator values can be added by btf__add_enum64_value()
2169 * immediately after btf__add_enum64() succeeds.
2172 * - >0, type ID of newly added BTF type;
2175 int btf__add_enum64(struct btf *btf, const char *name, __u32 byte_sz,
2178 return btf_add_enum_common(btf, name, byte_sz, is_signed,
2183 * Append new enum value for the current ENUM64 type with:
2184 * - *name* - name of the enumerator value, can't be NULL or empty;
2185 * - *value* - integer value corresponding to enum value *name*;
2190 int btf__add_enum64_value(struct btf *btf, const char *name, __u64 value)
2192 struct btf_enum64 *v;
2196 /* last type should be BTF_KIND_ENUM64 */
2197 if (btf->nr_types == 0)
2198 return libbpf_err(-EINVAL);
2199 t = btf_last_type(btf);
2200 if (!btf_is_enum64(t))
2201 return libbpf_err(-EINVAL);
2203 /* non-empty name */
2204 if (!name || !name[0])
2205 return libbpf_err(-EINVAL);
2207 /* decompose and invalidate raw data */
2208 if (btf_ensure_modifiable(btf))
2209 return libbpf_err(-ENOMEM);
2211 sz = sizeof(struct btf_enum64);
2212 v = btf_add_type_mem(btf, sz);
2214 return libbpf_err(-ENOMEM);
2216 name_off = btf__add_str(btf, name);
2220 v->name_off = name_off;
2221 v->val_lo32 = (__u32)value;
2222 v->val_hi32 = value >> 32;
2224 /* update parent type's vlen */
2225 t = btf_last_type(btf);
2226 btf_type_inc_vlen(t);
2228 btf->hdr->type_len += sz;
2229 btf->hdr->str_off += sz;
2234 * Append new BTF_KIND_FWD type with:
2235 * - *name*, non-empty/non-NULL name;
2236 * - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2237 * BTF_FWD_UNION, or BTF_FWD_ENUM;
2239 * - >0, type ID of newly added BTF type;
2242 int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2244 if (!name || !name[0])
2245 return libbpf_err(-EINVAL);
2248 case BTF_FWD_STRUCT:
2249 case BTF_FWD_UNION: {
2253 id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
2256 t = btf_type_by_id(btf, id);
2257 t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2261 /* enum forward in BTF currently is just an enum with no enum
2262 * values; we also assume a standard 4-byte size for it
2264 return btf__add_enum(btf, name, sizeof(int));
2266 return libbpf_err(-EINVAL);
2271 * Append new BTF_KING_TYPEDEF type with:
2272 * - *name*, non-empty/non-NULL name;
2273 * - *ref_type_id* - referenced type ID, it might not exist yet;
2275 * - >0, type ID of newly added BTF type;
2278 int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2280 if (!name || !name[0])
2281 return libbpf_err(-EINVAL);
2283 return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2287 * Append new BTF_KIND_VOLATILE type with:
2288 * - *ref_type_id* - referenced type ID, it might not exist yet;
2290 * - >0, type ID of newly added BTF type;
2293 int btf__add_volatile(struct btf *btf, int ref_type_id)
2295 return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2299 * Append new BTF_KIND_CONST type with:
2300 * - *ref_type_id* - referenced type ID, it might not exist yet;
2302 * - >0, type ID of newly added BTF type;
2305 int btf__add_const(struct btf *btf, int ref_type_id)
2307 return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2311 * Append new BTF_KIND_RESTRICT type with:
2312 * - *ref_type_id* - referenced type ID, it might not exist yet;
2314 * - >0, type ID of newly added BTF type;
2317 int btf__add_restrict(struct btf *btf, int ref_type_id)
2319 return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2323 * Append new BTF_KIND_TYPE_TAG type with:
2324 * - *value*, non-empty/non-NULL tag value;
2325 * - *ref_type_id* - referenced type ID, it might not exist yet;
2327 * - >0, type ID of newly added BTF type;
2330 int btf__add_type_tag(struct btf *btf, const char *value, int ref_type_id)
2332 if (!value|| !value[0])
2333 return libbpf_err(-EINVAL);
2335 return btf_add_ref_kind(btf, BTF_KIND_TYPE_TAG, value, ref_type_id);
2339 * Append new BTF_KIND_FUNC type with:
2340 * - *name*, non-empty/non-NULL name;
2341 * - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2343 * - >0, type ID of newly added BTF type;
2346 int btf__add_func(struct btf *btf, const char *name,
2347 enum btf_func_linkage linkage, int proto_type_id)
2351 if (!name || !name[0])
2352 return libbpf_err(-EINVAL);
2353 if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2354 linkage != BTF_FUNC_EXTERN)
2355 return libbpf_err(-EINVAL);
2357 id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2359 struct btf_type *t = btf_type_by_id(btf, id);
2361 t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2363 return libbpf_err(id);
2367 * Append new BTF_KIND_FUNC_PROTO with:
2368 * - *ret_type_id* - type ID for return result of a function.
2370 * Function prototype initially has no arguments, but they can be added by
2371 * btf__add_func_param() one by one, immediately after
2372 * btf__add_func_proto() succeeded.
2375 * - >0, type ID of newly added BTF type;
2378 int btf__add_func_proto(struct btf *btf, int ret_type_id)
2383 if (validate_type_id(ret_type_id))
2384 return libbpf_err(-EINVAL);
2386 if (btf_ensure_modifiable(btf))
2387 return libbpf_err(-ENOMEM);
2389 sz = sizeof(struct btf_type);
2390 t = btf_add_type_mem(btf, sz);
2392 return libbpf_err(-ENOMEM);
2394 /* start out with vlen=0; this will be adjusted when adding enum
2395 * values, if necessary
2398 t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2399 t->type = ret_type_id;
2401 return btf_commit_type(btf, sz);
2405 * Append new function parameter for current FUNC_PROTO type with:
2406 * - *name* - parameter name, can be NULL or empty;
2407 * - *type_id* - type ID describing the type of the parameter.
2412 int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2415 struct btf_param *p;
2416 int sz, name_off = 0;
2418 if (validate_type_id(type_id))
2419 return libbpf_err(-EINVAL);
2421 /* last type should be BTF_KIND_FUNC_PROTO */
2422 if (btf->nr_types == 0)
2423 return libbpf_err(-EINVAL);
2424 t = btf_last_type(btf);
2425 if (!btf_is_func_proto(t))
2426 return libbpf_err(-EINVAL);
2428 /* decompose and invalidate raw data */
2429 if (btf_ensure_modifiable(btf))
2430 return libbpf_err(-ENOMEM);
2432 sz = sizeof(struct btf_param);
2433 p = btf_add_type_mem(btf, sz);
2435 return libbpf_err(-ENOMEM);
2437 if (name && name[0]) {
2438 name_off = btf__add_str(btf, name);
2443 p->name_off = name_off;
2446 /* update parent type's vlen */
2447 t = btf_last_type(btf);
2448 btf_type_inc_vlen(t);
2450 btf->hdr->type_len += sz;
2451 btf->hdr->str_off += sz;
2456 * Append new BTF_KIND_VAR type with:
2457 * - *name* - non-empty/non-NULL name;
2458 * - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2459 * BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2460 * - *type_id* - type ID of the type describing the type of the variable.
2462 * - >0, type ID of newly added BTF type;
2465 int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2471 /* non-empty name */
2472 if (!name || !name[0])
2473 return libbpf_err(-EINVAL);
2474 if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2475 linkage != BTF_VAR_GLOBAL_EXTERN)
2476 return libbpf_err(-EINVAL);
2477 if (validate_type_id(type_id))
2478 return libbpf_err(-EINVAL);
2480 /* deconstruct BTF, if necessary, and invalidate raw_data */
2481 if (btf_ensure_modifiable(btf))
2482 return libbpf_err(-ENOMEM);
2484 sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2485 t = btf_add_type_mem(btf, sz);
2487 return libbpf_err(-ENOMEM);
2489 name_off = btf__add_str(btf, name);
2493 t->name_off = name_off;
2494 t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2498 v->linkage = linkage;
2500 return btf_commit_type(btf, sz);
2504 * Append new BTF_KIND_DATASEC type with:
2505 * - *name* - non-empty/non-NULL name;
2506 * - *byte_sz* - data section size, in bytes.
2508 * Data section is initially empty. Variables info can be added with
2509 * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2512 * - >0, type ID of newly added BTF type;
2515 int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2520 /* non-empty name */
2521 if (!name || !name[0])
2522 return libbpf_err(-EINVAL);
2524 if (btf_ensure_modifiable(btf))
2525 return libbpf_err(-ENOMEM);
2527 sz = sizeof(struct btf_type);
2528 t = btf_add_type_mem(btf, sz);
2530 return libbpf_err(-ENOMEM);
2532 name_off = btf__add_str(btf, name);
2536 /* start with vlen=0, which will be update as var_secinfos are added */
2537 t->name_off = name_off;
2538 t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2541 return btf_commit_type(btf, sz);
2545 * Append new data section variable information entry for current DATASEC type:
2546 * - *var_type_id* - type ID, describing type of the variable;
2547 * - *offset* - variable offset within data section, in bytes;
2548 * - *byte_sz* - variable size, in bytes.
2554 int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2557 struct btf_var_secinfo *v;
2560 /* last type should be BTF_KIND_DATASEC */
2561 if (btf->nr_types == 0)
2562 return libbpf_err(-EINVAL);
2563 t = btf_last_type(btf);
2564 if (!btf_is_datasec(t))
2565 return libbpf_err(-EINVAL);
2567 if (validate_type_id(var_type_id))
2568 return libbpf_err(-EINVAL);
2570 /* decompose and invalidate raw data */
2571 if (btf_ensure_modifiable(btf))
2572 return libbpf_err(-ENOMEM);
2574 sz = sizeof(struct btf_var_secinfo);
2575 v = btf_add_type_mem(btf, sz);
2577 return libbpf_err(-ENOMEM);
2579 v->type = var_type_id;
2583 /* update parent type's vlen */
2584 t = btf_last_type(btf);
2585 btf_type_inc_vlen(t);
2587 btf->hdr->type_len += sz;
2588 btf->hdr->str_off += sz;
2593 * Append new BTF_KIND_DECL_TAG type with:
2594 * - *value* - non-empty/non-NULL string;
2595 * - *ref_type_id* - referenced type ID, it might not exist yet;
2596 * - *component_idx* - -1 for tagging reference type, otherwise struct/union
2597 * member or function argument index;
2599 * - >0, type ID of newly added BTF type;
2602 int btf__add_decl_tag(struct btf *btf, const char *value, int ref_type_id,
2608 if (!value || !value[0] || component_idx < -1)
2609 return libbpf_err(-EINVAL);
2611 if (validate_type_id(ref_type_id))
2612 return libbpf_err(-EINVAL);
2614 if (btf_ensure_modifiable(btf))
2615 return libbpf_err(-ENOMEM);
2617 sz = sizeof(struct btf_type) + sizeof(struct btf_decl_tag);
2618 t = btf_add_type_mem(btf, sz);
2620 return libbpf_err(-ENOMEM);
2622 value_off = btf__add_str(btf, value);
2626 t->name_off = value_off;
2627 t->info = btf_type_info(BTF_KIND_DECL_TAG, 0, false);
2628 t->type = ref_type_id;
2629 btf_decl_tag(t)->component_idx = component_idx;
2631 return btf_commit_type(btf, sz);
2634 struct btf_ext_sec_setup_param {
2638 struct btf_ext_info *ext_info;
2642 static int btf_ext_setup_info(struct btf_ext *btf_ext,
2643 struct btf_ext_sec_setup_param *ext_sec)
2645 const struct btf_ext_info_sec *sinfo;
2646 struct btf_ext_info *ext_info;
2647 __u32 info_left, record_size;
2649 /* The start of the info sec (including the __u32 record_size). */
2652 if (ext_sec->len == 0)
2655 if (ext_sec->off & 0x03) {
2656 pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2661 info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2662 info_left = ext_sec->len;
2664 if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2665 pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2666 ext_sec->desc, ext_sec->off, ext_sec->len);
2670 /* At least a record size */
2671 if (info_left < sizeof(__u32)) {
2672 pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2676 /* The record size needs to meet the minimum standard */
2677 record_size = *(__u32 *)info;
2678 if (record_size < ext_sec->min_rec_size ||
2679 record_size & 0x03) {
2680 pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2681 ext_sec->desc, record_size);
2685 sinfo = info + sizeof(__u32);
2686 info_left -= sizeof(__u32);
2688 /* If no records, return failure now so .BTF.ext won't be used. */
2690 pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
2695 unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2696 __u64 total_record_size;
2699 if (info_left < sec_hdrlen) {
2700 pr_debug("%s section header is not found in .BTF.ext\n",
2705 num_records = sinfo->num_info;
2706 if (num_records == 0) {
2707 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2712 total_record_size = sec_hdrlen + (__u64)num_records * record_size;
2713 if (info_left < total_record_size) {
2714 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2719 info_left -= total_record_size;
2720 sinfo = (void *)sinfo + total_record_size;
2724 ext_info = ext_sec->ext_info;
2725 ext_info->len = ext_sec->len - sizeof(__u32);
2726 ext_info->rec_size = record_size;
2727 ext_info->info = info + sizeof(__u32);
2728 ext_info->sec_cnt = sec_cnt;
2733 static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
2735 struct btf_ext_sec_setup_param param = {
2736 .off = btf_ext->hdr->func_info_off,
2737 .len = btf_ext->hdr->func_info_len,
2738 .min_rec_size = sizeof(struct bpf_func_info_min),
2739 .ext_info = &btf_ext->func_info,
2743 return btf_ext_setup_info(btf_ext, ¶m);
2746 static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
2748 struct btf_ext_sec_setup_param param = {
2749 .off = btf_ext->hdr->line_info_off,
2750 .len = btf_ext->hdr->line_info_len,
2751 .min_rec_size = sizeof(struct bpf_line_info_min),
2752 .ext_info = &btf_ext->line_info,
2753 .desc = "line_info",
2756 return btf_ext_setup_info(btf_ext, ¶m);
2759 static int btf_ext_setup_core_relos(struct btf_ext *btf_ext)
2761 struct btf_ext_sec_setup_param param = {
2762 .off = btf_ext->hdr->core_relo_off,
2763 .len = btf_ext->hdr->core_relo_len,
2764 .min_rec_size = sizeof(struct bpf_core_relo),
2765 .ext_info = &btf_ext->core_relo_info,
2766 .desc = "core_relo",
2769 return btf_ext_setup_info(btf_ext, ¶m);
2772 static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
2774 const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
2776 if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
2777 data_size < hdr->hdr_len) {
2778 pr_debug("BTF.ext header not found");
2782 if (hdr->magic == bswap_16(BTF_MAGIC)) {
2783 pr_warn("BTF.ext in non-native endianness is not supported\n");
2785 } else if (hdr->magic != BTF_MAGIC) {
2786 pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
2790 if (hdr->version != BTF_VERSION) {
2791 pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
2796 pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
2800 if (data_size == hdr->hdr_len) {
2801 pr_debug("BTF.ext has no data\n");
2808 void btf_ext__free(struct btf_ext *btf_ext)
2810 if (IS_ERR_OR_NULL(btf_ext))
2812 free(btf_ext->func_info.sec_idxs);
2813 free(btf_ext->line_info.sec_idxs);
2814 free(btf_ext->core_relo_info.sec_idxs);
2815 free(btf_ext->data);
2819 struct btf_ext *btf_ext__new(const __u8 *data, __u32 size)
2821 struct btf_ext *btf_ext;
2824 btf_ext = calloc(1, sizeof(struct btf_ext));
2826 return libbpf_err_ptr(-ENOMEM);
2828 btf_ext->data_size = size;
2829 btf_ext->data = malloc(size);
2830 if (!btf_ext->data) {
2834 memcpy(btf_ext->data, data, size);
2836 err = btf_ext_parse_hdr(btf_ext->data, size);
2840 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
2845 err = btf_ext_setup_func_info(btf_ext);
2849 err = btf_ext_setup_line_info(btf_ext);
2853 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
2854 goto done; /* skip core relos parsing */
2856 err = btf_ext_setup_core_relos(btf_ext);
2862 btf_ext__free(btf_ext);
2863 return libbpf_err_ptr(err);
2869 const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
2871 *size = btf_ext->data_size;
2872 return btf_ext->data;
2877 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts);
2878 static void btf_dedup_free(struct btf_dedup *d);
2879 static int btf_dedup_prep(struct btf_dedup *d);
2880 static int btf_dedup_strings(struct btf_dedup *d);
2881 static int btf_dedup_prim_types(struct btf_dedup *d);
2882 static int btf_dedup_struct_types(struct btf_dedup *d);
2883 static int btf_dedup_ref_types(struct btf_dedup *d);
2884 static int btf_dedup_compact_types(struct btf_dedup *d);
2885 static int btf_dedup_remap_types(struct btf_dedup *d);
2888 * Deduplicate BTF types and strings.
2890 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
2891 * section with all BTF type descriptors and string data. It overwrites that
2892 * memory in-place with deduplicated types and strings without any loss of
2893 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
2894 * is provided, all the strings referenced from .BTF.ext section are honored
2895 * and updated to point to the right offsets after deduplication.
2897 * If function returns with error, type/string data might be garbled and should
2900 * More verbose and detailed description of both problem btf_dedup is solving,
2901 * as well as solution could be found at:
2902 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
2904 * Problem description and justification
2905 * =====================================
2907 * BTF type information is typically emitted either as a result of conversion
2908 * from DWARF to BTF or directly by compiler. In both cases, each compilation
2909 * unit contains information about a subset of all the types that are used
2910 * in an application. These subsets are frequently overlapping and contain a lot
2911 * of duplicated information when later concatenated together into a single
2912 * binary. This algorithm ensures that each unique type is represented by single
2913 * BTF type descriptor, greatly reducing resulting size of BTF data.
2915 * Compilation unit isolation and subsequent duplication of data is not the only
2916 * problem. The same type hierarchy (e.g., struct and all the type that struct
2917 * references) in different compilation units can be represented in BTF to
2918 * various degrees of completeness (or, rather, incompleteness) due to
2919 * struct/union forward declarations.
2921 * Let's take a look at an example, that we'll use to better understand the
2922 * problem (and solution). Suppose we have two compilation units, each using
2923 * same `struct S`, but each of them having incomplete type information about
2952 * In case of CU #1, BTF data will know only that `struct B` exist (but no
2953 * more), but will know the complete type information about `struct A`. While
2954 * for CU #2, it will know full type information about `struct B`, but will
2955 * only know about forward declaration of `struct A` (in BTF terms, it will
2956 * have `BTF_KIND_FWD` type descriptor with name `B`).
2958 * This compilation unit isolation means that it's possible that there is no
2959 * single CU with complete type information describing structs `S`, `A`, and
2960 * `B`. Also, we might get tons of duplicated and redundant type information.
2962 * Additional complication we need to keep in mind comes from the fact that
2963 * types, in general, can form graphs containing cycles, not just DAGs.
2965 * While algorithm does deduplication, it also merges and resolves type
2966 * information (unless disabled throught `struct btf_opts`), whenever possible.
2967 * E.g., in the example above with two compilation units having partial type
2968 * information for structs `A` and `B`, the output of algorithm will emit
2969 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
2970 * (as well as type information for `int` and pointers), as if they were defined
2971 * in a single compilation unit as:
2991 * Algorithm completes its work in 6 separate passes:
2993 * 1. Strings deduplication.
2994 * 2. Primitive types deduplication (int, enum, fwd).
2995 * 3. Struct/union types deduplication.
2996 * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
2997 * protos, and const/volatile/restrict modifiers).
2998 * 5. Types compaction.
2999 * 6. Types remapping.
3001 * Algorithm determines canonical type descriptor, which is a single
3002 * representative type for each truly unique type. This canonical type is the
3003 * one that will go into final deduplicated BTF type information. For
3004 * struct/unions, it is also the type that algorithm will merge additional type
3005 * information into (while resolving FWDs), as it discovers it from data in
3006 * other CUs. Each input BTF type eventually gets either mapped to itself, if
3007 * that type is canonical, or to some other type, if that type is equivalent
3008 * and was chosen as canonical representative. This mapping is stored in
3009 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
3010 * FWD type got resolved to.
3012 * To facilitate fast discovery of canonical types, we also maintain canonical
3013 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
3014 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
3015 * that match that signature. With sufficiently good choice of type signature
3016 * hashing function, we can limit number of canonical types for each unique type
3017 * signature to a very small number, allowing to find canonical type for any
3018 * duplicated type very quickly.
3020 * Struct/union deduplication is the most critical part and algorithm for
3021 * deduplicating structs/unions is described in greater details in comments for
3022 * `btf_dedup_is_equiv` function.
3024 int btf__dedup(struct btf *btf, const struct btf_dedup_opts *opts)
3026 struct btf_dedup *d;
3029 if (!OPTS_VALID(opts, btf_dedup_opts))
3030 return libbpf_err(-EINVAL);
3032 d = btf_dedup_new(btf, opts);
3034 pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
3035 return libbpf_err(-EINVAL);
3038 if (btf_ensure_modifiable(btf)) {
3043 err = btf_dedup_prep(d);
3045 pr_debug("btf_dedup_prep failed:%d\n", err);
3048 err = btf_dedup_strings(d);
3050 pr_debug("btf_dedup_strings failed:%d\n", err);
3053 err = btf_dedup_prim_types(d);
3055 pr_debug("btf_dedup_prim_types failed:%d\n", err);
3058 err = btf_dedup_struct_types(d);
3060 pr_debug("btf_dedup_struct_types failed:%d\n", err);
3063 err = btf_dedup_ref_types(d);
3065 pr_debug("btf_dedup_ref_types failed:%d\n", err);
3068 err = btf_dedup_compact_types(d);
3070 pr_debug("btf_dedup_compact_types failed:%d\n", err);
3073 err = btf_dedup_remap_types(d);
3075 pr_debug("btf_dedup_remap_types failed:%d\n", err);
3081 return libbpf_err(err);
3084 #define BTF_UNPROCESSED_ID ((__u32)-1)
3085 #define BTF_IN_PROGRESS_ID ((__u32)-2)
3088 /* .BTF section to be deduped in-place */
3091 * Optional .BTF.ext section. When provided, any strings referenced
3092 * from it will be taken into account when deduping strings
3094 struct btf_ext *btf_ext;
3096 * This is a map from any type's signature hash to a list of possible
3097 * canonical representative type candidates. Hash collisions are
3098 * ignored, so even types of various kinds can share same list of
3099 * candidates, which is fine because we rely on subsequent
3100 * btf_xxx_equal() checks to authoritatively verify type equality.
3102 struct hashmap *dedup_table;
3103 /* Canonical types map */
3105 /* Hypothetical mapping, used during type graph equivalence checks */
3110 /* Whether hypothetical mapping, if successful, would need to adjust
3111 * already canonicalized types (due to a new forward declaration to
3112 * concrete type resolution). In such case, during split BTF dedup
3113 * candidate type would still be considered as different, because base
3114 * BTF is considered to be immutable.
3116 bool hypot_adjust_canon;
3117 /* Various option modifying behavior of algorithm */
3118 struct btf_dedup_opts opts;
3119 /* temporary strings deduplication state */
3120 struct strset *strs_set;
3123 static long hash_combine(long h, long value)
3125 return h * 31 + value;
3128 #define for_each_dedup_cand(d, node, hash) \
3129 hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
3131 static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
3133 return hashmap__append(d->dedup_table,
3134 (void *)hash, (void *)(long)type_id);
3137 static int btf_dedup_hypot_map_add(struct btf_dedup *d,
3138 __u32 from_id, __u32 to_id)
3140 if (d->hypot_cnt == d->hypot_cap) {
3143 d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3144 new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3147 d->hypot_list = new_list;
3149 d->hypot_list[d->hypot_cnt++] = from_id;
3150 d->hypot_map[from_id] = to_id;
3154 static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3158 for (i = 0; i < d->hypot_cnt; i++)
3159 d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3161 d->hypot_adjust_canon = false;
3164 static void btf_dedup_free(struct btf_dedup *d)
3166 hashmap__free(d->dedup_table);
3167 d->dedup_table = NULL;
3173 d->hypot_map = NULL;
3175 free(d->hypot_list);
3176 d->hypot_list = NULL;
3181 static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
3186 static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
3191 static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
3196 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts)
3198 struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3199 hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3200 int i, err = 0, type_cnt;
3203 return ERR_PTR(-ENOMEM);
3205 if (OPTS_GET(opts, force_collisions, false))
3206 hash_fn = btf_dedup_collision_hash_fn;
3209 d->btf_ext = OPTS_GET(opts, btf_ext, NULL);
3211 d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3212 if (IS_ERR(d->dedup_table)) {
3213 err = PTR_ERR(d->dedup_table);
3214 d->dedup_table = NULL;
3218 type_cnt = btf__type_cnt(btf);
3219 d->map = malloc(sizeof(__u32) * type_cnt);
3224 /* special BTF "void" type is made canonical immediately */
3226 for (i = 1; i < type_cnt; i++) {
3227 struct btf_type *t = btf_type_by_id(d->btf, i);
3229 /* VAR and DATASEC are never deduped and are self-canonical */
3230 if (btf_is_var(t) || btf_is_datasec(t))
3233 d->map[i] = BTF_UNPROCESSED_ID;
3236 d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3237 if (!d->hypot_map) {
3241 for (i = 0; i < type_cnt; i++)
3242 d->hypot_map[i] = BTF_UNPROCESSED_ID;
3247 return ERR_PTR(err);
3254 * Iterate over all possible places in .BTF and .BTF.ext that can reference
3255 * string and pass pointer to it to a provided callback `fn`.
3257 static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3261 for (i = 0; i < d->btf->nr_types; i++) {
3262 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3264 r = btf_type_visit_str_offs(t, fn, ctx);
3272 r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3279 static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3281 struct btf_dedup *d = ctx;
3282 __u32 str_off = *str_off_ptr;
3286 /* don't touch empty string or string in main BTF */
3287 if (str_off == 0 || str_off < d->btf->start_str_off)
3290 s = btf__str_by_offset(d->btf, str_off);
3291 if (d->btf->base_btf) {
3292 err = btf__find_str(d->btf->base_btf, s);
3301 off = strset__add_str(d->strs_set, s);
3305 *str_off_ptr = d->btf->start_str_off + off;
3310 * Dedup string and filter out those that are not referenced from either .BTF
3311 * or .BTF.ext (if provided) sections.
3313 * This is done by building index of all strings in BTF's string section,
3314 * then iterating over all entities that can reference strings (e.g., type
3315 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3316 * strings as used. After that all used strings are deduped and compacted into
3317 * sequential blob of memory and new offsets are calculated. Then all the string
3318 * references are iterated again and rewritten using new offsets.
3320 static int btf_dedup_strings(struct btf_dedup *d)
3324 if (d->btf->strs_deduped)
3327 d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3328 if (IS_ERR(d->strs_set)) {
3329 err = PTR_ERR(d->strs_set);
3333 if (!d->btf->base_btf) {
3334 /* insert empty string; we won't be looking it up during strings
3335 * dedup, but it's good to have it for generic BTF string lookups
3337 err = strset__add_str(d->strs_set, "");
3342 /* remap string offsets */
3343 err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3347 /* replace BTF string data and hash with deduped ones */
3348 strset__free(d->btf->strs_set);
3349 d->btf->hdr->str_len = strset__data_size(d->strs_set);
3350 d->btf->strs_set = d->strs_set;
3352 d->btf->strs_deduped = true;
3356 strset__free(d->strs_set);
3362 static long btf_hash_common(struct btf_type *t)
3366 h = hash_combine(0, t->name_off);
3367 h = hash_combine(h, t->info);
3368 h = hash_combine(h, t->size);
3372 static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3374 return t1->name_off == t2->name_off &&
3375 t1->info == t2->info &&
3376 t1->size == t2->size;
3379 /* Calculate type signature hash of INT or TAG. */
3380 static long btf_hash_int_decl_tag(struct btf_type *t)
3382 __u32 info = *(__u32 *)(t + 1);
3385 h = btf_hash_common(t);
3386 h = hash_combine(h, info);
3390 /* Check structural equality of two INTs or TAGs. */
3391 static bool btf_equal_int_tag(struct btf_type *t1, struct btf_type *t2)
3395 if (!btf_equal_common(t1, t2))
3397 info1 = *(__u32 *)(t1 + 1);
3398 info2 = *(__u32 *)(t2 + 1);
3399 return info1 == info2;
3402 /* Calculate type signature hash of ENUM/ENUM64. */
3403 static long btf_hash_enum(struct btf_type *t)
3407 /* don't hash vlen and enum members to support enum fwd resolving */
3408 h = hash_combine(0, t->name_off);
3409 h = hash_combine(h, t->info & ~0xffff);
3410 h = hash_combine(h, t->size);
3414 /* Check structural equality of two ENUMs. */
3415 static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3417 const struct btf_enum *m1, *m2;
3421 if (!btf_equal_common(t1, t2))
3424 vlen = btf_vlen(t1);
3427 for (i = 0; i < vlen; i++) {
3428 if (m1->name_off != m2->name_off || m1->val != m2->val)
3436 static bool btf_equal_enum64(struct btf_type *t1, struct btf_type *t2)
3438 const struct btf_enum64 *m1, *m2;
3442 if (!btf_equal_common(t1, t2))
3445 vlen = btf_vlen(t1);
3446 m1 = btf_enum64(t1);
3447 m2 = btf_enum64(t2);
3448 for (i = 0; i < vlen; i++) {
3449 if (m1->name_off != m2->name_off || m1->val_lo32 != m2->val_lo32 ||
3450 m1->val_hi32 != m2->val_hi32)
3458 static inline bool btf_is_enum_fwd(struct btf_type *t)
3460 return btf_is_any_enum(t) && btf_vlen(t) == 0;
3463 static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3465 if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3466 return btf_equal_enum(t1, t2);
3467 /* ignore vlen when comparing */
3468 return t1->name_off == t2->name_off &&
3469 (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
3470 t1->size == t2->size;
3473 static bool btf_compat_enum64(struct btf_type *t1, struct btf_type *t2)
3475 if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3476 return btf_equal_enum64(t1, t2);
3478 /* ignore vlen when comparing */
3479 return t1->name_off == t2->name_off &&
3480 (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
3481 t1->size == t2->size;
3485 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3486 * as referenced type IDs equivalence is established separately during type
3487 * graph equivalence check algorithm.
3489 static long btf_hash_struct(struct btf_type *t)
3491 const struct btf_member *member = btf_members(t);
3492 __u32 vlen = btf_vlen(t);
3493 long h = btf_hash_common(t);
3496 for (i = 0; i < vlen; i++) {
3497 h = hash_combine(h, member->name_off);
3498 h = hash_combine(h, member->offset);
3499 /* no hashing of referenced type ID, it can be unresolved yet */
3506 * Check structural compatibility of two STRUCTs/UNIONs, ignoring referenced
3507 * type IDs. This check is performed during type graph equivalence check and
3508 * referenced types equivalence is checked separately.
3510 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3512 const struct btf_member *m1, *m2;
3516 if (!btf_equal_common(t1, t2))
3519 vlen = btf_vlen(t1);
3520 m1 = btf_members(t1);
3521 m2 = btf_members(t2);
3522 for (i = 0; i < vlen; i++) {
3523 if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3532 * Calculate type signature hash of ARRAY, including referenced type IDs,
3533 * under assumption that they were already resolved to canonical type IDs and
3534 * are not going to change.
3536 static long btf_hash_array(struct btf_type *t)
3538 const struct btf_array *info = btf_array(t);
3539 long h = btf_hash_common(t);
3541 h = hash_combine(h, info->type);
3542 h = hash_combine(h, info->index_type);
3543 h = hash_combine(h, info->nelems);
3548 * Check exact equality of two ARRAYs, taking into account referenced
3549 * type IDs, under assumption that they were already resolved to canonical
3550 * type IDs and are not going to change.
3551 * This function is called during reference types deduplication to compare
3552 * ARRAY to potential canonical representative.
3554 static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3556 const struct btf_array *info1, *info2;
3558 if (!btf_equal_common(t1, t2))
3561 info1 = btf_array(t1);
3562 info2 = btf_array(t2);
3563 return info1->type == info2->type &&
3564 info1->index_type == info2->index_type &&
3565 info1->nelems == info2->nelems;
3569 * Check structural compatibility of two ARRAYs, ignoring referenced type
3570 * IDs. This check is performed during type graph equivalence check and
3571 * referenced types equivalence is checked separately.
3573 static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
3575 if (!btf_equal_common(t1, t2))
3578 return btf_array(t1)->nelems == btf_array(t2)->nelems;
3582 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
3583 * under assumption that they were already resolved to canonical type IDs and
3584 * are not going to change.
3586 static long btf_hash_fnproto(struct btf_type *t)
3588 const struct btf_param *member = btf_params(t);
3589 __u16 vlen = btf_vlen(t);
3590 long h = btf_hash_common(t);
3593 for (i = 0; i < vlen; i++) {
3594 h = hash_combine(h, member->name_off);
3595 h = hash_combine(h, member->type);
3602 * Check exact equality of two FUNC_PROTOs, taking into account referenced
3603 * type IDs, under assumption that they were already resolved to canonical
3604 * type IDs and are not going to change.
3605 * This function is called during reference types deduplication to compare
3606 * FUNC_PROTO to potential canonical representative.
3608 static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
3610 const struct btf_param *m1, *m2;
3614 if (!btf_equal_common(t1, t2))
3617 vlen = btf_vlen(t1);
3618 m1 = btf_params(t1);
3619 m2 = btf_params(t2);
3620 for (i = 0; i < vlen; i++) {
3621 if (m1->name_off != m2->name_off || m1->type != m2->type)
3630 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3631 * IDs. This check is performed during type graph equivalence check and
3632 * referenced types equivalence is checked separately.
3634 static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
3636 const struct btf_param *m1, *m2;
3640 /* skip return type ID */
3641 if (t1->name_off != t2->name_off || t1->info != t2->info)
3644 vlen = btf_vlen(t1);
3645 m1 = btf_params(t1);
3646 m2 = btf_params(t2);
3647 for (i = 0; i < vlen; i++) {
3648 if (m1->name_off != m2->name_off)
3656 /* Prepare split BTF for deduplication by calculating hashes of base BTF's
3657 * types and initializing the rest of the state (canonical type mapping) for
3658 * the fixed base BTF part.
3660 static int btf_dedup_prep(struct btf_dedup *d)
3666 if (!d->btf->base_btf)
3669 for (type_id = 1; type_id < d->btf->start_id; type_id++) {
3670 t = btf_type_by_id(d->btf, type_id);
3672 /* all base BTF types are self-canonical by definition */
3673 d->map[type_id] = type_id;
3675 switch (btf_kind(t)) {
3677 case BTF_KIND_DATASEC:
3678 /* VAR and DATASEC are never hash/deduplicated */
3680 case BTF_KIND_CONST:
3681 case BTF_KIND_VOLATILE:
3682 case BTF_KIND_RESTRICT:
3685 case BTF_KIND_TYPEDEF:
3687 case BTF_KIND_FLOAT:
3688 case BTF_KIND_TYPE_TAG:
3689 h = btf_hash_common(t);
3692 case BTF_KIND_DECL_TAG:
3693 h = btf_hash_int_decl_tag(t);
3696 case BTF_KIND_ENUM64:
3697 h = btf_hash_enum(t);
3699 case BTF_KIND_STRUCT:
3700 case BTF_KIND_UNION:
3701 h = btf_hash_struct(t);
3703 case BTF_KIND_ARRAY:
3704 h = btf_hash_array(t);
3706 case BTF_KIND_FUNC_PROTO:
3707 h = btf_hash_fnproto(t);
3710 pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
3713 if (btf_dedup_table_add(d, h, type_id))
3721 * Deduplicate primitive types, that can't reference other types, by calculating
3722 * their type signature hash and comparing them with any possible canonical
3723 * candidate. If no canonical candidate matches, type itself is marked as
3724 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
3726 static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
3728 struct btf_type *t = btf_type_by_id(d->btf, type_id);
3729 struct hashmap_entry *hash_entry;
3730 struct btf_type *cand;
3731 /* if we don't find equivalent type, then we are canonical */
3732 __u32 new_id = type_id;
3736 switch (btf_kind(t)) {
3737 case BTF_KIND_CONST:
3738 case BTF_KIND_VOLATILE:
3739 case BTF_KIND_RESTRICT:
3741 case BTF_KIND_TYPEDEF:
3742 case BTF_KIND_ARRAY:
3743 case BTF_KIND_STRUCT:
3744 case BTF_KIND_UNION:
3746 case BTF_KIND_FUNC_PROTO:
3748 case BTF_KIND_DATASEC:
3749 case BTF_KIND_DECL_TAG:
3750 case BTF_KIND_TYPE_TAG:
3754 h = btf_hash_int_decl_tag(t);
3755 for_each_dedup_cand(d, hash_entry, h) {
3756 cand_id = (__u32)(long)hash_entry->value;
3757 cand = btf_type_by_id(d->btf, cand_id);
3758 if (btf_equal_int_tag(t, cand)) {
3766 h = btf_hash_enum(t);
3767 for_each_dedup_cand(d, hash_entry, h) {
3768 cand_id = (__u32)(long)hash_entry->value;
3769 cand = btf_type_by_id(d->btf, cand_id);
3770 if (btf_equal_enum(t, cand)) {
3774 if (btf_compat_enum(t, cand)) {
3775 if (btf_is_enum_fwd(t)) {
3776 /* resolve fwd to full enum */
3780 /* resolve canonical enum fwd to full enum */
3781 d->map[cand_id] = type_id;
3786 case BTF_KIND_ENUM64:
3787 h = btf_hash_enum(t);
3788 for_each_dedup_cand(d, hash_entry, h) {
3789 cand_id = (__u32)(long)hash_entry->value;
3790 cand = btf_type_by_id(d->btf, cand_id);
3791 if (btf_equal_enum64(t, cand)) {
3795 if (btf_compat_enum64(t, cand)) {
3796 if (btf_is_enum_fwd(t)) {
3797 /* resolve fwd to full enum */
3801 /* resolve canonical enum fwd to full enum */
3802 d->map[cand_id] = type_id;
3808 case BTF_KIND_FLOAT:
3809 h = btf_hash_common(t);
3810 for_each_dedup_cand(d, hash_entry, h) {
3811 cand_id = (__u32)(long)hash_entry->value;
3812 cand = btf_type_by_id(d->btf, cand_id);
3813 if (btf_equal_common(t, cand)) {
3824 d->map[type_id] = new_id;
3825 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
3831 static int btf_dedup_prim_types(struct btf_dedup *d)
3835 for (i = 0; i < d->btf->nr_types; i++) {
3836 err = btf_dedup_prim_type(d, d->btf->start_id + i);
3844 * Check whether type is already mapped into canonical one (could be to itself).
3846 static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
3848 return d->map[type_id] <= BTF_MAX_NR_TYPES;
3852 * Resolve type ID into its canonical type ID, if any; otherwise return original
3853 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
3854 * STRUCT/UNION link and resolve it into canonical type ID as well.
3856 static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
3858 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3859 type_id = d->map[type_id];
3864 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
3867 static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
3869 __u32 orig_type_id = type_id;
3871 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3874 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3875 type_id = d->map[type_id];
3877 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3880 return orig_type_id;
3884 static inline __u16 btf_fwd_kind(struct btf_type *t)
3886 return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
3889 /* Check if given two types are identical ARRAY definitions */
3890 static int btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
3892 struct btf_type *t1, *t2;
3894 t1 = btf_type_by_id(d->btf, id1);
3895 t2 = btf_type_by_id(d->btf, id2);
3896 if (!btf_is_array(t1) || !btf_is_array(t2))
3899 return btf_equal_array(t1, t2);
3902 /* Check if given two types are identical STRUCT/UNION definitions */
3903 static bool btf_dedup_identical_structs(struct btf_dedup *d, __u32 id1, __u32 id2)
3905 const struct btf_member *m1, *m2;
3906 struct btf_type *t1, *t2;
3909 t1 = btf_type_by_id(d->btf, id1);
3910 t2 = btf_type_by_id(d->btf, id2);
3912 if (!btf_is_composite(t1) || btf_kind(t1) != btf_kind(t2))
3915 if (!btf_shallow_equal_struct(t1, t2))
3918 m1 = btf_members(t1);
3919 m2 = btf_members(t2);
3920 for (i = 0, n = btf_vlen(t1); i < n; i++, m1++, m2++) {
3921 if (m1->type != m2->type)
3928 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
3929 * call it "candidate graph" in this description for brevity) to a type graph
3930 * formed by (potential) canonical struct/union ("canonical graph" for brevity
3931 * here, though keep in mind that not all types in canonical graph are
3932 * necessarily canonical representatives themselves, some of them might be
3933 * duplicates or its uniqueness might not have been established yet).
3935 * - >0, if type graphs are equivalent;
3936 * - 0, if not equivalent;
3939 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
3940 * equivalence of BTF types at each step. If at any point BTF types in candidate
3941 * and canonical graphs are not compatible structurally, whole graphs are
3942 * incompatible. If types are structurally equivalent (i.e., all information
3943 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
3944 * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
3945 * If a type references other types, then those referenced types are checked
3946 * for equivalence recursively.
3948 * During DFS traversal, if we find that for current `canon_id` type we
3949 * already have some mapping in hypothetical map, we check for two possible
3951 * - `canon_id` is mapped to exactly the same type as `cand_id`. This will
3952 * happen when type graphs have cycles. In this case we assume those two
3953 * types are equivalent.
3954 * - `canon_id` is mapped to different type. This is contradiction in our
3955 * hypothetical mapping, because same graph in canonical graph corresponds
3956 * to two different types in candidate graph, which for equivalent type
3957 * graphs shouldn't happen. This condition terminates equivalence check
3958 * with negative result.
3960 * If type graphs traversal exhausts types to check and find no contradiction,
3961 * then type graphs are equivalent.
3963 * When checking types for equivalence, there is one special case: FWD types.
3964 * If FWD type resolution is allowed and one of the types (either from canonical
3965 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
3966 * flag) and their names match, hypothetical mapping is updated to point from
3967 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
3968 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
3970 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
3971 * if there are two exactly named (or anonymous) structs/unions that are
3972 * compatible structurally, one of which has FWD field, while other is concrete
3973 * STRUCT/UNION, but according to C sources they are different structs/unions
3974 * that are referencing different types with the same name. This is extremely
3975 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
3976 * this logic is causing problems.
3978 * Doing FWD resolution means that both candidate and/or canonical graphs can
3979 * consists of portions of the graph that come from multiple compilation units.
3980 * This is due to the fact that types within single compilation unit are always
3981 * deduplicated and FWDs are already resolved, if referenced struct/union
3982 * definiton is available. So, if we had unresolved FWD and found corresponding
3983 * STRUCT/UNION, they will be from different compilation units. This
3984 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
3985 * type graph will likely have at least two different BTF types that describe
3986 * same type (e.g., most probably there will be two different BTF types for the
3987 * same 'int' primitive type) and could even have "overlapping" parts of type
3988 * graph that describe same subset of types.
3990 * This in turn means that our assumption that each type in canonical graph
3991 * must correspond to exactly one type in candidate graph might not hold
3992 * anymore and will make it harder to detect contradictions using hypothetical
3993 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
3994 * resolution only in canonical graph. FWDs in candidate graphs are never
3995 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
3997 * - Both types in canonical and candidate graphs are FWDs. If they are
3998 * structurally equivalent, then they can either be both resolved to the
3999 * same STRUCT/UNION or not resolved at all. In both cases they are
4000 * equivalent and there is no need to resolve FWD on candidate side.
4001 * - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
4002 * so nothing to resolve as well, algorithm will check equivalence anyway.
4003 * - Type in canonical graph is FWD, while type in candidate is concrete
4004 * STRUCT/UNION. In this case candidate graph comes from single compilation
4005 * unit, so there is exactly one BTF type for each unique C type. After
4006 * resolving FWD into STRUCT/UNION, there might be more than one BTF type
4007 * in canonical graph mapping to single BTF type in candidate graph, but
4008 * because hypothetical mapping maps from canonical to candidate types, it's
4009 * alright, and we still maintain the property of having single `canon_id`
4010 * mapping to single `cand_id` (there could be two different `canon_id`
4011 * mapped to the same `cand_id`, but it's not contradictory).
4012 * - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
4013 * graph is FWD. In this case we are just going to check compatibility of
4014 * STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
4015 * assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
4016 * a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
4017 * turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
4020 static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
4023 struct btf_type *cand_type;
4024 struct btf_type *canon_type;
4025 __u32 hypot_type_id;
4030 /* if both resolve to the same canonical, they must be equivalent */
4031 if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
4034 canon_id = resolve_fwd_id(d, canon_id);
4036 hypot_type_id = d->hypot_map[canon_id];
4037 if (hypot_type_id <= BTF_MAX_NR_TYPES) {
4038 if (hypot_type_id == cand_id)
4040 /* In some cases compiler will generate different DWARF types
4041 * for *identical* array type definitions and use them for
4042 * different fields within the *same* struct. This breaks type
4043 * equivalence check, which makes an assumption that candidate
4044 * types sub-graph has a consistent and deduped-by-compiler
4045 * types within a single CU. So work around that by explicitly
4046 * allowing identical array types here.
4048 if (btf_dedup_identical_arrays(d, hypot_type_id, cand_id))
4050 /* It turns out that similar situation can happen with
4051 * struct/union sometimes, sigh... Handle the case where
4052 * structs/unions are exactly the same, down to the referenced
4053 * type IDs. Anything more complicated (e.g., if referenced
4054 * types are different, but equivalent) is *way more*
4055 * complicated and requires a many-to-many equivalence mapping.
4057 if (btf_dedup_identical_structs(d, hypot_type_id, cand_id))
4062 if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
4065 cand_type = btf_type_by_id(d->btf, cand_id);
4066 canon_type = btf_type_by_id(d->btf, canon_id);
4067 cand_kind = btf_kind(cand_type);
4068 canon_kind = btf_kind(canon_type);
4070 if (cand_type->name_off != canon_type->name_off)
4073 /* FWD <--> STRUCT/UNION equivalence check, if enabled */
4074 if ((cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
4075 && cand_kind != canon_kind) {
4079 if (cand_kind == BTF_KIND_FWD) {
4080 real_kind = canon_kind;
4081 fwd_kind = btf_fwd_kind(cand_type);
4083 real_kind = cand_kind;
4084 fwd_kind = btf_fwd_kind(canon_type);
4085 /* we'd need to resolve base FWD to STRUCT/UNION */
4086 if (fwd_kind == real_kind && canon_id < d->btf->start_id)
4087 d->hypot_adjust_canon = true;
4089 return fwd_kind == real_kind;
4092 if (cand_kind != canon_kind)
4095 switch (cand_kind) {
4097 return btf_equal_int_tag(cand_type, canon_type);
4100 return btf_compat_enum(cand_type, canon_type);
4102 case BTF_KIND_ENUM64:
4103 return btf_compat_enum64(cand_type, canon_type);
4106 case BTF_KIND_FLOAT:
4107 return btf_equal_common(cand_type, canon_type);
4109 case BTF_KIND_CONST:
4110 case BTF_KIND_VOLATILE:
4111 case BTF_KIND_RESTRICT:
4113 case BTF_KIND_TYPEDEF:
4115 case BTF_KIND_TYPE_TAG:
4116 if (cand_type->info != canon_type->info)
4118 return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4120 case BTF_KIND_ARRAY: {
4121 const struct btf_array *cand_arr, *canon_arr;
4123 if (!btf_compat_array(cand_type, canon_type))
4125 cand_arr = btf_array(cand_type);
4126 canon_arr = btf_array(canon_type);
4127 eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
4130 return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
4133 case BTF_KIND_STRUCT:
4134 case BTF_KIND_UNION: {
4135 const struct btf_member *cand_m, *canon_m;
4138 if (!btf_shallow_equal_struct(cand_type, canon_type))
4140 vlen = btf_vlen(cand_type);
4141 cand_m = btf_members(cand_type);
4142 canon_m = btf_members(canon_type);
4143 for (i = 0; i < vlen; i++) {
4144 eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
4154 case BTF_KIND_FUNC_PROTO: {
4155 const struct btf_param *cand_p, *canon_p;
4158 if (!btf_compat_fnproto(cand_type, canon_type))
4160 eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4163 vlen = btf_vlen(cand_type);
4164 cand_p = btf_params(cand_type);
4165 canon_p = btf_params(canon_type);
4166 for (i = 0; i < vlen; i++) {
4167 eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
4183 * Use hypothetical mapping, produced by successful type graph equivalence
4184 * check, to augment existing struct/union canonical mapping, where possible.
4186 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
4187 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
4188 * it doesn't matter if FWD type was part of canonical graph or candidate one,
4189 * we are recording the mapping anyway. As opposed to carefulness required
4190 * for struct/union correspondence mapping (described below), for FWD resolution
4191 * it's not important, as by the time that FWD type (reference type) will be
4192 * deduplicated all structs/unions will be deduped already anyway.
4194 * Recording STRUCT/UNION mapping is purely a performance optimization and is
4195 * not required for correctness. It needs to be done carefully to ensure that
4196 * struct/union from candidate's type graph is not mapped into corresponding
4197 * struct/union from canonical type graph that itself hasn't been resolved into
4198 * canonical representative. The only guarantee we have is that canonical
4199 * struct/union was determined as canonical and that won't change. But any
4200 * types referenced through that struct/union fields could have been not yet
4201 * resolved, so in case like that it's too early to establish any kind of
4202 * correspondence between structs/unions.
4204 * No canonical correspondence is derived for primitive types (they are already
4205 * deduplicated completely already anyway) or reference types (they rely on
4206 * stability of struct/union canonical relationship for equivalence checks).
4208 static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
4210 __u32 canon_type_id, targ_type_id;
4211 __u16 t_kind, c_kind;
4215 for (i = 0; i < d->hypot_cnt; i++) {
4216 canon_type_id = d->hypot_list[i];
4217 targ_type_id = d->hypot_map[canon_type_id];
4218 t_id = resolve_type_id(d, targ_type_id);
4219 c_id = resolve_type_id(d, canon_type_id);
4220 t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
4221 c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
4223 * Resolve FWD into STRUCT/UNION.
4224 * It's ok to resolve FWD into STRUCT/UNION that's not yet
4225 * mapped to canonical representative (as opposed to
4226 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
4227 * eventually that struct is going to be mapped and all resolved
4228 * FWDs will automatically resolve to correct canonical
4229 * representative. This will happen before ref type deduping,
4230 * which critically depends on stability of these mapping. This
4231 * stability is not a requirement for STRUCT/UNION equivalence
4235 /* if it's the split BTF case, we still need to point base FWD
4236 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4237 * will be resolved against base FWD. If we don't point base
4238 * canonical FWD to the resolved STRUCT/UNION, then all the
4239 * FWDs in split BTF won't be correctly resolved to a proper
4242 if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4243 d->map[c_id] = t_id;
4245 /* if graph equivalence determined that we'd need to adjust
4246 * base canonical types, then we need to only point base FWDs
4247 * to STRUCTs/UNIONs and do no more modifications. For all
4248 * other purposes the type graphs were not equivalent.
4250 if (d->hypot_adjust_canon)
4253 if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4254 d->map[t_id] = c_id;
4256 if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4257 c_kind != BTF_KIND_FWD &&
4258 is_type_mapped(d, c_id) &&
4259 !is_type_mapped(d, t_id)) {
4261 * as a perf optimization, we can map struct/union
4262 * that's part of type graph we just verified for
4263 * equivalence. We can do that for struct/union that has
4264 * canonical representative only, though.
4266 d->map[t_id] = c_id;
4272 * Deduplicate struct/union types.
4274 * For each struct/union type its type signature hash is calculated, taking
4275 * into account type's name, size, number, order and names of fields, but
4276 * ignoring type ID's referenced from fields, because they might not be deduped
4277 * completely until after reference types deduplication phase. This type hash
4278 * is used to iterate over all potential canonical types, sharing same hash.
4279 * For each canonical candidate we check whether type graphs that they form
4280 * (through referenced types in fields and so on) are equivalent using algorithm
4281 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4282 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4283 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4284 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4285 * potentially map other structs/unions to their canonical representatives,
4286 * if such relationship hasn't yet been established. This speeds up algorithm
4287 * by eliminating some of the duplicate work.
4289 * If no matching canonical representative was found, struct/union is marked
4290 * as canonical for itself and is added into btf_dedup->dedup_table hash map
4291 * for further look ups.
4293 static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4295 struct btf_type *cand_type, *t;
4296 struct hashmap_entry *hash_entry;
4297 /* if we don't find equivalent type, then we are canonical */
4298 __u32 new_id = type_id;
4302 /* already deduped or is in process of deduping (loop detected) */
4303 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4306 t = btf_type_by_id(d->btf, type_id);
4309 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4312 h = btf_hash_struct(t);
4313 for_each_dedup_cand(d, hash_entry, h) {
4314 __u32 cand_id = (__u32)(long)hash_entry->value;
4318 * Even though btf_dedup_is_equiv() checks for
4319 * btf_shallow_equal_struct() internally when checking two
4320 * structs (unions) for equivalence, we need to guard here
4321 * from picking matching FWD type as a dedup candidate.
4322 * This can happen due to hash collision. In such case just
4323 * relying on btf_dedup_is_equiv() would lead to potentially
4324 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4325 * FWD and compatible STRUCT/UNION are considered equivalent.
4327 cand_type = btf_type_by_id(d->btf, cand_id);
4328 if (!btf_shallow_equal_struct(t, cand_type))
4331 btf_dedup_clear_hypot_map(d);
4332 eq = btf_dedup_is_equiv(d, type_id, cand_id);
4337 btf_dedup_merge_hypot_map(d);
4338 if (d->hypot_adjust_canon) /* not really equivalent */
4344 d->map[type_id] = new_id;
4345 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4351 static int btf_dedup_struct_types(struct btf_dedup *d)
4355 for (i = 0; i < d->btf->nr_types; i++) {
4356 err = btf_dedup_struct_type(d, d->btf->start_id + i);
4364 * Deduplicate reference type.
4366 * Once all primitive and struct/union types got deduplicated, we can easily
4367 * deduplicate all other (reference) BTF types. This is done in two steps:
4369 * 1. Resolve all referenced type IDs into their canonical type IDs. This
4370 * resolution can be done either immediately for primitive or struct/union types
4371 * (because they were deduped in previous two phases) or recursively for
4372 * reference types. Recursion will always terminate at either primitive or
4373 * struct/union type, at which point we can "unwind" chain of reference types
4374 * one by one. There is no danger of encountering cycles because in C type
4375 * system the only way to form type cycle is through struct/union, so any chain
4376 * of reference types, even those taking part in a type cycle, will inevitably
4377 * reach struct/union at some point.
4379 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4380 * becomes "stable", in the sense that no further deduplication will cause
4381 * any changes to it. With that, it's now possible to calculate type's signature
4382 * hash (this time taking into account referenced type IDs) and loop over all
4383 * potential canonical representatives. If no match was found, current type
4384 * will become canonical representative of itself and will be added into
4385 * btf_dedup->dedup_table as another possible canonical representative.
4387 static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4389 struct hashmap_entry *hash_entry;
4390 __u32 new_id = type_id, cand_id;
4391 struct btf_type *t, *cand;
4392 /* if we don't find equivalent type, then we are representative type */
4396 if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4398 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4399 return resolve_type_id(d, type_id);
4401 t = btf_type_by_id(d->btf, type_id);
4402 d->map[type_id] = BTF_IN_PROGRESS_ID;
4404 switch (btf_kind(t)) {
4405 case BTF_KIND_CONST:
4406 case BTF_KIND_VOLATILE:
4407 case BTF_KIND_RESTRICT:
4409 case BTF_KIND_TYPEDEF:
4411 case BTF_KIND_TYPE_TAG:
4412 ref_type_id = btf_dedup_ref_type(d, t->type);
4413 if (ref_type_id < 0)
4415 t->type = ref_type_id;
4417 h = btf_hash_common(t);
4418 for_each_dedup_cand(d, hash_entry, h) {
4419 cand_id = (__u32)(long)hash_entry->value;
4420 cand = btf_type_by_id(d->btf, cand_id);
4421 if (btf_equal_common(t, cand)) {
4428 case BTF_KIND_DECL_TAG:
4429 ref_type_id = btf_dedup_ref_type(d, t->type);
4430 if (ref_type_id < 0)
4432 t->type = ref_type_id;
4434 h = btf_hash_int_decl_tag(t);
4435 for_each_dedup_cand(d, hash_entry, h) {
4436 cand_id = (__u32)(long)hash_entry->value;
4437 cand = btf_type_by_id(d->btf, cand_id);
4438 if (btf_equal_int_tag(t, cand)) {
4445 case BTF_KIND_ARRAY: {
4446 struct btf_array *info = btf_array(t);
4448 ref_type_id = btf_dedup_ref_type(d, info->type);
4449 if (ref_type_id < 0)
4451 info->type = ref_type_id;
4453 ref_type_id = btf_dedup_ref_type(d, info->index_type);
4454 if (ref_type_id < 0)
4456 info->index_type = ref_type_id;
4458 h = btf_hash_array(t);
4459 for_each_dedup_cand(d, hash_entry, h) {
4460 cand_id = (__u32)(long)hash_entry->value;
4461 cand = btf_type_by_id(d->btf, cand_id);
4462 if (btf_equal_array(t, cand)) {
4470 case BTF_KIND_FUNC_PROTO: {
4471 struct btf_param *param;
4475 ref_type_id = btf_dedup_ref_type(d, t->type);
4476 if (ref_type_id < 0)
4478 t->type = ref_type_id;
4481 param = btf_params(t);
4482 for (i = 0; i < vlen; i++) {
4483 ref_type_id = btf_dedup_ref_type(d, param->type);
4484 if (ref_type_id < 0)
4486 param->type = ref_type_id;
4490 h = btf_hash_fnproto(t);
4491 for_each_dedup_cand(d, hash_entry, h) {
4492 cand_id = (__u32)(long)hash_entry->value;
4493 cand = btf_type_by_id(d->btf, cand_id);
4494 if (btf_equal_fnproto(t, cand)) {
4506 d->map[type_id] = new_id;
4507 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4513 static int btf_dedup_ref_types(struct btf_dedup *d)
4517 for (i = 0; i < d->btf->nr_types; i++) {
4518 err = btf_dedup_ref_type(d, d->btf->start_id + i);
4522 /* we won't need d->dedup_table anymore */
4523 hashmap__free(d->dedup_table);
4524 d->dedup_table = NULL;
4531 * After we established for each type its corresponding canonical representative
4532 * type, we now can eliminate types that are not canonical and leave only
4533 * canonical ones layed out sequentially in memory by copying them over
4534 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
4535 * a map from original type ID to a new compacted type ID, which will be used
4536 * during next phase to "fix up" type IDs, referenced from struct/union and
4539 static int btf_dedup_compact_types(struct btf_dedup *d)
4542 __u32 next_type_id = d->btf->start_id;
4543 const struct btf_type *t;
4547 /* we are going to reuse hypot_map to store compaction remapping */
4548 d->hypot_map[0] = 0;
4549 /* base BTF types are not renumbered */
4550 for (id = 1; id < d->btf->start_id; id++)
4551 d->hypot_map[id] = id;
4552 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
4553 d->hypot_map[id] = BTF_UNPROCESSED_ID;
4555 p = d->btf->types_data;
4557 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
4558 if (d->map[id] != id)
4561 t = btf__type_by_id(d->btf, id);
4562 len = btf_type_size(t);
4567 d->hypot_map[id] = next_type_id;
4568 d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
4573 /* shrink struct btf's internal types index and update btf_header */
4574 d->btf->nr_types = next_type_id - d->btf->start_id;
4575 d->btf->type_offs_cap = d->btf->nr_types;
4576 d->btf->hdr->type_len = p - d->btf->types_data;
4577 new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
4579 if (d->btf->type_offs_cap && !new_offs)
4581 d->btf->type_offs = new_offs;
4582 d->btf->hdr->str_off = d->btf->hdr->type_len;
4583 d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
4588 * Figure out final (deduplicated and compacted) type ID for provided original
4589 * `type_id` by first resolving it into corresponding canonical type ID and
4590 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
4591 * which is populated during compaction phase.
4593 static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
4595 struct btf_dedup *d = ctx;
4596 __u32 resolved_type_id, new_type_id;
4598 resolved_type_id = resolve_type_id(d, *type_id);
4599 new_type_id = d->hypot_map[resolved_type_id];
4600 if (new_type_id > BTF_MAX_NR_TYPES)
4603 *type_id = new_type_id;
4608 * Remap referenced type IDs into deduped type IDs.
4610 * After BTF types are deduplicated and compacted, their final type IDs may
4611 * differ from original ones. The map from original to a corresponding
4612 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
4613 * compaction phase. During remapping phase we are rewriting all type IDs
4614 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
4615 * their final deduped type IDs.
4617 static int btf_dedup_remap_types(struct btf_dedup *d)
4621 for (i = 0; i < d->btf->nr_types; i++) {
4622 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
4624 r = btf_type_visit_type_ids(t, btf_dedup_remap_type_id, d);
4632 r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
4640 * Probe few well-known locations for vmlinux kernel image and try to load BTF
4641 * data out of it to use for target BTF.
4643 struct btf *btf__load_vmlinux_btf(void)
4645 const char *locations[] = {
4646 /* try canonical vmlinux BTF through sysfs first */
4647 "/sys/kernel/btf/vmlinux",
4648 /* fall back to trying to find vmlinux on disk otherwise */
4649 "/boot/vmlinux-%1$s",
4650 "/lib/modules/%1$s/vmlinux-%1$s",
4651 "/lib/modules/%1$s/build/vmlinux",
4652 "/usr/lib/modules/%1$s/kernel/vmlinux",
4653 "/usr/lib/debug/boot/vmlinux-%1$s",
4654 "/usr/lib/debug/boot/vmlinux-%1$s.debug",
4655 "/usr/lib/debug/lib/modules/%1$s/vmlinux",
4657 char path[PATH_MAX + 1];
4664 for (i = 0; i < ARRAY_SIZE(locations); i++) {
4665 snprintf(path, PATH_MAX, locations[i], buf.release);
4667 if (faccessat(AT_FDCWD, path, R_OK, AT_EACCESS))
4670 btf = btf__parse(path, NULL);
4671 err = libbpf_get_error(btf);
4672 pr_debug("loading kernel BTF '%s': %d\n", path, err);
4679 pr_warn("failed to find valid kernel BTF\n");
4680 return libbpf_err_ptr(-ESRCH);
4683 struct btf *libbpf_find_kernel_btf(void) __attribute__((alias("btf__load_vmlinux_btf")));
4685 struct btf *btf__load_module_btf(const char *module_name, struct btf *vmlinux_btf)
4689 snprintf(path, sizeof(path), "/sys/kernel/btf/%s", module_name);
4690 return btf__parse_split(path, vmlinux_btf);
4693 int btf_type_visit_type_ids(struct btf_type *t, type_id_visit_fn visit, void *ctx)
4697 switch (btf_kind(t)) {
4699 case BTF_KIND_FLOAT:
4701 case BTF_KIND_ENUM64:
4705 case BTF_KIND_CONST:
4706 case BTF_KIND_VOLATILE:
4707 case BTF_KIND_RESTRICT:
4709 case BTF_KIND_TYPEDEF:
4712 case BTF_KIND_DECL_TAG:
4713 case BTF_KIND_TYPE_TAG:
4714 return visit(&t->type, ctx);
4716 case BTF_KIND_ARRAY: {
4717 struct btf_array *a = btf_array(t);
4719 err = visit(&a->type, ctx);
4720 err = err ?: visit(&a->index_type, ctx);
4724 case BTF_KIND_STRUCT:
4725 case BTF_KIND_UNION: {
4726 struct btf_member *m = btf_members(t);
4728 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4729 err = visit(&m->type, ctx);
4736 case BTF_KIND_FUNC_PROTO: {
4737 struct btf_param *m = btf_params(t);
4739 err = visit(&t->type, ctx);
4742 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4743 err = visit(&m->type, ctx);
4750 case BTF_KIND_DATASEC: {
4751 struct btf_var_secinfo *m = btf_var_secinfos(t);
4753 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4754 err = visit(&m->type, ctx);
4766 int btf_type_visit_str_offs(struct btf_type *t, str_off_visit_fn visit, void *ctx)
4770 err = visit(&t->name_off, ctx);
4774 switch (btf_kind(t)) {
4775 case BTF_KIND_STRUCT:
4776 case BTF_KIND_UNION: {
4777 struct btf_member *m = btf_members(t);
4779 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4780 err = visit(&m->name_off, ctx);
4786 case BTF_KIND_ENUM: {
4787 struct btf_enum *m = btf_enum(t);
4789 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4790 err = visit(&m->name_off, ctx);
4796 case BTF_KIND_ENUM64: {
4797 struct btf_enum64 *m = btf_enum64(t);
4799 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4800 err = visit(&m->name_off, ctx);
4806 case BTF_KIND_FUNC_PROTO: {
4807 struct btf_param *m = btf_params(t);
4809 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4810 err = visit(&m->name_off, ctx);
4823 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
4825 const struct btf_ext_info *seg;
4826 struct btf_ext_info_sec *sec;
4829 seg = &btf_ext->func_info;
4830 for_each_btf_ext_sec(seg, sec) {
4831 struct bpf_func_info_min *rec;
4833 for_each_btf_ext_rec(seg, sec, i, rec) {
4834 err = visit(&rec->type_id, ctx);
4840 seg = &btf_ext->core_relo_info;
4841 for_each_btf_ext_sec(seg, sec) {
4842 struct bpf_core_relo *rec;
4844 for_each_btf_ext_rec(seg, sec, i, rec) {
4845 err = visit(&rec->type_id, ctx);
4854 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
4856 const struct btf_ext_info *seg;
4857 struct btf_ext_info_sec *sec;
4860 seg = &btf_ext->func_info;
4861 for_each_btf_ext_sec(seg, sec) {
4862 err = visit(&sec->sec_name_off, ctx);
4867 seg = &btf_ext->line_info;
4868 for_each_btf_ext_sec(seg, sec) {
4869 struct bpf_line_info_min *rec;
4871 err = visit(&sec->sec_name_off, ctx);
4875 for_each_btf_ext_rec(seg, sec, i, rec) {
4876 err = visit(&rec->file_name_off, ctx);
4879 err = visit(&rec->line_off, ctx);
4885 seg = &btf_ext->core_relo_info;
4886 for_each_btf_ext_sec(seg, sec) {
4887 struct bpf_core_relo *rec;
4889 err = visit(&sec->sec_name_off, ctx);
4893 for_each_btf_ext_rec(seg, sec, i, rec) {
4894 err = visit(&rec->access_str_off, ctx);