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"
25 #include "str_error.h"
27 #define BTF_MAX_NR_TYPES 0x7fffffffU
28 #define BTF_MAX_STR_OFFSET 0x7fffffffU
30 static struct btf_type btf_void;
33 /* raw BTF data in native endianness */
35 /* raw BTF data in non-native endianness */
36 void *raw_data_swapped;
38 /* whether target endianness differs from the native one */
42 * When BTF is loaded from an ELF or raw memory it is stored
43 * in a contiguous memory block. The hdr, type_data, and, strs_data
44 * point inside that memory region to their respective parts of BTF
47 * +--------------------------------+
48 * | Header | Types | Strings |
49 * +--------------------------------+
54 * strs_data------------+
56 * If BTF data is later modified, e.g., due to types added or
57 * removed, BTF deduplication performed, etc, this contiguous
58 * representation is broken up into three independently allocated
59 * memory regions to be able to modify them independently.
60 * raw_data is nulled out at that point, but can be later allocated
61 * and cached again if user calls btf__raw_data(), at which point
62 * raw_data will contain a contiguous copy of header, types, and
65 * +----------+ +---------+ +-----------+
66 * | Header | | Types | | Strings |
67 * +----------+ +---------+ +-----------+
72 * strset__data(strs_set)-----+
74 * +----------+---------+-----------+
75 * | Header | Types | Strings |
76 * raw_data----->+----------+---------+-----------+
78 struct btf_header *hdr;
81 size_t types_data_cap; /* used size stored in hdr->type_len */
83 /* type ID to `struct btf_type *` lookup index
84 * type_offs[0] corresponds to the first non-VOID type:
85 * - for base BTF it's type [1];
86 * - for split BTF it's the first non-base BTF type.
90 /* number of types in this BTF instance:
91 * - doesn't include special [0] void type;
92 * - for split BTF counts number of types added on top of base BTF.
95 /* if not NULL, points to the base BTF on top of which the current
99 /* BTF type ID of the first type in this BTF instance:
100 * - for base BTF it's equal to 1;
101 * - for split BTF it's equal to biggest type ID of base BTF plus 1.
104 /* logical string offset of this BTF instance:
105 * - for base BTF it's equal to 0;
106 * - for split BTF it's equal to total size of base BTF's string section size.
110 /* only one of strs_data or strs_set can be non-NULL, depending on
111 * whether BTF is in a modifiable state (strs_set is used) or not
112 * (strs_data points inside raw_data)
115 /* a set of unique strings */
116 struct strset *strs_set;
117 /* whether strings are already deduplicated */
120 /* whether base_btf should be freed in btf_free for this instance */
123 /* BTF object FD, if loaded into kernel */
126 /* Pointer size (in bytes) for a target architecture of this BTF */
130 static inline __u64 ptr_to_u64(const void *ptr)
132 return (__u64) (unsigned long) ptr;
135 /* Ensure given dynamically allocated memory region pointed to by *data* with
136 * capacity of *cap_cnt* elements each taking *elem_sz* bytes has enough
137 * memory to accommodate *add_cnt* new elements, assuming *cur_cnt* elements
138 * are already used. At most *max_cnt* elements can be ever allocated.
139 * If necessary, memory is reallocated and all existing data is copied over,
140 * new pointer to the memory region is stored at *data, new memory region
141 * capacity (in number of elements) is stored in *cap.
142 * On success, memory pointer to the beginning of unused memory is returned.
143 * On error, NULL is returned.
145 void *libbpf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz,
146 size_t cur_cnt, size_t max_cnt, size_t add_cnt)
151 if (cur_cnt + add_cnt <= *cap_cnt)
152 return *data + cur_cnt * elem_sz;
154 /* requested more than the set limit */
155 if (cur_cnt + add_cnt > max_cnt)
159 new_cnt += new_cnt / 4; /* expand by 25% */
160 if (new_cnt < 16) /* but at least 16 elements */
162 if (new_cnt > max_cnt) /* but not exceeding a set limit */
164 if (new_cnt < cur_cnt + add_cnt) /* also ensure we have enough memory */
165 new_cnt = cur_cnt + add_cnt;
167 new_data = libbpf_reallocarray(*data, new_cnt, elem_sz);
171 /* zero out newly allocated portion of memory */
172 memset(new_data + (*cap_cnt) * elem_sz, 0, (new_cnt - *cap_cnt) * elem_sz);
176 return new_data + cur_cnt * elem_sz;
179 /* Ensure given dynamically allocated memory region has enough allocated space
180 * to accommodate *need_cnt* elements of size *elem_sz* bytes each
182 int libbpf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt)
186 if (need_cnt <= *cap_cnt)
189 p = libbpf_add_mem(data, cap_cnt, elem_sz, *cap_cnt, SIZE_MAX, need_cnt - *cap_cnt);
196 static void *btf_add_type_offs_mem(struct btf *btf, size_t add_cnt)
198 return libbpf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
199 btf->nr_types, BTF_MAX_NR_TYPES, add_cnt);
202 static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
206 p = btf_add_type_offs_mem(btf, 1);
214 static void btf_bswap_hdr(struct btf_header *h)
216 h->magic = bswap_16(h->magic);
217 h->hdr_len = bswap_32(h->hdr_len);
218 h->type_off = bswap_32(h->type_off);
219 h->type_len = bswap_32(h->type_len);
220 h->str_off = bswap_32(h->str_off);
221 h->str_len = bswap_32(h->str_len);
224 static int btf_parse_hdr(struct btf *btf)
226 struct btf_header *hdr = btf->hdr;
229 if (btf->raw_size < sizeof(struct btf_header)) {
230 pr_debug("BTF header not found\n");
234 if (hdr->magic == bswap_16(BTF_MAGIC)) {
235 btf->swapped_endian = true;
236 if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
237 pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
238 bswap_32(hdr->hdr_len));
242 } else if (hdr->magic != BTF_MAGIC) {
243 pr_debug("Invalid BTF magic: %x\n", hdr->magic);
247 if (btf->raw_size < hdr->hdr_len) {
248 pr_debug("BTF header len %u larger than data size %u\n",
249 hdr->hdr_len, btf->raw_size);
253 meta_left = btf->raw_size - hdr->hdr_len;
254 if (meta_left < (long long)hdr->str_off + hdr->str_len) {
255 pr_debug("Invalid BTF total size: %u\n", btf->raw_size);
259 if ((long long)hdr->type_off + hdr->type_len > hdr->str_off) {
260 pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
261 hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
265 if (hdr->type_off % 4) {
266 pr_debug("BTF type section is not aligned to 4 bytes\n");
273 static int btf_parse_str_sec(struct btf *btf)
275 const struct btf_header *hdr = btf->hdr;
276 const char *start = btf->strs_data;
277 const char *end = start + btf->hdr->str_len;
279 if (btf->base_btf && hdr->str_len == 0)
281 if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET || end[-1]) {
282 pr_debug("Invalid BTF string section\n");
285 if (!btf->base_btf && start[0]) {
286 pr_debug("Invalid BTF string section\n");
292 static int btf_type_size(const struct btf_type *t)
294 const int base_size = sizeof(struct btf_type);
295 __u16 vlen = btf_vlen(t);
297 switch (btf_kind(t)) {
300 case BTF_KIND_VOLATILE:
301 case BTF_KIND_RESTRICT:
303 case BTF_KIND_TYPEDEF:
306 case BTF_KIND_TYPE_TAG:
309 return base_size + sizeof(__u32);
311 return base_size + vlen * sizeof(struct btf_enum);
312 case BTF_KIND_ENUM64:
313 return base_size + vlen * sizeof(struct btf_enum64);
315 return base_size + sizeof(struct btf_array);
316 case BTF_KIND_STRUCT:
318 return base_size + vlen * sizeof(struct btf_member);
319 case BTF_KIND_FUNC_PROTO:
320 return base_size + vlen * sizeof(struct btf_param);
322 return base_size + sizeof(struct btf_var);
323 case BTF_KIND_DATASEC:
324 return base_size + vlen * sizeof(struct btf_var_secinfo);
325 case BTF_KIND_DECL_TAG:
326 return base_size + sizeof(struct btf_decl_tag);
328 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
333 static void btf_bswap_type_base(struct btf_type *t)
335 t->name_off = bswap_32(t->name_off);
336 t->info = bswap_32(t->info);
337 t->type = bswap_32(t->type);
340 static int btf_bswap_type_rest(struct btf_type *t)
342 struct btf_var_secinfo *v;
343 struct btf_enum64 *e64;
344 struct btf_member *m;
348 __u16 vlen = btf_vlen(t);
351 switch (btf_kind(t)) {
354 case BTF_KIND_VOLATILE:
355 case BTF_KIND_RESTRICT:
357 case BTF_KIND_TYPEDEF:
360 case BTF_KIND_TYPE_TAG:
363 *(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
366 for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
367 e->name_off = bswap_32(e->name_off);
368 e->val = bswap_32(e->val);
371 case BTF_KIND_ENUM64:
372 for (i = 0, e64 = btf_enum64(t); i < vlen; i++, e64++) {
373 e64->name_off = bswap_32(e64->name_off);
374 e64->val_lo32 = bswap_32(e64->val_lo32);
375 e64->val_hi32 = bswap_32(e64->val_hi32);
380 a->type = bswap_32(a->type);
381 a->index_type = bswap_32(a->index_type);
382 a->nelems = bswap_32(a->nelems);
384 case BTF_KIND_STRUCT:
386 for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
387 m->name_off = bswap_32(m->name_off);
388 m->type = bswap_32(m->type);
389 m->offset = bswap_32(m->offset);
392 case BTF_KIND_FUNC_PROTO:
393 for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
394 p->name_off = bswap_32(p->name_off);
395 p->type = bswap_32(p->type);
399 btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
401 case BTF_KIND_DATASEC:
402 for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
403 v->type = bswap_32(v->type);
404 v->offset = bswap_32(v->offset);
405 v->size = bswap_32(v->size);
408 case BTF_KIND_DECL_TAG:
409 btf_decl_tag(t)->component_idx = bswap_32(btf_decl_tag(t)->component_idx);
412 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
417 static int btf_parse_type_sec(struct btf *btf)
419 struct btf_header *hdr = btf->hdr;
420 void *next_type = btf->types_data;
421 void *end_type = next_type + hdr->type_len;
424 while (next_type + sizeof(struct btf_type) <= end_type) {
425 if (btf->swapped_endian)
426 btf_bswap_type_base(next_type);
428 type_size = btf_type_size(next_type);
431 if (next_type + type_size > end_type) {
432 pr_warn("BTF type [%d] is malformed\n", btf->start_id + btf->nr_types);
436 if (btf->swapped_endian && btf_bswap_type_rest(next_type))
439 err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
443 next_type += type_size;
447 if (next_type != end_type) {
448 pr_warn("BTF types data is malformed\n");
455 static int btf_validate_str(const struct btf *btf, __u32 str_off, const char *what, __u32 type_id)
459 s = btf__str_by_offset(btf, str_off);
461 pr_warn("btf: type [%u]: invalid %s (string offset %u)\n", type_id, what, str_off);
468 static int btf_validate_id(const struct btf *btf, __u32 id, __u32 ctx_id)
470 const struct btf_type *t;
472 t = btf__type_by_id(btf, id);
474 pr_warn("btf: type [%u]: invalid referenced type ID %u\n", ctx_id, id);
481 static int btf_validate_type(const struct btf *btf, const struct btf_type *t, __u32 id)
483 __u32 kind = btf_kind(t);
486 err = btf_validate_str(btf, t->name_off, "type name", id);
497 case BTF_KIND_TYPEDEF:
498 case BTF_KIND_VOLATILE:
500 case BTF_KIND_RESTRICT:
502 case BTF_KIND_DECL_TAG:
503 case BTF_KIND_TYPE_TAG:
504 err = btf_validate_id(btf, t->type, id);
508 case BTF_KIND_ARRAY: {
509 const struct btf_array *a = btf_array(t);
511 err = btf_validate_id(btf, a->type, id);
512 err = err ?: btf_validate_id(btf, a->index_type, id);
517 case BTF_KIND_STRUCT:
518 case BTF_KIND_UNION: {
519 const struct btf_member *m = btf_members(t);
522 for (i = 0; i < n; i++, m++) {
523 err = btf_validate_str(btf, m->name_off, "field name", id);
524 err = err ?: btf_validate_id(btf, m->type, id);
530 case BTF_KIND_ENUM: {
531 const struct btf_enum *m = btf_enum(t);
534 for (i = 0; i < n; i++, m++) {
535 err = btf_validate_str(btf, m->name_off, "enum name", id);
541 case BTF_KIND_ENUM64: {
542 const struct btf_enum64 *m = btf_enum64(t);
545 for (i = 0; i < n; i++, m++) {
546 err = btf_validate_str(btf, m->name_off, "enum name", id);
552 case BTF_KIND_FUNC: {
553 const struct btf_type *ft;
555 err = btf_validate_id(btf, t->type, id);
558 ft = btf__type_by_id(btf, t->type);
559 if (btf_kind(ft) != BTF_KIND_FUNC_PROTO) {
560 pr_warn("btf: type [%u]: referenced type [%u] is not FUNC_PROTO\n", id, t->type);
565 case BTF_KIND_FUNC_PROTO: {
566 const struct btf_param *m = btf_params(t);
569 for (i = 0; i < n; i++, m++) {
570 err = btf_validate_str(btf, m->name_off, "param name", id);
571 err = err ?: btf_validate_id(btf, m->type, id);
577 case BTF_KIND_DATASEC: {
578 const struct btf_var_secinfo *m = btf_var_secinfos(t);
581 for (i = 0; i < n; i++, m++) {
582 err = btf_validate_id(btf, m->type, id);
589 pr_warn("btf: type [%u]: unrecognized kind %u\n", id, kind);
595 /* Validate basic sanity of BTF. It's intentionally less thorough than
596 * kernel's validation and validates only properties of BTF that libbpf relies
597 * on to be correct (e.g., valid type IDs, valid string offsets, etc)
599 static int btf_sanity_check(const struct btf *btf)
601 const struct btf_type *t;
602 __u32 i, n = btf__type_cnt(btf);
605 for (i = btf->start_id; i < n; i++) {
606 t = btf_type_by_id(btf, i);
607 err = btf_validate_type(btf, t, i);
614 __u32 btf__type_cnt(const struct btf *btf)
616 return btf->start_id + btf->nr_types;
619 const struct btf *btf__base_btf(const struct btf *btf)
621 return btf->base_btf;
624 /* internal helper returning non-const pointer to a type */
625 struct btf_type *btf_type_by_id(const struct btf *btf, __u32 type_id)
629 if (type_id < btf->start_id)
630 return btf_type_by_id(btf->base_btf, type_id);
631 return btf->types_data + btf->type_offs[type_id - btf->start_id];
634 const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
636 if (type_id >= btf->start_id + btf->nr_types)
637 return errno = EINVAL, NULL;
638 return btf_type_by_id((struct btf *)btf, type_id);
641 static int determine_ptr_size(const struct btf *btf)
643 static const char * const long_aliases[] = {
656 const struct btf_type *t;
660 if (btf->base_btf && btf->base_btf->ptr_sz > 0)
661 return btf->base_btf->ptr_sz;
663 n = btf__type_cnt(btf);
664 for (i = 1; i < n; i++) {
665 t = btf__type_by_id(btf, i);
669 if (t->size != 4 && t->size != 8)
672 name = btf__name_by_offset(btf, t->name_off);
676 for (j = 0; j < ARRAY_SIZE(long_aliases); j++) {
677 if (strcmp(name, long_aliases[j]) == 0)
685 static size_t btf_ptr_sz(const struct btf *btf)
688 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
689 return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
692 /* Return pointer size this BTF instance assumes. The size is heuristically
693 * determined by looking for 'long' or 'unsigned long' integer type and
694 * recording its size in bytes. If BTF type information doesn't have any such
695 * type, this function returns 0. In the latter case, native architecture's
696 * pointer size is assumed, so will be either 4 or 8, depending on
697 * architecture that libbpf was compiled for. It's possible to override
698 * guessed value by using btf__set_pointer_size() API.
700 size_t btf__pointer_size(const struct btf *btf)
703 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
706 /* not enough BTF type info to guess */
712 /* Override or set pointer size in bytes. Only values of 4 and 8 are
715 int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
717 if (ptr_sz != 4 && ptr_sz != 8)
718 return libbpf_err(-EINVAL);
719 btf->ptr_sz = ptr_sz;
723 static bool is_host_big_endian(void)
725 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
727 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
730 # error "Unrecognized __BYTE_ORDER__"
734 enum btf_endianness btf__endianness(const struct btf *btf)
736 if (is_host_big_endian())
737 return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
739 return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
742 int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
744 if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
745 return libbpf_err(-EINVAL);
747 btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
748 if (!btf->swapped_endian) {
749 free(btf->raw_data_swapped);
750 btf->raw_data_swapped = NULL;
755 static bool btf_type_is_void(const struct btf_type *t)
757 return t == &btf_void || btf_is_fwd(t);
760 static bool btf_type_is_void_or_null(const struct btf_type *t)
762 return !t || btf_type_is_void(t);
765 #define MAX_RESOLVE_DEPTH 32
767 __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
769 const struct btf_array *array;
770 const struct btf_type *t;
775 t = btf__type_by_id(btf, type_id);
776 for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t); i++) {
777 switch (btf_kind(t)) {
779 case BTF_KIND_STRUCT:
782 case BTF_KIND_ENUM64:
783 case BTF_KIND_DATASEC:
788 size = btf_ptr_sz(btf);
790 case BTF_KIND_TYPEDEF:
791 case BTF_KIND_VOLATILE:
793 case BTF_KIND_RESTRICT:
795 case BTF_KIND_DECL_TAG:
796 case BTF_KIND_TYPE_TAG:
800 array = btf_array(t);
801 if (nelems && array->nelems > UINT32_MAX / nelems)
802 return libbpf_err(-E2BIG);
803 nelems *= array->nelems;
804 type_id = array->type;
807 return libbpf_err(-EINVAL);
810 t = btf__type_by_id(btf, type_id);
815 return libbpf_err(-EINVAL);
816 if (nelems && size > UINT32_MAX / nelems)
817 return libbpf_err(-E2BIG);
819 return nelems * size;
822 int btf__align_of(const struct btf *btf, __u32 id)
824 const struct btf_type *t = btf__type_by_id(btf, id);
825 __u16 kind = btf_kind(t);
830 case BTF_KIND_ENUM64:
832 return min(btf_ptr_sz(btf), (size_t)t->size);
834 return btf_ptr_sz(btf);
835 case BTF_KIND_TYPEDEF:
836 case BTF_KIND_VOLATILE:
838 case BTF_KIND_RESTRICT:
839 case BTF_KIND_TYPE_TAG:
840 return btf__align_of(btf, t->type);
842 return btf__align_of(btf, btf_array(t)->type);
843 case BTF_KIND_STRUCT:
844 case BTF_KIND_UNION: {
845 const struct btf_member *m = btf_members(t);
846 __u16 vlen = btf_vlen(t);
847 int i, max_align = 1, align;
849 for (i = 0; i < vlen; i++, m++) {
850 align = btf__align_of(btf, m->type);
852 return libbpf_err(align);
853 max_align = max(max_align, align);
855 /* if field offset isn't aligned according to field
856 * type's alignment, then struct must be packed
858 if (btf_member_bitfield_size(t, i) == 0 &&
859 (m->offset % (8 * align)) != 0)
863 /* if struct/union size isn't a multiple of its alignment,
864 * then struct must be packed
866 if ((t->size % max_align) != 0)
872 pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
873 return errno = EINVAL, 0;
877 int btf__resolve_type(const struct btf *btf, __u32 type_id)
879 const struct btf_type *t;
882 t = btf__type_by_id(btf, type_id);
883 while (depth < MAX_RESOLVE_DEPTH &&
884 !btf_type_is_void_or_null(t) &&
885 (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
887 t = btf__type_by_id(btf, type_id);
891 if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
892 return libbpf_err(-EINVAL);
897 __s32 btf__find_by_name(const struct btf *btf, const char *type_name)
899 __u32 i, nr_types = btf__type_cnt(btf);
901 if (!strcmp(type_name, "void"))
904 for (i = 1; i < nr_types; i++) {
905 const struct btf_type *t = btf__type_by_id(btf, i);
906 const char *name = btf__name_by_offset(btf, t->name_off);
908 if (name && !strcmp(type_name, name))
912 return libbpf_err(-ENOENT);
915 static __s32 btf_find_by_name_kind(const struct btf *btf, int start_id,
916 const char *type_name, __u32 kind)
918 __u32 i, nr_types = btf__type_cnt(btf);
920 if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
923 for (i = start_id; i < nr_types; i++) {
924 const struct btf_type *t = btf__type_by_id(btf, i);
927 if (btf_kind(t) != kind)
929 name = btf__name_by_offset(btf, t->name_off);
930 if (name && !strcmp(type_name, name))
934 return libbpf_err(-ENOENT);
937 __s32 btf__find_by_name_kind_own(const struct btf *btf, const char *type_name,
940 return btf_find_by_name_kind(btf, btf->start_id, type_name, kind);
943 __s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
946 return btf_find_by_name_kind(btf, 1, type_name, kind);
949 static bool btf_is_modifiable(const struct btf *btf)
951 return (void *)btf->hdr != btf->raw_data;
954 void btf__free(struct btf *btf)
956 if (IS_ERR_OR_NULL(btf))
962 if (btf_is_modifiable(btf)) {
963 /* if BTF was modified after loading, it will have a split
964 * in-memory representation for header, types, and strings
965 * sections, so we need to free all of them individually. It
966 * might still have a cached contiguous raw data present,
967 * which will be unconditionally freed below.
970 free(btf->types_data);
971 strset__free(btf->strs_set);
974 free(btf->raw_data_swapped);
975 free(btf->type_offs);
977 btf__free(btf->base_btf);
981 static struct btf *btf_new_empty(struct btf *base_btf)
985 btf = calloc(1, sizeof(*btf));
987 return ERR_PTR(-ENOMEM);
991 btf->start_str_off = 0;
993 btf->ptr_sz = sizeof(void *);
994 btf->swapped_endian = false;
997 btf->base_btf = base_btf;
998 btf->start_id = btf__type_cnt(base_btf);
999 btf->start_str_off = base_btf->hdr->str_len;
1000 btf->swapped_endian = base_btf->swapped_endian;
1003 /* +1 for empty string at offset 0 */
1004 btf->raw_size = sizeof(struct btf_header) + (base_btf ? 0 : 1);
1005 btf->raw_data = calloc(1, btf->raw_size);
1006 if (!btf->raw_data) {
1008 return ERR_PTR(-ENOMEM);
1011 btf->hdr = btf->raw_data;
1012 btf->hdr->hdr_len = sizeof(struct btf_header);
1013 btf->hdr->magic = BTF_MAGIC;
1014 btf->hdr->version = BTF_VERSION;
1016 btf->types_data = btf->raw_data + btf->hdr->hdr_len;
1017 btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
1018 btf->hdr->str_len = base_btf ? 0 : 1; /* empty string at offset 0 */
1023 struct btf *btf__new_empty(void)
1025 return libbpf_ptr(btf_new_empty(NULL));
1028 struct btf *btf__new_empty_split(struct btf *base_btf)
1030 return libbpf_ptr(btf_new_empty(base_btf));
1033 static struct btf *btf_new(const void *data, __u32 size, struct btf *base_btf)
1038 btf = calloc(1, sizeof(struct btf));
1040 return ERR_PTR(-ENOMEM);
1044 btf->start_str_off = 0;
1048 btf->base_btf = base_btf;
1049 btf->start_id = btf__type_cnt(base_btf);
1050 btf->start_str_off = base_btf->hdr->str_len;
1053 btf->raw_data = malloc(size);
1054 if (!btf->raw_data) {
1058 memcpy(btf->raw_data, data, size);
1059 btf->raw_size = size;
1061 btf->hdr = btf->raw_data;
1062 err = btf_parse_hdr(btf);
1066 btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
1067 btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
1069 err = btf_parse_str_sec(btf);
1070 err = err ?: btf_parse_type_sec(btf);
1071 err = err ?: btf_sanity_check(btf);
1078 return ERR_PTR(err);
1084 struct btf *btf__new(const void *data, __u32 size)
1086 return libbpf_ptr(btf_new(data, size, NULL));
1089 struct btf *btf__new_split(const void *data, __u32 size, struct btf *base_btf)
1091 return libbpf_ptr(btf_new(data, size, base_btf));
1094 struct btf_elf_secs {
1096 Elf_Data *btf_ext_data;
1097 Elf_Data *btf_base_data;
1100 static int btf_find_elf_sections(Elf *elf, const char *path, struct btf_elf_secs *secs)
1102 Elf_Scn *scn = NULL;
1108 if (!gelf_getehdr(elf, &ehdr)) {
1109 pr_warn("failed to get EHDR from %s\n", path);
1113 if (elf_getshdrstrndx(elf, &shstrndx)) {
1114 pr_warn("failed to get section names section index for %s\n",
1119 if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
1120 pr_warn("failed to get e_shstrndx from %s\n", path);
1124 while ((scn = elf_nextscn(elf, scn)) != NULL) {
1130 if (gelf_getshdr(scn, &sh) != &sh) {
1131 pr_warn("failed to get section(%d) header from %s\n",
1135 name = elf_strptr(elf, shstrndx, sh.sh_name);
1137 pr_warn("failed to get section(%d) name from %s\n",
1142 if (strcmp(name, BTF_ELF_SEC) == 0)
1143 field = &secs->btf_data;
1144 else if (strcmp(name, BTF_EXT_ELF_SEC) == 0)
1145 field = &secs->btf_ext_data;
1146 else if (strcmp(name, BTF_BASE_ELF_SEC) == 0)
1147 field = &secs->btf_base_data;
1151 data = elf_getdata(scn, 0);
1153 pr_warn("failed to get section(%d, %s) data from %s\n",
1163 return -LIBBPF_ERRNO__FORMAT;
1166 static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,
1167 struct btf_ext **btf_ext)
1169 struct btf_elf_secs secs = {};
1170 struct btf *dist_base_btf = NULL;
1171 struct btf *btf = NULL;
1172 int err = 0, fd = -1;
1175 if (elf_version(EV_CURRENT) == EV_NONE) {
1176 pr_warn("failed to init libelf for %s\n", path);
1177 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1180 fd = open(path, O_RDONLY | O_CLOEXEC);
1183 pr_warn("failed to open %s: %s\n", path, errstr(err));
1184 return ERR_PTR(err);
1187 elf = elf_begin(fd, ELF_C_READ, NULL);
1189 pr_warn("failed to open %s as ELF file\n", path);
1193 err = btf_find_elf_sections(elf, path, &secs);
1197 if (!secs.btf_data) {
1198 pr_warn("failed to find '%s' ELF section in %s\n", BTF_ELF_SEC, path);
1203 if (secs.btf_base_data) {
1204 dist_base_btf = btf_new(secs.btf_base_data->d_buf, secs.btf_base_data->d_size,
1206 if (IS_ERR(dist_base_btf)) {
1207 err = PTR_ERR(dist_base_btf);
1208 dist_base_btf = NULL;
1213 btf = btf_new(secs.btf_data->d_buf, secs.btf_data->d_size,
1214 dist_base_btf ?: base_btf);
1219 if (dist_base_btf && base_btf) {
1220 err = btf__relocate(btf, base_btf);
1223 btf__free(dist_base_btf);
1224 dist_base_btf = NULL;
1228 btf->owns_base = true;
1230 switch (gelf_getclass(elf)) {
1232 btf__set_pointer_size(btf, 4);
1235 btf__set_pointer_size(btf, 8);
1238 pr_warn("failed to get ELF class (bitness) for %s\n", path);
1242 if (btf_ext && secs.btf_ext_data) {
1243 *btf_ext = btf_ext__new(secs.btf_ext_data->d_buf, secs.btf_ext_data->d_size);
1244 if (IS_ERR(*btf_ext)) {
1245 err = PTR_ERR(*btf_ext);
1248 } else if (btf_ext) {
1260 btf_ext__free(*btf_ext);
1261 btf__free(dist_base_btf);
1264 return ERR_PTR(err);
1267 struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
1269 return libbpf_ptr(btf_parse_elf(path, NULL, btf_ext));
1272 struct btf *btf__parse_elf_split(const char *path, struct btf *base_btf)
1274 return libbpf_ptr(btf_parse_elf(path, base_btf, NULL));
1277 static struct btf *btf_parse_raw(const char *path, struct btf *base_btf)
1279 struct btf *btf = NULL;
1286 f = fopen(path, "rbe");
1292 /* check BTF magic */
1293 if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
1297 if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
1298 /* definitely not a raw BTF */
1304 if (fseek(f, 0, SEEK_END)) {
1313 /* rewind to the start */
1314 if (fseek(f, 0, SEEK_SET)) {
1319 /* pre-alloc memory and read all of BTF data */
1325 if (fread(data, 1, sz, f) < sz) {
1330 /* finally parse BTF data */
1331 btf = btf_new(data, sz, base_btf);
1337 return err ? ERR_PTR(err) : btf;
1340 struct btf *btf__parse_raw(const char *path)
1342 return libbpf_ptr(btf_parse_raw(path, NULL));
1345 struct btf *btf__parse_raw_split(const char *path, struct btf *base_btf)
1347 return libbpf_ptr(btf_parse_raw(path, base_btf));
1350 static struct btf *btf_parse(const char *path, struct btf *base_btf, struct btf_ext **btf_ext)
1358 btf = btf_parse_raw(path, base_btf);
1359 err = libbpf_get_error(btf);
1363 return ERR_PTR(err);
1364 return btf_parse_elf(path, base_btf, btf_ext);
1367 struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
1369 return libbpf_ptr(btf_parse(path, NULL, btf_ext));
1372 struct btf *btf__parse_split(const char *path, struct btf *base_btf)
1374 return libbpf_ptr(btf_parse(path, base_btf, NULL));
1377 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1379 int btf_load_into_kernel(struct btf *btf,
1380 char *log_buf, size_t log_sz, __u32 log_level,
1383 LIBBPF_OPTS(bpf_btf_load_opts, opts);
1384 __u32 buf_sz = 0, raw_size;
1385 char *buf = NULL, *tmp;
1390 return libbpf_err(-EEXIST);
1391 if (log_sz && !log_buf)
1392 return libbpf_err(-EINVAL);
1394 /* cache native raw data representation */
1395 raw_data = btf_get_raw_data(btf, &raw_size, false);
1400 btf->raw_size = raw_size;
1401 btf->raw_data = raw_data;
1404 /* if log_level is 0, we won't provide log_buf/log_size to the kernel,
1405 * initially. Only if BTF loading fails, we bump log_level to 1 and
1406 * retry, using either auto-allocated or custom log_buf. This way
1407 * non-NULL custom log_buf provides a buffer just in case, but hopes
1408 * for successful load and no need for log_buf.
1411 /* if caller didn't provide custom log_buf, we'll keep
1412 * allocating our own progressively bigger buffers for BTF
1416 buf_sz = max((__u32)BPF_LOG_BUF_SIZE, buf_sz * 2);
1417 tmp = realloc(buf, buf_sz);
1426 opts.log_buf = log_buf ? log_buf : buf;
1427 opts.log_size = log_buf ? log_sz : buf_sz;
1428 opts.log_level = log_level;
1431 opts.token_fd = token_fd;
1433 opts.btf_flags |= BPF_F_TOKEN_FD;
1435 btf->fd = bpf_btf_load(raw_data, raw_size, &opts);
1437 /* time to turn on verbose mode and try again */
1438 if (log_level == 0) {
1442 /* only retry if caller didn't provide custom log_buf, but
1443 * make sure we can never overflow buf_sz
1445 if (!log_buf && errno == ENOSPC && buf_sz <= UINT_MAX / 2)
1449 pr_warn("BTF loading error: %s\n", errstr(err));
1450 /* don't print out contents of custom log_buf */
1451 if (!log_buf && buf[0])
1452 pr_warn("-- BEGIN BTF LOAD LOG ---\n%s\n-- END BTF LOAD LOG --\n", buf);
1457 return libbpf_err(err);
1460 int btf__load_into_kernel(struct btf *btf)
1462 return btf_load_into_kernel(btf, NULL, 0, 0, 0);
1465 int btf__fd(const struct btf *btf)
1470 void btf__set_fd(struct btf *btf, int fd)
1475 static const void *btf_strs_data(const struct btf *btf)
1477 return btf->strs_data ? btf->strs_data : strset__data(btf->strs_set);
1480 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1482 struct btf_header *hdr = btf->hdr;
1488 data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1490 *size = btf->raw_size;
1494 data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1495 data = calloc(1, data_sz);
1500 memcpy(p, hdr, hdr->hdr_len);
1505 memcpy(p, btf->types_data, hdr->type_len);
1507 for (i = 0; i < btf->nr_types; i++) {
1508 t = p + btf->type_offs[i];
1509 /* btf_bswap_type_rest() relies on native t->info, so
1510 * we swap base type info after we swapped all the
1511 * additional information
1513 if (btf_bswap_type_rest(t))
1515 btf_bswap_type_base(t);
1520 memcpy(p, btf_strs_data(btf), hdr->str_len);
1530 const void *btf__raw_data(const struct btf *btf_ro, __u32 *size)
1532 struct btf *btf = (struct btf *)btf_ro;
1536 data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1538 return errno = ENOMEM, NULL;
1540 btf->raw_size = data_sz;
1541 if (btf->swapped_endian)
1542 btf->raw_data_swapped = data;
1544 btf->raw_data = data;
1549 __attribute__((alias("btf__raw_data")))
1550 const void *btf__get_raw_data(const struct btf *btf, __u32 *size);
1552 const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
1554 if (offset < btf->start_str_off)
1555 return btf__str_by_offset(btf->base_btf, offset);
1556 else if (offset - btf->start_str_off < btf->hdr->str_len)
1557 return btf_strs_data(btf) + (offset - btf->start_str_off);
1559 return errno = EINVAL, NULL;
1562 const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1564 return btf__str_by_offset(btf, offset);
1567 struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf)
1569 struct bpf_btf_info btf_info;
1570 __u32 len = sizeof(btf_info);
1576 /* we won't know btf_size until we call bpf_btf_get_info_by_fd(). so
1577 * let's start with a sane default - 4KiB here - and resize it only if
1578 * bpf_btf_get_info_by_fd() needs a bigger buffer.
1581 ptr = malloc(last_size);
1583 return ERR_PTR(-ENOMEM);
1585 memset(&btf_info, 0, sizeof(btf_info));
1586 btf_info.btf = ptr_to_u64(ptr);
1587 btf_info.btf_size = last_size;
1588 err = bpf_btf_get_info_by_fd(btf_fd, &btf_info, &len);
1590 if (!err && btf_info.btf_size > last_size) {
1593 last_size = btf_info.btf_size;
1594 temp_ptr = realloc(ptr, last_size);
1596 btf = ERR_PTR(-ENOMEM);
1601 len = sizeof(btf_info);
1602 memset(&btf_info, 0, sizeof(btf_info));
1603 btf_info.btf = ptr_to_u64(ptr);
1604 btf_info.btf_size = last_size;
1606 err = bpf_btf_get_info_by_fd(btf_fd, &btf_info, &len);
1609 if (err || btf_info.btf_size > last_size) {
1610 btf = err ? ERR_PTR(-errno) : ERR_PTR(-E2BIG);
1614 btf = btf_new(ptr, btf_info.btf_size, base_btf);
1621 struct btf *btf__load_from_kernel_by_id_split(__u32 id, struct btf *base_btf)
1626 btf_fd = bpf_btf_get_fd_by_id(id);
1628 return libbpf_err_ptr(-errno);
1630 btf = btf_get_from_fd(btf_fd, base_btf);
1633 return libbpf_ptr(btf);
1636 struct btf *btf__load_from_kernel_by_id(__u32 id)
1638 return btf__load_from_kernel_by_id_split(id, NULL);
1641 static void btf_invalidate_raw_data(struct btf *btf)
1643 if (btf->raw_data) {
1644 free(btf->raw_data);
1645 btf->raw_data = NULL;
1647 if (btf->raw_data_swapped) {
1648 free(btf->raw_data_swapped);
1649 btf->raw_data_swapped = NULL;
1653 /* Ensure BTF is ready to be modified (by splitting into a three memory
1654 * regions for header, types, and strings). Also invalidate cached
1657 static int btf_ensure_modifiable(struct btf *btf)
1660 struct strset *set = NULL;
1663 if (btf_is_modifiable(btf)) {
1664 /* any BTF modification invalidates raw_data */
1665 btf_invalidate_raw_data(btf);
1669 /* split raw data into three memory regions */
1670 hdr = malloc(btf->hdr->hdr_len);
1671 types = malloc(btf->hdr->type_len);
1675 memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1676 memcpy(types, btf->types_data, btf->hdr->type_len);
1678 /* build lookup index for all strings */
1679 set = strset__new(BTF_MAX_STR_OFFSET, btf->strs_data, btf->hdr->str_len);
1685 /* only when everything was successful, update internal state */
1687 btf->types_data = types;
1688 btf->types_data_cap = btf->hdr->type_len;
1689 btf->strs_data = NULL;
1690 btf->strs_set = set;
1691 /* if BTF was created from scratch, all strings are guaranteed to be
1692 * unique and deduplicated
1694 if (btf->hdr->str_len == 0)
1695 btf->strs_deduped = true;
1696 if (!btf->base_btf && btf->hdr->str_len == 1)
1697 btf->strs_deduped = true;
1699 /* invalidate raw_data representation */
1700 btf_invalidate_raw_data(btf);
1711 /* Find an offset in BTF string section that corresponds to a given string *s*.
1713 * - >0 offset into string section, if string is found;
1714 * - -ENOENT, if string is not in the string section;
1715 * - <0, on any other error.
1717 int btf__find_str(struct btf *btf, const char *s)
1721 if (btf->base_btf) {
1722 off = btf__find_str(btf->base_btf, s);
1727 /* BTF needs to be in a modifiable state to build string lookup index */
1728 if (btf_ensure_modifiable(btf))
1729 return libbpf_err(-ENOMEM);
1731 off = strset__find_str(btf->strs_set, s);
1733 return libbpf_err(off);
1735 return btf->start_str_off + off;
1738 /* Add a string s to the BTF string section.
1740 * - > 0 offset into string section, on success;
1743 int btf__add_str(struct btf *btf, const char *s)
1747 if (btf->base_btf) {
1748 off = btf__find_str(btf->base_btf, s);
1753 if (btf_ensure_modifiable(btf))
1754 return libbpf_err(-ENOMEM);
1756 off = strset__add_str(btf->strs_set, s);
1758 return libbpf_err(off);
1760 btf->hdr->str_len = strset__data_size(btf->strs_set);
1762 return btf->start_str_off + off;
1765 static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1767 return libbpf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1768 btf->hdr->type_len, UINT_MAX, add_sz);
1771 static void btf_type_inc_vlen(struct btf_type *t)
1773 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1776 static int btf_commit_type(struct btf *btf, int data_sz)
1780 err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1782 return libbpf_err(err);
1784 btf->hdr->type_len += data_sz;
1785 btf->hdr->str_off += data_sz;
1787 return btf->start_id + btf->nr_types - 1;
1791 const struct btf *src;
1793 struct hashmap *str_off_map; /* map string offsets from src to dst */
1796 static int btf_rewrite_str(struct btf_pipe *p, __u32 *str_off)
1801 if (!*str_off) /* nothing to do for empty strings */
1804 if (p->str_off_map &&
1805 hashmap__find(p->str_off_map, *str_off, &mapped_off)) {
1806 *str_off = mapped_off;
1810 off = btf__add_str(p->dst, btf__str_by_offset(p->src, *str_off));
1814 /* Remember string mapping from src to dst. It avoids
1815 * performing expensive string comparisons.
1817 if (p->str_off_map) {
1818 err = hashmap__append(p->str_off_map, *str_off, off);
1827 static int btf_add_type(struct btf_pipe *p, const struct btf_type *src_type)
1829 struct btf_field_iter it;
1834 sz = btf_type_size(src_type);
1836 return libbpf_err(sz);
1838 /* deconstruct BTF, if necessary, and invalidate raw_data */
1839 if (btf_ensure_modifiable(p->dst))
1840 return libbpf_err(-ENOMEM);
1842 t = btf_add_type_mem(p->dst, sz);
1844 return libbpf_err(-ENOMEM);
1846 memcpy(t, src_type, sz);
1848 err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_STRS);
1850 return libbpf_err(err);
1852 while ((str_off = btf_field_iter_next(&it))) {
1853 err = btf_rewrite_str(p, str_off);
1855 return libbpf_err(err);
1858 return btf_commit_type(p->dst, sz);
1861 int btf__add_type(struct btf *btf, const struct btf *src_btf, const struct btf_type *src_type)
1863 struct btf_pipe p = { .src = src_btf, .dst = btf };
1865 return btf_add_type(&p, src_type);
1868 static size_t btf_dedup_identity_hash_fn(long key, void *ctx);
1869 static bool btf_dedup_equal_fn(long k1, long k2, void *ctx);
1871 int btf__add_btf(struct btf *btf, const struct btf *src_btf)
1873 struct btf_pipe p = { .src = src_btf, .dst = btf };
1874 int data_sz, sz, cnt, i, err, old_strs_len;
1878 /* appending split BTF isn't supported yet */
1879 if (src_btf->base_btf)
1880 return libbpf_err(-ENOTSUP);
1882 /* deconstruct BTF, if necessary, and invalidate raw_data */
1883 if (btf_ensure_modifiable(btf))
1884 return libbpf_err(-ENOMEM);
1886 /* remember original strings section size if we have to roll back
1887 * partial strings section changes
1889 old_strs_len = btf->hdr->str_len;
1891 data_sz = src_btf->hdr->type_len;
1892 cnt = btf__type_cnt(src_btf) - 1;
1894 /* pre-allocate enough memory for new types */
1895 t = btf_add_type_mem(btf, data_sz);
1897 return libbpf_err(-ENOMEM);
1899 /* pre-allocate enough memory for type offset index for new types */
1900 off = btf_add_type_offs_mem(btf, cnt);
1902 return libbpf_err(-ENOMEM);
1904 /* Map the string offsets from src_btf to the offsets from btf to improve performance */
1905 p.str_off_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
1906 if (IS_ERR(p.str_off_map))
1907 return libbpf_err(-ENOMEM);
1909 /* bulk copy types data for all types from src_btf */
1910 memcpy(t, src_btf->types_data, data_sz);
1912 for (i = 0; i < cnt; i++) {
1913 struct btf_field_iter it;
1914 __u32 *type_id, *str_off;
1916 sz = btf_type_size(t);
1918 /* unlikely, has to be corrupted src_btf */
1923 /* fill out type ID to type offset mapping for lookups by type ID */
1924 *off = t - btf->types_data;
1926 /* add, dedup, and remap strings referenced by this BTF type */
1927 err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_STRS);
1930 while ((str_off = btf_field_iter_next(&it))) {
1931 err = btf_rewrite_str(&p, str_off);
1936 /* remap all type IDs referenced from this BTF type */
1937 err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_IDS);
1941 while ((type_id = btf_field_iter_next(&it))) {
1942 if (!*type_id) /* nothing to do for VOID references */
1945 /* we haven't updated btf's type count yet, so
1946 * btf->start_id + btf->nr_types - 1 is the type ID offset we should
1947 * add to all newly added BTF types
1949 *type_id += btf->start_id + btf->nr_types - 1;
1952 /* go to next type data and type offset index entry */
1957 /* Up until now any of the copied type data was effectively invisible,
1958 * so if we exited early before this point due to error, BTF would be
1959 * effectively unmodified. There would be extra internal memory
1960 * pre-allocated, but it would not be available for querying. But now
1961 * that we've copied and rewritten all the data successfully, we can
1962 * update type count and various internal offsets and sizes to
1963 * "commit" the changes and made them visible to the outside world.
1965 btf->hdr->type_len += data_sz;
1966 btf->hdr->str_off += data_sz;
1967 btf->nr_types += cnt;
1969 hashmap__free(p.str_off_map);
1971 /* return type ID of the first added BTF type */
1972 return btf->start_id + btf->nr_types - cnt;
1974 /* zero out preallocated memory as if it was just allocated with
1977 memset(btf->types_data + btf->hdr->type_len, 0, data_sz);
1978 memset(btf->strs_data + old_strs_len, 0, btf->hdr->str_len - old_strs_len);
1980 /* and now restore original strings section size; types data size
1981 * wasn't modified, so doesn't need restoring, see big comment above
1983 btf->hdr->str_len = old_strs_len;
1985 hashmap__free(p.str_off_map);
1987 return libbpf_err(err);
1991 * Append new BTF_KIND_INT type with:
1992 * - *name* - non-empty, non-NULL type name;
1993 * - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1994 * - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1996 * - >0, type ID of newly added BTF type;
1999 int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
2004 /* non-empty name */
2005 if (!name || !name[0])
2006 return libbpf_err(-EINVAL);
2007 /* byte_sz must be power of 2 */
2008 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
2009 return libbpf_err(-EINVAL);
2010 if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
2011 return libbpf_err(-EINVAL);
2013 /* deconstruct BTF, if necessary, and invalidate raw_data */
2014 if (btf_ensure_modifiable(btf))
2015 return libbpf_err(-ENOMEM);
2017 sz = sizeof(struct btf_type) + sizeof(int);
2018 t = btf_add_type_mem(btf, sz);
2020 return libbpf_err(-ENOMEM);
2022 /* if something goes wrong later, we might end up with an extra string,
2023 * but that shouldn't be a problem, because BTF can't be constructed
2024 * completely anyway and will most probably be just discarded
2026 name_off = btf__add_str(btf, name);
2030 t->name_off = name_off;
2031 t->info = btf_type_info(BTF_KIND_INT, 0, 0);
2033 /* set INT info, we don't allow setting legacy bit offset/size */
2034 *(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
2036 return btf_commit_type(btf, sz);
2040 * Append new BTF_KIND_FLOAT type with:
2041 * - *name* - non-empty, non-NULL type name;
2042 * - *sz* - size of the type, in bytes;
2044 * - >0, type ID of newly added BTF type;
2047 int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
2052 /* non-empty name */
2053 if (!name || !name[0])
2054 return libbpf_err(-EINVAL);
2056 /* byte_sz must be one of the explicitly allowed values */
2057 if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
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 name_off = btf__add_str(btf, name);
2073 t->name_off = name_off;
2074 t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
2077 return btf_commit_type(btf, sz);
2080 /* it's completely legal to append BTF types with type IDs pointing forward to
2081 * types that haven't been appended yet, so we only make sure that id looks
2082 * sane, we can't guarantee that ID will always be valid
2084 static int validate_type_id(int id)
2086 if (id < 0 || id > BTF_MAX_NR_TYPES)
2091 /* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
2092 static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
2095 int sz, name_off = 0;
2097 if (validate_type_id(ref_type_id))
2098 return libbpf_err(-EINVAL);
2100 if (btf_ensure_modifiable(btf))
2101 return libbpf_err(-ENOMEM);
2103 sz = sizeof(struct btf_type);
2104 t = btf_add_type_mem(btf, sz);
2106 return libbpf_err(-ENOMEM);
2108 if (name && name[0]) {
2109 name_off = btf__add_str(btf, name);
2114 t->name_off = name_off;
2115 t->info = btf_type_info(kind, 0, 0);
2116 t->type = ref_type_id;
2118 return btf_commit_type(btf, sz);
2122 * Append new BTF_KIND_PTR type with:
2123 * - *ref_type_id* - referenced type ID, it might not exist yet;
2125 * - >0, type ID of newly added BTF type;
2128 int btf__add_ptr(struct btf *btf, int ref_type_id)
2130 return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
2134 * Append new BTF_KIND_ARRAY type with:
2135 * - *index_type_id* - type ID of the type describing array index;
2136 * - *elem_type_id* - type ID of the type describing array element;
2137 * - *nr_elems* - the size of the array;
2139 * - >0, type ID of newly added BTF type;
2142 int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
2145 struct btf_array *a;
2148 if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
2149 return libbpf_err(-EINVAL);
2151 if (btf_ensure_modifiable(btf))
2152 return libbpf_err(-ENOMEM);
2154 sz = sizeof(struct btf_type) + sizeof(struct btf_array);
2155 t = btf_add_type_mem(btf, sz);
2157 return libbpf_err(-ENOMEM);
2160 t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
2164 a->type = elem_type_id;
2165 a->index_type = index_type_id;
2166 a->nelems = nr_elems;
2168 return btf_commit_type(btf, sz);
2171 /* generic STRUCT/UNION append function */
2172 static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
2175 int sz, name_off = 0;
2177 if (btf_ensure_modifiable(btf))
2178 return libbpf_err(-ENOMEM);
2180 sz = sizeof(struct btf_type);
2181 t = btf_add_type_mem(btf, sz);
2183 return libbpf_err(-ENOMEM);
2185 if (name && name[0]) {
2186 name_off = btf__add_str(btf, name);
2191 /* start out with vlen=0 and no kflag; this will be adjusted when
2192 * adding each member
2194 t->name_off = name_off;
2195 t->info = btf_type_info(kind, 0, 0);
2198 return btf_commit_type(btf, sz);
2202 * Append new BTF_KIND_STRUCT type with:
2203 * - *name* - name of the struct, can be NULL or empty for anonymous structs;
2204 * - *byte_sz* - size of the struct, in bytes;
2206 * Struct initially has no fields in it. Fields can be added by
2207 * btf__add_field() right after btf__add_struct() succeeds.
2210 * - >0, type ID of newly added BTF type;
2213 int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
2215 return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
2219 * Append new BTF_KIND_UNION type with:
2220 * - *name* - name of the union, can be NULL or empty for anonymous union;
2221 * - *byte_sz* - size of the union, in bytes;
2223 * Union initially has no fields in it. Fields can be added by
2224 * btf__add_field() right after btf__add_union() succeeds. All fields
2225 * should have *bit_offset* of 0.
2228 * - >0, type ID of newly added BTF type;
2231 int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
2233 return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
2236 static struct btf_type *btf_last_type(struct btf *btf)
2238 return btf_type_by_id(btf, btf__type_cnt(btf) - 1);
2242 * Append new field for the current STRUCT/UNION type with:
2243 * - *name* - name of the field, can be NULL or empty for anonymous field;
2244 * - *type_id* - type ID for the type describing field type;
2245 * - *bit_offset* - bit offset of the start of the field within struct/union;
2246 * - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
2251 int btf__add_field(struct btf *btf, const char *name, int type_id,
2252 __u32 bit_offset, __u32 bit_size)
2255 struct btf_member *m;
2257 int sz, name_off = 0;
2259 /* last type should be union/struct */
2260 if (btf->nr_types == 0)
2261 return libbpf_err(-EINVAL);
2262 t = btf_last_type(btf);
2263 if (!btf_is_composite(t))
2264 return libbpf_err(-EINVAL);
2266 if (validate_type_id(type_id))
2267 return libbpf_err(-EINVAL);
2268 /* best-effort bit field offset/size enforcement */
2269 is_bitfield = bit_size || (bit_offset % 8 != 0);
2270 if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
2271 return libbpf_err(-EINVAL);
2273 /* only offset 0 is allowed for unions */
2274 if (btf_is_union(t) && bit_offset)
2275 return libbpf_err(-EINVAL);
2277 /* decompose and invalidate raw data */
2278 if (btf_ensure_modifiable(btf))
2279 return libbpf_err(-ENOMEM);
2281 sz = sizeof(struct btf_member);
2282 m = btf_add_type_mem(btf, sz);
2284 return libbpf_err(-ENOMEM);
2286 if (name && name[0]) {
2287 name_off = btf__add_str(btf, name);
2292 m->name_off = name_off;
2294 m->offset = bit_offset | (bit_size << 24);
2296 /* btf_add_type_mem can invalidate t pointer */
2297 t = btf_last_type(btf);
2298 /* update parent type's vlen and kflag */
2299 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
2301 btf->hdr->type_len += sz;
2302 btf->hdr->str_off += sz;
2306 static int btf_add_enum_common(struct btf *btf, const char *name, __u32 byte_sz,
2307 bool is_signed, __u8 kind)
2310 int sz, name_off = 0;
2312 /* byte_sz must be power of 2 */
2313 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2314 return libbpf_err(-EINVAL);
2316 if (btf_ensure_modifiable(btf))
2317 return libbpf_err(-ENOMEM);
2319 sz = sizeof(struct btf_type);
2320 t = btf_add_type_mem(btf, sz);
2322 return libbpf_err(-ENOMEM);
2324 if (name && name[0]) {
2325 name_off = btf__add_str(btf, name);
2330 /* start out with vlen=0; it will be adjusted when adding enum values */
2331 t->name_off = name_off;
2332 t->info = btf_type_info(kind, 0, is_signed);
2335 return btf_commit_type(btf, sz);
2339 * Append new BTF_KIND_ENUM type with:
2340 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
2341 * - *byte_sz* - size of the enum, in bytes.
2343 * Enum initially has no enum values in it (and corresponds to enum forward
2344 * declaration). Enumerator values can be added by btf__add_enum_value()
2345 * immediately after btf__add_enum() succeeds.
2348 * - >0, type ID of newly added BTF type;
2351 int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2354 * set the signedness to be unsigned, it will change to signed
2355 * if any later enumerator is negative.
2357 return btf_add_enum_common(btf, name, byte_sz, false, BTF_KIND_ENUM);
2361 * Append new enum value for the current ENUM type with:
2362 * - *name* - name of the enumerator value, can't be NULL or empty;
2363 * - *value* - integer value corresponding to enum value *name*;
2368 int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2374 /* last type should be BTF_KIND_ENUM */
2375 if (btf->nr_types == 0)
2376 return libbpf_err(-EINVAL);
2377 t = btf_last_type(btf);
2378 if (!btf_is_enum(t))
2379 return libbpf_err(-EINVAL);
2381 /* non-empty name */
2382 if (!name || !name[0])
2383 return libbpf_err(-EINVAL);
2384 if (value < INT_MIN || value > UINT_MAX)
2385 return libbpf_err(-E2BIG);
2387 /* decompose and invalidate raw data */
2388 if (btf_ensure_modifiable(btf))
2389 return libbpf_err(-ENOMEM);
2391 sz = sizeof(struct btf_enum);
2392 v = btf_add_type_mem(btf, sz);
2394 return libbpf_err(-ENOMEM);
2396 name_off = btf__add_str(btf, name);
2400 v->name_off = name_off;
2403 /* update parent type's vlen */
2404 t = btf_last_type(btf);
2405 btf_type_inc_vlen(t);
2407 /* if negative value, set signedness to signed */
2409 t->info = btf_type_info(btf_kind(t), btf_vlen(t), true);
2411 btf->hdr->type_len += sz;
2412 btf->hdr->str_off += sz;
2417 * Append new BTF_KIND_ENUM64 type with:
2418 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
2419 * - *byte_sz* - size of the enum, in bytes.
2420 * - *is_signed* - whether the enum values are signed or not;
2422 * Enum initially has no enum values in it (and corresponds to enum forward
2423 * declaration). Enumerator values can be added by btf__add_enum64_value()
2424 * immediately after btf__add_enum64() succeeds.
2427 * - >0, type ID of newly added BTF type;
2430 int btf__add_enum64(struct btf *btf, const char *name, __u32 byte_sz,
2433 return btf_add_enum_common(btf, name, byte_sz, is_signed,
2438 * Append new enum value for the current ENUM64 type with:
2439 * - *name* - name of the enumerator value, can't be NULL or empty;
2440 * - *value* - integer value corresponding to enum value *name*;
2445 int btf__add_enum64_value(struct btf *btf, const char *name, __u64 value)
2447 struct btf_enum64 *v;
2451 /* last type should be BTF_KIND_ENUM64 */
2452 if (btf->nr_types == 0)
2453 return libbpf_err(-EINVAL);
2454 t = btf_last_type(btf);
2455 if (!btf_is_enum64(t))
2456 return libbpf_err(-EINVAL);
2458 /* non-empty name */
2459 if (!name || !name[0])
2460 return libbpf_err(-EINVAL);
2462 /* decompose and invalidate raw data */
2463 if (btf_ensure_modifiable(btf))
2464 return libbpf_err(-ENOMEM);
2466 sz = sizeof(struct btf_enum64);
2467 v = btf_add_type_mem(btf, sz);
2469 return libbpf_err(-ENOMEM);
2471 name_off = btf__add_str(btf, name);
2475 v->name_off = name_off;
2476 v->val_lo32 = (__u32)value;
2477 v->val_hi32 = value >> 32;
2479 /* update parent type's vlen */
2480 t = btf_last_type(btf);
2481 btf_type_inc_vlen(t);
2483 btf->hdr->type_len += sz;
2484 btf->hdr->str_off += sz;
2489 * Append new BTF_KIND_FWD type with:
2490 * - *name*, non-empty/non-NULL name;
2491 * - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2492 * BTF_FWD_UNION, or BTF_FWD_ENUM;
2494 * - >0, type ID of newly added BTF type;
2497 int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2499 if (!name || !name[0])
2500 return libbpf_err(-EINVAL);
2503 case BTF_FWD_STRUCT:
2504 case BTF_FWD_UNION: {
2508 id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
2511 t = btf_type_by_id(btf, id);
2512 t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2516 /* enum forward in BTF currently is just an enum with no enum
2517 * values; we also assume a standard 4-byte size for it
2519 return btf__add_enum(btf, name, sizeof(int));
2521 return libbpf_err(-EINVAL);
2526 * Append new BTF_KING_TYPEDEF type with:
2527 * - *name*, non-empty/non-NULL name;
2528 * - *ref_type_id* - referenced type ID, it might not exist yet;
2530 * - >0, type ID of newly added BTF type;
2533 int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2535 if (!name || !name[0])
2536 return libbpf_err(-EINVAL);
2538 return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2542 * Append new BTF_KIND_VOLATILE type with:
2543 * - *ref_type_id* - referenced type ID, it might not exist yet;
2545 * - >0, type ID of newly added BTF type;
2548 int btf__add_volatile(struct btf *btf, int ref_type_id)
2550 return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2554 * Append new BTF_KIND_CONST type with:
2555 * - *ref_type_id* - referenced type ID, it might not exist yet;
2557 * - >0, type ID of newly added BTF type;
2560 int btf__add_const(struct btf *btf, int ref_type_id)
2562 return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2566 * Append new BTF_KIND_RESTRICT type with:
2567 * - *ref_type_id* - referenced type ID, it might not exist yet;
2569 * - >0, type ID of newly added BTF type;
2572 int btf__add_restrict(struct btf *btf, int ref_type_id)
2574 return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2578 * Append new BTF_KIND_TYPE_TAG type with:
2579 * - *value*, non-empty/non-NULL tag value;
2580 * - *ref_type_id* - referenced type ID, it might not exist yet;
2582 * - >0, type ID of newly added BTF type;
2585 int btf__add_type_tag(struct btf *btf, const char *value, int ref_type_id)
2587 if (!value || !value[0])
2588 return libbpf_err(-EINVAL);
2590 return btf_add_ref_kind(btf, BTF_KIND_TYPE_TAG, value, ref_type_id);
2594 * Append new BTF_KIND_FUNC type with:
2595 * - *name*, non-empty/non-NULL name;
2596 * - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2598 * - >0, type ID of newly added BTF type;
2601 int btf__add_func(struct btf *btf, const char *name,
2602 enum btf_func_linkage linkage, int proto_type_id)
2606 if (!name || !name[0])
2607 return libbpf_err(-EINVAL);
2608 if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2609 linkage != BTF_FUNC_EXTERN)
2610 return libbpf_err(-EINVAL);
2612 id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2614 struct btf_type *t = btf_type_by_id(btf, id);
2616 t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2618 return libbpf_err(id);
2622 * Append new BTF_KIND_FUNC_PROTO with:
2623 * - *ret_type_id* - type ID for return result of a function.
2625 * Function prototype initially has no arguments, but they can be added by
2626 * btf__add_func_param() one by one, immediately after
2627 * btf__add_func_proto() succeeded.
2630 * - >0, type ID of newly added BTF type;
2633 int btf__add_func_proto(struct btf *btf, int ret_type_id)
2638 if (validate_type_id(ret_type_id))
2639 return libbpf_err(-EINVAL);
2641 if (btf_ensure_modifiable(btf))
2642 return libbpf_err(-ENOMEM);
2644 sz = sizeof(struct btf_type);
2645 t = btf_add_type_mem(btf, sz);
2647 return libbpf_err(-ENOMEM);
2649 /* start out with vlen=0; this will be adjusted when adding enum
2650 * values, if necessary
2653 t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2654 t->type = ret_type_id;
2656 return btf_commit_type(btf, sz);
2660 * Append new function parameter for current FUNC_PROTO type with:
2661 * - *name* - parameter name, can be NULL or empty;
2662 * - *type_id* - type ID describing the type of the parameter.
2667 int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2670 struct btf_param *p;
2671 int sz, name_off = 0;
2673 if (validate_type_id(type_id))
2674 return libbpf_err(-EINVAL);
2676 /* last type should be BTF_KIND_FUNC_PROTO */
2677 if (btf->nr_types == 0)
2678 return libbpf_err(-EINVAL);
2679 t = btf_last_type(btf);
2680 if (!btf_is_func_proto(t))
2681 return libbpf_err(-EINVAL);
2683 /* decompose and invalidate raw data */
2684 if (btf_ensure_modifiable(btf))
2685 return libbpf_err(-ENOMEM);
2687 sz = sizeof(struct btf_param);
2688 p = btf_add_type_mem(btf, sz);
2690 return libbpf_err(-ENOMEM);
2692 if (name && name[0]) {
2693 name_off = btf__add_str(btf, name);
2698 p->name_off = name_off;
2701 /* update parent type's vlen */
2702 t = btf_last_type(btf);
2703 btf_type_inc_vlen(t);
2705 btf->hdr->type_len += sz;
2706 btf->hdr->str_off += sz;
2711 * Append new BTF_KIND_VAR type with:
2712 * - *name* - non-empty/non-NULL name;
2713 * - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2714 * BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2715 * - *type_id* - type ID of the type describing the type of the variable.
2717 * - >0, type ID of newly added BTF type;
2720 int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2726 /* non-empty name */
2727 if (!name || !name[0])
2728 return libbpf_err(-EINVAL);
2729 if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2730 linkage != BTF_VAR_GLOBAL_EXTERN)
2731 return libbpf_err(-EINVAL);
2732 if (validate_type_id(type_id))
2733 return libbpf_err(-EINVAL);
2735 /* deconstruct BTF, if necessary, and invalidate raw_data */
2736 if (btf_ensure_modifiable(btf))
2737 return libbpf_err(-ENOMEM);
2739 sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2740 t = btf_add_type_mem(btf, sz);
2742 return libbpf_err(-ENOMEM);
2744 name_off = btf__add_str(btf, name);
2748 t->name_off = name_off;
2749 t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2753 v->linkage = linkage;
2755 return btf_commit_type(btf, sz);
2759 * Append new BTF_KIND_DATASEC type with:
2760 * - *name* - non-empty/non-NULL name;
2761 * - *byte_sz* - data section size, in bytes.
2763 * Data section is initially empty. Variables info can be added with
2764 * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2767 * - >0, type ID of newly added BTF type;
2770 int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2775 /* non-empty name */
2776 if (!name || !name[0])
2777 return libbpf_err(-EINVAL);
2779 if (btf_ensure_modifiable(btf))
2780 return libbpf_err(-ENOMEM);
2782 sz = sizeof(struct btf_type);
2783 t = btf_add_type_mem(btf, sz);
2785 return libbpf_err(-ENOMEM);
2787 name_off = btf__add_str(btf, name);
2791 /* start with vlen=0, which will be update as var_secinfos are added */
2792 t->name_off = name_off;
2793 t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2796 return btf_commit_type(btf, sz);
2800 * Append new data section variable information entry for current DATASEC type:
2801 * - *var_type_id* - type ID, describing type of the variable;
2802 * - *offset* - variable offset within data section, in bytes;
2803 * - *byte_sz* - variable size, in bytes.
2809 int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2812 struct btf_var_secinfo *v;
2815 /* last type should be BTF_KIND_DATASEC */
2816 if (btf->nr_types == 0)
2817 return libbpf_err(-EINVAL);
2818 t = btf_last_type(btf);
2819 if (!btf_is_datasec(t))
2820 return libbpf_err(-EINVAL);
2822 if (validate_type_id(var_type_id))
2823 return libbpf_err(-EINVAL);
2825 /* decompose and invalidate raw data */
2826 if (btf_ensure_modifiable(btf))
2827 return libbpf_err(-ENOMEM);
2829 sz = sizeof(struct btf_var_secinfo);
2830 v = btf_add_type_mem(btf, sz);
2832 return libbpf_err(-ENOMEM);
2834 v->type = var_type_id;
2838 /* update parent type's vlen */
2839 t = btf_last_type(btf);
2840 btf_type_inc_vlen(t);
2842 btf->hdr->type_len += sz;
2843 btf->hdr->str_off += sz;
2848 * Append new BTF_KIND_DECL_TAG type with:
2849 * - *value* - non-empty/non-NULL string;
2850 * - *ref_type_id* - referenced type ID, it might not exist yet;
2851 * - *component_idx* - -1 for tagging reference type, otherwise struct/union
2852 * member or function argument index;
2854 * - >0, type ID of newly added BTF type;
2857 int btf__add_decl_tag(struct btf *btf, const char *value, int ref_type_id,
2863 if (!value || !value[0] || component_idx < -1)
2864 return libbpf_err(-EINVAL);
2866 if (validate_type_id(ref_type_id))
2867 return libbpf_err(-EINVAL);
2869 if (btf_ensure_modifiable(btf))
2870 return libbpf_err(-ENOMEM);
2872 sz = sizeof(struct btf_type) + sizeof(struct btf_decl_tag);
2873 t = btf_add_type_mem(btf, sz);
2875 return libbpf_err(-ENOMEM);
2877 value_off = btf__add_str(btf, value);
2881 t->name_off = value_off;
2882 t->info = btf_type_info(BTF_KIND_DECL_TAG, 0, false);
2883 t->type = ref_type_id;
2884 btf_decl_tag(t)->component_idx = component_idx;
2886 return btf_commit_type(btf, sz);
2889 struct btf_ext_sec_info_param {
2893 struct btf_ext_info *ext_info;
2898 * Parse a single info subsection of the BTF.ext info data:
2899 * - validate subsection structure and elements
2900 * - save info subsection start and sizing details in struct btf_ext
2901 * - endian-independent operation, for calling before byte-swapping
2903 static int btf_ext_parse_sec_info(struct btf_ext *btf_ext,
2904 struct btf_ext_sec_info_param *ext_sec,
2907 const struct btf_ext_info_sec *sinfo;
2908 struct btf_ext_info *ext_info;
2909 __u32 info_left, record_size;
2913 if (ext_sec->len == 0)
2916 if (ext_sec->off & 0x03) {
2917 pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2922 /* The start of the info sec (including the __u32 record_size). */
2923 info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2924 info_left = ext_sec->len;
2926 if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2927 pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2928 ext_sec->desc, ext_sec->off, ext_sec->len);
2932 /* At least a record size */
2933 if (info_left < sizeof(__u32)) {
2934 pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2938 /* The record size needs to meet either the minimum standard or, when
2939 * handling non-native endianness data, the exact standard so as
2940 * to allow safe byte-swapping.
2942 record_size = is_native ? *(__u32 *)info : bswap_32(*(__u32 *)info);
2943 if (record_size < ext_sec->min_rec_size ||
2944 (!is_native && record_size != ext_sec->min_rec_size) ||
2945 record_size & 0x03) {
2946 pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2947 ext_sec->desc, record_size);
2951 sinfo = info + sizeof(__u32);
2952 info_left -= sizeof(__u32);
2954 /* If no records, return failure now so .BTF.ext won't be used. */
2956 pr_debug("%s section in .BTF.ext has no records\n", ext_sec->desc);
2961 unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2962 __u64 total_record_size;
2965 if (info_left < sec_hdrlen) {
2966 pr_debug("%s section header is not found in .BTF.ext\n",
2971 num_records = is_native ? sinfo->num_info : bswap_32(sinfo->num_info);
2972 if (num_records == 0) {
2973 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2978 total_record_size = sec_hdrlen + (__u64)num_records * record_size;
2979 if (info_left < total_record_size) {
2980 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2985 info_left -= total_record_size;
2986 sinfo = (void *)sinfo + total_record_size;
2990 ext_info = ext_sec->ext_info;
2991 ext_info->len = ext_sec->len - sizeof(__u32);
2992 ext_info->rec_size = record_size;
2993 ext_info->info = info + sizeof(__u32);
2994 ext_info->sec_cnt = sec_cnt;
2999 /* Parse all info secs in the BTF.ext info data */
3000 static int btf_ext_parse_info(struct btf_ext *btf_ext, bool is_native)
3002 struct btf_ext_sec_info_param func_info = {
3003 .off = btf_ext->hdr->func_info_off,
3004 .len = btf_ext->hdr->func_info_len,
3005 .min_rec_size = sizeof(struct bpf_func_info_min),
3006 .ext_info = &btf_ext->func_info,
3009 struct btf_ext_sec_info_param line_info = {
3010 .off = btf_ext->hdr->line_info_off,
3011 .len = btf_ext->hdr->line_info_len,
3012 .min_rec_size = sizeof(struct bpf_line_info_min),
3013 .ext_info = &btf_ext->line_info,
3014 .desc = "line_info",
3016 struct btf_ext_sec_info_param core_relo = {
3017 .off = btf_ext->hdr->core_relo_off,
3018 .len = btf_ext->hdr->core_relo_len,
3019 .min_rec_size = sizeof(struct bpf_core_relo),
3020 .ext_info = &btf_ext->core_relo_info,
3021 .desc = "core_relo",
3025 err = btf_ext_parse_sec_info(btf_ext, &func_info, is_native);
3029 err = btf_ext_parse_sec_info(btf_ext, &line_info, is_native);
3033 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
3034 return 0; /* skip core relos parsing */
3036 err = btf_ext_parse_sec_info(btf_ext, &core_relo, is_native);
3043 /* Swap byte-order of BTF.ext header with any endianness */
3044 static void btf_ext_bswap_hdr(struct btf_ext_header *h)
3046 bool is_native = h->magic == BTF_MAGIC;
3049 hdr_len = is_native ? h->hdr_len : bswap_32(h->hdr_len);
3051 h->magic = bswap_16(h->magic);
3052 h->hdr_len = bswap_32(h->hdr_len);
3053 h->func_info_off = bswap_32(h->func_info_off);
3054 h->func_info_len = bswap_32(h->func_info_len);
3055 h->line_info_off = bswap_32(h->line_info_off);
3056 h->line_info_len = bswap_32(h->line_info_len);
3058 if (hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
3061 h->core_relo_off = bswap_32(h->core_relo_off);
3062 h->core_relo_len = bswap_32(h->core_relo_len);
3065 /* Swap byte-order of generic info subsection */
3066 static void btf_ext_bswap_info_sec(void *info, __u32 len, bool is_native,
3067 info_rec_bswap_fn bswap_fn)
3069 struct btf_ext_info_sec *sec;
3070 __u32 info_left, rec_size, *rs;
3075 rs = info; /* info record size */
3076 rec_size = is_native ? *rs : bswap_32(*rs);
3077 *rs = bswap_32(*rs);
3079 sec = info + sizeof(__u32); /* info sec #1 */
3080 info_left = len - sizeof(__u32);
3082 unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
3086 num_recs = is_native ? sec->num_info : bswap_32(sec->num_info);
3087 sec->sec_name_off = bswap_32(sec->sec_name_off);
3088 sec->num_info = bswap_32(sec->num_info);
3089 p = sec->data; /* info rec #1 */
3090 for (i = 0; i < num_recs; i++, p += rec_size)
3093 info_left -= sec_hdrlen + (__u64)rec_size * num_recs;
3098 * Swap byte-order of all info data in a BTF.ext section
3099 * - requires BTF.ext hdr in native endianness
3101 static void btf_ext_bswap_info(struct btf_ext *btf_ext, void *data)
3103 const bool is_native = btf_ext->swapped_endian;
3104 const struct btf_ext_header *h = data;
3107 /* Swap func_info subsection byte-order */
3108 info = data + h->hdr_len + h->func_info_off;
3109 btf_ext_bswap_info_sec(info, h->func_info_len, is_native,
3110 (info_rec_bswap_fn)bpf_func_info_bswap);
3112 /* Swap line_info subsection byte-order */
3113 info = data + h->hdr_len + h->line_info_off;
3114 btf_ext_bswap_info_sec(info, h->line_info_len, is_native,
3115 (info_rec_bswap_fn)bpf_line_info_bswap);
3117 /* Swap core_relo subsection byte-order (if present) */
3118 if (h->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
3121 info = data + h->hdr_len + h->core_relo_off;
3122 btf_ext_bswap_info_sec(info, h->core_relo_len, is_native,
3123 (info_rec_bswap_fn)bpf_core_relo_bswap);
3126 /* Parse hdr data and info sections: check and convert to native endianness */
3127 static int btf_ext_parse(struct btf_ext *btf_ext)
3129 __u32 hdr_len, data_size = btf_ext->data_size;
3130 struct btf_ext_header *hdr = btf_ext->hdr;
3131 bool swapped_endian = false;
3134 if (data_size < offsetofend(struct btf_ext_header, hdr_len)) {
3135 pr_debug("BTF.ext header too short\n");
3139 hdr_len = hdr->hdr_len;
3140 if (hdr->magic == bswap_16(BTF_MAGIC)) {
3141 swapped_endian = true;
3142 hdr_len = bswap_32(hdr_len);
3143 } else if (hdr->magic != BTF_MAGIC) {
3144 pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
3148 /* Ensure known version of structs, current BTF_VERSION == 1 */
3149 if (hdr->version != 1) {
3150 pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
3155 pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
3159 if (data_size < hdr_len) {
3160 pr_debug("BTF.ext header not found\n");
3162 } else if (data_size == hdr_len) {
3163 pr_debug("BTF.ext has no data\n");
3167 /* Verify mandatory hdr info details present */
3168 if (hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
3169 pr_warn("BTF.ext header missing func_info, line_info\n");
3173 /* Keep hdr native byte-order in memory for introspection */
3175 btf_ext_bswap_hdr(btf_ext->hdr);
3177 /* Validate info subsections and cache key metadata */
3178 err = btf_ext_parse_info(btf_ext, !swapped_endian);
3182 /* Keep infos native byte-order in memory for introspection */
3184 btf_ext_bswap_info(btf_ext, btf_ext->data);
3187 * Set btf_ext->swapped_endian only after all header and info data has
3188 * been swapped, helping bswap functions determine if their data are
3189 * in native byte-order when called.
3191 btf_ext->swapped_endian = swapped_endian;
3195 void btf_ext__free(struct btf_ext *btf_ext)
3197 if (IS_ERR_OR_NULL(btf_ext))
3199 free(btf_ext->func_info.sec_idxs);
3200 free(btf_ext->line_info.sec_idxs);
3201 free(btf_ext->core_relo_info.sec_idxs);
3202 free(btf_ext->data);
3203 free(btf_ext->data_swapped);
3207 struct btf_ext *btf_ext__new(const __u8 *data, __u32 size)
3209 struct btf_ext *btf_ext;
3212 btf_ext = calloc(1, sizeof(struct btf_ext));
3214 return libbpf_err_ptr(-ENOMEM);
3216 btf_ext->data_size = size;
3217 btf_ext->data = malloc(size);
3218 if (!btf_ext->data) {
3222 memcpy(btf_ext->data, data, size);
3224 err = btf_ext_parse(btf_ext);
3228 btf_ext__free(btf_ext);
3229 return libbpf_err_ptr(err);
3235 static void *btf_ext_raw_data(const struct btf_ext *btf_ext_ro, bool swap_endian)
3237 struct btf_ext *btf_ext = (struct btf_ext *)btf_ext_ro;
3238 const __u32 data_sz = btf_ext->data_size;
3241 /* Return native data (always present) or swapped data if present */
3243 return btf_ext->data;
3244 else if (btf_ext->data_swapped)
3245 return btf_ext->data_swapped;
3247 /* Recreate missing swapped data, then cache and return */
3248 data = calloc(1, data_sz);
3251 memcpy(data, btf_ext->data, data_sz);
3253 btf_ext_bswap_info(btf_ext, data);
3254 btf_ext_bswap_hdr(data);
3255 btf_ext->data_swapped = data;
3259 const void *btf_ext__raw_data(const struct btf_ext *btf_ext, __u32 *size)
3263 data = btf_ext_raw_data(btf_ext, btf_ext->swapped_endian);
3265 return errno = ENOMEM, NULL;
3267 *size = btf_ext->data_size;
3271 __attribute__((alias("btf_ext__raw_data")))
3272 const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size);
3274 enum btf_endianness btf_ext__endianness(const struct btf_ext *btf_ext)
3276 if (is_host_big_endian())
3277 return btf_ext->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
3279 return btf_ext->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
3282 int btf_ext__set_endianness(struct btf_ext *btf_ext, enum btf_endianness endian)
3284 if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
3285 return libbpf_err(-EINVAL);
3287 btf_ext->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
3289 if (!btf_ext->swapped_endian) {
3290 free(btf_ext->data_swapped);
3291 btf_ext->data_swapped = NULL;
3298 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts);
3299 static void btf_dedup_free(struct btf_dedup *d);
3300 static int btf_dedup_prep(struct btf_dedup *d);
3301 static int btf_dedup_strings(struct btf_dedup *d);
3302 static int btf_dedup_prim_types(struct btf_dedup *d);
3303 static int btf_dedup_struct_types(struct btf_dedup *d);
3304 static int btf_dedup_ref_types(struct btf_dedup *d);
3305 static int btf_dedup_resolve_fwds(struct btf_dedup *d);
3306 static int btf_dedup_compact_types(struct btf_dedup *d);
3307 static int btf_dedup_remap_types(struct btf_dedup *d);
3310 * Deduplicate BTF types and strings.
3312 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
3313 * section with all BTF type descriptors and string data. It overwrites that
3314 * memory in-place with deduplicated types and strings without any loss of
3315 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
3316 * is provided, all the strings referenced from .BTF.ext section are honored
3317 * and updated to point to the right offsets after deduplication.
3319 * If function returns with error, type/string data might be garbled and should
3322 * More verbose and detailed description of both problem btf_dedup is solving,
3323 * as well as solution could be found at:
3324 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
3326 * Problem description and justification
3327 * =====================================
3329 * BTF type information is typically emitted either as a result of conversion
3330 * from DWARF to BTF or directly by compiler. In both cases, each compilation
3331 * unit contains information about a subset of all the types that are used
3332 * in an application. These subsets are frequently overlapping and contain a lot
3333 * of duplicated information when later concatenated together into a single
3334 * binary. This algorithm ensures that each unique type is represented by single
3335 * BTF type descriptor, greatly reducing resulting size of BTF data.
3337 * Compilation unit isolation and subsequent duplication of data is not the only
3338 * problem. The same type hierarchy (e.g., struct and all the type that struct
3339 * references) in different compilation units can be represented in BTF to
3340 * various degrees of completeness (or, rather, incompleteness) due to
3341 * struct/union forward declarations.
3343 * Let's take a look at an example, that we'll use to better understand the
3344 * problem (and solution). Suppose we have two compilation units, each using
3345 * same `struct S`, but each of them having incomplete type information about
3374 * In case of CU #1, BTF data will know only that `struct B` exist (but no
3375 * more), but will know the complete type information about `struct A`. While
3376 * for CU #2, it will know full type information about `struct B`, but will
3377 * only know about forward declaration of `struct A` (in BTF terms, it will
3378 * have `BTF_KIND_FWD` type descriptor with name `B`).
3380 * This compilation unit isolation means that it's possible that there is no
3381 * single CU with complete type information describing structs `S`, `A`, and
3382 * `B`. Also, we might get tons of duplicated and redundant type information.
3384 * Additional complication we need to keep in mind comes from the fact that
3385 * types, in general, can form graphs containing cycles, not just DAGs.
3387 * While algorithm does deduplication, it also merges and resolves type
3388 * information (unless disabled throught `struct btf_opts`), whenever possible.
3389 * E.g., in the example above with two compilation units having partial type
3390 * information for structs `A` and `B`, the output of algorithm will emit
3391 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
3392 * (as well as type information for `int` and pointers), as if they were defined
3393 * in a single compilation unit as:
3413 * Algorithm completes its work in 7 separate passes:
3415 * 1. Strings deduplication.
3416 * 2. Primitive types deduplication (int, enum, fwd).
3417 * 3. Struct/union types deduplication.
3418 * 4. Resolve unambiguous forward declarations.
3419 * 5. Reference types deduplication (pointers, typedefs, arrays, funcs, func
3420 * protos, and const/volatile/restrict modifiers).
3421 * 6. Types compaction.
3422 * 7. Types remapping.
3424 * Algorithm determines canonical type descriptor, which is a single
3425 * representative type for each truly unique type. This canonical type is the
3426 * one that will go into final deduplicated BTF type information. For
3427 * struct/unions, it is also the type that algorithm will merge additional type
3428 * information into (while resolving FWDs), as it discovers it from data in
3429 * other CUs. Each input BTF type eventually gets either mapped to itself, if
3430 * that type is canonical, or to some other type, if that type is equivalent
3431 * and was chosen as canonical representative. This mapping is stored in
3432 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
3433 * FWD type got resolved to.
3435 * To facilitate fast discovery of canonical types, we also maintain canonical
3436 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
3437 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
3438 * that match that signature. With sufficiently good choice of type signature
3439 * hashing function, we can limit number of canonical types for each unique type
3440 * signature to a very small number, allowing to find canonical type for any
3441 * duplicated type very quickly.
3443 * Struct/union deduplication is the most critical part and algorithm for
3444 * deduplicating structs/unions is described in greater details in comments for
3445 * `btf_dedup_is_equiv` function.
3447 int btf__dedup(struct btf *btf, const struct btf_dedup_opts *opts)
3449 struct btf_dedup *d;
3452 if (!OPTS_VALID(opts, btf_dedup_opts))
3453 return libbpf_err(-EINVAL);
3455 d = btf_dedup_new(btf, opts);
3457 pr_debug("btf_dedup_new failed: %ld\n", PTR_ERR(d));
3458 return libbpf_err(-EINVAL);
3461 if (btf_ensure_modifiable(btf)) {
3466 err = btf_dedup_prep(d);
3468 pr_debug("btf_dedup_prep failed: %s\n", errstr(err));
3471 err = btf_dedup_strings(d);
3473 pr_debug("btf_dedup_strings failed: %s\n", errstr(err));
3476 err = btf_dedup_prim_types(d);
3478 pr_debug("btf_dedup_prim_types failed: %s\n", errstr(err));
3481 err = btf_dedup_struct_types(d);
3483 pr_debug("btf_dedup_struct_types failed: %s\n", errstr(err));
3486 err = btf_dedup_resolve_fwds(d);
3488 pr_debug("btf_dedup_resolve_fwds failed: %s\n", errstr(err));
3491 err = btf_dedup_ref_types(d);
3493 pr_debug("btf_dedup_ref_types failed: %s\n", errstr(err));
3496 err = btf_dedup_compact_types(d);
3498 pr_debug("btf_dedup_compact_types failed: %s\n", errstr(err));
3501 err = btf_dedup_remap_types(d);
3503 pr_debug("btf_dedup_remap_types failed: %s\n", errstr(err));
3509 return libbpf_err(err);
3512 #define BTF_UNPROCESSED_ID ((__u32)-1)
3513 #define BTF_IN_PROGRESS_ID ((__u32)-2)
3516 /* .BTF section to be deduped in-place */
3519 * Optional .BTF.ext section. When provided, any strings referenced
3520 * from it will be taken into account when deduping strings
3522 struct btf_ext *btf_ext;
3524 * This is a map from any type's signature hash to a list of possible
3525 * canonical representative type candidates. Hash collisions are
3526 * ignored, so even types of various kinds can share same list of
3527 * candidates, which is fine because we rely on subsequent
3528 * btf_xxx_equal() checks to authoritatively verify type equality.
3530 struct hashmap *dedup_table;
3531 /* Canonical types map */
3533 /* Hypothetical mapping, used during type graph equivalence checks */
3538 /* Whether hypothetical mapping, if successful, would need to adjust
3539 * already canonicalized types (due to a new forward declaration to
3540 * concrete type resolution). In such case, during split BTF dedup
3541 * candidate type would still be considered as different, because base
3542 * BTF is considered to be immutable.
3544 bool hypot_adjust_canon;
3545 /* Various option modifying behavior of algorithm */
3546 struct btf_dedup_opts opts;
3547 /* temporary strings deduplication state */
3548 struct strset *strs_set;
3551 static unsigned long hash_combine(unsigned long h, unsigned long value)
3553 return h * 31 + value;
3556 #define for_each_dedup_cand(d, node, hash) \
3557 hashmap__for_each_key_entry(d->dedup_table, node, hash)
3559 static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
3561 return hashmap__append(d->dedup_table, hash, type_id);
3564 static int btf_dedup_hypot_map_add(struct btf_dedup *d,
3565 __u32 from_id, __u32 to_id)
3567 if (d->hypot_cnt == d->hypot_cap) {
3570 d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3571 new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3574 d->hypot_list = new_list;
3576 d->hypot_list[d->hypot_cnt++] = from_id;
3577 d->hypot_map[from_id] = to_id;
3581 static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3585 for (i = 0; i < d->hypot_cnt; i++)
3586 d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3588 d->hypot_adjust_canon = false;
3591 static void btf_dedup_free(struct btf_dedup *d)
3593 hashmap__free(d->dedup_table);
3594 d->dedup_table = NULL;
3600 d->hypot_map = NULL;
3602 free(d->hypot_list);
3603 d->hypot_list = NULL;
3608 static size_t btf_dedup_identity_hash_fn(long key, void *ctx)
3613 static size_t btf_dedup_collision_hash_fn(long key, void *ctx)
3618 static bool btf_dedup_equal_fn(long k1, long k2, void *ctx)
3623 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts)
3625 struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3626 hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3627 int i, err = 0, type_cnt;
3630 return ERR_PTR(-ENOMEM);
3632 if (OPTS_GET(opts, force_collisions, false))
3633 hash_fn = btf_dedup_collision_hash_fn;
3636 d->btf_ext = OPTS_GET(opts, btf_ext, NULL);
3638 d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3639 if (IS_ERR(d->dedup_table)) {
3640 err = PTR_ERR(d->dedup_table);
3641 d->dedup_table = NULL;
3645 type_cnt = btf__type_cnt(btf);
3646 d->map = malloc(sizeof(__u32) * type_cnt);
3651 /* special BTF "void" type is made canonical immediately */
3653 for (i = 1; i < type_cnt; i++) {
3654 struct btf_type *t = btf_type_by_id(d->btf, i);
3656 /* VAR and DATASEC are never deduped and are self-canonical */
3657 if (btf_is_var(t) || btf_is_datasec(t))
3660 d->map[i] = BTF_UNPROCESSED_ID;
3663 d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3664 if (!d->hypot_map) {
3668 for (i = 0; i < type_cnt; i++)
3669 d->hypot_map[i] = BTF_UNPROCESSED_ID;
3674 return ERR_PTR(err);
3681 * Iterate over all possible places in .BTF and .BTF.ext that can reference
3682 * string and pass pointer to it to a provided callback `fn`.
3684 static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3688 for (i = 0; i < d->btf->nr_types; i++) {
3689 struct btf_field_iter it;
3690 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3693 r = btf_field_iter_init(&it, t, BTF_FIELD_ITER_STRS);
3697 while ((str_off = btf_field_iter_next(&it))) {
3698 r = fn(str_off, ctx);
3707 r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3714 static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3716 struct btf_dedup *d = ctx;
3717 __u32 str_off = *str_off_ptr;
3721 /* don't touch empty string or string in main BTF */
3722 if (str_off == 0 || str_off < d->btf->start_str_off)
3725 s = btf__str_by_offset(d->btf, str_off);
3726 if (d->btf->base_btf) {
3727 err = btf__find_str(d->btf->base_btf, s);
3736 off = strset__add_str(d->strs_set, s);
3740 *str_off_ptr = d->btf->start_str_off + off;
3745 * Dedup string and filter out those that are not referenced from either .BTF
3746 * or .BTF.ext (if provided) sections.
3748 * This is done by building index of all strings in BTF's string section,
3749 * then iterating over all entities that can reference strings (e.g., type
3750 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3751 * strings as used. After that all used strings are deduped and compacted into
3752 * sequential blob of memory and new offsets are calculated. Then all the string
3753 * references are iterated again and rewritten using new offsets.
3755 static int btf_dedup_strings(struct btf_dedup *d)
3759 if (d->btf->strs_deduped)
3762 d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3763 if (IS_ERR(d->strs_set)) {
3764 err = PTR_ERR(d->strs_set);
3768 if (!d->btf->base_btf) {
3769 /* insert empty string; we won't be looking it up during strings
3770 * dedup, but it's good to have it for generic BTF string lookups
3772 err = strset__add_str(d->strs_set, "");
3777 /* remap string offsets */
3778 err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3782 /* replace BTF string data and hash with deduped ones */
3783 strset__free(d->btf->strs_set);
3784 d->btf->hdr->str_len = strset__data_size(d->strs_set);
3785 d->btf->strs_set = d->strs_set;
3787 d->btf->strs_deduped = true;
3791 strset__free(d->strs_set);
3797 static long btf_hash_common(struct btf_type *t)
3801 h = hash_combine(0, t->name_off);
3802 h = hash_combine(h, t->info);
3803 h = hash_combine(h, t->size);
3807 static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3809 return t1->name_off == t2->name_off &&
3810 t1->info == t2->info &&
3811 t1->size == t2->size;
3814 /* Calculate type signature hash of INT or TAG. */
3815 static long btf_hash_int_decl_tag(struct btf_type *t)
3817 __u32 info = *(__u32 *)(t + 1);
3820 h = btf_hash_common(t);
3821 h = hash_combine(h, info);
3825 /* Check structural equality of two INTs or TAGs. */
3826 static bool btf_equal_int_tag(struct btf_type *t1, struct btf_type *t2)
3830 if (!btf_equal_common(t1, t2))
3832 info1 = *(__u32 *)(t1 + 1);
3833 info2 = *(__u32 *)(t2 + 1);
3834 return info1 == info2;
3837 /* Calculate type signature hash of ENUM/ENUM64. */
3838 static long btf_hash_enum(struct btf_type *t)
3842 /* don't hash vlen, enum members and size to support enum fwd resolving */
3843 h = hash_combine(0, t->name_off);
3847 static bool btf_equal_enum_members(struct btf_type *t1, struct btf_type *t2)
3849 const struct btf_enum *m1, *m2;
3853 vlen = btf_vlen(t1);
3856 for (i = 0; i < vlen; i++) {
3857 if (m1->name_off != m2->name_off || m1->val != m2->val)
3865 static bool btf_equal_enum64_members(struct btf_type *t1, struct btf_type *t2)
3867 const struct btf_enum64 *m1, *m2;
3871 vlen = btf_vlen(t1);
3872 m1 = btf_enum64(t1);
3873 m2 = btf_enum64(t2);
3874 for (i = 0; i < vlen; i++) {
3875 if (m1->name_off != m2->name_off || m1->val_lo32 != m2->val_lo32 ||
3876 m1->val_hi32 != m2->val_hi32)
3884 /* Check structural equality of two ENUMs or ENUM64s. */
3885 static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3887 if (!btf_equal_common(t1, t2))
3890 /* t1 & t2 kinds are identical because of btf_equal_common */
3891 if (btf_kind(t1) == BTF_KIND_ENUM)
3892 return btf_equal_enum_members(t1, t2);
3894 return btf_equal_enum64_members(t1, t2);
3897 static inline bool btf_is_enum_fwd(struct btf_type *t)
3899 return btf_is_any_enum(t) && btf_vlen(t) == 0;
3902 static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3904 if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3905 return btf_equal_enum(t1, t2);
3906 /* At this point either t1 or t2 or both are forward declarations, thus:
3907 * - skip comparing vlen because it is zero for forward declarations;
3908 * - skip comparing size to allow enum forward declarations
3909 * to be compatible with enum64 full declarations;
3910 * - skip comparing kind for the same reason.
3912 return t1->name_off == t2->name_off &&
3913 btf_is_any_enum(t1) && btf_is_any_enum(t2);
3917 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3918 * as referenced type IDs equivalence is established separately during type
3919 * graph equivalence check algorithm.
3921 static long btf_hash_struct(struct btf_type *t)
3923 const struct btf_member *member = btf_members(t);
3924 __u32 vlen = btf_vlen(t);
3925 long h = btf_hash_common(t);
3928 for (i = 0; i < vlen; i++) {
3929 h = hash_combine(h, member->name_off);
3930 h = hash_combine(h, member->offset);
3931 /* no hashing of referenced type ID, it can be unresolved yet */
3938 * Check structural compatibility of two STRUCTs/UNIONs, ignoring referenced
3939 * type IDs. This check is performed during type graph equivalence check and
3940 * referenced types equivalence is checked separately.
3942 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3944 const struct btf_member *m1, *m2;
3948 if (!btf_equal_common(t1, t2))
3951 vlen = btf_vlen(t1);
3952 m1 = btf_members(t1);
3953 m2 = btf_members(t2);
3954 for (i = 0; i < vlen; i++) {
3955 if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3964 * Calculate type signature hash of ARRAY, including referenced type IDs,
3965 * under assumption that they were already resolved to canonical type IDs and
3966 * are not going to change.
3968 static long btf_hash_array(struct btf_type *t)
3970 const struct btf_array *info = btf_array(t);
3971 long h = btf_hash_common(t);
3973 h = hash_combine(h, info->type);
3974 h = hash_combine(h, info->index_type);
3975 h = hash_combine(h, info->nelems);
3980 * Check exact equality of two ARRAYs, taking into account referenced
3981 * type IDs, under assumption that they were already resolved to canonical
3982 * type IDs and are not going to change.
3983 * This function is called during reference types deduplication to compare
3984 * ARRAY to potential canonical representative.
3986 static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3988 const struct btf_array *info1, *info2;
3990 if (!btf_equal_common(t1, t2))
3993 info1 = btf_array(t1);
3994 info2 = btf_array(t2);
3995 return info1->type == info2->type &&
3996 info1->index_type == info2->index_type &&
3997 info1->nelems == info2->nelems;
4001 * Check structural compatibility of two ARRAYs, ignoring referenced type
4002 * IDs. This check is performed during type graph equivalence check and
4003 * referenced types equivalence is checked separately.
4005 static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
4007 if (!btf_equal_common(t1, t2))
4010 return btf_array(t1)->nelems == btf_array(t2)->nelems;
4014 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
4015 * under assumption that they were already resolved to canonical type IDs and
4016 * are not going to change.
4018 static long btf_hash_fnproto(struct btf_type *t)
4020 const struct btf_param *member = btf_params(t);
4021 __u16 vlen = btf_vlen(t);
4022 long h = btf_hash_common(t);
4025 for (i = 0; i < vlen; i++) {
4026 h = hash_combine(h, member->name_off);
4027 h = hash_combine(h, member->type);
4034 * Check exact equality of two FUNC_PROTOs, taking into account referenced
4035 * type IDs, under assumption that they were already resolved to canonical
4036 * type IDs and are not going to change.
4037 * This function is called during reference types deduplication to compare
4038 * FUNC_PROTO to potential canonical representative.
4040 static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
4042 const struct btf_param *m1, *m2;
4046 if (!btf_equal_common(t1, t2))
4049 vlen = btf_vlen(t1);
4050 m1 = btf_params(t1);
4051 m2 = btf_params(t2);
4052 for (i = 0; i < vlen; i++) {
4053 if (m1->name_off != m2->name_off || m1->type != m2->type)
4062 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
4063 * IDs. This check is performed during type graph equivalence check and
4064 * referenced types equivalence is checked separately.
4066 static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
4068 const struct btf_param *m1, *m2;
4072 /* skip return type ID */
4073 if (t1->name_off != t2->name_off || t1->info != t2->info)
4076 vlen = btf_vlen(t1);
4077 m1 = btf_params(t1);
4078 m2 = btf_params(t2);
4079 for (i = 0; i < vlen; i++) {
4080 if (m1->name_off != m2->name_off)
4088 /* Prepare split BTF for deduplication by calculating hashes of base BTF's
4089 * types and initializing the rest of the state (canonical type mapping) for
4090 * the fixed base BTF part.
4092 static int btf_dedup_prep(struct btf_dedup *d)
4098 if (!d->btf->base_btf)
4101 for (type_id = 1; type_id < d->btf->start_id; type_id++) {
4102 t = btf_type_by_id(d->btf, type_id);
4104 /* all base BTF types are self-canonical by definition */
4105 d->map[type_id] = type_id;
4107 switch (btf_kind(t)) {
4109 case BTF_KIND_DATASEC:
4110 /* VAR and DATASEC are never hash/deduplicated */
4112 case BTF_KIND_CONST:
4113 case BTF_KIND_VOLATILE:
4114 case BTF_KIND_RESTRICT:
4117 case BTF_KIND_TYPEDEF:
4119 case BTF_KIND_FLOAT:
4120 case BTF_KIND_TYPE_TAG:
4121 h = btf_hash_common(t);
4124 case BTF_KIND_DECL_TAG:
4125 h = btf_hash_int_decl_tag(t);
4128 case BTF_KIND_ENUM64:
4129 h = btf_hash_enum(t);
4131 case BTF_KIND_STRUCT:
4132 case BTF_KIND_UNION:
4133 h = btf_hash_struct(t);
4135 case BTF_KIND_ARRAY:
4136 h = btf_hash_array(t);
4138 case BTF_KIND_FUNC_PROTO:
4139 h = btf_hash_fnproto(t);
4142 pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
4145 if (btf_dedup_table_add(d, h, type_id))
4153 * Deduplicate primitive types, that can't reference other types, by calculating
4154 * their type signature hash and comparing them with any possible canonical
4155 * candidate. If no canonical candidate matches, type itself is marked as
4156 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
4158 static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
4160 struct btf_type *t = btf_type_by_id(d->btf, type_id);
4161 struct hashmap_entry *hash_entry;
4162 struct btf_type *cand;
4163 /* if we don't find equivalent type, then we are canonical */
4164 __u32 new_id = type_id;
4168 switch (btf_kind(t)) {
4169 case BTF_KIND_CONST:
4170 case BTF_KIND_VOLATILE:
4171 case BTF_KIND_RESTRICT:
4173 case BTF_KIND_TYPEDEF:
4174 case BTF_KIND_ARRAY:
4175 case BTF_KIND_STRUCT:
4176 case BTF_KIND_UNION:
4178 case BTF_KIND_FUNC_PROTO:
4180 case BTF_KIND_DATASEC:
4181 case BTF_KIND_DECL_TAG:
4182 case BTF_KIND_TYPE_TAG:
4186 h = btf_hash_int_decl_tag(t);
4187 for_each_dedup_cand(d, hash_entry, h) {
4188 cand_id = hash_entry->value;
4189 cand = btf_type_by_id(d->btf, cand_id);
4190 if (btf_equal_int_tag(t, cand)) {
4198 case BTF_KIND_ENUM64:
4199 h = btf_hash_enum(t);
4200 for_each_dedup_cand(d, hash_entry, h) {
4201 cand_id = hash_entry->value;
4202 cand = btf_type_by_id(d->btf, cand_id);
4203 if (btf_equal_enum(t, cand)) {
4207 if (btf_compat_enum(t, cand)) {
4208 if (btf_is_enum_fwd(t)) {
4209 /* resolve fwd to full enum */
4213 /* resolve canonical enum fwd to full enum */
4214 d->map[cand_id] = type_id;
4220 case BTF_KIND_FLOAT:
4221 h = btf_hash_common(t);
4222 for_each_dedup_cand(d, hash_entry, h) {
4223 cand_id = hash_entry->value;
4224 cand = btf_type_by_id(d->btf, cand_id);
4225 if (btf_equal_common(t, cand)) {
4236 d->map[type_id] = new_id;
4237 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4243 static int btf_dedup_prim_types(struct btf_dedup *d)
4247 for (i = 0; i < d->btf->nr_types; i++) {
4248 err = btf_dedup_prim_type(d, d->btf->start_id + i);
4256 * Check whether type is already mapped into canonical one (could be to itself).
4258 static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
4260 return d->map[type_id] <= BTF_MAX_NR_TYPES;
4264 * Resolve type ID into its canonical type ID, if any; otherwise return original
4265 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
4266 * STRUCT/UNION link and resolve it into canonical type ID as well.
4268 static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
4270 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
4271 type_id = d->map[type_id];
4276 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
4279 static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
4281 __u32 orig_type_id = type_id;
4283 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
4286 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
4287 type_id = d->map[type_id];
4289 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
4292 return orig_type_id;
4296 static inline __u16 btf_fwd_kind(struct btf_type *t)
4298 return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
4301 /* Check if given two types are identical ARRAY definitions */
4302 static bool btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
4304 struct btf_type *t1, *t2;
4306 t1 = btf_type_by_id(d->btf, id1);
4307 t2 = btf_type_by_id(d->btf, id2);
4308 if (!btf_is_array(t1) || !btf_is_array(t2))
4311 return btf_equal_array(t1, t2);
4314 /* Check if given two types are identical STRUCT/UNION definitions */
4315 static bool btf_dedup_identical_structs(struct btf_dedup *d, __u32 id1, __u32 id2)
4317 const struct btf_member *m1, *m2;
4318 struct btf_type *t1, *t2;
4321 t1 = btf_type_by_id(d->btf, id1);
4322 t2 = btf_type_by_id(d->btf, id2);
4324 if (!btf_is_composite(t1) || btf_kind(t1) != btf_kind(t2))
4327 if (!btf_shallow_equal_struct(t1, t2))
4330 m1 = btf_members(t1);
4331 m2 = btf_members(t2);
4332 for (i = 0, n = btf_vlen(t1); i < n; i++, m1++, m2++) {
4333 if (m1->type != m2->type &&
4334 !btf_dedup_identical_arrays(d, m1->type, m2->type) &&
4335 !btf_dedup_identical_structs(d, m1->type, m2->type))
4342 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
4343 * call it "candidate graph" in this description for brevity) to a type graph
4344 * formed by (potential) canonical struct/union ("canonical graph" for brevity
4345 * here, though keep in mind that not all types in canonical graph are
4346 * necessarily canonical representatives themselves, some of them might be
4347 * duplicates or its uniqueness might not have been established yet).
4349 * - >0, if type graphs are equivalent;
4350 * - 0, if not equivalent;
4353 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
4354 * equivalence of BTF types at each step. If at any point BTF types in candidate
4355 * and canonical graphs are not compatible structurally, whole graphs are
4356 * incompatible. If types are structurally equivalent (i.e., all information
4357 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
4358 * a `cand_id` is recoded in hypothetical mapping (`btf_dedup->hypot_map`).
4359 * If a type references other types, then those referenced types are checked
4360 * for equivalence recursively.
4362 * During DFS traversal, if we find that for current `canon_id` type we
4363 * already have some mapping in hypothetical map, we check for two possible
4365 * - `canon_id` is mapped to exactly the same type as `cand_id`. This will
4366 * happen when type graphs have cycles. In this case we assume those two
4367 * types are equivalent.
4368 * - `canon_id` is mapped to different type. This is contradiction in our
4369 * hypothetical mapping, because same graph in canonical graph corresponds
4370 * to two different types in candidate graph, which for equivalent type
4371 * graphs shouldn't happen. This condition terminates equivalence check
4372 * with negative result.
4374 * If type graphs traversal exhausts types to check and find no contradiction,
4375 * then type graphs are equivalent.
4377 * When checking types for equivalence, there is one special case: FWD types.
4378 * If FWD type resolution is allowed and one of the types (either from canonical
4379 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
4380 * flag) and their names match, hypothetical mapping is updated to point from
4381 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
4382 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
4384 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
4385 * if there are two exactly named (or anonymous) structs/unions that are
4386 * compatible structurally, one of which has FWD field, while other is concrete
4387 * STRUCT/UNION, but according to C sources they are different structs/unions
4388 * that are referencing different types with the same name. This is extremely
4389 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
4390 * this logic is causing problems.
4392 * Doing FWD resolution means that both candidate and/or canonical graphs can
4393 * consists of portions of the graph that come from multiple compilation units.
4394 * This is due to the fact that types within single compilation unit are always
4395 * deduplicated and FWDs are already resolved, if referenced struct/union
4396 * definition is available. So, if we had unresolved FWD and found corresponding
4397 * STRUCT/UNION, they will be from different compilation units. This
4398 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
4399 * type graph will likely have at least two different BTF types that describe
4400 * same type (e.g., most probably there will be two different BTF types for the
4401 * same 'int' primitive type) and could even have "overlapping" parts of type
4402 * graph that describe same subset of types.
4404 * This in turn means that our assumption that each type in canonical graph
4405 * must correspond to exactly one type in candidate graph might not hold
4406 * anymore and will make it harder to detect contradictions using hypothetical
4407 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
4408 * resolution only in canonical graph. FWDs in candidate graphs are never
4409 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
4411 * - Both types in canonical and candidate graphs are FWDs. If they are
4412 * structurally equivalent, then they can either be both resolved to the
4413 * same STRUCT/UNION or not resolved at all. In both cases they are
4414 * equivalent and there is no need to resolve FWD on candidate side.
4415 * - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
4416 * so nothing to resolve as well, algorithm will check equivalence anyway.
4417 * - Type in canonical graph is FWD, while type in candidate is concrete
4418 * STRUCT/UNION. In this case candidate graph comes from single compilation
4419 * unit, so there is exactly one BTF type for each unique C type. After
4420 * resolving FWD into STRUCT/UNION, there might be more than one BTF type
4421 * in canonical graph mapping to single BTF type in candidate graph, but
4422 * because hypothetical mapping maps from canonical to candidate types, it's
4423 * alright, and we still maintain the property of having single `canon_id`
4424 * mapping to single `cand_id` (there could be two different `canon_id`
4425 * mapped to the same `cand_id`, but it's not contradictory).
4426 * - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
4427 * graph is FWD. In this case we are just going to check compatibility of
4428 * STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
4429 * assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
4430 * a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
4431 * turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
4434 static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
4437 struct btf_type *cand_type;
4438 struct btf_type *canon_type;
4439 __u32 hypot_type_id;
4444 /* if both resolve to the same canonical, they must be equivalent */
4445 if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
4448 canon_id = resolve_fwd_id(d, canon_id);
4450 hypot_type_id = d->hypot_map[canon_id];
4451 if (hypot_type_id <= BTF_MAX_NR_TYPES) {
4452 if (hypot_type_id == cand_id)
4454 /* In some cases compiler will generate different DWARF types
4455 * for *identical* array type definitions and use them for
4456 * different fields within the *same* struct. This breaks type
4457 * equivalence check, which makes an assumption that candidate
4458 * types sub-graph has a consistent and deduped-by-compiler
4459 * types within a single CU. So work around that by explicitly
4460 * allowing identical array types here.
4462 if (btf_dedup_identical_arrays(d, hypot_type_id, cand_id))
4464 /* It turns out that similar situation can happen with
4465 * struct/union sometimes, sigh... Handle the case where
4466 * structs/unions are exactly the same, down to the referenced
4467 * type IDs. Anything more complicated (e.g., if referenced
4468 * types are different, but equivalent) is *way more*
4469 * complicated and requires a many-to-many equivalence mapping.
4471 if (btf_dedup_identical_structs(d, hypot_type_id, cand_id))
4476 if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
4479 cand_type = btf_type_by_id(d->btf, cand_id);
4480 canon_type = btf_type_by_id(d->btf, canon_id);
4481 cand_kind = btf_kind(cand_type);
4482 canon_kind = btf_kind(canon_type);
4484 if (cand_type->name_off != canon_type->name_off)
4487 /* FWD <--> STRUCT/UNION equivalence check, if enabled */
4488 if ((cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
4489 && cand_kind != canon_kind) {
4493 if (cand_kind == BTF_KIND_FWD) {
4494 real_kind = canon_kind;
4495 fwd_kind = btf_fwd_kind(cand_type);
4497 real_kind = cand_kind;
4498 fwd_kind = btf_fwd_kind(canon_type);
4499 /* we'd need to resolve base FWD to STRUCT/UNION */
4500 if (fwd_kind == real_kind && canon_id < d->btf->start_id)
4501 d->hypot_adjust_canon = true;
4503 return fwd_kind == real_kind;
4506 if (cand_kind != canon_kind)
4509 switch (cand_kind) {
4511 return btf_equal_int_tag(cand_type, canon_type);
4514 case BTF_KIND_ENUM64:
4515 return btf_compat_enum(cand_type, canon_type);
4518 case BTF_KIND_FLOAT:
4519 return btf_equal_common(cand_type, canon_type);
4521 case BTF_KIND_CONST:
4522 case BTF_KIND_VOLATILE:
4523 case BTF_KIND_RESTRICT:
4525 case BTF_KIND_TYPEDEF:
4527 case BTF_KIND_TYPE_TAG:
4528 if (cand_type->info != canon_type->info)
4530 return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4532 case BTF_KIND_ARRAY: {
4533 const struct btf_array *cand_arr, *canon_arr;
4535 if (!btf_compat_array(cand_type, canon_type))
4537 cand_arr = btf_array(cand_type);
4538 canon_arr = btf_array(canon_type);
4539 eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
4542 return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
4545 case BTF_KIND_STRUCT:
4546 case BTF_KIND_UNION: {
4547 const struct btf_member *cand_m, *canon_m;
4550 if (!btf_shallow_equal_struct(cand_type, canon_type))
4552 vlen = btf_vlen(cand_type);
4553 cand_m = btf_members(cand_type);
4554 canon_m = btf_members(canon_type);
4555 for (i = 0; i < vlen; i++) {
4556 eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
4566 case BTF_KIND_FUNC_PROTO: {
4567 const struct btf_param *cand_p, *canon_p;
4570 if (!btf_compat_fnproto(cand_type, canon_type))
4572 eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4575 vlen = btf_vlen(cand_type);
4576 cand_p = btf_params(cand_type);
4577 canon_p = btf_params(canon_type);
4578 for (i = 0; i < vlen; i++) {
4579 eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
4595 * Use hypothetical mapping, produced by successful type graph equivalence
4596 * check, to augment existing struct/union canonical mapping, where possible.
4598 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
4599 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
4600 * it doesn't matter if FWD type was part of canonical graph or candidate one,
4601 * we are recording the mapping anyway. As opposed to carefulness required
4602 * for struct/union correspondence mapping (described below), for FWD resolution
4603 * it's not important, as by the time that FWD type (reference type) will be
4604 * deduplicated all structs/unions will be deduped already anyway.
4606 * Recording STRUCT/UNION mapping is purely a performance optimization and is
4607 * not required for correctness. It needs to be done carefully to ensure that
4608 * struct/union from candidate's type graph is not mapped into corresponding
4609 * struct/union from canonical type graph that itself hasn't been resolved into
4610 * canonical representative. The only guarantee we have is that canonical
4611 * struct/union was determined as canonical and that won't change. But any
4612 * types referenced through that struct/union fields could have been not yet
4613 * resolved, so in case like that it's too early to establish any kind of
4614 * correspondence between structs/unions.
4616 * No canonical correspondence is derived for primitive types (they are already
4617 * deduplicated completely already anyway) or reference types (they rely on
4618 * stability of struct/union canonical relationship for equivalence checks).
4620 static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
4622 __u32 canon_type_id, targ_type_id;
4623 __u16 t_kind, c_kind;
4627 for (i = 0; i < d->hypot_cnt; i++) {
4628 canon_type_id = d->hypot_list[i];
4629 targ_type_id = d->hypot_map[canon_type_id];
4630 t_id = resolve_type_id(d, targ_type_id);
4631 c_id = resolve_type_id(d, canon_type_id);
4632 t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
4633 c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
4635 * Resolve FWD into STRUCT/UNION.
4636 * It's ok to resolve FWD into STRUCT/UNION that's not yet
4637 * mapped to canonical representative (as opposed to
4638 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
4639 * eventually that struct is going to be mapped and all resolved
4640 * FWDs will automatically resolve to correct canonical
4641 * representative. This will happen before ref type deduping,
4642 * which critically depends on stability of these mapping. This
4643 * stability is not a requirement for STRUCT/UNION equivalence
4647 /* if it's the split BTF case, we still need to point base FWD
4648 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4649 * will be resolved against base FWD. If we don't point base
4650 * canonical FWD to the resolved STRUCT/UNION, then all the
4651 * FWDs in split BTF won't be correctly resolved to a proper
4654 if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4655 d->map[c_id] = t_id;
4657 /* if graph equivalence determined that we'd need to adjust
4658 * base canonical types, then we need to only point base FWDs
4659 * to STRUCTs/UNIONs and do no more modifications. For all
4660 * other purposes the type graphs were not equivalent.
4662 if (d->hypot_adjust_canon)
4665 if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4666 d->map[t_id] = c_id;
4668 if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4669 c_kind != BTF_KIND_FWD &&
4670 is_type_mapped(d, c_id) &&
4671 !is_type_mapped(d, t_id)) {
4673 * as a perf optimization, we can map struct/union
4674 * that's part of type graph we just verified for
4675 * equivalence. We can do that for struct/union that has
4676 * canonical representative only, though.
4678 d->map[t_id] = c_id;
4684 * Deduplicate struct/union types.
4686 * For each struct/union type its type signature hash is calculated, taking
4687 * into account type's name, size, number, order and names of fields, but
4688 * ignoring type ID's referenced from fields, because they might not be deduped
4689 * completely until after reference types deduplication phase. This type hash
4690 * is used to iterate over all potential canonical types, sharing same hash.
4691 * For each canonical candidate we check whether type graphs that they form
4692 * (through referenced types in fields and so on) are equivalent using algorithm
4693 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4694 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4695 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4696 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4697 * potentially map other structs/unions to their canonical representatives,
4698 * if such relationship hasn't yet been established. This speeds up algorithm
4699 * by eliminating some of the duplicate work.
4701 * If no matching canonical representative was found, struct/union is marked
4702 * as canonical for itself and is added into btf_dedup->dedup_table hash map
4703 * for further look ups.
4705 static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4707 struct btf_type *cand_type, *t;
4708 struct hashmap_entry *hash_entry;
4709 /* if we don't find equivalent type, then we are canonical */
4710 __u32 new_id = type_id;
4714 /* already deduped or is in process of deduping (loop detected) */
4715 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4718 t = btf_type_by_id(d->btf, type_id);
4721 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4724 h = btf_hash_struct(t);
4725 for_each_dedup_cand(d, hash_entry, h) {
4726 __u32 cand_id = hash_entry->value;
4730 * Even though btf_dedup_is_equiv() checks for
4731 * btf_shallow_equal_struct() internally when checking two
4732 * structs (unions) for equivalence, we need to guard here
4733 * from picking matching FWD type as a dedup candidate.
4734 * This can happen due to hash collision. In such case just
4735 * relying on btf_dedup_is_equiv() would lead to potentially
4736 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4737 * FWD and compatible STRUCT/UNION are considered equivalent.
4739 cand_type = btf_type_by_id(d->btf, cand_id);
4740 if (!btf_shallow_equal_struct(t, cand_type))
4743 btf_dedup_clear_hypot_map(d);
4744 eq = btf_dedup_is_equiv(d, type_id, cand_id);
4749 btf_dedup_merge_hypot_map(d);
4750 if (d->hypot_adjust_canon) /* not really equivalent */
4756 d->map[type_id] = new_id;
4757 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4763 static int btf_dedup_struct_types(struct btf_dedup *d)
4767 for (i = 0; i < d->btf->nr_types; i++) {
4768 err = btf_dedup_struct_type(d, d->btf->start_id + i);
4776 * Deduplicate reference type.
4778 * Once all primitive and struct/union types got deduplicated, we can easily
4779 * deduplicate all other (reference) BTF types. This is done in two steps:
4781 * 1. Resolve all referenced type IDs into their canonical type IDs. This
4782 * resolution can be done either immediately for primitive or struct/union types
4783 * (because they were deduped in previous two phases) or recursively for
4784 * reference types. Recursion will always terminate at either primitive or
4785 * struct/union type, at which point we can "unwind" chain of reference types
4786 * one by one. There is no danger of encountering cycles because in C type
4787 * system the only way to form type cycle is through struct/union, so any chain
4788 * of reference types, even those taking part in a type cycle, will inevitably
4789 * reach struct/union at some point.
4791 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4792 * becomes "stable", in the sense that no further deduplication will cause
4793 * any changes to it. With that, it's now possible to calculate type's signature
4794 * hash (this time taking into account referenced type IDs) and loop over all
4795 * potential canonical representatives. If no match was found, current type
4796 * will become canonical representative of itself and will be added into
4797 * btf_dedup->dedup_table as another possible canonical representative.
4799 static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4801 struct hashmap_entry *hash_entry;
4802 __u32 new_id = type_id, cand_id;
4803 struct btf_type *t, *cand;
4804 /* if we don't find equivalent type, then we are representative type */
4808 if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4810 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4811 return resolve_type_id(d, type_id);
4813 t = btf_type_by_id(d->btf, type_id);
4814 d->map[type_id] = BTF_IN_PROGRESS_ID;
4816 switch (btf_kind(t)) {
4817 case BTF_KIND_CONST:
4818 case BTF_KIND_VOLATILE:
4819 case BTF_KIND_RESTRICT:
4821 case BTF_KIND_TYPEDEF:
4823 case BTF_KIND_TYPE_TAG:
4824 ref_type_id = btf_dedup_ref_type(d, t->type);
4825 if (ref_type_id < 0)
4827 t->type = ref_type_id;
4829 h = btf_hash_common(t);
4830 for_each_dedup_cand(d, hash_entry, h) {
4831 cand_id = hash_entry->value;
4832 cand = btf_type_by_id(d->btf, cand_id);
4833 if (btf_equal_common(t, cand)) {
4840 case BTF_KIND_DECL_TAG:
4841 ref_type_id = btf_dedup_ref_type(d, t->type);
4842 if (ref_type_id < 0)
4844 t->type = ref_type_id;
4846 h = btf_hash_int_decl_tag(t);
4847 for_each_dedup_cand(d, hash_entry, h) {
4848 cand_id = hash_entry->value;
4849 cand = btf_type_by_id(d->btf, cand_id);
4850 if (btf_equal_int_tag(t, cand)) {
4857 case BTF_KIND_ARRAY: {
4858 struct btf_array *info = btf_array(t);
4860 ref_type_id = btf_dedup_ref_type(d, info->type);
4861 if (ref_type_id < 0)
4863 info->type = ref_type_id;
4865 ref_type_id = btf_dedup_ref_type(d, info->index_type);
4866 if (ref_type_id < 0)
4868 info->index_type = ref_type_id;
4870 h = btf_hash_array(t);
4871 for_each_dedup_cand(d, hash_entry, h) {
4872 cand_id = hash_entry->value;
4873 cand = btf_type_by_id(d->btf, cand_id);
4874 if (btf_equal_array(t, cand)) {
4882 case BTF_KIND_FUNC_PROTO: {
4883 struct btf_param *param;
4887 ref_type_id = btf_dedup_ref_type(d, t->type);
4888 if (ref_type_id < 0)
4890 t->type = ref_type_id;
4893 param = btf_params(t);
4894 for (i = 0; i < vlen; i++) {
4895 ref_type_id = btf_dedup_ref_type(d, param->type);
4896 if (ref_type_id < 0)
4898 param->type = ref_type_id;
4902 h = btf_hash_fnproto(t);
4903 for_each_dedup_cand(d, hash_entry, h) {
4904 cand_id = hash_entry->value;
4905 cand = btf_type_by_id(d->btf, cand_id);
4906 if (btf_equal_fnproto(t, cand)) {
4918 d->map[type_id] = new_id;
4919 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4925 static int btf_dedup_ref_types(struct btf_dedup *d)
4929 for (i = 0; i < d->btf->nr_types; i++) {
4930 err = btf_dedup_ref_type(d, d->btf->start_id + i);
4934 /* we won't need d->dedup_table anymore */
4935 hashmap__free(d->dedup_table);
4936 d->dedup_table = NULL;
4941 * Collect a map from type names to type ids for all canonical structs
4942 * and unions. If the same name is shared by several canonical types
4943 * use a special value 0 to indicate this fact.
4945 static int btf_dedup_fill_unique_names_map(struct btf_dedup *d, struct hashmap *names_map)
4947 __u32 nr_types = btf__type_cnt(d->btf);
4954 * Iterate over base and split module ids in order to get all
4955 * available structs in the map.
4957 for (type_id = 1; type_id < nr_types; ++type_id) {
4958 t = btf_type_by_id(d->btf, type_id);
4961 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4964 /* Skip non-canonical types */
4965 if (type_id != d->map[type_id])
4968 err = hashmap__add(names_map, t->name_off, type_id);
4970 err = hashmap__set(names_map, t->name_off, 0, NULL, NULL);
4979 static int btf_dedup_resolve_fwd(struct btf_dedup *d, struct hashmap *names_map, __u32 type_id)
4981 struct btf_type *t = btf_type_by_id(d->btf, type_id);
4982 enum btf_fwd_kind fwd_kind = btf_kflag(t);
4983 __u16 cand_kind, kind = btf_kind(t);
4984 struct btf_type *cand_t;
4987 if (kind != BTF_KIND_FWD)
4990 /* Skip if this FWD already has a mapping */
4991 if (type_id != d->map[type_id])
4994 if (!hashmap__find(names_map, t->name_off, &cand_id))
4997 /* Zero is a special value indicating that name is not unique */
5001 cand_t = btf_type_by_id(d->btf, cand_id);
5002 cand_kind = btf_kind(cand_t);
5003 if ((cand_kind == BTF_KIND_STRUCT && fwd_kind != BTF_FWD_STRUCT) ||
5004 (cand_kind == BTF_KIND_UNION && fwd_kind != BTF_FWD_UNION))
5007 d->map[type_id] = cand_id;
5013 * Resolve unambiguous forward declarations.
5015 * The lion's share of all FWD declarations is resolved during
5016 * `btf_dedup_struct_types` phase when different type graphs are
5017 * compared against each other. However, if in some compilation unit a
5018 * FWD declaration is not a part of a type graph compared against
5019 * another type graph that declaration's canonical type would not be
5025 * struct foo *some_global;
5029 * struct foo { int u; };
5030 * struct foo *another_global;
5032 * After `btf_dedup_struct_types` the BTF looks as follows:
5034 * [1] STRUCT 'foo' size=4 vlen=1 ...
5035 * [2] INT 'int' size=4 ...
5036 * [3] PTR '(anon)' type_id=1
5037 * [4] FWD 'foo' fwd_kind=struct
5038 * [5] PTR '(anon)' type_id=4
5040 * This pass assumes that such FWD declarations should be mapped to
5041 * structs or unions with identical name in case if the name is not
5044 static int btf_dedup_resolve_fwds(struct btf_dedup *d)
5047 struct hashmap *names_map;
5049 names_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
5050 if (IS_ERR(names_map))
5051 return PTR_ERR(names_map);
5053 err = btf_dedup_fill_unique_names_map(d, names_map);
5057 for (i = 0; i < d->btf->nr_types; i++) {
5058 err = btf_dedup_resolve_fwd(d, names_map, d->btf->start_id + i);
5064 hashmap__free(names_map);
5071 * After we established for each type its corresponding canonical representative
5072 * type, we now can eliminate types that are not canonical and leave only
5073 * canonical ones layed out sequentially in memory by copying them over
5074 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
5075 * a map from original type ID to a new compacted type ID, which will be used
5076 * during next phase to "fix up" type IDs, referenced from struct/union and
5079 static int btf_dedup_compact_types(struct btf_dedup *d)
5082 __u32 next_type_id = d->btf->start_id;
5083 const struct btf_type *t;
5087 /* we are going to reuse hypot_map to store compaction remapping */
5088 d->hypot_map[0] = 0;
5089 /* base BTF types are not renumbered */
5090 for (id = 1; id < d->btf->start_id; id++)
5091 d->hypot_map[id] = id;
5092 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
5093 d->hypot_map[id] = BTF_UNPROCESSED_ID;
5095 p = d->btf->types_data;
5097 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
5098 if (d->map[id] != id)
5101 t = btf__type_by_id(d->btf, id);
5102 len = btf_type_size(t);
5107 d->hypot_map[id] = next_type_id;
5108 d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
5113 /* shrink struct btf's internal types index and update btf_header */
5114 d->btf->nr_types = next_type_id - d->btf->start_id;
5115 d->btf->type_offs_cap = d->btf->nr_types;
5116 d->btf->hdr->type_len = p - d->btf->types_data;
5117 new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
5119 if (d->btf->type_offs_cap && !new_offs)
5121 d->btf->type_offs = new_offs;
5122 d->btf->hdr->str_off = d->btf->hdr->type_len;
5123 d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
5128 * Figure out final (deduplicated and compacted) type ID for provided original
5129 * `type_id` by first resolving it into corresponding canonical type ID and
5130 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
5131 * which is populated during compaction phase.
5133 static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
5135 struct btf_dedup *d = ctx;
5136 __u32 resolved_type_id, new_type_id;
5138 resolved_type_id = resolve_type_id(d, *type_id);
5139 new_type_id = d->hypot_map[resolved_type_id];
5140 if (new_type_id > BTF_MAX_NR_TYPES)
5143 *type_id = new_type_id;
5148 * Remap referenced type IDs into deduped type IDs.
5150 * After BTF types are deduplicated and compacted, their final type IDs may
5151 * differ from original ones. The map from original to a corresponding
5152 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
5153 * compaction phase. During remapping phase we are rewriting all type IDs
5154 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
5155 * their final deduped type IDs.
5157 static int btf_dedup_remap_types(struct btf_dedup *d)
5161 for (i = 0; i < d->btf->nr_types; i++) {
5162 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
5163 struct btf_field_iter it;
5166 r = btf_field_iter_init(&it, t, BTF_FIELD_ITER_IDS);
5170 while ((type_id = btf_field_iter_next(&it))) {
5171 __u32 resolved_id, new_id;
5173 resolved_id = resolve_type_id(d, *type_id);
5174 new_id = d->hypot_map[resolved_id];
5175 if (new_id > BTF_MAX_NR_TYPES)
5185 r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
5193 * Probe few well-known locations for vmlinux kernel image and try to load BTF
5194 * data out of it to use for target BTF.
5196 struct btf *btf__load_vmlinux_btf(void)
5198 const char *sysfs_btf_path = "/sys/kernel/btf/vmlinux";
5199 /* fall back locations, trying to find vmlinux on disk */
5200 const char *locations[] = {
5201 "/boot/vmlinux-%1$s",
5202 "/lib/modules/%1$s/vmlinux-%1$s",
5203 "/lib/modules/%1$s/build/vmlinux",
5204 "/usr/lib/modules/%1$s/kernel/vmlinux",
5205 "/usr/lib/debug/boot/vmlinux-%1$s",
5206 "/usr/lib/debug/boot/vmlinux-%1$s.debug",
5207 "/usr/lib/debug/lib/modules/%1$s/vmlinux",
5209 char path[PATH_MAX + 1];
5214 /* is canonical sysfs location accessible? */
5215 if (faccessat(AT_FDCWD, sysfs_btf_path, F_OK, AT_EACCESS) < 0) {
5216 pr_warn("kernel BTF is missing at '%s', was CONFIG_DEBUG_INFO_BTF enabled?\n",
5219 btf = btf__parse(sysfs_btf_path, NULL);
5222 pr_warn("failed to read kernel BTF from '%s': %s\n",
5223 sysfs_btf_path, errstr(err));
5224 return libbpf_err_ptr(err);
5226 pr_debug("loaded kernel BTF from '%s'\n", sysfs_btf_path);
5230 /* try fallback locations */
5232 for (i = 0; i < ARRAY_SIZE(locations); i++) {
5233 snprintf(path, PATH_MAX, locations[i], buf.release);
5235 if (faccessat(AT_FDCWD, path, R_OK, AT_EACCESS))
5238 btf = btf__parse(path, NULL);
5239 err = libbpf_get_error(btf);
5240 pr_debug("loading kernel BTF '%s': %s\n", path, errstr(err));
5247 pr_warn("failed to find valid kernel BTF\n");
5248 return libbpf_err_ptr(-ESRCH);
5251 struct btf *libbpf_find_kernel_btf(void) __attribute__((alias("btf__load_vmlinux_btf")));
5253 struct btf *btf__load_module_btf(const char *module_name, struct btf *vmlinux_btf)
5257 snprintf(path, sizeof(path), "/sys/kernel/btf/%s", module_name);
5258 return btf__parse_split(path, vmlinux_btf);
5261 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
5263 const struct btf_ext_info *seg;
5264 struct btf_ext_info_sec *sec;
5267 seg = &btf_ext->func_info;
5268 for_each_btf_ext_sec(seg, sec) {
5269 struct bpf_func_info_min *rec;
5271 for_each_btf_ext_rec(seg, sec, i, rec) {
5272 err = visit(&rec->type_id, ctx);
5278 seg = &btf_ext->core_relo_info;
5279 for_each_btf_ext_sec(seg, sec) {
5280 struct bpf_core_relo *rec;
5282 for_each_btf_ext_rec(seg, sec, i, rec) {
5283 err = visit(&rec->type_id, ctx);
5292 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
5294 const struct btf_ext_info *seg;
5295 struct btf_ext_info_sec *sec;
5298 seg = &btf_ext->func_info;
5299 for_each_btf_ext_sec(seg, sec) {
5300 err = visit(&sec->sec_name_off, ctx);
5305 seg = &btf_ext->line_info;
5306 for_each_btf_ext_sec(seg, sec) {
5307 struct bpf_line_info_min *rec;
5309 err = visit(&sec->sec_name_off, ctx);
5313 for_each_btf_ext_rec(seg, sec, i, rec) {
5314 err = visit(&rec->file_name_off, ctx);
5317 err = visit(&rec->line_off, ctx);
5323 seg = &btf_ext->core_relo_info;
5324 for_each_btf_ext_sec(seg, sec) {
5325 struct bpf_core_relo *rec;
5327 err = visit(&sec->sec_name_off, ctx);
5331 for_each_btf_ext_rec(seg, sec, i, rec) {
5332 err = visit(&rec->access_str_off, ctx);
5341 struct btf_distill {
5342 struct btf_pipe pipe;
5344 unsigned int split_start_id;
5345 unsigned int split_start_str;
5349 static int btf_add_distilled_type_ids(struct btf_distill *dist, __u32 i)
5351 struct btf_type *split_t = btf_type_by_id(dist->pipe.src, i);
5352 struct btf_field_iter it;
5356 err = btf_field_iter_init(&it, split_t, BTF_FIELD_ITER_IDS);
5359 while ((id = btf_field_iter_next(&it))) {
5360 struct btf_type *base_t;
5364 /* split BTF id, not needed */
5365 if (*id >= dist->split_start_id)
5367 /* already added ? */
5368 if (dist->id_map[*id] > 0)
5371 /* only a subset of base BTF types should be referenced from
5372 * split BTF; ensure nothing unexpected is referenced.
5374 base_t = btf_type_by_id(dist->pipe.src, *id);
5375 switch (btf_kind(base_t)) {
5377 case BTF_KIND_FLOAT:
5379 case BTF_KIND_ARRAY:
5380 case BTF_KIND_STRUCT:
5381 case BTF_KIND_UNION:
5382 case BTF_KIND_TYPEDEF:
5384 case BTF_KIND_ENUM64:
5386 case BTF_KIND_CONST:
5387 case BTF_KIND_RESTRICT:
5388 case BTF_KIND_VOLATILE:
5389 case BTF_KIND_FUNC_PROTO:
5390 case BTF_KIND_TYPE_TAG:
5391 dist->id_map[*id] = *id;
5394 pr_warn("unexpected reference to base type[%u] of kind [%u] when creating distilled base BTF.\n",
5395 *id, btf_kind(base_t));
5398 /* If a base type is used, ensure types it refers to are
5399 * marked as used also; so for example if we find a PTR to INT
5400 * we need both the PTR and INT.
5402 * The only exception is named struct/unions, since distilled
5403 * base BTF composite types have no members.
5405 if (btf_is_composite(base_t) && base_t->name_off)
5407 err = btf_add_distilled_type_ids(dist, *id);
5414 static int btf_add_distilled_types(struct btf_distill *dist)
5416 bool adding_to_base = dist->pipe.dst->start_id == 1;
5417 int id = btf__type_cnt(dist->pipe.dst);
5422 /* Add types for each of the required references to either distilled
5423 * base or split BTF, depending on type characteristics.
5425 for (i = 1; i < dist->split_start_id; i++) {
5429 if (!dist->id_map[i])
5431 t = btf_type_by_id(dist->pipe.src, i);
5433 name = btf__name_by_offset(dist->pipe.src, t->name_off);
5437 case BTF_KIND_FLOAT:
5439 /* Named int, float, fwd are added to base. */
5440 if (!adding_to_base)
5442 err = btf_add_type(&dist->pipe, t);
5444 case BTF_KIND_STRUCT:
5445 case BTF_KIND_UNION:
5446 /* Named struct/union are added to base as 0-vlen
5447 * struct/union of same size. Anonymous struct/unions
5448 * are added to split BTF as-is.
5450 if (adding_to_base) {
5453 err = btf_add_composite(dist->pipe.dst, kind, name, t->size);
5457 err = btf_add_type(&dist->pipe, t);
5461 case BTF_KIND_ENUM64:
5462 /* Named enum[64]s are added to base as a sized
5463 * enum; relocation will match with appropriately-named
5464 * and sized enum or enum64.
5466 * Anonymous enums are added to split BTF as-is.
5468 if (adding_to_base) {
5471 err = btf__add_enum(dist->pipe.dst, name, t->size);
5475 err = btf_add_type(&dist->pipe, t);
5478 case BTF_KIND_ARRAY:
5479 case BTF_KIND_TYPEDEF:
5481 case BTF_KIND_CONST:
5482 case BTF_KIND_RESTRICT:
5483 case BTF_KIND_VOLATILE:
5484 case BTF_KIND_FUNC_PROTO:
5485 case BTF_KIND_TYPE_TAG:
5486 /* All other types are added to split BTF. */
5489 err = btf_add_type(&dist->pipe, t);
5492 pr_warn("unexpected kind when adding base type '%s'[%u] of kind [%u] to distilled base BTF.\n",
5499 dist->id_map[i] = id++;
5504 /* Split BTF ids without a mapping will be shifted downwards since distilled
5505 * base BTF is smaller than the original base BTF. For those that have a
5506 * mapping (either to base or updated split BTF), update the id based on
5509 static int btf_update_distilled_type_ids(struct btf_distill *dist, __u32 i)
5511 struct btf_type *t = btf_type_by_id(dist->pipe.dst, i);
5512 struct btf_field_iter it;
5516 err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_IDS);
5519 while ((id = btf_field_iter_next(&it))) {
5520 if (dist->id_map[*id])
5521 *id = dist->id_map[*id];
5522 else if (*id >= dist->split_start_id)
5523 *id -= dist->diff_id;
5528 /* Create updated split BTF with distilled base BTF; distilled base BTF
5529 * consists of BTF information required to clarify the types that split
5530 * BTF refers to, omitting unneeded details. Specifically it will contain
5531 * base types and memberless definitions of named structs, unions and enumerated
5532 * types. Associated reference types like pointers, arrays and anonymous
5533 * structs, unions and enumerated types will be added to split BTF.
5534 * Size is recorded for named struct/unions to help guide matching to the
5535 * target base BTF during later relocation.
5537 * The only case where structs, unions or enumerated types are fully represented
5538 * is when they are anonymous; in such cases, the anonymous type is added to
5539 * split BTF in full.
5541 * We return newly-created split BTF where the split BTF refers to a newly-created
5542 * distilled base BTF. Both must be freed separately by the caller.
5544 int btf__distill_base(const struct btf *src_btf, struct btf **new_base_btf,
5545 struct btf **new_split_btf)
5547 struct btf *new_base = NULL, *new_split = NULL;
5548 const struct btf *old_base;
5549 unsigned int n = btf__type_cnt(src_btf);
5550 struct btf_distill dist = {};
5554 /* src BTF must be split BTF. */
5555 old_base = btf__base_btf(src_btf);
5556 if (!new_base_btf || !new_split_btf || !old_base)
5557 return libbpf_err(-EINVAL);
5559 new_base = btf__new_empty();
5561 return libbpf_err(-ENOMEM);
5563 btf__set_endianness(new_base, btf__endianness(src_btf));
5565 dist.id_map = calloc(n, sizeof(*dist.id_map));
5570 dist.pipe.src = src_btf;
5571 dist.pipe.dst = new_base;
5572 dist.pipe.str_off_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
5573 if (IS_ERR(dist.pipe.str_off_map)) {
5577 dist.split_start_id = btf__type_cnt(old_base);
5578 dist.split_start_str = old_base->hdr->str_len;
5580 /* Pass over src split BTF; generate the list of base BTF type ids it
5581 * references; these will constitute our distilled BTF set to be
5582 * distributed over base and split BTF as appropriate.
5584 for (i = src_btf->start_id; i < n; i++) {
5585 err = btf_add_distilled_type_ids(&dist, i);
5589 /* Next add types for each of the required references to base BTF and split BTF
5592 err = btf_add_distilled_types(&dist);
5596 /* Create new split BTF with distilled base BTF as its base; the final
5597 * state is split BTF with distilled base BTF that represents enough
5598 * about its base references to allow it to be relocated with the base
5601 new_split = btf__new_empty_split(new_base);
5606 dist.pipe.dst = new_split;
5607 /* First add all split types */
5608 for (i = src_btf->start_id; i < n; i++) {
5609 t = btf_type_by_id(src_btf, i);
5610 err = btf_add_type(&dist.pipe, t);
5614 /* Now add distilled types to split BTF that are not added to base. */
5615 err = btf_add_distilled_types(&dist);
5619 /* All split BTF ids will be shifted downwards since there are less base
5620 * BTF ids in distilled base BTF.
5622 dist.diff_id = dist.split_start_id - btf__type_cnt(new_base);
5624 n = btf__type_cnt(new_split);
5625 /* Now update base/split BTF ids. */
5626 for (i = 1; i < n; i++) {
5627 err = btf_update_distilled_type_ids(&dist, i);
5633 hashmap__free(dist.pipe.str_off_map);
5635 btf__free(new_split);
5636 btf__free(new_base);
5637 return libbpf_err(err);
5639 *new_base_btf = new_base;
5640 *new_split_btf = new_split;
5645 const struct btf_header *btf_header(const struct btf *btf)
5650 void btf_set_base_btf(struct btf *btf, const struct btf *base_btf)
5652 btf->base_btf = (struct btf *)base_btf;
5653 btf->start_id = btf__type_cnt(base_btf);
5654 btf->start_str_off = base_btf->hdr->str_len;
5657 int btf__relocate(struct btf *btf, const struct btf *base_btf)
5659 int err = btf_relocate(btf, base_btf, NULL);
5662 btf->owns_base = false;
5663 return libbpf_err(err);