1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 /* Copyright (c) 2018 Facebook */
12 #include <sys/utsname.h>
13 #include <sys/param.h>
15 #include <linux/kernel.h>
16 #include <linux/err.h>
17 #include <linux/btf.h>
22 #include "libbpf_internal.h"
26 #define BTF_MAX_NR_TYPES 0x7fffffffU
27 #define BTF_MAX_STR_OFFSET 0x7fffffffU
29 static struct btf_type btf_void;
32 /* raw BTF data in native endianness */
34 /* raw BTF data in non-native endianness */
35 void *raw_data_swapped;
37 /* whether target endianness differs from the native one */
41 * When BTF is loaded from an ELF or raw memory it is stored
42 * in a contiguous memory block. The hdr, type_data, and, strs_data
43 * point inside that memory region to their respective parts of BTF
46 * +--------------------------------+
47 * | Header | Types | Strings |
48 * +--------------------------------+
53 * strs_data------------+
55 * If BTF data is later modified, e.g., due to types added or
56 * removed, BTF deduplication performed, etc, this contiguous
57 * representation is broken up into three independently allocated
58 * memory regions to be able to modify them independently.
59 * raw_data is nulled out at that point, but can be later allocated
60 * and cached again if user calls btf__raw_data(), at which point
61 * raw_data will contain a contiguous copy of header, types, and
64 * +----------+ +---------+ +-----------+
65 * | Header | | Types | | Strings |
66 * +----------+ +---------+ +-----------+
71 * strset__data(strs_set)-----+
73 * +----------+---------+-----------+
74 * | Header | Types | Strings |
75 * raw_data----->+----------+---------+-----------+
77 struct btf_header *hdr;
80 size_t types_data_cap; /* used size stored in hdr->type_len */
82 /* type ID to `struct btf_type *` lookup index
83 * type_offs[0] corresponds to the first non-VOID type:
84 * - for base BTF it's type [1];
85 * - for split BTF it's the first non-base BTF type.
89 /* number of types in this BTF instance:
90 * - doesn't include special [0] void type;
91 * - for split BTF counts number of types added on top of base BTF.
94 /* if not NULL, points to the base BTF on top of which the current
98 /* BTF type ID of the first type in this BTF instance:
99 * - for base BTF it's equal to 1;
100 * - for split BTF it's equal to biggest type ID of base BTF plus 1.
103 /* logical string offset of this BTF instance:
104 * - for base BTF it's equal to 0;
105 * - for split BTF it's equal to total size of base BTF's string section size.
109 /* only one of strs_data or strs_set can be non-NULL, depending on
110 * whether BTF is in a modifiable state (strs_set is used) or not
111 * (strs_data points inside raw_data)
114 /* a set of unique strings */
115 struct strset *strs_set;
116 /* whether strings are already deduplicated */
119 /* BTF object FD, if loaded into kernel */
122 /* Pointer size (in bytes) for a target architecture of this BTF */
126 static inline __u64 ptr_to_u64(const void *ptr)
128 return (__u64) (unsigned long) ptr;
131 /* Ensure given dynamically allocated memory region pointed to by *data* with
132 * capacity of *cap_cnt* elements each taking *elem_sz* bytes has enough
133 * memory to accommodate *add_cnt* new elements, assuming *cur_cnt* elements
134 * are already used. At most *max_cnt* elements can be ever allocated.
135 * If necessary, memory is reallocated and all existing data is copied over,
136 * new pointer to the memory region is stored at *data, new memory region
137 * capacity (in number of elements) is stored in *cap.
138 * On success, memory pointer to the beginning of unused memory is returned.
139 * On error, NULL is returned.
141 void *libbpf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz,
142 size_t cur_cnt, size_t max_cnt, size_t add_cnt)
147 if (cur_cnt + add_cnt <= *cap_cnt)
148 return *data + cur_cnt * elem_sz;
150 /* requested more than the set limit */
151 if (cur_cnt + add_cnt > max_cnt)
155 new_cnt += new_cnt / 4; /* expand by 25% */
156 if (new_cnt < 16) /* but at least 16 elements */
158 if (new_cnt > max_cnt) /* but not exceeding a set limit */
160 if (new_cnt < cur_cnt + add_cnt) /* also ensure we have enough memory */
161 new_cnt = cur_cnt + add_cnt;
163 new_data = libbpf_reallocarray(*data, new_cnt, elem_sz);
167 /* zero out newly allocated portion of memory */
168 memset(new_data + (*cap_cnt) * elem_sz, 0, (new_cnt - *cap_cnt) * elem_sz);
172 return new_data + cur_cnt * elem_sz;
175 /* Ensure given dynamically allocated memory region has enough allocated space
176 * to accommodate *need_cnt* elements of size *elem_sz* bytes each
178 int libbpf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt)
182 if (need_cnt <= *cap_cnt)
185 p = libbpf_add_mem(data, cap_cnt, elem_sz, *cap_cnt, SIZE_MAX, need_cnt - *cap_cnt);
192 static void *btf_add_type_offs_mem(struct btf *btf, size_t add_cnt)
194 return libbpf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
195 btf->nr_types, BTF_MAX_NR_TYPES, add_cnt);
198 static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
202 p = btf_add_type_offs_mem(btf, 1);
210 static void btf_bswap_hdr(struct btf_header *h)
212 h->magic = bswap_16(h->magic);
213 h->hdr_len = bswap_32(h->hdr_len);
214 h->type_off = bswap_32(h->type_off);
215 h->type_len = bswap_32(h->type_len);
216 h->str_off = bswap_32(h->str_off);
217 h->str_len = bswap_32(h->str_len);
220 static int btf_parse_hdr(struct btf *btf)
222 struct btf_header *hdr = btf->hdr;
225 if (btf->raw_size < sizeof(struct btf_header)) {
226 pr_debug("BTF header not found\n");
230 if (hdr->magic == bswap_16(BTF_MAGIC)) {
231 btf->swapped_endian = true;
232 if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
233 pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
234 bswap_32(hdr->hdr_len));
238 } else if (hdr->magic != BTF_MAGIC) {
239 pr_debug("Invalid BTF magic: %x\n", hdr->magic);
243 if (btf->raw_size < hdr->hdr_len) {
244 pr_debug("BTF header len %u larger than data size %u\n",
245 hdr->hdr_len, btf->raw_size);
249 meta_left = btf->raw_size - hdr->hdr_len;
250 if (meta_left < (long long)hdr->str_off + hdr->str_len) {
251 pr_debug("Invalid BTF total size: %u\n", btf->raw_size);
255 if ((long long)hdr->type_off + hdr->type_len > hdr->str_off) {
256 pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
257 hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
261 if (hdr->type_off % 4) {
262 pr_debug("BTF type section is not aligned to 4 bytes\n");
269 static int btf_parse_str_sec(struct btf *btf)
271 const struct btf_header *hdr = btf->hdr;
272 const char *start = btf->strs_data;
273 const char *end = start + btf->hdr->str_len;
275 if (btf->base_btf && hdr->str_len == 0)
277 if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET || end[-1]) {
278 pr_debug("Invalid BTF string section\n");
281 if (!btf->base_btf && start[0]) {
282 pr_debug("Invalid BTF string section\n");
288 static int btf_type_size(const struct btf_type *t)
290 const int base_size = sizeof(struct btf_type);
291 __u16 vlen = btf_vlen(t);
293 switch (btf_kind(t)) {
296 case BTF_KIND_VOLATILE:
297 case BTF_KIND_RESTRICT:
299 case BTF_KIND_TYPEDEF:
302 case BTF_KIND_TYPE_TAG:
305 return base_size + sizeof(__u32);
307 return base_size + vlen * sizeof(struct btf_enum);
309 return base_size + sizeof(struct btf_array);
310 case BTF_KIND_STRUCT:
312 return base_size + vlen * sizeof(struct btf_member);
313 case BTF_KIND_FUNC_PROTO:
314 return base_size + vlen * sizeof(struct btf_param);
316 return base_size + sizeof(struct btf_var);
317 case BTF_KIND_DATASEC:
318 return base_size + vlen * sizeof(struct btf_var_secinfo);
319 case BTF_KIND_DECL_TAG:
320 return base_size + sizeof(struct btf_decl_tag);
322 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
327 static void btf_bswap_type_base(struct btf_type *t)
329 t->name_off = bswap_32(t->name_off);
330 t->info = bswap_32(t->info);
331 t->type = bswap_32(t->type);
334 static int btf_bswap_type_rest(struct btf_type *t)
336 struct btf_var_secinfo *v;
337 struct btf_member *m;
341 __u16 vlen = btf_vlen(t);
344 switch (btf_kind(t)) {
347 case BTF_KIND_VOLATILE:
348 case BTF_KIND_RESTRICT:
350 case BTF_KIND_TYPEDEF:
353 case BTF_KIND_TYPE_TAG:
356 *(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
359 for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
360 e->name_off = bswap_32(e->name_off);
361 e->val = bswap_32(e->val);
366 a->type = bswap_32(a->type);
367 a->index_type = bswap_32(a->index_type);
368 a->nelems = bswap_32(a->nelems);
370 case BTF_KIND_STRUCT:
372 for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
373 m->name_off = bswap_32(m->name_off);
374 m->type = bswap_32(m->type);
375 m->offset = bswap_32(m->offset);
378 case BTF_KIND_FUNC_PROTO:
379 for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
380 p->name_off = bswap_32(p->name_off);
381 p->type = bswap_32(p->type);
385 btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
387 case BTF_KIND_DATASEC:
388 for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
389 v->type = bswap_32(v->type);
390 v->offset = bswap_32(v->offset);
391 v->size = bswap_32(v->size);
394 case BTF_KIND_DECL_TAG:
395 btf_decl_tag(t)->component_idx = bswap_32(btf_decl_tag(t)->component_idx);
398 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
403 static int btf_parse_type_sec(struct btf *btf)
405 struct btf_header *hdr = btf->hdr;
406 void *next_type = btf->types_data;
407 void *end_type = next_type + hdr->type_len;
410 while (next_type + sizeof(struct btf_type) <= end_type) {
411 if (btf->swapped_endian)
412 btf_bswap_type_base(next_type);
414 type_size = btf_type_size(next_type);
417 if (next_type + type_size > end_type) {
418 pr_warn("BTF type [%d] is malformed\n", btf->start_id + btf->nr_types);
422 if (btf->swapped_endian && btf_bswap_type_rest(next_type))
425 err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
429 next_type += type_size;
433 if (next_type != end_type) {
434 pr_warn("BTF types data is malformed\n");
441 __u32 btf__get_nr_types(const struct btf *btf)
443 return btf->start_id + btf->nr_types - 1;
446 __u32 btf__type_cnt(const struct btf *btf)
448 return btf->start_id + btf->nr_types;
451 const struct btf *btf__base_btf(const struct btf *btf)
453 return btf->base_btf;
456 /* internal helper returning non-const pointer to a type */
457 struct btf_type *btf_type_by_id(const struct btf *btf, __u32 type_id)
461 if (type_id < btf->start_id)
462 return btf_type_by_id(btf->base_btf, type_id);
463 return btf->types_data + btf->type_offs[type_id - btf->start_id];
466 const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
468 if (type_id >= btf->start_id + btf->nr_types)
469 return errno = EINVAL, NULL;
470 return btf_type_by_id((struct btf *)btf, type_id);
473 static int determine_ptr_size(const struct btf *btf)
475 static const char * const long_aliases[] = {
488 const struct btf_type *t;
492 if (btf->base_btf && btf->base_btf->ptr_sz > 0)
493 return btf->base_btf->ptr_sz;
495 n = btf__type_cnt(btf);
496 for (i = 1; i < n; i++) {
497 t = btf__type_by_id(btf, i);
501 if (t->size != 4 && t->size != 8)
504 name = btf__name_by_offset(btf, t->name_off);
508 for (j = 0; j < ARRAY_SIZE(long_aliases); j++) {
509 if (strcmp(name, long_aliases[j]) == 0)
517 static size_t btf_ptr_sz(const struct btf *btf)
520 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
521 return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
524 /* Return pointer size this BTF instance assumes. The size is heuristically
525 * determined by looking for 'long' or 'unsigned long' integer type and
526 * recording its size in bytes. If BTF type information doesn't have any such
527 * type, this function returns 0. In the latter case, native architecture's
528 * pointer size is assumed, so will be either 4 or 8, depending on
529 * architecture that libbpf was compiled for. It's possible to override
530 * guessed value by using btf__set_pointer_size() API.
532 size_t btf__pointer_size(const struct btf *btf)
535 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
538 /* not enough BTF type info to guess */
544 /* Override or set pointer size in bytes. Only values of 4 and 8 are
547 int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
549 if (ptr_sz != 4 && ptr_sz != 8)
550 return libbpf_err(-EINVAL);
551 btf->ptr_sz = ptr_sz;
555 static bool is_host_big_endian(void)
557 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
559 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
562 # error "Unrecognized __BYTE_ORDER__"
566 enum btf_endianness btf__endianness(const struct btf *btf)
568 if (is_host_big_endian())
569 return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
571 return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
574 int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
576 if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
577 return libbpf_err(-EINVAL);
579 btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
580 if (!btf->swapped_endian) {
581 free(btf->raw_data_swapped);
582 btf->raw_data_swapped = NULL;
587 static bool btf_type_is_void(const struct btf_type *t)
589 return t == &btf_void || btf_is_fwd(t);
592 static bool btf_type_is_void_or_null(const struct btf_type *t)
594 return !t || btf_type_is_void(t);
597 #define MAX_RESOLVE_DEPTH 32
599 __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
601 const struct btf_array *array;
602 const struct btf_type *t;
607 t = btf__type_by_id(btf, type_id);
608 for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t); i++) {
609 switch (btf_kind(t)) {
611 case BTF_KIND_STRUCT:
614 case BTF_KIND_DATASEC:
619 size = btf_ptr_sz(btf);
621 case BTF_KIND_TYPEDEF:
622 case BTF_KIND_VOLATILE:
624 case BTF_KIND_RESTRICT:
626 case BTF_KIND_DECL_TAG:
627 case BTF_KIND_TYPE_TAG:
631 array = btf_array(t);
632 if (nelems && array->nelems > UINT32_MAX / nelems)
633 return libbpf_err(-E2BIG);
634 nelems *= array->nelems;
635 type_id = array->type;
638 return libbpf_err(-EINVAL);
641 t = btf__type_by_id(btf, type_id);
646 return libbpf_err(-EINVAL);
647 if (nelems && size > UINT32_MAX / nelems)
648 return libbpf_err(-E2BIG);
650 return nelems * size;
653 int btf__align_of(const struct btf *btf, __u32 id)
655 const struct btf_type *t = btf__type_by_id(btf, id);
656 __u16 kind = btf_kind(t);
662 return min(btf_ptr_sz(btf), (size_t)t->size);
664 return btf_ptr_sz(btf);
665 case BTF_KIND_TYPEDEF:
666 case BTF_KIND_VOLATILE:
668 case BTF_KIND_RESTRICT:
669 case BTF_KIND_TYPE_TAG:
670 return btf__align_of(btf, t->type);
672 return btf__align_of(btf, btf_array(t)->type);
673 case BTF_KIND_STRUCT:
674 case BTF_KIND_UNION: {
675 const struct btf_member *m = btf_members(t);
676 __u16 vlen = btf_vlen(t);
677 int i, max_align = 1, align;
679 for (i = 0; i < vlen; i++, m++) {
680 align = btf__align_of(btf, m->type);
682 return libbpf_err(align);
683 max_align = max(max_align, align);
689 pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
690 return errno = EINVAL, 0;
694 int btf__resolve_type(const struct btf *btf, __u32 type_id)
696 const struct btf_type *t;
699 t = btf__type_by_id(btf, type_id);
700 while (depth < MAX_RESOLVE_DEPTH &&
701 !btf_type_is_void_or_null(t) &&
702 (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
704 t = btf__type_by_id(btf, type_id);
708 if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
709 return libbpf_err(-EINVAL);
714 __s32 btf__find_by_name(const struct btf *btf, const char *type_name)
716 __u32 i, nr_types = btf__type_cnt(btf);
718 if (!strcmp(type_name, "void"))
721 for (i = 1; i < nr_types; i++) {
722 const struct btf_type *t = btf__type_by_id(btf, i);
723 const char *name = btf__name_by_offset(btf, t->name_off);
725 if (name && !strcmp(type_name, name))
729 return libbpf_err(-ENOENT);
732 static __s32 btf_find_by_name_kind(const struct btf *btf, int start_id,
733 const char *type_name, __u32 kind)
735 __u32 i, nr_types = btf__type_cnt(btf);
737 if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
740 for (i = start_id; i < nr_types; i++) {
741 const struct btf_type *t = btf__type_by_id(btf, i);
744 if (btf_kind(t) != kind)
746 name = btf__name_by_offset(btf, t->name_off);
747 if (name && !strcmp(type_name, name))
751 return libbpf_err(-ENOENT);
754 __s32 btf__find_by_name_kind_own(const struct btf *btf, const char *type_name,
757 return btf_find_by_name_kind(btf, btf->start_id, type_name, kind);
760 __s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
763 return btf_find_by_name_kind(btf, 1, type_name, kind);
766 static bool btf_is_modifiable(const struct btf *btf)
768 return (void *)btf->hdr != btf->raw_data;
771 void btf__free(struct btf *btf)
773 if (IS_ERR_OR_NULL(btf))
779 if (btf_is_modifiable(btf)) {
780 /* if BTF was modified after loading, it will have a split
781 * in-memory representation for header, types, and strings
782 * sections, so we need to free all of them individually. It
783 * might still have a cached contiguous raw data present,
784 * which will be unconditionally freed below.
787 free(btf->types_data);
788 strset__free(btf->strs_set);
791 free(btf->raw_data_swapped);
792 free(btf->type_offs);
796 static struct btf *btf_new_empty(struct btf *base_btf)
800 btf = calloc(1, sizeof(*btf));
802 return ERR_PTR(-ENOMEM);
806 btf->start_str_off = 0;
808 btf->ptr_sz = sizeof(void *);
809 btf->swapped_endian = false;
812 btf->base_btf = base_btf;
813 btf->start_id = btf__type_cnt(base_btf);
814 btf->start_str_off = base_btf->hdr->str_len;
817 /* +1 for empty string at offset 0 */
818 btf->raw_size = sizeof(struct btf_header) + (base_btf ? 0 : 1);
819 btf->raw_data = calloc(1, btf->raw_size);
820 if (!btf->raw_data) {
822 return ERR_PTR(-ENOMEM);
825 btf->hdr = btf->raw_data;
826 btf->hdr->hdr_len = sizeof(struct btf_header);
827 btf->hdr->magic = BTF_MAGIC;
828 btf->hdr->version = BTF_VERSION;
830 btf->types_data = btf->raw_data + btf->hdr->hdr_len;
831 btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
832 btf->hdr->str_len = base_btf ? 0 : 1; /* empty string at offset 0 */
837 struct btf *btf__new_empty(void)
839 return libbpf_ptr(btf_new_empty(NULL));
842 struct btf *btf__new_empty_split(struct btf *base_btf)
844 return libbpf_ptr(btf_new_empty(base_btf));
847 static struct btf *btf_new(const void *data, __u32 size, struct btf *base_btf)
852 btf = calloc(1, sizeof(struct btf));
854 return ERR_PTR(-ENOMEM);
858 btf->start_str_off = 0;
862 btf->base_btf = base_btf;
863 btf->start_id = btf__type_cnt(base_btf);
864 btf->start_str_off = base_btf->hdr->str_len;
867 btf->raw_data = malloc(size);
868 if (!btf->raw_data) {
872 memcpy(btf->raw_data, data, size);
873 btf->raw_size = size;
875 btf->hdr = btf->raw_data;
876 err = btf_parse_hdr(btf);
880 btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
881 btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
883 err = btf_parse_str_sec(btf);
884 err = err ?: btf_parse_type_sec(btf);
897 struct btf *btf__new(const void *data, __u32 size)
899 return libbpf_ptr(btf_new(data, size, NULL));
902 static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,
903 struct btf_ext **btf_ext)
905 Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
906 int err = 0, fd = -1, idx = 0;
907 struct btf *btf = NULL;
913 if (elf_version(EV_CURRENT) == EV_NONE) {
914 pr_warn("failed to init libelf for %s\n", path);
915 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
918 fd = open(path, O_RDONLY | O_CLOEXEC);
921 pr_warn("failed to open %s: %s\n", path, strerror(errno));
925 err = -LIBBPF_ERRNO__FORMAT;
927 elf = elf_begin(fd, ELF_C_READ, NULL);
929 pr_warn("failed to open %s as ELF file\n", path);
932 if (!gelf_getehdr(elf, &ehdr)) {
933 pr_warn("failed to get EHDR from %s\n", path);
937 if (elf_getshdrstrndx(elf, &shstrndx)) {
938 pr_warn("failed to get section names section index for %s\n",
943 if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
944 pr_warn("failed to get e_shstrndx from %s\n", path);
948 while ((scn = elf_nextscn(elf, scn)) != NULL) {
953 if (gelf_getshdr(scn, &sh) != &sh) {
954 pr_warn("failed to get section(%d) header from %s\n",
958 name = elf_strptr(elf, shstrndx, sh.sh_name);
960 pr_warn("failed to get section(%d) name from %s\n",
964 if (strcmp(name, BTF_ELF_SEC) == 0) {
965 btf_data = elf_getdata(scn, 0);
967 pr_warn("failed to get section(%d, %s) data from %s\n",
972 } else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
973 btf_ext_data = elf_getdata(scn, 0);
975 pr_warn("failed to get section(%d, %s) data from %s\n",
989 btf = btf_new(btf_data->d_buf, btf_data->d_size, base_btf);
990 err = libbpf_get_error(btf);
994 switch (gelf_getclass(elf)) {
996 btf__set_pointer_size(btf, 4);
999 btf__set_pointer_size(btf, 8);
1002 pr_warn("failed to get ELF class (bitness) for %s\n", path);
1006 if (btf_ext && btf_ext_data) {
1007 *btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
1008 err = libbpf_get_error(*btf_ext);
1011 } else if (btf_ext) {
1023 btf_ext__free(*btf_ext);
1026 return ERR_PTR(err);
1029 struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
1031 return libbpf_ptr(btf_parse_elf(path, NULL, btf_ext));
1034 struct btf *btf__parse_elf_split(const char *path, struct btf *base_btf)
1036 return libbpf_ptr(btf_parse_elf(path, base_btf, NULL));
1039 static struct btf *btf_parse_raw(const char *path, struct btf *base_btf)
1041 struct btf *btf = NULL;
1048 f = fopen(path, "rb");
1054 /* check BTF magic */
1055 if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
1059 if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
1060 /* definitely not a raw BTF */
1066 if (fseek(f, 0, SEEK_END)) {
1075 /* rewind to the start */
1076 if (fseek(f, 0, SEEK_SET)) {
1081 /* pre-alloc memory and read all of BTF data */
1087 if (fread(data, 1, sz, f) < sz) {
1092 /* finally parse BTF data */
1093 btf = btf_new(data, sz, base_btf);
1099 return err ? ERR_PTR(err) : btf;
1102 struct btf *btf__parse_raw(const char *path)
1104 return libbpf_ptr(btf_parse_raw(path, NULL));
1107 struct btf *btf__parse_raw_split(const char *path, struct btf *base_btf)
1109 return libbpf_ptr(btf_parse_raw(path, base_btf));
1112 static struct btf *btf_parse(const char *path, struct btf *base_btf, struct btf_ext **btf_ext)
1120 btf = btf_parse_raw(path, base_btf);
1121 err = libbpf_get_error(btf);
1125 return ERR_PTR(err);
1126 return btf_parse_elf(path, base_btf, btf_ext);
1129 struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
1131 return libbpf_ptr(btf_parse(path, NULL, btf_ext));
1134 struct btf *btf__parse_split(const char *path, struct btf *base_btf)
1136 return libbpf_ptr(btf_parse(path, base_btf, NULL));
1139 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1141 int btf_load_into_kernel(struct btf *btf, char *log_buf, size_t log_sz, __u32 log_level)
1143 LIBBPF_OPTS(bpf_btf_load_opts, opts);
1144 __u32 buf_sz = 0, raw_size;
1145 char *buf = NULL, *tmp;
1150 return libbpf_err(-EEXIST);
1151 if (log_sz && !log_buf)
1152 return libbpf_err(-EINVAL);
1154 /* cache native raw data representation */
1155 raw_data = btf_get_raw_data(btf, &raw_size, false);
1160 btf->raw_size = raw_size;
1161 btf->raw_data = raw_data;
1164 /* if log_level is 0, we won't provide log_buf/log_size to the kernel,
1165 * initially. Only if BTF loading fails, we bump log_level to 1 and
1166 * retry, using either auto-allocated or custom log_buf. This way
1167 * non-NULL custom log_buf provides a buffer just in case, but hopes
1168 * for successful load and no need for log_buf.
1171 /* if caller didn't provide custom log_buf, we'll keep
1172 * allocating our own progressively bigger buffers for BTF
1176 buf_sz = max((__u32)BPF_LOG_BUF_SIZE, buf_sz * 2);
1177 tmp = realloc(buf, buf_sz);
1186 opts.log_buf = log_buf ? log_buf : buf;
1187 opts.log_size = log_buf ? log_sz : buf_sz;
1188 opts.log_level = log_level;
1191 btf->fd = bpf_btf_load(raw_data, raw_size, &opts);
1193 /* time to turn on verbose mode and try again */
1194 if (log_level == 0) {
1198 /* only retry if caller didn't provide custom log_buf, but
1199 * make sure we can never overflow buf_sz
1201 if (!log_buf && errno == ENOSPC && buf_sz <= UINT_MAX / 2)
1205 pr_warn("BTF loading error: %d\n", err);
1206 /* don't print out contents of custom log_buf */
1207 if (!log_buf && buf[0])
1208 pr_warn("-- BEGIN BTF LOAD LOG ---\n%s\n-- END BTF LOAD LOG --\n", buf);
1213 return libbpf_err(err);
1216 int btf__load_into_kernel(struct btf *btf)
1218 return btf_load_into_kernel(btf, NULL, 0, 0);
1221 int btf__load(struct btf *) __attribute__((alias("btf__load_into_kernel")));
1223 int btf__fd(const struct btf *btf)
1228 void btf__set_fd(struct btf *btf, int fd)
1233 static const void *btf_strs_data(const struct btf *btf)
1235 return btf->strs_data ? btf->strs_data : strset__data(btf->strs_set);
1238 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1240 struct btf_header *hdr = btf->hdr;
1246 data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1248 *size = btf->raw_size;
1252 data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1253 data = calloc(1, data_sz);
1258 memcpy(p, hdr, hdr->hdr_len);
1263 memcpy(p, btf->types_data, hdr->type_len);
1265 for (i = 0; i < btf->nr_types; i++) {
1266 t = p + btf->type_offs[i];
1267 /* btf_bswap_type_rest() relies on native t->info, so
1268 * we swap base type info after we swapped all the
1269 * additional information
1271 if (btf_bswap_type_rest(t))
1273 btf_bswap_type_base(t);
1278 memcpy(p, btf_strs_data(btf), hdr->str_len);
1288 const void *btf__raw_data(const struct btf *btf_ro, __u32 *size)
1290 struct btf *btf = (struct btf *)btf_ro;
1294 data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1296 return errno = ENOMEM, NULL;
1298 btf->raw_size = data_sz;
1299 if (btf->swapped_endian)
1300 btf->raw_data_swapped = data;
1302 btf->raw_data = data;
1307 __attribute__((alias("btf__raw_data")))
1308 const void *btf__get_raw_data(const struct btf *btf, __u32 *size);
1310 const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
1312 if (offset < btf->start_str_off)
1313 return btf__str_by_offset(btf->base_btf, offset);
1314 else if (offset - btf->start_str_off < btf->hdr->str_len)
1315 return btf_strs_data(btf) + (offset - btf->start_str_off);
1317 return errno = EINVAL, NULL;
1320 const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1322 return btf__str_by_offset(btf, offset);
1325 struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf)
1327 struct bpf_btf_info btf_info;
1328 __u32 len = sizeof(btf_info);
1334 /* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
1335 * let's start with a sane default - 4KiB here - and resize it only if
1336 * bpf_obj_get_info_by_fd() needs a bigger buffer.
1339 ptr = malloc(last_size);
1341 return ERR_PTR(-ENOMEM);
1343 memset(&btf_info, 0, sizeof(btf_info));
1344 btf_info.btf = ptr_to_u64(ptr);
1345 btf_info.btf_size = last_size;
1346 err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1348 if (!err && btf_info.btf_size > last_size) {
1351 last_size = btf_info.btf_size;
1352 temp_ptr = realloc(ptr, last_size);
1354 btf = ERR_PTR(-ENOMEM);
1359 len = sizeof(btf_info);
1360 memset(&btf_info, 0, sizeof(btf_info));
1361 btf_info.btf = ptr_to_u64(ptr);
1362 btf_info.btf_size = last_size;
1364 err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1367 if (err || btf_info.btf_size > last_size) {
1368 btf = err ? ERR_PTR(-errno) : ERR_PTR(-E2BIG);
1372 btf = btf_new(ptr, btf_info.btf_size, base_btf);
1379 struct btf *btf__load_from_kernel_by_id_split(__u32 id, struct btf *base_btf)
1384 btf_fd = bpf_btf_get_fd_by_id(id);
1386 return libbpf_err_ptr(-errno);
1388 btf = btf_get_from_fd(btf_fd, base_btf);
1391 return libbpf_ptr(btf);
1394 struct btf *btf__load_from_kernel_by_id(__u32 id)
1396 return btf__load_from_kernel_by_id_split(id, NULL);
1399 int btf__get_from_id(__u32 id, struct btf **btf)
1405 res = btf__load_from_kernel_by_id(id);
1406 err = libbpf_get_error(res);
1409 return libbpf_err(err);
1415 int btf__get_map_kv_tids(const struct btf *btf, const char *map_name,
1416 __u32 expected_key_size, __u32 expected_value_size,
1417 __u32 *key_type_id, __u32 *value_type_id)
1419 const struct btf_type *container_type;
1420 const struct btf_member *key, *value;
1421 const size_t max_name = 256;
1422 char container_name[max_name];
1423 __s64 key_size, value_size;
1426 if (snprintf(container_name, max_name, "____btf_map_%s", map_name) == max_name) {
1427 pr_warn("map:%s length of '____btf_map_%s' is too long\n",
1428 map_name, map_name);
1429 return libbpf_err(-EINVAL);
1432 container_id = btf__find_by_name(btf, container_name);
1433 if (container_id < 0) {
1434 pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
1435 map_name, container_name);
1436 return libbpf_err(container_id);
1439 container_type = btf__type_by_id(btf, container_id);
1440 if (!container_type) {
1441 pr_warn("map:%s cannot find BTF type for container_id:%u\n",
1442 map_name, container_id);
1443 return libbpf_err(-EINVAL);
1446 if (!btf_is_struct(container_type) || btf_vlen(container_type) < 2) {
1447 pr_warn("map:%s container_name:%s is an invalid container struct\n",
1448 map_name, container_name);
1449 return libbpf_err(-EINVAL);
1452 key = btf_members(container_type);
1455 key_size = btf__resolve_size(btf, key->type);
1457 pr_warn("map:%s invalid BTF key_type_size\n", map_name);
1458 return libbpf_err(key_size);
1461 if (expected_key_size != key_size) {
1462 pr_warn("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
1463 map_name, (__u32)key_size, expected_key_size);
1464 return libbpf_err(-EINVAL);
1467 value_size = btf__resolve_size(btf, value->type);
1468 if (value_size < 0) {
1469 pr_warn("map:%s invalid BTF value_type_size\n", map_name);
1470 return libbpf_err(value_size);
1473 if (expected_value_size != value_size) {
1474 pr_warn("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1475 map_name, (__u32)value_size, expected_value_size);
1476 return libbpf_err(-EINVAL);
1479 *key_type_id = key->type;
1480 *value_type_id = value->type;
1485 static void btf_invalidate_raw_data(struct btf *btf)
1487 if (btf->raw_data) {
1488 free(btf->raw_data);
1489 btf->raw_data = NULL;
1491 if (btf->raw_data_swapped) {
1492 free(btf->raw_data_swapped);
1493 btf->raw_data_swapped = NULL;
1497 /* Ensure BTF is ready to be modified (by splitting into a three memory
1498 * regions for header, types, and strings). Also invalidate cached
1501 static int btf_ensure_modifiable(struct btf *btf)
1504 struct strset *set = NULL;
1507 if (btf_is_modifiable(btf)) {
1508 /* any BTF modification invalidates raw_data */
1509 btf_invalidate_raw_data(btf);
1513 /* split raw data into three memory regions */
1514 hdr = malloc(btf->hdr->hdr_len);
1515 types = malloc(btf->hdr->type_len);
1519 memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1520 memcpy(types, btf->types_data, btf->hdr->type_len);
1522 /* build lookup index for all strings */
1523 set = strset__new(BTF_MAX_STR_OFFSET, btf->strs_data, btf->hdr->str_len);
1529 /* only when everything was successful, update internal state */
1531 btf->types_data = types;
1532 btf->types_data_cap = btf->hdr->type_len;
1533 btf->strs_data = NULL;
1534 btf->strs_set = set;
1535 /* if BTF was created from scratch, all strings are guaranteed to be
1536 * unique and deduplicated
1538 if (btf->hdr->str_len == 0)
1539 btf->strs_deduped = true;
1540 if (!btf->base_btf && btf->hdr->str_len == 1)
1541 btf->strs_deduped = true;
1543 /* invalidate raw_data representation */
1544 btf_invalidate_raw_data(btf);
1555 /* Find an offset in BTF string section that corresponds to a given string *s*.
1557 * - >0 offset into string section, if string is found;
1558 * - -ENOENT, if string is not in the string section;
1559 * - <0, on any other error.
1561 int btf__find_str(struct btf *btf, const char *s)
1565 if (btf->base_btf) {
1566 off = btf__find_str(btf->base_btf, s);
1571 /* BTF needs to be in a modifiable state to build string lookup index */
1572 if (btf_ensure_modifiable(btf))
1573 return libbpf_err(-ENOMEM);
1575 off = strset__find_str(btf->strs_set, s);
1577 return libbpf_err(off);
1579 return btf->start_str_off + off;
1582 /* Add a string s to the BTF string section.
1584 * - > 0 offset into string section, on success;
1587 int btf__add_str(struct btf *btf, const char *s)
1591 if (btf->base_btf) {
1592 off = btf__find_str(btf->base_btf, s);
1597 if (btf_ensure_modifiable(btf))
1598 return libbpf_err(-ENOMEM);
1600 off = strset__add_str(btf->strs_set, s);
1602 return libbpf_err(off);
1604 btf->hdr->str_len = strset__data_size(btf->strs_set);
1606 return btf->start_str_off + off;
1609 static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1611 return libbpf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1612 btf->hdr->type_len, UINT_MAX, add_sz);
1615 static void btf_type_inc_vlen(struct btf_type *t)
1617 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1620 static int btf_commit_type(struct btf *btf, int data_sz)
1624 err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1626 return libbpf_err(err);
1628 btf->hdr->type_len += data_sz;
1629 btf->hdr->str_off += data_sz;
1631 return btf->start_id + btf->nr_types - 1;
1635 const struct btf *src;
1637 struct hashmap *str_off_map; /* map string offsets from src to dst */
1640 static int btf_rewrite_str(__u32 *str_off, void *ctx)
1642 struct btf_pipe *p = ctx;
1646 if (!*str_off) /* nothing to do for empty strings */
1649 if (p->str_off_map &&
1650 hashmap__find(p->str_off_map, (void *)(long)*str_off, &mapped_off)) {
1651 *str_off = (__u32)(long)mapped_off;
1655 off = btf__add_str(p->dst, btf__str_by_offset(p->src, *str_off));
1659 /* Remember string mapping from src to dst. It avoids
1660 * performing expensive string comparisons.
1662 if (p->str_off_map) {
1663 err = hashmap__append(p->str_off_map, (void *)(long)*str_off, (void *)(long)off);
1672 int btf__add_type(struct btf *btf, const struct btf *src_btf, const struct btf_type *src_type)
1674 struct btf_pipe p = { .src = src_btf, .dst = btf };
1678 sz = btf_type_size(src_type);
1680 return libbpf_err(sz);
1682 /* deconstruct BTF, if necessary, and invalidate raw_data */
1683 if (btf_ensure_modifiable(btf))
1684 return libbpf_err(-ENOMEM);
1686 t = btf_add_type_mem(btf, sz);
1688 return libbpf_err(-ENOMEM);
1690 memcpy(t, src_type, sz);
1692 err = btf_type_visit_str_offs(t, btf_rewrite_str, &p);
1694 return libbpf_err(err);
1696 return btf_commit_type(btf, sz);
1699 static int btf_rewrite_type_ids(__u32 *type_id, void *ctx)
1701 struct btf *btf = ctx;
1703 if (!*type_id) /* nothing to do for VOID references */
1706 /* we haven't updated btf's type count yet, so
1707 * btf->start_id + btf->nr_types - 1 is the type ID offset we should
1708 * add to all newly added BTF types
1710 *type_id += btf->start_id + btf->nr_types - 1;
1714 static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx);
1715 static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx);
1717 int btf__add_btf(struct btf *btf, const struct btf *src_btf)
1719 struct btf_pipe p = { .src = src_btf, .dst = btf };
1720 int data_sz, sz, cnt, i, err, old_strs_len;
1724 /* appending split BTF isn't supported yet */
1725 if (src_btf->base_btf)
1726 return libbpf_err(-ENOTSUP);
1728 /* deconstruct BTF, if necessary, and invalidate raw_data */
1729 if (btf_ensure_modifiable(btf))
1730 return libbpf_err(-ENOMEM);
1732 /* remember original strings section size if we have to roll back
1733 * partial strings section changes
1735 old_strs_len = btf->hdr->str_len;
1737 data_sz = src_btf->hdr->type_len;
1738 cnt = btf__type_cnt(src_btf) - 1;
1740 /* pre-allocate enough memory for new types */
1741 t = btf_add_type_mem(btf, data_sz);
1743 return libbpf_err(-ENOMEM);
1745 /* pre-allocate enough memory for type offset index for new types */
1746 off = btf_add_type_offs_mem(btf, cnt);
1748 return libbpf_err(-ENOMEM);
1750 /* Map the string offsets from src_btf to the offsets from btf to improve performance */
1751 p.str_off_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
1752 if (IS_ERR(p.str_off_map))
1753 return libbpf_err(-ENOMEM);
1755 /* bulk copy types data for all types from src_btf */
1756 memcpy(t, src_btf->types_data, data_sz);
1758 for (i = 0; i < cnt; i++) {
1759 sz = btf_type_size(t);
1761 /* unlikely, has to be corrupted src_btf */
1766 /* fill out type ID to type offset mapping for lookups by type ID */
1767 *off = t - btf->types_data;
1769 /* add, dedup, and remap strings referenced by this BTF type */
1770 err = btf_type_visit_str_offs(t, btf_rewrite_str, &p);
1774 /* remap all type IDs referenced from this BTF type */
1775 err = btf_type_visit_type_ids(t, btf_rewrite_type_ids, btf);
1779 /* go to next type data and type offset index entry */
1784 /* Up until now any of the copied type data was effectively invisible,
1785 * so if we exited early before this point due to error, BTF would be
1786 * effectively unmodified. There would be extra internal memory
1787 * pre-allocated, but it would not be available for querying. But now
1788 * that we've copied and rewritten all the data successfully, we can
1789 * update type count and various internal offsets and sizes to
1790 * "commit" the changes and made them visible to the outside world.
1792 btf->hdr->type_len += data_sz;
1793 btf->hdr->str_off += data_sz;
1794 btf->nr_types += cnt;
1796 hashmap__free(p.str_off_map);
1798 /* return type ID of the first added BTF type */
1799 return btf->start_id + btf->nr_types - cnt;
1801 /* zero out preallocated memory as if it was just allocated with
1804 memset(btf->types_data + btf->hdr->type_len, 0, data_sz);
1805 memset(btf->strs_data + old_strs_len, 0, btf->hdr->str_len - old_strs_len);
1807 /* and now restore original strings section size; types data size
1808 * wasn't modified, so doesn't need restoring, see big comment above */
1809 btf->hdr->str_len = old_strs_len;
1811 hashmap__free(p.str_off_map);
1813 return libbpf_err(err);
1817 * Append new BTF_KIND_INT type with:
1818 * - *name* - non-empty, non-NULL type name;
1819 * - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1820 * - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1822 * - >0, type ID of newly added BTF type;
1825 int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
1830 /* non-empty name */
1831 if (!name || !name[0])
1832 return libbpf_err(-EINVAL);
1833 /* byte_sz must be power of 2 */
1834 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
1835 return libbpf_err(-EINVAL);
1836 if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
1837 return libbpf_err(-EINVAL);
1839 /* deconstruct BTF, if necessary, and invalidate raw_data */
1840 if (btf_ensure_modifiable(btf))
1841 return libbpf_err(-ENOMEM);
1843 sz = sizeof(struct btf_type) + sizeof(int);
1844 t = btf_add_type_mem(btf, sz);
1846 return libbpf_err(-ENOMEM);
1848 /* if something goes wrong later, we might end up with an extra string,
1849 * but that shouldn't be a problem, because BTF can't be constructed
1850 * completely anyway and will most probably be just discarded
1852 name_off = btf__add_str(btf, name);
1856 t->name_off = name_off;
1857 t->info = btf_type_info(BTF_KIND_INT, 0, 0);
1859 /* set INT info, we don't allow setting legacy bit offset/size */
1860 *(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
1862 return btf_commit_type(btf, sz);
1866 * Append new BTF_KIND_FLOAT type with:
1867 * - *name* - non-empty, non-NULL type name;
1868 * - *sz* - size of the type, in bytes;
1870 * - >0, type ID of newly added BTF type;
1873 int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
1878 /* non-empty name */
1879 if (!name || !name[0])
1880 return libbpf_err(-EINVAL);
1882 /* byte_sz must be one of the explicitly allowed values */
1883 if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
1885 return libbpf_err(-EINVAL);
1887 if (btf_ensure_modifiable(btf))
1888 return libbpf_err(-ENOMEM);
1890 sz = sizeof(struct btf_type);
1891 t = btf_add_type_mem(btf, sz);
1893 return libbpf_err(-ENOMEM);
1895 name_off = btf__add_str(btf, name);
1899 t->name_off = name_off;
1900 t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
1903 return btf_commit_type(btf, sz);
1906 /* it's completely legal to append BTF types with type IDs pointing forward to
1907 * types that haven't been appended yet, so we only make sure that id looks
1908 * sane, we can't guarantee that ID will always be valid
1910 static int validate_type_id(int id)
1912 if (id < 0 || id > BTF_MAX_NR_TYPES)
1917 /* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
1918 static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
1921 int sz, name_off = 0;
1923 if (validate_type_id(ref_type_id))
1924 return libbpf_err(-EINVAL);
1926 if (btf_ensure_modifiable(btf))
1927 return libbpf_err(-ENOMEM);
1929 sz = sizeof(struct btf_type);
1930 t = btf_add_type_mem(btf, sz);
1932 return libbpf_err(-ENOMEM);
1934 if (name && name[0]) {
1935 name_off = btf__add_str(btf, name);
1940 t->name_off = name_off;
1941 t->info = btf_type_info(kind, 0, 0);
1942 t->type = ref_type_id;
1944 return btf_commit_type(btf, sz);
1948 * Append new BTF_KIND_PTR type with:
1949 * - *ref_type_id* - referenced type ID, it might not exist yet;
1951 * - >0, type ID of newly added BTF type;
1954 int btf__add_ptr(struct btf *btf, int ref_type_id)
1956 return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
1960 * Append new BTF_KIND_ARRAY type with:
1961 * - *index_type_id* - type ID of the type describing array index;
1962 * - *elem_type_id* - type ID of the type describing array element;
1963 * - *nr_elems* - the size of the array;
1965 * - >0, type ID of newly added BTF type;
1968 int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
1971 struct btf_array *a;
1974 if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
1975 return libbpf_err(-EINVAL);
1977 if (btf_ensure_modifiable(btf))
1978 return libbpf_err(-ENOMEM);
1980 sz = sizeof(struct btf_type) + sizeof(struct btf_array);
1981 t = btf_add_type_mem(btf, sz);
1983 return libbpf_err(-ENOMEM);
1986 t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
1990 a->type = elem_type_id;
1991 a->index_type = index_type_id;
1992 a->nelems = nr_elems;
1994 return btf_commit_type(btf, sz);
1997 /* generic STRUCT/UNION append function */
1998 static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
2001 int sz, name_off = 0;
2003 if (btf_ensure_modifiable(btf))
2004 return libbpf_err(-ENOMEM);
2006 sz = sizeof(struct btf_type);
2007 t = btf_add_type_mem(btf, sz);
2009 return libbpf_err(-ENOMEM);
2011 if (name && name[0]) {
2012 name_off = btf__add_str(btf, name);
2017 /* start out with vlen=0 and no kflag; this will be adjusted when
2018 * adding each member
2020 t->name_off = name_off;
2021 t->info = btf_type_info(kind, 0, 0);
2024 return btf_commit_type(btf, sz);
2028 * Append new BTF_KIND_STRUCT type with:
2029 * - *name* - name of the struct, can be NULL or empty for anonymous structs;
2030 * - *byte_sz* - size of the struct, in bytes;
2032 * Struct initially has no fields in it. Fields can be added by
2033 * btf__add_field() right after btf__add_struct() succeeds.
2036 * - >0, type ID of newly added BTF type;
2039 int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
2041 return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
2045 * Append new BTF_KIND_UNION type with:
2046 * - *name* - name of the union, can be NULL or empty for anonymous union;
2047 * - *byte_sz* - size of the union, in bytes;
2049 * Union initially has no fields in it. Fields can be added by
2050 * btf__add_field() right after btf__add_union() succeeds. All fields
2051 * should have *bit_offset* of 0.
2054 * - >0, type ID of newly added BTF type;
2057 int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
2059 return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
2062 static struct btf_type *btf_last_type(struct btf *btf)
2064 return btf_type_by_id(btf, btf__type_cnt(btf) - 1);
2068 * Append new field for the current STRUCT/UNION type with:
2069 * - *name* - name of the field, can be NULL or empty for anonymous field;
2070 * - *type_id* - type ID for the type describing field type;
2071 * - *bit_offset* - bit offset of the start of the field within struct/union;
2072 * - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
2077 int btf__add_field(struct btf *btf, const char *name, int type_id,
2078 __u32 bit_offset, __u32 bit_size)
2081 struct btf_member *m;
2083 int sz, name_off = 0;
2085 /* last type should be union/struct */
2086 if (btf->nr_types == 0)
2087 return libbpf_err(-EINVAL);
2088 t = btf_last_type(btf);
2089 if (!btf_is_composite(t))
2090 return libbpf_err(-EINVAL);
2092 if (validate_type_id(type_id))
2093 return libbpf_err(-EINVAL);
2094 /* best-effort bit field offset/size enforcement */
2095 is_bitfield = bit_size || (bit_offset % 8 != 0);
2096 if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
2097 return libbpf_err(-EINVAL);
2099 /* only offset 0 is allowed for unions */
2100 if (btf_is_union(t) && bit_offset)
2101 return libbpf_err(-EINVAL);
2103 /* decompose and invalidate raw data */
2104 if (btf_ensure_modifiable(btf))
2105 return libbpf_err(-ENOMEM);
2107 sz = sizeof(struct btf_member);
2108 m = btf_add_type_mem(btf, sz);
2110 return libbpf_err(-ENOMEM);
2112 if (name && name[0]) {
2113 name_off = btf__add_str(btf, name);
2118 m->name_off = name_off;
2120 m->offset = bit_offset | (bit_size << 24);
2122 /* btf_add_type_mem can invalidate t pointer */
2123 t = btf_last_type(btf);
2124 /* update parent type's vlen and kflag */
2125 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
2127 btf->hdr->type_len += sz;
2128 btf->hdr->str_off += sz;
2133 * Append new BTF_KIND_ENUM type with:
2134 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
2135 * - *byte_sz* - size of the enum, in bytes.
2137 * Enum initially has no enum values in it (and corresponds to enum forward
2138 * declaration). Enumerator values can be added by btf__add_enum_value()
2139 * immediately after btf__add_enum() succeeds.
2142 * - >0, type ID of newly added BTF type;
2145 int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2148 int sz, name_off = 0;
2150 /* byte_sz must be power of 2 */
2151 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2152 return libbpf_err(-EINVAL);
2154 if (btf_ensure_modifiable(btf))
2155 return libbpf_err(-ENOMEM);
2157 sz = sizeof(struct btf_type);
2158 t = btf_add_type_mem(btf, sz);
2160 return libbpf_err(-ENOMEM);
2162 if (name && name[0]) {
2163 name_off = btf__add_str(btf, name);
2168 /* start out with vlen=0; it will be adjusted when adding enum values */
2169 t->name_off = name_off;
2170 t->info = btf_type_info(BTF_KIND_ENUM, 0, 0);
2173 return btf_commit_type(btf, sz);
2177 * Append new enum value for the current ENUM type with:
2178 * - *name* - name of the enumerator value, can't be NULL or empty;
2179 * - *value* - integer value corresponding to enum value *name*;
2184 int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2190 /* last type should be BTF_KIND_ENUM */
2191 if (btf->nr_types == 0)
2192 return libbpf_err(-EINVAL);
2193 t = btf_last_type(btf);
2194 if (!btf_is_enum(t))
2195 return libbpf_err(-EINVAL);
2197 /* non-empty name */
2198 if (!name || !name[0])
2199 return libbpf_err(-EINVAL);
2200 if (value < INT_MIN || value > UINT_MAX)
2201 return libbpf_err(-E2BIG);
2203 /* decompose and invalidate raw data */
2204 if (btf_ensure_modifiable(btf))
2205 return libbpf_err(-ENOMEM);
2207 sz = sizeof(struct btf_enum);
2208 v = btf_add_type_mem(btf, sz);
2210 return libbpf_err(-ENOMEM);
2212 name_off = btf__add_str(btf, name);
2216 v->name_off = name_off;
2219 /* update parent type's vlen */
2220 t = btf_last_type(btf);
2221 btf_type_inc_vlen(t);
2223 btf->hdr->type_len += sz;
2224 btf->hdr->str_off += sz;
2229 * Append new BTF_KIND_FWD type with:
2230 * - *name*, non-empty/non-NULL name;
2231 * - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2232 * BTF_FWD_UNION, or BTF_FWD_ENUM;
2234 * - >0, type ID of newly added BTF type;
2237 int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2239 if (!name || !name[0])
2240 return libbpf_err(-EINVAL);
2243 case BTF_FWD_STRUCT:
2244 case BTF_FWD_UNION: {
2248 id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
2251 t = btf_type_by_id(btf, id);
2252 t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2256 /* enum forward in BTF currently is just an enum with no enum
2257 * values; we also assume a standard 4-byte size for it
2259 return btf__add_enum(btf, name, sizeof(int));
2261 return libbpf_err(-EINVAL);
2266 * Append new BTF_KING_TYPEDEF type with:
2267 * - *name*, non-empty/non-NULL name;
2268 * - *ref_type_id* - referenced type ID, it might not exist yet;
2270 * - >0, type ID of newly added BTF type;
2273 int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2275 if (!name || !name[0])
2276 return libbpf_err(-EINVAL);
2278 return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2282 * Append new BTF_KIND_VOLATILE type with:
2283 * - *ref_type_id* - referenced type ID, it might not exist yet;
2285 * - >0, type ID of newly added BTF type;
2288 int btf__add_volatile(struct btf *btf, int ref_type_id)
2290 return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2294 * Append new BTF_KIND_CONST type with:
2295 * - *ref_type_id* - referenced type ID, it might not exist yet;
2297 * - >0, type ID of newly added BTF type;
2300 int btf__add_const(struct btf *btf, int ref_type_id)
2302 return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2306 * Append new BTF_KIND_RESTRICT type with:
2307 * - *ref_type_id* - referenced type ID, it might not exist yet;
2309 * - >0, type ID of newly added BTF type;
2312 int btf__add_restrict(struct btf *btf, int ref_type_id)
2314 return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2318 * Append new BTF_KIND_TYPE_TAG type with:
2319 * - *value*, non-empty/non-NULL tag value;
2320 * - *ref_type_id* - referenced type ID, it might not exist yet;
2322 * - >0, type ID of newly added BTF type;
2325 int btf__add_type_tag(struct btf *btf, const char *value, int ref_type_id)
2327 if (!value|| !value[0])
2328 return libbpf_err(-EINVAL);
2330 return btf_add_ref_kind(btf, BTF_KIND_TYPE_TAG, value, ref_type_id);
2334 * Append new BTF_KIND_FUNC type with:
2335 * - *name*, non-empty/non-NULL name;
2336 * - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2338 * - >0, type ID of newly added BTF type;
2341 int btf__add_func(struct btf *btf, const char *name,
2342 enum btf_func_linkage linkage, int proto_type_id)
2346 if (!name || !name[0])
2347 return libbpf_err(-EINVAL);
2348 if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2349 linkage != BTF_FUNC_EXTERN)
2350 return libbpf_err(-EINVAL);
2352 id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2354 struct btf_type *t = btf_type_by_id(btf, id);
2356 t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2358 return libbpf_err(id);
2362 * Append new BTF_KIND_FUNC_PROTO with:
2363 * - *ret_type_id* - type ID for return result of a function.
2365 * Function prototype initially has no arguments, but they can be added by
2366 * btf__add_func_param() one by one, immediately after
2367 * btf__add_func_proto() succeeded.
2370 * - >0, type ID of newly added BTF type;
2373 int btf__add_func_proto(struct btf *btf, int ret_type_id)
2378 if (validate_type_id(ret_type_id))
2379 return libbpf_err(-EINVAL);
2381 if (btf_ensure_modifiable(btf))
2382 return libbpf_err(-ENOMEM);
2384 sz = sizeof(struct btf_type);
2385 t = btf_add_type_mem(btf, sz);
2387 return libbpf_err(-ENOMEM);
2389 /* start out with vlen=0; this will be adjusted when adding enum
2390 * values, if necessary
2393 t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2394 t->type = ret_type_id;
2396 return btf_commit_type(btf, sz);
2400 * Append new function parameter for current FUNC_PROTO type with:
2401 * - *name* - parameter name, can be NULL or empty;
2402 * - *type_id* - type ID describing the type of the parameter.
2407 int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2410 struct btf_param *p;
2411 int sz, name_off = 0;
2413 if (validate_type_id(type_id))
2414 return libbpf_err(-EINVAL);
2416 /* last type should be BTF_KIND_FUNC_PROTO */
2417 if (btf->nr_types == 0)
2418 return libbpf_err(-EINVAL);
2419 t = btf_last_type(btf);
2420 if (!btf_is_func_proto(t))
2421 return libbpf_err(-EINVAL);
2423 /* decompose and invalidate raw data */
2424 if (btf_ensure_modifiable(btf))
2425 return libbpf_err(-ENOMEM);
2427 sz = sizeof(struct btf_param);
2428 p = btf_add_type_mem(btf, sz);
2430 return libbpf_err(-ENOMEM);
2432 if (name && name[0]) {
2433 name_off = btf__add_str(btf, name);
2438 p->name_off = name_off;
2441 /* update parent type's vlen */
2442 t = btf_last_type(btf);
2443 btf_type_inc_vlen(t);
2445 btf->hdr->type_len += sz;
2446 btf->hdr->str_off += sz;
2451 * Append new BTF_KIND_VAR type with:
2452 * - *name* - non-empty/non-NULL name;
2453 * - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2454 * BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2455 * - *type_id* - type ID of the type describing the type of the variable.
2457 * - >0, type ID of newly added BTF type;
2460 int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2466 /* non-empty name */
2467 if (!name || !name[0])
2468 return libbpf_err(-EINVAL);
2469 if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2470 linkage != BTF_VAR_GLOBAL_EXTERN)
2471 return libbpf_err(-EINVAL);
2472 if (validate_type_id(type_id))
2473 return libbpf_err(-EINVAL);
2475 /* deconstruct BTF, if necessary, and invalidate raw_data */
2476 if (btf_ensure_modifiable(btf))
2477 return libbpf_err(-ENOMEM);
2479 sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2480 t = btf_add_type_mem(btf, sz);
2482 return libbpf_err(-ENOMEM);
2484 name_off = btf__add_str(btf, name);
2488 t->name_off = name_off;
2489 t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2493 v->linkage = linkage;
2495 return btf_commit_type(btf, sz);
2499 * Append new BTF_KIND_DATASEC type with:
2500 * - *name* - non-empty/non-NULL name;
2501 * - *byte_sz* - data section size, in bytes.
2503 * Data section is initially empty. Variables info can be added with
2504 * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2507 * - >0, type ID of newly added BTF type;
2510 int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2515 /* non-empty name */
2516 if (!name || !name[0])
2517 return libbpf_err(-EINVAL);
2519 if (btf_ensure_modifiable(btf))
2520 return libbpf_err(-ENOMEM);
2522 sz = sizeof(struct btf_type);
2523 t = btf_add_type_mem(btf, sz);
2525 return libbpf_err(-ENOMEM);
2527 name_off = btf__add_str(btf, name);
2531 /* start with vlen=0, which will be update as var_secinfos are added */
2532 t->name_off = name_off;
2533 t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2536 return btf_commit_type(btf, sz);
2540 * Append new data section variable information entry for current DATASEC type:
2541 * - *var_type_id* - type ID, describing type of the variable;
2542 * - *offset* - variable offset within data section, in bytes;
2543 * - *byte_sz* - variable size, in bytes.
2549 int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2552 struct btf_var_secinfo *v;
2555 /* last type should be BTF_KIND_DATASEC */
2556 if (btf->nr_types == 0)
2557 return libbpf_err(-EINVAL);
2558 t = btf_last_type(btf);
2559 if (!btf_is_datasec(t))
2560 return libbpf_err(-EINVAL);
2562 if (validate_type_id(var_type_id))
2563 return libbpf_err(-EINVAL);
2565 /* decompose and invalidate raw data */
2566 if (btf_ensure_modifiable(btf))
2567 return libbpf_err(-ENOMEM);
2569 sz = sizeof(struct btf_var_secinfo);
2570 v = btf_add_type_mem(btf, sz);
2572 return libbpf_err(-ENOMEM);
2574 v->type = var_type_id;
2578 /* update parent type's vlen */
2579 t = btf_last_type(btf);
2580 btf_type_inc_vlen(t);
2582 btf->hdr->type_len += sz;
2583 btf->hdr->str_off += sz;
2588 * Append new BTF_KIND_DECL_TAG type with:
2589 * - *value* - non-empty/non-NULL string;
2590 * - *ref_type_id* - referenced type ID, it might not exist yet;
2591 * - *component_idx* - -1 for tagging reference type, otherwise struct/union
2592 * member or function argument index;
2594 * - >0, type ID of newly added BTF type;
2597 int btf__add_decl_tag(struct btf *btf, const char *value, int ref_type_id,
2603 if (!value || !value[0] || component_idx < -1)
2604 return libbpf_err(-EINVAL);
2606 if (validate_type_id(ref_type_id))
2607 return libbpf_err(-EINVAL);
2609 if (btf_ensure_modifiable(btf))
2610 return libbpf_err(-ENOMEM);
2612 sz = sizeof(struct btf_type) + sizeof(struct btf_decl_tag);
2613 t = btf_add_type_mem(btf, sz);
2615 return libbpf_err(-ENOMEM);
2617 value_off = btf__add_str(btf, value);
2621 t->name_off = value_off;
2622 t->info = btf_type_info(BTF_KIND_DECL_TAG, 0, false);
2623 t->type = ref_type_id;
2624 btf_decl_tag(t)->component_idx = component_idx;
2626 return btf_commit_type(btf, sz);
2629 struct btf_ext_sec_setup_param {
2633 struct btf_ext_info *ext_info;
2637 static int btf_ext_setup_info(struct btf_ext *btf_ext,
2638 struct btf_ext_sec_setup_param *ext_sec)
2640 const struct btf_ext_info_sec *sinfo;
2641 struct btf_ext_info *ext_info;
2642 __u32 info_left, record_size;
2644 /* The start of the info sec (including the __u32 record_size). */
2647 if (ext_sec->len == 0)
2650 if (ext_sec->off & 0x03) {
2651 pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2656 info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2657 info_left = ext_sec->len;
2659 if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2660 pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2661 ext_sec->desc, ext_sec->off, ext_sec->len);
2665 /* At least a record size */
2666 if (info_left < sizeof(__u32)) {
2667 pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2671 /* The record size needs to meet the minimum standard */
2672 record_size = *(__u32 *)info;
2673 if (record_size < ext_sec->min_rec_size ||
2674 record_size & 0x03) {
2675 pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2676 ext_sec->desc, record_size);
2680 sinfo = info + sizeof(__u32);
2681 info_left -= sizeof(__u32);
2683 /* If no records, return failure now so .BTF.ext won't be used. */
2685 pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
2690 unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2691 __u64 total_record_size;
2694 if (info_left < sec_hdrlen) {
2695 pr_debug("%s section header is not found in .BTF.ext\n",
2700 num_records = sinfo->num_info;
2701 if (num_records == 0) {
2702 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2707 total_record_size = sec_hdrlen + (__u64)num_records * record_size;
2708 if (info_left < total_record_size) {
2709 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2714 info_left -= total_record_size;
2715 sinfo = (void *)sinfo + total_record_size;
2719 ext_info = ext_sec->ext_info;
2720 ext_info->len = ext_sec->len - sizeof(__u32);
2721 ext_info->rec_size = record_size;
2722 ext_info->info = info + sizeof(__u32);
2723 ext_info->sec_cnt = sec_cnt;
2728 static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
2730 struct btf_ext_sec_setup_param param = {
2731 .off = btf_ext->hdr->func_info_off,
2732 .len = btf_ext->hdr->func_info_len,
2733 .min_rec_size = sizeof(struct bpf_func_info_min),
2734 .ext_info = &btf_ext->func_info,
2738 return btf_ext_setup_info(btf_ext, ¶m);
2741 static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
2743 struct btf_ext_sec_setup_param param = {
2744 .off = btf_ext->hdr->line_info_off,
2745 .len = btf_ext->hdr->line_info_len,
2746 .min_rec_size = sizeof(struct bpf_line_info_min),
2747 .ext_info = &btf_ext->line_info,
2748 .desc = "line_info",
2751 return btf_ext_setup_info(btf_ext, ¶m);
2754 static int btf_ext_setup_core_relos(struct btf_ext *btf_ext)
2756 struct btf_ext_sec_setup_param param = {
2757 .off = btf_ext->hdr->core_relo_off,
2758 .len = btf_ext->hdr->core_relo_len,
2759 .min_rec_size = sizeof(struct bpf_core_relo),
2760 .ext_info = &btf_ext->core_relo_info,
2761 .desc = "core_relo",
2764 return btf_ext_setup_info(btf_ext, ¶m);
2767 static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
2769 const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
2771 if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
2772 data_size < hdr->hdr_len) {
2773 pr_debug("BTF.ext header not found");
2777 if (hdr->magic == bswap_16(BTF_MAGIC)) {
2778 pr_warn("BTF.ext in non-native endianness is not supported\n");
2780 } else if (hdr->magic != BTF_MAGIC) {
2781 pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
2785 if (hdr->version != BTF_VERSION) {
2786 pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
2791 pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
2795 if (data_size == hdr->hdr_len) {
2796 pr_debug("BTF.ext has no data\n");
2803 void btf_ext__free(struct btf_ext *btf_ext)
2805 if (IS_ERR_OR_NULL(btf_ext))
2807 free(btf_ext->func_info.sec_idxs);
2808 free(btf_ext->line_info.sec_idxs);
2809 free(btf_ext->core_relo_info.sec_idxs);
2810 free(btf_ext->data);
2814 struct btf_ext *btf_ext__new(const __u8 *data, __u32 size)
2816 struct btf_ext *btf_ext;
2819 btf_ext = calloc(1, sizeof(struct btf_ext));
2821 return libbpf_err_ptr(-ENOMEM);
2823 btf_ext->data_size = size;
2824 btf_ext->data = malloc(size);
2825 if (!btf_ext->data) {
2829 memcpy(btf_ext->data, data, size);
2831 err = btf_ext_parse_hdr(btf_ext->data, size);
2835 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
2840 err = btf_ext_setup_func_info(btf_ext);
2844 err = btf_ext_setup_line_info(btf_ext);
2848 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
2849 goto done; /* skip core relos parsing */
2851 err = btf_ext_setup_core_relos(btf_ext);
2857 btf_ext__free(btf_ext);
2858 return libbpf_err_ptr(err);
2864 const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
2866 *size = btf_ext->data_size;
2867 return btf_ext->data;
2870 static int btf_ext_reloc_info(const struct btf *btf,
2871 const struct btf_ext_info *ext_info,
2872 const char *sec_name, __u32 insns_cnt,
2873 void **info, __u32 *cnt)
2875 __u32 sec_hdrlen = sizeof(struct btf_ext_info_sec);
2876 __u32 i, record_size, existing_len, records_len;
2877 struct btf_ext_info_sec *sinfo;
2878 const char *info_sec_name;
2882 record_size = ext_info->rec_size;
2883 sinfo = ext_info->info;
2884 remain_len = ext_info->len;
2885 while (remain_len > 0) {
2886 records_len = sinfo->num_info * record_size;
2887 info_sec_name = btf__name_by_offset(btf, sinfo->sec_name_off);
2888 if (strcmp(info_sec_name, sec_name)) {
2889 remain_len -= sec_hdrlen + records_len;
2890 sinfo = (void *)sinfo + sec_hdrlen + records_len;
2894 existing_len = (*cnt) * record_size;
2895 data = realloc(*info, existing_len + records_len);
2897 return libbpf_err(-ENOMEM);
2899 memcpy(data + existing_len, sinfo->data, records_len);
2900 /* adjust insn_off only, the rest data will be passed
2903 for (i = 0; i < sinfo->num_info; i++) {
2906 insn_off = data + existing_len + (i * record_size);
2907 *insn_off = *insn_off / sizeof(struct bpf_insn) + insns_cnt;
2910 *cnt += sinfo->num_info;
2914 return libbpf_err(-ENOENT);
2917 int btf_ext__reloc_func_info(const struct btf *btf,
2918 const struct btf_ext *btf_ext,
2919 const char *sec_name, __u32 insns_cnt,
2920 void **func_info, __u32 *cnt)
2922 return btf_ext_reloc_info(btf, &btf_ext->func_info, sec_name,
2923 insns_cnt, func_info, cnt);
2926 int btf_ext__reloc_line_info(const struct btf *btf,
2927 const struct btf_ext *btf_ext,
2928 const char *sec_name, __u32 insns_cnt,
2929 void **line_info, __u32 *cnt)
2931 return btf_ext_reloc_info(btf, &btf_ext->line_info, sec_name,
2932 insns_cnt, line_info, cnt);
2935 __u32 btf_ext__func_info_rec_size(const struct btf_ext *btf_ext)
2937 return btf_ext->func_info.rec_size;
2940 __u32 btf_ext__line_info_rec_size(const struct btf_ext *btf_ext)
2942 return btf_ext->line_info.rec_size;
2947 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts);
2948 static void btf_dedup_free(struct btf_dedup *d);
2949 static int btf_dedup_prep(struct btf_dedup *d);
2950 static int btf_dedup_strings(struct btf_dedup *d);
2951 static int btf_dedup_prim_types(struct btf_dedup *d);
2952 static int btf_dedup_struct_types(struct btf_dedup *d);
2953 static int btf_dedup_ref_types(struct btf_dedup *d);
2954 static int btf_dedup_compact_types(struct btf_dedup *d);
2955 static int btf_dedup_remap_types(struct btf_dedup *d);
2958 * Deduplicate BTF types and strings.
2960 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
2961 * section with all BTF type descriptors and string data. It overwrites that
2962 * memory in-place with deduplicated types and strings without any loss of
2963 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
2964 * is provided, all the strings referenced from .BTF.ext section are honored
2965 * and updated to point to the right offsets after deduplication.
2967 * If function returns with error, type/string data might be garbled and should
2970 * More verbose and detailed description of both problem btf_dedup is solving,
2971 * as well as solution could be found at:
2972 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
2974 * Problem description and justification
2975 * =====================================
2977 * BTF type information is typically emitted either as a result of conversion
2978 * from DWARF to BTF or directly by compiler. In both cases, each compilation
2979 * unit contains information about a subset of all the types that are used
2980 * in an application. These subsets are frequently overlapping and contain a lot
2981 * of duplicated information when later concatenated together into a single
2982 * binary. This algorithm ensures that each unique type is represented by single
2983 * BTF type descriptor, greatly reducing resulting size of BTF data.
2985 * Compilation unit isolation and subsequent duplication of data is not the only
2986 * problem. The same type hierarchy (e.g., struct and all the type that struct
2987 * references) in different compilation units can be represented in BTF to
2988 * various degrees of completeness (or, rather, incompleteness) due to
2989 * struct/union forward declarations.
2991 * Let's take a look at an example, that we'll use to better understand the
2992 * problem (and solution). Suppose we have two compilation units, each using
2993 * same `struct S`, but each of them having incomplete type information about
3022 * In case of CU #1, BTF data will know only that `struct B` exist (but no
3023 * more), but will know the complete type information about `struct A`. While
3024 * for CU #2, it will know full type information about `struct B`, but will
3025 * only know about forward declaration of `struct A` (in BTF terms, it will
3026 * have `BTF_KIND_FWD` type descriptor with name `B`).
3028 * This compilation unit isolation means that it's possible that there is no
3029 * single CU with complete type information describing structs `S`, `A`, and
3030 * `B`. Also, we might get tons of duplicated and redundant type information.
3032 * Additional complication we need to keep in mind comes from the fact that
3033 * types, in general, can form graphs containing cycles, not just DAGs.
3035 * While algorithm does deduplication, it also merges and resolves type
3036 * information (unless disabled throught `struct btf_opts`), whenever possible.
3037 * E.g., in the example above with two compilation units having partial type
3038 * information for structs `A` and `B`, the output of algorithm will emit
3039 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
3040 * (as well as type information for `int` and pointers), as if they were defined
3041 * in a single compilation unit as:
3061 * Algorithm completes its work in 6 separate passes:
3063 * 1. Strings deduplication.
3064 * 2. Primitive types deduplication (int, enum, fwd).
3065 * 3. Struct/union types deduplication.
3066 * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
3067 * protos, and const/volatile/restrict modifiers).
3068 * 5. Types compaction.
3069 * 6. Types remapping.
3071 * Algorithm determines canonical type descriptor, which is a single
3072 * representative type for each truly unique type. This canonical type is the
3073 * one that will go into final deduplicated BTF type information. For
3074 * struct/unions, it is also the type that algorithm will merge additional type
3075 * information into (while resolving FWDs), as it discovers it from data in
3076 * other CUs. Each input BTF type eventually gets either mapped to itself, if
3077 * that type is canonical, or to some other type, if that type is equivalent
3078 * and was chosen as canonical representative. This mapping is stored in
3079 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
3080 * FWD type got resolved to.
3082 * To facilitate fast discovery of canonical types, we also maintain canonical
3083 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
3084 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
3085 * that match that signature. With sufficiently good choice of type signature
3086 * hashing function, we can limit number of canonical types for each unique type
3087 * signature to a very small number, allowing to find canonical type for any
3088 * duplicated type very quickly.
3090 * Struct/union deduplication is the most critical part and algorithm for
3091 * deduplicating structs/unions is described in greater details in comments for
3092 * `btf_dedup_is_equiv` function.
3095 DEFAULT_VERSION(btf__dedup_v0_6_0, btf__dedup, LIBBPF_0.6.0)
3096 int btf__dedup_v0_6_0(struct btf *btf, const struct btf_dedup_opts *opts)
3098 struct btf_dedup *d;
3101 if (!OPTS_VALID(opts, btf_dedup_opts))
3102 return libbpf_err(-EINVAL);
3104 d = btf_dedup_new(btf, opts);
3106 pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
3107 return libbpf_err(-EINVAL);
3110 if (btf_ensure_modifiable(btf)) {
3115 err = btf_dedup_prep(d);
3117 pr_debug("btf_dedup_prep failed:%d\n", err);
3120 err = btf_dedup_strings(d);
3122 pr_debug("btf_dedup_strings failed:%d\n", err);
3125 err = btf_dedup_prim_types(d);
3127 pr_debug("btf_dedup_prim_types failed:%d\n", err);
3130 err = btf_dedup_struct_types(d);
3132 pr_debug("btf_dedup_struct_types failed:%d\n", err);
3135 err = btf_dedup_ref_types(d);
3137 pr_debug("btf_dedup_ref_types failed:%d\n", err);
3140 err = btf_dedup_compact_types(d);
3142 pr_debug("btf_dedup_compact_types failed:%d\n", err);
3145 err = btf_dedup_remap_types(d);
3147 pr_debug("btf_dedup_remap_types failed:%d\n", err);
3153 return libbpf_err(err);
3156 COMPAT_VERSION(btf__dedup_deprecated, btf__dedup, LIBBPF_0.0.2)
3157 int btf__dedup_deprecated(struct btf *btf, struct btf_ext *btf_ext, const void *unused_opts)
3159 LIBBPF_OPTS(btf_dedup_opts, opts, .btf_ext = btf_ext);
3162 pr_warn("please use new version of btf__dedup() that supports options\n");
3163 return libbpf_err(-ENOTSUP);
3166 return btf__dedup(btf, &opts);
3169 #define BTF_UNPROCESSED_ID ((__u32)-1)
3170 #define BTF_IN_PROGRESS_ID ((__u32)-2)
3173 /* .BTF section to be deduped in-place */
3176 * Optional .BTF.ext section. When provided, any strings referenced
3177 * from it will be taken into account when deduping strings
3179 struct btf_ext *btf_ext;
3181 * This is a map from any type's signature hash to a list of possible
3182 * canonical representative type candidates. Hash collisions are
3183 * ignored, so even types of various kinds can share same list of
3184 * candidates, which is fine because we rely on subsequent
3185 * btf_xxx_equal() checks to authoritatively verify type equality.
3187 struct hashmap *dedup_table;
3188 /* Canonical types map */
3190 /* Hypothetical mapping, used during type graph equivalence checks */
3195 /* Whether hypothetical mapping, if successful, would need to adjust
3196 * already canonicalized types (due to a new forward declaration to
3197 * concrete type resolution). In such case, during split BTF dedup
3198 * candidate type would still be considered as different, because base
3199 * BTF is considered to be immutable.
3201 bool hypot_adjust_canon;
3202 /* Various option modifying behavior of algorithm */
3203 struct btf_dedup_opts opts;
3204 /* temporary strings deduplication state */
3205 struct strset *strs_set;
3208 static long hash_combine(long h, long value)
3210 return h * 31 + value;
3213 #define for_each_dedup_cand(d, node, hash) \
3214 hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
3216 static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
3218 return hashmap__append(d->dedup_table,
3219 (void *)hash, (void *)(long)type_id);
3222 static int btf_dedup_hypot_map_add(struct btf_dedup *d,
3223 __u32 from_id, __u32 to_id)
3225 if (d->hypot_cnt == d->hypot_cap) {
3228 d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3229 new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3232 d->hypot_list = new_list;
3234 d->hypot_list[d->hypot_cnt++] = from_id;
3235 d->hypot_map[from_id] = to_id;
3239 static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3243 for (i = 0; i < d->hypot_cnt; i++)
3244 d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3246 d->hypot_adjust_canon = false;
3249 static void btf_dedup_free(struct btf_dedup *d)
3251 hashmap__free(d->dedup_table);
3252 d->dedup_table = NULL;
3258 d->hypot_map = NULL;
3260 free(d->hypot_list);
3261 d->hypot_list = NULL;
3266 static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
3271 static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
3276 static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
3281 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts)
3283 struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3284 hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3285 int i, err = 0, type_cnt;
3288 return ERR_PTR(-ENOMEM);
3290 if (OPTS_GET(opts, force_collisions, false))
3291 hash_fn = btf_dedup_collision_hash_fn;
3294 d->btf_ext = OPTS_GET(opts, btf_ext, NULL);
3296 d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3297 if (IS_ERR(d->dedup_table)) {
3298 err = PTR_ERR(d->dedup_table);
3299 d->dedup_table = NULL;
3303 type_cnt = btf__type_cnt(btf);
3304 d->map = malloc(sizeof(__u32) * type_cnt);
3309 /* special BTF "void" type is made canonical immediately */
3311 for (i = 1; i < type_cnt; i++) {
3312 struct btf_type *t = btf_type_by_id(d->btf, i);
3314 /* VAR and DATASEC are never deduped and are self-canonical */
3315 if (btf_is_var(t) || btf_is_datasec(t))
3318 d->map[i] = BTF_UNPROCESSED_ID;
3321 d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3322 if (!d->hypot_map) {
3326 for (i = 0; i < type_cnt; i++)
3327 d->hypot_map[i] = BTF_UNPROCESSED_ID;
3332 return ERR_PTR(err);
3339 * Iterate over all possible places in .BTF and .BTF.ext that can reference
3340 * string and pass pointer to it to a provided callback `fn`.
3342 static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3346 for (i = 0; i < d->btf->nr_types; i++) {
3347 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3349 r = btf_type_visit_str_offs(t, fn, ctx);
3357 r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3364 static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3366 struct btf_dedup *d = ctx;
3367 __u32 str_off = *str_off_ptr;
3371 /* don't touch empty string or string in main BTF */
3372 if (str_off == 0 || str_off < d->btf->start_str_off)
3375 s = btf__str_by_offset(d->btf, str_off);
3376 if (d->btf->base_btf) {
3377 err = btf__find_str(d->btf->base_btf, s);
3386 off = strset__add_str(d->strs_set, s);
3390 *str_off_ptr = d->btf->start_str_off + off;
3395 * Dedup string and filter out those that are not referenced from either .BTF
3396 * or .BTF.ext (if provided) sections.
3398 * This is done by building index of all strings in BTF's string section,
3399 * then iterating over all entities that can reference strings (e.g., type
3400 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3401 * strings as used. After that all used strings are deduped and compacted into
3402 * sequential blob of memory and new offsets are calculated. Then all the string
3403 * references are iterated again and rewritten using new offsets.
3405 static int btf_dedup_strings(struct btf_dedup *d)
3409 if (d->btf->strs_deduped)
3412 d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3413 if (IS_ERR(d->strs_set)) {
3414 err = PTR_ERR(d->strs_set);
3418 if (!d->btf->base_btf) {
3419 /* insert empty string; we won't be looking it up during strings
3420 * dedup, but it's good to have it for generic BTF string lookups
3422 err = strset__add_str(d->strs_set, "");
3427 /* remap string offsets */
3428 err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3432 /* replace BTF string data and hash with deduped ones */
3433 strset__free(d->btf->strs_set);
3434 d->btf->hdr->str_len = strset__data_size(d->strs_set);
3435 d->btf->strs_set = d->strs_set;
3437 d->btf->strs_deduped = true;
3441 strset__free(d->strs_set);
3447 static long btf_hash_common(struct btf_type *t)
3451 h = hash_combine(0, t->name_off);
3452 h = hash_combine(h, t->info);
3453 h = hash_combine(h, t->size);
3457 static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3459 return t1->name_off == t2->name_off &&
3460 t1->info == t2->info &&
3461 t1->size == t2->size;
3464 /* Calculate type signature hash of INT or TAG. */
3465 static long btf_hash_int_decl_tag(struct btf_type *t)
3467 __u32 info = *(__u32 *)(t + 1);
3470 h = btf_hash_common(t);
3471 h = hash_combine(h, info);
3475 /* Check structural equality of two INTs or TAGs. */
3476 static bool btf_equal_int_tag(struct btf_type *t1, struct btf_type *t2)
3480 if (!btf_equal_common(t1, t2))
3482 info1 = *(__u32 *)(t1 + 1);
3483 info2 = *(__u32 *)(t2 + 1);
3484 return info1 == info2;
3487 /* Calculate type signature hash of ENUM. */
3488 static long btf_hash_enum(struct btf_type *t)
3492 /* don't hash vlen and enum members to support enum fwd resolving */
3493 h = hash_combine(0, t->name_off);
3494 h = hash_combine(h, t->info & ~0xffff);
3495 h = hash_combine(h, t->size);
3499 /* Check structural equality of two ENUMs. */
3500 static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3502 const struct btf_enum *m1, *m2;
3506 if (!btf_equal_common(t1, t2))
3509 vlen = btf_vlen(t1);
3512 for (i = 0; i < vlen; i++) {
3513 if (m1->name_off != m2->name_off || m1->val != m2->val)
3521 static inline bool btf_is_enum_fwd(struct btf_type *t)
3523 return btf_is_enum(t) && btf_vlen(t) == 0;
3526 static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3528 if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3529 return btf_equal_enum(t1, t2);
3530 /* ignore vlen when comparing */
3531 return t1->name_off == t2->name_off &&
3532 (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
3533 t1->size == t2->size;
3537 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3538 * as referenced type IDs equivalence is established separately during type
3539 * graph equivalence check algorithm.
3541 static long btf_hash_struct(struct btf_type *t)
3543 const struct btf_member *member = btf_members(t);
3544 __u32 vlen = btf_vlen(t);
3545 long h = btf_hash_common(t);
3548 for (i = 0; i < vlen; i++) {
3549 h = hash_combine(h, member->name_off);
3550 h = hash_combine(h, member->offset);
3551 /* no hashing of referenced type ID, it can be unresolved yet */
3558 * Check structural compatibility of two STRUCTs/UNIONs, ignoring referenced
3559 * type IDs. This check is performed during type graph equivalence check and
3560 * referenced types equivalence is checked separately.
3562 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3564 const struct btf_member *m1, *m2;
3568 if (!btf_equal_common(t1, t2))
3571 vlen = btf_vlen(t1);
3572 m1 = btf_members(t1);
3573 m2 = btf_members(t2);
3574 for (i = 0; i < vlen; i++) {
3575 if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3584 * Calculate type signature hash of ARRAY, including referenced type IDs,
3585 * under assumption that they were already resolved to canonical type IDs and
3586 * are not going to change.
3588 static long btf_hash_array(struct btf_type *t)
3590 const struct btf_array *info = btf_array(t);
3591 long h = btf_hash_common(t);
3593 h = hash_combine(h, info->type);
3594 h = hash_combine(h, info->index_type);
3595 h = hash_combine(h, info->nelems);
3600 * Check exact equality of two ARRAYs, taking into account referenced
3601 * type IDs, under assumption that they were already resolved to canonical
3602 * type IDs and are not going to change.
3603 * This function is called during reference types deduplication to compare
3604 * ARRAY to potential canonical representative.
3606 static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3608 const struct btf_array *info1, *info2;
3610 if (!btf_equal_common(t1, t2))
3613 info1 = btf_array(t1);
3614 info2 = btf_array(t2);
3615 return info1->type == info2->type &&
3616 info1->index_type == info2->index_type &&
3617 info1->nelems == info2->nelems;
3621 * Check structural compatibility of two ARRAYs, ignoring referenced type
3622 * IDs. This check is performed during type graph equivalence check and
3623 * referenced types equivalence is checked separately.
3625 static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
3627 if (!btf_equal_common(t1, t2))
3630 return btf_array(t1)->nelems == btf_array(t2)->nelems;
3634 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
3635 * under assumption that they were already resolved to canonical type IDs and
3636 * are not going to change.
3638 static long btf_hash_fnproto(struct btf_type *t)
3640 const struct btf_param *member = btf_params(t);
3641 __u16 vlen = btf_vlen(t);
3642 long h = btf_hash_common(t);
3645 for (i = 0; i < vlen; i++) {
3646 h = hash_combine(h, member->name_off);
3647 h = hash_combine(h, member->type);
3654 * Check exact equality of two FUNC_PROTOs, taking into account referenced
3655 * type IDs, under assumption that they were already resolved to canonical
3656 * type IDs and are not going to change.
3657 * This function is called during reference types deduplication to compare
3658 * FUNC_PROTO to potential canonical representative.
3660 static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
3662 const struct btf_param *m1, *m2;
3666 if (!btf_equal_common(t1, t2))
3669 vlen = btf_vlen(t1);
3670 m1 = btf_params(t1);
3671 m2 = btf_params(t2);
3672 for (i = 0; i < vlen; i++) {
3673 if (m1->name_off != m2->name_off || m1->type != m2->type)
3682 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3683 * IDs. This check is performed during type graph equivalence check and
3684 * referenced types equivalence is checked separately.
3686 static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
3688 const struct btf_param *m1, *m2;
3692 /* skip return type ID */
3693 if (t1->name_off != t2->name_off || t1->info != t2->info)
3696 vlen = btf_vlen(t1);
3697 m1 = btf_params(t1);
3698 m2 = btf_params(t2);
3699 for (i = 0; i < vlen; i++) {
3700 if (m1->name_off != m2->name_off)
3708 /* Prepare split BTF for deduplication by calculating hashes of base BTF's
3709 * types and initializing the rest of the state (canonical type mapping) for
3710 * the fixed base BTF part.
3712 static int btf_dedup_prep(struct btf_dedup *d)
3718 if (!d->btf->base_btf)
3721 for (type_id = 1; type_id < d->btf->start_id; type_id++) {
3722 t = btf_type_by_id(d->btf, type_id);
3724 /* all base BTF types are self-canonical by definition */
3725 d->map[type_id] = type_id;
3727 switch (btf_kind(t)) {
3729 case BTF_KIND_DATASEC:
3730 /* VAR and DATASEC are never hash/deduplicated */
3732 case BTF_KIND_CONST:
3733 case BTF_KIND_VOLATILE:
3734 case BTF_KIND_RESTRICT:
3737 case BTF_KIND_TYPEDEF:
3739 case BTF_KIND_FLOAT:
3740 case BTF_KIND_TYPE_TAG:
3741 h = btf_hash_common(t);
3744 case BTF_KIND_DECL_TAG:
3745 h = btf_hash_int_decl_tag(t);
3748 h = btf_hash_enum(t);
3750 case BTF_KIND_STRUCT:
3751 case BTF_KIND_UNION:
3752 h = btf_hash_struct(t);
3754 case BTF_KIND_ARRAY:
3755 h = btf_hash_array(t);
3757 case BTF_KIND_FUNC_PROTO:
3758 h = btf_hash_fnproto(t);
3761 pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
3764 if (btf_dedup_table_add(d, h, type_id))
3772 * Deduplicate primitive types, that can't reference other types, by calculating
3773 * their type signature hash and comparing them with any possible canonical
3774 * candidate. If no canonical candidate matches, type itself is marked as
3775 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
3777 static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
3779 struct btf_type *t = btf_type_by_id(d->btf, type_id);
3780 struct hashmap_entry *hash_entry;
3781 struct btf_type *cand;
3782 /* if we don't find equivalent type, then we are canonical */
3783 __u32 new_id = type_id;
3787 switch (btf_kind(t)) {
3788 case BTF_KIND_CONST:
3789 case BTF_KIND_VOLATILE:
3790 case BTF_KIND_RESTRICT:
3792 case BTF_KIND_TYPEDEF:
3793 case BTF_KIND_ARRAY:
3794 case BTF_KIND_STRUCT:
3795 case BTF_KIND_UNION:
3797 case BTF_KIND_FUNC_PROTO:
3799 case BTF_KIND_DATASEC:
3800 case BTF_KIND_DECL_TAG:
3801 case BTF_KIND_TYPE_TAG:
3805 h = btf_hash_int_decl_tag(t);
3806 for_each_dedup_cand(d, hash_entry, h) {
3807 cand_id = (__u32)(long)hash_entry->value;
3808 cand = btf_type_by_id(d->btf, cand_id);
3809 if (btf_equal_int_tag(t, cand)) {
3817 h = btf_hash_enum(t);
3818 for_each_dedup_cand(d, hash_entry, h) {
3819 cand_id = (__u32)(long)hash_entry->value;
3820 cand = btf_type_by_id(d->btf, cand_id);
3821 if (btf_equal_enum(t, cand)) {
3825 if (btf_compat_enum(t, cand)) {
3826 if (btf_is_enum_fwd(t)) {
3827 /* resolve fwd to full enum */
3831 /* resolve canonical enum fwd to full enum */
3832 d->map[cand_id] = type_id;
3838 case BTF_KIND_FLOAT:
3839 h = btf_hash_common(t);
3840 for_each_dedup_cand(d, hash_entry, h) {
3841 cand_id = (__u32)(long)hash_entry->value;
3842 cand = btf_type_by_id(d->btf, cand_id);
3843 if (btf_equal_common(t, cand)) {
3854 d->map[type_id] = new_id;
3855 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
3861 static int btf_dedup_prim_types(struct btf_dedup *d)
3865 for (i = 0; i < d->btf->nr_types; i++) {
3866 err = btf_dedup_prim_type(d, d->btf->start_id + i);
3874 * Check whether type is already mapped into canonical one (could be to itself).
3876 static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
3878 return d->map[type_id] <= BTF_MAX_NR_TYPES;
3882 * Resolve type ID into its canonical type ID, if any; otherwise return original
3883 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
3884 * STRUCT/UNION link and resolve it into canonical type ID as well.
3886 static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
3888 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3889 type_id = d->map[type_id];
3894 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
3897 static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
3899 __u32 orig_type_id = type_id;
3901 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3904 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3905 type_id = d->map[type_id];
3907 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3910 return orig_type_id;
3914 static inline __u16 btf_fwd_kind(struct btf_type *t)
3916 return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
3919 /* Check if given two types are identical ARRAY definitions */
3920 static int btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
3922 struct btf_type *t1, *t2;
3924 t1 = btf_type_by_id(d->btf, id1);
3925 t2 = btf_type_by_id(d->btf, id2);
3926 if (!btf_is_array(t1) || !btf_is_array(t2))
3929 return btf_equal_array(t1, t2);
3932 /* Check if given two types are identical STRUCT/UNION definitions */
3933 static bool btf_dedup_identical_structs(struct btf_dedup *d, __u32 id1, __u32 id2)
3935 const struct btf_member *m1, *m2;
3936 struct btf_type *t1, *t2;
3939 t1 = btf_type_by_id(d->btf, id1);
3940 t2 = btf_type_by_id(d->btf, id2);
3942 if (!btf_is_composite(t1) || btf_kind(t1) != btf_kind(t2))
3945 if (!btf_shallow_equal_struct(t1, t2))
3948 m1 = btf_members(t1);
3949 m2 = btf_members(t2);
3950 for (i = 0, n = btf_vlen(t1); i < n; i++, m1++, m2++) {
3951 if (m1->type != m2->type)
3958 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
3959 * call it "candidate graph" in this description for brevity) to a type graph
3960 * formed by (potential) canonical struct/union ("canonical graph" for brevity
3961 * here, though keep in mind that not all types in canonical graph are
3962 * necessarily canonical representatives themselves, some of them might be
3963 * duplicates or its uniqueness might not have been established yet).
3965 * - >0, if type graphs are equivalent;
3966 * - 0, if not equivalent;
3969 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
3970 * equivalence of BTF types at each step. If at any point BTF types in candidate
3971 * and canonical graphs are not compatible structurally, whole graphs are
3972 * incompatible. If types are structurally equivalent (i.e., all information
3973 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
3974 * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
3975 * If a type references other types, then those referenced types are checked
3976 * for equivalence recursively.
3978 * During DFS traversal, if we find that for current `canon_id` type we
3979 * already have some mapping in hypothetical map, we check for two possible
3981 * - `canon_id` is mapped to exactly the same type as `cand_id`. This will
3982 * happen when type graphs have cycles. In this case we assume those two
3983 * types are equivalent.
3984 * - `canon_id` is mapped to different type. This is contradiction in our
3985 * hypothetical mapping, because same graph in canonical graph corresponds
3986 * to two different types in candidate graph, which for equivalent type
3987 * graphs shouldn't happen. This condition terminates equivalence check
3988 * with negative result.
3990 * If type graphs traversal exhausts types to check and find no contradiction,
3991 * then type graphs are equivalent.
3993 * When checking types for equivalence, there is one special case: FWD types.
3994 * If FWD type resolution is allowed and one of the types (either from canonical
3995 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
3996 * flag) and their names match, hypothetical mapping is updated to point from
3997 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
3998 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
4000 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
4001 * if there are two exactly named (or anonymous) structs/unions that are
4002 * compatible structurally, one of which has FWD field, while other is concrete
4003 * STRUCT/UNION, but according to C sources they are different structs/unions
4004 * that are referencing different types with the same name. This is extremely
4005 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
4006 * this logic is causing problems.
4008 * Doing FWD resolution means that both candidate and/or canonical graphs can
4009 * consists of portions of the graph that come from multiple compilation units.
4010 * This is due to the fact that types within single compilation unit are always
4011 * deduplicated and FWDs are already resolved, if referenced struct/union
4012 * definiton is available. So, if we had unresolved FWD and found corresponding
4013 * STRUCT/UNION, they will be from different compilation units. This
4014 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
4015 * type graph will likely have at least two different BTF types that describe
4016 * same type (e.g., most probably there will be two different BTF types for the
4017 * same 'int' primitive type) and could even have "overlapping" parts of type
4018 * graph that describe same subset of types.
4020 * This in turn means that our assumption that each type in canonical graph
4021 * must correspond to exactly one type in candidate graph might not hold
4022 * anymore and will make it harder to detect contradictions using hypothetical
4023 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
4024 * resolution only in canonical graph. FWDs in candidate graphs are never
4025 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
4027 * - Both types in canonical and candidate graphs are FWDs. If they are
4028 * structurally equivalent, then they can either be both resolved to the
4029 * same STRUCT/UNION or not resolved at all. In both cases they are
4030 * equivalent and there is no need to resolve FWD on candidate side.
4031 * - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
4032 * so nothing to resolve as well, algorithm will check equivalence anyway.
4033 * - Type in canonical graph is FWD, while type in candidate is concrete
4034 * STRUCT/UNION. In this case candidate graph comes from single compilation
4035 * unit, so there is exactly one BTF type for each unique C type. After
4036 * resolving FWD into STRUCT/UNION, there might be more than one BTF type
4037 * in canonical graph mapping to single BTF type in candidate graph, but
4038 * because hypothetical mapping maps from canonical to candidate types, it's
4039 * alright, and we still maintain the property of having single `canon_id`
4040 * mapping to single `cand_id` (there could be two different `canon_id`
4041 * mapped to the same `cand_id`, but it's not contradictory).
4042 * - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
4043 * graph is FWD. In this case we are just going to check compatibility of
4044 * STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
4045 * assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
4046 * a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
4047 * turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
4050 static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
4053 struct btf_type *cand_type;
4054 struct btf_type *canon_type;
4055 __u32 hypot_type_id;
4060 /* if both resolve to the same canonical, they must be equivalent */
4061 if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
4064 canon_id = resolve_fwd_id(d, canon_id);
4066 hypot_type_id = d->hypot_map[canon_id];
4067 if (hypot_type_id <= BTF_MAX_NR_TYPES) {
4068 if (hypot_type_id == cand_id)
4070 /* In some cases compiler will generate different DWARF types
4071 * for *identical* array type definitions and use them for
4072 * different fields within the *same* struct. This breaks type
4073 * equivalence check, which makes an assumption that candidate
4074 * types sub-graph has a consistent and deduped-by-compiler
4075 * types within a single CU. So work around that by explicitly
4076 * allowing identical array types here.
4078 if (btf_dedup_identical_arrays(d, hypot_type_id, cand_id))
4080 /* It turns out that similar situation can happen with
4081 * struct/union sometimes, sigh... Handle the case where
4082 * structs/unions are exactly the same, down to the referenced
4083 * type IDs. Anything more complicated (e.g., if referenced
4084 * types are different, but equivalent) is *way more*
4085 * complicated and requires a many-to-many equivalence mapping.
4087 if (btf_dedup_identical_structs(d, hypot_type_id, cand_id))
4092 if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
4095 cand_type = btf_type_by_id(d->btf, cand_id);
4096 canon_type = btf_type_by_id(d->btf, canon_id);
4097 cand_kind = btf_kind(cand_type);
4098 canon_kind = btf_kind(canon_type);
4100 if (cand_type->name_off != canon_type->name_off)
4103 /* FWD <--> STRUCT/UNION equivalence check, if enabled */
4104 if ((cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
4105 && cand_kind != canon_kind) {
4109 if (cand_kind == BTF_KIND_FWD) {
4110 real_kind = canon_kind;
4111 fwd_kind = btf_fwd_kind(cand_type);
4113 real_kind = cand_kind;
4114 fwd_kind = btf_fwd_kind(canon_type);
4115 /* we'd need to resolve base FWD to STRUCT/UNION */
4116 if (fwd_kind == real_kind && canon_id < d->btf->start_id)
4117 d->hypot_adjust_canon = true;
4119 return fwd_kind == real_kind;
4122 if (cand_kind != canon_kind)
4125 switch (cand_kind) {
4127 return btf_equal_int_tag(cand_type, canon_type);
4130 return btf_compat_enum(cand_type, canon_type);
4133 case BTF_KIND_FLOAT:
4134 return btf_equal_common(cand_type, canon_type);
4136 case BTF_KIND_CONST:
4137 case BTF_KIND_VOLATILE:
4138 case BTF_KIND_RESTRICT:
4140 case BTF_KIND_TYPEDEF:
4142 case BTF_KIND_TYPE_TAG:
4143 if (cand_type->info != canon_type->info)
4145 return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4147 case BTF_KIND_ARRAY: {
4148 const struct btf_array *cand_arr, *canon_arr;
4150 if (!btf_compat_array(cand_type, canon_type))
4152 cand_arr = btf_array(cand_type);
4153 canon_arr = btf_array(canon_type);
4154 eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
4157 return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
4160 case BTF_KIND_STRUCT:
4161 case BTF_KIND_UNION: {
4162 const struct btf_member *cand_m, *canon_m;
4165 if (!btf_shallow_equal_struct(cand_type, canon_type))
4167 vlen = btf_vlen(cand_type);
4168 cand_m = btf_members(cand_type);
4169 canon_m = btf_members(canon_type);
4170 for (i = 0; i < vlen; i++) {
4171 eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
4181 case BTF_KIND_FUNC_PROTO: {
4182 const struct btf_param *cand_p, *canon_p;
4185 if (!btf_compat_fnproto(cand_type, canon_type))
4187 eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4190 vlen = btf_vlen(cand_type);
4191 cand_p = btf_params(cand_type);
4192 canon_p = btf_params(canon_type);
4193 for (i = 0; i < vlen; i++) {
4194 eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
4210 * Use hypothetical mapping, produced by successful type graph equivalence
4211 * check, to augment existing struct/union canonical mapping, where possible.
4213 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
4214 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
4215 * it doesn't matter if FWD type was part of canonical graph or candidate one,
4216 * we are recording the mapping anyway. As opposed to carefulness required
4217 * for struct/union correspondence mapping (described below), for FWD resolution
4218 * it's not important, as by the time that FWD type (reference type) will be
4219 * deduplicated all structs/unions will be deduped already anyway.
4221 * Recording STRUCT/UNION mapping is purely a performance optimization and is
4222 * not required for correctness. It needs to be done carefully to ensure that
4223 * struct/union from candidate's type graph is not mapped into corresponding
4224 * struct/union from canonical type graph that itself hasn't been resolved into
4225 * canonical representative. The only guarantee we have is that canonical
4226 * struct/union was determined as canonical and that won't change. But any
4227 * types referenced through that struct/union fields could have been not yet
4228 * resolved, so in case like that it's too early to establish any kind of
4229 * correspondence between structs/unions.
4231 * No canonical correspondence is derived for primitive types (they are already
4232 * deduplicated completely already anyway) or reference types (they rely on
4233 * stability of struct/union canonical relationship for equivalence checks).
4235 static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
4237 __u32 canon_type_id, targ_type_id;
4238 __u16 t_kind, c_kind;
4242 for (i = 0; i < d->hypot_cnt; i++) {
4243 canon_type_id = d->hypot_list[i];
4244 targ_type_id = d->hypot_map[canon_type_id];
4245 t_id = resolve_type_id(d, targ_type_id);
4246 c_id = resolve_type_id(d, canon_type_id);
4247 t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
4248 c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
4250 * Resolve FWD into STRUCT/UNION.
4251 * It's ok to resolve FWD into STRUCT/UNION that's not yet
4252 * mapped to canonical representative (as opposed to
4253 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
4254 * eventually that struct is going to be mapped and all resolved
4255 * FWDs will automatically resolve to correct canonical
4256 * representative. This will happen before ref type deduping,
4257 * which critically depends on stability of these mapping. This
4258 * stability is not a requirement for STRUCT/UNION equivalence
4262 /* if it's the split BTF case, we still need to point base FWD
4263 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4264 * will be resolved against base FWD. If we don't point base
4265 * canonical FWD to the resolved STRUCT/UNION, then all the
4266 * FWDs in split BTF won't be correctly resolved to a proper
4269 if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4270 d->map[c_id] = t_id;
4272 /* if graph equivalence determined that we'd need to adjust
4273 * base canonical types, then we need to only point base FWDs
4274 * to STRUCTs/UNIONs and do no more modifications. For all
4275 * other purposes the type graphs were not equivalent.
4277 if (d->hypot_adjust_canon)
4280 if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4281 d->map[t_id] = c_id;
4283 if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4284 c_kind != BTF_KIND_FWD &&
4285 is_type_mapped(d, c_id) &&
4286 !is_type_mapped(d, t_id)) {
4288 * as a perf optimization, we can map struct/union
4289 * that's part of type graph we just verified for
4290 * equivalence. We can do that for struct/union that has
4291 * canonical representative only, though.
4293 d->map[t_id] = c_id;
4299 * Deduplicate struct/union types.
4301 * For each struct/union type its type signature hash is calculated, taking
4302 * into account type's name, size, number, order and names of fields, but
4303 * ignoring type ID's referenced from fields, because they might not be deduped
4304 * completely until after reference types deduplication phase. This type hash
4305 * is used to iterate over all potential canonical types, sharing same hash.
4306 * For each canonical candidate we check whether type graphs that they form
4307 * (through referenced types in fields and so on) are equivalent using algorithm
4308 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4309 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4310 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4311 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4312 * potentially map other structs/unions to their canonical representatives,
4313 * if such relationship hasn't yet been established. This speeds up algorithm
4314 * by eliminating some of the duplicate work.
4316 * If no matching canonical representative was found, struct/union is marked
4317 * as canonical for itself and is added into btf_dedup->dedup_table hash map
4318 * for further look ups.
4320 static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4322 struct btf_type *cand_type, *t;
4323 struct hashmap_entry *hash_entry;
4324 /* if we don't find equivalent type, then we are canonical */
4325 __u32 new_id = type_id;
4329 /* already deduped or is in process of deduping (loop detected) */
4330 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4333 t = btf_type_by_id(d->btf, type_id);
4336 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4339 h = btf_hash_struct(t);
4340 for_each_dedup_cand(d, hash_entry, h) {
4341 __u32 cand_id = (__u32)(long)hash_entry->value;
4345 * Even though btf_dedup_is_equiv() checks for
4346 * btf_shallow_equal_struct() internally when checking two
4347 * structs (unions) for equivalence, we need to guard here
4348 * from picking matching FWD type as a dedup candidate.
4349 * This can happen due to hash collision. In such case just
4350 * relying on btf_dedup_is_equiv() would lead to potentially
4351 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4352 * FWD and compatible STRUCT/UNION are considered equivalent.
4354 cand_type = btf_type_by_id(d->btf, cand_id);
4355 if (!btf_shallow_equal_struct(t, cand_type))
4358 btf_dedup_clear_hypot_map(d);
4359 eq = btf_dedup_is_equiv(d, type_id, cand_id);
4364 btf_dedup_merge_hypot_map(d);
4365 if (d->hypot_adjust_canon) /* not really equivalent */
4371 d->map[type_id] = new_id;
4372 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4378 static int btf_dedup_struct_types(struct btf_dedup *d)
4382 for (i = 0; i < d->btf->nr_types; i++) {
4383 err = btf_dedup_struct_type(d, d->btf->start_id + i);
4391 * Deduplicate reference type.
4393 * Once all primitive and struct/union types got deduplicated, we can easily
4394 * deduplicate all other (reference) BTF types. This is done in two steps:
4396 * 1. Resolve all referenced type IDs into their canonical type IDs. This
4397 * resolution can be done either immediately for primitive or struct/union types
4398 * (because they were deduped in previous two phases) or recursively for
4399 * reference types. Recursion will always terminate at either primitive or
4400 * struct/union type, at which point we can "unwind" chain of reference types
4401 * one by one. There is no danger of encountering cycles because in C type
4402 * system the only way to form type cycle is through struct/union, so any chain
4403 * of reference types, even those taking part in a type cycle, will inevitably
4404 * reach struct/union at some point.
4406 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4407 * becomes "stable", in the sense that no further deduplication will cause
4408 * any changes to it. With that, it's now possible to calculate type's signature
4409 * hash (this time taking into account referenced type IDs) and loop over all
4410 * potential canonical representatives. If no match was found, current type
4411 * will become canonical representative of itself and will be added into
4412 * btf_dedup->dedup_table as another possible canonical representative.
4414 static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4416 struct hashmap_entry *hash_entry;
4417 __u32 new_id = type_id, cand_id;
4418 struct btf_type *t, *cand;
4419 /* if we don't find equivalent type, then we are representative type */
4423 if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4425 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4426 return resolve_type_id(d, type_id);
4428 t = btf_type_by_id(d->btf, type_id);
4429 d->map[type_id] = BTF_IN_PROGRESS_ID;
4431 switch (btf_kind(t)) {
4432 case BTF_KIND_CONST:
4433 case BTF_KIND_VOLATILE:
4434 case BTF_KIND_RESTRICT:
4436 case BTF_KIND_TYPEDEF:
4438 case BTF_KIND_TYPE_TAG:
4439 ref_type_id = btf_dedup_ref_type(d, t->type);
4440 if (ref_type_id < 0)
4442 t->type = ref_type_id;
4444 h = btf_hash_common(t);
4445 for_each_dedup_cand(d, hash_entry, h) {
4446 cand_id = (__u32)(long)hash_entry->value;
4447 cand = btf_type_by_id(d->btf, cand_id);
4448 if (btf_equal_common(t, cand)) {
4455 case BTF_KIND_DECL_TAG:
4456 ref_type_id = btf_dedup_ref_type(d, t->type);
4457 if (ref_type_id < 0)
4459 t->type = ref_type_id;
4461 h = btf_hash_int_decl_tag(t);
4462 for_each_dedup_cand(d, hash_entry, h) {
4463 cand_id = (__u32)(long)hash_entry->value;
4464 cand = btf_type_by_id(d->btf, cand_id);
4465 if (btf_equal_int_tag(t, cand)) {
4472 case BTF_KIND_ARRAY: {
4473 struct btf_array *info = btf_array(t);
4475 ref_type_id = btf_dedup_ref_type(d, info->type);
4476 if (ref_type_id < 0)
4478 info->type = ref_type_id;
4480 ref_type_id = btf_dedup_ref_type(d, info->index_type);
4481 if (ref_type_id < 0)
4483 info->index_type = ref_type_id;
4485 h = btf_hash_array(t);
4486 for_each_dedup_cand(d, hash_entry, h) {
4487 cand_id = (__u32)(long)hash_entry->value;
4488 cand = btf_type_by_id(d->btf, cand_id);
4489 if (btf_equal_array(t, cand)) {
4497 case BTF_KIND_FUNC_PROTO: {
4498 struct btf_param *param;
4502 ref_type_id = btf_dedup_ref_type(d, t->type);
4503 if (ref_type_id < 0)
4505 t->type = ref_type_id;
4508 param = btf_params(t);
4509 for (i = 0; i < vlen; i++) {
4510 ref_type_id = btf_dedup_ref_type(d, param->type);
4511 if (ref_type_id < 0)
4513 param->type = ref_type_id;
4517 h = btf_hash_fnproto(t);
4518 for_each_dedup_cand(d, hash_entry, h) {
4519 cand_id = (__u32)(long)hash_entry->value;
4520 cand = btf_type_by_id(d->btf, cand_id);
4521 if (btf_equal_fnproto(t, cand)) {
4533 d->map[type_id] = new_id;
4534 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4540 static int btf_dedup_ref_types(struct btf_dedup *d)
4544 for (i = 0; i < d->btf->nr_types; i++) {
4545 err = btf_dedup_ref_type(d, d->btf->start_id + i);
4549 /* we won't need d->dedup_table anymore */
4550 hashmap__free(d->dedup_table);
4551 d->dedup_table = NULL;
4558 * After we established for each type its corresponding canonical representative
4559 * type, we now can eliminate types that are not canonical and leave only
4560 * canonical ones layed out sequentially in memory by copying them over
4561 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
4562 * a map from original type ID to a new compacted type ID, which will be used
4563 * during next phase to "fix up" type IDs, referenced from struct/union and
4566 static int btf_dedup_compact_types(struct btf_dedup *d)
4569 __u32 next_type_id = d->btf->start_id;
4570 const struct btf_type *t;
4574 /* we are going to reuse hypot_map to store compaction remapping */
4575 d->hypot_map[0] = 0;
4576 /* base BTF types are not renumbered */
4577 for (id = 1; id < d->btf->start_id; id++)
4578 d->hypot_map[id] = id;
4579 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
4580 d->hypot_map[id] = BTF_UNPROCESSED_ID;
4582 p = d->btf->types_data;
4584 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
4585 if (d->map[id] != id)
4588 t = btf__type_by_id(d->btf, id);
4589 len = btf_type_size(t);
4594 d->hypot_map[id] = next_type_id;
4595 d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
4600 /* shrink struct btf's internal types index and update btf_header */
4601 d->btf->nr_types = next_type_id - d->btf->start_id;
4602 d->btf->type_offs_cap = d->btf->nr_types;
4603 d->btf->hdr->type_len = p - d->btf->types_data;
4604 new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
4606 if (d->btf->type_offs_cap && !new_offs)
4608 d->btf->type_offs = new_offs;
4609 d->btf->hdr->str_off = d->btf->hdr->type_len;
4610 d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
4615 * Figure out final (deduplicated and compacted) type ID for provided original
4616 * `type_id` by first resolving it into corresponding canonical type ID and
4617 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
4618 * which is populated during compaction phase.
4620 static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
4622 struct btf_dedup *d = ctx;
4623 __u32 resolved_type_id, new_type_id;
4625 resolved_type_id = resolve_type_id(d, *type_id);
4626 new_type_id = d->hypot_map[resolved_type_id];
4627 if (new_type_id > BTF_MAX_NR_TYPES)
4630 *type_id = new_type_id;
4635 * Remap referenced type IDs into deduped type IDs.
4637 * After BTF types are deduplicated and compacted, their final type IDs may
4638 * differ from original ones. The map from original to a corresponding
4639 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
4640 * compaction phase. During remapping phase we are rewriting all type IDs
4641 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
4642 * their final deduped type IDs.
4644 static int btf_dedup_remap_types(struct btf_dedup *d)
4648 for (i = 0; i < d->btf->nr_types; i++) {
4649 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
4651 r = btf_type_visit_type_ids(t, btf_dedup_remap_type_id, d);
4659 r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
4667 * Probe few well-known locations for vmlinux kernel image and try to load BTF
4668 * data out of it to use for target BTF.
4670 struct btf *btf__load_vmlinux_btf(void)
4673 const char *path_fmt;
4676 /* try canonical vmlinux BTF through sysfs first */
4677 { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
4678 /* fall back to trying to find vmlinux ELF on disk otherwise */
4679 { "/boot/vmlinux-%1$s" },
4680 { "/lib/modules/%1$s/vmlinux-%1$s" },
4681 { "/lib/modules/%1$s/build/vmlinux" },
4682 { "/usr/lib/modules/%1$s/kernel/vmlinux" },
4683 { "/usr/lib/debug/boot/vmlinux-%1$s" },
4684 { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
4685 { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
4687 char path[PATH_MAX + 1];
4694 for (i = 0; i < ARRAY_SIZE(locations); i++) {
4695 snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
4697 if (access(path, R_OK))
4700 if (locations[i].raw_btf)
4701 btf = btf__parse_raw(path);
4703 btf = btf__parse_elf(path, NULL);
4704 err = libbpf_get_error(btf);
4705 pr_debug("loading kernel BTF '%s': %d\n", path, err);
4712 pr_warn("failed to find valid kernel BTF\n");
4713 return libbpf_err_ptr(-ESRCH);
4716 struct btf *libbpf_find_kernel_btf(void) __attribute__((alias("btf__load_vmlinux_btf")));
4718 struct btf *btf__load_module_btf(const char *module_name, struct btf *vmlinux_btf)
4722 snprintf(path, sizeof(path), "/sys/kernel/btf/%s", module_name);
4723 return btf__parse_split(path, vmlinux_btf);
4726 int btf_type_visit_type_ids(struct btf_type *t, type_id_visit_fn visit, void *ctx)
4730 switch (btf_kind(t)) {
4732 case BTF_KIND_FLOAT:
4737 case BTF_KIND_CONST:
4738 case BTF_KIND_VOLATILE:
4739 case BTF_KIND_RESTRICT:
4741 case BTF_KIND_TYPEDEF:
4744 case BTF_KIND_DECL_TAG:
4745 case BTF_KIND_TYPE_TAG:
4746 return visit(&t->type, ctx);
4748 case BTF_KIND_ARRAY: {
4749 struct btf_array *a = btf_array(t);
4751 err = visit(&a->type, ctx);
4752 err = err ?: visit(&a->index_type, ctx);
4756 case BTF_KIND_STRUCT:
4757 case BTF_KIND_UNION: {
4758 struct btf_member *m = btf_members(t);
4760 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4761 err = visit(&m->type, ctx);
4768 case BTF_KIND_FUNC_PROTO: {
4769 struct btf_param *m = btf_params(t);
4771 err = visit(&t->type, ctx);
4774 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4775 err = visit(&m->type, ctx);
4782 case BTF_KIND_DATASEC: {
4783 struct btf_var_secinfo *m = btf_var_secinfos(t);
4785 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4786 err = visit(&m->type, ctx);
4798 int btf_type_visit_str_offs(struct btf_type *t, str_off_visit_fn visit, void *ctx)
4802 err = visit(&t->name_off, ctx);
4806 switch (btf_kind(t)) {
4807 case BTF_KIND_STRUCT:
4808 case BTF_KIND_UNION: {
4809 struct btf_member *m = btf_members(t);
4811 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4812 err = visit(&m->name_off, ctx);
4818 case BTF_KIND_ENUM: {
4819 struct btf_enum *m = btf_enum(t);
4821 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4822 err = visit(&m->name_off, ctx);
4828 case BTF_KIND_FUNC_PROTO: {
4829 struct btf_param *m = btf_params(t);
4831 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4832 err = visit(&m->name_off, ctx);
4845 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
4847 const struct btf_ext_info *seg;
4848 struct btf_ext_info_sec *sec;
4851 seg = &btf_ext->func_info;
4852 for_each_btf_ext_sec(seg, sec) {
4853 struct bpf_func_info_min *rec;
4855 for_each_btf_ext_rec(seg, sec, i, rec) {
4856 err = visit(&rec->type_id, ctx);
4862 seg = &btf_ext->core_relo_info;
4863 for_each_btf_ext_sec(seg, sec) {
4864 struct bpf_core_relo *rec;
4866 for_each_btf_ext_rec(seg, sec, i, rec) {
4867 err = visit(&rec->type_id, ctx);
4876 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
4878 const struct btf_ext_info *seg;
4879 struct btf_ext_info_sec *sec;
4882 seg = &btf_ext->func_info;
4883 for_each_btf_ext_sec(seg, sec) {
4884 err = visit(&sec->sec_name_off, ctx);
4889 seg = &btf_ext->line_info;
4890 for_each_btf_ext_sec(seg, sec) {
4891 struct bpf_line_info_min *rec;
4893 err = visit(&sec->sec_name_off, ctx);
4897 for_each_btf_ext_rec(seg, sec, i, rec) {
4898 err = visit(&rec->file_name_off, ctx);
4901 err = visit(&rec->line_off, ctx);
4907 seg = &btf_ext->core_relo_info;
4908 for_each_btf_ext_sec(seg, sec) {
4909 struct bpf_core_relo *rec;
4911 err = visit(&sec->sec_name_off, ctx);
4915 for_each_btf_ext_rec(seg, sec, i, rec) {
4916 err = visit(&rec->access_str_off, ctx);