1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 /* Copyright (c) 2018 Facebook */
12 #include <sys/utsname.h>
13 #include <sys/param.h>
15 #include <linux/kernel.h>
16 #include <linux/err.h>
17 #include <linux/btf.h>
22 #include "libbpf_internal.h"
26 #define BTF_MAX_NR_TYPES 0x7fffffffU
27 #define BTF_MAX_STR_OFFSET 0x7fffffffU
29 static struct btf_type btf_void;
32 /* raw BTF data in native endianness */
34 /* raw BTF data in non-native endianness */
35 void *raw_data_swapped;
37 /* whether target endianness differs from the native one */
41 * When BTF is loaded from an ELF or raw memory it is stored
42 * in a contiguous memory block. The hdr, type_data, and, strs_data
43 * point inside that memory region to their respective parts of BTF
46 * +--------------------------------+
47 * | Header | Types | Strings |
48 * +--------------------------------+
53 * strs_data------------+
55 * If BTF data is later modified, e.g., due to types added or
56 * removed, BTF deduplication performed, etc, this contiguous
57 * representation is broken up into three independently allocated
58 * memory regions to be able to modify them independently.
59 * raw_data is nulled out at that point, but can be later allocated
60 * and cached again if user calls btf__get_raw_data(), at which point
61 * raw_data will contain a contiguous copy of header, types, and
64 * +----------+ +---------+ +-----------+
65 * | Header | | Types | | Strings |
66 * +----------+ +---------+ +-----------+
71 * strset__data(strs_set)-----+
73 * +----------+---------+-----------+
74 * | Header | Types | Strings |
75 * raw_data----->+----------+---------+-----------+
77 struct btf_header *hdr;
80 size_t types_data_cap; /* used size stored in hdr->type_len */
82 /* type ID to `struct btf_type *` lookup index
83 * type_offs[0] corresponds to the first non-VOID type:
84 * - for base BTF it's type [1];
85 * - for split BTF it's the first non-base BTF type.
89 /* number of types in this BTF instance:
90 * - doesn't include special [0] void type;
91 * - for split BTF counts number of types added on top of base BTF.
94 /* if not NULL, points to the base BTF on top of which the current
98 /* BTF type ID of the first type in this BTF instance:
99 * - for base BTF it's equal to 1;
100 * - for split BTF it's equal to biggest type ID of base BTF plus 1.
103 /* logical string offset of this BTF instance:
104 * - for base BTF it's equal to 0;
105 * - for split BTF it's equal to total size of base BTF's string section size.
109 /* only one of strs_data or strs_set can be non-NULL, depending on
110 * whether BTF is in a modifiable state (strs_set is used) or not
111 * (strs_data points inside raw_data)
114 /* a set of unique strings */
115 struct strset *strs_set;
116 /* whether strings are already deduplicated */
119 /* BTF object FD, if loaded into kernel */
122 /* Pointer size (in bytes) for a target architecture of this BTF */
126 static inline __u64 ptr_to_u64(const void *ptr)
128 return (__u64) (unsigned long) ptr;
131 /* Ensure given dynamically allocated memory region pointed to by *data* with
132 * capacity of *cap_cnt* elements each taking *elem_sz* bytes has enough
133 * memory to accomodate *add_cnt* new elements, assuming *cur_cnt* elements
134 * are already used. At most *max_cnt* elements can be ever allocated.
135 * If necessary, memory is reallocated and all existing data is copied over,
136 * new pointer to the memory region is stored at *data, new memory region
137 * capacity (in number of elements) is stored in *cap.
138 * On success, memory pointer to the beginning of unused memory is returned.
139 * On error, NULL is returned.
141 void *libbpf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz,
142 size_t cur_cnt, size_t max_cnt, size_t add_cnt)
147 if (cur_cnt + add_cnt <= *cap_cnt)
148 return *data + cur_cnt * elem_sz;
150 /* requested more than the set limit */
151 if (cur_cnt + add_cnt > max_cnt)
155 new_cnt += new_cnt / 4; /* expand by 25% */
156 if (new_cnt < 16) /* but at least 16 elements */
158 if (new_cnt > max_cnt) /* but not exceeding a set limit */
160 if (new_cnt < cur_cnt + add_cnt) /* also ensure we have enough memory */
161 new_cnt = cur_cnt + add_cnt;
163 new_data = libbpf_reallocarray(*data, new_cnt, elem_sz);
167 /* zero out newly allocated portion of memory */
168 memset(new_data + (*cap_cnt) * elem_sz, 0, (new_cnt - *cap_cnt) * elem_sz);
172 return new_data + cur_cnt * elem_sz;
175 /* Ensure given dynamically allocated memory region has enough allocated space
176 * to accommodate *need_cnt* elements of size *elem_sz* bytes each
178 int libbpf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt)
182 if (need_cnt <= *cap_cnt)
185 p = libbpf_add_mem(data, cap_cnt, elem_sz, *cap_cnt, SIZE_MAX, need_cnt - *cap_cnt);
192 static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
196 p = libbpf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
197 btf->nr_types, BTF_MAX_NR_TYPES, 1);
205 static void btf_bswap_hdr(struct btf_header *h)
207 h->magic = bswap_16(h->magic);
208 h->hdr_len = bswap_32(h->hdr_len);
209 h->type_off = bswap_32(h->type_off);
210 h->type_len = bswap_32(h->type_len);
211 h->str_off = bswap_32(h->str_off);
212 h->str_len = bswap_32(h->str_len);
215 static int btf_parse_hdr(struct btf *btf)
217 struct btf_header *hdr = btf->hdr;
220 if (btf->raw_size < sizeof(struct btf_header)) {
221 pr_debug("BTF header not found\n");
225 if (hdr->magic == bswap_16(BTF_MAGIC)) {
226 btf->swapped_endian = true;
227 if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
228 pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
229 bswap_32(hdr->hdr_len));
233 } else if (hdr->magic != BTF_MAGIC) {
234 pr_debug("Invalid BTF magic:%x\n", hdr->magic);
238 meta_left = btf->raw_size - sizeof(*hdr);
239 if (meta_left < hdr->str_off + hdr->str_len) {
240 pr_debug("Invalid BTF total size:%u\n", btf->raw_size);
244 if (hdr->type_off + hdr->type_len > hdr->str_off) {
245 pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
246 hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
250 if (hdr->type_off % 4) {
251 pr_debug("BTF type section is not aligned to 4 bytes\n");
258 static int btf_parse_str_sec(struct btf *btf)
260 const struct btf_header *hdr = btf->hdr;
261 const char *start = btf->strs_data;
262 const char *end = start + btf->hdr->str_len;
264 if (btf->base_btf && hdr->str_len == 0)
266 if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET || end[-1]) {
267 pr_debug("Invalid BTF string section\n");
270 if (!btf->base_btf && start[0]) {
271 pr_debug("Invalid BTF string section\n");
277 static int btf_type_size(const struct btf_type *t)
279 const int base_size = sizeof(struct btf_type);
280 __u16 vlen = btf_vlen(t);
282 switch (btf_kind(t)) {
285 case BTF_KIND_VOLATILE:
286 case BTF_KIND_RESTRICT:
288 case BTF_KIND_TYPEDEF:
293 return base_size + sizeof(__u32);
295 return base_size + vlen * sizeof(struct btf_enum);
297 return base_size + sizeof(struct btf_array);
298 case BTF_KIND_STRUCT:
300 return base_size + vlen * sizeof(struct btf_member);
301 case BTF_KIND_FUNC_PROTO:
302 return base_size + vlen * sizeof(struct btf_param);
304 return base_size + sizeof(struct btf_var);
305 case BTF_KIND_DATASEC:
306 return base_size + vlen * sizeof(struct btf_var_secinfo);
308 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
313 static void btf_bswap_type_base(struct btf_type *t)
315 t->name_off = bswap_32(t->name_off);
316 t->info = bswap_32(t->info);
317 t->type = bswap_32(t->type);
320 static int btf_bswap_type_rest(struct btf_type *t)
322 struct btf_var_secinfo *v;
323 struct btf_member *m;
327 __u16 vlen = btf_vlen(t);
330 switch (btf_kind(t)) {
333 case BTF_KIND_VOLATILE:
334 case BTF_KIND_RESTRICT:
336 case BTF_KIND_TYPEDEF:
341 *(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
344 for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
345 e->name_off = bswap_32(e->name_off);
346 e->val = bswap_32(e->val);
351 a->type = bswap_32(a->type);
352 a->index_type = bswap_32(a->index_type);
353 a->nelems = bswap_32(a->nelems);
355 case BTF_KIND_STRUCT:
357 for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
358 m->name_off = bswap_32(m->name_off);
359 m->type = bswap_32(m->type);
360 m->offset = bswap_32(m->offset);
363 case BTF_KIND_FUNC_PROTO:
364 for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
365 p->name_off = bswap_32(p->name_off);
366 p->type = bswap_32(p->type);
370 btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
372 case BTF_KIND_DATASEC:
373 for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
374 v->type = bswap_32(v->type);
375 v->offset = bswap_32(v->offset);
376 v->size = bswap_32(v->size);
380 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
385 static int btf_parse_type_sec(struct btf *btf)
387 struct btf_header *hdr = btf->hdr;
388 void *next_type = btf->types_data;
389 void *end_type = next_type + hdr->type_len;
392 while (next_type + sizeof(struct btf_type) <= end_type) {
393 if (btf->swapped_endian)
394 btf_bswap_type_base(next_type);
396 type_size = btf_type_size(next_type);
399 if (next_type + type_size > end_type) {
400 pr_warn("BTF type [%d] is malformed\n", btf->start_id + btf->nr_types);
404 if (btf->swapped_endian && btf_bswap_type_rest(next_type))
407 err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
411 next_type += type_size;
415 if (next_type != end_type) {
416 pr_warn("BTF types data is malformed\n");
423 __u32 btf__get_nr_types(const struct btf *btf)
425 return btf->start_id + btf->nr_types - 1;
428 const struct btf *btf__base_btf(const struct btf *btf)
430 return btf->base_btf;
433 /* internal helper returning non-const pointer to a type */
434 struct btf_type *btf_type_by_id(struct btf *btf, __u32 type_id)
438 if (type_id < btf->start_id)
439 return btf_type_by_id(btf->base_btf, type_id);
440 return btf->types_data + btf->type_offs[type_id - btf->start_id];
443 const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
445 if (type_id >= btf->start_id + btf->nr_types)
447 return btf_type_by_id((struct btf *)btf, type_id);
450 static int determine_ptr_size(const struct btf *btf)
452 const struct btf_type *t;
456 if (btf->base_btf && btf->base_btf->ptr_sz > 0)
457 return btf->base_btf->ptr_sz;
459 n = btf__get_nr_types(btf);
460 for (i = 1; i <= n; i++) {
461 t = btf__type_by_id(btf, i);
465 name = btf__name_by_offset(btf, t->name_off);
469 if (strcmp(name, "long int") == 0 ||
470 strcmp(name, "long unsigned int") == 0) {
471 if (t->size != 4 && t->size != 8)
480 static size_t btf_ptr_sz(const struct btf *btf)
483 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
484 return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
487 /* Return pointer size this BTF instance assumes. The size is heuristically
488 * determined by looking for 'long' or 'unsigned long' integer type and
489 * recording its size in bytes. If BTF type information doesn't have any such
490 * type, this function returns 0. In the latter case, native architecture's
491 * pointer size is assumed, so will be either 4 or 8, depending on
492 * architecture that libbpf was compiled for. It's possible to override
493 * guessed value by using btf__set_pointer_size() API.
495 size_t btf__pointer_size(const struct btf *btf)
498 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
501 /* not enough BTF type info to guess */
507 /* Override or set pointer size in bytes. Only values of 4 and 8 are
510 int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
512 if (ptr_sz != 4 && ptr_sz != 8)
514 btf->ptr_sz = ptr_sz;
518 static bool is_host_big_endian(void)
520 #if __BYTE_ORDER == __LITTLE_ENDIAN
522 #elif __BYTE_ORDER == __BIG_ENDIAN
525 # error "Unrecognized __BYTE_ORDER__"
529 enum btf_endianness btf__endianness(const struct btf *btf)
531 if (is_host_big_endian())
532 return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
534 return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
537 int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
539 if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
542 btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
543 if (!btf->swapped_endian) {
544 free(btf->raw_data_swapped);
545 btf->raw_data_swapped = NULL;
550 static bool btf_type_is_void(const struct btf_type *t)
552 return t == &btf_void || btf_is_fwd(t);
555 static bool btf_type_is_void_or_null(const struct btf_type *t)
557 return !t || btf_type_is_void(t);
560 #define MAX_RESOLVE_DEPTH 32
562 __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
564 const struct btf_array *array;
565 const struct btf_type *t;
570 t = btf__type_by_id(btf, type_id);
571 for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t);
573 switch (btf_kind(t)) {
575 case BTF_KIND_STRUCT:
578 case BTF_KIND_DATASEC:
583 size = btf_ptr_sz(btf);
585 case BTF_KIND_TYPEDEF:
586 case BTF_KIND_VOLATILE:
588 case BTF_KIND_RESTRICT:
593 array = btf_array(t);
594 if (nelems && array->nelems > UINT32_MAX / nelems)
596 nelems *= array->nelems;
597 type_id = array->type;
603 t = btf__type_by_id(btf, type_id);
609 if (nelems && size > UINT32_MAX / nelems)
612 return nelems * size;
615 int btf__align_of(const struct btf *btf, __u32 id)
617 const struct btf_type *t = btf__type_by_id(btf, id);
618 __u16 kind = btf_kind(t);
624 return min(btf_ptr_sz(btf), (size_t)t->size);
626 return btf_ptr_sz(btf);
627 case BTF_KIND_TYPEDEF:
628 case BTF_KIND_VOLATILE:
630 case BTF_KIND_RESTRICT:
631 return btf__align_of(btf, t->type);
633 return btf__align_of(btf, btf_array(t)->type);
634 case BTF_KIND_STRUCT:
635 case BTF_KIND_UNION: {
636 const struct btf_member *m = btf_members(t);
637 __u16 vlen = btf_vlen(t);
638 int i, max_align = 1, align;
640 for (i = 0; i < vlen; i++, m++) {
641 align = btf__align_of(btf, m->type);
644 max_align = max(max_align, align);
650 pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
655 int btf__resolve_type(const struct btf *btf, __u32 type_id)
657 const struct btf_type *t;
660 t = btf__type_by_id(btf, type_id);
661 while (depth < MAX_RESOLVE_DEPTH &&
662 !btf_type_is_void_or_null(t) &&
663 (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
665 t = btf__type_by_id(btf, type_id);
669 if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
675 __s32 btf__find_by_name(const struct btf *btf, const char *type_name)
677 __u32 i, nr_types = btf__get_nr_types(btf);
679 if (!strcmp(type_name, "void"))
682 for (i = 1; i <= nr_types; i++) {
683 const struct btf_type *t = btf__type_by_id(btf, i);
684 const char *name = btf__name_by_offset(btf, t->name_off);
686 if (name && !strcmp(type_name, name))
693 __s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
696 __u32 i, nr_types = btf__get_nr_types(btf);
698 if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
701 for (i = 1; i <= nr_types; i++) {
702 const struct btf_type *t = btf__type_by_id(btf, i);
705 if (btf_kind(t) != kind)
707 name = btf__name_by_offset(btf, t->name_off);
708 if (name && !strcmp(type_name, name))
715 static bool btf_is_modifiable(const struct btf *btf)
717 return (void *)btf->hdr != btf->raw_data;
720 void btf__free(struct btf *btf)
722 if (IS_ERR_OR_NULL(btf))
728 if (btf_is_modifiable(btf)) {
729 /* if BTF was modified after loading, it will have a split
730 * in-memory representation for header, types, and strings
731 * sections, so we need to free all of them individually. It
732 * might still have a cached contiguous raw data present,
733 * which will be unconditionally freed below.
736 free(btf->types_data);
737 strset__free(btf->strs_set);
740 free(btf->raw_data_swapped);
741 free(btf->type_offs);
745 static struct btf *btf_new_empty(struct btf *base_btf)
749 btf = calloc(1, sizeof(*btf));
751 return ERR_PTR(-ENOMEM);
755 btf->start_str_off = 0;
757 btf->ptr_sz = sizeof(void *);
758 btf->swapped_endian = false;
761 btf->base_btf = base_btf;
762 btf->start_id = btf__get_nr_types(base_btf) + 1;
763 btf->start_str_off = base_btf->hdr->str_len;
766 /* +1 for empty string at offset 0 */
767 btf->raw_size = sizeof(struct btf_header) + (base_btf ? 0 : 1);
768 btf->raw_data = calloc(1, btf->raw_size);
769 if (!btf->raw_data) {
771 return ERR_PTR(-ENOMEM);
774 btf->hdr = btf->raw_data;
775 btf->hdr->hdr_len = sizeof(struct btf_header);
776 btf->hdr->magic = BTF_MAGIC;
777 btf->hdr->version = BTF_VERSION;
779 btf->types_data = btf->raw_data + btf->hdr->hdr_len;
780 btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
781 btf->hdr->str_len = base_btf ? 0 : 1; /* empty string at offset 0 */
786 struct btf *btf__new_empty(void)
788 return btf_new_empty(NULL);
791 struct btf *btf__new_empty_split(struct btf *base_btf)
793 return btf_new_empty(base_btf);
796 static struct btf *btf_new(const void *data, __u32 size, struct btf *base_btf)
801 btf = calloc(1, sizeof(struct btf));
803 return ERR_PTR(-ENOMEM);
807 btf->start_str_off = 0;
810 btf->base_btf = base_btf;
811 btf->start_id = btf__get_nr_types(base_btf) + 1;
812 btf->start_str_off = base_btf->hdr->str_len;
815 btf->raw_data = malloc(size);
816 if (!btf->raw_data) {
820 memcpy(btf->raw_data, data, size);
821 btf->raw_size = size;
823 btf->hdr = btf->raw_data;
824 err = btf_parse_hdr(btf);
828 btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
829 btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
831 err = btf_parse_str_sec(btf);
832 err = err ?: btf_parse_type_sec(btf);
847 struct btf *btf__new(const void *data, __u32 size)
849 return btf_new(data, size, NULL);
852 static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,
853 struct btf_ext **btf_ext)
855 Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
856 int err = 0, fd = -1, idx = 0;
857 struct btf *btf = NULL;
863 if (elf_version(EV_CURRENT) == EV_NONE) {
864 pr_warn("failed to init libelf for %s\n", path);
865 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
868 fd = open(path, O_RDONLY);
871 pr_warn("failed to open %s: %s\n", path, strerror(errno));
875 err = -LIBBPF_ERRNO__FORMAT;
877 elf = elf_begin(fd, ELF_C_READ, NULL);
879 pr_warn("failed to open %s as ELF file\n", path);
882 if (!gelf_getehdr(elf, &ehdr)) {
883 pr_warn("failed to get EHDR from %s\n", path);
887 if (elf_getshdrstrndx(elf, &shstrndx)) {
888 pr_warn("failed to get section names section index for %s\n",
893 if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
894 pr_warn("failed to get e_shstrndx from %s\n", path);
898 while ((scn = elf_nextscn(elf, scn)) != NULL) {
903 if (gelf_getshdr(scn, &sh) != &sh) {
904 pr_warn("failed to get section(%d) header from %s\n",
908 name = elf_strptr(elf, shstrndx, sh.sh_name);
910 pr_warn("failed to get section(%d) name from %s\n",
914 if (strcmp(name, BTF_ELF_SEC) == 0) {
915 btf_data = elf_getdata(scn, 0);
917 pr_warn("failed to get section(%d, %s) data from %s\n",
922 } else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
923 btf_ext_data = elf_getdata(scn, 0);
925 pr_warn("failed to get section(%d, %s) data from %s\n",
939 btf = btf_new(btf_data->d_buf, btf_data->d_size, base_btf);
943 switch (gelf_getclass(elf)) {
945 btf__set_pointer_size(btf, 4);
948 btf__set_pointer_size(btf, 8);
951 pr_warn("failed to get ELF class (bitness) for %s\n", path);
955 if (btf_ext && btf_ext_data) {
956 *btf_ext = btf_ext__new(btf_ext_data->d_buf,
957 btf_ext_data->d_size);
958 if (IS_ERR(*btf_ext))
960 } else if (btf_ext) {
971 * btf is always parsed before btf_ext, so no need to clean up
972 * btf_ext, if btf loading failed
976 if (btf_ext && IS_ERR(*btf_ext)) {
978 err = PTR_ERR(*btf_ext);
984 struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
986 return btf_parse_elf(path, NULL, btf_ext);
989 struct btf *btf__parse_elf_split(const char *path, struct btf *base_btf)
991 return btf_parse_elf(path, base_btf, NULL);
994 static struct btf *btf_parse_raw(const char *path, struct btf *base_btf)
996 struct btf *btf = NULL;
1003 f = fopen(path, "rb");
1009 /* check BTF magic */
1010 if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
1014 if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
1015 /* definitely not a raw BTF */
1021 if (fseek(f, 0, SEEK_END)) {
1030 /* rewind to the start */
1031 if (fseek(f, 0, SEEK_SET)) {
1036 /* pre-alloc memory and read all of BTF data */
1042 if (fread(data, 1, sz, f) < sz) {
1047 /* finally parse BTF data */
1048 btf = btf_new(data, sz, base_btf);
1054 return err ? ERR_PTR(err) : btf;
1057 struct btf *btf__parse_raw(const char *path)
1059 return btf_parse_raw(path, NULL);
1062 struct btf *btf__parse_raw_split(const char *path, struct btf *base_btf)
1064 return btf_parse_raw(path, base_btf);
1067 static struct btf *btf_parse(const char *path, struct btf *base_btf, struct btf_ext **btf_ext)
1074 btf = btf_parse_raw(path, base_btf);
1075 if (!IS_ERR(btf) || PTR_ERR(btf) != -EPROTO)
1078 return btf_parse_elf(path, base_btf, btf_ext);
1081 struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
1083 return btf_parse(path, NULL, btf_ext);
1086 struct btf *btf__parse_split(const char *path, struct btf *base_btf)
1088 return btf_parse(path, base_btf, NULL);
1091 static int compare_vsi_off(const void *_a, const void *_b)
1093 const struct btf_var_secinfo *a = _a;
1094 const struct btf_var_secinfo *b = _b;
1096 return a->offset - b->offset;
1099 static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
1102 __u32 size = 0, off = 0, i, vars = btf_vlen(t);
1103 const char *name = btf__name_by_offset(btf, t->name_off);
1104 const struct btf_type *t_var;
1105 struct btf_var_secinfo *vsi;
1106 const struct btf_var *var;
1110 pr_debug("No name found in string section for DATASEC kind.\n");
1114 /* .extern datasec size and var offsets were set correctly during
1115 * extern collection step, so just skip straight to sorting variables
1120 ret = bpf_object__section_size(obj, name, &size);
1121 if (ret || !size || (t->size && t->size != size)) {
1122 pr_debug("Invalid size for section %s: %u bytes\n", name, size);
1128 for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
1129 t_var = btf__type_by_id(btf, vsi->type);
1130 var = btf_var(t_var);
1132 if (!btf_is_var(t_var)) {
1133 pr_debug("Non-VAR type seen in section %s\n", name);
1137 if (var->linkage == BTF_VAR_STATIC)
1140 name = btf__name_by_offset(btf, t_var->name_off);
1142 pr_debug("No name found in string section for VAR kind\n");
1146 ret = bpf_object__variable_offset(obj, name, &off);
1148 pr_debug("No offset found in symbol table for VAR %s\n",
1157 qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
1161 int btf__finalize_data(struct bpf_object *obj, struct btf *btf)
1166 for (i = 1; i <= btf->nr_types; i++) {
1167 struct btf_type *t = btf_type_by_id(btf, i);
1169 /* Loader needs to fix up some of the things compiler
1170 * couldn't get its hands on while emitting BTF. This
1171 * is section size and global variable offset. We use
1172 * the info from the ELF itself for this purpose.
1174 if (btf_is_datasec(t)) {
1175 err = btf_fixup_datasec(obj, btf, t);
1184 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1186 int btf__load(struct btf *btf)
1188 __u32 log_buf_size = 0, raw_size;
1189 char *log_buf = NULL;
1198 log_buf = malloc(log_buf_size);
1205 raw_data = btf_get_raw_data(btf, &raw_size, false);
1210 /* cache native raw data representation */
1211 btf->raw_size = raw_size;
1212 btf->raw_data = raw_data;
1214 btf->fd = bpf_load_btf(raw_data, raw_size, log_buf, log_buf_size, false);
1216 if (!log_buf || errno == ENOSPC) {
1217 log_buf_size = max((__u32)BPF_LOG_BUF_SIZE,
1224 pr_warn("Error loading BTF: %s(%d)\n", strerror(errno), errno);
1226 pr_warn("%s\n", log_buf);
1235 int btf__fd(const struct btf *btf)
1240 void btf__set_fd(struct btf *btf, int fd)
1245 static const void *btf_strs_data(const struct btf *btf)
1247 return btf->strs_data ? btf->strs_data : strset__data(btf->strs_set);
1250 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1252 struct btf_header *hdr = btf->hdr;
1258 data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1260 *size = btf->raw_size;
1264 data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1265 data = calloc(1, data_sz);
1270 memcpy(p, hdr, hdr->hdr_len);
1275 memcpy(p, btf->types_data, hdr->type_len);
1277 for (i = 0; i < btf->nr_types; i++) {
1278 t = p + btf->type_offs[i];
1279 /* btf_bswap_type_rest() relies on native t->info, so
1280 * we swap base type info after we swapped all the
1281 * additional information
1283 if (btf_bswap_type_rest(t))
1285 btf_bswap_type_base(t);
1290 memcpy(p, btf_strs_data(btf), hdr->str_len);
1300 const void *btf__get_raw_data(const struct btf *btf_ro, __u32 *size)
1302 struct btf *btf = (struct btf *)btf_ro;
1306 data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1310 btf->raw_size = data_sz;
1311 if (btf->swapped_endian)
1312 btf->raw_data_swapped = data;
1314 btf->raw_data = data;
1319 const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
1321 if (offset < btf->start_str_off)
1322 return btf__str_by_offset(btf->base_btf, offset);
1323 else if (offset - btf->start_str_off < btf->hdr->str_len)
1324 return btf_strs_data(btf) + (offset - btf->start_str_off);
1329 const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1331 return btf__str_by_offset(btf, offset);
1334 struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf)
1336 struct bpf_btf_info btf_info;
1337 __u32 len = sizeof(btf_info);
1343 /* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
1344 * let's start with a sane default - 4KiB here - and resize it only if
1345 * bpf_obj_get_info_by_fd() needs a bigger buffer.
1348 ptr = malloc(last_size);
1350 return ERR_PTR(-ENOMEM);
1352 memset(&btf_info, 0, sizeof(btf_info));
1353 btf_info.btf = ptr_to_u64(ptr);
1354 btf_info.btf_size = last_size;
1355 err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1357 if (!err && btf_info.btf_size > last_size) {
1360 last_size = btf_info.btf_size;
1361 temp_ptr = realloc(ptr, last_size);
1363 btf = ERR_PTR(-ENOMEM);
1368 len = sizeof(btf_info);
1369 memset(&btf_info, 0, sizeof(btf_info));
1370 btf_info.btf = ptr_to_u64(ptr);
1371 btf_info.btf_size = last_size;
1373 err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1376 if (err || btf_info.btf_size > last_size) {
1377 btf = err ? ERR_PTR(-errno) : ERR_PTR(-E2BIG);
1381 btf = btf_new(ptr, btf_info.btf_size, base_btf);
1388 int btf__get_from_id(__u32 id, struct btf **btf)
1394 btf_fd = bpf_btf_get_fd_by_id(id);
1398 res = btf_get_from_fd(btf_fd, NULL);
1401 return PTR_ERR(res);
1407 int btf__get_map_kv_tids(const struct btf *btf, const char *map_name,
1408 __u32 expected_key_size, __u32 expected_value_size,
1409 __u32 *key_type_id, __u32 *value_type_id)
1411 const struct btf_type *container_type;
1412 const struct btf_member *key, *value;
1413 const size_t max_name = 256;
1414 char container_name[max_name];
1415 __s64 key_size, value_size;
1418 if (snprintf(container_name, max_name, "____btf_map_%s", map_name) ==
1420 pr_warn("map:%s length of '____btf_map_%s' is too long\n",
1421 map_name, map_name);
1425 container_id = btf__find_by_name(btf, container_name);
1426 if (container_id < 0) {
1427 pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
1428 map_name, container_name);
1429 return container_id;
1432 container_type = btf__type_by_id(btf, container_id);
1433 if (!container_type) {
1434 pr_warn("map:%s cannot find BTF type for container_id:%u\n",
1435 map_name, container_id);
1439 if (!btf_is_struct(container_type) || btf_vlen(container_type) < 2) {
1440 pr_warn("map:%s container_name:%s is an invalid container struct\n",
1441 map_name, container_name);
1445 key = btf_members(container_type);
1448 key_size = btf__resolve_size(btf, key->type);
1450 pr_warn("map:%s invalid BTF key_type_size\n", map_name);
1454 if (expected_key_size != key_size) {
1455 pr_warn("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
1456 map_name, (__u32)key_size, expected_key_size);
1460 value_size = btf__resolve_size(btf, value->type);
1461 if (value_size < 0) {
1462 pr_warn("map:%s invalid BTF value_type_size\n", map_name);
1466 if (expected_value_size != value_size) {
1467 pr_warn("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1468 map_name, (__u32)value_size, expected_value_size);
1472 *key_type_id = key->type;
1473 *value_type_id = value->type;
1478 static void btf_invalidate_raw_data(struct btf *btf)
1480 if (btf->raw_data) {
1481 free(btf->raw_data);
1482 btf->raw_data = NULL;
1484 if (btf->raw_data_swapped) {
1485 free(btf->raw_data_swapped);
1486 btf->raw_data_swapped = NULL;
1490 /* Ensure BTF is ready to be modified (by splitting into a three memory
1491 * regions for header, types, and strings). Also invalidate cached
1494 static int btf_ensure_modifiable(struct btf *btf)
1497 struct strset *set = NULL;
1500 if (btf_is_modifiable(btf)) {
1501 /* any BTF modification invalidates raw_data */
1502 btf_invalidate_raw_data(btf);
1506 /* split raw data into three memory regions */
1507 hdr = malloc(btf->hdr->hdr_len);
1508 types = malloc(btf->hdr->type_len);
1512 memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1513 memcpy(types, btf->types_data, btf->hdr->type_len);
1515 /* build lookup index for all strings */
1516 set = strset__new(BTF_MAX_STR_OFFSET, btf->strs_data, btf->hdr->str_len);
1522 /* only when everything was successful, update internal state */
1524 btf->types_data = types;
1525 btf->types_data_cap = btf->hdr->type_len;
1526 btf->strs_data = NULL;
1527 btf->strs_set = set;
1528 /* if BTF was created from scratch, all strings are guaranteed to be
1529 * unique and deduplicated
1531 if (btf->hdr->str_len == 0)
1532 btf->strs_deduped = true;
1533 if (!btf->base_btf && btf->hdr->str_len == 1)
1534 btf->strs_deduped = true;
1536 /* invalidate raw_data representation */
1537 btf_invalidate_raw_data(btf);
1548 /* Find an offset in BTF string section that corresponds to a given string *s*.
1550 * - >0 offset into string section, if string is found;
1551 * - -ENOENT, if string is not in the string section;
1552 * - <0, on any other error.
1554 int btf__find_str(struct btf *btf, const char *s)
1558 if (btf->base_btf) {
1559 off = btf__find_str(btf->base_btf, s);
1564 /* BTF needs to be in a modifiable state to build string lookup index */
1565 if (btf_ensure_modifiable(btf))
1568 off = strset__find_str(btf->strs_set, s);
1572 return btf->start_str_off + off;
1575 /* Add a string s to the BTF string section.
1577 * - > 0 offset into string section, on success;
1580 int btf__add_str(struct btf *btf, const char *s)
1584 if (btf->base_btf) {
1585 off = btf__find_str(btf->base_btf, s);
1590 if (btf_ensure_modifiable(btf))
1593 off = strset__add_str(btf->strs_set, s);
1597 btf->hdr->str_len = strset__data_size(btf->strs_set);
1599 return btf->start_str_off + off;
1602 static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1604 return libbpf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1605 btf->hdr->type_len, UINT_MAX, add_sz);
1608 static void btf_type_inc_vlen(struct btf_type *t)
1610 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1613 static int btf_commit_type(struct btf *btf, int data_sz)
1617 err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1621 btf->hdr->type_len += data_sz;
1622 btf->hdr->str_off += data_sz;
1624 return btf->start_id + btf->nr_types - 1;
1628 const struct btf *src;
1632 static int btf_rewrite_str(__u32 *str_off, void *ctx)
1634 struct btf_pipe *p = ctx;
1637 if (!*str_off) /* nothing to do for empty strings */
1640 off = btf__add_str(p->dst, btf__str_by_offset(p->src, *str_off));
1648 int btf__add_type(struct btf *btf, const struct btf *src_btf, const struct btf_type *src_type)
1650 struct btf_pipe p = { .src = src_btf, .dst = btf };
1654 sz = btf_type_size(src_type);
1658 /* deconstruct BTF, if necessary, and invalidate raw_data */
1659 if (btf_ensure_modifiable(btf))
1662 t = btf_add_type_mem(btf, sz);
1666 memcpy(t, src_type, sz);
1668 err = btf_type_visit_str_offs(t, btf_rewrite_str, &p);
1672 return btf_commit_type(btf, sz);
1676 * Append new BTF_KIND_INT type with:
1677 * - *name* - non-empty, non-NULL type name;
1678 * - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1679 * - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1681 * - >0, type ID of newly added BTF type;
1684 int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
1689 /* non-empty name */
1690 if (!name || !name[0])
1692 /* byte_sz must be power of 2 */
1693 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
1695 if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
1698 /* deconstruct BTF, if necessary, and invalidate raw_data */
1699 if (btf_ensure_modifiable(btf))
1702 sz = sizeof(struct btf_type) + sizeof(int);
1703 t = btf_add_type_mem(btf, sz);
1707 /* if something goes wrong later, we might end up with an extra string,
1708 * but that shouldn't be a problem, because BTF can't be constructed
1709 * completely anyway and will most probably be just discarded
1711 name_off = btf__add_str(btf, name);
1715 t->name_off = name_off;
1716 t->info = btf_type_info(BTF_KIND_INT, 0, 0);
1718 /* set INT info, we don't allow setting legacy bit offset/size */
1719 *(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
1721 return btf_commit_type(btf, sz);
1725 * Append new BTF_KIND_FLOAT type with:
1726 * - *name* - non-empty, non-NULL type name;
1727 * - *sz* - size of the type, in bytes;
1729 * - >0, type ID of newly added BTF type;
1732 int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
1737 /* non-empty name */
1738 if (!name || !name[0])
1741 /* byte_sz must be one of the explicitly allowed values */
1742 if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
1746 if (btf_ensure_modifiable(btf))
1749 sz = sizeof(struct btf_type);
1750 t = btf_add_type_mem(btf, sz);
1754 name_off = btf__add_str(btf, name);
1758 t->name_off = name_off;
1759 t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
1762 return btf_commit_type(btf, sz);
1765 /* it's completely legal to append BTF types with type IDs pointing forward to
1766 * types that haven't been appended yet, so we only make sure that id looks
1767 * sane, we can't guarantee that ID will always be valid
1769 static int validate_type_id(int id)
1771 if (id < 0 || id > BTF_MAX_NR_TYPES)
1776 /* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
1777 static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
1780 int sz, name_off = 0;
1782 if (validate_type_id(ref_type_id))
1785 if (btf_ensure_modifiable(btf))
1788 sz = sizeof(struct btf_type);
1789 t = btf_add_type_mem(btf, sz);
1793 if (name && name[0]) {
1794 name_off = btf__add_str(btf, name);
1799 t->name_off = name_off;
1800 t->info = btf_type_info(kind, 0, 0);
1801 t->type = ref_type_id;
1803 return btf_commit_type(btf, sz);
1807 * Append new BTF_KIND_PTR type with:
1808 * - *ref_type_id* - referenced type ID, it might not exist yet;
1810 * - >0, type ID of newly added BTF type;
1813 int btf__add_ptr(struct btf *btf, int ref_type_id)
1815 return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
1819 * Append new BTF_KIND_ARRAY type with:
1820 * - *index_type_id* - type ID of the type describing array index;
1821 * - *elem_type_id* - type ID of the type describing array element;
1822 * - *nr_elems* - the size of the array;
1824 * - >0, type ID of newly added BTF type;
1827 int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
1830 struct btf_array *a;
1833 if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
1836 if (btf_ensure_modifiable(btf))
1839 sz = sizeof(struct btf_type) + sizeof(struct btf_array);
1840 t = btf_add_type_mem(btf, sz);
1845 t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
1849 a->type = elem_type_id;
1850 a->index_type = index_type_id;
1851 a->nelems = nr_elems;
1853 return btf_commit_type(btf, sz);
1856 /* generic STRUCT/UNION append function */
1857 static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
1860 int sz, name_off = 0;
1862 if (btf_ensure_modifiable(btf))
1865 sz = sizeof(struct btf_type);
1866 t = btf_add_type_mem(btf, sz);
1870 if (name && name[0]) {
1871 name_off = btf__add_str(btf, name);
1876 /* start out with vlen=0 and no kflag; this will be adjusted when
1877 * adding each member
1879 t->name_off = name_off;
1880 t->info = btf_type_info(kind, 0, 0);
1883 return btf_commit_type(btf, sz);
1887 * Append new BTF_KIND_STRUCT type with:
1888 * - *name* - name of the struct, can be NULL or empty for anonymous structs;
1889 * - *byte_sz* - size of the struct, in bytes;
1891 * Struct initially has no fields in it. Fields can be added by
1892 * btf__add_field() right after btf__add_struct() succeeds.
1895 * - >0, type ID of newly added BTF type;
1898 int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
1900 return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
1904 * Append new BTF_KIND_UNION type with:
1905 * - *name* - name of the union, can be NULL or empty for anonymous union;
1906 * - *byte_sz* - size of the union, in bytes;
1908 * Union initially has no fields in it. Fields can be added by
1909 * btf__add_field() right after btf__add_union() succeeds. All fields
1910 * should have *bit_offset* of 0.
1913 * - >0, type ID of newly added BTF type;
1916 int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
1918 return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
1921 static struct btf_type *btf_last_type(struct btf *btf)
1923 return btf_type_by_id(btf, btf__get_nr_types(btf));
1927 * Append new field for the current STRUCT/UNION type with:
1928 * - *name* - name of the field, can be NULL or empty for anonymous field;
1929 * - *type_id* - type ID for the type describing field type;
1930 * - *bit_offset* - bit offset of the start of the field within struct/union;
1931 * - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
1936 int btf__add_field(struct btf *btf, const char *name, int type_id,
1937 __u32 bit_offset, __u32 bit_size)
1940 struct btf_member *m;
1942 int sz, name_off = 0;
1944 /* last type should be union/struct */
1945 if (btf->nr_types == 0)
1947 t = btf_last_type(btf);
1948 if (!btf_is_composite(t))
1951 if (validate_type_id(type_id))
1953 /* best-effort bit field offset/size enforcement */
1954 is_bitfield = bit_size || (bit_offset % 8 != 0);
1955 if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
1958 /* only offset 0 is allowed for unions */
1959 if (btf_is_union(t) && bit_offset)
1962 /* decompose and invalidate raw data */
1963 if (btf_ensure_modifiable(btf))
1966 sz = sizeof(struct btf_member);
1967 m = btf_add_type_mem(btf, sz);
1971 if (name && name[0]) {
1972 name_off = btf__add_str(btf, name);
1977 m->name_off = name_off;
1979 m->offset = bit_offset | (bit_size << 24);
1981 /* btf_add_type_mem can invalidate t pointer */
1982 t = btf_last_type(btf);
1983 /* update parent type's vlen and kflag */
1984 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
1986 btf->hdr->type_len += sz;
1987 btf->hdr->str_off += sz;
1992 * Append new BTF_KIND_ENUM type with:
1993 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
1994 * - *byte_sz* - size of the enum, in bytes.
1996 * Enum initially has no enum values in it (and corresponds to enum forward
1997 * declaration). Enumerator values can be added by btf__add_enum_value()
1998 * immediately after btf__add_enum() succeeds.
2001 * - >0, type ID of newly added BTF type;
2004 int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2007 int sz, name_off = 0;
2009 /* byte_sz must be power of 2 */
2010 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2013 if (btf_ensure_modifiable(btf))
2016 sz = sizeof(struct btf_type);
2017 t = btf_add_type_mem(btf, sz);
2021 if (name && name[0]) {
2022 name_off = btf__add_str(btf, name);
2027 /* start out with vlen=0; it will be adjusted when adding enum values */
2028 t->name_off = name_off;
2029 t->info = btf_type_info(BTF_KIND_ENUM, 0, 0);
2032 return btf_commit_type(btf, sz);
2036 * Append new enum value for the current ENUM type with:
2037 * - *name* - name of the enumerator value, can't be NULL or empty;
2038 * - *value* - integer value corresponding to enum value *name*;
2043 int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2049 /* last type should be BTF_KIND_ENUM */
2050 if (btf->nr_types == 0)
2052 t = btf_last_type(btf);
2053 if (!btf_is_enum(t))
2056 /* non-empty name */
2057 if (!name || !name[0])
2059 if (value < INT_MIN || value > UINT_MAX)
2062 /* decompose and invalidate raw data */
2063 if (btf_ensure_modifiable(btf))
2066 sz = sizeof(struct btf_enum);
2067 v = btf_add_type_mem(btf, sz);
2071 name_off = btf__add_str(btf, name);
2075 v->name_off = name_off;
2078 /* update parent type's vlen */
2079 t = btf_last_type(btf);
2080 btf_type_inc_vlen(t);
2082 btf->hdr->type_len += sz;
2083 btf->hdr->str_off += sz;
2088 * Append new BTF_KIND_FWD type with:
2089 * - *name*, non-empty/non-NULL name;
2090 * - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2091 * BTF_FWD_UNION, or BTF_FWD_ENUM;
2093 * - >0, type ID of newly added BTF type;
2096 int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2098 if (!name || !name[0])
2102 case BTF_FWD_STRUCT:
2103 case BTF_FWD_UNION: {
2107 id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
2110 t = btf_type_by_id(btf, id);
2111 t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2115 /* enum forward in BTF currently is just an enum with no enum
2116 * values; we also assume a standard 4-byte size for it
2118 return btf__add_enum(btf, name, sizeof(int));
2125 * Append new BTF_KING_TYPEDEF type with:
2126 * - *name*, non-empty/non-NULL name;
2127 * - *ref_type_id* - referenced type ID, it might not exist yet;
2129 * - >0, type ID of newly added BTF type;
2132 int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2134 if (!name || !name[0])
2137 return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2141 * Append new BTF_KIND_VOLATILE type with:
2142 * - *ref_type_id* - referenced type ID, it might not exist yet;
2144 * - >0, type ID of newly added BTF type;
2147 int btf__add_volatile(struct btf *btf, int ref_type_id)
2149 return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2153 * Append new BTF_KIND_CONST type with:
2154 * - *ref_type_id* - referenced type ID, it might not exist yet;
2156 * - >0, type ID of newly added BTF type;
2159 int btf__add_const(struct btf *btf, int ref_type_id)
2161 return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2165 * Append new BTF_KIND_RESTRICT type with:
2166 * - *ref_type_id* - referenced type ID, it might not exist yet;
2168 * - >0, type ID of newly added BTF type;
2171 int btf__add_restrict(struct btf *btf, int ref_type_id)
2173 return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2177 * Append new BTF_KIND_FUNC type with:
2178 * - *name*, non-empty/non-NULL name;
2179 * - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2181 * - >0, type ID of newly added BTF type;
2184 int btf__add_func(struct btf *btf, const char *name,
2185 enum btf_func_linkage linkage, int proto_type_id)
2189 if (!name || !name[0])
2191 if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2192 linkage != BTF_FUNC_EXTERN)
2195 id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2197 struct btf_type *t = btf_type_by_id(btf, id);
2199 t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2205 * Append new BTF_KIND_FUNC_PROTO with:
2206 * - *ret_type_id* - type ID for return result of a function.
2208 * Function prototype initially has no arguments, but they can be added by
2209 * btf__add_func_param() one by one, immediately after
2210 * btf__add_func_proto() succeeded.
2213 * - >0, type ID of newly added BTF type;
2216 int btf__add_func_proto(struct btf *btf, int ret_type_id)
2221 if (validate_type_id(ret_type_id))
2224 if (btf_ensure_modifiable(btf))
2227 sz = sizeof(struct btf_type);
2228 t = btf_add_type_mem(btf, sz);
2232 /* start out with vlen=0; this will be adjusted when adding enum
2233 * values, if necessary
2236 t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2237 t->type = ret_type_id;
2239 return btf_commit_type(btf, sz);
2243 * Append new function parameter for current FUNC_PROTO type with:
2244 * - *name* - parameter name, can be NULL or empty;
2245 * - *type_id* - type ID describing the type of the parameter.
2250 int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2253 struct btf_param *p;
2254 int sz, name_off = 0;
2256 if (validate_type_id(type_id))
2259 /* last type should be BTF_KIND_FUNC_PROTO */
2260 if (btf->nr_types == 0)
2262 t = btf_last_type(btf);
2263 if (!btf_is_func_proto(t))
2266 /* decompose and invalidate raw data */
2267 if (btf_ensure_modifiable(btf))
2270 sz = sizeof(struct btf_param);
2271 p = btf_add_type_mem(btf, sz);
2275 if (name && name[0]) {
2276 name_off = btf__add_str(btf, name);
2281 p->name_off = name_off;
2284 /* update parent type's vlen */
2285 t = btf_last_type(btf);
2286 btf_type_inc_vlen(t);
2288 btf->hdr->type_len += sz;
2289 btf->hdr->str_off += sz;
2294 * Append new BTF_KIND_VAR type with:
2295 * - *name* - non-empty/non-NULL name;
2296 * - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2297 * BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2298 * - *type_id* - type ID of the type describing the type of the variable.
2300 * - >0, type ID of newly added BTF type;
2303 int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2309 /* non-empty name */
2310 if (!name || !name[0])
2312 if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2313 linkage != BTF_VAR_GLOBAL_EXTERN)
2315 if (validate_type_id(type_id))
2318 /* deconstruct BTF, if necessary, and invalidate raw_data */
2319 if (btf_ensure_modifiable(btf))
2322 sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2323 t = btf_add_type_mem(btf, sz);
2327 name_off = btf__add_str(btf, name);
2331 t->name_off = name_off;
2332 t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2336 v->linkage = linkage;
2338 return btf_commit_type(btf, sz);
2342 * Append new BTF_KIND_DATASEC type with:
2343 * - *name* - non-empty/non-NULL name;
2344 * - *byte_sz* - data section size, in bytes.
2346 * Data section is initially empty. Variables info can be added with
2347 * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2350 * - >0, type ID of newly added BTF type;
2353 int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2358 /* non-empty name */
2359 if (!name || !name[0])
2362 if (btf_ensure_modifiable(btf))
2365 sz = sizeof(struct btf_type);
2366 t = btf_add_type_mem(btf, sz);
2370 name_off = btf__add_str(btf, name);
2374 /* start with vlen=0, which will be update as var_secinfos are added */
2375 t->name_off = name_off;
2376 t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2379 return btf_commit_type(btf, sz);
2383 * Append new data section variable information entry for current DATASEC type:
2384 * - *var_type_id* - type ID, describing type of the variable;
2385 * - *offset* - variable offset within data section, in bytes;
2386 * - *byte_sz* - variable size, in bytes.
2392 int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2395 struct btf_var_secinfo *v;
2398 /* last type should be BTF_KIND_DATASEC */
2399 if (btf->nr_types == 0)
2401 t = btf_last_type(btf);
2402 if (!btf_is_datasec(t))
2405 if (validate_type_id(var_type_id))
2408 /* decompose and invalidate raw data */
2409 if (btf_ensure_modifiable(btf))
2412 sz = sizeof(struct btf_var_secinfo);
2413 v = btf_add_type_mem(btf, sz);
2417 v->type = var_type_id;
2421 /* update parent type's vlen */
2422 t = btf_last_type(btf);
2423 btf_type_inc_vlen(t);
2425 btf->hdr->type_len += sz;
2426 btf->hdr->str_off += sz;
2430 struct btf_ext_sec_setup_param {
2434 struct btf_ext_info *ext_info;
2438 static int btf_ext_setup_info(struct btf_ext *btf_ext,
2439 struct btf_ext_sec_setup_param *ext_sec)
2441 const struct btf_ext_info_sec *sinfo;
2442 struct btf_ext_info *ext_info;
2443 __u32 info_left, record_size;
2444 /* The start of the info sec (including the __u32 record_size). */
2447 if (ext_sec->len == 0)
2450 if (ext_sec->off & 0x03) {
2451 pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2456 info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2457 info_left = ext_sec->len;
2459 if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2460 pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2461 ext_sec->desc, ext_sec->off, ext_sec->len);
2465 /* At least a record size */
2466 if (info_left < sizeof(__u32)) {
2467 pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2471 /* The record size needs to meet the minimum standard */
2472 record_size = *(__u32 *)info;
2473 if (record_size < ext_sec->min_rec_size ||
2474 record_size & 0x03) {
2475 pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2476 ext_sec->desc, record_size);
2480 sinfo = info + sizeof(__u32);
2481 info_left -= sizeof(__u32);
2483 /* If no records, return failure now so .BTF.ext won't be used. */
2485 pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
2490 unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2491 __u64 total_record_size;
2494 if (info_left < sec_hdrlen) {
2495 pr_debug("%s section header is not found in .BTF.ext\n",
2500 num_records = sinfo->num_info;
2501 if (num_records == 0) {
2502 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2507 total_record_size = sec_hdrlen +
2508 (__u64)num_records * record_size;
2509 if (info_left < total_record_size) {
2510 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2515 info_left -= total_record_size;
2516 sinfo = (void *)sinfo + total_record_size;
2519 ext_info = ext_sec->ext_info;
2520 ext_info->len = ext_sec->len - sizeof(__u32);
2521 ext_info->rec_size = record_size;
2522 ext_info->info = info + sizeof(__u32);
2527 static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
2529 struct btf_ext_sec_setup_param param = {
2530 .off = btf_ext->hdr->func_info_off,
2531 .len = btf_ext->hdr->func_info_len,
2532 .min_rec_size = sizeof(struct bpf_func_info_min),
2533 .ext_info = &btf_ext->func_info,
2537 return btf_ext_setup_info(btf_ext, ¶m);
2540 static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
2542 struct btf_ext_sec_setup_param param = {
2543 .off = btf_ext->hdr->line_info_off,
2544 .len = btf_ext->hdr->line_info_len,
2545 .min_rec_size = sizeof(struct bpf_line_info_min),
2546 .ext_info = &btf_ext->line_info,
2547 .desc = "line_info",
2550 return btf_ext_setup_info(btf_ext, ¶m);
2553 static int btf_ext_setup_core_relos(struct btf_ext *btf_ext)
2555 struct btf_ext_sec_setup_param param = {
2556 .off = btf_ext->hdr->core_relo_off,
2557 .len = btf_ext->hdr->core_relo_len,
2558 .min_rec_size = sizeof(struct bpf_core_relo),
2559 .ext_info = &btf_ext->core_relo_info,
2560 .desc = "core_relo",
2563 return btf_ext_setup_info(btf_ext, ¶m);
2566 static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
2568 const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
2570 if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
2571 data_size < hdr->hdr_len) {
2572 pr_debug("BTF.ext header not found");
2576 if (hdr->magic == bswap_16(BTF_MAGIC)) {
2577 pr_warn("BTF.ext in non-native endianness is not supported\n");
2579 } else if (hdr->magic != BTF_MAGIC) {
2580 pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
2584 if (hdr->version != BTF_VERSION) {
2585 pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
2590 pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
2594 if (data_size == hdr->hdr_len) {
2595 pr_debug("BTF.ext has no data\n");
2602 void btf_ext__free(struct btf_ext *btf_ext)
2604 if (IS_ERR_OR_NULL(btf_ext))
2606 free(btf_ext->data);
2610 struct btf_ext *btf_ext__new(__u8 *data, __u32 size)
2612 struct btf_ext *btf_ext;
2615 err = btf_ext_parse_hdr(data, size);
2617 return ERR_PTR(err);
2619 btf_ext = calloc(1, sizeof(struct btf_ext));
2621 return ERR_PTR(-ENOMEM);
2623 btf_ext->data_size = size;
2624 btf_ext->data = malloc(size);
2625 if (!btf_ext->data) {
2629 memcpy(btf_ext->data, data, size);
2631 if (btf_ext->hdr->hdr_len <
2632 offsetofend(struct btf_ext_header, line_info_len))
2634 err = btf_ext_setup_func_info(btf_ext);
2638 err = btf_ext_setup_line_info(btf_ext);
2642 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
2644 err = btf_ext_setup_core_relos(btf_ext);
2650 btf_ext__free(btf_ext);
2651 return ERR_PTR(err);
2657 const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
2659 *size = btf_ext->data_size;
2660 return btf_ext->data;
2663 static int btf_ext_reloc_info(const struct btf *btf,
2664 const struct btf_ext_info *ext_info,
2665 const char *sec_name, __u32 insns_cnt,
2666 void **info, __u32 *cnt)
2668 __u32 sec_hdrlen = sizeof(struct btf_ext_info_sec);
2669 __u32 i, record_size, existing_len, records_len;
2670 struct btf_ext_info_sec *sinfo;
2671 const char *info_sec_name;
2675 record_size = ext_info->rec_size;
2676 sinfo = ext_info->info;
2677 remain_len = ext_info->len;
2678 while (remain_len > 0) {
2679 records_len = sinfo->num_info * record_size;
2680 info_sec_name = btf__name_by_offset(btf, sinfo->sec_name_off);
2681 if (strcmp(info_sec_name, sec_name)) {
2682 remain_len -= sec_hdrlen + records_len;
2683 sinfo = (void *)sinfo + sec_hdrlen + records_len;
2687 existing_len = (*cnt) * record_size;
2688 data = realloc(*info, existing_len + records_len);
2692 memcpy(data + existing_len, sinfo->data, records_len);
2693 /* adjust insn_off only, the rest data will be passed
2696 for (i = 0; i < sinfo->num_info; i++) {
2699 insn_off = data + existing_len + (i * record_size);
2700 *insn_off = *insn_off / sizeof(struct bpf_insn) +
2704 *cnt += sinfo->num_info;
2711 int btf_ext__reloc_func_info(const struct btf *btf,
2712 const struct btf_ext *btf_ext,
2713 const char *sec_name, __u32 insns_cnt,
2714 void **func_info, __u32 *cnt)
2716 return btf_ext_reloc_info(btf, &btf_ext->func_info, sec_name,
2717 insns_cnt, func_info, cnt);
2720 int btf_ext__reloc_line_info(const struct btf *btf,
2721 const struct btf_ext *btf_ext,
2722 const char *sec_name, __u32 insns_cnt,
2723 void **line_info, __u32 *cnt)
2725 return btf_ext_reloc_info(btf, &btf_ext->line_info, sec_name,
2726 insns_cnt, line_info, cnt);
2729 __u32 btf_ext__func_info_rec_size(const struct btf_ext *btf_ext)
2731 return btf_ext->func_info.rec_size;
2734 __u32 btf_ext__line_info_rec_size(const struct btf_ext *btf_ext)
2736 return btf_ext->line_info.rec_size;
2741 static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
2742 const struct btf_dedup_opts *opts);
2743 static void btf_dedup_free(struct btf_dedup *d);
2744 static int btf_dedup_prep(struct btf_dedup *d);
2745 static int btf_dedup_strings(struct btf_dedup *d);
2746 static int btf_dedup_prim_types(struct btf_dedup *d);
2747 static int btf_dedup_struct_types(struct btf_dedup *d);
2748 static int btf_dedup_ref_types(struct btf_dedup *d);
2749 static int btf_dedup_compact_types(struct btf_dedup *d);
2750 static int btf_dedup_remap_types(struct btf_dedup *d);
2753 * Deduplicate BTF types and strings.
2755 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
2756 * section with all BTF type descriptors and string data. It overwrites that
2757 * memory in-place with deduplicated types and strings without any loss of
2758 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
2759 * is provided, all the strings referenced from .BTF.ext section are honored
2760 * and updated to point to the right offsets after deduplication.
2762 * If function returns with error, type/string data might be garbled and should
2765 * More verbose and detailed description of both problem btf_dedup is solving,
2766 * as well as solution could be found at:
2767 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
2769 * Problem description and justification
2770 * =====================================
2772 * BTF type information is typically emitted either as a result of conversion
2773 * from DWARF to BTF or directly by compiler. In both cases, each compilation
2774 * unit contains information about a subset of all the types that are used
2775 * in an application. These subsets are frequently overlapping and contain a lot
2776 * of duplicated information when later concatenated together into a single
2777 * binary. This algorithm ensures that each unique type is represented by single
2778 * BTF type descriptor, greatly reducing resulting size of BTF data.
2780 * Compilation unit isolation and subsequent duplication of data is not the only
2781 * problem. The same type hierarchy (e.g., struct and all the type that struct
2782 * references) in different compilation units can be represented in BTF to
2783 * various degrees of completeness (or, rather, incompleteness) due to
2784 * struct/union forward declarations.
2786 * Let's take a look at an example, that we'll use to better understand the
2787 * problem (and solution). Suppose we have two compilation units, each using
2788 * same `struct S`, but each of them having incomplete type information about
2817 * In case of CU #1, BTF data will know only that `struct B` exist (but no
2818 * more), but will know the complete type information about `struct A`. While
2819 * for CU #2, it will know full type information about `struct B`, but will
2820 * only know about forward declaration of `struct A` (in BTF terms, it will
2821 * have `BTF_KIND_FWD` type descriptor with name `B`).
2823 * This compilation unit isolation means that it's possible that there is no
2824 * single CU with complete type information describing structs `S`, `A`, and
2825 * `B`. Also, we might get tons of duplicated and redundant type information.
2827 * Additional complication we need to keep in mind comes from the fact that
2828 * types, in general, can form graphs containing cycles, not just DAGs.
2830 * While algorithm does deduplication, it also merges and resolves type
2831 * information (unless disabled throught `struct btf_opts`), whenever possible.
2832 * E.g., in the example above with two compilation units having partial type
2833 * information for structs `A` and `B`, the output of algorithm will emit
2834 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
2835 * (as well as type information for `int` and pointers), as if they were defined
2836 * in a single compilation unit as:
2856 * Algorithm completes its work in 6 separate passes:
2858 * 1. Strings deduplication.
2859 * 2. Primitive types deduplication (int, enum, fwd).
2860 * 3. Struct/union types deduplication.
2861 * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
2862 * protos, and const/volatile/restrict modifiers).
2863 * 5. Types compaction.
2864 * 6. Types remapping.
2866 * Algorithm determines canonical type descriptor, which is a single
2867 * representative type for each truly unique type. This canonical type is the
2868 * one that will go into final deduplicated BTF type information. For
2869 * struct/unions, it is also the type that algorithm will merge additional type
2870 * information into (while resolving FWDs), as it discovers it from data in
2871 * other CUs. Each input BTF type eventually gets either mapped to itself, if
2872 * that type is canonical, or to some other type, if that type is equivalent
2873 * and was chosen as canonical representative. This mapping is stored in
2874 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
2875 * FWD type got resolved to.
2877 * To facilitate fast discovery of canonical types, we also maintain canonical
2878 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
2879 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
2880 * that match that signature. With sufficiently good choice of type signature
2881 * hashing function, we can limit number of canonical types for each unique type
2882 * signature to a very small number, allowing to find canonical type for any
2883 * duplicated type very quickly.
2885 * Struct/union deduplication is the most critical part and algorithm for
2886 * deduplicating structs/unions is described in greater details in comments for
2887 * `btf_dedup_is_equiv` function.
2889 int btf__dedup(struct btf *btf, struct btf_ext *btf_ext,
2890 const struct btf_dedup_opts *opts)
2892 struct btf_dedup *d = btf_dedup_new(btf, btf_ext, opts);
2896 pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
2900 if (btf_ensure_modifiable(btf))
2903 err = btf_dedup_prep(d);
2905 pr_debug("btf_dedup_prep failed:%d\n", err);
2908 err = btf_dedup_strings(d);
2910 pr_debug("btf_dedup_strings failed:%d\n", err);
2913 err = btf_dedup_prim_types(d);
2915 pr_debug("btf_dedup_prim_types failed:%d\n", err);
2918 err = btf_dedup_struct_types(d);
2920 pr_debug("btf_dedup_struct_types failed:%d\n", err);
2923 err = btf_dedup_ref_types(d);
2925 pr_debug("btf_dedup_ref_types failed:%d\n", err);
2928 err = btf_dedup_compact_types(d);
2930 pr_debug("btf_dedup_compact_types failed:%d\n", err);
2933 err = btf_dedup_remap_types(d);
2935 pr_debug("btf_dedup_remap_types failed:%d\n", err);
2944 #define BTF_UNPROCESSED_ID ((__u32)-1)
2945 #define BTF_IN_PROGRESS_ID ((__u32)-2)
2948 /* .BTF section to be deduped in-place */
2951 * Optional .BTF.ext section. When provided, any strings referenced
2952 * from it will be taken into account when deduping strings
2954 struct btf_ext *btf_ext;
2956 * This is a map from any type's signature hash to a list of possible
2957 * canonical representative type candidates. Hash collisions are
2958 * ignored, so even types of various kinds can share same list of
2959 * candidates, which is fine because we rely on subsequent
2960 * btf_xxx_equal() checks to authoritatively verify type equality.
2962 struct hashmap *dedup_table;
2963 /* Canonical types map */
2965 /* Hypothetical mapping, used during type graph equivalence checks */
2970 /* Whether hypothetical mapping, if successful, would need to adjust
2971 * already canonicalized types (due to a new forward declaration to
2972 * concrete type resolution). In such case, during split BTF dedup
2973 * candidate type would still be considered as different, because base
2974 * BTF is considered to be immutable.
2976 bool hypot_adjust_canon;
2977 /* Various option modifying behavior of algorithm */
2978 struct btf_dedup_opts opts;
2979 /* temporary strings deduplication state */
2980 struct strset *strs_set;
2983 static long hash_combine(long h, long value)
2985 return h * 31 + value;
2988 #define for_each_dedup_cand(d, node, hash) \
2989 hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
2991 static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
2993 return hashmap__append(d->dedup_table,
2994 (void *)hash, (void *)(long)type_id);
2997 static int btf_dedup_hypot_map_add(struct btf_dedup *d,
2998 __u32 from_id, __u32 to_id)
3000 if (d->hypot_cnt == d->hypot_cap) {
3003 d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3004 new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3007 d->hypot_list = new_list;
3009 d->hypot_list[d->hypot_cnt++] = from_id;
3010 d->hypot_map[from_id] = to_id;
3014 static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3018 for (i = 0; i < d->hypot_cnt; i++)
3019 d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3021 d->hypot_adjust_canon = false;
3024 static void btf_dedup_free(struct btf_dedup *d)
3026 hashmap__free(d->dedup_table);
3027 d->dedup_table = NULL;
3033 d->hypot_map = NULL;
3035 free(d->hypot_list);
3036 d->hypot_list = NULL;
3041 static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
3046 static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
3051 static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
3056 static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
3057 const struct btf_dedup_opts *opts)
3059 struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3060 hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3061 int i, err = 0, type_cnt;
3064 return ERR_PTR(-ENOMEM);
3066 d->opts.dont_resolve_fwds = opts && opts->dont_resolve_fwds;
3067 /* dedup_table_size is now used only to force collisions in tests */
3068 if (opts && opts->dedup_table_size == 1)
3069 hash_fn = btf_dedup_collision_hash_fn;
3072 d->btf_ext = btf_ext;
3074 d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3075 if (IS_ERR(d->dedup_table)) {
3076 err = PTR_ERR(d->dedup_table);
3077 d->dedup_table = NULL;
3081 type_cnt = btf__get_nr_types(btf) + 1;
3082 d->map = malloc(sizeof(__u32) * type_cnt);
3087 /* special BTF "void" type is made canonical immediately */
3089 for (i = 1; i < type_cnt; i++) {
3090 struct btf_type *t = btf_type_by_id(d->btf, i);
3092 /* VAR and DATASEC are never deduped and are self-canonical */
3093 if (btf_is_var(t) || btf_is_datasec(t))
3096 d->map[i] = BTF_UNPROCESSED_ID;
3099 d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3100 if (!d->hypot_map) {
3104 for (i = 0; i < type_cnt; i++)
3105 d->hypot_map[i] = BTF_UNPROCESSED_ID;
3110 return ERR_PTR(err);
3117 * Iterate over all possible places in .BTF and .BTF.ext that can reference
3118 * string and pass pointer to it to a provided callback `fn`.
3120 static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3124 for (i = 0; i < d->btf->nr_types; i++) {
3125 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3127 r = btf_type_visit_str_offs(t, fn, ctx);
3135 r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3142 static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3144 struct btf_dedup *d = ctx;
3145 __u32 str_off = *str_off_ptr;
3149 /* don't touch empty string or string in main BTF */
3150 if (str_off == 0 || str_off < d->btf->start_str_off)
3153 s = btf__str_by_offset(d->btf, str_off);
3154 if (d->btf->base_btf) {
3155 err = btf__find_str(d->btf->base_btf, s);
3164 off = strset__add_str(d->strs_set, s);
3168 *str_off_ptr = d->btf->start_str_off + off;
3173 * Dedup string and filter out those that are not referenced from either .BTF
3174 * or .BTF.ext (if provided) sections.
3176 * This is done by building index of all strings in BTF's string section,
3177 * then iterating over all entities that can reference strings (e.g., type
3178 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3179 * strings as used. After that all used strings are deduped and compacted into
3180 * sequential blob of memory and new offsets are calculated. Then all the string
3181 * references are iterated again and rewritten using new offsets.
3183 static int btf_dedup_strings(struct btf_dedup *d)
3187 if (d->btf->strs_deduped)
3190 d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3191 if (IS_ERR(d->strs_set)) {
3192 err = PTR_ERR(d->strs_set);
3196 if (!d->btf->base_btf) {
3197 /* insert empty string; we won't be looking it up during strings
3198 * dedup, but it's good to have it for generic BTF string lookups
3200 err = strset__add_str(d->strs_set, "");
3205 /* remap string offsets */
3206 err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3210 /* replace BTF string data and hash with deduped ones */
3211 strset__free(d->btf->strs_set);
3212 d->btf->hdr->str_len = strset__data_size(d->strs_set);
3213 d->btf->strs_set = d->strs_set;
3215 d->btf->strs_deduped = true;
3219 strset__free(d->strs_set);
3225 static long btf_hash_common(struct btf_type *t)
3229 h = hash_combine(0, t->name_off);
3230 h = hash_combine(h, t->info);
3231 h = hash_combine(h, t->size);
3235 static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3237 return t1->name_off == t2->name_off &&
3238 t1->info == t2->info &&
3239 t1->size == t2->size;
3242 /* Calculate type signature hash of INT. */
3243 static long btf_hash_int(struct btf_type *t)
3245 __u32 info = *(__u32 *)(t + 1);
3248 h = btf_hash_common(t);
3249 h = hash_combine(h, info);
3253 /* Check structural equality of two INTs. */
3254 static bool btf_equal_int(struct btf_type *t1, struct btf_type *t2)
3258 if (!btf_equal_common(t1, t2))
3260 info1 = *(__u32 *)(t1 + 1);
3261 info2 = *(__u32 *)(t2 + 1);
3262 return info1 == info2;
3265 /* Calculate type signature hash of ENUM. */
3266 static long btf_hash_enum(struct btf_type *t)
3270 /* don't hash vlen and enum members to support enum fwd resolving */
3271 h = hash_combine(0, t->name_off);
3272 h = hash_combine(h, t->info & ~0xffff);
3273 h = hash_combine(h, t->size);
3277 /* Check structural equality of two ENUMs. */
3278 static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3280 const struct btf_enum *m1, *m2;
3284 if (!btf_equal_common(t1, t2))
3287 vlen = btf_vlen(t1);
3290 for (i = 0; i < vlen; i++) {
3291 if (m1->name_off != m2->name_off || m1->val != m2->val)
3299 static inline bool btf_is_enum_fwd(struct btf_type *t)
3301 return btf_is_enum(t) && btf_vlen(t) == 0;
3304 static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3306 if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3307 return btf_equal_enum(t1, t2);
3308 /* ignore vlen when comparing */
3309 return t1->name_off == t2->name_off &&
3310 (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
3311 t1->size == t2->size;
3315 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3316 * as referenced type IDs equivalence is established separately during type
3317 * graph equivalence check algorithm.
3319 static long btf_hash_struct(struct btf_type *t)
3321 const struct btf_member *member = btf_members(t);
3322 __u32 vlen = btf_vlen(t);
3323 long h = btf_hash_common(t);
3326 for (i = 0; i < vlen; i++) {
3327 h = hash_combine(h, member->name_off);
3328 h = hash_combine(h, member->offset);
3329 /* no hashing of referenced type ID, it can be unresolved yet */
3336 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3337 * IDs. This check is performed during type graph equivalence check and
3338 * referenced types equivalence is checked separately.
3340 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3342 const struct btf_member *m1, *m2;
3346 if (!btf_equal_common(t1, t2))
3349 vlen = btf_vlen(t1);
3350 m1 = btf_members(t1);
3351 m2 = btf_members(t2);
3352 for (i = 0; i < vlen; i++) {
3353 if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3362 * Calculate type signature hash of ARRAY, including referenced type IDs,
3363 * under assumption that they were already resolved to canonical type IDs and
3364 * are not going to change.
3366 static long btf_hash_array(struct btf_type *t)
3368 const struct btf_array *info = btf_array(t);
3369 long h = btf_hash_common(t);
3371 h = hash_combine(h, info->type);
3372 h = hash_combine(h, info->index_type);
3373 h = hash_combine(h, info->nelems);
3378 * Check exact equality of two ARRAYs, taking into account referenced
3379 * type IDs, under assumption that they were already resolved to canonical
3380 * type IDs and are not going to change.
3381 * This function is called during reference types deduplication to compare
3382 * ARRAY to potential canonical representative.
3384 static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3386 const struct btf_array *info1, *info2;
3388 if (!btf_equal_common(t1, t2))
3391 info1 = btf_array(t1);
3392 info2 = btf_array(t2);
3393 return info1->type == info2->type &&
3394 info1->index_type == info2->index_type &&
3395 info1->nelems == info2->nelems;
3399 * Check structural compatibility of two ARRAYs, ignoring referenced type
3400 * IDs. This check is performed during type graph equivalence check and
3401 * referenced types equivalence is checked separately.
3403 static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
3405 if (!btf_equal_common(t1, t2))
3408 return btf_array(t1)->nelems == btf_array(t2)->nelems;
3412 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
3413 * under assumption that they were already resolved to canonical type IDs and
3414 * are not going to change.
3416 static long btf_hash_fnproto(struct btf_type *t)
3418 const struct btf_param *member = btf_params(t);
3419 __u16 vlen = btf_vlen(t);
3420 long h = btf_hash_common(t);
3423 for (i = 0; i < vlen; i++) {
3424 h = hash_combine(h, member->name_off);
3425 h = hash_combine(h, member->type);
3432 * Check exact equality of two FUNC_PROTOs, taking into account referenced
3433 * type IDs, under assumption that they were already resolved to canonical
3434 * type IDs and are not going to change.
3435 * This function is called during reference types deduplication to compare
3436 * FUNC_PROTO to potential canonical representative.
3438 static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
3440 const struct btf_param *m1, *m2;
3444 if (!btf_equal_common(t1, t2))
3447 vlen = btf_vlen(t1);
3448 m1 = btf_params(t1);
3449 m2 = btf_params(t2);
3450 for (i = 0; i < vlen; i++) {
3451 if (m1->name_off != m2->name_off || m1->type != m2->type)
3460 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3461 * IDs. This check is performed during type graph equivalence check and
3462 * referenced types equivalence is checked separately.
3464 static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
3466 const struct btf_param *m1, *m2;
3470 /* skip return type ID */
3471 if (t1->name_off != t2->name_off || t1->info != t2->info)
3474 vlen = btf_vlen(t1);
3475 m1 = btf_params(t1);
3476 m2 = btf_params(t2);
3477 for (i = 0; i < vlen; i++) {
3478 if (m1->name_off != m2->name_off)
3486 /* Prepare split BTF for deduplication by calculating hashes of base BTF's
3487 * types and initializing the rest of the state (canonical type mapping) for
3488 * the fixed base BTF part.
3490 static int btf_dedup_prep(struct btf_dedup *d)
3496 if (!d->btf->base_btf)
3499 for (type_id = 1; type_id < d->btf->start_id; type_id++) {
3500 t = btf_type_by_id(d->btf, type_id);
3502 /* all base BTF types are self-canonical by definition */
3503 d->map[type_id] = type_id;
3505 switch (btf_kind(t)) {
3507 case BTF_KIND_DATASEC:
3508 /* VAR and DATASEC are never hash/deduplicated */
3510 case BTF_KIND_CONST:
3511 case BTF_KIND_VOLATILE:
3512 case BTF_KIND_RESTRICT:
3515 case BTF_KIND_TYPEDEF:
3517 case BTF_KIND_FLOAT:
3518 h = btf_hash_common(t);
3521 h = btf_hash_int(t);
3524 h = btf_hash_enum(t);
3526 case BTF_KIND_STRUCT:
3527 case BTF_KIND_UNION:
3528 h = btf_hash_struct(t);
3530 case BTF_KIND_ARRAY:
3531 h = btf_hash_array(t);
3533 case BTF_KIND_FUNC_PROTO:
3534 h = btf_hash_fnproto(t);
3537 pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
3540 if (btf_dedup_table_add(d, h, type_id))
3548 * Deduplicate primitive types, that can't reference other types, by calculating
3549 * their type signature hash and comparing them with any possible canonical
3550 * candidate. If no canonical candidate matches, type itself is marked as
3551 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
3553 static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
3555 struct btf_type *t = btf_type_by_id(d->btf, type_id);
3556 struct hashmap_entry *hash_entry;
3557 struct btf_type *cand;
3558 /* if we don't find equivalent type, then we are canonical */
3559 __u32 new_id = type_id;
3563 switch (btf_kind(t)) {
3564 case BTF_KIND_CONST:
3565 case BTF_KIND_VOLATILE:
3566 case BTF_KIND_RESTRICT:
3568 case BTF_KIND_TYPEDEF:
3569 case BTF_KIND_ARRAY:
3570 case BTF_KIND_STRUCT:
3571 case BTF_KIND_UNION:
3573 case BTF_KIND_FUNC_PROTO:
3575 case BTF_KIND_DATASEC:
3579 h = btf_hash_int(t);
3580 for_each_dedup_cand(d, hash_entry, h) {
3581 cand_id = (__u32)(long)hash_entry->value;
3582 cand = btf_type_by_id(d->btf, cand_id);
3583 if (btf_equal_int(t, cand)) {
3591 h = btf_hash_enum(t);
3592 for_each_dedup_cand(d, hash_entry, h) {
3593 cand_id = (__u32)(long)hash_entry->value;
3594 cand = btf_type_by_id(d->btf, cand_id);
3595 if (btf_equal_enum(t, cand)) {
3599 if (d->opts.dont_resolve_fwds)
3601 if (btf_compat_enum(t, cand)) {
3602 if (btf_is_enum_fwd(t)) {
3603 /* resolve fwd to full enum */
3607 /* resolve canonical enum fwd to full enum */
3608 d->map[cand_id] = type_id;
3614 case BTF_KIND_FLOAT:
3615 h = btf_hash_common(t);
3616 for_each_dedup_cand(d, hash_entry, h) {
3617 cand_id = (__u32)(long)hash_entry->value;
3618 cand = btf_type_by_id(d->btf, cand_id);
3619 if (btf_equal_common(t, cand)) {
3630 d->map[type_id] = new_id;
3631 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
3637 static int btf_dedup_prim_types(struct btf_dedup *d)
3641 for (i = 0; i < d->btf->nr_types; i++) {
3642 err = btf_dedup_prim_type(d, d->btf->start_id + i);
3650 * Check whether type is already mapped into canonical one (could be to itself).
3652 static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
3654 return d->map[type_id] <= BTF_MAX_NR_TYPES;
3658 * Resolve type ID into its canonical type ID, if any; otherwise return original
3659 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
3660 * STRUCT/UNION link and resolve it into canonical type ID as well.
3662 static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
3664 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3665 type_id = d->map[type_id];
3670 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
3673 static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
3675 __u32 orig_type_id = type_id;
3677 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3680 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3681 type_id = d->map[type_id];
3683 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3686 return orig_type_id;
3690 static inline __u16 btf_fwd_kind(struct btf_type *t)
3692 return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
3695 /* Check if given two types are identical ARRAY definitions */
3696 static int btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
3698 struct btf_type *t1, *t2;
3700 t1 = btf_type_by_id(d->btf, id1);
3701 t2 = btf_type_by_id(d->btf, id2);
3702 if (!btf_is_array(t1) || !btf_is_array(t2))
3705 return btf_equal_array(t1, t2);
3709 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
3710 * call it "candidate graph" in this description for brevity) to a type graph
3711 * formed by (potential) canonical struct/union ("canonical graph" for brevity
3712 * here, though keep in mind that not all types in canonical graph are
3713 * necessarily canonical representatives themselves, some of them might be
3714 * duplicates or its uniqueness might not have been established yet).
3716 * - >0, if type graphs are equivalent;
3717 * - 0, if not equivalent;
3720 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
3721 * equivalence of BTF types at each step. If at any point BTF types in candidate
3722 * and canonical graphs are not compatible structurally, whole graphs are
3723 * incompatible. If types are structurally equivalent (i.e., all information
3724 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
3725 * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
3726 * If a type references other types, then those referenced types are checked
3727 * for equivalence recursively.
3729 * During DFS traversal, if we find that for current `canon_id` type we
3730 * already have some mapping in hypothetical map, we check for two possible
3732 * - `canon_id` is mapped to exactly the same type as `cand_id`. This will
3733 * happen when type graphs have cycles. In this case we assume those two
3734 * types are equivalent.
3735 * - `canon_id` is mapped to different type. This is contradiction in our
3736 * hypothetical mapping, because same graph in canonical graph corresponds
3737 * to two different types in candidate graph, which for equivalent type
3738 * graphs shouldn't happen. This condition terminates equivalence check
3739 * with negative result.
3741 * If type graphs traversal exhausts types to check and find no contradiction,
3742 * then type graphs are equivalent.
3744 * When checking types for equivalence, there is one special case: FWD types.
3745 * If FWD type resolution is allowed and one of the types (either from canonical
3746 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
3747 * flag) and their names match, hypothetical mapping is updated to point from
3748 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
3749 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
3751 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
3752 * if there are two exactly named (or anonymous) structs/unions that are
3753 * compatible structurally, one of which has FWD field, while other is concrete
3754 * STRUCT/UNION, but according to C sources they are different structs/unions
3755 * that are referencing different types with the same name. This is extremely
3756 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
3757 * this logic is causing problems.
3759 * Doing FWD resolution means that both candidate and/or canonical graphs can
3760 * consists of portions of the graph that come from multiple compilation units.
3761 * This is due to the fact that types within single compilation unit are always
3762 * deduplicated and FWDs are already resolved, if referenced struct/union
3763 * definiton is available. So, if we had unresolved FWD and found corresponding
3764 * STRUCT/UNION, they will be from different compilation units. This
3765 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
3766 * type graph will likely have at least two different BTF types that describe
3767 * same type (e.g., most probably there will be two different BTF types for the
3768 * same 'int' primitive type) and could even have "overlapping" parts of type
3769 * graph that describe same subset of types.
3771 * This in turn means that our assumption that each type in canonical graph
3772 * must correspond to exactly one type in candidate graph might not hold
3773 * anymore and will make it harder to detect contradictions using hypothetical
3774 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
3775 * resolution only in canonical graph. FWDs in candidate graphs are never
3776 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
3778 * - Both types in canonical and candidate graphs are FWDs. If they are
3779 * structurally equivalent, then they can either be both resolved to the
3780 * same STRUCT/UNION or not resolved at all. In both cases they are
3781 * equivalent and there is no need to resolve FWD on candidate side.
3782 * - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
3783 * so nothing to resolve as well, algorithm will check equivalence anyway.
3784 * - Type in canonical graph is FWD, while type in candidate is concrete
3785 * STRUCT/UNION. In this case candidate graph comes from single compilation
3786 * unit, so there is exactly one BTF type for each unique C type. After
3787 * resolving FWD into STRUCT/UNION, there might be more than one BTF type
3788 * in canonical graph mapping to single BTF type in candidate graph, but
3789 * because hypothetical mapping maps from canonical to candidate types, it's
3790 * alright, and we still maintain the property of having single `canon_id`
3791 * mapping to single `cand_id` (there could be two different `canon_id`
3792 * mapped to the same `cand_id`, but it's not contradictory).
3793 * - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
3794 * graph is FWD. In this case we are just going to check compatibility of
3795 * STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
3796 * assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
3797 * a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
3798 * turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
3801 static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
3804 struct btf_type *cand_type;
3805 struct btf_type *canon_type;
3806 __u32 hypot_type_id;
3811 /* if both resolve to the same canonical, they must be equivalent */
3812 if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
3815 canon_id = resolve_fwd_id(d, canon_id);
3817 hypot_type_id = d->hypot_map[canon_id];
3818 if (hypot_type_id <= BTF_MAX_NR_TYPES) {
3819 /* In some cases compiler will generate different DWARF types
3820 * for *identical* array type definitions and use them for
3821 * different fields within the *same* struct. This breaks type
3822 * equivalence check, which makes an assumption that candidate
3823 * types sub-graph has a consistent and deduped-by-compiler
3824 * types within a single CU. So work around that by explicitly
3825 * allowing identical array types here.
3827 return hypot_type_id == cand_id ||
3828 btf_dedup_identical_arrays(d, hypot_type_id, cand_id);
3831 if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
3834 cand_type = btf_type_by_id(d->btf, cand_id);
3835 canon_type = btf_type_by_id(d->btf, canon_id);
3836 cand_kind = btf_kind(cand_type);
3837 canon_kind = btf_kind(canon_type);
3839 if (cand_type->name_off != canon_type->name_off)
3842 /* FWD <--> STRUCT/UNION equivalence check, if enabled */
3843 if (!d->opts.dont_resolve_fwds
3844 && (cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
3845 && cand_kind != canon_kind) {
3849 if (cand_kind == BTF_KIND_FWD) {
3850 real_kind = canon_kind;
3851 fwd_kind = btf_fwd_kind(cand_type);
3853 real_kind = cand_kind;
3854 fwd_kind = btf_fwd_kind(canon_type);
3855 /* we'd need to resolve base FWD to STRUCT/UNION */
3856 if (fwd_kind == real_kind && canon_id < d->btf->start_id)
3857 d->hypot_adjust_canon = true;
3859 return fwd_kind == real_kind;
3862 if (cand_kind != canon_kind)
3865 switch (cand_kind) {
3867 return btf_equal_int(cand_type, canon_type);
3870 if (d->opts.dont_resolve_fwds)
3871 return btf_equal_enum(cand_type, canon_type);
3873 return btf_compat_enum(cand_type, canon_type);
3876 case BTF_KIND_FLOAT:
3877 return btf_equal_common(cand_type, canon_type);
3879 case BTF_KIND_CONST:
3880 case BTF_KIND_VOLATILE:
3881 case BTF_KIND_RESTRICT:
3883 case BTF_KIND_TYPEDEF:
3885 if (cand_type->info != canon_type->info)
3887 return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
3889 case BTF_KIND_ARRAY: {
3890 const struct btf_array *cand_arr, *canon_arr;
3892 if (!btf_compat_array(cand_type, canon_type))
3894 cand_arr = btf_array(cand_type);
3895 canon_arr = btf_array(canon_type);
3896 eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
3899 return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
3902 case BTF_KIND_STRUCT:
3903 case BTF_KIND_UNION: {
3904 const struct btf_member *cand_m, *canon_m;
3907 if (!btf_shallow_equal_struct(cand_type, canon_type))
3909 vlen = btf_vlen(cand_type);
3910 cand_m = btf_members(cand_type);
3911 canon_m = btf_members(canon_type);
3912 for (i = 0; i < vlen; i++) {
3913 eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
3923 case BTF_KIND_FUNC_PROTO: {
3924 const struct btf_param *cand_p, *canon_p;
3927 if (!btf_compat_fnproto(cand_type, canon_type))
3929 eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
3932 vlen = btf_vlen(cand_type);
3933 cand_p = btf_params(cand_type);
3934 canon_p = btf_params(canon_type);
3935 for (i = 0; i < vlen; i++) {
3936 eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
3952 * Use hypothetical mapping, produced by successful type graph equivalence
3953 * check, to augment existing struct/union canonical mapping, where possible.
3955 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
3956 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
3957 * it doesn't matter if FWD type was part of canonical graph or candidate one,
3958 * we are recording the mapping anyway. As opposed to carefulness required
3959 * for struct/union correspondence mapping (described below), for FWD resolution
3960 * it's not important, as by the time that FWD type (reference type) will be
3961 * deduplicated all structs/unions will be deduped already anyway.
3963 * Recording STRUCT/UNION mapping is purely a performance optimization and is
3964 * not required for correctness. It needs to be done carefully to ensure that
3965 * struct/union from candidate's type graph is not mapped into corresponding
3966 * struct/union from canonical type graph that itself hasn't been resolved into
3967 * canonical representative. The only guarantee we have is that canonical
3968 * struct/union was determined as canonical and that won't change. But any
3969 * types referenced through that struct/union fields could have been not yet
3970 * resolved, so in case like that it's too early to establish any kind of
3971 * correspondence between structs/unions.
3973 * No canonical correspondence is derived for primitive types (they are already
3974 * deduplicated completely already anyway) or reference types (they rely on
3975 * stability of struct/union canonical relationship for equivalence checks).
3977 static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
3979 __u32 canon_type_id, targ_type_id;
3980 __u16 t_kind, c_kind;
3984 for (i = 0; i < d->hypot_cnt; i++) {
3985 canon_type_id = d->hypot_list[i];
3986 targ_type_id = d->hypot_map[canon_type_id];
3987 t_id = resolve_type_id(d, targ_type_id);
3988 c_id = resolve_type_id(d, canon_type_id);
3989 t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
3990 c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
3992 * Resolve FWD into STRUCT/UNION.
3993 * It's ok to resolve FWD into STRUCT/UNION that's not yet
3994 * mapped to canonical representative (as opposed to
3995 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
3996 * eventually that struct is going to be mapped and all resolved
3997 * FWDs will automatically resolve to correct canonical
3998 * representative. This will happen before ref type deduping,
3999 * which critically depends on stability of these mapping. This
4000 * stability is not a requirement for STRUCT/UNION equivalence
4004 /* if it's the split BTF case, we still need to point base FWD
4005 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4006 * will be resolved against base FWD. If we don't point base
4007 * canonical FWD to the resolved STRUCT/UNION, then all the
4008 * FWDs in split BTF won't be correctly resolved to a proper
4011 if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4012 d->map[c_id] = t_id;
4014 /* if graph equivalence determined that we'd need to adjust
4015 * base canonical types, then we need to only point base FWDs
4016 * to STRUCTs/UNIONs and do no more modifications. For all
4017 * other purposes the type graphs were not equivalent.
4019 if (d->hypot_adjust_canon)
4022 if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4023 d->map[t_id] = c_id;
4025 if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4026 c_kind != BTF_KIND_FWD &&
4027 is_type_mapped(d, c_id) &&
4028 !is_type_mapped(d, t_id)) {
4030 * as a perf optimization, we can map struct/union
4031 * that's part of type graph we just verified for
4032 * equivalence. We can do that for struct/union that has
4033 * canonical representative only, though.
4035 d->map[t_id] = c_id;
4041 * Deduplicate struct/union types.
4043 * For each struct/union type its type signature hash is calculated, taking
4044 * into account type's name, size, number, order and names of fields, but
4045 * ignoring type ID's referenced from fields, because they might not be deduped
4046 * completely until after reference types deduplication phase. This type hash
4047 * is used to iterate over all potential canonical types, sharing same hash.
4048 * For each canonical candidate we check whether type graphs that they form
4049 * (through referenced types in fields and so on) are equivalent using algorithm
4050 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4051 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4052 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4053 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4054 * potentially map other structs/unions to their canonical representatives,
4055 * if such relationship hasn't yet been established. This speeds up algorithm
4056 * by eliminating some of the duplicate work.
4058 * If no matching canonical representative was found, struct/union is marked
4059 * as canonical for itself and is added into btf_dedup->dedup_table hash map
4060 * for further look ups.
4062 static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4064 struct btf_type *cand_type, *t;
4065 struct hashmap_entry *hash_entry;
4066 /* if we don't find equivalent type, then we are canonical */
4067 __u32 new_id = type_id;
4071 /* already deduped or is in process of deduping (loop detected) */
4072 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4075 t = btf_type_by_id(d->btf, type_id);
4078 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4081 h = btf_hash_struct(t);
4082 for_each_dedup_cand(d, hash_entry, h) {
4083 __u32 cand_id = (__u32)(long)hash_entry->value;
4087 * Even though btf_dedup_is_equiv() checks for
4088 * btf_shallow_equal_struct() internally when checking two
4089 * structs (unions) for equivalence, we need to guard here
4090 * from picking matching FWD type as a dedup candidate.
4091 * This can happen due to hash collision. In such case just
4092 * relying on btf_dedup_is_equiv() would lead to potentially
4093 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4094 * FWD and compatible STRUCT/UNION are considered equivalent.
4096 cand_type = btf_type_by_id(d->btf, cand_id);
4097 if (!btf_shallow_equal_struct(t, cand_type))
4100 btf_dedup_clear_hypot_map(d);
4101 eq = btf_dedup_is_equiv(d, type_id, cand_id);
4106 btf_dedup_merge_hypot_map(d);
4107 if (d->hypot_adjust_canon) /* not really equivalent */
4113 d->map[type_id] = new_id;
4114 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4120 static int btf_dedup_struct_types(struct btf_dedup *d)
4124 for (i = 0; i < d->btf->nr_types; i++) {
4125 err = btf_dedup_struct_type(d, d->btf->start_id + i);
4133 * Deduplicate reference type.
4135 * Once all primitive and struct/union types got deduplicated, we can easily
4136 * deduplicate all other (reference) BTF types. This is done in two steps:
4138 * 1. Resolve all referenced type IDs into their canonical type IDs. This
4139 * resolution can be done either immediately for primitive or struct/union types
4140 * (because they were deduped in previous two phases) or recursively for
4141 * reference types. Recursion will always terminate at either primitive or
4142 * struct/union type, at which point we can "unwind" chain of reference types
4143 * one by one. There is no danger of encountering cycles because in C type
4144 * system the only way to form type cycle is through struct/union, so any chain
4145 * of reference types, even those taking part in a type cycle, will inevitably
4146 * reach struct/union at some point.
4148 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4149 * becomes "stable", in the sense that no further deduplication will cause
4150 * any changes to it. With that, it's now possible to calculate type's signature
4151 * hash (this time taking into account referenced type IDs) and loop over all
4152 * potential canonical representatives. If no match was found, current type
4153 * will become canonical representative of itself and will be added into
4154 * btf_dedup->dedup_table as another possible canonical representative.
4156 static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4158 struct hashmap_entry *hash_entry;
4159 __u32 new_id = type_id, cand_id;
4160 struct btf_type *t, *cand;
4161 /* if we don't find equivalent type, then we are representative type */
4165 if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4167 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4168 return resolve_type_id(d, type_id);
4170 t = btf_type_by_id(d->btf, type_id);
4171 d->map[type_id] = BTF_IN_PROGRESS_ID;
4173 switch (btf_kind(t)) {
4174 case BTF_KIND_CONST:
4175 case BTF_KIND_VOLATILE:
4176 case BTF_KIND_RESTRICT:
4178 case BTF_KIND_TYPEDEF:
4180 ref_type_id = btf_dedup_ref_type(d, t->type);
4181 if (ref_type_id < 0)
4183 t->type = ref_type_id;
4185 h = btf_hash_common(t);
4186 for_each_dedup_cand(d, hash_entry, h) {
4187 cand_id = (__u32)(long)hash_entry->value;
4188 cand = btf_type_by_id(d->btf, cand_id);
4189 if (btf_equal_common(t, cand)) {
4196 case BTF_KIND_ARRAY: {
4197 struct btf_array *info = btf_array(t);
4199 ref_type_id = btf_dedup_ref_type(d, info->type);
4200 if (ref_type_id < 0)
4202 info->type = ref_type_id;
4204 ref_type_id = btf_dedup_ref_type(d, info->index_type);
4205 if (ref_type_id < 0)
4207 info->index_type = ref_type_id;
4209 h = btf_hash_array(t);
4210 for_each_dedup_cand(d, hash_entry, h) {
4211 cand_id = (__u32)(long)hash_entry->value;
4212 cand = btf_type_by_id(d->btf, cand_id);
4213 if (btf_equal_array(t, cand)) {
4221 case BTF_KIND_FUNC_PROTO: {
4222 struct btf_param *param;
4226 ref_type_id = btf_dedup_ref_type(d, t->type);
4227 if (ref_type_id < 0)
4229 t->type = ref_type_id;
4232 param = btf_params(t);
4233 for (i = 0; i < vlen; i++) {
4234 ref_type_id = btf_dedup_ref_type(d, param->type);
4235 if (ref_type_id < 0)
4237 param->type = ref_type_id;
4241 h = btf_hash_fnproto(t);
4242 for_each_dedup_cand(d, hash_entry, h) {
4243 cand_id = (__u32)(long)hash_entry->value;
4244 cand = btf_type_by_id(d->btf, cand_id);
4245 if (btf_equal_fnproto(t, cand)) {
4257 d->map[type_id] = new_id;
4258 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4264 static int btf_dedup_ref_types(struct btf_dedup *d)
4268 for (i = 0; i < d->btf->nr_types; i++) {
4269 err = btf_dedup_ref_type(d, d->btf->start_id + i);
4273 /* we won't need d->dedup_table anymore */
4274 hashmap__free(d->dedup_table);
4275 d->dedup_table = NULL;
4282 * After we established for each type its corresponding canonical representative
4283 * type, we now can eliminate types that are not canonical and leave only
4284 * canonical ones layed out sequentially in memory by copying them over
4285 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
4286 * a map from original type ID to a new compacted type ID, which will be used
4287 * during next phase to "fix up" type IDs, referenced from struct/union and
4290 static int btf_dedup_compact_types(struct btf_dedup *d)
4293 __u32 next_type_id = d->btf->start_id;
4294 const struct btf_type *t;
4298 /* we are going to reuse hypot_map to store compaction remapping */
4299 d->hypot_map[0] = 0;
4300 /* base BTF types are not renumbered */
4301 for (id = 1; id < d->btf->start_id; id++)
4302 d->hypot_map[id] = id;
4303 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
4304 d->hypot_map[id] = BTF_UNPROCESSED_ID;
4306 p = d->btf->types_data;
4308 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
4309 if (d->map[id] != id)
4312 t = btf__type_by_id(d->btf, id);
4313 len = btf_type_size(t);
4318 d->hypot_map[id] = next_type_id;
4319 d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
4324 /* shrink struct btf's internal types index and update btf_header */
4325 d->btf->nr_types = next_type_id - d->btf->start_id;
4326 d->btf->type_offs_cap = d->btf->nr_types;
4327 d->btf->hdr->type_len = p - d->btf->types_data;
4328 new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
4330 if (d->btf->type_offs_cap && !new_offs)
4332 d->btf->type_offs = new_offs;
4333 d->btf->hdr->str_off = d->btf->hdr->type_len;
4334 d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
4339 * Figure out final (deduplicated and compacted) type ID for provided original
4340 * `type_id` by first resolving it into corresponding canonical type ID and
4341 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
4342 * which is populated during compaction phase.
4344 static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
4346 struct btf_dedup *d = ctx;
4347 __u32 resolved_type_id, new_type_id;
4349 resolved_type_id = resolve_type_id(d, *type_id);
4350 new_type_id = d->hypot_map[resolved_type_id];
4351 if (new_type_id > BTF_MAX_NR_TYPES)
4354 *type_id = new_type_id;
4359 * Remap referenced type IDs into deduped type IDs.
4361 * After BTF types are deduplicated and compacted, their final type IDs may
4362 * differ from original ones. The map from original to a corresponding
4363 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
4364 * compaction phase. During remapping phase we are rewriting all type IDs
4365 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
4366 * their final deduped type IDs.
4368 static int btf_dedup_remap_types(struct btf_dedup *d)
4372 for (i = 0; i < d->btf->nr_types; i++) {
4373 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
4375 r = btf_type_visit_type_ids(t, btf_dedup_remap_type_id, d);
4383 r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
4391 * Probe few well-known locations for vmlinux kernel image and try to load BTF
4392 * data out of it to use for target BTF.
4394 struct btf *libbpf_find_kernel_btf(void)
4397 const char *path_fmt;
4400 /* try canonical vmlinux BTF through sysfs first */
4401 { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
4402 /* fall back to trying to find vmlinux ELF on disk otherwise */
4403 { "/boot/vmlinux-%1$s" },
4404 { "/lib/modules/%1$s/vmlinux-%1$s" },
4405 { "/lib/modules/%1$s/build/vmlinux" },
4406 { "/usr/lib/modules/%1$s/kernel/vmlinux" },
4407 { "/usr/lib/debug/boot/vmlinux-%1$s" },
4408 { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
4409 { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
4411 char path[PATH_MAX + 1];
4418 for (i = 0; i < ARRAY_SIZE(locations); i++) {
4419 snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
4421 if (access(path, R_OK))
4424 if (locations[i].raw_btf)
4425 btf = btf__parse_raw(path);
4427 btf = btf__parse_elf(path, NULL);
4429 pr_debug("loading kernel BTF '%s': %ld\n",
4430 path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
4437 pr_warn("failed to find valid kernel BTF\n");
4438 return ERR_PTR(-ESRCH);
4441 int btf_type_visit_type_ids(struct btf_type *t, type_id_visit_fn visit, void *ctx)
4445 switch (btf_kind(t)) {
4447 case BTF_KIND_FLOAT:
4452 case BTF_KIND_CONST:
4453 case BTF_KIND_VOLATILE:
4454 case BTF_KIND_RESTRICT:
4456 case BTF_KIND_TYPEDEF:
4459 return visit(&t->type, ctx);
4461 case BTF_KIND_ARRAY: {
4462 struct btf_array *a = btf_array(t);
4464 err = visit(&a->type, ctx);
4465 err = err ?: visit(&a->index_type, ctx);
4469 case BTF_KIND_STRUCT:
4470 case BTF_KIND_UNION: {
4471 struct btf_member *m = btf_members(t);
4473 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4474 err = visit(&m->type, ctx);
4481 case BTF_KIND_FUNC_PROTO: {
4482 struct btf_param *m = btf_params(t);
4484 err = visit(&t->type, ctx);
4487 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4488 err = visit(&m->type, ctx);
4495 case BTF_KIND_DATASEC: {
4496 struct btf_var_secinfo *m = btf_var_secinfos(t);
4498 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4499 err = visit(&m->type, ctx);
4511 int btf_type_visit_str_offs(struct btf_type *t, str_off_visit_fn visit, void *ctx)
4515 err = visit(&t->name_off, ctx);
4519 switch (btf_kind(t)) {
4520 case BTF_KIND_STRUCT:
4521 case BTF_KIND_UNION: {
4522 struct btf_member *m = btf_members(t);
4524 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4525 err = visit(&m->name_off, ctx);
4531 case BTF_KIND_ENUM: {
4532 struct btf_enum *m = btf_enum(t);
4534 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4535 err = visit(&m->name_off, ctx);
4541 case BTF_KIND_FUNC_PROTO: {
4542 struct btf_param *m = btf_params(t);
4544 for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4545 err = visit(&m->name_off, ctx);
4558 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
4560 const struct btf_ext_info *seg;
4561 struct btf_ext_info_sec *sec;
4564 seg = &btf_ext->func_info;
4565 for_each_btf_ext_sec(seg, sec) {
4566 struct bpf_func_info_min *rec;
4568 for_each_btf_ext_rec(seg, sec, i, rec) {
4569 err = visit(&rec->type_id, ctx);
4575 seg = &btf_ext->core_relo_info;
4576 for_each_btf_ext_sec(seg, sec) {
4577 struct bpf_core_relo *rec;
4579 for_each_btf_ext_rec(seg, sec, i, rec) {
4580 err = visit(&rec->type_id, ctx);
4589 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
4591 const struct btf_ext_info *seg;
4592 struct btf_ext_info_sec *sec;
4595 seg = &btf_ext->func_info;
4596 for_each_btf_ext_sec(seg, sec) {
4597 err = visit(&sec->sec_name_off, ctx);
4602 seg = &btf_ext->line_info;
4603 for_each_btf_ext_sec(seg, sec) {
4604 struct bpf_line_info_min *rec;
4606 err = visit(&sec->sec_name_off, ctx);
4610 for_each_btf_ext_rec(seg, sec, i, rec) {
4611 err = visit(&rec->file_name_off, ctx);
4614 err = visit(&rec->line_off, ctx);
4620 seg = &btf_ext->core_relo_info;
4621 for_each_btf_ext_sec(seg, sec) {
4622 struct bpf_core_relo *rec;
4624 err = visit(&sec->sec_name_off, ctx);
4628 for_each_btf_ext_rec(seg, sec, i, rec) {
4629 err = visit(&rec->access_str_off, ctx);