1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 /* Copyright (c) 2018 Facebook */
11 #include <linux/err.h>
12 #include <linux/btf.h>
17 #include "libbpf_internal.h"
20 #define BTF_MAX_NR_TYPES 0x7fffffff
21 #define BTF_MAX_STR_OFFSET 0x7fffffff
23 #define IS_MODIFIER(k) (((k) == BTF_KIND_TYPEDEF) || \
24 ((k) == BTF_KIND_VOLATILE) || \
25 ((k) == BTF_KIND_CONST) || \
26 ((k) == BTF_KIND_RESTRICT))
28 #define IS_VAR(k) ((k) == BTF_KIND_VAR)
30 static struct btf_type btf_void;
34 struct btf_header *hdr;
37 struct btf_type **types;
48 * info points to the individual info section (e.g. func_info and
49 * line_info) from the .BTF.ext. It does not include the __u32 rec_size.
58 struct btf_ext_header *hdr;
61 struct btf_ext_info func_info;
62 struct btf_ext_info line_info;
66 struct btf_ext_info_sec {
69 /* Followed by num_info * record_size number of bytes */
73 /* The minimum bpf_func_info checked by the loader */
74 struct bpf_func_info_min {
79 /* The minimum bpf_line_info checked by the loader */
80 struct bpf_line_info_min {
87 static inline __u64 ptr_to_u64(const void *ptr)
89 return (__u64) (unsigned long) ptr;
92 static int btf_add_type(struct btf *btf, struct btf_type *t)
94 if (btf->types_size - btf->nr_types < 2) {
95 struct btf_type **new_types;
96 __u32 expand_by, new_size;
98 if (btf->types_size == BTF_MAX_NR_TYPES)
101 expand_by = max(btf->types_size >> 2, 16);
102 new_size = min(BTF_MAX_NR_TYPES, btf->types_size + expand_by);
104 new_types = realloc(btf->types, sizeof(*new_types) * new_size);
108 if (btf->nr_types == 0)
109 new_types[0] = &btf_void;
111 btf->types = new_types;
112 btf->types_size = new_size;
115 btf->types[++(btf->nr_types)] = t;
120 static int btf_parse_hdr(struct btf *btf)
122 const struct btf_header *hdr = btf->hdr;
125 if (btf->data_size < sizeof(struct btf_header)) {
126 pr_debug("BTF header not found\n");
130 if (hdr->magic != BTF_MAGIC) {
131 pr_debug("Invalid BTF magic:%x\n", hdr->magic);
135 if (hdr->version != BTF_VERSION) {
136 pr_debug("Unsupported BTF version:%u\n", hdr->version);
141 pr_debug("Unsupported BTF flags:%x\n", hdr->flags);
145 meta_left = btf->data_size - sizeof(*hdr);
147 pr_debug("BTF has no data\n");
151 if (meta_left < hdr->type_off) {
152 pr_debug("Invalid BTF type section offset:%u\n", hdr->type_off);
156 if (meta_left < hdr->str_off) {
157 pr_debug("Invalid BTF string section offset:%u\n", hdr->str_off);
161 if (hdr->type_off >= hdr->str_off) {
162 pr_debug("BTF type section offset >= string section offset. No type?\n");
166 if (hdr->type_off & 0x02) {
167 pr_debug("BTF type section is not aligned to 4 bytes\n");
171 btf->nohdr_data = btf->hdr + 1;
176 static int btf_parse_str_sec(struct btf *btf)
178 const struct btf_header *hdr = btf->hdr;
179 const char *start = btf->nohdr_data + hdr->str_off;
180 const char *end = start + btf->hdr->str_len;
182 if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET ||
183 start[0] || end[-1]) {
184 pr_debug("Invalid BTF string section\n");
188 btf->strings = start;
193 static int btf_type_size(struct btf_type *t)
195 int base_size = sizeof(struct btf_type);
196 __u16 vlen = BTF_INFO_VLEN(t->info);
198 switch (BTF_INFO_KIND(t->info)) {
201 case BTF_KIND_VOLATILE:
202 case BTF_KIND_RESTRICT:
204 case BTF_KIND_TYPEDEF:
208 return base_size + sizeof(__u32);
210 return base_size + vlen * sizeof(struct btf_enum);
212 return base_size + sizeof(struct btf_array);
213 case BTF_KIND_STRUCT:
215 return base_size + vlen * sizeof(struct btf_member);
216 case BTF_KIND_FUNC_PROTO:
217 return base_size + vlen * sizeof(struct btf_param);
219 return base_size + sizeof(struct btf_var);
220 case BTF_KIND_DATASEC:
221 return base_size + vlen * sizeof(struct btf_var_secinfo);
223 pr_debug("Unsupported BTF_KIND:%u\n", BTF_INFO_KIND(t->info));
228 static int btf_parse_type_sec(struct btf *btf)
230 struct btf_header *hdr = btf->hdr;
231 void *nohdr_data = btf->nohdr_data;
232 void *next_type = nohdr_data + hdr->type_off;
233 void *end_type = nohdr_data + hdr->str_off;
235 while (next_type < end_type) {
236 struct btf_type *t = next_type;
240 type_size = btf_type_size(t);
243 next_type += type_size;
244 err = btf_add_type(btf, t);
252 __u32 btf__get_nr_types(const struct btf *btf)
254 return btf->nr_types;
257 const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
259 if (type_id > btf->nr_types)
262 return btf->types[type_id];
265 static bool btf_type_is_void(const struct btf_type *t)
267 return t == &btf_void || BTF_INFO_KIND(t->info) == BTF_KIND_FWD;
270 static bool btf_type_is_void_or_null(const struct btf_type *t)
272 return !t || btf_type_is_void(t);
275 #define MAX_RESOLVE_DEPTH 32
277 __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
279 const struct btf_array *array;
280 const struct btf_type *t;
285 t = btf__type_by_id(btf, type_id);
286 for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t);
288 switch (BTF_INFO_KIND(t->info)) {
290 case BTF_KIND_STRUCT:
293 case BTF_KIND_DATASEC:
297 size = sizeof(void *);
299 case BTF_KIND_TYPEDEF:
300 case BTF_KIND_VOLATILE:
302 case BTF_KIND_RESTRICT:
307 array = (const struct btf_array *)(t + 1);
308 if (nelems && array->nelems > UINT32_MAX / nelems)
310 nelems *= array->nelems;
311 type_id = array->type;
317 t = btf__type_by_id(btf, type_id);
324 if (nelems && size > UINT32_MAX / nelems)
327 return nelems * size;
330 int btf__resolve_type(const struct btf *btf, __u32 type_id)
332 const struct btf_type *t;
335 t = btf__type_by_id(btf, type_id);
336 while (depth < MAX_RESOLVE_DEPTH &&
337 !btf_type_is_void_or_null(t) &&
338 (IS_MODIFIER(BTF_INFO_KIND(t->info)) ||
339 IS_VAR(BTF_INFO_KIND(t->info)))) {
341 t = btf__type_by_id(btf, type_id);
345 if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
351 __s32 btf__find_by_name(const struct btf *btf, const char *type_name)
355 if (!strcmp(type_name, "void"))
358 for (i = 1; i <= btf->nr_types; i++) {
359 const struct btf_type *t = btf->types[i];
360 const char *name = btf__name_by_offset(btf, t->name_off);
362 if (name && !strcmp(type_name, name))
369 void btf__free(struct btf *btf)
382 struct btf *btf__new(__u8 *data, __u32 size)
387 btf = calloc(1, sizeof(struct btf));
389 return ERR_PTR(-ENOMEM);
393 btf->data = malloc(size);
399 memcpy(btf->data, data, size);
400 btf->data_size = size;
402 err = btf_parse_hdr(btf);
406 err = btf_parse_str_sec(btf);
410 err = btf_parse_type_sec(btf);
421 static bool btf_check_endianness(const GElf_Ehdr *ehdr)
423 #if __BYTE_ORDER == __LITTLE_ENDIAN
424 return ehdr->e_ident[EI_DATA] == ELFDATA2LSB;
425 #elif __BYTE_ORDER == __BIG_ENDIAN
426 return ehdr->e_ident[EI_DATA] == ELFDATA2MSB;
428 # error "Unrecognized __BYTE_ORDER__"
432 struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
434 Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
435 int err = 0, fd = -1, idx = 0;
436 struct btf *btf = NULL;
441 if (elf_version(EV_CURRENT) == EV_NONE) {
442 pr_warning("failed to init libelf for %s\n", path);
443 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
446 fd = open(path, O_RDONLY);
449 pr_warning("failed to open %s: %s\n", path, strerror(errno));
453 err = -LIBBPF_ERRNO__FORMAT;
455 elf = elf_begin(fd, ELF_C_READ, NULL);
457 pr_warning("failed to open %s as ELF file\n", path);
460 if (!gelf_getehdr(elf, &ehdr)) {
461 pr_warning("failed to get EHDR from %s\n", path);
464 if (!btf_check_endianness(&ehdr)) {
465 pr_warning("non-native ELF endianness is not supported\n");
468 if (!elf_rawdata(elf_getscn(elf, ehdr.e_shstrndx), NULL)) {
469 pr_warning("failed to get e_shstrndx from %s\n", path);
473 while ((scn = elf_nextscn(elf, scn)) != NULL) {
478 if (gelf_getshdr(scn, &sh) != &sh) {
479 pr_warning("failed to get section(%d) header from %s\n",
483 name = elf_strptr(elf, ehdr.e_shstrndx, sh.sh_name);
485 pr_warning("failed to get section(%d) name from %s\n",
489 if (strcmp(name, BTF_ELF_SEC) == 0) {
490 btf_data = elf_getdata(scn, 0);
492 pr_warning("failed to get section(%d, %s) data from %s\n",
497 } else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
498 btf_ext_data = elf_getdata(scn, 0);
500 pr_warning("failed to get section(%d, %s) data from %s\n",
514 btf = btf__new(btf_data->d_buf, btf_data->d_size);
518 if (btf_ext && btf_ext_data) {
519 *btf_ext = btf_ext__new(btf_ext_data->d_buf,
520 btf_ext_data->d_size);
521 if (IS_ERR(*btf_ext))
523 } else if (btf_ext) {
534 * btf is always parsed before btf_ext, so no need to clean up
535 * btf_ext, if btf loading failed
539 if (btf_ext && IS_ERR(*btf_ext)) {
541 err = PTR_ERR(*btf_ext);
547 static int compare_vsi_off(const void *_a, const void *_b)
549 const struct btf_var_secinfo *a = _a;
550 const struct btf_var_secinfo *b = _b;
552 return a->offset - b->offset;
555 static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
558 __u32 size = 0, off = 0, i, vars = BTF_INFO_VLEN(t->info);
559 const char *name = btf__name_by_offset(btf, t->name_off);
560 const struct btf_type *t_var;
561 struct btf_var_secinfo *vsi;
566 pr_debug("No name found in string section for DATASEC kind.\n");
570 ret = bpf_object__section_size(obj, name, &size);
571 if (ret || !size || (t->size && t->size != size)) {
572 pr_debug("Invalid size for section %s: %u bytes\n", name, size);
578 for (i = 0, vsi = (struct btf_var_secinfo *)(t + 1);
579 i < vars; i++, vsi++) {
580 t_var = btf__type_by_id(btf, vsi->type);
581 var = (struct btf_var *)(t_var + 1);
583 if (BTF_INFO_KIND(t_var->info) != BTF_KIND_VAR) {
584 pr_debug("Non-VAR type seen in section %s\n", name);
588 if (var->linkage == BTF_VAR_STATIC)
591 name = btf__name_by_offset(btf, t_var->name_off);
593 pr_debug("No name found in string section for VAR kind\n");
597 ret = bpf_object__variable_offset(obj, name, &off);
599 pr_debug("No offset found in symbol table for VAR %s\n", name);
606 qsort(t + 1, vars, sizeof(*vsi), compare_vsi_off);
610 int btf__finalize_data(struct bpf_object *obj, struct btf *btf)
615 for (i = 1; i <= btf->nr_types; i++) {
616 struct btf_type *t = btf->types[i];
618 /* Loader needs to fix up some of the things compiler
619 * couldn't get its hands on while emitting BTF. This
620 * is section size and global variable offset. We use
621 * the info from the ELF itself for this purpose.
623 if (BTF_INFO_KIND(t->info) == BTF_KIND_DATASEC) {
624 err = btf_fixup_datasec(obj, btf, t);
633 int btf__load(struct btf *btf)
635 __u32 log_buf_size = BPF_LOG_BUF_SIZE;
636 char *log_buf = NULL;
642 log_buf = malloc(log_buf_size);
648 btf->fd = bpf_load_btf(btf->data, btf->data_size,
649 log_buf, log_buf_size, false);
652 pr_warning("Error loading BTF: %s(%d)\n", strerror(errno), errno);
654 pr_warning("%s\n", log_buf);
663 int btf__fd(const struct btf *btf)
668 const void *btf__get_raw_data(const struct btf *btf, __u32 *size)
670 *size = btf->data_size;
674 const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
676 if (offset < btf->hdr->str_len)
677 return &btf->strings[offset];
682 int btf__get_from_id(__u32 id, struct btf **btf)
684 struct bpf_btf_info btf_info = { 0 };
685 __u32 len = sizeof(btf_info);
693 btf_fd = bpf_btf_get_fd_by_id(id);
697 /* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
698 * let's start with a sane default - 4KiB here - and resize it only if
699 * bpf_obj_get_info_by_fd() needs a bigger buffer.
701 btf_info.btf_size = 4096;
702 last_size = btf_info.btf_size;
703 ptr = malloc(last_size);
709 memset(ptr, 0, last_size);
710 btf_info.btf = ptr_to_u64(ptr);
711 err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
713 if (!err && btf_info.btf_size > last_size) {
716 last_size = btf_info.btf_size;
717 temp_ptr = realloc(ptr, last_size);
723 memset(ptr, 0, last_size);
724 btf_info.btf = ptr_to_u64(ptr);
725 err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
728 if (err || btf_info.btf_size > last_size) {
733 *btf = btf__new((__u8 *)(long)btf_info.btf, btf_info.btf_size);
746 int btf__get_map_kv_tids(const struct btf *btf, const char *map_name,
747 __u32 expected_key_size, __u32 expected_value_size,
748 __u32 *key_type_id, __u32 *value_type_id)
750 const struct btf_type *container_type;
751 const struct btf_member *key, *value;
752 const size_t max_name = 256;
753 char container_name[max_name];
754 __s64 key_size, value_size;
757 if (snprintf(container_name, max_name, "____btf_map_%s", map_name) ==
759 pr_warning("map:%s length of '____btf_map_%s' is too long\n",
764 container_id = btf__find_by_name(btf, container_name);
765 if (container_id < 0) {
766 pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
767 map_name, container_name);
771 container_type = btf__type_by_id(btf, container_id);
772 if (!container_type) {
773 pr_warning("map:%s cannot find BTF type for container_id:%u\n",
774 map_name, container_id);
778 if (BTF_INFO_KIND(container_type->info) != BTF_KIND_STRUCT ||
779 BTF_INFO_VLEN(container_type->info) < 2) {
780 pr_warning("map:%s container_name:%s is an invalid container struct\n",
781 map_name, container_name);
785 key = (struct btf_member *)(container_type + 1);
788 key_size = btf__resolve_size(btf, key->type);
790 pr_warning("map:%s invalid BTF key_type_size\n", map_name);
794 if (expected_key_size != key_size) {
795 pr_warning("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
796 map_name, (__u32)key_size, expected_key_size);
800 value_size = btf__resolve_size(btf, value->type);
801 if (value_size < 0) {
802 pr_warning("map:%s invalid BTF value_type_size\n", map_name);
806 if (expected_value_size != value_size) {
807 pr_warning("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
808 map_name, (__u32)value_size, expected_value_size);
812 *key_type_id = key->type;
813 *value_type_id = value->type;
818 struct btf_ext_sec_setup_param {
822 struct btf_ext_info *ext_info;
826 static int btf_ext_setup_info(struct btf_ext *btf_ext,
827 struct btf_ext_sec_setup_param *ext_sec)
829 const struct btf_ext_info_sec *sinfo;
830 struct btf_ext_info *ext_info;
831 __u32 info_left, record_size;
832 /* The start of the info sec (including the __u32 record_size). */
835 if (ext_sec->off & 0x03) {
836 pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
841 info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
842 info_left = ext_sec->len;
844 if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
845 pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
846 ext_sec->desc, ext_sec->off, ext_sec->len);
850 /* At least a record size */
851 if (info_left < sizeof(__u32)) {
852 pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
856 /* The record size needs to meet the minimum standard */
857 record_size = *(__u32 *)info;
858 if (record_size < ext_sec->min_rec_size ||
859 record_size & 0x03) {
860 pr_debug("%s section in .BTF.ext has invalid record size %u\n",
861 ext_sec->desc, record_size);
865 sinfo = info + sizeof(__u32);
866 info_left -= sizeof(__u32);
868 /* If no records, return failure now so .BTF.ext won't be used. */
870 pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
875 unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
876 __u64 total_record_size;
879 if (info_left < sec_hdrlen) {
880 pr_debug("%s section header is not found in .BTF.ext\n",
885 num_records = sinfo->num_info;
886 if (num_records == 0) {
887 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
892 total_record_size = sec_hdrlen +
893 (__u64)num_records * record_size;
894 if (info_left < total_record_size) {
895 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
900 info_left -= total_record_size;
901 sinfo = (void *)sinfo + total_record_size;
904 ext_info = ext_sec->ext_info;
905 ext_info->len = ext_sec->len - sizeof(__u32);
906 ext_info->rec_size = record_size;
907 ext_info->info = info + sizeof(__u32);
912 static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
914 struct btf_ext_sec_setup_param param = {
915 .off = btf_ext->hdr->func_info_off,
916 .len = btf_ext->hdr->func_info_len,
917 .min_rec_size = sizeof(struct bpf_func_info_min),
918 .ext_info = &btf_ext->func_info,
922 return btf_ext_setup_info(btf_ext, ¶m);
925 static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
927 struct btf_ext_sec_setup_param param = {
928 .off = btf_ext->hdr->line_info_off,
929 .len = btf_ext->hdr->line_info_len,
930 .min_rec_size = sizeof(struct bpf_line_info_min),
931 .ext_info = &btf_ext->line_info,
935 return btf_ext_setup_info(btf_ext, ¶m);
938 static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
940 const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
942 if (data_size < offsetof(struct btf_ext_header, func_info_off) ||
943 data_size < hdr->hdr_len) {
944 pr_debug("BTF.ext header not found");
948 if (hdr->magic != BTF_MAGIC) {
949 pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
953 if (hdr->version != BTF_VERSION) {
954 pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
959 pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
963 if (data_size == hdr->hdr_len) {
964 pr_debug("BTF.ext has no data\n");
971 void btf_ext__free(struct btf_ext *btf_ext)
979 struct btf_ext *btf_ext__new(__u8 *data, __u32 size)
981 struct btf_ext *btf_ext;
984 err = btf_ext_parse_hdr(data, size);
988 btf_ext = calloc(1, sizeof(struct btf_ext));
990 return ERR_PTR(-ENOMEM);
992 btf_ext->data_size = size;
993 btf_ext->data = malloc(size);
994 if (!btf_ext->data) {
998 memcpy(btf_ext->data, data, size);
1000 err = btf_ext_setup_func_info(btf_ext);
1004 err = btf_ext_setup_line_info(btf_ext);
1010 btf_ext__free(btf_ext);
1011 return ERR_PTR(err);
1017 const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
1019 *size = btf_ext->data_size;
1020 return btf_ext->data;
1023 static int btf_ext_reloc_info(const struct btf *btf,
1024 const struct btf_ext_info *ext_info,
1025 const char *sec_name, __u32 insns_cnt,
1026 void **info, __u32 *cnt)
1028 __u32 sec_hdrlen = sizeof(struct btf_ext_info_sec);
1029 __u32 i, record_size, existing_len, records_len;
1030 struct btf_ext_info_sec *sinfo;
1031 const char *info_sec_name;
1035 record_size = ext_info->rec_size;
1036 sinfo = ext_info->info;
1037 remain_len = ext_info->len;
1038 while (remain_len > 0) {
1039 records_len = sinfo->num_info * record_size;
1040 info_sec_name = btf__name_by_offset(btf, sinfo->sec_name_off);
1041 if (strcmp(info_sec_name, sec_name)) {
1042 remain_len -= sec_hdrlen + records_len;
1043 sinfo = (void *)sinfo + sec_hdrlen + records_len;
1047 existing_len = (*cnt) * record_size;
1048 data = realloc(*info, existing_len + records_len);
1052 memcpy(data + existing_len, sinfo->data, records_len);
1053 /* adjust insn_off only, the rest data will be passed
1056 for (i = 0; i < sinfo->num_info; i++) {
1059 insn_off = data + existing_len + (i * record_size);
1060 *insn_off = *insn_off / sizeof(struct bpf_insn) +
1064 *cnt += sinfo->num_info;
1071 int btf_ext__reloc_func_info(const struct btf *btf,
1072 const struct btf_ext *btf_ext,
1073 const char *sec_name, __u32 insns_cnt,
1074 void **func_info, __u32 *cnt)
1076 return btf_ext_reloc_info(btf, &btf_ext->func_info, sec_name,
1077 insns_cnt, func_info, cnt);
1080 int btf_ext__reloc_line_info(const struct btf *btf,
1081 const struct btf_ext *btf_ext,
1082 const char *sec_name, __u32 insns_cnt,
1083 void **line_info, __u32 *cnt)
1085 return btf_ext_reloc_info(btf, &btf_ext->line_info, sec_name,
1086 insns_cnt, line_info, cnt);
1089 __u32 btf_ext__func_info_rec_size(const struct btf_ext *btf_ext)
1091 return btf_ext->func_info.rec_size;
1094 __u32 btf_ext__line_info_rec_size(const struct btf_ext *btf_ext)
1096 return btf_ext->line_info.rec_size;
1101 static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
1102 const struct btf_dedup_opts *opts);
1103 static void btf_dedup_free(struct btf_dedup *d);
1104 static int btf_dedup_strings(struct btf_dedup *d);
1105 static int btf_dedup_prim_types(struct btf_dedup *d);
1106 static int btf_dedup_struct_types(struct btf_dedup *d);
1107 static int btf_dedup_ref_types(struct btf_dedup *d);
1108 static int btf_dedup_compact_types(struct btf_dedup *d);
1109 static int btf_dedup_remap_types(struct btf_dedup *d);
1112 * Deduplicate BTF types and strings.
1114 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
1115 * section with all BTF type descriptors and string data. It overwrites that
1116 * memory in-place with deduplicated types and strings without any loss of
1117 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
1118 * is provided, all the strings referenced from .BTF.ext section are honored
1119 * and updated to point to the right offsets after deduplication.
1121 * If function returns with error, type/string data might be garbled and should
1124 * More verbose and detailed description of both problem btf_dedup is solving,
1125 * as well as solution could be found at:
1126 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
1128 * Problem description and justification
1129 * =====================================
1131 * BTF type information is typically emitted either as a result of conversion
1132 * from DWARF to BTF or directly by compiler. In both cases, each compilation
1133 * unit contains information about a subset of all the types that are used
1134 * in an application. These subsets are frequently overlapping and contain a lot
1135 * of duplicated information when later concatenated together into a single
1136 * binary. This algorithm ensures that each unique type is represented by single
1137 * BTF type descriptor, greatly reducing resulting size of BTF data.
1139 * Compilation unit isolation and subsequent duplication of data is not the only
1140 * problem. The same type hierarchy (e.g., struct and all the type that struct
1141 * references) in different compilation units can be represented in BTF to
1142 * various degrees of completeness (or, rather, incompleteness) due to
1143 * struct/union forward declarations.
1145 * Let's take a look at an example, that we'll use to better understand the
1146 * problem (and solution). Suppose we have two compilation units, each using
1147 * same `struct S`, but each of them having incomplete type information about
1176 * In case of CU #1, BTF data will know only that `struct B` exist (but no
1177 * more), but will know the complete type information about `struct A`. While
1178 * for CU #2, it will know full type information about `struct B`, but will
1179 * only know about forward declaration of `struct A` (in BTF terms, it will
1180 * have `BTF_KIND_FWD` type descriptor with name `B`).
1182 * This compilation unit isolation means that it's possible that there is no
1183 * single CU with complete type information describing structs `S`, `A`, and
1184 * `B`. Also, we might get tons of duplicated and redundant type information.
1186 * Additional complication we need to keep in mind comes from the fact that
1187 * types, in general, can form graphs containing cycles, not just DAGs.
1189 * While algorithm does deduplication, it also merges and resolves type
1190 * information (unless disabled throught `struct btf_opts`), whenever possible.
1191 * E.g., in the example above with two compilation units having partial type
1192 * information for structs `A` and `B`, the output of algorithm will emit
1193 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
1194 * (as well as type information for `int` and pointers), as if they were defined
1195 * in a single compilation unit as:
1215 * Algorithm completes its work in 6 separate passes:
1217 * 1. Strings deduplication.
1218 * 2. Primitive types deduplication (int, enum, fwd).
1219 * 3. Struct/union types deduplication.
1220 * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
1221 * protos, and const/volatile/restrict modifiers).
1222 * 5. Types compaction.
1223 * 6. Types remapping.
1225 * Algorithm determines canonical type descriptor, which is a single
1226 * representative type for each truly unique type. This canonical type is the
1227 * one that will go into final deduplicated BTF type information. For
1228 * struct/unions, it is also the type that algorithm will merge additional type
1229 * information into (while resolving FWDs), as it discovers it from data in
1230 * other CUs. Each input BTF type eventually gets either mapped to itself, if
1231 * that type is canonical, or to some other type, if that type is equivalent
1232 * and was chosen as canonical representative. This mapping is stored in
1233 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
1234 * FWD type got resolved to.
1236 * To facilitate fast discovery of canonical types, we also maintain canonical
1237 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
1238 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
1239 * that match that signature. With sufficiently good choice of type signature
1240 * hashing function, we can limit number of canonical types for each unique type
1241 * signature to a very small number, allowing to find canonical type for any
1242 * duplicated type very quickly.
1244 * Struct/union deduplication is the most critical part and algorithm for
1245 * deduplicating structs/unions is described in greater details in comments for
1246 * `btf_dedup_is_equiv` function.
1248 int btf__dedup(struct btf *btf, struct btf_ext *btf_ext,
1249 const struct btf_dedup_opts *opts)
1251 struct btf_dedup *d = btf_dedup_new(btf, btf_ext, opts);
1255 pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
1259 err = btf_dedup_strings(d);
1261 pr_debug("btf_dedup_strings failed:%d\n", err);
1264 err = btf_dedup_prim_types(d);
1266 pr_debug("btf_dedup_prim_types failed:%d\n", err);
1269 err = btf_dedup_struct_types(d);
1271 pr_debug("btf_dedup_struct_types failed:%d\n", err);
1274 err = btf_dedup_ref_types(d);
1276 pr_debug("btf_dedup_ref_types failed:%d\n", err);
1279 err = btf_dedup_compact_types(d);
1281 pr_debug("btf_dedup_compact_types failed:%d\n", err);
1284 err = btf_dedup_remap_types(d);
1286 pr_debug("btf_dedup_remap_types failed:%d\n", err);
1295 #define BTF_UNPROCESSED_ID ((__u32)-1)
1296 #define BTF_IN_PROGRESS_ID ((__u32)-2)
1299 /* .BTF section to be deduped in-place */
1302 * Optional .BTF.ext section. When provided, any strings referenced
1303 * from it will be taken into account when deduping strings
1305 struct btf_ext *btf_ext;
1307 * This is a map from any type's signature hash to a list of possible
1308 * canonical representative type candidates. Hash collisions are
1309 * ignored, so even types of various kinds can share same list of
1310 * candidates, which is fine because we rely on subsequent
1311 * btf_xxx_equal() checks to authoritatively verify type equality.
1313 struct hashmap *dedup_table;
1314 /* Canonical types map */
1316 /* Hypothetical mapping, used during type graph equivalence checks */
1321 /* Various option modifying behavior of algorithm */
1322 struct btf_dedup_opts opts;
1325 struct btf_str_ptr {
1331 struct btf_str_ptrs {
1332 struct btf_str_ptr *ptrs;
1338 static long hash_combine(long h, long value)
1340 return h * 31 + value;
1343 #define for_each_dedup_cand(d, node, hash) \
1344 hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
1346 static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
1348 return hashmap__append(d->dedup_table,
1349 (void *)hash, (void *)(long)type_id);
1352 static int btf_dedup_hypot_map_add(struct btf_dedup *d,
1353 __u32 from_id, __u32 to_id)
1355 if (d->hypot_cnt == d->hypot_cap) {
1358 d->hypot_cap += max(16, d->hypot_cap / 2);
1359 new_list = realloc(d->hypot_list, sizeof(__u32) * d->hypot_cap);
1362 d->hypot_list = new_list;
1364 d->hypot_list[d->hypot_cnt++] = from_id;
1365 d->hypot_map[from_id] = to_id;
1369 static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
1373 for (i = 0; i < d->hypot_cnt; i++)
1374 d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
1378 static void btf_dedup_free(struct btf_dedup *d)
1380 hashmap__free(d->dedup_table);
1381 d->dedup_table = NULL;
1387 d->hypot_map = NULL;
1389 free(d->hypot_list);
1390 d->hypot_list = NULL;
1395 static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
1400 static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
1405 static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
1410 static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
1411 const struct btf_dedup_opts *opts)
1413 struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
1414 hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
1418 return ERR_PTR(-ENOMEM);
1420 d->opts.dont_resolve_fwds = opts && opts->dont_resolve_fwds;
1421 /* dedup_table_size is now used only to force collisions in tests */
1422 if (opts && opts->dedup_table_size == 1)
1423 hash_fn = btf_dedup_collision_hash_fn;
1426 d->btf_ext = btf_ext;
1428 d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
1429 if (IS_ERR(d->dedup_table)) {
1430 err = PTR_ERR(d->dedup_table);
1431 d->dedup_table = NULL;
1435 d->map = malloc(sizeof(__u32) * (1 + btf->nr_types));
1440 /* special BTF "void" type is made canonical immediately */
1442 for (i = 1; i <= btf->nr_types; i++) {
1443 struct btf_type *t = d->btf->types[i];
1444 __u16 kind = BTF_INFO_KIND(t->info);
1446 /* VAR and DATASEC are never deduped and are self-canonical */
1447 if (kind == BTF_KIND_VAR || kind == BTF_KIND_DATASEC)
1450 d->map[i] = BTF_UNPROCESSED_ID;
1453 d->hypot_map = malloc(sizeof(__u32) * (1 + btf->nr_types));
1454 if (!d->hypot_map) {
1458 for (i = 0; i <= btf->nr_types; i++)
1459 d->hypot_map[i] = BTF_UNPROCESSED_ID;
1464 return ERR_PTR(err);
1470 typedef int (*str_off_fn_t)(__u32 *str_off_ptr, void *ctx);
1473 * Iterate over all possible places in .BTF and .BTF.ext that can reference
1474 * string and pass pointer to it to a provided callback `fn`.
1476 static int btf_for_each_str_off(struct btf_dedup *d, str_off_fn_t fn, void *ctx)
1478 void *line_data_cur, *line_data_end;
1479 int i, j, r, rec_size;
1482 for (i = 1; i <= d->btf->nr_types; i++) {
1483 t = d->btf->types[i];
1484 r = fn(&t->name_off, ctx);
1488 switch (BTF_INFO_KIND(t->info)) {
1489 case BTF_KIND_STRUCT:
1490 case BTF_KIND_UNION: {
1491 struct btf_member *m = (struct btf_member *)(t + 1);
1492 __u16 vlen = BTF_INFO_VLEN(t->info);
1494 for (j = 0; j < vlen; j++) {
1495 r = fn(&m->name_off, ctx);
1502 case BTF_KIND_ENUM: {
1503 struct btf_enum *m = (struct btf_enum *)(t + 1);
1504 __u16 vlen = BTF_INFO_VLEN(t->info);
1506 for (j = 0; j < vlen; j++) {
1507 r = fn(&m->name_off, ctx);
1514 case BTF_KIND_FUNC_PROTO: {
1515 struct btf_param *m = (struct btf_param *)(t + 1);
1516 __u16 vlen = BTF_INFO_VLEN(t->info);
1518 for (j = 0; j < vlen; j++) {
1519 r = fn(&m->name_off, ctx);
1534 line_data_cur = d->btf_ext->line_info.info;
1535 line_data_end = d->btf_ext->line_info.info + d->btf_ext->line_info.len;
1536 rec_size = d->btf_ext->line_info.rec_size;
1538 while (line_data_cur < line_data_end) {
1539 struct btf_ext_info_sec *sec = line_data_cur;
1540 struct bpf_line_info_min *line_info;
1541 __u32 num_info = sec->num_info;
1543 r = fn(&sec->sec_name_off, ctx);
1547 line_data_cur += sizeof(struct btf_ext_info_sec);
1548 for (i = 0; i < num_info; i++) {
1549 line_info = line_data_cur;
1550 r = fn(&line_info->file_name_off, ctx);
1553 r = fn(&line_info->line_off, ctx);
1556 line_data_cur += rec_size;
1563 static int str_sort_by_content(const void *a1, const void *a2)
1565 const struct btf_str_ptr *p1 = a1;
1566 const struct btf_str_ptr *p2 = a2;
1568 return strcmp(p1->str, p2->str);
1571 static int str_sort_by_offset(const void *a1, const void *a2)
1573 const struct btf_str_ptr *p1 = a1;
1574 const struct btf_str_ptr *p2 = a2;
1576 if (p1->str != p2->str)
1577 return p1->str < p2->str ? -1 : 1;
1581 static int btf_dedup_str_ptr_cmp(const void *str_ptr, const void *pelem)
1583 const struct btf_str_ptr *p = pelem;
1585 if (str_ptr != p->str)
1586 return (const char *)str_ptr < p->str ? -1 : 1;
1590 static int btf_str_mark_as_used(__u32 *str_off_ptr, void *ctx)
1592 struct btf_str_ptrs *strs;
1593 struct btf_str_ptr *s;
1595 if (*str_off_ptr == 0)
1599 s = bsearch(strs->data + *str_off_ptr, strs->ptrs, strs->cnt,
1600 sizeof(struct btf_str_ptr), btf_dedup_str_ptr_cmp);
1607 static int btf_str_remap_offset(__u32 *str_off_ptr, void *ctx)
1609 struct btf_str_ptrs *strs;
1610 struct btf_str_ptr *s;
1612 if (*str_off_ptr == 0)
1616 s = bsearch(strs->data + *str_off_ptr, strs->ptrs, strs->cnt,
1617 sizeof(struct btf_str_ptr), btf_dedup_str_ptr_cmp);
1620 *str_off_ptr = s->new_off;
1625 * Dedup string and filter out those that are not referenced from either .BTF
1626 * or .BTF.ext (if provided) sections.
1628 * This is done by building index of all strings in BTF's string section,
1629 * then iterating over all entities that can reference strings (e.g., type
1630 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
1631 * strings as used. After that all used strings are deduped and compacted into
1632 * sequential blob of memory and new offsets are calculated. Then all the string
1633 * references are iterated again and rewritten using new offsets.
1635 static int btf_dedup_strings(struct btf_dedup *d)
1637 const struct btf_header *hdr = d->btf->hdr;
1638 char *start = (char *)d->btf->nohdr_data + hdr->str_off;
1639 char *end = start + d->btf->hdr->str_len;
1640 char *p = start, *tmp_strs = NULL;
1641 struct btf_str_ptrs strs = {
1647 int i, j, err = 0, grp_idx;
1650 /* build index of all strings */
1652 if (strs.cnt + 1 > strs.cap) {
1653 struct btf_str_ptr *new_ptrs;
1655 strs.cap += max(strs.cnt / 2, 16);
1656 new_ptrs = realloc(strs.ptrs,
1657 sizeof(strs.ptrs[0]) * strs.cap);
1662 strs.ptrs = new_ptrs;
1665 strs.ptrs[strs.cnt].str = p;
1666 strs.ptrs[strs.cnt].used = false;
1672 /* temporary storage for deduplicated strings */
1673 tmp_strs = malloc(d->btf->hdr->str_len);
1679 /* mark all used strings */
1680 strs.ptrs[0].used = true;
1681 err = btf_for_each_str_off(d, btf_str_mark_as_used, &strs);
1685 /* sort strings by context, so that we can identify duplicates */
1686 qsort(strs.ptrs, strs.cnt, sizeof(strs.ptrs[0]), str_sort_by_content);
1689 * iterate groups of equal strings and if any instance in a group was
1690 * referenced, emit single instance and remember new offset
1694 grp_used = strs.ptrs[0].used;
1695 /* iterate past end to avoid code duplication after loop */
1696 for (i = 1; i <= strs.cnt; i++) {
1698 * when i == strs.cnt, we want to skip string comparison and go
1699 * straight to handling last group of strings (otherwise we'd
1700 * need to handle last group after the loop w/ duplicated code)
1703 !strcmp(strs.ptrs[i].str, strs.ptrs[grp_idx].str)) {
1704 grp_used = grp_used || strs.ptrs[i].used;
1709 * this check would have been required after the loop to handle
1710 * last group of strings, but due to <= condition in a loop
1711 * we avoid that duplication
1714 int new_off = p - tmp_strs;
1715 __u32 len = strlen(strs.ptrs[grp_idx].str);
1717 memmove(p, strs.ptrs[grp_idx].str, len + 1);
1718 for (j = grp_idx; j < i; j++)
1719 strs.ptrs[j].new_off = new_off;
1725 grp_used = strs.ptrs[i].used;
1729 /* replace original strings with deduped ones */
1730 d->btf->hdr->str_len = p - tmp_strs;
1731 memmove(start, tmp_strs, d->btf->hdr->str_len);
1732 end = start + d->btf->hdr->str_len;
1734 /* restore original order for further binary search lookups */
1735 qsort(strs.ptrs, strs.cnt, sizeof(strs.ptrs[0]), str_sort_by_offset);
1737 /* remap string offsets */
1738 err = btf_for_each_str_off(d, btf_str_remap_offset, &strs);
1742 d->btf->hdr->str_len = end - start;
1750 static long btf_hash_common(struct btf_type *t)
1754 h = hash_combine(0, t->name_off);
1755 h = hash_combine(h, t->info);
1756 h = hash_combine(h, t->size);
1760 static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
1762 return t1->name_off == t2->name_off &&
1763 t1->info == t2->info &&
1764 t1->size == t2->size;
1767 /* Calculate type signature hash of INT. */
1768 static long btf_hash_int(struct btf_type *t)
1770 __u32 info = *(__u32 *)(t + 1);
1773 h = btf_hash_common(t);
1774 h = hash_combine(h, info);
1778 /* Check structural equality of two INTs. */
1779 static bool btf_equal_int(struct btf_type *t1, struct btf_type *t2)
1783 if (!btf_equal_common(t1, t2))
1785 info1 = *(__u32 *)(t1 + 1);
1786 info2 = *(__u32 *)(t2 + 1);
1787 return info1 == info2;
1790 /* Calculate type signature hash of ENUM. */
1791 static long btf_hash_enum(struct btf_type *t)
1795 /* don't hash vlen and enum members to support enum fwd resolving */
1796 h = hash_combine(0, t->name_off);
1797 h = hash_combine(h, t->info & ~0xffff);
1798 h = hash_combine(h, t->size);
1802 /* Check structural equality of two ENUMs. */
1803 static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
1805 struct btf_enum *m1, *m2;
1809 if (!btf_equal_common(t1, t2))
1812 vlen = BTF_INFO_VLEN(t1->info);
1813 m1 = (struct btf_enum *)(t1 + 1);
1814 m2 = (struct btf_enum *)(t2 + 1);
1815 for (i = 0; i < vlen; i++) {
1816 if (m1->name_off != m2->name_off || m1->val != m2->val)
1824 static inline bool btf_is_enum_fwd(struct btf_type *t)
1826 return BTF_INFO_KIND(t->info) == BTF_KIND_ENUM &&
1827 BTF_INFO_VLEN(t->info) == 0;
1830 static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
1832 if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
1833 return btf_equal_enum(t1, t2);
1834 /* ignore vlen when comparing */
1835 return t1->name_off == t2->name_off &&
1836 (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
1837 t1->size == t2->size;
1841 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
1842 * as referenced type IDs equivalence is established separately during type
1843 * graph equivalence check algorithm.
1845 static long btf_hash_struct(struct btf_type *t)
1847 struct btf_member *member = (struct btf_member *)(t + 1);
1848 __u32 vlen = BTF_INFO_VLEN(t->info);
1849 long h = btf_hash_common(t);
1852 for (i = 0; i < vlen; i++) {
1853 h = hash_combine(h, member->name_off);
1854 h = hash_combine(h, member->offset);
1855 /* no hashing of referenced type ID, it can be unresolved yet */
1862 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
1863 * IDs. This check is performed during type graph equivalence check and
1864 * referenced types equivalence is checked separately.
1866 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
1868 struct btf_member *m1, *m2;
1872 if (!btf_equal_common(t1, t2))
1875 vlen = BTF_INFO_VLEN(t1->info);
1876 m1 = (struct btf_member *)(t1 + 1);
1877 m2 = (struct btf_member *)(t2 + 1);
1878 for (i = 0; i < vlen; i++) {
1879 if (m1->name_off != m2->name_off || m1->offset != m2->offset)
1888 * Calculate type signature hash of ARRAY, including referenced type IDs,
1889 * under assumption that they were already resolved to canonical type IDs and
1890 * are not going to change.
1892 static long btf_hash_array(struct btf_type *t)
1894 struct btf_array *info = (struct btf_array *)(t + 1);
1895 long h = btf_hash_common(t);
1897 h = hash_combine(h, info->type);
1898 h = hash_combine(h, info->index_type);
1899 h = hash_combine(h, info->nelems);
1904 * Check exact equality of two ARRAYs, taking into account referenced
1905 * type IDs, under assumption that they were already resolved to canonical
1906 * type IDs and are not going to change.
1907 * This function is called during reference types deduplication to compare
1908 * ARRAY to potential canonical representative.
1910 static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
1912 struct btf_array *info1, *info2;
1914 if (!btf_equal_common(t1, t2))
1917 info1 = (struct btf_array *)(t1 + 1);
1918 info2 = (struct btf_array *)(t2 + 1);
1919 return info1->type == info2->type &&
1920 info1->index_type == info2->index_type &&
1921 info1->nelems == info2->nelems;
1925 * Check structural compatibility of two ARRAYs, ignoring referenced type
1926 * IDs. This check is performed during type graph equivalence check and
1927 * referenced types equivalence is checked separately.
1929 static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
1931 struct btf_array *info1, *info2;
1933 if (!btf_equal_common(t1, t2))
1936 info1 = (struct btf_array *)(t1 + 1);
1937 info2 = (struct btf_array *)(t2 + 1);
1938 return info1->nelems == info2->nelems;
1942 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
1943 * under assumption that they were already resolved to canonical type IDs and
1944 * are not going to change.
1946 static long btf_hash_fnproto(struct btf_type *t)
1948 struct btf_param *member = (struct btf_param *)(t + 1);
1949 __u16 vlen = BTF_INFO_VLEN(t->info);
1950 long h = btf_hash_common(t);
1953 for (i = 0; i < vlen; i++) {
1954 h = hash_combine(h, member->name_off);
1955 h = hash_combine(h, member->type);
1962 * Check exact equality of two FUNC_PROTOs, taking into account referenced
1963 * type IDs, under assumption that they were already resolved to canonical
1964 * type IDs and are not going to change.
1965 * This function is called during reference types deduplication to compare
1966 * FUNC_PROTO to potential canonical representative.
1968 static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
1970 struct btf_param *m1, *m2;
1974 if (!btf_equal_common(t1, t2))
1977 vlen = BTF_INFO_VLEN(t1->info);
1978 m1 = (struct btf_param *)(t1 + 1);
1979 m2 = (struct btf_param *)(t2 + 1);
1980 for (i = 0; i < vlen; i++) {
1981 if (m1->name_off != m2->name_off || m1->type != m2->type)
1990 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
1991 * IDs. This check is performed during type graph equivalence check and
1992 * referenced types equivalence is checked separately.
1994 static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
1996 struct btf_param *m1, *m2;
2000 /* skip return type ID */
2001 if (t1->name_off != t2->name_off || t1->info != t2->info)
2004 vlen = BTF_INFO_VLEN(t1->info);
2005 m1 = (struct btf_param *)(t1 + 1);
2006 m2 = (struct btf_param *)(t2 + 1);
2007 for (i = 0; i < vlen; i++) {
2008 if (m1->name_off != m2->name_off)
2017 * Deduplicate primitive types, that can't reference other types, by calculating
2018 * their type signature hash and comparing them with any possible canonical
2019 * candidate. If no canonical candidate matches, type itself is marked as
2020 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
2022 static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
2024 struct btf_type *t = d->btf->types[type_id];
2025 struct hashmap_entry *hash_entry;
2026 struct btf_type *cand;
2027 /* if we don't find equivalent type, then we are canonical */
2028 __u32 new_id = type_id;
2032 switch (BTF_INFO_KIND(t->info)) {
2033 case BTF_KIND_CONST:
2034 case BTF_KIND_VOLATILE:
2035 case BTF_KIND_RESTRICT:
2037 case BTF_KIND_TYPEDEF:
2038 case BTF_KIND_ARRAY:
2039 case BTF_KIND_STRUCT:
2040 case BTF_KIND_UNION:
2042 case BTF_KIND_FUNC_PROTO:
2044 case BTF_KIND_DATASEC:
2048 h = btf_hash_int(t);
2049 for_each_dedup_cand(d, hash_entry, h) {
2050 cand_id = (__u32)(long)hash_entry->value;
2051 cand = d->btf->types[cand_id];
2052 if (btf_equal_int(t, cand)) {
2060 h = btf_hash_enum(t);
2061 for_each_dedup_cand(d, hash_entry, h) {
2062 cand_id = (__u32)(long)hash_entry->value;
2063 cand = d->btf->types[cand_id];
2064 if (btf_equal_enum(t, cand)) {
2068 if (d->opts.dont_resolve_fwds)
2070 if (btf_compat_enum(t, cand)) {
2071 if (btf_is_enum_fwd(t)) {
2072 /* resolve fwd to full enum */
2076 /* resolve canonical enum fwd to full enum */
2077 d->map[cand_id] = type_id;
2083 h = btf_hash_common(t);
2084 for_each_dedup_cand(d, hash_entry, h) {
2085 cand_id = (__u32)(long)hash_entry->value;
2086 cand = d->btf->types[cand_id];
2087 if (btf_equal_common(t, cand)) {
2098 d->map[type_id] = new_id;
2099 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
2105 static int btf_dedup_prim_types(struct btf_dedup *d)
2109 for (i = 1; i <= d->btf->nr_types; i++) {
2110 err = btf_dedup_prim_type(d, i);
2118 * Check whether type is already mapped into canonical one (could be to itself).
2120 static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
2122 return d->map[type_id] <= BTF_MAX_NR_TYPES;
2126 * Resolve type ID into its canonical type ID, if any; otherwise return original
2127 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
2128 * STRUCT/UNION link and resolve it into canonical type ID as well.
2130 static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
2132 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
2133 type_id = d->map[type_id];
2138 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
2141 static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
2143 __u32 orig_type_id = type_id;
2145 if (BTF_INFO_KIND(d->btf->types[type_id]->info) != BTF_KIND_FWD)
2148 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
2149 type_id = d->map[type_id];
2151 if (BTF_INFO_KIND(d->btf->types[type_id]->info) != BTF_KIND_FWD)
2154 return orig_type_id;
2158 static inline __u16 btf_fwd_kind(struct btf_type *t)
2160 return BTF_INFO_KFLAG(t->info) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
2164 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
2165 * call it "candidate graph" in this description for brevity) to a type graph
2166 * formed by (potential) canonical struct/union ("canonical graph" for brevity
2167 * here, though keep in mind that not all types in canonical graph are
2168 * necessarily canonical representatives themselves, some of them might be
2169 * duplicates or its uniqueness might not have been established yet).
2171 * - >0, if type graphs are equivalent;
2172 * - 0, if not equivalent;
2175 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
2176 * equivalence of BTF types at each step. If at any point BTF types in candidate
2177 * and canonical graphs are not compatible structurally, whole graphs are
2178 * incompatible. If types are structurally equivalent (i.e., all information
2179 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
2180 * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
2181 * If a type references other types, then those referenced types are checked
2182 * for equivalence recursively.
2184 * During DFS traversal, if we find that for current `canon_id` type we
2185 * already have some mapping in hypothetical map, we check for two possible
2187 * - `canon_id` is mapped to exactly the same type as `cand_id`. This will
2188 * happen when type graphs have cycles. In this case we assume those two
2189 * types are equivalent.
2190 * - `canon_id` is mapped to different type. This is contradiction in our
2191 * hypothetical mapping, because same graph in canonical graph corresponds
2192 * to two different types in candidate graph, which for equivalent type
2193 * graphs shouldn't happen. This condition terminates equivalence check
2194 * with negative result.
2196 * If type graphs traversal exhausts types to check and find no contradiction,
2197 * then type graphs are equivalent.
2199 * When checking types for equivalence, there is one special case: FWD types.
2200 * If FWD type resolution is allowed and one of the types (either from canonical
2201 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
2202 * flag) and their names match, hypothetical mapping is updated to point from
2203 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
2204 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
2206 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
2207 * if there are two exactly named (or anonymous) structs/unions that are
2208 * compatible structurally, one of which has FWD field, while other is concrete
2209 * STRUCT/UNION, but according to C sources they are different structs/unions
2210 * that are referencing different types with the same name. This is extremely
2211 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
2212 * this logic is causing problems.
2214 * Doing FWD resolution means that both candidate and/or canonical graphs can
2215 * consists of portions of the graph that come from multiple compilation units.
2216 * This is due to the fact that types within single compilation unit are always
2217 * deduplicated and FWDs are already resolved, if referenced struct/union
2218 * definiton is available. So, if we had unresolved FWD and found corresponding
2219 * STRUCT/UNION, they will be from different compilation units. This
2220 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
2221 * type graph will likely have at least two different BTF types that describe
2222 * same type (e.g., most probably there will be two different BTF types for the
2223 * same 'int' primitive type) and could even have "overlapping" parts of type
2224 * graph that describe same subset of types.
2226 * This in turn means that our assumption that each type in canonical graph
2227 * must correspond to exactly one type in candidate graph might not hold
2228 * anymore and will make it harder to detect contradictions using hypothetical
2229 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
2230 * resolution only in canonical graph. FWDs in candidate graphs are never
2231 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
2233 * - Both types in canonical and candidate graphs are FWDs. If they are
2234 * structurally equivalent, then they can either be both resolved to the
2235 * same STRUCT/UNION or not resolved at all. In both cases they are
2236 * equivalent and there is no need to resolve FWD on candidate side.
2237 * - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
2238 * so nothing to resolve as well, algorithm will check equivalence anyway.
2239 * - Type in canonical graph is FWD, while type in candidate is concrete
2240 * STRUCT/UNION. In this case candidate graph comes from single compilation
2241 * unit, so there is exactly one BTF type for each unique C type. After
2242 * resolving FWD into STRUCT/UNION, there might be more than one BTF type
2243 * in canonical graph mapping to single BTF type in candidate graph, but
2244 * because hypothetical mapping maps from canonical to candidate types, it's
2245 * alright, and we still maintain the property of having single `canon_id`
2246 * mapping to single `cand_id` (there could be two different `canon_id`
2247 * mapped to the same `cand_id`, but it's not contradictory).
2248 * - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
2249 * graph is FWD. In this case we are just going to check compatibility of
2250 * STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
2251 * assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
2252 * a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
2253 * turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
2256 static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
2259 struct btf_type *cand_type;
2260 struct btf_type *canon_type;
2261 __u32 hypot_type_id;
2266 /* if both resolve to the same canonical, they must be equivalent */
2267 if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
2270 canon_id = resolve_fwd_id(d, canon_id);
2272 hypot_type_id = d->hypot_map[canon_id];
2273 if (hypot_type_id <= BTF_MAX_NR_TYPES)
2274 return hypot_type_id == cand_id;
2276 if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
2279 cand_type = d->btf->types[cand_id];
2280 canon_type = d->btf->types[canon_id];
2281 cand_kind = BTF_INFO_KIND(cand_type->info);
2282 canon_kind = BTF_INFO_KIND(canon_type->info);
2284 if (cand_type->name_off != canon_type->name_off)
2287 /* FWD <--> STRUCT/UNION equivalence check, if enabled */
2288 if (!d->opts.dont_resolve_fwds
2289 && (cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
2290 && cand_kind != canon_kind) {
2294 if (cand_kind == BTF_KIND_FWD) {
2295 real_kind = canon_kind;
2296 fwd_kind = btf_fwd_kind(cand_type);
2298 real_kind = cand_kind;
2299 fwd_kind = btf_fwd_kind(canon_type);
2301 return fwd_kind == real_kind;
2304 if (cand_kind != canon_kind)
2307 switch (cand_kind) {
2309 return btf_equal_int(cand_type, canon_type);
2312 if (d->opts.dont_resolve_fwds)
2313 return btf_equal_enum(cand_type, canon_type);
2315 return btf_compat_enum(cand_type, canon_type);
2318 return btf_equal_common(cand_type, canon_type);
2320 case BTF_KIND_CONST:
2321 case BTF_KIND_VOLATILE:
2322 case BTF_KIND_RESTRICT:
2324 case BTF_KIND_TYPEDEF:
2326 if (cand_type->info != canon_type->info)
2328 return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
2330 case BTF_KIND_ARRAY: {
2331 struct btf_array *cand_arr, *canon_arr;
2333 if (!btf_compat_array(cand_type, canon_type))
2335 cand_arr = (struct btf_array *)(cand_type + 1);
2336 canon_arr = (struct btf_array *)(canon_type + 1);
2337 eq = btf_dedup_is_equiv(d,
2338 cand_arr->index_type, canon_arr->index_type);
2341 return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
2344 case BTF_KIND_STRUCT:
2345 case BTF_KIND_UNION: {
2346 struct btf_member *cand_m, *canon_m;
2349 if (!btf_shallow_equal_struct(cand_type, canon_type))
2351 vlen = BTF_INFO_VLEN(cand_type->info);
2352 cand_m = (struct btf_member *)(cand_type + 1);
2353 canon_m = (struct btf_member *)(canon_type + 1);
2354 for (i = 0; i < vlen; i++) {
2355 eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
2365 case BTF_KIND_FUNC_PROTO: {
2366 struct btf_param *cand_p, *canon_p;
2369 if (!btf_compat_fnproto(cand_type, canon_type))
2371 eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
2374 vlen = BTF_INFO_VLEN(cand_type->info);
2375 cand_p = (struct btf_param *)(cand_type + 1);
2376 canon_p = (struct btf_param *)(canon_type + 1);
2377 for (i = 0; i < vlen; i++) {
2378 eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
2394 * Use hypothetical mapping, produced by successful type graph equivalence
2395 * check, to augment existing struct/union canonical mapping, where possible.
2397 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
2398 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
2399 * it doesn't matter if FWD type was part of canonical graph or candidate one,
2400 * we are recording the mapping anyway. As opposed to carefulness required
2401 * for struct/union correspondence mapping (described below), for FWD resolution
2402 * it's not important, as by the time that FWD type (reference type) will be
2403 * deduplicated all structs/unions will be deduped already anyway.
2405 * Recording STRUCT/UNION mapping is purely a performance optimization and is
2406 * not required for correctness. It needs to be done carefully to ensure that
2407 * struct/union from candidate's type graph is not mapped into corresponding
2408 * struct/union from canonical type graph that itself hasn't been resolved into
2409 * canonical representative. The only guarantee we have is that canonical
2410 * struct/union was determined as canonical and that won't change. But any
2411 * types referenced through that struct/union fields could have been not yet
2412 * resolved, so in case like that it's too early to establish any kind of
2413 * correspondence between structs/unions.
2415 * No canonical correspondence is derived for primitive types (they are already
2416 * deduplicated completely already anyway) or reference types (they rely on
2417 * stability of struct/union canonical relationship for equivalence checks).
2419 static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
2421 __u32 cand_type_id, targ_type_id;
2422 __u16 t_kind, c_kind;
2426 for (i = 0; i < d->hypot_cnt; i++) {
2427 cand_type_id = d->hypot_list[i];
2428 targ_type_id = d->hypot_map[cand_type_id];
2429 t_id = resolve_type_id(d, targ_type_id);
2430 c_id = resolve_type_id(d, cand_type_id);
2431 t_kind = BTF_INFO_KIND(d->btf->types[t_id]->info);
2432 c_kind = BTF_INFO_KIND(d->btf->types[c_id]->info);
2434 * Resolve FWD into STRUCT/UNION.
2435 * It's ok to resolve FWD into STRUCT/UNION that's not yet
2436 * mapped to canonical representative (as opposed to
2437 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
2438 * eventually that struct is going to be mapped and all resolved
2439 * FWDs will automatically resolve to correct canonical
2440 * representative. This will happen before ref type deduping,
2441 * which critically depends on stability of these mapping. This
2442 * stability is not a requirement for STRUCT/UNION equivalence
2445 if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
2446 d->map[c_id] = t_id;
2447 else if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
2448 d->map[t_id] = c_id;
2450 if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
2451 c_kind != BTF_KIND_FWD &&
2452 is_type_mapped(d, c_id) &&
2453 !is_type_mapped(d, t_id)) {
2455 * as a perf optimization, we can map struct/union
2456 * that's part of type graph we just verified for
2457 * equivalence. We can do that for struct/union that has
2458 * canonical representative only, though.
2460 d->map[t_id] = c_id;
2466 * Deduplicate struct/union types.
2468 * For each struct/union type its type signature hash is calculated, taking
2469 * into account type's name, size, number, order and names of fields, but
2470 * ignoring type ID's referenced from fields, because they might not be deduped
2471 * completely until after reference types deduplication phase. This type hash
2472 * is used to iterate over all potential canonical types, sharing same hash.
2473 * For each canonical candidate we check whether type graphs that they form
2474 * (through referenced types in fields and so on) are equivalent using algorithm
2475 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
2476 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
2477 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
2478 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
2479 * potentially map other structs/unions to their canonical representatives,
2480 * if such relationship hasn't yet been established. This speeds up algorithm
2481 * by eliminating some of the duplicate work.
2483 * If no matching canonical representative was found, struct/union is marked
2484 * as canonical for itself and is added into btf_dedup->dedup_table hash map
2485 * for further look ups.
2487 static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
2489 struct btf_type *cand_type, *t;
2490 struct hashmap_entry *hash_entry;
2491 /* if we don't find equivalent type, then we are canonical */
2492 __u32 new_id = type_id;
2496 /* already deduped or is in process of deduping (loop detected) */
2497 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
2500 t = d->btf->types[type_id];
2501 kind = BTF_INFO_KIND(t->info);
2503 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
2506 h = btf_hash_struct(t);
2507 for_each_dedup_cand(d, hash_entry, h) {
2508 __u32 cand_id = (__u32)(long)hash_entry->value;
2512 * Even though btf_dedup_is_equiv() checks for
2513 * btf_shallow_equal_struct() internally when checking two
2514 * structs (unions) for equivalence, we need to guard here
2515 * from picking matching FWD type as a dedup candidate.
2516 * This can happen due to hash collision. In such case just
2517 * relying on btf_dedup_is_equiv() would lead to potentially
2518 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
2519 * FWD and compatible STRUCT/UNION are considered equivalent.
2521 cand_type = d->btf->types[cand_id];
2522 if (!btf_shallow_equal_struct(t, cand_type))
2525 btf_dedup_clear_hypot_map(d);
2526 eq = btf_dedup_is_equiv(d, type_id, cand_id);
2532 btf_dedup_merge_hypot_map(d);
2536 d->map[type_id] = new_id;
2537 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
2543 static int btf_dedup_struct_types(struct btf_dedup *d)
2547 for (i = 1; i <= d->btf->nr_types; i++) {
2548 err = btf_dedup_struct_type(d, i);
2556 * Deduplicate reference type.
2558 * Once all primitive and struct/union types got deduplicated, we can easily
2559 * deduplicate all other (reference) BTF types. This is done in two steps:
2561 * 1. Resolve all referenced type IDs into their canonical type IDs. This
2562 * resolution can be done either immediately for primitive or struct/union types
2563 * (because they were deduped in previous two phases) or recursively for
2564 * reference types. Recursion will always terminate at either primitive or
2565 * struct/union type, at which point we can "unwind" chain of reference types
2566 * one by one. There is no danger of encountering cycles because in C type
2567 * system the only way to form type cycle is through struct/union, so any chain
2568 * of reference types, even those taking part in a type cycle, will inevitably
2569 * reach struct/union at some point.
2571 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
2572 * becomes "stable", in the sense that no further deduplication will cause
2573 * any changes to it. With that, it's now possible to calculate type's signature
2574 * hash (this time taking into account referenced type IDs) and loop over all
2575 * potential canonical representatives. If no match was found, current type
2576 * will become canonical representative of itself and will be added into
2577 * btf_dedup->dedup_table as another possible canonical representative.
2579 static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
2581 struct hashmap_entry *hash_entry;
2582 __u32 new_id = type_id, cand_id;
2583 struct btf_type *t, *cand;
2584 /* if we don't find equivalent type, then we are representative type */
2588 if (d->map[type_id] == BTF_IN_PROGRESS_ID)
2590 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
2591 return resolve_type_id(d, type_id);
2593 t = d->btf->types[type_id];
2594 d->map[type_id] = BTF_IN_PROGRESS_ID;
2596 switch (BTF_INFO_KIND(t->info)) {
2597 case BTF_KIND_CONST:
2598 case BTF_KIND_VOLATILE:
2599 case BTF_KIND_RESTRICT:
2601 case BTF_KIND_TYPEDEF:
2603 ref_type_id = btf_dedup_ref_type(d, t->type);
2604 if (ref_type_id < 0)
2606 t->type = ref_type_id;
2608 h = btf_hash_common(t);
2609 for_each_dedup_cand(d, hash_entry, h) {
2610 cand_id = (__u32)(long)hash_entry->value;
2611 cand = d->btf->types[cand_id];
2612 if (btf_equal_common(t, cand)) {
2619 case BTF_KIND_ARRAY: {
2620 struct btf_array *info = (struct btf_array *)(t + 1);
2622 ref_type_id = btf_dedup_ref_type(d, info->type);
2623 if (ref_type_id < 0)
2625 info->type = ref_type_id;
2627 ref_type_id = btf_dedup_ref_type(d, info->index_type);
2628 if (ref_type_id < 0)
2630 info->index_type = ref_type_id;
2632 h = btf_hash_array(t);
2633 for_each_dedup_cand(d, hash_entry, h) {
2634 cand_id = (__u32)(long)hash_entry->value;
2635 cand = d->btf->types[cand_id];
2636 if (btf_equal_array(t, cand)) {
2644 case BTF_KIND_FUNC_PROTO: {
2645 struct btf_param *param;
2649 ref_type_id = btf_dedup_ref_type(d, t->type);
2650 if (ref_type_id < 0)
2652 t->type = ref_type_id;
2654 vlen = BTF_INFO_VLEN(t->info);
2655 param = (struct btf_param *)(t + 1);
2656 for (i = 0; i < vlen; i++) {
2657 ref_type_id = btf_dedup_ref_type(d, param->type);
2658 if (ref_type_id < 0)
2660 param->type = ref_type_id;
2664 h = btf_hash_fnproto(t);
2665 for_each_dedup_cand(d, hash_entry, h) {
2666 cand_id = (__u32)(long)hash_entry->value;
2667 cand = d->btf->types[cand_id];
2668 if (btf_equal_fnproto(t, cand)) {
2680 d->map[type_id] = new_id;
2681 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
2687 static int btf_dedup_ref_types(struct btf_dedup *d)
2691 for (i = 1; i <= d->btf->nr_types; i++) {
2692 err = btf_dedup_ref_type(d, i);
2696 /* we won't need d->dedup_table anymore */
2697 hashmap__free(d->dedup_table);
2698 d->dedup_table = NULL;
2705 * After we established for each type its corresponding canonical representative
2706 * type, we now can eliminate types that are not canonical and leave only
2707 * canonical ones layed out sequentially in memory by copying them over
2708 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
2709 * a map from original type ID to a new compacted type ID, which will be used
2710 * during next phase to "fix up" type IDs, referenced from struct/union and
2713 static int btf_dedup_compact_types(struct btf_dedup *d)
2715 struct btf_type **new_types;
2716 __u32 next_type_id = 1;
2717 char *types_start, *p;
2720 /* we are going to reuse hypot_map to store compaction remapping */
2721 d->hypot_map[0] = 0;
2722 for (i = 1; i <= d->btf->nr_types; i++)
2723 d->hypot_map[i] = BTF_UNPROCESSED_ID;
2725 types_start = d->btf->nohdr_data + d->btf->hdr->type_off;
2728 for (i = 1; i <= d->btf->nr_types; i++) {
2732 len = btf_type_size(d->btf->types[i]);
2736 memmove(p, d->btf->types[i], len);
2737 d->hypot_map[i] = next_type_id;
2738 d->btf->types[next_type_id] = (struct btf_type *)p;
2743 /* shrink struct btf's internal types index and update btf_header */
2744 d->btf->nr_types = next_type_id - 1;
2745 d->btf->types_size = d->btf->nr_types;
2746 d->btf->hdr->type_len = p - types_start;
2747 new_types = realloc(d->btf->types,
2748 (1 + d->btf->nr_types) * sizeof(struct btf_type *));
2751 d->btf->types = new_types;
2753 /* make sure string section follows type information without gaps */
2754 d->btf->hdr->str_off = p - (char *)d->btf->nohdr_data;
2755 memmove(p, d->btf->strings, d->btf->hdr->str_len);
2756 d->btf->strings = p;
2757 p += d->btf->hdr->str_len;
2759 d->btf->data_size = p - (char *)d->btf->data;
2764 * Figure out final (deduplicated and compacted) type ID for provided original
2765 * `type_id` by first resolving it into corresponding canonical type ID and
2766 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
2767 * which is populated during compaction phase.
2769 static int btf_dedup_remap_type_id(struct btf_dedup *d, __u32 type_id)
2771 __u32 resolved_type_id, new_type_id;
2773 resolved_type_id = resolve_type_id(d, type_id);
2774 new_type_id = d->hypot_map[resolved_type_id];
2775 if (new_type_id > BTF_MAX_NR_TYPES)
2781 * Remap referenced type IDs into deduped type IDs.
2783 * After BTF types are deduplicated and compacted, their final type IDs may
2784 * differ from original ones. The map from original to a corresponding
2785 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
2786 * compaction phase. During remapping phase we are rewriting all type IDs
2787 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
2788 * their final deduped type IDs.
2790 static int btf_dedup_remap_type(struct btf_dedup *d, __u32 type_id)
2792 struct btf_type *t = d->btf->types[type_id];
2795 switch (BTF_INFO_KIND(t->info)) {
2801 case BTF_KIND_CONST:
2802 case BTF_KIND_VOLATILE:
2803 case BTF_KIND_RESTRICT:
2805 case BTF_KIND_TYPEDEF:
2808 r = btf_dedup_remap_type_id(d, t->type);
2814 case BTF_KIND_ARRAY: {
2815 struct btf_array *arr_info = (struct btf_array *)(t + 1);
2817 r = btf_dedup_remap_type_id(d, arr_info->type);
2821 r = btf_dedup_remap_type_id(d, arr_info->index_type);
2824 arr_info->index_type = r;
2828 case BTF_KIND_STRUCT:
2829 case BTF_KIND_UNION: {
2830 struct btf_member *member = (struct btf_member *)(t + 1);
2831 __u16 vlen = BTF_INFO_VLEN(t->info);
2833 for (i = 0; i < vlen; i++) {
2834 r = btf_dedup_remap_type_id(d, member->type);
2843 case BTF_KIND_FUNC_PROTO: {
2844 struct btf_param *param = (struct btf_param *)(t + 1);
2845 __u16 vlen = BTF_INFO_VLEN(t->info);
2847 r = btf_dedup_remap_type_id(d, t->type);
2852 for (i = 0; i < vlen; i++) {
2853 r = btf_dedup_remap_type_id(d, param->type);
2862 case BTF_KIND_DATASEC: {
2863 struct btf_var_secinfo *var = (struct btf_var_secinfo *)(t + 1);
2864 __u16 vlen = BTF_INFO_VLEN(t->info);
2866 for (i = 0; i < vlen; i++) {
2867 r = btf_dedup_remap_type_id(d, var->type);
2883 static int btf_dedup_remap_types(struct btf_dedup *d)
2887 for (i = 1; i <= d->btf->nr_types; i++) {
2888 r = btf_dedup_remap_type(d, i);