1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * elf.c - ELF access library
5 * Adapted from kpatch (https://github.com/dynup/kpatch):
6 * Copyright (C) 2013-2015 Josh Poimboeuf <jpoimboe@redhat.com>
7 * Copyright (C) 2014 Seth Jennings <sjenning@redhat.com>
10 #include <sys/types.h>
23 #define MAX_NAME_LEN 128
25 static inline u32 str_hash(const char *str)
27 return jhash(str, strlen(str), 0);
30 static inline int elf_hash_bits(void)
32 return vmlinux ? ELF_HASH_BITS : 16;
35 #define elf_hash_add(hashtable, node, key) \
36 hlist_add_head(node, &hashtable[hash_min(key, elf_hash_bits())])
38 static void elf_hash_init(struct hlist_head *table)
40 __hash_init(table, 1U << elf_hash_bits());
43 #define elf_hash_for_each_possible(name, obj, member, key) \
44 hlist_for_each_entry(obj, &name[hash_min(key, elf_hash_bits())], member)
46 static void rb_add(struct rb_root *tree, struct rb_node *node,
47 int (*cmp)(struct rb_node *, const struct rb_node *))
49 struct rb_node **link = &tree->rb_node;
50 struct rb_node *parent = NULL;
54 if (cmp(node, parent) < 0)
55 link = &parent->rb_left;
57 link = &parent->rb_right;
60 rb_link_node(node, parent, link);
61 rb_insert_color(node, tree);
64 static struct rb_node *rb_find_first(const struct rb_root *tree, const void *key,
65 int (*cmp)(const void *key, const struct rb_node *))
67 struct rb_node *node = tree->rb_node;
68 struct rb_node *match = NULL;
71 int c = cmp(key, node);
77 node = node->rb_right;
84 static struct rb_node *rb_next_match(struct rb_node *node, const void *key,
85 int (*cmp)(const void *key, const struct rb_node *))
88 if (node && cmp(key, node))
93 #define rb_for_each(tree, node, key, cmp) \
94 for ((node) = rb_find_first((tree), (key), (cmp)); \
95 (node); (node) = rb_next_match((node), (key), (cmp)))
97 static int symbol_to_offset(struct rb_node *a, const struct rb_node *b)
99 struct symbol *sa = rb_entry(a, struct symbol, node);
100 struct symbol *sb = rb_entry(b, struct symbol, node);
102 if (sa->offset < sb->offset)
104 if (sa->offset > sb->offset)
107 if (sa->len < sb->len)
109 if (sa->len > sb->len)
117 static int symbol_by_offset(const void *key, const struct rb_node *node)
119 const struct symbol *s = rb_entry(node, struct symbol, node);
120 const unsigned long *o = key;
124 if (*o >= s->offset + s->len)
130 struct section *find_section_by_name(const struct elf *elf, const char *name)
134 elf_hash_for_each_possible(elf->section_name_hash, sec, name_hash, str_hash(name))
135 if (!strcmp(sec->name, name))
141 static struct section *find_section_by_index(struct elf *elf,
146 elf_hash_for_each_possible(elf->section_hash, sec, hash, idx)
153 static struct symbol *find_symbol_by_index(struct elf *elf, unsigned int idx)
157 elf_hash_for_each_possible(elf->symbol_hash, sym, hash, idx)
164 struct symbol *find_symbol_by_offset(struct section *sec, unsigned long offset)
166 struct rb_node *node;
168 rb_for_each(&sec->symbol_tree, node, &offset, symbol_by_offset) {
169 struct symbol *s = rb_entry(node, struct symbol, node);
171 if (s->offset == offset && s->type != STT_SECTION)
178 struct symbol *find_func_by_offset(struct section *sec, unsigned long offset)
180 struct rb_node *node;
182 rb_for_each(&sec->symbol_tree, node, &offset, symbol_by_offset) {
183 struct symbol *s = rb_entry(node, struct symbol, node);
185 if (s->offset == offset && s->type == STT_FUNC)
192 struct symbol *find_symbol_containing(const struct section *sec, unsigned long offset)
194 struct rb_node *node;
196 rb_for_each(&sec->symbol_tree, node, &offset, symbol_by_offset) {
197 struct symbol *s = rb_entry(node, struct symbol, node);
199 if (s->type != STT_SECTION)
206 struct symbol *find_func_containing(struct section *sec, unsigned long offset)
208 struct rb_node *node;
210 rb_for_each(&sec->symbol_tree, node, &offset, symbol_by_offset) {
211 struct symbol *s = rb_entry(node, struct symbol, node);
213 if (s->type == STT_FUNC)
220 struct symbol *find_symbol_by_name(const struct elf *elf, const char *name)
224 elf_hash_for_each_possible(elf->symbol_name_hash, sym, name_hash, str_hash(name))
225 if (!strcmp(sym->name, name))
231 struct reloc *find_reloc_by_dest_range(const struct elf *elf, struct section *sec,
232 unsigned long offset, unsigned int len)
234 struct reloc *reloc, *r = NULL;
242 for_offset_range(o, offset, offset + len) {
243 elf_hash_for_each_possible(elf->reloc_hash, reloc, hash,
244 sec_offset_hash(sec, o)) {
245 if (reloc->sec != sec)
248 if (reloc->offset >= offset && reloc->offset < offset + len) {
249 if (!r || reloc->offset < r->offset)
260 struct reloc *find_reloc_by_dest(const struct elf *elf, struct section *sec, unsigned long offset)
262 return find_reloc_by_dest_range(elf, sec, offset, 1);
265 void insn_to_reloc_sym_addend(struct section *sec, unsigned long offset,
269 reloc->sym = sec->sym;
270 reloc->addend = offset;
275 * The Clang assembler strips section symbols, so we have to reference
276 * the function symbol instead:
278 reloc->sym = find_symbol_containing(sec, offset);
281 * Hack alert. This happens when we need to reference the NOP
282 * pad insn immediately after the function.
284 reloc->sym = find_symbol_containing(sec, offset - 1);
288 reloc->addend = offset - reloc->sym->offset;
291 static int read_sections(struct elf *elf)
295 size_t shstrndx, sections_nr;
298 if (elf_getshdrnum(elf->elf, §ions_nr)) {
299 WARN_ELF("elf_getshdrnum");
303 if (elf_getshdrstrndx(elf->elf, &shstrndx)) {
304 WARN_ELF("elf_getshdrstrndx");
308 for (i = 0; i < sections_nr; i++) {
309 sec = malloc(sizeof(*sec));
314 memset(sec, 0, sizeof(*sec));
316 INIT_LIST_HEAD(&sec->symbol_list);
317 INIT_LIST_HEAD(&sec->reloc_list);
319 s = elf_getscn(elf->elf, i);
321 WARN_ELF("elf_getscn");
325 sec->idx = elf_ndxscn(s);
327 if (!gelf_getshdr(s, &sec->sh)) {
328 WARN_ELF("gelf_getshdr");
332 sec->name = elf_strptr(elf->elf, shstrndx, sec->sh.sh_name);
334 WARN_ELF("elf_strptr");
338 if (sec->sh.sh_size != 0) {
339 sec->data = elf_getdata(s, NULL);
341 WARN_ELF("elf_getdata");
344 if (sec->data->d_off != 0 ||
345 sec->data->d_size != sec->sh.sh_size) {
346 WARN("unexpected data attributes for %s",
351 sec->len = sec->sh.sh_size;
353 list_add_tail(&sec->list, &elf->sections);
354 elf_hash_add(elf->section_hash, &sec->hash, sec->idx);
355 elf_hash_add(elf->section_name_hash, &sec->name_hash, str_hash(sec->name));
359 printf("nr_sections: %lu\n", (unsigned long)sections_nr);
361 /* sanity check, one more call to elf_nextscn() should return NULL */
362 if (elf_nextscn(elf->elf, s)) {
363 WARN("section entry mismatch");
370 static int read_symbols(struct elf *elf)
372 struct section *symtab, *symtab_shndx, *sec;
373 struct symbol *sym, *pfunc;
374 struct list_head *entry;
375 struct rb_node *pnode;
378 Elf_Data *shndx_data = NULL;
381 symtab = find_section_by_name(elf, ".symtab");
384 * A missing symbol table is actually possible if it's an empty
385 * .o file. This can happen for thunk_64.o.
390 symtab_shndx = find_section_by_name(elf, ".symtab_shndx");
392 shndx_data = symtab_shndx->data;
394 symbols_nr = symtab->sh.sh_size / symtab->sh.sh_entsize;
396 for (i = 0; i < symbols_nr; i++) {
397 sym = malloc(sizeof(*sym));
402 memset(sym, 0, sizeof(*sym));
407 if (!gelf_getsymshndx(symtab->data, shndx_data, i, &sym->sym,
409 WARN_ELF("gelf_getsymshndx");
413 sym->name = elf_strptr(elf->elf, symtab->sh.sh_link,
416 WARN_ELF("elf_strptr");
420 sym->type = GELF_ST_TYPE(sym->sym.st_info);
421 sym->bind = GELF_ST_BIND(sym->sym.st_info);
423 if ((sym->sym.st_shndx > SHN_UNDEF &&
424 sym->sym.st_shndx < SHN_LORESERVE) ||
425 (shndx_data && sym->sym.st_shndx == SHN_XINDEX)) {
426 if (sym->sym.st_shndx != SHN_XINDEX)
427 shndx = sym->sym.st_shndx;
429 sym->sec = find_section_by_index(elf, shndx);
431 WARN("couldn't find section for symbol %s",
435 if (sym->type == STT_SECTION) {
436 sym->name = sym->sec->name;
440 sym->sec = find_section_by_index(elf, 0);
442 sym->offset = sym->sym.st_value;
443 sym->len = sym->sym.st_size;
445 rb_add(&sym->sec->symbol_tree, &sym->node, symbol_to_offset);
446 pnode = rb_prev(&sym->node);
448 entry = &rb_entry(pnode, struct symbol, node)->list;
450 entry = &sym->sec->symbol_list;
451 list_add(&sym->list, entry);
452 elf_hash_add(elf->symbol_hash, &sym->hash, sym->idx);
453 elf_hash_add(elf->symbol_name_hash, &sym->name_hash, str_hash(sym->name));
456 * Don't store empty STT_NOTYPE symbols in the rbtree. They
457 * can exist within a function, confusing the sorting.
460 rb_erase(&sym->node, &sym->sec->symbol_tree);
464 printf("nr_symbols: %lu\n", (unsigned long)symbols_nr);
466 /* Create parent/child links for any cold subfunctions */
467 list_for_each_entry(sec, &elf->sections, list) {
468 list_for_each_entry(sym, &sec->symbol_list, list) {
469 char pname[MAX_NAME_LEN + 1];
471 if (sym->type != STT_FUNC)
474 if (sym->pfunc == NULL)
477 if (sym->cfunc == NULL)
480 coldstr = strstr(sym->name, ".cold");
484 pnamelen = coldstr - sym->name;
485 if (pnamelen > MAX_NAME_LEN) {
486 WARN("%s(): parent function name exceeds maximum length of %d characters",
487 sym->name, MAX_NAME_LEN);
491 strncpy(pname, sym->name, pnamelen);
492 pname[pnamelen] = '\0';
493 pfunc = find_symbol_by_name(elf, pname);
496 WARN("%s(): can't find parent function",
505 * Unfortunately, -fnoreorder-functions puts the child
506 * inside the parent. Remove the overlap so we can
507 * have sane assumptions.
509 * Note that pfunc->len now no longer matches
510 * pfunc->sym.st_size.
512 if (sym->sec == pfunc->sec &&
513 sym->offset >= pfunc->offset &&
514 sym->offset + sym->len == pfunc->offset + pfunc->len) {
515 pfunc->len -= sym->len;
527 void elf_add_reloc(struct elf *elf, struct reloc *reloc)
529 struct section *sec = reloc->sec;
531 list_add_tail(&reloc->list, &sec->reloc_list);
532 elf_hash_add(elf->reloc_hash, &reloc->hash, reloc_hash(reloc));
535 static int read_rel_reloc(struct section *sec, int i, struct reloc *reloc, unsigned int *symndx)
537 if (!gelf_getrel(sec->data, i, &reloc->rel)) {
538 WARN_ELF("gelf_getrel");
541 reloc->type = GELF_R_TYPE(reloc->rel.r_info);
543 reloc->offset = reloc->rel.r_offset;
544 *symndx = GELF_R_SYM(reloc->rel.r_info);
548 static int read_rela_reloc(struct section *sec, int i, struct reloc *reloc, unsigned int *symndx)
550 if (!gelf_getrela(sec->data, i, &reloc->rela)) {
551 WARN_ELF("gelf_getrela");
554 reloc->type = GELF_R_TYPE(reloc->rela.r_info);
555 reloc->addend = reloc->rela.r_addend;
556 reloc->offset = reloc->rela.r_offset;
557 *symndx = GELF_R_SYM(reloc->rela.r_info);
561 static int read_relocs(struct elf *elf)
567 unsigned long nr_reloc, max_reloc = 0, tot_reloc = 0;
569 list_for_each_entry(sec, &elf->sections, list) {
570 if ((sec->sh.sh_type != SHT_RELA) &&
571 (sec->sh.sh_type != SHT_REL))
574 sec->base = find_section_by_index(elf, sec->sh.sh_info);
576 WARN("can't find base section for reloc section %s",
581 sec->base->reloc = sec;
584 for (i = 0; i < sec->sh.sh_size / sec->sh.sh_entsize; i++) {
585 reloc = malloc(sizeof(*reloc));
590 memset(reloc, 0, sizeof(*reloc));
591 switch (sec->sh.sh_type) {
593 if (read_rel_reloc(sec, i, reloc, &symndx))
597 if (read_rela_reloc(sec, i, reloc, &symndx))
605 reloc->sym = find_symbol_by_index(elf, symndx);
607 WARN("can't find reloc entry symbol %d for %s",
612 elf_add_reloc(elf, reloc);
615 max_reloc = max(max_reloc, nr_reloc);
616 tot_reloc += nr_reloc;
620 printf("max_reloc: %lu\n", max_reloc);
621 printf("tot_reloc: %lu\n", tot_reloc);
627 struct elf *elf_open_read(const char *name, int flags)
632 elf_version(EV_CURRENT);
634 elf = malloc(sizeof(*elf));
639 memset(elf, 0, offsetof(struct elf, sections));
641 INIT_LIST_HEAD(&elf->sections);
643 elf_hash_init(elf->symbol_hash);
644 elf_hash_init(elf->symbol_name_hash);
645 elf_hash_init(elf->section_hash);
646 elf_hash_init(elf->section_name_hash);
647 elf_hash_init(elf->reloc_hash);
649 elf->fd = open(name, flags);
651 fprintf(stderr, "objtool: Can't open '%s': %s\n",
652 name, strerror(errno));
656 if ((flags & O_ACCMODE) == O_RDONLY)
657 cmd = ELF_C_READ_MMAP;
658 else if ((flags & O_ACCMODE) == O_RDWR)
663 elf->elf = elf_begin(elf->fd, cmd, NULL);
665 WARN_ELF("elf_begin");
669 if (!gelf_getehdr(elf->elf, &elf->ehdr)) {
670 WARN_ELF("gelf_getehdr");
674 if (read_sections(elf))
677 if (read_symbols(elf))
680 if (read_relocs(elf))
690 struct section *elf_create_section(struct elf *elf, const char *name,
691 unsigned int sh_flags, size_t entsize, int nr)
693 struct section *sec, *shstrtab;
694 size_t size = entsize * nr;
698 sec = malloc(sizeof(*sec));
703 memset(sec, 0, sizeof(*sec));
705 INIT_LIST_HEAD(&sec->symbol_list);
706 INIT_LIST_HEAD(&sec->reloc_list);
708 s = elf_newscn(elf->elf);
710 WARN_ELF("elf_newscn");
714 sec->name = strdup(name);
720 sec->idx = elf_ndxscn(s);
724 sec->data = elf_newdata(s);
726 WARN_ELF("elf_newdata");
730 sec->data->d_size = size;
731 sec->data->d_align = 1;
734 sec->data->d_buf = malloc(size);
735 if (!sec->data->d_buf) {
739 memset(sec->data->d_buf, 0, size);
742 if (!gelf_getshdr(s, &sec->sh)) {
743 WARN_ELF("gelf_getshdr");
747 sec->sh.sh_size = size;
748 sec->sh.sh_entsize = entsize;
749 sec->sh.sh_type = SHT_PROGBITS;
750 sec->sh.sh_addralign = 1;
751 sec->sh.sh_flags = SHF_ALLOC | sh_flags;
754 /* Add section name to .shstrtab (or .strtab for Clang) */
755 shstrtab = find_section_by_name(elf, ".shstrtab");
757 shstrtab = find_section_by_name(elf, ".strtab");
759 WARN("can't find .shstrtab or .strtab section");
763 s = elf_getscn(elf->elf, shstrtab->idx);
765 WARN_ELF("elf_getscn");
769 data = elf_newdata(s);
771 WARN_ELF("elf_newdata");
775 data->d_buf = sec->name;
776 data->d_size = strlen(name) + 1;
779 sec->sh.sh_name = shstrtab->len;
781 shstrtab->len += strlen(name) + 1;
782 shstrtab->changed = true;
784 list_add_tail(&sec->list, &elf->sections);
785 elf_hash_add(elf->section_hash, &sec->hash, sec->idx);
786 elf_hash_add(elf->section_name_hash, &sec->name_hash, str_hash(sec->name));
793 static struct section *elf_create_rel_reloc_section(struct elf *elf, struct section *base)
798 relocname = malloc(strlen(base->name) + strlen(".rel") + 1);
803 strcpy(relocname, ".rel");
804 strcat(relocname, base->name);
806 sec = elf_create_section(elf, relocname, 0, sizeof(GElf_Rel), 0);
814 sec->sh.sh_type = SHT_REL;
815 sec->sh.sh_addralign = 8;
816 sec->sh.sh_link = find_section_by_name(elf, ".symtab")->idx;
817 sec->sh.sh_info = base->idx;
818 sec->sh.sh_flags = SHF_INFO_LINK;
823 static struct section *elf_create_rela_reloc_section(struct elf *elf, struct section *base)
828 relocname = malloc(strlen(base->name) + strlen(".rela") + 1);
833 strcpy(relocname, ".rela");
834 strcat(relocname, base->name);
836 sec = elf_create_section(elf, relocname, 0, sizeof(GElf_Rela), 0);
844 sec->sh.sh_type = SHT_RELA;
845 sec->sh.sh_addralign = 8;
846 sec->sh.sh_link = find_section_by_name(elf, ".symtab")->idx;
847 sec->sh.sh_info = base->idx;
848 sec->sh.sh_flags = SHF_INFO_LINK;
853 struct section *elf_create_reloc_section(struct elf *elf,
854 struct section *base,
858 case SHT_REL: return elf_create_rel_reloc_section(elf, base);
859 case SHT_RELA: return elf_create_rela_reloc_section(elf, base);
860 default: return NULL;
864 static int elf_rebuild_rel_reloc_section(struct section *sec, int nr)
870 /* Allocate a buffer for relocations */
871 size = nr * sizeof(*relocs);
872 relocs = malloc(size);
878 sec->data->d_buf = relocs;
879 sec->data->d_size = size;
881 sec->sh.sh_size = size;
884 list_for_each_entry(reloc, &sec->reloc_list, list) {
885 relocs[idx].r_offset = reloc->offset;
886 relocs[idx].r_info = GELF_R_INFO(reloc->sym->idx, reloc->type);
893 static int elf_rebuild_rela_reloc_section(struct section *sec, int nr)
899 /* Allocate a buffer for relocations with addends */
900 size = nr * sizeof(*relocs);
901 relocs = malloc(size);
907 sec->data->d_buf = relocs;
908 sec->data->d_size = size;
910 sec->sh.sh_size = size;
913 list_for_each_entry(reloc, &sec->reloc_list, list) {
914 relocs[idx].r_offset = reloc->offset;
915 relocs[idx].r_addend = reloc->addend;
916 relocs[idx].r_info = GELF_R_INFO(reloc->sym->idx, reloc->type);
923 int elf_rebuild_reloc_section(struct elf *elf, struct section *sec)
932 list_for_each_entry(reloc, &sec->reloc_list, list)
935 switch (sec->sh.sh_type) {
936 case SHT_REL: return elf_rebuild_rel_reloc_section(sec, nr);
937 case SHT_RELA: return elf_rebuild_rela_reloc_section(sec, nr);
942 int elf_write_insn(struct elf *elf, struct section *sec,
943 unsigned long offset, unsigned int len,
946 Elf_Data *data = sec->data;
948 if (data->d_type != ELF_T_BYTE || data->d_off) {
949 WARN("write to unexpected data for section: %s", sec->name);
953 memcpy(data->d_buf + offset, insn, len);
954 elf_flagdata(data, ELF_C_SET, ELF_F_DIRTY);
961 int elf_write_reloc(struct elf *elf, struct reloc *reloc)
963 struct section *sec = reloc->sec;
965 if (sec->sh.sh_type == SHT_REL) {
966 reloc->rel.r_info = GELF_R_INFO(reloc->sym->idx, reloc->type);
967 reloc->rel.r_offset = reloc->offset;
969 if (!gelf_update_rel(sec->data, reloc->idx, &reloc->rel)) {
970 WARN_ELF("gelf_update_rel");
974 reloc->rela.r_info = GELF_R_INFO(reloc->sym->idx, reloc->type);
975 reloc->rela.r_addend = reloc->addend;
976 reloc->rela.r_offset = reloc->offset;
978 if (!gelf_update_rela(sec->data, reloc->idx, &reloc->rela)) {
979 WARN_ELF("gelf_update_rela");
989 int elf_write(struct elf *elf)
994 /* Update section headers for changed sections: */
995 list_for_each_entry(sec, &elf->sections, list) {
997 s = elf_getscn(elf->elf, sec->idx);
999 WARN_ELF("elf_getscn");
1002 if (!gelf_update_shdr(s, &sec->sh)) {
1003 WARN_ELF("gelf_update_shdr");
1007 sec->changed = false;
1011 /* Make sure the new section header entries get updated properly. */
1012 elf_flagelf(elf->elf, ELF_C_SET, ELF_F_DIRTY);
1014 /* Write all changes to the file. */
1015 if (elf_update(elf->elf, ELF_C_WRITE) < 0) {
1016 WARN_ELF("elf_update");
1020 elf->changed = false;
1025 void elf_close(struct elf *elf)
1027 struct section *sec, *tmpsec;
1028 struct symbol *sym, *tmpsym;
1029 struct reloc *reloc, *tmpreloc;
1037 list_for_each_entry_safe(sec, tmpsec, &elf->sections, list) {
1038 list_for_each_entry_safe(sym, tmpsym, &sec->symbol_list, list) {
1039 list_del(&sym->list);
1040 hash_del(&sym->hash);
1043 list_for_each_entry_safe(reloc, tmpreloc, &sec->reloc_list, list) {
1044 list_del(&reloc->list);
1045 hash_del(&reloc->hash);
1048 list_del(&sec->list);