1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * elf.c - ELF access library
4 *
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>
8 */
9
10 #include <sys/types.h>
11 #include <sys/stat.h>
12 #include <fcntl.h>
13 #include <stdio.h>
14 #include <stdlib.h>
15 #include <string.h>
16 #include <unistd.h>
17 #include <errno.h>
18 #include "builtin.h"
19
20 #include "elf.h"
21 #include "warn.h"
22
23 #define MAX_NAME_LEN 128
24
str_hash(const char * str)25 static inline u32 str_hash(const char *str)
26 {
27 return jhash(str, strlen(str), 0);
28 }
29
elf_hash_bits(void)30 static inline int elf_hash_bits(void)
31 {
32 return vmlinux ? ELF_HASH_BITS : 16;
33 }
34
35 #define elf_hash_add(hashtable, node, key) \
36 hlist_add_head(node, &hashtable[hash_min(key, elf_hash_bits())])
37
elf_hash_init(struct hlist_head * table)38 static void elf_hash_init(struct hlist_head *table)
39 {
40 __hash_init(table, 1U << elf_hash_bits());
41 }
42
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)
45
rb_add(struct rb_root * tree,struct rb_node * node,int (* cmp)(struct rb_node *,const struct rb_node *))46 static void rb_add(struct rb_root *tree, struct rb_node *node,
47 int (*cmp)(struct rb_node *, const struct rb_node *))
48 {
49 struct rb_node **link = &tree->rb_node;
50 struct rb_node *parent = NULL;
51
52 while (*link) {
53 parent = *link;
54 if (cmp(node, parent) < 0)
55 link = &parent->rb_left;
56 else
57 link = &parent->rb_right;
58 }
59
60 rb_link_node(node, parent, link);
61 rb_insert_color(node, tree);
62 }
63
rb_find_first(const struct rb_root * tree,const void * key,int (* cmp)(const void * key,const struct rb_node *))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 *))
66 {
67 struct rb_node *node = tree->rb_node;
68 struct rb_node *match = NULL;
69
70 while (node) {
71 int c = cmp(key, node);
72 if (c <= 0) {
73 if (!c)
74 match = node;
75 node = node->rb_left;
76 } else if (c > 0) {
77 node = node->rb_right;
78 }
79 }
80
81 return match;
82 }
83
rb_next_match(struct rb_node * node,const void * key,int (* cmp)(const void * key,const struct rb_node *))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 *))
86 {
87 node = rb_next(node);
88 if (node && cmp(key, node))
89 node = NULL;
90 return node;
91 }
92
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)))
96
symbol_to_offset(struct rb_node * a,const struct rb_node * b)97 static int symbol_to_offset(struct rb_node *a, const struct rb_node *b)
98 {
99 struct symbol *sa = rb_entry(a, struct symbol, node);
100 struct symbol *sb = rb_entry(b, struct symbol, node);
101
102 if (sa->offset < sb->offset)
103 return -1;
104 if (sa->offset > sb->offset)
105 return 1;
106
107 if (sa->len < sb->len)
108 return -1;
109 if (sa->len > sb->len)
110 return 1;
111
112 sa->alias = sb;
113
114 return 0;
115 }
116
symbol_by_offset(const void * key,const struct rb_node * node)117 static int symbol_by_offset(const void *key, const struct rb_node *node)
118 {
119 const struct symbol *s = rb_entry(node, struct symbol, node);
120 const unsigned long *o = key;
121
122 if (*o < s->offset)
123 return -1;
124 if (*o >= s->offset + s->len)
125 return 1;
126
127 return 0;
128 }
129
find_section_by_name(const struct elf * elf,const char * name)130 struct section *find_section_by_name(const struct elf *elf, const char *name)
131 {
132 struct section *sec;
133
134 elf_hash_for_each_possible(elf->section_name_hash, sec, name_hash, str_hash(name))
135 if (!strcmp(sec->name, name))
136 return sec;
137
138 return NULL;
139 }
140
find_section_by_index(struct elf * elf,unsigned int idx)141 static struct section *find_section_by_index(struct elf *elf,
142 unsigned int idx)
143 {
144 struct section *sec;
145
146 elf_hash_for_each_possible(elf->section_hash, sec, hash, idx)
147 if (sec->idx == idx)
148 return sec;
149
150 return NULL;
151 }
152
find_symbol_by_index(struct elf * elf,unsigned int idx)153 static struct symbol *find_symbol_by_index(struct elf *elf, unsigned int idx)
154 {
155 struct symbol *sym;
156
157 elf_hash_for_each_possible(elf->symbol_hash, sym, hash, idx)
158 if (sym->idx == idx)
159 return sym;
160
161 return NULL;
162 }
163
find_symbol_by_offset(struct section * sec,unsigned long offset)164 struct symbol *find_symbol_by_offset(struct section *sec, unsigned long offset)
165 {
166 struct rb_node *node;
167
168 rb_for_each(&sec->symbol_tree, node, &offset, symbol_by_offset) {
169 struct symbol *s = rb_entry(node, struct symbol, node);
170
171 if (s->offset == offset && s->type != STT_SECTION)
172 return s;
173 }
174
175 return NULL;
176 }
177
find_func_by_offset(struct section * sec,unsigned long offset)178 struct symbol *find_func_by_offset(struct section *sec, unsigned long offset)
179 {
180 struct rb_node *node;
181
182 rb_for_each(&sec->symbol_tree, node, &offset, symbol_by_offset) {
183 struct symbol *s = rb_entry(node, struct symbol, node);
184
185 if (s->offset == offset && s->type == STT_FUNC)
186 return s;
187 }
188
189 return NULL;
190 }
191
find_symbol_containing(const struct section * sec,unsigned long offset)192 struct symbol *find_symbol_containing(const struct section *sec, unsigned long offset)
193 {
194 struct rb_node *node;
195
196 rb_for_each(&sec->symbol_tree, node, &offset, symbol_by_offset) {
197 struct symbol *s = rb_entry(node, struct symbol, node);
198
199 if (s->type != STT_SECTION)
200 return s;
201 }
202
203 return NULL;
204 }
205
find_func_containing(struct section * sec,unsigned long offset)206 struct symbol *find_func_containing(struct section *sec, unsigned long offset)
207 {
208 struct rb_node *node;
209
210 rb_for_each(&sec->symbol_tree, node, &offset, symbol_by_offset) {
211 struct symbol *s = rb_entry(node, struct symbol, node);
212
213 if (s->type == STT_FUNC)
214 return s;
215 }
216
217 return NULL;
218 }
219
find_symbol_by_name(const struct elf * elf,const char * name)220 struct symbol *find_symbol_by_name(const struct elf *elf, const char *name)
221 {
222 struct symbol *sym;
223
224 elf_hash_for_each_possible(elf->symbol_name_hash, sym, name_hash, str_hash(name))
225 if (!strcmp(sym->name, name))
226 return sym;
227
228 return NULL;
229 }
230
find_reloc_by_dest_range(const struct elf * elf,struct section * sec,unsigned long offset,unsigned int len)231 struct reloc *find_reloc_by_dest_range(const struct elf *elf, struct section *sec,
232 unsigned long offset, unsigned int len)
233 {
234 struct reloc *reloc, *r = NULL;
235 unsigned long o;
236
237 if (!sec->reloc)
238 return NULL;
239
240 sec = sec->reloc;
241
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)
246 continue;
247
248 if (reloc->offset >= offset && reloc->offset < offset + len) {
249 if (!r || reloc->offset < r->offset)
250 r = reloc;
251 }
252 }
253 if (r)
254 return r;
255 }
256
257 return NULL;
258 }
259
find_reloc_by_dest(const struct elf * elf,struct section * sec,unsigned long offset)260 struct reloc *find_reloc_by_dest(const struct elf *elf, struct section *sec, unsigned long offset)
261 {
262 return find_reloc_by_dest_range(elf, sec, offset, 1);
263 }
264
read_sections(struct elf * elf)265 static int read_sections(struct elf *elf)
266 {
267 Elf_Scn *s = NULL;
268 struct section *sec;
269 size_t shstrndx, sections_nr;
270 int i;
271
272 if (elf_getshdrnum(elf->elf, §ions_nr)) {
273 WARN_ELF("elf_getshdrnum");
274 return -1;
275 }
276
277 if (elf_getshdrstrndx(elf->elf, &shstrndx)) {
278 WARN_ELF("elf_getshdrstrndx");
279 return -1;
280 }
281
282 for (i = 0; i < sections_nr; i++) {
283 sec = malloc(sizeof(*sec));
284 if (!sec) {
285 perror("malloc");
286 return -1;
287 }
288 memset(sec, 0, sizeof(*sec));
289
290 INIT_LIST_HEAD(&sec->symbol_list);
291 INIT_LIST_HEAD(&sec->reloc_list);
292
293 s = elf_getscn(elf->elf, i);
294 if (!s) {
295 WARN_ELF("elf_getscn");
296 return -1;
297 }
298
299 sec->idx = elf_ndxscn(s);
300
301 if (!gelf_getshdr(s, &sec->sh)) {
302 WARN_ELF("gelf_getshdr");
303 return -1;
304 }
305
306 sec->name = elf_strptr(elf->elf, shstrndx, sec->sh.sh_name);
307 if (!sec->name) {
308 WARN_ELF("elf_strptr");
309 return -1;
310 }
311
312 if (sec->sh.sh_size != 0) {
313 sec->data = elf_getdata(s, NULL);
314 if (!sec->data) {
315 WARN_ELF("elf_getdata");
316 return -1;
317 }
318 if (sec->data->d_off != 0 ||
319 sec->data->d_size != sec->sh.sh_size) {
320 WARN("unexpected data attributes for %s",
321 sec->name);
322 return -1;
323 }
324 }
325 sec->len = sec->sh.sh_size;
326
327 list_add_tail(&sec->list, &elf->sections);
328 elf_hash_add(elf->section_hash, &sec->hash, sec->idx);
329 elf_hash_add(elf->section_name_hash, &sec->name_hash, str_hash(sec->name));
330 }
331
332 if (stats)
333 printf("nr_sections: %lu\n", (unsigned long)sections_nr);
334
335 /* sanity check, one more call to elf_nextscn() should return NULL */
336 if (elf_nextscn(elf->elf, s)) {
337 WARN("section entry mismatch");
338 return -1;
339 }
340
341 return 0;
342 }
343
elf_add_symbol(struct elf * elf,struct symbol * sym)344 static void elf_add_symbol(struct elf *elf, struct symbol *sym)
345 {
346 struct list_head *entry;
347 struct rb_node *pnode;
348
349 sym->alias = sym;
350
351 sym->type = GELF_ST_TYPE(sym->sym.st_info);
352 sym->bind = GELF_ST_BIND(sym->sym.st_info);
353
354 sym->offset = sym->sym.st_value;
355 sym->len = sym->sym.st_size;
356
357 rb_add(&sym->sec->symbol_tree, &sym->node, symbol_to_offset);
358 pnode = rb_prev(&sym->node);
359 if (pnode)
360 entry = &rb_entry(pnode, struct symbol, node)->list;
361 else
362 entry = &sym->sec->symbol_list;
363 list_add(&sym->list, entry);
364 elf_hash_add(elf->symbol_hash, &sym->hash, sym->idx);
365 elf_hash_add(elf->symbol_name_hash, &sym->name_hash, str_hash(sym->name));
366
367 /*
368 * Don't store empty STT_NOTYPE symbols in the rbtree. They
369 * can exist within a function, confusing the sorting.
370 */
371 if (!sym->len)
372 rb_erase(&sym->node, &sym->sec->symbol_tree);
373 }
374
read_symbols(struct elf * elf)375 static int read_symbols(struct elf *elf)
376 {
377 struct section *symtab, *symtab_shndx, *sec;
378 struct symbol *sym, *pfunc;
379 int symbols_nr, i;
380 char *coldstr;
381 Elf_Data *shndx_data = NULL;
382 Elf32_Word shndx;
383
384 symtab = find_section_by_name(elf, ".symtab");
385 if (!symtab) {
386 /*
387 * A missing symbol table is actually possible if it's an empty
388 * .o file. This can happen for thunk_64.o.
389 */
390 return 0;
391 }
392
393 symtab_shndx = find_section_by_name(elf, ".symtab_shndx");
394 if (symtab_shndx)
395 shndx_data = symtab_shndx->data;
396
397 symbols_nr = symtab->sh.sh_size / symtab->sh.sh_entsize;
398
399 for (i = 0; i < symbols_nr; i++) {
400 sym = malloc(sizeof(*sym));
401 if (!sym) {
402 perror("malloc");
403 return -1;
404 }
405 memset(sym, 0, sizeof(*sym));
406
407 sym->idx = i;
408
409 if (!gelf_getsymshndx(symtab->data, shndx_data, i, &sym->sym,
410 &shndx)) {
411 WARN_ELF("gelf_getsymshndx");
412 goto err;
413 }
414
415 sym->name = elf_strptr(elf->elf, symtab->sh.sh_link,
416 sym->sym.st_name);
417 if (!sym->name) {
418 WARN_ELF("elf_strptr");
419 goto err;
420 }
421
422 if ((sym->sym.st_shndx > SHN_UNDEF &&
423 sym->sym.st_shndx < SHN_LORESERVE) ||
424 (shndx_data && sym->sym.st_shndx == SHN_XINDEX)) {
425 if (sym->sym.st_shndx != SHN_XINDEX)
426 shndx = sym->sym.st_shndx;
427
428 sym->sec = find_section_by_index(elf, shndx);
429 if (!sym->sec) {
430 WARN("couldn't find section for symbol %s",
431 sym->name);
432 goto err;
433 }
434 if (GELF_ST_TYPE(sym->sym.st_info) == STT_SECTION) {
435 sym->name = sym->sec->name;
436 sym->sec->sym = sym;
437 }
438 } else
439 sym->sec = find_section_by_index(elf, 0);
440
441 elf_add_symbol(elf, sym);
442 }
443
444 if (stats)
445 printf("nr_symbols: %lu\n", (unsigned long)symbols_nr);
446
447 /* Create parent/child links for any cold subfunctions */
448 list_for_each_entry(sec, &elf->sections, list) {
449 list_for_each_entry(sym, &sec->symbol_list, list) {
450 char pname[MAX_NAME_LEN + 1];
451 size_t pnamelen;
452 if (sym->type != STT_FUNC)
453 continue;
454
455 if (sym->pfunc == NULL)
456 sym->pfunc = sym;
457
458 if (sym->cfunc == NULL)
459 sym->cfunc = sym;
460
461 coldstr = strstr(sym->name, ".cold");
462 if (!coldstr)
463 continue;
464
465 pnamelen = coldstr - sym->name;
466 if (pnamelen > MAX_NAME_LEN) {
467 WARN("%s(): parent function name exceeds maximum length of %d characters",
468 sym->name, MAX_NAME_LEN);
469 return -1;
470 }
471
472 strncpy(pname, sym->name, pnamelen);
473 pname[pnamelen] = '\0';
474 pfunc = find_symbol_by_name(elf, pname);
475
476 if (!pfunc) {
477 WARN("%s(): can't find parent function",
478 sym->name);
479 return -1;
480 }
481
482 sym->pfunc = pfunc;
483 pfunc->cfunc = sym;
484
485 /*
486 * Unfortunately, -fnoreorder-functions puts the child
487 * inside the parent. Remove the overlap so we can
488 * have sane assumptions.
489 *
490 * Note that pfunc->len now no longer matches
491 * pfunc->sym.st_size.
492 */
493 if (sym->sec == pfunc->sec &&
494 sym->offset >= pfunc->offset &&
495 sym->offset + sym->len == pfunc->offset + pfunc->len) {
496 pfunc->len -= sym->len;
497 }
498 }
499 }
500
501 return 0;
502
503 err:
504 free(sym);
505 return -1;
506 }
507
508 static struct section *elf_create_reloc_section(struct elf *elf,
509 struct section *base,
510 int reltype);
511
elf_add_reloc(struct elf * elf,struct section * sec,unsigned long offset,unsigned int type,struct symbol * sym,s64 addend)512 int elf_add_reloc(struct elf *elf, struct section *sec, unsigned long offset,
513 unsigned int type, struct symbol *sym, s64 addend)
514 {
515 struct reloc *reloc;
516
517 if (!sec->reloc && !elf_create_reloc_section(elf, sec, SHT_RELA))
518 return -1;
519
520 reloc = malloc(sizeof(*reloc));
521 if (!reloc) {
522 perror("malloc");
523 return -1;
524 }
525 memset(reloc, 0, sizeof(*reloc));
526
527 reloc->sec = sec->reloc;
528 reloc->offset = offset;
529 reloc->type = type;
530 reloc->sym = sym;
531 reloc->addend = addend;
532
533 list_add_tail(&reloc->list, &sec->reloc->reloc_list);
534 elf_hash_add(elf->reloc_hash, &reloc->hash, reloc_hash(reloc));
535
536 sec->reloc->changed = true;
537
538 return 0;
539 }
540
541 /*
542 * Ensure that any reloc section containing references to @sym is marked
543 * changed such that it will get re-generated in elf_rebuild_reloc_sections()
544 * with the new symbol index.
545 */
elf_dirty_reloc_sym(struct elf * elf,struct symbol * sym)546 static void elf_dirty_reloc_sym(struct elf *elf, struct symbol *sym)
547 {
548 struct section *sec;
549
550 list_for_each_entry(sec, &elf->sections, list) {
551 struct reloc *reloc;
552
553 if (sec->changed)
554 continue;
555
556 list_for_each_entry(reloc, &sec->reloc_list, list) {
557 if (reloc->sym == sym) {
558 sec->changed = true;
559 break;
560 }
561 }
562 }
563 }
564
565 /*
566 * The libelf API is terrible; gelf_update_sym*() takes a data block relative
567 * index value, *NOT* the symbol index. As such, iterate the data blocks and
568 * adjust index until it fits.
569 *
570 * If no data block is found, allow adding a new data block provided the index
571 * is only one past the end.
572 */
elf_update_symbol(struct elf * elf,struct section * symtab,struct section * symtab_shndx,struct symbol * sym)573 static int elf_update_symbol(struct elf *elf, struct section *symtab,
574 struct section *symtab_shndx, struct symbol *sym)
575 {
576 Elf32_Word shndx = sym->sec ? sym->sec->idx : SHN_UNDEF;
577 Elf_Data *symtab_data = NULL, *shndx_data = NULL;
578 Elf64_Xword entsize = symtab->sh.sh_entsize;
579 int max_idx, idx = sym->idx;
580 Elf_Scn *s, *t = NULL;
581 bool is_special_shndx = sym->sym.st_shndx >= SHN_LORESERVE &&
582 sym->sym.st_shndx != SHN_XINDEX;
583
584 if (is_special_shndx)
585 shndx = sym->sym.st_shndx;
586
587 s = elf_getscn(elf->elf, symtab->idx);
588 if (!s) {
589 WARN_ELF("elf_getscn");
590 return -1;
591 }
592
593 if (symtab_shndx) {
594 t = elf_getscn(elf->elf, symtab_shndx->idx);
595 if (!t) {
596 WARN_ELF("elf_getscn");
597 return -1;
598 }
599 }
600
601 for (;;) {
602 /* get next data descriptor for the relevant sections */
603 symtab_data = elf_getdata(s, symtab_data);
604 if (t)
605 shndx_data = elf_getdata(t, shndx_data);
606
607 /* end-of-list */
608 if (!symtab_data) {
609 void *buf;
610
611 if (idx) {
612 /* we don't do holes in symbol tables */
613 WARN("index out of range");
614 return -1;
615 }
616
617 /* if @idx == 0, it's the next contiguous entry, create it */
618 symtab_data = elf_newdata(s);
619 if (t)
620 shndx_data = elf_newdata(t);
621
622 buf = calloc(1, entsize);
623 if (!buf) {
624 WARN("malloc");
625 return -1;
626 }
627
628 symtab_data->d_buf = buf;
629 symtab_data->d_size = entsize;
630 symtab_data->d_align = 1;
631 symtab_data->d_type = ELF_T_SYM;
632
633 symtab->sh.sh_size += entsize;
634 symtab->changed = true;
635
636 if (t) {
637 shndx_data->d_buf = &sym->sec->idx;
638 shndx_data->d_size = sizeof(Elf32_Word);
639 shndx_data->d_align = sizeof(Elf32_Word);
640 shndx_data->d_type = ELF_T_WORD;
641
642 symtab_shndx->sh.sh_size += sizeof(Elf32_Word);
643 symtab_shndx->changed = true;
644 }
645
646 break;
647 }
648
649 /* empty blocks should not happen */
650 if (!symtab_data->d_size) {
651 WARN("zero size data");
652 return -1;
653 }
654
655 /* is this the right block? */
656 max_idx = symtab_data->d_size / entsize;
657 if (idx < max_idx)
658 break;
659
660 /* adjust index and try again */
661 idx -= max_idx;
662 }
663
664 /* something went side-ways */
665 if (idx < 0) {
666 WARN("negative index");
667 return -1;
668 }
669
670 /* setup extended section index magic and write the symbol */
671 if ((shndx >= SHN_UNDEF && shndx < SHN_LORESERVE) || is_special_shndx) {
672 sym->sym.st_shndx = shndx;
673 if (!shndx_data)
674 shndx = 0;
675 } else {
676 sym->sym.st_shndx = SHN_XINDEX;
677 if (!shndx_data) {
678 WARN("no .symtab_shndx");
679 return -1;
680 }
681 }
682
683 if (!gelf_update_symshndx(symtab_data, shndx_data, idx, &sym->sym, shndx)) {
684 WARN_ELF("gelf_update_symshndx");
685 return -1;
686 }
687
688 return 0;
689 }
690
691 static struct symbol *
elf_create_section_symbol(struct elf * elf,struct section * sec)692 elf_create_section_symbol(struct elf *elf, struct section *sec)
693 {
694 struct section *symtab, *symtab_shndx;
695 Elf32_Word first_non_local, new_idx;
696 struct symbol *sym, *old;
697
698 symtab = find_section_by_name(elf, ".symtab");
699 if (symtab) {
700 symtab_shndx = find_section_by_name(elf, ".symtab_shndx");
701 } else {
702 WARN("no .symtab");
703 return NULL;
704 }
705
706 sym = calloc(1, sizeof(*sym));
707 if (!sym) {
708 perror("malloc");
709 return NULL;
710 }
711
712 sym->name = sec->name;
713 sym->sec = sec;
714
715 // st_name 0
716 sym->sym.st_info = GELF_ST_INFO(STB_LOCAL, STT_SECTION);
717 // st_other 0
718 // st_value 0
719 // st_size 0
720
721 /*
722 * Move the first global symbol, as per sh_info, into a new, higher
723 * symbol index. This fees up a spot for a new local symbol.
724 */
725 first_non_local = symtab->sh.sh_info;
726 new_idx = symtab->sh.sh_size / symtab->sh.sh_entsize;
727 old = find_symbol_by_index(elf, first_non_local);
728 if (old) {
729 old->idx = new_idx;
730
731 hlist_del(&old->hash);
732 elf_hash_add(elf->symbol_hash, &old->hash, old->idx);
733
734 elf_dirty_reloc_sym(elf, old);
735
736 if (elf_update_symbol(elf, symtab, symtab_shndx, old)) {
737 WARN("elf_update_symbol move");
738 return NULL;
739 }
740
741 new_idx = first_non_local;
742 }
743
744 sym->idx = new_idx;
745 if (elf_update_symbol(elf, symtab, symtab_shndx, sym)) {
746 WARN("elf_update_symbol");
747 return NULL;
748 }
749
750 /*
751 * Either way, we added a LOCAL symbol.
752 */
753 symtab->sh.sh_info += 1;
754
755 elf_add_symbol(elf, sym);
756
757 return sym;
758 }
759
elf_add_reloc_to_insn(struct elf * elf,struct section * sec,unsigned long offset,unsigned int type,struct section * insn_sec,unsigned long insn_off)760 int elf_add_reloc_to_insn(struct elf *elf, struct section *sec,
761 unsigned long offset, unsigned int type,
762 struct section *insn_sec, unsigned long insn_off)
763 {
764 struct symbol *sym = insn_sec->sym;
765 int addend = insn_off;
766
767 if (!sym) {
768 /*
769 * Due to how weak functions work, we must use section based
770 * relocations. Symbol based relocations would result in the
771 * weak and non-weak function annotations being overlaid on the
772 * non-weak function after linking.
773 */
774 sym = elf_create_section_symbol(elf, insn_sec);
775 if (!sym)
776 return -1;
777
778 insn_sec->sym = sym;
779 }
780
781 return elf_add_reloc(elf, sec, offset, type, sym, addend);
782 }
783
read_rel_reloc(struct section * sec,int i,struct reloc * reloc,unsigned int * symndx)784 static int read_rel_reloc(struct section *sec, int i, struct reloc *reloc, unsigned int *symndx)
785 {
786 if (!gelf_getrel(sec->data, i, &reloc->rel)) {
787 WARN_ELF("gelf_getrel");
788 return -1;
789 }
790 reloc->type = GELF_R_TYPE(reloc->rel.r_info);
791 reloc->addend = 0;
792 reloc->offset = reloc->rel.r_offset;
793 *symndx = GELF_R_SYM(reloc->rel.r_info);
794 return 0;
795 }
796
read_rela_reloc(struct section * sec,int i,struct reloc * reloc,unsigned int * symndx)797 static int read_rela_reloc(struct section *sec, int i, struct reloc *reloc, unsigned int *symndx)
798 {
799 if (!gelf_getrela(sec->data, i, &reloc->rela)) {
800 WARN_ELF("gelf_getrela");
801 return -1;
802 }
803 reloc->type = GELF_R_TYPE(reloc->rela.r_info);
804 reloc->addend = reloc->rela.r_addend;
805 reloc->offset = reloc->rela.r_offset;
806 *symndx = GELF_R_SYM(reloc->rela.r_info);
807 return 0;
808 }
809
read_relocs(struct elf * elf)810 static int read_relocs(struct elf *elf)
811 {
812 struct section *sec;
813 struct reloc *reloc;
814 int i;
815 unsigned int symndx;
816 unsigned long nr_reloc, max_reloc = 0, tot_reloc = 0;
817
818 list_for_each_entry(sec, &elf->sections, list) {
819 if ((sec->sh.sh_type != SHT_RELA) &&
820 (sec->sh.sh_type != SHT_REL))
821 continue;
822
823 sec->base = find_section_by_index(elf, sec->sh.sh_info);
824 if (!sec->base) {
825 WARN("can't find base section for reloc section %s",
826 sec->name);
827 return -1;
828 }
829
830 sec->base->reloc = sec;
831
832 nr_reloc = 0;
833 for (i = 0; i < sec->sh.sh_size / sec->sh.sh_entsize; i++) {
834 reloc = malloc(sizeof(*reloc));
835 if (!reloc) {
836 perror("malloc");
837 return -1;
838 }
839 memset(reloc, 0, sizeof(*reloc));
840 switch (sec->sh.sh_type) {
841 case SHT_REL:
842 if (read_rel_reloc(sec, i, reloc, &symndx))
843 return -1;
844 break;
845 case SHT_RELA:
846 if (read_rela_reloc(sec, i, reloc, &symndx))
847 return -1;
848 break;
849 default: return -1;
850 }
851
852 reloc->sec = sec;
853 reloc->idx = i;
854 reloc->sym = find_symbol_by_index(elf, symndx);
855 if (!reloc->sym) {
856 WARN("can't find reloc entry symbol %d for %s",
857 symndx, sec->name);
858 return -1;
859 }
860
861 list_add_tail(&reloc->list, &sec->reloc_list);
862 elf_hash_add(elf->reloc_hash, &reloc->hash, reloc_hash(reloc));
863
864 nr_reloc++;
865 }
866 max_reloc = max(max_reloc, nr_reloc);
867 tot_reloc += nr_reloc;
868 }
869
870 if (stats) {
871 printf("max_reloc: %lu\n", max_reloc);
872 printf("tot_reloc: %lu\n", tot_reloc);
873 }
874
875 return 0;
876 }
877
elf_open_read(const char * name,int flags)878 struct elf *elf_open_read(const char *name, int flags)
879 {
880 struct elf *elf;
881 Elf_Cmd cmd;
882
883 elf_version(EV_CURRENT);
884
885 elf = malloc(sizeof(*elf));
886 if (!elf) {
887 perror("malloc");
888 return NULL;
889 }
890 memset(elf, 0, offsetof(struct elf, sections));
891
892 INIT_LIST_HEAD(&elf->sections);
893
894 elf_hash_init(elf->symbol_hash);
895 elf_hash_init(elf->symbol_name_hash);
896 elf_hash_init(elf->section_hash);
897 elf_hash_init(elf->section_name_hash);
898 elf_hash_init(elf->reloc_hash);
899
900 elf->fd = open(name, flags);
901 if (elf->fd == -1) {
902 fprintf(stderr, "objtool: Can't open '%s': %s\n",
903 name, strerror(errno));
904 goto err;
905 }
906
907 if ((flags & O_ACCMODE) == O_RDONLY)
908 cmd = ELF_C_READ_MMAP;
909 else if ((flags & O_ACCMODE) == O_RDWR)
910 cmd = ELF_C_RDWR;
911 else /* O_WRONLY */
912 cmd = ELF_C_WRITE;
913
914 elf->elf = elf_begin(elf->fd, cmd, NULL);
915 if (!elf->elf) {
916 WARN_ELF("elf_begin");
917 goto err;
918 }
919
920 if (!gelf_getehdr(elf->elf, &elf->ehdr)) {
921 WARN_ELF("gelf_getehdr");
922 goto err;
923 }
924
925 if (read_sections(elf))
926 goto err;
927
928 if (read_symbols(elf))
929 goto err;
930
931 if (read_relocs(elf))
932 goto err;
933
934 return elf;
935
936 err:
937 elf_close(elf);
938 return NULL;
939 }
940
elf_add_string(struct elf * elf,struct section * strtab,char * str)941 static int elf_add_string(struct elf *elf, struct section *strtab, char *str)
942 {
943 Elf_Data *data;
944 Elf_Scn *s;
945 int len;
946
947 if (!strtab)
948 strtab = find_section_by_name(elf, ".strtab");
949 if (!strtab) {
950 WARN("can't find .strtab section");
951 return -1;
952 }
953
954 s = elf_getscn(elf->elf, strtab->idx);
955 if (!s) {
956 WARN_ELF("elf_getscn");
957 return -1;
958 }
959
960 data = elf_newdata(s);
961 if (!data) {
962 WARN_ELF("elf_newdata");
963 return -1;
964 }
965
966 data->d_buf = str;
967 data->d_size = strlen(str) + 1;
968 data->d_align = 1;
969 data->d_type = ELF_T_SYM;
970
971 len = strtab->len;
972 strtab->len += data->d_size;
973 strtab->changed = true;
974
975 return len;
976 }
977
elf_create_section(struct elf * elf,const char * name,unsigned int sh_flags,size_t entsize,int nr)978 struct section *elf_create_section(struct elf *elf, const char *name,
979 unsigned int sh_flags, size_t entsize, int nr)
980 {
981 struct section *sec, *shstrtab;
982 size_t size = entsize * nr;
983 Elf_Scn *s;
984
985 sec = malloc(sizeof(*sec));
986 if (!sec) {
987 perror("malloc");
988 return NULL;
989 }
990 memset(sec, 0, sizeof(*sec));
991
992 INIT_LIST_HEAD(&sec->symbol_list);
993 INIT_LIST_HEAD(&sec->reloc_list);
994
995 s = elf_newscn(elf->elf);
996 if (!s) {
997 WARN_ELF("elf_newscn");
998 return NULL;
999 }
1000
1001 sec->name = strdup(name);
1002 if (!sec->name) {
1003 perror("strdup");
1004 return NULL;
1005 }
1006
1007 sec->idx = elf_ndxscn(s);
1008 sec->len = size;
1009 sec->changed = true;
1010
1011 sec->data = elf_newdata(s);
1012 if (!sec->data) {
1013 WARN_ELF("elf_newdata");
1014 return NULL;
1015 }
1016
1017 sec->data->d_size = size;
1018 sec->data->d_align = 1;
1019
1020 if (size) {
1021 sec->data->d_buf = malloc(size);
1022 if (!sec->data->d_buf) {
1023 perror("malloc");
1024 return NULL;
1025 }
1026 memset(sec->data->d_buf, 0, size);
1027 }
1028
1029 if (!gelf_getshdr(s, &sec->sh)) {
1030 WARN_ELF("gelf_getshdr");
1031 return NULL;
1032 }
1033
1034 sec->sh.sh_size = size;
1035 sec->sh.sh_entsize = entsize;
1036 sec->sh.sh_type = SHT_PROGBITS;
1037 sec->sh.sh_addralign = 1;
1038 sec->sh.sh_flags = SHF_ALLOC | sh_flags;
1039
1040 /* Add section name to .shstrtab (or .strtab for Clang) */
1041 shstrtab = find_section_by_name(elf, ".shstrtab");
1042 if (!shstrtab)
1043 shstrtab = find_section_by_name(elf, ".strtab");
1044 if (!shstrtab) {
1045 WARN("can't find .shstrtab or .strtab section");
1046 return NULL;
1047 }
1048 sec->sh.sh_name = elf_add_string(elf, shstrtab, sec->name);
1049 if (sec->sh.sh_name == -1)
1050 return NULL;
1051
1052 list_add_tail(&sec->list, &elf->sections);
1053 elf_hash_add(elf->section_hash, &sec->hash, sec->idx);
1054 elf_hash_add(elf->section_name_hash, &sec->name_hash, str_hash(sec->name));
1055
1056 elf->changed = true;
1057
1058 return sec;
1059 }
1060
elf_create_rel_reloc_section(struct elf * elf,struct section * base)1061 static struct section *elf_create_rel_reloc_section(struct elf *elf, struct section *base)
1062 {
1063 char *relocname;
1064 struct section *sec;
1065
1066 relocname = malloc(strlen(base->name) + strlen(".rel") + 1);
1067 if (!relocname) {
1068 perror("malloc");
1069 return NULL;
1070 }
1071 strcpy(relocname, ".rel");
1072 strcat(relocname, base->name);
1073
1074 sec = elf_create_section(elf, relocname, 0, sizeof(GElf_Rel), 0);
1075 free(relocname);
1076 if (!sec)
1077 return NULL;
1078
1079 base->reloc = sec;
1080 sec->base = base;
1081
1082 sec->sh.sh_type = SHT_REL;
1083 sec->sh.sh_addralign = 8;
1084 sec->sh.sh_link = find_section_by_name(elf, ".symtab")->idx;
1085 sec->sh.sh_info = base->idx;
1086 sec->sh.sh_flags = SHF_INFO_LINK;
1087
1088 return sec;
1089 }
1090
elf_create_rela_reloc_section(struct elf * elf,struct section * base)1091 static struct section *elf_create_rela_reloc_section(struct elf *elf, struct section *base)
1092 {
1093 char *relocname;
1094 struct section *sec;
1095
1096 relocname = malloc(strlen(base->name) + strlen(".rela") + 1);
1097 if (!relocname) {
1098 perror("malloc");
1099 return NULL;
1100 }
1101 strcpy(relocname, ".rela");
1102 strcat(relocname, base->name);
1103
1104 sec = elf_create_section(elf, relocname, 0, sizeof(GElf_Rela), 0);
1105 free(relocname);
1106 if (!sec)
1107 return NULL;
1108
1109 base->reloc = sec;
1110 sec->base = base;
1111
1112 sec->sh.sh_type = SHT_RELA;
1113 sec->sh.sh_addralign = 8;
1114 sec->sh.sh_link = find_section_by_name(elf, ".symtab")->idx;
1115 sec->sh.sh_info = base->idx;
1116 sec->sh.sh_flags = SHF_INFO_LINK;
1117
1118 return sec;
1119 }
1120
elf_create_reloc_section(struct elf * elf,struct section * base,int reltype)1121 static struct section *elf_create_reloc_section(struct elf *elf,
1122 struct section *base,
1123 int reltype)
1124 {
1125 switch (reltype) {
1126 case SHT_REL: return elf_create_rel_reloc_section(elf, base);
1127 case SHT_RELA: return elf_create_rela_reloc_section(elf, base);
1128 default: return NULL;
1129 }
1130 }
1131
elf_rebuild_rel_reloc_section(struct section * sec,int nr)1132 static int elf_rebuild_rel_reloc_section(struct section *sec, int nr)
1133 {
1134 struct reloc *reloc;
1135 int idx = 0, size;
1136 GElf_Rel *relocs;
1137
1138 /* Allocate a buffer for relocations */
1139 size = nr * sizeof(*relocs);
1140 relocs = malloc(size);
1141 if (!relocs) {
1142 perror("malloc");
1143 return -1;
1144 }
1145
1146 sec->data->d_buf = relocs;
1147 sec->data->d_size = size;
1148
1149 sec->sh.sh_size = size;
1150
1151 idx = 0;
1152 list_for_each_entry(reloc, &sec->reloc_list, list) {
1153 relocs[idx].r_offset = reloc->offset;
1154 relocs[idx].r_info = GELF_R_INFO(reloc->sym->idx, reloc->type);
1155 idx++;
1156 }
1157
1158 return 0;
1159 }
1160
elf_rebuild_rela_reloc_section(struct section * sec,int nr)1161 static int elf_rebuild_rela_reloc_section(struct section *sec, int nr)
1162 {
1163 struct reloc *reloc;
1164 int idx = 0, size;
1165 GElf_Rela *relocs;
1166
1167 /* Allocate a buffer for relocations with addends */
1168 size = nr * sizeof(*relocs);
1169 relocs = malloc(size);
1170 if (!relocs) {
1171 perror("malloc");
1172 return -1;
1173 }
1174
1175 sec->data->d_buf = relocs;
1176 sec->data->d_size = size;
1177
1178 sec->sh.sh_size = size;
1179
1180 idx = 0;
1181 list_for_each_entry(reloc, &sec->reloc_list, list) {
1182 relocs[idx].r_offset = reloc->offset;
1183 relocs[idx].r_addend = reloc->addend;
1184 relocs[idx].r_info = GELF_R_INFO(reloc->sym->idx, reloc->type);
1185 idx++;
1186 }
1187
1188 return 0;
1189 }
1190
elf_rebuild_reloc_section(struct elf * elf,struct section * sec)1191 static int elf_rebuild_reloc_section(struct elf *elf, struct section *sec)
1192 {
1193 struct reloc *reloc;
1194 int nr;
1195
1196 nr = 0;
1197 list_for_each_entry(reloc, &sec->reloc_list, list)
1198 nr++;
1199
1200 switch (sec->sh.sh_type) {
1201 case SHT_REL: return elf_rebuild_rel_reloc_section(sec, nr);
1202 case SHT_RELA: return elf_rebuild_rela_reloc_section(sec, nr);
1203 default: return -1;
1204 }
1205 }
1206
elf_write_insn(struct elf * elf,struct section * sec,unsigned long offset,unsigned int len,const char * insn)1207 int elf_write_insn(struct elf *elf, struct section *sec,
1208 unsigned long offset, unsigned int len,
1209 const char *insn)
1210 {
1211 Elf_Data *data = sec->data;
1212
1213 if (data->d_type != ELF_T_BYTE || data->d_off) {
1214 WARN("write to unexpected data for section: %s", sec->name);
1215 return -1;
1216 }
1217
1218 memcpy(data->d_buf + offset, insn, len);
1219 elf_flagdata(data, ELF_C_SET, ELF_F_DIRTY);
1220
1221 elf->changed = true;
1222
1223 return 0;
1224 }
1225
elf_write_reloc(struct elf * elf,struct reloc * reloc)1226 int elf_write_reloc(struct elf *elf, struct reloc *reloc)
1227 {
1228 struct section *sec = reloc->sec;
1229
1230 if (sec->sh.sh_type == SHT_REL) {
1231 reloc->rel.r_info = GELF_R_INFO(reloc->sym->idx, reloc->type);
1232 reloc->rel.r_offset = reloc->offset;
1233
1234 if (!gelf_update_rel(sec->data, reloc->idx, &reloc->rel)) {
1235 WARN_ELF("gelf_update_rel");
1236 return -1;
1237 }
1238 } else {
1239 reloc->rela.r_info = GELF_R_INFO(reloc->sym->idx, reloc->type);
1240 reloc->rela.r_addend = reloc->addend;
1241 reloc->rela.r_offset = reloc->offset;
1242
1243 if (!gelf_update_rela(sec->data, reloc->idx, &reloc->rela)) {
1244 WARN_ELF("gelf_update_rela");
1245 return -1;
1246 }
1247 }
1248
1249 elf->changed = true;
1250
1251 return 0;
1252 }
1253
elf_write(struct elf * elf)1254 int elf_write(struct elf *elf)
1255 {
1256 struct section *sec;
1257 Elf_Scn *s;
1258
1259 /* Update changed relocation sections and section headers: */
1260 list_for_each_entry(sec, &elf->sections, list) {
1261 if (sec->changed) {
1262 if (sec->base &&
1263 elf_rebuild_reloc_section(elf, sec)) {
1264 WARN("elf_rebuild_reloc_section");
1265 return -1;
1266 }
1267
1268 s = elf_getscn(elf->elf, sec->idx);
1269 if (!s) {
1270 WARN_ELF("elf_getscn");
1271 return -1;
1272 }
1273 if (!gelf_update_shdr(s, &sec->sh)) {
1274 WARN_ELF("gelf_update_shdr");
1275 return -1;
1276 }
1277
1278 sec->changed = false;
1279 elf->changed = true;
1280 }
1281 }
1282
1283 /* Make sure the new section header entries get updated properly. */
1284 elf_flagelf(elf->elf, ELF_C_SET, ELF_F_DIRTY);
1285
1286 /* Write all changes to the file. */
1287 if (elf_update(elf->elf, ELF_C_WRITE) < 0) {
1288 WARN_ELF("elf_update");
1289 return -1;
1290 }
1291
1292 elf->changed = false;
1293
1294 return 0;
1295 }
1296
elf_close(struct elf * elf)1297 void elf_close(struct elf *elf)
1298 {
1299 struct section *sec, *tmpsec;
1300 struct symbol *sym, *tmpsym;
1301 struct reloc *reloc, *tmpreloc;
1302
1303 if (elf->elf)
1304 elf_end(elf->elf);
1305
1306 if (elf->fd > 0)
1307 close(elf->fd);
1308
1309 list_for_each_entry_safe(sec, tmpsec, &elf->sections, list) {
1310 list_for_each_entry_safe(sym, tmpsym, &sec->symbol_list, list) {
1311 list_del(&sym->list);
1312 hash_del(&sym->hash);
1313 free(sym);
1314 }
1315 list_for_each_entry_safe(reloc, tmpreloc, &sec->reloc_list, list) {
1316 list_del(&reloc->list);
1317 hash_del(&reloc->hash);
1318 free(reloc);
1319 }
1320 list_del(&sec->list);
1321 free(sec);
1322 }
1323
1324 free(elf);
1325 }
1326