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 <sys/mman.h>
13 #include <fcntl.h>
14 #include <stdio.h>
15 #include <stdlib.h>
16 #include <string.h>
17 #include <unistd.h>
18 #include <errno.h>
19 #include <ctype.h>
20 #include <linux/align.h>
21 #include <linux/kernel.h>
22 #include <linux/interval_tree_generic.h>
23 #include <linux/log2.h>
24 #include <objtool/builtin.h>
25 #include <objtool/elf.h>
26 #include <objtool/warn.h>
27
28 static ssize_t demangled_name_len(const char *name);
29
str_hash(const char * str)30 static inline u32 str_hash(const char *str)
31 {
32 return jhash(str, strlen(str), 0);
33 }
34
str_hash_demangled(const char * str)35 static inline u32 str_hash_demangled(const char *str)
36 {
37 return jhash(str, demangled_name_len(str), 0);
38 }
39
40 #define __elf_table(name) (elf->name##_hash)
41 #define __elf_bits(name) (elf->name##_bits)
42
43 #define __elf_table_entry(name, key) \
44 __elf_table(name)[hash_min(key, __elf_bits(name))]
45
46 #define elf_hash_add(name, node, key) \
47 ({ \
48 struct elf_hash_node *__node = node; \
49 __node->next = __elf_table_entry(name, key); \
50 __elf_table_entry(name, key) = __node; \
51 })
52
__elf_hash_del(struct elf_hash_node * node,struct elf_hash_node ** head)53 static inline void __elf_hash_del(struct elf_hash_node *node,
54 struct elf_hash_node **head)
55 {
56 struct elf_hash_node *cur, *prev;
57
58 if (node == *head) {
59 *head = node->next;
60 return;
61 }
62
63 for (prev = NULL, cur = *head; cur; prev = cur, cur = cur->next) {
64 if (cur == node) {
65 prev->next = cur->next;
66 break;
67 }
68 }
69 }
70
71 #define elf_hash_del(name, node, key) \
72 __elf_hash_del(node, &__elf_table_entry(name, key))
73
74 #define elf_list_entry(ptr, type, member) \
75 ({ \
76 typeof(ptr) __ptr = (ptr); \
77 __ptr ? container_of(__ptr, type, member) : NULL; \
78 })
79
80 #define elf_hash_for_each_possible(name, obj, member, key) \
81 for (obj = elf_list_entry(__elf_table_entry(name, key), typeof(*obj), member); \
82 obj; \
83 obj = elf_list_entry(obj->member.next, typeof(*(obj)), member))
84
85 #define elf_alloc_hash(name, size) \
86 ({ \
87 __elf_bits(name) = max(10, ilog2(size)); \
88 __elf_table(name) = mmap(NULL, sizeof(struct elf_hash_node *) << __elf_bits(name), \
89 PROT_READ|PROT_WRITE, \
90 MAP_PRIVATE|MAP_ANON, -1, 0); \
91 if (__elf_table(name) == (void *)-1L) { \
92 ERROR_GLIBC("mmap fail " #name); \
93 __elf_table(name) = NULL; \
94 } \
95 __elf_table(name); \
96 })
97
__sym_start(struct symbol * s)98 static inline unsigned long __sym_start(struct symbol *s)
99 {
100 return s->offset;
101 }
102
__sym_last(struct symbol * s)103 static inline unsigned long __sym_last(struct symbol *s)
104 {
105 return s->offset + (s->len ? s->len - 1 : 0);
106 }
107
108 INTERVAL_TREE_DEFINE(struct symbol, node, unsigned long, __subtree_last,
109 __sym_start, __sym_last, static inline __maybe_unused,
110 __sym)
111
112 #define __sym_for_each(_iter, _tree, _start, _end) \
113 for (_iter = __sym_iter_first((_tree), (_start), (_end)); \
114 _iter; _iter = __sym_iter_next(_iter, (_start), (_end)))
115
116 struct symbol_hole {
117 unsigned long key;
118 const struct symbol *sym;
119 };
120
121 /*
122 * Find the last symbol before @offset.
123 */
symbol_hole_by_offset(const void * key,const struct rb_node * node)124 static int symbol_hole_by_offset(const void *key, const struct rb_node *node)
125 {
126 const struct symbol *s = rb_entry(node, struct symbol, node);
127 struct symbol_hole *sh = (void *)key;
128
129 if (sh->key < s->offset)
130 return -1;
131
132 if (sh->key >= s->offset + s->len) {
133 sh->sym = s;
134 return 1;
135 }
136
137 return 0;
138 }
139
find_section_by_name(const struct elf * elf,const char * name)140 struct section *find_section_by_name(const struct elf *elf, const char *name)
141 {
142 struct section *sec;
143
144 elf_hash_for_each_possible(section_name, sec, name_hash, str_hash(name)) {
145 if (!strcmp(sec->name, name))
146 return sec;
147 }
148
149 return NULL;
150 }
151
find_section_by_index(struct elf * elf,unsigned int idx)152 static struct section *find_section_by_index(struct elf *elf,
153 unsigned int idx)
154 {
155 struct section *sec;
156
157 elf_hash_for_each_possible(section, sec, hash, idx) {
158 if (sec->idx == idx)
159 return sec;
160 }
161
162 return NULL;
163 }
164
find_symbol_by_index(struct elf * elf,unsigned int idx)165 static struct symbol *find_symbol_by_index(struct elf *elf, unsigned int idx)
166 {
167 struct symbol *sym;
168
169 elf_hash_for_each_possible(symbol, sym, hash, idx) {
170 if (sym->idx == idx)
171 return sym;
172 }
173
174 return NULL;
175 }
176
find_symbol_by_offset(struct section * sec,unsigned long offset)177 struct symbol *find_symbol_by_offset(struct section *sec, unsigned long offset)
178 {
179 struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
180 struct symbol *sym;
181
182 __sym_for_each(sym, tree, offset, offset) {
183 if (sym->offset == offset && !is_sec_sym(sym))
184 return sym->alias;
185 }
186
187 return NULL;
188 }
189
find_func_by_offset(struct section * sec,unsigned long offset)190 struct symbol *find_func_by_offset(struct section *sec, unsigned long offset)
191 {
192 struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
193 struct symbol *func;
194
195 __sym_for_each(func, tree, offset, offset) {
196 if (func->offset == offset && is_func_sym(func))
197 return func->alias;
198 }
199
200 return NULL;
201 }
202
find_symbol_containing(const struct section * sec,unsigned long offset)203 struct symbol *find_symbol_containing(const struct section *sec, unsigned long offset)
204 {
205 struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
206 struct symbol *sym = NULL, *tmp;
207
208 __sym_for_each(tmp, tree, offset, offset) {
209 if (tmp->len) {
210 if (!sym) {
211 sym = tmp;
212 continue;
213 }
214
215 if (sym->offset != tmp->offset || sym->len != tmp->len) {
216 /*
217 * In the rare case of overlapping symbols,
218 * pick the smaller one.
219 *
220 * TODO: outlaw overlapping symbols
221 */
222 if (tmp->len < sym->len)
223 sym = tmp;
224 }
225 }
226 }
227
228 return sym ? sym->alias : NULL;
229 }
230
231 /*
232 * Returns size of hole starting at @offset.
233 */
find_symbol_hole_containing(const struct section * sec,unsigned long offset)234 int find_symbol_hole_containing(const struct section *sec, unsigned long offset)
235 {
236 struct symbol_hole hole = {
237 .key = offset,
238 .sym = NULL,
239 };
240 struct rb_node *n;
241 struct symbol *s;
242
243 /*
244 * Find the rightmost symbol for which @offset is after it.
245 */
246 n = rb_find(&hole, &sec->symbol_tree.rb_root, symbol_hole_by_offset);
247
248 /* found a symbol that contains @offset */
249 if (n)
250 return 0; /* not a hole */
251
252 /*
253 * @offset >= sym->offset + sym->len, find symbol after it.
254 * When hole.sym is empty, use the first node to compute the hole.
255 * If there is no symbol in the section, the first node will be NULL,
256 * in which case, -1 is returned to skip the whole section.
257 */
258 if (hole.sym)
259 n = rb_next(&hole.sym->node);
260 else
261 n = rb_first_cached(&sec->symbol_tree);
262
263 if (!n)
264 return -1; /* until end of address space */
265
266 /* hole until start of next symbol */
267 s = rb_entry(n, struct symbol, node);
268 return s->offset - offset;
269 }
270
find_func_containing(struct section * sec,unsigned long offset)271 struct symbol *find_func_containing(struct section *sec, unsigned long offset)
272 {
273 struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
274 struct symbol *func;
275
276 __sym_for_each(func, tree, offset, offset) {
277 if (is_func_sym(func))
278 return func->alias;
279 }
280
281 return NULL;
282 }
283
find_symbol_by_name(const struct elf * elf,const char * name)284 struct symbol *find_symbol_by_name(const struct elf *elf, const char *name)
285 {
286 struct symbol *sym;
287
288 elf_hash_for_each_possible(symbol_name, sym, name_hash, str_hash(name)) {
289 if (!strcmp(sym->name, name))
290 return sym;
291 }
292
293 return NULL;
294 }
295
296 /* Find local symbol with matching STT_FILE */
find_local_symbol_by_file_and_name(const struct elf * elf,struct symbol * file,const char * name)297 static struct symbol *find_local_symbol_by_file_and_name(const struct elf *elf,
298 struct symbol *file,
299 const char *name)
300 {
301 struct symbol *sym;
302
303 elf_hash_for_each_possible(symbol_name, sym, name_hash, str_hash_demangled(name)) {
304 if (sym->bind == STB_LOCAL && sym->file == file &&
305 !strcmp(sym->name, name)) {
306 return sym;
307 }
308 }
309
310 return NULL;
311 }
312
find_global_symbol_by_name(const struct elf * elf,const char * name)313 struct symbol *find_global_symbol_by_name(const struct elf *elf, const char *name)
314 {
315 struct symbol *sym;
316
317 elf_hash_for_each_possible(symbol_name, sym, name_hash, str_hash_demangled(name)) {
318 if (!strcmp(sym->name, name) && !is_local_sym(sym))
319 return sym;
320 }
321
322 return NULL;
323 }
324
iterate_global_symbol_by_demangled_name(const struct elf * elf,const char * demangled_name,void (* process)(struct symbol * sym,void * data),void * data)325 void iterate_global_symbol_by_demangled_name(const struct elf *elf,
326 const char *demangled_name,
327 void (*process)(struct symbol *sym, void *data),
328 void *data)
329 {
330 struct symbol *sym;
331
332 elf_hash_for_each_possible(symbol_name, sym, name_hash, str_hash(demangled_name)) {
333 if (!strcmp(sym->demangled_name, demangled_name) && !is_local_sym(sym))
334 process(sym, data);
335 }
336 }
337
find_reloc_by_dest_range(const struct elf * elf,struct section * sec,unsigned long offset,unsigned int len)338 struct reloc *find_reloc_by_dest_range(const struct elf *elf, struct section *sec,
339 unsigned long offset, unsigned int len)
340 {
341 struct reloc *reloc, *r = NULL;
342 struct section *rsec;
343 unsigned long o;
344
345 rsec = sec->rsec;
346 if (!rsec)
347 return NULL;
348
349 for_offset_range(o, offset, offset + len) {
350 elf_hash_for_each_possible(reloc, reloc, hash,
351 sec_offset_hash(rsec, o)) {
352 if (reloc->sec != rsec)
353 continue;
354
355 if (reloc_offset(reloc) >= offset &&
356 reloc_offset(reloc) < offset + len) {
357 if (!r || reloc_offset(reloc) < reloc_offset(r))
358 r = reloc;
359 }
360 }
361 if (r)
362 return r;
363 }
364
365 return NULL;
366 }
367
find_reloc_by_dest(const struct elf * elf,struct section * sec,unsigned long offset)368 struct reloc *find_reloc_by_dest(const struct elf *elf, struct section *sec, unsigned long offset)
369 {
370 return find_reloc_by_dest_range(elf, sec, offset, 1);
371 }
372
is_dwarf_section(struct section * sec)373 static bool is_dwarf_section(struct section *sec)
374 {
375 return !strncmp(sec->name, ".debug_", 7);
376 }
377
read_sections(struct elf * elf)378 static int read_sections(struct elf *elf)
379 {
380 Elf_Scn *s = NULL;
381 struct section *sec;
382 size_t shstrndx, sections_nr;
383 int i;
384
385 if (elf_getshdrnum(elf->elf, §ions_nr)) {
386 ERROR_ELF("elf_getshdrnum");
387 return -1;
388 }
389
390 if (elf_getshdrstrndx(elf->elf, &shstrndx)) {
391 ERROR_ELF("elf_getshdrstrndx");
392 return -1;
393 }
394
395 if (!elf_alloc_hash(section, sections_nr) ||
396 !elf_alloc_hash(section_name, sections_nr))
397 return -1;
398
399 elf->section_data = calloc(sections_nr, sizeof(*sec));
400 if (!elf->section_data) {
401 ERROR_GLIBC("calloc");
402 return -1;
403 }
404 for (i = 0; i < sections_nr; i++) {
405 sec = &elf->section_data[i];
406
407 INIT_LIST_HEAD(&sec->symbol_list);
408
409 s = elf_getscn(elf->elf, i);
410 if (!s) {
411 ERROR_ELF("elf_getscn");
412 return -1;
413 }
414
415 sec->idx = elf_ndxscn(s);
416
417 if (!gelf_getshdr(s, &sec->sh)) {
418 ERROR_ELF("gelf_getshdr");
419 return -1;
420 }
421
422 sec->name = elf_strptr(elf->elf, shstrndx, sec->sh.sh_name);
423 if (!sec->name) {
424 ERROR_ELF("elf_strptr");
425 return -1;
426 }
427
428 if (sec_size(sec) != 0 && !is_dwarf_section(sec)) {
429 sec->data = elf_getdata(s, NULL);
430 if (!sec->data) {
431 ERROR_ELF("elf_getdata");
432 return -1;
433 }
434 if (sec->data->d_off != 0 ||
435 sec->data->d_size != sec_size(sec)) {
436 ERROR("unexpected data attributes for %s", sec->name);
437 return -1;
438 }
439 }
440
441 list_add_tail(&sec->list, &elf->sections);
442 elf_hash_add(section, &sec->hash, sec->idx);
443 elf_hash_add(section_name, &sec->name_hash, str_hash(sec->name));
444
445 if (is_reloc_sec(sec))
446 elf->num_relocs += sec_num_entries(sec);
447 }
448
449 if (opts.stats) {
450 printf("nr_sections: %lu\n", (unsigned long)sections_nr);
451 printf("section_bits: %d\n", elf->section_bits);
452 }
453
454 /* sanity check, one more call to elf_nextscn() should return NULL */
455 if (elf_nextscn(elf->elf, s)) {
456 ERROR("section entry mismatch");
457 return -1;
458 }
459
460 return 0;
461 }
462
463 /*
464 * Returns desired length of the demangled name.
465 * If name doesn't need demangling, return strlen(name).
466 */
demangled_name_len(const char * name)467 static ssize_t demangled_name_len(const char *name)
468 {
469 ssize_t idx;
470 const char *p;
471
472 p = strstr(name, ".llvm.");
473 if (p)
474 return p - name;
475
476 if (!strstarts(name, "__UNIQUE_ID_") && !strchr(name, '.'))
477 return strlen(name);
478
479 for (idx = strlen(name) - 1; idx >= 0; idx--) {
480 char c = name[idx];
481
482 if (!isdigit(c) && c != '.' && c != '_')
483 break;
484 }
485 if (idx <= 0)
486 return strlen(name);
487 return idx + 1;
488 }
489
490 /*
491 * Remove number suffix of a symbol.
492 *
493 * Specifically, remove trailing numbers for "__UNIQUE_ID_" symbols and
494 * symbols with '.'.
495 *
496 * With CONFIG_LTO_CLANG_THIN, it is possible to have nested __UNIQUE_ID_,
497 * such as
498 *
499 * __UNIQUE_ID_addressable___UNIQUE_ID_pci_invalid_bar_694_695
500 *
501 * to remove both trailing numbers, also remove trailing '_'.
502 *
503 * For symbols with llvm suffix, i.e., foo.llvm.<hash>, remove the
504 * .llvm.<hash> part.
505 */
demangle_name(struct symbol * sym)506 static const char *demangle_name(struct symbol *sym)
507 {
508 char *str;
509 ssize_t len;
510
511 if (!is_func_sym(sym) && !is_object_sym(sym))
512 return sym->name;
513
514 len = demangled_name_len(sym->name);
515 if (len == strlen(sym->name))
516 return sym->name;
517
518 str = strndup(sym->name, len);
519 if (!str) {
520 ERROR_GLIBC("strdup");
521 return NULL;
522 }
523
524 return str;
525 }
526
elf_add_symbol(struct elf * elf,struct symbol * sym)527 static int elf_add_symbol(struct elf *elf, struct symbol *sym)
528 {
529 struct list_head *entry;
530 struct rb_node *pnode;
531 struct symbol *iter;
532
533 INIT_LIST_HEAD(&sym->pv_target);
534 sym->alias = sym;
535
536 sym->type = GELF_ST_TYPE(sym->sym.st_info);
537 sym->bind = GELF_ST_BIND(sym->sym.st_info);
538
539 if (is_file_sym(sym))
540 elf->num_files++;
541
542 sym->offset = sym->sym.st_value;
543 sym->len = sym->sym.st_size;
544
545 __sym_for_each(iter, &sym->sec->symbol_tree, sym->offset, sym->offset) {
546 if (!is_undef_sym(iter) && iter->offset == sym->offset &&
547 iter->type == sym->type && iter->len == sym->len)
548 iter->alias = sym;
549 }
550
551 __sym_insert(sym, &sym->sec->symbol_tree);
552 pnode = rb_prev(&sym->node);
553 if (pnode)
554 entry = &rb_entry(pnode, struct symbol, node)->list;
555 else
556 entry = &sym->sec->symbol_list;
557 list_add(&sym->list, entry);
558
559 sym->demangled_name = demangle_name(sym);
560 if (!sym->demangled_name)
561 return -1;
562
563 list_add_tail(&sym->global_list, &elf->symbols);
564 elf_hash_add(symbol, &sym->hash, sym->idx);
565 elf_hash_add(symbol_name, &sym->name_hash, str_hash(sym->demangled_name));
566
567 if (is_func_sym(sym) &&
568 (strstarts(sym->name, "__pfx_") ||
569 strstarts(sym->name, "__cfi_") ||
570 strstarts(sym->name, "__pi___pfx_") ||
571 strstarts(sym->name, "__pi___cfi_")))
572 sym->prefix = 1;
573
574 if (strstarts(sym->name, ".klp.sym"))
575 sym->klp = 1;
576
577 if (!sym->klp && !is_sec_sym(sym) && strstr(sym->name, ".cold")) {
578 sym->cold = 1;
579
580 /*
581 * Clang doesn't mark cold subfunctions as STT_FUNC, which
582 * breaks several objtool assumptions. Fake it.
583 */
584 sym->type = STT_FUNC;
585 }
586
587 sym->pfunc = sym->cfunc = sym;
588
589 return 0;
590 }
591
read_symbols(struct elf * elf)592 static int read_symbols(struct elf *elf)
593 {
594 struct section *symtab, *symtab_shndx, *sec;
595 struct symbol *sym, *pfunc, *file = NULL;
596 int symbols_nr, i;
597 char *coldstr;
598 Elf_Data *shndx_data = NULL;
599 Elf32_Word shndx;
600
601 symtab = find_section_by_name(elf, ".symtab");
602 if (symtab) {
603 symtab_shndx = find_section_by_name(elf, ".symtab_shndx");
604 if (symtab_shndx)
605 shndx_data = symtab_shndx->data;
606
607 symbols_nr = sec_num_entries(symtab);
608 } else {
609 /*
610 * A missing symbol table is actually possible if it's an empty
611 * .o file. This can happen for thunk_64.o. Make sure to at
612 * least allocate the symbol hash tables so we can do symbol
613 * lookups without crashing.
614 */
615 symbols_nr = 0;
616 }
617
618 if (!elf_alloc_hash(symbol, symbols_nr) ||
619 !elf_alloc_hash(symbol_name, symbols_nr))
620 return -1;
621
622 elf->symbol_data = calloc(symbols_nr, sizeof(*sym));
623 if (!elf->symbol_data) {
624 ERROR_GLIBC("calloc");
625 return -1;
626 }
627
628 INIT_LIST_HEAD(&elf->symbols);
629
630 for (i = 0; i < symbols_nr; i++) {
631 sym = &elf->symbol_data[i];
632
633 sym->idx = i;
634
635 if (!gelf_getsymshndx(symtab->data, shndx_data, i, &sym->sym,
636 &shndx)) {
637 ERROR_ELF("gelf_getsymshndx");
638 return -1;
639 }
640
641 sym->name = elf_strptr(elf->elf, symtab->sh.sh_link,
642 sym->sym.st_name);
643 if (!sym->name) {
644 ERROR_ELF("elf_strptr");
645 return -1;
646 }
647
648 if ((sym->sym.st_shndx > SHN_UNDEF &&
649 sym->sym.st_shndx < SHN_LORESERVE) ||
650 (shndx_data && sym->sym.st_shndx == SHN_XINDEX)) {
651 if (sym->sym.st_shndx != SHN_XINDEX)
652 shndx = sym->sym.st_shndx;
653
654 sym->sec = find_section_by_index(elf, shndx);
655 if (!sym->sec) {
656 ERROR("couldn't find section for symbol %s", sym->name);
657 return -1;
658 }
659 if (GELF_ST_TYPE(sym->sym.st_info) == STT_SECTION) {
660 sym->name = sym->sec->name;
661 sym->sec->sym = sym;
662 }
663 } else
664 sym->sec = find_section_by_index(elf, 0);
665
666 if (elf_add_symbol(elf, sym))
667 return -1;
668
669 if (is_file_sym(sym))
670 file = sym;
671 else if (sym->bind == STB_LOCAL)
672 sym->file = file;
673 }
674
675 if (opts.stats) {
676 printf("nr_symbols: %lu\n", (unsigned long)symbols_nr);
677 printf("symbol_bits: %d\n", elf->symbol_bits);
678 }
679
680 /* Create parent/child links for any cold subfunctions */
681 list_for_each_entry(sec, &elf->sections, list) {
682 sec_for_each_sym(sec, sym) {
683 char *pname;
684 size_t pnamelen;
685
686 if (!sym->cold)
687 continue;
688
689 coldstr = strstr(sym->name, ".cold");
690 if (!coldstr) {
691 ERROR("%s(): cold subfunction without \".cold\"?", sym->name);
692 return -1;
693 }
694
695 pnamelen = coldstr - sym->name;
696 pname = strndup(sym->name, pnamelen);
697 if (!pname) {
698 ERROR("%s(): failed to allocate memory", sym->name);
699 return -1;
700 }
701
702 pfunc = find_local_symbol_by_file_and_name(elf, sym->file, pname);
703 if (!pfunc)
704 pfunc = find_global_symbol_by_name(elf, pname);
705 free(pname);
706
707 if (!pfunc) {
708 ERROR("%s(): can't find parent function", sym->name);
709 return -1;
710 }
711
712 sym->pfunc = pfunc->alias;
713 pfunc->cfunc = sym;
714 pfunc->alias->cfunc = sym;
715
716 /*
717 * Unfortunately, -fnoreorder-functions puts the child
718 * inside the parent. Remove the overlap so we can
719 * have sane assumptions.
720 *
721 * Note that pfunc->len now no longer matches
722 * pfunc->sym.st_size.
723 */
724 if (sym->sec == pfunc->sec &&
725 sym->offset >= pfunc->offset &&
726 sym->offset + sym->len == pfunc->offset + pfunc->len) {
727 pfunc->len -= sym->len;
728 }
729 }
730 }
731
732 return 0;
733 }
734
mark_group_syms(struct elf * elf)735 static int mark_group_syms(struct elf *elf)
736 {
737 struct section *symtab, *sec;
738 struct symbol *sym;
739
740 symtab = find_section_by_name(elf, ".symtab");
741 if (!symtab) {
742 ERROR("no .symtab");
743 return -1;
744 }
745
746 for_each_sec(elf, sec) {
747 if (sec->sh.sh_type == SHT_GROUP &&
748 sec->sh.sh_link == symtab->idx) {
749 sym = find_symbol_by_index(elf, sec->sh.sh_info);
750 if (!sym) {
751 ERROR("%s: can't find SHT_GROUP signature symbol",
752 sec->name);
753 return -1;
754 }
755
756 sym->group_sec = sec;
757 }
758 }
759
760 return 0;
761 }
762
763 /*
764 * @sym's idx has changed. Update the relocs which reference it.
765 */
elf_update_sym_relocs(struct elf * elf,struct symbol * sym)766 static int elf_update_sym_relocs(struct elf *elf, struct symbol *sym)
767 {
768 struct reloc *reloc;
769
770 for (reloc = sym->relocs; reloc; reloc = sym_next_reloc(reloc))
771 set_reloc_sym(elf, reloc, reloc->sym->idx);
772
773 return 0;
774 }
775
776 /*
777 * The libelf API is terrible; gelf_update_sym*() takes a data block relative
778 * index value, *NOT* the symbol index. As such, iterate the data blocks and
779 * adjust index until it fits.
780 *
781 * If no data block is found, allow adding a new data block provided the index
782 * is only one past the end.
783 */
elf_update_symbol(struct elf * elf,struct section * symtab,struct section * symtab_shndx,struct symbol * sym)784 static int elf_update_symbol(struct elf *elf, struct section *symtab,
785 struct section *symtab_shndx, struct symbol *sym)
786 {
787 Elf32_Word shndx;
788 Elf_Data *symtab_data = NULL, *shndx_data = NULL;
789 Elf64_Xword entsize = symtab->sh.sh_entsize;
790 int max_idx, idx = sym->idx;
791 Elf_Scn *s, *t = NULL;
792 bool is_special_shndx = sym->sym.st_shndx >= SHN_LORESERVE &&
793 sym->sym.st_shndx != SHN_XINDEX;
794
795 shndx = is_special_shndx ? sym->sym.st_shndx : sym->sec->idx;
796
797 s = elf_getscn(elf->elf, symtab->idx);
798 if (!s) {
799 ERROR_ELF("elf_getscn");
800 return -1;
801 }
802
803 if (symtab_shndx) {
804 t = elf_getscn(elf->elf, symtab_shndx->idx);
805 if (!t) {
806 ERROR_ELF("elf_getscn");
807 return -1;
808 }
809 }
810
811 for (;;) {
812 /* get next data descriptor for the relevant sections */
813 symtab_data = elf_getdata(s, symtab_data);
814 if (t)
815 shndx_data = elf_getdata(t, shndx_data);
816
817 /* end-of-list */
818 if (!symtab_data) {
819 /*
820 * Over-allocate to avoid O(n^2) symbol creation
821 * behaviour. The down side is that libelf doesn't
822 * like this; see elf_truncate_section() for the fixup.
823 */
824 int num = max(1U, sym->idx/3);
825 void *buf;
826
827 if (idx) {
828 /* we don't do holes in symbol tables */
829 ERROR("index out of range");
830 return -1;
831 }
832
833 /* if @idx == 0, it's the next contiguous entry, create it */
834 symtab_data = elf_newdata(s);
835 if (t)
836 shndx_data = elf_newdata(t);
837
838 buf = calloc(num, entsize);
839 if (!buf) {
840 ERROR_GLIBC("calloc");
841 return -1;
842 }
843
844 symtab_data->d_buf = buf;
845 symtab_data->d_size = num * entsize;
846 symtab_data->d_align = 1;
847 symtab_data->d_type = ELF_T_SYM;
848
849 mark_sec_changed(elf, symtab, true);
850 symtab->truncate = true;
851
852 if (t) {
853 buf = calloc(num, sizeof(Elf32_Word));
854 if (!buf) {
855 ERROR_GLIBC("calloc");
856 return -1;
857 }
858
859 shndx_data->d_buf = buf;
860 shndx_data->d_size = num * sizeof(Elf32_Word);
861 shndx_data->d_align = sizeof(Elf32_Word);
862 shndx_data->d_type = ELF_T_WORD;
863
864 mark_sec_changed(elf, symtab_shndx, true);
865 symtab_shndx->truncate = true;
866 }
867
868 break;
869 }
870
871 /* empty blocks should not happen */
872 if (!symtab_data->d_size) {
873 ERROR("zero size data");
874 return -1;
875 }
876
877 /* is this the right block? */
878 max_idx = symtab_data->d_size / entsize;
879 if (idx < max_idx)
880 break;
881
882 /* adjust index and try again */
883 idx -= max_idx;
884 }
885
886 /* something went side-ways */
887 if (idx < 0) {
888 ERROR("negative index");
889 return -1;
890 }
891
892 /* setup extended section index magic and write the symbol */
893 if (shndx < SHN_LORESERVE || is_special_shndx) {
894 sym->sym.st_shndx = shndx;
895 if (!shndx_data)
896 shndx = 0;
897 } else {
898 sym->sym.st_shndx = SHN_XINDEX;
899 if (!shndx_data) {
900 ERROR("no .symtab_shndx");
901 return -1;
902 }
903 }
904
905 if (!gelf_update_symshndx(symtab_data, shndx_data, idx, &sym->sym, shndx)) {
906 ERROR_ELF("gelf_update_symshndx");
907 return -1;
908 }
909
910 return 0;
911 }
912
elf_create_symbol(struct elf * elf,const char * name,struct section * sec,unsigned int bind,unsigned int type,unsigned long offset,size_t size)913 struct symbol *elf_create_symbol(struct elf *elf, const char *name,
914 struct section *sec, unsigned int bind,
915 unsigned int type, unsigned long offset,
916 size_t size)
917 {
918 struct section *symtab, *symtab_shndx;
919 Elf32_Word first_non_local, new_idx;
920 struct symbol *old, *sym;
921
922 sym = calloc(1, sizeof(*sym));
923 if (!sym) {
924 ERROR_GLIBC("calloc");
925 return NULL;
926 }
927
928 sym->name = strdup(name);
929 if (!sym->name) {
930 ERROR_GLIBC("strdup");
931 return NULL;
932 }
933
934 if (type != STT_SECTION) {
935 sym->sym.st_name = elf_add_string(elf, NULL, sym->name);
936 if (sym->sym.st_name == -1)
937 return NULL;
938 }
939
940 if (sec) {
941 sym->sec = sec;
942 } else {
943 sym->sec = find_section_by_index(elf, 0);
944 if (!sym->sec) {
945 ERROR("no NULL section");
946 return NULL;
947 }
948 }
949
950 sym->sym.st_info = GELF_ST_INFO(bind, type);
951 sym->sym.st_value = offset;
952 sym->sym.st_size = size;
953
954 symtab = find_section_by_name(elf, ".symtab");
955 if (!symtab) {
956 ERROR("no .symtab");
957 return NULL;
958 }
959
960 symtab_shndx = find_section_by_name(elf, ".symtab_shndx");
961
962 new_idx = sec_num_entries(symtab);
963
964 if (bind != STB_LOCAL)
965 goto non_local;
966
967 /*
968 * Move the first global symbol, as per sh_info, into a new, higher
969 * symbol index. This frees up a spot for a new local symbol.
970 */
971 first_non_local = symtab->sh.sh_info;
972 old = find_symbol_by_index(elf, first_non_local);
973 if (old) {
974
975 elf_hash_del(symbol, &old->hash, old->idx);
976 elf_hash_add(symbol, &old->hash, new_idx);
977 old->idx = new_idx;
978
979 if (elf_update_symbol(elf, symtab, symtab_shndx, old)) {
980 ERROR("elf_update_symbol move");
981 return NULL;
982 }
983
984 if (elf_update_sym_relocs(elf, old))
985 return NULL;
986
987 if (old->group_sec) {
988 old->group_sec->sh.sh_info = new_idx;
989 mark_sec_changed(elf, old->group_sec, true);
990 }
991
992 new_idx = first_non_local;
993 }
994
995 /*
996 * Either way, we will add a LOCAL symbol.
997 */
998 symtab->sh.sh_info += 1;
999
1000 non_local:
1001 sym->idx = new_idx;
1002 if (sym->idx && elf_update_symbol(elf, symtab, symtab_shndx, sym))
1003 return NULL;
1004
1005 symtab->sh.sh_size += symtab->sh.sh_entsize;
1006 mark_sec_changed(elf, symtab, true);
1007
1008 if (symtab_shndx) {
1009 symtab_shndx->sh.sh_size += sizeof(Elf32_Word);
1010 mark_sec_changed(elf, symtab_shndx, true);
1011 }
1012
1013 if (elf_add_symbol(elf, sym))
1014 return NULL;
1015
1016 return sym;
1017 }
1018
elf_create_section_symbol(struct elf * elf,struct section * sec)1019 struct symbol *elf_create_section_symbol(struct elf *elf, struct section *sec)
1020 {
1021 struct symbol *sym = calloc(1, sizeof(*sym));
1022
1023 sym = elf_create_symbol(elf, sec->name, sec, STB_LOCAL, STT_SECTION, 0, 0);
1024 if (!sym)
1025 return NULL;
1026
1027 sec->sym = sym;
1028
1029 return sym;
1030 }
1031
elf_init_reloc(struct elf * elf,struct section * rsec,unsigned int reloc_idx,unsigned long offset,struct symbol * sym,s64 addend,unsigned int type)1032 struct reloc *elf_init_reloc(struct elf *elf, struct section *rsec,
1033 unsigned int reloc_idx, unsigned long offset,
1034 struct symbol *sym, s64 addend, unsigned int type)
1035 {
1036 struct reloc *reloc, empty = { 0 };
1037
1038 if (reloc_idx >= sec_num_entries(rsec)) {
1039 ERROR("%s: bad reloc_idx %u for %s with %d relocs",
1040 __func__, reloc_idx, rsec->name, sec_num_entries(rsec));
1041 return NULL;
1042 }
1043
1044 reloc = &rsec->relocs[reloc_idx];
1045
1046 if (memcmp(reloc, &empty, sizeof(empty))) {
1047 ERROR("%s: %s: reloc %d already initialized!",
1048 __func__, rsec->name, reloc_idx);
1049 return NULL;
1050 }
1051
1052 reloc->sec = rsec;
1053 reloc->sym = sym;
1054
1055 set_reloc_offset(elf, reloc, offset);
1056 set_reloc_sym(elf, reloc, sym->idx);
1057 set_reloc_type(elf, reloc, type);
1058 set_reloc_addend(elf, reloc, addend);
1059
1060 elf_hash_add(reloc, &reloc->hash, reloc_hash(reloc));
1061 set_sym_next_reloc(reloc, sym->relocs);
1062 sym->relocs = reloc;
1063
1064 return reloc;
1065 }
1066
elf_init_reloc_text_sym(struct elf * elf,struct section * sec,unsigned long offset,unsigned int reloc_idx,struct section * insn_sec,unsigned long insn_off)1067 struct reloc *elf_init_reloc_text_sym(struct elf *elf, struct section *sec,
1068 unsigned long offset,
1069 unsigned int reloc_idx,
1070 struct section *insn_sec,
1071 unsigned long insn_off)
1072 {
1073 struct symbol *sym = insn_sec->sym;
1074 s64 addend = insn_off;
1075
1076 if (!is_text_sec(insn_sec)) {
1077 ERROR("bad call to %s() for data symbol %s", __func__, sym->name);
1078 return NULL;
1079 }
1080
1081 if (!sym) {
1082 /*
1083 * Due to how weak functions work, we must use section based
1084 * relocations. Symbol based relocations would result in the
1085 * weak and non-weak function annotations being overlaid on the
1086 * non-weak function after linking.
1087 */
1088 sym = elf_create_section_symbol(elf, insn_sec);
1089 if (!sym)
1090 return NULL;
1091 }
1092
1093 return elf_init_reloc(elf, sec->rsec, reloc_idx, offset, sym, addend,
1094 elf_text_rela_type(elf));
1095 }
1096
elf_init_reloc_data_sym(struct elf * elf,struct section * sec,unsigned long offset,unsigned int reloc_idx,struct symbol * sym,s64 addend)1097 struct reloc *elf_init_reloc_data_sym(struct elf *elf, struct section *sec,
1098 unsigned long offset,
1099 unsigned int reloc_idx,
1100 struct symbol *sym,
1101 s64 addend)
1102 {
1103 if (is_text_sec(sec)) {
1104 ERROR("bad call to %s() for text symbol %s", __func__, sym->name);
1105 return NULL;
1106 }
1107
1108 return elf_init_reloc(elf, sec->rsec, reloc_idx, offset, sym, addend,
1109 elf_data_rela_type(elf));
1110 }
1111
read_relocs(struct elf * elf)1112 static int read_relocs(struct elf *elf)
1113 {
1114 unsigned long nr_reloc, max_reloc = 0;
1115 struct section *rsec;
1116 struct reloc *reloc;
1117 unsigned int symndx;
1118 struct symbol *sym;
1119 int i;
1120
1121 if (!elf_alloc_hash(reloc, elf->num_relocs))
1122 return -1;
1123
1124 list_for_each_entry(rsec, &elf->sections, list) {
1125 if (!is_reloc_sec(rsec))
1126 continue;
1127
1128 rsec->base = find_section_by_index(elf, rsec->sh.sh_info);
1129 if (!rsec->base) {
1130 ERROR("can't find base section for reloc section %s", rsec->name);
1131 return -1;
1132 }
1133
1134 rsec->base->rsec = rsec;
1135
1136 /* nr_alloc_relocs=0: libelf owns d_buf */
1137 rsec->nr_alloc_relocs = 0;
1138
1139 rsec->relocs = calloc(sec_num_entries(rsec), sizeof(*reloc));
1140 if (!rsec->relocs) {
1141 ERROR_GLIBC("calloc");
1142 return -1;
1143 }
1144
1145 nr_reloc = 0;
1146 for (i = 0; i < sec_num_entries(rsec); i++) {
1147 reloc = &rsec->relocs[i];
1148
1149 reloc->sec = rsec;
1150 symndx = reloc_sym(reloc);
1151 reloc->sym = sym = find_symbol_by_index(elf, symndx);
1152 if (!reloc->sym) {
1153 ERROR("can't find reloc entry symbol %d for %s", symndx, rsec->name);
1154 return -1;
1155 }
1156
1157 elf_hash_add(reloc, &reloc->hash, reloc_hash(reloc));
1158 set_sym_next_reloc(reloc, sym->relocs);
1159 sym->relocs = reloc;
1160
1161 nr_reloc++;
1162 }
1163 max_reloc = max(max_reloc, nr_reloc);
1164 }
1165
1166 if (opts.stats) {
1167 printf("max_reloc: %lu\n", max_reloc);
1168 printf("num_relocs: %lu\n", elf->num_relocs);
1169 printf("reloc_bits: %d\n", elf->reloc_bits);
1170 }
1171
1172 return 0;
1173 }
1174
elf_open_read(const char * name,int flags)1175 struct elf *elf_open_read(const char *name, int flags)
1176 {
1177 struct elf *elf;
1178 Elf_Cmd cmd;
1179
1180 elf_version(EV_CURRENT);
1181
1182 elf = malloc(sizeof(*elf));
1183 if (!elf) {
1184 ERROR_GLIBC("malloc");
1185 return NULL;
1186 }
1187 memset(elf, 0, sizeof(*elf));
1188
1189 INIT_LIST_HEAD(&elf->sections);
1190
1191 elf->fd = open(name, flags);
1192 if (elf->fd == -1) {
1193 fprintf(stderr, "objtool: Can't open '%s': %s\n",
1194 name, strerror(errno));
1195 goto err;
1196 }
1197
1198 elf->name = strdup(name);
1199 if (!elf->name) {
1200 ERROR_GLIBC("strdup");
1201 return NULL;
1202 }
1203
1204 if ((flags & O_ACCMODE) == O_RDONLY)
1205 cmd = ELF_C_READ_MMAP;
1206 else if ((flags & O_ACCMODE) == O_RDWR)
1207 cmd = ELF_C_RDWR;
1208 else /* O_WRONLY */
1209 cmd = ELF_C_WRITE;
1210
1211 elf->elf = elf_begin(elf->fd, cmd, NULL);
1212 if (!elf->elf) {
1213 ERROR_ELF("elf_begin");
1214 goto err;
1215 }
1216
1217 if (!gelf_getehdr(elf->elf, &elf->ehdr)) {
1218 ERROR_ELF("gelf_getehdr");
1219 goto err;
1220 }
1221
1222 if (read_sections(elf))
1223 goto err;
1224
1225 if (read_symbols(elf))
1226 goto err;
1227
1228 if (mark_group_syms(elf))
1229 goto err;
1230
1231 if (read_relocs(elf))
1232 goto err;
1233
1234 return elf;
1235
1236 err:
1237 elf_close(elf);
1238 return NULL;
1239 }
1240
elf_create_file(GElf_Ehdr * ehdr,const char * name)1241 struct elf *elf_create_file(GElf_Ehdr *ehdr, const char *name)
1242 {
1243 struct section *null, *symtab, *strtab, *shstrtab;
1244 char *tmp_name;
1245 struct symbol *sym;
1246 struct elf *elf;
1247
1248 elf_version(EV_CURRENT);
1249
1250 elf = calloc(1, sizeof(*elf));
1251 if (!elf) {
1252 ERROR_GLIBC("calloc");
1253 return NULL;
1254 }
1255
1256 INIT_LIST_HEAD(&elf->sections);
1257
1258 tmp_name = malloc(strlen(name) + 8);
1259 if (!tmp_name) {
1260 ERROR_GLIBC("malloc");
1261 return NULL;
1262 }
1263
1264 sprintf(tmp_name, "%s.XXXXXX", name);
1265
1266 elf->fd = mkstemp(tmp_name);
1267 if (elf->fd == -1) {
1268 ERROR_GLIBC("can't create tmp file");
1269 exit(1);
1270 }
1271
1272 elf->tmp_name = tmp_name;
1273
1274 elf->name = strdup(name);
1275 if (!elf->name) {
1276 ERROR_GLIBC("strdup");
1277 return NULL;
1278 }
1279
1280 elf->elf = elf_begin(elf->fd, ELF_C_WRITE, NULL);
1281 if (!elf->elf) {
1282 ERROR_ELF("elf_begin");
1283 return NULL;
1284 }
1285
1286 if (!gelf_newehdr(elf->elf, ELFCLASS64)) {
1287 ERROR_ELF("gelf_newehdr");
1288 return NULL;
1289 }
1290
1291 memcpy(&elf->ehdr, ehdr, sizeof(elf->ehdr));
1292
1293 if (!gelf_update_ehdr(elf->elf, &elf->ehdr)) {
1294 ERROR_ELF("gelf_update_ehdr");
1295 return NULL;
1296 }
1297
1298 INIT_LIST_HEAD(&elf->symbols);
1299
1300 if (!elf_alloc_hash(section, 1000) ||
1301 !elf_alloc_hash(section_name, 1000) ||
1302 !elf_alloc_hash(symbol, 10000) ||
1303 !elf_alloc_hash(symbol_name, 10000) ||
1304 !elf_alloc_hash(reloc, 100000))
1305 return NULL;
1306
1307 null = elf_create_section(elf, NULL, 0, 0, SHT_NULL, 0, 0);
1308 shstrtab = elf_create_section(elf, NULL, 0, 0, SHT_STRTAB, 1, 0);
1309 strtab = elf_create_section(elf, NULL, 0, 0, SHT_STRTAB, 1, 0);
1310
1311 if (!null || !shstrtab || !strtab)
1312 return NULL;
1313
1314 null->name = "";
1315 shstrtab->name = ".shstrtab";
1316 strtab->name = ".strtab";
1317
1318 null->sh.sh_name = elf_add_string(elf, shstrtab, null->name);
1319 shstrtab->sh.sh_name = elf_add_string(elf, shstrtab, shstrtab->name);
1320 strtab->sh.sh_name = elf_add_string(elf, shstrtab, strtab->name);
1321
1322 if (null->sh.sh_name == -1 || shstrtab->sh.sh_name == -1 || strtab->sh.sh_name == -1)
1323 return NULL;
1324
1325 elf_hash_add(section_name, &null->name_hash, str_hash(null->name));
1326 elf_hash_add(section_name, &strtab->name_hash, str_hash(strtab->name));
1327 elf_hash_add(section_name, &shstrtab->name_hash, str_hash(shstrtab->name));
1328
1329 if (elf_add_string(elf, strtab, "") == -1)
1330 return NULL;
1331
1332 symtab = elf_create_section(elf, ".symtab", 0x18, 0x18, SHT_SYMTAB, 0x8, 0);
1333 if (!symtab)
1334 return NULL;
1335
1336 symtab->sh.sh_link = strtab->idx;
1337 symtab->sh.sh_info = 1;
1338
1339 elf->ehdr.e_shstrndx = shstrtab->idx;
1340 if (!gelf_update_ehdr(elf->elf, &elf->ehdr)) {
1341 ERROR_ELF("gelf_update_ehdr");
1342 return NULL;
1343 }
1344
1345 sym = calloc(1, sizeof(*sym));
1346 if (!sym) {
1347 ERROR_GLIBC("calloc");
1348 return NULL;
1349 }
1350
1351 sym->name = "";
1352 sym->sec = null;
1353 elf_add_symbol(elf, sym);
1354
1355 return elf;
1356 }
1357
elf_add_string(struct elf * elf,struct section * strtab,const char * str)1358 unsigned int elf_add_string(struct elf *elf, struct section *strtab, const char *str)
1359 {
1360 unsigned int offset;
1361
1362 if (!strtab)
1363 strtab = find_section_by_name(elf, ".strtab");
1364 if (!strtab) {
1365 ERROR("can't find .strtab section");
1366 return -1;
1367 }
1368
1369 if (!strtab->sh.sh_addralign) {
1370 ERROR("'%s': invalid sh_addralign", strtab->name);
1371 return -1;
1372 }
1373
1374 offset = ALIGN(sec_size(strtab), strtab->sh.sh_addralign);
1375
1376 if (!elf_add_data(elf, strtab, str, strlen(str) + 1))
1377 return -1;
1378
1379 return offset;
1380 }
1381
elf_add_data(struct elf * elf,struct section * sec,const void * data,size_t size)1382 void *elf_add_data(struct elf *elf, struct section *sec, const void *data, size_t size)
1383 {
1384 unsigned long offset;
1385 Elf_Scn *s;
1386
1387 if (!sec->sh.sh_addralign) {
1388 ERROR("'%s': invalid sh_addralign", sec->name);
1389 return NULL;
1390 }
1391
1392 s = elf_getscn(elf->elf, sec->idx);
1393 if (!s) {
1394 ERROR_ELF("elf_getscn");
1395 return NULL;
1396 }
1397
1398 sec->data = elf_newdata(s);
1399 if (!sec->data) {
1400 ERROR_ELF("elf_newdata");
1401 return NULL;
1402 }
1403
1404 sec->data->d_buf = calloc(1, size);
1405 if (!sec->data->d_buf) {
1406 ERROR_GLIBC("calloc");
1407 return NULL;
1408 }
1409
1410 if (data)
1411 memcpy(sec->data->d_buf, data, size);
1412
1413 sec->data->d_size = size;
1414 sec->data->d_align = sec->sh.sh_addralign;
1415
1416 offset = ALIGN(sec_size(sec), sec->sh.sh_addralign);
1417 sec->sh.sh_size = offset + size;
1418
1419 mark_sec_changed(elf, sec, true);
1420
1421 return sec->data->d_buf;
1422 }
1423
elf_create_section(struct elf * elf,const char * name,size_t size,size_t entsize,unsigned int type,unsigned int align,unsigned int flags)1424 struct section *elf_create_section(struct elf *elf, const char *name,
1425 size_t size, size_t entsize,
1426 unsigned int type, unsigned int align,
1427 unsigned int flags)
1428 {
1429 struct section *sec, *shstrtab;
1430 Elf_Scn *s;
1431
1432 if (name && find_section_by_name(elf, name)) {
1433 ERROR("section '%s' already exists", name);
1434 return NULL;
1435 }
1436
1437 sec = calloc(1, sizeof(*sec));
1438 if (!sec) {
1439 ERROR_GLIBC("calloc");
1440 return NULL;
1441 }
1442
1443 INIT_LIST_HEAD(&sec->symbol_list);
1444
1445 /* don't actually create the section, just the data structures */
1446 if (type == SHT_NULL)
1447 goto add;
1448
1449 s = elf_newscn(elf->elf);
1450 if (!s) {
1451 ERROR_ELF("elf_newscn");
1452 return NULL;
1453 }
1454
1455 sec->idx = elf_ndxscn(s);
1456
1457 if (size) {
1458 sec->data = elf_newdata(s);
1459 if (!sec->data) {
1460 ERROR_ELF("elf_newdata");
1461 return NULL;
1462 }
1463
1464 sec->data->d_size = size;
1465 sec->data->d_align = 1;
1466
1467 sec->data->d_buf = calloc(1, size);
1468 if (!sec->data->d_buf) {
1469 ERROR_GLIBC("calloc");
1470 return NULL;
1471 }
1472 }
1473
1474 if (!gelf_getshdr(s, &sec->sh)) {
1475 ERROR_ELF("gelf_getshdr");
1476 return NULL;
1477 }
1478
1479 sec->sh.sh_size = size;
1480 sec->sh.sh_entsize = entsize;
1481 sec->sh.sh_type = type;
1482 sec->sh.sh_addralign = align;
1483 sec->sh.sh_flags = flags;
1484
1485 if (name) {
1486 sec->name = strdup(name);
1487 if (!sec->name) {
1488 ERROR("strdup");
1489 return NULL;
1490 }
1491
1492 /* Add section name to .shstrtab (or .strtab for Clang) */
1493 shstrtab = find_section_by_name(elf, ".shstrtab");
1494 if (!shstrtab) {
1495 shstrtab = find_section_by_name(elf, ".strtab");
1496 if (!shstrtab) {
1497 ERROR("can't find .shstrtab or .strtab");
1498 return NULL;
1499 }
1500 }
1501 sec->sh.sh_name = elf_add_string(elf, shstrtab, sec->name);
1502 if (sec->sh.sh_name == -1)
1503 return NULL;
1504
1505 elf_hash_add(section_name, &sec->name_hash, str_hash(sec->name));
1506 }
1507
1508 add:
1509 list_add_tail(&sec->list, &elf->sections);
1510 elf_hash_add(section, &sec->hash, sec->idx);
1511
1512 mark_sec_changed(elf, sec, true);
1513
1514 return sec;
1515 }
1516
elf_alloc_reloc(struct elf * elf,struct section * rsec)1517 static int elf_alloc_reloc(struct elf *elf, struct section *rsec)
1518 {
1519 struct reloc *old_relocs, *old_relocs_end, *new_relocs;
1520 unsigned int nr_relocs_old = sec_num_entries(rsec);
1521 unsigned int nr_relocs_new = nr_relocs_old + 1;
1522 unsigned long nr_alloc;
1523 struct symbol *sym;
1524
1525 if (!rsec->data) {
1526 rsec->data = elf_newdata(elf_getscn(elf->elf, rsec->idx));
1527 if (!rsec->data) {
1528 ERROR_ELF("elf_newdata");
1529 return -1;
1530 }
1531
1532 rsec->data->d_align = 1;
1533 rsec->data->d_type = ELF_T_RELA;
1534 rsec->data->d_buf = NULL;
1535 }
1536
1537 rsec->data->d_size = nr_relocs_new * elf_rela_size(elf);
1538 rsec->sh.sh_size = rsec->data->d_size;
1539
1540 nr_alloc = max(64UL, roundup_pow_of_two(nr_relocs_new));
1541 if (nr_alloc <= rsec->nr_alloc_relocs)
1542 return 0;
1543
1544 if (rsec->data->d_buf && !rsec->nr_alloc_relocs) {
1545 void *orig_buf = rsec->data->d_buf;
1546
1547 /*
1548 * The original d_buf is owned by libelf so it can't be
1549 * realloced.
1550 */
1551 rsec->data->d_buf = malloc(nr_alloc * elf_rela_size(elf));
1552 if (!rsec->data->d_buf) {
1553 ERROR_GLIBC("malloc");
1554 return -1;
1555 }
1556 memcpy(rsec->data->d_buf, orig_buf,
1557 nr_relocs_old * elf_rela_size(elf));
1558 } else {
1559 rsec->data->d_buf = realloc(rsec->data->d_buf,
1560 nr_alloc * elf_rela_size(elf));
1561 if (!rsec->data->d_buf) {
1562 ERROR_GLIBC("realloc");
1563 return -1;
1564 }
1565 }
1566
1567 rsec->nr_alloc_relocs = nr_alloc;
1568
1569 old_relocs = rsec->relocs;
1570 new_relocs = calloc(nr_alloc, sizeof(struct reloc));
1571 if (!new_relocs) {
1572 ERROR_GLIBC("calloc");
1573 return -1;
1574 }
1575
1576 if (!old_relocs)
1577 goto done;
1578
1579 /*
1580 * The struct reloc's address has changed. Update all the symbols and
1581 * relocs which reference it.
1582 */
1583
1584 old_relocs_end = &old_relocs[nr_relocs_old];
1585 for_each_sym(elf, sym) {
1586 struct reloc *reloc;
1587
1588 reloc = sym->relocs;
1589 if (!reloc)
1590 continue;
1591
1592 if (reloc >= old_relocs && reloc < old_relocs_end)
1593 sym->relocs = &new_relocs[reloc - old_relocs];
1594
1595 while (1) {
1596 struct reloc *next_reloc = sym_next_reloc(reloc);
1597
1598 if (!next_reloc)
1599 break;
1600
1601 if (next_reloc >= old_relocs && next_reloc < old_relocs_end)
1602 set_sym_next_reloc(reloc, &new_relocs[next_reloc - old_relocs]);
1603
1604 reloc = next_reloc;
1605 }
1606 }
1607
1608 memcpy(new_relocs, old_relocs, nr_relocs_old * sizeof(struct reloc));
1609
1610 for (int i = 0; i < nr_relocs_old; i++) {
1611 struct reloc *old = &old_relocs[i];
1612 struct reloc *new = &new_relocs[i];
1613 u32 key = reloc_hash(old);
1614
1615 elf_hash_del(reloc, &old->hash, key);
1616 elf_hash_add(reloc, &new->hash, key);
1617 }
1618
1619 free(old_relocs);
1620 done:
1621 rsec->relocs = new_relocs;
1622 return 0;
1623 }
1624
elf_create_rela_section(struct elf * elf,struct section * sec,unsigned int nr_relocs)1625 struct section *elf_create_rela_section(struct elf *elf, struct section *sec,
1626 unsigned int nr_relocs)
1627 {
1628 struct section *rsec;
1629 char *rsec_name;
1630
1631 rsec_name = malloc(strlen(sec->name) + strlen(".rela") + 1);
1632 if (!rsec_name) {
1633 ERROR_GLIBC("malloc");
1634 return NULL;
1635 }
1636 strcpy(rsec_name, ".rela");
1637 strcat(rsec_name, sec->name);
1638
1639 rsec = elf_create_section(elf, rsec_name, nr_relocs * elf_rela_size(elf),
1640 elf_rela_size(elf), SHT_RELA, elf_addr_size(elf),
1641 SHF_INFO_LINK);
1642 free(rsec_name);
1643 if (!rsec)
1644 return NULL;
1645
1646 if (nr_relocs) {
1647 rsec->data->d_type = ELF_T_RELA;
1648
1649 rsec->nr_alloc_relocs = nr_relocs;
1650 rsec->relocs = calloc(nr_relocs, sizeof(struct reloc));
1651 if (!rsec->relocs) {
1652 ERROR_GLIBC("calloc");
1653 return NULL;
1654 }
1655 }
1656
1657 rsec->sh.sh_link = find_section_by_name(elf, ".symtab")->idx;
1658 rsec->sh.sh_info = sec->idx;
1659
1660 sec->rsec = rsec;
1661 rsec->base = sec;
1662
1663 return rsec;
1664 }
1665
elf_create_reloc(struct elf * elf,struct section * sec,unsigned long offset,struct symbol * sym,s64 addend,unsigned int type)1666 struct reloc *elf_create_reloc(struct elf *elf, struct section *sec,
1667 unsigned long offset,
1668 struct symbol *sym, s64 addend,
1669 unsigned int type)
1670 {
1671 struct section *rsec = sec->rsec;
1672
1673 if (!rsec) {
1674 rsec = elf_create_rela_section(elf, sec, 0);
1675 if (!rsec)
1676 return NULL;
1677 }
1678
1679 if (find_reloc_by_dest(elf, sec, offset)) {
1680 ERROR_FUNC(sec, offset, "duplicate reloc");
1681 return NULL;
1682 }
1683
1684 if (elf_alloc_reloc(elf, rsec))
1685 return NULL;
1686
1687 mark_sec_changed(elf, rsec, true);
1688
1689 return elf_init_reloc(elf, rsec, sec_num_entries(rsec) - 1, offset, sym,
1690 addend, type);
1691 }
1692
elf_create_section_pair(struct elf * elf,const char * name,size_t entsize,unsigned int nr,unsigned int nr_relocs)1693 struct section *elf_create_section_pair(struct elf *elf, const char *name,
1694 size_t entsize, unsigned int nr,
1695 unsigned int nr_relocs)
1696 {
1697 struct section *sec;
1698
1699 sec = elf_create_section(elf, name, nr * entsize, entsize,
1700 SHT_PROGBITS, 1, SHF_ALLOC);
1701 if (!sec)
1702 return NULL;
1703
1704 if (!elf_create_rela_section(elf, sec, nr_relocs))
1705 return NULL;
1706
1707 return sec;
1708 }
1709
elf_write_insn(struct elf * elf,struct section * sec,unsigned long offset,unsigned int len,const char * insn)1710 int elf_write_insn(struct elf *elf, struct section *sec,
1711 unsigned long offset, unsigned int len,
1712 const char *insn)
1713 {
1714 Elf_Data *data = sec->data;
1715
1716 if (data->d_type != ELF_T_BYTE || data->d_off) {
1717 ERROR("write to unexpected data for section: %s", sec->name);
1718 return -1;
1719 }
1720
1721 memcpy(data->d_buf + offset, insn, len);
1722
1723 mark_sec_changed(elf, sec, true);
1724
1725 return 0;
1726 }
1727
1728 /*
1729 * When Elf_Scn::sh_size is smaller than the combined Elf_Data::d_size
1730 * do you:
1731 *
1732 * A) adhere to the section header and truncate the data, or
1733 * B) ignore the section header and write out all the data you've got?
1734 *
1735 * Yes, libelf sucks and we need to manually truncate if we over-allocate data.
1736 */
elf_truncate_section(struct elf * elf,struct section * sec)1737 static int elf_truncate_section(struct elf *elf, struct section *sec)
1738 {
1739 u64 size = sec_size(sec);
1740 bool truncated = false;
1741 Elf_Data *data = NULL;
1742 Elf_Scn *s;
1743
1744 s = elf_getscn(elf->elf, sec->idx);
1745 if (!s) {
1746 ERROR_ELF("elf_getscn");
1747 return -1;
1748 }
1749
1750 for (;;) {
1751 /* get next data descriptor for the relevant section */
1752 data = elf_getdata(s, data);
1753 if (!data) {
1754 if (size) {
1755 ERROR("end of section data but non-zero size left\n");
1756 return -1;
1757 }
1758 return 0;
1759 }
1760
1761 if (truncated) {
1762 /* when we remove symbols */
1763 ERROR("truncated; but more data\n");
1764 return -1;
1765 }
1766
1767 if (!data->d_size) {
1768 ERROR("zero size data");
1769 return -1;
1770 }
1771
1772 if (data->d_size > size) {
1773 truncated = true;
1774 data->d_size = size;
1775 }
1776
1777 size -= data->d_size;
1778 }
1779 }
1780
elf_write(struct elf * elf)1781 int elf_write(struct elf *elf)
1782 {
1783 struct section *sec;
1784 Elf_Scn *s;
1785
1786 /* Update changed relocation sections and section headers: */
1787 list_for_each_entry(sec, &elf->sections, list) {
1788 if (sec->truncate && elf_truncate_section(elf, sec))
1789 return -1;
1790
1791 if (sec_changed(sec)) {
1792 s = elf_getscn(elf->elf, sec->idx);
1793 if (!s) {
1794 ERROR_ELF("elf_getscn");
1795 return -1;
1796 }
1797
1798 /* Note this also flags the section dirty */
1799 if (!gelf_update_shdr(s, &sec->sh)) {
1800 ERROR_ELF("gelf_update_shdr");
1801 return -1;
1802 }
1803
1804 mark_sec_changed(elf, sec, false);
1805 }
1806 }
1807
1808 /* Make sure the new section header entries get updated properly. */
1809 elf_flagelf(elf->elf, ELF_C_SET, ELF_F_DIRTY);
1810
1811 /* Write all changes to the file. */
1812 if (elf_update(elf->elf, ELF_C_WRITE) < 0) {
1813 ERROR_ELF("elf_update");
1814 return -1;
1815 }
1816
1817 elf->changed = false;
1818
1819 return 0;
1820 }
1821
elf_close(struct elf * elf)1822 int elf_close(struct elf *elf)
1823 {
1824 if (elf->elf)
1825 elf_end(elf->elf);
1826
1827 if (elf->fd > 0)
1828 close(elf->fd);
1829
1830 if (elf->tmp_name && rename(elf->tmp_name, elf->name))
1831 return -1;
1832
1833 /*
1834 * NOTE: All remaining allocations are leaked on purpose. Objtool is
1835 * about to exit anyway.
1836 */
1837 return 0;
1838 }
1839