xref: /linux/tools/objtool/elf.c (revision 3e01ab44af207a2d8c197653e7278083b039a274)
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 
30 u32 str_hash_demangled(const char *str)
31 {
32 	return jhash(str, demangled_name_len(str), 0);
33 }
34 
35 #define elf_hash_add(name, node, key)					\
36 ({									\
37 	struct elf_hash_node *__node = node;				\
38 	__node->next = __elf_table_entry(elf, name, key);		\
39 	__elf_table_entry(elf, name, key) = __node;			\
40 })
41 
42 static inline void __elf_hash_del(struct elf_hash_node *node,
43 				  struct elf_hash_node **head)
44 {
45 	struct elf_hash_node *cur, *prev;
46 
47 	if (node == *head) {
48 		*head = node->next;
49 		return;
50 	}
51 
52 	for (prev = NULL, cur = *head; cur; prev = cur, cur = cur->next) {
53 		if (cur == node) {
54 			prev->next = cur->next;
55 			break;
56 		}
57 	}
58 }
59 
60 #define elf_hash_del(name, node, key) \
61 	__elf_hash_del(node, &__elf_table_entry(elf, name, key))
62 
63 #define elf_alloc_hash(name, size)					\
64 ({									\
65 	__elf_bits(elf, name) = max(10, ilog2(size));			\
66 	__elf_table(elf, name) = mmap(NULL,				\
67 				 sizeof(struct elf_hash_node *) << __elf_bits(elf, name), \
68 				 PROT_READ|PROT_WRITE,			\
69 				 MAP_PRIVATE|MAP_ANON, -1, 0);		\
70 	if (__elf_table(elf, name) == (void *)-1L) {			\
71 		ERROR_GLIBC("mmap fail " #name);			\
72 		__elf_table(elf, name) = NULL;				\
73 	}								\
74 	__elf_table(elf, name);						\
75 })
76 
77 static inline unsigned long __sym_start(struct symbol *s)
78 {
79 	return s->offset;
80 }
81 
82 static inline unsigned long __sym_last(struct symbol *s)
83 {
84 	return s->offset + (s->len ? s->len - 1 : 0);
85 }
86 
87 INTERVAL_TREE_DEFINE(struct symbol, node, unsigned long, __subtree_last,
88 		     __sym_start, __sym_last, static inline __maybe_unused,
89 		     __sym)
90 
91 #define __sym_for_each(_iter, _tree, _start, _end)			\
92 	for (_iter = __sym_iter_first((_tree), (_start), (_end));	\
93 	     _iter; _iter = __sym_iter_next(_iter, (_start), (_end)))
94 
95 struct symbol_hole {
96 	unsigned long key;
97 	const struct symbol *sym;
98 };
99 
100 /*
101  * Find the last symbol before @offset.
102  */
103 static int symbol_hole_by_offset(const void *key, const struct rb_node *node)
104 {
105 	const struct symbol *s = rb_entry(node, struct symbol, node);
106 	struct symbol_hole *sh = (void *)key;
107 
108 	if (sh->key < s->offset)
109 		return -1;
110 
111 	if (sh->key >= s->offset + s->len) {
112 		sh->sym = s;
113 		return 1;
114 	}
115 
116 	return 0;
117 }
118 
119 struct section *find_section_by_name(const struct elf *elf, const char *name)
120 {
121 	struct section *sec;
122 
123 	elf_hash_for_each_possible(elf, section_name, sec, name_hash, str_hash(name)) {
124 		if (!strcmp(sec->name, name))
125 			return sec;
126 	}
127 
128 	return NULL;
129 }
130 
131 static struct section *find_section_by_index(struct elf *elf,
132 					     unsigned int idx)
133 {
134 	struct section *sec;
135 
136 	elf_hash_for_each_possible(elf, section, sec, hash, idx) {
137 		if (sec->idx == idx)
138 			return sec;
139 	}
140 
141 	return NULL;
142 }
143 
144 static struct symbol *find_symbol_by_index(struct elf *elf, unsigned int idx)
145 {
146 	struct symbol *sym;
147 
148 	elf_hash_for_each_possible(elf, symbol, sym, hash, idx) {
149 		if (sym->idx == idx)
150 			return sym;
151 	}
152 
153 	return NULL;
154 }
155 
156 struct symbol *find_symbol_by_offset(struct section *sec, unsigned long offset)
157 {
158 	struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
159 	struct symbol *sym;
160 
161 	__sym_for_each(sym, tree, offset, offset) {
162 		if (sym->offset == offset && !is_sec_sym(sym))
163 			return sym->alias;
164 	}
165 
166 	return NULL;
167 }
168 
169 struct symbol *find_func_by_offset(struct section *sec, unsigned long offset)
170 {
171 	struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
172 	struct symbol *func;
173 
174 	__sym_for_each(func, tree, offset, offset) {
175 		if (func->offset == offset && is_func_sym(func))
176 			return func->alias;
177 	}
178 
179 	return NULL;
180 }
181 
182 struct symbol *find_symbol_containing(const struct section *sec, unsigned long offset)
183 {
184 	struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
185 	struct symbol *sym = NULL, *tmp;
186 
187 	__sym_for_each(tmp, tree, offset, offset) {
188 		if (tmp->len) {
189 			if (!sym) {
190 				sym = tmp;
191 				continue;
192 			}
193 
194 			if (sym->offset != tmp->offset || sym->len != tmp->len) {
195 				/*
196 				 * In the rare case of overlapping symbols,
197 				 * pick the smaller one.
198 				 *
199 				 * TODO: outlaw overlapping symbols
200 				 */
201 				if (tmp->len < sym->len)
202 					sym = tmp;
203 			}
204 		}
205 	}
206 
207 	return sym ? sym->alias : NULL;
208 }
209 
210 /*
211  * Returns size of hole starting at @offset.
212  */
213 int find_symbol_hole_containing(const struct section *sec, unsigned long offset)
214 {
215 	struct symbol_hole hole = {
216 		.key = offset,
217 		.sym = NULL,
218 	};
219 	struct rb_node *n;
220 	struct symbol *s;
221 
222 	/*
223 	 * Find the rightmost symbol for which @offset is after it.
224 	 */
225 	n = rb_find(&hole, &sec->symbol_tree.rb_root, symbol_hole_by_offset);
226 
227 	/* found a symbol that contains @offset */
228 	if (n)
229 		return 0; /* not a hole */
230 
231 	/*
232 	 * @offset >= sym->offset + sym->len, find symbol after it.
233 	 * When hole.sym is empty, use the first node to compute the hole.
234 	 * If there is no symbol in the section, the first node will be NULL,
235 	 * in which case, -1 is returned to skip the whole section.
236 	 */
237 	if (hole.sym)
238 		n = rb_next(&hole.sym->node);
239 	else
240 		n = rb_first_cached(&sec->symbol_tree);
241 
242 	if (!n)
243 		return -1; /* until end of address space */
244 
245 	/* hole until start of next symbol */
246 	s = rb_entry(n, struct symbol, node);
247 	return s->offset - offset;
248 }
249 
250 struct symbol *find_func_containing(struct section *sec, unsigned long offset)
251 {
252 	struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
253 	struct symbol *func;
254 
255 	__sym_for_each(func, tree, offset, offset) {
256 		if (is_func_sym(func))
257 			return func->alias;
258 	}
259 
260 	return NULL;
261 }
262 
263 struct symbol *find_symbol_by_name(const struct elf *elf, const char *name)
264 {
265 	struct symbol *sym;
266 
267 	elf_hash_for_each_possible(elf, symbol_name, sym, name_hash, str_hash(name)) {
268 		if (!strcmp(sym->name, name))
269 			return sym;
270 	}
271 
272 	return NULL;
273 }
274 
275 /* Find local symbol with matching STT_FILE */
276 static struct symbol *find_local_symbol_by_file_and_name(const struct elf *elf,
277 							 struct symbol *file,
278 							 const char *name)
279 {
280 	struct symbol *sym;
281 
282 	elf_hash_for_each_possible(elf, symbol_name, sym, name_hash, str_hash_demangled(name)) {
283 		if (sym->bind == STB_LOCAL && sym->file == file &&
284 		    !strcmp(sym->name, name)) {
285 			return sym;
286 		}
287 	}
288 
289 	return NULL;
290 }
291 
292 struct symbol *find_global_symbol_by_name(const struct elf *elf, const char *name)
293 {
294 	struct symbol *sym;
295 
296 	elf_hash_for_each_possible(elf, symbol_name, sym, name_hash, str_hash_demangled(name)) {
297 		if (!strcmp(sym->name, name) && !is_local_sym(sym))
298 			return sym;
299 	}
300 
301 	return NULL;
302 }
303 
304 struct reloc *find_reloc_by_dest_range(const struct elf *elf, struct section *sec,
305 				     unsigned long offset, unsigned int len)
306 {
307 	struct reloc *reloc, *r = NULL;
308 	struct section *rsec;
309 	unsigned long o;
310 
311 	rsec = sec->rsec;
312 	if (!rsec)
313 		return NULL;
314 
315 	for_offset_range(o, offset, offset + len) {
316 		elf_hash_for_each_possible(elf, reloc, reloc, hash,
317 					   sec_offset_hash(rsec, o)) {
318 			if (reloc->sec != rsec)
319 				continue;
320 
321 			if (reloc_offset(reloc) >= offset &&
322 			    reloc_offset(reloc) < offset + len) {
323 				if (!r || reloc_offset(reloc) < reloc_offset(r))
324 					r = reloc;
325 			}
326 		}
327 		if (r)
328 			return r;
329 	}
330 
331 	return NULL;
332 }
333 
334 struct reloc *find_reloc_by_dest(const struct elf *elf, struct section *sec, unsigned long offset)
335 {
336 	return find_reloc_by_dest_range(elf, sec, offset, 1);
337 }
338 
339 static bool is_dwarf_section(struct section *sec)
340 {
341 	return !strncmp(sec->name, ".debug_", 7);
342 }
343 
344 static int read_sections(struct elf *elf)
345 {
346 	Elf_Scn *s = NULL;
347 	struct section *sec;
348 	size_t shstrndx, sections_nr;
349 	int i;
350 
351 	if (elf_getshdrnum(elf->elf, &sections_nr)) {
352 		ERROR_ELF("elf_getshdrnum");
353 		return -1;
354 	}
355 
356 	if (elf_getshdrstrndx(elf->elf, &shstrndx)) {
357 		ERROR_ELF("elf_getshdrstrndx");
358 		return -1;
359 	}
360 
361 	if (!elf_alloc_hash(section, sections_nr) ||
362 	    !elf_alloc_hash(section_name, sections_nr))
363 		return -1;
364 
365 	elf->section_data = calloc(sections_nr, sizeof(*sec));
366 	if (!elf->section_data) {
367 		ERROR_GLIBC("calloc");
368 		return -1;
369 	}
370 	for (i = 0; i < sections_nr; i++) {
371 		sec = &elf->section_data[i];
372 
373 		INIT_LIST_HEAD(&sec->symbol_list);
374 
375 		s = elf_getscn(elf->elf, i);
376 		if (!s) {
377 			ERROR_ELF("elf_getscn");
378 			return -1;
379 		}
380 
381 		sec->idx = elf_ndxscn(s);
382 
383 		if (!gelf_getshdr(s, &sec->sh)) {
384 			ERROR_ELF("gelf_getshdr");
385 			return -1;
386 		}
387 
388 		sec->name = elf_strptr(elf->elf, shstrndx, sec->sh.sh_name);
389 		if (!sec->name) {
390 			ERROR_ELF("elf_strptr");
391 			return -1;
392 		}
393 
394 		if (sec_size(sec) != 0 && !is_dwarf_section(sec)) {
395 			sec->data = elf_getdata(s, NULL);
396 			if (!sec->data) {
397 				ERROR_ELF("elf_getdata");
398 				return -1;
399 			}
400 			if (sec->data->d_off != 0 ||
401 			    sec->data->d_size != sec_size(sec)) {
402 				ERROR("unexpected data attributes for %s", sec->name);
403 				return -1;
404 			}
405 		}
406 
407 		list_add_tail(&sec->list, &elf->sections);
408 		elf_hash_add(section, &sec->hash, sec->idx);
409 		elf_hash_add(section_name, &sec->name_hash, str_hash(sec->name));
410 
411 		if (is_reloc_sec(sec))
412 			elf->num_relocs += sec_num_entries(sec);
413 	}
414 
415 	if (opts.stats) {
416 		printf("nr_sections: %lu\n", (unsigned long)sections_nr);
417 		printf("section_bits: %d\n", elf->section_bits);
418 	}
419 
420 	/* sanity check, one more call to elf_nextscn() should return NULL */
421 	if (elf_nextscn(elf->elf, s)) {
422 		ERROR("section entry mismatch");
423 		return -1;
424 	}
425 
426 	return 0;
427 }
428 
429 /*
430  * Returns desired length of the demangled name.
431  * If name doesn't need demangling, return strlen(name).
432  */
433 static ssize_t demangled_name_len(const char *name)
434 {
435 	ssize_t idx;
436 	const char *p;
437 
438 	p = strstr(name, ".llvm.");
439 	if (p)
440 		return p - name;
441 
442 	if (!strstarts(name, "__UNIQUE_ID_") && !strchr(name, '.'))
443 		return strlen(name);
444 
445 	for (idx = strlen(name) - 1; idx >= 0; idx--) {
446 		char c = name[idx];
447 
448 		if (!isdigit(c) && c != '.' && c != '_')
449 			break;
450 	}
451 	if (idx <= 0)
452 		return strlen(name);
453 	return idx + 1;
454 }
455 
456 /*
457  * Remove number suffix of a symbol.
458  *
459  * Specifically, remove trailing numbers for "__UNIQUE_ID_" symbols and
460  * symbols with '.'.
461  *
462  * With CONFIG_LTO_CLANG_THIN, it is possible to have nested __UNIQUE_ID_,
463  * such as
464  *
465  *   __UNIQUE_ID_addressable___UNIQUE_ID_pci_invalid_bar_694_695
466  *
467  * to remove both trailing numbers, also remove trailing '_'.
468  *
469  * For symbols with llvm suffix, i.e., foo.llvm.<hash>, remove the
470  * .llvm.<hash> part.
471  */
472 static const char *demangle_name(struct symbol *sym)
473 {
474 	char *str;
475 	ssize_t len;
476 
477 	if (!is_func_sym(sym) && !is_object_sym(sym))
478 		return sym->name;
479 
480 	len = demangled_name_len(sym->name);
481 	if (len == strlen(sym->name))
482 		return sym->name;
483 
484 	str = strndup(sym->name, len);
485 	if (!str) {
486 		ERROR_GLIBC("strdup");
487 		return NULL;
488 	}
489 
490 	return str;
491 }
492 
493 static int elf_add_symbol(struct elf *elf, struct symbol *sym)
494 {
495 	struct list_head *entry;
496 	struct rb_node *pnode;
497 	struct symbol *iter;
498 
499 	INIT_LIST_HEAD(&sym->pv_target);
500 	sym->alias = sym;
501 
502 	sym->type = GELF_ST_TYPE(sym->sym.st_info);
503 	sym->bind = GELF_ST_BIND(sym->sym.st_info);
504 
505 	if (is_file_sym(sym))
506 		elf->num_files++;
507 
508 	sym->offset = sym->sym.st_value;
509 	sym->len = sym->sym.st_size;
510 
511 	__sym_for_each(iter, &sym->sec->symbol_tree, sym->offset, sym->offset) {
512 		if (!is_undef_sym(iter) && iter->offset == sym->offset &&
513 		    iter->type == sym->type && iter->len == sym->len)
514 			iter->alias = sym;
515 	}
516 
517 	__sym_insert(sym, &sym->sec->symbol_tree);
518 	pnode = rb_prev(&sym->node);
519 	if (pnode)
520 		entry = &rb_entry(pnode, struct symbol, node)->list;
521 	else
522 		entry = &sym->sec->symbol_list;
523 	list_add(&sym->list, entry);
524 
525 	sym->demangled_name = demangle_name(sym);
526 	if (!sym->demangled_name)
527 		return -1;
528 
529 	list_add_tail(&sym->global_list, &elf->symbols);
530 	elf_hash_add(symbol, &sym->hash, sym->idx);
531 	elf_hash_add(symbol_name, &sym->name_hash, str_hash(sym->demangled_name));
532 
533 	if (is_func_sym(sym) &&
534 	    (strstarts(sym->name, "__pfx_") ||
535 	     strstarts(sym->name, "__cfi_") ||
536 	     strstarts(sym->name, "__pi___pfx_") ||
537 	     strstarts(sym->name, "__pi___cfi_")))
538 		sym->prefix = 1;
539 
540 	if (strstarts(sym->name, ".klp.sym"))
541 		sym->klp = 1;
542 
543 	if (!sym->klp && !is_sec_sym(sym) && strstr(sym->name, ".cold")) {
544 		sym->cold = 1;
545 
546 		/*
547 		 * Clang doesn't mark cold subfunctions as STT_FUNC, which
548 		 * breaks several objtool assumptions.  Fake it.
549 		 */
550 		sym->type = STT_FUNC;
551 	}
552 
553 	sym->pfunc = sym->cfunc = sym;
554 
555 	return 0;
556 }
557 
558 static int read_symbols(struct elf *elf)
559 {
560 	struct section *symtab, *symtab_shndx, *sec;
561 	struct symbol *sym, *pfunc, *file = NULL;
562 	int symbols_nr, i;
563 	char *coldstr;
564 	Elf_Data *shndx_data = NULL;
565 	Elf32_Word shndx;
566 
567 	symtab = find_section_by_name(elf, ".symtab");
568 	if (symtab) {
569 		symtab_shndx = find_section_by_name(elf, ".symtab_shndx");
570 		if (symtab_shndx)
571 			shndx_data = symtab_shndx->data;
572 
573 		symbols_nr = sec_num_entries(symtab);
574 	} else {
575 		/*
576 		 * A missing symbol table is actually possible if it's an empty
577 		 * .o file. This can happen for thunk_64.o. Make sure to at
578 		 * least allocate the symbol hash tables so we can do symbol
579 		 * lookups without crashing.
580 		 */
581 		symbols_nr = 0;
582 	}
583 
584 	if (!elf_alloc_hash(symbol, symbols_nr) ||
585 	    !elf_alloc_hash(symbol_name, symbols_nr))
586 		return -1;
587 
588 	elf->symbol_data = calloc(symbols_nr, sizeof(*sym));
589 	if (!elf->symbol_data) {
590 		ERROR_GLIBC("calloc");
591 		return -1;
592 	}
593 
594 	INIT_LIST_HEAD(&elf->symbols);
595 
596 	for (i = 0; i < symbols_nr; i++) {
597 		sym = &elf->symbol_data[i];
598 
599 		sym->idx = i;
600 
601 		if (!gelf_getsymshndx(symtab->data, shndx_data, i, &sym->sym,
602 				      &shndx)) {
603 			ERROR_ELF("gelf_getsymshndx");
604 			return -1;
605 		}
606 
607 		sym->name = elf_strptr(elf->elf, symtab->sh.sh_link,
608 				       sym->sym.st_name);
609 		if (!sym->name) {
610 			ERROR_ELF("elf_strptr");
611 			return -1;
612 		}
613 
614 		if ((sym->sym.st_shndx > SHN_UNDEF &&
615 		     sym->sym.st_shndx < SHN_LORESERVE) ||
616 		    (shndx_data && sym->sym.st_shndx == SHN_XINDEX)) {
617 			if (sym->sym.st_shndx != SHN_XINDEX)
618 				shndx = sym->sym.st_shndx;
619 
620 			sym->sec = find_section_by_index(elf, shndx);
621 			if (!sym->sec) {
622 				ERROR("couldn't find section for symbol %s", sym->name);
623 				return -1;
624 			}
625 			if (GELF_ST_TYPE(sym->sym.st_info) == STT_SECTION) {
626 				sym->name = sym->sec->name;
627 				sym->sec->sym = sym;
628 			}
629 		} else
630 			sym->sec = find_section_by_index(elf, 0);
631 
632 		if (elf_add_symbol(elf, sym))
633 			return -1;
634 
635 		if (is_file_sym(sym))
636 			file = sym;
637 		else if (sym->bind == STB_LOCAL)
638 			sym->file = file;
639 	}
640 
641 	if (opts.stats) {
642 		printf("nr_symbols: %lu\n", (unsigned long)symbols_nr);
643 		printf("symbol_bits: %d\n", elf->symbol_bits);
644 	}
645 
646 	/* Create parent/child links for any cold subfunctions */
647 	list_for_each_entry(sec, &elf->sections, list) {
648 		sec_for_each_sym(sec, sym) {
649 			char *pname;
650 			size_t pnamelen;
651 
652 			if (!sym->cold)
653 				continue;
654 
655 			coldstr = strstr(sym->name, ".cold");
656 			if (!coldstr) {
657 				ERROR("%s(): cold subfunction without \".cold\"?", sym->name);
658 				return -1;
659 			}
660 
661 			pnamelen = coldstr - sym->name;
662 			pname = strndup(sym->name, pnamelen);
663 			if (!pname) {
664 				ERROR("%s(): failed to allocate memory", sym->name);
665 				return -1;
666 			}
667 
668 			pfunc = find_local_symbol_by_file_and_name(elf, sym->file, pname);
669 			if (!pfunc)
670 				pfunc = find_global_symbol_by_name(elf, pname);
671 			free(pname);
672 
673 			if (!pfunc) {
674 				ERROR("%s(): can't find parent function", sym->name);
675 				return -1;
676 			}
677 
678 			sym->pfunc = pfunc->alias;
679 			pfunc->cfunc = sym;
680 			pfunc->alias->cfunc = sym;
681 
682 			/*
683 			 * Unfortunately, -fnoreorder-functions puts the child
684 			 * inside the parent.  Remove the overlap so we can
685 			 * have sane assumptions.
686 			 *
687 			 * Note that pfunc->len now no longer matches
688 			 * pfunc->sym.st_size.
689 			 */
690 			if (sym->sec == pfunc->sec &&
691 			    sym->offset >= pfunc->offset &&
692 			    sym->offset + sym->len == pfunc->offset + pfunc->len) {
693 				pfunc->len -= sym->len;
694 			}
695 		}
696 	}
697 
698 	return 0;
699 }
700 
701 static int mark_group_syms(struct elf *elf)
702 {
703 	struct section *symtab, *sec;
704 	struct symbol *sym;
705 
706 	symtab = find_section_by_name(elf, ".symtab");
707 	if (!symtab) {
708 		ERROR("no .symtab");
709 		return -1;
710 	}
711 
712 	for_each_sec(elf, sec) {
713 		if (sec->sh.sh_type == SHT_GROUP &&
714 		    sec->sh.sh_link == symtab->idx) {
715 			sym = find_symbol_by_index(elf, sec->sh.sh_info);
716 			if (!sym) {
717 				ERROR("%s: can't find SHT_GROUP signature symbol",
718 				      sec->name);
719 				return -1;
720 			}
721 
722 			sym->group_sec = sec;
723 		}
724 	}
725 
726 	return 0;
727 }
728 
729 /*
730  * @sym's idx has changed.  Update the relocs which reference it.
731  */
732 static int elf_update_sym_relocs(struct elf *elf, struct symbol *sym)
733 {
734 	struct reloc *reloc;
735 
736 	for (reloc = sym->relocs; reloc; reloc = sym_next_reloc(reloc))
737 		set_reloc_sym(elf, reloc, reloc->sym->idx);
738 
739 	return 0;
740 }
741 
742 /*
743  * The libelf API is terrible; gelf_update_sym*() takes a data block relative
744  * index value, *NOT* the symbol index. As such, iterate the data blocks and
745  * adjust index until it fits.
746  *
747  * If no data block is found, allow adding a new data block provided the index
748  * is only one past the end.
749  */
750 static int elf_update_symbol(struct elf *elf, struct section *symtab,
751 			     struct section *symtab_shndx, struct symbol *sym)
752 {
753 	Elf32_Word shndx;
754 	Elf_Data *symtab_data = NULL, *shndx_data = NULL;
755 	Elf64_Xword entsize = symtab->sh.sh_entsize;
756 	int max_idx, idx = sym->idx;
757 	Elf_Scn *s, *t = NULL;
758 	bool is_special_shndx = sym->sym.st_shndx >= SHN_LORESERVE &&
759 				sym->sym.st_shndx != SHN_XINDEX;
760 
761 	shndx = is_special_shndx ? sym->sym.st_shndx : sym->sec->idx;
762 
763 	s = elf_getscn(elf->elf, symtab->idx);
764 	if (!s) {
765 		ERROR_ELF("elf_getscn");
766 		return -1;
767 	}
768 
769 	if (symtab_shndx) {
770 		t = elf_getscn(elf->elf, symtab_shndx->idx);
771 		if (!t) {
772 			ERROR_ELF("elf_getscn");
773 			return -1;
774 		}
775 	}
776 
777 	for (;;) {
778 		/* get next data descriptor for the relevant sections */
779 		symtab_data = elf_getdata(s, symtab_data);
780 		if (t)
781 			shndx_data = elf_getdata(t, shndx_data);
782 
783 		/* end-of-list */
784 		if (!symtab_data) {
785 			/*
786 			 * Over-allocate to avoid O(n^2) symbol creation
787 			 * behaviour.  The down side is that libelf doesn't
788 			 * like this; see elf_truncate_section() for the fixup.
789 			 */
790 			int num = max(1U, sym->idx/3);
791 			void *buf;
792 
793 			if (idx) {
794 				/* we don't do holes in symbol tables */
795 				ERROR("index out of range");
796 				return -1;
797 			}
798 
799 			/* if @idx == 0, it's the next contiguous entry, create it */
800 			symtab_data = elf_newdata(s);
801 			if (t)
802 				shndx_data = elf_newdata(t);
803 
804 			buf = calloc(num, entsize);
805 			if (!buf) {
806 				ERROR_GLIBC("calloc");
807 				return -1;
808 			}
809 
810 			symtab_data->d_buf = buf;
811 			symtab_data->d_size = num * entsize;
812 			symtab_data->d_align = 1;
813 			symtab_data->d_type = ELF_T_SYM;
814 
815 			mark_sec_changed(elf, symtab, true);
816 			symtab->truncate = true;
817 
818 			if (t) {
819 				buf = calloc(num, sizeof(Elf32_Word));
820 				if (!buf) {
821 					ERROR_GLIBC("calloc");
822 					return -1;
823 				}
824 
825 				shndx_data->d_buf = buf;
826 				shndx_data->d_size = num * sizeof(Elf32_Word);
827 				shndx_data->d_align = sizeof(Elf32_Word);
828 				shndx_data->d_type = ELF_T_WORD;
829 
830 				mark_sec_changed(elf, symtab_shndx, true);
831 				symtab_shndx->truncate = true;
832 			}
833 
834 			break;
835 		}
836 
837 		/* empty blocks should not happen */
838 		if (!symtab_data->d_size) {
839 			ERROR("zero size data");
840 			return -1;
841 		}
842 
843 		/* is this the right block? */
844 		max_idx = symtab_data->d_size / entsize;
845 		if (idx < max_idx)
846 			break;
847 
848 		/* adjust index and try again */
849 		idx -= max_idx;
850 	}
851 
852 	/* something went side-ways */
853 	if (idx < 0) {
854 		ERROR("negative index");
855 		return -1;
856 	}
857 
858 	/* setup extended section index magic and write the symbol */
859 	if (shndx < SHN_LORESERVE || is_special_shndx) {
860 		sym->sym.st_shndx = shndx;
861 		if (!shndx_data)
862 			shndx = 0;
863 	} else {
864 		sym->sym.st_shndx = SHN_XINDEX;
865 		if (!shndx_data) {
866 			ERROR("no .symtab_shndx");
867 			return -1;
868 		}
869 	}
870 
871 	if (!gelf_update_symshndx(symtab_data, shndx_data, idx, &sym->sym, shndx)) {
872 		ERROR_ELF("gelf_update_symshndx");
873 		return -1;
874 	}
875 
876 	return 0;
877 }
878 
879 struct symbol *elf_create_symbol(struct elf *elf, const char *name,
880 				 struct section *sec, unsigned int bind,
881 				 unsigned int type, unsigned long offset,
882 				 size_t size)
883 {
884 	struct section *symtab, *symtab_shndx;
885 	Elf32_Word first_non_local, new_idx;
886 	struct symbol *old, *sym;
887 
888 	sym = calloc(1, sizeof(*sym));
889 	if (!sym) {
890 		ERROR_GLIBC("calloc");
891 		return NULL;
892 	}
893 
894 	sym->name = strdup(name);
895 	if (!sym->name) {
896 		ERROR_GLIBC("strdup");
897 		return NULL;
898 	}
899 
900 	if (type != STT_SECTION) {
901 		sym->sym.st_name = elf_add_string(elf, NULL, sym->name);
902 		if (sym->sym.st_name == -1)
903 			return NULL;
904 	}
905 
906 	if (sec) {
907 		sym->sec = sec;
908 	} else {
909 		sym->sec = find_section_by_index(elf, 0);
910 		if (!sym->sec) {
911 			ERROR("no NULL section");
912 			return NULL;
913 		}
914 	}
915 
916 	sym->sym.st_info  = GELF_ST_INFO(bind, type);
917 	sym->sym.st_value = offset;
918 	sym->sym.st_size  = size;
919 
920 	symtab = find_section_by_name(elf, ".symtab");
921 	if (!symtab) {
922 		ERROR("no .symtab");
923 		return NULL;
924 	}
925 
926 	symtab_shndx = find_section_by_name(elf, ".symtab_shndx");
927 
928 	new_idx = sec_num_entries(symtab);
929 
930 	if (bind != STB_LOCAL)
931 		goto non_local;
932 
933 	/*
934 	 * Move the first global symbol, as per sh_info, into a new, higher
935 	 * symbol index. This frees up a spot for a new local symbol.
936 	 */
937 	first_non_local = symtab->sh.sh_info;
938 	old = find_symbol_by_index(elf, first_non_local);
939 	if (old) {
940 
941 		elf_hash_del(symbol, &old->hash, old->idx);
942 		elf_hash_add(symbol, &old->hash, new_idx);
943 		old->idx = new_idx;
944 
945 		if (elf_update_symbol(elf, symtab, symtab_shndx, old)) {
946 			ERROR("elf_update_symbol move");
947 			return NULL;
948 		}
949 
950 		if (elf_update_sym_relocs(elf, old))
951 			return NULL;
952 
953 		if (old->group_sec) {
954 			old->group_sec->sh.sh_info = new_idx;
955 			mark_sec_changed(elf, old->group_sec, true);
956 		}
957 
958 		new_idx = first_non_local;
959 	}
960 
961 	/*
962 	 * Either way, we will add a LOCAL symbol.
963 	 */
964 	symtab->sh.sh_info += 1;
965 
966 non_local:
967 	sym->idx = new_idx;
968 	if (sym->idx && elf_update_symbol(elf, symtab, symtab_shndx, sym))
969 		return NULL;
970 
971 	symtab->sh.sh_size += symtab->sh.sh_entsize;
972 	mark_sec_changed(elf, symtab, true);
973 
974 	if (symtab_shndx) {
975 		symtab_shndx->sh.sh_size += sizeof(Elf32_Word);
976 		mark_sec_changed(elf, symtab_shndx, true);
977 	}
978 
979 	if (elf_add_symbol(elf, sym))
980 		return NULL;
981 
982 	return sym;
983 }
984 
985 struct symbol *elf_create_section_symbol(struct elf *elf, struct section *sec)
986 {
987 	struct symbol *sym = calloc(1, sizeof(*sym));
988 
989 	sym = elf_create_symbol(elf, sec->name, sec, STB_LOCAL, STT_SECTION, 0, 0);
990 	if (!sym)
991 		return NULL;
992 
993 	sec->sym = sym;
994 
995 	return sym;
996 }
997 
998 struct reloc *elf_init_reloc(struct elf *elf, struct section *rsec,
999 			     unsigned int reloc_idx, unsigned long offset,
1000 			     struct symbol *sym, s64 addend, unsigned int type)
1001 {
1002 	struct reloc *reloc, empty = { 0 };
1003 
1004 	if (reloc_idx >= sec_num_entries(rsec)) {
1005 		ERROR("%s: bad reloc_idx %u for %s with %d relocs",
1006 		      __func__, reloc_idx, rsec->name, sec_num_entries(rsec));
1007 		return NULL;
1008 	}
1009 
1010 	reloc = &rsec->relocs[reloc_idx];
1011 
1012 	if (memcmp(reloc, &empty, sizeof(empty))) {
1013 		ERROR("%s: %s: reloc %d already initialized!",
1014 		      __func__, rsec->name, reloc_idx);
1015 		return NULL;
1016 	}
1017 
1018 	reloc->sec = rsec;
1019 	reloc->sym = sym;
1020 
1021 	set_reloc_offset(elf, reloc, offset);
1022 	set_reloc_sym(elf, reloc, sym->idx);
1023 	set_reloc_type(elf, reloc, type);
1024 	set_reloc_addend(elf, reloc, addend);
1025 
1026 	elf_hash_add(reloc, &reloc->hash, reloc_hash(reloc));
1027 	set_sym_next_reloc(reloc, sym->relocs);
1028 	sym->relocs = reloc;
1029 
1030 	return reloc;
1031 }
1032 
1033 struct reloc *elf_init_reloc_text_sym(struct elf *elf, struct section *sec,
1034 				      unsigned long offset,
1035 				      unsigned int reloc_idx,
1036 				      struct section *insn_sec,
1037 				      unsigned long insn_off)
1038 {
1039 	struct symbol *sym = insn_sec->sym;
1040 	s64 addend = insn_off;
1041 
1042 	if (!is_text_sec(insn_sec)) {
1043 		ERROR("bad call to %s() for data symbol %s", __func__, sym->name);
1044 		return NULL;
1045 	}
1046 
1047 	if (!sym) {
1048 		/*
1049 		 * Due to how weak functions work, we must use section based
1050 		 * relocations. Symbol based relocations would result in the
1051 		 * weak and non-weak function annotations being overlaid on the
1052 		 * non-weak function after linking.
1053 		 */
1054 		sym = elf_create_section_symbol(elf, insn_sec);
1055 		if (!sym)
1056 			return NULL;
1057 	}
1058 
1059 	return elf_init_reloc(elf, sec->rsec, reloc_idx, offset, sym, addend,
1060 			      elf_text_rela_type(elf));
1061 }
1062 
1063 struct reloc *elf_init_reloc_data_sym(struct elf *elf, struct section *sec,
1064 				      unsigned long offset,
1065 				      unsigned int reloc_idx,
1066 				      struct symbol *sym,
1067 				      s64 addend)
1068 {
1069 	if (is_text_sec(sec)) {
1070 		ERROR("bad call to %s() for text symbol %s", __func__, sym->name);
1071 		return NULL;
1072 	}
1073 
1074 	return elf_init_reloc(elf, sec->rsec, reloc_idx, offset, sym, addend,
1075 			      elf_data_rela_type(elf));
1076 }
1077 
1078 static int read_relocs(struct elf *elf)
1079 {
1080 	unsigned long nr_reloc, max_reloc = 0;
1081 	struct section *rsec;
1082 	struct reloc *reloc;
1083 	unsigned int symndx;
1084 	struct symbol *sym;
1085 	int i;
1086 
1087 	if (!elf_alloc_hash(reloc, elf->num_relocs))
1088 		return -1;
1089 
1090 	list_for_each_entry(rsec, &elf->sections, list) {
1091 		if (!is_reloc_sec(rsec))
1092 			continue;
1093 
1094 		rsec->base = find_section_by_index(elf, rsec->sh.sh_info);
1095 		if (!rsec->base) {
1096 			ERROR("can't find base section for reloc section %s", rsec->name);
1097 			return -1;
1098 		}
1099 
1100 		rsec->base->rsec = rsec;
1101 
1102 		/* nr_alloc_relocs=0: libelf owns d_buf */
1103 		rsec->nr_alloc_relocs = 0;
1104 
1105 		rsec->relocs = calloc(sec_num_entries(rsec), sizeof(*reloc));
1106 		if (!rsec->relocs) {
1107 			ERROR_GLIBC("calloc");
1108 			return -1;
1109 		}
1110 
1111 		nr_reloc = 0;
1112 		for (i = 0; i < sec_num_entries(rsec); i++) {
1113 			reloc = &rsec->relocs[i];
1114 
1115 			reloc->sec = rsec;
1116 			symndx = reloc_sym(reloc);
1117 			reloc->sym = sym = find_symbol_by_index(elf, symndx);
1118 			if (!reloc->sym) {
1119 				ERROR("can't find reloc entry symbol %d for %s", symndx, rsec->name);
1120 				return -1;
1121 			}
1122 
1123 			elf_hash_add(reloc, &reloc->hash, reloc_hash(reloc));
1124 			set_sym_next_reloc(reloc, sym->relocs);
1125 			sym->relocs = reloc;
1126 
1127 			nr_reloc++;
1128 		}
1129 		max_reloc = max(max_reloc, nr_reloc);
1130 	}
1131 
1132 	if (opts.stats) {
1133 		printf("max_reloc: %lu\n", max_reloc);
1134 		printf("num_relocs: %lu\n", elf->num_relocs);
1135 		printf("reloc_bits: %d\n", elf->reloc_bits);
1136 	}
1137 
1138 	return 0;
1139 }
1140 
1141 static void mark_rodata(struct elf *elf)
1142 {
1143 	struct section *sec;
1144 
1145 	for_each_sec(elf, sec) {
1146 		if ((strstarts(sec->name, ".rodata") && !strstr(sec->name, ".str1.")) ||
1147 		    strstarts(sec->name, ".data.rel.ro"))
1148 			sec->rodata = true;
1149 	}
1150 }
1151 
1152 struct elf *elf_open_read(const char *name, int flags)
1153 {
1154 	struct elf *elf;
1155 	Elf_Cmd cmd;
1156 
1157 	elf_version(EV_CURRENT);
1158 
1159 	elf = malloc(sizeof(*elf));
1160 	if (!elf) {
1161 		ERROR_GLIBC("malloc");
1162 		return NULL;
1163 	}
1164 	memset(elf, 0, sizeof(*elf));
1165 
1166 	INIT_LIST_HEAD(&elf->sections);
1167 
1168 	elf->fd = open(name, flags);
1169 	if (elf->fd == -1) {
1170 		fprintf(stderr, "objtool: Can't open '%s': %s\n",
1171 			name, strerror(errno));
1172 		goto err;
1173 	}
1174 
1175 	elf->name = strdup(name);
1176 	if (!elf->name) {
1177 		ERROR_GLIBC("strdup");
1178 		return NULL;
1179 	}
1180 
1181 	if ((flags & O_ACCMODE) == O_RDONLY)
1182 		cmd = ELF_C_READ_MMAP;
1183 	else if ((flags & O_ACCMODE) == O_RDWR)
1184 		cmd = ELF_C_RDWR;
1185 	else /* O_WRONLY */
1186 		cmd = ELF_C_WRITE;
1187 
1188 	elf->elf = elf_begin(elf->fd, cmd, NULL);
1189 	if (!elf->elf) {
1190 		ERROR_ELF("elf_begin");
1191 		goto err;
1192 	}
1193 
1194 	if (!gelf_getehdr(elf->elf, &elf->ehdr)) {
1195 		ERROR_ELF("gelf_getehdr");
1196 		goto err;
1197 	}
1198 
1199 	if (read_sections(elf))
1200 		goto err;
1201 
1202 	mark_rodata(elf);
1203 
1204 	if (read_symbols(elf))
1205 		goto err;
1206 
1207 	if (mark_group_syms(elf))
1208 		goto err;
1209 
1210 	if (read_relocs(elf))
1211 		goto err;
1212 
1213 	return elf;
1214 
1215 err:
1216 	elf_close(elf);
1217 	return NULL;
1218 }
1219 
1220 struct elf *elf_create_file(GElf_Ehdr *ehdr, const char *name)
1221 {
1222 	struct section *null, *symtab, *strtab, *shstrtab;
1223 	char *tmp_name;
1224 	struct symbol *sym;
1225 	struct elf *elf;
1226 
1227 	elf_version(EV_CURRENT);
1228 
1229 	elf = calloc(1, sizeof(*elf));
1230 	if (!elf) {
1231 		ERROR_GLIBC("calloc");
1232 		return NULL;
1233 	}
1234 
1235 	INIT_LIST_HEAD(&elf->sections);
1236 
1237 	tmp_name = malloc(strlen(name) + 8);
1238 	if (!tmp_name) {
1239 		ERROR_GLIBC("malloc");
1240 		return NULL;
1241 	}
1242 
1243 	sprintf(tmp_name, "%s.XXXXXX", name);
1244 
1245 	elf->fd = mkstemp(tmp_name);
1246 	if (elf->fd == -1) {
1247 		ERROR_GLIBC("can't create tmp file");
1248 		exit(1);
1249 	}
1250 
1251 	elf->tmp_name = tmp_name;
1252 
1253 	elf->name = strdup(name);
1254 	if (!elf->name) {
1255 		ERROR_GLIBC("strdup");
1256 		return NULL;
1257 	}
1258 
1259 	elf->elf = elf_begin(elf->fd, ELF_C_WRITE, NULL);
1260 	if (!elf->elf) {
1261 		ERROR_ELF("elf_begin");
1262 		return NULL;
1263 	}
1264 
1265 	if (!gelf_newehdr(elf->elf, ELFCLASS64)) {
1266 		ERROR_ELF("gelf_newehdr");
1267 		return NULL;
1268 	}
1269 
1270 	memcpy(&elf->ehdr, ehdr, sizeof(elf->ehdr));
1271 
1272 	if (!gelf_update_ehdr(elf->elf, &elf->ehdr)) {
1273 		ERROR_ELF("gelf_update_ehdr");
1274 		return NULL;
1275 	}
1276 
1277 	INIT_LIST_HEAD(&elf->symbols);
1278 
1279 	if (!elf_alloc_hash(section,		1000) ||
1280 	    !elf_alloc_hash(section_name,	1000) ||
1281 	    !elf_alloc_hash(symbol,		10000) ||
1282 	    !elf_alloc_hash(symbol_name,	10000) ||
1283 	    !elf_alloc_hash(reloc,		100000))
1284 		return NULL;
1285 
1286 	null		= elf_create_section(elf, NULL, 0, 0, SHT_NULL, 0, 0);
1287 	shstrtab	= elf_create_section(elf, NULL, 0, 0, SHT_STRTAB, 1, 0);
1288 	strtab		= elf_create_section(elf, NULL, 0, 0, SHT_STRTAB, 1, 0);
1289 
1290 	if (!null || !shstrtab || !strtab)
1291 		return NULL;
1292 
1293 	null->name	= "";
1294 	shstrtab->name	= ".shstrtab";
1295 	strtab->name	= ".strtab";
1296 
1297 	null->sh.sh_name	= elf_add_string(elf, shstrtab, null->name);
1298 	shstrtab->sh.sh_name	= elf_add_string(elf, shstrtab, shstrtab->name);
1299 	strtab->sh.sh_name	= elf_add_string(elf, shstrtab, strtab->name);
1300 
1301 	if (null->sh.sh_name == -1 || shstrtab->sh.sh_name == -1 || strtab->sh.sh_name == -1)
1302 		return NULL;
1303 
1304 	elf_hash_add(section_name, &null->name_hash,		str_hash(null->name));
1305 	elf_hash_add(section_name, &strtab->name_hash,		str_hash(strtab->name));
1306 	elf_hash_add(section_name, &shstrtab->name_hash,	str_hash(shstrtab->name));
1307 
1308 	if (elf_add_string(elf, strtab, "") == -1)
1309 		return NULL;
1310 
1311 	symtab = elf_create_section(elf, ".symtab", 0x18, 0x18, SHT_SYMTAB, 0x8, 0);
1312 	if (!symtab)
1313 		return NULL;
1314 
1315 	symtab->sh.sh_link = strtab->idx;
1316 	symtab->sh.sh_info = 1;
1317 
1318 	elf->ehdr.e_shstrndx = shstrtab->idx;
1319 	if (!gelf_update_ehdr(elf->elf, &elf->ehdr)) {
1320 		ERROR_ELF("gelf_update_ehdr");
1321 		return NULL;
1322 	}
1323 
1324 	sym = calloc(1, sizeof(*sym));
1325 	if (!sym) {
1326 		ERROR_GLIBC("calloc");
1327 		return NULL;
1328 	}
1329 
1330 	sym->name = "";
1331 	sym->sec = null;
1332 	elf_add_symbol(elf, sym);
1333 
1334 	return elf;
1335 }
1336 
1337 unsigned int elf_add_string(struct elf *elf, struct section *strtab, const char *str)
1338 {
1339 	unsigned int offset;
1340 
1341 	if (!strtab)
1342 		strtab = find_section_by_name(elf, ".strtab");
1343 	if (!strtab) {
1344 		ERROR("can't find .strtab section");
1345 		return -1;
1346 	}
1347 
1348 	if (!strtab->sh.sh_addralign) {
1349 		ERROR("'%s': invalid sh_addralign", strtab->name);
1350 		return -1;
1351 	}
1352 
1353 	offset = ALIGN(sec_size(strtab), strtab->sh.sh_addralign);
1354 
1355 	if (!elf_add_data(elf, strtab, str, strlen(str) + 1))
1356 		return -1;
1357 
1358 	return offset;
1359 }
1360 
1361 void *elf_add_data(struct elf *elf, struct section *sec, const void *data, size_t size)
1362 {
1363 	unsigned long offset;
1364 	Elf_Scn *s;
1365 
1366 	if (!sec->sh.sh_addralign) {
1367 		ERROR("'%s': invalid sh_addralign", sec->name);
1368 		return NULL;
1369 	}
1370 
1371 	s = elf_getscn(elf->elf, sec->idx);
1372 	if (!s) {
1373 		ERROR_ELF("elf_getscn");
1374 		return NULL;
1375 	}
1376 
1377 	sec->data = elf_newdata(s);
1378 	if (!sec->data) {
1379 		ERROR_ELF("elf_newdata");
1380 		return NULL;
1381 	}
1382 
1383 	sec->data->d_buf = calloc(1, size);
1384 	if (!sec->data->d_buf) {
1385 		ERROR_GLIBC("calloc");
1386 		return NULL;
1387 	}
1388 
1389 	if (data)
1390 		memcpy(sec->data->d_buf, data, size);
1391 
1392 	sec->data->d_size = size;
1393 	sec->data->d_align = sec->sh.sh_addralign;
1394 
1395 	offset = ALIGN(sec_size(sec), sec->sh.sh_addralign);
1396 	sec->sh.sh_size = offset + size;
1397 
1398 	mark_sec_changed(elf, sec, true);
1399 
1400 	return sec->data->d_buf;
1401 }
1402 
1403 struct section *elf_create_section(struct elf *elf, const char *name,
1404 				   size_t size, size_t entsize,
1405 				   unsigned int type, unsigned int align,
1406 				   unsigned int flags)
1407 {
1408 	struct section *sec, *shstrtab;
1409 	Elf_Scn *s;
1410 
1411 	if (name && find_section_by_name(elf, name)) {
1412 		ERROR("section '%s' already exists", name);
1413 		return NULL;
1414 	}
1415 
1416 	sec = calloc(1, sizeof(*sec));
1417 	if (!sec) {
1418 		ERROR_GLIBC("calloc");
1419 		return NULL;
1420 	}
1421 
1422 	INIT_LIST_HEAD(&sec->symbol_list);
1423 
1424 	/* don't actually create the section, just the data structures */
1425 	if (type == SHT_NULL)
1426 		goto add;
1427 
1428 	s = elf_newscn(elf->elf);
1429 	if (!s) {
1430 		ERROR_ELF("elf_newscn");
1431 		return NULL;
1432 	}
1433 
1434 	sec->idx = elf_ndxscn(s);
1435 
1436 	if (size) {
1437 		sec->data = elf_newdata(s);
1438 		if (!sec->data) {
1439 			ERROR_ELF("elf_newdata");
1440 			return NULL;
1441 		}
1442 
1443 		sec->data->d_size = size;
1444 		sec->data->d_align = 1;
1445 
1446 		sec->data->d_buf = calloc(1, size);
1447 		if (!sec->data->d_buf) {
1448 			ERROR_GLIBC("calloc");
1449 			return NULL;
1450 		}
1451 	}
1452 
1453 	if (!gelf_getshdr(s, &sec->sh)) {
1454 		ERROR_ELF("gelf_getshdr");
1455 		return NULL;
1456 	}
1457 
1458 	sec->sh.sh_size = size;
1459 	sec->sh.sh_entsize = entsize;
1460 	sec->sh.sh_type = type;
1461 	sec->sh.sh_addralign = align;
1462 	sec->sh.sh_flags = flags;
1463 
1464 	if (name) {
1465 		sec->name = strdup(name);
1466 		if (!sec->name) {
1467 			ERROR("strdup");
1468 			return NULL;
1469 		}
1470 
1471 		/* Add section name to .shstrtab (or .strtab for Clang) */
1472 		shstrtab = find_section_by_name(elf, ".shstrtab");
1473 		if (!shstrtab) {
1474 			shstrtab = find_section_by_name(elf, ".strtab");
1475 			if (!shstrtab) {
1476 				ERROR("can't find .shstrtab or .strtab");
1477 				return NULL;
1478 			}
1479 		}
1480 		sec->sh.sh_name = elf_add_string(elf, shstrtab, sec->name);
1481 		if (sec->sh.sh_name == -1)
1482 			return NULL;
1483 
1484 		elf_hash_add(section_name, &sec->name_hash, str_hash(sec->name));
1485 	}
1486 
1487 add:
1488 	list_add_tail(&sec->list, &elf->sections);
1489 	elf_hash_add(section, &sec->hash, sec->idx);
1490 
1491 	mark_sec_changed(elf, sec, true);
1492 
1493 	return sec;
1494 }
1495 
1496 static int elf_alloc_reloc(struct elf *elf, struct section *rsec)
1497 {
1498 	struct reloc *old_relocs, *old_relocs_end, *new_relocs;
1499 	unsigned int nr_relocs_old = sec_num_entries(rsec);
1500 	unsigned int nr_relocs_new = nr_relocs_old + 1;
1501 	unsigned long nr_alloc;
1502 	struct symbol *sym;
1503 
1504 	if (!rsec->data) {
1505 		rsec->data = elf_newdata(elf_getscn(elf->elf, rsec->idx));
1506 		if (!rsec->data) {
1507 			ERROR_ELF("elf_newdata");
1508 			return -1;
1509 		}
1510 
1511 		rsec->data->d_align = 1;
1512 		rsec->data->d_type = ELF_T_RELA;
1513 		rsec->data->d_buf = NULL;
1514 	}
1515 
1516 	rsec->data->d_size = nr_relocs_new * elf_rela_size(elf);
1517 	rsec->sh.sh_size   = rsec->data->d_size;
1518 
1519 	nr_alloc = max(64UL, roundup_pow_of_two(nr_relocs_new));
1520 	if (nr_alloc <= rsec->nr_alloc_relocs)
1521 		return 0;
1522 
1523 	if (rsec->data->d_buf && !rsec->nr_alloc_relocs) {
1524 		void *orig_buf = rsec->data->d_buf;
1525 
1526 		/*
1527 		 * The original d_buf is owned by libelf so it can't be
1528 		 * realloced.
1529 		 */
1530 		rsec->data->d_buf = malloc(nr_alloc * elf_rela_size(elf));
1531 		if (!rsec->data->d_buf) {
1532 			ERROR_GLIBC("malloc");
1533 			return -1;
1534 		}
1535 		memcpy(rsec->data->d_buf, orig_buf,
1536 		       nr_relocs_old * elf_rela_size(elf));
1537 	} else {
1538 		rsec->data->d_buf = realloc(rsec->data->d_buf,
1539 					    nr_alloc * elf_rela_size(elf));
1540 		if (!rsec->data->d_buf) {
1541 			ERROR_GLIBC("realloc");
1542 			return -1;
1543 		}
1544 	}
1545 
1546 	rsec->nr_alloc_relocs = nr_alloc;
1547 
1548 	old_relocs = rsec->relocs;
1549 	new_relocs = calloc(nr_alloc, sizeof(struct reloc));
1550 	if (!new_relocs) {
1551 		ERROR_GLIBC("calloc");
1552 		return -1;
1553 	}
1554 
1555 	if (!old_relocs)
1556 		goto done;
1557 
1558 	/*
1559 	 * The struct reloc's address has changed.  Update all the symbols and
1560 	 * relocs which reference it.
1561 	 */
1562 
1563 	old_relocs_end = &old_relocs[nr_relocs_old];
1564 	for_each_sym(elf, sym) {
1565 		struct reloc *reloc;
1566 
1567 		reloc = sym->relocs;
1568 		if (!reloc)
1569 			continue;
1570 
1571 		if (reloc >= old_relocs && reloc < old_relocs_end)
1572 			sym->relocs = &new_relocs[reloc - old_relocs];
1573 
1574 		while (1) {
1575 			struct reloc *next_reloc = sym_next_reloc(reloc);
1576 
1577 			if (!next_reloc)
1578 				break;
1579 
1580 			if (next_reloc >= old_relocs && next_reloc < old_relocs_end)
1581 				set_sym_next_reloc(reloc, &new_relocs[next_reloc - old_relocs]);
1582 
1583 			reloc = next_reloc;
1584 		}
1585 	}
1586 
1587 	memcpy(new_relocs, old_relocs, nr_relocs_old * sizeof(struct reloc));
1588 
1589 	for (int i = 0; i < nr_relocs_old; i++) {
1590 		struct reloc *old = &old_relocs[i];
1591 		struct reloc *new = &new_relocs[i];
1592 		u32 key = reloc_hash(old);
1593 
1594 		elf_hash_del(reloc, &old->hash, key);
1595 		elf_hash_add(reloc, &new->hash, key);
1596 	}
1597 
1598 	free(old_relocs);
1599 done:
1600 	rsec->relocs = new_relocs;
1601 	return 0;
1602 }
1603 
1604 struct section *elf_create_rela_section(struct elf *elf, struct section *sec,
1605 					unsigned int nr_relocs)
1606 {
1607 	struct section *rsec;
1608 	char *rsec_name;
1609 
1610 	rsec_name = malloc(strlen(sec->name) + strlen(".rela") + 1);
1611 	if (!rsec_name) {
1612 		ERROR_GLIBC("malloc");
1613 		return NULL;
1614 	}
1615 	strcpy(rsec_name, ".rela");
1616 	strcat(rsec_name, sec->name);
1617 
1618 	rsec = elf_create_section(elf, rsec_name, nr_relocs * elf_rela_size(elf),
1619 				  elf_rela_size(elf), SHT_RELA, elf_addr_size(elf),
1620 				  SHF_INFO_LINK);
1621 	free(rsec_name);
1622 	if (!rsec)
1623 		return NULL;
1624 
1625 	if (nr_relocs) {
1626 		rsec->data->d_type = ELF_T_RELA;
1627 
1628 		rsec->nr_alloc_relocs = nr_relocs;
1629 		rsec->relocs = calloc(nr_relocs, sizeof(struct reloc));
1630 		if (!rsec->relocs) {
1631 			ERROR_GLIBC("calloc");
1632 			return NULL;
1633 		}
1634 	}
1635 
1636 	rsec->sh.sh_link = find_section_by_name(elf, ".symtab")->idx;
1637 	rsec->sh.sh_info = sec->idx;
1638 
1639 	sec->rsec = rsec;
1640 	rsec->base = sec;
1641 
1642 	return rsec;
1643 }
1644 
1645 struct reloc *elf_create_reloc(struct elf *elf, struct section *sec,
1646 			       unsigned long offset,
1647 			       struct symbol *sym, s64 addend,
1648 			       unsigned int type)
1649 {
1650 	struct section *rsec = sec->rsec;
1651 
1652 	if (!rsec) {
1653 		rsec = elf_create_rela_section(elf, sec, 0);
1654 		if (!rsec)
1655 			return NULL;
1656 	}
1657 
1658 	if (find_reloc_by_dest(elf, sec, offset)) {
1659 		ERROR_FUNC(sec, offset, "duplicate reloc");
1660 		return NULL;
1661 	}
1662 
1663 	if (elf_alloc_reloc(elf, rsec))
1664 		return NULL;
1665 
1666 	mark_sec_changed(elf, rsec, true);
1667 
1668 	return elf_init_reloc(elf, rsec, sec_num_entries(rsec) - 1, offset, sym,
1669 			      addend, type);
1670 }
1671 
1672 struct section *elf_create_section_pair(struct elf *elf, const char *name,
1673 					size_t entsize, unsigned int nr,
1674 					unsigned int nr_relocs)
1675 {
1676 	struct section *sec;
1677 
1678 	sec = elf_create_section(elf, name, nr * entsize, entsize,
1679 				 SHT_PROGBITS, 1, SHF_ALLOC);
1680 	if (!sec)
1681 		return NULL;
1682 
1683 	if (!elf_create_rela_section(elf, sec, nr_relocs))
1684 		return NULL;
1685 
1686 	return sec;
1687 }
1688 
1689 int elf_write_insn(struct elf *elf, struct section *sec,
1690 		   unsigned long offset, unsigned int len,
1691 		   const char *insn)
1692 {
1693 	Elf_Data *data = sec->data;
1694 
1695 	if (data->d_type != ELF_T_BYTE || data->d_off) {
1696 		ERROR("write to unexpected data for section: %s", sec->name);
1697 		return -1;
1698 	}
1699 
1700 	memcpy(data->d_buf + offset, insn, len);
1701 
1702 	mark_sec_changed(elf, sec, true);
1703 
1704 	return 0;
1705 }
1706 
1707 /*
1708  * When Elf_Scn::sh_size is smaller than the combined Elf_Data::d_size
1709  * do you:
1710  *
1711  *   A) adhere to the section header and truncate the data, or
1712  *   B) ignore the section header and write out all the data you've got?
1713  *
1714  * Yes, libelf sucks and we need to manually truncate if we over-allocate data.
1715  */
1716 static int elf_truncate_section(struct elf *elf, struct section *sec)
1717 {
1718 	u64 size = sec_size(sec);
1719 	bool truncated = false;
1720 	Elf_Data *data = NULL;
1721 	Elf_Scn *s;
1722 
1723 	s = elf_getscn(elf->elf, sec->idx);
1724 	if (!s) {
1725 		ERROR_ELF("elf_getscn");
1726 		return -1;
1727 	}
1728 
1729 	for (;;) {
1730 		/* get next data descriptor for the relevant section */
1731 		data = elf_getdata(s, data);
1732 		if (!data) {
1733 			if (size) {
1734 				ERROR("end of section data but non-zero size left\n");
1735 				return -1;
1736 			}
1737 			return 0;
1738 		}
1739 
1740 		if (truncated) {
1741 			/* when we remove symbols */
1742 			ERROR("truncated; but more data\n");
1743 			return -1;
1744 		}
1745 
1746 		if (!data->d_size) {
1747 			ERROR("zero size data");
1748 			return -1;
1749 		}
1750 
1751 		if (data->d_size > size) {
1752 			truncated = true;
1753 			data->d_size = size;
1754 		}
1755 
1756 		size -= data->d_size;
1757 	}
1758 }
1759 
1760 int elf_write(struct elf *elf)
1761 {
1762 	struct section *sec;
1763 	Elf_Scn *s;
1764 
1765 	/* Update changed relocation sections and section headers: */
1766 	list_for_each_entry(sec, &elf->sections, list) {
1767 		if (sec->truncate && elf_truncate_section(elf, sec))
1768 			return -1;
1769 
1770 		if (sec_changed(sec)) {
1771 			s = elf_getscn(elf->elf, sec->idx);
1772 			if (!s) {
1773 				ERROR_ELF("elf_getscn");
1774 				return -1;
1775 			}
1776 
1777 			/* Note this also flags the section dirty */
1778 			if (!gelf_update_shdr(s, &sec->sh)) {
1779 				ERROR_ELF("gelf_update_shdr");
1780 				return -1;
1781 			}
1782 
1783 			mark_sec_changed(elf, sec, false);
1784 		}
1785 	}
1786 
1787 	/* Make sure the new section header entries get updated properly. */
1788 	elf_flagelf(elf->elf, ELF_C_SET, ELF_F_DIRTY);
1789 
1790 	/* Write all changes to the file. */
1791 	if (elf_update(elf->elf, ELF_C_WRITE) < 0) {
1792 		ERROR_ELF("elf_update");
1793 		return -1;
1794 	}
1795 
1796 	elf->changed = false;
1797 
1798 	return 0;
1799 }
1800 
1801 int elf_close(struct elf *elf)
1802 {
1803 	if (elf->elf)
1804 		elf_end(elf->elf);
1805 
1806 	if (elf->fd > 0)
1807 		close(elf->fd);
1808 
1809 	if (elf->tmp_name && rename(elf->tmp_name, elf->name))
1810 		return -1;
1811 
1812 	/*
1813 	 * NOTE: All remaining allocations are leaked on purpose.  Objtool is
1814 	 * about to exit anyway.
1815 	 */
1816 	return 0;
1817 }
1818