xref: /linux/tools/objtool/elf.c (revision dea622e183d34e6a4f90acfee9abb605885432bf)
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 
str_hash(const char * str)28 static inline u32 str_hash(const char *str)
29 {
30 	return jhash(str, strlen(str), 0);
31 }
32 
33 #define __elf_table(name)	(elf->name##_hash)
34 #define __elf_bits(name)	(elf->name##_bits)
35 
36 #define __elf_table_entry(name, key) \
37 	__elf_table(name)[hash_min(key, __elf_bits(name))]
38 
39 #define elf_hash_add(name, node, key)					\
40 ({									\
41 	struct elf_hash_node *__node = node;				\
42 	__node->next = __elf_table_entry(name, key);			\
43 	__elf_table_entry(name, key) = __node;				\
44 })
45 
__elf_hash_del(struct elf_hash_node * node,struct elf_hash_node ** head)46 static inline void __elf_hash_del(struct elf_hash_node *node,
47 				  struct elf_hash_node **head)
48 {
49 	struct elf_hash_node *cur, *prev;
50 
51 	if (node == *head) {
52 		*head = node->next;
53 		return;
54 	}
55 
56 	for (prev = NULL, cur = *head; cur; prev = cur, cur = cur->next) {
57 		if (cur == node) {
58 			prev->next = cur->next;
59 			break;
60 		}
61 	}
62 }
63 
64 #define elf_hash_del(name, node, key) \
65 	__elf_hash_del(node, &__elf_table_entry(name, key))
66 
67 #define elf_list_entry(ptr, type, member)				\
68 ({									\
69 	typeof(ptr) __ptr = (ptr);					\
70 	__ptr ? container_of(__ptr, type, member) : NULL;		\
71 })
72 
73 #define elf_hash_for_each_possible(name, obj, member, key)		\
74 	for (obj = elf_list_entry(__elf_table_entry(name, key), typeof(*obj), member); \
75 	     obj;							\
76 	     obj = elf_list_entry(obj->member.next, typeof(*(obj)), member))
77 
78 #define elf_alloc_hash(name, size)					\
79 ({									\
80 	__elf_bits(name) = max(10, ilog2(size));			\
81 	__elf_table(name) = mmap(NULL, sizeof(struct elf_hash_node *) << __elf_bits(name), \
82 				 PROT_READ|PROT_WRITE,			\
83 				 MAP_PRIVATE|MAP_ANON, -1, 0);		\
84 	if (__elf_table(name) == (void *)-1L) {				\
85 		ERROR_GLIBC("mmap fail " #name);			\
86 		__elf_table(name) = NULL;				\
87 	}								\
88 	__elf_table(name);						\
89 })
90 
__sym_start(struct symbol * s)91 static inline unsigned long __sym_start(struct symbol *s)
92 {
93 	return s->offset;
94 }
95 
__sym_last(struct symbol * s)96 static inline unsigned long __sym_last(struct symbol *s)
97 {
98 	return s->offset + (s->len ? s->len - 1 : 0);
99 }
100 
101 INTERVAL_TREE_DEFINE(struct symbol, node, unsigned long, __subtree_last,
102 		     __sym_start, __sym_last, static inline __maybe_unused,
103 		     __sym)
104 
105 #define __sym_for_each(_iter, _tree, _start, _end)			\
106 	for (_iter = __sym_iter_first((_tree), (_start), (_end));	\
107 	     _iter; _iter = __sym_iter_next(_iter, (_start), (_end)))
108 
109 struct symbol_hole {
110 	unsigned long key;
111 	const struct symbol *sym;
112 };
113 
114 /*
115  * Find the last symbol before @offset.
116  */
symbol_hole_by_offset(const void * key,const struct rb_node * node)117 static int symbol_hole_by_offset(const void *key, const struct rb_node *node)
118 {
119 	const struct symbol *s = rb_entry(node, struct symbol, node);
120 	struct symbol_hole *sh = (void *)key;
121 
122 	if (sh->key < s->offset)
123 		return -1;
124 
125 	if (sh->key >= s->offset + s->len) {
126 		sh->sym = s;
127 		return 1;
128 	}
129 
130 	return 0;
131 }
132 
find_section_by_name(const struct elf * elf,const char * name)133 struct section *find_section_by_name(const struct elf *elf, const char *name)
134 {
135 	struct section *sec;
136 
137 	elf_hash_for_each_possible(section_name, sec, name_hash, str_hash(name)) {
138 		if (!strcmp(sec->name, name))
139 			return sec;
140 	}
141 
142 	return NULL;
143 }
144 
find_section_by_index(struct elf * elf,unsigned int idx)145 static struct section *find_section_by_index(struct elf *elf,
146 					     unsigned int idx)
147 {
148 	struct section *sec;
149 
150 	elf_hash_for_each_possible(section, sec, hash, idx) {
151 		if (sec->idx == idx)
152 			return sec;
153 	}
154 
155 	return NULL;
156 }
157 
find_symbol_by_index(struct elf * elf,unsigned int idx)158 static struct symbol *find_symbol_by_index(struct elf *elf, unsigned int idx)
159 {
160 	struct symbol *sym;
161 
162 	elf_hash_for_each_possible(symbol, sym, hash, idx) {
163 		if (sym->idx == idx)
164 			return sym;
165 	}
166 
167 	return NULL;
168 }
169 
find_symbol_by_offset(struct section * sec,unsigned long offset)170 struct symbol *find_symbol_by_offset(struct section *sec, unsigned long offset)
171 {
172 	struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
173 	struct symbol *sym;
174 
175 	__sym_for_each(sym, tree, offset, offset) {
176 		if (sym->offset == offset && !is_sec_sym(sym))
177 			return sym->alias;
178 	}
179 
180 	return NULL;
181 }
182 
find_func_by_offset(struct section * sec,unsigned long offset)183 struct symbol *find_func_by_offset(struct section *sec, unsigned long offset)
184 {
185 	struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
186 	struct symbol *func;
187 
188 	__sym_for_each(func, tree, offset, offset) {
189 		if (func->offset == offset && is_func_sym(func))
190 			return func->alias;
191 	}
192 
193 	return NULL;
194 }
195 
find_symbol_containing(const struct section * sec,unsigned long offset)196 struct symbol *find_symbol_containing(const struct section *sec, unsigned long offset)
197 {
198 	struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
199 	struct symbol *sym = NULL, *tmp;
200 
201 	__sym_for_each(tmp, tree, offset, offset) {
202 		if (tmp->len) {
203 			if (!sym) {
204 				sym = tmp;
205 				continue;
206 			}
207 
208 			if (sym->offset != tmp->offset || sym->len != tmp->len) {
209 				/*
210 				 * In the rare case of overlapping symbols,
211 				 * pick the smaller one.
212 				 *
213 				 * TODO: outlaw overlapping symbols
214 				 */
215 				if (tmp->len < sym->len)
216 					sym = tmp;
217 			}
218 		}
219 	}
220 
221 	return sym ? sym->alias : NULL;
222 }
223 
224 /*
225  * Returns size of hole starting at @offset.
226  */
find_symbol_hole_containing(const struct section * sec,unsigned long offset)227 int find_symbol_hole_containing(const struct section *sec, unsigned long offset)
228 {
229 	struct symbol_hole hole = {
230 		.key = offset,
231 		.sym = NULL,
232 	};
233 	struct rb_node *n;
234 	struct symbol *s;
235 
236 	/*
237 	 * Find the rightmost symbol for which @offset is after it.
238 	 */
239 	n = rb_find(&hole, &sec->symbol_tree.rb_root, symbol_hole_by_offset);
240 
241 	/* found a symbol that contains @offset */
242 	if (n)
243 		return 0; /* not a hole */
244 
245 	/*
246 	 * @offset >= sym->offset + sym->len, find symbol after it.
247 	 * When hole.sym is empty, use the first node to compute the hole.
248 	 * If there is no symbol in the section, the first node will be NULL,
249 	 * in which case, -1 is returned to skip the whole section.
250 	 */
251 	if (hole.sym)
252 		n = rb_next(&hole.sym->node);
253 	else
254 		n = rb_first_cached(&sec->symbol_tree);
255 
256 	if (!n)
257 		return -1; /* until end of address space */
258 
259 	/* hole until start of next symbol */
260 	s = rb_entry(n, struct symbol, node);
261 	return s->offset - offset;
262 }
263 
find_func_containing(struct section * sec,unsigned long offset)264 struct symbol *find_func_containing(struct section *sec, unsigned long offset)
265 {
266 	struct rb_root_cached *tree = (struct rb_root_cached *)&sec->symbol_tree;
267 	struct symbol *func;
268 
269 	__sym_for_each(func, tree, offset, offset) {
270 		if (is_func_sym(func))
271 			return func->alias;
272 	}
273 
274 	return NULL;
275 }
276 
find_symbol_by_name(const struct elf * elf,const char * name)277 struct symbol *find_symbol_by_name(const struct elf *elf, const char *name)
278 {
279 	struct symbol *sym;
280 
281 	elf_hash_for_each_possible(symbol_name, sym, name_hash, str_hash(name)) {
282 		if (!strcmp(sym->name, name))
283 			return sym;
284 	}
285 
286 	return NULL;
287 }
288 
289 /* Find local symbol with matching STT_FILE */
find_local_symbol_by_file_and_name(const struct elf * elf,struct symbol * file,const char * name)290 static struct symbol *find_local_symbol_by_file_and_name(const struct elf *elf,
291 							 struct symbol *file,
292 							 const char *name)
293 {
294 	struct symbol *sym;
295 
296 	elf_hash_for_each_possible(symbol_name, sym, name_hash, str_hash(name)) {
297 		if (sym->bind == STB_LOCAL && sym->file == file &&
298 		    !strcmp(sym->name, name)) {
299 			return sym;
300 		}
301 	}
302 
303 	return NULL;
304 }
305 
find_global_symbol_by_name(const struct elf * elf,const char * name)306 struct symbol *find_global_symbol_by_name(const struct elf *elf, const char *name)
307 {
308 	struct symbol *sym;
309 
310 	elf_hash_for_each_possible(symbol_name, sym, name_hash, str_hash(name)) {
311 		if (!strcmp(sym->name, name) && !is_local_sym(sym))
312 			return sym;
313 	}
314 
315 	return NULL;
316 }
317 
find_reloc_by_dest_range(const struct elf * elf,struct section * sec,unsigned long offset,unsigned int len)318 struct reloc *find_reloc_by_dest_range(const struct elf *elf, struct section *sec,
319 				     unsigned long offset, unsigned int len)
320 {
321 	struct reloc *reloc, *r = NULL;
322 	struct section *rsec;
323 	unsigned long o;
324 
325 	rsec = sec->rsec;
326 	if (!rsec)
327 		return NULL;
328 
329 	for_offset_range(o, offset, offset + len) {
330 		elf_hash_for_each_possible(reloc, reloc, hash,
331 					   sec_offset_hash(rsec, o)) {
332 			if (reloc->sec != rsec)
333 				continue;
334 
335 			if (reloc_offset(reloc) >= offset &&
336 			    reloc_offset(reloc) < offset + len) {
337 				if (!r || reloc_offset(reloc) < reloc_offset(r))
338 					r = reloc;
339 			}
340 		}
341 		if (r)
342 			return r;
343 	}
344 
345 	return NULL;
346 }
347 
find_reloc_by_dest(const struct elf * elf,struct section * sec,unsigned long offset)348 struct reloc *find_reloc_by_dest(const struct elf *elf, struct section *sec, unsigned long offset)
349 {
350 	return find_reloc_by_dest_range(elf, sec, offset, 1);
351 }
352 
is_dwarf_section(struct section * sec)353 static bool is_dwarf_section(struct section *sec)
354 {
355 	return !strncmp(sec->name, ".debug_", 7);
356 }
357 
read_sections(struct elf * elf)358 static int read_sections(struct elf *elf)
359 {
360 	Elf_Scn *s = NULL;
361 	struct section *sec;
362 	size_t shstrndx, sections_nr;
363 	int i;
364 
365 	if (elf_getshdrnum(elf->elf, &sections_nr)) {
366 		ERROR_ELF("elf_getshdrnum");
367 		return -1;
368 	}
369 
370 	if (elf_getshdrstrndx(elf->elf, &shstrndx)) {
371 		ERROR_ELF("elf_getshdrstrndx");
372 		return -1;
373 	}
374 
375 	if (!elf_alloc_hash(section, sections_nr) ||
376 	    !elf_alloc_hash(section_name, sections_nr))
377 		return -1;
378 
379 	elf->section_data = calloc(sections_nr, sizeof(*sec));
380 	if (!elf->section_data) {
381 		ERROR_GLIBC("calloc");
382 		return -1;
383 	}
384 	for (i = 0; i < sections_nr; i++) {
385 		sec = &elf->section_data[i];
386 
387 		INIT_LIST_HEAD(&sec->symbol_list);
388 
389 		s = elf_getscn(elf->elf, i);
390 		if (!s) {
391 			ERROR_ELF("elf_getscn");
392 			return -1;
393 		}
394 
395 		sec->idx = elf_ndxscn(s);
396 
397 		if (!gelf_getshdr(s, &sec->sh)) {
398 			ERROR_ELF("gelf_getshdr");
399 			return -1;
400 		}
401 
402 		sec->name = elf_strptr(elf->elf, shstrndx, sec->sh.sh_name);
403 		if (!sec->name) {
404 			ERROR_ELF("elf_strptr");
405 			return -1;
406 		}
407 
408 		if (sec_size(sec) != 0 && !is_dwarf_section(sec)) {
409 			sec->data = elf_getdata(s, NULL);
410 			if (!sec->data) {
411 				ERROR_ELF("elf_getdata");
412 				return -1;
413 			}
414 			if (sec->data->d_off != 0 ||
415 			    sec->data->d_size != sec_size(sec)) {
416 				ERROR("unexpected data attributes for %s", sec->name);
417 				return -1;
418 			}
419 		}
420 
421 		list_add_tail(&sec->list, &elf->sections);
422 		elf_hash_add(section, &sec->hash, sec->idx);
423 		elf_hash_add(section_name, &sec->name_hash, str_hash(sec->name));
424 
425 		if (is_reloc_sec(sec))
426 			elf->num_relocs += sec_num_entries(sec);
427 	}
428 
429 	if (opts.stats) {
430 		printf("nr_sections: %lu\n", (unsigned long)sections_nr);
431 		printf("section_bits: %d\n", elf->section_bits);
432 	}
433 
434 	/* sanity check, one more call to elf_nextscn() should return NULL */
435 	if (elf_nextscn(elf->elf, s)) {
436 		ERROR("section entry mismatch");
437 		return -1;
438 	}
439 
440 	return 0;
441 }
442 
demangle_name(struct symbol * sym)443 static const char *demangle_name(struct symbol *sym)
444 {
445 	char *str;
446 
447 	if (!is_local_sym(sym))
448 		return sym->name;
449 
450 	if (!is_func_sym(sym) && !is_object_sym(sym))
451 		return sym->name;
452 
453 	if (!strstarts(sym->name, "__UNIQUE_ID_") && !strchr(sym->name, '.'))
454 		return sym->name;
455 
456 	str = strdup(sym->name);
457 	if (!str) {
458 		ERROR_GLIBC("strdup");
459 		return NULL;
460 	}
461 
462 	for (int i = strlen(str) - 1; i >= 0; i--) {
463 		char c = str[i];
464 
465 		if (!isdigit(c) && c != '.') {
466 			str[i + 1] = '\0';
467 			break;
468 		}
469 	}
470 
471 	return str;
472 }
473 
elf_add_symbol(struct elf * elf,struct symbol * sym)474 static int elf_add_symbol(struct elf *elf, struct symbol *sym)
475 {
476 	struct list_head *entry;
477 	struct rb_node *pnode;
478 	struct symbol *iter;
479 
480 	INIT_LIST_HEAD(&sym->pv_target);
481 	sym->alias = sym;
482 
483 	sym->type = GELF_ST_TYPE(sym->sym.st_info);
484 	sym->bind = GELF_ST_BIND(sym->sym.st_info);
485 
486 	if (is_file_sym(sym))
487 		elf->num_files++;
488 
489 	sym->offset = sym->sym.st_value;
490 	sym->len = sym->sym.st_size;
491 
492 	__sym_for_each(iter, &sym->sec->symbol_tree, sym->offset, sym->offset) {
493 		if (!is_undef_sym(iter) && iter->offset == sym->offset &&
494 		    iter->type == sym->type && iter->len == sym->len)
495 			iter->alias = sym;
496 	}
497 
498 	__sym_insert(sym, &sym->sec->symbol_tree);
499 	pnode = rb_prev(&sym->node);
500 	if (pnode)
501 		entry = &rb_entry(pnode, struct symbol, node)->list;
502 	else
503 		entry = &sym->sec->symbol_list;
504 	list_add(&sym->list, entry);
505 
506 	list_add_tail(&sym->global_list, &elf->symbols);
507 	elf_hash_add(symbol, &sym->hash, sym->idx);
508 	elf_hash_add(symbol_name, &sym->name_hash, str_hash(sym->name));
509 
510 	if (is_func_sym(sym) &&
511 	    (strstarts(sym->name, "__pfx_") ||
512 	     strstarts(sym->name, "__cfi_") ||
513 	     strstarts(sym->name, "__pi___pfx_") ||
514 	     strstarts(sym->name, "__pi___cfi_")))
515 		sym->prefix = 1;
516 
517 	if (strstarts(sym->name, ".klp.sym"))
518 		sym->klp = 1;
519 
520 	if (!sym->klp && !is_sec_sym(sym) && strstr(sym->name, ".cold")) {
521 		sym->cold = 1;
522 
523 		/*
524 		 * Clang doesn't mark cold subfunctions as STT_FUNC, which
525 		 * breaks several objtool assumptions.  Fake it.
526 		 */
527 		sym->type = STT_FUNC;
528 	}
529 
530 	sym->pfunc = sym->cfunc = sym;
531 
532 	sym->demangled_name = demangle_name(sym);
533 	if (!sym->demangled_name)
534 		return -1;
535 
536 	return 0;
537 }
538 
read_symbols(struct elf * elf)539 static int read_symbols(struct elf *elf)
540 {
541 	struct section *symtab, *symtab_shndx, *sec;
542 	struct symbol *sym, *pfunc, *file = NULL;
543 	int symbols_nr, i;
544 	char *coldstr;
545 	Elf_Data *shndx_data = NULL;
546 	Elf32_Word shndx;
547 
548 	symtab = find_section_by_name(elf, ".symtab");
549 	if (symtab) {
550 		symtab_shndx = find_section_by_name(elf, ".symtab_shndx");
551 		if (symtab_shndx)
552 			shndx_data = symtab_shndx->data;
553 
554 		symbols_nr = sec_num_entries(symtab);
555 	} else {
556 		/*
557 		 * A missing symbol table is actually possible if it's an empty
558 		 * .o file. This can happen for thunk_64.o. Make sure to at
559 		 * least allocate the symbol hash tables so we can do symbol
560 		 * lookups without crashing.
561 		 */
562 		symbols_nr = 0;
563 	}
564 
565 	if (!elf_alloc_hash(symbol, symbols_nr) ||
566 	    !elf_alloc_hash(symbol_name, symbols_nr))
567 		return -1;
568 
569 	elf->symbol_data = calloc(symbols_nr, sizeof(*sym));
570 	if (!elf->symbol_data) {
571 		ERROR_GLIBC("calloc");
572 		return -1;
573 	}
574 
575 	INIT_LIST_HEAD(&elf->symbols);
576 
577 	for (i = 0; i < symbols_nr; i++) {
578 		sym = &elf->symbol_data[i];
579 
580 		sym->idx = i;
581 
582 		if (!gelf_getsymshndx(symtab->data, shndx_data, i, &sym->sym,
583 				      &shndx)) {
584 			ERROR_ELF("gelf_getsymshndx");
585 			return -1;
586 		}
587 
588 		sym->name = elf_strptr(elf->elf, symtab->sh.sh_link,
589 				       sym->sym.st_name);
590 		if (!sym->name) {
591 			ERROR_ELF("elf_strptr");
592 			return -1;
593 		}
594 
595 		if ((sym->sym.st_shndx > SHN_UNDEF &&
596 		     sym->sym.st_shndx < SHN_LORESERVE) ||
597 		    (shndx_data && sym->sym.st_shndx == SHN_XINDEX)) {
598 			if (sym->sym.st_shndx != SHN_XINDEX)
599 				shndx = sym->sym.st_shndx;
600 
601 			sym->sec = find_section_by_index(elf, shndx);
602 			if (!sym->sec) {
603 				ERROR("couldn't find section for symbol %s", sym->name);
604 				return -1;
605 			}
606 			if (GELF_ST_TYPE(sym->sym.st_info) == STT_SECTION) {
607 				sym->name = sym->sec->name;
608 				sym->sec->sym = sym;
609 			}
610 		} else
611 			sym->sec = find_section_by_index(elf, 0);
612 
613 		if (elf_add_symbol(elf, sym))
614 			return -1;
615 
616 		if (sym->type == STT_FILE)
617 			file = sym;
618 		else if (sym->bind == STB_LOCAL)
619 			sym->file = file;
620 	}
621 
622 	if (opts.stats) {
623 		printf("nr_symbols: %lu\n", (unsigned long)symbols_nr);
624 		printf("symbol_bits: %d\n", elf->symbol_bits);
625 	}
626 
627 	/* Create parent/child links for any cold subfunctions */
628 	list_for_each_entry(sec, &elf->sections, list) {
629 		sec_for_each_sym(sec, sym) {
630 			char *pname;
631 			size_t pnamelen;
632 
633 			if (!sym->cold)
634 				continue;
635 
636 			coldstr = strstr(sym->name, ".cold");
637 			if (!coldstr) {
638 				ERROR("%s(): cold subfunction without \".cold\"?", sym->name);
639 				return -1;
640 			}
641 
642 			pnamelen = coldstr - sym->name;
643 			pname = strndup(sym->name, pnamelen);
644 			if (!pname) {
645 				ERROR("%s(): failed to allocate memory", sym->name);
646 				return -1;
647 			}
648 
649 			pfunc = find_local_symbol_by_file_and_name(elf, sym->file, pname);
650 			if (!pfunc)
651 				pfunc = find_global_symbol_by_name(elf, pname);
652 			free(pname);
653 
654 			if (!pfunc) {
655 				ERROR("%s(): can't find parent function", sym->name);
656 				return -1;
657 			}
658 
659 			sym->pfunc = pfunc->alias;
660 			pfunc->cfunc = sym;
661 			pfunc->alias->cfunc = sym;
662 
663 			/*
664 			 * Unfortunately, -fnoreorder-functions puts the child
665 			 * inside the parent.  Remove the overlap so we can
666 			 * have sane assumptions.
667 			 *
668 			 * Note that pfunc->len now no longer matches
669 			 * pfunc->sym.st_size.
670 			 */
671 			if (sym->sec == pfunc->sec &&
672 			    sym->offset >= pfunc->offset &&
673 			    sym->offset + sym->len == pfunc->offset + pfunc->len) {
674 				pfunc->len -= sym->len;
675 			}
676 		}
677 	}
678 
679 	return 0;
680 }
681 
mark_group_syms(struct elf * elf)682 static int mark_group_syms(struct elf *elf)
683 {
684 	struct section *symtab, *sec;
685 	struct symbol *sym;
686 
687 	symtab = find_section_by_name(elf, ".symtab");
688 	if (!symtab) {
689 		ERROR("no .symtab");
690 		return -1;
691 	}
692 
693 	for_each_sec(elf, sec) {
694 		if (sec->sh.sh_type == SHT_GROUP &&
695 		    sec->sh.sh_link == symtab->idx) {
696 			sym = find_symbol_by_index(elf, sec->sh.sh_info);
697 			if (!sym) {
698 				ERROR("%s: can't find SHT_GROUP signature symbol",
699 				      sec->name);
700 				return -1;
701 			}
702 
703 			sym->group_sec = sec;
704 		}
705 	}
706 
707 	return 0;
708 }
709 
710 /*
711  * @sym's idx has changed.  Update the relocs which reference it.
712  */
elf_update_sym_relocs(struct elf * elf,struct symbol * sym)713 static int elf_update_sym_relocs(struct elf *elf, struct symbol *sym)
714 {
715 	struct reloc *reloc;
716 
717 	for (reloc = sym->relocs; reloc; reloc = sym_next_reloc(reloc))
718 		set_reloc_sym(elf, reloc, reloc->sym->idx);
719 
720 	return 0;
721 }
722 
723 /*
724  * The libelf API is terrible; gelf_update_sym*() takes a data block relative
725  * index value, *NOT* the symbol index. As such, iterate the data blocks and
726  * adjust index until it fits.
727  *
728  * If no data block is found, allow adding a new data block provided the index
729  * is only one past the end.
730  */
elf_update_symbol(struct elf * elf,struct section * symtab,struct section * symtab_shndx,struct symbol * sym)731 static int elf_update_symbol(struct elf *elf, struct section *symtab,
732 			     struct section *symtab_shndx, struct symbol *sym)
733 {
734 	Elf32_Word shndx;
735 	Elf_Data *symtab_data = NULL, *shndx_data = NULL;
736 	Elf64_Xword entsize = symtab->sh.sh_entsize;
737 	int max_idx, idx = sym->idx;
738 	Elf_Scn *s, *t = NULL;
739 	bool is_special_shndx = sym->sym.st_shndx >= SHN_LORESERVE &&
740 				sym->sym.st_shndx != SHN_XINDEX;
741 
742 	shndx = is_special_shndx ? sym->sym.st_shndx : sym->sec->idx;
743 
744 	s = elf_getscn(elf->elf, symtab->idx);
745 	if (!s) {
746 		ERROR_ELF("elf_getscn");
747 		return -1;
748 	}
749 
750 	if (symtab_shndx) {
751 		t = elf_getscn(elf->elf, symtab_shndx->idx);
752 		if (!t) {
753 			ERROR_ELF("elf_getscn");
754 			return -1;
755 		}
756 	}
757 
758 	for (;;) {
759 		/* get next data descriptor for the relevant sections */
760 		symtab_data = elf_getdata(s, symtab_data);
761 		if (t)
762 			shndx_data = elf_getdata(t, shndx_data);
763 
764 		/* end-of-list */
765 		if (!symtab_data) {
766 			/*
767 			 * Over-allocate to avoid O(n^2) symbol creation
768 			 * behaviour.  The down side is that libelf doesn't
769 			 * like this; see elf_truncate_section() for the fixup.
770 			 */
771 			int num = max(1U, sym->idx/3);
772 			void *buf;
773 
774 			if (idx) {
775 				/* we don't do holes in symbol tables */
776 				ERROR("index out of range");
777 				return -1;
778 			}
779 
780 			/* if @idx == 0, it's the next contiguous entry, create it */
781 			symtab_data = elf_newdata(s);
782 			if (t)
783 				shndx_data = elf_newdata(t);
784 
785 			buf = calloc(num, entsize);
786 			if (!buf) {
787 				ERROR_GLIBC("calloc");
788 				return -1;
789 			}
790 
791 			symtab_data->d_buf = buf;
792 			symtab_data->d_size = num * entsize;
793 			symtab_data->d_align = 1;
794 			symtab_data->d_type = ELF_T_SYM;
795 
796 			mark_sec_changed(elf, symtab, true);
797 			symtab->truncate = true;
798 
799 			if (t) {
800 				buf = calloc(num, sizeof(Elf32_Word));
801 				if (!buf) {
802 					ERROR_GLIBC("calloc");
803 					return -1;
804 				}
805 
806 				shndx_data->d_buf = buf;
807 				shndx_data->d_size = num * sizeof(Elf32_Word);
808 				shndx_data->d_align = sizeof(Elf32_Word);
809 				shndx_data->d_type = ELF_T_WORD;
810 
811 				mark_sec_changed(elf, symtab_shndx, true);
812 				symtab_shndx->truncate = true;
813 			}
814 
815 			break;
816 		}
817 
818 		/* empty blocks should not happen */
819 		if (!symtab_data->d_size) {
820 			ERROR("zero size data");
821 			return -1;
822 		}
823 
824 		/* is this the right block? */
825 		max_idx = symtab_data->d_size / entsize;
826 		if (idx < max_idx)
827 			break;
828 
829 		/* adjust index and try again */
830 		idx -= max_idx;
831 	}
832 
833 	/* something went side-ways */
834 	if (idx < 0) {
835 		ERROR("negative index");
836 		return -1;
837 	}
838 
839 	/* setup extended section index magic and write the symbol */
840 	if (shndx < SHN_LORESERVE || is_special_shndx) {
841 		sym->sym.st_shndx = shndx;
842 		if (!shndx_data)
843 			shndx = 0;
844 	} else {
845 		sym->sym.st_shndx = SHN_XINDEX;
846 		if (!shndx_data) {
847 			ERROR("no .symtab_shndx");
848 			return -1;
849 		}
850 	}
851 
852 	if (!gelf_update_symshndx(symtab_data, shndx_data, idx, &sym->sym, shndx)) {
853 		ERROR_ELF("gelf_update_symshndx");
854 		return -1;
855 	}
856 
857 	return 0;
858 }
859 
elf_create_symbol(struct elf * elf,const char * name,struct section * sec,unsigned int bind,unsigned int type,unsigned long offset,size_t size)860 struct symbol *elf_create_symbol(struct elf *elf, const char *name,
861 				 struct section *sec, unsigned int bind,
862 				 unsigned int type, unsigned long offset,
863 				 size_t size)
864 {
865 	struct section *symtab, *symtab_shndx;
866 	Elf32_Word first_non_local, new_idx;
867 	struct symbol *old, *sym;
868 
869 	sym = calloc(1, sizeof(*sym));
870 	if (!sym) {
871 		ERROR_GLIBC("calloc");
872 		return NULL;
873 	}
874 
875 	sym->name = strdup(name);
876 	if (!sym->name) {
877 		ERROR_GLIBC("strdup");
878 		return NULL;
879 	}
880 
881 	if (type != STT_SECTION) {
882 		sym->sym.st_name = elf_add_string(elf, NULL, sym->name);
883 		if (sym->sym.st_name == -1)
884 			return NULL;
885 	}
886 
887 	if (sec) {
888 		sym->sec = sec;
889 	} else {
890 		sym->sec = find_section_by_index(elf, 0);
891 		if (!sym->sec) {
892 			ERROR("no NULL section");
893 			return NULL;
894 		}
895 	}
896 
897 	sym->sym.st_info  = GELF_ST_INFO(bind, type);
898 	sym->sym.st_value = offset;
899 	sym->sym.st_size  = size;
900 
901 	symtab = find_section_by_name(elf, ".symtab");
902 	if (!symtab) {
903 		ERROR("no .symtab");
904 		return NULL;
905 	}
906 
907 	symtab_shndx = find_section_by_name(elf, ".symtab_shndx");
908 
909 	new_idx = sec_num_entries(symtab);
910 
911 	if (bind != STB_LOCAL)
912 		goto non_local;
913 
914 	/*
915 	 * Move the first global symbol, as per sh_info, into a new, higher
916 	 * symbol index. This frees up a spot for a new local symbol.
917 	 */
918 	first_non_local = symtab->sh.sh_info;
919 	old = find_symbol_by_index(elf, first_non_local);
920 	if (old) {
921 
922 		elf_hash_del(symbol, &old->hash, old->idx);
923 		elf_hash_add(symbol, &old->hash, new_idx);
924 		old->idx = new_idx;
925 
926 		if (elf_update_symbol(elf, symtab, symtab_shndx, old)) {
927 			ERROR("elf_update_symbol move");
928 			return NULL;
929 		}
930 
931 		if (elf_update_sym_relocs(elf, old))
932 			return NULL;
933 
934 		if (old->group_sec) {
935 			old->group_sec->sh.sh_info = new_idx;
936 			mark_sec_changed(elf, old->group_sec, true);
937 		}
938 
939 		new_idx = first_non_local;
940 	}
941 
942 	/*
943 	 * Either way, we will add a LOCAL symbol.
944 	 */
945 	symtab->sh.sh_info += 1;
946 
947 non_local:
948 	sym->idx = new_idx;
949 	if (sym->idx && elf_update_symbol(elf, symtab, symtab_shndx, sym))
950 		return NULL;
951 
952 	symtab->sh.sh_size += symtab->sh.sh_entsize;
953 	mark_sec_changed(elf, symtab, true);
954 
955 	if (symtab_shndx) {
956 		symtab_shndx->sh.sh_size += sizeof(Elf32_Word);
957 		mark_sec_changed(elf, symtab_shndx, true);
958 	}
959 
960 	if (elf_add_symbol(elf, sym))
961 		return NULL;
962 
963 	return sym;
964 }
965 
elf_create_section_symbol(struct elf * elf,struct section * sec)966 struct symbol *elf_create_section_symbol(struct elf *elf, struct section *sec)
967 {
968 	struct symbol *sym = calloc(1, sizeof(*sym));
969 
970 	sym = elf_create_symbol(elf, sec->name, sec, STB_LOCAL, STT_SECTION, 0, 0);
971 	if (!sym)
972 		return NULL;
973 
974 	sec->sym = sym;
975 
976 	return sym;
977 }
978 
elf_init_reloc(struct elf * elf,struct section * rsec,unsigned int reloc_idx,unsigned long offset,struct symbol * sym,s64 addend,unsigned int type)979 struct reloc *elf_init_reloc(struct elf *elf, struct section *rsec,
980 			     unsigned int reloc_idx, unsigned long offset,
981 			     struct symbol *sym, s64 addend, unsigned int type)
982 {
983 	struct reloc *reloc, empty = { 0 };
984 
985 	if (reloc_idx >= sec_num_entries(rsec)) {
986 		ERROR("%s: bad reloc_idx %u for %s with %d relocs",
987 		      __func__, reloc_idx, rsec->name, sec_num_entries(rsec));
988 		return NULL;
989 	}
990 
991 	reloc = &rsec->relocs[reloc_idx];
992 
993 	if (memcmp(reloc, &empty, sizeof(empty))) {
994 		ERROR("%s: %s: reloc %d already initialized!",
995 		      __func__, rsec->name, reloc_idx);
996 		return NULL;
997 	}
998 
999 	reloc->sec = rsec;
1000 	reloc->sym = sym;
1001 
1002 	set_reloc_offset(elf, reloc, offset);
1003 	set_reloc_sym(elf, reloc, sym->idx);
1004 	set_reloc_type(elf, reloc, type);
1005 	set_reloc_addend(elf, reloc, addend);
1006 
1007 	elf_hash_add(reloc, &reloc->hash, reloc_hash(reloc));
1008 	set_sym_next_reloc(reloc, sym->relocs);
1009 	sym->relocs = reloc;
1010 
1011 	return reloc;
1012 }
1013 
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)1014 struct reloc *elf_init_reloc_text_sym(struct elf *elf, struct section *sec,
1015 				      unsigned long offset,
1016 				      unsigned int reloc_idx,
1017 				      struct section *insn_sec,
1018 				      unsigned long insn_off)
1019 {
1020 	struct symbol *sym = insn_sec->sym;
1021 	s64 addend = insn_off;
1022 
1023 	if (!is_text_sec(insn_sec)) {
1024 		ERROR("bad call to %s() for data symbol %s", __func__, sym->name);
1025 		return NULL;
1026 	}
1027 
1028 	if (!sym) {
1029 		/*
1030 		 * Due to how weak functions work, we must use section based
1031 		 * relocations. Symbol based relocations would result in the
1032 		 * weak and non-weak function annotations being overlaid on the
1033 		 * non-weak function after linking.
1034 		 */
1035 		sym = elf_create_section_symbol(elf, insn_sec);
1036 		if (!sym)
1037 			return NULL;
1038 	}
1039 
1040 	return elf_init_reloc(elf, sec->rsec, reloc_idx, offset, sym, addend,
1041 			      elf_text_rela_type(elf));
1042 }
1043 
elf_init_reloc_data_sym(struct elf * elf,struct section * sec,unsigned long offset,unsigned int reloc_idx,struct symbol * sym,s64 addend)1044 struct reloc *elf_init_reloc_data_sym(struct elf *elf, struct section *sec,
1045 				      unsigned long offset,
1046 				      unsigned int reloc_idx,
1047 				      struct symbol *sym,
1048 				      s64 addend)
1049 {
1050 	if (is_text_sec(sec)) {
1051 		ERROR("bad call to %s() for text symbol %s", __func__, sym->name);
1052 		return NULL;
1053 	}
1054 
1055 	return elf_init_reloc(elf, sec->rsec, reloc_idx, offset, sym, addend,
1056 			      elf_data_rela_type(elf));
1057 }
1058 
read_relocs(struct elf * elf)1059 static int read_relocs(struct elf *elf)
1060 {
1061 	unsigned long nr_reloc, max_reloc = 0;
1062 	struct section *rsec;
1063 	struct reloc *reloc;
1064 	unsigned int symndx;
1065 	struct symbol *sym;
1066 	int i;
1067 
1068 	if (!elf_alloc_hash(reloc, elf->num_relocs))
1069 		return -1;
1070 
1071 	list_for_each_entry(rsec, &elf->sections, list) {
1072 		if (!is_reloc_sec(rsec))
1073 			continue;
1074 
1075 		rsec->base = find_section_by_index(elf, rsec->sh.sh_info);
1076 		if (!rsec->base) {
1077 			ERROR("can't find base section for reloc section %s", rsec->name);
1078 			return -1;
1079 		}
1080 
1081 		rsec->base->rsec = rsec;
1082 
1083 		/* nr_alloc_relocs=0: libelf owns d_buf */
1084 		rsec->nr_alloc_relocs = 0;
1085 
1086 		rsec->relocs = calloc(sec_num_entries(rsec), sizeof(*reloc));
1087 		if (!rsec->relocs) {
1088 			ERROR_GLIBC("calloc");
1089 			return -1;
1090 		}
1091 
1092 		nr_reloc = 0;
1093 		for (i = 0; i < sec_num_entries(rsec); i++) {
1094 			reloc = &rsec->relocs[i];
1095 
1096 			reloc->sec = rsec;
1097 			symndx = reloc_sym(reloc);
1098 			reloc->sym = sym = find_symbol_by_index(elf, symndx);
1099 			if (!reloc->sym) {
1100 				ERROR("can't find reloc entry symbol %d for %s", symndx, rsec->name);
1101 				return -1;
1102 			}
1103 
1104 			elf_hash_add(reloc, &reloc->hash, reloc_hash(reloc));
1105 			set_sym_next_reloc(reloc, sym->relocs);
1106 			sym->relocs = reloc;
1107 
1108 			nr_reloc++;
1109 		}
1110 		max_reloc = max(max_reloc, nr_reloc);
1111 	}
1112 
1113 	if (opts.stats) {
1114 		printf("max_reloc: %lu\n", max_reloc);
1115 		printf("num_relocs: %lu\n", elf->num_relocs);
1116 		printf("reloc_bits: %d\n", elf->reloc_bits);
1117 	}
1118 
1119 	return 0;
1120 }
1121 
elf_open_read(const char * name,int flags)1122 struct elf *elf_open_read(const char *name, int flags)
1123 {
1124 	struct elf *elf;
1125 	Elf_Cmd cmd;
1126 
1127 	elf_version(EV_CURRENT);
1128 
1129 	elf = malloc(sizeof(*elf));
1130 	if (!elf) {
1131 		ERROR_GLIBC("malloc");
1132 		return NULL;
1133 	}
1134 	memset(elf, 0, sizeof(*elf));
1135 
1136 	INIT_LIST_HEAD(&elf->sections);
1137 
1138 	elf->fd = open(name, flags);
1139 	if (elf->fd == -1) {
1140 		fprintf(stderr, "objtool: Can't open '%s': %s\n",
1141 			name, strerror(errno));
1142 		goto err;
1143 	}
1144 
1145 	elf->name = strdup(name);
1146 	if (!elf->name) {
1147 		ERROR_GLIBC("strdup");
1148 		return NULL;
1149 	}
1150 
1151 	if ((flags & O_ACCMODE) == O_RDONLY)
1152 		cmd = ELF_C_READ_MMAP;
1153 	else if ((flags & O_ACCMODE) == O_RDWR)
1154 		cmd = ELF_C_RDWR;
1155 	else /* O_WRONLY */
1156 		cmd = ELF_C_WRITE;
1157 
1158 	elf->elf = elf_begin(elf->fd, cmd, NULL);
1159 	if (!elf->elf) {
1160 		ERROR_ELF("elf_begin");
1161 		goto err;
1162 	}
1163 
1164 	if (!gelf_getehdr(elf->elf, &elf->ehdr)) {
1165 		ERROR_ELF("gelf_getehdr");
1166 		goto err;
1167 	}
1168 
1169 	if (read_sections(elf))
1170 		goto err;
1171 
1172 	if (read_symbols(elf))
1173 		goto err;
1174 
1175 	if (mark_group_syms(elf))
1176 		goto err;
1177 
1178 	if (read_relocs(elf))
1179 		goto err;
1180 
1181 	return elf;
1182 
1183 err:
1184 	elf_close(elf);
1185 	return NULL;
1186 }
1187 
elf_create_file(GElf_Ehdr * ehdr,const char * name)1188 struct elf *elf_create_file(GElf_Ehdr *ehdr, const char *name)
1189 {
1190 	struct section *null, *symtab, *strtab, *shstrtab;
1191 	char *tmp_name;
1192 	struct symbol *sym;
1193 	struct elf *elf;
1194 
1195 	elf_version(EV_CURRENT);
1196 
1197 	elf = calloc(1, sizeof(*elf));
1198 	if (!elf) {
1199 		ERROR_GLIBC("calloc");
1200 		return NULL;
1201 	}
1202 
1203 	INIT_LIST_HEAD(&elf->sections);
1204 
1205 	tmp_name = malloc(strlen(name) + 8);
1206 	if (!tmp_name) {
1207 		ERROR_GLIBC("malloc");
1208 		return NULL;
1209 	}
1210 
1211 	sprintf(tmp_name, "%s.XXXXXX", name);
1212 
1213 	elf->fd = mkstemp(tmp_name);
1214 	if (elf->fd == -1) {
1215 		ERROR_GLIBC("can't create tmp file");
1216 		exit(1);
1217 	}
1218 
1219 	elf->tmp_name = tmp_name;
1220 
1221 	elf->name = strdup(name);
1222 	if (!elf->name) {
1223 		ERROR_GLIBC("strdup");
1224 		return NULL;
1225 	}
1226 
1227 	elf->elf = elf_begin(elf->fd, ELF_C_WRITE, NULL);
1228 	if (!elf->elf) {
1229 		ERROR_ELF("elf_begin");
1230 		return NULL;
1231 	}
1232 
1233 	if (!gelf_newehdr(elf->elf, ELFCLASS64)) {
1234 		ERROR_ELF("gelf_newehdr");
1235 		return NULL;
1236 	}
1237 
1238 	memcpy(&elf->ehdr, ehdr, sizeof(elf->ehdr));
1239 
1240 	if (!gelf_update_ehdr(elf->elf, &elf->ehdr)) {
1241 		ERROR_ELF("gelf_update_ehdr");
1242 		return NULL;
1243 	}
1244 
1245 	INIT_LIST_HEAD(&elf->symbols);
1246 
1247 	if (!elf_alloc_hash(section,		1000) ||
1248 	    !elf_alloc_hash(section_name,	1000) ||
1249 	    !elf_alloc_hash(symbol,		10000) ||
1250 	    !elf_alloc_hash(symbol_name,	10000) ||
1251 	    !elf_alloc_hash(reloc,		100000))
1252 		return NULL;
1253 
1254 	null		= elf_create_section(elf, NULL, 0, 0, SHT_NULL, 0, 0);
1255 	shstrtab	= elf_create_section(elf, NULL, 0, 0, SHT_STRTAB, 1, 0);
1256 	strtab		= elf_create_section(elf, NULL, 0, 0, SHT_STRTAB, 1, 0);
1257 
1258 	if (!null || !shstrtab || !strtab)
1259 		return NULL;
1260 
1261 	null->name	= "";
1262 	shstrtab->name	= ".shstrtab";
1263 	strtab->name	= ".strtab";
1264 
1265 	null->sh.sh_name	= elf_add_string(elf, shstrtab, null->name);
1266 	shstrtab->sh.sh_name	= elf_add_string(elf, shstrtab, shstrtab->name);
1267 	strtab->sh.sh_name	= elf_add_string(elf, shstrtab, strtab->name);
1268 
1269 	if (null->sh.sh_name == -1 || shstrtab->sh.sh_name == -1 || strtab->sh.sh_name == -1)
1270 		return NULL;
1271 
1272 	elf_hash_add(section_name, &null->name_hash,		str_hash(null->name));
1273 	elf_hash_add(section_name, &strtab->name_hash,		str_hash(strtab->name));
1274 	elf_hash_add(section_name, &shstrtab->name_hash,	str_hash(shstrtab->name));
1275 
1276 	if (elf_add_string(elf, strtab, "") == -1)
1277 		return NULL;
1278 
1279 	symtab = elf_create_section(elf, ".symtab", 0x18, 0x18, SHT_SYMTAB, 0x8, 0);
1280 	if (!symtab)
1281 		return NULL;
1282 
1283 	symtab->sh.sh_link = strtab->idx;
1284 	symtab->sh.sh_info = 1;
1285 
1286 	elf->ehdr.e_shstrndx = shstrtab->idx;
1287 	if (!gelf_update_ehdr(elf->elf, &elf->ehdr)) {
1288 		ERROR_ELF("gelf_update_ehdr");
1289 		return NULL;
1290 	}
1291 
1292 	sym = calloc(1, sizeof(*sym));
1293 	if (!sym) {
1294 		ERROR_GLIBC("calloc");
1295 		return NULL;
1296 	}
1297 
1298 	sym->name = "";
1299 	sym->sec = null;
1300 	elf_add_symbol(elf, sym);
1301 
1302 	return elf;
1303 }
1304 
elf_add_string(struct elf * elf,struct section * strtab,const char * str)1305 unsigned int elf_add_string(struct elf *elf, struct section *strtab, const char *str)
1306 {
1307 	unsigned int offset;
1308 
1309 	if (!strtab)
1310 		strtab = find_section_by_name(elf, ".strtab");
1311 	if (!strtab) {
1312 		ERROR("can't find .strtab section");
1313 		return -1;
1314 	}
1315 
1316 	if (!strtab->sh.sh_addralign) {
1317 		ERROR("'%s': invalid sh_addralign", strtab->name);
1318 		return -1;
1319 	}
1320 
1321 	offset = ALIGN(strtab->sh.sh_size, strtab->sh.sh_addralign);
1322 
1323 	if (!elf_add_data(elf, strtab, str, strlen(str) + 1))
1324 		return -1;
1325 
1326 	return offset;
1327 }
1328 
elf_add_data(struct elf * elf,struct section * sec,const void * data,size_t size)1329 void *elf_add_data(struct elf *elf, struct section *sec, const void *data, size_t size)
1330 {
1331 	unsigned long offset;
1332 	Elf_Scn *s;
1333 
1334 	if (!sec->sh.sh_addralign) {
1335 		ERROR("'%s': invalid sh_addralign", sec->name);
1336 		return NULL;
1337 	}
1338 
1339 	s = elf_getscn(elf->elf, sec->idx);
1340 	if (!s) {
1341 		ERROR_ELF("elf_getscn");
1342 		return NULL;
1343 	}
1344 
1345 	sec->data = elf_newdata(s);
1346 	if (!sec->data) {
1347 		ERROR_ELF("elf_newdata");
1348 		return NULL;
1349 	}
1350 
1351 	sec->data->d_buf = calloc(1, size);
1352 	if (!sec->data->d_buf) {
1353 		ERROR_GLIBC("calloc");
1354 		return NULL;
1355 	}
1356 
1357 	if (data)
1358 		memcpy(sec->data->d_buf, data, size);
1359 
1360 	sec->data->d_size = size;
1361 	sec->data->d_align = sec->sh.sh_addralign;
1362 
1363 	offset = ALIGN(sec->sh.sh_size, sec->sh.sh_addralign);
1364 	sec->sh.sh_size = offset + size;
1365 
1366 	mark_sec_changed(elf, sec, true);
1367 
1368 	return sec->data->d_buf;
1369 }
1370 
elf_create_section(struct elf * elf,const char * name,size_t size,size_t entsize,unsigned int type,unsigned int align,unsigned int flags)1371 struct section *elf_create_section(struct elf *elf, const char *name,
1372 				   size_t size, size_t entsize,
1373 				   unsigned int type, unsigned int align,
1374 				   unsigned int flags)
1375 {
1376 	struct section *sec, *shstrtab;
1377 	Elf_Scn *s;
1378 
1379 	if (name && find_section_by_name(elf, name)) {
1380 		ERROR("section '%s' already exists", name);
1381 		return NULL;
1382 	}
1383 
1384 	sec = calloc(1, sizeof(*sec));
1385 	if (!sec) {
1386 		ERROR_GLIBC("calloc");
1387 		return NULL;
1388 	}
1389 
1390 	INIT_LIST_HEAD(&sec->symbol_list);
1391 
1392 	/* don't actually create the section, just the data structures */
1393 	if (type == SHT_NULL)
1394 		goto add;
1395 
1396 	s = elf_newscn(elf->elf);
1397 	if (!s) {
1398 		ERROR_ELF("elf_newscn");
1399 		return NULL;
1400 	}
1401 
1402 	sec->idx = elf_ndxscn(s);
1403 
1404 	if (size) {
1405 		sec->data = elf_newdata(s);
1406 		if (!sec->data) {
1407 			ERROR_ELF("elf_newdata");
1408 			return NULL;
1409 		}
1410 
1411 		sec->data->d_size = size;
1412 		sec->data->d_align = 1;
1413 
1414 		sec->data->d_buf = calloc(1, size);
1415 		if (!sec->data->d_buf) {
1416 			ERROR_GLIBC("calloc");
1417 			return NULL;
1418 		}
1419 	}
1420 
1421 	if (!gelf_getshdr(s, &sec->sh)) {
1422 		ERROR_ELF("gelf_getshdr");
1423 		return NULL;
1424 	}
1425 
1426 	sec->sh.sh_size = size;
1427 	sec->sh.sh_entsize = entsize;
1428 	sec->sh.sh_type = type;
1429 	sec->sh.sh_addralign = align;
1430 	sec->sh.sh_flags = flags;
1431 
1432 	if (name) {
1433 		sec->name = strdup(name);
1434 		if (!sec->name) {
1435 			ERROR("strdup");
1436 			return NULL;
1437 		}
1438 
1439 		/* Add section name to .shstrtab (or .strtab for Clang) */
1440 		shstrtab = find_section_by_name(elf, ".shstrtab");
1441 		if (!shstrtab) {
1442 			shstrtab = find_section_by_name(elf, ".strtab");
1443 			if (!shstrtab) {
1444 				ERROR("can't find .shstrtab or .strtab");
1445 				return NULL;
1446 			}
1447 		}
1448 		sec->sh.sh_name = elf_add_string(elf, shstrtab, sec->name);
1449 		if (sec->sh.sh_name == -1)
1450 			return NULL;
1451 
1452 		elf_hash_add(section_name, &sec->name_hash, str_hash(sec->name));
1453 	}
1454 
1455 add:
1456 	list_add_tail(&sec->list, &elf->sections);
1457 	elf_hash_add(section, &sec->hash, sec->idx);
1458 
1459 	mark_sec_changed(elf, sec, true);
1460 
1461 	return sec;
1462 }
1463 
elf_alloc_reloc(struct elf * elf,struct section * rsec)1464 static int elf_alloc_reloc(struct elf *elf, struct section *rsec)
1465 {
1466 	struct reloc *old_relocs, *old_relocs_end, *new_relocs;
1467 	unsigned int nr_relocs_old = sec_num_entries(rsec);
1468 	unsigned int nr_relocs_new = nr_relocs_old + 1;
1469 	unsigned long nr_alloc;
1470 	struct symbol *sym;
1471 
1472 	if (!rsec->data) {
1473 		rsec->data = elf_newdata(elf_getscn(elf->elf, rsec->idx));
1474 		if (!rsec->data) {
1475 			ERROR_ELF("elf_newdata");
1476 			return -1;
1477 		}
1478 
1479 		rsec->data->d_align = 1;
1480 		rsec->data->d_type = ELF_T_RELA;
1481 		rsec->data->d_buf = NULL;
1482 	}
1483 
1484 	rsec->data->d_size = nr_relocs_new * elf_rela_size(elf);
1485 	rsec->sh.sh_size   = rsec->data->d_size;
1486 
1487 	nr_alloc = max(64UL, roundup_pow_of_two(nr_relocs_new));
1488 	if (nr_alloc <= rsec->nr_alloc_relocs)
1489 		return 0;
1490 
1491 	if (rsec->data->d_buf && !rsec->nr_alloc_relocs) {
1492 		void *orig_buf = rsec->data->d_buf;
1493 
1494 		/*
1495 		 * The original d_buf is owned by libelf so it can't be
1496 		 * realloced.
1497 		 */
1498 		rsec->data->d_buf = malloc(nr_alloc * elf_rela_size(elf));
1499 		if (!rsec->data->d_buf) {
1500 			ERROR_GLIBC("malloc");
1501 			return -1;
1502 		}
1503 		memcpy(rsec->data->d_buf, orig_buf,
1504 		       nr_relocs_old * elf_rela_size(elf));
1505 	} else {
1506 		rsec->data->d_buf = realloc(rsec->data->d_buf,
1507 					    nr_alloc * elf_rela_size(elf));
1508 		if (!rsec->data->d_buf) {
1509 			ERROR_GLIBC("realloc");
1510 			return -1;
1511 		}
1512 	}
1513 
1514 	rsec->nr_alloc_relocs = nr_alloc;
1515 
1516 	old_relocs = rsec->relocs;
1517 	new_relocs = calloc(nr_alloc, sizeof(struct reloc));
1518 	if (!new_relocs) {
1519 		ERROR_GLIBC("calloc");
1520 		return -1;
1521 	}
1522 
1523 	if (!old_relocs)
1524 		goto done;
1525 
1526 	/*
1527 	 * The struct reloc's address has changed.  Update all the symbols and
1528 	 * relocs which reference it.
1529 	 */
1530 
1531 	old_relocs_end = &old_relocs[nr_relocs_old];
1532 	for_each_sym(elf, sym) {
1533 		struct reloc *reloc;
1534 
1535 		reloc = sym->relocs;
1536 		if (!reloc)
1537 			continue;
1538 
1539 		if (reloc >= old_relocs && reloc < old_relocs_end)
1540 			sym->relocs = &new_relocs[reloc - old_relocs];
1541 
1542 		while (1) {
1543 			struct reloc *next_reloc = sym_next_reloc(reloc);
1544 
1545 			if (!next_reloc)
1546 				break;
1547 
1548 			if (next_reloc >= old_relocs && next_reloc < old_relocs_end)
1549 				set_sym_next_reloc(reloc, &new_relocs[next_reloc - old_relocs]);
1550 
1551 			reloc = next_reloc;
1552 		}
1553 	}
1554 
1555 	memcpy(new_relocs, old_relocs, nr_relocs_old * sizeof(struct reloc));
1556 
1557 	for (int i = 0; i < nr_relocs_old; i++) {
1558 		struct reloc *old = &old_relocs[i];
1559 		struct reloc *new = &new_relocs[i];
1560 		u32 key = reloc_hash(old);
1561 
1562 		elf_hash_del(reloc, &old->hash, key);
1563 		elf_hash_add(reloc, &new->hash, key);
1564 	}
1565 
1566 	free(old_relocs);
1567 done:
1568 	rsec->relocs = new_relocs;
1569 	return 0;
1570 }
1571 
elf_create_rela_section(struct elf * elf,struct section * sec,unsigned int nr_relocs)1572 struct section *elf_create_rela_section(struct elf *elf, struct section *sec,
1573 					unsigned int nr_relocs)
1574 {
1575 	struct section *rsec;
1576 	char *rsec_name;
1577 
1578 	rsec_name = malloc(strlen(sec->name) + strlen(".rela") + 1);
1579 	if (!rsec_name) {
1580 		ERROR_GLIBC("malloc");
1581 		return NULL;
1582 	}
1583 	strcpy(rsec_name, ".rela");
1584 	strcat(rsec_name, sec->name);
1585 
1586 	rsec = elf_create_section(elf, rsec_name, nr_relocs * elf_rela_size(elf),
1587 				  elf_rela_size(elf), SHT_RELA, elf_addr_size(elf),
1588 				  SHF_INFO_LINK);
1589 	free(rsec_name);
1590 	if (!rsec)
1591 		return NULL;
1592 
1593 	if (nr_relocs) {
1594 		rsec->data->d_type = ELF_T_RELA;
1595 
1596 		rsec->nr_alloc_relocs = nr_relocs;
1597 		rsec->relocs = calloc(nr_relocs, sizeof(struct reloc));
1598 		if (!rsec->relocs) {
1599 			ERROR_GLIBC("calloc");
1600 			return NULL;
1601 		}
1602 	}
1603 
1604 	rsec->sh.sh_link = find_section_by_name(elf, ".symtab")->idx;
1605 	rsec->sh.sh_info = sec->idx;
1606 
1607 	sec->rsec = rsec;
1608 	rsec->base = sec;
1609 
1610 	return rsec;
1611 }
1612 
elf_create_reloc(struct elf * elf,struct section * sec,unsigned long offset,struct symbol * sym,s64 addend,unsigned int type)1613 struct reloc *elf_create_reloc(struct elf *elf, struct section *sec,
1614 			       unsigned long offset,
1615 			       struct symbol *sym, s64 addend,
1616 			       unsigned int type)
1617 {
1618 	struct section *rsec = sec->rsec;
1619 
1620 	if (!rsec) {
1621 		rsec = elf_create_rela_section(elf, sec, 0);
1622 		if (!rsec)
1623 			return NULL;
1624 	}
1625 
1626 	if (find_reloc_by_dest(elf, sec, offset)) {
1627 		ERROR_FUNC(sec, offset, "duplicate reloc");
1628 		return NULL;
1629 	}
1630 
1631 	if (elf_alloc_reloc(elf, rsec))
1632 		return NULL;
1633 
1634 	mark_sec_changed(elf, rsec, true);
1635 
1636 	return elf_init_reloc(elf, rsec, sec_num_entries(rsec) - 1, offset, sym,
1637 			      addend, type);
1638 }
1639 
elf_create_section_pair(struct elf * elf,const char * name,size_t entsize,unsigned int nr,unsigned int nr_relocs)1640 struct section *elf_create_section_pair(struct elf *elf, const char *name,
1641 					size_t entsize, unsigned int nr,
1642 					unsigned int nr_relocs)
1643 {
1644 	struct section *sec;
1645 
1646 	sec = elf_create_section(elf, name, nr * entsize, entsize,
1647 				 SHT_PROGBITS, 1, SHF_ALLOC);
1648 	if (!sec)
1649 		return NULL;
1650 
1651 	if (!elf_create_rela_section(elf, sec, nr_relocs))
1652 		return NULL;
1653 
1654 	return sec;
1655 }
1656 
elf_write_insn(struct elf * elf,struct section * sec,unsigned long offset,unsigned int len,const char * insn)1657 int elf_write_insn(struct elf *elf, struct section *sec,
1658 		   unsigned long offset, unsigned int len,
1659 		   const char *insn)
1660 {
1661 	Elf_Data *data = sec->data;
1662 
1663 	if (data->d_type != ELF_T_BYTE || data->d_off) {
1664 		ERROR("write to unexpected data for section: %s", sec->name);
1665 		return -1;
1666 	}
1667 
1668 	memcpy(data->d_buf + offset, insn, len);
1669 
1670 	mark_sec_changed(elf, sec, true);
1671 
1672 	return 0;
1673 }
1674 
1675 /*
1676  * When Elf_Scn::sh_size is smaller than the combined Elf_Data::d_size
1677  * do you:
1678  *
1679  *   A) adhere to the section header and truncate the data, or
1680  *   B) ignore the section header and write out all the data you've got?
1681  *
1682  * Yes, libelf sucks and we need to manually truncate if we over-allocate data.
1683  */
elf_truncate_section(struct elf * elf,struct section * sec)1684 static int elf_truncate_section(struct elf *elf, struct section *sec)
1685 {
1686 	u64 size = sec_size(sec);
1687 	bool truncated = false;
1688 	Elf_Data *data = NULL;
1689 	Elf_Scn *s;
1690 
1691 	s = elf_getscn(elf->elf, sec->idx);
1692 	if (!s) {
1693 		ERROR_ELF("elf_getscn");
1694 		return -1;
1695 	}
1696 
1697 	for (;;) {
1698 		/* get next data descriptor for the relevant section */
1699 		data = elf_getdata(s, data);
1700 		if (!data) {
1701 			if (size) {
1702 				ERROR("end of section data but non-zero size left\n");
1703 				return -1;
1704 			}
1705 			return 0;
1706 		}
1707 
1708 		if (truncated) {
1709 			/* when we remove symbols */
1710 			ERROR("truncated; but more data\n");
1711 			return -1;
1712 		}
1713 
1714 		if (!data->d_size) {
1715 			ERROR("zero size data");
1716 			return -1;
1717 		}
1718 
1719 		if (data->d_size > size) {
1720 			truncated = true;
1721 			data->d_size = size;
1722 		}
1723 
1724 		size -= data->d_size;
1725 	}
1726 }
1727 
elf_write(struct elf * elf)1728 int elf_write(struct elf *elf)
1729 {
1730 	struct section *sec;
1731 	Elf_Scn *s;
1732 
1733 	/* Update changed relocation sections and section headers: */
1734 	list_for_each_entry(sec, &elf->sections, list) {
1735 		if (sec->truncate && elf_truncate_section(elf, sec))
1736 			return -1;
1737 
1738 		if (sec_changed(sec)) {
1739 			s = elf_getscn(elf->elf, sec->idx);
1740 			if (!s) {
1741 				ERROR_ELF("elf_getscn");
1742 				return -1;
1743 			}
1744 
1745 			/* Note this also flags the section dirty */
1746 			if (!gelf_update_shdr(s, &sec->sh)) {
1747 				ERROR_ELF("gelf_update_shdr");
1748 				return -1;
1749 			}
1750 
1751 			mark_sec_changed(elf, sec, false);
1752 		}
1753 	}
1754 
1755 	/* Make sure the new section header entries get updated properly. */
1756 	elf_flagelf(elf->elf, ELF_C_SET, ELF_F_DIRTY);
1757 
1758 	/* Write all changes to the file. */
1759 	if (elf_update(elf->elf, ELF_C_WRITE) < 0) {
1760 		ERROR_ELF("elf_update");
1761 		return -1;
1762 	}
1763 
1764 	elf->changed = false;
1765 
1766 	return 0;
1767 }
1768 
elf_close(struct elf * elf)1769 int elf_close(struct elf *elf)
1770 {
1771 	if (elf->elf)
1772 		elf_end(elf->elf);
1773 
1774 	if (elf->fd > 0)
1775 		close(elf->fd);
1776 
1777 	if (elf->tmp_name && rename(elf->tmp_name, elf->name))
1778 		return -1;
1779 
1780 	/*
1781 	 * NOTE: All remaining allocations are leaked on purpose.  Objtool is
1782 	 * about to exit anyway.
1783 	 */
1784 	return 0;
1785 }
1786