1 #include <fcntl.h>
2 #include <stdio.h>
3 #include <errno.h>
4 #include <string.h>
5 #include <unistd.h>
6 #include <inttypes.h>
7 
8 #include "symbol.h"
9 #include "debug.h"
10 
11 #ifndef HAVE_ELF_GETPHDRNUM
elf_getphdrnum(Elf * elf,size_t * dst)12 static int elf_getphdrnum(Elf *elf, size_t *dst)
13 {
14 	GElf_Ehdr gehdr;
15 	GElf_Ehdr *ehdr;
16 
17 	ehdr = gelf_getehdr(elf, &gehdr);
18 	if (!ehdr)
19 		return -1;
20 
21 	*dst = ehdr->e_phnum;
22 
23 	return 0;
24 }
25 #endif
26 
27 #ifndef NT_GNU_BUILD_ID
28 #define NT_GNU_BUILD_ID 3
29 #endif
30 
31 /**
32  * elf_symtab__for_each_symbol - iterate thru all the symbols
33  *
34  * @syms: struct elf_symtab instance to iterate
35  * @idx: uint32_t idx
36  * @sym: GElf_Sym iterator
37  */
38 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
39 	for (idx = 0, gelf_getsym(syms, idx, &sym);\
40 	     idx < nr_syms; \
41 	     idx++, gelf_getsym(syms, idx, &sym))
42 
elf_sym__type(const GElf_Sym * sym)43 static inline uint8_t elf_sym__type(const GElf_Sym *sym)
44 {
45 	return GELF_ST_TYPE(sym->st_info);
46 }
47 
elf_sym__is_function(const GElf_Sym * sym)48 static inline int elf_sym__is_function(const GElf_Sym *sym)
49 {
50 	return elf_sym__type(sym) == STT_FUNC &&
51 	       sym->st_name != 0 &&
52 	       sym->st_shndx != SHN_UNDEF;
53 }
54 
elf_sym__is_object(const GElf_Sym * sym)55 static inline bool elf_sym__is_object(const GElf_Sym *sym)
56 {
57 	return elf_sym__type(sym) == STT_OBJECT &&
58 		sym->st_name != 0 &&
59 		sym->st_shndx != SHN_UNDEF;
60 }
61 
elf_sym__is_label(const GElf_Sym * sym)62 static inline int elf_sym__is_label(const GElf_Sym *sym)
63 {
64 	return elf_sym__type(sym) == STT_NOTYPE &&
65 		sym->st_name != 0 &&
66 		sym->st_shndx != SHN_UNDEF &&
67 		sym->st_shndx != SHN_ABS;
68 }
69 
elf_sym__is_a(GElf_Sym * sym,enum map_type type)70 static bool elf_sym__is_a(GElf_Sym *sym, enum map_type type)
71 {
72 	switch (type) {
73 	case MAP__FUNCTION:
74 		return elf_sym__is_function(sym);
75 	case MAP__VARIABLE:
76 		return elf_sym__is_object(sym);
77 	default:
78 		return false;
79 	}
80 }
81 
elf_sym__name(const GElf_Sym * sym,const Elf_Data * symstrs)82 static inline const char *elf_sym__name(const GElf_Sym *sym,
83 					const Elf_Data *symstrs)
84 {
85 	return symstrs->d_buf + sym->st_name;
86 }
87 
elf_sec__name(const GElf_Shdr * shdr,const Elf_Data * secstrs)88 static inline const char *elf_sec__name(const GElf_Shdr *shdr,
89 					const Elf_Data *secstrs)
90 {
91 	return secstrs->d_buf + shdr->sh_name;
92 }
93 
elf_sec__is_text(const GElf_Shdr * shdr,const Elf_Data * secstrs)94 static inline int elf_sec__is_text(const GElf_Shdr *shdr,
95 					const Elf_Data *secstrs)
96 {
97 	return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
98 }
99 
elf_sec__is_data(const GElf_Shdr * shdr,const Elf_Data * secstrs)100 static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
101 				    const Elf_Data *secstrs)
102 {
103 	return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
104 }
105 
elf_sec__is_a(GElf_Shdr * shdr,Elf_Data * secstrs,enum map_type type)106 static bool elf_sec__is_a(GElf_Shdr *shdr, Elf_Data *secstrs,
107 			  enum map_type type)
108 {
109 	switch (type) {
110 	case MAP__FUNCTION:
111 		return elf_sec__is_text(shdr, secstrs);
112 	case MAP__VARIABLE:
113 		return elf_sec__is_data(shdr, secstrs);
114 	default:
115 		return false;
116 	}
117 }
118 
elf_addr_to_index(Elf * elf,GElf_Addr addr)119 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
120 {
121 	Elf_Scn *sec = NULL;
122 	GElf_Shdr shdr;
123 	size_t cnt = 1;
124 
125 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
126 		gelf_getshdr(sec, &shdr);
127 
128 		if ((addr >= shdr.sh_addr) &&
129 		    (addr < (shdr.sh_addr + shdr.sh_size)))
130 			return cnt;
131 
132 		++cnt;
133 	}
134 
135 	return -1;
136 }
137 
elf_section_by_name(Elf * elf,GElf_Ehdr * ep,GElf_Shdr * shp,const char * name,size_t * idx)138 static Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
139 				    GElf_Shdr *shp, const char *name,
140 				    size_t *idx)
141 {
142 	Elf_Scn *sec = NULL;
143 	size_t cnt = 1;
144 
145 	/* Elf is corrupted/truncated, avoid calling elf_strptr. */
146 	if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL))
147 		return NULL;
148 
149 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
150 		char *str;
151 
152 		gelf_getshdr(sec, shp);
153 		str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
154 		if (!strcmp(name, str)) {
155 			if (idx)
156 				*idx = cnt;
157 			break;
158 		}
159 		++cnt;
160 	}
161 
162 	return sec;
163 }
164 
165 #define elf_section__for_each_rel(reldata, pos, pos_mem, idx, nr_entries) \
166 	for (idx = 0, pos = gelf_getrel(reldata, 0, &pos_mem); \
167 	     idx < nr_entries; \
168 	     ++idx, pos = gelf_getrel(reldata, idx, &pos_mem))
169 
170 #define elf_section__for_each_rela(reldata, pos, pos_mem, idx, nr_entries) \
171 	for (idx = 0, pos = gelf_getrela(reldata, 0, &pos_mem); \
172 	     idx < nr_entries; \
173 	     ++idx, pos = gelf_getrela(reldata, idx, &pos_mem))
174 
175 /*
176  * We need to check if we have a .dynsym, so that we can handle the
177  * .plt, synthesizing its symbols, that aren't on the symtabs (be it
178  * .dynsym or .symtab).
179  * And always look at the original dso, not at debuginfo packages, that
180  * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
181  */
dso__synthesize_plt_symbols(struct dso * dso,struct symsrc * ss,struct map * map,symbol_filter_t filter)182 int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss, struct map *map,
183 				symbol_filter_t filter)
184 {
185 	uint32_t nr_rel_entries, idx;
186 	GElf_Sym sym;
187 	u64 plt_offset;
188 	GElf_Shdr shdr_plt;
189 	struct symbol *f;
190 	GElf_Shdr shdr_rel_plt, shdr_dynsym;
191 	Elf_Data *reldata, *syms, *symstrs;
192 	Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
193 	size_t dynsym_idx;
194 	GElf_Ehdr ehdr;
195 	char sympltname[1024];
196 	Elf *elf;
197 	int nr = 0, symidx, err = 0;
198 
199 	if (!ss->dynsym)
200 		return 0;
201 
202 	elf = ss->elf;
203 	ehdr = ss->ehdr;
204 
205 	scn_dynsym = ss->dynsym;
206 	shdr_dynsym = ss->dynshdr;
207 	dynsym_idx = ss->dynsym_idx;
208 
209 	if (scn_dynsym == NULL)
210 		goto out_elf_end;
211 
212 	scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
213 					  ".rela.plt", NULL);
214 	if (scn_plt_rel == NULL) {
215 		scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
216 						  ".rel.plt", NULL);
217 		if (scn_plt_rel == NULL)
218 			goto out_elf_end;
219 	}
220 
221 	err = -1;
222 
223 	if (shdr_rel_plt.sh_link != dynsym_idx)
224 		goto out_elf_end;
225 
226 	if (elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL) == NULL)
227 		goto out_elf_end;
228 
229 	/*
230 	 * Fetch the relocation section to find the idxes to the GOT
231 	 * and the symbols in the .dynsym they refer to.
232 	 */
233 	reldata = elf_getdata(scn_plt_rel, NULL);
234 	if (reldata == NULL)
235 		goto out_elf_end;
236 
237 	syms = elf_getdata(scn_dynsym, NULL);
238 	if (syms == NULL)
239 		goto out_elf_end;
240 
241 	scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
242 	if (scn_symstrs == NULL)
243 		goto out_elf_end;
244 
245 	symstrs = elf_getdata(scn_symstrs, NULL);
246 	if (symstrs == NULL)
247 		goto out_elf_end;
248 
249 	if (symstrs->d_size == 0)
250 		goto out_elf_end;
251 
252 	nr_rel_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
253 	plt_offset = shdr_plt.sh_offset;
254 
255 	if (shdr_rel_plt.sh_type == SHT_RELA) {
256 		GElf_Rela pos_mem, *pos;
257 
258 		elf_section__for_each_rela(reldata, pos, pos_mem, idx,
259 					   nr_rel_entries) {
260 			symidx = GELF_R_SYM(pos->r_info);
261 			plt_offset += shdr_plt.sh_entsize;
262 			gelf_getsym(syms, symidx, &sym);
263 			snprintf(sympltname, sizeof(sympltname),
264 				 "%s@plt", elf_sym__name(&sym, symstrs));
265 
266 			f = symbol__new(plt_offset, shdr_plt.sh_entsize,
267 					STB_GLOBAL, sympltname);
268 			if (!f)
269 				goto out_elf_end;
270 
271 			if (filter && filter(map, f))
272 				symbol__delete(f);
273 			else {
274 				symbols__insert(&dso->symbols[map->type], f);
275 				++nr;
276 			}
277 		}
278 	} else if (shdr_rel_plt.sh_type == SHT_REL) {
279 		GElf_Rel pos_mem, *pos;
280 		elf_section__for_each_rel(reldata, pos, pos_mem, idx,
281 					  nr_rel_entries) {
282 			symidx = GELF_R_SYM(pos->r_info);
283 			plt_offset += shdr_plt.sh_entsize;
284 			gelf_getsym(syms, symidx, &sym);
285 			snprintf(sympltname, sizeof(sympltname),
286 				 "%s@plt", elf_sym__name(&sym, symstrs));
287 
288 			f = symbol__new(plt_offset, shdr_plt.sh_entsize,
289 					STB_GLOBAL, sympltname);
290 			if (!f)
291 				goto out_elf_end;
292 
293 			if (filter && filter(map, f))
294 				symbol__delete(f);
295 			else {
296 				symbols__insert(&dso->symbols[map->type], f);
297 				++nr;
298 			}
299 		}
300 	}
301 
302 	err = 0;
303 out_elf_end:
304 	if (err == 0)
305 		return nr;
306 	pr_debug("%s: problems reading %s PLT info.\n",
307 		 __func__, dso->long_name);
308 	return 0;
309 }
310 
311 /*
312  * Align offset to 4 bytes as needed for note name and descriptor data.
313  */
314 #define NOTE_ALIGN(n) (((n) + 3) & -4U)
315 
elf_read_build_id(Elf * elf,void * bf,size_t size)316 static int elf_read_build_id(Elf *elf, void *bf, size_t size)
317 {
318 	int err = -1;
319 	GElf_Ehdr ehdr;
320 	GElf_Shdr shdr;
321 	Elf_Data *data;
322 	Elf_Scn *sec;
323 	Elf_Kind ek;
324 	void *ptr;
325 
326 	if (size < BUILD_ID_SIZE)
327 		goto out;
328 
329 	ek = elf_kind(elf);
330 	if (ek != ELF_K_ELF)
331 		goto out;
332 
333 	if (gelf_getehdr(elf, &ehdr) == NULL) {
334 		pr_err("%s: cannot get elf header.\n", __func__);
335 		goto out;
336 	}
337 
338 	/*
339 	 * Check following sections for notes:
340 	 *   '.note.gnu.build-id'
341 	 *   '.notes'
342 	 *   '.note' (VDSO specific)
343 	 */
344 	do {
345 		sec = elf_section_by_name(elf, &ehdr, &shdr,
346 					  ".note.gnu.build-id", NULL);
347 		if (sec)
348 			break;
349 
350 		sec = elf_section_by_name(elf, &ehdr, &shdr,
351 					  ".notes", NULL);
352 		if (sec)
353 			break;
354 
355 		sec = elf_section_by_name(elf, &ehdr, &shdr,
356 					  ".note", NULL);
357 		if (sec)
358 			break;
359 
360 		return err;
361 
362 	} while (0);
363 
364 	data = elf_getdata(sec, NULL);
365 	if (data == NULL)
366 		goto out;
367 
368 	ptr = data->d_buf;
369 	while (ptr < (data->d_buf + data->d_size)) {
370 		GElf_Nhdr *nhdr = ptr;
371 		size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
372 		       descsz = NOTE_ALIGN(nhdr->n_descsz);
373 		const char *name;
374 
375 		ptr += sizeof(*nhdr);
376 		name = ptr;
377 		ptr += namesz;
378 		if (nhdr->n_type == NT_GNU_BUILD_ID &&
379 		    nhdr->n_namesz == sizeof("GNU")) {
380 			if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
381 				size_t sz = min(size, descsz);
382 				memcpy(bf, ptr, sz);
383 				memset(bf + sz, 0, size - sz);
384 				err = descsz;
385 				break;
386 			}
387 		}
388 		ptr += descsz;
389 	}
390 
391 out:
392 	return err;
393 }
394 
filename__read_build_id(const char * filename,void * bf,size_t size)395 int filename__read_build_id(const char *filename, void *bf, size_t size)
396 {
397 	int fd, err = -1;
398 	Elf *elf;
399 
400 	if (size < BUILD_ID_SIZE)
401 		goto out;
402 
403 	fd = open(filename, O_RDONLY);
404 	if (fd < 0)
405 		goto out;
406 
407 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
408 	if (elf == NULL) {
409 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
410 		goto out_close;
411 	}
412 
413 	err = elf_read_build_id(elf, bf, size);
414 
415 	elf_end(elf);
416 out_close:
417 	close(fd);
418 out:
419 	return err;
420 }
421 
sysfs__read_build_id(const char * filename,void * build_id,size_t size)422 int sysfs__read_build_id(const char *filename, void *build_id, size_t size)
423 {
424 	int fd, err = -1;
425 
426 	if (size < BUILD_ID_SIZE)
427 		goto out;
428 
429 	fd = open(filename, O_RDONLY);
430 	if (fd < 0)
431 		goto out;
432 
433 	while (1) {
434 		char bf[BUFSIZ];
435 		GElf_Nhdr nhdr;
436 		size_t namesz, descsz;
437 
438 		if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
439 			break;
440 
441 		namesz = NOTE_ALIGN(nhdr.n_namesz);
442 		descsz = NOTE_ALIGN(nhdr.n_descsz);
443 		if (nhdr.n_type == NT_GNU_BUILD_ID &&
444 		    nhdr.n_namesz == sizeof("GNU")) {
445 			if (read(fd, bf, namesz) != (ssize_t)namesz)
446 				break;
447 			if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
448 				size_t sz = min(descsz, size);
449 				if (read(fd, build_id, sz) == (ssize_t)sz) {
450 					memset(build_id + sz, 0, size - sz);
451 					err = 0;
452 					break;
453 				}
454 			} else if (read(fd, bf, descsz) != (ssize_t)descsz)
455 				break;
456 		} else {
457 			int n = namesz + descsz;
458 			if (read(fd, bf, n) != n)
459 				break;
460 		}
461 	}
462 	close(fd);
463 out:
464 	return err;
465 }
466 
filename__read_debuglink(const char * filename,char * debuglink,size_t size)467 int filename__read_debuglink(const char *filename, char *debuglink,
468 			     size_t size)
469 {
470 	int fd, err = -1;
471 	Elf *elf;
472 	GElf_Ehdr ehdr;
473 	GElf_Shdr shdr;
474 	Elf_Data *data;
475 	Elf_Scn *sec;
476 	Elf_Kind ek;
477 
478 	fd = open(filename, O_RDONLY);
479 	if (fd < 0)
480 		goto out;
481 
482 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
483 	if (elf == NULL) {
484 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
485 		goto out_close;
486 	}
487 
488 	ek = elf_kind(elf);
489 	if (ek != ELF_K_ELF)
490 		goto out_close;
491 
492 	if (gelf_getehdr(elf, &ehdr) == NULL) {
493 		pr_err("%s: cannot get elf header.\n", __func__);
494 		goto out_close;
495 	}
496 
497 	sec = elf_section_by_name(elf, &ehdr, &shdr,
498 				  ".gnu_debuglink", NULL);
499 	if (sec == NULL)
500 		goto out_close;
501 
502 	data = elf_getdata(sec, NULL);
503 	if (data == NULL)
504 		goto out_close;
505 
506 	/* the start of this section is a zero-terminated string */
507 	strncpy(debuglink, data->d_buf, size);
508 
509 	elf_end(elf);
510 
511 out_close:
512 	close(fd);
513 out:
514 	return err;
515 }
516 
dso__swap_init(struct dso * dso,unsigned char eidata)517 static int dso__swap_init(struct dso *dso, unsigned char eidata)
518 {
519 	static unsigned int const endian = 1;
520 
521 	dso->needs_swap = DSO_SWAP__NO;
522 
523 	switch (eidata) {
524 	case ELFDATA2LSB:
525 		/* We are big endian, DSO is little endian. */
526 		if (*(unsigned char const *)&endian != 1)
527 			dso->needs_swap = DSO_SWAP__YES;
528 		break;
529 
530 	case ELFDATA2MSB:
531 		/* We are little endian, DSO is big endian. */
532 		if (*(unsigned char const *)&endian != 0)
533 			dso->needs_swap = DSO_SWAP__YES;
534 		break;
535 
536 	default:
537 		pr_err("unrecognized DSO data encoding %d\n", eidata);
538 		return -EINVAL;
539 	}
540 
541 	return 0;
542 }
543 
symsrc__possibly_runtime(struct symsrc * ss)544 bool symsrc__possibly_runtime(struct symsrc *ss)
545 {
546 	return ss->dynsym || ss->opdsec;
547 }
548 
symsrc__has_symtab(struct symsrc * ss)549 bool symsrc__has_symtab(struct symsrc *ss)
550 {
551 	return ss->symtab != NULL;
552 }
553 
symsrc__destroy(struct symsrc * ss)554 void symsrc__destroy(struct symsrc *ss)
555 {
556 	free(ss->name);
557 	elf_end(ss->elf);
558 	close(ss->fd);
559 }
560 
symsrc__init(struct symsrc * ss,struct dso * dso,const char * name,enum dso_binary_type type)561 int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
562 		 enum dso_binary_type type)
563 {
564 	int err = -1;
565 	GElf_Ehdr ehdr;
566 	Elf *elf;
567 	int fd;
568 
569 	fd = open(name, O_RDONLY);
570 	if (fd < 0)
571 		return -1;
572 
573 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
574 	if (elf == NULL) {
575 		pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
576 		goto out_close;
577 	}
578 
579 	if (gelf_getehdr(elf, &ehdr) == NULL) {
580 		pr_debug("%s: cannot get elf header.\n", __func__);
581 		goto out_elf_end;
582 	}
583 
584 	if (dso__swap_init(dso, ehdr.e_ident[EI_DATA]))
585 		goto out_elf_end;
586 
587 	/* Always reject images with a mismatched build-id: */
588 	if (dso->has_build_id) {
589 		u8 build_id[BUILD_ID_SIZE];
590 
591 		if (elf_read_build_id(elf, build_id, BUILD_ID_SIZE) < 0)
592 			goto out_elf_end;
593 
594 		if (!dso__build_id_equal(dso, build_id))
595 			goto out_elf_end;
596 	}
597 
598 	ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab",
599 			NULL);
600 	if (ss->symshdr.sh_type != SHT_SYMTAB)
601 		ss->symtab = NULL;
602 
603 	ss->dynsym_idx = 0;
604 	ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym",
605 			&ss->dynsym_idx);
606 	if (ss->dynshdr.sh_type != SHT_DYNSYM)
607 		ss->dynsym = NULL;
608 
609 	ss->opdidx = 0;
610 	ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd",
611 			&ss->opdidx);
612 	if (ss->opdshdr.sh_type != SHT_PROGBITS)
613 		ss->opdsec = NULL;
614 
615 	if (dso->kernel == DSO_TYPE_USER) {
616 		GElf_Shdr shdr;
617 		ss->adjust_symbols = (ehdr.e_type == ET_EXEC ||
618 				ehdr.e_type == ET_REL ||
619 				elf_section_by_name(elf, &ehdr, &shdr,
620 						     ".gnu.prelink_undo",
621 						     NULL) != NULL);
622 	} else {
623 		ss->adjust_symbols = ehdr.e_type == ET_EXEC ||
624 				     ehdr.e_type == ET_REL;
625 	}
626 
627 	ss->name   = strdup(name);
628 	if (!ss->name)
629 		goto out_elf_end;
630 
631 	ss->elf    = elf;
632 	ss->fd     = fd;
633 	ss->ehdr   = ehdr;
634 	ss->type   = type;
635 
636 	return 0;
637 
638 out_elf_end:
639 	elf_end(elf);
640 out_close:
641 	close(fd);
642 	return err;
643 }
644 
645 /**
646  * ref_reloc_sym_not_found - has kernel relocation symbol been found.
647  * @kmap: kernel maps and relocation reference symbol
648  *
649  * This function returns %true if we are dealing with the kernel maps and the
650  * relocation reference symbol has not yet been found.  Otherwise %false is
651  * returned.
652  */
ref_reloc_sym_not_found(struct kmap * kmap)653 static bool ref_reloc_sym_not_found(struct kmap *kmap)
654 {
655 	return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
656 	       !kmap->ref_reloc_sym->unrelocated_addr;
657 }
658 
659 /**
660  * ref_reloc - kernel relocation offset.
661  * @kmap: kernel maps and relocation reference symbol
662  *
663  * This function returns the offset of kernel addresses as determined by using
664  * the relocation reference symbol i.e. if the kernel has not been relocated
665  * then the return value is zero.
666  */
ref_reloc(struct kmap * kmap)667 static u64 ref_reloc(struct kmap *kmap)
668 {
669 	if (kmap && kmap->ref_reloc_sym &&
670 	    kmap->ref_reloc_sym->unrelocated_addr)
671 		return kmap->ref_reloc_sym->addr -
672 		       kmap->ref_reloc_sym->unrelocated_addr;
673 	return 0;
674 }
675 
dso__load_sym(struct dso * dso,struct map * map,struct symsrc * syms_ss,struct symsrc * runtime_ss,symbol_filter_t filter,int kmodule)676 int dso__load_sym(struct dso *dso, struct map *map,
677 		  struct symsrc *syms_ss, struct symsrc *runtime_ss,
678 		  symbol_filter_t filter, int kmodule)
679 {
680 	struct kmap *kmap = dso->kernel ? map__kmap(map) : NULL;
681 	struct map *curr_map = map;
682 	struct dso *curr_dso = dso;
683 	Elf_Data *symstrs, *secstrs;
684 	uint32_t nr_syms;
685 	int err = -1;
686 	uint32_t idx;
687 	GElf_Ehdr ehdr;
688 	GElf_Shdr shdr;
689 	Elf_Data *syms, *opddata = NULL;
690 	GElf_Sym sym;
691 	Elf_Scn *sec, *sec_strndx;
692 	Elf *elf;
693 	int nr = 0;
694 	bool remap_kernel = false, adjust_kernel_syms = false;
695 
696 	dso->symtab_type = syms_ss->type;
697 	dso->rel = syms_ss->ehdr.e_type == ET_REL;
698 
699 	/*
700 	 * Modules may already have symbols from kallsyms, but those symbols
701 	 * have the wrong values for the dso maps, so remove them.
702 	 */
703 	if (kmodule && syms_ss->symtab)
704 		symbols__delete(&dso->symbols[map->type]);
705 
706 	if (!syms_ss->symtab) {
707 		syms_ss->symtab  = syms_ss->dynsym;
708 		syms_ss->symshdr = syms_ss->dynshdr;
709 	}
710 
711 	elf = syms_ss->elf;
712 	ehdr = syms_ss->ehdr;
713 	sec = syms_ss->symtab;
714 	shdr = syms_ss->symshdr;
715 
716 	if (runtime_ss->opdsec)
717 		opddata = elf_rawdata(runtime_ss->opdsec, NULL);
718 
719 	syms = elf_getdata(sec, NULL);
720 	if (syms == NULL)
721 		goto out_elf_end;
722 
723 	sec = elf_getscn(elf, shdr.sh_link);
724 	if (sec == NULL)
725 		goto out_elf_end;
726 
727 	symstrs = elf_getdata(sec, NULL);
728 	if (symstrs == NULL)
729 		goto out_elf_end;
730 
731 	sec_strndx = elf_getscn(elf, ehdr.e_shstrndx);
732 	if (sec_strndx == NULL)
733 		goto out_elf_end;
734 
735 	secstrs = elf_getdata(sec_strndx, NULL);
736 	if (secstrs == NULL)
737 		goto out_elf_end;
738 
739 	nr_syms = shdr.sh_size / shdr.sh_entsize;
740 
741 	memset(&sym, 0, sizeof(sym));
742 
743 	/*
744 	 * The kernel relocation symbol is needed in advance in order to adjust
745 	 * kernel maps correctly.
746 	 */
747 	if (ref_reloc_sym_not_found(kmap)) {
748 		elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
749 			const char *elf_name = elf_sym__name(&sym, symstrs);
750 
751 			if (strcmp(elf_name, kmap->ref_reloc_sym->name))
752 				continue;
753 			kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
754 			break;
755 		}
756 	}
757 
758 	dso->adjust_symbols = runtime_ss->adjust_symbols || ref_reloc(kmap);
759 	/*
760 	 * Initial kernel and module mappings do not map to the dso.  For
761 	 * function mappings, flag the fixups.
762 	 */
763 	if (map->type == MAP__FUNCTION && (dso->kernel || kmodule)) {
764 		remap_kernel = true;
765 		adjust_kernel_syms = dso->adjust_symbols;
766 	}
767 	elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
768 		struct symbol *f;
769 		const char *elf_name = elf_sym__name(&sym, symstrs);
770 		char *demangled = NULL;
771 		int is_label = elf_sym__is_label(&sym);
772 		const char *section_name;
773 		bool used_opd = false;
774 
775 		if (!is_label && !elf_sym__is_a(&sym, map->type))
776 			continue;
777 
778 		/* Reject ARM ELF "mapping symbols": these aren't unique and
779 		 * don't identify functions, so will confuse the profile
780 		 * output: */
781 		if (ehdr.e_machine == EM_ARM) {
782 			if (!strcmp(elf_name, "$a") ||
783 			    !strcmp(elf_name, "$d") ||
784 			    !strcmp(elf_name, "$t"))
785 				continue;
786 		}
787 
788 		if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) {
789 			u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr;
790 			u64 *opd = opddata->d_buf + offset;
791 			sym.st_value = DSO__SWAP(dso, u64, *opd);
792 			sym.st_shndx = elf_addr_to_index(runtime_ss->elf,
793 					sym.st_value);
794 			used_opd = true;
795 		}
796 		/*
797 		 * When loading symbols in a data mapping, ABS symbols (which
798 		 * has a value of SHN_ABS in its st_shndx) failed at
799 		 * elf_getscn().  And it marks the loading as a failure so
800 		 * already loaded symbols cannot be fixed up.
801 		 *
802 		 * I'm not sure what should be done. Just ignore them for now.
803 		 * - Namhyung Kim
804 		 */
805 		if (sym.st_shndx == SHN_ABS)
806 			continue;
807 
808 		sec = elf_getscn(runtime_ss->elf, sym.st_shndx);
809 		if (!sec)
810 			goto out_elf_end;
811 
812 		gelf_getshdr(sec, &shdr);
813 
814 		if (is_label && !elf_sec__is_a(&shdr, secstrs, map->type))
815 			continue;
816 
817 		section_name = elf_sec__name(&shdr, secstrs);
818 
819 		/* On ARM, symbols for thumb functions have 1 added to
820 		 * the symbol address as a flag - remove it */
821 		if ((ehdr.e_machine == EM_ARM) &&
822 		    (map->type == MAP__FUNCTION) &&
823 		    (sym.st_value & 1))
824 			--sym.st_value;
825 
826 		if (dso->kernel || kmodule) {
827 			char dso_name[PATH_MAX];
828 
829 			/* Adjust symbol to map to file offset */
830 			if (adjust_kernel_syms)
831 				sym.st_value -= shdr.sh_addr - shdr.sh_offset;
832 
833 			if (strcmp(section_name,
834 				   (curr_dso->short_name +
835 				    dso->short_name_len)) == 0)
836 				goto new_symbol;
837 
838 			if (strcmp(section_name, ".text") == 0) {
839 				/*
840 				 * The initial kernel mapping is based on
841 				 * kallsyms and identity maps.  Overwrite it to
842 				 * map to the kernel dso.
843 				 */
844 				if (remap_kernel && dso->kernel) {
845 					remap_kernel = false;
846 					map->start = shdr.sh_addr +
847 						     ref_reloc(kmap);
848 					map->end = map->start + shdr.sh_size;
849 					map->pgoff = shdr.sh_offset;
850 					map->map_ip = map__map_ip;
851 					map->unmap_ip = map__unmap_ip;
852 					/* Ensure maps are correctly ordered */
853 					map_groups__remove(kmap->kmaps, map);
854 					map_groups__insert(kmap->kmaps, map);
855 				}
856 
857 				/*
858 				 * The initial module mapping is based on
859 				 * /proc/modules mapped to offset zero.
860 				 * Overwrite it to map to the module dso.
861 				 */
862 				if (remap_kernel && kmodule) {
863 					remap_kernel = false;
864 					map->pgoff = shdr.sh_offset;
865 				}
866 
867 				curr_map = map;
868 				curr_dso = dso;
869 				goto new_symbol;
870 			}
871 
872 			if (!kmap)
873 				goto new_symbol;
874 
875 			snprintf(dso_name, sizeof(dso_name),
876 				 "%s%s", dso->short_name, section_name);
877 
878 			curr_map = map_groups__find_by_name(kmap->kmaps, map->type, dso_name);
879 			if (curr_map == NULL) {
880 				u64 start = sym.st_value;
881 
882 				if (kmodule)
883 					start += map->start + shdr.sh_offset;
884 
885 				curr_dso = dso__new(dso_name);
886 				if (curr_dso == NULL)
887 					goto out_elf_end;
888 				curr_dso->kernel = dso->kernel;
889 				curr_dso->long_name = dso->long_name;
890 				curr_dso->long_name_len = dso->long_name_len;
891 				curr_map = map__new2(start, curr_dso,
892 						     map->type);
893 				if (curr_map == NULL) {
894 					dso__delete(curr_dso);
895 					goto out_elf_end;
896 				}
897 				if (adjust_kernel_syms) {
898 					curr_map->start = shdr.sh_addr +
899 							  ref_reloc(kmap);
900 					curr_map->end = curr_map->start +
901 							shdr.sh_size;
902 					curr_map->pgoff = shdr.sh_offset;
903 				} else {
904 					curr_map->map_ip = identity__map_ip;
905 					curr_map->unmap_ip = identity__map_ip;
906 				}
907 				curr_dso->symtab_type = dso->symtab_type;
908 				map_groups__insert(kmap->kmaps, curr_map);
909 				dsos__add(&dso->node, curr_dso);
910 				dso__set_loaded(curr_dso, map->type);
911 			} else
912 				curr_dso = curr_map->dso;
913 
914 			goto new_symbol;
915 		}
916 
917 		if ((used_opd && runtime_ss->adjust_symbols)
918 				|| (!used_opd && syms_ss->adjust_symbols)) {
919 			pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
920 				  "sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n", __func__,
921 				  (u64)sym.st_value, (u64)shdr.sh_addr,
922 				  (u64)shdr.sh_offset);
923 			sym.st_value -= shdr.sh_addr - shdr.sh_offset;
924 		}
925 		/*
926 		 * We need to figure out if the object was created from C++ sources
927 		 * DWARF DW_compile_unit has this, but we don't always have access
928 		 * to it...
929 		 */
930 		if (symbol_conf.demangle) {
931 			demangled = bfd_demangle(NULL, elf_name,
932 						 DMGL_PARAMS | DMGL_ANSI);
933 			if (demangled != NULL)
934 				elf_name = demangled;
935 		}
936 new_symbol:
937 		f = symbol__new(sym.st_value, sym.st_size,
938 				GELF_ST_BIND(sym.st_info), elf_name);
939 		free(demangled);
940 		if (!f)
941 			goto out_elf_end;
942 
943 		if (filter && filter(curr_map, f))
944 			symbol__delete(f);
945 		else {
946 			symbols__insert(&curr_dso->symbols[curr_map->type], f);
947 			nr++;
948 		}
949 	}
950 
951 	/*
952 	 * For misannotated, zeroed, ASM function sizes.
953 	 */
954 	if (nr > 0) {
955 		symbols__fixup_duplicate(&dso->symbols[map->type]);
956 		symbols__fixup_end(&dso->symbols[map->type]);
957 		if (kmap) {
958 			/*
959 			 * We need to fixup this here too because we create new
960 			 * maps here, for things like vsyscall sections.
961 			 */
962 			__map_groups__fixup_end(kmap->kmaps, map->type);
963 		}
964 	}
965 	err = nr;
966 out_elf_end:
967 	return err;
968 }
969 
elf_read_maps(Elf * elf,bool exe,mapfn_t mapfn,void * data)970 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
971 {
972 	GElf_Phdr phdr;
973 	size_t i, phdrnum;
974 	int err;
975 	u64 sz;
976 
977 	if (elf_getphdrnum(elf, &phdrnum))
978 		return -1;
979 
980 	for (i = 0; i < phdrnum; i++) {
981 		if (gelf_getphdr(elf, i, &phdr) == NULL)
982 			return -1;
983 		if (phdr.p_type != PT_LOAD)
984 			continue;
985 		if (exe) {
986 			if (!(phdr.p_flags & PF_X))
987 				continue;
988 		} else {
989 			if (!(phdr.p_flags & PF_R))
990 				continue;
991 		}
992 		sz = min(phdr.p_memsz, phdr.p_filesz);
993 		if (!sz)
994 			continue;
995 		err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
996 		if (err)
997 			return err;
998 	}
999 	return 0;
1000 }
1001 
file__read_maps(int fd,bool exe,mapfn_t mapfn,void * data,bool * is_64_bit)1002 int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data,
1003 		    bool *is_64_bit)
1004 {
1005 	int err;
1006 	Elf *elf;
1007 
1008 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1009 	if (elf == NULL)
1010 		return -1;
1011 
1012 	if (is_64_bit)
1013 		*is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1014 
1015 	err = elf_read_maps(elf, exe, mapfn, data);
1016 
1017 	elf_end(elf);
1018 	return err;
1019 }
1020 
symbol__elf_init(void)1021 void symbol__elf_init(void)
1022 {
1023 	elf_version(EV_CURRENT);
1024 }
1025