xref: /openbmc/linux/tools/perf/util/symbol-elf.c (revision 9cfc5c90)
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 "machine.h"
10 #include "vdso.h"
11 #include <symbol/kallsyms.h>
12 #include "debug.h"
13 
14 #ifndef EM_AARCH64
15 #define EM_AARCH64	183  /* ARM 64 bit */
16 #endif
17 
18 
19 #ifdef HAVE_CPLUS_DEMANGLE_SUPPORT
20 extern char *cplus_demangle(const char *, int);
21 
22 static inline char *bfd_demangle(void __maybe_unused *v, const char *c, int i)
23 {
24 	return cplus_demangle(c, i);
25 }
26 #else
27 #ifdef NO_DEMANGLE
28 static inline char *bfd_demangle(void __maybe_unused *v,
29 				 const char __maybe_unused *c,
30 				 int __maybe_unused i)
31 {
32 	return NULL;
33 }
34 #else
35 #define PACKAGE 'perf'
36 #include <bfd.h>
37 #endif
38 #endif
39 
40 #ifndef HAVE_ELF_GETPHDRNUM_SUPPORT
41 static int elf_getphdrnum(Elf *elf, size_t *dst)
42 {
43 	GElf_Ehdr gehdr;
44 	GElf_Ehdr *ehdr;
45 
46 	ehdr = gelf_getehdr(elf, &gehdr);
47 	if (!ehdr)
48 		return -1;
49 
50 	*dst = ehdr->e_phnum;
51 
52 	return 0;
53 }
54 #endif
55 
56 #ifndef NT_GNU_BUILD_ID
57 #define NT_GNU_BUILD_ID 3
58 #endif
59 
60 /**
61  * elf_symtab__for_each_symbol - iterate thru all the symbols
62  *
63  * @syms: struct elf_symtab instance to iterate
64  * @idx: uint32_t idx
65  * @sym: GElf_Sym iterator
66  */
67 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
68 	for (idx = 0, gelf_getsym(syms, idx, &sym);\
69 	     idx < nr_syms; \
70 	     idx++, gelf_getsym(syms, idx, &sym))
71 
72 static inline uint8_t elf_sym__type(const GElf_Sym *sym)
73 {
74 	return GELF_ST_TYPE(sym->st_info);
75 }
76 
77 #ifndef STT_GNU_IFUNC
78 #define STT_GNU_IFUNC 10
79 #endif
80 
81 static inline int elf_sym__is_function(const GElf_Sym *sym)
82 {
83 	return (elf_sym__type(sym) == STT_FUNC ||
84 		elf_sym__type(sym) == STT_GNU_IFUNC) &&
85 	       sym->st_name != 0 &&
86 	       sym->st_shndx != SHN_UNDEF;
87 }
88 
89 static inline bool elf_sym__is_object(const GElf_Sym *sym)
90 {
91 	return elf_sym__type(sym) == STT_OBJECT &&
92 		sym->st_name != 0 &&
93 		sym->st_shndx != SHN_UNDEF;
94 }
95 
96 static inline int elf_sym__is_label(const GElf_Sym *sym)
97 {
98 	return elf_sym__type(sym) == STT_NOTYPE &&
99 		sym->st_name != 0 &&
100 		sym->st_shndx != SHN_UNDEF &&
101 		sym->st_shndx != SHN_ABS;
102 }
103 
104 static bool elf_sym__is_a(GElf_Sym *sym, enum map_type type)
105 {
106 	switch (type) {
107 	case MAP__FUNCTION:
108 		return elf_sym__is_function(sym);
109 	case MAP__VARIABLE:
110 		return elf_sym__is_object(sym);
111 	default:
112 		return false;
113 	}
114 }
115 
116 static inline const char *elf_sym__name(const GElf_Sym *sym,
117 					const Elf_Data *symstrs)
118 {
119 	return symstrs->d_buf + sym->st_name;
120 }
121 
122 static inline const char *elf_sec__name(const GElf_Shdr *shdr,
123 					const Elf_Data *secstrs)
124 {
125 	return secstrs->d_buf + shdr->sh_name;
126 }
127 
128 static inline int elf_sec__is_text(const GElf_Shdr *shdr,
129 					const Elf_Data *secstrs)
130 {
131 	return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
132 }
133 
134 static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
135 				    const Elf_Data *secstrs)
136 {
137 	return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
138 }
139 
140 static bool elf_sec__is_a(GElf_Shdr *shdr, Elf_Data *secstrs,
141 			  enum map_type type)
142 {
143 	switch (type) {
144 	case MAP__FUNCTION:
145 		return elf_sec__is_text(shdr, secstrs);
146 	case MAP__VARIABLE:
147 		return elf_sec__is_data(shdr, secstrs);
148 	default:
149 		return false;
150 	}
151 }
152 
153 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
154 {
155 	Elf_Scn *sec = NULL;
156 	GElf_Shdr shdr;
157 	size_t cnt = 1;
158 
159 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
160 		gelf_getshdr(sec, &shdr);
161 
162 		if ((addr >= shdr.sh_addr) &&
163 		    (addr < (shdr.sh_addr + shdr.sh_size)))
164 			return cnt;
165 
166 		++cnt;
167 	}
168 
169 	return -1;
170 }
171 
172 Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
173 			     GElf_Shdr *shp, const char *name, size_t *idx)
174 {
175 	Elf_Scn *sec = NULL;
176 	size_t cnt = 1;
177 
178 	/* Elf is corrupted/truncated, avoid calling elf_strptr. */
179 	if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL))
180 		return NULL;
181 
182 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
183 		char *str;
184 
185 		gelf_getshdr(sec, shp);
186 		str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
187 		if (str && !strcmp(name, str)) {
188 			if (idx)
189 				*idx = cnt;
190 			return sec;
191 		}
192 		++cnt;
193 	}
194 
195 	return NULL;
196 }
197 
198 #define elf_section__for_each_rel(reldata, pos, pos_mem, idx, nr_entries) \
199 	for (idx = 0, pos = gelf_getrel(reldata, 0, &pos_mem); \
200 	     idx < nr_entries; \
201 	     ++idx, pos = gelf_getrel(reldata, idx, &pos_mem))
202 
203 #define elf_section__for_each_rela(reldata, pos, pos_mem, idx, nr_entries) \
204 	for (idx = 0, pos = gelf_getrela(reldata, 0, &pos_mem); \
205 	     idx < nr_entries; \
206 	     ++idx, pos = gelf_getrela(reldata, idx, &pos_mem))
207 
208 /*
209  * We need to check if we have a .dynsym, so that we can handle the
210  * .plt, synthesizing its symbols, that aren't on the symtabs (be it
211  * .dynsym or .symtab).
212  * And always look at the original dso, not at debuginfo packages, that
213  * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
214  */
215 int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss, struct map *map,
216 				symbol_filter_t filter)
217 {
218 	uint32_t nr_rel_entries, idx;
219 	GElf_Sym sym;
220 	u64 plt_offset;
221 	GElf_Shdr shdr_plt;
222 	struct symbol *f;
223 	GElf_Shdr shdr_rel_plt, shdr_dynsym;
224 	Elf_Data *reldata, *syms, *symstrs;
225 	Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
226 	size_t dynsym_idx;
227 	GElf_Ehdr ehdr;
228 	char sympltname[1024];
229 	Elf *elf;
230 	int nr = 0, symidx, err = 0;
231 
232 	if (!ss->dynsym)
233 		return 0;
234 
235 	elf = ss->elf;
236 	ehdr = ss->ehdr;
237 
238 	scn_dynsym = ss->dynsym;
239 	shdr_dynsym = ss->dynshdr;
240 	dynsym_idx = ss->dynsym_idx;
241 
242 	if (scn_dynsym == NULL)
243 		goto out_elf_end;
244 
245 	scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
246 					  ".rela.plt", NULL);
247 	if (scn_plt_rel == NULL) {
248 		scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
249 						  ".rel.plt", NULL);
250 		if (scn_plt_rel == NULL)
251 			goto out_elf_end;
252 	}
253 
254 	err = -1;
255 
256 	if (shdr_rel_plt.sh_link != dynsym_idx)
257 		goto out_elf_end;
258 
259 	if (elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL) == NULL)
260 		goto out_elf_end;
261 
262 	/*
263 	 * Fetch the relocation section to find the idxes to the GOT
264 	 * and the symbols in the .dynsym they refer to.
265 	 */
266 	reldata = elf_getdata(scn_plt_rel, NULL);
267 	if (reldata == NULL)
268 		goto out_elf_end;
269 
270 	syms = elf_getdata(scn_dynsym, NULL);
271 	if (syms == NULL)
272 		goto out_elf_end;
273 
274 	scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
275 	if (scn_symstrs == NULL)
276 		goto out_elf_end;
277 
278 	symstrs = elf_getdata(scn_symstrs, NULL);
279 	if (symstrs == NULL)
280 		goto out_elf_end;
281 
282 	if (symstrs->d_size == 0)
283 		goto out_elf_end;
284 
285 	nr_rel_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
286 	plt_offset = shdr_plt.sh_offset;
287 
288 	if (shdr_rel_plt.sh_type == SHT_RELA) {
289 		GElf_Rela pos_mem, *pos;
290 
291 		elf_section__for_each_rela(reldata, pos, pos_mem, idx,
292 					   nr_rel_entries) {
293 			symidx = GELF_R_SYM(pos->r_info);
294 			plt_offset += shdr_plt.sh_entsize;
295 			gelf_getsym(syms, symidx, &sym);
296 			snprintf(sympltname, sizeof(sympltname),
297 				 "%s@plt", elf_sym__name(&sym, symstrs));
298 
299 			f = symbol__new(plt_offset, shdr_plt.sh_entsize,
300 					STB_GLOBAL, sympltname);
301 			if (!f)
302 				goto out_elf_end;
303 
304 			if (filter && filter(map, f))
305 				symbol__delete(f);
306 			else {
307 				symbols__insert(&dso->symbols[map->type], f);
308 				++nr;
309 			}
310 		}
311 	} else if (shdr_rel_plt.sh_type == SHT_REL) {
312 		GElf_Rel pos_mem, *pos;
313 		elf_section__for_each_rel(reldata, pos, pos_mem, idx,
314 					  nr_rel_entries) {
315 			symidx = GELF_R_SYM(pos->r_info);
316 			plt_offset += shdr_plt.sh_entsize;
317 			gelf_getsym(syms, symidx, &sym);
318 			snprintf(sympltname, sizeof(sympltname),
319 				 "%s@plt", elf_sym__name(&sym, symstrs));
320 
321 			f = symbol__new(plt_offset, shdr_plt.sh_entsize,
322 					STB_GLOBAL, sympltname);
323 			if (!f)
324 				goto out_elf_end;
325 
326 			if (filter && filter(map, f))
327 				symbol__delete(f);
328 			else {
329 				symbols__insert(&dso->symbols[map->type], f);
330 				++nr;
331 			}
332 		}
333 	}
334 
335 	err = 0;
336 out_elf_end:
337 	if (err == 0)
338 		return nr;
339 	pr_debug("%s: problems reading %s PLT info.\n",
340 		 __func__, dso->long_name);
341 	return 0;
342 }
343 
344 /*
345  * Align offset to 4 bytes as needed for note name and descriptor data.
346  */
347 #define NOTE_ALIGN(n) (((n) + 3) & -4U)
348 
349 static int elf_read_build_id(Elf *elf, void *bf, size_t size)
350 {
351 	int err = -1;
352 	GElf_Ehdr ehdr;
353 	GElf_Shdr shdr;
354 	Elf_Data *data;
355 	Elf_Scn *sec;
356 	Elf_Kind ek;
357 	void *ptr;
358 
359 	if (size < BUILD_ID_SIZE)
360 		goto out;
361 
362 	ek = elf_kind(elf);
363 	if (ek != ELF_K_ELF)
364 		goto out;
365 
366 	if (gelf_getehdr(elf, &ehdr) == NULL) {
367 		pr_err("%s: cannot get elf header.\n", __func__);
368 		goto out;
369 	}
370 
371 	/*
372 	 * Check following sections for notes:
373 	 *   '.note.gnu.build-id'
374 	 *   '.notes'
375 	 *   '.note' (VDSO specific)
376 	 */
377 	do {
378 		sec = elf_section_by_name(elf, &ehdr, &shdr,
379 					  ".note.gnu.build-id", NULL);
380 		if (sec)
381 			break;
382 
383 		sec = elf_section_by_name(elf, &ehdr, &shdr,
384 					  ".notes", NULL);
385 		if (sec)
386 			break;
387 
388 		sec = elf_section_by_name(elf, &ehdr, &shdr,
389 					  ".note", NULL);
390 		if (sec)
391 			break;
392 
393 		return err;
394 
395 	} while (0);
396 
397 	data = elf_getdata(sec, NULL);
398 	if (data == NULL)
399 		goto out;
400 
401 	ptr = data->d_buf;
402 	while (ptr < (data->d_buf + data->d_size)) {
403 		GElf_Nhdr *nhdr = ptr;
404 		size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
405 		       descsz = NOTE_ALIGN(nhdr->n_descsz);
406 		const char *name;
407 
408 		ptr += sizeof(*nhdr);
409 		name = ptr;
410 		ptr += namesz;
411 		if (nhdr->n_type == NT_GNU_BUILD_ID &&
412 		    nhdr->n_namesz == sizeof("GNU")) {
413 			if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
414 				size_t sz = min(size, descsz);
415 				memcpy(bf, ptr, sz);
416 				memset(bf + sz, 0, size - sz);
417 				err = descsz;
418 				break;
419 			}
420 		}
421 		ptr += descsz;
422 	}
423 
424 out:
425 	return err;
426 }
427 
428 int filename__read_build_id(const char *filename, void *bf, size_t size)
429 {
430 	int fd, err = -1;
431 	Elf *elf;
432 
433 	if (size < BUILD_ID_SIZE)
434 		goto out;
435 
436 	fd = open(filename, O_RDONLY);
437 	if (fd < 0)
438 		goto out;
439 
440 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
441 	if (elf == NULL) {
442 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
443 		goto out_close;
444 	}
445 
446 	err = elf_read_build_id(elf, bf, size);
447 
448 	elf_end(elf);
449 out_close:
450 	close(fd);
451 out:
452 	return err;
453 }
454 
455 int sysfs__read_build_id(const char *filename, void *build_id, size_t size)
456 {
457 	int fd, err = -1;
458 
459 	if (size < BUILD_ID_SIZE)
460 		goto out;
461 
462 	fd = open(filename, O_RDONLY);
463 	if (fd < 0)
464 		goto out;
465 
466 	while (1) {
467 		char bf[BUFSIZ];
468 		GElf_Nhdr nhdr;
469 		size_t namesz, descsz;
470 
471 		if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
472 			break;
473 
474 		namesz = NOTE_ALIGN(nhdr.n_namesz);
475 		descsz = NOTE_ALIGN(nhdr.n_descsz);
476 		if (nhdr.n_type == NT_GNU_BUILD_ID &&
477 		    nhdr.n_namesz == sizeof("GNU")) {
478 			if (read(fd, bf, namesz) != (ssize_t)namesz)
479 				break;
480 			if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
481 				size_t sz = min(descsz, size);
482 				if (read(fd, build_id, sz) == (ssize_t)sz) {
483 					memset(build_id + sz, 0, size - sz);
484 					err = 0;
485 					break;
486 				}
487 			} else if (read(fd, bf, descsz) != (ssize_t)descsz)
488 				break;
489 		} else {
490 			int n = namesz + descsz;
491 			if (read(fd, bf, n) != n)
492 				break;
493 		}
494 	}
495 	close(fd);
496 out:
497 	return err;
498 }
499 
500 int filename__read_debuglink(const char *filename, char *debuglink,
501 			     size_t size)
502 {
503 	int fd, err = -1;
504 	Elf *elf;
505 	GElf_Ehdr ehdr;
506 	GElf_Shdr shdr;
507 	Elf_Data *data;
508 	Elf_Scn *sec;
509 	Elf_Kind ek;
510 
511 	fd = open(filename, O_RDONLY);
512 	if (fd < 0)
513 		goto out;
514 
515 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
516 	if (elf == NULL) {
517 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
518 		goto out_close;
519 	}
520 
521 	ek = elf_kind(elf);
522 	if (ek != ELF_K_ELF)
523 		goto out_elf_end;
524 
525 	if (gelf_getehdr(elf, &ehdr) == NULL) {
526 		pr_err("%s: cannot get elf header.\n", __func__);
527 		goto out_elf_end;
528 	}
529 
530 	sec = elf_section_by_name(elf, &ehdr, &shdr,
531 				  ".gnu_debuglink", NULL);
532 	if (sec == NULL)
533 		goto out_elf_end;
534 
535 	data = elf_getdata(sec, NULL);
536 	if (data == NULL)
537 		goto out_elf_end;
538 
539 	/* the start of this section is a zero-terminated string */
540 	strncpy(debuglink, data->d_buf, size);
541 
542 	err = 0;
543 
544 out_elf_end:
545 	elf_end(elf);
546 out_close:
547 	close(fd);
548 out:
549 	return err;
550 }
551 
552 static int dso__swap_init(struct dso *dso, unsigned char eidata)
553 {
554 	static unsigned int const endian = 1;
555 
556 	dso->needs_swap = DSO_SWAP__NO;
557 
558 	switch (eidata) {
559 	case ELFDATA2LSB:
560 		/* We are big endian, DSO is little endian. */
561 		if (*(unsigned char const *)&endian != 1)
562 			dso->needs_swap = DSO_SWAP__YES;
563 		break;
564 
565 	case ELFDATA2MSB:
566 		/* We are little endian, DSO is big endian. */
567 		if (*(unsigned char const *)&endian != 0)
568 			dso->needs_swap = DSO_SWAP__YES;
569 		break;
570 
571 	default:
572 		pr_err("unrecognized DSO data encoding %d\n", eidata);
573 		return -EINVAL;
574 	}
575 
576 	return 0;
577 }
578 
579 static int decompress_kmodule(struct dso *dso, const char *name,
580 			      enum dso_binary_type type)
581 {
582 	int fd = -1;
583 	char tmpbuf[] = "/tmp/perf-kmod-XXXXXX";
584 	struct kmod_path m;
585 
586 	if (type != DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP &&
587 	    type != DSO_BINARY_TYPE__GUEST_KMODULE_COMP &&
588 	    type != DSO_BINARY_TYPE__BUILD_ID_CACHE)
589 		return -1;
590 
591 	if (type == DSO_BINARY_TYPE__BUILD_ID_CACHE)
592 		name = dso->long_name;
593 
594 	if (kmod_path__parse_ext(&m, name) || !m.comp)
595 		return -1;
596 
597 	fd = mkstemp(tmpbuf);
598 	if (fd < 0) {
599 		dso->load_errno = errno;
600 		goto out;
601 	}
602 
603 	if (!decompress_to_file(m.ext, name, fd)) {
604 		dso->load_errno = DSO_LOAD_ERRNO__DECOMPRESSION_FAILURE;
605 		close(fd);
606 		fd = -1;
607 	}
608 
609 	unlink(tmpbuf);
610 
611 out:
612 	free(m.ext);
613 	return fd;
614 }
615 
616 bool symsrc__possibly_runtime(struct symsrc *ss)
617 {
618 	return ss->dynsym || ss->opdsec;
619 }
620 
621 bool symsrc__has_symtab(struct symsrc *ss)
622 {
623 	return ss->symtab != NULL;
624 }
625 
626 void symsrc__destroy(struct symsrc *ss)
627 {
628 	zfree(&ss->name);
629 	elf_end(ss->elf);
630 	close(ss->fd);
631 }
632 
633 bool __weak elf__needs_adjust_symbols(GElf_Ehdr ehdr)
634 {
635 	return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL;
636 }
637 
638 int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
639 		 enum dso_binary_type type)
640 {
641 	int err = -1;
642 	GElf_Ehdr ehdr;
643 	Elf *elf;
644 	int fd;
645 
646 	if (dso__needs_decompress(dso)) {
647 		fd = decompress_kmodule(dso, name, type);
648 		if (fd < 0)
649 			return -1;
650 	} else {
651 		fd = open(name, O_RDONLY);
652 		if (fd < 0) {
653 			dso->load_errno = errno;
654 			return -1;
655 		}
656 	}
657 
658 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
659 	if (elf == NULL) {
660 		pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
661 		dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
662 		goto out_close;
663 	}
664 
665 	if (gelf_getehdr(elf, &ehdr) == NULL) {
666 		dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
667 		pr_debug("%s: cannot get elf header.\n", __func__);
668 		goto out_elf_end;
669 	}
670 
671 	if (dso__swap_init(dso, ehdr.e_ident[EI_DATA])) {
672 		dso->load_errno = DSO_LOAD_ERRNO__INTERNAL_ERROR;
673 		goto out_elf_end;
674 	}
675 
676 	/* Always reject images with a mismatched build-id: */
677 	if (dso->has_build_id) {
678 		u8 build_id[BUILD_ID_SIZE];
679 
680 		if (elf_read_build_id(elf, build_id, BUILD_ID_SIZE) < 0) {
681 			dso->load_errno = DSO_LOAD_ERRNO__CANNOT_READ_BUILDID;
682 			goto out_elf_end;
683 		}
684 
685 		if (!dso__build_id_equal(dso, build_id)) {
686 			pr_debug("%s: build id mismatch for %s.\n", __func__, name);
687 			dso->load_errno = DSO_LOAD_ERRNO__MISMATCHING_BUILDID;
688 			goto out_elf_end;
689 		}
690 	}
691 
692 	ss->is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
693 
694 	ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab",
695 			NULL);
696 	if (ss->symshdr.sh_type != SHT_SYMTAB)
697 		ss->symtab = NULL;
698 
699 	ss->dynsym_idx = 0;
700 	ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym",
701 			&ss->dynsym_idx);
702 	if (ss->dynshdr.sh_type != SHT_DYNSYM)
703 		ss->dynsym = NULL;
704 
705 	ss->opdidx = 0;
706 	ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd",
707 			&ss->opdidx);
708 	if (ss->opdshdr.sh_type != SHT_PROGBITS)
709 		ss->opdsec = NULL;
710 
711 	if (dso->kernel == DSO_TYPE_USER) {
712 		GElf_Shdr shdr;
713 		ss->adjust_symbols = (ehdr.e_type == ET_EXEC ||
714 				ehdr.e_type == ET_REL ||
715 				dso__is_vdso(dso) ||
716 				elf_section_by_name(elf, &ehdr, &shdr,
717 						     ".gnu.prelink_undo",
718 						     NULL) != NULL);
719 	} else {
720 		ss->adjust_symbols = elf__needs_adjust_symbols(ehdr);
721 	}
722 
723 	ss->name   = strdup(name);
724 	if (!ss->name) {
725 		dso->load_errno = errno;
726 		goto out_elf_end;
727 	}
728 
729 	ss->elf    = elf;
730 	ss->fd     = fd;
731 	ss->ehdr   = ehdr;
732 	ss->type   = type;
733 
734 	return 0;
735 
736 out_elf_end:
737 	elf_end(elf);
738 out_close:
739 	close(fd);
740 	return err;
741 }
742 
743 /**
744  * ref_reloc_sym_not_found - has kernel relocation symbol been found.
745  * @kmap: kernel maps and relocation reference symbol
746  *
747  * This function returns %true if we are dealing with the kernel maps and the
748  * relocation reference symbol has not yet been found.  Otherwise %false is
749  * returned.
750  */
751 static bool ref_reloc_sym_not_found(struct kmap *kmap)
752 {
753 	return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
754 	       !kmap->ref_reloc_sym->unrelocated_addr;
755 }
756 
757 /**
758  * ref_reloc - kernel relocation offset.
759  * @kmap: kernel maps and relocation reference symbol
760  *
761  * This function returns the offset of kernel addresses as determined by using
762  * the relocation reference symbol i.e. if the kernel has not been relocated
763  * then the return value is zero.
764  */
765 static u64 ref_reloc(struct kmap *kmap)
766 {
767 	if (kmap && kmap->ref_reloc_sym &&
768 	    kmap->ref_reloc_sym->unrelocated_addr)
769 		return kmap->ref_reloc_sym->addr -
770 		       kmap->ref_reloc_sym->unrelocated_addr;
771 	return 0;
772 }
773 
774 static bool want_demangle(bool is_kernel_sym)
775 {
776 	return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle;
777 }
778 
779 void __weak arch__elf_sym_adjust(GElf_Sym *sym __maybe_unused) { }
780 
781 int dso__load_sym(struct dso *dso, struct map *map,
782 		  struct symsrc *syms_ss, struct symsrc *runtime_ss,
783 		  symbol_filter_t filter, int kmodule)
784 {
785 	struct kmap *kmap = dso->kernel ? map__kmap(map) : NULL;
786 	struct map_groups *kmaps = kmap ? map__kmaps(map) : NULL;
787 	struct map *curr_map = map;
788 	struct dso *curr_dso = dso;
789 	Elf_Data *symstrs, *secstrs;
790 	uint32_t nr_syms;
791 	int err = -1;
792 	uint32_t idx;
793 	GElf_Ehdr ehdr;
794 	GElf_Shdr shdr;
795 	Elf_Data *syms, *opddata = NULL;
796 	GElf_Sym sym;
797 	Elf_Scn *sec, *sec_strndx;
798 	Elf *elf;
799 	int nr = 0;
800 	bool remap_kernel = false, adjust_kernel_syms = false;
801 
802 	if (kmap && !kmaps)
803 		return -1;
804 
805 	dso->symtab_type = syms_ss->type;
806 	dso->is_64_bit = syms_ss->is_64_bit;
807 	dso->rel = syms_ss->ehdr.e_type == ET_REL;
808 
809 	/*
810 	 * Modules may already have symbols from kallsyms, but those symbols
811 	 * have the wrong values for the dso maps, so remove them.
812 	 */
813 	if (kmodule && syms_ss->symtab)
814 		symbols__delete(&dso->symbols[map->type]);
815 
816 	if (!syms_ss->symtab) {
817 		/*
818 		 * If the vmlinux is stripped, fail so we will fall back
819 		 * to using kallsyms. The vmlinux runtime symbols aren't
820 		 * of much use.
821 		 */
822 		if (dso->kernel)
823 			goto out_elf_end;
824 
825 		syms_ss->symtab  = syms_ss->dynsym;
826 		syms_ss->symshdr = syms_ss->dynshdr;
827 	}
828 
829 	elf = syms_ss->elf;
830 	ehdr = syms_ss->ehdr;
831 	sec = syms_ss->symtab;
832 	shdr = syms_ss->symshdr;
833 
834 	if (runtime_ss->opdsec)
835 		opddata = elf_rawdata(runtime_ss->opdsec, NULL);
836 
837 	syms = elf_getdata(sec, NULL);
838 	if (syms == NULL)
839 		goto out_elf_end;
840 
841 	sec = elf_getscn(elf, shdr.sh_link);
842 	if (sec == NULL)
843 		goto out_elf_end;
844 
845 	symstrs = elf_getdata(sec, NULL);
846 	if (symstrs == NULL)
847 		goto out_elf_end;
848 
849 	sec_strndx = elf_getscn(runtime_ss->elf, runtime_ss->ehdr.e_shstrndx);
850 	if (sec_strndx == NULL)
851 		goto out_elf_end;
852 
853 	secstrs = elf_getdata(sec_strndx, NULL);
854 	if (secstrs == NULL)
855 		goto out_elf_end;
856 
857 	nr_syms = shdr.sh_size / shdr.sh_entsize;
858 
859 	memset(&sym, 0, sizeof(sym));
860 
861 	/*
862 	 * The kernel relocation symbol is needed in advance in order to adjust
863 	 * kernel maps correctly.
864 	 */
865 	if (ref_reloc_sym_not_found(kmap)) {
866 		elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
867 			const char *elf_name = elf_sym__name(&sym, symstrs);
868 
869 			if (strcmp(elf_name, kmap->ref_reloc_sym->name))
870 				continue;
871 			kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
872 			map->reloc = kmap->ref_reloc_sym->addr -
873 				     kmap->ref_reloc_sym->unrelocated_addr;
874 			break;
875 		}
876 	}
877 
878 	/*
879 	 * Handle any relocation of vdso necessary because older kernels
880 	 * attempted to prelink vdso to its virtual address.
881 	 */
882 	if (dso__is_vdso(dso)) {
883 		GElf_Shdr tshdr;
884 
885 		if (elf_section_by_name(elf, &ehdr, &tshdr, ".text", NULL))
886 			map->reloc = map->start - tshdr.sh_addr + tshdr.sh_offset;
887 	}
888 
889 	dso->adjust_symbols = runtime_ss->adjust_symbols || ref_reloc(kmap);
890 	/*
891 	 * Initial kernel and module mappings do not map to the dso.  For
892 	 * function mappings, flag the fixups.
893 	 */
894 	if (map->type == MAP__FUNCTION && (dso->kernel || kmodule)) {
895 		remap_kernel = true;
896 		adjust_kernel_syms = dso->adjust_symbols;
897 	}
898 	elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
899 		struct symbol *f;
900 		const char *elf_name = elf_sym__name(&sym, symstrs);
901 		char *demangled = NULL;
902 		int is_label = elf_sym__is_label(&sym);
903 		const char *section_name;
904 		bool used_opd = false;
905 
906 		if (!is_label && !elf_sym__is_a(&sym, map->type))
907 			continue;
908 
909 		/* Reject ARM ELF "mapping symbols": these aren't unique and
910 		 * don't identify functions, so will confuse the profile
911 		 * output: */
912 		if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) {
913 			if (elf_name[0] == '$' && strchr("adtx", elf_name[1])
914 			    && (elf_name[2] == '\0' || elf_name[2] == '.'))
915 				continue;
916 		}
917 
918 		if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) {
919 			u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr;
920 			u64 *opd = opddata->d_buf + offset;
921 			sym.st_value = DSO__SWAP(dso, u64, *opd);
922 			sym.st_shndx = elf_addr_to_index(runtime_ss->elf,
923 					sym.st_value);
924 			used_opd = true;
925 		}
926 		/*
927 		 * When loading symbols in a data mapping, ABS symbols (which
928 		 * has a value of SHN_ABS in its st_shndx) failed at
929 		 * elf_getscn().  And it marks the loading as a failure so
930 		 * already loaded symbols cannot be fixed up.
931 		 *
932 		 * I'm not sure what should be done. Just ignore them for now.
933 		 * - Namhyung Kim
934 		 */
935 		if (sym.st_shndx == SHN_ABS)
936 			continue;
937 
938 		sec = elf_getscn(runtime_ss->elf, sym.st_shndx);
939 		if (!sec)
940 			goto out_elf_end;
941 
942 		gelf_getshdr(sec, &shdr);
943 
944 		if (is_label && !elf_sec__is_a(&shdr, secstrs, map->type))
945 			continue;
946 
947 		section_name = elf_sec__name(&shdr, secstrs);
948 
949 		/* On ARM, symbols for thumb functions have 1 added to
950 		 * the symbol address as a flag - remove it */
951 		if ((ehdr.e_machine == EM_ARM) &&
952 		    (map->type == MAP__FUNCTION) &&
953 		    (sym.st_value & 1))
954 			--sym.st_value;
955 
956 		arch__elf_sym_adjust(&sym);
957 
958 		if (dso->kernel || kmodule) {
959 			char dso_name[PATH_MAX];
960 
961 			/* Adjust symbol to map to file offset */
962 			if (adjust_kernel_syms)
963 				sym.st_value -= shdr.sh_addr - shdr.sh_offset;
964 
965 			if (strcmp(section_name,
966 				   (curr_dso->short_name +
967 				    dso->short_name_len)) == 0)
968 				goto new_symbol;
969 
970 			if (strcmp(section_name, ".text") == 0) {
971 				/*
972 				 * The initial kernel mapping is based on
973 				 * kallsyms and identity maps.  Overwrite it to
974 				 * map to the kernel dso.
975 				 */
976 				if (remap_kernel && dso->kernel) {
977 					remap_kernel = false;
978 					map->start = shdr.sh_addr +
979 						     ref_reloc(kmap);
980 					map->end = map->start + shdr.sh_size;
981 					map->pgoff = shdr.sh_offset;
982 					map->map_ip = map__map_ip;
983 					map->unmap_ip = map__unmap_ip;
984 					/* Ensure maps are correctly ordered */
985 					if (kmaps) {
986 						map__get(map);
987 						map_groups__remove(kmaps, map);
988 						map_groups__insert(kmaps, map);
989 						map__put(map);
990 					}
991 				}
992 
993 				/*
994 				 * The initial module mapping is based on
995 				 * /proc/modules mapped to offset zero.
996 				 * Overwrite it to map to the module dso.
997 				 */
998 				if (remap_kernel && kmodule) {
999 					remap_kernel = false;
1000 					map->pgoff = shdr.sh_offset;
1001 				}
1002 
1003 				curr_map = map;
1004 				curr_dso = dso;
1005 				goto new_symbol;
1006 			}
1007 
1008 			if (!kmap)
1009 				goto new_symbol;
1010 
1011 			snprintf(dso_name, sizeof(dso_name),
1012 				 "%s%s", dso->short_name, section_name);
1013 
1014 			curr_map = map_groups__find_by_name(kmaps, map->type, dso_name);
1015 			if (curr_map == NULL) {
1016 				u64 start = sym.st_value;
1017 
1018 				if (kmodule)
1019 					start += map->start + shdr.sh_offset;
1020 
1021 				curr_dso = dso__new(dso_name);
1022 				if (curr_dso == NULL)
1023 					goto out_elf_end;
1024 				curr_dso->kernel = dso->kernel;
1025 				curr_dso->long_name = dso->long_name;
1026 				curr_dso->long_name_len = dso->long_name_len;
1027 				curr_map = map__new2(start, curr_dso,
1028 						     map->type);
1029 				if (curr_map == NULL) {
1030 					dso__put(curr_dso);
1031 					goto out_elf_end;
1032 				}
1033 				if (adjust_kernel_syms) {
1034 					curr_map->start = shdr.sh_addr +
1035 							  ref_reloc(kmap);
1036 					curr_map->end = curr_map->start +
1037 							shdr.sh_size;
1038 					curr_map->pgoff = shdr.sh_offset;
1039 				} else {
1040 					curr_map->map_ip = identity__map_ip;
1041 					curr_map->unmap_ip = identity__map_ip;
1042 				}
1043 				curr_dso->symtab_type = dso->symtab_type;
1044 				map_groups__insert(kmaps, curr_map);
1045 				dsos__add(&map->groups->machine->dsos, curr_dso);
1046 				dso__set_loaded(curr_dso, map->type);
1047 			} else
1048 				curr_dso = curr_map->dso;
1049 
1050 			goto new_symbol;
1051 		}
1052 
1053 		if ((used_opd && runtime_ss->adjust_symbols)
1054 				|| (!used_opd && syms_ss->adjust_symbols)) {
1055 			pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1056 				  "sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n", __func__,
1057 				  (u64)sym.st_value, (u64)shdr.sh_addr,
1058 				  (u64)shdr.sh_offset);
1059 			sym.st_value -= shdr.sh_addr - shdr.sh_offset;
1060 		}
1061 new_symbol:
1062 		/*
1063 		 * We need to figure out if the object was created from C++ sources
1064 		 * DWARF DW_compile_unit has this, but we don't always have access
1065 		 * to it...
1066 		 */
1067 		if (want_demangle(dso->kernel || kmodule)) {
1068 			int demangle_flags = DMGL_NO_OPTS;
1069 			if (verbose)
1070 				demangle_flags = DMGL_PARAMS | DMGL_ANSI;
1071 
1072 			demangled = bfd_demangle(NULL, elf_name, demangle_flags);
1073 			if (demangled != NULL)
1074 				elf_name = demangled;
1075 		}
1076 		f = symbol__new(sym.st_value, sym.st_size,
1077 				GELF_ST_BIND(sym.st_info), elf_name);
1078 		free(demangled);
1079 		if (!f)
1080 			goto out_elf_end;
1081 
1082 		if (filter && filter(curr_map, f))
1083 			symbol__delete(f);
1084 		else {
1085 			symbols__insert(&curr_dso->symbols[curr_map->type], f);
1086 			nr++;
1087 		}
1088 	}
1089 
1090 	/*
1091 	 * For misannotated, zeroed, ASM function sizes.
1092 	 */
1093 	if (nr > 0) {
1094 		if (!symbol_conf.allow_aliases)
1095 			symbols__fixup_duplicate(&dso->symbols[map->type]);
1096 		symbols__fixup_end(&dso->symbols[map->type]);
1097 		if (kmap) {
1098 			/*
1099 			 * We need to fixup this here too because we create new
1100 			 * maps here, for things like vsyscall sections.
1101 			 */
1102 			__map_groups__fixup_end(kmaps, map->type);
1103 		}
1104 	}
1105 	err = nr;
1106 out_elf_end:
1107 	return err;
1108 }
1109 
1110 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
1111 {
1112 	GElf_Phdr phdr;
1113 	size_t i, phdrnum;
1114 	int err;
1115 	u64 sz;
1116 
1117 	if (elf_getphdrnum(elf, &phdrnum))
1118 		return -1;
1119 
1120 	for (i = 0; i < phdrnum; i++) {
1121 		if (gelf_getphdr(elf, i, &phdr) == NULL)
1122 			return -1;
1123 		if (phdr.p_type != PT_LOAD)
1124 			continue;
1125 		if (exe) {
1126 			if (!(phdr.p_flags & PF_X))
1127 				continue;
1128 		} else {
1129 			if (!(phdr.p_flags & PF_R))
1130 				continue;
1131 		}
1132 		sz = min(phdr.p_memsz, phdr.p_filesz);
1133 		if (!sz)
1134 			continue;
1135 		err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
1136 		if (err)
1137 			return err;
1138 	}
1139 	return 0;
1140 }
1141 
1142 int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data,
1143 		    bool *is_64_bit)
1144 {
1145 	int err;
1146 	Elf *elf;
1147 
1148 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1149 	if (elf == NULL)
1150 		return -1;
1151 
1152 	if (is_64_bit)
1153 		*is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1154 
1155 	err = elf_read_maps(elf, exe, mapfn, data);
1156 
1157 	elf_end(elf);
1158 	return err;
1159 }
1160 
1161 enum dso_type dso__type_fd(int fd)
1162 {
1163 	enum dso_type dso_type = DSO__TYPE_UNKNOWN;
1164 	GElf_Ehdr ehdr;
1165 	Elf_Kind ek;
1166 	Elf *elf;
1167 
1168 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1169 	if (elf == NULL)
1170 		goto out;
1171 
1172 	ek = elf_kind(elf);
1173 	if (ek != ELF_K_ELF)
1174 		goto out_end;
1175 
1176 	if (gelf_getclass(elf) == ELFCLASS64) {
1177 		dso_type = DSO__TYPE_64BIT;
1178 		goto out_end;
1179 	}
1180 
1181 	if (gelf_getehdr(elf, &ehdr) == NULL)
1182 		goto out_end;
1183 
1184 	if (ehdr.e_machine == EM_X86_64)
1185 		dso_type = DSO__TYPE_X32BIT;
1186 	else
1187 		dso_type = DSO__TYPE_32BIT;
1188 out_end:
1189 	elf_end(elf);
1190 out:
1191 	return dso_type;
1192 }
1193 
1194 static int copy_bytes(int from, off_t from_offs, int to, off_t to_offs, u64 len)
1195 {
1196 	ssize_t r;
1197 	size_t n;
1198 	int err = -1;
1199 	char *buf = malloc(page_size);
1200 
1201 	if (buf == NULL)
1202 		return -1;
1203 
1204 	if (lseek(to, to_offs, SEEK_SET) != to_offs)
1205 		goto out;
1206 
1207 	if (lseek(from, from_offs, SEEK_SET) != from_offs)
1208 		goto out;
1209 
1210 	while (len) {
1211 		n = page_size;
1212 		if (len < n)
1213 			n = len;
1214 		/* Use read because mmap won't work on proc files */
1215 		r = read(from, buf, n);
1216 		if (r < 0)
1217 			goto out;
1218 		if (!r)
1219 			break;
1220 		n = r;
1221 		r = write(to, buf, n);
1222 		if (r < 0)
1223 			goto out;
1224 		if ((size_t)r != n)
1225 			goto out;
1226 		len -= n;
1227 	}
1228 
1229 	err = 0;
1230 out:
1231 	free(buf);
1232 	return err;
1233 }
1234 
1235 struct kcore {
1236 	int fd;
1237 	int elfclass;
1238 	Elf *elf;
1239 	GElf_Ehdr ehdr;
1240 };
1241 
1242 static int kcore__open(struct kcore *kcore, const char *filename)
1243 {
1244 	GElf_Ehdr *ehdr;
1245 
1246 	kcore->fd = open(filename, O_RDONLY);
1247 	if (kcore->fd == -1)
1248 		return -1;
1249 
1250 	kcore->elf = elf_begin(kcore->fd, ELF_C_READ, NULL);
1251 	if (!kcore->elf)
1252 		goto out_close;
1253 
1254 	kcore->elfclass = gelf_getclass(kcore->elf);
1255 	if (kcore->elfclass == ELFCLASSNONE)
1256 		goto out_end;
1257 
1258 	ehdr = gelf_getehdr(kcore->elf, &kcore->ehdr);
1259 	if (!ehdr)
1260 		goto out_end;
1261 
1262 	return 0;
1263 
1264 out_end:
1265 	elf_end(kcore->elf);
1266 out_close:
1267 	close(kcore->fd);
1268 	return -1;
1269 }
1270 
1271 static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
1272 		       bool temp)
1273 {
1274 	kcore->elfclass = elfclass;
1275 
1276 	if (temp)
1277 		kcore->fd = mkstemp(filename);
1278 	else
1279 		kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
1280 	if (kcore->fd == -1)
1281 		return -1;
1282 
1283 	kcore->elf = elf_begin(kcore->fd, ELF_C_WRITE, NULL);
1284 	if (!kcore->elf)
1285 		goto out_close;
1286 
1287 	if (!gelf_newehdr(kcore->elf, elfclass))
1288 		goto out_end;
1289 
1290 	memset(&kcore->ehdr, 0, sizeof(GElf_Ehdr));
1291 
1292 	return 0;
1293 
1294 out_end:
1295 	elf_end(kcore->elf);
1296 out_close:
1297 	close(kcore->fd);
1298 	unlink(filename);
1299 	return -1;
1300 }
1301 
1302 static void kcore__close(struct kcore *kcore)
1303 {
1304 	elf_end(kcore->elf);
1305 	close(kcore->fd);
1306 }
1307 
1308 static int kcore__copy_hdr(struct kcore *from, struct kcore *to, size_t count)
1309 {
1310 	GElf_Ehdr *ehdr = &to->ehdr;
1311 	GElf_Ehdr *kehdr = &from->ehdr;
1312 
1313 	memcpy(ehdr->e_ident, kehdr->e_ident, EI_NIDENT);
1314 	ehdr->e_type      = kehdr->e_type;
1315 	ehdr->e_machine   = kehdr->e_machine;
1316 	ehdr->e_version   = kehdr->e_version;
1317 	ehdr->e_entry     = 0;
1318 	ehdr->e_shoff     = 0;
1319 	ehdr->e_flags     = kehdr->e_flags;
1320 	ehdr->e_phnum     = count;
1321 	ehdr->e_shentsize = 0;
1322 	ehdr->e_shnum     = 0;
1323 	ehdr->e_shstrndx  = 0;
1324 
1325 	if (from->elfclass == ELFCLASS32) {
1326 		ehdr->e_phoff     = sizeof(Elf32_Ehdr);
1327 		ehdr->e_ehsize    = sizeof(Elf32_Ehdr);
1328 		ehdr->e_phentsize = sizeof(Elf32_Phdr);
1329 	} else {
1330 		ehdr->e_phoff     = sizeof(Elf64_Ehdr);
1331 		ehdr->e_ehsize    = sizeof(Elf64_Ehdr);
1332 		ehdr->e_phentsize = sizeof(Elf64_Phdr);
1333 	}
1334 
1335 	if (!gelf_update_ehdr(to->elf, ehdr))
1336 		return -1;
1337 
1338 	if (!gelf_newphdr(to->elf, count))
1339 		return -1;
1340 
1341 	return 0;
1342 }
1343 
1344 static int kcore__add_phdr(struct kcore *kcore, int idx, off_t offset,
1345 			   u64 addr, u64 len)
1346 {
1347 	GElf_Phdr phdr = {
1348 		.p_type		= PT_LOAD,
1349 		.p_flags	= PF_R | PF_W | PF_X,
1350 		.p_offset	= offset,
1351 		.p_vaddr	= addr,
1352 		.p_paddr	= 0,
1353 		.p_filesz	= len,
1354 		.p_memsz	= len,
1355 		.p_align	= page_size,
1356 	};
1357 
1358 	if (!gelf_update_phdr(kcore->elf, idx, &phdr))
1359 		return -1;
1360 
1361 	return 0;
1362 }
1363 
1364 static off_t kcore__write(struct kcore *kcore)
1365 {
1366 	return elf_update(kcore->elf, ELF_C_WRITE);
1367 }
1368 
1369 struct phdr_data {
1370 	off_t offset;
1371 	u64 addr;
1372 	u64 len;
1373 };
1374 
1375 struct kcore_copy_info {
1376 	u64 stext;
1377 	u64 etext;
1378 	u64 first_symbol;
1379 	u64 last_symbol;
1380 	u64 first_module;
1381 	u64 last_module_symbol;
1382 	struct phdr_data kernel_map;
1383 	struct phdr_data modules_map;
1384 };
1385 
1386 static int kcore_copy__process_kallsyms(void *arg, const char *name, char type,
1387 					u64 start)
1388 {
1389 	struct kcore_copy_info *kci = arg;
1390 
1391 	if (!symbol_type__is_a(type, MAP__FUNCTION))
1392 		return 0;
1393 
1394 	if (strchr(name, '[')) {
1395 		if (start > kci->last_module_symbol)
1396 			kci->last_module_symbol = start;
1397 		return 0;
1398 	}
1399 
1400 	if (!kci->first_symbol || start < kci->first_symbol)
1401 		kci->first_symbol = start;
1402 
1403 	if (!kci->last_symbol || start > kci->last_symbol)
1404 		kci->last_symbol = start;
1405 
1406 	if (!strcmp(name, "_stext")) {
1407 		kci->stext = start;
1408 		return 0;
1409 	}
1410 
1411 	if (!strcmp(name, "_etext")) {
1412 		kci->etext = start;
1413 		return 0;
1414 	}
1415 
1416 	return 0;
1417 }
1418 
1419 static int kcore_copy__parse_kallsyms(struct kcore_copy_info *kci,
1420 				      const char *dir)
1421 {
1422 	char kallsyms_filename[PATH_MAX];
1423 
1424 	scnprintf(kallsyms_filename, PATH_MAX, "%s/kallsyms", dir);
1425 
1426 	if (symbol__restricted_filename(kallsyms_filename, "/proc/kallsyms"))
1427 		return -1;
1428 
1429 	if (kallsyms__parse(kallsyms_filename, kci,
1430 			    kcore_copy__process_kallsyms) < 0)
1431 		return -1;
1432 
1433 	return 0;
1434 }
1435 
1436 static int kcore_copy__process_modules(void *arg,
1437 				       const char *name __maybe_unused,
1438 				       u64 start)
1439 {
1440 	struct kcore_copy_info *kci = arg;
1441 
1442 	if (!kci->first_module || start < kci->first_module)
1443 		kci->first_module = start;
1444 
1445 	return 0;
1446 }
1447 
1448 static int kcore_copy__parse_modules(struct kcore_copy_info *kci,
1449 				     const char *dir)
1450 {
1451 	char modules_filename[PATH_MAX];
1452 
1453 	scnprintf(modules_filename, PATH_MAX, "%s/modules", dir);
1454 
1455 	if (symbol__restricted_filename(modules_filename, "/proc/modules"))
1456 		return -1;
1457 
1458 	if (modules__parse(modules_filename, kci,
1459 			   kcore_copy__process_modules) < 0)
1460 		return -1;
1461 
1462 	return 0;
1463 }
1464 
1465 static void kcore_copy__map(struct phdr_data *p, u64 start, u64 end, u64 pgoff,
1466 			    u64 s, u64 e)
1467 {
1468 	if (p->addr || s < start || s >= end)
1469 		return;
1470 
1471 	p->addr = s;
1472 	p->offset = (s - start) + pgoff;
1473 	p->len = e < end ? e - s : end - s;
1474 }
1475 
1476 static int kcore_copy__read_map(u64 start, u64 len, u64 pgoff, void *data)
1477 {
1478 	struct kcore_copy_info *kci = data;
1479 	u64 end = start + len;
1480 
1481 	kcore_copy__map(&kci->kernel_map, start, end, pgoff, kci->stext,
1482 			kci->etext);
1483 
1484 	kcore_copy__map(&kci->modules_map, start, end, pgoff, kci->first_module,
1485 			kci->last_module_symbol);
1486 
1487 	return 0;
1488 }
1489 
1490 static int kcore_copy__read_maps(struct kcore_copy_info *kci, Elf *elf)
1491 {
1492 	if (elf_read_maps(elf, true, kcore_copy__read_map, kci) < 0)
1493 		return -1;
1494 
1495 	return 0;
1496 }
1497 
1498 static int kcore_copy__calc_maps(struct kcore_copy_info *kci, const char *dir,
1499 				 Elf *elf)
1500 {
1501 	if (kcore_copy__parse_kallsyms(kci, dir))
1502 		return -1;
1503 
1504 	if (kcore_copy__parse_modules(kci, dir))
1505 		return -1;
1506 
1507 	if (kci->stext)
1508 		kci->stext = round_down(kci->stext, page_size);
1509 	else
1510 		kci->stext = round_down(kci->first_symbol, page_size);
1511 
1512 	if (kci->etext) {
1513 		kci->etext = round_up(kci->etext, page_size);
1514 	} else if (kci->last_symbol) {
1515 		kci->etext = round_up(kci->last_symbol, page_size);
1516 		kci->etext += page_size;
1517 	}
1518 
1519 	kci->first_module = round_down(kci->first_module, page_size);
1520 
1521 	if (kci->last_module_symbol) {
1522 		kci->last_module_symbol = round_up(kci->last_module_symbol,
1523 						   page_size);
1524 		kci->last_module_symbol += page_size;
1525 	}
1526 
1527 	if (!kci->stext || !kci->etext)
1528 		return -1;
1529 
1530 	if (kci->first_module && !kci->last_module_symbol)
1531 		return -1;
1532 
1533 	return kcore_copy__read_maps(kci, elf);
1534 }
1535 
1536 static int kcore_copy__copy_file(const char *from_dir, const char *to_dir,
1537 				 const char *name)
1538 {
1539 	char from_filename[PATH_MAX];
1540 	char to_filename[PATH_MAX];
1541 
1542 	scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
1543 	scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
1544 
1545 	return copyfile_mode(from_filename, to_filename, 0400);
1546 }
1547 
1548 static int kcore_copy__unlink(const char *dir, const char *name)
1549 {
1550 	char filename[PATH_MAX];
1551 
1552 	scnprintf(filename, PATH_MAX, "%s/%s", dir, name);
1553 
1554 	return unlink(filename);
1555 }
1556 
1557 static int kcore_copy__compare_fds(int from, int to)
1558 {
1559 	char *buf_from;
1560 	char *buf_to;
1561 	ssize_t ret;
1562 	size_t len;
1563 	int err = -1;
1564 
1565 	buf_from = malloc(page_size);
1566 	buf_to = malloc(page_size);
1567 	if (!buf_from || !buf_to)
1568 		goto out;
1569 
1570 	while (1) {
1571 		/* Use read because mmap won't work on proc files */
1572 		ret = read(from, buf_from, page_size);
1573 		if (ret < 0)
1574 			goto out;
1575 
1576 		if (!ret)
1577 			break;
1578 
1579 		len = ret;
1580 
1581 		if (readn(to, buf_to, len) != (int)len)
1582 			goto out;
1583 
1584 		if (memcmp(buf_from, buf_to, len))
1585 			goto out;
1586 	}
1587 
1588 	err = 0;
1589 out:
1590 	free(buf_to);
1591 	free(buf_from);
1592 	return err;
1593 }
1594 
1595 static int kcore_copy__compare_files(const char *from_filename,
1596 				     const char *to_filename)
1597 {
1598 	int from, to, err = -1;
1599 
1600 	from = open(from_filename, O_RDONLY);
1601 	if (from < 0)
1602 		return -1;
1603 
1604 	to = open(to_filename, O_RDONLY);
1605 	if (to < 0)
1606 		goto out_close_from;
1607 
1608 	err = kcore_copy__compare_fds(from, to);
1609 
1610 	close(to);
1611 out_close_from:
1612 	close(from);
1613 	return err;
1614 }
1615 
1616 static int kcore_copy__compare_file(const char *from_dir, const char *to_dir,
1617 				    const char *name)
1618 {
1619 	char from_filename[PATH_MAX];
1620 	char to_filename[PATH_MAX];
1621 
1622 	scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
1623 	scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
1624 
1625 	return kcore_copy__compare_files(from_filename, to_filename);
1626 }
1627 
1628 /**
1629  * kcore_copy - copy kallsyms, modules and kcore from one directory to another.
1630  * @from_dir: from directory
1631  * @to_dir: to directory
1632  *
1633  * This function copies kallsyms, modules and kcore files from one directory to
1634  * another.  kallsyms and modules are copied entirely.  Only code segments are
1635  * copied from kcore.  It is assumed that two segments suffice: one for the
1636  * kernel proper and one for all the modules.  The code segments are determined
1637  * from kallsyms and modules files.  The kernel map starts at _stext or the
1638  * lowest function symbol, and ends at _etext or the highest function symbol.
1639  * The module map starts at the lowest module address and ends at the highest
1640  * module symbol.  Start addresses are rounded down to the nearest page.  End
1641  * addresses are rounded up to the nearest page.  An extra page is added to the
1642  * highest kernel symbol and highest module symbol to, hopefully, encompass that
1643  * symbol too.  Because it contains only code sections, the resulting kcore is
1644  * unusual.  One significant peculiarity is that the mapping (start -> pgoff)
1645  * is not the same for the kernel map and the modules map.  That happens because
1646  * the data is copied adjacently whereas the original kcore has gaps.  Finally,
1647  * kallsyms and modules files are compared with their copies to check that
1648  * modules have not been loaded or unloaded while the copies were taking place.
1649  *
1650  * Return: %0 on success, %-1 on failure.
1651  */
1652 int kcore_copy(const char *from_dir, const char *to_dir)
1653 {
1654 	struct kcore kcore;
1655 	struct kcore extract;
1656 	size_t count = 2;
1657 	int idx = 0, err = -1;
1658 	off_t offset = page_size, sz, modules_offset = 0;
1659 	struct kcore_copy_info kci = { .stext = 0, };
1660 	char kcore_filename[PATH_MAX];
1661 	char extract_filename[PATH_MAX];
1662 
1663 	if (kcore_copy__copy_file(from_dir, to_dir, "kallsyms"))
1664 		return -1;
1665 
1666 	if (kcore_copy__copy_file(from_dir, to_dir, "modules"))
1667 		goto out_unlink_kallsyms;
1668 
1669 	scnprintf(kcore_filename, PATH_MAX, "%s/kcore", from_dir);
1670 	scnprintf(extract_filename, PATH_MAX, "%s/kcore", to_dir);
1671 
1672 	if (kcore__open(&kcore, kcore_filename))
1673 		goto out_unlink_modules;
1674 
1675 	if (kcore_copy__calc_maps(&kci, from_dir, kcore.elf))
1676 		goto out_kcore_close;
1677 
1678 	if (kcore__init(&extract, extract_filename, kcore.elfclass, false))
1679 		goto out_kcore_close;
1680 
1681 	if (!kci.modules_map.addr)
1682 		count -= 1;
1683 
1684 	if (kcore__copy_hdr(&kcore, &extract, count))
1685 		goto out_extract_close;
1686 
1687 	if (kcore__add_phdr(&extract, idx++, offset, kci.kernel_map.addr,
1688 			    kci.kernel_map.len))
1689 		goto out_extract_close;
1690 
1691 	if (kci.modules_map.addr) {
1692 		modules_offset = offset + kci.kernel_map.len;
1693 		if (kcore__add_phdr(&extract, idx, modules_offset,
1694 				    kci.modules_map.addr, kci.modules_map.len))
1695 			goto out_extract_close;
1696 	}
1697 
1698 	sz = kcore__write(&extract);
1699 	if (sz < 0 || sz > offset)
1700 		goto out_extract_close;
1701 
1702 	if (copy_bytes(kcore.fd, kci.kernel_map.offset, extract.fd, offset,
1703 		       kci.kernel_map.len))
1704 		goto out_extract_close;
1705 
1706 	if (modules_offset && copy_bytes(kcore.fd, kci.modules_map.offset,
1707 					 extract.fd, modules_offset,
1708 					 kci.modules_map.len))
1709 		goto out_extract_close;
1710 
1711 	if (kcore_copy__compare_file(from_dir, to_dir, "modules"))
1712 		goto out_extract_close;
1713 
1714 	if (kcore_copy__compare_file(from_dir, to_dir, "kallsyms"))
1715 		goto out_extract_close;
1716 
1717 	err = 0;
1718 
1719 out_extract_close:
1720 	kcore__close(&extract);
1721 	if (err)
1722 		unlink(extract_filename);
1723 out_kcore_close:
1724 	kcore__close(&kcore);
1725 out_unlink_modules:
1726 	if (err)
1727 		kcore_copy__unlink(to_dir, "modules");
1728 out_unlink_kallsyms:
1729 	if (err)
1730 		kcore_copy__unlink(to_dir, "kallsyms");
1731 
1732 	return err;
1733 }
1734 
1735 int kcore_extract__create(struct kcore_extract *kce)
1736 {
1737 	struct kcore kcore;
1738 	struct kcore extract;
1739 	size_t count = 1;
1740 	int idx = 0, err = -1;
1741 	off_t offset = page_size, sz;
1742 
1743 	if (kcore__open(&kcore, kce->kcore_filename))
1744 		return -1;
1745 
1746 	strcpy(kce->extract_filename, PERF_KCORE_EXTRACT);
1747 	if (kcore__init(&extract, kce->extract_filename, kcore.elfclass, true))
1748 		goto out_kcore_close;
1749 
1750 	if (kcore__copy_hdr(&kcore, &extract, count))
1751 		goto out_extract_close;
1752 
1753 	if (kcore__add_phdr(&extract, idx, offset, kce->addr, kce->len))
1754 		goto out_extract_close;
1755 
1756 	sz = kcore__write(&extract);
1757 	if (sz < 0 || sz > offset)
1758 		goto out_extract_close;
1759 
1760 	if (copy_bytes(kcore.fd, kce->offs, extract.fd, offset, kce->len))
1761 		goto out_extract_close;
1762 
1763 	err = 0;
1764 
1765 out_extract_close:
1766 	kcore__close(&extract);
1767 	if (err)
1768 		unlink(kce->extract_filename);
1769 out_kcore_close:
1770 	kcore__close(&kcore);
1771 
1772 	return err;
1773 }
1774 
1775 void kcore_extract__delete(struct kcore_extract *kce)
1776 {
1777 	unlink(kce->extract_filename);
1778 }
1779 
1780 void symbol__elf_init(void)
1781 {
1782 	elf_version(EV_CURRENT);
1783 }
1784