xref: /openbmc/linux/tools/perf/util/symbol-elf.c (revision 78700c0a)
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 "demangle-java.h"
10 #include "machine.h"
11 #include "vdso.h"
12 #include <symbol/kallsyms.h>
13 #include "debug.h"
14 
15 #ifndef EM_AARCH64
16 #define EM_AARCH64	183  /* ARM 64 bit */
17 #endif
18 
19 
20 #ifdef HAVE_CPLUS_DEMANGLE_SUPPORT
21 extern char *cplus_demangle(const char *, int);
22 
23 static inline char *bfd_demangle(void __maybe_unused *v, const char *c, int i)
24 {
25 	return cplus_demangle(c, i);
26 }
27 #else
28 #ifdef NO_DEMANGLE
29 static inline char *bfd_demangle(void __maybe_unused *v,
30 				 const char __maybe_unused *c,
31 				 int __maybe_unused i)
32 {
33 	return NULL;
34 }
35 #else
36 #define PACKAGE 'perf'
37 #include <bfd.h>
38 #endif
39 #endif
40 
41 #ifndef HAVE_ELF_GETPHDRNUM_SUPPORT
42 static int elf_getphdrnum(Elf *elf, size_t *dst)
43 {
44 	GElf_Ehdr gehdr;
45 	GElf_Ehdr *ehdr;
46 
47 	ehdr = gelf_getehdr(elf, &gehdr);
48 	if (!ehdr)
49 		return -1;
50 
51 	*dst = ehdr->e_phnum;
52 
53 	return 0;
54 }
55 #endif
56 
57 #ifndef NT_GNU_BUILD_ID
58 #define NT_GNU_BUILD_ID 3
59 #endif
60 
61 /**
62  * elf_symtab__for_each_symbol - iterate thru all the symbols
63  *
64  * @syms: struct elf_symtab instance to iterate
65  * @idx: uint32_t idx
66  * @sym: GElf_Sym iterator
67  */
68 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
69 	for (idx = 0, gelf_getsym(syms, idx, &sym);\
70 	     idx < nr_syms; \
71 	     idx++, gelf_getsym(syms, idx, &sym))
72 
73 static inline uint8_t elf_sym__type(const GElf_Sym *sym)
74 {
75 	return GELF_ST_TYPE(sym->st_info);
76 }
77 
78 #ifndef STT_GNU_IFUNC
79 #define STT_GNU_IFUNC 10
80 #endif
81 
82 static inline int elf_sym__is_function(const GElf_Sym *sym)
83 {
84 	return (elf_sym__type(sym) == STT_FUNC ||
85 		elf_sym__type(sym) == STT_GNU_IFUNC) &&
86 	       sym->st_name != 0 &&
87 	       sym->st_shndx != SHN_UNDEF;
88 }
89 
90 static inline bool elf_sym__is_object(const GElf_Sym *sym)
91 {
92 	return elf_sym__type(sym) == STT_OBJECT &&
93 		sym->st_name != 0 &&
94 		sym->st_shndx != SHN_UNDEF;
95 }
96 
97 static inline int elf_sym__is_label(const GElf_Sym *sym)
98 {
99 	return elf_sym__type(sym) == STT_NOTYPE &&
100 		sym->st_name != 0 &&
101 		sym->st_shndx != SHN_UNDEF &&
102 		sym->st_shndx != SHN_ABS;
103 }
104 
105 static bool elf_sym__is_a(GElf_Sym *sym, enum map_type type)
106 {
107 	switch (type) {
108 	case MAP__FUNCTION:
109 		return elf_sym__is_function(sym);
110 	case MAP__VARIABLE:
111 		return elf_sym__is_object(sym);
112 	default:
113 		return false;
114 	}
115 }
116 
117 static inline const char *elf_sym__name(const GElf_Sym *sym,
118 					const Elf_Data *symstrs)
119 {
120 	return symstrs->d_buf + sym->st_name;
121 }
122 
123 static inline const char *elf_sec__name(const GElf_Shdr *shdr,
124 					const Elf_Data *secstrs)
125 {
126 	return secstrs->d_buf + shdr->sh_name;
127 }
128 
129 static inline int elf_sec__is_text(const GElf_Shdr *shdr,
130 					const Elf_Data *secstrs)
131 {
132 	return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
133 }
134 
135 static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
136 				    const Elf_Data *secstrs)
137 {
138 	return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
139 }
140 
141 static bool elf_sec__is_a(GElf_Shdr *shdr, Elf_Data *secstrs,
142 			  enum map_type type)
143 {
144 	switch (type) {
145 	case MAP__FUNCTION:
146 		return elf_sec__is_text(shdr, secstrs);
147 	case MAP__VARIABLE:
148 		return elf_sec__is_data(shdr, secstrs);
149 	default:
150 		return false;
151 	}
152 }
153 
154 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
155 {
156 	Elf_Scn *sec = NULL;
157 	GElf_Shdr shdr;
158 	size_t cnt = 1;
159 
160 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
161 		gelf_getshdr(sec, &shdr);
162 
163 		if ((addr >= shdr.sh_addr) &&
164 		    (addr < (shdr.sh_addr + shdr.sh_size)))
165 			return cnt;
166 
167 		++cnt;
168 	}
169 
170 	return -1;
171 }
172 
173 Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
174 			     GElf_Shdr *shp, const char *name, size_t *idx)
175 {
176 	Elf_Scn *sec = NULL;
177 	size_t cnt = 1;
178 
179 	/* Elf is corrupted/truncated, avoid calling elf_strptr. */
180 	if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL))
181 		return NULL;
182 
183 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
184 		char *str;
185 
186 		gelf_getshdr(sec, shp);
187 		str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
188 		if (str && !strcmp(name, str)) {
189 			if (idx)
190 				*idx = cnt;
191 			return sec;
192 		}
193 		++cnt;
194 	}
195 
196 	return NULL;
197 }
198 
199 #define elf_section__for_each_rel(reldata, pos, pos_mem, idx, nr_entries) \
200 	for (idx = 0, pos = gelf_getrel(reldata, 0, &pos_mem); \
201 	     idx < nr_entries; \
202 	     ++idx, pos = gelf_getrel(reldata, idx, &pos_mem))
203 
204 #define elf_section__for_each_rela(reldata, pos, pos_mem, idx, nr_entries) \
205 	for (idx = 0, pos = gelf_getrela(reldata, 0, &pos_mem); \
206 	     idx < nr_entries; \
207 	     ++idx, pos = gelf_getrela(reldata, idx, &pos_mem))
208 
209 /*
210  * We need to check if we have a .dynsym, so that we can handle the
211  * .plt, synthesizing its symbols, that aren't on the symtabs (be it
212  * .dynsym or .symtab).
213  * And always look at the original dso, not at debuginfo packages, that
214  * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
215  */
216 int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss, struct map *map,
217 				symbol_filter_t filter)
218 {
219 	uint32_t nr_rel_entries, idx;
220 	GElf_Sym sym;
221 	u64 plt_offset;
222 	GElf_Shdr shdr_plt;
223 	struct symbol *f;
224 	GElf_Shdr shdr_rel_plt, shdr_dynsym;
225 	Elf_Data *reldata, *syms, *symstrs;
226 	Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
227 	size_t dynsym_idx;
228 	GElf_Ehdr ehdr;
229 	char sympltname[1024];
230 	Elf *elf;
231 	int nr = 0, symidx, err = 0;
232 
233 	if (!ss->dynsym)
234 		return 0;
235 
236 	elf = ss->elf;
237 	ehdr = ss->ehdr;
238 
239 	scn_dynsym = ss->dynsym;
240 	shdr_dynsym = ss->dynshdr;
241 	dynsym_idx = ss->dynsym_idx;
242 
243 	if (scn_dynsym == NULL)
244 		goto out_elf_end;
245 
246 	scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
247 					  ".rela.plt", NULL);
248 	if (scn_plt_rel == NULL) {
249 		scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
250 						  ".rel.plt", NULL);
251 		if (scn_plt_rel == NULL)
252 			goto out_elf_end;
253 	}
254 
255 	err = -1;
256 
257 	if (shdr_rel_plt.sh_link != dynsym_idx)
258 		goto out_elf_end;
259 
260 	if (elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL) == NULL)
261 		goto out_elf_end;
262 
263 	/*
264 	 * Fetch the relocation section to find the idxes to the GOT
265 	 * and the symbols in the .dynsym they refer to.
266 	 */
267 	reldata = elf_getdata(scn_plt_rel, NULL);
268 	if (reldata == NULL)
269 		goto out_elf_end;
270 
271 	syms = elf_getdata(scn_dynsym, NULL);
272 	if (syms == NULL)
273 		goto out_elf_end;
274 
275 	scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
276 	if (scn_symstrs == NULL)
277 		goto out_elf_end;
278 
279 	symstrs = elf_getdata(scn_symstrs, NULL);
280 	if (symstrs == NULL)
281 		goto out_elf_end;
282 
283 	if (symstrs->d_size == 0)
284 		goto out_elf_end;
285 
286 	nr_rel_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
287 	plt_offset = shdr_plt.sh_offset;
288 
289 	if (shdr_rel_plt.sh_type == SHT_RELA) {
290 		GElf_Rela pos_mem, *pos;
291 
292 		elf_section__for_each_rela(reldata, pos, pos_mem, idx,
293 					   nr_rel_entries) {
294 			symidx = GELF_R_SYM(pos->r_info);
295 			plt_offset += shdr_plt.sh_entsize;
296 			gelf_getsym(syms, symidx, &sym);
297 			snprintf(sympltname, sizeof(sympltname),
298 				 "%s@plt", elf_sym__name(&sym, symstrs));
299 
300 			f = symbol__new(plt_offset, shdr_plt.sh_entsize,
301 					STB_GLOBAL, sympltname);
302 			if (!f)
303 				goto out_elf_end;
304 
305 			if (filter && filter(map, f))
306 				symbol__delete(f);
307 			else {
308 				symbols__insert(&dso->symbols[map->type], f);
309 				++nr;
310 			}
311 		}
312 	} else if (shdr_rel_plt.sh_type == SHT_REL) {
313 		GElf_Rel pos_mem, *pos;
314 		elf_section__for_each_rel(reldata, pos, pos_mem, idx,
315 					  nr_rel_entries) {
316 			symidx = GELF_R_SYM(pos->r_info);
317 			plt_offset += shdr_plt.sh_entsize;
318 			gelf_getsym(syms, symidx, &sym);
319 			snprintf(sympltname, sizeof(sympltname),
320 				 "%s@plt", elf_sym__name(&sym, symstrs));
321 
322 			f = symbol__new(plt_offset, shdr_plt.sh_entsize,
323 					STB_GLOBAL, sympltname);
324 			if (!f)
325 				goto out_elf_end;
326 
327 			if (filter && filter(map, f))
328 				symbol__delete(f);
329 			else {
330 				symbols__insert(&dso->symbols[map->type], f);
331 				++nr;
332 			}
333 		}
334 	}
335 
336 	err = 0;
337 out_elf_end:
338 	if (err == 0)
339 		return nr;
340 	pr_debug("%s: problems reading %s PLT info.\n",
341 		 __func__, dso->long_name);
342 	return 0;
343 }
344 
345 /*
346  * Align offset to 4 bytes as needed for note name and descriptor data.
347  */
348 #define NOTE_ALIGN(n) (((n) + 3) & -4U)
349 
350 static int elf_read_build_id(Elf *elf, void *bf, size_t size)
351 {
352 	int err = -1;
353 	GElf_Ehdr ehdr;
354 	GElf_Shdr shdr;
355 	Elf_Data *data;
356 	Elf_Scn *sec;
357 	Elf_Kind ek;
358 	void *ptr;
359 
360 	if (size < BUILD_ID_SIZE)
361 		goto out;
362 
363 	ek = elf_kind(elf);
364 	if (ek != ELF_K_ELF)
365 		goto out;
366 
367 	if (gelf_getehdr(elf, &ehdr) == NULL) {
368 		pr_err("%s: cannot get elf header.\n", __func__);
369 		goto out;
370 	}
371 
372 	/*
373 	 * Check following sections for notes:
374 	 *   '.note.gnu.build-id'
375 	 *   '.notes'
376 	 *   '.note' (VDSO specific)
377 	 */
378 	do {
379 		sec = elf_section_by_name(elf, &ehdr, &shdr,
380 					  ".note.gnu.build-id", NULL);
381 		if (sec)
382 			break;
383 
384 		sec = elf_section_by_name(elf, &ehdr, &shdr,
385 					  ".notes", NULL);
386 		if (sec)
387 			break;
388 
389 		sec = elf_section_by_name(elf, &ehdr, &shdr,
390 					  ".note", NULL);
391 		if (sec)
392 			break;
393 
394 		return err;
395 
396 	} while (0);
397 
398 	data = elf_getdata(sec, NULL);
399 	if (data == NULL)
400 		goto out;
401 
402 	ptr = data->d_buf;
403 	while (ptr < (data->d_buf + data->d_size)) {
404 		GElf_Nhdr *nhdr = ptr;
405 		size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
406 		       descsz = NOTE_ALIGN(nhdr->n_descsz);
407 		const char *name;
408 
409 		ptr += sizeof(*nhdr);
410 		name = ptr;
411 		ptr += namesz;
412 		if (nhdr->n_type == NT_GNU_BUILD_ID &&
413 		    nhdr->n_namesz == sizeof("GNU")) {
414 			if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
415 				size_t sz = min(size, descsz);
416 				memcpy(bf, ptr, sz);
417 				memset(bf + sz, 0, size - sz);
418 				err = descsz;
419 				break;
420 			}
421 		}
422 		ptr += descsz;
423 	}
424 
425 out:
426 	return err;
427 }
428 
429 int filename__read_build_id(const char *filename, void *bf, size_t size)
430 {
431 	int fd, err = -1;
432 	Elf *elf;
433 
434 	if (size < BUILD_ID_SIZE)
435 		goto out;
436 
437 	fd = open(filename, O_RDONLY);
438 	if (fd < 0)
439 		goto out;
440 
441 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
442 	if (elf == NULL) {
443 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
444 		goto out_close;
445 	}
446 
447 	err = elf_read_build_id(elf, bf, size);
448 
449 	elf_end(elf);
450 out_close:
451 	close(fd);
452 out:
453 	return err;
454 }
455 
456 int sysfs__read_build_id(const char *filename, void *build_id, size_t size)
457 {
458 	int fd, err = -1;
459 
460 	if (size < BUILD_ID_SIZE)
461 		goto out;
462 
463 	fd = open(filename, O_RDONLY);
464 	if (fd < 0)
465 		goto out;
466 
467 	while (1) {
468 		char bf[BUFSIZ];
469 		GElf_Nhdr nhdr;
470 		size_t namesz, descsz;
471 
472 		if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
473 			break;
474 
475 		namesz = NOTE_ALIGN(nhdr.n_namesz);
476 		descsz = NOTE_ALIGN(nhdr.n_descsz);
477 		if (nhdr.n_type == NT_GNU_BUILD_ID &&
478 		    nhdr.n_namesz == sizeof("GNU")) {
479 			if (read(fd, bf, namesz) != (ssize_t)namesz)
480 				break;
481 			if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
482 				size_t sz = min(descsz, size);
483 				if (read(fd, build_id, sz) == (ssize_t)sz) {
484 					memset(build_id + sz, 0, size - sz);
485 					err = 0;
486 					break;
487 				}
488 			} else if (read(fd, bf, descsz) != (ssize_t)descsz)
489 				break;
490 		} else {
491 			int n = namesz + descsz;
492 			if (read(fd, bf, n) != n)
493 				break;
494 		}
495 	}
496 	close(fd);
497 out:
498 	return err;
499 }
500 
501 int filename__read_debuglink(const char *filename, char *debuglink,
502 			     size_t size)
503 {
504 	int fd, err = -1;
505 	Elf *elf;
506 	GElf_Ehdr ehdr;
507 	GElf_Shdr shdr;
508 	Elf_Data *data;
509 	Elf_Scn *sec;
510 	Elf_Kind ek;
511 
512 	fd = open(filename, O_RDONLY);
513 	if (fd < 0)
514 		goto out;
515 
516 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
517 	if (elf == NULL) {
518 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
519 		goto out_close;
520 	}
521 
522 	ek = elf_kind(elf);
523 	if (ek != ELF_K_ELF)
524 		goto out_elf_end;
525 
526 	if (gelf_getehdr(elf, &ehdr) == NULL) {
527 		pr_err("%s: cannot get elf header.\n", __func__);
528 		goto out_elf_end;
529 	}
530 
531 	sec = elf_section_by_name(elf, &ehdr, &shdr,
532 				  ".gnu_debuglink", NULL);
533 	if (sec == NULL)
534 		goto out_elf_end;
535 
536 	data = elf_getdata(sec, NULL);
537 	if (data == NULL)
538 		goto out_elf_end;
539 
540 	/* the start of this section is a zero-terminated string */
541 	strncpy(debuglink, data->d_buf, size);
542 
543 	err = 0;
544 
545 out_elf_end:
546 	elf_end(elf);
547 out_close:
548 	close(fd);
549 out:
550 	return err;
551 }
552 
553 static int dso__swap_init(struct dso *dso, unsigned char eidata)
554 {
555 	static unsigned int const endian = 1;
556 
557 	dso->needs_swap = DSO_SWAP__NO;
558 
559 	switch (eidata) {
560 	case ELFDATA2LSB:
561 		/* We are big endian, DSO is little endian. */
562 		if (*(unsigned char const *)&endian != 1)
563 			dso->needs_swap = DSO_SWAP__YES;
564 		break;
565 
566 	case ELFDATA2MSB:
567 		/* We are little endian, DSO is big endian. */
568 		if (*(unsigned char const *)&endian != 0)
569 			dso->needs_swap = DSO_SWAP__YES;
570 		break;
571 
572 	default:
573 		pr_err("unrecognized DSO data encoding %d\n", eidata);
574 		return -EINVAL;
575 	}
576 
577 	return 0;
578 }
579 
580 static int decompress_kmodule(struct dso *dso, const char *name,
581 			      enum dso_binary_type type)
582 {
583 	int fd = -1;
584 	char tmpbuf[] = "/tmp/perf-kmod-XXXXXX";
585 	struct kmod_path m;
586 
587 	if (type != DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP &&
588 	    type != DSO_BINARY_TYPE__GUEST_KMODULE_COMP &&
589 	    type != DSO_BINARY_TYPE__BUILD_ID_CACHE)
590 		return -1;
591 
592 	if (type == DSO_BINARY_TYPE__BUILD_ID_CACHE)
593 		name = dso->long_name;
594 
595 	if (kmod_path__parse_ext(&m, name) || !m.comp)
596 		return -1;
597 
598 	fd = mkstemp(tmpbuf);
599 	if (fd < 0) {
600 		dso->load_errno = errno;
601 		goto out;
602 	}
603 
604 	if (!decompress_to_file(m.ext, name, fd)) {
605 		dso->load_errno = DSO_LOAD_ERRNO__DECOMPRESSION_FAILURE;
606 		close(fd);
607 		fd = -1;
608 	}
609 
610 	unlink(tmpbuf);
611 
612 out:
613 	free(m.ext);
614 	return fd;
615 }
616 
617 bool symsrc__possibly_runtime(struct symsrc *ss)
618 {
619 	return ss->dynsym || ss->opdsec;
620 }
621 
622 bool symsrc__has_symtab(struct symsrc *ss)
623 {
624 	return ss->symtab != NULL;
625 }
626 
627 void symsrc__destroy(struct symsrc *ss)
628 {
629 	zfree(&ss->name);
630 	elf_end(ss->elf);
631 	close(ss->fd);
632 }
633 
634 bool __weak elf__needs_adjust_symbols(GElf_Ehdr ehdr)
635 {
636 	return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL;
637 }
638 
639 int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
640 		 enum dso_binary_type type)
641 {
642 	int err = -1;
643 	GElf_Ehdr ehdr;
644 	Elf *elf;
645 	int fd;
646 
647 	if (dso__needs_decompress(dso)) {
648 		fd = decompress_kmodule(dso, name, type);
649 		if (fd < 0)
650 			return -1;
651 	} else {
652 		fd = open(name, O_RDONLY);
653 		if (fd < 0) {
654 			dso->load_errno = errno;
655 			return -1;
656 		}
657 	}
658 
659 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
660 	if (elf == NULL) {
661 		pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
662 		dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
663 		goto out_close;
664 	}
665 
666 	if (gelf_getehdr(elf, &ehdr) == NULL) {
667 		dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
668 		pr_debug("%s: cannot get elf header.\n", __func__);
669 		goto out_elf_end;
670 	}
671 
672 	if (dso__swap_init(dso, ehdr.e_ident[EI_DATA])) {
673 		dso->load_errno = DSO_LOAD_ERRNO__INTERNAL_ERROR;
674 		goto out_elf_end;
675 	}
676 
677 	/* Always reject images with a mismatched build-id: */
678 	if (dso->has_build_id) {
679 		u8 build_id[BUILD_ID_SIZE];
680 
681 		if (elf_read_build_id(elf, build_id, BUILD_ID_SIZE) < 0) {
682 			dso->load_errno = DSO_LOAD_ERRNO__CANNOT_READ_BUILDID;
683 			goto out_elf_end;
684 		}
685 
686 		if (!dso__build_id_equal(dso, build_id)) {
687 			pr_debug("%s: build id mismatch for %s.\n", __func__, name);
688 			dso->load_errno = DSO_LOAD_ERRNO__MISMATCHING_BUILDID;
689 			goto out_elf_end;
690 		}
691 	}
692 
693 	ss->is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
694 
695 	ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab",
696 			NULL);
697 	if (ss->symshdr.sh_type != SHT_SYMTAB)
698 		ss->symtab = NULL;
699 
700 	ss->dynsym_idx = 0;
701 	ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym",
702 			&ss->dynsym_idx);
703 	if (ss->dynshdr.sh_type != SHT_DYNSYM)
704 		ss->dynsym = NULL;
705 
706 	ss->opdidx = 0;
707 	ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd",
708 			&ss->opdidx);
709 	if (ss->opdshdr.sh_type != SHT_PROGBITS)
710 		ss->opdsec = NULL;
711 
712 	if (dso->kernel == DSO_TYPE_USER)
713 		ss->adjust_symbols = true;
714 	else
715 		ss->adjust_symbols = elf__needs_adjust_symbols(ehdr);
716 
717 	ss->name   = strdup(name);
718 	if (!ss->name) {
719 		dso->load_errno = errno;
720 		goto out_elf_end;
721 	}
722 
723 	ss->elf    = elf;
724 	ss->fd     = fd;
725 	ss->ehdr   = ehdr;
726 	ss->type   = type;
727 
728 	return 0;
729 
730 out_elf_end:
731 	elf_end(elf);
732 out_close:
733 	close(fd);
734 	return err;
735 }
736 
737 /**
738  * ref_reloc_sym_not_found - has kernel relocation symbol been found.
739  * @kmap: kernel maps and relocation reference symbol
740  *
741  * This function returns %true if we are dealing with the kernel maps and the
742  * relocation reference symbol has not yet been found.  Otherwise %false is
743  * returned.
744  */
745 static bool ref_reloc_sym_not_found(struct kmap *kmap)
746 {
747 	return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
748 	       !kmap->ref_reloc_sym->unrelocated_addr;
749 }
750 
751 /**
752  * ref_reloc - kernel relocation offset.
753  * @kmap: kernel maps and relocation reference symbol
754  *
755  * This function returns the offset of kernel addresses as determined by using
756  * the relocation reference symbol i.e. if the kernel has not been relocated
757  * then the return value is zero.
758  */
759 static u64 ref_reloc(struct kmap *kmap)
760 {
761 	if (kmap && kmap->ref_reloc_sym &&
762 	    kmap->ref_reloc_sym->unrelocated_addr)
763 		return kmap->ref_reloc_sym->addr -
764 		       kmap->ref_reloc_sym->unrelocated_addr;
765 	return 0;
766 }
767 
768 static bool want_demangle(bool is_kernel_sym)
769 {
770 	return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle;
771 }
772 
773 void __weak arch__sym_update(struct symbol *s __maybe_unused,
774 		GElf_Sym *sym __maybe_unused) { }
775 
776 int dso__load_sym(struct dso *dso, struct map *map,
777 		  struct symsrc *syms_ss, struct symsrc *runtime_ss,
778 		  symbol_filter_t filter, int kmodule)
779 {
780 	struct kmap *kmap = dso->kernel ? map__kmap(map) : NULL;
781 	struct map_groups *kmaps = kmap ? map__kmaps(map) : NULL;
782 	struct map *curr_map = map;
783 	struct dso *curr_dso = dso;
784 	Elf_Data *symstrs, *secstrs;
785 	uint32_t nr_syms;
786 	int err = -1;
787 	uint32_t idx;
788 	GElf_Ehdr ehdr;
789 	GElf_Shdr shdr;
790 	GElf_Shdr tshdr;
791 	Elf_Data *syms, *opddata = NULL;
792 	GElf_Sym sym;
793 	Elf_Scn *sec, *sec_strndx;
794 	Elf *elf;
795 	int nr = 0;
796 	bool remap_kernel = false, adjust_kernel_syms = false;
797 
798 	if (kmap && !kmaps)
799 		return -1;
800 
801 	dso->symtab_type = syms_ss->type;
802 	dso->is_64_bit = syms_ss->is_64_bit;
803 	dso->rel = syms_ss->ehdr.e_type == ET_REL;
804 
805 	/*
806 	 * Modules may already have symbols from kallsyms, but those symbols
807 	 * have the wrong values for the dso maps, so remove them.
808 	 */
809 	if (kmodule && syms_ss->symtab)
810 		symbols__delete(&dso->symbols[map->type]);
811 
812 	if (!syms_ss->symtab) {
813 		/*
814 		 * If the vmlinux is stripped, fail so we will fall back
815 		 * to using kallsyms. The vmlinux runtime symbols aren't
816 		 * of much use.
817 		 */
818 		if (dso->kernel)
819 			goto out_elf_end;
820 
821 		syms_ss->symtab  = syms_ss->dynsym;
822 		syms_ss->symshdr = syms_ss->dynshdr;
823 	}
824 
825 	elf = syms_ss->elf;
826 	ehdr = syms_ss->ehdr;
827 	sec = syms_ss->symtab;
828 	shdr = syms_ss->symshdr;
829 
830 	if (elf_section_by_name(elf, &ehdr, &tshdr, ".text", NULL))
831 		dso->text_offset = tshdr.sh_addr - tshdr.sh_offset;
832 
833 	if (runtime_ss->opdsec)
834 		opddata = elf_rawdata(runtime_ss->opdsec, NULL);
835 
836 	syms = elf_getdata(sec, NULL);
837 	if (syms == NULL)
838 		goto out_elf_end;
839 
840 	sec = elf_getscn(elf, shdr.sh_link);
841 	if (sec == NULL)
842 		goto out_elf_end;
843 
844 	symstrs = elf_getdata(sec, NULL);
845 	if (symstrs == NULL)
846 		goto out_elf_end;
847 
848 	sec_strndx = elf_getscn(runtime_ss->elf, runtime_ss->ehdr.e_shstrndx);
849 	if (sec_strndx == NULL)
850 		goto out_elf_end;
851 
852 	secstrs = elf_getdata(sec_strndx, NULL);
853 	if (secstrs == NULL)
854 		goto out_elf_end;
855 
856 	nr_syms = shdr.sh_size / shdr.sh_entsize;
857 
858 	memset(&sym, 0, sizeof(sym));
859 
860 	/*
861 	 * The kernel relocation symbol is needed in advance in order to adjust
862 	 * kernel maps correctly.
863 	 */
864 	if (ref_reloc_sym_not_found(kmap)) {
865 		elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
866 			const char *elf_name = elf_sym__name(&sym, symstrs);
867 
868 			if (strcmp(elf_name, kmap->ref_reloc_sym->name))
869 				continue;
870 			kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
871 			map->reloc = kmap->ref_reloc_sym->addr -
872 				     kmap->ref_reloc_sym->unrelocated_addr;
873 			break;
874 		}
875 	}
876 
877 	/*
878 	 * Handle any relocation of vdso necessary because older kernels
879 	 * attempted to prelink vdso to its virtual address.
880 	 */
881 	if (dso__is_vdso(dso))
882 		map->reloc = map->start - dso->text_offset;
883 
884 	dso->adjust_symbols = runtime_ss->adjust_symbols || ref_reloc(kmap);
885 	/*
886 	 * Initial kernel and module mappings do not map to the dso.  For
887 	 * function mappings, flag the fixups.
888 	 */
889 	if (map->type == MAP__FUNCTION && (dso->kernel || kmodule)) {
890 		remap_kernel = true;
891 		adjust_kernel_syms = dso->adjust_symbols;
892 	}
893 	elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
894 		struct symbol *f;
895 		const char *elf_name = elf_sym__name(&sym, symstrs);
896 		char *demangled = NULL;
897 		int is_label = elf_sym__is_label(&sym);
898 		const char *section_name;
899 		bool used_opd = false;
900 
901 		if (!is_label && !elf_sym__is_a(&sym, map->type))
902 			continue;
903 
904 		/* Reject ARM ELF "mapping symbols": these aren't unique and
905 		 * don't identify functions, so will confuse the profile
906 		 * output: */
907 		if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) {
908 			if (elf_name[0] == '$' && strchr("adtx", elf_name[1])
909 			    && (elf_name[2] == '\0' || elf_name[2] == '.'))
910 				continue;
911 		}
912 
913 		if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) {
914 			u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr;
915 			u64 *opd = opddata->d_buf + offset;
916 			sym.st_value = DSO__SWAP(dso, u64, *opd);
917 			sym.st_shndx = elf_addr_to_index(runtime_ss->elf,
918 					sym.st_value);
919 			used_opd = true;
920 		}
921 		/*
922 		 * When loading symbols in a data mapping, ABS symbols (which
923 		 * has a value of SHN_ABS in its st_shndx) failed at
924 		 * elf_getscn().  And it marks the loading as a failure so
925 		 * already loaded symbols cannot be fixed up.
926 		 *
927 		 * I'm not sure what should be done. Just ignore them for now.
928 		 * - Namhyung Kim
929 		 */
930 		if (sym.st_shndx == SHN_ABS)
931 			continue;
932 
933 		sec = elf_getscn(runtime_ss->elf, sym.st_shndx);
934 		if (!sec)
935 			goto out_elf_end;
936 
937 		gelf_getshdr(sec, &shdr);
938 
939 		if (is_label && !elf_sec__is_a(&shdr, secstrs, map->type))
940 			continue;
941 
942 		section_name = elf_sec__name(&shdr, secstrs);
943 
944 		/* On ARM, symbols for thumb functions have 1 added to
945 		 * the symbol address as a flag - remove it */
946 		if ((ehdr.e_machine == EM_ARM) &&
947 		    (map->type == MAP__FUNCTION) &&
948 		    (sym.st_value & 1))
949 			--sym.st_value;
950 
951 		if (dso->kernel || kmodule) {
952 			char dso_name[PATH_MAX];
953 
954 			/* Adjust symbol to map to file offset */
955 			if (adjust_kernel_syms)
956 				sym.st_value -= shdr.sh_addr - shdr.sh_offset;
957 
958 			if (strcmp(section_name,
959 				   (curr_dso->short_name +
960 				    dso->short_name_len)) == 0)
961 				goto new_symbol;
962 
963 			if (strcmp(section_name, ".text") == 0) {
964 				/*
965 				 * The initial kernel mapping is based on
966 				 * kallsyms and identity maps.  Overwrite it to
967 				 * map to the kernel dso.
968 				 */
969 				if (remap_kernel && dso->kernel) {
970 					remap_kernel = false;
971 					map->start = shdr.sh_addr +
972 						     ref_reloc(kmap);
973 					map->end = map->start + shdr.sh_size;
974 					map->pgoff = shdr.sh_offset;
975 					map->map_ip = map__map_ip;
976 					map->unmap_ip = map__unmap_ip;
977 					/* Ensure maps are correctly ordered */
978 					if (kmaps) {
979 						map__get(map);
980 						map_groups__remove(kmaps, map);
981 						map_groups__insert(kmaps, map);
982 						map__put(map);
983 					}
984 				}
985 
986 				/*
987 				 * The initial module mapping is based on
988 				 * /proc/modules mapped to offset zero.
989 				 * Overwrite it to map to the module dso.
990 				 */
991 				if (remap_kernel && kmodule) {
992 					remap_kernel = false;
993 					map->pgoff = shdr.sh_offset;
994 				}
995 
996 				curr_map = map;
997 				curr_dso = dso;
998 				goto new_symbol;
999 			}
1000 
1001 			if (!kmap)
1002 				goto new_symbol;
1003 
1004 			snprintf(dso_name, sizeof(dso_name),
1005 				 "%s%s", dso->short_name, section_name);
1006 
1007 			curr_map = map_groups__find_by_name(kmaps, map->type, dso_name);
1008 			if (curr_map == NULL) {
1009 				u64 start = sym.st_value;
1010 
1011 				if (kmodule)
1012 					start += map->start + shdr.sh_offset;
1013 
1014 				curr_dso = dso__new(dso_name);
1015 				if (curr_dso == NULL)
1016 					goto out_elf_end;
1017 				curr_dso->kernel = dso->kernel;
1018 				curr_dso->long_name = dso->long_name;
1019 				curr_dso->long_name_len = dso->long_name_len;
1020 				curr_map = map__new2(start, curr_dso,
1021 						     map->type);
1022 				dso__put(curr_dso);
1023 				if (curr_map == NULL) {
1024 					goto out_elf_end;
1025 				}
1026 				if (adjust_kernel_syms) {
1027 					curr_map->start = shdr.sh_addr +
1028 							  ref_reloc(kmap);
1029 					curr_map->end = curr_map->start +
1030 							shdr.sh_size;
1031 					curr_map->pgoff = shdr.sh_offset;
1032 				} else {
1033 					curr_map->map_ip = identity__map_ip;
1034 					curr_map->unmap_ip = identity__map_ip;
1035 				}
1036 				curr_dso->symtab_type = dso->symtab_type;
1037 				map_groups__insert(kmaps, curr_map);
1038 				/*
1039 				 * Add it before we drop the referece to curr_map,
1040 				 * i.e. while we still are sure to have a reference
1041 				 * to this DSO via curr_map->dso.
1042 				 */
1043 				dsos__add(&map->groups->machine->dsos, curr_dso);
1044 				/* kmaps already got it */
1045 				map__put(curr_map);
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 				demangled = java_demangle_sym(elf_name, JAVA_DEMANGLE_NORET);
1075 			if (demangled != NULL)
1076 				elf_name = demangled;
1077 		}
1078 		f = symbol__new(sym.st_value, sym.st_size,
1079 				GELF_ST_BIND(sym.st_info), elf_name);
1080 		free(demangled);
1081 		if (!f)
1082 			goto out_elf_end;
1083 
1084 		arch__sym_update(f, &sym);
1085 
1086 		if (filter && filter(curr_map, f))
1087 			symbol__delete(f);
1088 		else {
1089 			symbols__insert(&curr_dso->symbols[curr_map->type], f);
1090 			nr++;
1091 		}
1092 	}
1093 
1094 	/*
1095 	 * For misannotated, zeroed, ASM function sizes.
1096 	 */
1097 	if (nr > 0) {
1098 		if (!symbol_conf.allow_aliases)
1099 			symbols__fixup_duplicate(&dso->symbols[map->type]);
1100 		symbols__fixup_end(&dso->symbols[map->type]);
1101 		if (kmap) {
1102 			/*
1103 			 * We need to fixup this here too because we create new
1104 			 * maps here, for things like vsyscall sections.
1105 			 */
1106 			__map_groups__fixup_end(kmaps, map->type);
1107 		}
1108 	}
1109 	err = nr;
1110 out_elf_end:
1111 	return err;
1112 }
1113 
1114 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
1115 {
1116 	GElf_Phdr phdr;
1117 	size_t i, phdrnum;
1118 	int err;
1119 	u64 sz;
1120 
1121 	if (elf_getphdrnum(elf, &phdrnum))
1122 		return -1;
1123 
1124 	for (i = 0; i < phdrnum; i++) {
1125 		if (gelf_getphdr(elf, i, &phdr) == NULL)
1126 			return -1;
1127 		if (phdr.p_type != PT_LOAD)
1128 			continue;
1129 		if (exe) {
1130 			if (!(phdr.p_flags & PF_X))
1131 				continue;
1132 		} else {
1133 			if (!(phdr.p_flags & PF_R))
1134 				continue;
1135 		}
1136 		sz = min(phdr.p_memsz, phdr.p_filesz);
1137 		if (!sz)
1138 			continue;
1139 		err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
1140 		if (err)
1141 			return err;
1142 	}
1143 	return 0;
1144 }
1145 
1146 int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data,
1147 		    bool *is_64_bit)
1148 {
1149 	int err;
1150 	Elf *elf;
1151 
1152 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1153 	if (elf == NULL)
1154 		return -1;
1155 
1156 	if (is_64_bit)
1157 		*is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1158 
1159 	err = elf_read_maps(elf, exe, mapfn, data);
1160 
1161 	elf_end(elf);
1162 	return err;
1163 }
1164 
1165 enum dso_type dso__type_fd(int fd)
1166 {
1167 	enum dso_type dso_type = DSO__TYPE_UNKNOWN;
1168 	GElf_Ehdr ehdr;
1169 	Elf_Kind ek;
1170 	Elf *elf;
1171 
1172 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1173 	if (elf == NULL)
1174 		goto out;
1175 
1176 	ek = elf_kind(elf);
1177 	if (ek != ELF_K_ELF)
1178 		goto out_end;
1179 
1180 	if (gelf_getclass(elf) == ELFCLASS64) {
1181 		dso_type = DSO__TYPE_64BIT;
1182 		goto out_end;
1183 	}
1184 
1185 	if (gelf_getehdr(elf, &ehdr) == NULL)
1186 		goto out_end;
1187 
1188 	if (ehdr.e_machine == EM_X86_64)
1189 		dso_type = DSO__TYPE_X32BIT;
1190 	else
1191 		dso_type = DSO__TYPE_32BIT;
1192 out_end:
1193 	elf_end(elf);
1194 out:
1195 	return dso_type;
1196 }
1197 
1198 static int copy_bytes(int from, off_t from_offs, int to, off_t to_offs, u64 len)
1199 {
1200 	ssize_t r;
1201 	size_t n;
1202 	int err = -1;
1203 	char *buf = malloc(page_size);
1204 
1205 	if (buf == NULL)
1206 		return -1;
1207 
1208 	if (lseek(to, to_offs, SEEK_SET) != to_offs)
1209 		goto out;
1210 
1211 	if (lseek(from, from_offs, SEEK_SET) != from_offs)
1212 		goto out;
1213 
1214 	while (len) {
1215 		n = page_size;
1216 		if (len < n)
1217 			n = len;
1218 		/* Use read because mmap won't work on proc files */
1219 		r = read(from, buf, n);
1220 		if (r < 0)
1221 			goto out;
1222 		if (!r)
1223 			break;
1224 		n = r;
1225 		r = write(to, buf, n);
1226 		if (r < 0)
1227 			goto out;
1228 		if ((size_t)r != n)
1229 			goto out;
1230 		len -= n;
1231 	}
1232 
1233 	err = 0;
1234 out:
1235 	free(buf);
1236 	return err;
1237 }
1238 
1239 struct kcore {
1240 	int fd;
1241 	int elfclass;
1242 	Elf *elf;
1243 	GElf_Ehdr ehdr;
1244 };
1245 
1246 static int kcore__open(struct kcore *kcore, const char *filename)
1247 {
1248 	GElf_Ehdr *ehdr;
1249 
1250 	kcore->fd = open(filename, O_RDONLY);
1251 	if (kcore->fd == -1)
1252 		return -1;
1253 
1254 	kcore->elf = elf_begin(kcore->fd, ELF_C_READ, NULL);
1255 	if (!kcore->elf)
1256 		goto out_close;
1257 
1258 	kcore->elfclass = gelf_getclass(kcore->elf);
1259 	if (kcore->elfclass == ELFCLASSNONE)
1260 		goto out_end;
1261 
1262 	ehdr = gelf_getehdr(kcore->elf, &kcore->ehdr);
1263 	if (!ehdr)
1264 		goto out_end;
1265 
1266 	return 0;
1267 
1268 out_end:
1269 	elf_end(kcore->elf);
1270 out_close:
1271 	close(kcore->fd);
1272 	return -1;
1273 }
1274 
1275 static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
1276 		       bool temp)
1277 {
1278 	kcore->elfclass = elfclass;
1279 
1280 	if (temp)
1281 		kcore->fd = mkstemp(filename);
1282 	else
1283 		kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
1284 	if (kcore->fd == -1)
1285 		return -1;
1286 
1287 	kcore->elf = elf_begin(kcore->fd, ELF_C_WRITE, NULL);
1288 	if (!kcore->elf)
1289 		goto out_close;
1290 
1291 	if (!gelf_newehdr(kcore->elf, elfclass))
1292 		goto out_end;
1293 
1294 	memset(&kcore->ehdr, 0, sizeof(GElf_Ehdr));
1295 
1296 	return 0;
1297 
1298 out_end:
1299 	elf_end(kcore->elf);
1300 out_close:
1301 	close(kcore->fd);
1302 	unlink(filename);
1303 	return -1;
1304 }
1305 
1306 static void kcore__close(struct kcore *kcore)
1307 {
1308 	elf_end(kcore->elf);
1309 	close(kcore->fd);
1310 }
1311 
1312 static int kcore__copy_hdr(struct kcore *from, struct kcore *to, size_t count)
1313 {
1314 	GElf_Ehdr *ehdr = &to->ehdr;
1315 	GElf_Ehdr *kehdr = &from->ehdr;
1316 
1317 	memcpy(ehdr->e_ident, kehdr->e_ident, EI_NIDENT);
1318 	ehdr->e_type      = kehdr->e_type;
1319 	ehdr->e_machine   = kehdr->e_machine;
1320 	ehdr->e_version   = kehdr->e_version;
1321 	ehdr->e_entry     = 0;
1322 	ehdr->e_shoff     = 0;
1323 	ehdr->e_flags     = kehdr->e_flags;
1324 	ehdr->e_phnum     = count;
1325 	ehdr->e_shentsize = 0;
1326 	ehdr->e_shnum     = 0;
1327 	ehdr->e_shstrndx  = 0;
1328 
1329 	if (from->elfclass == ELFCLASS32) {
1330 		ehdr->e_phoff     = sizeof(Elf32_Ehdr);
1331 		ehdr->e_ehsize    = sizeof(Elf32_Ehdr);
1332 		ehdr->e_phentsize = sizeof(Elf32_Phdr);
1333 	} else {
1334 		ehdr->e_phoff     = sizeof(Elf64_Ehdr);
1335 		ehdr->e_ehsize    = sizeof(Elf64_Ehdr);
1336 		ehdr->e_phentsize = sizeof(Elf64_Phdr);
1337 	}
1338 
1339 	if (!gelf_update_ehdr(to->elf, ehdr))
1340 		return -1;
1341 
1342 	if (!gelf_newphdr(to->elf, count))
1343 		return -1;
1344 
1345 	return 0;
1346 }
1347 
1348 static int kcore__add_phdr(struct kcore *kcore, int idx, off_t offset,
1349 			   u64 addr, u64 len)
1350 {
1351 	GElf_Phdr phdr = {
1352 		.p_type		= PT_LOAD,
1353 		.p_flags	= PF_R | PF_W | PF_X,
1354 		.p_offset	= offset,
1355 		.p_vaddr	= addr,
1356 		.p_paddr	= 0,
1357 		.p_filesz	= len,
1358 		.p_memsz	= len,
1359 		.p_align	= page_size,
1360 	};
1361 
1362 	if (!gelf_update_phdr(kcore->elf, idx, &phdr))
1363 		return -1;
1364 
1365 	return 0;
1366 }
1367 
1368 static off_t kcore__write(struct kcore *kcore)
1369 {
1370 	return elf_update(kcore->elf, ELF_C_WRITE);
1371 }
1372 
1373 struct phdr_data {
1374 	off_t offset;
1375 	u64 addr;
1376 	u64 len;
1377 };
1378 
1379 struct kcore_copy_info {
1380 	u64 stext;
1381 	u64 etext;
1382 	u64 first_symbol;
1383 	u64 last_symbol;
1384 	u64 first_module;
1385 	u64 last_module_symbol;
1386 	struct phdr_data kernel_map;
1387 	struct phdr_data modules_map;
1388 };
1389 
1390 static int kcore_copy__process_kallsyms(void *arg, const char *name, char type,
1391 					u64 start)
1392 {
1393 	struct kcore_copy_info *kci = arg;
1394 
1395 	if (!symbol_type__is_a(type, MAP__FUNCTION))
1396 		return 0;
1397 
1398 	if (strchr(name, '[')) {
1399 		if (start > kci->last_module_symbol)
1400 			kci->last_module_symbol = start;
1401 		return 0;
1402 	}
1403 
1404 	if (!kci->first_symbol || start < kci->first_symbol)
1405 		kci->first_symbol = start;
1406 
1407 	if (!kci->last_symbol || start > kci->last_symbol)
1408 		kci->last_symbol = start;
1409 
1410 	if (!strcmp(name, "_stext")) {
1411 		kci->stext = start;
1412 		return 0;
1413 	}
1414 
1415 	if (!strcmp(name, "_etext")) {
1416 		kci->etext = start;
1417 		return 0;
1418 	}
1419 
1420 	return 0;
1421 }
1422 
1423 static int kcore_copy__parse_kallsyms(struct kcore_copy_info *kci,
1424 				      const char *dir)
1425 {
1426 	char kallsyms_filename[PATH_MAX];
1427 
1428 	scnprintf(kallsyms_filename, PATH_MAX, "%s/kallsyms", dir);
1429 
1430 	if (symbol__restricted_filename(kallsyms_filename, "/proc/kallsyms"))
1431 		return -1;
1432 
1433 	if (kallsyms__parse(kallsyms_filename, kci,
1434 			    kcore_copy__process_kallsyms) < 0)
1435 		return -1;
1436 
1437 	return 0;
1438 }
1439 
1440 static int kcore_copy__process_modules(void *arg,
1441 				       const char *name __maybe_unused,
1442 				       u64 start)
1443 {
1444 	struct kcore_copy_info *kci = arg;
1445 
1446 	if (!kci->first_module || start < kci->first_module)
1447 		kci->first_module = start;
1448 
1449 	return 0;
1450 }
1451 
1452 static int kcore_copy__parse_modules(struct kcore_copy_info *kci,
1453 				     const char *dir)
1454 {
1455 	char modules_filename[PATH_MAX];
1456 
1457 	scnprintf(modules_filename, PATH_MAX, "%s/modules", dir);
1458 
1459 	if (symbol__restricted_filename(modules_filename, "/proc/modules"))
1460 		return -1;
1461 
1462 	if (modules__parse(modules_filename, kci,
1463 			   kcore_copy__process_modules) < 0)
1464 		return -1;
1465 
1466 	return 0;
1467 }
1468 
1469 static void kcore_copy__map(struct phdr_data *p, u64 start, u64 end, u64 pgoff,
1470 			    u64 s, u64 e)
1471 {
1472 	if (p->addr || s < start || s >= end)
1473 		return;
1474 
1475 	p->addr = s;
1476 	p->offset = (s - start) + pgoff;
1477 	p->len = e < end ? e - s : end - s;
1478 }
1479 
1480 static int kcore_copy__read_map(u64 start, u64 len, u64 pgoff, void *data)
1481 {
1482 	struct kcore_copy_info *kci = data;
1483 	u64 end = start + len;
1484 
1485 	kcore_copy__map(&kci->kernel_map, start, end, pgoff, kci->stext,
1486 			kci->etext);
1487 
1488 	kcore_copy__map(&kci->modules_map, start, end, pgoff, kci->first_module,
1489 			kci->last_module_symbol);
1490 
1491 	return 0;
1492 }
1493 
1494 static int kcore_copy__read_maps(struct kcore_copy_info *kci, Elf *elf)
1495 {
1496 	if (elf_read_maps(elf, true, kcore_copy__read_map, kci) < 0)
1497 		return -1;
1498 
1499 	return 0;
1500 }
1501 
1502 static int kcore_copy__calc_maps(struct kcore_copy_info *kci, const char *dir,
1503 				 Elf *elf)
1504 {
1505 	if (kcore_copy__parse_kallsyms(kci, dir))
1506 		return -1;
1507 
1508 	if (kcore_copy__parse_modules(kci, dir))
1509 		return -1;
1510 
1511 	if (kci->stext)
1512 		kci->stext = round_down(kci->stext, page_size);
1513 	else
1514 		kci->stext = round_down(kci->first_symbol, page_size);
1515 
1516 	if (kci->etext) {
1517 		kci->etext = round_up(kci->etext, page_size);
1518 	} else if (kci->last_symbol) {
1519 		kci->etext = round_up(kci->last_symbol, page_size);
1520 		kci->etext += page_size;
1521 	}
1522 
1523 	kci->first_module = round_down(kci->first_module, page_size);
1524 
1525 	if (kci->last_module_symbol) {
1526 		kci->last_module_symbol = round_up(kci->last_module_symbol,
1527 						   page_size);
1528 		kci->last_module_symbol += page_size;
1529 	}
1530 
1531 	if (!kci->stext || !kci->etext)
1532 		return -1;
1533 
1534 	if (kci->first_module && !kci->last_module_symbol)
1535 		return -1;
1536 
1537 	return kcore_copy__read_maps(kci, elf);
1538 }
1539 
1540 static int kcore_copy__copy_file(const char *from_dir, const char *to_dir,
1541 				 const char *name)
1542 {
1543 	char from_filename[PATH_MAX];
1544 	char to_filename[PATH_MAX];
1545 
1546 	scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
1547 	scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
1548 
1549 	return copyfile_mode(from_filename, to_filename, 0400);
1550 }
1551 
1552 static int kcore_copy__unlink(const char *dir, const char *name)
1553 {
1554 	char filename[PATH_MAX];
1555 
1556 	scnprintf(filename, PATH_MAX, "%s/%s", dir, name);
1557 
1558 	return unlink(filename);
1559 }
1560 
1561 static int kcore_copy__compare_fds(int from, int to)
1562 {
1563 	char *buf_from;
1564 	char *buf_to;
1565 	ssize_t ret;
1566 	size_t len;
1567 	int err = -1;
1568 
1569 	buf_from = malloc(page_size);
1570 	buf_to = malloc(page_size);
1571 	if (!buf_from || !buf_to)
1572 		goto out;
1573 
1574 	while (1) {
1575 		/* Use read because mmap won't work on proc files */
1576 		ret = read(from, buf_from, page_size);
1577 		if (ret < 0)
1578 			goto out;
1579 
1580 		if (!ret)
1581 			break;
1582 
1583 		len = ret;
1584 
1585 		if (readn(to, buf_to, len) != (int)len)
1586 			goto out;
1587 
1588 		if (memcmp(buf_from, buf_to, len))
1589 			goto out;
1590 	}
1591 
1592 	err = 0;
1593 out:
1594 	free(buf_to);
1595 	free(buf_from);
1596 	return err;
1597 }
1598 
1599 static int kcore_copy__compare_files(const char *from_filename,
1600 				     const char *to_filename)
1601 {
1602 	int from, to, err = -1;
1603 
1604 	from = open(from_filename, O_RDONLY);
1605 	if (from < 0)
1606 		return -1;
1607 
1608 	to = open(to_filename, O_RDONLY);
1609 	if (to < 0)
1610 		goto out_close_from;
1611 
1612 	err = kcore_copy__compare_fds(from, to);
1613 
1614 	close(to);
1615 out_close_from:
1616 	close(from);
1617 	return err;
1618 }
1619 
1620 static int kcore_copy__compare_file(const char *from_dir, const char *to_dir,
1621 				    const char *name)
1622 {
1623 	char from_filename[PATH_MAX];
1624 	char to_filename[PATH_MAX];
1625 
1626 	scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
1627 	scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
1628 
1629 	return kcore_copy__compare_files(from_filename, to_filename);
1630 }
1631 
1632 /**
1633  * kcore_copy - copy kallsyms, modules and kcore from one directory to another.
1634  * @from_dir: from directory
1635  * @to_dir: to directory
1636  *
1637  * This function copies kallsyms, modules and kcore files from one directory to
1638  * another.  kallsyms and modules are copied entirely.  Only code segments are
1639  * copied from kcore.  It is assumed that two segments suffice: one for the
1640  * kernel proper and one for all the modules.  The code segments are determined
1641  * from kallsyms and modules files.  The kernel map starts at _stext or the
1642  * lowest function symbol, and ends at _etext or the highest function symbol.
1643  * The module map starts at the lowest module address and ends at the highest
1644  * module symbol.  Start addresses are rounded down to the nearest page.  End
1645  * addresses are rounded up to the nearest page.  An extra page is added to the
1646  * highest kernel symbol and highest module symbol to, hopefully, encompass that
1647  * symbol too.  Because it contains only code sections, the resulting kcore is
1648  * unusual.  One significant peculiarity is that the mapping (start -> pgoff)
1649  * is not the same for the kernel map and the modules map.  That happens because
1650  * the data is copied adjacently whereas the original kcore has gaps.  Finally,
1651  * kallsyms and modules files are compared with their copies to check that
1652  * modules have not been loaded or unloaded while the copies were taking place.
1653  *
1654  * Return: %0 on success, %-1 on failure.
1655  */
1656 int kcore_copy(const char *from_dir, const char *to_dir)
1657 {
1658 	struct kcore kcore;
1659 	struct kcore extract;
1660 	size_t count = 2;
1661 	int idx = 0, err = -1;
1662 	off_t offset = page_size, sz, modules_offset = 0;
1663 	struct kcore_copy_info kci = { .stext = 0, };
1664 	char kcore_filename[PATH_MAX];
1665 	char extract_filename[PATH_MAX];
1666 
1667 	if (kcore_copy__copy_file(from_dir, to_dir, "kallsyms"))
1668 		return -1;
1669 
1670 	if (kcore_copy__copy_file(from_dir, to_dir, "modules"))
1671 		goto out_unlink_kallsyms;
1672 
1673 	scnprintf(kcore_filename, PATH_MAX, "%s/kcore", from_dir);
1674 	scnprintf(extract_filename, PATH_MAX, "%s/kcore", to_dir);
1675 
1676 	if (kcore__open(&kcore, kcore_filename))
1677 		goto out_unlink_modules;
1678 
1679 	if (kcore_copy__calc_maps(&kci, from_dir, kcore.elf))
1680 		goto out_kcore_close;
1681 
1682 	if (kcore__init(&extract, extract_filename, kcore.elfclass, false))
1683 		goto out_kcore_close;
1684 
1685 	if (!kci.modules_map.addr)
1686 		count -= 1;
1687 
1688 	if (kcore__copy_hdr(&kcore, &extract, count))
1689 		goto out_extract_close;
1690 
1691 	if (kcore__add_phdr(&extract, idx++, offset, kci.kernel_map.addr,
1692 			    kci.kernel_map.len))
1693 		goto out_extract_close;
1694 
1695 	if (kci.modules_map.addr) {
1696 		modules_offset = offset + kci.kernel_map.len;
1697 		if (kcore__add_phdr(&extract, idx, modules_offset,
1698 				    kci.modules_map.addr, kci.modules_map.len))
1699 			goto out_extract_close;
1700 	}
1701 
1702 	sz = kcore__write(&extract);
1703 	if (sz < 0 || sz > offset)
1704 		goto out_extract_close;
1705 
1706 	if (copy_bytes(kcore.fd, kci.kernel_map.offset, extract.fd, offset,
1707 		       kci.kernel_map.len))
1708 		goto out_extract_close;
1709 
1710 	if (modules_offset && copy_bytes(kcore.fd, kci.modules_map.offset,
1711 					 extract.fd, modules_offset,
1712 					 kci.modules_map.len))
1713 		goto out_extract_close;
1714 
1715 	if (kcore_copy__compare_file(from_dir, to_dir, "modules"))
1716 		goto out_extract_close;
1717 
1718 	if (kcore_copy__compare_file(from_dir, to_dir, "kallsyms"))
1719 		goto out_extract_close;
1720 
1721 	err = 0;
1722 
1723 out_extract_close:
1724 	kcore__close(&extract);
1725 	if (err)
1726 		unlink(extract_filename);
1727 out_kcore_close:
1728 	kcore__close(&kcore);
1729 out_unlink_modules:
1730 	if (err)
1731 		kcore_copy__unlink(to_dir, "modules");
1732 out_unlink_kallsyms:
1733 	if (err)
1734 		kcore_copy__unlink(to_dir, "kallsyms");
1735 
1736 	return err;
1737 }
1738 
1739 int kcore_extract__create(struct kcore_extract *kce)
1740 {
1741 	struct kcore kcore;
1742 	struct kcore extract;
1743 	size_t count = 1;
1744 	int idx = 0, err = -1;
1745 	off_t offset = page_size, sz;
1746 
1747 	if (kcore__open(&kcore, kce->kcore_filename))
1748 		return -1;
1749 
1750 	strcpy(kce->extract_filename, PERF_KCORE_EXTRACT);
1751 	if (kcore__init(&extract, kce->extract_filename, kcore.elfclass, true))
1752 		goto out_kcore_close;
1753 
1754 	if (kcore__copy_hdr(&kcore, &extract, count))
1755 		goto out_extract_close;
1756 
1757 	if (kcore__add_phdr(&extract, idx, offset, kce->addr, kce->len))
1758 		goto out_extract_close;
1759 
1760 	sz = kcore__write(&extract);
1761 	if (sz < 0 || sz > offset)
1762 		goto out_extract_close;
1763 
1764 	if (copy_bytes(kcore.fd, kce->offs, extract.fd, offset, kce->len))
1765 		goto out_extract_close;
1766 
1767 	err = 0;
1768 
1769 out_extract_close:
1770 	kcore__close(&extract);
1771 	if (err)
1772 		unlink(kce->extract_filename);
1773 out_kcore_close:
1774 	kcore__close(&kcore);
1775 
1776 	return err;
1777 }
1778 
1779 void kcore_extract__delete(struct kcore_extract *kce)
1780 {
1781 	unlink(kce->extract_filename);
1782 }
1783 
1784 void symbol__elf_init(void)
1785 {
1786 	elf_version(EV_CURRENT);
1787 }
1788