xref: /linux/tools/perf/util/symbol-elf.c (revision 17e548405a81665fd14cee960db7d093d1396400)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <fcntl.h>
3 #include <stdio.h>
4 #include <errno.h>
5 #include <stdlib.h>
6 #include <string.h>
7 #include <unistd.h>
8 #include <inttypes.h>
9 
10 #include "compress.h"
11 #include "dso.h"
12 #include "map.h"
13 #include "maps.h"
14 #include "symbol.h"
15 #include "symsrc.h"
16 #include "demangle-cxx.h"
17 #include "demangle-ocaml.h"
18 #include "demangle-java.h"
19 #include "demangle-rust-v0.h"
20 #include "machine.h"
21 #include "vdso.h"
22 #include "debug.h"
23 #include "util/copyfile.h"
24 #include <linux/ctype.h>
25 #include <linux/kernel.h>
26 #include <linux/log2.h>
27 #include <linux/zalloc.h>
28 #include <linux/string.h>
29 #include <symbol/kallsyms.h>
30 #include <internal/lib.h>
31 
32 #ifdef HAVE_LIBBFD_SUPPORT
33 #define PACKAGE 'perf'
34 #include <bfd.h>
35 #endif
36 
37 #if defined(HAVE_LIBBFD_SUPPORT) || defined(HAVE_CPLUS_DEMANGLE_SUPPORT)
38 #ifndef DMGL_PARAMS
39 #define DMGL_PARAMS     (1 << 0)  /* Include function args */
40 #define DMGL_ANSI       (1 << 1)  /* Include const, volatile, etc */
41 #endif
42 #endif
43 
44 #ifndef EM_AARCH64
45 #define EM_AARCH64	183  /* ARM 64 bit */
46 #endif
47 
48 #ifndef EM_LOONGARCH
49 #define EM_LOONGARCH	258
50 #endif
51 
52 #ifndef ELF32_ST_VISIBILITY
53 #define ELF32_ST_VISIBILITY(o)	((o) & 0x03)
54 #endif
55 
56 /* For ELF64 the definitions are the same.  */
57 #ifndef ELF64_ST_VISIBILITY
58 #define ELF64_ST_VISIBILITY(o)	ELF32_ST_VISIBILITY (o)
59 #endif
60 
61 /* How to extract information held in the st_other field.  */
62 #ifndef GELF_ST_VISIBILITY
63 #define GELF_ST_VISIBILITY(val)	ELF64_ST_VISIBILITY (val)
64 #endif
65 
66 typedef Elf64_Nhdr GElf_Nhdr;
67 
68 
69 #ifndef HAVE_ELF_GETPHDRNUM_SUPPORT
70 static int elf_getphdrnum(Elf *elf, size_t *dst)
71 {
72 	GElf_Ehdr gehdr;
73 	GElf_Ehdr *ehdr;
74 
75 	ehdr = gelf_getehdr(elf, &gehdr);
76 	if (!ehdr)
77 		return -1;
78 
79 	*dst = ehdr->e_phnum;
80 
81 	return 0;
82 }
83 #endif
84 
85 #ifndef HAVE_ELF_GETSHDRSTRNDX_SUPPORT
86 static int elf_getshdrstrndx(Elf *elf __maybe_unused, size_t *dst __maybe_unused)
87 {
88 	pr_err("%s: update your libelf to > 0.140, this one lacks elf_getshdrstrndx().\n", __func__);
89 	return -1;
90 }
91 #endif
92 
93 #ifndef NT_GNU_BUILD_ID
94 #define NT_GNU_BUILD_ID 3
95 #endif
96 
97 /**
98  * elf_symtab__for_each_symbol - iterate thru all the symbols
99  *
100  * @syms: struct elf_symtab instance to iterate
101  * @idx: uint32_t idx
102  * @sym: GElf_Sym iterator
103  */
104 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
105 	for (idx = 0, gelf_getsym(syms, idx, &sym);\
106 	     idx < nr_syms; \
107 	     idx++, gelf_getsym(syms, idx, &sym))
108 
109 static inline uint8_t elf_sym__type(const GElf_Sym *sym)
110 {
111 	return GELF_ST_TYPE(sym->st_info);
112 }
113 
114 static inline uint8_t elf_sym__visibility(const GElf_Sym *sym)
115 {
116 	return GELF_ST_VISIBILITY(sym->st_other);
117 }
118 
119 #ifndef STT_GNU_IFUNC
120 #define STT_GNU_IFUNC 10
121 #endif
122 
123 static inline int elf_sym__is_function(const GElf_Sym *sym)
124 {
125 	return (elf_sym__type(sym) == STT_FUNC ||
126 		elf_sym__type(sym) == STT_GNU_IFUNC) &&
127 	       sym->st_name != 0 &&
128 	       sym->st_shndx != SHN_UNDEF;
129 }
130 
131 static inline bool elf_sym__is_object(const GElf_Sym *sym)
132 {
133 	return elf_sym__type(sym) == STT_OBJECT &&
134 		sym->st_name != 0 &&
135 		sym->st_shndx != SHN_UNDEF;
136 }
137 
138 static inline int elf_sym__is_label(const GElf_Sym *sym)
139 {
140 	return elf_sym__type(sym) == STT_NOTYPE &&
141 		sym->st_name != 0 &&
142 		sym->st_shndx != SHN_UNDEF &&
143 		sym->st_shndx != SHN_ABS &&
144 		elf_sym__visibility(sym) != STV_HIDDEN &&
145 		elf_sym__visibility(sym) != STV_INTERNAL;
146 }
147 
148 static bool elf_sym__filter(GElf_Sym *sym)
149 {
150 	return elf_sym__is_function(sym) || elf_sym__is_object(sym);
151 }
152 
153 static inline const char *elf_sym__name(const GElf_Sym *sym,
154 					const Elf_Data *symstrs)
155 {
156 	return symstrs->d_buf + sym->st_name;
157 }
158 
159 static inline const char *elf_sec__name(const GElf_Shdr *shdr,
160 					const Elf_Data *secstrs)
161 {
162 	return secstrs->d_buf + shdr->sh_name;
163 }
164 
165 static inline int elf_sec__is_text(const GElf_Shdr *shdr,
166 					const Elf_Data *secstrs)
167 {
168 	return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
169 }
170 
171 static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
172 				    const Elf_Data *secstrs)
173 {
174 	return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
175 }
176 
177 static bool elf_sec__filter(GElf_Shdr *shdr, Elf_Data *secstrs)
178 {
179 	return elf_sec__is_text(shdr, secstrs) ||
180 	       elf_sec__is_data(shdr, secstrs);
181 }
182 
183 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
184 {
185 	Elf_Scn *sec = NULL;
186 	GElf_Shdr shdr;
187 	size_t cnt = 1;
188 
189 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
190 		gelf_getshdr(sec, &shdr);
191 
192 		if ((addr >= shdr.sh_addr) &&
193 		    (addr < (shdr.sh_addr + shdr.sh_size)))
194 			return cnt;
195 
196 		++cnt;
197 	}
198 
199 	return -1;
200 }
201 
202 Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
203 			     GElf_Shdr *shp, const char *name, size_t *idx)
204 {
205 	Elf_Scn *sec = NULL;
206 	size_t cnt = 1;
207 
208 	/* ELF is corrupted/truncated, avoid calling elf_strptr. */
209 	if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL))
210 		return NULL;
211 
212 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
213 		char *str;
214 
215 		gelf_getshdr(sec, shp);
216 		str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
217 		if (str && !strcmp(name, str)) {
218 			if (idx)
219 				*idx = cnt;
220 			return sec;
221 		}
222 		++cnt;
223 	}
224 
225 	return NULL;
226 }
227 
228 bool filename__has_section(const char *filename, const char *sec)
229 {
230 	int fd;
231 	Elf *elf;
232 	GElf_Ehdr ehdr;
233 	GElf_Shdr shdr;
234 	bool found = false;
235 
236 	fd = open(filename, O_RDONLY);
237 	if (fd < 0)
238 		return false;
239 
240 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
241 	if (elf == NULL)
242 		goto out;
243 
244 	if (gelf_getehdr(elf, &ehdr) == NULL)
245 		goto elf_out;
246 
247 	found = !!elf_section_by_name(elf, &ehdr, &shdr, sec, NULL);
248 
249 elf_out:
250 	elf_end(elf);
251 out:
252 	close(fd);
253 	return found;
254 }
255 
256 static int elf_read_program_header(Elf *elf, u64 vaddr, GElf_Phdr *phdr)
257 {
258 	size_t i, phdrnum;
259 	u64 sz;
260 
261 	if (elf_getphdrnum(elf, &phdrnum))
262 		return -1;
263 
264 	for (i = 0; i < phdrnum; i++) {
265 		if (gelf_getphdr(elf, i, phdr) == NULL)
266 			return -1;
267 
268 		if (phdr->p_type != PT_LOAD)
269 			continue;
270 
271 		sz = max(phdr->p_memsz, phdr->p_filesz);
272 		if (!sz)
273 			continue;
274 
275 		if (vaddr >= phdr->p_vaddr && (vaddr < phdr->p_vaddr + sz))
276 			return 0;
277 	}
278 
279 	/* Not found any valid program header */
280 	return -1;
281 }
282 
283 static bool want_demangle(bool is_kernel_sym)
284 {
285 	return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle;
286 }
287 
288 /*
289  * Demangle C++ function signature, typically replaced by demangle-cxx.cpp
290  * version.
291  */
292 #ifndef HAVE_CXA_DEMANGLE_SUPPORT
293 char *cxx_demangle_sym(const char *str __maybe_unused, bool params __maybe_unused,
294 		       bool modifiers __maybe_unused)
295 {
296 #ifdef HAVE_LIBBFD_SUPPORT
297 	int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0);
298 
299 	return bfd_demangle(NULL, str, flags);
300 #elif defined(HAVE_CPLUS_DEMANGLE_SUPPORT)
301 	int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0);
302 
303 	return cplus_demangle(str, flags);
304 #else
305 	return NULL;
306 #endif
307 }
308 #endif /* !HAVE_CXA_DEMANGLE_SUPPORT */
309 
310 static char *demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
311 {
312 	struct demangle rust_demangle = {
313 		.style = DemangleStyleUnknown,
314 	};
315 	char *demangled = NULL;
316 
317 	/*
318 	 * We need to figure out if the object was created from C++ sources
319 	 * DWARF DW_compile_unit has this, but we don't always have access
320 	 * to it...
321 	 */
322 	if (!want_demangle((dso && dso__kernel(dso)) || kmodule))
323 		return demangled;
324 
325 	rust_demangle_demangle(elf_name, &rust_demangle);
326 	if (rust_demangle_is_known(&rust_demangle)) {
327 		/* A rust mangled name. */
328 		if (rust_demangle.mangled_len == 0)
329 			return demangled;
330 
331 		for (size_t buf_len = roundup_pow_of_two(rust_demangle.mangled_len * 2);
332 		     buf_len < 1024 * 1024; buf_len += 32) {
333 			char *tmp = realloc(demangled, buf_len);
334 
335 			if (!tmp) {
336 				/* Failure to grow output buffer, return what is there. */
337 				return demangled;
338 			}
339 			demangled = tmp;
340 			if (rust_demangle_display_demangle(&rust_demangle, demangled, buf_len,
341 							   /*alternate=*/true) == OverflowOk)
342 				return demangled;
343 		}
344 		/* Buffer exceeded sensible bounds, return what is there. */
345 		return demangled;
346 	}
347 
348 	demangled = cxx_demangle_sym(elf_name, verbose > 0, verbose > 0);
349 	if (demangled)
350 		return demangled;
351 
352 	demangled = ocaml_demangle_sym(elf_name);
353 	if (demangled)
354 		return demangled;
355 
356 	return java_demangle_sym(elf_name, JAVA_DEMANGLE_NORET);
357 }
358 
359 struct rel_info {
360 	u32		nr_entries;
361 	u32		*sorted;
362 	bool		is_rela;
363 	Elf_Data	*reldata;
364 	GElf_Rela	rela;
365 	GElf_Rel	rel;
366 };
367 
368 static u32 get_rel_symidx(struct rel_info *ri, u32 idx)
369 {
370 	idx = ri->sorted ? ri->sorted[idx] : idx;
371 	if (ri->is_rela) {
372 		gelf_getrela(ri->reldata, idx, &ri->rela);
373 		return GELF_R_SYM(ri->rela.r_info);
374 	}
375 	gelf_getrel(ri->reldata, idx, &ri->rel);
376 	return GELF_R_SYM(ri->rel.r_info);
377 }
378 
379 static u64 get_rel_offset(struct rel_info *ri, u32 x)
380 {
381 	if (ri->is_rela) {
382 		GElf_Rela rela;
383 
384 		gelf_getrela(ri->reldata, x, &rela);
385 		return rela.r_offset;
386 	} else {
387 		GElf_Rel rel;
388 
389 		gelf_getrel(ri->reldata, x, &rel);
390 		return rel.r_offset;
391 	}
392 }
393 
394 static int rel_cmp(const void *a, const void *b, void *r)
395 {
396 	struct rel_info *ri = r;
397 	u64 a_offset = get_rel_offset(ri, *(const u32 *)a);
398 	u64 b_offset = get_rel_offset(ri, *(const u32 *)b);
399 
400 	return a_offset < b_offset ? -1 : (a_offset > b_offset ? 1 : 0);
401 }
402 
403 static int sort_rel(struct rel_info *ri)
404 {
405 	size_t sz = sizeof(ri->sorted[0]);
406 	u32 i;
407 
408 	ri->sorted = calloc(ri->nr_entries, sz);
409 	if (!ri->sorted)
410 		return -1;
411 	for (i = 0; i < ri->nr_entries; i++)
412 		ri->sorted[i] = i;
413 	qsort_r(ri->sorted, ri->nr_entries, sz, rel_cmp, ri);
414 	return 0;
415 }
416 
417 /*
418  * For x86_64, the GNU linker is putting IFUNC information in the relocation
419  * addend.
420  */
421 static bool addend_may_be_ifunc(GElf_Ehdr *ehdr, struct rel_info *ri)
422 {
423 	return ehdr->e_machine == EM_X86_64 && ri->is_rela &&
424 	       GELF_R_TYPE(ri->rela.r_info) == R_X86_64_IRELATIVE;
425 }
426 
427 static bool get_ifunc_name(Elf *elf, struct dso *dso, GElf_Ehdr *ehdr,
428 			   struct rel_info *ri, char *buf, size_t buf_sz)
429 {
430 	u64 addr = ri->rela.r_addend;
431 	struct symbol *sym;
432 	GElf_Phdr phdr;
433 
434 	if (!addend_may_be_ifunc(ehdr, ri))
435 		return false;
436 
437 	if (elf_read_program_header(elf, addr, &phdr))
438 		return false;
439 
440 	addr -= phdr.p_vaddr - phdr.p_offset;
441 
442 	sym = dso__find_symbol_nocache(dso, addr);
443 
444 	/* Expecting the address to be an IFUNC or IFUNC alias */
445 	if (!sym || sym->start != addr || (sym->type != STT_GNU_IFUNC && !sym->ifunc_alias))
446 		return false;
447 
448 	snprintf(buf, buf_sz, "%s@plt", sym->name);
449 
450 	return true;
451 }
452 
453 static void exit_rel(struct rel_info *ri)
454 {
455 	zfree(&ri->sorted);
456 }
457 
458 static bool get_plt_sizes(struct dso *dso, GElf_Ehdr *ehdr, GElf_Shdr *shdr_plt,
459 			  u64 *plt_header_size, u64 *plt_entry_size)
460 {
461 	switch (ehdr->e_machine) {
462 	case EM_ARM:
463 		*plt_header_size = 20;
464 		*plt_entry_size = 12;
465 		return true;
466 	case EM_AARCH64:
467 		*plt_header_size = 32;
468 		*plt_entry_size = 16;
469 		return true;
470 	case EM_LOONGARCH:
471 		*plt_header_size = 32;
472 		*plt_entry_size = 16;
473 		return true;
474 	case EM_SPARC:
475 		*plt_header_size = 48;
476 		*plt_entry_size = 12;
477 		return true;
478 	case EM_SPARCV9:
479 		*plt_header_size = 128;
480 		*plt_entry_size = 32;
481 		return true;
482 	case EM_386:
483 	case EM_X86_64:
484 		*plt_entry_size = shdr_plt->sh_entsize;
485 		/* Size is 8 or 16, if not, assume alignment indicates size */
486 		if (*plt_entry_size != 8 && *plt_entry_size != 16)
487 			*plt_entry_size = shdr_plt->sh_addralign == 8 ? 8 : 16;
488 		*plt_header_size = *plt_entry_size;
489 		break;
490 	default: /* FIXME: s390/alpha/mips/parisc/poperpc/sh/xtensa need to be checked */
491 		*plt_header_size = shdr_plt->sh_entsize;
492 		*plt_entry_size = shdr_plt->sh_entsize;
493 		break;
494 	}
495 	if (*plt_entry_size)
496 		return true;
497 	pr_debug("Missing PLT entry size for %s\n", dso__long_name(dso));
498 	return false;
499 }
500 
501 static bool machine_is_x86(GElf_Half e_machine)
502 {
503 	return e_machine == EM_386 || e_machine == EM_X86_64;
504 }
505 
506 struct rela_dyn {
507 	GElf_Addr	offset;
508 	u32		sym_idx;
509 };
510 
511 struct rela_dyn_info {
512 	struct dso	*dso;
513 	Elf_Data	*plt_got_data;
514 	u32		nr_entries;
515 	struct rela_dyn	*sorted;
516 	Elf_Data	*dynsym_data;
517 	Elf_Data	*dynstr_data;
518 	Elf_Data	*rela_dyn_data;
519 };
520 
521 static void exit_rela_dyn(struct rela_dyn_info *di)
522 {
523 	zfree(&di->sorted);
524 }
525 
526 static int cmp_offset(const void *a, const void *b)
527 {
528 	const struct rela_dyn *va = a;
529 	const struct rela_dyn *vb = b;
530 
531 	return va->offset < vb->offset ? -1 : (va->offset > vb->offset ? 1 : 0);
532 }
533 
534 static int sort_rela_dyn(struct rela_dyn_info *di)
535 {
536 	u32 i, n;
537 
538 	di->sorted = calloc(di->nr_entries, sizeof(di->sorted[0]));
539 	if (!di->sorted)
540 		return -1;
541 
542 	/* Get data for sorting: the offset and symbol index */
543 	for (i = 0, n = 0; i < di->nr_entries; i++) {
544 		GElf_Rela rela;
545 		u32 sym_idx;
546 
547 		gelf_getrela(di->rela_dyn_data, i, &rela);
548 		sym_idx = GELF_R_SYM(rela.r_info);
549 		if (sym_idx) {
550 			di->sorted[n].sym_idx = sym_idx;
551 			di->sorted[n].offset = rela.r_offset;
552 			n += 1;
553 		}
554 	}
555 
556 	/* Sort by offset */
557 	di->nr_entries = n;
558 	qsort(di->sorted, n, sizeof(di->sorted[0]), cmp_offset);
559 
560 	return 0;
561 }
562 
563 static void get_rela_dyn_info(Elf *elf, GElf_Ehdr *ehdr, struct rela_dyn_info *di, Elf_Scn *scn)
564 {
565 	GElf_Shdr rela_dyn_shdr;
566 	GElf_Shdr shdr;
567 
568 	di->plt_got_data = elf_getdata(scn, NULL);
569 
570 	scn = elf_section_by_name(elf, ehdr, &rela_dyn_shdr, ".rela.dyn", NULL);
571 	if (!scn || !rela_dyn_shdr.sh_link || !rela_dyn_shdr.sh_entsize)
572 		return;
573 
574 	di->nr_entries = rela_dyn_shdr.sh_size / rela_dyn_shdr.sh_entsize;
575 	di->rela_dyn_data = elf_getdata(scn, NULL);
576 
577 	scn = elf_getscn(elf, rela_dyn_shdr.sh_link);
578 	if (!scn || !gelf_getshdr(scn, &shdr) || !shdr.sh_link)
579 		return;
580 
581 	di->dynsym_data = elf_getdata(scn, NULL);
582 	di->dynstr_data = elf_getdata(elf_getscn(elf, shdr.sh_link), NULL);
583 
584 	if (!di->plt_got_data || !di->dynstr_data || !di->dynsym_data || !di->rela_dyn_data)
585 		return;
586 
587 	/* Sort into offset order */
588 	sort_rela_dyn(di);
589 }
590 
591 /* Get instruction displacement from a plt entry for x86_64 */
592 static u32 get_x86_64_plt_disp(const u8 *p)
593 {
594 	u8 endbr64[] = {0xf3, 0x0f, 0x1e, 0xfa};
595 	int n = 0;
596 
597 	/* Skip endbr64 */
598 	if (!memcmp(p, endbr64, sizeof(endbr64)))
599 		n += sizeof(endbr64);
600 	/* Skip bnd prefix */
601 	if (p[n] == 0xf2)
602 		n += 1;
603 	/* jmp with 4-byte displacement */
604 	if (p[n] == 0xff && p[n + 1] == 0x25) {
605 		u32 disp;
606 
607 		n += 2;
608 		/* Also add offset from start of entry to end of instruction */
609 		memcpy(&disp, p + n, sizeof(disp));
610 		return n + 4 + le32toh(disp);
611 	}
612 	return 0;
613 }
614 
615 static bool get_plt_got_name(GElf_Shdr *shdr, size_t i,
616 			     struct rela_dyn_info *di,
617 			     char *buf, size_t buf_sz)
618 {
619 	struct rela_dyn vi, *vr;
620 	const char *sym_name;
621 	char *demangled;
622 	GElf_Sym sym;
623 	bool result;
624 	u32 disp;
625 
626 	if (!di->sorted)
627 		return false;
628 
629 	disp = get_x86_64_plt_disp(di->plt_got_data->d_buf + i);
630 	if (!disp)
631 		return false;
632 
633 	/* Compute target offset of the .plt.got entry */
634 	vi.offset = shdr->sh_offset + di->plt_got_data->d_off + i + disp;
635 
636 	/* Find that offset in .rela.dyn (sorted by offset) */
637 	vr = bsearch(&vi, di->sorted, di->nr_entries, sizeof(di->sorted[0]), cmp_offset);
638 	if (!vr)
639 		return false;
640 
641 	/* Get the associated symbol */
642 	gelf_getsym(di->dynsym_data, vr->sym_idx, &sym);
643 	sym_name = elf_sym__name(&sym, di->dynstr_data);
644 	demangled = demangle_sym(di->dso, 0, sym_name);
645 	if (demangled != NULL)
646 		sym_name = demangled;
647 
648 	snprintf(buf, buf_sz, "%s@plt", sym_name);
649 
650 	result = *sym_name;
651 
652 	free(demangled);
653 
654 	return result;
655 }
656 
657 static int dso__synthesize_plt_got_symbols(struct dso *dso, Elf *elf,
658 					   GElf_Ehdr *ehdr,
659 					   char *buf, size_t buf_sz)
660 {
661 	struct rela_dyn_info di = { .dso = dso };
662 	struct symbol *sym;
663 	GElf_Shdr shdr;
664 	Elf_Scn *scn;
665 	int err = -1;
666 	size_t i;
667 
668 	scn = elf_section_by_name(elf, ehdr, &shdr, ".plt.got", NULL);
669 	if (!scn || !shdr.sh_entsize)
670 		return 0;
671 
672 	if (ehdr->e_machine == EM_X86_64)
673 		get_rela_dyn_info(elf, ehdr, &di, scn);
674 
675 	for (i = 0; i < shdr.sh_size; i += shdr.sh_entsize) {
676 		if (!get_plt_got_name(&shdr, i, &di, buf, buf_sz))
677 			snprintf(buf, buf_sz, "offset_%#" PRIx64 "@plt", (u64)shdr.sh_offset + i);
678 		sym = symbol__new(shdr.sh_offset + i, shdr.sh_entsize, STB_GLOBAL, STT_FUNC, buf);
679 		if (!sym)
680 			goto out;
681 		symbols__insert(dso__symbols(dso), sym);
682 	}
683 	err = 0;
684 out:
685 	exit_rela_dyn(&di);
686 	return err;
687 }
688 
689 /*
690  * We need to check if we have a .dynsym, so that we can handle the
691  * .plt, synthesizing its symbols, that aren't on the symtabs (be it
692  * .dynsym or .symtab).
693  * And always look at the original dso, not at debuginfo packages, that
694  * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
695  */
696 int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss)
697 {
698 	uint32_t idx;
699 	GElf_Sym sym;
700 	u64 plt_offset, plt_header_size, plt_entry_size;
701 	GElf_Shdr shdr_plt, plt_sec_shdr;
702 	struct symbol *f, *plt_sym;
703 	GElf_Shdr shdr_rel_plt, shdr_dynsym;
704 	Elf_Data *syms, *symstrs;
705 	Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
706 	GElf_Ehdr ehdr;
707 	char sympltname[1024];
708 	Elf *elf;
709 	int nr = 0, err = -1;
710 	struct rel_info ri = { .is_rela = false };
711 	bool lazy_plt;
712 
713 	elf = ss->elf;
714 	ehdr = ss->ehdr;
715 
716 	if (!elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL))
717 		return 0;
718 
719 	/*
720 	 * A symbol from a previous section (e.g. .init) can have been expanded
721 	 * by symbols__fixup_end() to overlap .plt. Truncate it before adding
722 	 * a symbol for .plt header.
723 	 */
724 	f = dso__find_symbol_nocache(dso, shdr_plt.sh_offset);
725 	if (f && f->start < shdr_plt.sh_offset && f->end > shdr_plt.sh_offset)
726 		f->end = shdr_plt.sh_offset;
727 
728 	if (!get_plt_sizes(dso, &ehdr, &shdr_plt, &plt_header_size, &plt_entry_size))
729 		return 0;
730 
731 	/* Add a symbol for .plt header */
732 	plt_sym = symbol__new(shdr_plt.sh_offset, plt_header_size, STB_GLOBAL, STT_FUNC, ".plt");
733 	if (!plt_sym)
734 		goto out_elf_end;
735 	symbols__insert(dso__symbols(dso), plt_sym);
736 
737 	/* Only x86 has .plt.got */
738 	if (machine_is_x86(ehdr.e_machine) &&
739 	    dso__synthesize_plt_got_symbols(dso, elf, &ehdr, sympltname, sizeof(sympltname)))
740 		goto out_elf_end;
741 
742 	/* Only x86 has .plt.sec */
743 	if (machine_is_x86(ehdr.e_machine) &&
744 	    elf_section_by_name(elf, &ehdr, &plt_sec_shdr, ".plt.sec", NULL)) {
745 		if (!get_plt_sizes(dso, &ehdr, &plt_sec_shdr, &plt_header_size, &plt_entry_size))
746 			return 0;
747 		/* Extend .plt symbol to entire .plt */
748 		plt_sym->end = plt_sym->start + shdr_plt.sh_size;
749 		/* Use .plt.sec offset */
750 		plt_offset = plt_sec_shdr.sh_offset;
751 		lazy_plt = false;
752 	} else {
753 		plt_offset = shdr_plt.sh_offset;
754 		lazy_plt = true;
755 	}
756 
757 	scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
758 					  ".rela.plt", NULL);
759 	if (scn_plt_rel == NULL) {
760 		scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
761 						  ".rel.plt", NULL);
762 		if (scn_plt_rel == NULL)
763 			return 0;
764 	}
765 
766 	if (shdr_rel_plt.sh_type != SHT_RELA &&
767 	    shdr_rel_plt.sh_type != SHT_REL)
768 		return 0;
769 
770 	if (!shdr_rel_plt.sh_link)
771 		return 0;
772 
773 	if (shdr_rel_plt.sh_link == ss->dynsym_idx) {
774 		scn_dynsym = ss->dynsym;
775 		shdr_dynsym = ss->dynshdr;
776 	} else if (shdr_rel_plt.sh_link == ss->symtab_idx) {
777 		/*
778 		 * A static executable can have a .plt due to IFUNCs, in which
779 		 * case .symtab is used not .dynsym.
780 		 */
781 		scn_dynsym = ss->symtab;
782 		shdr_dynsym = ss->symshdr;
783 	} else {
784 		goto out_elf_end;
785 	}
786 
787 	if (!scn_dynsym)
788 		return 0;
789 
790 	/*
791 	 * Fetch the relocation section to find the idxes to the GOT
792 	 * and the symbols in the .dynsym they refer to.
793 	 */
794 	ri.reldata = elf_getdata(scn_plt_rel, NULL);
795 	if (!ri.reldata)
796 		goto out_elf_end;
797 
798 	syms = elf_getdata(scn_dynsym, NULL);
799 	if (syms == NULL)
800 		goto out_elf_end;
801 
802 	scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
803 	if (scn_symstrs == NULL)
804 		goto out_elf_end;
805 
806 	symstrs = elf_getdata(scn_symstrs, NULL);
807 	if (symstrs == NULL)
808 		goto out_elf_end;
809 
810 	if (symstrs->d_size == 0)
811 		goto out_elf_end;
812 
813 	ri.nr_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
814 
815 	ri.is_rela = shdr_rel_plt.sh_type == SHT_RELA;
816 
817 	if (lazy_plt) {
818 		/*
819 		 * Assume a .plt with the same number of entries as the number
820 		 * of relocation entries is not lazy and does not have a header.
821 		 */
822 		if (ri.nr_entries * plt_entry_size == shdr_plt.sh_size)
823 			dso__delete_symbol(dso, plt_sym);
824 		else
825 			plt_offset += plt_header_size;
826 	}
827 
828 	/*
829 	 * x86 doesn't insert IFUNC relocations in .plt order, so sort to get
830 	 * back in order.
831 	 */
832 	if (machine_is_x86(ehdr.e_machine) && sort_rel(&ri))
833 		goto out_elf_end;
834 
835 	for (idx = 0; idx < ri.nr_entries; idx++) {
836 		const char *elf_name = NULL;
837 		char *demangled = NULL;
838 
839 		gelf_getsym(syms, get_rel_symidx(&ri, idx), &sym);
840 
841 		elf_name = elf_sym__name(&sym, symstrs);
842 		demangled = demangle_sym(dso, 0, elf_name);
843 		if (demangled)
844 			elf_name = demangled;
845 		if (*elf_name)
846 			snprintf(sympltname, sizeof(sympltname), "%s@plt", elf_name);
847 		else if (!get_ifunc_name(elf, dso, &ehdr, &ri, sympltname, sizeof(sympltname)))
848 			snprintf(sympltname, sizeof(sympltname),
849 				 "offset_%#" PRIx64 "@plt", plt_offset);
850 		free(demangled);
851 
852 		f = symbol__new(plt_offset, plt_entry_size, STB_GLOBAL, STT_FUNC, sympltname);
853 		if (!f)
854 			goto out_elf_end;
855 
856 		plt_offset += plt_entry_size;
857 		symbols__insert(dso__symbols(dso), f);
858 		++nr;
859 	}
860 
861 	err = 0;
862 out_elf_end:
863 	exit_rel(&ri);
864 	if (err == 0)
865 		return nr;
866 	pr_debug("%s: problems reading %s PLT info.\n",
867 		 __func__, dso__long_name(dso));
868 	return 0;
869 }
870 
871 char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
872 {
873 	return demangle_sym(dso, kmodule, elf_name);
874 }
875 
876 /*
877  * Align offset to 4 bytes as needed for note name and descriptor data.
878  */
879 #define NOTE_ALIGN(n) (((n) + 3) & -4U)
880 
881 static int elf_read_build_id(Elf *elf, void *bf, size_t size)
882 {
883 	int err = -1;
884 	GElf_Ehdr ehdr;
885 	GElf_Shdr shdr;
886 	Elf_Data *data;
887 	Elf_Scn *sec;
888 	Elf_Kind ek;
889 	void *ptr;
890 
891 	if (size < BUILD_ID_SIZE)
892 		goto out;
893 
894 	ek = elf_kind(elf);
895 	if (ek != ELF_K_ELF)
896 		goto out;
897 
898 	if (gelf_getehdr(elf, &ehdr) == NULL) {
899 		pr_err("%s: cannot get elf header.\n", __func__);
900 		goto out;
901 	}
902 
903 	/*
904 	 * Check following sections for notes:
905 	 *   '.note.gnu.build-id'
906 	 *   '.notes'
907 	 *   '.note' (VDSO specific)
908 	 */
909 	do {
910 		sec = elf_section_by_name(elf, &ehdr, &shdr,
911 					  ".note.gnu.build-id", NULL);
912 		if (sec)
913 			break;
914 
915 		sec = elf_section_by_name(elf, &ehdr, &shdr,
916 					  ".notes", NULL);
917 		if (sec)
918 			break;
919 
920 		sec = elf_section_by_name(elf, &ehdr, &shdr,
921 					  ".note", NULL);
922 		if (sec)
923 			break;
924 
925 		return err;
926 
927 	} while (0);
928 
929 	data = elf_getdata(sec, NULL);
930 	if (data == NULL)
931 		goto out;
932 
933 	ptr = data->d_buf;
934 	while (ptr < (data->d_buf + data->d_size)) {
935 		GElf_Nhdr *nhdr = ptr;
936 		size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
937 		       descsz = NOTE_ALIGN(nhdr->n_descsz);
938 		const char *name;
939 
940 		ptr += sizeof(*nhdr);
941 		name = ptr;
942 		ptr += namesz;
943 		if (nhdr->n_type == NT_GNU_BUILD_ID &&
944 		    nhdr->n_namesz == sizeof("GNU")) {
945 			if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
946 				size_t sz = min(size, descsz);
947 				memcpy(bf, ptr, sz);
948 				memset(bf + sz, 0, size - sz);
949 				err = sz;
950 				break;
951 			}
952 		}
953 		ptr += descsz;
954 	}
955 
956 out:
957 	return err;
958 }
959 
960 #ifdef HAVE_LIBBFD_BUILDID_SUPPORT
961 
962 static int read_build_id(const char *filename, struct build_id *bid)
963 {
964 	size_t size = sizeof(bid->data);
965 	int err = -1;
966 	bfd *abfd;
967 
968 	abfd = bfd_openr(filename, NULL);
969 	if (!abfd)
970 		return -1;
971 
972 	if (!bfd_check_format(abfd, bfd_object)) {
973 		pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
974 		goto out_close;
975 	}
976 
977 	if (!abfd->build_id || abfd->build_id->size > size)
978 		goto out_close;
979 
980 	memcpy(bid->data, abfd->build_id->data, abfd->build_id->size);
981 	memset(bid->data + abfd->build_id->size, 0, size - abfd->build_id->size);
982 	err = bid->size = abfd->build_id->size;
983 
984 out_close:
985 	bfd_close(abfd);
986 	return err;
987 }
988 
989 #else // HAVE_LIBBFD_BUILDID_SUPPORT
990 
991 static int read_build_id(const char *filename, struct build_id *bid)
992 {
993 	size_t size = sizeof(bid->data);
994 	int fd, err = -1;
995 	Elf *elf;
996 
997 	if (size < BUILD_ID_SIZE)
998 		goto out;
999 
1000 	fd = open(filename, O_RDONLY);
1001 	if (fd < 0)
1002 		goto out;
1003 
1004 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1005 	if (elf == NULL) {
1006 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
1007 		goto out_close;
1008 	}
1009 
1010 	err = elf_read_build_id(elf, bid->data, size);
1011 	if (err > 0)
1012 		bid->size = err;
1013 
1014 	elf_end(elf);
1015 out_close:
1016 	close(fd);
1017 out:
1018 	return err;
1019 }
1020 
1021 #endif // HAVE_LIBBFD_BUILDID_SUPPORT
1022 
1023 int filename__read_build_id(const char *filename, struct build_id *bid)
1024 {
1025 	struct kmod_path m = { .name = NULL, };
1026 	char path[PATH_MAX];
1027 	int err;
1028 
1029 	if (!filename)
1030 		return -EFAULT;
1031 
1032 	err = kmod_path__parse(&m, filename);
1033 	if (err)
1034 		return -1;
1035 
1036 	if (m.comp) {
1037 		int error = 0, fd;
1038 
1039 		fd = filename__decompress(filename, path, sizeof(path), m.comp, &error);
1040 		if (fd < 0) {
1041 			pr_debug("Failed to decompress (error %d) %s\n",
1042 				 error, filename);
1043 			return -1;
1044 		}
1045 		close(fd);
1046 		filename = path;
1047 	}
1048 
1049 	err = read_build_id(filename, bid);
1050 
1051 	if (m.comp)
1052 		unlink(filename);
1053 	return err;
1054 }
1055 
1056 int sysfs__read_build_id(const char *filename, struct build_id *bid)
1057 {
1058 	size_t size = sizeof(bid->data);
1059 	int fd, err = -1;
1060 
1061 	fd = open(filename, O_RDONLY);
1062 	if (fd < 0)
1063 		goto out;
1064 
1065 	while (1) {
1066 		char bf[BUFSIZ];
1067 		GElf_Nhdr nhdr;
1068 		size_t namesz, descsz;
1069 
1070 		if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
1071 			break;
1072 
1073 		namesz = NOTE_ALIGN(nhdr.n_namesz);
1074 		descsz = NOTE_ALIGN(nhdr.n_descsz);
1075 		if (nhdr.n_type == NT_GNU_BUILD_ID &&
1076 		    nhdr.n_namesz == sizeof("GNU")) {
1077 			if (read(fd, bf, namesz) != (ssize_t)namesz)
1078 				break;
1079 			if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
1080 				size_t sz = min(descsz, size);
1081 				if (read(fd, bid->data, sz) == (ssize_t)sz) {
1082 					memset(bid->data + sz, 0, size - sz);
1083 					bid->size = sz;
1084 					err = 0;
1085 					break;
1086 				}
1087 			} else if (read(fd, bf, descsz) != (ssize_t)descsz)
1088 				break;
1089 		} else {
1090 			int n = namesz + descsz;
1091 
1092 			if (n > (int)sizeof(bf)) {
1093 				n = sizeof(bf);
1094 				pr_debug("%s: truncating reading of build id in sysfs file %s: n_namesz=%u, n_descsz=%u.\n",
1095 					 __func__, filename, nhdr.n_namesz, nhdr.n_descsz);
1096 			}
1097 			if (read(fd, bf, n) != n)
1098 				break;
1099 		}
1100 	}
1101 	close(fd);
1102 out:
1103 	return err;
1104 }
1105 
1106 #ifdef HAVE_LIBBFD_SUPPORT
1107 
1108 int filename__read_debuglink(const char *filename, char *debuglink,
1109 			     size_t size)
1110 {
1111 	int err = -1;
1112 	asection *section;
1113 	bfd *abfd;
1114 
1115 	abfd = bfd_openr(filename, NULL);
1116 	if (!abfd)
1117 		return -1;
1118 
1119 	if (!bfd_check_format(abfd, bfd_object)) {
1120 		pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
1121 		goto out_close;
1122 	}
1123 
1124 	section = bfd_get_section_by_name(abfd, ".gnu_debuglink");
1125 	if (!section)
1126 		goto out_close;
1127 
1128 	if (section->size > size)
1129 		goto out_close;
1130 
1131 	if (!bfd_get_section_contents(abfd, section, debuglink, 0,
1132 				      section->size))
1133 		goto out_close;
1134 
1135 	err = 0;
1136 
1137 out_close:
1138 	bfd_close(abfd);
1139 	return err;
1140 }
1141 
1142 #else
1143 
1144 int filename__read_debuglink(const char *filename, char *debuglink,
1145 			     size_t size)
1146 {
1147 	int fd, err = -1;
1148 	Elf *elf;
1149 	GElf_Ehdr ehdr;
1150 	GElf_Shdr shdr;
1151 	Elf_Data *data;
1152 	Elf_Scn *sec;
1153 	Elf_Kind ek;
1154 
1155 	fd = open(filename, O_RDONLY);
1156 	if (fd < 0)
1157 		goto out;
1158 
1159 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1160 	if (elf == NULL) {
1161 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
1162 		goto out_close;
1163 	}
1164 
1165 	ek = elf_kind(elf);
1166 	if (ek != ELF_K_ELF)
1167 		goto out_elf_end;
1168 
1169 	if (gelf_getehdr(elf, &ehdr) == NULL) {
1170 		pr_err("%s: cannot get elf header.\n", __func__);
1171 		goto out_elf_end;
1172 	}
1173 
1174 	sec = elf_section_by_name(elf, &ehdr, &shdr,
1175 				  ".gnu_debuglink", NULL);
1176 	if (sec == NULL)
1177 		goto out_elf_end;
1178 
1179 	data = elf_getdata(sec, NULL);
1180 	if (data == NULL)
1181 		goto out_elf_end;
1182 
1183 	/* the start of this section is a zero-terminated string */
1184 	strncpy(debuglink, data->d_buf, size);
1185 
1186 	err = 0;
1187 
1188 out_elf_end:
1189 	elf_end(elf);
1190 out_close:
1191 	close(fd);
1192 out:
1193 	return err;
1194 }
1195 
1196 #endif
1197 
1198 bool symsrc__possibly_runtime(struct symsrc *ss)
1199 {
1200 	return ss->dynsym || ss->opdsec;
1201 }
1202 
1203 bool symsrc__has_symtab(struct symsrc *ss)
1204 {
1205 	return ss->symtab != NULL;
1206 }
1207 
1208 void symsrc__destroy(struct symsrc *ss)
1209 {
1210 	zfree(&ss->name);
1211 	elf_end(ss->elf);
1212 	close(ss->fd);
1213 }
1214 
1215 bool elf__needs_adjust_symbols(GElf_Ehdr ehdr)
1216 {
1217 	/*
1218 	 * Usually vmlinux is an ELF file with type ET_EXEC for most
1219 	 * architectures; except Arm64 kernel is linked with option
1220 	 * '-share', so need to check type ET_DYN.
1221 	 */
1222 	return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL ||
1223 	       ehdr.e_type == ET_DYN;
1224 }
1225 
1226 static Elf *read_gnu_debugdata(struct dso *dso, Elf *elf, const char *name, int *fd_ret)
1227 {
1228 	Elf *elf_embedded;
1229 	GElf_Ehdr ehdr;
1230 	GElf_Shdr shdr;
1231 	Elf_Scn *scn;
1232 	Elf_Data *scn_data;
1233 	FILE *wrapped;
1234 	size_t shndx;
1235 	char temp_filename[] = "/tmp/perf.gnu_debugdata.elf.XXXXXX";
1236 	int ret, temp_fd;
1237 
1238 	if (gelf_getehdr(elf, &ehdr) == NULL) {
1239 		pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
1240 		*dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF;
1241 		return NULL;
1242 	}
1243 
1244 	scn = elf_section_by_name(elf, &ehdr, &shdr, ".gnu_debugdata", &shndx);
1245 	if (!scn) {
1246 		*dso__load_errno(dso) = -ENOENT;
1247 		return NULL;
1248 	}
1249 
1250 	if (shdr.sh_type == SHT_NOBITS) {
1251 		pr_debug("%s: .gnu_debugdata of ELF file %s has no data.\n", __func__, name);
1252 		*dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF;
1253 		return NULL;
1254 	}
1255 
1256 	scn_data = elf_rawdata(scn, NULL);
1257 	if (!scn_data) {
1258 		pr_debug("%s: error reading .gnu_debugdata of %s: %s\n", __func__,
1259 			 name, elf_errmsg(-1));
1260 		*dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF;
1261 		return NULL;
1262 	}
1263 
1264 	wrapped = fmemopen(scn_data->d_buf, scn_data->d_size, "r");
1265 	if (!wrapped) {
1266 		pr_debug("%s: fmemopen: %s\n", __func__, strerror(errno));
1267 		*dso__load_errno(dso) = -errno;
1268 		return NULL;
1269 	}
1270 
1271 	temp_fd = mkstemp(temp_filename);
1272 	if (temp_fd < 0) {
1273 		pr_debug("%s: mkstemp: %s\n", __func__, strerror(errno));
1274 		*dso__load_errno(dso) = -errno;
1275 		fclose(wrapped);
1276 		return NULL;
1277 	}
1278 	unlink(temp_filename);
1279 
1280 	ret = lzma_decompress_stream_to_file(wrapped, temp_fd);
1281 	fclose(wrapped);
1282 	if (ret < 0) {
1283 		*dso__load_errno(dso) = -errno;
1284 		close(temp_fd);
1285 		return NULL;
1286 	}
1287 
1288 	elf_embedded = elf_begin(temp_fd, PERF_ELF_C_READ_MMAP, NULL);
1289 	if (!elf_embedded) {
1290 		pr_debug("%s: error reading .gnu_debugdata of %s: %s\n", __func__,
1291 			 name, elf_errmsg(-1));
1292 		*dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF;
1293 		close(temp_fd);
1294 		return NULL;
1295 	}
1296 	pr_debug("%s: using .gnu_debugdata of %s\n", __func__, name);
1297 	*fd_ret = temp_fd;
1298 	return elf_embedded;
1299 }
1300 
1301 int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
1302 		 enum dso_binary_type type)
1303 {
1304 	GElf_Ehdr ehdr;
1305 	Elf *elf;
1306 	int fd;
1307 
1308 	if (dso__needs_decompress(dso)) {
1309 		fd = dso__decompress_kmodule_fd(dso, name);
1310 		if (fd < 0)
1311 			return -1;
1312 
1313 		type = dso__symtab_type(dso);
1314 	} else {
1315 		fd = open(name, O_RDONLY);
1316 		if (fd < 0) {
1317 			*dso__load_errno(dso) = errno;
1318 			return -1;
1319 		}
1320 	}
1321 
1322 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1323 	if (elf == NULL) {
1324 		pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
1325 		*dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF;
1326 		goto out_close;
1327 	}
1328 
1329 	if (type == DSO_BINARY_TYPE__GNU_DEBUGDATA) {
1330 		int new_fd;
1331 		Elf *embedded = read_gnu_debugdata(dso, elf, name, &new_fd);
1332 
1333 		if (!embedded)
1334 			goto out_close;
1335 
1336 		elf_end(elf);
1337 		close(fd);
1338 		fd = new_fd;
1339 		elf = embedded;
1340 	}
1341 
1342 	if (gelf_getehdr(elf, &ehdr) == NULL) {
1343 		*dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF;
1344 		pr_debug("%s: cannot get elf header.\n", __func__);
1345 		goto out_elf_end;
1346 	}
1347 
1348 	if (dso__swap_init(dso, ehdr.e_ident[EI_DATA])) {
1349 		*dso__load_errno(dso) = DSO_LOAD_ERRNO__INTERNAL_ERROR;
1350 		goto out_elf_end;
1351 	}
1352 
1353 	/* Always reject images with a mismatched build-id: */
1354 	if (dso__has_build_id(dso) && !symbol_conf.ignore_vmlinux_buildid) {
1355 		u8 build_id[BUILD_ID_SIZE];
1356 		struct build_id bid;
1357 		int size;
1358 
1359 		size = elf_read_build_id(elf, build_id, BUILD_ID_SIZE);
1360 		if (size <= 0) {
1361 			*dso__load_errno(dso) = DSO_LOAD_ERRNO__CANNOT_READ_BUILDID;
1362 			goto out_elf_end;
1363 		}
1364 
1365 		build_id__init(&bid, build_id, size);
1366 		if (!dso__build_id_equal(dso, &bid)) {
1367 			pr_debug("%s: build id mismatch for %s.\n", __func__, name);
1368 			*dso__load_errno(dso) = DSO_LOAD_ERRNO__MISMATCHING_BUILDID;
1369 			goto out_elf_end;
1370 		}
1371 	}
1372 
1373 	ss->is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1374 
1375 	ss->symtab_idx = 0;
1376 	ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab",
1377 			&ss->symtab_idx);
1378 	if (ss->symshdr.sh_type != SHT_SYMTAB)
1379 		ss->symtab = NULL;
1380 
1381 	ss->dynsym_idx = 0;
1382 	ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym",
1383 			&ss->dynsym_idx);
1384 	if (ss->dynshdr.sh_type != SHT_DYNSYM)
1385 		ss->dynsym = NULL;
1386 
1387 	ss->opdidx = 0;
1388 	ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd",
1389 			&ss->opdidx);
1390 	if (ss->opdshdr.sh_type != SHT_PROGBITS)
1391 		ss->opdsec = NULL;
1392 
1393 	if (dso__kernel(dso) == DSO_SPACE__USER)
1394 		ss->adjust_symbols = true;
1395 	else
1396 		ss->adjust_symbols = elf__needs_adjust_symbols(ehdr);
1397 
1398 	ss->name   = strdup(name);
1399 	if (!ss->name) {
1400 		*dso__load_errno(dso) = errno;
1401 		goto out_elf_end;
1402 	}
1403 
1404 	ss->elf    = elf;
1405 	ss->fd     = fd;
1406 	ss->ehdr   = ehdr;
1407 	ss->type   = type;
1408 
1409 	return 0;
1410 
1411 out_elf_end:
1412 	elf_end(elf);
1413 out_close:
1414 	close(fd);
1415 	return -1;
1416 }
1417 
1418 static bool is_exe_text(int flags)
1419 {
1420 	return (flags & (SHF_ALLOC | SHF_EXECINSTR)) == (SHF_ALLOC | SHF_EXECINSTR);
1421 }
1422 
1423 /*
1424  * Some executable module sections like .noinstr.text might be laid out with
1425  * .text so they can use the same mapping (memory address to file offset).
1426  * Check if that is the case. Refer to kernel layout_sections(). Return the
1427  * maximum offset.
1428  */
1429 static u64 max_text_section(Elf *elf, GElf_Ehdr *ehdr)
1430 {
1431 	Elf_Scn *sec = NULL;
1432 	GElf_Shdr shdr;
1433 	u64 offs = 0;
1434 
1435 	/* Doesn't work for some arch */
1436 	if (ehdr->e_machine == EM_PARISC ||
1437 	    ehdr->e_machine == EM_ALPHA)
1438 		return 0;
1439 
1440 	/* ELF is corrupted/truncated, avoid calling elf_strptr. */
1441 	if (!elf_rawdata(elf_getscn(elf, ehdr->e_shstrndx), NULL))
1442 		return 0;
1443 
1444 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
1445 		char *sec_name;
1446 
1447 		if (!gelf_getshdr(sec, &shdr))
1448 			break;
1449 
1450 		if (!is_exe_text(shdr.sh_flags))
1451 			continue;
1452 
1453 		/* .init and .exit sections are not placed with .text */
1454 		sec_name = elf_strptr(elf, ehdr->e_shstrndx, shdr.sh_name);
1455 		if (!sec_name ||
1456 		    strstarts(sec_name, ".init") ||
1457 		    strstarts(sec_name, ".exit"))
1458 			break;
1459 
1460 		/* Must be next to previous, assumes .text is first */
1461 		if (offs && PERF_ALIGN(offs, shdr.sh_addralign ?: 1) != shdr.sh_offset)
1462 			break;
1463 
1464 		offs = shdr.sh_offset + shdr.sh_size;
1465 	}
1466 
1467 	return offs;
1468 }
1469 
1470 /**
1471  * ref_reloc_sym_not_found - has kernel relocation symbol been found.
1472  * @kmap: kernel maps and relocation reference symbol
1473  *
1474  * This function returns %true if we are dealing with the kernel maps and the
1475  * relocation reference symbol has not yet been found.  Otherwise %false is
1476  * returned.
1477  */
1478 static bool ref_reloc_sym_not_found(struct kmap *kmap)
1479 {
1480 	return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
1481 	       !kmap->ref_reloc_sym->unrelocated_addr;
1482 }
1483 
1484 /**
1485  * ref_reloc - kernel relocation offset.
1486  * @kmap: kernel maps and relocation reference symbol
1487  *
1488  * This function returns the offset of kernel addresses as determined by using
1489  * the relocation reference symbol i.e. if the kernel has not been relocated
1490  * then the return value is zero.
1491  */
1492 static u64 ref_reloc(struct kmap *kmap)
1493 {
1494 	if (kmap && kmap->ref_reloc_sym &&
1495 	    kmap->ref_reloc_sym->unrelocated_addr)
1496 		return kmap->ref_reloc_sym->addr -
1497 		       kmap->ref_reloc_sym->unrelocated_addr;
1498 	return 0;
1499 }
1500 
1501 void __weak arch__sym_update(struct symbol *s __maybe_unused,
1502 		GElf_Sym *sym __maybe_unused) { }
1503 
1504 static int dso__process_kernel_symbol(struct dso *dso, struct map *map,
1505 				      GElf_Sym *sym, GElf_Shdr *shdr,
1506 				      struct maps *kmaps, struct kmap *kmap,
1507 				      struct dso **curr_dsop,
1508 				      const char *section_name,
1509 				      bool adjust_kernel_syms, bool kmodule, bool *remap_kernel,
1510 				      u64 max_text_sh_offset)
1511 {
1512 	struct dso *curr_dso = *curr_dsop;
1513 	struct map *curr_map;
1514 	char dso_name[PATH_MAX];
1515 
1516 	/* Adjust symbol to map to file offset */
1517 	if (adjust_kernel_syms)
1518 		sym->st_value -= shdr->sh_addr - shdr->sh_offset;
1519 
1520 	if (strcmp(section_name, (dso__short_name(curr_dso) + dso__short_name_len(dso))) == 0)
1521 		return 0;
1522 
1523 	if (strcmp(section_name, ".text") == 0) {
1524 		/*
1525 		 * The initial kernel mapping is based on
1526 		 * kallsyms and identity maps.  Overwrite it to
1527 		 * map to the kernel dso.
1528 		 */
1529 		if (*remap_kernel && dso__kernel(dso) && !kmodule) {
1530 			*remap_kernel = false;
1531 			map__set_start(map, shdr->sh_addr + ref_reloc(kmap));
1532 			map__set_end(map, map__start(map) + shdr->sh_size);
1533 			map__set_pgoff(map, shdr->sh_offset);
1534 			map__set_mapping_type(map, MAPPING_TYPE__DSO);
1535 			/* Ensure maps are correctly ordered */
1536 			if (kmaps) {
1537 				int err;
1538 				struct map *tmp = map__get(map);
1539 
1540 				maps__remove(kmaps, map);
1541 				err = maps__insert(kmaps, map);
1542 				map__put(tmp);
1543 				if (err)
1544 					return err;
1545 			}
1546 		}
1547 
1548 		/*
1549 		 * The initial module mapping is based on
1550 		 * /proc/modules mapped to offset zero.
1551 		 * Overwrite it to map to the module dso.
1552 		 */
1553 		if (*remap_kernel && kmodule) {
1554 			*remap_kernel = false;
1555 			map__set_pgoff(map, shdr->sh_offset);
1556 		}
1557 
1558 		dso__put(*curr_dsop);
1559 		*curr_dsop = dso__get(dso);
1560 		return 0;
1561 	}
1562 
1563 	if (!kmap)
1564 		return 0;
1565 
1566 	/*
1567 	 * perf does not record module section addresses except for .text, but
1568 	 * some sections can use the same mapping as .text.
1569 	 */
1570 	if (kmodule && adjust_kernel_syms && is_exe_text(shdr->sh_flags) &&
1571 	    shdr->sh_offset <= max_text_sh_offset) {
1572 		dso__put(*curr_dsop);
1573 		*curr_dsop = dso__get(dso);
1574 		return 0;
1575 	}
1576 
1577 	snprintf(dso_name, sizeof(dso_name), "%s%s", dso__short_name(dso), section_name);
1578 
1579 	curr_map = maps__find_by_name(kmaps, dso_name);
1580 	if (curr_map == NULL) {
1581 		u64 start = sym->st_value;
1582 
1583 		if (kmodule)
1584 			start += map__start(map) + shdr->sh_offset;
1585 
1586 		curr_dso = dso__new(dso_name);
1587 		if (curr_dso == NULL)
1588 			return -1;
1589 		dso__set_kernel(curr_dso, dso__kernel(dso));
1590 		RC_CHK_ACCESS(curr_dso)->long_name = dso__long_name(dso);
1591 		RC_CHK_ACCESS(curr_dso)->long_name_len = dso__long_name_len(dso);
1592 		dso__set_binary_type(curr_dso, dso__binary_type(dso));
1593 		dso__set_adjust_symbols(curr_dso, dso__adjust_symbols(dso));
1594 		curr_map = map__new2(start, curr_dso);
1595 		if (curr_map == NULL) {
1596 			dso__put(curr_dso);
1597 			return -1;
1598 		}
1599 		if (dso__kernel(curr_dso))
1600 			map__kmap(curr_map)->kmaps = kmaps;
1601 
1602 		if (adjust_kernel_syms) {
1603 			map__set_start(curr_map, shdr->sh_addr + ref_reloc(kmap));
1604 			map__set_end(curr_map, map__start(curr_map) + shdr->sh_size);
1605 			map__set_pgoff(curr_map, shdr->sh_offset);
1606 		} else {
1607 			map__set_mapping_type(curr_map, MAPPING_TYPE__IDENTITY);
1608 		}
1609 		dso__set_symtab_type(curr_dso, dso__symtab_type(dso));
1610 		if (maps__insert(kmaps, curr_map))
1611 			return -1;
1612 		dsos__add(&maps__machine(kmaps)->dsos, curr_dso);
1613 		dso__set_loaded(curr_dso);
1614 		dso__put(*curr_dsop);
1615 		*curr_dsop = curr_dso;
1616 	} else {
1617 		dso__put(*curr_dsop);
1618 		*curr_dsop = dso__get(map__dso(curr_map));
1619 	}
1620 	map__put(curr_map);
1621 
1622 	return 0;
1623 }
1624 
1625 static int
1626 dso__load_sym_internal(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1627 		       struct symsrc *runtime_ss, int kmodule, int dynsym)
1628 {
1629 	struct kmap *kmap = dso__kernel(dso) ? map__kmap(map) : NULL;
1630 	struct maps *kmaps = kmap ? map__kmaps(map) : NULL;
1631 	struct dso *curr_dso = NULL;
1632 	Elf_Data *symstrs, *secstrs, *secstrs_run, *secstrs_sym;
1633 	uint32_t nr_syms;
1634 	uint32_t idx;
1635 	GElf_Ehdr ehdr;
1636 	GElf_Shdr shdr;
1637 	GElf_Shdr tshdr;
1638 	Elf_Data *syms, *opddata = NULL;
1639 	GElf_Sym sym;
1640 	Elf_Scn *sec, *sec_strndx;
1641 	Elf *elf;
1642 	int nr = 0;
1643 	bool remap_kernel = false, adjust_kernel_syms = false;
1644 	u64 max_text_sh_offset = 0;
1645 
1646 	if (kmap && !kmaps)
1647 		return -1;
1648 
1649 	elf = syms_ss->elf;
1650 	ehdr = syms_ss->ehdr;
1651 	if (dynsym) {
1652 		sec  = syms_ss->dynsym;
1653 		shdr = syms_ss->dynshdr;
1654 	} else {
1655 		sec =  syms_ss->symtab;
1656 		shdr = syms_ss->symshdr;
1657 	}
1658 
1659 	if (elf_section_by_name(runtime_ss->elf, &runtime_ss->ehdr, &tshdr,
1660 				".text", NULL)) {
1661 		dso__set_text_offset(dso, tshdr.sh_addr - tshdr.sh_offset);
1662 		dso__set_text_end(dso, tshdr.sh_offset + tshdr.sh_size);
1663 	}
1664 
1665 	if (runtime_ss->opdsec)
1666 		opddata = elf_rawdata(runtime_ss->opdsec, NULL);
1667 
1668 	syms = elf_getdata(sec, NULL);
1669 	if (syms == NULL)
1670 		goto out_elf_end;
1671 
1672 	sec = elf_getscn(elf, shdr.sh_link);
1673 	if (sec == NULL)
1674 		goto out_elf_end;
1675 
1676 	symstrs = elf_getdata(sec, NULL);
1677 	if (symstrs == NULL)
1678 		goto out_elf_end;
1679 
1680 	sec_strndx = elf_getscn(runtime_ss->elf, runtime_ss->ehdr.e_shstrndx);
1681 	if (sec_strndx == NULL)
1682 		goto out_elf_end;
1683 
1684 	secstrs_run = elf_getdata(sec_strndx, NULL);
1685 	if (secstrs_run == NULL)
1686 		goto out_elf_end;
1687 
1688 	sec_strndx = elf_getscn(elf, ehdr.e_shstrndx);
1689 	if (sec_strndx == NULL)
1690 		goto out_elf_end;
1691 
1692 	secstrs_sym = elf_getdata(sec_strndx, NULL);
1693 	if (secstrs_sym == NULL)
1694 		goto out_elf_end;
1695 
1696 	nr_syms = shdr.sh_size / shdr.sh_entsize;
1697 
1698 	memset(&sym, 0, sizeof(sym));
1699 
1700 	/*
1701 	 * The kernel relocation symbol is needed in advance in order to adjust
1702 	 * kernel maps correctly.
1703 	 */
1704 	if (ref_reloc_sym_not_found(kmap)) {
1705 		elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1706 			const char *elf_name = elf_sym__name(&sym, symstrs);
1707 
1708 			if (strcmp(elf_name, kmap->ref_reloc_sym->name))
1709 				continue;
1710 			kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
1711 			map__set_reloc(map, kmap->ref_reloc_sym->addr - kmap->ref_reloc_sym->unrelocated_addr);
1712 			break;
1713 		}
1714 	}
1715 
1716 	/*
1717 	 * Handle any relocation of vdso necessary because older kernels
1718 	 * attempted to prelink vdso to its virtual address.
1719 	 */
1720 	if (dso__is_vdso(dso))
1721 		map__set_reloc(map, map__start(map) - dso__text_offset(dso));
1722 
1723 	dso__set_adjust_symbols(dso, runtime_ss->adjust_symbols || ref_reloc(kmap));
1724 	/*
1725 	 * Initial kernel and module mappings do not map to the dso.
1726 	 * Flag the fixups.
1727 	 */
1728 	if (dso__kernel(dso)) {
1729 		remap_kernel = true;
1730 		adjust_kernel_syms = dso__adjust_symbols(dso);
1731 	}
1732 
1733 	if (kmodule && adjust_kernel_syms)
1734 		max_text_sh_offset = max_text_section(runtime_ss->elf, &runtime_ss->ehdr);
1735 
1736 	curr_dso = dso__get(dso);
1737 	elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1738 		struct symbol *f;
1739 		const char *elf_name = elf_sym__name(&sym, symstrs);
1740 		char *demangled = NULL;
1741 		int is_label = elf_sym__is_label(&sym);
1742 		const char *section_name;
1743 		bool used_opd = false;
1744 
1745 		if (!is_label && !elf_sym__filter(&sym))
1746 			continue;
1747 
1748 		/* Reject ARM ELF "mapping symbols": these aren't unique and
1749 		 * don't identify functions, so will confuse the profile
1750 		 * output: */
1751 		if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) {
1752 			if (elf_name[0] == '$' && strchr("adtx", elf_name[1])
1753 			    && (elf_name[2] == '\0' || elf_name[2] == '.'))
1754 				continue;
1755 		}
1756 
1757 		if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) {
1758 			u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr;
1759 			u64 *opd = opddata->d_buf + offset;
1760 			sym.st_value = DSO__SWAP(dso, u64, *opd);
1761 			sym.st_shndx = elf_addr_to_index(runtime_ss->elf,
1762 					sym.st_value);
1763 			used_opd = true;
1764 		}
1765 
1766 		/*
1767 		 * When loading symbols in a data mapping, ABS symbols (which
1768 		 * has a value of SHN_ABS in its st_shndx) failed at
1769 		 * elf_getscn().  And it marks the loading as a failure so
1770 		 * already loaded symbols cannot be fixed up.
1771 		 *
1772 		 * I'm not sure what should be done. Just ignore them for now.
1773 		 * - Namhyung Kim
1774 		 */
1775 		if (sym.st_shndx == SHN_ABS)
1776 			continue;
1777 
1778 		sec = elf_getscn(syms_ss->elf, sym.st_shndx);
1779 		if (!sec)
1780 			goto out_elf_end;
1781 
1782 		gelf_getshdr(sec, &shdr);
1783 
1784 		/*
1785 		 * If the attribute bit SHF_ALLOC is not set, the section
1786 		 * doesn't occupy memory during process execution.
1787 		 * E.g. ".gnu.warning.*" section is used by linker to generate
1788 		 * warnings when calling deprecated functions, the symbols in
1789 		 * the section aren't loaded to memory during process execution,
1790 		 * so skip them.
1791 		 */
1792 		if (!(shdr.sh_flags & SHF_ALLOC))
1793 			continue;
1794 
1795 		secstrs = secstrs_sym;
1796 
1797 		/*
1798 		 * We have to fallback to runtime when syms' section header has
1799 		 * NOBITS set. NOBITS results in file offset (sh_offset) not
1800 		 * being incremented. So sh_offset used below has different
1801 		 * values for syms (invalid) and runtime (valid).
1802 		 */
1803 		if (shdr.sh_type == SHT_NOBITS) {
1804 			sec = elf_getscn(runtime_ss->elf, sym.st_shndx);
1805 			if (!sec)
1806 				goto out_elf_end;
1807 
1808 			gelf_getshdr(sec, &shdr);
1809 			secstrs = secstrs_run;
1810 		}
1811 
1812 		if (is_label && !elf_sec__filter(&shdr, secstrs))
1813 			continue;
1814 
1815 		section_name = elf_sec__name(&shdr, secstrs);
1816 
1817 		/* On ARM, symbols for thumb functions have 1 added to
1818 		 * the symbol address as a flag - remove it */
1819 		if ((ehdr.e_machine == EM_ARM) &&
1820 		    (GELF_ST_TYPE(sym.st_info) == STT_FUNC) &&
1821 		    (sym.st_value & 1))
1822 			--sym.st_value;
1823 
1824 		if (dso__kernel(dso)) {
1825 			if (dso__process_kernel_symbol(dso, map, &sym, &shdr,
1826 						       kmaps, kmap, &curr_dso,
1827 						       section_name,
1828 						       adjust_kernel_syms,
1829 						       kmodule,
1830 						       &remap_kernel,
1831 						       max_text_sh_offset))
1832 				goto out_elf_end;
1833 		} else if ((used_opd && runtime_ss->adjust_symbols) ||
1834 			   (!used_opd && syms_ss->adjust_symbols)) {
1835 			GElf_Phdr phdr;
1836 
1837 			if (elf_read_program_header(runtime_ss->elf,
1838 						    (u64)sym.st_value, &phdr)) {
1839 				pr_debug4("%s: failed to find program header for "
1840 					   "symbol: %s st_value: %#" PRIx64 "\n",
1841 					   __func__, elf_name, (u64)sym.st_value);
1842 				pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1843 					"sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n",
1844 					__func__, (u64)sym.st_value, (u64)shdr.sh_addr,
1845 					(u64)shdr.sh_offset);
1846 				/*
1847 				 * Fail to find program header, let's rollback
1848 				 * to use shdr.sh_addr and shdr.sh_offset to
1849 				 * calibrate symbol's file address, though this
1850 				 * is not necessary for normal C ELF file, we
1851 				 * still need to handle java JIT symbols in this
1852 				 * case.
1853 				 */
1854 				sym.st_value -= shdr.sh_addr - shdr.sh_offset;
1855 			} else {
1856 				pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1857 					"p_vaddr: %#" PRIx64 " p_offset: %#" PRIx64 "\n",
1858 					__func__, (u64)sym.st_value, (u64)phdr.p_vaddr,
1859 					(u64)phdr.p_offset);
1860 				sym.st_value -= phdr.p_vaddr - phdr.p_offset;
1861 			}
1862 		}
1863 
1864 		demangled = demangle_sym(dso, kmodule, elf_name);
1865 		if (demangled != NULL)
1866 			elf_name = demangled;
1867 
1868 		f = symbol__new(sym.st_value, sym.st_size,
1869 				GELF_ST_BIND(sym.st_info),
1870 				GELF_ST_TYPE(sym.st_info), elf_name);
1871 		free(demangled);
1872 		if (!f)
1873 			goto out_elf_end;
1874 
1875 		arch__sym_update(f, &sym);
1876 
1877 		__symbols__insert(dso__symbols(curr_dso), f, dso__kernel(dso));
1878 		nr++;
1879 	}
1880 	dso__put(curr_dso);
1881 
1882 	/*
1883 	 * For misannotated, zeroed, ASM function sizes.
1884 	 */
1885 	if (nr > 0) {
1886 		symbols__fixup_end(dso__symbols(dso), false);
1887 		symbols__fixup_duplicate(dso__symbols(dso));
1888 		if (kmap) {
1889 			/*
1890 			 * We need to fixup this here too because we create new
1891 			 * maps here, for things like vsyscall sections.
1892 			 */
1893 			maps__fixup_end(kmaps);
1894 		}
1895 	}
1896 	return nr;
1897 out_elf_end:
1898 	dso__put(curr_dso);
1899 	return -1;
1900 }
1901 
1902 int dso__load_sym(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1903 		  struct symsrc *runtime_ss, int kmodule)
1904 {
1905 	int nr = 0;
1906 	int err = -1;
1907 
1908 	dso__set_symtab_type(dso, syms_ss->type);
1909 	dso__set_is_64_bit(dso, syms_ss->is_64_bit);
1910 	dso__set_rel(dso, syms_ss->ehdr.e_type == ET_REL);
1911 
1912 	/*
1913 	 * Modules may already have symbols from kallsyms, but those symbols
1914 	 * have the wrong values for the dso maps, so remove them.
1915 	 */
1916 	if (kmodule && syms_ss->symtab)
1917 		symbols__delete(dso__symbols(dso));
1918 
1919 	if (!syms_ss->symtab) {
1920 		/*
1921 		 * If the vmlinux is stripped, fail so we will fall back
1922 		 * to using kallsyms. The vmlinux runtime symbols aren't
1923 		 * of much use.
1924 		 */
1925 		if (dso__kernel(dso))
1926 			return err;
1927 	} else  {
1928 		err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss,
1929 					     kmodule, 0);
1930 		if (err < 0)
1931 			return err;
1932 		nr = err;
1933 	}
1934 
1935 	if (syms_ss->dynsym) {
1936 		err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss,
1937 					     kmodule, 1);
1938 		if (err < 0)
1939 			return err;
1940 		nr += err;
1941 	}
1942 
1943 	/*
1944 	 * The .gnu_debugdata is a special situation: it contains a symbol
1945 	 * table, but the runtime file may also contain dynsym entries which are
1946 	 * not present there. We need to load both.
1947 	 */
1948 	if (syms_ss->type == DSO_BINARY_TYPE__GNU_DEBUGDATA && runtime_ss->dynsym) {
1949 		err = dso__load_sym_internal(dso, map, runtime_ss, runtime_ss,
1950 					     kmodule, 1);
1951 		if (err < 0)
1952 			return err;
1953 		nr += err;
1954 	}
1955 
1956 	return nr;
1957 }
1958 
1959 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
1960 {
1961 	GElf_Phdr phdr;
1962 	size_t i, phdrnum;
1963 	int err;
1964 	u64 sz;
1965 
1966 	if (elf_getphdrnum(elf, &phdrnum))
1967 		return -1;
1968 
1969 	for (i = 0; i < phdrnum; i++) {
1970 		if (gelf_getphdr(elf, i, &phdr) == NULL)
1971 			return -1;
1972 		if (phdr.p_type != PT_LOAD)
1973 			continue;
1974 		if (exe) {
1975 			if (!(phdr.p_flags & PF_X))
1976 				continue;
1977 		} else {
1978 			if (!(phdr.p_flags & PF_R))
1979 				continue;
1980 		}
1981 		sz = min(phdr.p_memsz, phdr.p_filesz);
1982 		if (!sz)
1983 			continue;
1984 		err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
1985 		if (err)
1986 			return err;
1987 	}
1988 	return 0;
1989 }
1990 
1991 int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data,
1992 		    bool *is_64_bit)
1993 {
1994 	int err;
1995 	Elf *elf;
1996 
1997 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1998 	if (elf == NULL)
1999 		return -1;
2000 
2001 	if (is_64_bit)
2002 		*is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
2003 
2004 	err = elf_read_maps(elf, exe, mapfn, data);
2005 
2006 	elf_end(elf);
2007 	return err;
2008 }
2009 
2010 enum dso_type dso__type_fd(int fd)
2011 {
2012 	enum dso_type dso_type = DSO__TYPE_UNKNOWN;
2013 	GElf_Ehdr ehdr;
2014 	Elf_Kind ek;
2015 	Elf *elf;
2016 
2017 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
2018 	if (elf == NULL)
2019 		goto out;
2020 
2021 	ek = elf_kind(elf);
2022 	if (ek != ELF_K_ELF)
2023 		goto out_end;
2024 
2025 	if (gelf_getclass(elf) == ELFCLASS64) {
2026 		dso_type = DSO__TYPE_64BIT;
2027 		goto out_end;
2028 	}
2029 
2030 	if (gelf_getehdr(elf, &ehdr) == NULL)
2031 		goto out_end;
2032 
2033 	if (ehdr.e_machine == EM_X86_64)
2034 		dso_type = DSO__TYPE_X32BIT;
2035 	else
2036 		dso_type = DSO__TYPE_32BIT;
2037 out_end:
2038 	elf_end(elf);
2039 out:
2040 	return dso_type;
2041 }
2042 
2043 static int copy_bytes(int from, off_t from_offs, int to, off_t to_offs, u64 len)
2044 {
2045 	ssize_t r;
2046 	size_t n;
2047 	int err = -1;
2048 	char *buf = malloc(page_size);
2049 
2050 	if (buf == NULL)
2051 		return -1;
2052 
2053 	if (lseek(to, to_offs, SEEK_SET) != to_offs)
2054 		goto out;
2055 
2056 	if (lseek(from, from_offs, SEEK_SET) != from_offs)
2057 		goto out;
2058 
2059 	while (len) {
2060 		n = page_size;
2061 		if (len < n)
2062 			n = len;
2063 		/* Use read because mmap won't work on proc files */
2064 		r = read(from, buf, n);
2065 		if (r < 0)
2066 			goto out;
2067 		if (!r)
2068 			break;
2069 		n = r;
2070 		r = write(to, buf, n);
2071 		if (r < 0)
2072 			goto out;
2073 		if ((size_t)r != n)
2074 			goto out;
2075 		len -= n;
2076 	}
2077 
2078 	err = 0;
2079 out:
2080 	free(buf);
2081 	return err;
2082 }
2083 
2084 struct kcore {
2085 	int fd;
2086 	int elfclass;
2087 	Elf *elf;
2088 	GElf_Ehdr ehdr;
2089 };
2090 
2091 static int kcore__open(struct kcore *kcore, const char *filename)
2092 {
2093 	GElf_Ehdr *ehdr;
2094 
2095 	kcore->fd = open(filename, O_RDONLY);
2096 	if (kcore->fd == -1)
2097 		return -1;
2098 
2099 	kcore->elf = elf_begin(kcore->fd, ELF_C_READ, NULL);
2100 	if (!kcore->elf)
2101 		goto out_close;
2102 
2103 	kcore->elfclass = gelf_getclass(kcore->elf);
2104 	if (kcore->elfclass == ELFCLASSNONE)
2105 		goto out_end;
2106 
2107 	ehdr = gelf_getehdr(kcore->elf, &kcore->ehdr);
2108 	if (!ehdr)
2109 		goto out_end;
2110 
2111 	return 0;
2112 
2113 out_end:
2114 	elf_end(kcore->elf);
2115 out_close:
2116 	close(kcore->fd);
2117 	return -1;
2118 }
2119 
2120 static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
2121 		       bool temp)
2122 {
2123 	kcore->elfclass = elfclass;
2124 
2125 	if (temp)
2126 		kcore->fd = mkstemp(filename);
2127 	else
2128 		kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
2129 	if (kcore->fd == -1)
2130 		return -1;
2131 
2132 	kcore->elf = elf_begin(kcore->fd, ELF_C_WRITE, NULL);
2133 	if (!kcore->elf)
2134 		goto out_close;
2135 
2136 	if (!gelf_newehdr(kcore->elf, elfclass))
2137 		goto out_end;
2138 
2139 	memset(&kcore->ehdr, 0, sizeof(GElf_Ehdr));
2140 
2141 	return 0;
2142 
2143 out_end:
2144 	elf_end(kcore->elf);
2145 out_close:
2146 	close(kcore->fd);
2147 	unlink(filename);
2148 	return -1;
2149 }
2150 
2151 static void kcore__close(struct kcore *kcore)
2152 {
2153 	elf_end(kcore->elf);
2154 	close(kcore->fd);
2155 }
2156 
2157 static int kcore__copy_hdr(struct kcore *from, struct kcore *to, size_t count)
2158 {
2159 	GElf_Ehdr *ehdr = &to->ehdr;
2160 	GElf_Ehdr *kehdr = &from->ehdr;
2161 
2162 	memcpy(ehdr->e_ident, kehdr->e_ident, EI_NIDENT);
2163 	ehdr->e_type      = kehdr->e_type;
2164 	ehdr->e_machine   = kehdr->e_machine;
2165 	ehdr->e_version   = kehdr->e_version;
2166 	ehdr->e_entry     = 0;
2167 	ehdr->e_shoff     = 0;
2168 	ehdr->e_flags     = kehdr->e_flags;
2169 	ehdr->e_phnum     = count;
2170 	ehdr->e_shentsize = 0;
2171 	ehdr->e_shnum     = 0;
2172 	ehdr->e_shstrndx  = 0;
2173 
2174 	if (from->elfclass == ELFCLASS32) {
2175 		ehdr->e_phoff     = sizeof(Elf32_Ehdr);
2176 		ehdr->e_ehsize    = sizeof(Elf32_Ehdr);
2177 		ehdr->e_phentsize = sizeof(Elf32_Phdr);
2178 	} else {
2179 		ehdr->e_phoff     = sizeof(Elf64_Ehdr);
2180 		ehdr->e_ehsize    = sizeof(Elf64_Ehdr);
2181 		ehdr->e_phentsize = sizeof(Elf64_Phdr);
2182 	}
2183 
2184 	if (!gelf_update_ehdr(to->elf, ehdr))
2185 		return -1;
2186 
2187 	if (!gelf_newphdr(to->elf, count))
2188 		return -1;
2189 
2190 	return 0;
2191 }
2192 
2193 static int kcore__add_phdr(struct kcore *kcore, int idx, off_t offset,
2194 			   u64 addr, u64 len)
2195 {
2196 	GElf_Phdr phdr = {
2197 		.p_type		= PT_LOAD,
2198 		.p_flags	= PF_R | PF_W | PF_X,
2199 		.p_offset	= offset,
2200 		.p_vaddr	= addr,
2201 		.p_paddr	= 0,
2202 		.p_filesz	= len,
2203 		.p_memsz	= len,
2204 		.p_align	= page_size,
2205 	};
2206 
2207 	if (!gelf_update_phdr(kcore->elf, idx, &phdr))
2208 		return -1;
2209 
2210 	return 0;
2211 }
2212 
2213 static off_t kcore__write(struct kcore *kcore)
2214 {
2215 	return elf_update(kcore->elf, ELF_C_WRITE);
2216 }
2217 
2218 struct phdr_data {
2219 	off_t offset;
2220 	off_t rel;
2221 	u64 addr;
2222 	u64 len;
2223 	struct list_head node;
2224 	struct phdr_data *remaps;
2225 };
2226 
2227 struct sym_data {
2228 	u64 addr;
2229 	struct list_head node;
2230 };
2231 
2232 struct kcore_copy_info {
2233 	u64 stext;
2234 	u64 etext;
2235 	u64 first_symbol;
2236 	u64 last_symbol;
2237 	u64 first_module;
2238 	u64 first_module_symbol;
2239 	u64 last_module_symbol;
2240 	size_t phnum;
2241 	struct list_head phdrs;
2242 	struct list_head syms;
2243 };
2244 
2245 #define kcore_copy__for_each_phdr(k, p) \
2246 	list_for_each_entry((p), &(k)->phdrs, node)
2247 
2248 static struct phdr_data *phdr_data__new(u64 addr, u64 len, off_t offset)
2249 {
2250 	struct phdr_data *p = zalloc(sizeof(*p));
2251 
2252 	if (p) {
2253 		p->addr   = addr;
2254 		p->len    = len;
2255 		p->offset = offset;
2256 	}
2257 
2258 	return p;
2259 }
2260 
2261 static struct phdr_data *kcore_copy_info__addnew(struct kcore_copy_info *kci,
2262 						 u64 addr, u64 len,
2263 						 off_t offset)
2264 {
2265 	struct phdr_data *p = phdr_data__new(addr, len, offset);
2266 
2267 	if (p)
2268 		list_add_tail(&p->node, &kci->phdrs);
2269 
2270 	return p;
2271 }
2272 
2273 static void kcore_copy__free_phdrs(struct kcore_copy_info *kci)
2274 {
2275 	struct phdr_data *p, *tmp;
2276 
2277 	list_for_each_entry_safe(p, tmp, &kci->phdrs, node) {
2278 		list_del_init(&p->node);
2279 		free(p);
2280 	}
2281 }
2282 
2283 static struct sym_data *kcore_copy__new_sym(struct kcore_copy_info *kci,
2284 					    u64 addr)
2285 {
2286 	struct sym_data *s = zalloc(sizeof(*s));
2287 
2288 	if (s) {
2289 		s->addr = addr;
2290 		list_add_tail(&s->node, &kci->syms);
2291 	}
2292 
2293 	return s;
2294 }
2295 
2296 static void kcore_copy__free_syms(struct kcore_copy_info *kci)
2297 {
2298 	struct sym_data *s, *tmp;
2299 
2300 	list_for_each_entry_safe(s, tmp, &kci->syms, node) {
2301 		list_del_init(&s->node);
2302 		free(s);
2303 	}
2304 }
2305 
2306 static int kcore_copy__process_kallsyms(void *arg, const char *name, char type,
2307 					u64 start)
2308 {
2309 	struct kcore_copy_info *kci = arg;
2310 
2311 	if (!kallsyms__is_function(type))
2312 		return 0;
2313 
2314 	if (strchr(name, '[')) {
2315 		if (!kci->first_module_symbol || start < kci->first_module_symbol)
2316 			kci->first_module_symbol = start;
2317 		if (start > kci->last_module_symbol)
2318 			kci->last_module_symbol = start;
2319 		return 0;
2320 	}
2321 
2322 	if (!kci->first_symbol || start < kci->first_symbol)
2323 		kci->first_symbol = start;
2324 
2325 	if (!kci->last_symbol || start > kci->last_symbol)
2326 		kci->last_symbol = start;
2327 
2328 	if (!strcmp(name, "_stext")) {
2329 		kci->stext = start;
2330 		return 0;
2331 	}
2332 
2333 	if (!strcmp(name, "_etext")) {
2334 		kci->etext = start;
2335 		return 0;
2336 	}
2337 
2338 	if (is_entry_trampoline(name) && !kcore_copy__new_sym(kci, start))
2339 		return -1;
2340 
2341 	return 0;
2342 }
2343 
2344 static int kcore_copy__parse_kallsyms(struct kcore_copy_info *kci,
2345 				      const char *dir)
2346 {
2347 	char kallsyms_filename[PATH_MAX];
2348 
2349 	scnprintf(kallsyms_filename, PATH_MAX, "%s/kallsyms", dir);
2350 
2351 	if (symbol__restricted_filename(kallsyms_filename, "/proc/kallsyms"))
2352 		return -1;
2353 
2354 	if (kallsyms__parse(kallsyms_filename, kci,
2355 			    kcore_copy__process_kallsyms) < 0)
2356 		return -1;
2357 
2358 	return 0;
2359 }
2360 
2361 static int kcore_copy__process_modules(void *arg,
2362 				       const char *name __maybe_unused,
2363 				       u64 start, u64 size __maybe_unused)
2364 {
2365 	struct kcore_copy_info *kci = arg;
2366 
2367 	if (!kci->first_module || start < kci->first_module)
2368 		kci->first_module = start;
2369 
2370 	return 0;
2371 }
2372 
2373 static int kcore_copy__parse_modules(struct kcore_copy_info *kci,
2374 				     const char *dir)
2375 {
2376 	char modules_filename[PATH_MAX];
2377 
2378 	scnprintf(modules_filename, PATH_MAX, "%s/modules", dir);
2379 
2380 	if (symbol__restricted_filename(modules_filename, "/proc/modules"))
2381 		return -1;
2382 
2383 	if (modules__parse(modules_filename, kci,
2384 			   kcore_copy__process_modules) < 0)
2385 		return -1;
2386 
2387 	return 0;
2388 }
2389 
2390 static int kcore_copy__map(struct kcore_copy_info *kci, u64 start, u64 end,
2391 			   u64 pgoff, u64 s, u64 e)
2392 {
2393 	u64 len, offset;
2394 
2395 	if (s < start || s >= end)
2396 		return 0;
2397 
2398 	offset = (s - start) + pgoff;
2399 	len = e < end ? e - s : end - s;
2400 
2401 	return kcore_copy_info__addnew(kci, s, len, offset) ? 0 : -1;
2402 }
2403 
2404 static int kcore_copy__read_map(u64 start, u64 len, u64 pgoff, void *data)
2405 {
2406 	struct kcore_copy_info *kci = data;
2407 	u64 end = start + len;
2408 	struct sym_data *sdat;
2409 
2410 	if (kcore_copy__map(kci, start, end, pgoff, kci->stext, kci->etext))
2411 		return -1;
2412 
2413 	if (kcore_copy__map(kci, start, end, pgoff, kci->first_module,
2414 			    kci->last_module_symbol))
2415 		return -1;
2416 
2417 	list_for_each_entry(sdat, &kci->syms, node) {
2418 		u64 s = round_down(sdat->addr, page_size);
2419 
2420 		if (kcore_copy__map(kci, start, end, pgoff, s, s + len))
2421 			return -1;
2422 	}
2423 
2424 	return 0;
2425 }
2426 
2427 static int kcore_copy__read_maps(struct kcore_copy_info *kci, Elf *elf)
2428 {
2429 	if (elf_read_maps(elf, true, kcore_copy__read_map, kci) < 0)
2430 		return -1;
2431 
2432 	return 0;
2433 }
2434 
2435 static void kcore_copy__find_remaps(struct kcore_copy_info *kci)
2436 {
2437 	struct phdr_data *p, *k = NULL;
2438 	u64 kend;
2439 
2440 	if (!kci->stext)
2441 		return;
2442 
2443 	/* Find phdr that corresponds to the kernel map (contains stext) */
2444 	kcore_copy__for_each_phdr(kci, p) {
2445 		u64 pend = p->addr + p->len - 1;
2446 
2447 		if (p->addr <= kci->stext && pend >= kci->stext) {
2448 			k = p;
2449 			break;
2450 		}
2451 	}
2452 
2453 	if (!k)
2454 		return;
2455 
2456 	kend = k->offset + k->len;
2457 
2458 	/* Find phdrs that remap the kernel */
2459 	kcore_copy__for_each_phdr(kci, p) {
2460 		u64 pend = p->offset + p->len;
2461 
2462 		if (p == k)
2463 			continue;
2464 
2465 		if (p->offset >= k->offset && pend <= kend)
2466 			p->remaps = k;
2467 	}
2468 }
2469 
2470 static void kcore_copy__layout(struct kcore_copy_info *kci)
2471 {
2472 	struct phdr_data *p;
2473 	off_t rel = 0;
2474 
2475 	kcore_copy__find_remaps(kci);
2476 
2477 	kcore_copy__for_each_phdr(kci, p) {
2478 		if (!p->remaps) {
2479 			p->rel = rel;
2480 			rel += p->len;
2481 		}
2482 		kci->phnum += 1;
2483 	}
2484 
2485 	kcore_copy__for_each_phdr(kci, p) {
2486 		struct phdr_data *k = p->remaps;
2487 
2488 		if (k)
2489 			p->rel = p->offset - k->offset + k->rel;
2490 	}
2491 }
2492 
2493 static int kcore_copy__calc_maps(struct kcore_copy_info *kci, const char *dir,
2494 				 Elf *elf)
2495 {
2496 	if (kcore_copy__parse_kallsyms(kci, dir))
2497 		return -1;
2498 
2499 	if (kcore_copy__parse_modules(kci, dir))
2500 		return -1;
2501 
2502 	if (kci->stext)
2503 		kci->stext = round_down(kci->stext, page_size);
2504 	else
2505 		kci->stext = round_down(kci->first_symbol, page_size);
2506 
2507 	if (kci->etext) {
2508 		kci->etext = round_up(kci->etext, page_size);
2509 	} else if (kci->last_symbol) {
2510 		kci->etext = round_up(kci->last_symbol, page_size);
2511 		kci->etext += page_size;
2512 	}
2513 
2514 	if (kci->first_module_symbol &&
2515 	    (!kci->first_module || kci->first_module_symbol < kci->first_module))
2516 		kci->first_module = kci->first_module_symbol;
2517 
2518 	kci->first_module = round_down(kci->first_module, page_size);
2519 
2520 	if (kci->last_module_symbol) {
2521 		kci->last_module_symbol = round_up(kci->last_module_symbol,
2522 						   page_size);
2523 		kci->last_module_symbol += page_size;
2524 	}
2525 
2526 	if (!kci->stext || !kci->etext)
2527 		return -1;
2528 
2529 	if (kci->first_module && !kci->last_module_symbol)
2530 		return -1;
2531 
2532 	if (kcore_copy__read_maps(kci, elf))
2533 		return -1;
2534 
2535 	kcore_copy__layout(kci);
2536 
2537 	return 0;
2538 }
2539 
2540 static int kcore_copy__copy_file(const char *from_dir, const char *to_dir,
2541 				 const char *name)
2542 {
2543 	char from_filename[PATH_MAX];
2544 	char to_filename[PATH_MAX];
2545 
2546 	scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
2547 	scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
2548 
2549 	return copyfile_mode(from_filename, to_filename, 0400);
2550 }
2551 
2552 static int kcore_copy__unlink(const char *dir, const char *name)
2553 {
2554 	char filename[PATH_MAX];
2555 
2556 	scnprintf(filename, PATH_MAX, "%s/%s", dir, name);
2557 
2558 	return unlink(filename);
2559 }
2560 
2561 static int kcore_copy__compare_fds(int from, int to)
2562 {
2563 	char *buf_from;
2564 	char *buf_to;
2565 	ssize_t ret;
2566 	size_t len;
2567 	int err = -1;
2568 
2569 	buf_from = malloc(page_size);
2570 	buf_to = malloc(page_size);
2571 	if (!buf_from || !buf_to)
2572 		goto out;
2573 
2574 	while (1) {
2575 		/* Use read because mmap won't work on proc files */
2576 		ret = read(from, buf_from, page_size);
2577 		if (ret < 0)
2578 			goto out;
2579 
2580 		if (!ret)
2581 			break;
2582 
2583 		len = ret;
2584 
2585 		if (readn(to, buf_to, len) != (int)len)
2586 			goto out;
2587 
2588 		if (memcmp(buf_from, buf_to, len))
2589 			goto out;
2590 	}
2591 
2592 	err = 0;
2593 out:
2594 	free(buf_to);
2595 	free(buf_from);
2596 	return err;
2597 }
2598 
2599 static int kcore_copy__compare_files(const char *from_filename,
2600 				     const char *to_filename)
2601 {
2602 	int from, to, err = -1;
2603 
2604 	from = open(from_filename, O_RDONLY);
2605 	if (from < 0)
2606 		return -1;
2607 
2608 	to = open(to_filename, O_RDONLY);
2609 	if (to < 0)
2610 		goto out_close_from;
2611 
2612 	err = kcore_copy__compare_fds(from, to);
2613 
2614 	close(to);
2615 out_close_from:
2616 	close(from);
2617 	return err;
2618 }
2619 
2620 static int kcore_copy__compare_file(const char *from_dir, const char *to_dir,
2621 				    const char *name)
2622 {
2623 	char from_filename[PATH_MAX];
2624 	char to_filename[PATH_MAX];
2625 
2626 	scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
2627 	scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
2628 
2629 	return kcore_copy__compare_files(from_filename, to_filename);
2630 }
2631 
2632 /**
2633  * kcore_copy - copy kallsyms, modules and kcore from one directory to another.
2634  * @from_dir: from directory
2635  * @to_dir: to directory
2636  *
2637  * This function copies kallsyms, modules and kcore files from one directory to
2638  * another.  kallsyms and modules are copied entirely.  Only code segments are
2639  * copied from kcore.  It is assumed that two segments suffice: one for the
2640  * kernel proper and one for all the modules.  The code segments are determined
2641  * from kallsyms and modules files.  The kernel map starts at _stext or the
2642  * lowest function symbol, and ends at _etext or the highest function symbol.
2643  * The module map starts at the lowest module address and ends at the highest
2644  * module symbol.  Start addresses are rounded down to the nearest page.  End
2645  * addresses are rounded up to the nearest page.  An extra page is added to the
2646  * highest kernel symbol and highest module symbol to, hopefully, encompass that
2647  * symbol too.  Because it contains only code sections, the resulting kcore is
2648  * unusual.  One significant peculiarity is that the mapping (start -> pgoff)
2649  * is not the same for the kernel map and the modules map.  That happens because
2650  * the data is copied adjacently whereas the original kcore has gaps.  Finally,
2651  * kallsyms file is compared with its copy to check that modules have not been
2652  * loaded or unloaded while the copies were taking place.
2653  *
2654  * Return: %0 on success, %-1 on failure.
2655  */
2656 int kcore_copy(const char *from_dir, const char *to_dir)
2657 {
2658 	struct kcore kcore;
2659 	struct kcore extract;
2660 	int idx = 0, err = -1;
2661 	off_t offset, sz;
2662 	struct kcore_copy_info kci = { .stext = 0, };
2663 	char kcore_filename[PATH_MAX];
2664 	char extract_filename[PATH_MAX];
2665 	struct phdr_data *p;
2666 
2667 	INIT_LIST_HEAD(&kci.phdrs);
2668 	INIT_LIST_HEAD(&kci.syms);
2669 
2670 	if (kcore_copy__copy_file(from_dir, to_dir, "kallsyms"))
2671 		return -1;
2672 
2673 	if (kcore_copy__copy_file(from_dir, to_dir, "modules"))
2674 		goto out_unlink_kallsyms;
2675 
2676 	scnprintf(kcore_filename, PATH_MAX, "%s/kcore", from_dir);
2677 	scnprintf(extract_filename, PATH_MAX, "%s/kcore", to_dir);
2678 
2679 	if (kcore__open(&kcore, kcore_filename))
2680 		goto out_unlink_modules;
2681 
2682 	if (kcore_copy__calc_maps(&kci, from_dir, kcore.elf))
2683 		goto out_kcore_close;
2684 
2685 	if (kcore__init(&extract, extract_filename, kcore.elfclass, false))
2686 		goto out_kcore_close;
2687 
2688 	if (kcore__copy_hdr(&kcore, &extract, kci.phnum))
2689 		goto out_extract_close;
2690 
2691 	offset = gelf_fsize(extract.elf, ELF_T_EHDR, 1, EV_CURRENT) +
2692 		 gelf_fsize(extract.elf, ELF_T_PHDR, kci.phnum, EV_CURRENT);
2693 	offset = round_up(offset, page_size);
2694 
2695 	kcore_copy__for_each_phdr(&kci, p) {
2696 		off_t offs = p->rel + offset;
2697 
2698 		if (kcore__add_phdr(&extract, idx++, offs, p->addr, p->len))
2699 			goto out_extract_close;
2700 	}
2701 
2702 	sz = kcore__write(&extract);
2703 	if (sz < 0 || sz > offset)
2704 		goto out_extract_close;
2705 
2706 	kcore_copy__for_each_phdr(&kci, p) {
2707 		off_t offs = p->rel + offset;
2708 
2709 		if (p->remaps)
2710 			continue;
2711 		if (copy_bytes(kcore.fd, p->offset, extract.fd, offs, p->len))
2712 			goto out_extract_close;
2713 	}
2714 
2715 	if (kcore_copy__compare_file(from_dir, to_dir, "kallsyms"))
2716 		goto out_extract_close;
2717 
2718 	err = 0;
2719 
2720 out_extract_close:
2721 	kcore__close(&extract);
2722 	if (err)
2723 		unlink(extract_filename);
2724 out_kcore_close:
2725 	kcore__close(&kcore);
2726 out_unlink_modules:
2727 	if (err)
2728 		kcore_copy__unlink(to_dir, "modules");
2729 out_unlink_kallsyms:
2730 	if (err)
2731 		kcore_copy__unlink(to_dir, "kallsyms");
2732 
2733 	kcore_copy__free_phdrs(&kci);
2734 	kcore_copy__free_syms(&kci);
2735 
2736 	return err;
2737 }
2738 
2739 int kcore_extract__create(struct kcore_extract *kce)
2740 {
2741 	struct kcore kcore;
2742 	struct kcore extract;
2743 	size_t count = 1;
2744 	int idx = 0, err = -1;
2745 	off_t offset = page_size, sz;
2746 
2747 	if (kcore__open(&kcore, kce->kcore_filename))
2748 		return -1;
2749 
2750 	strcpy(kce->extract_filename, PERF_KCORE_EXTRACT);
2751 	if (kcore__init(&extract, kce->extract_filename, kcore.elfclass, true))
2752 		goto out_kcore_close;
2753 
2754 	if (kcore__copy_hdr(&kcore, &extract, count))
2755 		goto out_extract_close;
2756 
2757 	if (kcore__add_phdr(&extract, idx, offset, kce->addr, kce->len))
2758 		goto out_extract_close;
2759 
2760 	sz = kcore__write(&extract);
2761 	if (sz < 0 || sz > offset)
2762 		goto out_extract_close;
2763 
2764 	if (copy_bytes(kcore.fd, kce->offs, extract.fd, offset, kce->len))
2765 		goto out_extract_close;
2766 
2767 	err = 0;
2768 
2769 out_extract_close:
2770 	kcore__close(&extract);
2771 	if (err)
2772 		unlink(kce->extract_filename);
2773 out_kcore_close:
2774 	kcore__close(&kcore);
2775 
2776 	return err;
2777 }
2778 
2779 void kcore_extract__delete(struct kcore_extract *kce)
2780 {
2781 	unlink(kce->extract_filename);
2782 }
2783 
2784 #ifdef HAVE_GELF_GETNOTE_SUPPORT
2785 
2786 static void sdt_adjust_loc(struct sdt_note *tmp, GElf_Addr base_off)
2787 {
2788 	if (!base_off)
2789 		return;
2790 
2791 	if (tmp->bit32)
2792 		tmp->addr.a32[SDT_NOTE_IDX_LOC] =
2793 			tmp->addr.a32[SDT_NOTE_IDX_LOC] + base_off -
2794 			tmp->addr.a32[SDT_NOTE_IDX_BASE];
2795 	else
2796 		tmp->addr.a64[SDT_NOTE_IDX_LOC] =
2797 			tmp->addr.a64[SDT_NOTE_IDX_LOC] + base_off -
2798 			tmp->addr.a64[SDT_NOTE_IDX_BASE];
2799 }
2800 
2801 static void sdt_adjust_refctr(struct sdt_note *tmp, GElf_Addr base_addr,
2802 			      GElf_Addr base_off)
2803 {
2804 	if (!base_off)
2805 		return;
2806 
2807 	if (tmp->bit32 && tmp->addr.a32[SDT_NOTE_IDX_REFCTR])
2808 		tmp->addr.a32[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2809 	else if (tmp->addr.a64[SDT_NOTE_IDX_REFCTR])
2810 		tmp->addr.a64[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2811 }
2812 
2813 /**
2814  * populate_sdt_note : Parse raw data and identify SDT note
2815  * @elf: elf of the opened file
2816  * @data: raw data of a section with description offset applied
2817  * @len: note description size
2818  * @type: type of the note
2819  * @sdt_notes: List to add the SDT note
2820  *
2821  * Responsible for parsing the @data in section .note.stapsdt in @elf and
2822  * if its an SDT note, it appends to @sdt_notes list.
2823  */
2824 static int populate_sdt_note(Elf **elf, const char *data, size_t len,
2825 			     struct list_head *sdt_notes)
2826 {
2827 	const char *provider, *name, *args;
2828 	struct sdt_note *tmp = NULL;
2829 	GElf_Ehdr ehdr;
2830 	GElf_Shdr shdr;
2831 	int ret = -EINVAL;
2832 
2833 	union {
2834 		Elf64_Addr a64[NR_ADDR];
2835 		Elf32_Addr a32[NR_ADDR];
2836 	} buf;
2837 
2838 	Elf_Data dst = {
2839 		.d_buf = &buf, .d_type = ELF_T_ADDR, .d_version = EV_CURRENT,
2840 		.d_size = gelf_fsize((*elf), ELF_T_ADDR, NR_ADDR, EV_CURRENT),
2841 		.d_off = 0, .d_align = 0
2842 	};
2843 	Elf_Data src = {
2844 		.d_buf = (void *) data, .d_type = ELF_T_ADDR,
2845 		.d_version = EV_CURRENT, .d_size = dst.d_size, .d_off = 0,
2846 		.d_align = 0
2847 	};
2848 
2849 	tmp = (struct sdt_note *)calloc(1, sizeof(struct sdt_note));
2850 	if (!tmp) {
2851 		ret = -ENOMEM;
2852 		goto out_err;
2853 	}
2854 
2855 	INIT_LIST_HEAD(&tmp->note_list);
2856 
2857 	if (len < dst.d_size + 3)
2858 		goto out_free_note;
2859 
2860 	/* Translation from file representation to memory representation */
2861 	if (gelf_xlatetom(*elf, &dst, &src,
2862 			  elf_getident(*elf, NULL)[EI_DATA]) == NULL) {
2863 		pr_err("gelf_xlatetom : %s\n", elf_errmsg(-1));
2864 		goto out_free_note;
2865 	}
2866 
2867 	/* Populate the fields of sdt_note */
2868 	provider = data + dst.d_size;
2869 
2870 	name = (const char *)memchr(provider, '\0', data + len - provider);
2871 	if (name++ == NULL)
2872 		goto out_free_note;
2873 
2874 	tmp->provider = strdup(provider);
2875 	if (!tmp->provider) {
2876 		ret = -ENOMEM;
2877 		goto out_free_note;
2878 	}
2879 	tmp->name = strdup(name);
2880 	if (!tmp->name) {
2881 		ret = -ENOMEM;
2882 		goto out_free_prov;
2883 	}
2884 
2885 	args = memchr(name, '\0', data + len - name);
2886 
2887 	/*
2888 	 * There is no argument if:
2889 	 * - We reached the end of the note;
2890 	 * - There is not enough room to hold a potential string;
2891 	 * - The argument string is empty or just contains ':'.
2892 	 */
2893 	if (args == NULL || data + len - args < 2 ||
2894 		args[1] == ':' || args[1] == '\0')
2895 		tmp->args = NULL;
2896 	else {
2897 		tmp->args = strdup(++args);
2898 		if (!tmp->args) {
2899 			ret = -ENOMEM;
2900 			goto out_free_name;
2901 		}
2902 	}
2903 
2904 	if (gelf_getclass(*elf) == ELFCLASS32) {
2905 		memcpy(&tmp->addr, &buf, 3 * sizeof(Elf32_Addr));
2906 		tmp->bit32 = true;
2907 	} else {
2908 		memcpy(&tmp->addr, &buf, 3 * sizeof(Elf64_Addr));
2909 		tmp->bit32 = false;
2910 	}
2911 
2912 	if (!gelf_getehdr(*elf, &ehdr)) {
2913 		pr_debug("%s : cannot get elf header.\n", __func__);
2914 		ret = -EBADF;
2915 		goto out_free_args;
2916 	}
2917 
2918 	/* Adjust the prelink effect :
2919 	 * Find out the .stapsdt.base section.
2920 	 * This scn will help us to handle prelinking (if present).
2921 	 * Compare the retrieved file offset of the base section with the
2922 	 * base address in the description of the SDT note. If its different,
2923 	 * then accordingly, adjust the note location.
2924 	 */
2925 	if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_BASE_SCN, NULL))
2926 		sdt_adjust_loc(tmp, shdr.sh_offset);
2927 
2928 	/* Adjust reference counter offset */
2929 	if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_PROBES_SCN, NULL))
2930 		sdt_adjust_refctr(tmp, shdr.sh_addr, shdr.sh_offset);
2931 
2932 	list_add_tail(&tmp->note_list, sdt_notes);
2933 	return 0;
2934 
2935 out_free_args:
2936 	zfree(&tmp->args);
2937 out_free_name:
2938 	zfree(&tmp->name);
2939 out_free_prov:
2940 	zfree(&tmp->provider);
2941 out_free_note:
2942 	free(tmp);
2943 out_err:
2944 	return ret;
2945 }
2946 
2947 /**
2948  * construct_sdt_notes_list : constructs a list of SDT notes
2949  * @elf : elf to look into
2950  * @sdt_notes : empty list_head
2951  *
2952  * Scans the sections in 'elf' for the section
2953  * .note.stapsdt. It, then calls populate_sdt_note to find
2954  * out the SDT events and populates the 'sdt_notes'.
2955  */
2956 static int construct_sdt_notes_list(Elf *elf, struct list_head *sdt_notes)
2957 {
2958 	GElf_Ehdr ehdr;
2959 	Elf_Scn *scn = NULL;
2960 	Elf_Data *data;
2961 	GElf_Shdr shdr;
2962 	size_t shstrndx, next;
2963 	GElf_Nhdr nhdr;
2964 	size_t name_off, desc_off, offset;
2965 	int ret = 0;
2966 
2967 	if (gelf_getehdr(elf, &ehdr) == NULL) {
2968 		ret = -EBADF;
2969 		goto out_ret;
2970 	}
2971 	if (elf_getshdrstrndx(elf, &shstrndx) != 0) {
2972 		ret = -EBADF;
2973 		goto out_ret;
2974 	}
2975 
2976 	/* Look for the required section */
2977 	scn = elf_section_by_name(elf, &ehdr, &shdr, SDT_NOTE_SCN, NULL);
2978 	if (!scn) {
2979 		ret = -ENOENT;
2980 		goto out_ret;
2981 	}
2982 
2983 	if ((shdr.sh_type != SHT_NOTE) || (shdr.sh_flags & SHF_ALLOC)) {
2984 		ret = -ENOENT;
2985 		goto out_ret;
2986 	}
2987 
2988 	data = elf_getdata(scn, NULL);
2989 
2990 	/* Get the SDT notes */
2991 	for (offset = 0; (next = gelf_getnote(data, offset, &nhdr, &name_off,
2992 					      &desc_off)) > 0; offset = next) {
2993 		if (nhdr.n_namesz == sizeof(SDT_NOTE_NAME) &&
2994 		    !memcmp(data->d_buf + name_off, SDT_NOTE_NAME,
2995 			    sizeof(SDT_NOTE_NAME))) {
2996 			/* Check the type of the note */
2997 			if (nhdr.n_type != SDT_NOTE_TYPE)
2998 				goto out_ret;
2999 
3000 			ret = populate_sdt_note(&elf, ((data->d_buf) + desc_off),
3001 						nhdr.n_descsz, sdt_notes);
3002 			if (ret < 0)
3003 				goto out_ret;
3004 		}
3005 	}
3006 	if (list_empty(sdt_notes))
3007 		ret = -ENOENT;
3008 
3009 out_ret:
3010 	return ret;
3011 }
3012 
3013 /**
3014  * get_sdt_note_list : Wrapper to construct a list of sdt notes
3015  * @head : empty list_head
3016  * @target : file to find SDT notes from
3017  *
3018  * This opens the file, initializes
3019  * the ELF and then calls construct_sdt_notes_list.
3020  */
3021 int get_sdt_note_list(struct list_head *head, const char *target)
3022 {
3023 	Elf *elf;
3024 	int fd, ret;
3025 
3026 	fd = open(target, O_RDONLY);
3027 	if (fd < 0)
3028 		return -EBADF;
3029 
3030 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
3031 	if (!elf) {
3032 		ret = -EBADF;
3033 		goto out_close;
3034 	}
3035 	ret = construct_sdt_notes_list(elf, head);
3036 	elf_end(elf);
3037 out_close:
3038 	close(fd);
3039 	return ret;
3040 }
3041 
3042 /**
3043  * cleanup_sdt_note_list : free the sdt notes' list
3044  * @sdt_notes: sdt notes' list
3045  *
3046  * Free up the SDT notes in @sdt_notes.
3047  * Returns the number of SDT notes free'd.
3048  */
3049 int cleanup_sdt_note_list(struct list_head *sdt_notes)
3050 {
3051 	struct sdt_note *tmp, *pos;
3052 	int nr_free = 0;
3053 
3054 	list_for_each_entry_safe(pos, tmp, sdt_notes, note_list) {
3055 		list_del_init(&pos->note_list);
3056 		zfree(&pos->args);
3057 		zfree(&pos->name);
3058 		zfree(&pos->provider);
3059 		free(pos);
3060 		nr_free++;
3061 	}
3062 	return nr_free;
3063 }
3064 
3065 /**
3066  * sdt_notes__get_count: Counts the number of sdt events
3067  * @start: list_head to sdt_notes list
3068  *
3069  * Returns the number of SDT notes in a list
3070  */
3071 int sdt_notes__get_count(struct list_head *start)
3072 {
3073 	struct sdt_note *sdt_ptr;
3074 	int count = 0;
3075 
3076 	list_for_each_entry(sdt_ptr, start, note_list)
3077 		count++;
3078 	return count;
3079 }
3080 #endif
3081 
3082 void symbol__elf_init(void)
3083 {
3084 	elf_version(EV_CURRENT);
3085 }
3086