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 "dso.h"
11 #include "map.h"
12 #include "maps.h"
13 #include "symbol.h"
14 #include "symsrc.h"
15 #include "demangle-ocaml.h"
16 #include "demangle-java.h"
17 #include "demangle-rust.h"
18 #include "machine.h"
19 #include "vdso.h"
20 #include "debug.h"
21 #include "util/copyfile.h"
22 #include <linux/ctype.h>
23 #include <linux/kernel.h>
24 #include <linux/zalloc.h>
25 #include <symbol/kallsyms.h>
26 #include <internal/lib.h>
27 
28 #ifndef EM_AARCH64
29 #define EM_AARCH64	183  /* ARM 64 bit */
30 #endif
31 
32 #ifndef ELF32_ST_VISIBILITY
33 #define ELF32_ST_VISIBILITY(o)	((o) & 0x03)
34 #endif
35 
36 /* For ELF64 the definitions are the same.  */
37 #ifndef ELF64_ST_VISIBILITY
38 #define ELF64_ST_VISIBILITY(o)	ELF32_ST_VISIBILITY (o)
39 #endif
40 
41 /* How to extract information held in the st_other field.  */
42 #ifndef GELF_ST_VISIBILITY
43 #define GELF_ST_VISIBILITY(val)	ELF64_ST_VISIBILITY (val)
44 #endif
45 
46 typedef Elf64_Nhdr GElf_Nhdr;
47 
48 #ifndef DMGL_PARAMS
49 #define DMGL_NO_OPTS     0              /* For readability... */
50 #define DMGL_PARAMS      (1 << 0)       /* Include function args */
51 #define DMGL_ANSI        (1 << 1)       /* Include const, volatile, etc */
52 #endif
53 
54 #ifdef HAVE_LIBBFD_SUPPORT
55 #define PACKAGE 'perf'
56 #include <bfd.h>
57 #else
58 #ifdef HAVE_CPLUS_DEMANGLE_SUPPORT
59 extern char *cplus_demangle(const char *, int);
60 
bfd_demangle(void __maybe_unused * v,const char * c,int i)61 static inline char *bfd_demangle(void __maybe_unused *v, const char *c, int i)
62 {
63 	return cplus_demangle(c, i);
64 }
65 #else
66 #ifdef NO_DEMANGLE
bfd_demangle(void __maybe_unused * v,const char __maybe_unused * c,int __maybe_unused i)67 static inline char *bfd_demangle(void __maybe_unused *v,
68 				 const char __maybe_unused *c,
69 				 int __maybe_unused i)
70 {
71 	return NULL;
72 }
73 #endif
74 #endif
75 #endif
76 
77 #ifndef HAVE_ELF_GETPHDRNUM_SUPPORT
elf_getphdrnum(Elf * elf,size_t * dst)78 static int elf_getphdrnum(Elf *elf, size_t *dst)
79 {
80 	GElf_Ehdr gehdr;
81 	GElf_Ehdr *ehdr;
82 
83 	ehdr = gelf_getehdr(elf, &gehdr);
84 	if (!ehdr)
85 		return -1;
86 
87 	*dst = ehdr->e_phnum;
88 
89 	return 0;
90 }
91 #endif
92 
93 #ifndef HAVE_ELF_GETSHDRSTRNDX_SUPPORT
elf_getshdrstrndx(Elf * elf __maybe_unused,size_t * dst __maybe_unused)94 static int elf_getshdrstrndx(Elf *elf __maybe_unused, size_t *dst __maybe_unused)
95 {
96 	pr_err("%s: update your libelf to > 0.140, this one lacks elf_getshdrstrndx().\n", __func__);
97 	return -1;
98 }
99 #endif
100 
101 #ifndef NT_GNU_BUILD_ID
102 #define NT_GNU_BUILD_ID 3
103 #endif
104 
105 /**
106  * elf_symtab__for_each_symbol - iterate thru all the symbols
107  *
108  * @syms: struct elf_symtab instance to iterate
109  * @idx: uint32_t idx
110  * @sym: GElf_Sym iterator
111  */
112 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
113 	for (idx = 0, gelf_getsym(syms, idx, &sym);\
114 	     idx < nr_syms; \
115 	     idx++, gelf_getsym(syms, idx, &sym))
116 
elf_sym__type(const GElf_Sym * sym)117 static inline uint8_t elf_sym__type(const GElf_Sym *sym)
118 {
119 	return GELF_ST_TYPE(sym->st_info);
120 }
121 
elf_sym__visibility(const GElf_Sym * sym)122 static inline uint8_t elf_sym__visibility(const GElf_Sym *sym)
123 {
124 	return GELF_ST_VISIBILITY(sym->st_other);
125 }
126 
127 #ifndef STT_GNU_IFUNC
128 #define STT_GNU_IFUNC 10
129 #endif
130 
elf_sym__is_function(const GElf_Sym * sym)131 static inline int elf_sym__is_function(const GElf_Sym *sym)
132 {
133 	return (elf_sym__type(sym) == STT_FUNC ||
134 		elf_sym__type(sym) == STT_GNU_IFUNC) &&
135 	       sym->st_name != 0 &&
136 	       sym->st_shndx != SHN_UNDEF;
137 }
138 
elf_sym__is_object(const GElf_Sym * sym)139 static inline bool elf_sym__is_object(const GElf_Sym *sym)
140 {
141 	return elf_sym__type(sym) == STT_OBJECT &&
142 		sym->st_name != 0 &&
143 		sym->st_shndx != SHN_UNDEF;
144 }
145 
elf_sym__is_label(const GElf_Sym * sym)146 static inline int elf_sym__is_label(const GElf_Sym *sym)
147 {
148 	return elf_sym__type(sym) == STT_NOTYPE &&
149 		sym->st_name != 0 &&
150 		sym->st_shndx != SHN_UNDEF &&
151 		sym->st_shndx != SHN_ABS &&
152 		elf_sym__visibility(sym) != STV_HIDDEN &&
153 		elf_sym__visibility(sym) != STV_INTERNAL;
154 }
155 
elf_sym__filter(GElf_Sym * sym)156 static bool elf_sym__filter(GElf_Sym *sym)
157 {
158 	return elf_sym__is_function(sym) || elf_sym__is_object(sym);
159 }
160 
elf_sym__name(const GElf_Sym * sym,const Elf_Data * symstrs)161 static inline const char *elf_sym__name(const GElf_Sym *sym,
162 					const Elf_Data *symstrs)
163 {
164 	return symstrs->d_buf + sym->st_name;
165 }
166 
elf_sec__name(const GElf_Shdr * shdr,const Elf_Data * secstrs)167 static inline const char *elf_sec__name(const GElf_Shdr *shdr,
168 					const Elf_Data *secstrs)
169 {
170 	return secstrs->d_buf + shdr->sh_name;
171 }
172 
elf_sec__is_text(const GElf_Shdr * shdr,const Elf_Data * secstrs)173 static inline int elf_sec__is_text(const GElf_Shdr *shdr,
174 					const Elf_Data *secstrs)
175 {
176 	return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
177 }
178 
elf_sec__is_data(const GElf_Shdr * shdr,const Elf_Data * secstrs)179 static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
180 				    const Elf_Data *secstrs)
181 {
182 	return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
183 }
184 
elf_sec__filter(GElf_Shdr * shdr,Elf_Data * secstrs)185 static bool elf_sec__filter(GElf_Shdr *shdr, Elf_Data *secstrs)
186 {
187 	return elf_sec__is_text(shdr, secstrs) ||
188 	       elf_sec__is_data(shdr, secstrs);
189 }
190 
elf_addr_to_index(Elf * elf,GElf_Addr addr)191 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
192 {
193 	Elf_Scn *sec = NULL;
194 	GElf_Shdr shdr;
195 	size_t cnt = 1;
196 
197 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
198 		gelf_getshdr(sec, &shdr);
199 
200 		if ((addr >= shdr.sh_addr) &&
201 		    (addr < (shdr.sh_addr + shdr.sh_size)))
202 			return cnt;
203 
204 		++cnt;
205 	}
206 
207 	return -1;
208 }
209 
elf_section_by_name(Elf * elf,GElf_Ehdr * ep,GElf_Shdr * shp,const char * name,size_t * idx)210 Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
211 			     GElf_Shdr *shp, const char *name, size_t *idx)
212 {
213 	Elf_Scn *sec = NULL;
214 	size_t cnt = 1;
215 
216 	/* Elf is corrupted/truncated, avoid calling elf_strptr. */
217 	if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL))
218 		return NULL;
219 
220 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
221 		char *str;
222 
223 		gelf_getshdr(sec, shp);
224 		str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
225 		if (str && !strcmp(name, str)) {
226 			if (idx)
227 				*idx = cnt;
228 			return sec;
229 		}
230 		++cnt;
231 	}
232 
233 	return NULL;
234 }
235 
elf_read_program_header(Elf * elf,u64 vaddr,GElf_Phdr * phdr)236 static int elf_read_program_header(Elf *elf, u64 vaddr, GElf_Phdr *phdr)
237 {
238 	size_t i, phdrnum;
239 	u64 sz;
240 
241 	if (elf_getphdrnum(elf, &phdrnum))
242 		return -1;
243 
244 	for (i = 0; i < phdrnum; i++) {
245 		if (gelf_getphdr(elf, i, phdr) == NULL)
246 			return -1;
247 
248 		if (phdr->p_type != PT_LOAD)
249 			continue;
250 
251 		sz = max(phdr->p_memsz, phdr->p_filesz);
252 		if (!sz)
253 			continue;
254 
255 		if (vaddr >= phdr->p_vaddr && (vaddr < phdr->p_vaddr + sz))
256 			return 0;
257 	}
258 
259 	/* Not found any valid program header */
260 	return -1;
261 }
262 
want_demangle(bool is_kernel_sym)263 static bool want_demangle(bool is_kernel_sym)
264 {
265 	return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle;
266 }
267 
demangle_sym(struct dso * dso,int kmodule,const char * elf_name)268 static char *demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
269 {
270 	int demangle_flags = verbose > 0 ? (DMGL_PARAMS | DMGL_ANSI) : DMGL_NO_OPTS;
271 	char *demangled = NULL;
272 
273 	/*
274 	 * We need to figure out if the object was created from C++ sources
275 	 * DWARF DW_compile_unit has this, but we don't always have access
276 	 * to it...
277 	 */
278 	if (!want_demangle(dso->kernel || kmodule))
279 	    return demangled;
280 
281 	demangled = bfd_demangle(NULL, elf_name, demangle_flags);
282 	if (demangled == NULL) {
283 		demangled = ocaml_demangle_sym(elf_name);
284 		if (demangled == NULL) {
285 			demangled = java_demangle_sym(elf_name, JAVA_DEMANGLE_NORET);
286 		}
287 	}
288 	else if (rust_is_mangled(demangled))
289 		/*
290 		    * Input to Rust demangling is the BFD-demangled
291 		    * name which it Rust-demangles in place.
292 		    */
293 		rust_demangle_sym(demangled);
294 
295 	return demangled;
296 }
297 
298 #define elf_section__for_each_rel(reldata, pos, pos_mem, idx, nr_entries) \
299 	for (idx = 0, pos = gelf_getrel(reldata, 0, &pos_mem); \
300 	     idx < nr_entries; \
301 	     ++idx, pos = gelf_getrel(reldata, idx, &pos_mem))
302 
303 #define elf_section__for_each_rela(reldata, pos, pos_mem, idx, nr_entries) \
304 	for (idx = 0, pos = gelf_getrela(reldata, 0, &pos_mem); \
305 	     idx < nr_entries; \
306 	     ++idx, pos = gelf_getrela(reldata, idx, &pos_mem))
307 
308 /*
309  * We need to check if we have a .dynsym, so that we can handle the
310  * .plt, synthesizing its symbols, that aren't on the symtabs (be it
311  * .dynsym or .symtab).
312  * And always look at the original dso, not at debuginfo packages, that
313  * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
314  */
dso__synthesize_plt_symbols(struct dso * dso,struct symsrc * ss)315 int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss)
316 {
317 	uint32_t nr_rel_entries, idx;
318 	GElf_Sym sym;
319 	u64 plt_offset, plt_header_size, plt_entry_size;
320 	GElf_Shdr shdr_plt;
321 	struct symbol *f;
322 	GElf_Shdr shdr_rel_plt, shdr_dynsym;
323 	Elf_Data *reldata, *syms, *symstrs;
324 	Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
325 	size_t dynsym_idx;
326 	GElf_Ehdr ehdr;
327 	char sympltname[1024];
328 	Elf *elf;
329 	int nr = 0, symidx, err = 0;
330 
331 	if (!ss->dynsym)
332 		return 0;
333 
334 	elf = ss->elf;
335 	ehdr = ss->ehdr;
336 
337 	scn_dynsym = ss->dynsym;
338 	shdr_dynsym = ss->dynshdr;
339 	dynsym_idx = ss->dynsym_idx;
340 
341 	if (scn_dynsym == NULL)
342 		goto out_elf_end;
343 
344 	scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
345 					  ".rela.plt", NULL);
346 	if (scn_plt_rel == NULL) {
347 		scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
348 						  ".rel.plt", NULL);
349 		if (scn_plt_rel == NULL)
350 			goto out_elf_end;
351 	}
352 
353 	err = -1;
354 
355 	if (shdr_rel_plt.sh_link != dynsym_idx)
356 		goto out_elf_end;
357 
358 	if (elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL) == NULL)
359 		goto out_elf_end;
360 
361 	/*
362 	 * Fetch the relocation section to find the idxes to the GOT
363 	 * and the symbols in the .dynsym they refer to.
364 	 */
365 	reldata = elf_getdata(scn_plt_rel, NULL);
366 	if (reldata == NULL)
367 		goto out_elf_end;
368 
369 	syms = elf_getdata(scn_dynsym, NULL);
370 	if (syms == NULL)
371 		goto out_elf_end;
372 
373 	scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
374 	if (scn_symstrs == NULL)
375 		goto out_elf_end;
376 
377 	symstrs = elf_getdata(scn_symstrs, NULL);
378 	if (symstrs == NULL)
379 		goto out_elf_end;
380 
381 	if (symstrs->d_size == 0)
382 		goto out_elf_end;
383 
384 	nr_rel_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
385 	plt_offset = shdr_plt.sh_offset;
386 	switch (ehdr.e_machine) {
387 		case EM_ARM:
388 			plt_header_size = 20;
389 			plt_entry_size = 12;
390 			break;
391 
392 		case EM_AARCH64:
393 			plt_header_size = 32;
394 			plt_entry_size = 16;
395 			break;
396 
397 		case EM_SPARC:
398 			plt_header_size = 48;
399 			plt_entry_size = 12;
400 			break;
401 
402 		case EM_SPARCV9:
403 			plt_header_size = 128;
404 			plt_entry_size = 32;
405 			break;
406 
407 		default: /* FIXME: s390/alpha/mips/parisc/poperpc/sh/xtensa need to be checked */
408 			plt_header_size = shdr_plt.sh_entsize;
409 			plt_entry_size = shdr_plt.sh_entsize;
410 			break;
411 	}
412 	plt_offset += plt_header_size;
413 
414 	if (shdr_rel_plt.sh_type == SHT_RELA) {
415 		GElf_Rela pos_mem, *pos;
416 
417 		elf_section__for_each_rela(reldata, pos, pos_mem, idx,
418 					   nr_rel_entries) {
419 			const char *elf_name = NULL;
420 			char *demangled = NULL;
421 			symidx = GELF_R_SYM(pos->r_info);
422 			gelf_getsym(syms, symidx, &sym);
423 
424 			elf_name = elf_sym__name(&sym, symstrs);
425 			demangled = demangle_sym(dso, 0, elf_name);
426 			if (demangled != NULL)
427 				elf_name = demangled;
428 			snprintf(sympltname, sizeof(sympltname),
429 				 "%s@plt", elf_name);
430 			free(demangled);
431 
432 			f = symbol__new(plt_offset, plt_entry_size,
433 					STB_GLOBAL, STT_FUNC, sympltname);
434 			if (!f)
435 				goto out_elf_end;
436 
437 			plt_offset += plt_entry_size;
438 			symbols__insert(&dso->symbols, f);
439 			++nr;
440 		}
441 	} else if (shdr_rel_plt.sh_type == SHT_REL) {
442 		GElf_Rel pos_mem, *pos;
443 		elf_section__for_each_rel(reldata, pos, pos_mem, idx,
444 					  nr_rel_entries) {
445 			const char *elf_name = NULL;
446 			char *demangled = NULL;
447 			symidx = GELF_R_SYM(pos->r_info);
448 			gelf_getsym(syms, symidx, &sym);
449 
450 			elf_name = elf_sym__name(&sym, symstrs);
451 			demangled = demangle_sym(dso, 0, elf_name);
452 			if (demangled != NULL)
453 				elf_name = demangled;
454 			snprintf(sympltname, sizeof(sympltname),
455 				 "%s@plt", elf_name);
456 			free(demangled);
457 
458 			f = symbol__new(plt_offset, plt_entry_size,
459 					STB_GLOBAL, STT_FUNC, sympltname);
460 			if (!f)
461 				goto out_elf_end;
462 
463 			plt_offset += plt_entry_size;
464 			symbols__insert(&dso->symbols, f);
465 			++nr;
466 		}
467 	}
468 
469 	err = 0;
470 out_elf_end:
471 	if (err == 0)
472 		return nr;
473 	pr_debug("%s: problems reading %s PLT info.\n",
474 		 __func__, dso->long_name);
475 	return 0;
476 }
477 
dso__demangle_sym(struct dso * dso,int kmodule,const char * elf_name)478 char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
479 {
480 	return demangle_sym(dso, kmodule, elf_name);
481 }
482 
483 /*
484  * Align offset to 4 bytes as needed for note name and descriptor data.
485  */
486 #define NOTE_ALIGN(n) (((n) + 3) & -4U)
487 
elf_read_build_id(Elf * elf,void * bf,size_t size)488 static int elf_read_build_id(Elf *elf, void *bf, size_t size)
489 {
490 	int err = -1;
491 	GElf_Ehdr ehdr;
492 	GElf_Shdr shdr;
493 	Elf_Data *data;
494 	Elf_Scn *sec;
495 	Elf_Kind ek;
496 	void *ptr;
497 
498 	if (size < BUILD_ID_SIZE)
499 		goto out;
500 
501 	ek = elf_kind(elf);
502 	if (ek != ELF_K_ELF)
503 		goto out;
504 
505 	if (gelf_getehdr(elf, &ehdr) == NULL) {
506 		pr_err("%s: cannot get elf header.\n", __func__);
507 		goto out;
508 	}
509 
510 	/*
511 	 * Check following sections for notes:
512 	 *   '.note.gnu.build-id'
513 	 *   '.notes'
514 	 *   '.note' (VDSO specific)
515 	 */
516 	do {
517 		sec = elf_section_by_name(elf, &ehdr, &shdr,
518 					  ".note.gnu.build-id", NULL);
519 		if (sec)
520 			break;
521 
522 		sec = elf_section_by_name(elf, &ehdr, &shdr,
523 					  ".notes", NULL);
524 		if (sec)
525 			break;
526 
527 		sec = elf_section_by_name(elf, &ehdr, &shdr,
528 					  ".note", NULL);
529 		if (sec)
530 			break;
531 
532 		return err;
533 
534 	} while (0);
535 
536 	data = elf_getdata(sec, NULL);
537 	if (data == NULL)
538 		goto out;
539 
540 	ptr = data->d_buf;
541 	while (ptr < (data->d_buf + data->d_size)) {
542 		GElf_Nhdr *nhdr = ptr;
543 		size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
544 		       descsz = NOTE_ALIGN(nhdr->n_descsz);
545 		const char *name;
546 
547 		ptr += sizeof(*nhdr);
548 		name = ptr;
549 		ptr += namesz;
550 		if (nhdr->n_type == NT_GNU_BUILD_ID &&
551 		    nhdr->n_namesz == sizeof("GNU")) {
552 			if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
553 				size_t sz = min(size, descsz);
554 				memcpy(bf, ptr, sz);
555 				memset(bf + sz, 0, size - sz);
556 				err = descsz;
557 				break;
558 			}
559 		}
560 		ptr += descsz;
561 	}
562 
563 out:
564 	return err;
565 }
566 
567 #ifdef HAVE_LIBBFD_BUILDID_SUPPORT
568 
read_build_id(const char * filename,struct build_id * bid)569 static int read_build_id(const char *filename, struct build_id *bid)
570 {
571 	size_t size = sizeof(bid->data);
572 	int err = -1;
573 	bfd *abfd;
574 
575 	abfd = bfd_openr(filename, NULL);
576 	if (!abfd)
577 		return -1;
578 
579 	if (!bfd_check_format(abfd, bfd_object)) {
580 		pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
581 		goto out_close;
582 	}
583 
584 	if (!abfd->build_id || abfd->build_id->size > size)
585 		goto out_close;
586 
587 	memcpy(bid->data, abfd->build_id->data, abfd->build_id->size);
588 	memset(bid->data + abfd->build_id->size, 0, size - abfd->build_id->size);
589 	err = bid->size = abfd->build_id->size;
590 
591 out_close:
592 	bfd_close(abfd);
593 	return err;
594 }
595 
596 #else // HAVE_LIBBFD_BUILDID_SUPPORT
597 
read_build_id(const char * filename,struct build_id * bid)598 static int read_build_id(const char *filename, struct build_id *bid)
599 {
600 	size_t size = sizeof(bid->data);
601 	int fd, err = -1;
602 	Elf *elf;
603 
604 	if (size < BUILD_ID_SIZE)
605 		goto out;
606 
607 	fd = open(filename, O_RDONLY);
608 	if (fd < 0)
609 		goto out;
610 
611 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
612 	if (elf == NULL) {
613 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
614 		goto out_close;
615 	}
616 
617 	err = elf_read_build_id(elf, bid->data, size);
618 	if (err > 0)
619 		bid->size = err;
620 
621 	elf_end(elf);
622 out_close:
623 	close(fd);
624 out:
625 	return err;
626 }
627 
628 #endif // HAVE_LIBBFD_BUILDID_SUPPORT
629 
filename__read_build_id(const char * filename,struct build_id * bid)630 int filename__read_build_id(const char *filename, struct build_id *bid)
631 {
632 	struct kmod_path m = { .name = NULL, };
633 	char path[PATH_MAX];
634 	int err;
635 
636 	if (!filename)
637 		return -EFAULT;
638 
639 	err = kmod_path__parse(&m, filename);
640 	if (err)
641 		return -1;
642 
643 	if (m.comp) {
644 		int error = 0, fd;
645 
646 		fd = filename__decompress(filename, path, sizeof(path), m.comp, &error);
647 		if (fd < 0) {
648 			pr_debug("Failed to decompress (error %d) %s\n",
649 				 error, filename);
650 			return -1;
651 		}
652 		close(fd);
653 		filename = path;
654 	}
655 
656 	err = read_build_id(filename, bid);
657 
658 	if (m.comp)
659 		unlink(filename);
660 	return err;
661 }
662 
sysfs__read_build_id(const char * filename,struct build_id * bid)663 int sysfs__read_build_id(const char *filename, struct build_id *bid)
664 {
665 	size_t size = sizeof(bid->data);
666 	int fd, err = -1;
667 
668 	fd = open(filename, O_RDONLY);
669 	if (fd < 0)
670 		goto out;
671 
672 	while (1) {
673 		char bf[BUFSIZ];
674 		GElf_Nhdr nhdr;
675 		size_t namesz, descsz;
676 
677 		if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
678 			break;
679 
680 		namesz = NOTE_ALIGN(nhdr.n_namesz);
681 		descsz = NOTE_ALIGN(nhdr.n_descsz);
682 		if (nhdr.n_type == NT_GNU_BUILD_ID &&
683 		    nhdr.n_namesz == sizeof("GNU")) {
684 			if (read(fd, bf, namesz) != (ssize_t)namesz)
685 				break;
686 			if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
687 				size_t sz = min(descsz, size);
688 				if (read(fd, bid->data, sz) == (ssize_t)sz) {
689 					memset(bid->data + sz, 0, size - sz);
690 					bid->size = sz;
691 					err = 0;
692 					break;
693 				}
694 			} else if (read(fd, bf, descsz) != (ssize_t)descsz)
695 				break;
696 		} else {
697 			int n = namesz + descsz;
698 
699 			if (n > (int)sizeof(bf)) {
700 				n = sizeof(bf);
701 				pr_debug("%s: truncating reading of build id in sysfs file %s: n_namesz=%u, n_descsz=%u.\n",
702 					 __func__, filename, nhdr.n_namesz, nhdr.n_descsz);
703 			}
704 			if (read(fd, bf, n) != n)
705 				break;
706 		}
707 	}
708 	close(fd);
709 out:
710 	return err;
711 }
712 
713 #ifdef HAVE_LIBBFD_SUPPORT
714 
filename__read_debuglink(const char * filename,char * debuglink,size_t size)715 int filename__read_debuglink(const char *filename, char *debuglink,
716 			     size_t size)
717 {
718 	int err = -1;
719 	asection *section;
720 	bfd *abfd;
721 
722 	abfd = bfd_openr(filename, NULL);
723 	if (!abfd)
724 		return -1;
725 
726 	if (!bfd_check_format(abfd, bfd_object)) {
727 		pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
728 		goto out_close;
729 	}
730 
731 	section = bfd_get_section_by_name(abfd, ".gnu_debuglink");
732 	if (!section)
733 		goto out_close;
734 
735 	if (section->size > size)
736 		goto out_close;
737 
738 	if (!bfd_get_section_contents(abfd, section, debuglink, 0,
739 				      section->size))
740 		goto out_close;
741 
742 	err = 0;
743 
744 out_close:
745 	bfd_close(abfd);
746 	return err;
747 }
748 
749 #else
750 
filename__read_debuglink(const char * filename,char * debuglink,size_t size)751 int filename__read_debuglink(const char *filename, char *debuglink,
752 			     size_t size)
753 {
754 	int fd, err = -1;
755 	Elf *elf;
756 	GElf_Ehdr ehdr;
757 	GElf_Shdr shdr;
758 	Elf_Data *data;
759 	Elf_Scn *sec;
760 	Elf_Kind ek;
761 
762 	fd = open(filename, O_RDONLY);
763 	if (fd < 0)
764 		goto out;
765 
766 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
767 	if (elf == NULL) {
768 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
769 		goto out_close;
770 	}
771 
772 	ek = elf_kind(elf);
773 	if (ek != ELF_K_ELF)
774 		goto out_elf_end;
775 
776 	if (gelf_getehdr(elf, &ehdr) == NULL) {
777 		pr_err("%s: cannot get elf header.\n", __func__);
778 		goto out_elf_end;
779 	}
780 
781 	sec = elf_section_by_name(elf, &ehdr, &shdr,
782 				  ".gnu_debuglink", NULL);
783 	if (sec == NULL)
784 		goto out_elf_end;
785 
786 	data = elf_getdata(sec, NULL);
787 	if (data == NULL)
788 		goto out_elf_end;
789 
790 	/* the start of this section is a zero-terminated string */
791 	strncpy(debuglink, data->d_buf, size);
792 
793 	err = 0;
794 
795 out_elf_end:
796 	elf_end(elf);
797 out_close:
798 	close(fd);
799 out:
800 	return err;
801 }
802 
803 #endif
804 
dso__swap_init(struct dso * dso,unsigned char eidata)805 static int dso__swap_init(struct dso *dso, unsigned char eidata)
806 {
807 	static unsigned int const endian = 1;
808 
809 	dso->needs_swap = DSO_SWAP__NO;
810 
811 	switch (eidata) {
812 	case ELFDATA2LSB:
813 		/* We are big endian, DSO is little endian. */
814 		if (*(unsigned char const *)&endian != 1)
815 			dso->needs_swap = DSO_SWAP__YES;
816 		break;
817 
818 	case ELFDATA2MSB:
819 		/* We are little endian, DSO is big endian. */
820 		if (*(unsigned char const *)&endian != 0)
821 			dso->needs_swap = DSO_SWAP__YES;
822 		break;
823 
824 	default:
825 		pr_err("unrecognized DSO data encoding %d\n", eidata);
826 		return -EINVAL;
827 	}
828 
829 	return 0;
830 }
831 
symsrc__possibly_runtime(struct symsrc * ss)832 bool symsrc__possibly_runtime(struct symsrc *ss)
833 {
834 	return ss->dynsym || ss->opdsec;
835 }
836 
symsrc__has_symtab(struct symsrc * ss)837 bool symsrc__has_symtab(struct symsrc *ss)
838 {
839 	return ss->symtab != NULL;
840 }
841 
symsrc__destroy(struct symsrc * ss)842 void symsrc__destroy(struct symsrc *ss)
843 {
844 	zfree(&ss->name);
845 	elf_end(ss->elf);
846 	close(ss->fd);
847 }
848 
elf__needs_adjust_symbols(GElf_Ehdr ehdr)849 bool elf__needs_adjust_symbols(GElf_Ehdr ehdr)
850 {
851 	/*
852 	 * Usually vmlinux is an ELF file with type ET_EXEC for most
853 	 * architectures; except Arm64 kernel is linked with option
854 	 * '-share', so need to check type ET_DYN.
855 	 */
856 	return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL ||
857 	       ehdr.e_type == ET_DYN;
858 }
859 
symsrc__init(struct symsrc * ss,struct dso * dso,const char * name,enum dso_binary_type type)860 int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
861 		 enum dso_binary_type type)
862 {
863 	GElf_Ehdr ehdr;
864 	Elf *elf;
865 	int fd;
866 
867 	if (dso__needs_decompress(dso)) {
868 		fd = dso__decompress_kmodule_fd(dso, name);
869 		if (fd < 0)
870 			return -1;
871 
872 		type = dso->symtab_type;
873 	} else {
874 		fd = open(name, O_RDONLY);
875 		if (fd < 0) {
876 			dso->load_errno = errno;
877 			return -1;
878 		}
879 	}
880 
881 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
882 	if (elf == NULL) {
883 		pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
884 		dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
885 		goto out_close;
886 	}
887 
888 	if (gelf_getehdr(elf, &ehdr) == NULL) {
889 		dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
890 		pr_debug("%s: cannot get elf header.\n", __func__);
891 		goto out_elf_end;
892 	}
893 
894 	if (dso__swap_init(dso, ehdr.e_ident[EI_DATA])) {
895 		dso->load_errno = DSO_LOAD_ERRNO__INTERNAL_ERROR;
896 		goto out_elf_end;
897 	}
898 
899 	/* Always reject images with a mismatched build-id: */
900 	if (dso->has_build_id && !symbol_conf.ignore_vmlinux_buildid) {
901 		u8 build_id[BUILD_ID_SIZE];
902 		struct build_id bid;
903 		int size;
904 
905 		size = elf_read_build_id(elf, build_id, BUILD_ID_SIZE);
906 		if (size <= 0) {
907 			dso->load_errno = DSO_LOAD_ERRNO__CANNOT_READ_BUILDID;
908 			goto out_elf_end;
909 		}
910 
911 		build_id__init(&bid, build_id, size);
912 		if (!dso__build_id_equal(dso, &bid)) {
913 			pr_debug("%s: build id mismatch for %s.\n", __func__, name);
914 			dso->load_errno = DSO_LOAD_ERRNO__MISMATCHING_BUILDID;
915 			goto out_elf_end;
916 		}
917 	}
918 
919 	ss->is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
920 
921 	ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab",
922 			NULL);
923 	if (ss->symshdr.sh_type != SHT_SYMTAB)
924 		ss->symtab = NULL;
925 
926 	ss->dynsym_idx = 0;
927 	ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym",
928 			&ss->dynsym_idx);
929 	if (ss->dynshdr.sh_type != SHT_DYNSYM)
930 		ss->dynsym = NULL;
931 
932 	ss->opdidx = 0;
933 	ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd",
934 			&ss->opdidx);
935 	if (ss->opdshdr.sh_type != SHT_PROGBITS)
936 		ss->opdsec = NULL;
937 
938 	if (dso->kernel == DSO_SPACE__USER)
939 		ss->adjust_symbols = true;
940 	else
941 		ss->adjust_symbols = elf__needs_adjust_symbols(ehdr);
942 
943 	ss->name   = strdup(name);
944 	if (!ss->name) {
945 		dso->load_errno = errno;
946 		goto out_elf_end;
947 	}
948 
949 	ss->elf    = elf;
950 	ss->fd     = fd;
951 	ss->ehdr   = ehdr;
952 	ss->type   = type;
953 
954 	return 0;
955 
956 out_elf_end:
957 	elf_end(elf);
958 out_close:
959 	close(fd);
960 	return -1;
961 }
962 
963 /**
964  * ref_reloc_sym_not_found - has kernel relocation symbol been found.
965  * @kmap: kernel maps and relocation reference symbol
966  *
967  * This function returns %true if we are dealing with the kernel maps and the
968  * relocation reference symbol has not yet been found.  Otherwise %false is
969  * returned.
970  */
ref_reloc_sym_not_found(struct kmap * kmap)971 static bool ref_reloc_sym_not_found(struct kmap *kmap)
972 {
973 	return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
974 	       !kmap->ref_reloc_sym->unrelocated_addr;
975 }
976 
977 /**
978  * ref_reloc - kernel relocation offset.
979  * @kmap: kernel maps and relocation reference symbol
980  *
981  * This function returns the offset of kernel addresses as determined by using
982  * the relocation reference symbol i.e. if the kernel has not been relocated
983  * then the return value is zero.
984  */
ref_reloc(struct kmap * kmap)985 static u64 ref_reloc(struct kmap *kmap)
986 {
987 	if (kmap && kmap->ref_reloc_sym &&
988 	    kmap->ref_reloc_sym->unrelocated_addr)
989 		return kmap->ref_reloc_sym->addr -
990 		       kmap->ref_reloc_sym->unrelocated_addr;
991 	return 0;
992 }
993 
arch__sym_update(struct symbol * s __maybe_unused,GElf_Sym * sym __maybe_unused)994 void __weak arch__sym_update(struct symbol *s __maybe_unused,
995 		GElf_Sym *sym __maybe_unused) { }
996 
dso__process_kernel_symbol(struct dso * dso,struct map * map,GElf_Sym * sym,GElf_Shdr * shdr,struct maps * kmaps,struct kmap * kmap,struct dso ** curr_dsop,struct map ** curr_mapp,const char * section_name,bool adjust_kernel_syms,bool kmodule,bool * remap_kernel)997 static int dso__process_kernel_symbol(struct dso *dso, struct map *map,
998 				      GElf_Sym *sym, GElf_Shdr *shdr,
999 				      struct maps *kmaps, struct kmap *kmap,
1000 				      struct dso **curr_dsop, struct map **curr_mapp,
1001 				      const char *section_name,
1002 				      bool adjust_kernel_syms, bool kmodule, bool *remap_kernel)
1003 {
1004 	struct dso *curr_dso = *curr_dsop;
1005 	struct map *curr_map;
1006 	char dso_name[PATH_MAX];
1007 
1008 	/* Adjust symbol to map to file offset */
1009 	if (adjust_kernel_syms)
1010 		sym->st_value -= shdr->sh_addr - shdr->sh_offset;
1011 
1012 	if (strcmp(section_name, (curr_dso->short_name + dso->short_name_len)) == 0)
1013 		return 0;
1014 
1015 	if (strcmp(section_name, ".text") == 0) {
1016 		/*
1017 		 * The initial kernel mapping is based on
1018 		 * kallsyms and identity maps.  Overwrite it to
1019 		 * map to the kernel dso.
1020 		 */
1021 		if (*remap_kernel && dso->kernel && !kmodule) {
1022 			*remap_kernel = false;
1023 			map->start = shdr->sh_addr + ref_reloc(kmap);
1024 			map->end = map->start + shdr->sh_size;
1025 			map->pgoff = shdr->sh_offset;
1026 			map->map_ip = map__map_ip;
1027 			map->unmap_ip = map__unmap_ip;
1028 			/* Ensure maps are correctly ordered */
1029 			if (kmaps) {
1030 				map__get(map);
1031 				maps__remove(kmaps, map);
1032 				maps__insert(kmaps, map);
1033 				map__put(map);
1034 			}
1035 		}
1036 
1037 		/*
1038 		 * The initial module mapping is based on
1039 		 * /proc/modules mapped to offset zero.
1040 		 * Overwrite it to map to the module dso.
1041 		 */
1042 		if (*remap_kernel && kmodule) {
1043 			*remap_kernel = false;
1044 			map->pgoff = shdr->sh_offset;
1045 		}
1046 
1047 		*curr_mapp = map;
1048 		*curr_dsop = dso;
1049 		return 0;
1050 	}
1051 
1052 	if (!kmap)
1053 		return 0;
1054 
1055 	snprintf(dso_name, sizeof(dso_name), "%s%s", dso->short_name, section_name);
1056 
1057 	curr_map = maps__find_by_name(kmaps, dso_name);
1058 	if (curr_map == NULL) {
1059 		u64 start = sym->st_value;
1060 
1061 		if (kmodule)
1062 			start += map->start + shdr->sh_offset;
1063 
1064 		curr_dso = dso__new(dso_name);
1065 		if (curr_dso == NULL)
1066 			return -1;
1067 		curr_dso->kernel = dso->kernel;
1068 		curr_dso->long_name = dso->long_name;
1069 		curr_dso->long_name_len = dso->long_name_len;
1070 		curr_map = map__new2(start, curr_dso);
1071 		dso__put(curr_dso);
1072 		if (curr_map == NULL)
1073 			return -1;
1074 
1075 		if (curr_dso->kernel)
1076 			map__kmap(curr_map)->kmaps = kmaps;
1077 
1078 		if (adjust_kernel_syms) {
1079 			curr_map->start  = shdr->sh_addr + ref_reloc(kmap);
1080 			curr_map->end	 = curr_map->start + shdr->sh_size;
1081 			curr_map->pgoff	 = shdr->sh_offset;
1082 		} else {
1083 			curr_map->map_ip = curr_map->unmap_ip = identity__map_ip;
1084 		}
1085 		curr_dso->symtab_type = dso->symtab_type;
1086 		maps__insert(kmaps, curr_map);
1087 		/*
1088 		 * Add it before we drop the reference to curr_map, i.e. while
1089 		 * we still are sure to have a reference to this DSO via
1090 		 * *curr_map->dso.
1091 		 */
1092 		dsos__add(&kmaps->machine->dsos, curr_dso);
1093 		/* kmaps already got it */
1094 		map__put(curr_map);
1095 		dso__set_loaded(curr_dso);
1096 		*curr_mapp = curr_map;
1097 		*curr_dsop = curr_dso;
1098 	} else
1099 		*curr_dsop = curr_map->dso;
1100 
1101 	return 0;
1102 }
1103 
1104 static int
dso__load_sym_internal(struct dso * dso,struct map * map,struct symsrc * syms_ss,struct symsrc * runtime_ss,int kmodule,int dynsym)1105 dso__load_sym_internal(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1106 		       struct symsrc *runtime_ss, int kmodule, int dynsym)
1107 {
1108 	struct kmap *kmap = dso->kernel ? map__kmap(map) : NULL;
1109 	struct maps *kmaps = kmap ? map__kmaps(map) : NULL;
1110 	struct map *curr_map = map;
1111 	struct dso *curr_dso = dso;
1112 	Elf_Data *symstrs, *secstrs, *secstrs_run, *secstrs_sym;
1113 	uint32_t nr_syms;
1114 	int err = -1;
1115 	uint32_t idx;
1116 	GElf_Ehdr ehdr;
1117 	GElf_Shdr shdr;
1118 	GElf_Shdr tshdr;
1119 	Elf_Data *syms, *opddata = NULL;
1120 	GElf_Sym sym;
1121 	Elf_Scn *sec, *sec_strndx;
1122 	Elf *elf;
1123 	int nr = 0;
1124 	bool remap_kernel = false, adjust_kernel_syms = false;
1125 
1126 	if (kmap && !kmaps)
1127 		return -1;
1128 
1129 	elf = syms_ss->elf;
1130 	ehdr = syms_ss->ehdr;
1131 	if (dynsym) {
1132 		sec  = syms_ss->dynsym;
1133 		shdr = syms_ss->dynshdr;
1134 	} else {
1135 		sec =  syms_ss->symtab;
1136 		shdr = syms_ss->symshdr;
1137 	}
1138 
1139 	if (elf_section_by_name(runtime_ss->elf, &runtime_ss->ehdr, &tshdr,
1140 				".text", NULL))
1141 		dso->text_offset = tshdr.sh_addr - tshdr.sh_offset;
1142 
1143 	if (runtime_ss->opdsec)
1144 		opddata = elf_rawdata(runtime_ss->opdsec, NULL);
1145 
1146 	syms = elf_getdata(sec, NULL);
1147 	if (syms == NULL)
1148 		goto out_elf_end;
1149 
1150 	sec = elf_getscn(elf, shdr.sh_link);
1151 	if (sec == NULL)
1152 		goto out_elf_end;
1153 
1154 	symstrs = elf_getdata(sec, NULL);
1155 	if (symstrs == NULL)
1156 		goto out_elf_end;
1157 
1158 	sec_strndx = elf_getscn(runtime_ss->elf, runtime_ss->ehdr.e_shstrndx);
1159 	if (sec_strndx == NULL)
1160 		goto out_elf_end;
1161 
1162 	secstrs_run = elf_getdata(sec_strndx, NULL);
1163 	if (secstrs_run == NULL)
1164 		goto out_elf_end;
1165 
1166 	sec_strndx = elf_getscn(elf, ehdr.e_shstrndx);
1167 	if (sec_strndx == NULL)
1168 		goto out_elf_end;
1169 
1170 	secstrs_sym = elf_getdata(sec_strndx, NULL);
1171 	if (secstrs_sym == NULL)
1172 		goto out_elf_end;
1173 
1174 	nr_syms = shdr.sh_size / shdr.sh_entsize;
1175 
1176 	memset(&sym, 0, sizeof(sym));
1177 
1178 	/*
1179 	 * The kernel relocation symbol is needed in advance in order to adjust
1180 	 * kernel maps correctly.
1181 	 */
1182 	if (ref_reloc_sym_not_found(kmap)) {
1183 		elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1184 			const char *elf_name = elf_sym__name(&sym, symstrs);
1185 
1186 			if (strcmp(elf_name, kmap->ref_reloc_sym->name))
1187 				continue;
1188 			kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
1189 			map->reloc = kmap->ref_reloc_sym->addr -
1190 				     kmap->ref_reloc_sym->unrelocated_addr;
1191 			break;
1192 		}
1193 	}
1194 
1195 	/*
1196 	 * Handle any relocation of vdso necessary because older kernels
1197 	 * attempted to prelink vdso to its virtual address.
1198 	 */
1199 	if (dso__is_vdso(dso))
1200 		map->reloc = map->start - dso->text_offset;
1201 
1202 	dso->adjust_symbols = runtime_ss->adjust_symbols || ref_reloc(kmap);
1203 	/*
1204 	 * Initial kernel and module mappings do not map to the dso.
1205 	 * Flag the fixups.
1206 	 */
1207 	if (dso->kernel) {
1208 		remap_kernel = true;
1209 		adjust_kernel_syms = dso->adjust_symbols;
1210 	}
1211 	elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1212 		struct symbol *f;
1213 		const char *elf_name = elf_sym__name(&sym, symstrs);
1214 		char *demangled = NULL;
1215 		int is_label = elf_sym__is_label(&sym);
1216 		const char *section_name;
1217 		bool used_opd = false;
1218 
1219 		if (!is_label && !elf_sym__filter(&sym))
1220 			continue;
1221 
1222 		/* Reject ARM ELF "mapping symbols": these aren't unique and
1223 		 * don't identify functions, so will confuse the profile
1224 		 * output: */
1225 		if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) {
1226 			if (elf_name[0] == '$' && strchr("adtx", elf_name[1])
1227 			    && (elf_name[2] == '\0' || elf_name[2] == '.'))
1228 				continue;
1229 		}
1230 
1231 		if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) {
1232 			u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr;
1233 			u64 *opd = opddata->d_buf + offset;
1234 			sym.st_value = DSO__SWAP(dso, u64, *opd);
1235 			sym.st_shndx = elf_addr_to_index(runtime_ss->elf,
1236 					sym.st_value);
1237 			used_opd = true;
1238 		}
1239 
1240 		/*
1241 		 * When loading symbols in a data mapping, ABS symbols (which
1242 		 * has a value of SHN_ABS in its st_shndx) failed at
1243 		 * elf_getscn().  And it marks the loading as a failure so
1244 		 * already loaded symbols cannot be fixed up.
1245 		 *
1246 		 * I'm not sure what should be done. Just ignore them for now.
1247 		 * - Namhyung Kim
1248 		 */
1249 		if (sym.st_shndx == SHN_ABS)
1250 			continue;
1251 
1252 		sec = elf_getscn(syms_ss->elf, sym.st_shndx);
1253 		if (!sec)
1254 			goto out_elf_end;
1255 
1256 		gelf_getshdr(sec, &shdr);
1257 
1258 		/*
1259 		 * If the attribute bit SHF_ALLOC is not set, the section
1260 		 * doesn't occupy memory during process execution.
1261 		 * E.g. ".gnu.warning.*" section is used by linker to generate
1262 		 * warnings when calling deprecated functions, the symbols in
1263 		 * the section aren't loaded to memory during process execution,
1264 		 * so skip them.
1265 		 */
1266 		if (!(shdr.sh_flags & SHF_ALLOC))
1267 			continue;
1268 
1269 		secstrs = secstrs_sym;
1270 
1271 		/*
1272 		 * We have to fallback to runtime when syms' section header has
1273 		 * NOBITS set. NOBITS results in file offset (sh_offset) not
1274 		 * being incremented. So sh_offset used below has different
1275 		 * values for syms (invalid) and runtime (valid).
1276 		 */
1277 		if (shdr.sh_type == SHT_NOBITS) {
1278 			sec = elf_getscn(runtime_ss->elf, sym.st_shndx);
1279 			if (!sec)
1280 				goto out_elf_end;
1281 
1282 			gelf_getshdr(sec, &shdr);
1283 			secstrs = secstrs_run;
1284 		}
1285 
1286 		if (is_label && !elf_sec__filter(&shdr, secstrs))
1287 			continue;
1288 
1289 		section_name = elf_sec__name(&shdr, secstrs);
1290 
1291 		/* On ARM, symbols for thumb functions have 1 added to
1292 		 * the symbol address as a flag - remove it */
1293 		if ((ehdr.e_machine == EM_ARM) &&
1294 		    (GELF_ST_TYPE(sym.st_info) == STT_FUNC) &&
1295 		    (sym.st_value & 1))
1296 			--sym.st_value;
1297 
1298 		if (dso->kernel) {
1299 			if (dso__process_kernel_symbol(dso, map, &sym, &shdr, kmaps, kmap, &curr_dso, &curr_map,
1300 						       section_name, adjust_kernel_syms, kmodule, &remap_kernel))
1301 				goto out_elf_end;
1302 		} else if ((used_opd && runtime_ss->adjust_symbols) ||
1303 			   (!used_opd && syms_ss->adjust_symbols)) {
1304 			GElf_Phdr phdr;
1305 
1306 			if (elf_read_program_header(syms_ss->elf,
1307 						    (u64)sym.st_value, &phdr)) {
1308 				pr_debug4("%s: failed to find program header for "
1309 					   "symbol: %s st_value: %#" PRIx64 "\n",
1310 					   __func__, elf_name, (u64)sym.st_value);
1311 				pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1312 					"sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n",
1313 					__func__, (u64)sym.st_value, (u64)shdr.sh_addr,
1314 					(u64)shdr.sh_offset);
1315 				/*
1316 				 * Fail to find program header, let's rollback
1317 				 * to use shdr.sh_addr and shdr.sh_offset to
1318 				 * calibrate symbol's file address, though this
1319 				 * is not necessary for normal C ELF file, we
1320 				 * still need to handle java JIT symbols in this
1321 				 * case.
1322 				 */
1323 				sym.st_value -= shdr.sh_addr - shdr.sh_offset;
1324 			} else {
1325 				pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1326 					"p_vaddr: %#" PRIx64 " p_offset: %#" PRIx64 "\n",
1327 					__func__, (u64)sym.st_value, (u64)phdr.p_vaddr,
1328 					(u64)phdr.p_offset);
1329 				sym.st_value -= phdr.p_vaddr - phdr.p_offset;
1330 			}
1331 		}
1332 
1333 		demangled = demangle_sym(dso, kmodule, elf_name);
1334 		if (demangled != NULL)
1335 			elf_name = demangled;
1336 
1337 		f = symbol__new(sym.st_value, sym.st_size,
1338 				GELF_ST_BIND(sym.st_info),
1339 				GELF_ST_TYPE(sym.st_info), elf_name);
1340 		free(demangled);
1341 		if (!f)
1342 			goto out_elf_end;
1343 
1344 		arch__sym_update(f, &sym);
1345 
1346 		__symbols__insert(&curr_dso->symbols, f, dso->kernel);
1347 		nr++;
1348 	}
1349 
1350 	/*
1351 	 * For misannotated, zeroed, ASM function sizes.
1352 	 */
1353 	if (nr > 0) {
1354 		symbols__fixup_end(&dso->symbols, false);
1355 		symbols__fixup_duplicate(&dso->symbols);
1356 		if (kmap) {
1357 			/*
1358 			 * We need to fixup this here too because we create new
1359 			 * maps here, for things like vsyscall sections.
1360 			 */
1361 			maps__fixup_end(kmaps);
1362 		}
1363 	}
1364 	err = nr;
1365 out_elf_end:
1366 	return err;
1367 }
1368 
dso__load_sym(struct dso * dso,struct map * map,struct symsrc * syms_ss,struct symsrc * runtime_ss,int kmodule)1369 int dso__load_sym(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1370 		  struct symsrc *runtime_ss, int kmodule)
1371 {
1372 	int nr = 0;
1373 	int err = -1;
1374 
1375 	dso->symtab_type = syms_ss->type;
1376 	dso->is_64_bit = syms_ss->is_64_bit;
1377 	dso->rel = syms_ss->ehdr.e_type == ET_REL;
1378 
1379 	/*
1380 	 * Modules may already have symbols from kallsyms, but those symbols
1381 	 * have the wrong values for the dso maps, so remove them.
1382 	 */
1383 	if (kmodule && syms_ss->symtab)
1384 		symbols__delete(&dso->symbols);
1385 
1386 	if (!syms_ss->symtab) {
1387 		/*
1388 		 * If the vmlinux is stripped, fail so we will fall back
1389 		 * to using kallsyms. The vmlinux runtime symbols aren't
1390 		 * of much use.
1391 		 */
1392 		if (dso->kernel)
1393 			return err;
1394 	} else  {
1395 		err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss,
1396 					     kmodule, 0);
1397 		if (err < 0)
1398 			return err;
1399 		nr = err;
1400 	}
1401 
1402 	if (syms_ss->dynsym) {
1403 		err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss,
1404 					     kmodule, 1);
1405 		if (err < 0)
1406 			return err;
1407 		err += nr;
1408 	}
1409 
1410 	return err;
1411 }
1412 
elf_read_maps(Elf * elf,bool exe,mapfn_t mapfn,void * data)1413 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
1414 {
1415 	GElf_Phdr phdr;
1416 	size_t i, phdrnum;
1417 	int err;
1418 	u64 sz;
1419 
1420 	if (elf_getphdrnum(elf, &phdrnum))
1421 		return -1;
1422 
1423 	for (i = 0; i < phdrnum; i++) {
1424 		if (gelf_getphdr(elf, i, &phdr) == NULL)
1425 			return -1;
1426 		if (phdr.p_type != PT_LOAD)
1427 			continue;
1428 		if (exe) {
1429 			if (!(phdr.p_flags & PF_X))
1430 				continue;
1431 		} else {
1432 			if (!(phdr.p_flags & PF_R))
1433 				continue;
1434 		}
1435 		sz = min(phdr.p_memsz, phdr.p_filesz);
1436 		if (!sz)
1437 			continue;
1438 		err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
1439 		if (err)
1440 			return err;
1441 	}
1442 	return 0;
1443 }
1444 
file__read_maps(int fd,bool exe,mapfn_t mapfn,void * data,bool * is_64_bit)1445 int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data,
1446 		    bool *is_64_bit)
1447 {
1448 	int err;
1449 	Elf *elf;
1450 
1451 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1452 	if (elf == NULL)
1453 		return -1;
1454 
1455 	if (is_64_bit)
1456 		*is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1457 
1458 	err = elf_read_maps(elf, exe, mapfn, data);
1459 
1460 	elf_end(elf);
1461 	return err;
1462 }
1463 
dso__type_fd(int fd)1464 enum dso_type dso__type_fd(int fd)
1465 {
1466 	enum dso_type dso_type = DSO__TYPE_UNKNOWN;
1467 	GElf_Ehdr ehdr;
1468 	Elf_Kind ek;
1469 	Elf *elf;
1470 
1471 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1472 	if (elf == NULL)
1473 		goto out;
1474 
1475 	ek = elf_kind(elf);
1476 	if (ek != ELF_K_ELF)
1477 		goto out_end;
1478 
1479 	if (gelf_getclass(elf) == ELFCLASS64) {
1480 		dso_type = DSO__TYPE_64BIT;
1481 		goto out_end;
1482 	}
1483 
1484 	if (gelf_getehdr(elf, &ehdr) == NULL)
1485 		goto out_end;
1486 
1487 	if (ehdr.e_machine == EM_X86_64)
1488 		dso_type = DSO__TYPE_X32BIT;
1489 	else
1490 		dso_type = DSO__TYPE_32BIT;
1491 out_end:
1492 	elf_end(elf);
1493 out:
1494 	return dso_type;
1495 }
1496 
copy_bytes(int from,off_t from_offs,int to,off_t to_offs,u64 len)1497 static int copy_bytes(int from, off_t from_offs, int to, off_t to_offs, u64 len)
1498 {
1499 	ssize_t r;
1500 	size_t n;
1501 	int err = -1;
1502 	char *buf = malloc(page_size);
1503 
1504 	if (buf == NULL)
1505 		return -1;
1506 
1507 	if (lseek(to, to_offs, SEEK_SET) != to_offs)
1508 		goto out;
1509 
1510 	if (lseek(from, from_offs, SEEK_SET) != from_offs)
1511 		goto out;
1512 
1513 	while (len) {
1514 		n = page_size;
1515 		if (len < n)
1516 			n = len;
1517 		/* Use read because mmap won't work on proc files */
1518 		r = read(from, buf, n);
1519 		if (r < 0)
1520 			goto out;
1521 		if (!r)
1522 			break;
1523 		n = r;
1524 		r = write(to, buf, n);
1525 		if (r < 0)
1526 			goto out;
1527 		if ((size_t)r != n)
1528 			goto out;
1529 		len -= n;
1530 	}
1531 
1532 	err = 0;
1533 out:
1534 	free(buf);
1535 	return err;
1536 }
1537 
1538 struct kcore {
1539 	int fd;
1540 	int elfclass;
1541 	Elf *elf;
1542 	GElf_Ehdr ehdr;
1543 };
1544 
kcore__open(struct kcore * kcore,const char * filename)1545 static int kcore__open(struct kcore *kcore, const char *filename)
1546 {
1547 	GElf_Ehdr *ehdr;
1548 
1549 	kcore->fd = open(filename, O_RDONLY);
1550 	if (kcore->fd == -1)
1551 		return -1;
1552 
1553 	kcore->elf = elf_begin(kcore->fd, ELF_C_READ, NULL);
1554 	if (!kcore->elf)
1555 		goto out_close;
1556 
1557 	kcore->elfclass = gelf_getclass(kcore->elf);
1558 	if (kcore->elfclass == ELFCLASSNONE)
1559 		goto out_end;
1560 
1561 	ehdr = gelf_getehdr(kcore->elf, &kcore->ehdr);
1562 	if (!ehdr)
1563 		goto out_end;
1564 
1565 	return 0;
1566 
1567 out_end:
1568 	elf_end(kcore->elf);
1569 out_close:
1570 	close(kcore->fd);
1571 	return -1;
1572 }
1573 
kcore__init(struct kcore * kcore,char * filename,int elfclass,bool temp)1574 static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
1575 		       bool temp)
1576 {
1577 	kcore->elfclass = elfclass;
1578 
1579 	if (temp)
1580 		kcore->fd = mkstemp(filename);
1581 	else
1582 		kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
1583 	if (kcore->fd == -1)
1584 		return -1;
1585 
1586 	kcore->elf = elf_begin(kcore->fd, ELF_C_WRITE, NULL);
1587 	if (!kcore->elf)
1588 		goto out_close;
1589 
1590 	if (!gelf_newehdr(kcore->elf, elfclass))
1591 		goto out_end;
1592 
1593 	memset(&kcore->ehdr, 0, sizeof(GElf_Ehdr));
1594 
1595 	return 0;
1596 
1597 out_end:
1598 	elf_end(kcore->elf);
1599 out_close:
1600 	close(kcore->fd);
1601 	unlink(filename);
1602 	return -1;
1603 }
1604 
kcore__close(struct kcore * kcore)1605 static void kcore__close(struct kcore *kcore)
1606 {
1607 	elf_end(kcore->elf);
1608 	close(kcore->fd);
1609 }
1610 
kcore__copy_hdr(struct kcore * from,struct kcore * to,size_t count)1611 static int kcore__copy_hdr(struct kcore *from, struct kcore *to, size_t count)
1612 {
1613 	GElf_Ehdr *ehdr = &to->ehdr;
1614 	GElf_Ehdr *kehdr = &from->ehdr;
1615 
1616 	memcpy(ehdr->e_ident, kehdr->e_ident, EI_NIDENT);
1617 	ehdr->e_type      = kehdr->e_type;
1618 	ehdr->e_machine   = kehdr->e_machine;
1619 	ehdr->e_version   = kehdr->e_version;
1620 	ehdr->e_entry     = 0;
1621 	ehdr->e_shoff     = 0;
1622 	ehdr->e_flags     = kehdr->e_flags;
1623 	ehdr->e_phnum     = count;
1624 	ehdr->e_shentsize = 0;
1625 	ehdr->e_shnum     = 0;
1626 	ehdr->e_shstrndx  = 0;
1627 
1628 	if (from->elfclass == ELFCLASS32) {
1629 		ehdr->e_phoff     = sizeof(Elf32_Ehdr);
1630 		ehdr->e_ehsize    = sizeof(Elf32_Ehdr);
1631 		ehdr->e_phentsize = sizeof(Elf32_Phdr);
1632 	} else {
1633 		ehdr->e_phoff     = sizeof(Elf64_Ehdr);
1634 		ehdr->e_ehsize    = sizeof(Elf64_Ehdr);
1635 		ehdr->e_phentsize = sizeof(Elf64_Phdr);
1636 	}
1637 
1638 	if (!gelf_update_ehdr(to->elf, ehdr))
1639 		return -1;
1640 
1641 	if (!gelf_newphdr(to->elf, count))
1642 		return -1;
1643 
1644 	return 0;
1645 }
1646 
kcore__add_phdr(struct kcore * kcore,int idx,off_t offset,u64 addr,u64 len)1647 static int kcore__add_phdr(struct kcore *kcore, int idx, off_t offset,
1648 			   u64 addr, u64 len)
1649 {
1650 	GElf_Phdr phdr = {
1651 		.p_type		= PT_LOAD,
1652 		.p_flags	= PF_R | PF_W | PF_X,
1653 		.p_offset	= offset,
1654 		.p_vaddr	= addr,
1655 		.p_paddr	= 0,
1656 		.p_filesz	= len,
1657 		.p_memsz	= len,
1658 		.p_align	= page_size,
1659 	};
1660 
1661 	if (!gelf_update_phdr(kcore->elf, idx, &phdr))
1662 		return -1;
1663 
1664 	return 0;
1665 }
1666 
kcore__write(struct kcore * kcore)1667 static off_t kcore__write(struct kcore *kcore)
1668 {
1669 	return elf_update(kcore->elf, ELF_C_WRITE);
1670 }
1671 
1672 struct phdr_data {
1673 	off_t offset;
1674 	off_t rel;
1675 	u64 addr;
1676 	u64 len;
1677 	struct list_head node;
1678 	struct phdr_data *remaps;
1679 };
1680 
1681 struct sym_data {
1682 	u64 addr;
1683 	struct list_head node;
1684 };
1685 
1686 struct kcore_copy_info {
1687 	u64 stext;
1688 	u64 etext;
1689 	u64 first_symbol;
1690 	u64 last_symbol;
1691 	u64 first_module;
1692 	u64 first_module_symbol;
1693 	u64 last_module_symbol;
1694 	size_t phnum;
1695 	struct list_head phdrs;
1696 	struct list_head syms;
1697 };
1698 
1699 #define kcore_copy__for_each_phdr(k, p) \
1700 	list_for_each_entry((p), &(k)->phdrs, node)
1701 
phdr_data__new(u64 addr,u64 len,off_t offset)1702 static struct phdr_data *phdr_data__new(u64 addr, u64 len, off_t offset)
1703 {
1704 	struct phdr_data *p = zalloc(sizeof(*p));
1705 
1706 	if (p) {
1707 		p->addr   = addr;
1708 		p->len    = len;
1709 		p->offset = offset;
1710 	}
1711 
1712 	return p;
1713 }
1714 
kcore_copy_info__addnew(struct kcore_copy_info * kci,u64 addr,u64 len,off_t offset)1715 static struct phdr_data *kcore_copy_info__addnew(struct kcore_copy_info *kci,
1716 						 u64 addr, u64 len,
1717 						 off_t offset)
1718 {
1719 	struct phdr_data *p = phdr_data__new(addr, len, offset);
1720 
1721 	if (p)
1722 		list_add_tail(&p->node, &kci->phdrs);
1723 
1724 	return p;
1725 }
1726 
kcore_copy__free_phdrs(struct kcore_copy_info * kci)1727 static void kcore_copy__free_phdrs(struct kcore_copy_info *kci)
1728 {
1729 	struct phdr_data *p, *tmp;
1730 
1731 	list_for_each_entry_safe(p, tmp, &kci->phdrs, node) {
1732 		list_del_init(&p->node);
1733 		free(p);
1734 	}
1735 }
1736 
kcore_copy__new_sym(struct kcore_copy_info * kci,u64 addr)1737 static struct sym_data *kcore_copy__new_sym(struct kcore_copy_info *kci,
1738 					    u64 addr)
1739 {
1740 	struct sym_data *s = zalloc(sizeof(*s));
1741 
1742 	if (s) {
1743 		s->addr = addr;
1744 		list_add_tail(&s->node, &kci->syms);
1745 	}
1746 
1747 	return s;
1748 }
1749 
kcore_copy__free_syms(struct kcore_copy_info * kci)1750 static void kcore_copy__free_syms(struct kcore_copy_info *kci)
1751 {
1752 	struct sym_data *s, *tmp;
1753 
1754 	list_for_each_entry_safe(s, tmp, &kci->syms, node) {
1755 		list_del_init(&s->node);
1756 		free(s);
1757 	}
1758 }
1759 
kcore_copy__process_kallsyms(void * arg,const char * name,char type,u64 start)1760 static int kcore_copy__process_kallsyms(void *arg, const char *name, char type,
1761 					u64 start)
1762 {
1763 	struct kcore_copy_info *kci = arg;
1764 
1765 	if (!kallsyms__is_function(type))
1766 		return 0;
1767 
1768 	if (strchr(name, '[')) {
1769 		if (!kci->first_module_symbol || start < kci->first_module_symbol)
1770 			kci->first_module_symbol = start;
1771 		if (start > kci->last_module_symbol)
1772 			kci->last_module_symbol = start;
1773 		return 0;
1774 	}
1775 
1776 	if (!kci->first_symbol || start < kci->first_symbol)
1777 		kci->first_symbol = start;
1778 
1779 	if (!kci->last_symbol || start > kci->last_symbol)
1780 		kci->last_symbol = start;
1781 
1782 	if (!strcmp(name, "_stext")) {
1783 		kci->stext = start;
1784 		return 0;
1785 	}
1786 
1787 	if (!strcmp(name, "_etext")) {
1788 		kci->etext = start;
1789 		return 0;
1790 	}
1791 
1792 	if (is_entry_trampoline(name) && !kcore_copy__new_sym(kci, start))
1793 		return -1;
1794 
1795 	return 0;
1796 }
1797 
kcore_copy__parse_kallsyms(struct kcore_copy_info * kci,const char * dir)1798 static int kcore_copy__parse_kallsyms(struct kcore_copy_info *kci,
1799 				      const char *dir)
1800 {
1801 	char kallsyms_filename[PATH_MAX];
1802 
1803 	scnprintf(kallsyms_filename, PATH_MAX, "%s/kallsyms", dir);
1804 
1805 	if (symbol__restricted_filename(kallsyms_filename, "/proc/kallsyms"))
1806 		return -1;
1807 
1808 	if (kallsyms__parse(kallsyms_filename, kci,
1809 			    kcore_copy__process_kallsyms) < 0)
1810 		return -1;
1811 
1812 	return 0;
1813 }
1814 
kcore_copy__process_modules(void * arg,const char * name __maybe_unused,u64 start,u64 size __maybe_unused)1815 static int kcore_copy__process_modules(void *arg,
1816 				       const char *name __maybe_unused,
1817 				       u64 start, u64 size __maybe_unused)
1818 {
1819 	struct kcore_copy_info *kci = arg;
1820 
1821 	if (!kci->first_module || start < kci->first_module)
1822 		kci->first_module = start;
1823 
1824 	return 0;
1825 }
1826 
kcore_copy__parse_modules(struct kcore_copy_info * kci,const char * dir)1827 static int kcore_copy__parse_modules(struct kcore_copy_info *kci,
1828 				     const char *dir)
1829 {
1830 	char modules_filename[PATH_MAX];
1831 
1832 	scnprintf(modules_filename, PATH_MAX, "%s/modules", dir);
1833 
1834 	if (symbol__restricted_filename(modules_filename, "/proc/modules"))
1835 		return -1;
1836 
1837 	if (modules__parse(modules_filename, kci,
1838 			   kcore_copy__process_modules) < 0)
1839 		return -1;
1840 
1841 	return 0;
1842 }
1843 
kcore_copy__map(struct kcore_copy_info * kci,u64 start,u64 end,u64 pgoff,u64 s,u64 e)1844 static int kcore_copy__map(struct kcore_copy_info *kci, u64 start, u64 end,
1845 			   u64 pgoff, u64 s, u64 e)
1846 {
1847 	u64 len, offset;
1848 
1849 	if (s < start || s >= end)
1850 		return 0;
1851 
1852 	offset = (s - start) + pgoff;
1853 	len = e < end ? e - s : end - s;
1854 
1855 	return kcore_copy_info__addnew(kci, s, len, offset) ? 0 : -1;
1856 }
1857 
kcore_copy__read_map(u64 start,u64 len,u64 pgoff,void * data)1858 static int kcore_copy__read_map(u64 start, u64 len, u64 pgoff, void *data)
1859 {
1860 	struct kcore_copy_info *kci = data;
1861 	u64 end = start + len;
1862 	struct sym_data *sdat;
1863 
1864 	if (kcore_copy__map(kci, start, end, pgoff, kci->stext, kci->etext))
1865 		return -1;
1866 
1867 	if (kcore_copy__map(kci, start, end, pgoff, kci->first_module,
1868 			    kci->last_module_symbol))
1869 		return -1;
1870 
1871 	list_for_each_entry(sdat, &kci->syms, node) {
1872 		u64 s = round_down(sdat->addr, page_size);
1873 
1874 		if (kcore_copy__map(kci, start, end, pgoff, s, s + len))
1875 			return -1;
1876 	}
1877 
1878 	return 0;
1879 }
1880 
kcore_copy__read_maps(struct kcore_copy_info * kci,Elf * elf)1881 static int kcore_copy__read_maps(struct kcore_copy_info *kci, Elf *elf)
1882 {
1883 	if (elf_read_maps(elf, true, kcore_copy__read_map, kci) < 0)
1884 		return -1;
1885 
1886 	return 0;
1887 }
1888 
kcore_copy__find_remaps(struct kcore_copy_info * kci)1889 static void kcore_copy__find_remaps(struct kcore_copy_info *kci)
1890 {
1891 	struct phdr_data *p, *k = NULL;
1892 	u64 kend;
1893 
1894 	if (!kci->stext)
1895 		return;
1896 
1897 	/* Find phdr that corresponds to the kernel map (contains stext) */
1898 	kcore_copy__for_each_phdr(kci, p) {
1899 		u64 pend = p->addr + p->len - 1;
1900 
1901 		if (p->addr <= kci->stext && pend >= kci->stext) {
1902 			k = p;
1903 			break;
1904 		}
1905 	}
1906 
1907 	if (!k)
1908 		return;
1909 
1910 	kend = k->offset + k->len;
1911 
1912 	/* Find phdrs that remap the kernel */
1913 	kcore_copy__for_each_phdr(kci, p) {
1914 		u64 pend = p->offset + p->len;
1915 
1916 		if (p == k)
1917 			continue;
1918 
1919 		if (p->offset >= k->offset && pend <= kend)
1920 			p->remaps = k;
1921 	}
1922 }
1923 
kcore_copy__layout(struct kcore_copy_info * kci)1924 static void kcore_copy__layout(struct kcore_copy_info *kci)
1925 {
1926 	struct phdr_data *p;
1927 	off_t rel = 0;
1928 
1929 	kcore_copy__find_remaps(kci);
1930 
1931 	kcore_copy__for_each_phdr(kci, p) {
1932 		if (!p->remaps) {
1933 			p->rel = rel;
1934 			rel += p->len;
1935 		}
1936 		kci->phnum += 1;
1937 	}
1938 
1939 	kcore_copy__for_each_phdr(kci, p) {
1940 		struct phdr_data *k = p->remaps;
1941 
1942 		if (k)
1943 			p->rel = p->offset - k->offset + k->rel;
1944 	}
1945 }
1946 
kcore_copy__calc_maps(struct kcore_copy_info * kci,const char * dir,Elf * elf)1947 static int kcore_copy__calc_maps(struct kcore_copy_info *kci, const char *dir,
1948 				 Elf *elf)
1949 {
1950 	if (kcore_copy__parse_kallsyms(kci, dir))
1951 		return -1;
1952 
1953 	if (kcore_copy__parse_modules(kci, dir))
1954 		return -1;
1955 
1956 	if (kci->stext)
1957 		kci->stext = round_down(kci->stext, page_size);
1958 	else
1959 		kci->stext = round_down(kci->first_symbol, page_size);
1960 
1961 	if (kci->etext) {
1962 		kci->etext = round_up(kci->etext, page_size);
1963 	} else if (kci->last_symbol) {
1964 		kci->etext = round_up(kci->last_symbol, page_size);
1965 		kci->etext += page_size;
1966 	}
1967 
1968 	if (kci->first_module_symbol &&
1969 	    (!kci->first_module || kci->first_module_symbol < kci->first_module))
1970 		kci->first_module = kci->first_module_symbol;
1971 
1972 	kci->first_module = round_down(kci->first_module, page_size);
1973 
1974 	if (kci->last_module_symbol) {
1975 		kci->last_module_symbol = round_up(kci->last_module_symbol,
1976 						   page_size);
1977 		kci->last_module_symbol += page_size;
1978 	}
1979 
1980 	if (!kci->stext || !kci->etext)
1981 		return -1;
1982 
1983 	if (kci->first_module && !kci->last_module_symbol)
1984 		return -1;
1985 
1986 	if (kcore_copy__read_maps(kci, elf))
1987 		return -1;
1988 
1989 	kcore_copy__layout(kci);
1990 
1991 	return 0;
1992 }
1993 
kcore_copy__copy_file(const char * from_dir,const char * to_dir,const char * name)1994 static int kcore_copy__copy_file(const char *from_dir, const char *to_dir,
1995 				 const char *name)
1996 {
1997 	char from_filename[PATH_MAX];
1998 	char to_filename[PATH_MAX];
1999 
2000 	scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
2001 	scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
2002 
2003 	return copyfile_mode(from_filename, to_filename, 0400);
2004 }
2005 
kcore_copy__unlink(const char * dir,const char * name)2006 static int kcore_copy__unlink(const char *dir, const char *name)
2007 {
2008 	char filename[PATH_MAX];
2009 
2010 	scnprintf(filename, PATH_MAX, "%s/%s", dir, name);
2011 
2012 	return unlink(filename);
2013 }
2014 
kcore_copy__compare_fds(int from,int to)2015 static int kcore_copy__compare_fds(int from, int to)
2016 {
2017 	char *buf_from;
2018 	char *buf_to;
2019 	ssize_t ret;
2020 	size_t len;
2021 	int err = -1;
2022 
2023 	buf_from = malloc(page_size);
2024 	buf_to = malloc(page_size);
2025 	if (!buf_from || !buf_to)
2026 		goto out;
2027 
2028 	while (1) {
2029 		/* Use read because mmap won't work on proc files */
2030 		ret = read(from, buf_from, page_size);
2031 		if (ret < 0)
2032 			goto out;
2033 
2034 		if (!ret)
2035 			break;
2036 
2037 		len = ret;
2038 
2039 		if (readn(to, buf_to, len) != (int)len)
2040 			goto out;
2041 
2042 		if (memcmp(buf_from, buf_to, len))
2043 			goto out;
2044 	}
2045 
2046 	err = 0;
2047 out:
2048 	free(buf_to);
2049 	free(buf_from);
2050 	return err;
2051 }
2052 
kcore_copy__compare_files(const char * from_filename,const char * to_filename)2053 static int kcore_copy__compare_files(const char *from_filename,
2054 				     const char *to_filename)
2055 {
2056 	int from, to, err = -1;
2057 
2058 	from = open(from_filename, O_RDONLY);
2059 	if (from < 0)
2060 		return -1;
2061 
2062 	to = open(to_filename, O_RDONLY);
2063 	if (to < 0)
2064 		goto out_close_from;
2065 
2066 	err = kcore_copy__compare_fds(from, to);
2067 
2068 	close(to);
2069 out_close_from:
2070 	close(from);
2071 	return err;
2072 }
2073 
kcore_copy__compare_file(const char * from_dir,const char * to_dir,const char * name)2074 static int kcore_copy__compare_file(const char *from_dir, const char *to_dir,
2075 				    const char *name)
2076 {
2077 	char from_filename[PATH_MAX];
2078 	char to_filename[PATH_MAX];
2079 
2080 	scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
2081 	scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
2082 
2083 	return kcore_copy__compare_files(from_filename, to_filename);
2084 }
2085 
2086 /**
2087  * kcore_copy - copy kallsyms, modules and kcore from one directory to another.
2088  * @from_dir: from directory
2089  * @to_dir: to directory
2090  *
2091  * This function copies kallsyms, modules and kcore files from one directory to
2092  * another.  kallsyms and modules are copied entirely.  Only code segments are
2093  * copied from kcore.  It is assumed that two segments suffice: one for the
2094  * kernel proper and one for all the modules.  The code segments are determined
2095  * from kallsyms and modules files.  The kernel map starts at _stext or the
2096  * lowest function symbol, and ends at _etext or the highest function symbol.
2097  * The module map starts at the lowest module address and ends at the highest
2098  * module symbol.  Start addresses are rounded down to the nearest page.  End
2099  * addresses are rounded up to the nearest page.  An extra page is added to the
2100  * highest kernel symbol and highest module symbol to, hopefully, encompass that
2101  * symbol too.  Because it contains only code sections, the resulting kcore is
2102  * unusual.  One significant peculiarity is that the mapping (start -> pgoff)
2103  * is not the same for the kernel map and the modules map.  That happens because
2104  * the data is copied adjacently whereas the original kcore has gaps.  Finally,
2105  * kallsyms and modules files are compared with their copies to check that
2106  * modules have not been loaded or unloaded while the copies were taking place.
2107  *
2108  * Return: %0 on success, %-1 on failure.
2109  */
kcore_copy(const char * from_dir,const char * to_dir)2110 int kcore_copy(const char *from_dir, const char *to_dir)
2111 {
2112 	struct kcore kcore;
2113 	struct kcore extract;
2114 	int idx = 0, err = -1;
2115 	off_t offset, sz;
2116 	struct kcore_copy_info kci = { .stext = 0, };
2117 	char kcore_filename[PATH_MAX];
2118 	char extract_filename[PATH_MAX];
2119 	struct phdr_data *p;
2120 
2121 	INIT_LIST_HEAD(&kci.phdrs);
2122 	INIT_LIST_HEAD(&kci.syms);
2123 
2124 	if (kcore_copy__copy_file(from_dir, to_dir, "kallsyms"))
2125 		return -1;
2126 
2127 	if (kcore_copy__copy_file(from_dir, to_dir, "modules"))
2128 		goto out_unlink_kallsyms;
2129 
2130 	scnprintf(kcore_filename, PATH_MAX, "%s/kcore", from_dir);
2131 	scnprintf(extract_filename, PATH_MAX, "%s/kcore", to_dir);
2132 
2133 	if (kcore__open(&kcore, kcore_filename))
2134 		goto out_unlink_modules;
2135 
2136 	if (kcore_copy__calc_maps(&kci, from_dir, kcore.elf))
2137 		goto out_kcore_close;
2138 
2139 	if (kcore__init(&extract, extract_filename, kcore.elfclass, false))
2140 		goto out_kcore_close;
2141 
2142 	if (kcore__copy_hdr(&kcore, &extract, kci.phnum))
2143 		goto out_extract_close;
2144 
2145 	offset = gelf_fsize(extract.elf, ELF_T_EHDR, 1, EV_CURRENT) +
2146 		 gelf_fsize(extract.elf, ELF_T_PHDR, kci.phnum, EV_CURRENT);
2147 	offset = round_up(offset, page_size);
2148 
2149 	kcore_copy__for_each_phdr(&kci, p) {
2150 		off_t offs = p->rel + offset;
2151 
2152 		if (kcore__add_phdr(&extract, idx++, offs, p->addr, p->len))
2153 			goto out_extract_close;
2154 	}
2155 
2156 	sz = kcore__write(&extract);
2157 	if (sz < 0 || sz > offset)
2158 		goto out_extract_close;
2159 
2160 	kcore_copy__for_each_phdr(&kci, p) {
2161 		off_t offs = p->rel + offset;
2162 
2163 		if (p->remaps)
2164 			continue;
2165 		if (copy_bytes(kcore.fd, p->offset, extract.fd, offs, p->len))
2166 			goto out_extract_close;
2167 	}
2168 
2169 	if (kcore_copy__compare_file(from_dir, to_dir, "modules"))
2170 		goto out_extract_close;
2171 
2172 	if (kcore_copy__compare_file(from_dir, to_dir, "kallsyms"))
2173 		goto out_extract_close;
2174 
2175 	err = 0;
2176 
2177 out_extract_close:
2178 	kcore__close(&extract);
2179 	if (err)
2180 		unlink(extract_filename);
2181 out_kcore_close:
2182 	kcore__close(&kcore);
2183 out_unlink_modules:
2184 	if (err)
2185 		kcore_copy__unlink(to_dir, "modules");
2186 out_unlink_kallsyms:
2187 	if (err)
2188 		kcore_copy__unlink(to_dir, "kallsyms");
2189 
2190 	kcore_copy__free_phdrs(&kci);
2191 	kcore_copy__free_syms(&kci);
2192 
2193 	return err;
2194 }
2195 
kcore_extract__create(struct kcore_extract * kce)2196 int kcore_extract__create(struct kcore_extract *kce)
2197 {
2198 	struct kcore kcore;
2199 	struct kcore extract;
2200 	size_t count = 1;
2201 	int idx = 0, err = -1;
2202 	off_t offset = page_size, sz;
2203 
2204 	if (kcore__open(&kcore, kce->kcore_filename))
2205 		return -1;
2206 
2207 	strcpy(kce->extract_filename, PERF_KCORE_EXTRACT);
2208 	if (kcore__init(&extract, kce->extract_filename, kcore.elfclass, true))
2209 		goto out_kcore_close;
2210 
2211 	if (kcore__copy_hdr(&kcore, &extract, count))
2212 		goto out_extract_close;
2213 
2214 	if (kcore__add_phdr(&extract, idx, offset, kce->addr, kce->len))
2215 		goto out_extract_close;
2216 
2217 	sz = kcore__write(&extract);
2218 	if (sz < 0 || sz > offset)
2219 		goto out_extract_close;
2220 
2221 	if (copy_bytes(kcore.fd, kce->offs, extract.fd, offset, kce->len))
2222 		goto out_extract_close;
2223 
2224 	err = 0;
2225 
2226 out_extract_close:
2227 	kcore__close(&extract);
2228 	if (err)
2229 		unlink(kce->extract_filename);
2230 out_kcore_close:
2231 	kcore__close(&kcore);
2232 
2233 	return err;
2234 }
2235 
kcore_extract__delete(struct kcore_extract * kce)2236 void kcore_extract__delete(struct kcore_extract *kce)
2237 {
2238 	unlink(kce->extract_filename);
2239 }
2240 
2241 #ifdef HAVE_GELF_GETNOTE_SUPPORT
2242 
sdt_adjust_loc(struct sdt_note * tmp,GElf_Addr base_off)2243 static void sdt_adjust_loc(struct sdt_note *tmp, GElf_Addr base_off)
2244 {
2245 	if (!base_off)
2246 		return;
2247 
2248 	if (tmp->bit32)
2249 		tmp->addr.a32[SDT_NOTE_IDX_LOC] =
2250 			tmp->addr.a32[SDT_NOTE_IDX_LOC] + base_off -
2251 			tmp->addr.a32[SDT_NOTE_IDX_BASE];
2252 	else
2253 		tmp->addr.a64[SDT_NOTE_IDX_LOC] =
2254 			tmp->addr.a64[SDT_NOTE_IDX_LOC] + base_off -
2255 			tmp->addr.a64[SDT_NOTE_IDX_BASE];
2256 }
2257 
sdt_adjust_refctr(struct sdt_note * tmp,GElf_Addr base_addr,GElf_Addr base_off)2258 static void sdt_adjust_refctr(struct sdt_note *tmp, GElf_Addr base_addr,
2259 			      GElf_Addr base_off)
2260 {
2261 	if (!base_off)
2262 		return;
2263 
2264 	if (tmp->bit32 && tmp->addr.a32[SDT_NOTE_IDX_REFCTR])
2265 		tmp->addr.a32[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2266 	else if (tmp->addr.a64[SDT_NOTE_IDX_REFCTR])
2267 		tmp->addr.a64[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2268 }
2269 
2270 /**
2271  * populate_sdt_note : Parse raw data and identify SDT note
2272  * @elf: elf of the opened file
2273  * @data: raw data of a section with description offset applied
2274  * @len: note description size
2275  * @type: type of the note
2276  * @sdt_notes: List to add the SDT note
2277  *
2278  * Responsible for parsing the @data in section .note.stapsdt in @elf and
2279  * if its an SDT note, it appends to @sdt_notes list.
2280  */
populate_sdt_note(Elf ** elf,const char * data,size_t len,struct list_head * sdt_notes)2281 static int populate_sdt_note(Elf **elf, const char *data, size_t len,
2282 			     struct list_head *sdt_notes)
2283 {
2284 	const char *provider, *name, *args;
2285 	struct sdt_note *tmp = NULL;
2286 	GElf_Ehdr ehdr;
2287 	GElf_Shdr shdr;
2288 	int ret = -EINVAL;
2289 
2290 	union {
2291 		Elf64_Addr a64[NR_ADDR];
2292 		Elf32_Addr a32[NR_ADDR];
2293 	} buf;
2294 
2295 	Elf_Data dst = {
2296 		.d_buf = &buf, .d_type = ELF_T_ADDR, .d_version = EV_CURRENT,
2297 		.d_size = gelf_fsize((*elf), ELF_T_ADDR, NR_ADDR, EV_CURRENT),
2298 		.d_off = 0, .d_align = 0
2299 	};
2300 	Elf_Data src = {
2301 		.d_buf = (void *) data, .d_type = ELF_T_ADDR,
2302 		.d_version = EV_CURRENT, .d_size = dst.d_size, .d_off = 0,
2303 		.d_align = 0
2304 	};
2305 
2306 	tmp = (struct sdt_note *)calloc(1, sizeof(struct sdt_note));
2307 	if (!tmp) {
2308 		ret = -ENOMEM;
2309 		goto out_err;
2310 	}
2311 
2312 	INIT_LIST_HEAD(&tmp->note_list);
2313 
2314 	if (len < dst.d_size + 3)
2315 		goto out_free_note;
2316 
2317 	/* Translation from file representation to memory representation */
2318 	if (gelf_xlatetom(*elf, &dst, &src,
2319 			  elf_getident(*elf, NULL)[EI_DATA]) == NULL) {
2320 		pr_err("gelf_xlatetom : %s\n", elf_errmsg(-1));
2321 		goto out_free_note;
2322 	}
2323 
2324 	/* Populate the fields of sdt_note */
2325 	provider = data + dst.d_size;
2326 
2327 	name = (const char *)memchr(provider, '\0', data + len - provider);
2328 	if (name++ == NULL)
2329 		goto out_free_note;
2330 
2331 	tmp->provider = strdup(provider);
2332 	if (!tmp->provider) {
2333 		ret = -ENOMEM;
2334 		goto out_free_note;
2335 	}
2336 	tmp->name = strdup(name);
2337 	if (!tmp->name) {
2338 		ret = -ENOMEM;
2339 		goto out_free_prov;
2340 	}
2341 
2342 	args = memchr(name, '\0', data + len - name);
2343 
2344 	/*
2345 	 * There is no argument if:
2346 	 * - We reached the end of the note;
2347 	 * - There is not enough room to hold a potential string;
2348 	 * - The argument string is empty or just contains ':'.
2349 	 */
2350 	if (args == NULL || data + len - args < 2 ||
2351 		args[1] == ':' || args[1] == '\0')
2352 		tmp->args = NULL;
2353 	else {
2354 		tmp->args = strdup(++args);
2355 		if (!tmp->args) {
2356 			ret = -ENOMEM;
2357 			goto out_free_name;
2358 		}
2359 	}
2360 
2361 	if (gelf_getclass(*elf) == ELFCLASS32) {
2362 		memcpy(&tmp->addr, &buf, 3 * sizeof(Elf32_Addr));
2363 		tmp->bit32 = true;
2364 	} else {
2365 		memcpy(&tmp->addr, &buf, 3 * sizeof(Elf64_Addr));
2366 		tmp->bit32 = false;
2367 	}
2368 
2369 	if (!gelf_getehdr(*elf, &ehdr)) {
2370 		pr_debug("%s : cannot get elf header.\n", __func__);
2371 		ret = -EBADF;
2372 		goto out_free_args;
2373 	}
2374 
2375 	/* Adjust the prelink effect :
2376 	 * Find out the .stapsdt.base section.
2377 	 * This scn will help us to handle prelinking (if present).
2378 	 * Compare the retrieved file offset of the base section with the
2379 	 * base address in the description of the SDT note. If its different,
2380 	 * then accordingly, adjust the note location.
2381 	 */
2382 	if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_BASE_SCN, NULL))
2383 		sdt_adjust_loc(tmp, shdr.sh_offset);
2384 
2385 	/* Adjust reference counter offset */
2386 	if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_PROBES_SCN, NULL))
2387 		sdt_adjust_refctr(tmp, shdr.sh_addr, shdr.sh_offset);
2388 
2389 	list_add_tail(&tmp->note_list, sdt_notes);
2390 	return 0;
2391 
2392 out_free_args:
2393 	zfree(&tmp->args);
2394 out_free_name:
2395 	zfree(&tmp->name);
2396 out_free_prov:
2397 	zfree(&tmp->provider);
2398 out_free_note:
2399 	free(tmp);
2400 out_err:
2401 	return ret;
2402 }
2403 
2404 /**
2405  * construct_sdt_notes_list : constructs a list of SDT notes
2406  * @elf : elf to look into
2407  * @sdt_notes : empty list_head
2408  *
2409  * Scans the sections in 'elf' for the section
2410  * .note.stapsdt. It, then calls populate_sdt_note to find
2411  * out the SDT events and populates the 'sdt_notes'.
2412  */
construct_sdt_notes_list(Elf * elf,struct list_head * sdt_notes)2413 static int construct_sdt_notes_list(Elf *elf, struct list_head *sdt_notes)
2414 {
2415 	GElf_Ehdr ehdr;
2416 	Elf_Scn *scn = NULL;
2417 	Elf_Data *data;
2418 	GElf_Shdr shdr;
2419 	size_t shstrndx, next;
2420 	GElf_Nhdr nhdr;
2421 	size_t name_off, desc_off, offset;
2422 	int ret = 0;
2423 
2424 	if (gelf_getehdr(elf, &ehdr) == NULL) {
2425 		ret = -EBADF;
2426 		goto out_ret;
2427 	}
2428 	if (elf_getshdrstrndx(elf, &shstrndx) != 0) {
2429 		ret = -EBADF;
2430 		goto out_ret;
2431 	}
2432 
2433 	/* Look for the required section */
2434 	scn = elf_section_by_name(elf, &ehdr, &shdr, SDT_NOTE_SCN, NULL);
2435 	if (!scn) {
2436 		ret = -ENOENT;
2437 		goto out_ret;
2438 	}
2439 
2440 	if ((shdr.sh_type != SHT_NOTE) || (shdr.sh_flags & SHF_ALLOC)) {
2441 		ret = -ENOENT;
2442 		goto out_ret;
2443 	}
2444 
2445 	data = elf_getdata(scn, NULL);
2446 
2447 	/* Get the SDT notes */
2448 	for (offset = 0; (next = gelf_getnote(data, offset, &nhdr, &name_off,
2449 					      &desc_off)) > 0; offset = next) {
2450 		if (nhdr.n_namesz == sizeof(SDT_NOTE_NAME) &&
2451 		    !memcmp(data->d_buf + name_off, SDT_NOTE_NAME,
2452 			    sizeof(SDT_NOTE_NAME))) {
2453 			/* Check the type of the note */
2454 			if (nhdr.n_type != SDT_NOTE_TYPE)
2455 				goto out_ret;
2456 
2457 			ret = populate_sdt_note(&elf, ((data->d_buf) + desc_off),
2458 						nhdr.n_descsz, sdt_notes);
2459 			if (ret < 0)
2460 				goto out_ret;
2461 		}
2462 	}
2463 	if (list_empty(sdt_notes))
2464 		ret = -ENOENT;
2465 
2466 out_ret:
2467 	return ret;
2468 }
2469 
2470 /**
2471  * get_sdt_note_list : Wrapper to construct a list of sdt notes
2472  * @head : empty list_head
2473  * @target : file to find SDT notes from
2474  *
2475  * This opens the file, initializes
2476  * the ELF and then calls construct_sdt_notes_list.
2477  */
get_sdt_note_list(struct list_head * head,const char * target)2478 int get_sdt_note_list(struct list_head *head, const char *target)
2479 {
2480 	Elf *elf;
2481 	int fd, ret;
2482 
2483 	fd = open(target, O_RDONLY);
2484 	if (fd < 0)
2485 		return -EBADF;
2486 
2487 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
2488 	if (!elf) {
2489 		ret = -EBADF;
2490 		goto out_close;
2491 	}
2492 	ret = construct_sdt_notes_list(elf, head);
2493 	elf_end(elf);
2494 out_close:
2495 	close(fd);
2496 	return ret;
2497 }
2498 
2499 /**
2500  * cleanup_sdt_note_list : free the sdt notes' list
2501  * @sdt_notes: sdt notes' list
2502  *
2503  * Free up the SDT notes in @sdt_notes.
2504  * Returns the number of SDT notes free'd.
2505  */
cleanup_sdt_note_list(struct list_head * sdt_notes)2506 int cleanup_sdt_note_list(struct list_head *sdt_notes)
2507 {
2508 	struct sdt_note *tmp, *pos;
2509 	int nr_free = 0;
2510 
2511 	list_for_each_entry_safe(pos, tmp, sdt_notes, note_list) {
2512 		list_del_init(&pos->note_list);
2513 		zfree(&pos->args);
2514 		zfree(&pos->name);
2515 		zfree(&pos->provider);
2516 		free(pos);
2517 		nr_free++;
2518 	}
2519 	return nr_free;
2520 }
2521 
2522 /**
2523  * sdt_notes__get_count: Counts the number of sdt events
2524  * @start: list_head to sdt_notes list
2525  *
2526  * Returns the number of SDT notes in a list
2527  */
sdt_notes__get_count(struct list_head * start)2528 int sdt_notes__get_count(struct list_head *start)
2529 {
2530 	struct sdt_note *sdt_ptr;
2531 	int count = 0;
2532 
2533 	list_for_each_entry(sdt_ptr, start, note_list)
2534 		count++;
2535 	return count;
2536 }
2537 #endif
2538 
symbol__elf_init(void)2539 void symbol__elf_init(void)
2540 {
2541 	elf_version(EV_CURRENT);
2542 }
2543