xref: /openbmc/linux/tools/perf/util/symbol-elf.c (revision e59fea47)
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-cxx.h"
16 #include "demangle-ocaml.h"
17 #include "demangle-java.h"
18 #include "demangle-rust.h"
19 #include "machine.h"
20 #include "vdso.h"
21 #include "debug.h"
22 #include "util/copyfile.h"
23 #include <linux/ctype.h>
24 #include <linux/kernel.h>
25 #include <linux/zalloc.h>
26 #include <symbol/kallsyms.h>
27 #include <internal/lib.h>
28 
29 #ifdef HAVE_LIBBFD_SUPPORT
30 #define PACKAGE 'perf'
31 #include <bfd.h>
32 #endif
33 
34 #if defined(HAVE_LIBBFD_SUPPORT) || defined(HAVE_CPLUS_DEMANGLE_SUPPORT)
35 #ifndef DMGL_PARAMS
36 #define DMGL_PARAMS     (1 << 0)  /* Include function args */
37 #define DMGL_ANSI       (1 << 1)  /* Include const, volatile, etc */
38 #endif
39 #endif
40 
41 #ifndef EM_AARCH64
42 #define EM_AARCH64	183  /* ARM 64 bit */
43 #endif
44 
45 #ifndef EM_LOONGARCH
46 #define EM_LOONGARCH	258
47 #endif
48 
49 #ifndef ELF32_ST_VISIBILITY
50 #define ELF32_ST_VISIBILITY(o)	((o) & 0x03)
51 #endif
52 
53 /* For ELF64 the definitions are the same.  */
54 #ifndef ELF64_ST_VISIBILITY
55 #define ELF64_ST_VISIBILITY(o)	ELF32_ST_VISIBILITY (o)
56 #endif
57 
58 /* How to extract information held in the st_other field.  */
59 #ifndef GELF_ST_VISIBILITY
60 #define GELF_ST_VISIBILITY(val)	ELF64_ST_VISIBILITY (val)
61 #endif
62 
63 typedef Elf64_Nhdr GElf_Nhdr;
64 
65 
66 #ifndef HAVE_ELF_GETPHDRNUM_SUPPORT
elf_getphdrnum(Elf * elf,size_t * dst)67 static int elf_getphdrnum(Elf *elf, size_t *dst)
68 {
69 	GElf_Ehdr gehdr;
70 	GElf_Ehdr *ehdr;
71 
72 	ehdr = gelf_getehdr(elf, &gehdr);
73 	if (!ehdr)
74 		return -1;
75 
76 	*dst = ehdr->e_phnum;
77 
78 	return 0;
79 }
80 #endif
81 
82 #ifndef HAVE_ELF_GETSHDRSTRNDX_SUPPORT
elf_getshdrstrndx(Elf * elf __maybe_unused,size_t * dst __maybe_unused)83 static int elf_getshdrstrndx(Elf *elf __maybe_unused, size_t *dst __maybe_unused)
84 {
85 	pr_err("%s: update your libelf to > 0.140, this one lacks elf_getshdrstrndx().\n", __func__);
86 	return -1;
87 }
88 #endif
89 
90 #ifndef NT_GNU_BUILD_ID
91 #define NT_GNU_BUILD_ID 3
92 #endif
93 
94 /**
95  * elf_symtab__for_each_symbol - iterate thru all the symbols
96  *
97  * @syms: struct elf_symtab instance to iterate
98  * @idx: uint32_t idx
99  * @sym: GElf_Sym iterator
100  */
101 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
102 	for (idx = 0, gelf_getsym(syms, idx, &sym);\
103 	     idx < nr_syms; \
104 	     idx++, gelf_getsym(syms, idx, &sym))
105 
elf_sym__type(const GElf_Sym * sym)106 static inline uint8_t elf_sym__type(const GElf_Sym *sym)
107 {
108 	return GELF_ST_TYPE(sym->st_info);
109 }
110 
elf_sym__visibility(const GElf_Sym * sym)111 static inline uint8_t elf_sym__visibility(const GElf_Sym *sym)
112 {
113 	return GELF_ST_VISIBILITY(sym->st_other);
114 }
115 
116 #ifndef STT_GNU_IFUNC
117 #define STT_GNU_IFUNC 10
118 #endif
119 
elf_sym__is_function(const GElf_Sym * sym)120 static inline int elf_sym__is_function(const GElf_Sym *sym)
121 {
122 	return (elf_sym__type(sym) == STT_FUNC ||
123 		elf_sym__type(sym) == STT_GNU_IFUNC) &&
124 	       sym->st_name != 0 &&
125 	       sym->st_shndx != SHN_UNDEF;
126 }
127 
elf_sym__is_object(const GElf_Sym * sym)128 static inline bool elf_sym__is_object(const GElf_Sym *sym)
129 {
130 	return elf_sym__type(sym) == STT_OBJECT &&
131 		sym->st_name != 0 &&
132 		sym->st_shndx != SHN_UNDEF;
133 }
134 
elf_sym__is_label(const GElf_Sym * sym)135 static inline int elf_sym__is_label(const GElf_Sym *sym)
136 {
137 	return elf_sym__type(sym) == STT_NOTYPE &&
138 		sym->st_name != 0 &&
139 		sym->st_shndx != SHN_UNDEF &&
140 		sym->st_shndx != SHN_ABS &&
141 		elf_sym__visibility(sym) != STV_HIDDEN &&
142 		elf_sym__visibility(sym) != STV_INTERNAL;
143 }
144 
elf_sym__filter(GElf_Sym * sym)145 static bool elf_sym__filter(GElf_Sym *sym)
146 {
147 	return elf_sym__is_function(sym) || elf_sym__is_object(sym);
148 }
149 
elf_sym__name(const GElf_Sym * sym,const Elf_Data * symstrs)150 static inline const char *elf_sym__name(const GElf_Sym *sym,
151 					const Elf_Data *symstrs)
152 {
153 	return symstrs->d_buf + sym->st_name;
154 }
155 
elf_sec__name(const GElf_Shdr * shdr,const Elf_Data * secstrs)156 static inline const char *elf_sec__name(const GElf_Shdr *shdr,
157 					const Elf_Data *secstrs)
158 {
159 	return secstrs->d_buf + shdr->sh_name;
160 }
161 
elf_sec__is_text(const GElf_Shdr * shdr,const Elf_Data * secstrs)162 static inline int elf_sec__is_text(const GElf_Shdr *shdr,
163 					const Elf_Data *secstrs)
164 {
165 	return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
166 }
167 
elf_sec__is_data(const GElf_Shdr * shdr,const Elf_Data * secstrs)168 static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
169 				    const Elf_Data *secstrs)
170 {
171 	return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
172 }
173 
elf_sec__filter(GElf_Shdr * shdr,Elf_Data * secstrs)174 static bool elf_sec__filter(GElf_Shdr *shdr, Elf_Data *secstrs)
175 {
176 	return elf_sec__is_text(shdr, secstrs) ||
177 	       elf_sec__is_data(shdr, secstrs);
178 }
179 
elf_addr_to_index(Elf * elf,GElf_Addr addr)180 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
181 {
182 	Elf_Scn *sec = NULL;
183 	GElf_Shdr shdr;
184 	size_t cnt = 1;
185 
186 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
187 		gelf_getshdr(sec, &shdr);
188 
189 		if ((addr >= shdr.sh_addr) &&
190 		    (addr < (shdr.sh_addr + shdr.sh_size)))
191 			return cnt;
192 
193 		++cnt;
194 	}
195 
196 	return -1;
197 }
198 
elf_section_by_name(Elf * elf,GElf_Ehdr * ep,GElf_Shdr * shp,const char * name,size_t * idx)199 Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
200 			     GElf_Shdr *shp, const char *name, size_t *idx)
201 {
202 	Elf_Scn *sec = NULL;
203 	size_t cnt = 1;
204 
205 	/* ELF is corrupted/truncated, avoid calling elf_strptr. */
206 	if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL))
207 		return NULL;
208 
209 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
210 		char *str;
211 
212 		gelf_getshdr(sec, shp);
213 		str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
214 		if (str && !strcmp(name, str)) {
215 			if (idx)
216 				*idx = cnt;
217 			return sec;
218 		}
219 		++cnt;
220 	}
221 
222 	return NULL;
223 }
224 
filename__has_section(const char * filename,const char * sec)225 bool filename__has_section(const char *filename, const char *sec)
226 {
227 	int fd;
228 	Elf *elf;
229 	GElf_Ehdr ehdr;
230 	GElf_Shdr shdr;
231 	bool found = false;
232 
233 	fd = open(filename, O_RDONLY);
234 	if (fd < 0)
235 		return false;
236 
237 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
238 	if (elf == NULL)
239 		goto out;
240 
241 	if (gelf_getehdr(elf, &ehdr) == NULL)
242 		goto elf_out;
243 
244 	found = !!elf_section_by_name(elf, &ehdr, &shdr, sec, NULL);
245 
246 elf_out:
247 	elf_end(elf);
248 out:
249 	close(fd);
250 	return found;
251 }
252 
elf_read_program_header(Elf * elf,u64 vaddr,GElf_Phdr * phdr)253 static int elf_read_program_header(Elf *elf, u64 vaddr, GElf_Phdr *phdr)
254 {
255 	size_t i, phdrnum;
256 	u64 sz;
257 
258 	if (elf_getphdrnum(elf, &phdrnum))
259 		return -1;
260 
261 	for (i = 0; i < phdrnum; i++) {
262 		if (gelf_getphdr(elf, i, phdr) == NULL)
263 			return -1;
264 
265 		if (phdr->p_type != PT_LOAD)
266 			continue;
267 
268 		sz = max(phdr->p_memsz, phdr->p_filesz);
269 		if (!sz)
270 			continue;
271 
272 		if (vaddr >= phdr->p_vaddr && (vaddr < phdr->p_vaddr + sz))
273 			return 0;
274 	}
275 
276 	/* Not found any valid program header */
277 	return -1;
278 }
279 
want_demangle(bool is_kernel_sym)280 static bool want_demangle(bool is_kernel_sym)
281 {
282 	return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle;
283 }
284 
285 /*
286  * Demangle C++ function signature, typically replaced by demangle-cxx.cpp
287  * version.
288  */
cxx_demangle_sym(const char * str __maybe_unused,bool params __maybe_unused,bool modifiers __maybe_unused)289 __weak char *cxx_demangle_sym(const char *str __maybe_unused, bool params __maybe_unused,
290 			      bool modifiers __maybe_unused)
291 {
292 #ifdef HAVE_LIBBFD_SUPPORT
293 	int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0);
294 
295 	return bfd_demangle(NULL, str, flags);
296 #elif defined(HAVE_CPLUS_DEMANGLE_SUPPORT)
297 	int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0);
298 
299 	return cplus_demangle(str, flags);
300 #else
301 	return NULL;
302 #endif
303 }
304 
demangle_sym(struct dso * dso,int kmodule,const char * elf_name)305 static char *demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
306 {
307 	char *demangled = NULL;
308 
309 	/*
310 	 * We need to figure out if the object was created from C++ sources
311 	 * DWARF DW_compile_unit has this, but we don't always have access
312 	 * to it...
313 	 */
314 	if (!want_demangle(dso->kernel || kmodule))
315 	    return demangled;
316 
317 	demangled = cxx_demangle_sym(elf_name, verbose > 0, verbose > 0);
318 	if (demangled == NULL) {
319 		demangled = ocaml_demangle_sym(elf_name);
320 		if (demangled == NULL) {
321 			demangled = java_demangle_sym(elf_name, JAVA_DEMANGLE_NORET);
322 		}
323 	}
324 	else if (rust_is_mangled(demangled))
325 		/*
326 		    * Input to Rust demangling is the BFD-demangled
327 		    * name which it Rust-demangles in place.
328 		    */
329 		rust_demangle_sym(demangled);
330 
331 	return demangled;
332 }
333 
334 struct rel_info {
335 	u32		nr_entries;
336 	u32		*sorted;
337 	bool		is_rela;
338 	Elf_Data	*reldata;
339 	GElf_Rela	rela;
340 	GElf_Rel	rel;
341 };
342 
get_rel_symidx(struct rel_info * ri,u32 idx)343 static u32 get_rel_symidx(struct rel_info *ri, u32 idx)
344 {
345 	idx = ri->sorted ? ri->sorted[idx] : idx;
346 	if (ri->is_rela) {
347 		gelf_getrela(ri->reldata, idx, &ri->rela);
348 		return GELF_R_SYM(ri->rela.r_info);
349 	}
350 	gelf_getrel(ri->reldata, idx, &ri->rel);
351 	return GELF_R_SYM(ri->rel.r_info);
352 }
353 
get_rel_offset(struct rel_info * ri,u32 x)354 static u64 get_rel_offset(struct rel_info *ri, u32 x)
355 {
356 	if (ri->is_rela) {
357 		GElf_Rela rela;
358 
359 		gelf_getrela(ri->reldata, x, &rela);
360 		return rela.r_offset;
361 	} else {
362 		GElf_Rel rel;
363 
364 		gelf_getrel(ri->reldata, x, &rel);
365 		return rel.r_offset;
366 	}
367 }
368 
rel_cmp(const void * a,const void * b,void * r)369 static int rel_cmp(const void *a, const void *b, void *r)
370 {
371 	struct rel_info *ri = r;
372 	u64 a_offset = get_rel_offset(ri, *(const u32 *)a);
373 	u64 b_offset = get_rel_offset(ri, *(const u32 *)b);
374 
375 	return a_offset < b_offset ? -1 : (a_offset > b_offset ? 1 : 0);
376 }
377 
sort_rel(struct rel_info * ri)378 static int sort_rel(struct rel_info *ri)
379 {
380 	size_t sz = sizeof(ri->sorted[0]);
381 	u32 i;
382 
383 	ri->sorted = calloc(ri->nr_entries, sz);
384 	if (!ri->sorted)
385 		return -1;
386 	for (i = 0; i < ri->nr_entries; i++)
387 		ri->sorted[i] = i;
388 	qsort_r(ri->sorted, ri->nr_entries, sz, rel_cmp, ri);
389 	return 0;
390 }
391 
392 /*
393  * For x86_64, the GNU linker is putting IFUNC information in the relocation
394  * addend.
395  */
addend_may_be_ifunc(GElf_Ehdr * ehdr,struct rel_info * ri)396 static bool addend_may_be_ifunc(GElf_Ehdr *ehdr, struct rel_info *ri)
397 {
398 	return ehdr->e_machine == EM_X86_64 && ri->is_rela &&
399 	       GELF_R_TYPE(ri->rela.r_info) == R_X86_64_IRELATIVE;
400 }
401 
get_ifunc_name(Elf * elf,struct dso * dso,GElf_Ehdr * ehdr,struct rel_info * ri,char * buf,size_t buf_sz)402 static bool get_ifunc_name(Elf *elf, struct dso *dso, GElf_Ehdr *ehdr,
403 			   struct rel_info *ri, char *buf, size_t buf_sz)
404 {
405 	u64 addr = ri->rela.r_addend;
406 	struct symbol *sym;
407 	GElf_Phdr phdr;
408 
409 	if (!addend_may_be_ifunc(ehdr, ri))
410 		return false;
411 
412 	if (elf_read_program_header(elf, addr, &phdr))
413 		return false;
414 
415 	addr -= phdr.p_vaddr - phdr.p_offset;
416 
417 	sym = dso__find_symbol_nocache(dso, addr);
418 
419 	/* Expecting the address to be an IFUNC or IFUNC alias */
420 	if (!sym || sym->start != addr || (sym->type != STT_GNU_IFUNC && !sym->ifunc_alias))
421 		return false;
422 
423 	snprintf(buf, buf_sz, "%s@plt", sym->name);
424 
425 	return true;
426 }
427 
exit_rel(struct rel_info * ri)428 static void exit_rel(struct rel_info *ri)
429 {
430 	zfree(&ri->sorted);
431 }
432 
get_plt_sizes(struct dso * dso,GElf_Ehdr * ehdr,GElf_Shdr * shdr_plt,u64 * plt_header_size,u64 * plt_entry_size)433 static bool get_plt_sizes(struct dso *dso, GElf_Ehdr *ehdr, GElf_Shdr *shdr_plt,
434 			  u64 *plt_header_size, u64 *plt_entry_size)
435 {
436 	switch (ehdr->e_machine) {
437 	case EM_ARM:
438 		*plt_header_size = 20;
439 		*plt_entry_size = 12;
440 		return true;
441 	case EM_AARCH64:
442 		*plt_header_size = 32;
443 		*plt_entry_size = 16;
444 		return true;
445 	case EM_LOONGARCH:
446 		*plt_header_size = 32;
447 		*plt_entry_size = 16;
448 		return true;
449 	case EM_SPARC:
450 		*plt_header_size = 48;
451 		*plt_entry_size = 12;
452 		return true;
453 	case EM_SPARCV9:
454 		*plt_header_size = 128;
455 		*plt_entry_size = 32;
456 		return true;
457 	case EM_386:
458 	case EM_X86_64:
459 		*plt_entry_size = shdr_plt->sh_entsize;
460 		/* Size is 8 or 16, if not, assume alignment indicates size */
461 		if (*plt_entry_size != 8 && *plt_entry_size != 16)
462 			*plt_entry_size = shdr_plt->sh_addralign == 8 ? 8 : 16;
463 		*plt_header_size = *plt_entry_size;
464 		break;
465 	default: /* FIXME: s390/alpha/mips/parisc/poperpc/sh/xtensa need to be checked */
466 		*plt_header_size = shdr_plt->sh_entsize;
467 		*plt_entry_size = shdr_plt->sh_entsize;
468 		break;
469 	}
470 	if (*plt_entry_size)
471 		return true;
472 	pr_debug("Missing PLT entry size for %s\n", dso->long_name);
473 	return false;
474 }
475 
machine_is_x86(GElf_Half e_machine)476 static bool machine_is_x86(GElf_Half e_machine)
477 {
478 	return e_machine == EM_386 || e_machine == EM_X86_64;
479 }
480 
481 struct rela_dyn {
482 	GElf_Addr	offset;
483 	u32		sym_idx;
484 };
485 
486 struct rela_dyn_info {
487 	struct dso	*dso;
488 	Elf_Data	*plt_got_data;
489 	u32		nr_entries;
490 	struct rela_dyn	*sorted;
491 	Elf_Data	*dynsym_data;
492 	Elf_Data	*dynstr_data;
493 	Elf_Data	*rela_dyn_data;
494 };
495 
exit_rela_dyn(struct rela_dyn_info * di)496 static void exit_rela_dyn(struct rela_dyn_info *di)
497 {
498 	zfree(&di->sorted);
499 }
500 
cmp_offset(const void * a,const void * b)501 static int cmp_offset(const void *a, const void *b)
502 {
503 	const struct rela_dyn *va = a;
504 	const struct rela_dyn *vb = b;
505 
506 	return va->offset < vb->offset ? -1 : (va->offset > vb->offset ? 1 : 0);
507 }
508 
sort_rela_dyn(struct rela_dyn_info * di)509 static int sort_rela_dyn(struct rela_dyn_info *di)
510 {
511 	u32 i, n;
512 
513 	di->sorted = calloc(di->nr_entries, sizeof(di->sorted[0]));
514 	if (!di->sorted)
515 		return -1;
516 
517 	/* Get data for sorting: the offset and symbol index */
518 	for (i = 0, n = 0; i < di->nr_entries; i++) {
519 		GElf_Rela rela;
520 		u32 sym_idx;
521 
522 		gelf_getrela(di->rela_dyn_data, i, &rela);
523 		sym_idx = GELF_R_SYM(rela.r_info);
524 		if (sym_idx) {
525 			di->sorted[n].sym_idx = sym_idx;
526 			di->sorted[n].offset = rela.r_offset;
527 			n += 1;
528 		}
529 	}
530 
531 	/* Sort by offset */
532 	di->nr_entries = n;
533 	qsort(di->sorted, n, sizeof(di->sorted[0]), cmp_offset);
534 
535 	return 0;
536 }
537 
get_rela_dyn_info(Elf * elf,GElf_Ehdr * ehdr,struct rela_dyn_info * di,Elf_Scn * scn)538 static void get_rela_dyn_info(Elf *elf, GElf_Ehdr *ehdr, struct rela_dyn_info *di, Elf_Scn *scn)
539 {
540 	GElf_Shdr rela_dyn_shdr;
541 	GElf_Shdr shdr;
542 
543 	di->plt_got_data = elf_getdata(scn, NULL);
544 
545 	scn = elf_section_by_name(elf, ehdr, &rela_dyn_shdr, ".rela.dyn", NULL);
546 	if (!scn || !rela_dyn_shdr.sh_link || !rela_dyn_shdr.sh_entsize)
547 		return;
548 
549 	di->nr_entries = rela_dyn_shdr.sh_size / rela_dyn_shdr.sh_entsize;
550 	di->rela_dyn_data = elf_getdata(scn, NULL);
551 
552 	scn = elf_getscn(elf, rela_dyn_shdr.sh_link);
553 	if (!scn || !gelf_getshdr(scn, &shdr) || !shdr.sh_link)
554 		return;
555 
556 	di->dynsym_data = elf_getdata(scn, NULL);
557 	di->dynstr_data = elf_getdata(elf_getscn(elf, shdr.sh_link), NULL);
558 
559 	if (!di->plt_got_data || !di->dynstr_data || !di->dynsym_data || !di->rela_dyn_data)
560 		return;
561 
562 	/* Sort into offset order */
563 	sort_rela_dyn(di);
564 }
565 
566 /* Get instruction displacement from a plt entry for x86_64 */
get_x86_64_plt_disp(const u8 * p)567 static u32 get_x86_64_plt_disp(const u8 *p)
568 {
569 	u8 endbr64[] = {0xf3, 0x0f, 0x1e, 0xfa};
570 	int n = 0;
571 
572 	/* Skip endbr64 */
573 	if (!memcmp(p, endbr64, sizeof(endbr64)))
574 		n += sizeof(endbr64);
575 	/* Skip bnd prefix */
576 	if (p[n] == 0xf2)
577 		n += 1;
578 	/* jmp with 4-byte displacement */
579 	if (p[n] == 0xff && p[n + 1] == 0x25) {
580 		u32 disp;
581 
582 		n += 2;
583 		/* Also add offset from start of entry to end of instruction */
584 		memcpy(&disp, p + n, sizeof(disp));
585 		return n + 4 + le32toh(disp);
586 	}
587 	return 0;
588 }
589 
get_plt_got_name(GElf_Shdr * shdr,size_t i,struct rela_dyn_info * di,char * buf,size_t buf_sz)590 static bool get_plt_got_name(GElf_Shdr *shdr, size_t i,
591 			     struct rela_dyn_info *di,
592 			     char *buf, size_t buf_sz)
593 {
594 	struct rela_dyn vi, *vr;
595 	const char *sym_name;
596 	char *demangled;
597 	GElf_Sym sym;
598 	bool result;
599 	u32 disp;
600 
601 	if (!di->sorted)
602 		return false;
603 
604 	disp = get_x86_64_plt_disp(di->plt_got_data->d_buf + i);
605 	if (!disp)
606 		return false;
607 
608 	/* Compute target offset of the .plt.got entry */
609 	vi.offset = shdr->sh_offset + di->plt_got_data->d_off + i + disp;
610 
611 	/* Find that offset in .rela.dyn (sorted by offset) */
612 	vr = bsearch(&vi, di->sorted, di->nr_entries, sizeof(di->sorted[0]), cmp_offset);
613 	if (!vr)
614 		return false;
615 
616 	/* Get the associated symbol */
617 	gelf_getsym(di->dynsym_data, vr->sym_idx, &sym);
618 	sym_name = elf_sym__name(&sym, di->dynstr_data);
619 	demangled = demangle_sym(di->dso, 0, sym_name);
620 	if (demangled != NULL)
621 		sym_name = demangled;
622 
623 	snprintf(buf, buf_sz, "%s@plt", sym_name);
624 
625 	result = *sym_name;
626 
627 	free(demangled);
628 
629 	return result;
630 }
631 
dso__synthesize_plt_got_symbols(struct dso * dso,Elf * elf,GElf_Ehdr * ehdr,char * buf,size_t buf_sz)632 static int dso__synthesize_plt_got_symbols(struct dso *dso, Elf *elf,
633 					   GElf_Ehdr *ehdr,
634 					   char *buf, size_t buf_sz)
635 {
636 	struct rela_dyn_info di = { .dso = dso };
637 	struct symbol *sym;
638 	GElf_Shdr shdr;
639 	Elf_Scn *scn;
640 	int err = -1;
641 	size_t i;
642 
643 	scn = elf_section_by_name(elf, ehdr, &shdr, ".plt.got", NULL);
644 	if (!scn || !shdr.sh_entsize)
645 		return 0;
646 
647 	if (ehdr->e_machine == EM_X86_64)
648 		get_rela_dyn_info(elf, ehdr, &di, scn);
649 
650 	for (i = 0; i < shdr.sh_size; i += shdr.sh_entsize) {
651 		if (!get_plt_got_name(&shdr, i, &di, buf, buf_sz))
652 			snprintf(buf, buf_sz, "offset_%#" PRIx64 "@plt", (u64)shdr.sh_offset + i);
653 		sym = symbol__new(shdr.sh_offset + i, shdr.sh_entsize, STB_GLOBAL, STT_FUNC, buf);
654 		if (!sym)
655 			goto out;
656 		symbols__insert(&dso->symbols, sym);
657 	}
658 	err = 0;
659 out:
660 	exit_rela_dyn(&di);
661 	return err;
662 }
663 
664 /*
665  * We need to check if we have a .dynsym, so that we can handle the
666  * .plt, synthesizing its symbols, that aren't on the symtabs (be it
667  * .dynsym or .symtab).
668  * And always look at the original dso, not at debuginfo packages, that
669  * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
670  */
dso__synthesize_plt_symbols(struct dso * dso,struct symsrc * ss)671 int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss)
672 {
673 	uint32_t idx;
674 	GElf_Sym sym;
675 	u64 plt_offset, plt_header_size, plt_entry_size;
676 	GElf_Shdr shdr_plt, plt_sec_shdr;
677 	struct symbol *f, *plt_sym;
678 	GElf_Shdr shdr_rel_plt, shdr_dynsym;
679 	Elf_Data *syms, *symstrs;
680 	Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
681 	GElf_Ehdr ehdr;
682 	char sympltname[1024];
683 	Elf *elf;
684 	int nr = 0, err = -1;
685 	struct rel_info ri = { .is_rela = false };
686 	bool lazy_plt;
687 
688 	elf = ss->elf;
689 	ehdr = ss->ehdr;
690 
691 	if (!elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL))
692 		return 0;
693 
694 	/*
695 	 * A symbol from a previous section (e.g. .init) can have been expanded
696 	 * by symbols__fixup_end() to overlap .plt. Truncate it before adding
697 	 * a symbol for .plt header.
698 	 */
699 	f = dso__find_symbol_nocache(dso, shdr_plt.sh_offset);
700 	if (f && f->start < shdr_plt.sh_offset && f->end > shdr_plt.sh_offset)
701 		f->end = shdr_plt.sh_offset;
702 
703 	if (!get_plt_sizes(dso, &ehdr, &shdr_plt, &plt_header_size, &plt_entry_size))
704 		return 0;
705 
706 	/* Add a symbol for .plt header */
707 	plt_sym = symbol__new(shdr_plt.sh_offset, plt_header_size, STB_GLOBAL, STT_FUNC, ".plt");
708 	if (!plt_sym)
709 		goto out_elf_end;
710 	symbols__insert(&dso->symbols, plt_sym);
711 
712 	/* Only x86 has .plt.got */
713 	if (machine_is_x86(ehdr.e_machine) &&
714 	    dso__synthesize_plt_got_symbols(dso, elf, &ehdr, sympltname, sizeof(sympltname)))
715 		goto out_elf_end;
716 
717 	/* Only x86 has .plt.sec */
718 	if (machine_is_x86(ehdr.e_machine) &&
719 	    elf_section_by_name(elf, &ehdr, &plt_sec_shdr, ".plt.sec", NULL)) {
720 		if (!get_plt_sizes(dso, &ehdr, &plt_sec_shdr, &plt_header_size, &plt_entry_size))
721 			return 0;
722 		/* Extend .plt symbol to entire .plt */
723 		plt_sym->end = plt_sym->start + shdr_plt.sh_size;
724 		/* Use .plt.sec offset */
725 		plt_offset = plt_sec_shdr.sh_offset;
726 		lazy_plt = false;
727 	} else {
728 		plt_offset = shdr_plt.sh_offset;
729 		lazy_plt = true;
730 	}
731 
732 	scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
733 					  ".rela.plt", NULL);
734 	if (scn_plt_rel == NULL) {
735 		scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
736 						  ".rel.plt", NULL);
737 		if (scn_plt_rel == NULL)
738 			return 0;
739 	}
740 
741 	if (shdr_rel_plt.sh_type != SHT_RELA &&
742 	    shdr_rel_plt.sh_type != SHT_REL)
743 		return 0;
744 
745 	if (!shdr_rel_plt.sh_link)
746 		return 0;
747 
748 	if (shdr_rel_plt.sh_link == ss->dynsym_idx) {
749 		scn_dynsym = ss->dynsym;
750 		shdr_dynsym = ss->dynshdr;
751 	} else if (shdr_rel_plt.sh_link == ss->symtab_idx) {
752 		/*
753 		 * A static executable can have a .plt due to IFUNCs, in which
754 		 * case .symtab is used not .dynsym.
755 		 */
756 		scn_dynsym = ss->symtab;
757 		shdr_dynsym = ss->symshdr;
758 	} else {
759 		goto out_elf_end;
760 	}
761 
762 	if (!scn_dynsym)
763 		return 0;
764 
765 	/*
766 	 * Fetch the relocation section to find the idxes to the GOT
767 	 * and the symbols in the .dynsym they refer to.
768 	 */
769 	ri.reldata = elf_getdata(scn_plt_rel, NULL);
770 	if (!ri.reldata)
771 		goto out_elf_end;
772 
773 	syms = elf_getdata(scn_dynsym, NULL);
774 	if (syms == NULL)
775 		goto out_elf_end;
776 
777 	scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
778 	if (scn_symstrs == NULL)
779 		goto out_elf_end;
780 
781 	symstrs = elf_getdata(scn_symstrs, NULL);
782 	if (symstrs == NULL)
783 		goto out_elf_end;
784 
785 	if (symstrs->d_size == 0)
786 		goto out_elf_end;
787 
788 	ri.nr_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
789 
790 	ri.is_rela = shdr_rel_plt.sh_type == SHT_RELA;
791 
792 	if (lazy_plt) {
793 		/*
794 		 * Assume a .plt with the same number of entries as the number
795 		 * of relocation entries is not lazy and does not have a header.
796 		 */
797 		if (ri.nr_entries * plt_entry_size == shdr_plt.sh_size)
798 			dso__delete_symbol(dso, plt_sym);
799 		else
800 			plt_offset += plt_header_size;
801 	}
802 
803 	/*
804 	 * x86 doesn't insert IFUNC relocations in .plt order, so sort to get
805 	 * back in order.
806 	 */
807 	if (machine_is_x86(ehdr.e_machine) && sort_rel(&ri))
808 		goto out_elf_end;
809 
810 	for (idx = 0; idx < ri.nr_entries; idx++) {
811 		const char *elf_name = NULL;
812 		char *demangled = NULL;
813 
814 		gelf_getsym(syms, get_rel_symidx(&ri, idx), &sym);
815 
816 		elf_name = elf_sym__name(&sym, symstrs);
817 		demangled = demangle_sym(dso, 0, elf_name);
818 		if (demangled)
819 			elf_name = demangled;
820 		if (*elf_name)
821 			snprintf(sympltname, sizeof(sympltname), "%s@plt", elf_name);
822 		else if (!get_ifunc_name(elf, dso, &ehdr, &ri, sympltname, sizeof(sympltname)))
823 			snprintf(sympltname, sizeof(sympltname),
824 				 "offset_%#" PRIx64 "@plt", plt_offset);
825 		free(demangled);
826 
827 		f = symbol__new(plt_offset, plt_entry_size, STB_GLOBAL, STT_FUNC, sympltname);
828 		if (!f)
829 			goto out_elf_end;
830 
831 		plt_offset += plt_entry_size;
832 		symbols__insert(&dso->symbols, f);
833 		++nr;
834 	}
835 
836 	err = 0;
837 out_elf_end:
838 	exit_rel(&ri);
839 	if (err == 0)
840 		return nr;
841 	pr_debug("%s: problems reading %s PLT info.\n",
842 		 __func__, dso->long_name);
843 	return 0;
844 }
845 
dso__demangle_sym(struct dso * dso,int kmodule,const char * elf_name)846 char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
847 {
848 	return demangle_sym(dso, kmodule, elf_name);
849 }
850 
851 /*
852  * Align offset to 4 bytes as needed for note name and descriptor data.
853  */
854 #define NOTE_ALIGN(n) (((n) + 3) & -4U)
855 
elf_read_build_id(Elf * elf,void * bf,size_t size)856 static int elf_read_build_id(Elf *elf, void *bf, size_t size)
857 {
858 	int err = -1;
859 	GElf_Ehdr ehdr;
860 	GElf_Shdr shdr;
861 	Elf_Data *data;
862 	Elf_Scn *sec;
863 	Elf_Kind ek;
864 	void *ptr;
865 
866 	if (size < BUILD_ID_SIZE)
867 		goto out;
868 
869 	ek = elf_kind(elf);
870 	if (ek != ELF_K_ELF)
871 		goto out;
872 
873 	if (gelf_getehdr(elf, &ehdr) == NULL) {
874 		pr_err("%s: cannot get elf header.\n", __func__);
875 		goto out;
876 	}
877 
878 	/*
879 	 * Check following sections for notes:
880 	 *   '.note.gnu.build-id'
881 	 *   '.notes'
882 	 *   '.note' (VDSO specific)
883 	 */
884 	do {
885 		sec = elf_section_by_name(elf, &ehdr, &shdr,
886 					  ".note.gnu.build-id", NULL);
887 		if (sec)
888 			break;
889 
890 		sec = elf_section_by_name(elf, &ehdr, &shdr,
891 					  ".notes", NULL);
892 		if (sec)
893 			break;
894 
895 		sec = elf_section_by_name(elf, &ehdr, &shdr,
896 					  ".note", NULL);
897 		if (sec)
898 			break;
899 
900 		return err;
901 
902 	} while (0);
903 
904 	data = elf_getdata(sec, NULL);
905 	if (data == NULL)
906 		goto out;
907 
908 	ptr = data->d_buf;
909 	while (ptr < (data->d_buf + data->d_size)) {
910 		GElf_Nhdr *nhdr = ptr;
911 		size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
912 		       descsz = NOTE_ALIGN(nhdr->n_descsz);
913 		const char *name;
914 
915 		ptr += sizeof(*nhdr);
916 		name = ptr;
917 		ptr += namesz;
918 		if (nhdr->n_type == NT_GNU_BUILD_ID &&
919 		    nhdr->n_namesz == sizeof("GNU")) {
920 			if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
921 				size_t sz = min(size, descsz);
922 				memcpy(bf, ptr, sz);
923 				memset(bf + sz, 0, size - sz);
924 				err = sz;
925 				break;
926 			}
927 		}
928 		ptr += descsz;
929 	}
930 
931 out:
932 	return err;
933 }
934 
935 #ifdef HAVE_LIBBFD_BUILDID_SUPPORT
936 
read_build_id(const char * filename,struct build_id * bid)937 static int read_build_id(const char *filename, struct build_id *bid)
938 {
939 	size_t size = sizeof(bid->data);
940 	int err = -1;
941 	bfd *abfd;
942 
943 	abfd = bfd_openr(filename, NULL);
944 	if (!abfd)
945 		return -1;
946 
947 	if (!bfd_check_format(abfd, bfd_object)) {
948 		pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
949 		goto out_close;
950 	}
951 
952 	if (!abfd->build_id || abfd->build_id->size > size)
953 		goto out_close;
954 
955 	memcpy(bid->data, abfd->build_id->data, abfd->build_id->size);
956 	memset(bid->data + abfd->build_id->size, 0, size - abfd->build_id->size);
957 	err = bid->size = abfd->build_id->size;
958 
959 out_close:
960 	bfd_close(abfd);
961 	return err;
962 }
963 
964 #else // HAVE_LIBBFD_BUILDID_SUPPORT
965 
read_build_id(const char * filename,struct build_id * bid)966 static int read_build_id(const char *filename, struct build_id *bid)
967 {
968 	size_t size = sizeof(bid->data);
969 	int fd, err = -1;
970 	Elf *elf;
971 
972 	if (size < BUILD_ID_SIZE)
973 		goto out;
974 
975 	fd = open(filename, O_RDONLY);
976 	if (fd < 0)
977 		goto out;
978 
979 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
980 	if (elf == NULL) {
981 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
982 		goto out_close;
983 	}
984 
985 	err = elf_read_build_id(elf, bid->data, size);
986 	if (err > 0)
987 		bid->size = err;
988 
989 	elf_end(elf);
990 out_close:
991 	close(fd);
992 out:
993 	return err;
994 }
995 
996 #endif // HAVE_LIBBFD_BUILDID_SUPPORT
997 
filename__read_build_id(const char * filename,struct build_id * bid)998 int filename__read_build_id(const char *filename, struct build_id *bid)
999 {
1000 	struct kmod_path m = { .name = NULL, };
1001 	char path[PATH_MAX];
1002 	int err;
1003 
1004 	if (!filename)
1005 		return -EFAULT;
1006 
1007 	err = kmod_path__parse(&m, filename);
1008 	if (err)
1009 		return -1;
1010 
1011 	if (m.comp) {
1012 		int error = 0, fd;
1013 
1014 		fd = filename__decompress(filename, path, sizeof(path), m.comp, &error);
1015 		if (fd < 0) {
1016 			pr_debug("Failed to decompress (error %d) %s\n",
1017 				 error, filename);
1018 			return -1;
1019 		}
1020 		close(fd);
1021 		filename = path;
1022 	}
1023 
1024 	err = read_build_id(filename, bid);
1025 
1026 	if (m.comp)
1027 		unlink(filename);
1028 	return err;
1029 }
1030 
sysfs__read_build_id(const char * filename,struct build_id * bid)1031 int sysfs__read_build_id(const char *filename, struct build_id *bid)
1032 {
1033 	size_t size = sizeof(bid->data);
1034 	int fd, err = -1;
1035 
1036 	fd = open(filename, O_RDONLY);
1037 	if (fd < 0)
1038 		goto out;
1039 
1040 	while (1) {
1041 		char bf[BUFSIZ];
1042 		GElf_Nhdr nhdr;
1043 		size_t namesz, descsz;
1044 
1045 		if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
1046 			break;
1047 
1048 		namesz = NOTE_ALIGN(nhdr.n_namesz);
1049 		descsz = NOTE_ALIGN(nhdr.n_descsz);
1050 		if (nhdr.n_type == NT_GNU_BUILD_ID &&
1051 		    nhdr.n_namesz == sizeof("GNU")) {
1052 			if (read(fd, bf, namesz) != (ssize_t)namesz)
1053 				break;
1054 			if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
1055 				size_t sz = min(descsz, size);
1056 				if (read(fd, bid->data, sz) == (ssize_t)sz) {
1057 					memset(bid->data + sz, 0, size - sz);
1058 					bid->size = sz;
1059 					err = 0;
1060 					break;
1061 				}
1062 			} else if (read(fd, bf, descsz) != (ssize_t)descsz)
1063 				break;
1064 		} else {
1065 			int n = namesz + descsz;
1066 
1067 			if (n > (int)sizeof(bf)) {
1068 				n = sizeof(bf);
1069 				pr_debug("%s: truncating reading of build id in sysfs file %s: n_namesz=%u, n_descsz=%u.\n",
1070 					 __func__, filename, nhdr.n_namesz, nhdr.n_descsz);
1071 			}
1072 			if (read(fd, bf, n) != n)
1073 				break;
1074 		}
1075 	}
1076 	close(fd);
1077 out:
1078 	return err;
1079 }
1080 
1081 #ifdef HAVE_LIBBFD_SUPPORT
1082 
filename__read_debuglink(const char * filename,char * debuglink,size_t size)1083 int filename__read_debuglink(const char *filename, char *debuglink,
1084 			     size_t size)
1085 {
1086 	int err = -1;
1087 	asection *section;
1088 	bfd *abfd;
1089 
1090 	abfd = bfd_openr(filename, NULL);
1091 	if (!abfd)
1092 		return -1;
1093 
1094 	if (!bfd_check_format(abfd, bfd_object)) {
1095 		pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
1096 		goto out_close;
1097 	}
1098 
1099 	section = bfd_get_section_by_name(abfd, ".gnu_debuglink");
1100 	if (!section)
1101 		goto out_close;
1102 
1103 	if (section->size > size)
1104 		goto out_close;
1105 
1106 	if (!bfd_get_section_contents(abfd, section, debuglink, 0,
1107 				      section->size))
1108 		goto out_close;
1109 
1110 	err = 0;
1111 
1112 out_close:
1113 	bfd_close(abfd);
1114 	return err;
1115 }
1116 
1117 #else
1118 
filename__read_debuglink(const char * filename,char * debuglink,size_t size)1119 int filename__read_debuglink(const char *filename, char *debuglink,
1120 			     size_t size)
1121 {
1122 	int fd, err = -1;
1123 	Elf *elf;
1124 	GElf_Ehdr ehdr;
1125 	GElf_Shdr shdr;
1126 	Elf_Data *data;
1127 	Elf_Scn *sec;
1128 	Elf_Kind ek;
1129 
1130 	fd = open(filename, O_RDONLY);
1131 	if (fd < 0)
1132 		goto out;
1133 
1134 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1135 	if (elf == NULL) {
1136 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
1137 		goto out_close;
1138 	}
1139 
1140 	ek = elf_kind(elf);
1141 	if (ek != ELF_K_ELF)
1142 		goto out_elf_end;
1143 
1144 	if (gelf_getehdr(elf, &ehdr) == NULL) {
1145 		pr_err("%s: cannot get elf header.\n", __func__);
1146 		goto out_elf_end;
1147 	}
1148 
1149 	sec = elf_section_by_name(elf, &ehdr, &shdr,
1150 				  ".gnu_debuglink", NULL);
1151 	if (sec == NULL)
1152 		goto out_elf_end;
1153 
1154 	data = elf_getdata(sec, NULL);
1155 	if (data == NULL)
1156 		goto out_elf_end;
1157 
1158 	/* the start of this section is a zero-terminated string */
1159 	strncpy(debuglink, data->d_buf, size);
1160 
1161 	err = 0;
1162 
1163 out_elf_end:
1164 	elf_end(elf);
1165 out_close:
1166 	close(fd);
1167 out:
1168 	return err;
1169 }
1170 
1171 #endif
1172 
dso__swap_init(struct dso * dso,unsigned char eidata)1173 static int dso__swap_init(struct dso *dso, unsigned char eidata)
1174 {
1175 	static unsigned int const endian = 1;
1176 
1177 	dso->needs_swap = DSO_SWAP__NO;
1178 
1179 	switch (eidata) {
1180 	case ELFDATA2LSB:
1181 		/* We are big endian, DSO is little endian. */
1182 		if (*(unsigned char const *)&endian != 1)
1183 			dso->needs_swap = DSO_SWAP__YES;
1184 		break;
1185 
1186 	case ELFDATA2MSB:
1187 		/* We are little endian, DSO is big endian. */
1188 		if (*(unsigned char const *)&endian != 0)
1189 			dso->needs_swap = DSO_SWAP__YES;
1190 		break;
1191 
1192 	default:
1193 		pr_err("unrecognized DSO data encoding %d\n", eidata);
1194 		return -EINVAL;
1195 	}
1196 
1197 	return 0;
1198 }
1199 
symsrc__possibly_runtime(struct symsrc * ss)1200 bool symsrc__possibly_runtime(struct symsrc *ss)
1201 {
1202 	return ss->dynsym || ss->opdsec;
1203 }
1204 
symsrc__has_symtab(struct symsrc * ss)1205 bool symsrc__has_symtab(struct symsrc *ss)
1206 {
1207 	return ss->symtab != NULL;
1208 }
1209 
symsrc__destroy(struct symsrc * ss)1210 void symsrc__destroy(struct symsrc *ss)
1211 {
1212 	zfree(&ss->name);
1213 	elf_end(ss->elf);
1214 	close(ss->fd);
1215 }
1216 
elf__needs_adjust_symbols(GElf_Ehdr ehdr)1217 bool elf__needs_adjust_symbols(GElf_Ehdr ehdr)
1218 {
1219 	/*
1220 	 * Usually vmlinux is an ELF file with type ET_EXEC for most
1221 	 * architectures; except Arm64 kernel is linked with option
1222 	 * '-share', so need to check type ET_DYN.
1223 	 */
1224 	return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL ||
1225 	       ehdr.e_type == ET_DYN;
1226 }
1227 
symsrc__init(struct symsrc * ss,struct dso * dso,const char * name,enum dso_binary_type type)1228 int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
1229 		 enum dso_binary_type type)
1230 {
1231 	GElf_Ehdr ehdr;
1232 	Elf *elf;
1233 	int fd;
1234 
1235 	if (dso__needs_decompress(dso)) {
1236 		fd = dso__decompress_kmodule_fd(dso, name);
1237 		if (fd < 0)
1238 			return -1;
1239 
1240 		type = dso->symtab_type;
1241 	} else {
1242 		fd = open(name, O_RDONLY);
1243 		if (fd < 0) {
1244 			dso->load_errno = errno;
1245 			return -1;
1246 		}
1247 	}
1248 
1249 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1250 	if (elf == NULL) {
1251 		pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
1252 		dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
1253 		goto out_close;
1254 	}
1255 
1256 	if (gelf_getehdr(elf, &ehdr) == NULL) {
1257 		dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
1258 		pr_debug("%s: cannot get elf header.\n", __func__);
1259 		goto out_elf_end;
1260 	}
1261 
1262 	if (dso__swap_init(dso, ehdr.e_ident[EI_DATA])) {
1263 		dso->load_errno = DSO_LOAD_ERRNO__INTERNAL_ERROR;
1264 		goto out_elf_end;
1265 	}
1266 
1267 	/* Always reject images with a mismatched build-id: */
1268 	if (dso->has_build_id && !symbol_conf.ignore_vmlinux_buildid) {
1269 		u8 build_id[BUILD_ID_SIZE];
1270 		struct build_id bid;
1271 		int size;
1272 
1273 		size = elf_read_build_id(elf, build_id, BUILD_ID_SIZE);
1274 		if (size <= 0) {
1275 			dso->load_errno = DSO_LOAD_ERRNO__CANNOT_READ_BUILDID;
1276 			goto out_elf_end;
1277 		}
1278 
1279 		build_id__init(&bid, build_id, size);
1280 		if (!dso__build_id_equal(dso, &bid)) {
1281 			pr_debug("%s: build id mismatch for %s.\n", __func__, name);
1282 			dso->load_errno = DSO_LOAD_ERRNO__MISMATCHING_BUILDID;
1283 			goto out_elf_end;
1284 		}
1285 	}
1286 
1287 	ss->is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1288 
1289 	ss->symtab_idx = 0;
1290 	ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab",
1291 			&ss->symtab_idx);
1292 	if (ss->symshdr.sh_type != SHT_SYMTAB)
1293 		ss->symtab = NULL;
1294 
1295 	ss->dynsym_idx = 0;
1296 	ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym",
1297 			&ss->dynsym_idx);
1298 	if (ss->dynshdr.sh_type != SHT_DYNSYM)
1299 		ss->dynsym = NULL;
1300 
1301 	ss->opdidx = 0;
1302 	ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd",
1303 			&ss->opdidx);
1304 	if (ss->opdshdr.sh_type != SHT_PROGBITS)
1305 		ss->opdsec = NULL;
1306 
1307 	if (dso->kernel == DSO_SPACE__USER)
1308 		ss->adjust_symbols = true;
1309 	else
1310 		ss->adjust_symbols = elf__needs_adjust_symbols(ehdr);
1311 
1312 	ss->name   = strdup(name);
1313 	if (!ss->name) {
1314 		dso->load_errno = errno;
1315 		goto out_elf_end;
1316 	}
1317 
1318 	ss->elf    = elf;
1319 	ss->fd     = fd;
1320 	ss->ehdr   = ehdr;
1321 	ss->type   = type;
1322 
1323 	return 0;
1324 
1325 out_elf_end:
1326 	elf_end(elf);
1327 out_close:
1328 	close(fd);
1329 	return -1;
1330 }
1331 
1332 /**
1333  * ref_reloc_sym_not_found - has kernel relocation symbol been found.
1334  * @kmap: kernel maps and relocation reference symbol
1335  *
1336  * This function returns %true if we are dealing with the kernel maps and the
1337  * relocation reference symbol has not yet been found.  Otherwise %false is
1338  * returned.
1339  */
ref_reloc_sym_not_found(struct kmap * kmap)1340 static bool ref_reloc_sym_not_found(struct kmap *kmap)
1341 {
1342 	return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
1343 	       !kmap->ref_reloc_sym->unrelocated_addr;
1344 }
1345 
1346 /**
1347  * ref_reloc - kernel relocation offset.
1348  * @kmap: kernel maps and relocation reference symbol
1349  *
1350  * This function returns the offset of kernel addresses as determined by using
1351  * the relocation reference symbol i.e. if the kernel has not been relocated
1352  * then the return value is zero.
1353  */
ref_reloc(struct kmap * kmap)1354 static u64 ref_reloc(struct kmap *kmap)
1355 {
1356 	if (kmap && kmap->ref_reloc_sym &&
1357 	    kmap->ref_reloc_sym->unrelocated_addr)
1358 		return kmap->ref_reloc_sym->addr -
1359 		       kmap->ref_reloc_sym->unrelocated_addr;
1360 	return 0;
1361 }
1362 
arch__sym_update(struct symbol * s __maybe_unused,GElf_Sym * sym __maybe_unused)1363 void __weak arch__sym_update(struct symbol *s __maybe_unused,
1364 		GElf_Sym *sym __maybe_unused) { }
1365 
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)1366 static int dso__process_kernel_symbol(struct dso *dso, struct map *map,
1367 				      GElf_Sym *sym, GElf_Shdr *shdr,
1368 				      struct maps *kmaps, struct kmap *kmap,
1369 				      struct dso **curr_dsop, struct map **curr_mapp,
1370 				      const char *section_name,
1371 				      bool adjust_kernel_syms, bool kmodule, bool *remap_kernel)
1372 {
1373 	struct dso *curr_dso = *curr_dsop;
1374 	struct map *curr_map;
1375 	char dso_name[PATH_MAX];
1376 
1377 	/* Adjust symbol to map to file offset */
1378 	if (adjust_kernel_syms)
1379 		sym->st_value -= shdr->sh_addr - shdr->sh_offset;
1380 
1381 	if (strcmp(section_name, (curr_dso->short_name + dso->short_name_len)) == 0)
1382 		return 0;
1383 
1384 	if (strcmp(section_name, ".text") == 0) {
1385 		/*
1386 		 * The initial kernel mapping is based on
1387 		 * kallsyms and identity maps.  Overwrite it to
1388 		 * map to the kernel dso.
1389 		 */
1390 		if (*remap_kernel && dso->kernel && !kmodule) {
1391 			*remap_kernel = false;
1392 			map__set_start(map, shdr->sh_addr + ref_reloc(kmap));
1393 			map__set_end(map, map__start(map) + shdr->sh_size);
1394 			map__set_pgoff(map, shdr->sh_offset);
1395 			map__set_map_ip(map, map__dso_map_ip);
1396 			map__set_unmap_ip(map, map__dso_unmap_ip);
1397 			/* Ensure maps are correctly ordered */
1398 			if (kmaps) {
1399 				int err;
1400 				struct map *tmp = map__get(map);
1401 
1402 				maps__remove(kmaps, map);
1403 				err = maps__insert(kmaps, map);
1404 				map__put(tmp);
1405 				if (err)
1406 					return err;
1407 			}
1408 		}
1409 
1410 		/*
1411 		 * The initial module mapping is based on
1412 		 * /proc/modules mapped to offset zero.
1413 		 * Overwrite it to map to the module dso.
1414 		 */
1415 		if (*remap_kernel && kmodule) {
1416 			*remap_kernel = false;
1417 			map__set_pgoff(map, shdr->sh_offset);
1418 		}
1419 
1420 		*curr_mapp = map;
1421 		*curr_dsop = dso;
1422 		return 0;
1423 	}
1424 
1425 	if (!kmap)
1426 		return 0;
1427 
1428 	snprintf(dso_name, sizeof(dso_name), "%s%s", dso->short_name, section_name);
1429 
1430 	curr_map = maps__find_by_name(kmaps, dso_name);
1431 	if (curr_map == NULL) {
1432 		u64 start = sym->st_value;
1433 
1434 		if (kmodule)
1435 			start += map__start(map) + shdr->sh_offset;
1436 
1437 		curr_dso = dso__new(dso_name);
1438 		if (curr_dso == NULL)
1439 			return -1;
1440 		curr_dso->kernel = dso->kernel;
1441 		curr_dso->long_name = dso->long_name;
1442 		curr_dso->long_name_len = dso->long_name_len;
1443 		curr_dso->binary_type = dso->binary_type;
1444 		curr_dso->adjust_symbols = dso->adjust_symbols;
1445 		curr_map = map__new2(start, curr_dso);
1446 		dso__put(curr_dso);
1447 		if (curr_map == NULL)
1448 			return -1;
1449 
1450 		if (curr_dso->kernel)
1451 			map__kmap(curr_map)->kmaps = kmaps;
1452 
1453 		if (adjust_kernel_syms) {
1454 			map__set_start(curr_map, shdr->sh_addr + ref_reloc(kmap));
1455 			map__set_end(curr_map, map__start(curr_map) + shdr->sh_size);
1456 			map__set_pgoff(curr_map, shdr->sh_offset);
1457 		} else {
1458 			map__set_map_ip(curr_map, identity__map_ip);
1459 			map__set_unmap_ip(curr_map, identity__map_ip);
1460 		}
1461 		curr_dso->symtab_type = dso->symtab_type;
1462 		if (maps__insert(kmaps, curr_map))
1463 			return -1;
1464 		/*
1465 		 * Add it before we drop the reference to curr_map, i.e. while
1466 		 * we still are sure to have a reference to this DSO via
1467 		 * *curr_map->dso.
1468 		 */
1469 		dsos__add(&maps__machine(kmaps)->dsos, curr_dso);
1470 		/* kmaps already got it */
1471 		map__put(curr_map);
1472 		dso__set_loaded(curr_dso);
1473 		*curr_mapp = curr_map;
1474 		*curr_dsop = curr_dso;
1475 	} else
1476 		*curr_dsop = map__dso(curr_map);
1477 
1478 	return 0;
1479 }
1480 
1481 static int
dso__load_sym_internal(struct dso * dso,struct map * map,struct symsrc * syms_ss,struct symsrc * runtime_ss,int kmodule,int dynsym)1482 dso__load_sym_internal(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1483 		       struct symsrc *runtime_ss, int kmodule, int dynsym)
1484 {
1485 	struct kmap *kmap = dso->kernel ? map__kmap(map) : NULL;
1486 	struct maps *kmaps = kmap ? map__kmaps(map) : NULL;
1487 	struct map *curr_map = map;
1488 	struct dso *curr_dso = dso;
1489 	Elf_Data *symstrs, *secstrs, *secstrs_run, *secstrs_sym;
1490 	uint32_t nr_syms;
1491 	int err = -1;
1492 	uint32_t idx;
1493 	GElf_Ehdr ehdr;
1494 	GElf_Shdr shdr;
1495 	GElf_Shdr tshdr;
1496 	Elf_Data *syms, *opddata = NULL;
1497 	GElf_Sym sym;
1498 	Elf_Scn *sec, *sec_strndx;
1499 	Elf *elf;
1500 	int nr = 0;
1501 	bool remap_kernel = false, adjust_kernel_syms = false;
1502 
1503 	if (kmap && !kmaps)
1504 		return -1;
1505 
1506 	elf = syms_ss->elf;
1507 	ehdr = syms_ss->ehdr;
1508 	if (dynsym) {
1509 		sec  = syms_ss->dynsym;
1510 		shdr = syms_ss->dynshdr;
1511 	} else {
1512 		sec =  syms_ss->symtab;
1513 		shdr = syms_ss->symshdr;
1514 	}
1515 
1516 	if (elf_section_by_name(runtime_ss->elf, &runtime_ss->ehdr, &tshdr,
1517 				".text", NULL))
1518 		dso->text_offset = tshdr.sh_addr - tshdr.sh_offset;
1519 
1520 	if (runtime_ss->opdsec)
1521 		opddata = elf_rawdata(runtime_ss->opdsec, NULL);
1522 
1523 	syms = elf_getdata(sec, NULL);
1524 	if (syms == NULL)
1525 		goto out_elf_end;
1526 
1527 	sec = elf_getscn(elf, shdr.sh_link);
1528 	if (sec == NULL)
1529 		goto out_elf_end;
1530 
1531 	symstrs = elf_getdata(sec, NULL);
1532 	if (symstrs == NULL)
1533 		goto out_elf_end;
1534 
1535 	sec_strndx = elf_getscn(runtime_ss->elf, runtime_ss->ehdr.e_shstrndx);
1536 	if (sec_strndx == NULL)
1537 		goto out_elf_end;
1538 
1539 	secstrs_run = elf_getdata(sec_strndx, NULL);
1540 	if (secstrs_run == NULL)
1541 		goto out_elf_end;
1542 
1543 	sec_strndx = elf_getscn(elf, ehdr.e_shstrndx);
1544 	if (sec_strndx == NULL)
1545 		goto out_elf_end;
1546 
1547 	secstrs_sym = elf_getdata(sec_strndx, NULL);
1548 	if (secstrs_sym == NULL)
1549 		goto out_elf_end;
1550 
1551 	nr_syms = shdr.sh_size / shdr.sh_entsize;
1552 
1553 	memset(&sym, 0, sizeof(sym));
1554 
1555 	/*
1556 	 * The kernel relocation symbol is needed in advance in order to adjust
1557 	 * kernel maps correctly.
1558 	 */
1559 	if (ref_reloc_sym_not_found(kmap)) {
1560 		elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1561 			const char *elf_name = elf_sym__name(&sym, symstrs);
1562 
1563 			if (strcmp(elf_name, kmap->ref_reloc_sym->name))
1564 				continue;
1565 			kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
1566 			map__set_reloc(map, kmap->ref_reloc_sym->addr - kmap->ref_reloc_sym->unrelocated_addr);
1567 			break;
1568 		}
1569 	}
1570 
1571 	/*
1572 	 * Handle any relocation of vdso necessary because older kernels
1573 	 * attempted to prelink vdso to its virtual address.
1574 	 */
1575 	if (dso__is_vdso(dso))
1576 		map__set_reloc(map, map__start(map) - dso->text_offset);
1577 
1578 	dso->adjust_symbols = runtime_ss->adjust_symbols || ref_reloc(kmap);
1579 	/*
1580 	 * Initial kernel and module mappings do not map to the dso.
1581 	 * Flag the fixups.
1582 	 */
1583 	if (dso->kernel) {
1584 		remap_kernel = true;
1585 		adjust_kernel_syms = dso->adjust_symbols;
1586 	}
1587 	elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1588 		struct symbol *f;
1589 		const char *elf_name = elf_sym__name(&sym, symstrs);
1590 		char *demangled = NULL;
1591 		int is_label = elf_sym__is_label(&sym);
1592 		const char *section_name;
1593 		bool used_opd = false;
1594 
1595 		if (!is_label && !elf_sym__filter(&sym))
1596 			continue;
1597 
1598 		/* Reject ARM ELF "mapping symbols": these aren't unique and
1599 		 * don't identify functions, so will confuse the profile
1600 		 * output: */
1601 		if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) {
1602 			if (elf_name[0] == '$' && strchr("adtx", elf_name[1])
1603 			    && (elf_name[2] == '\0' || elf_name[2] == '.'))
1604 				continue;
1605 		}
1606 
1607 		if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) {
1608 			u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr;
1609 			u64 *opd = opddata->d_buf + offset;
1610 			sym.st_value = DSO__SWAP(dso, u64, *opd);
1611 			sym.st_shndx = elf_addr_to_index(runtime_ss->elf,
1612 					sym.st_value);
1613 			used_opd = true;
1614 		}
1615 
1616 		/*
1617 		 * When loading symbols in a data mapping, ABS symbols (which
1618 		 * has a value of SHN_ABS in its st_shndx) failed at
1619 		 * elf_getscn().  And it marks the loading as a failure so
1620 		 * already loaded symbols cannot be fixed up.
1621 		 *
1622 		 * I'm not sure what should be done. Just ignore them for now.
1623 		 * - Namhyung Kim
1624 		 */
1625 		if (sym.st_shndx == SHN_ABS)
1626 			continue;
1627 
1628 		sec = elf_getscn(syms_ss->elf, sym.st_shndx);
1629 		if (!sec)
1630 			goto out_elf_end;
1631 
1632 		gelf_getshdr(sec, &shdr);
1633 
1634 		/*
1635 		 * If the attribute bit SHF_ALLOC is not set, the section
1636 		 * doesn't occupy memory during process execution.
1637 		 * E.g. ".gnu.warning.*" section is used by linker to generate
1638 		 * warnings when calling deprecated functions, the symbols in
1639 		 * the section aren't loaded to memory during process execution,
1640 		 * so skip them.
1641 		 */
1642 		if (!(shdr.sh_flags & SHF_ALLOC))
1643 			continue;
1644 
1645 		secstrs = secstrs_sym;
1646 
1647 		/*
1648 		 * We have to fallback to runtime when syms' section header has
1649 		 * NOBITS set. NOBITS results in file offset (sh_offset) not
1650 		 * being incremented. So sh_offset used below has different
1651 		 * values for syms (invalid) and runtime (valid).
1652 		 */
1653 		if (shdr.sh_type == SHT_NOBITS) {
1654 			sec = elf_getscn(runtime_ss->elf, sym.st_shndx);
1655 			if (!sec)
1656 				goto out_elf_end;
1657 
1658 			gelf_getshdr(sec, &shdr);
1659 			secstrs = secstrs_run;
1660 		}
1661 
1662 		if (is_label && !elf_sec__filter(&shdr, secstrs))
1663 			continue;
1664 
1665 		section_name = elf_sec__name(&shdr, secstrs);
1666 
1667 		/* On ARM, symbols for thumb functions have 1 added to
1668 		 * the symbol address as a flag - remove it */
1669 		if ((ehdr.e_machine == EM_ARM) &&
1670 		    (GELF_ST_TYPE(sym.st_info) == STT_FUNC) &&
1671 		    (sym.st_value & 1))
1672 			--sym.st_value;
1673 
1674 		if (dso->kernel) {
1675 			if (dso__process_kernel_symbol(dso, map, &sym, &shdr, kmaps, kmap, &curr_dso, &curr_map,
1676 						       section_name, adjust_kernel_syms, kmodule, &remap_kernel))
1677 				goto out_elf_end;
1678 		} else if ((used_opd && runtime_ss->adjust_symbols) ||
1679 			   (!used_opd && syms_ss->adjust_symbols)) {
1680 			GElf_Phdr phdr;
1681 
1682 			if (elf_read_program_header(runtime_ss->elf,
1683 						    (u64)sym.st_value, &phdr)) {
1684 				pr_debug4("%s: failed to find program header for "
1685 					   "symbol: %s st_value: %#" PRIx64 "\n",
1686 					   __func__, elf_name, (u64)sym.st_value);
1687 				pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1688 					"sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n",
1689 					__func__, (u64)sym.st_value, (u64)shdr.sh_addr,
1690 					(u64)shdr.sh_offset);
1691 				/*
1692 				 * Fail to find program header, let's rollback
1693 				 * to use shdr.sh_addr and shdr.sh_offset to
1694 				 * calibrate symbol's file address, though this
1695 				 * is not necessary for normal C ELF file, we
1696 				 * still need to handle java JIT symbols in this
1697 				 * case.
1698 				 */
1699 				sym.st_value -= shdr.sh_addr - shdr.sh_offset;
1700 			} else {
1701 				pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1702 					"p_vaddr: %#" PRIx64 " p_offset: %#" PRIx64 "\n",
1703 					__func__, (u64)sym.st_value, (u64)phdr.p_vaddr,
1704 					(u64)phdr.p_offset);
1705 				sym.st_value -= phdr.p_vaddr - phdr.p_offset;
1706 			}
1707 		}
1708 
1709 		demangled = demangle_sym(dso, kmodule, elf_name);
1710 		if (demangled != NULL)
1711 			elf_name = demangled;
1712 
1713 		f = symbol__new(sym.st_value, sym.st_size,
1714 				GELF_ST_BIND(sym.st_info),
1715 				GELF_ST_TYPE(sym.st_info), elf_name);
1716 		free(demangled);
1717 		if (!f)
1718 			goto out_elf_end;
1719 
1720 		arch__sym_update(f, &sym);
1721 
1722 		__symbols__insert(&curr_dso->symbols, f, dso->kernel);
1723 		nr++;
1724 	}
1725 
1726 	/*
1727 	 * For misannotated, zeroed, ASM function sizes.
1728 	 */
1729 	if (nr > 0) {
1730 		symbols__fixup_end(&dso->symbols, false);
1731 		symbols__fixup_duplicate(&dso->symbols);
1732 		if (kmap) {
1733 			/*
1734 			 * We need to fixup this here too because we create new
1735 			 * maps here, for things like vsyscall sections.
1736 			 */
1737 			maps__fixup_end(kmaps);
1738 		}
1739 	}
1740 	err = nr;
1741 out_elf_end:
1742 	return err;
1743 }
1744 
dso__load_sym(struct dso * dso,struct map * map,struct symsrc * syms_ss,struct symsrc * runtime_ss,int kmodule)1745 int dso__load_sym(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1746 		  struct symsrc *runtime_ss, int kmodule)
1747 {
1748 	int nr = 0;
1749 	int err = -1;
1750 
1751 	dso->symtab_type = syms_ss->type;
1752 	dso->is_64_bit = syms_ss->is_64_bit;
1753 	dso->rel = syms_ss->ehdr.e_type == ET_REL;
1754 
1755 	/*
1756 	 * Modules may already have symbols from kallsyms, but those symbols
1757 	 * have the wrong values for the dso maps, so remove them.
1758 	 */
1759 	if (kmodule && syms_ss->symtab)
1760 		symbols__delete(&dso->symbols);
1761 
1762 	if (!syms_ss->symtab) {
1763 		/*
1764 		 * If the vmlinux is stripped, fail so we will fall back
1765 		 * to using kallsyms. The vmlinux runtime symbols aren't
1766 		 * of much use.
1767 		 */
1768 		if (dso->kernel)
1769 			return err;
1770 	} else  {
1771 		err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss,
1772 					     kmodule, 0);
1773 		if (err < 0)
1774 			return err;
1775 		nr = err;
1776 	}
1777 
1778 	if (syms_ss->dynsym) {
1779 		err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss,
1780 					     kmodule, 1);
1781 		if (err < 0)
1782 			return err;
1783 		err += nr;
1784 	}
1785 
1786 	return err;
1787 }
1788 
elf_read_maps(Elf * elf,bool exe,mapfn_t mapfn,void * data)1789 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
1790 {
1791 	GElf_Phdr phdr;
1792 	size_t i, phdrnum;
1793 	int err;
1794 	u64 sz;
1795 
1796 	if (elf_getphdrnum(elf, &phdrnum))
1797 		return -1;
1798 
1799 	for (i = 0; i < phdrnum; i++) {
1800 		if (gelf_getphdr(elf, i, &phdr) == NULL)
1801 			return -1;
1802 		if (phdr.p_type != PT_LOAD)
1803 			continue;
1804 		if (exe) {
1805 			if (!(phdr.p_flags & PF_X))
1806 				continue;
1807 		} else {
1808 			if (!(phdr.p_flags & PF_R))
1809 				continue;
1810 		}
1811 		sz = min(phdr.p_memsz, phdr.p_filesz);
1812 		if (!sz)
1813 			continue;
1814 		err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
1815 		if (err)
1816 			return err;
1817 	}
1818 	return 0;
1819 }
1820 
file__read_maps(int fd,bool exe,mapfn_t mapfn,void * data,bool * is_64_bit)1821 int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data,
1822 		    bool *is_64_bit)
1823 {
1824 	int err;
1825 	Elf *elf;
1826 
1827 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1828 	if (elf == NULL)
1829 		return -1;
1830 
1831 	if (is_64_bit)
1832 		*is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1833 
1834 	err = elf_read_maps(elf, exe, mapfn, data);
1835 
1836 	elf_end(elf);
1837 	return err;
1838 }
1839 
dso__type_fd(int fd)1840 enum dso_type dso__type_fd(int fd)
1841 {
1842 	enum dso_type dso_type = DSO__TYPE_UNKNOWN;
1843 	GElf_Ehdr ehdr;
1844 	Elf_Kind ek;
1845 	Elf *elf;
1846 
1847 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1848 	if (elf == NULL)
1849 		goto out;
1850 
1851 	ek = elf_kind(elf);
1852 	if (ek != ELF_K_ELF)
1853 		goto out_end;
1854 
1855 	if (gelf_getclass(elf) == ELFCLASS64) {
1856 		dso_type = DSO__TYPE_64BIT;
1857 		goto out_end;
1858 	}
1859 
1860 	if (gelf_getehdr(elf, &ehdr) == NULL)
1861 		goto out_end;
1862 
1863 	if (ehdr.e_machine == EM_X86_64)
1864 		dso_type = DSO__TYPE_X32BIT;
1865 	else
1866 		dso_type = DSO__TYPE_32BIT;
1867 out_end:
1868 	elf_end(elf);
1869 out:
1870 	return dso_type;
1871 }
1872 
copy_bytes(int from,off_t from_offs,int to,off_t to_offs,u64 len)1873 static int copy_bytes(int from, off_t from_offs, int to, off_t to_offs, u64 len)
1874 {
1875 	ssize_t r;
1876 	size_t n;
1877 	int err = -1;
1878 	char *buf = malloc(page_size);
1879 
1880 	if (buf == NULL)
1881 		return -1;
1882 
1883 	if (lseek(to, to_offs, SEEK_SET) != to_offs)
1884 		goto out;
1885 
1886 	if (lseek(from, from_offs, SEEK_SET) != from_offs)
1887 		goto out;
1888 
1889 	while (len) {
1890 		n = page_size;
1891 		if (len < n)
1892 			n = len;
1893 		/* Use read because mmap won't work on proc files */
1894 		r = read(from, buf, n);
1895 		if (r < 0)
1896 			goto out;
1897 		if (!r)
1898 			break;
1899 		n = r;
1900 		r = write(to, buf, n);
1901 		if (r < 0)
1902 			goto out;
1903 		if ((size_t)r != n)
1904 			goto out;
1905 		len -= n;
1906 	}
1907 
1908 	err = 0;
1909 out:
1910 	free(buf);
1911 	return err;
1912 }
1913 
1914 struct kcore {
1915 	int fd;
1916 	int elfclass;
1917 	Elf *elf;
1918 	GElf_Ehdr ehdr;
1919 };
1920 
kcore__open(struct kcore * kcore,const char * filename)1921 static int kcore__open(struct kcore *kcore, const char *filename)
1922 {
1923 	GElf_Ehdr *ehdr;
1924 
1925 	kcore->fd = open(filename, O_RDONLY);
1926 	if (kcore->fd == -1)
1927 		return -1;
1928 
1929 	kcore->elf = elf_begin(kcore->fd, ELF_C_READ, NULL);
1930 	if (!kcore->elf)
1931 		goto out_close;
1932 
1933 	kcore->elfclass = gelf_getclass(kcore->elf);
1934 	if (kcore->elfclass == ELFCLASSNONE)
1935 		goto out_end;
1936 
1937 	ehdr = gelf_getehdr(kcore->elf, &kcore->ehdr);
1938 	if (!ehdr)
1939 		goto out_end;
1940 
1941 	return 0;
1942 
1943 out_end:
1944 	elf_end(kcore->elf);
1945 out_close:
1946 	close(kcore->fd);
1947 	return -1;
1948 }
1949 
kcore__init(struct kcore * kcore,char * filename,int elfclass,bool temp)1950 static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
1951 		       bool temp)
1952 {
1953 	kcore->elfclass = elfclass;
1954 
1955 	if (temp)
1956 		kcore->fd = mkstemp(filename);
1957 	else
1958 		kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
1959 	if (kcore->fd == -1)
1960 		return -1;
1961 
1962 	kcore->elf = elf_begin(kcore->fd, ELF_C_WRITE, NULL);
1963 	if (!kcore->elf)
1964 		goto out_close;
1965 
1966 	if (!gelf_newehdr(kcore->elf, elfclass))
1967 		goto out_end;
1968 
1969 	memset(&kcore->ehdr, 0, sizeof(GElf_Ehdr));
1970 
1971 	return 0;
1972 
1973 out_end:
1974 	elf_end(kcore->elf);
1975 out_close:
1976 	close(kcore->fd);
1977 	unlink(filename);
1978 	return -1;
1979 }
1980 
kcore__close(struct kcore * kcore)1981 static void kcore__close(struct kcore *kcore)
1982 {
1983 	elf_end(kcore->elf);
1984 	close(kcore->fd);
1985 }
1986 
kcore__copy_hdr(struct kcore * from,struct kcore * to,size_t count)1987 static int kcore__copy_hdr(struct kcore *from, struct kcore *to, size_t count)
1988 {
1989 	GElf_Ehdr *ehdr = &to->ehdr;
1990 	GElf_Ehdr *kehdr = &from->ehdr;
1991 
1992 	memcpy(ehdr->e_ident, kehdr->e_ident, EI_NIDENT);
1993 	ehdr->e_type      = kehdr->e_type;
1994 	ehdr->e_machine   = kehdr->e_machine;
1995 	ehdr->e_version   = kehdr->e_version;
1996 	ehdr->e_entry     = 0;
1997 	ehdr->e_shoff     = 0;
1998 	ehdr->e_flags     = kehdr->e_flags;
1999 	ehdr->e_phnum     = count;
2000 	ehdr->e_shentsize = 0;
2001 	ehdr->e_shnum     = 0;
2002 	ehdr->e_shstrndx  = 0;
2003 
2004 	if (from->elfclass == ELFCLASS32) {
2005 		ehdr->e_phoff     = sizeof(Elf32_Ehdr);
2006 		ehdr->e_ehsize    = sizeof(Elf32_Ehdr);
2007 		ehdr->e_phentsize = sizeof(Elf32_Phdr);
2008 	} else {
2009 		ehdr->e_phoff     = sizeof(Elf64_Ehdr);
2010 		ehdr->e_ehsize    = sizeof(Elf64_Ehdr);
2011 		ehdr->e_phentsize = sizeof(Elf64_Phdr);
2012 	}
2013 
2014 	if (!gelf_update_ehdr(to->elf, ehdr))
2015 		return -1;
2016 
2017 	if (!gelf_newphdr(to->elf, count))
2018 		return -1;
2019 
2020 	return 0;
2021 }
2022 
kcore__add_phdr(struct kcore * kcore,int idx,off_t offset,u64 addr,u64 len)2023 static int kcore__add_phdr(struct kcore *kcore, int idx, off_t offset,
2024 			   u64 addr, u64 len)
2025 {
2026 	GElf_Phdr phdr = {
2027 		.p_type		= PT_LOAD,
2028 		.p_flags	= PF_R | PF_W | PF_X,
2029 		.p_offset	= offset,
2030 		.p_vaddr	= addr,
2031 		.p_paddr	= 0,
2032 		.p_filesz	= len,
2033 		.p_memsz	= len,
2034 		.p_align	= page_size,
2035 	};
2036 
2037 	if (!gelf_update_phdr(kcore->elf, idx, &phdr))
2038 		return -1;
2039 
2040 	return 0;
2041 }
2042 
kcore__write(struct kcore * kcore)2043 static off_t kcore__write(struct kcore *kcore)
2044 {
2045 	return elf_update(kcore->elf, ELF_C_WRITE);
2046 }
2047 
2048 struct phdr_data {
2049 	off_t offset;
2050 	off_t rel;
2051 	u64 addr;
2052 	u64 len;
2053 	struct list_head node;
2054 	struct phdr_data *remaps;
2055 };
2056 
2057 struct sym_data {
2058 	u64 addr;
2059 	struct list_head node;
2060 };
2061 
2062 struct kcore_copy_info {
2063 	u64 stext;
2064 	u64 etext;
2065 	u64 first_symbol;
2066 	u64 last_symbol;
2067 	u64 first_module;
2068 	u64 first_module_symbol;
2069 	u64 last_module_symbol;
2070 	size_t phnum;
2071 	struct list_head phdrs;
2072 	struct list_head syms;
2073 };
2074 
2075 #define kcore_copy__for_each_phdr(k, p) \
2076 	list_for_each_entry((p), &(k)->phdrs, node)
2077 
phdr_data__new(u64 addr,u64 len,off_t offset)2078 static struct phdr_data *phdr_data__new(u64 addr, u64 len, off_t offset)
2079 {
2080 	struct phdr_data *p = zalloc(sizeof(*p));
2081 
2082 	if (p) {
2083 		p->addr   = addr;
2084 		p->len    = len;
2085 		p->offset = offset;
2086 	}
2087 
2088 	return p;
2089 }
2090 
kcore_copy_info__addnew(struct kcore_copy_info * kci,u64 addr,u64 len,off_t offset)2091 static struct phdr_data *kcore_copy_info__addnew(struct kcore_copy_info *kci,
2092 						 u64 addr, u64 len,
2093 						 off_t offset)
2094 {
2095 	struct phdr_data *p = phdr_data__new(addr, len, offset);
2096 
2097 	if (p)
2098 		list_add_tail(&p->node, &kci->phdrs);
2099 
2100 	return p;
2101 }
2102 
kcore_copy__free_phdrs(struct kcore_copy_info * kci)2103 static void kcore_copy__free_phdrs(struct kcore_copy_info *kci)
2104 {
2105 	struct phdr_data *p, *tmp;
2106 
2107 	list_for_each_entry_safe(p, tmp, &kci->phdrs, node) {
2108 		list_del_init(&p->node);
2109 		free(p);
2110 	}
2111 }
2112 
kcore_copy__new_sym(struct kcore_copy_info * kci,u64 addr)2113 static struct sym_data *kcore_copy__new_sym(struct kcore_copy_info *kci,
2114 					    u64 addr)
2115 {
2116 	struct sym_data *s = zalloc(sizeof(*s));
2117 
2118 	if (s) {
2119 		s->addr = addr;
2120 		list_add_tail(&s->node, &kci->syms);
2121 	}
2122 
2123 	return s;
2124 }
2125 
kcore_copy__free_syms(struct kcore_copy_info * kci)2126 static void kcore_copy__free_syms(struct kcore_copy_info *kci)
2127 {
2128 	struct sym_data *s, *tmp;
2129 
2130 	list_for_each_entry_safe(s, tmp, &kci->syms, node) {
2131 		list_del_init(&s->node);
2132 		free(s);
2133 	}
2134 }
2135 
kcore_copy__process_kallsyms(void * arg,const char * name,char type,u64 start)2136 static int kcore_copy__process_kallsyms(void *arg, const char *name, char type,
2137 					u64 start)
2138 {
2139 	struct kcore_copy_info *kci = arg;
2140 
2141 	if (!kallsyms__is_function(type))
2142 		return 0;
2143 
2144 	if (strchr(name, '[')) {
2145 		if (!kci->first_module_symbol || start < kci->first_module_symbol)
2146 			kci->first_module_symbol = start;
2147 		if (start > kci->last_module_symbol)
2148 			kci->last_module_symbol = start;
2149 		return 0;
2150 	}
2151 
2152 	if (!kci->first_symbol || start < kci->first_symbol)
2153 		kci->first_symbol = start;
2154 
2155 	if (!kci->last_symbol || start > kci->last_symbol)
2156 		kci->last_symbol = start;
2157 
2158 	if (!strcmp(name, "_stext")) {
2159 		kci->stext = start;
2160 		return 0;
2161 	}
2162 
2163 	if (!strcmp(name, "_etext")) {
2164 		kci->etext = start;
2165 		return 0;
2166 	}
2167 
2168 	if (is_entry_trampoline(name) && !kcore_copy__new_sym(kci, start))
2169 		return -1;
2170 
2171 	return 0;
2172 }
2173 
kcore_copy__parse_kallsyms(struct kcore_copy_info * kci,const char * dir)2174 static int kcore_copy__parse_kallsyms(struct kcore_copy_info *kci,
2175 				      const char *dir)
2176 {
2177 	char kallsyms_filename[PATH_MAX];
2178 
2179 	scnprintf(kallsyms_filename, PATH_MAX, "%s/kallsyms", dir);
2180 
2181 	if (symbol__restricted_filename(kallsyms_filename, "/proc/kallsyms"))
2182 		return -1;
2183 
2184 	if (kallsyms__parse(kallsyms_filename, kci,
2185 			    kcore_copy__process_kallsyms) < 0)
2186 		return -1;
2187 
2188 	return 0;
2189 }
2190 
kcore_copy__process_modules(void * arg,const char * name __maybe_unused,u64 start,u64 size __maybe_unused)2191 static int kcore_copy__process_modules(void *arg,
2192 				       const char *name __maybe_unused,
2193 				       u64 start, u64 size __maybe_unused)
2194 {
2195 	struct kcore_copy_info *kci = arg;
2196 
2197 	if (!kci->first_module || start < kci->first_module)
2198 		kci->first_module = start;
2199 
2200 	return 0;
2201 }
2202 
kcore_copy__parse_modules(struct kcore_copy_info * kci,const char * dir)2203 static int kcore_copy__parse_modules(struct kcore_copy_info *kci,
2204 				     const char *dir)
2205 {
2206 	char modules_filename[PATH_MAX];
2207 
2208 	scnprintf(modules_filename, PATH_MAX, "%s/modules", dir);
2209 
2210 	if (symbol__restricted_filename(modules_filename, "/proc/modules"))
2211 		return -1;
2212 
2213 	if (modules__parse(modules_filename, kci,
2214 			   kcore_copy__process_modules) < 0)
2215 		return -1;
2216 
2217 	return 0;
2218 }
2219 
kcore_copy__map(struct kcore_copy_info * kci,u64 start,u64 end,u64 pgoff,u64 s,u64 e)2220 static int kcore_copy__map(struct kcore_copy_info *kci, u64 start, u64 end,
2221 			   u64 pgoff, u64 s, u64 e)
2222 {
2223 	u64 len, offset;
2224 
2225 	if (s < start || s >= end)
2226 		return 0;
2227 
2228 	offset = (s - start) + pgoff;
2229 	len = e < end ? e - s : end - s;
2230 
2231 	return kcore_copy_info__addnew(kci, s, len, offset) ? 0 : -1;
2232 }
2233 
kcore_copy__read_map(u64 start,u64 len,u64 pgoff,void * data)2234 static int kcore_copy__read_map(u64 start, u64 len, u64 pgoff, void *data)
2235 {
2236 	struct kcore_copy_info *kci = data;
2237 	u64 end = start + len;
2238 	struct sym_data *sdat;
2239 
2240 	if (kcore_copy__map(kci, start, end, pgoff, kci->stext, kci->etext))
2241 		return -1;
2242 
2243 	if (kcore_copy__map(kci, start, end, pgoff, kci->first_module,
2244 			    kci->last_module_symbol))
2245 		return -1;
2246 
2247 	list_for_each_entry(sdat, &kci->syms, node) {
2248 		u64 s = round_down(sdat->addr, page_size);
2249 
2250 		if (kcore_copy__map(kci, start, end, pgoff, s, s + len))
2251 			return -1;
2252 	}
2253 
2254 	return 0;
2255 }
2256 
kcore_copy__read_maps(struct kcore_copy_info * kci,Elf * elf)2257 static int kcore_copy__read_maps(struct kcore_copy_info *kci, Elf *elf)
2258 {
2259 	if (elf_read_maps(elf, true, kcore_copy__read_map, kci) < 0)
2260 		return -1;
2261 
2262 	return 0;
2263 }
2264 
kcore_copy__find_remaps(struct kcore_copy_info * kci)2265 static void kcore_copy__find_remaps(struct kcore_copy_info *kci)
2266 {
2267 	struct phdr_data *p, *k = NULL;
2268 	u64 kend;
2269 
2270 	if (!kci->stext)
2271 		return;
2272 
2273 	/* Find phdr that corresponds to the kernel map (contains stext) */
2274 	kcore_copy__for_each_phdr(kci, p) {
2275 		u64 pend = p->addr + p->len - 1;
2276 
2277 		if (p->addr <= kci->stext && pend >= kci->stext) {
2278 			k = p;
2279 			break;
2280 		}
2281 	}
2282 
2283 	if (!k)
2284 		return;
2285 
2286 	kend = k->offset + k->len;
2287 
2288 	/* Find phdrs that remap the kernel */
2289 	kcore_copy__for_each_phdr(kci, p) {
2290 		u64 pend = p->offset + p->len;
2291 
2292 		if (p == k)
2293 			continue;
2294 
2295 		if (p->offset >= k->offset && pend <= kend)
2296 			p->remaps = k;
2297 	}
2298 }
2299 
kcore_copy__layout(struct kcore_copy_info * kci)2300 static void kcore_copy__layout(struct kcore_copy_info *kci)
2301 {
2302 	struct phdr_data *p;
2303 	off_t rel = 0;
2304 
2305 	kcore_copy__find_remaps(kci);
2306 
2307 	kcore_copy__for_each_phdr(kci, p) {
2308 		if (!p->remaps) {
2309 			p->rel = rel;
2310 			rel += p->len;
2311 		}
2312 		kci->phnum += 1;
2313 	}
2314 
2315 	kcore_copy__for_each_phdr(kci, p) {
2316 		struct phdr_data *k = p->remaps;
2317 
2318 		if (k)
2319 			p->rel = p->offset - k->offset + k->rel;
2320 	}
2321 }
2322 
kcore_copy__calc_maps(struct kcore_copy_info * kci,const char * dir,Elf * elf)2323 static int kcore_copy__calc_maps(struct kcore_copy_info *kci, const char *dir,
2324 				 Elf *elf)
2325 {
2326 	if (kcore_copy__parse_kallsyms(kci, dir))
2327 		return -1;
2328 
2329 	if (kcore_copy__parse_modules(kci, dir))
2330 		return -1;
2331 
2332 	if (kci->stext)
2333 		kci->stext = round_down(kci->stext, page_size);
2334 	else
2335 		kci->stext = round_down(kci->first_symbol, page_size);
2336 
2337 	if (kci->etext) {
2338 		kci->etext = round_up(kci->etext, page_size);
2339 	} else if (kci->last_symbol) {
2340 		kci->etext = round_up(kci->last_symbol, page_size);
2341 		kci->etext += page_size;
2342 	}
2343 
2344 	if (kci->first_module_symbol &&
2345 	    (!kci->first_module || kci->first_module_symbol < kci->first_module))
2346 		kci->first_module = kci->first_module_symbol;
2347 
2348 	kci->first_module = round_down(kci->first_module, page_size);
2349 
2350 	if (kci->last_module_symbol) {
2351 		kci->last_module_symbol = round_up(kci->last_module_symbol,
2352 						   page_size);
2353 		kci->last_module_symbol += page_size;
2354 	}
2355 
2356 	if (!kci->stext || !kci->etext)
2357 		return -1;
2358 
2359 	if (kci->first_module && !kci->last_module_symbol)
2360 		return -1;
2361 
2362 	if (kcore_copy__read_maps(kci, elf))
2363 		return -1;
2364 
2365 	kcore_copy__layout(kci);
2366 
2367 	return 0;
2368 }
2369 
kcore_copy__copy_file(const char * from_dir,const char * to_dir,const char * name)2370 static int kcore_copy__copy_file(const char *from_dir, const char *to_dir,
2371 				 const char *name)
2372 {
2373 	char from_filename[PATH_MAX];
2374 	char to_filename[PATH_MAX];
2375 
2376 	scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
2377 	scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
2378 
2379 	return copyfile_mode(from_filename, to_filename, 0400);
2380 }
2381 
kcore_copy__unlink(const char * dir,const char * name)2382 static int kcore_copy__unlink(const char *dir, const char *name)
2383 {
2384 	char filename[PATH_MAX];
2385 
2386 	scnprintf(filename, PATH_MAX, "%s/%s", dir, name);
2387 
2388 	return unlink(filename);
2389 }
2390 
kcore_copy__compare_fds(int from,int to)2391 static int kcore_copy__compare_fds(int from, int to)
2392 {
2393 	char *buf_from;
2394 	char *buf_to;
2395 	ssize_t ret;
2396 	size_t len;
2397 	int err = -1;
2398 
2399 	buf_from = malloc(page_size);
2400 	buf_to = malloc(page_size);
2401 	if (!buf_from || !buf_to)
2402 		goto out;
2403 
2404 	while (1) {
2405 		/* Use read because mmap won't work on proc files */
2406 		ret = read(from, buf_from, page_size);
2407 		if (ret < 0)
2408 			goto out;
2409 
2410 		if (!ret)
2411 			break;
2412 
2413 		len = ret;
2414 
2415 		if (readn(to, buf_to, len) != (int)len)
2416 			goto out;
2417 
2418 		if (memcmp(buf_from, buf_to, len))
2419 			goto out;
2420 	}
2421 
2422 	err = 0;
2423 out:
2424 	free(buf_to);
2425 	free(buf_from);
2426 	return err;
2427 }
2428 
kcore_copy__compare_files(const char * from_filename,const char * to_filename)2429 static int kcore_copy__compare_files(const char *from_filename,
2430 				     const char *to_filename)
2431 {
2432 	int from, to, err = -1;
2433 
2434 	from = open(from_filename, O_RDONLY);
2435 	if (from < 0)
2436 		return -1;
2437 
2438 	to = open(to_filename, O_RDONLY);
2439 	if (to < 0)
2440 		goto out_close_from;
2441 
2442 	err = kcore_copy__compare_fds(from, to);
2443 
2444 	close(to);
2445 out_close_from:
2446 	close(from);
2447 	return err;
2448 }
2449 
kcore_copy__compare_file(const char * from_dir,const char * to_dir,const char * name)2450 static int kcore_copy__compare_file(const char *from_dir, const char *to_dir,
2451 				    const char *name)
2452 {
2453 	char from_filename[PATH_MAX];
2454 	char to_filename[PATH_MAX];
2455 
2456 	scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
2457 	scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
2458 
2459 	return kcore_copy__compare_files(from_filename, to_filename);
2460 }
2461 
2462 /**
2463  * kcore_copy - copy kallsyms, modules and kcore from one directory to another.
2464  * @from_dir: from directory
2465  * @to_dir: to directory
2466  *
2467  * This function copies kallsyms, modules and kcore files from one directory to
2468  * another.  kallsyms and modules are copied entirely.  Only code segments are
2469  * copied from kcore.  It is assumed that two segments suffice: one for the
2470  * kernel proper and one for all the modules.  The code segments are determined
2471  * from kallsyms and modules files.  The kernel map starts at _stext or the
2472  * lowest function symbol, and ends at _etext or the highest function symbol.
2473  * The module map starts at the lowest module address and ends at the highest
2474  * module symbol.  Start addresses are rounded down to the nearest page.  End
2475  * addresses are rounded up to the nearest page.  An extra page is added to the
2476  * highest kernel symbol and highest module symbol to, hopefully, encompass that
2477  * symbol too.  Because it contains only code sections, the resulting kcore is
2478  * unusual.  One significant peculiarity is that the mapping (start -> pgoff)
2479  * is not the same for the kernel map and the modules map.  That happens because
2480  * the data is copied adjacently whereas the original kcore has gaps.  Finally,
2481  * kallsyms file is compared with its copy to check that modules have not been
2482  * loaded or unloaded while the copies were taking place.
2483  *
2484  * Return: %0 on success, %-1 on failure.
2485  */
kcore_copy(const char * from_dir,const char * to_dir)2486 int kcore_copy(const char *from_dir, const char *to_dir)
2487 {
2488 	struct kcore kcore;
2489 	struct kcore extract;
2490 	int idx = 0, err = -1;
2491 	off_t offset, sz;
2492 	struct kcore_copy_info kci = { .stext = 0, };
2493 	char kcore_filename[PATH_MAX];
2494 	char extract_filename[PATH_MAX];
2495 	struct phdr_data *p;
2496 
2497 	INIT_LIST_HEAD(&kci.phdrs);
2498 	INIT_LIST_HEAD(&kci.syms);
2499 
2500 	if (kcore_copy__copy_file(from_dir, to_dir, "kallsyms"))
2501 		return -1;
2502 
2503 	if (kcore_copy__copy_file(from_dir, to_dir, "modules"))
2504 		goto out_unlink_kallsyms;
2505 
2506 	scnprintf(kcore_filename, PATH_MAX, "%s/kcore", from_dir);
2507 	scnprintf(extract_filename, PATH_MAX, "%s/kcore", to_dir);
2508 
2509 	if (kcore__open(&kcore, kcore_filename))
2510 		goto out_unlink_modules;
2511 
2512 	if (kcore_copy__calc_maps(&kci, from_dir, kcore.elf))
2513 		goto out_kcore_close;
2514 
2515 	if (kcore__init(&extract, extract_filename, kcore.elfclass, false))
2516 		goto out_kcore_close;
2517 
2518 	if (kcore__copy_hdr(&kcore, &extract, kci.phnum))
2519 		goto out_extract_close;
2520 
2521 	offset = gelf_fsize(extract.elf, ELF_T_EHDR, 1, EV_CURRENT) +
2522 		 gelf_fsize(extract.elf, ELF_T_PHDR, kci.phnum, EV_CURRENT);
2523 	offset = round_up(offset, page_size);
2524 
2525 	kcore_copy__for_each_phdr(&kci, p) {
2526 		off_t offs = p->rel + offset;
2527 
2528 		if (kcore__add_phdr(&extract, idx++, offs, p->addr, p->len))
2529 			goto out_extract_close;
2530 	}
2531 
2532 	sz = kcore__write(&extract);
2533 	if (sz < 0 || sz > offset)
2534 		goto out_extract_close;
2535 
2536 	kcore_copy__for_each_phdr(&kci, p) {
2537 		off_t offs = p->rel + offset;
2538 
2539 		if (p->remaps)
2540 			continue;
2541 		if (copy_bytes(kcore.fd, p->offset, extract.fd, offs, p->len))
2542 			goto out_extract_close;
2543 	}
2544 
2545 	if (kcore_copy__compare_file(from_dir, to_dir, "kallsyms"))
2546 		goto out_extract_close;
2547 
2548 	err = 0;
2549 
2550 out_extract_close:
2551 	kcore__close(&extract);
2552 	if (err)
2553 		unlink(extract_filename);
2554 out_kcore_close:
2555 	kcore__close(&kcore);
2556 out_unlink_modules:
2557 	if (err)
2558 		kcore_copy__unlink(to_dir, "modules");
2559 out_unlink_kallsyms:
2560 	if (err)
2561 		kcore_copy__unlink(to_dir, "kallsyms");
2562 
2563 	kcore_copy__free_phdrs(&kci);
2564 	kcore_copy__free_syms(&kci);
2565 
2566 	return err;
2567 }
2568 
kcore_extract__create(struct kcore_extract * kce)2569 int kcore_extract__create(struct kcore_extract *kce)
2570 {
2571 	struct kcore kcore;
2572 	struct kcore extract;
2573 	size_t count = 1;
2574 	int idx = 0, err = -1;
2575 	off_t offset = page_size, sz;
2576 
2577 	if (kcore__open(&kcore, kce->kcore_filename))
2578 		return -1;
2579 
2580 	strcpy(kce->extract_filename, PERF_KCORE_EXTRACT);
2581 	if (kcore__init(&extract, kce->extract_filename, kcore.elfclass, true))
2582 		goto out_kcore_close;
2583 
2584 	if (kcore__copy_hdr(&kcore, &extract, count))
2585 		goto out_extract_close;
2586 
2587 	if (kcore__add_phdr(&extract, idx, offset, kce->addr, kce->len))
2588 		goto out_extract_close;
2589 
2590 	sz = kcore__write(&extract);
2591 	if (sz < 0 || sz > offset)
2592 		goto out_extract_close;
2593 
2594 	if (copy_bytes(kcore.fd, kce->offs, extract.fd, offset, kce->len))
2595 		goto out_extract_close;
2596 
2597 	err = 0;
2598 
2599 out_extract_close:
2600 	kcore__close(&extract);
2601 	if (err)
2602 		unlink(kce->extract_filename);
2603 out_kcore_close:
2604 	kcore__close(&kcore);
2605 
2606 	return err;
2607 }
2608 
kcore_extract__delete(struct kcore_extract * kce)2609 void kcore_extract__delete(struct kcore_extract *kce)
2610 {
2611 	unlink(kce->extract_filename);
2612 }
2613 
2614 #ifdef HAVE_GELF_GETNOTE_SUPPORT
2615 
sdt_adjust_loc(struct sdt_note * tmp,GElf_Addr base_off)2616 static void sdt_adjust_loc(struct sdt_note *tmp, GElf_Addr base_off)
2617 {
2618 	if (!base_off)
2619 		return;
2620 
2621 	if (tmp->bit32)
2622 		tmp->addr.a32[SDT_NOTE_IDX_LOC] =
2623 			tmp->addr.a32[SDT_NOTE_IDX_LOC] + base_off -
2624 			tmp->addr.a32[SDT_NOTE_IDX_BASE];
2625 	else
2626 		tmp->addr.a64[SDT_NOTE_IDX_LOC] =
2627 			tmp->addr.a64[SDT_NOTE_IDX_LOC] + base_off -
2628 			tmp->addr.a64[SDT_NOTE_IDX_BASE];
2629 }
2630 
sdt_adjust_refctr(struct sdt_note * tmp,GElf_Addr base_addr,GElf_Addr base_off)2631 static void sdt_adjust_refctr(struct sdt_note *tmp, GElf_Addr base_addr,
2632 			      GElf_Addr base_off)
2633 {
2634 	if (!base_off)
2635 		return;
2636 
2637 	if (tmp->bit32 && tmp->addr.a32[SDT_NOTE_IDX_REFCTR])
2638 		tmp->addr.a32[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2639 	else if (tmp->addr.a64[SDT_NOTE_IDX_REFCTR])
2640 		tmp->addr.a64[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2641 }
2642 
2643 /**
2644  * populate_sdt_note : Parse raw data and identify SDT note
2645  * @elf: elf of the opened file
2646  * @data: raw data of a section with description offset applied
2647  * @len: note description size
2648  * @type: type of the note
2649  * @sdt_notes: List to add the SDT note
2650  *
2651  * Responsible for parsing the @data in section .note.stapsdt in @elf and
2652  * if its an SDT note, it appends to @sdt_notes list.
2653  */
populate_sdt_note(Elf ** elf,const char * data,size_t len,struct list_head * sdt_notes)2654 static int populate_sdt_note(Elf **elf, const char *data, size_t len,
2655 			     struct list_head *sdt_notes)
2656 {
2657 	const char *provider, *name, *args;
2658 	struct sdt_note *tmp = NULL;
2659 	GElf_Ehdr ehdr;
2660 	GElf_Shdr shdr;
2661 	int ret = -EINVAL;
2662 
2663 	union {
2664 		Elf64_Addr a64[NR_ADDR];
2665 		Elf32_Addr a32[NR_ADDR];
2666 	} buf;
2667 
2668 	Elf_Data dst = {
2669 		.d_buf = &buf, .d_type = ELF_T_ADDR, .d_version = EV_CURRENT,
2670 		.d_size = gelf_fsize((*elf), ELF_T_ADDR, NR_ADDR, EV_CURRENT),
2671 		.d_off = 0, .d_align = 0
2672 	};
2673 	Elf_Data src = {
2674 		.d_buf = (void *) data, .d_type = ELF_T_ADDR,
2675 		.d_version = EV_CURRENT, .d_size = dst.d_size, .d_off = 0,
2676 		.d_align = 0
2677 	};
2678 
2679 	tmp = (struct sdt_note *)calloc(1, sizeof(struct sdt_note));
2680 	if (!tmp) {
2681 		ret = -ENOMEM;
2682 		goto out_err;
2683 	}
2684 
2685 	INIT_LIST_HEAD(&tmp->note_list);
2686 
2687 	if (len < dst.d_size + 3)
2688 		goto out_free_note;
2689 
2690 	/* Translation from file representation to memory representation */
2691 	if (gelf_xlatetom(*elf, &dst, &src,
2692 			  elf_getident(*elf, NULL)[EI_DATA]) == NULL) {
2693 		pr_err("gelf_xlatetom : %s\n", elf_errmsg(-1));
2694 		goto out_free_note;
2695 	}
2696 
2697 	/* Populate the fields of sdt_note */
2698 	provider = data + dst.d_size;
2699 
2700 	name = (const char *)memchr(provider, '\0', data + len - provider);
2701 	if (name++ == NULL)
2702 		goto out_free_note;
2703 
2704 	tmp->provider = strdup(provider);
2705 	if (!tmp->provider) {
2706 		ret = -ENOMEM;
2707 		goto out_free_note;
2708 	}
2709 	tmp->name = strdup(name);
2710 	if (!tmp->name) {
2711 		ret = -ENOMEM;
2712 		goto out_free_prov;
2713 	}
2714 
2715 	args = memchr(name, '\0', data + len - name);
2716 
2717 	/*
2718 	 * There is no argument if:
2719 	 * - We reached the end of the note;
2720 	 * - There is not enough room to hold a potential string;
2721 	 * - The argument string is empty or just contains ':'.
2722 	 */
2723 	if (args == NULL || data + len - args < 2 ||
2724 		args[1] == ':' || args[1] == '\0')
2725 		tmp->args = NULL;
2726 	else {
2727 		tmp->args = strdup(++args);
2728 		if (!tmp->args) {
2729 			ret = -ENOMEM;
2730 			goto out_free_name;
2731 		}
2732 	}
2733 
2734 	if (gelf_getclass(*elf) == ELFCLASS32) {
2735 		memcpy(&tmp->addr, &buf, 3 * sizeof(Elf32_Addr));
2736 		tmp->bit32 = true;
2737 	} else {
2738 		memcpy(&tmp->addr, &buf, 3 * sizeof(Elf64_Addr));
2739 		tmp->bit32 = false;
2740 	}
2741 
2742 	if (!gelf_getehdr(*elf, &ehdr)) {
2743 		pr_debug("%s : cannot get elf header.\n", __func__);
2744 		ret = -EBADF;
2745 		goto out_free_args;
2746 	}
2747 
2748 	/* Adjust the prelink effect :
2749 	 * Find out the .stapsdt.base section.
2750 	 * This scn will help us to handle prelinking (if present).
2751 	 * Compare the retrieved file offset of the base section with the
2752 	 * base address in the description of the SDT note. If its different,
2753 	 * then accordingly, adjust the note location.
2754 	 */
2755 	if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_BASE_SCN, NULL))
2756 		sdt_adjust_loc(tmp, shdr.sh_offset);
2757 
2758 	/* Adjust reference counter offset */
2759 	if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_PROBES_SCN, NULL))
2760 		sdt_adjust_refctr(tmp, shdr.sh_addr, shdr.sh_offset);
2761 
2762 	list_add_tail(&tmp->note_list, sdt_notes);
2763 	return 0;
2764 
2765 out_free_args:
2766 	zfree(&tmp->args);
2767 out_free_name:
2768 	zfree(&tmp->name);
2769 out_free_prov:
2770 	zfree(&tmp->provider);
2771 out_free_note:
2772 	free(tmp);
2773 out_err:
2774 	return ret;
2775 }
2776 
2777 /**
2778  * construct_sdt_notes_list : constructs a list of SDT notes
2779  * @elf : elf to look into
2780  * @sdt_notes : empty list_head
2781  *
2782  * Scans the sections in 'elf' for the section
2783  * .note.stapsdt. It, then calls populate_sdt_note to find
2784  * out the SDT events and populates the 'sdt_notes'.
2785  */
construct_sdt_notes_list(Elf * elf,struct list_head * sdt_notes)2786 static int construct_sdt_notes_list(Elf *elf, struct list_head *sdt_notes)
2787 {
2788 	GElf_Ehdr ehdr;
2789 	Elf_Scn *scn = NULL;
2790 	Elf_Data *data;
2791 	GElf_Shdr shdr;
2792 	size_t shstrndx, next;
2793 	GElf_Nhdr nhdr;
2794 	size_t name_off, desc_off, offset;
2795 	int ret = 0;
2796 
2797 	if (gelf_getehdr(elf, &ehdr) == NULL) {
2798 		ret = -EBADF;
2799 		goto out_ret;
2800 	}
2801 	if (elf_getshdrstrndx(elf, &shstrndx) != 0) {
2802 		ret = -EBADF;
2803 		goto out_ret;
2804 	}
2805 
2806 	/* Look for the required section */
2807 	scn = elf_section_by_name(elf, &ehdr, &shdr, SDT_NOTE_SCN, NULL);
2808 	if (!scn) {
2809 		ret = -ENOENT;
2810 		goto out_ret;
2811 	}
2812 
2813 	if ((shdr.sh_type != SHT_NOTE) || (shdr.sh_flags & SHF_ALLOC)) {
2814 		ret = -ENOENT;
2815 		goto out_ret;
2816 	}
2817 
2818 	data = elf_getdata(scn, NULL);
2819 
2820 	/* Get the SDT notes */
2821 	for (offset = 0; (next = gelf_getnote(data, offset, &nhdr, &name_off,
2822 					      &desc_off)) > 0; offset = next) {
2823 		if (nhdr.n_namesz == sizeof(SDT_NOTE_NAME) &&
2824 		    !memcmp(data->d_buf + name_off, SDT_NOTE_NAME,
2825 			    sizeof(SDT_NOTE_NAME))) {
2826 			/* Check the type of the note */
2827 			if (nhdr.n_type != SDT_NOTE_TYPE)
2828 				goto out_ret;
2829 
2830 			ret = populate_sdt_note(&elf, ((data->d_buf) + desc_off),
2831 						nhdr.n_descsz, sdt_notes);
2832 			if (ret < 0)
2833 				goto out_ret;
2834 		}
2835 	}
2836 	if (list_empty(sdt_notes))
2837 		ret = -ENOENT;
2838 
2839 out_ret:
2840 	return ret;
2841 }
2842 
2843 /**
2844  * get_sdt_note_list : Wrapper to construct a list of sdt notes
2845  * @head : empty list_head
2846  * @target : file to find SDT notes from
2847  *
2848  * This opens the file, initializes
2849  * the ELF and then calls construct_sdt_notes_list.
2850  */
get_sdt_note_list(struct list_head * head,const char * target)2851 int get_sdt_note_list(struct list_head *head, const char *target)
2852 {
2853 	Elf *elf;
2854 	int fd, ret;
2855 
2856 	fd = open(target, O_RDONLY);
2857 	if (fd < 0)
2858 		return -EBADF;
2859 
2860 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
2861 	if (!elf) {
2862 		ret = -EBADF;
2863 		goto out_close;
2864 	}
2865 	ret = construct_sdt_notes_list(elf, head);
2866 	elf_end(elf);
2867 out_close:
2868 	close(fd);
2869 	return ret;
2870 }
2871 
2872 /**
2873  * cleanup_sdt_note_list : free the sdt notes' list
2874  * @sdt_notes: sdt notes' list
2875  *
2876  * Free up the SDT notes in @sdt_notes.
2877  * Returns the number of SDT notes free'd.
2878  */
cleanup_sdt_note_list(struct list_head * sdt_notes)2879 int cleanup_sdt_note_list(struct list_head *sdt_notes)
2880 {
2881 	struct sdt_note *tmp, *pos;
2882 	int nr_free = 0;
2883 
2884 	list_for_each_entry_safe(pos, tmp, sdt_notes, note_list) {
2885 		list_del_init(&pos->note_list);
2886 		zfree(&pos->args);
2887 		zfree(&pos->name);
2888 		zfree(&pos->provider);
2889 		free(pos);
2890 		nr_free++;
2891 	}
2892 	return nr_free;
2893 }
2894 
2895 /**
2896  * sdt_notes__get_count: Counts the number of sdt events
2897  * @start: list_head to sdt_notes list
2898  *
2899  * Returns the number of SDT notes in a list
2900  */
sdt_notes__get_count(struct list_head * start)2901 int sdt_notes__get_count(struct list_head *start)
2902 {
2903 	struct sdt_note *sdt_ptr;
2904 	int count = 0;
2905 
2906 	list_for_each_entry(sdt_ptr, start, note_list)
2907 		count++;
2908 	return count;
2909 }
2910 #endif
2911 
symbol__elf_init(void)2912 void symbol__elf_init(void)
2913 {
2914 	elf_version(EV_CURRENT);
2915 }
2916