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