Commit | Line | Data |
---|---|---|
a2928c42 ACM |
1 | #include "util.h" |
2 | #include "../perf.h" | |
a0055ae2 | 3 | #include "string.h" |
a2928c42 ACM |
4 | #include "symbol.h" |
5 | ||
6 | #include <libelf.h> | |
7 | #include <gelf.h> | |
8 | #include <elf.h> | |
9 | ||
0b73da3f IM |
10 | const char *sym_hist_filter; |
11 | ||
a2928c42 | 12 | static struct symbol *symbol__new(uint64_t start, uint64_t len, |
0b73da3f IM |
13 | const char *name, unsigned int priv_size, |
14 | uint64_t obj_start, int verbose) | |
a2928c42 | 15 | { |
0085c954 | 16 | size_t namelen = strlen(name) + 1; |
0b73da3f | 17 | struct symbol *self = calloc(1, priv_size + sizeof(*self) + namelen); |
a2928c42 | 18 | |
0b73da3f IM |
19 | if (!self) |
20 | return NULL; | |
21 | ||
22 | if (verbose >= 2) | |
23 | printf("new symbol: %016Lx [%08lx]: %s, hist: %p, obj_start: %p\n", | |
7d37a0cb | 24 | (__u64)start, (unsigned long)len, name, self->hist, (void *)(unsigned long)obj_start); |
0b73da3f IM |
25 | |
26 | self->obj_start= obj_start; | |
27 | self->hist = NULL; | |
28 | self->hist_sum = 0; | |
29 | ||
30 | if (sym_hist_filter && !strcmp(name, sym_hist_filter)) | |
31 | self->hist = calloc(sizeof(__u64), len); | |
32 | ||
33 | if (priv_size) { | |
34 | memset(self, 0, priv_size); | |
35 | self = ((void *)self) + priv_size; | |
a2928c42 | 36 | } |
0b73da3f IM |
37 | self->start = start; |
38 | self->end = start + len - 1; | |
39 | memcpy(self->name, name, namelen); | |
a2928c42 ACM |
40 | |
41 | return self; | |
42 | } | |
43 | ||
0085c954 | 44 | static void symbol__delete(struct symbol *self, unsigned int priv_size) |
a2928c42 | 45 | { |
0085c954 | 46 | free(((void *)self) - priv_size); |
a2928c42 ACM |
47 | } |
48 | ||
49 | static size_t symbol__fprintf(struct symbol *self, FILE *fp) | |
50 | { | |
51 | return fprintf(fp, " %llx-%llx %s\n", | |
52 | self->start, self->end, self->name); | |
53 | } | |
54 | ||
0085c954 | 55 | struct dso *dso__new(const char *name, unsigned int sym_priv_size) |
a2928c42 ACM |
56 | { |
57 | struct dso *self = malloc(sizeof(*self) + strlen(name) + 1); | |
58 | ||
59 | if (self != NULL) { | |
60 | strcpy(self->name, name); | |
61 | self->syms = RB_ROOT; | |
0085c954 | 62 | self->sym_priv_size = sym_priv_size; |
fc54db51 | 63 | self->find_symbol = dso__find_symbol; |
a2928c42 ACM |
64 | } |
65 | ||
66 | return self; | |
67 | } | |
68 | ||
69 | static void dso__delete_symbols(struct dso *self) | |
70 | { | |
71 | struct symbol *pos; | |
72 | struct rb_node *next = rb_first(&self->syms); | |
73 | ||
74 | while (next) { | |
75 | pos = rb_entry(next, struct symbol, rb_node); | |
76 | next = rb_next(&pos->rb_node); | |
c8c96525 | 77 | rb_erase(&pos->rb_node, &self->syms); |
0085c954 | 78 | symbol__delete(pos, self->sym_priv_size); |
a2928c42 ACM |
79 | } |
80 | } | |
81 | ||
82 | void dso__delete(struct dso *self) | |
83 | { | |
84 | dso__delete_symbols(self); | |
85 | free(self); | |
86 | } | |
87 | ||
88 | static void dso__insert_symbol(struct dso *self, struct symbol *sym) | |
89 | { | |
90 | struct rb_node **p = &self->syms.rb_node; | |
91 | struct rb_node *parent = NULL; | |
92 | const uint64_t ip = sym->start; | |
93 | struct symbol *s; | |
94 | ||
95 | while (*p != NULL) { | |
96 | parent = *p; | |
97 | s = rb_entry(parent, struct symbol, rb_node); | |
98 | if (ip < s->start) | |
99 | p = &(*p)->rb_left; | |
100 | else | |
101 | p = &(*p)->rb_right; | |
102 | } | |
103 | rb_link_node(&sym->rb_node, parent, p); | |
104 | rb_insert_color(&sym->rb_node, &self->syms); | |
105 | } | |
106 | ||
107 | struct symbol *dso__find_symbol(struct dso *self, uint64_t ip) | |
108 | { | |
109 | struct rb_node *n; | |
110 | ||
111 | if (self == NULL) | |
112 | return NULL; | |
113 | ||
114 | n = self->syms.rb_node; | |
115 | ||
116 | while (n) { | |
117 | struct symbol *s = rb_entry(n, struct symbol, rb_node); | |
118 | ||
119 | if (ip < s->start) | |
120 | n = n->rb_left; | |
121 | else if (ip > s->end) | |
122 | n = n->rb_right; | |
123 | else | |
124 | return s; | |
125 | } | |
126 | ||
127 | return NULL; | |
128 | } | |
129 | ||
130 | size_t dso__fprintf(struct dso *self, FILE *fp) | |
131 | { | |
132 | size_t ret = fprintf(fp, "dso: %s\n", self->name); | |
133 | ||
134 | struct rb_node *nd; | |
135 | for (nd = rb_first(&self->syms); nd; nd = rb_next(nd)) { | |
136 | struct symbol *pos = rb_entry(nd, struct symbol, rb_node); | |
137 | ret += symbol__fprintf(pos, fp); | |
138 | } | |
139 | ||
140 | return ret; | |
141 | } | |
142 | ||
bd74137e | 143 | static int dso__load_kallsyms(struct dso *self, symbol_filter_t filter, int verbose) |
a2928c42 ACM |
144 | { |
145 | struct rb_node *nd, *prevnd; | |
146 | char *line = NULL; | |
147 | size_t n; | |
148 | FILE *file = fopen("/proc/kallsyms", "r"); | |
149 | ||
150 | if (file == NULL) | |
151 | goto out_failure; | |
152 | ||
153 | while (!feof(file)) { | |
a0055ae2 | 154 | __u64 start; |
a2928c42 ACM |
155 | struct symbol *sym; |
156 | int line_len, len; | |
157 | char symbol_type; | |
158 | ||
159 | line_len = getline(&line, &n, file); | |
160 | if (line_len < 0) | |
161 | break; | |
162 | ||
163 | if (!line) | |
164 | goto out_failure; | |
165 | ||
166 | line[--line_len] = '\0'; /* \n */ | |
167 | ||
a0055ae2 | 168 | len = hex2u64(line, &start); |
a2928c42 ACM |
169 | |
170 | len++; | |
171 | if (len + 2 >= line_len) | |
172 | continue; | |
173 | ||
174 | symbol_type = toupper(line[len]); | |
175 | /* | |
176 | * We're interested only in code ('T'ext) | |
177 | */ | |
178 | if (symbol_type != 'T' && symbol_type != 'W') | |
179 | continue; | |
180 | /* | |
181 | * Well fix up the end later, when we have all sorted. | |
182 | */ | |
0085c954 | 183 | sym = symbol__new(start, 0xdead, line + len + 2, |
0b73da3f | 184 | self->sym_priv_size, 0, verbose); |
a2928c42 ACM |
185 | |
186 | if (sym == NULL) | |
187 | goto out_delete_line; | |
188 | ||
69ee69f6 ACM |
189 | if (filter && filter(self, sym)) |
190 | symbol__delete(sym, self->sym_priv_size); | |
191 | else | |
192 | dso__insert_symbol(self, sym); | |
a2928c42 ACM |
193 | } |
194 | ||
195 | /* | |
196 | * Now that we have all sorted out, just set the ->end of all | |
197 | * symbols | |
198 | */ | |
199 | prevnd = rb_first(&self->syms); | |
200 | ||
201 | if (prevnd == NULL) | |
202 | goto out_delete_line; | |
203 | ||
204 | for (nd = rb_next(prevnd); nd; nd = rb_next(nd)) { | |
205 | struct symbol *prev = rb_entry(prevnd, struct symbol, rb_node), | |
206 | *curr = rb_entry(nd, struct symbol, rb_node); | |
207 | ||
208 | prev->end = curr->start - 1; | |
209 | prevnd = nd; | |
210 | } | |
211 | ||
212 | free(line); | |
213 | fclose(file); | |
214 | ||
215 | return 0; | |
216 | ||
217 | out_delete_line: | |
218 | free(line); | |
219 | out_failure: | |
220 | return -1; | |
221 | } | |
222 | ||
223 | /** | |
224 | * elf_symtab__for_each_symbol - iterate thru all the symbols | |
225 | * | |
226 | * @self: struct elf_symtab instance to iterate | |
227 | * @index: uint32_t index | |
228 | * @sym: GElf_Sym iterator | |
229 | */ | |
230 | #define elf_symtab__for_each_symbol(syms, nr_syms, index, sym) \ | |
231 | for (index = 0, gelf_getsym(syms, index, &sym);\ | |
232 | index < nr_syms; \ | |
233 | index++, gelf_getsym(syms, index, &sym)) | |
234 | ||
235 | static inline uint8_t elf_sym__type(const GElf_Sym *sym) | |
236 | { | |
237 | return GELF_ST_TYPE(sym->st_info); | |
238 | } | |
239 | ||
240 | static inline int elf_sym__is_function(const GElf_Sym *sym) | |
241 | { | |
242 | return elf_sym__type(sym) == STT_FUNC && | |
243 | sym->st_name != 0 && | |
244 | sym->st_shndx != SHN_UNDEF && | |
245 | sym->st_size != 0; | |
246 | } | |
247 | ||
248 | static inline const char *elf_sym__name(const GElf_Sym *sym, | |
249 | const Elf_Data *symstrs) | |
250 | { | |
251 | return symstrs->d_buf + sym->st_name; | |
252 | } | |
253 | ||
254 | static Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep, | |
255 | GElf_Shdr *shp, const char *name, | |
256 | size_t *index) | |
257 | { | |
258 | Elf_Scn *sec = NULL; | |
259 | size_t cnt = 1; | |
260 | ||
261 | while ((sec = elf_nextscn(elf, sec)) != NULL) { | |
262 | char *str; | |
263 | ||
264 | gelf_getshdr(sec, shp); | |
265 | str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name); | |
266 | if (!strcmp(name, str)) { | |
267 | if (index) | |
268 | *index = cnt; | |
269 | break; | |
270 | } | |
271 | ++cnt; | |
272 | } | |
273 | ||
274 | return sec; | |
275 | } | |
276 | ||
8ce998d6 ACM |
277 | #define elf_section__for_each_rel(reldata, pos, pos_mem, idx, nr_entries) \ |
278 | for (idx = 0, pos = gelf_getrel(reldata, 0, &pos_mem); \ | |
279 | idx < nr_entries; \ | |
280 | ++idx, pos = gelf_getrel(reldata, idx, &pos_mem)) | |
281 | ||
282 | #define elf_section__for_each_rela(reldata, pos, pos_mem, idx, nr_entries) \ | |
283 | for (idx = 0, pos = gelf_getrela(reldata, 0, &pos_mem); \ | |
284 | idx < nr_entries; \ | |
285 | ++idx, pos = gelf_getrela(reldata, idx, &pos_mem)) | |
286 | ||
287 | static int dso__synthesize_plt_symbols(struct dso *self, Elf *elf, | |
288 | GElf_Ehdr *ehdr, Elf_Scn *scn_dynsym, | |
289 | GElf_Shdr *shdr_dynsym, | |
0b73da3f | 290 | size_t dynsym_idx, int verbose) |
8ce998d6 ACM |
291 | { |
292 | uint32_t nr_rel_entries, idx; | |
293 | GElf_Sym sym; | |
294 | __u64 plt_offset; | |
295 | GElf_Shdr shdr_plt; | |
296 | struct symbol *f; | |
297 | GElf_Shdr shdr_rel_plt; | |
298 | Elf_Data *reldata, *syms, *symstrs; | |
299 | Elf_Scn *scn_plt_rel, *scn_symstrs; | |
300 | char sympltname[1024]; | |
301 | int nr = 0, symidx; | |
302 | ||
303 | scn_plt_rel = elf_section_by_name(elf, ehdr, &shdr_rel_plt, | |
304 | ".rela.plt", NULL); | |
305 | if (scn_plt_rel == NULL) { | |
306 | scn_plt_rel = elf_section_by_name(elf, ehdr, &shdr_rel_plt, | |
307 | ".rel.plt", NULL); | |
308 | if (scn_plt_rel == NULL) | |
309 | return 0; | |
310 | } | |
311 | ||
312 | if (shdr_rel_plt.sh_link != dynsym_idx) | |
313 | return 0; | |
314 | ||
315 | if (elf_section_by_name(elf, ehdr, &shdr_plt, ".plt", NULL) == NULL) | |
316 | return 0; | |
317 | ||
318 | /* | |
319 | * Fetch the relocation section to find the indexes to the GOT | |
320 | * and the symbols in the .dynsym they refer to. | |
321 | */ | |
322 | reldata = elf_getdata(scn_plt_rel, NULL); | |
323 | if (reldata == NULL) | |
324 | return -1; | |
325 | ||
326 | syms = elf_getdata(scn_dynsym, NULL); | |
327 | if (syms == NULL) | |
328 | return -1; | |
329 | ||
330 | scn_symstrs = elf_getscn(elf, shdr_dynsym->sh_link); | |
331 | if (scn_symstrs == NULL) | |
332 | return -1; | |
333 | ||
334 | symstrs = elf_getdata(scn_symstrs, NULL); | |
335 | if (symstrs == NULL) | |
336 | return -1; | |
337 | ||
338 | nr_rel_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize; | |
339 | plt_offset = shdr_plt.sh_offset; | |
340 | ||
341 | if (shdr_rel_plt.sh_type == SHT_RELA) { | |
342 | GElf_Rela pos_mem, *pos; | |
343 | ||
344 | elf_section__for_each_rela(reldata, pos, pos_mem, idx, | |
345 | nr_rel_entries) { | |
346 | symidx = GELF_R_SYM(pos->r_info); | |
347 | plt_offset += shdr_plt.sh_entsize; | |
348 | gelf_getsym(syms, symidx, &sym); | |
349 | snprintf(sympltname, sizeof(sympltname), | |
350 | "%s@plt", elf_sym__name(&sym, symstrs)); | |
351 | ||
352 | f = symbol__new(plt_offset, shdr_plt.sh_entsize, | |
0b73da3f | 353 | sympltname, self->sym_priv_size, 0, verbose); |
8ce998d6 ACM |
354 | if (!f) |
355 | return -1; | |
356 | ||
357 | dso__insert_symbol(self, f); | |
358 | ++nr; | |
359 | } | |
360 | } else if (shdr_rel_plt.sh_type == SHT_REL) { | |
361 | GElf_Rel pos_mem, *pos; | |
362 | elf_section__for_each_rel(reldata, pos, pos_mem, idx, | |
363 | nr_rel_entries) { | |
364 | symidx = GELF_R_SYM(pos->r_info); | |
365 | plt_offset += shdr_plt.sh_entsize; | |
366 | gelf_getsym(syms, symidx, &sym); | |
367 | snprintf(sympltname, sizeof(sympltname), | |
368 | "%s@plt", elf_sym__name(&sym, symstrs)); | |
369 | ||
370 | f = symbol__new(plt_offset, shdr_plt.sh_entsize, | |
0b73da3f | 371 | sympltname, self->sym_priv_size, 0, verbose); |
8ce998d6 ACM |
372 | if (!f) |
373 | return -1; | |
374 | ||
375 | dso__insert_symbol(self, f); | |
376 | ++nr; | |
377 | } | |
378 | } else { | |
379 | /* | |
380 | * TODO: There are still one more shdr_rel_plt.sh_type | |
381 | * I have to investigate, but probably should be ignored. | |
382 | */ | |
383 | } | |
384 | ||
385 | return nr; | |
386 | } | |
387 | ||
69ee69f6 | 388 | static int dso__load_sym(struct dso *self, int fd, const char *name, |
bd74137e | 389 | symbol_filter_t filter, int verbose) |
a2928c42 ACM |
390 | { |
391 | Elf_Data *symstrs; | |
392 | uint32_t nr_syms; | |
393 | int err = -1; | |
394 | uint32_t index; | |
395 | GElf_Ehdr ehdr; | |
396 | GElf_Shdr shdr; | |
397 | Elf_Data *syms; | |
398 | GElf_Sym sym; | |
8ce998d6 | 399 | Elf_Scn *sec, *sec_dynsym; |
a2928c42 | 400 | Elf *elf; |
8ce998d6 | 401 | size_t dynsym_idx; |
a2928c42 ACM |
402 | int nr = 0; |
403 | ||
404 | elf = elf_begin(fd, ELF_C_READ_MMAP, NULL); | |
405 | if (elf == NULL) { | |
bd74137e IM |
406 | if (verbose) |
407 | fprintf(stderr, "%s: cannot read %s ELF file.\n", | |
408 | __func__, name); | |
a2928c42 ACM |
409 | goto out_close; |
410 | } | |
411 | ||
412 | if (gelf_getehdr(elf, &ehdr) == NULL) { | |
bd74137e IM |
413 | if (verbose) |
414 | fprintf(stderr, "%s: cannot get elf header.\n", __func__); | |
a2928c42 ACM |
415 | goto out_elf_end; |
416 | } | |
417 | ||
8ce998d6 ACM |
418 | /* |
419 | * We need to check if we have a .dynsym, so that we can handle the | |
420 | * .plt, synthesizing its symbols, that aren't on the symtabs (be it | |
421 | * .dynsym or .symtab) | |
422 | */ | |
423 | sec_dynsym = elf_section_by_name(elf, &ehdr, &shdr, | |
424 | ".dynsym", &dynsym_idx); | |
425 | if (sec_dynsym != NULL) { | |
426 | nr = dso__synthesize_plt_symbols(self, elf, &ehdr, | |
427 | sec_dynsym, &shdr, | |
0b73da3f | 428 | dynsym_idx, verbose); |
8ce998d6 ACM |
429 | if (nr < 0) |
430 | goto out_elf_end; | |
431 | } | |
432 | ||
433 | /* | |
434 | * But if we have a full .symtab (that is a superset of .dynsym) we | |
435 | * should add the symbols not in the .dynsyn | |
436 | */ | |
a2928c42 | 437 | sec = elf_section_by_name(elf, &ehdr, &shdr, ".symtab", NULL); |
8ce998d6 ACM |
438 | if (sec == NULL) { |
439 | if (sec_dynsym == NULL) | |
440 | goto out_elf_end; | |
a2928c42 | 441 | |
8ce998d6 ACM |
442 | sec = sec_dynsym; |
443 | gelf_getshdr(sec, &shdr); | |
444 | } | |
a2928c42 ACM |
445 | |
446 | syms = elf_getdata(sec, NULL); | |
447 | if (syms == NULL) | |
448 | goto out_elf_end; | |
449 | ||
450 | sec = elf_getscn(elf, shdr.sh_link); | |
451 | if (sec == NULL) | |
452 | goto out_elf_end; | |
453 | ||
454 | symstrs = elf_getdata(sec, NULL); | |
455 | if (symstrs == NULL) | |
456 | goto out_elf_end; | |
457 | ||
458 | nr_syms = shdr.sh_size / shdr.sh_entsize; | |
459 | ||
e9fbc9dc AV |
460 | memset(&sym, 0, sizeof(sym)); |
461 | ||
a2928c42 ACM |
462 | elf_symtab__for_each_symbol(syms, nr_syms, index, sym) { |
463 | struct symbol *f; | |
0b73da3f | 464 | uint64_t obj_start; |
a2928c42 ACM |
465 | |
466 | if (!elf_sym__is_function(&sym)) | |
467 | continue; | |
468 | ||
469 | sec = elf_getscn(elf, sym.st_shndx); | |
470 | if (!sec) | |
471 | goto out_elf_end; | |
472 | ||
473 | gelf_getshdr(sec, &shdr); | |
0b73da3f IM |
474 | obj_start = sym.st_value; |
475 | ||
a2928c42 ACM |
476 | sym.st_value -= shdr.sh_addr - shdr.sh_offset; |
477 | ||
478 | f = symbol__new(sym.st_value, sym.st_size, | |
0085c954 | 479 | elf_sym__name(&sym, symstrs), |
0b73da3f | 480 | self->sym_priv_size, obj_start, verbose); |
a2928c42 ACM |
481 | if (!f) |
482 | goto out_elf_end; | |
483 | ||
69ee69f6 ACM |
484 | if (filter && filter(self, f)) |
485 | symbol__delete(f, self->sym_priv_size); | |
486 | else { | |
487 | dso__insert_symbol(self, f); | |
488 | nr++; | |
489 | } | |
a2928c42 ACM |
490 | } |
491 | ||
492 | err = nr; | |
493 | out_elf_end: | |
494 | elf_end(elf); | |
495 | out_close: | |
496 | return err; | |
497 | } | |
498 | ||
bd74137e | 499 | int dso__load(struct dso *self, symbol_filter_t filter, int verbose) |
a2928c42 ACM |
500 | { |
501 | int size = strlen(self->name) + sizeof("/usr/lib/debug%s.debug"); | |
502 | char *name = malloc(size); | |
503 | int variant = 0; | |
504 | int ret = -1; | |
505 | int fd; | |
506 | ||
507 | if (!name) | |
508 | return -1; | |
509 | ||
510 | more: | |
511 | do { | |
512 | switch (variant) { | |
513 | case 0: /* Fedora */ | |
514 | snprintf(name, size, "/usr/lib/debug%s.debug", self->name); | |
515 | break; | |
516 | case 1: /* Ubuntu */ | |
517 | snprintf(name, size, "/usr/lib/debug%s", self->name); | |
518 | break; | |
519 | case 2: /* Sane people */ | |
520 | snprintf(name, size, "%s", self->name); | |
521 | break; | |
522 | ||
523 | default: | |
524 | goto out; | |
525 | } | |
526 | variant++; | |
527 | ||
528 | fd = open(name, O_RDONLY); | |
529 | } while (fd < 0); | |
530 | ||
bd74137e | 531 | ret = dso__load_sym(self, fd, name, filter, verbose); |
a2928c42 ACM |
532 | close(fd); |
533 | ||
534 | /* | |
535 | * Some people seem to have debuginfo files _WITHOUT_ debug info!?!? | |
536 | */ | |
537 | if (!ret) | |
538 | goto more; | |
539 | ||
540 | out: | |
541 | free(name); | |
542 | return ret; | |
543 | } | |
544 | ||
69ee69f6 | 545 | static int dso__load_vmlinux(struct dso *self, const char *vmlinux, |
bd74137e | 546 | symbol_filter_t filter, int verbose) |
a2928c42 ACM |
547 | { |
548 | int err, fd = open(vmlinux, O_RDONLY); | |
549 | ||
550 | if (fd < 0) | |
551 | return -1; | |
552 | ||
bd74137e | 553 | err = dso__load_sym(self, fd, vmlinux, filter, verbose); |
a2928c42 ACM |
554 | close(fd); |
555 | ||
556 | return err; | |
557 | } | |
558 | ||
bd74137e IM |
559 | int dso__load_kernel(struct dso *self, const char *vmlinux, |
560 | symbol_filter_t filter, int verbose) | |
a827c875 ACM |
561 | { |
562 | int err = -1; | |
563 | ||
564 | if (vmlinux) | |
bd74137e | 565 | err = dso__load_vmlinux(self, vmlinux, filter, verbose); |
a827c875 ACM |
566 | |
567 | if (err) | |
bd74137e | 568 | err = dso__load_kallsyms(self, filter, verbose); |
a827c875 ACM |
569 | |
570 | return err; | |
571 | } | |
572 | ||
a2928c42 ACM |
573 | void symbol__init(void) |
574 | { | |
575 | elf_version(EV_CURRENT); | |
576 | } |