Commit | Line | Data |
---|---|---|
5c742725 JO |
1 | // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) |
2 | ||
bb7fa093 HC |
3 | #ifndef _GNU_SOURCE |
4 | #define _GNU_SOURCE | |
5 | #endif | |
5c742725 JO |
6 | #include <libelf.h> |
7 | #include <gelf.h> | |
8 | #include <fcntl.h> | |
9 | #include <linux/kernel.h> | |
10 | ||
11 | #include "libbpf_internal.h" | |
12 | #include "str_error.h" | |
13 | ||
d17aff80 AN |
14 | #define STRERR_BUFSIZE 128 |
15 | ||
bb7fa093 HC |
16 | /* A SHT_GNU_versym section holds 16-bit words. This bit is set if |
17 | * the symbol is hidden and can only be seen when referenced using an | |
18 | * explicit version number. This is a GNU extension. | |
19 | */ | |
20 | #define VERSYM_HIDDEN 0x8000 | |
21 | ||
22 | /* This is the mask for the rest of the data in a word read from a | |
23 | * SHT_GNU_versym section. | |
24 | */ | |
25 | #define VERSYM_VERSION 0x7fff | |
26 | ||
f90eb70d JO |
27 | int elf_open(const char *binary_path, struct elf_fd *elf_fd) |
28 | { | |
29 | char errmsg[STRERR_BUFSIZE]; | |
30 | int fd, ret; | |
31 | Elf *elf; | |
32 | ||
33 | if (elf_version(EV_CURRENT) == EV_NONE) { | |
34 | pr_warn("elf: failed to init libelf for %s\n", binary_path); | |
35 | return -LIBBPF_ERRNO__LIBELF; | |
36 | } | |
37 | fd = open(binary_path, O_RDONLY | O_CLOEXEC); | |
38 | if (fd < 0) { | |
39 | ret = -errno; | |
40 | pr_warn("elf: failed to open %s: %s\n", binary_path, | |
41 | libbpf_strerror_r(ret, errmsg, sizeof(errmsg))); | |
42 | return ret; | |
43 | } | |
44 | elf = elf_begin(fd, ELF_C_READ_MMAP, NULL); | |
45 | if (!elf) { | |
46 | pr_warn("elf: could not read elf from %s: %s\n", binary_path, elf_errmsg(-1)); | |
47 | close(fd); | |
48 | return -LIBBPF_ERRNO__FORMAT; | |
49 | } | |
50 | elf_fd->fd = fd; | |
51 | elf_fd->elf = elf; | |
52 | return 0; | |
53 | } | |
54 | ||
55 | void elf_close(struct elf_fd *elf_fd) | |
56 | { | |
57 | if (!elf_fd) | |
58 | return; | |
59 | elf_end(elf_fd->elf); | |
60 | close(elf_fd->fd); | |
61 | } | |
62 | ||
5c742725 JO |
63 | /* Return next ELF section of sh_type after scn, or first of that type if scn is NULL. */ |
64 | static Elf_Scn *elf_find_next_scn_by_type(Elf *elf, int sh_type, Elf_Scn *scn) | |
65 | { | |
66 | while ((scn = elf_nextscn(elf, scn)) != NULL) { | |
67 | GElf_Shdr sh; | |
68 | ||
69 | if (!gelf_getshdr(scn, &sh)) | |
70 | continue; | |
71 | if (sh.sh_type == sh_type) | |
72 | return scn; | |
73 | } | |
74 | return NULL; | |
75 | } | |
76 | ||
3774705d JO |
77 | struct elf_sym { |
78 | const char *name; | |
79 | GElf_Sym sym; | |
80 | GElf_Shdr sh; | |
bb7fa093 HC |
81 | int ver; |
82 | bool hidden; | |
3774705d JO |
83 | }; |
84 | ||
85 | struct elf_sym_iter { | |
86 | Elf *elf; | |
87 | Elf_Data *syms; | |
bb7fa093 HC |
88 | Elf_Data *versyms; |
89 | Elf_Data *verdefs; | |
3774705d JO |
90 | size_t nr_syms; |
91 | size_t strtabidx; | |
bb7fa093 | 92 | size_t verdef_strtabidx; |
3774705d JO |
93 | size_t next_sym_idx; |
94 | struct elf_sym sym; | |
95 | int st_type; | |
96 | }; | |
97 | ||
98 | static int elf_sym_iter_new(struct elf_sym_iter *iter, | |
99 | Elf *elf, const char *binary_path, | |
100 | int sh_type, int st_type) | |
101 | { | |
102 | Elf_Scn *scn = NULL; | |
103 | GElf_Ehdr ehdr; | |
104 | GElf_Shdr sh; | |
105 | ||
106 | memset(iter, 0, sizeof(*iter)); | |
107 | ||
108 | if (!gelf_getehdr(elf, &ehdr)) { | |
109 | pr_warn("elf: failed to get ehdr from %s: %s\n", binary_path, elf_errmsg(-1)); | |
110 | return -EINVAL; | |
111 | } | |
112 | ||
113 | scn = elf_find_next_scn_by_type(elf, sh_type, NULL); | |
114 | if (!scn) { | |
115 | pr_debug("elf: failed to find symbol table ELF sections in '%s'\n", | |
116 | binary_path); | |
117 | return -ENOENT; | |
118 | } | |
119 | ||
120 | if (!gelf_getshdr(scn, &sh)) | |
121 | return -EINVAL; | |
122 | ||
123 | iter->strtabidx = sh.sh_link; | |
124 | iter->syms = elf_getdata(scn, 0); | |
125 | if (!iter->syms) { | |
126 | pr_warn("elf: failed to get symbols for symtab section in '%s': %s\n", | |
127 | binary_path, elf_errmsg(-1)); | |
128 | return -EINVAL; | |
129 | } | |
130 | iter->nr_syms = iter->syms->d_size / sh.sh_entsize; | |
131 | iter->elf = elf; | |
132 | iter->st_type = st_type; | |
bb7fa093 HC |
133 | |
134 | /* Version symbol table is meaningful to dynsym only */ | |
135 | if (sh_type != SHT_DYNSYM) | |
136 | return 0; | |
137 | ||
138 | scn = elf_find_next_scn_by_type(elf, SHT_GNU_versym, NULL); | |
139 | if (!scn) | |
140 | return 0; | |
141 | iter->versyms = elf_getdata(scn, 0); | |
142 | ||
143 | scn = elf_find_next_scn_by_type(elf, SHT_GNU_verdef, NULL); | |
137df118 AN |
144 | if (!scn) |
145 | return 0; | |
146 | ||
147 | iter->verdefs = elf_getdata(scn, 0); | |
148 | if (!iter->verdefs || !gelf_getshdr(scn, &sh)) { | |
149 | pr_warn("elf: failed to get verdef ELF section in '%s'\n", binary_path); | |
bb7fa093 | 150 | return -EINVAL; |
137df118 | 151 | } |
bb7fa093 | 152 | iter->verdef_strtabidx = sh.sh_link; |
bb7fa093 | 153 | |
3774705d JO |
154 | return 0; |
155 | } | |
156 | ||
157 | static struct elf_sym *elf_sym_iter_next(struct elf_sym_iter *iter) | |
158 | { | |
159 | struct elf_sym *ret = &iter->sym; | |
160 | GElf_Sym *sym = &ret->sym; | |
161 | const char *name = NULL; | |
bb7fa093 | 162 | GElf_Versym versym; |
3774705d JO |
163 | Elf_Scn *sym_scn; |
164 | size_t idx; | |
165 | ||
166 | for (idx = iter->next_sym_idx; idx < iter->nr_syms; idx++) { | |
167 | if (!gelf_getsym(iter->syms, idx, sym)) | |
168 | continue; | |
169 | if (GELF_ST_TYPE(sym->st_info) != iter->st_type) | |
170 | continue; | |
171 | name = elf_strptr(iter->elf, iter->strtabidx, sym->st_name); | |
172 | if (!name) | |
173 | continue; | |
174 | sym_scn = elf_getscn(iter->elf, sym->st_shndx); | |
175 | if (!sym_scn) | |
176 | continue; | |
177 | if (!gelf_getshdr(sym_scn, &ret->sh)) | |
178 | continue; | |
179 | ||
180 | iter->next_sym_idx = idx + 1; | |
181 | ret->name = name; | |
bb7fa093 HC |
182 | ret->ver = 0; |
183 | ret->hidden = false; | |
184 | ||
185 | if (iter->versyms) { | |
186 | if (!gelf_getversym(iter->versyms, idx, &versym)) | |
187 | continue; | |
188 | ret->ver = versym & VERSYM_VERSION; | |
189 | ret->hidden = versym & VERSYM_HIDDEN; | |
190 | } | |
3774705d JO |
191 | return ret; |
192 | } | |
193 | ||
194 | return NULL; | |
195 | } | |
196 | ||
bb7fa093 HC |
197 | static const char *elf_get_vername(struct elf_sym_iter *iter, int ver) |
198 | { | |
199 | GElf_Verdaux verdaux; | |
200 | GElf_Verdef verdef; | |
201 | int offset; | |
202 | ||
137df118 AN |
203 | if (!iter->verdefs) |
204 | return NULL; | |
205 | ||
bb7fa093 HC |
206 | offset = 0; |
207 | while (gelf_getverdef(iter->verdefs, offset, &verdef)) { | |
208 | if (verdef.vd_ndx != ver) { | |
209 | if (!verdef.vd_next) | |
210 | break; | |
211 | ||
212 | offset += verdef.vd_next; | |
213 | continue; | |
214 | } | |
215 | ||
216 | if (!gelf_getverdaux(iter->verdefs, offset + verdef.vd_aux, &verdaux)) | |
217 | break; | |
218 | ||
219 | return elf_strptr(iter->elf, iter->verdef_strtabidx, verdaux.vda_name); | |
220 | ||
221 | } | |
222 | return NULL; | |
223 | } | |
224 | ||
225 | static bool symbol_match(struct elf_sym_iter *iter, int sh_type, struct elf_sym *sym, | |
226 | const char *name, size_t name_len, const char *lib_ver) | |
227 | { | |
228 | const char *ver_name; | |
229 | ||
230 | /* Symbols are in forms of func, func@LIB_VER or func@@LIB_VER | |
231 | * make sure the func part matches the user specified name | |
232 | */ | |
233 | if (strncmp(sym->name, name, name_len) != 0) | |
234 | return false; | |
235 | ||
236 | /* ...but we don't want a search for "foo" to match 'foo2" also, so any | |
237 | * additional characters in sname should be of the form "@@LIB". | |
238 | */ | |
239 | if (sym->name[name_len] != '\0' && sym->name[name_len] != '@') | |
240 | return false; | |
241 | ||
242 | /* If user does not specify symbol version, then we got a match */ | |
243 | if (!lib_ver) | |
244 | return true; | |
245 | ||
246 | /* If user specifies symbol version, for dynamic symbols, | |
247 | * get version name from ELF verdef section for comparison. | |
248 | */ | |
249 | if (sh_type == SHT_DYNSYM) { | |
250 | ver_name = elf_get_vername(iter, sym->ver); | |
251 | if (!ver_name) | |
252 | return false; | |
253 | return strcmp(ver_name, lib_ver) == 0; | |
254 | } | |
255 | ||
256 | /* For normal symbols, it is already in form of func@LIB_VER */ | |
257 | return strcmp(sym->name, name) == 0; | |
258 | } | |
3774705d JO |
259 | |
260 | /* Transform symbol's virtual address (absolute for binaries and relative | |
261 | * for shared libs) into file offset, which is what kernel is expecting | |
262 | * for uprobe/uretprobe attachment. | |
263 | * See Documentation/trace/uprobetracer.rst for more details. This is done | |
264 | * by looking up symbol's containing section's header and using iter's virtual | |
265 | * address (sh_addr) and corresponding file offset (sh_offset) to transform | |
266 | * sym.st_value (virtual address) into desired final file offset. | |
267 | */ | |
268 | static unsigned long elf_sym_offset(struct elf_sym *sym) | |
269 | { | |
270 | return sym->sym.st_value - sym->sh.sh_addr + sym->sh.sh_offset; | |
271 | } | |
272 | ||
5c742725 JO |
273 | /* Find offset of function name in the provided ELF object. "binary_path" is |
274 | * the path to the ELF binary represented by "elf", and only used for error | |
275 | * reporting matters. "name" matches symbol name or name@@LIB for library | |
276 | * functions. | |
277 | */ | |
278 | long elf_find_func_offset(Elf *elf, const char *binary_path, const char *name) | |
279 | { | |
280 | int i, sh_types[2] = { SHT_DYNSYM, SHT_SYMTAB }; | |
bb7fa093 HC |
281 | const char *at_symbol, *lib_ver; |
282 | bool is_shared_lib; | |
5c742725 JO |
283 | long ret = -ENOENT; |
284 | size_t name_len; | |
285 | GElf_Ehdr ehdr; | |
286 | ||
287 | if (!gelf_getehdr(elf, &ehdr)) { | |
288 | pr_warn("elf: failed to get ehdr from %s: %s\n", binary_path, elf_errmsg(-1)); | |
289 | ret = -LIBBPF_ERRNO__FORMAT; | |
290 | goto out; | |
291 | } | |
292 | /* for shared lib case, we do not need to calculate relative offset */ | |
293 | is_shared_lib = ehdr.e_type == ET_DYN; | |
294 | ||
bb7fa093 HC |
295 | /* Does name specify "@@LIB_VER" or "@LIB_VER" ? */ |
296 | at_symbol = strchr(name, '@'); | |
297 | if (at_symbol) { | |
298 | name_len = at_symbol - name; | |
299 | /* skip second @ if it's @@LIB_VER case */ | |
300 | if (at_symbol[1] == '@') | |
301 | at_symbol++; | |
302 | lib_ver = at_symbol + 1; | |
303 | } else { | |
304 | name_len = strlen(name); | |
305 | lib_ver = NULL; | |
306 | } | |
5c742725 JO |
307 | |
308 | /* Search SHT_DYNSYM, SHT_SYMTAB for symbol. This search order is used because if | |
309 | * a binary is stripped, it may only have SHT_DYNSYM, and a fully-statically | |
310 | * linked binary may not have SHT_DYMSYM, so absence of a section should not be | |
311 | * reported as a warning/error. | |
312 | */ | |
313 | for (i = 0; i < ARRAY_SIZE(sh_types); i++) { | |
3774705d JO |
314 | struct elf_sym_iter iter; |
315 | struct elf_sym *sym; | |
5c742725 | 316 | int last_bind = -1; |
3774705d | 317 | int cur_bind; |
5c742725 | 318 | |
3774705d JO |
319 | ret = elf_sym_iter_new(&iter, elf, binary_path, sh_types[i], STT_FUNC); |
320 | if (ret == -ENOENT) | |
5c742725 | 321 | continue; |
3774705d | 322 | if (ret) |
5c742725 | 323 | goto out; |
5c742725 | 324 | |
3774705d | 325 | while ((sym = elf_sym_iter_next(&iter))) { |
bb7fa093 | 326 | if (!symbol_match(&iter, sh_types[i], sym, name, name_len, lib_ver)) |
5c742725 JO |
327 | continue; |
328 | ||
3774705d JO |
329 | cur_bind = GELF_ST_BIND(sym->sym.st_info); |
330 | ||
331 | if (ret > 0) { | |
5c742725 | 332 | /* handle multiple matches */ |
7257cee6 HC |
333 | if (elf_sym_offset(sym) == ret) { |
334 | /* same offset, no problem */ | |
335 | continue; | |
336 | } else if (last_bind != STB_WEAK && cur_bind != STB_WEAK) { | |
5c742725 JO |
337 | /* Only accept one non-weak bind. */ |
338 | pr_warn("elf: ambiguous match for '%s', '%s' in '%s'\n", | |
3774705d | 339 | sym->name, name, binary_path); |
5c742725 JO |
340 | ret = -LIBBPF_ERRNO__FORMAT; |
341 | goto out; | |
3774705d | 342 | } else if (cur_bind == STB_WEAK) { |
5c742725 JO |
343 | /* already have a non-weak bind, and |
344 | * this is a weak bind, so ignore. | |
345 | */ | |
346 | continue; | |
347 | } | |
348 | } | |
349 | ||
3774705d JO |
350 | ret = elf_sym_offset(sym); |
351 | last_bind = cur_bind; | |
5c742725 JO |
352 | } |
353 | if (ret > 0) | |
354 | break; | |
355 | } | |
356 | ||
357 | if (ret > 0) { | |
358 | pr_debug("elf: symbol address match for '%s' in '%s': 0x%lx\n", name, binary_path, | |
359 | ret); | |
360 | } else { | |
361 | if (ret == 0) { | |
362 | pr_warn("elf: '%s' is 0 in symtab for '%s': %s\n", name, binary_path, | |
363 | is_shared_lib ? "should not be 0 in a shared library" : | |
364 | "try using shared library path instead"); | |
365 | ret = -ENOENT; | |
366 | } else { | |
367 | pr_warn("elf: failed to find symbol '%s' in '%s'\n", name, binary_path); | |
368 | } | |
369 | } | |
370 | out: | |
371 | return ret; | |
372 | } | |
373 | ||
374 | /* Find offset of function name in ELF object specified by path. "name" matches | |
375 | * symbol name or name@@LIB for library functions. | |
376 | */ | |
377 | long elf_find_func_offset_from_file(const char *binary_path, const char *name) | |
378 | { | |
f90eb70d | 379 | struct elf_fd elf_fd; |
5c742725 | 380 | long ret = -ENOENT; |
5c742725 | 381 | |
f90eb70d JO |
382 | ret = elf_open(binary_path, &elf_fd); |
383 | if (ret) | |
5c742725 | 384 | return ret; |
f90eb70d JO |
385 | ret = elf_find_func_offset(elf_fd.elf, binary_path, name); |
386 | elf_close(&elf_fd); | |
5c742725 JO |
387 | return ret; |
388 | } | |
7ace84c6 JO |
389 | |
390 | struct symbol { | |
391 | const char *name; | |
392 | int bind; | |
393 | int idx; | |
394 | }; | |
395 | ||
396 | static int symbol_cmp(const void *a, const void *b) | |
397 | { | |
398 | const struct symbol *sym_a = a; | |
399 | const struct symbol *sym_b = b; | |
400 | ||
401 | return strcmp(sym_a->name, sym_b->name); | |
402 | } | |
403 | ||
404 | /* | |
405 | * Return offsets in @poffsets for symbols specified in @syms array argument. | |
406 | * On success returns 0 and offsets are returned in allocated array with @cnt | |
407 | * size, that needs to be released by the caller. | |
408 | */ | |
409 | int elf_resolve_syms_offsets(const char *binary_path, int cnt, | |
48f0dfd8 JO |
410 | const char **syms, unsigned long **poffsets, |
411 | int st_type) | |
7ace84c6 JO |
412 | { |
413 | int sh_types[2] = { SHT_DYNSYM, SHT_SYMTAB }; | |
414 | int err = 0, i, cnt_done = 0; | |
415 | unsigned long *offsets; | |
416 | struct symbol *symbols; | |
417 | struct elf_fd elf_fd; | |
418 | ||
419 | err = elf_open(binary_path, &elf_fd); | |
420 | if (err) | |
421 | return err; | |
422 | ||
423 | offsets = calloc(cnt, sizeof(*offsets)); | |
424 | symbols = calloc(cnt, sizeof(*symbols)); | |
425 | ||
426 | if (!offsets || !symbols) { | |
427 | err = -ENOMEM; | |
428 | goto out; | |
429 | } | |
430 | ||
431 | for (i = 0; i < cnt; i++) { | |
432 | symbols[i].name = syms[i]; | |
433 | symbols[i].idx = i; | |
434 | } | |
435 | ||
436 | qsort(symbols, cnt, sizeof(*symbols), symbol_cmp); | |
437 | ||
438 | for (i = 0; i < ARRAY_SIZE(sh_types); i++) { | |
439 | struct elf_sym_iter iter; | |
440 | struct elf_sym *sym; | |
441 | ||
48f0dfd8 | 442 | err = elf_sym_iter_new(&iter, elf_fd.elf, binary_path, sh_types[i], st_type); |
7ace84c6 JO |
443 | if (err == -ENOENT) |
444 | continue; | |
445 | if (err) | |
446 | goto out; | |
447 | ||
448 | while ((sym = elf_sym_iter_next(&iter))) { | |
449 | unsigned long sym_offset = elf_sym_offset(sym); | |
450 | int bind = GELF_ST_BIND(sym->sym.st_info); | |
451 | struct symbol *found, tmp = { | |
452 | .name = sym->name, | |
453 | }; | |
454 | unsigned long *offset; | |
455 | ||
456 | found = bsearch(&tmp, symbols, cnt, sizeof(*symbols), symbol_cmp); | |
457 | if (!found) | |
458 | continue; | |
459 | ||
460 | offset = &offsets[found->idx]; | |
461 | if (*offset > 0) { | |
462 | /* same offset, no problem */ | |
463 | if (*offset == sym_offset) | |
464 | continue; | |
465 | /* handle multiple matches */ | |
466 | if (found->bind != STB_WEAK && bind != STB_WEAK) { | |
467 | /* Only accept one non-weak bind. */ | |
468 | pr_warn("elf: ambiguous match found '%s@%lu' in '%s' previous offset %lu\n", | |
469 | sym->name, sym_offset, binary_path, *offset); | |
470 | err = -ESRCH; | |
471 | goto out; | |
472 | } else if (bind == STB_WEAK) { | |
473 | /* already have a non-weak bind, and | |
474 | * this is a weak bind, so ignore. | |
475 | */ | |
476 | continue; | |
477 | } | |
478 | } else { | |
479 | cnt_done++; | |
480 | } | |
481 | *offset = sym_offset; | |
482 | found->bind = bind; | |
483 | } | |
484 | } | |
485 | ||
486 | if (cnt != cnt_done) { | |
487 | err = -ENOENT; | |
488 | goto out; | |
489 | } | |
490 | ||
491 | *poffsets = offsets; | |
492 | ||
493 | out: | |
494 | free(symbols); | |
495 | if (err) | |
496 | free(offsets); | |
497 | elf_close(&elf_fd); | |
498 | return err; | |
499 | } | |
e613d1d0 JO |
500 | |
501 | /* | |
502 | * Return offsets in @poffsets for symbols specified by @pattern argument. | |
503 | * On success returns 0 and offsets are returned in allocated @poffsets | |
504 | * array with the @pctn size, that needs to be released by the caller. | |
505 | */ | |
506 | int elf_resolve_pattern_offsets(const char *binary_path, const char *pattern, | |
507 | unsigned long **poffsets, size_t *pcnt) | |
508 | { | |
509 | int sh_types[2] = { SHT_SYMTAB, SHT_DYNSYM }; | |
510 | unsigned long *offsets = NULL; | |
511 | size_t cap = 0, cnt = 0; | |
512 | struct elf_fd elf_fd; | |
513 | int err = 0, i; | |
514 | ||
515 | err = elf_open(binary_path, &elf_fd); | |
516 | if (err) | |
517 | return err; | |
518 | ||
519 | for (i = 0; i < ARRAY_SIZE(sh_types); i++) { | |
520 | struct elf_sym_iter iter; | |
521 | struct elf_sym *sym; | |
522 | ||
523 | err = elf_sym_iter_new(&iter, elf_fd.elf, binary_path, sh_types[i], STT_FUNC); | |
524 | if (err == -ENOENT) | |
525 | continue; | |
526 | if (err) | |
527 | goto out; | |
528 | ||
529 | while ((sym = elf_sym_iter_next(&iter))) { | |
530 | if (!glob_match(sym->name, pattern)) | |
531 | continue; | |
532 | ||
533 | err = libbpf_ensure_mem((void **) &offsets, &cap, sizeof(*offsets), | |
534 | cnt + 1); | |
535 | if (err) | |
536 | goto out; | |
537 | ||
538 | offsets[cnt++] = elf_sym_offset(sym); | |
539 | } | |
540 | ||
541 | /* If we found anything in the first symbol section, | |
542 | * do not search others to avoid duplicates. | |
543 | */ | |
544 | if (cnt) | |
545 | break; | |
546 | } | |
547 | ||
548 | if (cnt) { | |
549 | *poffsets = offsets; | |
550 | *pcnt = cnt; | |
551 | } else { | |
552 | err = -ENOENT; | |
553 | } | |
554 | ||
555 | out: | |
556 | if (err) | |
557 | free(offsets); | |
558 | elf_close(&elf_fd); | |
559 | return err; | |
560 | } |