1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2
3 #ifndef _GNU_SOURCE
4 #define _GNU_SOURCE
5 #endif
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
14 /* A SHT_GNU_versym section holds 16-bit words. This bit is set if
15 * the symbol is hidden and can only be seen when referenced using an
16 * explicit version number. This is a GNU extension.
17 */
18 #define VERSYM_HIDDEN 0x8000
19
20 /* This is the mask for the rest of the data in a word read from a
21 * SHT_GNU_versym section.
22 */
23 #define VERSYM_VERSION 0x7fff
24
elf_open(const char * binary_path,struct elf_fd * elf_fd)25 int elf_open(const char *binary_path, struct elf_fd *elf_fd)
26 {
27 char errmsg[STRERR_BUFSIZE];
28 int fd, ret;
29 Elf *elf;
30
31 elf_fd->elf = NULL;
32 elf_fd->fd = -1;
33
34 if (elf_version(EV_CURRENT) == EV_NONE) {
35 pr_warn("elf: failed to init libelf for %s\n", binary_path);
36 return -LIBBPF_ERRNO__LIBELF;
37 }
38 fd = open(binary_path, O_RDONLY | O_CLOEXEC);
39 if (fd < 0) {
40 ret = -errno;
41 pr_warn("elf: failed to open %s: %s\n", binary_path,
42 libbpf_strerror_r(ret, errmsg, sizeof(errmsg)));
43 return ret;
44 }
45 elf = elf_begin(fd, ELF_C_READ_MMAP, NULL);
46 if (!elf) {
47 pr_warn("elf: could not read elf from %s: %s\n", binary_path, elf_errmsg(-1));
48 close(fd);
49 return -LIBBPF_ERRNO__FORMAT;
50 }
51 elf_fd->fd = fd;
52 elf_fd->elf = elf;
53 return 0;
54 }
55
elf_close(struct elf_fd * elf_fd)56 void elf_close(struct elf_fd *elf_fd)
57 {
58 if (!elf_fd)
59 return;
60 elf_end(elf_fd->elf);
61 close(elf_fd->fd);
62 }
63
64 /* Return next ELF section of sh_type after scn, or first of that type if scn is NULL. */
elf_find_next_scn_by_type(Elf * elf,int sh_type,Elf_Scn * scn)65 static Elf_Scn *elf_find_next_scn_by_type(Elf *elf, int sh_type, Elf_Scn *scn)
66 {
67 while ((scn = elf_nextscn(elf, scn)) != NULL) {
68 GElf_Shdr sh;
69
70 if (!gelf_getshdr(scn, &sh))
71 continue;
72 if (sh.sh_type == sh_type)
73 return scn;
74 }
75 return NULL;
76 }
77
78 struct elf_sym {
79 const char *name;
80 GElf_Sym sym;
81 GElf_Shdr sh;
82 int ver;
83 bool hidden;
84 };
85
86 struct elf_sym_iter {
87 Elf *elf;
88 Elf_Data *syms;
89 Elf_Data *versyms;
90 Elf_Data *verdefs;
91 size_t nr_syms;
92 size_t strtabidx;
93 size_t verdef_strtabidx;
94 size_t next_sym_idx;
95 struct elf_sym sym;
96 int st_type;
97 };
98
elf_sym_iter_new(struct elf_sym_iter * iter,Elf * elf,const char * binary_path,int sh_type,int st_type)99 static int elf_sym_iter_new(struct elf_sym_iter *iter,
100 Elf *elf, const char *binary_path,
101 int sh_type, int st_type)
102 {
103 Elf_Scn *scn = NULL;
104 GElf_Ehdr ehdr;
105 GElf_Shdr sh;
106
107 memset(iter, 0, sizeof(*iter));
108
109 if (!gelf_getehdr(elf, &ehdr)) {
110 pr_warn("elf: failed to get ehdr from %s: %s\n", binary_path, elf_errmsg(-1));
111 return -EINVAL;
112 }
113
114 scn = elf_find_next_scn_by_type(elf, sh_type, NULL);
115 if (!scn) {
116 pr_debug("elf: failed to find symbol table ELF sections in '%s'\n",
117 binary_path);
118 return -ENOENT;
119 }
120
121 if (!gelf_getshdr(scn, &sh))
122 return -EINVAL;
123
124 iter->strtabidx = sh.sh_link;
125 iter->syms = elf_getdata(scn, 0);
126 if (!iter->syms) {
127 pr_warn("elf: failed to get symbols for symtab section in '%s': %s\n",
128 binary_path, elf_errmsg(-1));
129 return -EINVAL;
130 }
131 iter->nr_syms = iter->syms->d_size / sh.sh_entsize;
132 iter->elf = elf;
133 iter->st_type = st_type;
134
135 /* Version symbol table is meaningful to dynsym only */
136 if (sh_type != SHT_DYNSYM)
137 return 0;
138
139 scn = elf_find_next_scn_by_type(elf, SHT_GNU_versym, NULL);
140 if (!scn)
141 return 0;
142 iter->versyms = elf_getdata(scn, 0);
143
144 scn = elf_find_next_scn_by_type(elf, SHT_GNU_verdef, NULL);
145 if (!scn)
146 return 0;
147
148 iter->verdefs = elf_getdata(scn, 0);
149 if (!iter->verdefs || !gelf_getshdr(scn, &sh)) {
150 pr_warn("elf: failed to get verdef ELF section in '%s'\n", binary_path);
151 return -EINVAL;
152 }
153 iter->verdef_strtabidx = sh.sh_link;
154
155 return 0;
156 }
157
elf_sym_iter_next(struct elf_sym_iter * iter)158 static struct elf_sym *elf_sym_iter_next(struct elf_sym_iter *iter)
159 {
160 struct elf_sym *ret = &iter->sym;
161 GElf_Sym *sym = &ret->sym;
162 const char *name = NULL;
163 GElf_Versym versym;
164 Elf_Scn *sym_scn;
165 size_t idx;
166
167 for (idx = iter->next_sym_idx; idx < iter->nr_syms; idx++) {
168 if (!gelf_getsym(iter->syms, idx, sym))
169 continue;
170 if (GELF_ST_TYPE(sym->st_info) != iter->st_type)
171 continue;
172 name = elf_strptr(iter->elf, iter->strtabidx, sym->st_name);
173 if (!name)
174 continue;
175 sym_scn = elf_getscn(iter->elf, sym->st_shndx);
176 if (!sym_scn)
177 continue;
178 if (!gelf_getshdr(sym_scn, &ret->sh))
179 continue;
180
181 iter->next_sym_idx = idx + 1;
182 ret->name = name;
183 ret->ver = 0;
184 ret->hidden = false;
185
186 if (iter->versyms) {
187 if (!gelf_getversym(iter->versyms, idx, &versym))
188 continue;
189 ret->ver = versym & VERSYM_VERSION;
190 ret->hidden = versym & VERSYM_HIDDEN;
191 }
192 return ret;
193 }
194
195 return NULL;
196 }
197
elf_get_vername(struct elf_sym_iter * iter,int ver)198 static const char *elf_get_vername(struct elf_sym_iter *iter, int ver)
199 {
200 GElf_Verdaux verdaux;
201 GElf_Verdef verdef;
202 int offset;
203
204 if (!iter->verdefs)
205 return NULL;
206
207 offset = 0;
208 while (gelf_getverdef(iter->verdefs, offset, &verdef)) {
209 if (verdef.vd_ndx != ver) {
210 if (!verdef.vd_next)
211 break;
212
213 offset += verdef.vd_next;
214 continue;
215 }
216
217 if (!gelf_getverdaux(iter->verdefs, offset + verdef.vd_aux, &verdaux))
218 break;
219
220 return elf_strptr(iter->elf, iter->verdef_strtabidx, verdaux.vda_name);
221
222 }
223 return NULL;
224 }
225
symbol_match(struct elf_sym_iter * iter,int sh_type,struct elf_sym * sym,const char * name,size_t name_len,const char * lib_ver)226 static bool symbol_match(struct elf_sym_iter *iter, int sh_type, struct elf_sym *sym,
227 const char *name, size_t name_len, const char *lib_ver)
228 {
229 const char *ver_name;
230
231 /* Symbols are in forms of func, func@LIB_VER or func@@LIB_VER
232 * make sure the func part matches the user specified name
233 */
234 if (strncmp(sym->name, name, name_len) != 0)
235 return false;
236
237 /* ...but we don't want a search for "foo" to match 'foo2" also, so any
238 * additional characters in sname should be of the form "@@LIB".
239 */
240 if (sym->name[name_len] != '\0' && sym->name[name_len] != '@')
241 return false;
242
243 /* If user does not specify symbol version, then we got a match */
244 if (!lib_ver)
245 return true;
246
247 /* If user specifies symbol version, for dynamic symbols,
248 * get version name from ELF verdef section for comparison.
249 */
250 if (sh_type == SHT_DYNSYM) {
251 ver_name = elf_get_vername(iter, sym->ver);
252 if (!ver_name)
253 return false;
254 return strcmp(ver_name, lib_ver) == 0;
255 }
256
257 /* For normal symbols, it is already in form of func@LIB_VER */
258 return strcmp(sym->name, name) == 0;
259 }
260
261 /* Transform symbol's virtual address (absolute for binaries and relative
262 * for shared libs) into file offset, which is what kernel is expecting
263 * for uprobe/uretprobe attachment.
264 * See Documentation/trace/uprobetracer.rst for more details. This is done
265 * by looking up symbol's containing section's header and using iter's virtual
266 * address (sh_addr) and corresponding file offset (sh_offset) to transform
267 * sym.st_value (virtual address) into desired final file offset.
268 */
elf_sym_offset(struct elf_sym * sym)269 static unsigned long elf_sym_offset(struct elf_sym *sym)
270 {
271 return sym->sym.st_value - sym->sh.sh_addr + sym->sh.sh_offset;
272 }
273
274 /* Find offset of function name in the provided ELF object. "binary_path" is
275 * the path to the ELF binary represented by "elf", and only used for error
276 * reporting matters. "name" matches symbol name or name@@LIB for library
277 * functions.
278 */
elf_find_func_offset(Elf * elf,const char * binary_path,const char * name)279 long elf_find_func_offset(Elf *elf, const char *binary_path, const char *name)
280 {
281 int i, sh_types[2] = { SHT_DYNSYM, SHT_SYMTAB };
282 const char *at_symbol, *lib_ver;
283 bool is_shared_lib;
284 long ret = -ENOENT;
285 size_t name_len;
286 GElf_Ehdr ehdr;
287
288 if (!gelf_getehdr(elf, &ehdr)) {
289 pr_warn("elf: failed to get ehdr from %s: %s\n", binary_path, elf_errmsg(-1));
290 ret = -LIBBPF_ERRNO__FORMAT;
291 goto out;
292 }
293 /* for shared lib case, we do not need to calculate relative offset */
294 is_shared_lib = ehdr.e_type == ET_DYN;
295
296 /* Does name specify "@@LIB_VER" or "@LIB_VER" ? */
297 at_symbol = strchr(name, '@');
298 if (at_symbol) {
299 name_len = at_symbol - name;
300 /* skip second @ if it's @@LIB_VER case */
301 if (at_symbol[1] == '@')
302 at_symbol++;
303 lib_ver = at_symbol + 1;
304 } else {
305 name_len = strlen(name);
306 lib_ver = NULL;
307 }
308
309 /* Search SHT_DYNSYM, SHT_SYMTAB for symbol. This search order is used because if
310 * a binary is stripped, it may only have SHT_DYNSYM, and a fully-statically
311 * linked binary may not have SHT_DYMSYM, so absence of a section should not be
312 * reported as a warning/error.
313 */
314 for (i = 0; i < ARRAY_SIZE(sh_types); i++) {
315 struct elf_sym_iter iter;
316 struct elf_sym *sym;
317 int last_bind = -1;
318 int cur_bind;
319
320 ret = elf_sym_iter_new(&iter, elf, binary_path, sh_types[i], STT_FUNC);
321 if (ret == -ENOENT)
322 continue;
323 if (ret)
324 goto out;
325
326 while ((sym = elf_sym_iter_next(&iter))) {
327 if (!symbol_match(&iter, sh_types[i], sym, name, name_len, lib_ver))
328 continue;
329
330 cur_bind = GELF_ST_BIND(sym->sym.st_info);
331
332 if (ret > 0) {
333 /* handle multiple matches */
334 if (elf_sym_offset(sym) == ret) {
335 /* same offset, no problem */
336 continue;
337 } else if (last_bind != STB_WEAK && cur_bind != STB_WEAK) {
338 /* Only accept one non-weak bind. */
339 pr_warn("elf: ambiguous match for '%s', '%s' in '%s'\n",
340 sym->name, name, binary_path);
341 ret = -LIBBPF_ERRNO__FORMAT;
342 goto out;
343 } else if (cur_bind == STB_WEAK) {
344 /* already have a non-weak bind, and
345 * this is a weak bind, so ignore.
346 */
347 continue;
348 }
349 }
350
351 ret = elf_sym_offset(sym);
352 last_bind = cur_bind;
353 }
354 if (ret > 0)
355 break;
356 }
357
358 if (ret > 0) {
359 pr_debug("elf: symbol address match for '%s' in '%s': 0x%lx\n", name, binary_path,
360 ret);
361 } else {
362 if (ret == 0) {
363 pr_warn("elf: '%s' is 0 in symtab for '%s': %s\n", name, binary_path,
364 is_shared_lib ? "should not be 0 in a shared library" :
365 "try using shared library path instead");
366 ret = -ENOENT;
367 } else {
368 pr_warn("elf: failed to find symbol '%s' in '%s'\n", name, binary_path);
369 }
370 }
371 out:
372 return ret;
373 }
374
375 /* Find offset of function name in ELF object specified by path. "name" matches
376 * symbol name or name@@LIB for library functions.
377 */
elf_find_func_offset_from_file(const char * binary_path,const char * name)378 long elf_find_func_offset_from_file(const char *binary_path, const char *name)
379 {
380 struct elf_fd elf_fd;
381 long ret = -ENOENT;
382
383 ret = elf_open(binary_path, &elf_fd);
384 if (ret)
385 return ret;
386 ret = elf_find_func_offset(elf_fd.elf, binary_path, name);
387 elf_close(&elf_fd);
388 return ret;
389 }
390
391 struct symbol {
392 const char *name;
393 int bind;
394 int idx;
395 };
396
symbol_cmp(const void * a,const void * b)397 static int symbol_cmp(const void *a, const void *b)
398 {
399 const struct symbol *sym_a = a;
400 const struct symbol *sym_b = b;
401
402 return strcmp(sym_a->name, sym_b->name);
403 }
404
405 /*
406 * Return offsets in @poffsets for symbols specified in @syms array argument.
407 * On success returns 0 and offsets are returned in allocated array with @cnt
408 * size, that needs to be released by the caller.
409 */
elf_resolve_syms_offsets(const char * binary_path,int cnt,const char ** syms,unsigned long ** poffsets,int st_type)410 int elf_resolve_syms_offsets(const char *binary_path, int cnt,
411 const char **syms, unsigned long **poffsets,
412 int st_type)
413 {
414 int sh_types[2] = { SHT_DYNSYM, SHT_SYMTAB };
415 int err = 0, i, cnt_done = 0;
416 unsigned long *offsets;
417 struct symbol *symbols;
418 struct elf_fd elf_fd;
419
420 err = elf_open(binary_path, &elf_fd);
421 if (err)
422 return err;
423
424 offsets = calloc(cnt, sizeof(*offsets));
425 symbols = calloc(cnt, sizeof(*symbols));
426
427 if (!offsets || !symbols) {
428 err = -ENOMEM;
429 goto out;
430 }
431
432 for (i = 0; i < cnt; i++) {
433 symbols[i].name = syms[i];
434 symbols[i].idx = i;
435 }
436
437 qsort(symbols, cnt, sizeof(*symbols), symbol_cmp);
438
439 for (i = 0; i < ARRAY_SIZE(sh_types); i++) {
440 struct elf_sym_iter iter;
441 struct elf_sym *sym;
442
443 err = elf_sym_iter_new(&iter, elf_fd.elf, binary_path, sh_types[i], st_type);
444 if (err == -ENOENT)
445 continue;
446 if (err)
447 goto out;
448
449 while ((sym = elf_sym_iter_next(&iter))) {
450 unsigned long sym_offset = elf_sym_offset(sym);
451 int bind = GELF_ST_BIND(sym->sym.st_info);
452 struct symbol *found, tmp = {
453 .name = sym->name,
454 };
455 unsigned long *offset;
456
457 found = bsearch(&tmp, symbols, cnt, sizeof(*symbols), symbol_cmp);
458 if (!found)
459 continue;
460
461 offset = &offsets[found->idx];
462 if (*offset > 0) {
463 /* same offset, no problem */
464 if (*offset == sym_offset)
465 continue;
466 /* handle multiple matches */
467 if (found->bind != STB_WEAK && bind != STB_WEAK) {
468 /* Only accept one non-weak bind. */
469 pr_warn("elf: ambiguous match found '%s@%lu' in '%s' previous offset %lu\n",
470 sym->name, sym_offset, binary_path, *offset);
471 err = -ESRCH;
472 goto out;
473 } else if (bind == STB_WEAK) {
474 /* already have a non-weak bind, and
475 * this is a weak bind, so ignore.
476 */
477 continue;
478 }
479 } else {
480 cnt_done++;
481 }
482 *offset = sym_offset;
483 found->bind = bind;
484 }
485 }
486
487 if (cnt != cnt_done) {
488 err = -ENOENT;
489 goto out;
490 }
491
492 *poffsets = offsets;
493
494 out:
495 free(symbols);
496 if (err)
497 free(offsets);
498 elf_close(&elf_fd);
499 return err;
500 }
501
502 /*
503 * Return offsets in @poffsets for symbols specified by @pattern argument.
504 * On success returns 0 and offsets are returned in allocated @poffsets
505 * array with the @pctn size, that needs to be released by the caller.
506 */
elf_resolve_pattern_offsets(const char * binary_path,const char * pattern,unsigned long ** poffsets,size_t * pcnt)507 int elf_resolve_pattern_offsets(const char *binary_path, const char *pattern,
508 unsigned long **poffsets, size_t *pcnt)
509 {
510 int sh_types[2] = { SHT_SYMTAB, SHT_DYNSYM };
511 unsigned long *offsets = NULL;
512 size_t cap = 0, cnt = 0;
513 struct elf_fd elf_fd;
514 int err = 0, i;
515
516 err = elf_open(binary_path, &elf_fd);
517 if (err)
518 return err;
519
520 for (i = 0; i < ARRAY_SIZE(sh_types); i++) {
521 struct elf_sym_iter iter;
522 struct elf_sym *sym;
523
524 err = elf_sym_iter_new(&iter, elf_fd.elf, binary_path, sh_types[i], STT_FUNC);
525 if (err == -ENOENT)
526 continue;
527 if (err)
528 goto out;
529
530 while ((sym = elf_sym_iter_next(&iter))) {
531 if (!glob_match(sym->name, pattern))
532 continue;
533
534 err = libbpf_ensure_mem((void **) &offsets, &cap, sizeof(*offsets),
535 cnt + 1);
536 if (err)
537 goto out;
538
539 offsets[cnt++] = elf_sym_offset(sym);
540 }
541
542 /* If we found anything in the first symbol section,
543 * do not search others to avoid duplicates.
544 */
545 if (cnt)
546 break;
547 }
548
549 if (cnt) {
550 *poffsets = offsets;
551 *pcnt = cnt;
552 } else {
553 err = -ENOENT;
554 }
555
556 out:
557 if (err)
558 free(offsets);
559 elf_close(&elf_fd);
560 return err;
561 }
562