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