Re-organize sources
[babeltrace.git] / src / plugins / lttng-utils / debug-info / bin-info.c
1 /*
2 * bin-info.c
3 *
4 * Babeltrace - Executable and Shared Object Debug Info Reader
5 *
6 * Copyright 2015 Antoine Busque <abusque@efficios.com>
7 *
8 * Author: Antoine Busque <abusque@efficios.com>
9 *
10 * Permission is hereby granted, free of charge, to any person obtaining a copy
11 * of this software and associated documentation files (the "Software"), to deal
12 * in the Software without restriction, including without limitation the rights
13 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14 * copies of the Software, and to permit persons to whom the Software is
15 * furnished to do so, subject to the following conditions:
16 *
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
19 *
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
23 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
26 * SOFTWARE.
27 */
28
29 #define BT_LOG_TAG "PLUGIN-CTF-LTTNG-UTILS-DEBUG-INFO-FLT-BIN-INFO"
30 #include "logging.h"
31
32 #include <dwarf.h>
33 #include <errno.h>
34 #include <fcntl.h>
35 #include <inttypes.h>
36 #include <libgen.h>
37 #include <math.h>
38 #include <stdio.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <unistd.h>
42
43 #include <glib.h>
44
45 #include "common/common.h"
46
47 #include "bin-info.h"
48 #include "crc32.h"
49 #include "dwarf.h"
50 #include "utils.h"
51
52 /*
53 * An address printed in hex is at most 20 bytes (16 for 64-bits +
54 * leading 0x + optional leading '+' if addr is an offset + null
55 * character).
56 */
57 #define ADDR_STR_LEN 20
58 #define BUILD_ID_NOTE_NAME "GNU"
59
60 BT_HIDDEN
61 int bin_info_init(void)
62 {
63 int ret = 0;
64
65 if (elf_version(EV_CURRENT) == EV_NONE) {
66 BT_LOGD("ELF library initialization failed: %s.",
67 elf_errmsg(-1));
68 ret = -1;
69 }
70
71 return ret;
72 }
73
74 BT_HIDDEN
75 struct bin_info *bin_info_create(struct bt_fd_cache *fdc, const char *path,
76 uint64_t low_addr, uint64_t memsz, bool is_pic,
77 const char *debug_info_dir, const char *target_prefix)
78 {
79 struct bin_info *bin = NULL;
80
81 BT_ASSERT(fdc);
82
83 if (!path) {
84 goto error;
85 }
86
87 bin = g_new0(struct bin_info, 1);
88 if (!bin) {
89 goto error;
90 }
91
92 if (target_prefix) {
93 bin->elf_path = g_build_filename(target_prefix, path, NULL);
94 } else {
95 bin->elf_path = g_strdup(path);
96 }
97
98 if (!bin->elf_path) {
99 goto error;
100 }
101
102 if (debug_info_dir) {
103 bin->debug_info_dir = g_strdup(debug_info_dir);
104 if (!bin->debug_info_dir) {
105 goto error;
106 }
107 }
108
109 bin->is_pic = is_pic;
110 bin->memsz = memsz;
111 bin->low_addr = low_addr;
112 bin->high_addr = bin->low_addr + bin->memsz;
113 bin->build_id = NULL;
114 bin->build_id_len = 0;
115 bin->file_build_id_matches = false;
116 bin->fd_cache = fdc;
117
118 return bin;
119
120 error:
121 bin_info_destroy(bin);
122 return NULL;
123 }
124
125 BT_HIDDEN
126 void bin_info_destroy(struct bin_info *bin)
127 {
128 if (!bin) {
129 return;
130 }
131
132 dwarf_end(bin->dwarf_info);
133
134 g_free(bin->debug_info_dir);
135 g_free(bin->elf_path);
136 g_free(bin->dwarf_path);
137 g_free(bin->build_id);
138 g_free(bin->dbg_link_filename);
139
140 elf_end(bin->elf_file);
141
142 bt_fd_cache_put_handle(bin->fd_cache, bin->elf_handle);
143 bt_fd_cache_put_handle(bin->fd_cache, bin->dwarf_handle);
144
145 g_free(bin);
146 }
147
148 /**
149 * Initialize the ELF file for a given executable.
150 *
151 * @param bin bin_info instance
152 * @returns 0 on success, negative value on error.
153 */
154 static
155 int bin_info_set_elf_file(struct bin_info *bin)
156 {
157 struct bt_fd_cache_handle *elf_handle = NULL;
158 Elf *elf_file = NULL;
159
160 if (!bin) {
161 goto error;
162 }
163
164 elf_handle = bt_fd_cache_get_handle(bin->fd_cache, bin->elf_path);
165 if (!elf_handle) {
166 BT_LOGD("Failed to open %s", bin->elf_path);
167 goto error;
168 }
169 bin->elf_handle = elf_handle;
170
171 elf_file = elf_begin(bt_fd_cache_handle_get_fd(bin->elf_handle),
172 ELF_C_READ, NULL);
173 if (!elf_file) {
174 BT_LOGE("elf_begin failed: %s", elf_errmsg(-1));
175 goto error;
176 }
177
178 bin->elf_file = elf_file;
179
180 if (elf_kind(elf_file) != ELF_K_ELF) {
181 BT_LOGE("Error: %s is not an ELF object", bin->elf_path);
182 goto error;
183 }
184
185 return 0;
186
187 error:
188 bt_fd_cache_put_handle(bin->fd_cache, elf_handle);
189 elf_end(elf_file);
190 return -1;
191 }
192
193 /**
194 * From a note section data struct, check if it is a build id note.
195 *
196 * @param note_data Pointer to a note section
197 *
198 * @returns 1 on match, 0 if `buf` does not contain a
199 * valid build id note
200 */
201 static
202 int is_build_id_note_section(Elf_Data *note_data)
203 {
204 size_t name_offset, desc_offset;
205 GElf_Nhdr note_header;
206 int ret = 0;
207
208 /*
209 * Discard the return value as it contains the size of the note section
210 * and we don't need it.
211 */
212 (void) gelf_getnote(note_data, 0, &note_header, &name_offset,
213 &desc_offset);
214
215 /*
216 * Check the note name length. The name_sz field includes the
217 * terminating null byte.
218 */
219 if (note_header.n_namesz != sizeof(BUILD_ID_NOTE_NAME)) {
220 goto invalid;
221 }
222
223 /* Check the note type. */
224 if (note_header.n_type != NT_GNU_BUILD_ID) {
225 goto invalid;
226 }
227
228 /* Check the note name. */
229 if (memcmp(note_data->d_buf + name_offset, BUILD_ID_NOTE_NAME,
230 note_header.n_namesz) != 0) {
231 goto invalid;
232 }
233
234 ret = 1;
235
236 invalid:
237 return ret;
238 }
239
240 /**
241 * From a build id note section data struct, check if the build id it contains
242 * is identical to the build id passed as parameter.
243 *
244 * @param note_data Pointer to the file build id note section.
245 * @param build_id Pointer to a build id to compare to.
246 * @param build_id_len length of the build id.
247 *
248 * @returns 1 on match, 0 otherwise.
249 */
250 static
251 int is_build_id_note_section_matching(Elf_Data *note_data,
252 uint8_t *build_id, size_t build_id_len)
253 {
254 size_t name_offset, desc_offset;
255 GElf_Nhdr note_header;
256
257 if (build_id_len <= 0) {
258 goto end;
259 }
260
261 /*
262 * Discard the return value as it contains the size of the note section
263 * and we don't need it.
264 */
265 (void) gelf_getnote(note_data, 0, &note_header, &name_offset,
266 &desc_offset);
267
268 /*
269 * Compare the binary build id with the supplied build id.
270 */
271 if (memcmp(build_id, note_data->d_buf + desc_offset,
272 build_id_len) == 0) {
273 return 1;
274 }
275 end:
276 return 0;
277 }
278
279 /**
280 * Checks if the build id stored in `bin` (bin->build_id) is matching the build
281 * id of the ondisk file (bin->elf_file).
282 *
283 * @param bin bin_info instance
284 * @param build_id build id to compare ot the on disk file
285 * @param build_id_len length of the build id
286 *
287 * @returns 1 on if the build id of stored in `bin` matches
288 * the build id of the ondisk file.
289 * 0 on if they are different or an error occured.
290 */
291 static
292 int is_build_id_matching(struct bin_info *bin)
293 {
294 int ret, is_build_id, is_matching = 0;
295 Elf_Scn *curr_section = NULL, *next_section = NULL;
296 GElf_Shdr curr_section_hdr;
297
298 if (!bin->build_id) {
299 goto error;
300 }
301
302 /* Set ELF file if it hasn't been accessed yet. */
303 if (!bin->elf_file) {
304 ret = bin_info_set_elf_file(bin);
305 if (ret) {
306 /* Failed to set ELF file. */
307 goto error;
308 }
309 }
310
311 next_section = elf_nextscn(bin->elf_file, curr_section);
312 if (!next_section) {
313 goto error;
314 }
315
316 while (next_section) {
317 Elf_Data *note_data = NULL;
318
319 curr_section = next_section;
320 next_section = elf_nextscn(bin->elf_file, curr_section);
321
322 if (!gelf_getshdr(curr_section, &curr_section_hdr)) {
323 goto error;
324 }
325
326 if (curr_section_hdr.sh_type != SHT_NOTE) {
327 continue;
328 }
329
330 /*
331 * elf_getdata() translates the data to native byte order.
332 */
333 note_data = elf_getdata(curr_section, NULL);
334 if (!note_data) {
335 goto error;
336 }
337
338 /* Check if the note is of the build-id type. */
339 is_build_id = is_build_id_note_section(note_data);
340 if (!is_build_id) {
341 continue;
342 }
343
344 /*
345 * Compare the build id of the on-disk file and
346 * the build id recorded in the trace.
347 */
348 is_matching = is_build_id_note_section_matching(
349 note_data, bin->build_id, bin->build_id_len);
350 if (!is_matching) {
351 break;
352 }
353 }
354 error:
355 return is_matching;
356 }
357
358 BT_HIDDEN
359 int bin_info_set_build_id(struct bin_info *bin, uint8_t *build_id,
360 size_t build_id_len)
361 {
362 if (!bin || !build_id) {
363 goto error;
364 }
365
366 /* Set the build id. */
367 bin->build_id = g_new0(uint8_t, build_id_len);
368 if (!bin->build_id) {
369 goto error;
370 }
371
372 memcpy(bin->build_id, build_id, build_id_len);
373 bin->build_id_len = build_id_len;
374
375 /*
376 * Check if the file found on the file system has the same build id
377 * that what was recorded in the trace.
378 */
379 bin->file_build_id_matches = is_build_id_matching(bin);
380 if (!bin->file_build_id_matches) {
381 BT_LOGD_STR("Supplied Build ID does not match Build ID of the "
382 "binary or library found on the file system.");
383 goto error;
384 }
385
386 /*
387 * Reset the is_elf_only flag in case it had been set
388 * previously, because we might find separate debug info using
389 * the new build id information.
390 */
391 bin->is_elf_only = false;
392
393 return 0;
394
395 error:
396 return -1;
397 }
398
399 BT_HIDDEN
400 int bin_info_set_debug_link(struct bin_info *bin, const char *filename,
401 uint32_t crc)
402 {
403 if (!bin || !filename) {
404 goto error;
405 }
406
407 bin->dbg_link_filename = g_strdup(filename);
408 if (!bin->dbg_link_filename) {
409 goto error;
410 }
411
412 bin->dbg_link_crc = crc;
413
414 /*
415 * Reset the is_elf_only flag in case it had been set
416 * previously, because we might find separate debug info using
417 * the new build id information.
418 */
419 bin->is_elf_only = false;
420
421 return 0;
422
423 error:
424
425 return -1;
426 }
427
428 /**
429 * Tries to read DWARF info from the location given by path, and
430 * attach it to the given bin_info instance if it exists.
431 *
432 * @param bin bin_info instance for which to set DWARF info
433 * @param path Presumed location of the DWARF info
434 * @returns 0 on success, negative value on failure
435 */
436 static
437 int bin_info_set_dwarf_info_from_path(struct bin_info *bin, char *path)
438 {
439 int ret = 0;
440 struct bt_fd_cache_handle *dwarf_handle = NULL;
441 struct bt_dwarf_cu *cu = NULL;
442 Dwarf *dwarf_info = NULL;
443
444 if (!bin || !path) {
445 goto error;
446 }
447
448 dwarf_handle = bt_fd_cache_get_handle(bin->fd_cache, path);
449 if (!dwarf_handle) {
450 goto error;
451 }
452
453 dwarf_info = dwarf_begin(bt_fd_cache_handle_get_fd(dwarf_handle),
454 DWARF_C_READ);
455 if (!dwarf_info) {
456 goto error;
457 }
458
459 /*
460 * Check if the dwarf info has any CU. If not, the
461 * executable's object file contains no DWARF info.
462 */
463 cu = bt_dwarf_cu_create(dwarf_info);
464 if (!cu) {
465 goto error;
466 }
467
468 ret = bt_dwarf_cu_next(cu);
469 if (ret) {
470 goto error;
471 }
472
473 bin->dwarf_handle = dwarf_handle;
474 bin->dwarf_path = g_strdup(path);
475 if (!bin->dwarf_path) {
476 goto error;
477 }
478 bin->dwarf_info = dwarf_info;
479 free(cu);
480
481 return 0;
482
483 error:
484 bt_fd_cache_put_handle(bin->fd_cache, dwarf_handle);
485 dwarf_end(dwarf_info);
486 g_free(dwarf_info);
487 free(cu);
488
489 return -1;
490 }
491
492 /**
493 * Try to set the dwarf_info for a given bin_info instance via the
494 * build ID method.
495 *
496 * @param bin bin_info instance for which to retrieve the
497 * DWARF info via build ID
498 * @returns 0 on success (i.e. dwarf_info set), -1 on failure
499 */
500 static
501 int bin_info_set_dwarf_info_build_id(struct bin_info *bin)
502 {
503 int i = 0, ret = 0;
504 char *path = NULL, *build_id_prefix_dir = NULL, *build_id_file = NULL;
505 const char *dbg_dir = NULL;
506 size_t build_id_char_len, build_id_suffix_char_len, build_id_file_len;
507
508 if (!bin || !bin->build_id) {
509 goto error;
510 }
511
512 dbg_dir = bin->debug_info_dir ? bin->debug_info_dir : DEFAULT_DEBUG_DIR;
513
514 /*
515 * The prefix dir is the first byte of the build id, represented in
516 * lowercase hex as two characters per byte, +1 for '\0'.
517 */
518 build_id_prefix_dir = g_new0(gchar, BUILD_ID_PREFIX_DIR_LEN + 1);
519 if (!build_id_prefix_dir) {
520 goto error;
521 }
522 g_snprintf(build_id_prefix_dir, BUILD_ID_PREFIX_DIR_LEN + 1, "%02x", bin->build_id[0]);
523
524 /*
525 * The build id file is the remaining bytes of the build id,
526 * represented in lowercase hex, as two characters per byte.
527 */
528 build_id_char_len = (2 * (bin->build_id_len - 1));
529
530 /* To which the build id suffix is added, +1 for '\0'. */
531 build_id_suffix_char_len = strlen(BUILD_ID_SUFFIX) + 1;
532
533 /*
534 * The resulting filename string is the concatenation of the
535 * hex build id and the suffix.
536 */
537 build_id_file_len = build_id_char_len + build_id_suffix_char_len;
538 build_id_file = g_new0(gchar, build_id_file_len);
539 if (!build_id_file) {
540 goto error;
541 }
542
543 /*
544 * For each byte, starting at offset 1, append two characters
545 * in lowercase hex.
546 */
547 for (i = 1; i < bin->build_id_len; ++i) {
548 int path_idx = 2 * (i - 1);
549
550 g_snprintf(&build_id_file[path_idx], 3, "%02x", bin->build_id[i]);
551 }
552 /* Append the suffix to the generated string, including the '\0'. */
553 g_snprintf(&build_id_file[build_id_char_len], build_id_suffix_char_len,
554 BUILD_ID_SUFFIX);
555
556 path = g_build_filename(dbg_dir, BUILD_ID_SUBDIR, build_id_prefix_dir, build_id_file, NULL);
557 if (!path) {
558 goto error;
559 }
560
561 ret = bin_info_set_dwarf_info_from_path(bin, path);
562 if (ret) {
563 goto error;
564 }
565
566 goto end;
567
568 error:
569 ret = -1;
570 end:
571 g_free(build_id_prefix_dir);
572 g_free(build_id_file);
573 g_free(path);
574
575 return ret;
576 }
577
578 /**
579 * Tests whether the file located at path exists and has the expected
580 * checksum.
581 *
582 * This predicate is used when looking up separate debug info via the
583 * GNU debuglink method. The expected crc can be found .gnu_debuglink
584 * section in the original ELF file, along with the filename for the
585 * file containing the debug info.
586 *
587 * @param path Full path at which to look for the debug file
588 * @param crc Expected checksum for the debug file
589 * @returns 1 if the file exists and has the correct checksum,
590 * 0 otherwise
591 */
592 static
593 int is_valid_debug_file(struct bin_info *bin, char *path, uint32_t crc)
594 {
595 int ret = 0;
596 struct bt_fd_cache_handle *debug_handle = NULL;
597 uint32_t _crc = 0;
598
599 if (!path) {
600 goto end;
601 }
602
603 debug_handle = bt_fd_cache_get_handle(bin->fd_cache, path);
604 if (!debug_handle) {
605 goto end;
606 }
607
608 ret = crc32(bt_fd_cache_handle_get_fd(debug_handle), &_crc);
609 if (ret) {
610 ret = 0;
611 goto end;
612 }
613
614 ret = (crc == _crc);
615
616 end:
617 bt_fd_cache_put_handle(bin->fd_cache, debug_handle);
618 return ret;
619 }
620
621 /**
622 * Try to set the dwarf_info for a given bin_info instance via the
623 * debug-link method.
624 *
625 * @param bin bin_info instance for which to retrieve the
626 * DWARF info via debug link
627 * @returns 0 on success (i.e. dwarf_info set), -1 on failure
628 */
629 static
630 int bin_info_set_dwarf_info_debug_link(struct bin_info *bin)
631 {
632 int ret = 0;
633 const gchar *dbg_dir = NULL;
634 gchar *bin_dir = NULL, *path = NULL;
635
636 if (!bin || !bin->dbg_link_filename) {
637 goto error;
638 }
639
640 dbg_dir = bin->debug_info_dir ? bin->debug_info_dir : DEFAULT_DEBUG_DIR;
641 bin_dir = g_path_get_dirname(bin->elf_path);
642
643 /* First look in the executable's dir */
644 path = g_build_filename(bin_dir, bin->dbg_link_filename, NULL);
645
646 if (is_valid_debug_file(bin, path, bin->dbg_link_crc)) {
647 goto found;
648 }
649
650 /* If not found, look in .debug subdir */
651 g_free(path);
652 path = g_build_filename(bin_dir, DEBUG_SUBDIR, bin->dbg_link_filename, NULL);
653
654 if (is_valid_debug_file(bin, path, bin->dbg_link_crc)) {
655 goto found;
656 }
657
658 /* Lastly, look under the global debug directory */
659 g_free(path);
660
661 path = g_build_filename(dbg_dir, bin_dir, bin->dbg_link_filename, NULL);
662 if (is_valid_debug_file(bin, path, bin->dbg_link_crc)) {
663 goto found;
664 }
665
666 error:
667 ret = -1;
668 end:
669 g_free(bin_dir);
670 g_free(path);
671
672 return ret;
673
674 found:
675 ret = bin_info_set_dwarf_info_from_path(bin, path);
676 if (ret) {
677 goto error;
678 }
679
680 goto end;
681 }
682
683 /**
684 * Initialize the DWARF info for a given executable.
685 *
686 * @param bin bin_info instance
687 * @returns 0 on success, negative value on failure
688 */
689 static
690 int bin_info_set_dwarf_info(struct bin_info *bin)
691 {
692 int ret = 0;
693
694 if (!bin) {
695 ret = -1;
696 goto end;
697 }
698
699 /* First try to set the DWARF info from the ELF file */
700 ret = bin_info_set_dwarf_info_from_path(bin, bin->elf_path);
701 if (!ret) {
702 goto end;
703 }
704
705 /*
706 * If that fails, try to find separate debug info via build ID
707 * and debug link.
708 */
709 ret = bin_info_set_dwarf_info_build_id(bin);
710 if (!ret) {
711 goto end;
712 }
713
714 ret = bin_info_set_dwarf_info_debug_link(bin);
715 if (!ret) {
716 goto end;
717 }
718
719 end:
720 return ret;
721 }
722
723 BT_HIDDEN
724 void source_location_destroy(struct source_location *src_loc)
725 {
726 if (!src_loc) {
727 return;
728 }
729
730 free(src_loc->filename);
731 g_free(src_loc);
732 }
733
734 /**
735 * Append a string representation of an address offset to an existing
736 * string.
737 *
738 * On success, the out parameter `result` will contain the base string
739 * followed by the offset string of the form "+0x1234". On failure,
740 * `result` remains unchanged.
741 *
742 * @param base_str The string to which to append an offset string
743 * @param low_addr The lower virtual memory address, the base from
744 * which the offset is computed
745 * @param high_addr The higher virtual memory address
746 * @param result Out parameter, the base string followed by the
747 * offset string
748 * @returns 0 on success, -1 on failure
749 */
750 static
751 int bin_info_append_offset_str(const char *base_str, uint64_t low_addr,
752 uint64_t high_addr, char **result)
753 {
754 uint64_t offset;
755 char *_result = NULL;
756
757 if (!base_str || !result) {
758 goto error;
759 }
760
761 offset = high_addr - low_addr;
762
763 _result = g_strdup_printf("%s+%#0" PRIx64, base_str, offset);
764 if (!_result) {
765 goto error;
766 }
767 *result = _result;
768
769 return 0;
770
771 error:
772 free(_result);
773 return -1;
774 }
775
776 /**
777 * Try to find the symbol closest to an address within a given ELF
778 * section.
779 *
780 * Only function symbols are taken into account. The symbol's address
781 * must precede `addr`. A symbol with a closer address might exist
782 * after `addr` but is irrelevant because it cannot encompass `addr`.
783 *
784 * On success, if found, the out parameters `sym` and `shdr` are
785 * set. On failure or if none are found, they remain unchanged.
786 *
787 * @param scn ELF section in which to look for the address
788 * @param addr Virtual memory address for which to find the
789 * nearest function symbol
790 * @param sym Out parameter, the nearest function symbol
791 * @param shdr Out parameter, the section header for scn
792 * @returns 0 on success, -1 on failure
793 */
794 static
795 int bin_info_get_nearest_symbol_from_section(Elf_Scn *scn, uint64_t addr,
796 GElf_Sym **sym, GElf_Shdr **shdr)
797 {
798 int i;
799 size_t symbol_count;
800 Elf_Data *data = NULL;
801 GElf_Shdr *_shdr = NULL;
802 GElf_Sym *nearest_sym = NULL;
803
804 if (!scn || !sym || !shdr) {
805 goto error;
806 }
807
808 _shdr = g_new0(GElf_Shdr, 1);
809 if (!_shdr) {
810 goto error;
811 }
812
813 _shdr = gelf_getshdr(scn, _shdr);
814 if (!_shdr) {
815 goto error;
816 }
817
818 if (_shdr->sh_type != SHT_SYMTAB) {
819 /*
820 * We are only interested in symbol table (symtab)
821 * sections, skip this one.
822 */
823 goto end;
824 }
825
826 data = elf_getdata(scn, NULL);
827 if (!data) {
828 goto error;
829 }
830
831 symbol_count = _shdr->sh_size / _shdr->sh_entsize;
832
833 for (i = 0; i < symbol_count; ++i) {
834 GElf_Sym *cur_sym = NULL;
835
836 cur_sym = g_new0(GElf_Sym, 1);
837 if (!cur_sym) {
838 goto error;
839 }
840 cur_sym = gelf_getsym(data, i, cur_sym);
841 if (!cur_sym) {
842 goto error;
843 }
844 if (GELF_ST_TYPE(cur_sym->st_info) != STT_FUNC) {
845 /* We're only interested in the functions. */
846 g_free(cur_sym);
847 continue;
848 }
849
850 if (cur_sym->st_value <= addr &&
851 (!nearest_sym ||
852 cur_sym->st_value > nearest_sym->st_value)) {
853 g_free(nearest_sym);
854 nearest_sym = cur_sym;
855 } else {
856 g_free(cur_sym);
857 }
858 }
859
860 end:
861 if (nearest_sym) {
862 *sym = nearest_sym;
863 *shdr = _shdr;
864 } else {
865 g_free(_shdr);
866 }
867
868 return 0;
869
870 error:
871 g_free(nearest_sym);
872 g_free(_shdr);
873 return -1;
874 }
875
876 /**
877 * Get the name of the function containing a given address within an
878 * executable using ELF symbols.
879 *
880 * The function name is in fact the name of the nearest ELF symbol,
881 * followed by the offset in bytes between the address and the symbol
882 * (in hex), separated by a '+' character.
883 *
884 * If found, the out parameter `func_name` is set on success. On failure,
885 * it remains unchanged.
886 *
887 * @param bin bin_info instance for the executable containing
888 * the address
889 * @param addr Virtual memory address for which to find the
890 * function name
891 * @param func_name Out parameter, the function name
892 * @returns 0 on success, -1 on failure
893 */
894 static
895 int bin_info_lookup_elf_function_name(struct bin_info *bin, uint64_t addr,
896 char **func_name)
897 {
898 /*
899 * TODO (possible optimisation): if an ELF has no symtab
900 * section, it has been stripped. Therefore, it would be wise
901 * to store a flag indicating the stripped status after the
902 * first iteration to prevent subsequent ones.
903 */
904 int ret = 0;
905 Elf_Scn *scn = NULL;
906 GElf_Sym *sym = NULL;
907 GElf_Shdr *shdr = NULL;
908 char *sym_name = NULL;
909
910 /* Set ELF file if it hasn't been accessed yet. */
911 if (!bin->elf_file) {
912 ret = bin_info_set_elf_file(bin);
913 if (ret) {
914 /* Failed to set ELF file. */
915 goto error;
916 }
917 }
918
919 scn = elf_nextscn(bin->elf_file, scn);
920 if (!scn) {
921 goto error;
922 }
923
924 while (scn && !sym) {
925 ret = bin_info_get_nearest_symbol_from_section(
926 scn, addr, &sym, &shdr);
927 if (ret) {
928 goto error;
929 }
930
931 scn = elf_nextscn(bin->elf_file, scn);
932 }
933
934 if (sym) {
935 sym_name = elf_strptr(bin->elf_file, shdr->sh_link,
936 sym->st_name);
937 if (!sym_name) {
938 goto error;
939 }
940
941 ret = bin_info_append_offset_str(sym_name, sym->st_value, addr,
942 func_name);
943 if (ret) {
944 goto error;
945 }
946 }
947
948 g_free(shdr);
949 g_free(sym);
950 return 0;
951
952 error:
953 g_free(shdr);
954 g_free(sym);
955 return ret;
956 }
957
958 /**
959 * Get the name of the function containing a given address within a
960 * given compile unit (CU).
961 *
962 * If found, the out parameter `func_name` is set on success. On
963 * failure, it remains unchanged.
964 *
965 * @param cu bt_dwarf_cu instance which may contain the address
966 * @param addr Virtual memory address for which to find the
967 * function name
968 * @param func_name Out parameter, the function name
969 * @returns 0 on success, -1 on failure
970 */
971 static
972 int bin_info_lookup_cu_function_name(struct bt_dwarf_cu *cu, uint64_t addr,
973 char **func_name)
974 {
975 int ret = 0;
976 bool found = false;
977 struct bt_dwarf_die *die = NULL;
978
979 if (!cu || !func_name) {
980 goto error;
981 }
982
983 die = bt_dwarf_die_create(cu);
984 if (!die) {
985 goto error;
986 }
987
988 while (bt_dwarf_die_next(die) == 0) {
989 int tag;
990
991 ret = bt_dwarf_die_get_tag(die, &tag);
992 if (ret) {
993 goto error;
994 }
995
996 if (tag == DW_TAG_subprogram) {
997 ret = bt_dwarf_die_contains_addr(die, addr, &found);
998 if (ret) {
999 goto error;
1000 }
1001
1002 if (found) {
1003 break;
1004 }
1005 }
1006 }
1007
1008 if (found) {
1009 uint64_t low_addr = 0;
1010 char *die_name = NULL;
1011
1012 ret = bt_dwarf_die_get_name(die, &die_name);
1013 if (ret) {
1014 goto error;
1015 }
1016
1017 ret = dwarf_lowpc(die->dwarf_die, &low_addr);
1018 if (ret) {
1019 free(die_name);
1020 goto error;
1021 }
1022
1023 ret = bin_info_append_offset_str(die_name, low_addr, addr,
1024 func_name);
1025 free(die_name);
1026 if (ret) {
1027 goto error;
1028 }
1029 }
1030
1031 bt_dwarf_die_destroy(die);
1032 return 0;
1033
1034 error:
1035 bt_dwarf_die_destroy(die);
1036 return -1;
1037 }
1038
1039 /**
1040 * Get the name of the function containing a given address within an
1041 * executable using DWARF debug info.
1042 *
1043 * If found, the out parameter `func_name` is set on success. On
1044 * failure, it remains unchanged.
1045 *
1046 * @param bin bin_info instance for the executable containing
1047 * the address
1048 * @param addr Virtual memory address for which to find the
1049 * function name
1050 * @param func_name Out parameter, the function name
1051 * @returns 0 on success, -1 on failure
1052 */
1053 static
1054 int bin_info_lookup_dwarf_function_name(struct bin_info *bin, uint64_t addr,
1055 char **func_name)
1056 {
1057 int ret = 0;
1058 char *_func_name = NULL;
1059 struct bt_dwarf_cu *cu = NULL;
1060
1061 if (!bin || !func_name) {
1062 goto error;
1063 }
1064
1065 cu = bt_dwarf_cu_create(bin->dwarf_info);
1066 if (!cu) {
1067 goto error;
1068 }
1069
1070 while (bt_dwarf_cu_next(cu) == 0) {
1071 ret = bin_info_lookup_cu_function_name(cu, addr, &_func_name);
1072 if (ret) {
1073 goto error;
1074 }
1075
1076 if (_func_name) {
1077 break;
1078 }
1079 }
1080
1081 if (_func_name) {
1082 *func_name = _func_name;
1083 } else {
1084 goto error;
1085 }
1086
1087 bt_dwarf_cu_destroy(cu);
1088 return 0;
1089
1090 error:
1091 bt_dwarf_cu_destroy(cu);
1092 return -1;
1093 }
1094
1095 BT_HIDDEN
1096 int bin_info_lookup_function_name(struct bin_info *bin,
1097 uint64_t addr, char **func_name)
1098 {
1099 int ret = 0;
1100 char *_func_name = NULL;
1101
1102 if (!bin || !func_name) {
1103 goto error;
1104 }
1105
1106 /*
1107 * If the bin_info has a build id but it does not match the build id
1108 * that was found on the file system, return an error.
1109 */
1110 if (bin->build_id && !bin->file_build_id_matches) {
1111 goto error;
1112 }
1113
1114 /* Set DWARF info if it hasn't been accessed yet. */
1115 if (!bin->dwarf_info && !bin->is_elf_only) {
1116 ret = bin_info_set_dwarf_info(bin);
1117 if (ret) {
1118 BT_LOGD_STR("Failed to set bin dwarf info, falling "
1119 "back to ELF lookup.");
1120 /* Failed to set DWARF info, fallback to ELF. */
1121 bin->is_elf_only = true;
1122 }
1123 }
1124
1125 if (!bin_info_has_address(bin, addr)) {
1126 goto error;
1127 }
1128
1129 /*
1130 * Addresses in ELF and DWARF are relative to base address for
1131 * PIC, so make the address argument relative too if needed.
1132 */
1133 if (bin->is_pic) {
1134 addr -= bin->low_addr;
1135 }
1136
1137 if (bin->is_elf_only) {
1138 ret = bin_info_lookup_elf_function_name(bin, addr,
1139 &_func_name);
1140 if (ret) {
1141 BT_LOGD("Failed to lookup function name (ELF): "
1142 "ret=%d", ret);
1143 }
1144 } else {
1145 ret = bin_info_lookup_dwarf_function_name(bin, addr,
1146 &_func_name);
1147 if (ret) {
1148 BT_LOGD("Failed to lookup function name (DWARF): "
1149 "ret=%d", ret);
1150 }
1151 }
1152
1153 *func_name = _func_name;
1154 return 0;
1155
1156 error:
1157 return -1;
1158 }
1159
1160 BT_HIDDEN
1161 int bin_info_get_bin_loc(struct bin_info *bin, uint64_t addr, char **bin_loc)
1162 {
1163 gchar *_bin_loc = NULL;
1164
1165 if (!bin || !bin_loc) {
1166 goto error;
1167 }
1168
1169 /*
1170 * If the bin_info has a build id but it does not match the build id
1171 * that was found on the file system, return an error.
1172 */
1173 if (bin->build_id && !bin->file_build_id_matches) {
1174 goto error;
1175 }
1176
1177 if (bin->is_pic) {
1178 addr -= bin->low_addr;
1179 _bin_loc = g_strdup_printf("+%#0" PRIx64, addr);
1180 } else {
1181 _bin_loc = g_strdup_printf("@%#0" PRIx64, addr);
1182 }
1183
1184 if (!_bin_loc) {
1185 goto error;
1186 }
1187
1188 *bin_loc = _bin_loc;
1189 return 0;
1190
1191 error:
1192 return -1;
1193 }
1194
1195 /**
1196 * Predicate used to determine whether the children of a given DIE
1197 * contain a specific address.
1198 *
1199 * More specifically, the parameter `die` is expected to be a
1200 * subprogram (function) DIE, and this predicate tells whether any
1201 * subroutines are inlined within this function and would contain
1202 * `addr`.
1203 *
1204 * On success, the out parameter `contains` is set with the boolean
1205 * value indicating whether the DIE's range covers `addr`. On failure,
1206 * it remains unchanged.
1207 *
1208 * Do note that this function advances the position of `die`. If the
1209 * address is found within one of its children, `die` will be pointing
1210 * to that child upon returning from the function, allowing to extract
1211 * the information deemed necessary.
1212 *
1213 * @param die The parent DIE in whose children the address will be
1214 * looked for
1215 * @param addr The address for which to look for in the DIEs
1216 * @param contains Out parameter, true if addr is contained,
1217 * false if not
1218 * @returns Returns 0 on success, -1 on failure
1219 */
1220 static
1221 int bin_info_child_die_has_address(struct bt_dwarf_die *die, uint64_t addr, bool *contains)
1222 {
1223 int ret = 0;
1224 bool _contains = false;
1225
1226 if (!die) {
1227 goto error;
1228 }
1229
1230 ret = bt_dwarf_die_child(die);
1231 if (ret) {
1232 goto error;
1233 }
1234
1235 do {
1236 ret = bt_dwarf_die_contains_addr(die, addr, &_contains);
1237 if (ret) {
1238 goto error;
1239 }
1240
1241 if (_contains) {
1242 /*
1243 * The address is within the range of the current DIE
1244 * or its children.
1245 */
1246 int tag;
1247
1248 ret = bt_dwarf_die_get_tag(die, &tag);
1249 if (ret) {
1250 goto error;
1251 }
1252
1253 if (tag == DW_TAG_inlined_subroutine) {
1254 /* Found the tracepoint. */
1255 goto end;
1256 }
1257
1258 if (bt_dwarf_die_has_children(die)) {
1259 /*
1260 * Look for the address in the children DIEs.
1261 */
1262 ret = bt_dwarf_die_child(die);
1263 if (ret) {
1264 goto error;
1265 }
1266 }
1267 }
1268 } while (bt_dwarf_die_next(die) == 0);
1269
1270 end:
1271 *contains = _contains;
1272 return 0;
1273
1274 error:
1275 return -1;
1276 }
1277
1278 /**
1279 * Lookup the source location for a given address within a CU, making
1280 * the assumption that it is contained within an inline routine in a
1281 * function.
1282 *
1283 * @param cu bt_dwarf_cu instance in which to look for the address
1284 * @param addr The address for which to look for
1285 * @param src_loc Out parameter, the source location (filename and
1286 * line number) for the address
1287 * @returns 0 on success, -1 on failure
1288 */
1289 static
1290 int bin_info_lookup_cu_src_loc_inl(struct bt_dwarf_cu *cu, uint64_t addr,
1291 struct source_location **src_loc)
1292 {
1293 int ret = 0;
1294 bool found = false;
1295 struct bt_dwarf_die *die = NULL;
1296 struct source_location *_src_loc = NULL;
1297
1298 if (!cu || !src_loc) {
1299 goto error;
1300 }
1301
1302 die = bt_dwarf_die_create(cu);
1303 if (!die) {
1304 goto error;
1305 }
1306
1307 while (bt_dwarf_die_next(die) == 0) {
1308 int tag;
1309
1310 ret = bt_dwarf_die_get_tag(die, &tag);
1311 if (ret) {
1312 goto error;
1313 }
1314
1315 if (tag == DW_TAG_subprogram) {
1316 bool contains = false;
1317
1318 ret = bt_dwarf_die_contains_addr(die, addr, &contains);
1319 if (ret) {
1320 goto error;
1321 }
1322
1323 if (contains) {
1324 /*
1325 * Try to find an inlined subroutine
1326 * child of this DIE containing addr.
1327 */
1328 ret = bin_info_child_die_has_address(die, addr,
1329 &found);
1330 if(ret) {
1331 goto error;
1332 }
1333
1334 goto end;
1335 }
1336 }
1337 }
1338
1339 end:
1340 if (found) {
1341 char *filename = NULL;
1342 uint64_t line_no;
1343
1344 _src_loc = g_new0(struct source_location, 1);
1345 if (!_src_loc) {
1346 goto error;
1347 }
1348
1349 ret = bt_dwarf_die_get_call_file(die, &filename);
1350 if (ret) {
1351 goto error;
1352 }
1353 ret = bt_dwarf_die_get_call_line(die, &line_no);
1354 if (ret) {
1355 free(filename);
1356 goto error;
1357 }
1358
1359 _src_loc->filename = filename;
1360 _src_loc->line_no = line_no;
1361 *src_loc = _src_loc;
1362 }
1363
1364 bt_dwarf_die_destroy(die);
1365 return 0;
1366
1367 error:
1368 source_location_destroy(_src_loc);
1369 bt_dwarf_die_destroy(die);
1370 return -1;
1371 }
1372
1373 /**
1374 * Lookup the source location for a given address within a CU,
1375 * assuming that it is contained within an inlined function.
1376 *
1377 * A source location can be found regardless of inlining status for
1378 * this method, but in the case of an inlined function, the returned
1379 * source location will point not to the callsite but rather to the
1380 * definition site of the inline function.
1381 *
1382 * @param cu bt_dwarf_cu instance in which to look for the address
1383 * @param addr The address for which to look for
1384 * @param src_loc Out parameter, the source location (filename and
1385 * line number) for the address. Set only if the address
1386 * is found and resolved successfully
1387 *
1388 * @returns 0 on success, -1 on failure
1389 */
1390 static
1391 int bin_info_lookup_cu_src_loc_no_inl(struct bt_dwarf_cu *cu, uint64_t addr,
1392 struct source_location **src_loc)
1393 {
1394 struct source_location *_src_loc = NULL;
1395 struct bt_dwarf_die *die = NULL;
1396 const char *filename = NULL;
1397 Dwarf_Line *line = NULL;
1398 Dwarf_Addr line_addr;
1399 int ret = 0, line_no;
1400
1401 if (!cu || !src_loc) {
1402 goto error;
1403 }
1404
1405 die = bt_dwarf_die_create(cu);
1406 if (!die) {
1407 goto error;
1408 }
1409
1410 line = dwarf_getsrc_die(die->dwarf_die, addr);
1411 if (!line) {
1412 /* This is not an error. The caller needs to keep looking. */
1413 goto end;
1414 }
1415
1416 ret = dwarf_lineaddr(line, &line_addr);
1417 if (ret) {
1418 goto error;
1419 }
1420
1421 filename = dwarf_linesrc(line, NULL, NULL);
1422 if (!filename) {
1423 goto error;
1424 }
1425
1426 if (addr == line_addr) {
1427 _src_loc = g_new0(struct source_location, 1);
1428 if (!_src_loc) {
1429 goto error;
1430 }
1431
1432 ret = dwarf_lineno(line, &line_no);
1433 if (ret) {
1434 goto error;
1435 }
1436
1437 _src_loc->line_no = line_no;
1438 _src_loc->filename = g_strdup(filename);
1439 }
1440
1441 if (_src_loc) {
1442 *src_loc = _src_loc;
1443 }
1444
1445 goto end;
1446
1447 error:
1448 source_location_destroy(_src_loc);
1449 ret = -1;
1450 end:
1451 bt_dwarf_die_destroy(die);
1452 return ret;
1453 }
1454
1455 /**
1456 * Get the source location (file name and line number) for a given
1457 * address within a compile unit (CU).
1458 *
1459 * On success, the out parameter `src_loc` is set if found. On
1460 * failure, it remains unchanged.
1461 *
1462 * @param cu bt_dwarf_cu instance for the compile unit which
1463 * may contain the address
1464 * @param addr Virtual memory address for which to find the
1465 * source location
1466 * @param src_loc Out parameter, the source location
1467 * @returns 0 on success, -1 on failure
1468 */
1469 static
1470 int bin_info_lookup_cu_src_loc(struct bt_dwarf_cu *cu, uint64_t addr,
1471 struct source_location **src_loc)
1472 {
1473 int ret = 0;
1474 struct source_location *_src_loc = NULL;
1475
1476 if (!cu || !src_loc) {
1477 goto error;
1478 }
1479
1480 ret = bin_info_lookup_cu_src_loc_inl(cu, addr, &_src_loc);
1481 if (ret) {
1482 goto error;
1483 }
1484
1485 if (_src_loc) {
1486 goto end;
1487 }
1488
1489 ret = bin_info_lookup_cu_src_loc_no_inl(cu, addr, &_src_loc);
1490 if (ret) {
1491 goto error;
1492 }
1493
1494 if (_src_loc) {
1495 goto end;
1496 }
1497
1498 end:
1499 if (_src_loc) {
1500 *src_loc = _src_loc;
1501 }
1502
1503 return 0;
1504
1505 error:
1506 source_location_destroy(_src_loc);
1507 return -1;
1508 }
1509
1510 BT_HIDDEN
1511 int bin_info_lookup_source_location(struct bin_info *bin, uint64_t addr,
1512 struct source_location **src_loc)
1513 {
1514 struct bt_dwarf_cu *cu = NULL;
1515 struct source_location *_src_loc = NULL;
1516
1517 if (!bin || !src_loc) {
1518 goto error;
1519 }
1520
1521 /*
1522 * If the bin_info has a build id but it does not match the build id
1523 * that was found on the file system, return an error.
1524 */
1525 if (bin->build_id && !bin->file_build_id_matches) {
1526 goto error;
1527 }
1528
1529 /* Set DWARF info if it hasn't been accessed yet. */
1530 if (!bin->dwarf_info && !bin->is_elf_only) {
1531 if (bin_info_set_dwarf_info(bin)) {
1532 /* Failed to set DWARF info. */
1533 bin->is_elf_only = true;
1534 }
1535 }
1536
1537 if (bin->is_elf_only) {
1538 /* We cannot lookup source location without DWARF info. */
1539 goto error;
1540 }
1541
1542 if (!bin_info_has_address(bin, addr)) {
1543 goto error;
1544 }
1545
1546 /*
1547 * Addresses in ELF and DWARF are relative to base address for
1548 * PIC, so make the address argument relative too if needed.
1549 */
1550 if (bin->is_pic) {
1551 addr -= bin->low_addr;
1552 }
1553
1554 cu = bt_dwarf_cu_create(bin->dwarf_info);
1555 if (!cu) {
1556 goto error;
1557 }
1558
1559 while (bt_dwarf_cu_next(cu) == 0) {
1560 int ret;
1561
1562 ret = bin_info_lookup_cu_src_loc(cu, addr, &_src_loc);
1563 if (ret) {
1564 goto error;
1565 }
1566
1567 if (_src_loc) {
1568 break;
1569 }
1570 }
1571
1572 bt_dwarf_cu_destroy(cu);
1573 if (_src_loc) {
1574 *src_loc = _src_loc;
1575 }
1576
1577 return 0;
1578
1579 error:
1580 source_location_destroy(_src_loc);
1581 bt_dwarf_cu_destroy(cu);
1582 return -1;
1583 }
This page took 0.065952 seconds and 4 git commands to generate.