projects / babeltrace.git / blob
? search:


31c347ea7088348b32af3efed14d07b12d877f44
[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_COMP_LOG_SELF_COMP (bin->self_comp)
30 #define BT_LOG_OUTPUT_LEVEL (bin->log_level)
31 #define BT_LOG_TAG "PLUGIN/FLT.LTTNG-UTILS.DEBUG-INFO/BIN-INFO"
32 #include "logging/comp-logging.h"
33
34 #include <babeltrace2/logging.h>
35 #include <dwarf.h>
36 #include <errno.h>
37 #include <fcntl.h>
38 #include <inttypes.h>
39 #include <libgen.h>
40 #include <math.h>
41 #include <stdio.h>
42 #include <stdlib.h>
43 #include <string.h>
44 #include <unistd.h>
45
46 #include <glib.h>
47
48 #include "common/common.h"
49
50 #include "bin-info.h"
51 #include "crc32.h"
52 #include "dwarf.h"
53 #include "utils.h"
54
55 /*
56 * An address printed in hex is at most 20 bytes (16 for 64-bits +
57 * leading 0x + optional leading '+' if addr is an offset + null
58 * character).
59 */
60 #define ADDR_STR_LEN 20
61 #define BUILD_ID_NOTE_NAME "GNU"
62
63 BT_HIDDEN
64 int bin_info_init(bt_logging_level log_level, bt_self_component *self_comp)
65 {
66 int ret = 0;
67
68 if (elf_version(EV_CURRENT) == EV_NONE) {
69 BT_COMP_LOG_CUR_LVL(BT_LOG_INFO, log_level, self_comp,
70 "ELF library initialization failed: %s.",
71 elf_errmsg(-1));
72 ret = -1;
73 }
74
75 return ret;
76 }
77
78 BT_HIDDEN
79 struct bin_info *bin_info_create(struct bt_fd_cache *fdc, const char *path,
80 uint64_t low_addr, uint64_t memsz, bool is_pic,
81 const char *debug_info_dir, const char *target_prefix,
82 bt_logging_level log_level, bt_self_component *self_comp)
83 {
84 struct bin_info *bin = NULL;
85
86 BT_ASSERT(fdc);
87
88 if (!path) {
89 goto error;
90 }
91
92 bin = g_new0(struct bin_info, 1);
93 if (!bin) {
94 goto error;
95 }
96
97 bin->log_level = log_level;
98 bin->self_comp = self_comp;
99 if (target_prefix) {
100 bin->elf_path = g_build_filename(target_prefix, path, NULL);
101 } else {
102 bin->elf_path = g_strdup(path);
103 }
104
105 if (!bin->elf_path) {
106 goto error;
107 }
108
109 if (debug_info_dir) {
110 bin->debug_info_dir = g_strdup(debug_info_dir);
111 if (!bin->debug_info_dir) {
112 goto error;
113 }
114 }
115
116 bin->is_pic = is_pic;
117 bin->memsz = memsz;
118 bin->low_addr = low_addr;
119 bin->high_addr = bin->low_addr + bin->memsz;
120 bin->build_id = NULL;
121 bin->build_id_len = 0;
122 bin->file_build_id_matches = false;
123 bin->fd_cache = fdc;
124
125 return bin;
126
127 error:
128 bin_info_destroy(bin);
129 return NULL;
130 }
131
132 BT_HIDDEN
133 void bin_info_destroy(struct bin_info *bin)
134 {
135 if (!bin) {
136 return;
137 }
138
139 dwarf_end(bin->dwarf_info);
140
141 g_free(bin->debug_info_dir);
142 g_free(bin->elf_path);
143 g_free(bin->dwarf_path);
144 g_free(bin->build_id);
145 g_free(bin->dbg_link_filename);
146
147 elf_end(bin->elf_file);
148
149 bt_fd_cache_put_handle(bin->fd_cache, bin->elf_handle);
150 bt_fd_cache_put_handle(bin->fd_cache, bin->dwarf_handle);
151
152 g_free(bin);
153 }
154
155 /**
156 * Initialize the ELF file for a given executable.
157 *
158 * @param bin bin_info instance
159 * @returns 0 on success, negative value on error.
160 */
161 static
162 int bin_info_set_elf_file(struct bin_info *bin)
163 {
164 struct bt_fd_cache_handle *elf_handle = NULL;
165 Elf *elf_file = NULL;
166
167 if (!bin) {
168 goto error;
169 }
170
171 elf_handle = bt_fd_cache_get_handle(bin->fd_cache, bin->elf_path);
172 if (!elf_handle) {
173 BT_COMP_LOGI("Failed to open %s", bin->elf_path);
174 goto error;
175 }
176 bin->elf_handle = elf_handle;
177
178 elf_file = elf_begin(bt_fd_cache_handle_get_fd(bin->elf_handle),
179 ELF_C_READ, NULL);
180 if (!elf_file) {
181 BT_COMP_LOGE("elf_begin failed: %s", elf_errmsg(-1));
182 goto error;
183 }
184
185 bin->elf_file = elf_file;
186
187 if (elf_kind(elf_file) != ELF_K_ELF) {
188 BT_COMP_LOGE("Error: %s is not an ELF object", bin->elf_path);
189 goto error;
190 }
191
192 return 0;
193
194 error:
195 bt_fd_cache_put_handle(bin->fd_cache, elf_handle);
196 elf_end(elf_file);
197 return -1;
198 }
199
200 /**
201 * From a note section data struct, check if it is a build id note.
202 *
203 * @param note_data Pointer to a note section
204 *
205 * @returns 1 on match, 0 if `buf` does not contain a
206 * valid build id note
207 */
208 static
209 int is_build_id_note_section(Elf_Data *note_data)
210 {
211 size_t name_offset, desc_offset;
212 GElf_Nhdr note_header;
213 int ret = 0;
214
215 /*
216 * Discard the return value as it contains the size of the note section
217 * and we don't need it.
218 */
219 (void) gelf_getnote(note_data, 0, &note_header, &name_offset,
220 &desc_offset);
221
222 /*
223 * Check the note name length. The name_sz field includes the
224 * terminating null byte.
225 */
226 if (note_header.n_namesz != sizeof(BUILD_ID_NOTE_NAME)) {
227 goto invalid;
228 }
229
230 /* Check the note type. */
231 if (note_header.n_type != NT_GNU_BUILD_ID) {
232 goto invalid;
233 }
234
235 /* Check the note name. */
236 if (memcmp(note_data->d_buf + name_offset, BUILD_ID_NOTE_NAME,
237 note_header.n_namesz) != 0) {
238 goto invalid;
239 }
240
241 ret = 1;
242
243 invalid:
244 return ret;
245 }
246
247 /**
248 * From a build id note section data struct, check if the build id it contains
249 * is identical to the build id passed as parameter.
250 *
251 * @param note_data Pointer to the file build id note section.
252 * @param build_id Pointer to a build id to compare to.
253 * @param build_id_len length of the build id.
254 *
255 * @returns 1 on match, 0 otherwise.
256 */
257 static
258 int is_build_id_note_section_matching(Elf_Data *note_data,
259 uint8_t *build_id, size_t build_id_len)
260 {
261 size_t name_offset, desc_offset;
262 GElf_Nhdr note_header;
263
264 if (build_id_len <= 0) {
265 goto end;
266 }
267
268 /*
269 * Discard the return value as it contains the size of the note section
270 * and we don't need it.
271 */
272 (void) gelf_getnote(note_data, 0, &note_header, &name_offset,
273 &desc_offset);
274
275 /*
276 * Compare the binary build id with the supplied build id.
277 */
278 if (memcmp(build_id, note_data->d_buf + desc_offset,
279 build_id_len) == 0) {
280 return 1;
281 }
282 end:
283 return 0;
284 }
285
286 /**
287 * Checks if the build id stored in `bin` (bin->build_id) is matching the build
288 * id of the ondisk file (bin->elf_file).
289 *
290 * @param bin bin_info instance
291 * @param build_id build id to compare ot the on disk file
292 * @param build_id_len length of the build id
293 *
294 * @returns 1 on if the build id of stored in `bin` matches
295 * the build id of the ondisk file.
296 * 0 on if they are different or an error occured.
297 */
298 static
299 int is_build_id_matching(struct bin_info *bin)
300 {
301 int ret, is_build_id, is_matching = 0;
302 Elf_Scn *curr_section = NULL, *next_section = NULL;
303 GElf_Shdr curr_section_hdr;
304
305 if (!bin->build_id) {
306 goto error;
307 }
308
309 /* Set ELF file if it hasn't been accessed yet. */
310 if (!bin->elf_file) {
311 ret = bin_info_set_elf_file(bin);
312 if (ret) {
313 /* Failed to set ELF file. */
314 goto error;
315 }
316 }
317
318 next_section = elf_nextscn(bin->elf_file, curr_section);
319 if (!next_section) {
320 goto error;
321 }
322
323 while (next_section) {
324 Elf_Data *note_data = NULL;
325
326 curr_section = next_section;
327 next_section = elf_nextscn(bin->elf_file, curr_section);
328
329 if (!gelf_getshdr(curr_section, &curr_section_hdr)) {
330 goto error;
331 }
332
333 if (curr_section_hdr.sh_type != SHT_NOTE) {
334 continue;
335 }
336
337 /*
338 * elf_getdata() translates the data to native byte order.
339 */
340 note_data = elf_getdata(curr_section, NULL);
341 if (!note_data) {
342 goto error;
343 }
344
345 /* Check if the note is of the build-id type. */
346 is_build_id = is_build_id_note_section(note_data);
347 if (!is_build_id) {
348 continue;
349 }
350
351 /*
352 * Compare the build id of the on-disk file and
353 * the build id recorded in the trace.
354 */
355 is_matching = is_build_id_note_section_matching(
356 note_data, bin->build_id, bin->build_id_len);
357 if (!is_matching) {
358 break;
359 }
360 }
361 error:
362 return is_matching;
363 }
364
365 BT_HIDDEN
366 int bin_info_set_build_id(struct bin_info *bin, uint8_t *build_id,
367 size_t build_id_len)
368 {
369 if (!bin || !build_id) {
370 goto error;
371 }
372
373 /* Set the build id. */
374 bin->build_id = g_new0(uint8_t, build_id_len);
375 if (!bin->build_id) {
376 goto error;
377 }
378
379 memcpy(bin->build_id, build_id, build_id_len);
380 bin->build_id_len = build_id_len;
381
382 /*
383 * Check if the file found on the file system has the same build id
384 * that what was recorded in the trace.
385 */
386 bin->file_build_id_matches = is_build_id_matching(bin);
387 if (!bin->file_build_id_matches) {
388 BT_COMP_LOGI_STR("Supplied Build ID does not match Build ID of the "
389 "binary or library found on the file system.");
390 goto error;
391 }
392
393 /*
394 * Reset the is_elf_only flag in case it had been set
395 * previously, because we might find separate debug info using
396 * the new build id information.
397 */
398 bin->is_elf_only = false;
399
400 return 0;
401
402 error:
403 return -1;
404 }
405
406 BT_HIDDEN
407 int bin_info_set_debug_link(struct bin_info *bin, const char *filename,
408 uint32_t crc)
409 {
410 if (!bin || !filename) {
411 goto error;
412 }
413
414 bin->dbg_link_filename = g_strdup(filename);
415 if (!bin->dbg_link_filename) {
416 goto error;
417 }
418
419 bin->dbg_link_crc = crc;
420
421 /*
422 * Reset the is_elf_only flag in case it had been set
423 * previously, because we might find separate debug info using
424 * the new build id information.
425 */
426 bin->is_elf_only = false;
427
428 return 0;
429
430 error:
431
432 return -1;
433 }
434
435 /**
436 * Tries to read DWARF info from the location given by path, and
437 * attach it to the given bin_info instance if it exists.
438 *
439 * @param bin bin_info instance for which to set DWARF info
440 * @param path Presumed location of the DWARF info
441 * @returns 0 on success, negative value on failure
442 */
443 static
444 int bin_info_set_dwarf_info_from_path(struct bin_info *bin, char *path)
445 {
446 int ret = 0;
447 struct bt_fd_cache_handle *dwarf_handle = NULL;
448 struct bt_dwarf_cu *cu = NULL;
449 Dwarf *dwarf_info = NULL;
450
451 if (!bin || !path) {
452 goto error;
453 }
454
455 dwarf_handle = bt_fd_cache_get_handle(bin->fd_cache, path);
456 if (!dwarf_handle) {
457 goto error;
458 }
459
460 dwarf_info = dwarf_begin(bt_fd_cache_handle_get_fd(dwarf_handle),
461 DWARF_C_READ);
462 if (!dwarf_info) {
463 goto error;
464 }
465
466 /*
467 * Check if the dwarf info has any CU. If not, the
468 * executable's object file contains no DWARF info.
469 */
470 cu = bt_dwarf_cu_create(dwarf_info);
471 if (!cu) {
472 goto error;
473 }
474
475 ret = bt_dwarf_cu_next(cu);
476 if (ret) {
477 goto error;
478 }
479
480 bin->dwarf_handle = dwarf_handle;
481 bin->dwarf_path = g_strdup(path);
482 if (!bin->dwarf_path) {
483 goto error;
484 }
485 bin->dwarf_info = dwarf_info;
486 free(cu);
487
488 return 0;
489
490 error:
491 if (bin) {
492 bt_fd_cache_put_handle(bin->fd_cache, dwarf_handle);
493 }
494 dwarf_end(dwarf_info);
495 g_free(dwarf_info);
496 free(cu);
497
498 return -1;
499 }
500
501 /**
502 * Try to set the dwarf_info for a given bin_info instance via the
503 * build ID method.
504 *
505 * @param bin bin_info instance for which to retrieve the
506 * DWARF info via build ID
507 * @returns 0 on success (i.e. dwarf_info set), -1 on failure
508 */
509 static
510 int bin_info_set_dwarf_info_build_id(struct bin_info *bin)
511 {
512 int i = 0, ret = 0;
513 char *path = NULL, *build_id_prefix_dir = NULL, *build_id_file = NULL;
514 const char *dbg_dir = NULL;
515 size_t build_id_char_len, build_id_suffix_char_len, build_id_file_len;
516
517 if (!bin || !bin->build_id) {
518 goto error;
519 }
520
521 dbg_dir = bin->debug_info_dir ? bin->debug_info_dir : DEFAULT_DEBUG_DIR;
522
523 /*
524 * The prefix dir is the first byte of the build id, represented in
525 * lowercase hex as two characters per byte, +1 for '\0'.
526 */
527 build_id_prefix_dir = g_new0(gchar, BUILD_ID_PREFIX_DIR_LEN + 1);
528 if (!build_id_prefix_dir) {
529 goto error;
530 }
531 g_snprintf(build_id_prefix_dir, BUILD_ID_PREFIX_DIR_LEN + 1, "%02x", bin->build_id[0]);
532
533 /*
534 * The build id file is the remaining bytes of the build id,
535 * represented in lowercase hex, as two characters per byte.
536 */
537 build_id_char_len = (2 * (bin->build_id_len - 1));
538
539 /* To which the build id suffix is added, +1 for '\0'. */
540 build_id_suffix_char_len = strlen(BUILD_ID_SUFFIX) + 1;
541
542 /*
543 * The resulting filename string is the concatenation of the
544 * hex build id and the suffix.
545 */
546 build_id_file_len = build_id_char_len + build_id_suffix_char_len;
547 build_id_file = g_new0(gchar, build_id_file_len);
548 if (!build_id_file) {
549 goto error;
550 }
551
552 /*
553 * For each byte, starting at offset 1, append two characters
554 * in lowercase hex.
555 */
556 for (i = 1; i < bin->build_id_len; ++i) {
557 int path_idx = 2 * (i - 1);
558
559 g_snprintf(&build_id_file[path_idx], 3, "%02x", bin->build_id[i]);
560 }
561 /* Append the suffix to the generated string, including the '\0'. */
562 g_snprintf(&build_id_file[build_id_char_len], build_id_suffix_char_len,
563 BUILD_ID_SUFFIX);
564
565 path = g_build_filename(dbg_dir, BUILD_ID_SUBDIR, build_id_prefix_dir, build_id_file, NULL);
566 if (!path) {
567 goto error;
568 }
569
570 ret = bin_info_set_dwarf_info_from_path(bin, path);
571 if (ret) {
572 goto error;
573 }
574
575 goto end;
576
577 error:
578 ret = -1;
579 end:
580 g_free(build_id_prefix_dir);
581 g_free(build_id_file);
582 g_free(path);
583
584 return ret;
585 }
586
587 /**
588 * Tests whether the file located at path exists and has the expected
589 * checksum.
590 *
591 * This predicate is used when looking up separate debug info via the
592 * GNU debuglink method. The expected crc can be found .gnu_debuglink
593 * section in the original ELF file, along with the filename for the
594 * file containing the debug info.
595 *
596 * @param path Full path at which to look for the debug file
597 * @param crc Expected checksum for the debug file
598 * @returns 1 if the file exists and has the correct checksum,
599 * 0 otherwise
600 */
601 static
602 int is_valid_debug_file(struct bin_info *bin, char *path, uint32_t crc)
603 {
604 int ret = 0;
605 struct bt_fd_cache_handle *debug_handle = NULL;
606 uint32_t _crc = 0;
607
608 if (!path) {
609 goto end;
610 }
611
612 debug_handle = bt_fd_cache_get_handle(bin->fd_cache, path);
613 if (!debug_handle) {
614 goto end;
615 }
616
617 ret = crc32(bt_fd_cache_handle_get_fd(debug_handle), &_crc);
618 if (ret) {
619 ret = 0;
620 goto end;
621 }
622
623 ret = (crc == _crc);
624
625 end:
626 bt_fd_cache_put_handle(bin->fd_cache, debug_handle);
627 return ret;
628 }
629
630 /**
631 * Try to set the dwarf_info for a given bin_info instance via the
632 * debug-link method.
633 *
634 * @param bin bin_info instance for which to retrieve the
635 * DWARF info via debug link
636 * @returns 0 on success (i.e. dwarf_info set), -1 on failure
637 */
638 static
639 int bin_info_set_dwarf_info_debug_link(struct bin_info *bin)
640 {
641 int ret = 0;
642 const gchar *dbg_dir = NULL;
643 gchar *bin_dir = NULL, *path = NULL;
644
645 if (!bin || !bin->dbg_link_filename) {
646 goto error;
647 }
648
649 dbg_dir = bin->debug_info_dir ? bin->debug_info_dir : DEFAULT_DEBUG_DIR;
650 bin_dir = g_path_get_dirname(bin->elf_path);
651
652 /* First look in the executable's dir */
653 path = g_build_filename(bin_dir, bin->dbg_link_filename, NULL);
654
655 if (is_valid_debug_file(bin, path, bin->dbg_link_crc)) {
656 goto found;
657 }
658
659 /* If not found, look in .debug subdir */
660 g_free(path);
661 path = g_build_filename(bin_dir, DEBUG_SUBDIR, bin->dbg_link_filename, NULL);
662
663 if (is_valid_debug_file(bin, path, bin->dbg_link_crc)) {
664 goto found;
665 }
666
667 /* Lastly, look under the global debug directory */
668 g_free(path);
669
670 path = g_build_filename(dbg_dir, bin_dir, bin->dbg_link_filename, NULL);
671 if (is_valid_debug_file(bin, path, bin->dbg_link_crc)) {
672 goto found;
673 }
674
675 error:
676 ret = -1;
677 end:
678 g_free(bin_dir);
679 g_free(path);
680
681 return ret;
682
683 found:
684 ret = bin_info_set_dwarf_info_from_path(bin, path);
685 if (ret) {
686 goto error;
687 }
688
689 goto end;
690 }
691
692 /**
693 * Initialize the DWARF info for a given executable.
694 *
695 * @param bin bin_info instance
696 * @returns 0 on success, negative value on failure
697 */
698 static
699 int bin_info_set_dwarf_info(struct bin_info *bin)
700 {
701 int ret = 0;
702
703 if (!bin) {
704 ret = -1;
705 goto end;
706 }
707
708 /* First try to set the DWARF info from the ELF file */
709 ret = bin_info_set_dwarf_info_from_path(bin, bin->elf_path);
710 if (!ret) {
711 goto end;
712 }
713
714 /*
715 * If that fails, try to find separate debug info via build ID
716 * and debug link.
717 */
718 ret = bin_info_set_dwarf_info_build_id(bin);
719 if (!ret) {
720 goto end;
721 }
722
723 ret = bin_info_set_dwarf_info_debug_link(bin);
724 if (!ret) {
725 goto end;
726 }
727
728 end:
729 return ret;
730 }
731
732 BT_HIDDEN
733 void source_location_destroy(struct source_location *src_loc)
734 {
735 if (!src_loc) {
736 return;
737 }
738
739 free(src_loc->filename);
740 g_free(src_loc);
741 }
742
743 /**
744 * Append a string representation of an address offset to an existing
745 * string.
746 *
747 * On success, the out parameter `result` will contain the base string
748 * followed by the offset string of the form "+0x1234". On failure,
749 * `result` remains unchanged.
750 *
751 * @param base_str The string to which to append an offset string
752 * @param low_addr The lower virtual memory address, the base from
753 * which the offset is computed
754 * @param high_addr The higher virtual memory address
755 * @param result Out parameter, the base string followed by the
756 * offset string
757 * @returns 0 on success, -1 on failure
758 */
759 static
760 int bin_info_append_offset_str(const char *base_str, uint64_t low_addr,
761 uint64_t high_addr, char **result)
762 {
763 uint64_t offset;
764 char *_result = NULL;
765
766 if (!base_str || !result) {
767 goto error;
768 }
769
770 offset = high_addr - low_addr;
771
772 _result = g_strdup_printf("%s+%#0" PRIx64, base_str, offset);
773 if (!_result) {
774 goto error;
775 }
776 *result = _result;
777
778 return 0;
779
780 error:
781 free(_result);
782 return -1;
783 }
784
785 /**
786 * Try to find the symbol closest to an address within a given ELF
787 * section.
788 *
789 * Only function symbols are taken into account. The symbol's address
790 * must precede `addr`. A symbol with a closer address might exist
791 * after `addr` but is irrelevant because it cannot encompass `addr`.
792 *
793 * On success, if found, the out parameters `sym` and `shdr` are
794 * set. On failure or if none are found, they remain unchanged.
795 *
796 * @param scn ELF section in which to look for the address
797 * @param addr Virtual memory address for which to find the
798 * nearest function symbol
799 * @param sym Out parameter, the nearest function symbol
800 * @param shdr Out parameter, the section header for scn
801 * @returns 0 on success, -1 on failure
802 */
803 static
804 int bin_info_get_nearest_symbol_from_section(Elf_Scn *scn, uint64_t addr,
805 GElf_Sym **sym, GElf_Shdr **shdr)
806 {
807 int i;
808 size_t symbol_count;
809 Elf_Data *data = NULL;
810 GElf_Shdr *_shdr = NULL;
811 GElf_Sym *nearest_sym = NULL;
812
813 if (!scn || !sym || !shdr) {
814 goto error;
815 }
816
817 _shdr = g_new0(GElf_Shdr, 1);
818 if (!_shdr) {
819 goto error;
820 }
821
822 _shdr = gelf_getshdr(scn, _shdr);
823 if (!_shdr) {
824 goto error;
825 }
826
827 if (_shdr->sh_type != SHT_SYMTAB) {
828 /*
829 * We are only interested in symbol table (symtab)
830 * sections, skip this one.
831 */
832 goto end;
833 }
834
835 data = elf_getdata(scn, NULL);
836 if (!data) {
837 goto error;
838 }
839
840 symbol_count = _shdr->sh_size / _shdr->sh_entsize;
841
842 for (i = 0; i < symbol_count; ++i) {
843 GElf_Sym *cur_sym = NULL;
844
845 cur_sym = g_new0(GElf_Sym, 1);
846 if (!cur_sym) {
847 goto error;
848 }
849 cur_sym = gelf_getsym(data, i, cur_sym);
850 if (!cur_sym) {
851 goto error;
852 }
853 if (GELF_ST_TYPE(cur_sym->st_info) != STT_FUNC) {
854 /* We're only interested in the functions. */
855 g_free(cur_sym);
856 continue;
857 }
858
859 if (cur_sym->st_value <= addr &&
860 (!nearest_sym ||
861 cur_sym->st_value > nearest_sym->st_value)) {
862 g_free(nearest_sym);
863 nearest_sym = cur_sym;
864 } else {
865 g_free(cur_sym);
866 }
867 }
868
869 end:
870 if (nearest_sym) {
871 *sym = nearest_sym;
872 *shdr = _shdr;
873 } else {
874 g_free(_shdr);
875 }
876
877 return 0;
878
879 error:
880 g_free(nearest_sym);
881 g_free(_shdr);
882 return -1;
883 }
884
885 /**
886 * Get the name of the function containing a given address within an
887 * executable using ELF symbols.
888 *
889 * The function name is in fact the name of the nearest ELF symbol,
890 * followed by the offset in bytes between the address and the symbol
891 * (in hex), separated by a '+' character.
892 *
893 * If found, the out parameter `func_name` is set on success. On failure,
894 * it remains unchanged.
895 *
896 * @param bin bin_info instance for the executable containing
897 * the address
898 * @param addr Virtual memory address for which to find the
899 * function name
900 * @param func_name Out parameter, the function name
901 * @returns 0 on success, -1 on failure
902 */
903 static
904 int bin_info_lookup_elf_function_name(struct bin_info *bin, uint64_t addr,
905 char **func_name)
906 {
907 /*
908 * TODO (possible optimisation): if an ELF has no symtab
909 * section, it has been stripped. Therefore, it would be wise
910 * to store a flag indicating the stripped status after the
911 * first iteration to prevent subsequent ones.
912 */
913 int ret = 0;
914 Elf_Scn *scn = NULL;
915 GElf_Sym *sym = NULL;
916 GElf_Shdr *shdr = NULL;
917 char *sym_name = NULL;
918
919 /* Set ELF file if it hasn't been accessed yet. */
920 if (!bin->elf_file) {
921 ret = bin_info_set_elf_file(bin);
922 if (ret) {
923 /* Failed to set ELF file. */
924 goto error;
925 }
926 }
927
928 scn = elf_nextscn(bin->elf_file, scn);
929 if (!scn) {
930 goto error;
931 }
932
933 while (scn && !sym) {
934 ret = bin_info_get_nearest_symbol_from_section(
935 scn, addr, &sym, &shdr);
936 if (ret) {
937 goto error;
938 }
939
940 scn = elf_nextscn(bin->elf_file, scn);
941 }
942
943 if (sym) {
944 sym_name = elf_strptr(bin->elf_file, shdr->sh_link,
945 sym->st_name);
946 if (!sym_name) {
947 goto error;
948 }
949
950 ret = bin_info_append_offset_str(sym_name, sym->st_value, addr,
951 func_name);
952 if (ret) {
953 goto error;
954 }
955 }
956
957 g_free(shdr);
958 g_free(sym);
959 return 0;
960
961 error:
962 g_free(shdr);
963 g_free(sym);
964 return ret;
965 }
966
967 /**
968 * Get the name of the function containing a given address within a
969 * given compile unit (CU).
970 *
971 * If found, the out parameter `func_name` is set on success. On
972 * failure, it remains unchanged.
973 *
974 * @param cu bt_dwarf_cu instance which may contain the address
975 * @param addr Virtual memory address for which to find the
976 * function name
977 * @param func_name Out parameter, the function name
978 * @returns 0 on success, -1 on failure
979 */
980 static
981 int bin_info_lookup_cu_function_name(struct bt_dwarf_cu *cu, uint64_t addr,
982 char **func_name)
983 {
984 int ret = 0;
985 bool found = false;
986 struct bt_dwarf_die *die = NULL;
987
988 if (!cu || !func_name) {
989 goto error;
990 }
991
992 die = bt_dwarf_die_create(cu);
993 if (!die) {
994 goto error;
995 }
996
997 while (bt_dwarf_die_next(die) == 0) {
998 int tag;
999
1000 ret = bt_dwarf_die_get_tag(die, &tag);
1001 if (ret) {
1002 goto error;
1003 }
1004
1005 if (tag == DW_TAG_subprogram) {
1006 ret = bt_dwarf_die_contains_addr(die, addr, &found);
1007 if (ret) {
1008 goto error;
1009 }
1010
1011 if (found) {
1012 break;
1013 }
1014 }
1015 }
1016
1017 if (found) {
1018 uint64_t low_addr = 0;
1019 char *die_name = NULL;
1020
1021 ret = bt_dwarf_die_get_name(die, &die_name);
1022 if (ret) {
1023 goto error;
1024 }
1025
1026 ret = dwarf_lowpc(die->dwarf_die, &low_addr);
1027 if (ret) {
1028 free(die_name);
1029 goto error;
1030 }
1031
1032 ret = bin_info_append_offset_str(die_name, low_addr, addr,
1033 func_name);
1034 free(die_name);
1035 if (ret) {
1036 goto error;
1037 }
1038 }
1039
1040 bt_dwarf_die_destroy(die);
1041 return 0;
1042
1043 error:
1044 bt_dwarf_die_destroy(die);
1045 return -1;
1046 }
1047
1048 /**
1049 * Get the name of the function containing a given address within an
1050 * executable using DWARF debug info.
1051 *
1052 * If found, the out parameter `func_name` is set on success. On
1053 * failure, it remains unchanged.
1054 *
1055 * @param bin bin_info instance for the executable containing
1056 * the address
1057 * @param addr Virtual memory address for which to find the
1058 * function name
1059 * @param func_name Out parameter, the function name
1060 * @returns 0 on success, -1 on failure
1061 */
1062 static
1063 int bin_info_lookup_dwarf_function_name(struct bin_info *bin, uint64_t addr,
1064 char **func_name)
1065 {
1066 int ret = 0;
1067 char *_func_name = NULL;
1068 struct bt_dwarf_cu *cu = NULL;
1069
1070 if (!bin || !func_name) {
1071 goto error;
1072 }
1073
1074 cu = bt_dwarf_cu_create(bin->dwarf_info);
1075 if (!cu) {
1076 goto error;
1077 }
1078
1079 while (bt_dwarf_cu_next(cu) == 0) {
1080 ret = bin_info_lookup_cu_function_name(cu, addr, &_func_name);
1081 if (ret) {
1082 goto error;
1083 }
1084
1085 if (_func_name) {
1086 break;
1087 }
1088 }
1089
1090 if (_func_name) {
1091 *func_name = _func_name;
1092 } else {
1093 goto error;
1094 }
1095
1096 bt_dwarf_cu_destroy(cu);
1097 return 0;
1098
1099 error:
1100 bt_dwarf_cu_destroy(cu);
1101 return -1;
1102 }
1103
1104 BT_HIDDEN
1105 int bin_info_lookup_function_name(struct bin_info *bin,
1106 uint64_t addr, char **func_name)
1107 {
1108 int ret = 0;
1109 char *_func_name = NULL;
1110
1111 if (!bin || !func_name) {
1112 goto error;
1113 }
1114
1115 /*
1116 * If the bin_info has a build id but it does not match the build id
1117 * that was found on the file system, return an error.
1118 */
1119 if (bin->build_id && !bin->file_build_id_matches) {
1120 goto error;
1121 }
1122
1123 /* Set DWARF info if it hasn't been accessed yet. */
1124 if (!bin->dwarf_info && !bin->is_elf_only) {
1125 ret = bin_info_set_dwarf_info(bin);
1126 if (ret) {
1127 BT_COMP_LOGI_STR("Failed to set bin dwarf info, falling "
1128 "back to ELF lookup.");
1129 /* Failed to set DWARF info, fallback to ELF. */
1130 bin->is_elf_only = true;
1131 }
1132 }
1133
1134 if (!bin_info_has_address(bin, addr)) {
1135 goto error;
1136 }
1137
1138 /*
1139 * Addresses in ELF and DWARF are relative to base address for
1140 * PIC, so make the address argument relative too if needed.
1141 */
1142 if (bin->is_pic) {
1143 addr -= bin->low_addr;
1144 }
1145
1146 if (bin->is_elf_only) {
1147 ret = bin_info_lookup_elf_function_name(bin, addr,
1148 &_func_name);
1149 if (ret) {
1150 BT_COMP_LOGI("Failed to lookup function name (ELF): "
1151 "ret=%d", ret);
1152 }
1153 } else {
1154 ret = bin_info_lookup_dwarf_function_name(bin, addr,
1155 &_func_name);
1156 if (ret) {
1157 BT_COMP_LOGI("Failed to lookup function name (DWARF): "
1158 "ret=%d", ret);
1159 }
1160 }
1161
1162 *func_name = _func_name;
1163 return 0;
1164
1165 error:
1166 return -1;
1167 }
1168
1169 BT_HIDDEN
1170 int bin_info_get_bin_loc(struct bin_info *bin, uint64_t addr, char **bin_loc)
1171 {
1172 gchar *_bin_loc = NULL;
1173
1174 if (!bin || !bin_loc) {
1175 goto error;
1176 }
1177
1178 /*
1179 * If the bin_info has a build id but it does not match the build id
1180 * that was found on the file system, return an error.
1181 */
1182 if (bin->build_id && !bin->file_build_id_matches) {
1183 goto error;
1184 }
1185
1186 if (bin->is_pic) {
1187 addr -= bin->low_addr;
1188 _bin_loc = g_strdup_printf("+%#0" PRIx64, addr);
1189 } else {
1190 _bin_loc = g_strdup_printf("@%#0" PRIx64, addr);
1191 }
1192
1193 if (!_bin_loc) {
1194 goto error;
1195 }
1196
1197 *bin_loc = _bin_loc;
1198 return 0;
1199
1200 error:
1201 return -1;
1202 }
1203
1204 /**
1205 * Predicate used to determine whether the children of a given DIE
1206 * contain a specific address.
1207 *
1208 * More specifically, the parameter `die` is expected to be a
1209 * subprogram (function) DIE, and this predicate tells whether any
1210 * subroutines are inlined within this function and would contain
1211 * `addr`.
1212 *
1213 * On success, the out parameter `contains` is set with the boolean
1214 * value indicating whether the DIE's range covers `addr`. On failure,
1215 * it remains unchanged.
1216 *
1217 * Do note that this function advances the position of `die`. If the
1218 * address is found within one of its children, `die` will be pointing
1219 * to that child upon returning from the function, allowing to extract
1220 * the information deemed necessary.
1221 *
1222 * @param die The parent DIE in whose children the address will be
1223 * looked for
1224 * @param addr The address for which to look for in the DIEs
1225 * @param contains Out parameter, true if addr is contained,
1226 * false if not
1227 * @returns Returns 0 on success, -1 on failure
1228 */
1229 static
1230 int bin_info_child_die_has_address(struct bt_dwarf_die *die, uint64_t addr, bool *contains)
1231 {
1232 int ret = 0;
1233 bool _contains = false;
1234
1235 if (!die) {
1236 goto error;
1237 }
1238
1239 ret = bt_dwarf_die_child(die);
1240 if (ret) {
1241 goto error;
1242 }
1243
1244 do {
1245 ret = bt_dwarf_die_contains_addr(die, addr, &_contains);
1246 if (ret) {
1247 goto error;
1248 }
1249
1250 if (_contains) {
1251 /*
1252 * The address is within the range of the current DIE
1253 * or its children.
1254 */
1255 int tag;
1256
1257 ret = bt_dwarf_die_get_tag(die, &tag);
1258 if (ret) {
1259 goto error;
1260 }
1261
1262 if (tag == DW_TAG_inlined_subroutine) {
1263 /* Found the tracepoint. */
1264 goto end;
1265 }
1266
1267 if (bt_dwarf_die_has_children(die)) {
1268 /*
1269 * Look for the address in the children DIEs.
1270 */
1271 ret = bt_dwarf_die_child(die);
1272 if (ret) {
1273 goto error;
1274 }
1275 }
1276 }
1277 } while (bt_dwarf_die_next(die) == 0);
1278
1279 end:
1280 *contains = _contains;
1281 return 0;
1282
1283 error:
1284 return -1;
1285 }
1286
1287 /**
1288 * Lookup the source location for a given address within a CU, making
1289 * the assumption that it is contained within an inline routine in a
1290 * function.
1291 *
1292 * @param cu bt_dwarf_cu instance in which to look for the address
1293 * @param addr The address for which to look for
1294 * @param src_loc Out parameter, the source location (filename and
1295 * line number) for the address
1296 * @returns 0 on success, -1 on failure
1297 */
1298 static
1299 int bin_info_lookup_cu_src_loc_inl(struct bt_dwarf_cu *cu, uint64_t addr,
1300 struct source_location **src_loc)
1301 {
1302 int ret = 0;
1303 bool found = false;
1304 struct bt_dwarf_die *die = NULL;
1305 struct source_location *_src_loc = NULL;
1306
1307 if (!cu || !src_loc) {
1308 goto error;
1309 }
1310
1311 die = bt_dwarf_die_create(cu);
1312 if (!die) {
1313 goto error;
1314 }
1315
1316 while (bt_dwarf_die_next(die) == 0) {
1317 int tag;
1318
1319 ret = bt_dwarf_die_get_tag(die, &tag);
1320 if (ret) {
1321 goto error;
1322 }
1323
1324 if (tag == DW_TAG_subprogram) {
1325 bool contains = false;
1326
1327 ret = bt_dwarf_die_contains_addr(die, addr, &contains);
1328 if (ret) {
1329 goto error;
1330 }
1331
1332 if (contains) {
1333 /*
1334 * Try to find an inlined subroutine
1335 * child of this DIE containing addr.
1336 */
1337 ret = bin_info_child_die_has_address(die, addr,
1338 &found);
1339 if(ret) {
1340 goto error;
1341 }
1342
1343 goto end;
1344 }
1345 }
1346 }
1347
1348 end:
1349 if (found) {
1350 char *filename = NULL;
1351 uint64_t line_no;
1352
1353 _src_loc = g_new0(struct source_location, 1);
1354 if (!_src_loc) {
1355 goto error;
1356 }
1357
1358 ret = bt_dwarf_die_get_call_file(die, &filename);
1359 if (ret) {
1360 goto error;
1361 }
1362 ret = bt_dwarf_die_get_call_line(die, &line_no);
1363 if (ret) {
1364 free(filename);
1365 goto error;
1366 }
1367
1368 _src_loc->filename = filename;
1369 _src_loc->line_no = line_no;
1370 *src_loc = _src_loc;
1371 }
1372
1373 bt_dwarf_die_destroy(die);
1374 return 0;
1375
1376 error:
1377 source_location_destroy(_src_loc);
1378 bt_dwarf_die_destroy(die);
1379 return -1;
1380 }
1381
1382 /**
1383 * Lookup the source location for a given address within a CU,
1384 * assuming that it is contained within an inlined function.
1385 *
1386 * A source location can be found regardless of inlining status for
1387 * this method, but in the case of an inlined function, the returned
1388 * source location will point not to the callsite but rather to the
1389 * definition site of the inline function.
1390 *
1391 * @param cu bt_dwarf_cu instance in which to look for the address
1392 * @param addr The address for which to look for
1393 * @param src_loc Out parameter, the source location (filename and
1394 * line number) for the address. Set only if the address
1395 * is found and resolved successfully
1396 *
1397 * @returns 0 on success, -1 on failure
1398 */
1399 static
1400 int bin_info_lookup_cu_src_loc_no_inl(struct bt_dwarf_cu *cu, uint64_t addr,
1401 struct source_location **src_loc)
1402 {
1403 struct source_location *_src_loc = NULL;
1404 struct bt_dwarf_die *die = NULL;
1405 const char *filename = NULL;
1406 Dwarf_Line *line = NULL;
1407 Dwarf_Addr line_addr;
1408 int ret = 0, line_no;
1409
1410 if (!cu || !src_loc) {
1411 goto error;
1412 }
1413
1414 die = bt_dwarf_die_create(cu);
1415 if (!die) {
1416 goto error;
1417 }
1418
1419 line = dwarf_getsrc_die(die->dwarf_die, addr);
1420 if (!line) {
1421 /* This is not an error. The caller needs to keep looking. */
1422 goto end;
1423 }
1424
1425 ret = dwarf_lineaddr(line, &line_addr);
1426 if (ret) {
1427 goto error;
1428 }
1429
1430 filename = dwarf_linesrc(line, NULL, NULL);
1431 if (!filename) {
1432 goto error;
1433 }
1434
1435 if (addr == line_addr) {
1436 _src_loc = g_new0(struct source_location, 1);
1437 if (!_src_loc) {
1438 goto error;
1439 }
1440
1441 ret = dwarf_lineno(line, &line_no);
1442 if (ret) {
1443 goto error;
1444 }
1445
1446 _src_loc->line_no = line_no;
1447 _src_loc->filename = g_strdup(filename);
1448 }
1449
1450 if (_src_loc) {
1451 *src_loc = _src_loc;
1452 }
1453
1454 goto end;
1455
1456 error:
1457 source_location_destroy(_src_loc);
1458 ret = -1;
1459 end:
1460 bt_dwarf_die_destroy(die);
1461 return ret;
1462 }
1463
1464 /**
1465 * Get the source location (file name and line number) for a given
1466 * address within a compile unit (CU).
1467 *
1468 * On success, the out parameter `src_loc` is set if found. On
1469 * failure, it remains unchanged.
1470 *
1471 * @param cu bt_dwarf_cu instance for the compile unit which
1472 * may contain the address
1473 * @param addr Virtual memory address for which to find the
1474 * source location
1475 * @param src_loc Out parameter, the source location
1476 * @returns 0 on success, -1 on failure
1477 */
1478 static
1479 int bin_info_lookup_cu_src_loc(struct bt_dwarf_cu *cu, uint64_t addr,
1480 struct source_location **src_loc)
1481 {
1482 int ret = 0;
1483 struct source_location *_src_loc = NULL;
1484
1485 if (!cu || !src_loc) {
1486 goto error;
1487 }
1488
1489 ret = bin_info_lookup_cu_src_loc_inl(cu, addr, &_src_loc);
1490 if (ret) {
1491 goto error;
1492 }
1493
1494 if (_src_loc) {
1495 goto end;
1496 }
1497
1498 ret = bin_info_lookup_cu_src_loc_no_inl(cu, addr, &_src_loc);
1499 if (ret) {
1500 goto error;
1501 }
1502
1503 if (_src_loc) {
1504 goto end;
1505 }
1506
1507 end:
1508 if (_src_loc) {
1509 *src_loc = _src_loc;
1510 }
1511
1512 return 0;
1513
1514 error:
1515 source_location_destroy(_src_loc);
1516 return -1;
1517 }
1518
1519 BT_HIDDEN
1520 int bin_info_lookup_source_location(struct bin_info *bin, uint64_t addr,
1521 struct source_location **src_loc)
1522 {
1523 struct bt_dwarf_cu *cu = NULL;
1524 struct source_location *_src_loc = NULL;
1525
1526 if (!bin || !src_loc) {
1527 goto error;
1528 }
1529
1530 /*
1531 * If the bin_info has a build id but it does not match the build id
1532 * that was found on the file system, return an error.
1533 */
1534 if (bin->build_id && !bin->file_build_id_matches) {
1535 goto error;
1536 }
1537
1538 /* Set DWARF info if it hasn't been accessed yet. */
1539 if (!bin->dwarf_info && !bin->is_elf_only) {
1540 if (bin_info_set_dwarf_info(bin)) {
1541 /* Failed to set DWARF info. */
1542 bin->is_elf_only = true;
1543 }
1544 }
1545
1546 if (bin->is_elf_only) {
1547 /* We cannot lookup source location without DWARF info. */
1548 goto error;
1549 }
1550
1551 if (!bin_info_has_address(bin, addr)) {
1552 goto error;
1553 }
1554
1555 /*
1556 * Addresses in ELF and DWARF are relative to base address for
1557 * PIC, so make the address argument relative too if needed.
1558 */
1559 if (bin->is_pic) {
1560 addr -= bin->low_addr;
1561 }
1562
1563 cu = bt_dwarf_cu_create(bin->dwarf_info);
1564 if (!cu) {
1565 goto error;
1566 }
1567
1568 while (bt_dwarf_cu_next(cu) == 0) {
1569 int ret;
1570
1571 ret = bin_info_lookup_cu_src_loc(cu, addr, &_src_loc);
1572 if (ret) {
1573 goto error;
1574 }
1575
1576 if (_src_loc) {
1577 break;
1578 }
1579 }
1580
1581 bt_dwarf_cu_destroy(cu);
1582 if (_src_loc) {
1583 *src_loc = _src_loc;
1584 }
1585
1586 return 0;
1587
1588 error:
1589 source_location_destroy(_src_loc);
1590 bt_dwarf_cu_destroy(cu);
1591 return -1;
1592 }
BabelTrace - Trace Format Babel Tower
This page took 0.063781 seconds and 3 git commands to generate.