4 * Babeltrace - Executable and Shared Object Debug Info Reader
6 * Copyright 2015 Antoine Busque <abusque@efficios.com>
8 * Author: Antoine Busque <abusque@efficios.com>
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:
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
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
39 #include <babeltrace/dwarf.h>
40 #include <babeltrace/so-info.h>
41 #include <babeltrace/crc32.h>
42 #include <babeltrace/babeltrace-internal.h>
43 #include <babeltrace/utils.h>
46 * An address printed in hex is at most 20 bytes (16 for 64-bits +
47 * leading 0x + optional leading '+' if addr is an offset + null
50 #define ADDR_STR_LEN 20
53 int so_info_init(void)
57 if (elf_version(EV_CURRENT
) == EV_NONE
) {
58 fprintf(stderr
, "ELF library initialization failed: %s\n",
67 struct so_info
*so_info_create(const char *path
, uint64_t low_addr
,
68 uint64_t memsz
, bool is_pic
)
70 struct so_info
*so
= NULL
;
71 GElf_Ehdr
*ehdr
= NULL
;
77 so
= g_new0(struct so_info
, 1);
82 so
->elf_path
= strdup(path
);
87 so
->elf_fd
= open(path
, O_RDONLY
);
89 fprintf(stderr
, "Failed to open %s\n", path
);
93 so
->elf_file
= elf_begin(so
->elf_fd
, ELF_C_READ
, NULL
);
95 fprintf(stderr
, "elf_begin failed: %s\n", elf_errmsg(-1));
99 if (elf_kind(so
->elf_file
) != ELF_K_ELF
) {
100 fprintf(stderr
, "Error: %s is not an ELF object\n",
105 ehdr
= g_new0(GElf_Ehdr
, 1);
110 if (!gelf_getehdr(so
->elf_file
, ehdr
)) {
111 fprintf(stderr
, "Error: couldn't get ehdr for %s\n",
118 so
->low_addr
= low_addr
;
119 so
->high_addr
= so
->low_addr
+ so
->memsz
;
131 void so_info_destroy(struct so_info
*so
)
137 dwarf_end(so
->dwarf_info
);
140 free(so
->dwarf_path
);
142 free(so
->dbg_link_filename
);
144 elf_end(so
->elf_file
);
153 int so_info_set_build_id(struct so_info
*so
, uint8_t *build_id
,
156 if (!so
|| !build_id
) {
160 so
->build_id
= malloc(build_id_len
);
165 memcpy(so
->build_id
, build_id
, build_id_len
);
166 so
->build_id_len
= build_id_len
;
169 * Reset the is_elf_only flag in case it had been set
170 * previously, because we might find separate debug info using
171 * the new build id information.
173 so
->is_elf_only
= false;
183 int so_info_set_debug_link(struct so_info
*so
, char *filename
, uint32_t crc
)
185 if (!so
|| !filename
) {
189 so
->dbg_link_filename
= strdup(filename
);
190 if (!so
->dbg_link_filename
) {
194 so
->dbg_link_crc
= crc
;
197 * Reset the is_elf_only flag in case it had been set
198 * previously, because we might find separate debug info using
199 * the new build id information.
201 so
->is_elf_only
= false;
211 * Tries to read DWARF info from the location given by path, and
212 * attach it to the given so_info instance if it exists.
214 * @param so so_info instance for which to set DWARF info
215 * @param path Presumed location of the DWARF info
216 * @returns 0 on success, -1 on failure
219 int so_info_set_dwarf_info_from_path(struct so_info
*so
, char *path
)
221 int fd
= -1, ret
= 0;
222 struct bt_dwarf_cu
*cu
= NULL
;
223 Dwarf
*dwarf_info
= NULL
;
229 fd
= open(path
, O_RDONLY
);
234 dwarf_info
= dwarf_begin(fd
, DWARF_C_READ
);
240 * Check if the dwarf info has any CU. If not, the SO's object
241 * file contains no DWARF info.
243 cu
= bt_dwarf_cu_create(dwarf_info
);
248 ret
= bt_dwarf_cu_next(cu
);
254 so
->dwarf_path
= strdup(path
);
255 if (!so
->dwarf_path
) {
258 so
->dwarf_info
= dwarf_info
;
265 dwarf_end(dwarf_info
);
273 * Try to set the dwarf_info for a given so_info instance via the
276 * @param so so_info instance for which to retrieve the
277 * DWARF info via build ID
278 * @returns 0 on success (i.e. dwarf_info set), -1 on failure
281 int so_info_set_dwarf_info_build_id(struct so_info
*so
)
283 int i
= 0, ret
= 0, dbg_dir_trailing_slash
= 0;
284 char *path
= NULL
, *build_id_file
= NULL
;
285 const char *dbg_dir
= NULL
;
286 size_t build_id_file_len
, path_len
;
288 if (!so
|| !so
->build_id
) {
292 dbg_dir
= opt_debug_info_dir
? : DEFAULT_DEBUG_DIR
;
294 dbg_dir_trailing_slash
= dbg_dir
[strlen(dbg_dir
) - 1] == '/';
296 /* 2 characters per byte printed in hex, +2 for '/' and '\0' */
297 build_id_file_len
= (2 * so
->build_id_len
) + 2;
298 build_id_file
= malloc(build_id_file_len
);
299 if (!build_id_file
) {
303 snprintf(build_id_file
, 4, "%02x/", so
->build_id
[0]);
304 for (i
= 1; i
< so
->build_id_len
; ++i
) {
305 int path_idx
= 3 + 2 * (i
- 1);
307 snprintf(&build_id_file
[path_idx
], 3, "%02x", so
->build_id
[i
]);
310 path_len
= strlen(dbg_dir
) + strlen(BUILD_ID_SUBDIR
) +
311 strlen(build_id_file
) + strlen(BUILD_ID_SUFFIX
) + 1;
312 if (!dbg_dir_trailing_slash
) {
316 path
= malloc(path_len
);
321 strcpy(path
, dbg_dir
);
322 if (!dbg_dir_trailing_slash
) {
325 strcat(path
, BUILD_ID_SUBDIR
);
326 strcat(path
, build_id_file
);
327 strcat(path
, BUILD_ID_SUFFIX
);
329 ret
= so_info_set_dwarf_info_from_path(so
, path
);
346 * Tests whether the file located at path exists and has the expected
349 * This predicate is used when looking up separate debug info via the
350 * GNU debuglink method. The expected crc can be found .gnu_debuglink
351 * section in the original ELF file, along with the filename for the
352 * file containing the debug info.
354 * @param path Full path at which to look for the debug file
355 * @param crc Expected checksum for the debug file
356 * @returns 1 if the file exists and has the correct checksum,
360 int is_valid_debug_file(char *path
, uint32_t crc
)
362 int ret
= 0, fd
= -1;
369 fd
= open(path
, O_RDONLY
);
374 ret
= crc32(fd
, &_crc
);
388 * Try to set the dwarf_info for a given so_info instance via the
391 * @param so so_info instance for which to retrieve the
392 * DWARF info via debug link
393 * @returns 0 on success (i.e. dwarf_info set), -1 on failure
396 int so_info_set_dwarf_info_debug_link(struct so_info
*so
)
399 const char *dbg_dir
= NULL
;
400 char *dir_name
= NULL
, *so_dir
= NULL
, *path
= NULL
;
401 size_t max_path_len
= 0;
403 if (!so
|| !so
->dbg_link_filename
) {
407 dbg_dir
= opt_debug_info_dir
? : DEFAULT_DEBUG_DIR
;
409 dir_name
= dirname(so
->elf_path
);
414 /* so_dir is just dir_name with a trailing slash */
415 so_dir
= malloc(strlen(dir_name
) + 2);
420 strcpy(so_dir
, dir_name
);
423 max_path_len
= strlen(dbg_dir
) + strlen(so_dir
) +
424 strlen(DEBUG_SUBDIR
) + strlen(so
->dbg_link_filename
)
426 path
= malloc(max_path_len
);
431 /* First look in the SO's dir */
432 strcpy(path
, so_dir
);
433 strcat(path
, so
->dbg_link_filename
);
435 if (is_valid_debug_file(path
, so
->dbg_link_crc
)) {
439 /* If not found, look in .debug subdir */
440 strcpy(path
, so_dir
);
441 strcat(path
, DEBUG_SUBDIR
);
442 strcat(path
, so
->dbg_link_filename
);
444 if (is_valid_debug_file(path
, so
->dbg_link_crc
)) {
448 /* Lastly, look under the global debug directory */
449 strcpy(path
, dbg_dir
);
450 strcat(path
, so_dir
);
451 strcat(path
, so
->dbg_link_filename
);
453 if (is_valid_debug_file(path
, so
->dbg_link_crc
)) {
466 ret
= so_info_set_dwarf_info_from_path(so
, path
);
475 * Initialize the DWARF info for a given executable.
477 * @param so so_info instance
478 * @returns 0 on success, -1 on failure
481 int so_info_set_dwarf_info(struct so_info
*so
)
489 /* First try to set the DWARF info from the ELF file */
490 ret
= so_info_set_dwarf_info_from_path(so
, so
->elf_path
);
496 * If that fails, try to find separate debug info via build ID
499 ret
= so_info_set_dwarf_info_build_id(so
);
504 ret
= so_info_set_dwarf_info_debug_link(so
);
516 void source_location_destroy(struct source_location
*src_loc
)
522 free(src_loc
->filename
);
527 * Try to find the symbol closest to an address within a given ELF
530 * Only function symbols are taken into account. The symbol's address
531 * must precede `addr`. A symbol with a closer address might exist
532 * after `addr` but is irrelevant because it cannot encompass `addr`.
534 * On success, if found, the out parameters `sym` and `shdr` are
535 * set. On failure or if none are found, they remain unchanged.
537 * @param scn ELF section in which to look for the address
538 * @param addr Virtual memory address for which to find the
539 * nearest function symbol
540 * @param sym Out parameter, the nearest function symbol
541 * @param shdr Out parameter, the section header for scn
542 * @returns 0 on success, -1 on failure
545 int so_info_get_nearest_symbol_from_section(Elf_Scn
*scn
, uint64_t addr
,
546 GElf_Sym
**sym
, GElf_Shdr
**shdr
)
550 Elf_Data
*data
= NULL
;
551 GElf_Shdr
*_shdr
= NULL
;
552 GElf_Sym
*nearest_sym
= NULL
;
554 if (!scn
|| !sym
|| !shdr
) {
558 _shdr
= g_new0(GElf_Shdr
, 1);
563 _shdr
= gelf_getshdr(scn
, _shdr
);
568 if (_shdr
->sh_type
!= SHT_SYMTAB
) {
570 * We are only interested in symbol table (symtab)
571 * sections, skip this one.
576 data
= elf_getdata(scn
, NULL
);
581 symbol_count
= _shdr
->sh_size
/ _shdr
->sh_entsize
;
583 for (i
= 0; i
< symbol_count
; ++i
) {
584 GElf_Sym
*cur_sym
= NULL
;
586 cur_sym
= g_new0(GElf_Sym
, 1);
590 cur_sym
= gelf_getsym(data
, i
, cur_sym
);
594 if (GELF_ST_TYPE(cur_sym
->st_info
) != STT_FUNC
) {
595 /* We're only interested in the functions. */
600 if (cur_sym
->st_value
<= addr
&&
602 cur_sym
->st_value
> nearest_sym
->st_value
)) {
604 nearest_sym
= cur_sym
;
627 * Get the name of the function containing a given address within an
628 * executable using ELF symbols.
630 * The function name is in fact the name of the nearest ELF symbol,
631 * followed by the offset in bytes between the address and the symbol
632 * (in hex), separated by a '+' character.
634 * If found, the out parameter `func_name` is set on success. On failure,
635 * it remains unchanged.
637 * @param so so_info instance for the executable containing
639 * @param addr Virtual memory address for which to find the
641 * @param func_name Out parameter, the function name
642 * @returns 0 on success, -1 on failure
645 int so_info_lookup_elf_function_name(struct so_info
*so
, uint64_t addr
,
649 * TODO (possible optimisation): if an ELF has no symtab
650 * section, it has been stripped. Therefore, it would be wise
651 * to store a flag indicating the stripped status after the
652 * first iteration to prevent subsequent ones.
656 GElf_Sym
*sym
= NULL
;
657 GElf_Shdr
*shdr
= NULL
;
658 char *sym_name
= NULL
;
659 char *_func_name
= NULL
;
660 char offset_str
[ADDR_STR_LEN
];
662 scn
= elf_nextscn(so
->elf_file
, scn
);
667 while (scn
&& !sym
) {
668 ret
= so_info_get_nearest_symbol_from_section(
669 scn
, addr
, &sym
, &shdr
);
674 scn
= elf_nextscn(so
->elf_file
, scn
);
678 sym_name
= elf_strptr(so
->elf_file
, shdr
->sh_link
,
684 snprintf(offset_str
, ADDR_STR_LEN
, "+%#0" PRIx64
,
685 addr
- sym
->st_value
);
686 _func_name
= malloc(strlen(sym_name
) + ADDR_STR_LEN
);
691 strcpy(_func_name
, sym_name
);
692 strcat(_func_name
, offset_str
);
693 *func_name
= _func_name
;
708 * Get the name of the function containing a given address within a
709 * given compile unit (CU).
711 * If found, the out parameter `func_name` is set on success. On
712 * failure, it remains unchanged.
714 * @param cu bt_dwarf_cu instance which may contain the address
715 * @param addr Virtual memory address for which to find the
717 * @param func_name Out parameter, the function name
718 * @returns 0 on success, -1 on failure
721 int so_info_lookup_cu_function_name(struct bt_dwarf_cu
*cu
, uint64_t addr
,
724 int ret
= 0, found
= 0;
725 char *_func_name
= NULL
;
726 struct bt_dwarf_die
*die
= NULL
;
728 if (!cu
|| !func_name
) {
732 die
= bt_dwarf_die_create(cu
);
737 while (bt_dwarf_die_next(die
) == 0) {
740 ret
= bt_dwarf_die_get_tag(die
, &tag
);
745 if (tag
== DW_TAG_subprogram
) {
746 ret
= bt_dwarf_die_contains_addr(die
, addr
, &found
);
758 ret
= bt_dwarf_die_get_name(die
, &_func_name
);
763 *func_name
= _func_name
;
766 bt_dwarf_die_destroy(die
);
770 bt_dwarf_die_destroy(die
);
775 * Get the name of the function containing a given address within an
776 * executable using DWARF debug info.
778 * If found, the out parameter `func_name` is set on success. On
779 * failure, it remains unchanged.
781 * @param so so_info instance for the executable containing
783 * @param addr Virtual memory address for which to find the
785 * @param func_name Out parameter, the function name
786 * @returns 0 on success, -1 on failure
789 int so_info_lookup_dwarf_function_name(struct so_info
*so
, uint64_t addr
,
793 char *_func_name
= NULL
;
794 struct bt_dwarf_cu
*cu
= NULL
;
796 if (!so
|| !func_name
) {
800 cu
= bt_dwarf_cu_create(so
->dwarf_info
);
805 while (bt_dwarf_cu_next(cu
) == 0) {
806 ret
= so_info_lookup_cu_function_name(cu
, addr
, &_func_name
);
817 *func_name
= _func_name
;
820 bt_dwarf_cu_destroy(cu
);
824 bt_dwarf_cu_destroy(cu
);
829 int so_info_lookup_function_name(struct so_info
*so
, uint64_t ip
,
833 char *_func_name
= NULL
;
834 uint64_t relative_addr
;
836 if (!so
|| !func_name
) {
840 /* Set DWARF info if it hasn't been accessed yet. */
841 if (!so
->dwarf_info
&& !so
->is_elf_only
) {
842 ret
= so_info_set_dwarf_info(so
);
844 /* Failed to set DWARF info, fallback to ELF. */
845 so
->is_elf_only
= true;
849 if (!so_info_has_address(so
, ip
)) {
853 relative_addr
= ip
- so
->low_addr
;
855 * Addresses in ELF and DWARF are relative to base address for
856 * PIC, so make the address argument relative too if needed.
858 if (so
->is_elf_only
) {
859 ret
= so_info_lookup_elf_function_name(so
,
860 so
->is_pic
? relative_addr
: ip
,
863 ret
= so_info_lookup_dwarf_function_name(so
,
864 so
->is_pic
? relative_addr
: ip
,
874 * Can't map to a function; fallback to a generic output of the
875 * form binary+/@address.
877 * FIXME check position independence flag.
879 const char *binary_name
= get_filename_from_path(so
->elf_path
);
881 ret
= asprintf(&_func_name
, "%s+%#0" PRIx64
, binary_name
,
888 *func_name
= _func_name
;
896 * Predicate used to determine whether the children of a given DIE
897 * contain a specific address.
899 * More specifically, the parameter `die` is expected to be a
900 * subprogram (function) DIE, and this predicate tells whether any
901 * subroutines are inlined within this function and would contain
904 * Do note that this function advances the position of `die`. If the
905 * address is found within one of its children, `die` will be pointing
906 * to that child upon returning from the function, allowing to extract
907 * the information deemed necessary.
909 * @param die The parent DIE in whose children the address will be
911 * @param addr The address for which to look for in the DIEs
912 * @returns Returns 1 if the address was found, 0 if not
915 int so_info_child_die_has_address(struct bt_dwarf_die
*die
, uint64_t addr
)
917 int ret
= 0, contains
= 0;
923 ret
= bt_dwarf_die_child(die
);
931 ret
= bt_dwarf_die_get_tag(die
, &tag
);
936 if (tag
== DW_TAG_inlined_subroutine
) {
937 ret
= bt_dwarf_die_contains_addr(die
, addr
, &contains
);
947 } while (bt_dwarf_die_next(die
) == 0);
958 * Lookup the source location for a given address within a CU, making
959 * the assumption that it is contained within an inline routine in a
962 * @param cu bt_dwarf_cu instance in which to look for the address
963 * @param addr The address for which to look for
964 * @param src_loc Out parameter, the source location (filename and
965 * line number) for the address
966 * @returns 0 on success, -1 on failure
969 int so_info_lookup_cu_src_loc_inl(struct bt_dwarf_cu
*cu
, uint64_t addr
,
970 struct source_location
**src_loc
)
972 int ret
= 0, found
= 0;
973 struct bt_dwarf_die
*die
= NULL
;
974 struct source_location
*_src_loc
= NULL
;
976 if (!cu
|| !src_loc
) {
980 die
= bt_dwarf_die_create(cu
);
985 while (bt_dwarf_die_next(die
) == 0) {
988 ret
= bt_dwarf_die_get_tag(die
, &tag
);
993 if (tag
== DW_TAG_subprogram
) {
996 ret
= bt_dwarf_die_contains_addr(die
, addr
, &contains
);
1003 * Try to find an inlined subroutine
1004 * child of this DIE containing addr.
1006 found
= so_info_child_die_has_address(
1015 char *filename
= NULL
;
1018 _src_loc
= g_new0(struct source_location
, 1);
1023 ret
= bt_dwarf_die_get_call_file(die
, &filename
);
1027 ret
= bt_dwarf_die_get_call_line(die
, &line_no
);
1033 _src_loc
->filename
= filename
;
1034 _src_loc
->line_no
= line_no
;
1035 *src_loc
= _src_loc
;
1038 bt_dwarf_die_destroy(die
);
1042 source_location_destroy(_src_loc
);
1043 bt_dwarf_die_destroy(die
);
1048 * Lookup the source location for a given address within a CU,
1049 * assuming that it is contained within an inlined function.
1051 * A source location can be found regardless of inlining status for
1052 * this method, but in the case of an inlined function, the returned
1053 * source location will point not to the callsite but rather to the
1054 * definition site of the inline function.
1056 * @param cu bt_dwarf_cu instance in which to look for the address
1057 * @param addr The address for which to look for
1058 * @param src_loc Out parameter, the source location (filename and
1059 * line number) for the address
1060 * @returns 0 on success, -1 on failure
1063 int so_info_lookup_cu_src_loc_no_inl(struct bt_dwarf_cu
*cu
, uint64_t addr
,
1064 struct source_location
**src_loc
)
1066 struct source_location
*_src_loc
= NULL
;
1067 struct bt_dwarf_die
*die
= NULL
;
1068 const char *filename
= NULL
;
1069 Dwarf_Line
*line
= NULL
;
1070 Dwarf_Addr line_addr
;
1073 if (!cu
|| !src_loc
) {
1077 die
= bt_dwarf_die_create(cu
);
1082 line
= dwarf_getsrc_die(die
->dwarf_die
, addr
);
1087 ret
= dwarf_lineaddr(line
, &line_addr
);
1092 filename
= dwarf_linesrc(line
, NULL
, NULL
);
1097 if (addr
== line_addr
) {
1098 _src_loc
= g_new0(struct source_location
, 1);
1103 ret
= dwarf_lineno(line
, &line_no
);
1108 _src_loc
->line_no
= line_no
;
1109 _src_loc
->filename
= strdup(filename
);
1112 bt_dwarf_die_destroy(die
);
1115 *src_loc
= _src_loc
;
1121 source_location_destroy(_src_loc
);
1122 bt_dwarf_die_destroy(die
);
1127 * Get the source location (file name and line number) for a given
1128 * address within a compile unit (CU).
1130 * On success, the out parameter `src_loc` is set if found. On
1131 * failure, it remains unchanged.
1133 * @param so bt_dwarf_cu instance for the compile unit which
1134 * may contain the address
1135 * @param addr Virtual memory address for which to find the
1137 * @param src_loc Out parameter, the source location
1138 * @returns 0 on success, -1 on failure
1141 int so_info_lookup_cu_src_loc(struct bt_dwarf_cu
*cu
, uint64_t addr
,
1142 struct source_location
**src_loc
)
1145 struct source_location
*_src_loc
= NULL
;
1147 if (!cu
|| !src_loc
) {
1151 ret
= so_info_lookup_cu_src_loc_inl(cu
, addr
, &_src_loc
);
1160 ret
= so_info_lookup_cu_src_loc_no_inl(cu
, addr
, &_src_loc
);
1171 *src_loc
= _src_loc
;
1177 source_location_destroy(_src_loc
);
1182 int so_info_lookup_source_location(struct so_info
*so
, uint64_t addr
,
1183 struct source_location
**src_loc
)
1185 struct bt_dwarf_cu
*cu
= NULL
;
1186 struct source_location
*_src_loc
= NULL
;
1188 if (!so
|| !src_loc
) {
1192 /* Set DWARF info if it hasn't been accessed yet. */
1193 if (!so
->dwarf_info
&& !so
->is_elf_only
) {
1194 if (so_info_set_dwarf_info(so
)) {
1195 /* Failed to set DWARF info. */
1196 so
->is_elf_only
= true;
1200 if (so
->is_elf_only
) {
1201 /* We cannot lookup source location without DWARF info. */
1205 if (!so_info_has_address(so
, addr
)) {
1210 * Addresses in ELF and DWARF are relative to base address for
1211 * PIC, so make the address argument relative too if needed.
1214 addr
-= so
->low_addr
;
1217 cu
= bt_dwarf_cu_create(so
->dwarf_info
);
1222 while (bt_dwarf_cu_next(cu
) == 0) {
1225 ret
= so_info_lookup_cu_src_loc(cu
, addr
, &_src_loc
);
1235 bt_dwarf_cu_destroy(cu
);
1237 *src_loc
= _src_loc
;
1243 source_location_destroy(_src_loc
);
1244 bt_dwarf_cu_destroy(cu
);