forked from abseil/abseil-cpp
-
Notifications
You must be signed in to change notification settings - Fork 0
/
symbolize_elf.inc
1647 lines (1484 loc) · 56.1 KB
/
symbolize_elf.inc
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
// Copyright 2018 The Abseil Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// This library provides Symbolize() function that symbolizes program
// counters to their corresponding symbol names on linux platforms.
// This library has a minimal implementation of an ELF symbol table
// reader (i.e. it doesn't depend on libelf, etc.).
//
// The algorithm used in Symbolize() is as follows.
//
// 1. Go through a list of maps in /proc/self/maps and find the map
// containing the program counter.
//
// 2. Open the mapped file and find a regular symbol table inside.
// Iterate over symbols in the symbol table and look for the symbol
// containing the program counter. If such a symbol is found,
// obtain the symbol name, and demangle the symbol if possible.
// If the symbol isn't found in the regular symbol table (binary is
// stripped), try the same thing with a dynamic symbol table.
//
// Note that Symbolize() is originally implemented to be used in
// signal handlers, hence it doesn't use malloc() and other unsafe
// operations. It should be both thread-safe and async-signal-safe.
//
// Implementation note:
//
// We don't use heaps but only use stacks. We want to reduce the
// stack consumption so that the symbolizer can run on small stacks.
//
// Here are some numbers collected with GCC 4.1.0 on x86:
// - sizeof(Elf32_Sym) = 16
// - sizeof(Elf32_Shdr) = 40
// - sizeof(Elf64_Sym) = 24
// - sizeof(Elf64_Shdr) = 64
//
// This implementation is intended to be async-signal-safe but uses some
// functions which are not guaranteed to be so, such as memchr() and
// memmove(). We assume they are async-signal-safe.
#include <dlfcn.h>
#include <elf.h>
#include <fcntl.h>
#include <link.h> // For ElfW() macro.
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
#include <algorithm>
#include <array>
#include <atomic>
#include <cerrno>
#include <cinttypes>
#include <climits>
#include <cstdint>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include "absl/base/casts.h"
#include "absl/base/dynamic_annotations.h"
#include "absl/base/internal/low_level_alloc.h"
#include "absl/base/internal/raw_logging.h"
#include "absl/base/internal/spinlock.h"
#include "absl/base/port.h"
#include "absl/debugging/internal/demangle.h"
#include "absl/debugging/internal/vdso_support.h"
#include "absl/strings/string_view.h"
#if defined(__FreeBSD__) && !defined(ElfW)
#define ElfW(x) __ElfN(x)
#endif
namespace absl {
ABSL_NAMESPACE_BEGIN
// Value of argv[0]. Used by MaybeInitializeObjFile().
static char *argv0_value = nullptr;
void InitializeSymbolizer(const char *argv0) {
#ifdef ABSL_HAVE_VDSO_SUPPORT
// We need to make sure VDSOSupport::Init() is called before any setuid or
// chroot calls, so InitializeSymbolizer() should be called very early in the
// life of a program.
absl::debugging_internal::VDSOSupport::Init();
#endif
if (argv0_value != nullptr) {
free(argv0_value);
argv0_value = nullptr;
}
if (argv0 != nullptr && argv0[0] != '\0') {
argv0_value = strdup(argv0);
}
}
namespace debugging_internal {
namespace {
// Re-runs fn until it doesn't cause EINTR.
#define NO_INTR(fn) \
do { \
} while ((fn) < 0 && errno == EINTR)
// On Linux, ELF_ST_* are defined in <linux/elf.h>. To make this portable
// we define our own ELF_ST_BIND and ELF_ST_TYPE if not available.
#ifndef ELF_ST_BIND
#define ELF_ST_BIND(info) (((unsigned char)(info)) >> 4)
#endif
#ifndef ELF_ST_TYPE
#define ELF_ST_TYPE(info) (((unsigned char)(info)) & 0xF)
#endif
// Some platforms use a special .opd section to store function pointers.
const char kOpdSectionName[] = ".opd";
#if (defined(__powerpc__) && !(_CALL_ELF > 1)) || defined(__ia64)
// Use opd section for function descriptors on these platforms, the function
// address is the first word of the descriptor.
enum { kPlatformUsesOPDSections = 1 };
#else // not PPC or IA64
enum { kPlatformUsesOPDSections = 0 };
#endif
// This works for PowerPC & IA64 only. A function descriptor consist of two
// pointers and the first one is the function's entry.
const size_t kFunctionDescriptorSize = sizeof(void *) * 2;
const int kMaxDecorators = 10; // Seems like a reasonable upper limit.
struct InstalledSymbolDecorator {
SymbolDecorator fn;
void *arg;
int ticket;
};
int g_num_decorators;
InstalledSymbolDecorator g_decorators[kMaxDecorators];
struct FileMappingHint {
const void *start;
const void *end;
uint64_t offset;
const char *filename;
};
// Protects g_decorators.
// We are using SpinLock and not a Mutex here, because we may be called
// from inside Mutex::Lock itself, and it prohibits recursive calls.
// This happens in e.g. base/stacktrace_syscall_unittest.
// Moreover, we are using only TryLock(), if the decorator list
// is being modified (is busy), we skip all decorators, and possibly
// loose some info. Sorry, that's the best we could do.
ABSL_CONST_INIT absl::base_internal::SpinLock g_decorators_mu(
absl::kConstInit, absl::base_internal::SCHEDULE_KERNEL_ONLY);
const int kMaxFileMappingHints = 8;
int g_num_file_mapping_hints;
FileMappingHint g_file_mapping_hints[kMaxFileMappingHints];
// Protects g_file_mapping_hints.
ABSL_CONST_INIT absl::base_internal::SpinLock g_file_mapping_mu(
absl::kConstInit, absl::base_internal::SCHEDULE_KERNEL_ONLY);
// Async-signal-safe function to zero a buffer.
// memset() is not guaranteed to be async-signal-safe.
static void SafeMemZero(void* p, size_t size) {
unsigned char *c = static_cast<unsigned char *>(p);
while (size--) {
*c++ = 0;
}
}
struct ObjFile {
ObjFile()
: filename(nullptr),
start_addr(nullptr),
end_addr(nullptr),
offset(0),
fd(-1),
elf_type(-1) {
SafeMemZero(&elf_header, sizeof(elf_header));
SafeMemZero(&phdr[0], sizeof(phdr));
}
char *filename;
const void *start_addr;
const void *end_addr;
uint64_t offset;
// The following fields are initialized on the first access to the
// object file.
int fd;
int elf_type;
ElfW(Ehdr) elf_header;
// PT_LOAD program header describing executable code.
// Normally we expect just one, but SWIFT binaries have two.
// CUDA binaries have 3 (see cr/473913254 description).
std::array<ElfW(Phdr), 4> phdr;
};
// Build 4-way associative cache for symbols. Within each cache line, symbols
// are replaced in LRU order.
enum {
ASSOCIATIVITY = 4,
};
struct SymbolCacheLine {
const void *pc[ASSOCIATIVITY];
char *name[ASSOCIATIVITY];
// age[i] is incremented when a line is accessed. it's reset to zero if the
// i'th entry is read.
uint32_t age[ASSOCIATIVITY];
};
// ---------------------------------------------------------------
// An async-signal-safe arena for LowLevelAlloc
static std::atomic<base_internal::LowLevelAlloc::Arena *> g_sig_safe_arena;
static base_internal::LowLevelAlloc::Arena *SigSafeArena() {
return g_sig_safe_arena.load(std::memory_order_acquire);
}
static void InitSigSafeArena() {
if (SigSafeArena() == nullptr) {
base_internal::LowLevelAlloc::Arena *new_arena =
base_internal::LowLevelAlloc::NewArena(
base_internal::LowLevelAlloc::kAsyncSignalSafe);
base_internal::LowLevelAlloc::Arena *old_value = nullptr;
if (!g_sig_safe_arena.compare_exchange_strong(old_value, new_arena,
std::memory_order_release,
std::memory_order_relaxed)) {
// We lost a race to allocate an arena; deallocate.
base_internal::LowLevelAlloc::DeleteArena(new_arena);
}
}
}
// ---------------------------------------------------------------
// An AddrMap is a vector of ObjFile, using SigSafeArena() for allocation.
class AddrMap {
public:
AddrMap() : size_(0), allocated_(0), obj_(nullptr) {}
~AddrMap() { base_internal::LowLevelAlloc::Free(obj_); }
size_t Size() const { return size_; }
ObjFile *At(size_t i) { return &obj_[i]; }
ObjFile *Add();
void Clear();
private:
size_t size_; // count of valid elements (<= allocated_)
size_t allocated_; // count of allocated elements
ObjFile *obj_; // array of allocated_ elements
AddrMap(const AddrMap &) = delete;
AddrMap &operator=(const AddrMap &) = delete;
};
void AddrMap::Clear() {
for (size_t i = 0; i != size_; i++) {
At(i)->~ObjFile();
}
size_ = 0;
}
ObjFile *AddrMap::Add() {
if (size_ == allocated_) {
size_t new_allocated = allocated_ * 2 + 50;
ObjFile *new_obj_ =
static_cast<ObjFile *>(base_internal::LowLevelAlloc::AllocWithArena(
new_allocated * sizeof(*new_obj_), SigSafeArena()));
if (obj_) {
memcpy(new_obj_, obj_, allocated_ * sizeof(*new_obj_));
base_internal::LowLevelAlloc::Free(obj_);
}
obj_ = new_obj_;
allocated_ = new_allocated;
}
return new (&obj_[size_++]) ObjFile;
}
// ---------------------------------------------------------------
enum FindSymbolResult { SYMBOL_NOT_FOUND = 1, SYMBOL_TRUNCATED, SYMBOL_FOUND };
class Symbolizer {
public:
Symbolizer();
~Symbolizer();
const char *GetSymbol(const void *const pc);
private:
char *CopyString(const char *s) {
size_t len = strlen(s);
char *dst = static_cast<char *>(
base_internal::LowLevelAlloc::AllocWithArena(len + 1, SigSafeArena()));
ABSL_RAW_CHECK(dst != nullptr, "out of memory");
memcpy(dst, s, len + 1);
return dst;
}
ObjFile *FindObjFile(const void *const start,
size_t size) ABSL_ATTRIBUTE_NOINLINE;
static bool RegisterObjFile(const char *filename,
const void *const start_addr,
const void *const end_addr, uint64_t offset,
void *arg);
SymbolCacheLine *GetCacheLine(const void *const pc);
const char *FindSymbolInCache(const void *const pc);
const char *InsertSymbolInCache(const void *const pc, const char *name);
void AgeSymbols(SymbolCacheLine *line);
void ClearAddrMap();
FindSymbolResult GetSymbolFromObjectFile(const ObjFile &obj,
const void *const pc,
const ptrdiff_t relocation,
char *out, size_t out_size,
char *tmp_buf, size_t tmp_buf_size);
const char *GetUncachedSymbol(const void *pc);
enum {
SYMBOL_BUF_SIZE = 3072,
TMP_BUF_SIZE = 1024,
SYMBOL_CACHE_LINES = 128,
};
AddrMap addr_map_;
bool ok_;
bool addr_map_read_;
char symbol_buf_[SYMBOL_BUF_SIZE];
// tmp_buf_ will be used to store arrays of ElfW(Shdr) and ElfW(Sym)
// so we ensure that tmp_buf_ is properly aligned to store either.
alignas(16) char tmp_buf_[TMP_BUF_SIZE];
static_assert(alignof(ElfW(Shdr)) <= 16,
"alignment of tmp buf too small for Shdr");
static_assert(alignof(ElfW(Sym)) <= 16,
"alignment of tmp buf too small for Sym");
SymbolCacheLine symbol_cache_[SYMBOL_CACHE_LINES];
};
static std::atomic<Symbolizer *> g_cached_symbolizer;
} // namespace
static size_t SymbolizerSize() {
#if defined(__wasm__) || defined(__asmjs__)
auto pagesize = static_cast<size_t>(getpagesize());
#else
auto pagesize = static_cast<size_t>(sysconf(_SC_PAGESIZE));
#endif
return ((sizeof(Symbolizer) - 1) / pagesize + 1) * pagesize;
}
// Return (and set null) g_cached_symbolized_state if it is not null.
// Otherwise return a new symbolizer.
static Symbolizer *AllocateSymbolizer() {
InitSigSafeArena();
Symbolizer *symbolizer =
g_cached_symbolizer.exchange(nullptr, std::memory_order_acquire);
if (symbolizer != nullptr) {
return symbolizer;
}
return new (base_internal::LowLevelAlloc::AllocWithArena(
SymbolizerSize(), SigSafeArena())) Symbolizer();
}
// Set g_cached_symbolize_state to s if it is null, otherwise
// delete s.
static void FreeSymbolizer(Symbolizer *s) {
Symbolizer *old_cached_symbolizer = nullptr;
if (!g_cached_symbolizer.compare_exchange_strong(old_cached_symbolizer, s,
std::memory_order_release,
std::memory_order_relaxed)) {
s->~Symbolizer();
base_internal::LowLevelAlloc::Free(s);
}
}
Symbolizer::Symbolizer() : ok_(true), addr_map_read_(false) {
for (SymbolCacheLine &symbol_cache_line : symbol_cache_) {
for (size_t j = 0; j < ABSL_ARRAYSIZE(symbol_cache_line.name); ++j) {
symbol_cache_line.pc[j] = nullptr;
symbol_cache_line.name[j] = nullptr;
symbol_cache_line.age[j] = 0;
}
}
}
Symbolizer::~Symbolizer() {
for (SymbolCacheLine &symbol_cache_line : symbol_cache_) {
for (char *s : symbol_cache_line.name) {
base_internal::LowLevelAlloc::Free(s);
}
}
ClearAddrMap();
}
// We don't use assert() since it's not guaranteed to be
// async-signal-safe. Instead we define a minimal assertion
// macro. So far, we don't need pretty printing for __FILE__, etc.
#define SAFE_ASSERT(expr) ((expr) ? static_cast<void>(0) : abort())
// Read up to "count" bytes from file descriptor "fd" into the buffer
// starting at "buf" while handling short reads and EINTR. On
// success, return the number of bytes read. Otherwise, return -1.
static ssize_t ReadPersistent(int fd, void *buf, size_t count) {
SAFE_ASSERT(fd >= 0);
SAFE_ASSERT(count <= SSIZE_MAX);
char *buf0 = reinterpret_cast<char *>(buf);
size_t num_bytes = 0;
while (num_bytes < count) {
ssize_t len;
NO_INTR(len = read(fd, buf0 + num_bytes, count - num_bytes));
if (len < 0) { // There was an error other than EINTR.
ABSL_RAW_LOG(WARNING, "read failed: errno=%d", errno);
return -1;
}
if (len == 0) { // Reached EOF.
break;
}
num_bytes += static_cast<size_t>(len);
}
SAFE_ASSERT(num_bytes <= count);
return static_cast<ssize_t>(num_bytes);
}
// Read up to "count" bytes from "offset" in the file pointed by file
// descriptor "fd" into the buffer starting at "buf". On success,
// return the number of bytes read. Otherwise, return -1.
static ssize_t ReadFromOffset(const int fd, void *buf, const size_t count,
const off_t offset) {
off_t off = lseek(fd, offset, SEEK_SET);
if (off == (off_t)-1) {
ABSL_RAW_LOG(WARNING, "lseek(%d, %jd, SEEK_SET) failed: errno=%d", fd,
static_cast<intmax_t>(offset), errno);
return -1;
}
return ReadPersistent(fd, buf, count);
}
// Try reading exactly "count" bytes from "offset" bytes in a file
// pointed by "fd" into the buffer starting at "buf" while handling
// short reads and EINTR. On success, return true. Otherwise, return
// false.
static bool ReadFromOffsetExact(const int fd, void *buf, const size_t count,
const off_t offset) {
ssize_t len = ReadFromOffset(fd, buf, count, offset);
return len >= 0 && static_cast<size_t>(len) == count;
}
// Returns elf_header.e_type if the file pointed by fd is an ELF binary.
static int FileGetElfType(const int fd) {
ElfW(Ehdr) elf_header;
if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) {
return -1;
}
if (memcmp(elf_header.e_ident, ELFMAG, SELFMAG) != 0) {
return -1;
}
return elf_header.e_type;
}
// Read the section headers in the given ELF binary, and if a section
// of the specified type is found, set the output to this section header
// and return true. Otherwise, return false.
// To keep stack consumption low, we would like this function to not get
// inlined.
static ABSL_ATTRIBUTE_NOINLINE bool GetSectionHeaderByType(
const int fd, ElfW(Half) sh_num, const off_t sh_offset, ElfW(Word) type,
ElfW(Shdr) * out, char *tmp_buf, size_t tmp_buf_size) {
ElfW(Shdr) *buf = reinterpret_cast<ElfW(Shdr) *>(tmp_buf);
const size_t buf_entries = tmp_buf_size / sizeof(buf[0]);
const size_t buf_bytes = buf_entries * sizeof(buf[0]);
for (size_t i = 0; static_cast<int>(i) < sh_num;) {
const size_t num_bytes_left =
(static_cast<size_t>(sh_num) - i) * sizeof(buf[0]);
const size_t num_bytes_to_read =
(buf_bytes > num_bytes_left) ? num_bytes_left : buf_bytes;
const off_t offset = sh_offset + static_cast<off_t>(i * sizeof(buf[0]));
const ssize_t len = ReadFromOffset(fd, buf, num_bytes_to_read, offset);
if (len < 0) {
ABSL_RAW_LOG(
WARNING,
"Reading %zu bytes from offset %ju returned %zd which is negative.",
num_bytes_to_read, static_cast<intmax_t>(offset), len);
return false;
}
if (static_cast<size_t>(len) % sizeof(buf[0]) != 0) {
ABSL_RAW_LOG(
WARNING,
"Reading %zu bytes from offset %jd returned %zd which is not a "
"multiple of %zu.",
num_bytes_to_read, static_cast<intmax_t>(offset), len,
sizeof(buf[0]));
return false;
}
const size_t num_headers_in_buf = static_cast<size_t>(len) / sizeof(buf[0]);
SAFE_ASSERT(num_headers_in_buf <= buf_entries);
for (size_t j = 0; j < num_headers_in_buf; ++j) {
if (buf[j].sh_type == type) {
*out = buf[j];
return true;
}
}
i += num_headers_in_buf;
}
return false;
}
// There is no particular reason to limit section name to 63 characters,
// but there has (as yet) been no need for anything longer either.
const int kMaxSectionNameLen = 64;
bool ForEachSection(int fd,
const std::function<bool(absl::string_view name,
const ElfW(Shdr) &)> &callback) {
ElfW(Ehdr) elf_header;
if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) {
return false;
}
// Technically it can be larger, but in practice this never happens.
if (elf_header.e_shentsize != sizeof(ElfW(Shdr))) {
return false;
}
ElfW(Shdr) shstrtab;
off_t shstrtab_offset = static_cast<off_t>(elf_header.e_shoff) +
elf_header.e_shentsize * elf_header.e_shstrndx;
if (!ReadFromOffsetExact(fd, &shstrtab, sizeof(shstrtab), shstrtab_offset)) {
return false;
}
for (int i = 0; i < elf_header.e_shnum; ++i) {
ElfW(Shdr) out;
off_t section_header_offset =
static_cast<off_t>(elf_header.e_shoff) + elf_header.e_shentsize * i;
if (!ReadFromOffsetExact(fd, &out, sizeof(out), section_header_offset)) {
return false;
}
off_t name_offset = static_cast<off_t>(shstrtab.sh_offset) + out.sh_name;
char header_name[kMaxSectionNameLen];
ssize_t n_read =
ReadFromOffset(fd, &header_name, kMaxSectionNameLen, name_offset);
if (n_read < 0) {
return false;
} else if (n_read > kMaxSectionNameLen) {
// Long read?
return false;
}
absl::string_view name(header_name,
strnlen(header_name, static_cast<size_t>(n_read)));
if (!callback(name, out)) {
break;
}
}
return true;
}
// name_len should include terminating '\0'.
bool GetSectionHeaderByName(int fd, const char *name, size_t name_len,
ElfW(Shdr) * out) {
char header_name[kMaxSectionNameLen];
if (sizeof(header_name) < name_len) {
ABSL_RAW_LOG(WARNING,
"Section name '%s' is too long (%zu); "
"section will not be found (even if present).",
name, name_len);
// No point in even trying.
return false;
}
ElfW(Ehdr) elf_header;
if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) {
return false;
}
// Technically it can be larger, but in practice this never happens.
if (elf_header.e_shentsize != sizeof(ElfW(Shdr))) {
return false;
}
ElfW(Shdr) shstrtab;
off_t shstrtab_offset = static_cast<off_t>(elf_header.e_shoff) +
elf_header.e_shentsize * elf_header.e_shstrndx;
if (!ReadFromOffsetExact(fd, &shstrtab, sizeof(shstrtab), shstrtab_offset)) {
return false;
}
for (int i = 0; i < elf_header.e_shnum; ++i) {
off_t section_header_offset =
static_cast<off_t>(elf_header.e_shoff) + elf_header.e_shentsize * i;
if (!ReadFromOffsetExact(fd, out, sizeof(*out), section_header_offset)) {
return false;
}
off_t name_offset = static_cast<off_t>(shstrtab.sh_offset) + out->sh_name;
ssize_t n_read = ReadFromOffset(fd, &header_name, name_len, name_offset);
if (n_read < 0) {
return false;
} else if (static_cast<size_t>(n_read) != name_len) {
// Short read -- name could be at end of file.
continue;
}
if (memcmp(header_name, name, name_len) == 0) {
return true;
}
}
return false;
}
// Compare symbols at in the same address.
// Return true if we should pick symbol1.
static bool ShouldPickFirstSymbol(const ElfW(Sym) & symbol1,
const ElfW(Sym) & symbol2) {
// If one of the symbols is weak and the other is not, pick the one
// this is not a weak symbol.
char bind1 = ELF_ST_BIND(symbol1.st_info);
char bind2 = ELF_ST_BIND(symbol1.st_info);
if (bind1 == STB_WEAK && bind2 != STB_WEAK) return false;
if (bind2 == STB_WEAK && bind1 != STB_WEAK) return true;
// If one of the symbols has zero size and the other is not, pick the
// one that has non-zero size.
if (symbol1.st_size != 0 && symbol2.st_size == 0) {
return true;
}
if (symbol1.st_size == 0 && symbol2.st_size != 0) {
return false;
}
// If one of the symbols has no type and the other is not, pick the
// one that has a type.
char type1 = ELF_ST_TYPE(symbol1.st_info);
char type2 = ELF_ST_TYPE(symbol1.st_info);
if (type1 != STT_NOTYPE && type2 == STT_NOTYPE) {
return true;
}
if (type1 == STT_NOTYPE && type2 != STT_NOTYPE) {
return false;
}
// Pick the first one, if we still cannot decide.
return true;
}
// Return true if an address is inside a section.
static bool InSection(const void *address, const ElfW(Shdr) * section) {
const char *start = reinterpret_cast<const char *>(section->sh_addr);
size_t size = static_cast<size_t>(section->sh_size);
return start <= address && address < (start + size);
}
static const char *ComputeOffset(const char *base, ptrdiff_t offset) {
// Note: cast to intptr_t to avoid undefined behavior when base evaluates to
// zero and offset is non-zero.
return reinterpret_cast<const char *>(reinterpret_cast<intptr_t>(base) +
offset);
}
// Read a symbol table and look for the symbol containing the
// pc. Iterate over symbols in a symbol table and look for the symbol
// containing "pc". If the symbol is found, and its name fits in
// out_size, the name is written into out and SYMBOL_FOUND is returned.
// If the name does not fit, truncated name is written into out,
// and SYMBOL_TRUNCATED is returned. Out is NUL-terminated.
// If the symbol is not found, SYMBOL_NOT_FOUND is returned;
// To keep stack consumption low, we would like this function to not get
// inlined.
static ABSL_ATTRIBUTE_NOINLINE FindSymbolResult FindSymbol(
const void *const pc, const int fd, char *out, size_t out_size,
ptrdiff_t relocation, const ElfW(Shdr) * strtab, const ElfW(Shdr) * symtab,
const ElfW(Shdr) * opd, char *tmp_buf, size_t tmp_buf_size) {
if (symtab == nullptr) {
return SYMBOL_NOT_FOUND;
}
// Read multiple symbols at once to save read() calls.
ElfW(Sym) *buf = reinterpret_cast<ElfW(Sym) *>(tmp_buf);
const size_t buf_entries = tmp_buf_size / sizeof(buf[0]);
const size_t num_symbols = symtab->sh_size / symtab->sh_entsize;
// On platforms using an .opd section (PowerPC & IA64), a function symbol
// has the address of a function descriptor, which contains the real
// starting address. However, we do not always want to use the real
// starting address because we sometimes want to symbolize a function
// pointer into the .opd section, e.g. FindSymbol(&foo,...).
const bool pc_in_opd =
kPlatformUsesOPDSections && opd != nullptr && InSection(pc, opd);
const bool deref_function_descriptor_pointer =
kPlatformUsesOPDSections && opd != nullptr && !pc_in_opd;
ElfW(Sym) best_match;
SafeMemZero(&best_match, sizeof(best_match));
bool found_match = false;
for (size_t i = 0; i < num_symbols;) {
off_t offset =
static_cast<off_t>(symtab->sh_offset + i * symtab->sh_entsize);
const size_t num_remaining_symbols = num_symbols - i;
const size_t entries_in_chunk =
std::min(num_remaining_symbols, buf_entries);
const size_t bytes_in_chunk = entries_in_chunk * sizeof(buf[0]);
const ssize_t len = ReadFromOffset(fd, buf, bytes_in_chunk, offset);
SAFE_ASSERT(len >= 0);
SAFE_ASSERT(static_cast<size_t>(len) % sizeof(buf[0]) == 0);
const size_t num_symbols_in_buf = static_cast<size_t>(len) / sizeof(buf[0]);
SAFE_ASSERT(num_symbols_in_buf <= entries_in_chunk);
for (size_t j = 0; j < num_symbols_in_buf; ++j) {
const ElfW(Sym) &symbol = buf[j];
// For a DSO, a symbol address is relocated by the loading address.
// We keep the original address for opd redirection below.
const char *const original_start_address =
reinterpret_cast<const char *>(symbol.st_value);
const char *start_address =
ComputeOffset(original_start_address, relocation);
#ifdef __arm__
// ARM functions are always aligned to multiples of two bytes; the
// lowest-order bit in start_address is ignored by the CPU and indicates
// whether the function contains ARM (0) or Thumb (1) code. We don't care
// about what encoding is being used; we just want the real start address
// of the function.
start_address = reinterpret_cast<const char *>(
reinterpret_cast<uintptr_t>(start_address) & ~1u);
#endif
if (deref_function_descriptor_pointer &&
InSection(original_start_address, opd)) {
// The opd section is mapped into memory. Just dereference
// start_address to get the first double word, which points to the
// function entry.
start_address = *reinterpret_cast<const char *const *>(start_address);
}
// If pc is inside the .opd section, it points to a function descriptor.
const size_t size = pc_in_opd ? kFunctionDescriptorSize : symbol.st_size;
const void *const end_address =
ComputeOffset(start_address, static_cast<ptrdiff_t>(size));
if (symbol.st_value != 0 && // Skip null value symbols.
symbol.st_shndx != 0 && // Skip undefined symbols.
#ifdef STT_TLS
ELF_ST_TYPE(symbol.st_info) != STT_TLS && // Skip thread-local data.
#endif // STT_TLS
((start_address <= pc && pc < end_address) ||
(start_address == pc && pc == end_address))) {
if (!found_match || ShouldPickFirstSymbol(symbol, best_match)) {
found_match = true;
best_match = symbol;
}
}
}
i += num_symbols_in_buf;
}
if (found_match) {
const off_t off =
static_cast<off_t>(strtab->sh_offset) + best_match.st_name;
const ssize_t n_read = ReadFromOffset(fd, out, out_size, off);
if (n_read <= 0) {
// This should never happen.
ABSL_RAW_LOG(WARNING,
"Unable to read from fd %d at offset %lld: n_read = %zd", fd,
static_cast<long long>(off), n_read);
return SYMBOL_NOT_FOUND;
}
ABSL_RAW_CHECK(static_cast<size_t>(n_read) <= out_size,
"ReadFromOffset read too much data.");
// strtab->sh_offset points into .strtab-like section that contains
// NUL-terminated strings: '\0foo\0barbaz\0...".
//
// sh_offset+st_name points to the start of symbol name, but we don't know
// how long the symbol is, so we try to read as much as we have space for,
// and usually over-read (i.e. there is a NUL somewhere before n_read).
if (memchr(out, '\0', static_cast<size_t>(n_read)) == nullptr) {
// Either out_size was too small (n_read == out_size and no NUL), or
// we tried to read past the EOF (n_read < out_size) and .strtab is
// corrupt (missing terminating NUL; should never happen for valid ELF).
out[n_read - 1] = '\0';
return SYMBOL_TRUNCATED;
}
return SYMBOL_FOUND;
}
return SYMBOL_NOT_FOUND;
}
// Get the symbol name of "pc" from the file pointed by "fd". Process
// both regular and dynamic symbol tables if necessary.
// See FindSymbol() comment for description of return value.
FindSymbolResult Symbolizer::GetSymbolFromObjectFile(
const ObjFile &obj, const void *const pc, const ptrdiff_t relocation,
char *out, size_t out_size, char *tmp_buf, size_t tmp_buf_size) {
ElfW(Shdr) symtab;
ElfW(Shdr) strtab;
ElfW(Shdr) opd;
ElfW(Shdr) *opd_ptr = nullptr;
// On platforms using an .opd sections for function descriptor, read
// the section header. The .opd section is in data segment and should be
// loaded but we check that it is mapped just to be extra careful.
if (kPlatformUsesOPDSections) {
if (GetSectionHeaderByName(obj.fd, kOpdSectionName,
sizeof(kOpdSectionName) - 1, &opd) &&
FindObjFile(reinterpret_cast<const char *>(opd.sh_addr) + relocation,
opd.sh_size) != nullptr) {
opd_ptr = &opd;
} else {
return SYMBOL_NOT_FOUND;
}
}
// Consult a regular symbol table, then fall back to the dynamic symbol table.
for (const auto symbol_table_type : {SHT_SYMTAB, SHT_DYNSYM}) {
if (!GetSectionHeaderByType(obj.fd, obj.elf_header.e_shnum,
static_cast<off_t>(obj.elf_header.e_shoff),
static_cast<ElfW(Word)>(symbol_table_type),
&symtab, tmp_buf, tmp_buf_size)) {
continue;
}
if (!ReadFromOffsetExact(
obj.fd, &strtab, sizeof(strtab),
static_cast<off_t>(obj.elf_header.e_shoff +
symtab.sh_link * sizeof(symtab)))) {
continue;
}
const FindSymbolResult rc =
FindSymbol(pc, obj.fd, out, out_size, relocation, &strtab, &symtab,
opd_ptr, tmp_buf, tmp_buf_size);
if (rc != SYMBOL_NOT_FOUND) {
return rc;
}
}
return SYMBOL_NOT_FOUND;
}
namespace {
// Thin wrapper around a file descriptor so that the file descriptor
// gets closed for sure.
class FileDescriptor {
public:
explicit FileDescriptor(int fd) : fd_(fd) {}
FileDescriptor(const FileDescriptor &) = delete;
FileDescriptor &operator=(const FileDescriptor &) = delete;
~FileDescriptor() {
if (fd_ >= 0) {
close(fd_);
}
}
int get() const { return fd_; }
private:
const int fd_;
};
// Helper class for reading lines from file.
//
// Note: we don't use ProcMapsIterator since the object is big (it has
// a 5k array member) and uses async-unsafe functions such as sscanf()
// and snprintf().
class LineReader {
public:
explicit LineReader(int fd, char *buf, size_t buf_len)
: fd_(fd),
buf_len_(buf_len),
buf_(buf),
bol_(buf),
eol_(buf),
eod_(buf) {}
LineReader(const LineReader &) = delete;
LineReader &operator=(const LineReader &) = delete;
// Read '\n'-terminated line from file. On success, modify "bol"
// and "eol", then return true. Otherwise, return false.
//
// Note: if the last line doesn't end with '\n', the line will be
// dropped. It's an intentional behavior to make the code simple.
bool ReadLine(const char **bol, const char **eol) {
if (BufferIsEmpty()) { // First time.
const ssize_t num_bytes = ReadPersistent(fd_, buf_, buf_len_);
if (num_bytes <= 0) { // EOF or error.
return false;
}
eod_ = buf_ + num_bytes;
bol_ = buf_;
} else {
bol_ = eol_ + 1; // Advance to the next line in the buffer.
SAFE_ASSERT(bol_ <= eod_); // "bol_" can point to "eod_".
if (!HasCompleteLine()) {
const auto incomplete_line_length = static_cast<size_t>(eod_ - bol_);
// Move the trailing incomplete line to the beginning.
memmove(buf_, bol_, incomplete_line_length);
// Read text from file and append it.
char *const append_pos = buf_ + incomplete_line_length;
const size_t capacity_left = buf_len_ - incomplete_line_length;
const ssize_t num_bytes =
ReadPersistent(fd_, append_pos, capacity_left);
if (num_bytes <= 0) { // EOF or error.
return false;
}
eod_ = append_pos + num_bytes;
bol_ = buf_;
}
}
eol_ = FindLineFeed();
if (eol_ == nullptr) { // '\n' not found. Malformed line.
return false;
}
*eol_ = '\0'; // Replace '\n' with '\0'.
*bol = bol_;
*eol = eol_;
return true;
}
private:
char *FindLineFeed() const {
return reinterpret_cast<char *>(
memchr(bol_, '\n', static_cast<size_t>(eod_ - bol_)));
}
bool BufferIsEmpty() const { return buf_ == eod_; }
bool HasCompleteLine() const {
return !BufferIsEmpty() && FindLineFeed() != nullptr;
}
const int fd_;
const size_t buf_len_;
char *const buf_;
char *bol_;
char *eol_;
const char *eod_; // End of data in "buf_".
};
} // namespace
// Place the hex number read from "start" into "*hex". The pointer to
// the first non-hex character or "end" is returned.
static const char *GetHex(const char *start, const char *end,
uint64_t *const value) {
uint64_t hex = 0;
const char *p;
for (p = start; p < end; ++p) {
int ch = *p;
if ((ch >= '0' && ch <= '9') || (ch >= 'A' && ch <= 'F') ||
(ch >= 'a' && ch <= 'f')) {
hex = (hex << 4) |
static_cast<uint64_t>(ch < 'A' ? ch - '0' : (ch & 0xF) + 9);
} else { // Encountered the first non-hex character.
break;
}
}
SAFE_ASSERT(p <= end);
*value = hex;
return p;
}
static const char *GetHex(const char *start, const char *end,
const void **const addr) {
uint64_t hex = 0;
const char *p = GetHex(start, end, &hex);
*addr = reinterpret_cast<void *>(hex);
return p;
}
// Normally we are only interested in "r?x" maps.
// On the PowerPC, function pointers point to descriptors in the .opd
// section. The descriptors themselves are not executable code, so
// we need to relax the check below to "r??".
static bool ShouldUseMapping(const char *const flags) {
return flags[0] == 'r' && (kPlatformUsesOPDSections || flags[2] == 'x');
}
// Read /proc/self/maps and run "callback" for each mmapped file found. If
// "callback" returns false, stop scanning and return true. Else continue
// scanning /proc/self/maps. Return true if no parse error is found.
static ABSL_ATTRIBUTE_NOINLINE bool ReadAddrMap(
bool (*callback)(const char *filename, const void *const start_addr,
const void *const end_addr, uint64_t offset, void *arg),
void *arg, void *tmp_buf, size_t tmp_buf_size) {
// Use /proc/self/task/<pid>/maps instead of /proc/self/maps. The latter