1 // Copyright 2011 Google LLC
2 //
3 // Redistribution and use in source and binary forms, with or without
4 // modification, are permitted provided that the following conditions are
5 // met:
6 //
7 // * Redistributions of source code must retain the above copyright
8 // notice, this list of conditions and the following disclaimer.
9 // * Redistributions in binary form must reproduce the above
10 // copyright notice, this list of conditions and the following disclaimer
11 // in the documentation and/or other materials provided with the
12 // distribution.
13 // * Neither the name of Google LLC nor the names of its
14 // contributors may be used to endorse or promote products derived from
15 // this software without specific prior written permission.
16 //
17 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
18 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
19 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
20 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
21 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
22 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
23 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28
29 // Restructured in 2009 by: Jim Blandy <[email protected]> <[email protected]>
30
31 // dump_symbols.cc: implement google_breakpad::WriteSymbolFile:
32 // Find all the debugging info in a file and dump it as a Breakpad symbol file.
33
34 #ifdef HAVE_CONFIG_H
35 #include <config.h> // Must come first
36 #endif
37
38 #include "common/linux/dump_symbols.h"
39
40 #include <assert.h>
41 #include <elf.h>
42 #include <errno.h>
43 #include <fcntl.h>
44 #include <limits.h>
45 #include <link.h>
46 #include <stdint.h>
47 #include <stdio.h>
48 #include <stdlib.h>
49 #include <string.h>
50 #include <sys/mman.h>
51 #include <sys/stat.h>
52 #include <unistd.h>
53 #include <zlib.h>
54 #ifdef HAVE_LIBZSTD
55 #include <zstd.h>
56 #endif
57
58 #include <set>
59 #include <string>
60 #include <utility>
61 #include <vector>
62
63 #include "common/dwarf/bytereader-inl.h"
64 #include "common/dwarf/dwarf2diehandler.h"
65 #include "common/dwarf_cfi_to_module.h"
66 #include "common/dwarf_cu_to_module.h"
67 #include "common/dwarf_line_to_module.h"
68 #include "common/dwarf_range_list_handler.h"
69 #include "common/linux/crc32.h"
70 #include "common/linux/eintr_wrapper.h"
71 #include "common/linux/elfutils.h"
72 #include "common/linux/elfutils-inl.h"
73 #include "common/linux/elf_symbols_to_module.h"
74 #include "common/linux/file_id.h"
75 #include "common/memory_allocator.h"
76 #include "common/module.h"
77 #include "common/path_helper.h"
78 #include "common/scoped_ptr.h"
79 #ifndef NO_STABS_SUPPORT
80 #include "common/stabs_reader.h"
81 #include "common/stabs_to_module.h"
82 #endif
83 #include "common/using_std_string.h"
84
85 // This namespace contains helper functions.
86 namespace {
87
88 using google_breakpad::DumpOptions;
89 using google_breakpad::DwarfCFIToModule;
90 using google_breakpad::DwarfCUToModule;
91 using google_breakpad::DwarfLineToModule;
92 using google_breakpad::DwarfRangeListHandler;
93 using google_breakpad::ElfClass;
94 using google_breakpad::ElfClass32;
95 using google_breakpad::ElfClass64;
96 using google_breakpad::elf::FileID;
97 using google_breakpad::FindElfSectionByName;
98 using google_breakpad::GetOffset;
99 using google_breakpad::IsValidElf;
100 using google_breakpad::elf::kDefaultBuildIdSize;
101 using google_breakpad::Module;
102 using google_breakpad::PageAllocator;
103 #ifndef NO_STABS_SUPPORT
104 using google_breakpad::StabsToModule;
105 #endif
106 using google_breakpad::scoped_ptr;
107 using google_breakpad::wasteful_vector;
108
109 // Define AARCH64 ELF architecture if host machine does not include this define.
110 #ifndef EM_AARCH64
111 #define EM_AARCH64 183
112 #endif
113
114 // Define ZStd compression if host machine does not include this define.
115 #ifndef ELFCOMPRESS_ZSTD
116 #define ELFCOMPRESS_ZSTD 2
117 #endif
118
119 //
120 // FDWrapper
121 //
122 // Wrapper class to make sure opened file is closed.
123 //
124 class FDWrapper {
125 public:
FDWrapper(int fd)126 explicit FDWrapper(int fd) :
127 fd_(fd) {}
~FDWrapper()128 ~FDWrapper() {
129 if (fd_ != -1)
130 close(fd_);
131 }
get()132 int get() {
133 return fd_;
134 }
release()135 int release() {
136 int fd = fd_;
137 fd_ = -1;
138 return fd;
139 }
140 private:
141 int fd_;
142 };
143
144 //
145 // MmapWrapper
146 //
147 // Wrapper class to make sure mapped regions are unmapped.
148 //
149 class MmapWrapper {
150 public:
MmapWrapper()151 MmapWrapper() : is_set_(false) {}
~MmapWrapper()152 ~MmapWrapper() {
153 if (is_set_ && base_ != NULL) {
154 assert(size_ > 0);
155 munmap(base_, size_);
156 }
157 }
set(void * mapped_address,size_t mapped_size)158 void set(void* mapped_address, size_t mapped_size) {
159 is_set_ = true;
160 base_ = mapped_address;
161 size_ = mapped_size;
162 }
release()163 void release() {
164 assert(is_set_);
165 is_set_ = false;
166 base_ = NULL;
167 size_ = 0;
168 }
169
170 private:
171 bool is_set_;
172 void* base_;
173 size_t size_;
174 };
175
176 // Find the preferred loading address of the binary.
177 template<typename ElfClass>
GetLoadingAddress(const typename ElfClass::Phdr * program_headers,int nheader)178 typename ElfClass::Addr GetLoadingAddress(
179 const typename ElfClass::Phdr* program_headers,
180 int nheader) {
181 typedef typename ElfClass::Phdr Phdr;
182
183 // For non-PIC executables (e_type == ET_EXEC), the load address is
184 // the start address of the first PT_LOAD segment. (ELF requires
185 // the segments to be sorted by load address.) For PIC executables
186 // and dynamic libraries (e_type == ET_DYN), this address will
187 // normally be zero.
188 for (int i = 0; i < nheader; ++i) {
189 const Phdr& header = program_headers[i];
190 if (header.p_type == PT_LOAD)
191 return header.p_vaddr;
192 }
193 return 0;
194 }
195
196 // Find the set of address ranges for all PT_LOAD segments.
197 template <typename ElfClass>
GetPtLoadSegmentRanges(const typename ElfClass::Phdr * program_headers,int nheader)198 vector<Module::Range> GetPtLoadSegmentRanges(
199 const typename ElfClass::Phdr* program_headers,
200 int nheader) {
201 typedef typename ElfClass::Phdr Phdr;
202 vector<Module::Range> ranges;
203
204 for (int i = 0; i < nheader; ++i) {
205 const Phdr& header = program_headers[i];
206 if (header.p_type == PT_LOAD) {
207 ranges.push_back(Module::Range(header.p_vaddr, header.p_memsz));
208 }
209 }
210 return ranges;
211 }
212
213 #ifndef NO_STABS_SUPPORT
214 template<typename ElfClass>
LoadStabs(const typename ElfClass::Ehdr * elf_header,const typename ElfClass::Shdr * stab_section,const typename ElfClass::Shdr * stabstr_section,const bool big_endian,Module * module)215 bool LoadStabs(const typename ElfClass::Ehdr* elf_header,
216 const typename ElfClass::Shdr* stab_section,
217 const typename ElfClass::Shdr* stabstr_section,
218 const bool big_endian,
219 Module* module) {
220 // A callback object to handle data from the STABS reader.
221 StabsToModule handler(module);
222 // Find the addresses of the STABS data, and create a STABS reader object.
223 // On Linux, STABS entries always have 32-bit values, regardless of the
224 // address size of the architecture whose code they're describing, and
225 // the strings are always "unitized".
226 const uint8_t* stabs =
227 GetOffset<ElfClass, uint8_t>(elf_header, stab_section->sh_offset);
228 const uint8_t* stabstr =
229 GetOffset<ElfClass, uint8_t>(elf_header, stabstr_section->sh_offset);
230 google_breakpad::StabsReader reader(stabs, stab_section->sh_size,
231 stabstr, stabstr_section->sh_size,
232 big_endian, 4, true, &handler);
233 // Read the STABS data, and do post-processing.
234 if (!reader.Process())
235 return false;
236 handler.Finalize();
237 return true;
238 }
239 #endif // NO_STABS_SUPPORT
240
241 // A range handler that accepts rangelist data parsed by
242 // google_breakpad::RangeListReader and populates a range vector (typically
243 // owned by a function) with the results.
244 class DumperRangesHandler : public DwarfCUToModule::RangesHandler {
245 public:
DumperRangesHandler(google_breakpad::ByteReader * reader)246 DumperRangesHandler(google_breakpad::ByteReader* reader) :
247 reader_(reader) { }
248
ReadRanges(enum google_breakpad::DwarfForm form,uint64_t data,google_breakpad::RangeListReader::CURangesInfo * cu_info,vector<Module::Range> * ranges)249 bool ReadRanges(
250 enum google_breakpad::DwarfForm form, uint64_t data,
251 google_breakpad::RangeListReader::CURangesInfo* cu_info,
252 vector<Module::Range>* ranges) {
253 DwarfRangeListHandler handler(ranges);
254 google_breakpad::RangeListReader range_list_reader(reader_, cu_info,
255 &handler);
256 return range_list_reader.ReadRanges(form, data);
257 }
258
259 private:
260 google_breakpad::ByteReader* reader_;
261 };
262
263 // A line-to-module loader that accepts line number info parsed by
264 // google_breakpad::LineInfo and populates a Module and a line vector
265 // with the results.
266 class DumperLineToModule: public DwarfCUToModule::LineToModuleHandler {
267 public:
268 // Create a line-to-module converter using BYTE_READER.
DumperLineToModule(google_breakpad::ByteReader * byte_reader)269 explicit DumperLineToModule(google_breakpad::ByteReader* byte_reader)
270 : byte_reader_(byte_reader) { }
StartCompilationUnit(const string & compilation_dir)271 void StartCompilationUnit(const string& compilation_dir) {
272 compilation_dir_ = compilation_dir;
273 }
ReadProgram(const uint8_t * program,uint64_t length,const uint8_t * string_section,uint64_t string_section_length,const uint8_t * line_string_section,uint64_t line_string_section_length,Module * module,std::vector<Module::Line> * lines,std::map<uint32_t,Module::File * > * files)274 void ReadProgram(const uint8_t* program,
275 uint64_t length,
276 const uint8_t* string_section,
277 uint64_t string_section_length,
278 const uint8_t* line_string_section,
279 uint64_t line_string_section_length,
280 Module* module,
281 std::vector<Module::Line>* lines,
282 std::map<uint32_t, Module::File*>* files) {
283 DwarfLineToModule handler(module, compilation_dir_, lines, files);
284 google_breakpad::LineInfo parser(program, length, byte_reader_,
285 string_section, string_section_length,
286 line_string_section,
287 line_string_section_length,
288 &handler);
289 parser.Start();
290 }
291 private:
292 string compilation_dir_;
293 google_breakpad::ByteReader* byte_reader_;
294 };
295
296 template<typename ElfClass>
IsCompressedHeader(const typename ElfClass::Shdr * section)297 bool IsCompressedHeader(const typename ElfClass::Shdr* section) {
298 return (section->sh_flags & SHF_COMPRESSED) != 0;
299 }
300
301 template<typename ElfClass>
GetCompressionHeader(typename ElfClass::Chdr & compression_header,const uint8_t * content,uint64_t size)302 uint32_t GetCompressionHeader(
303 typename ElfClass::Chdr& compression_header,
304 const uint8_t* content, uint64_t size) {
305 const typename ElfClass::Chdr* header =
306 reinterpret_cast<const typename ElfClass::Chdr *>(content);
307
308 if (size < sizeof (*header)) {
309 return 0;
310 }
311
312 compression_header = *header;
313 return sizeof (*header);
314 }
315
UncompressZlibSectionContents(const uint8_t * compressed_buffer,uint64_t compressed_size,uint64_t uncompressed_size)316 std::pair<uint8_t *, uint64_t> UncompressZlibSectionContents(
317 const uint8_t* compressed_buffer, uint64_t compressed_size, uint64_t uncompressed_size) {
318 z_stream stream;
319 memset(&stream, 0, sizeof stream);
320
321 stream.avail_in = compressed_size;
322 stream.avail_out = uncompressed_size;
323 stream.next_in = const_cast<uint8_t *>(compressed_buffer);
324
325 google_breakpad::scoped_array<uint8_t> uncompressed_buffer(
326 new uint8_t[uncompressed_size]);
327
328 int status = inflateInit(&stream);
329 while (stream.avail_in != 0 && status == Z_OK) {
330 stream.next_out =
331 uncompressed_buffer.get() + uncompressed_size - stream.avail_out;
332
333 if ((status = inflate(&stream, Z_FINISH)) != Z_STREAM_END) {
334 break;
335 }
336
337 status = inflateReset(&stream);
338 }
339
340 return inflateEnd(&stream) != Z_OK || status != Z_OK || stream.avail_out != 0
341 ? std::make_pair(nullptr, 0)
342 : std::make_pair(uncompressed_buffer.release(), uncompressed_size);
343 }
344
345 #ifdef HAVE_LIBZSTD
UncompressZstdSectionContents(const uint8_t * compressed_buffer,uint64_t compressed_size,uint64_t uncompressed_size)346 std::pair<uint8_t *, uint64_t> UncompressZstdSectionContents(
347 const uint8_t* compressed_buffer, uint64_t compressed_size,uint64_t uncompressed_size) {
348
349 google_breakpad::scoped_array<uint8_t> uncompressed_buffer(new uint8_t[uncompressed_size]);
350 size_t out_size = ZSTD_decompress(uncompressed_buffer.get(), uncompressed_size,
351 compressed_buffer, compressed_size);
352 if (ZSTD_isError(out_size)) {
353 return std::make_pair(nullptr, 0);
354 }
355 assert(out_size == uncompressed_size);
356 return std::make_pair(uncompressed_buffer.release(), uncompressed_size);
357 }
358 #endif
359
UncompressSectionContents(uint64_t compression_type,const uint8_t * compressed_buffer,uint64_t compressed_size,uint64_t uncompressed_size)360 std::pair<uint8_t *, uint64_t> UncompressSectionContents(
361 uint64_t compression_type, const uint8_t* compressed_buffer,
362 uint64_t compressed_size, uint64_t uncompressed_size) {
363 if (compression_type == ELFCOMPRESS_ZLIB) {
364 return UncompressZlibSectionContents(compressed_buffer, compressed_size, uncompressed_size);
365 }
366
367 #ifdef HAVE_LIBZSTD
368 if (compression_type == ELFCOMPRESS_ZSTD) {
369 return UncompressZstdSectionContents(compressed_buffer, compressed_size, uncompressed_size);
370 }
371 #endif
372
373 return std::make_pair(nullptr, 0);
374 }
375
StartProcessSplitDwarf(google_breakpad::CompilationUnit * reader,Module * module,google_breakpad::Endianness endianness,bool handle_inter_cu_refs,bool handle_inline)376 void StartProcessSplitDwarf(google_breakpad::CompilationUnit* reader,
377 Module* module,
378 google_breakpad::Endianness endianness,
379 bool handle_inter_cu_refs,
380 bool handle_inline) {
381 std::string split_file;
382 google_breakpad::SectionMap split_sections;
383 google_breakpad::ByteReader split_byte_reader(endianness);
384 uint64_t cu_offset = 0;
385 if (!reader->ProcessSplitDwarf(split_file, split_sections, split_byte_reader,
386 cu_offset))
387 return;
388 DwarfCUToModule::FileContext file_context(split_file, module,
389 handle_inter_cu_refs);
390 for (auto section : split_sections)
391 file_context.AddSectionToSectionMap(section.first, section.second.first,
392 section.second.second);
393 // Because DWP/DWO file doesn't have .debug_addr/.debug_line/.debug_line_str,
394 // its debug info will refer to .debug_addr/.debug_line in the main binary.
395 if (file_context.section_map().find(".debug_addr") ==
396 file_context.section_map().end())
397 file_context.AddSectionToSectionMap(".debug_addr", reader->GetAddrBuffer(),
398 reader->GetAddrBufferLen());
399 if (file_context.section_map().find(".debug_line") ==
400 file_context.section_map().end())
401 file_context.AddSectionToSectionMap(".debug_line", reader->GetLineBuffer(),
402 reader->GetLineBufferLen());
403 if (file_context.section_map().find(".debug_line_str") ==
404 file_context.section_map().end())
405 file_context.AddSectionToSectionMap(".debug_line_str",
406 reader->GetLineStrBuffer(),
407 reader->GetLineStrBufferLen());
408
409 DumperRangesHandler ranges_handler(&split_byte_reader);
410 DumperLineToModule line_to_module(&split_byte_reader);
411 DwarfCUToModule::WarningReporter reporter(split_file, cu_offset);
412 DwarfCUToModule root_handler(
413 &file_context, &line_to_module, &ranges_handler, &reporter, handle_inline,
414 reader->GetLowPC(), reader->GetAddrBase(), reader->HasSourceLineInfo(),
415 reader->GetSourceLineOffset());
416 google_breakpad::DIEDispatcher die_dispatcher(&root_handler);
417 google_breakpad::CompilationUnit split_reader(
418 split_file, file_context.section_map(), cu_offset, &split_byte_reader,
419 &die_dispatcher);
420 split_reader.SetSplitDwarf(reader->GetAddrBase(), reader->GetDWOID());
421 split_reader.Start();
422 // Normally, it won't happen unless we have transitive reference.
423 if (split_reader.ShouldProcessSplitDwarf()) {
424 StartProcessSplitDwarf(&split_reader, module, endianness,
425 handle_inter_cu_refs, handle_inline);
426 }
427 }
428
429 template<typename ElfClass>
LoadDwarf(const string & dwarf_filename,const typename ElfClass::Ehdr * elf_header,const bool big_endian,bool handle_inter_cu_refs,bool handle_inline,Module * module)430 bool LoadDwarf(const string& dwarf_filename,
431 const typename ElfClass::Ehdr* elf_header,
432 const bool big_endian,
433 bool handle_inter_cu_refs,
434 bool handle_inline,
435 Module* module) {
436 typedef typename ElfClass::Shdr Shdr;
437
438 const google_breakpad::Endianness endianness = big_endian ?
439 google_breakpad::ENDIANNESS_BIG : google_breakpad::ENDIANNESS_LITTLE;
440 google_breakpad::ByteReader byte_reader(endianness);
441
442 // Construct a context for this file.
443 DwarfCUToModule::FileContext file_context(dwarf_filename,
444 module,
445 handle_inter_cu_refs);
446
447 // Build a map of the ELF file's sections.
448 const Shdr* sections =
449 GetOffset<ElfClass, Shdr>(elf_header, elf_header->e_shoff);
450 int num_sections = elf_header->e_shnum;
451 const Shdr* section_names = sections + elf_header->e_shstrndx;
452 for (int i = 0; i < num_sections; i++) {
453 const Shdr* section = §ions[i];
454 string name = GetOffset<ElfClass, char>(elf_header,
455 section_names->sh_offset) +
456 section->sh_name;
457 const uint8_t* contents = GetOffset<ElfClass, uint8_t>(elf_header,
458 section->sh_offset);
459 uint64_t size = section->sh_size;
460
461 if (!IsCompressedHeader<ElfClass>(section)) {
462 file_context.AddSectionToSectionMap(name, contents, size);
463 continue;
464 }
465
466 typename ElfClass::Chdr chdr;
467
468 uint32_t compression_header_size =
469 GetCompressionHeader<ElfClass>(chdr, contents, size);
470
471 if (compression_header_size == 0 || chdr.ch_size == 0) {
472 continue;
473 }
474
475 contents += compression_header_size;
476 size -= compression_header_size;
477
478 std::pair<uint8_t *, uint64_t> uncompressed =
479 UncompressSectionContents(chdr.ch_type, contents, size, chdr.ch_size);
480
481 if (uncompressed.first != nullptr && uncompressed.second != 0) {
482 file_context.AddManagedSectionToSectionMap(name, uncompressed.first, uncompressed.second);
483 }
484 }
485
486 // .debug_ranges and .debug_rnglists reader
487 DumperRangesHandler ranges_handler(&byte_reader);
488
489 // Parse all the compilation units in the .debug_info section.
490 DumperLineToModule line_to_module(&byte_reader);
491 google_breakpad::SectionMap::const_iterator debug_info_entry =
492 file_context.section_map().find(".debug_info");
493 assert(debug_info_entry != file_context.section_map().end());
494 const std::pair<const uint8_t*, uint64_t>& debug_info_section =
495 debug_info_entry->second;
496 // This should never have been called if the file doesn't have a
497 // .debug_info section.
498 assert(debug_info_section.first);
499 uint64_t debug_info_length = debug_info_section.second;
500 for (uint64_t offset = 0; offset < debug_info_length;) {
501 // Make a handler for the root DIE that populates MODULE with the
502 // data that was found.
503 DwarfCUToModule::WarningReporter reporter(dwarf_filename, offset);
504 DwarfCUToModule root_handler(&file_context, &line_to_module,
505 &ranges_handler, &reporter, handle_inline);
506 // Make a Dwarf2Handler that drives the DIEHandler.
507 google_breakpad::DIEDispatcher die_dispatcher(&root_handler);
508 // Make a DWARF parser for the compilation unit at OFFSET.
509 google_breakpad::CompilationUnit reader(dwarf_filename,
510 file_context.section_map(),
511 offset,
512 &byte_reader,
513 &die_dispatcher);
514 // Process the entire compilation unit; get the offset of the next.
515 offset += reader.Start();
516 // Start to process split dwarf file.
517 if (reader.ShouldProcessSplitDwarf()) {
518 StartProcessSplitDwarf(&reader, module, endianness, handle_inter_cu_refs,
519 handle_inline);
520 }
521 }
522 return true;
523 }
524
525 // Fill REGISTER_NAMES with the register names appropriate to the
526 // machine architecture given in HEADER, indexed by the register
527 // numbers used in DWARF call frame information. Return true on
528 // success, or false if HEADER's machine architecture is not
529 // supported.
530 template<typename ElfClass>
DwarfCFIRegisterNames(const typename ElfClass::Ehdr * elf_header,std::vector<string> * register_names)531 bool DwarfCFIRegisterNames(const typename ElfClass::Ehdr* elf_header,
532 std::vector<string>* register_names) {
533 switch (elf_header->e_machine) {
534 case EM_386:
535 *register_names = DwarfCFIToModule::RegisterNames::I386();
536 return true;
537 case EM_ARM:
538 *register_names = DwarfCFIToModule::RegisterNames::ARM();
539 return true;
540 case EM_AARCH64:
541 *register_names = DwarfCFIToModule::RegisterNames::ARM64();
542 return true;
543 case EM_MIPS:
544 *register_names = DwarfCFIToModule::RegisterNames::MIPS();
545 return true;
546 case EM_X86_64:
547 *register_names = DwarfCFIToModule::RegisterNames::X86_64();
548 return true;
549 case EM_RISCV:
550 *register_names = DwarfCFIToModule::RegisterNames::RISCV();
551 return true;
552 default:
553 return false;
554 }
555 }
556
557 template<typename ElfClass>
LoadDwarfCFI(const string & dwarf_filename,const typename ElfClass::Ehdr * elf_header,const char * section_name,const typename ElfClass::Shdr * section,const bool eh_frame,const typename ElfClass::Shdr * got_section,const typename ElfClass::Shdr * text_section,const bool big_endian,Module * module)558 bool LoadDwarfCFI(const string& dwarf_filename,
559 const typename ElfClass::Ehdr* elf_header,
560 const char* section_name,
561 const typename ElfClass::Shdr* section,
562 const bool eh_frame,
563 const typename ElfClass::Shdr* got_section,
564 const typename ElfClass::Shdr* text_section,
565 const bool big_endian,
566 Module* module) {
567 // Find the appropriate set of register names for this file's
568 // architecture.
569 std::vector<string> register_names;
570 if (!DwarfCFIRegisterNames<ElfClass>(elf_header, ®ister_names)) {
571 fprintf(stderr, "%s: unrecognized ELF machine architecture '%d';"
572 " cannot convert DWARF call frame information\n",
573 dwarf_filename.c_str(), elf_header->e_machine);
574 return false;
575 }
576
577 const google_breakpad::Endianness endianness = big_endian ?
578 google_breakpad::ENDIANNESS_BIG : google_breakpad::ENDIANNESS_LITTLE;
579
580 // Find the call frame information and its size.
581 const uint8_t* cfi =
582 GetOffset<ElfClass, uint8_t>(elf_header, section->sh_offset);
583 size_t cfi_size = section->sh_size;
584
585 // Plug together the parser, handler, and their entourages.
586 DwarfCFIToModule::Reporter module_reporter(dwarf_filename, section_name);
587 DwarfCFIToModule handler(module, register_names, &module_reporter);
588 google_breakpad::ByteReader byte_reader(endianness);
589
590 byte_reader.SetAddressSize(ElfClass::kAddrSize);
591
592 // Provide the base addresses for .eh_frame encoded pointers, if
593 // possible.
594 byte_reader.SetCFIDataBase(section->sh_addr, cfi);
595 if (got_section)
596 byte_reader.SetDataBase(got_section->sh_addr);
597 if (text_section)
598 byte_reader.SetTextBase(text_section->sh_addr);
599
600 google_breakpad::CallFrameInfo::Reporter dwarf_reporter(dwarf_filename,
601 section_name);
602 if (!IsCompressedHeader<ElfClass>(section)) {
603 google_breakpad::CallFrameInfo parser(cfi, cfi_size,
604 &byte_reader, &handler,
605 &dwarf_reporter, eh_frame);
606 parser.Start();
607 return true;
608 }
609
610 typename ElfClass::Chdr chdr;
611 uint32_t compression_header_size =
612 GetCompressionHeader<ElfClass>(chdr, cfi, cfi_size);
613
614 if (compression_header_size == 0 || chdr.ch_size == 0) {
615 fprintf(stderr, "%s: decompression failed at header\n",
616 dwarf_filename.c_str());
617 return false;
618 }
619 if (compression_header_size > cfi_size) {
620 fprintf(stderr, "%s: decompression error, compression_header too large\n",
621 dwarf_filename.c_str());
622 return false;
623 }
624
625 cfi += compression_header_size;
626 cfi_size -= compression_header_size;
627
628 std::pair<uint8_t *, uint64_t> uncompressed =
629 UncompressSectionContents(chdr.ch_type, cfi, cfi_size, chdr.ch_size);
630
631 if (uncompressed.first == nullptr || uncompressed.second == 0) {
632 fprintf(stderr, "%s: decompression failed\n", dwarf_filename.c_str());
633 return false;
634 }
635 google_breakpad::CallFrameInfo parser(uncompressed.first, uncompressed.second,
636 &byte_reader, &handler, &dwarf_reporter,
637 eh_frame);
638 parser.Start();
639 return true;
640 }
641
LoadELF(const string & obj_file,MmapWrapper * map_wrapper,void ** elf_header)642 bool LoadELF(const string& obj_file, MmapWrapper* map_wrapper,
643 void** elf_header) {
644 int obj_fd = open(obj_file.c_str(), O_RDONLY);
645 if (obj_fd < 0) {
646 fprintf(stderr, "Failed to open ELF file '%s': %s\n",
647 obj_file.c_str(), strerror(errno));
648 return false;
649 }
650 FDWrapper obj_fd_wrapper(obj_fd);
651 struct stat st;
652 if (fstat(obj_fd, &st) != 0 && st.st_size <= 0) {
653 fprintf(stderr, "Unable to fstat ELF file '%s': %s\n",
654 obj_file.c_str(), strerror(errno));
655 return false;
656 }
657 void* obj_base = mmap(NULL, st.st_size,
658 PROT_READ | PROT_WRITE, MAP_PRIVATE, obj_fd, 0);
659 if (obj_base == MAP_FAILED) {
660 fprintf(stderr, "Failed to mmap ELF file '%s': %s\n",
661 obj_file.c_str(), strerror(errno));
662 return false;
663 }
664 map_wrapper->set(obj_base, st.st_size);
665 *elf_header = obj_base;
666 if (!IsValidElf(*elf_header)) {
667 fprintf(stderr, "Not a valid ELF file: %s\n", obj_file.c_str());
668 return false;
669 }
670 return true;
671 }
672
673 // Get the endianness of ELF_HEADER. If it's invalid, return false.
674 template<typename ElfClass>
ElfEndianness(const typename ElfClass::Ehdr * elf_header,bool * big_endian)675 bool ElfEndianness(const typename ElfClass::Ehdr* elf_header,
676 bool* big_endian) {
677 if (elf_header->e_ident[EI_DATA] == ELFDATA2LSB) {
678 *big_endian = false;
679 return true;
680 }
681 if (elf_header->e_ident[EI_DATA] == ELFDATA2MSB) {
682 *big_endian = true;
683 return true;
684 }
685
686 fprintf(stderr, "bad data encoding in ELF header: %d\n",
687 elf_header->e_ident[EI_DATA]);
688 return false;
689 }
690
691 // Given |left_abspath|, find the absolute path for |right_path| and see if the
692 // two absolute paths are the same.
IsSameFile(const char * left_abspath,const string & right_path)693 bool IsSameFile(const char* left_abspath, const string& right_path) {
694 char right_abspath[PATH_MAX];
695 if (!realpath(right_path.c_str(), right_abspath))
696 return false;
697 return strcmp(left_abspath, right_abspath) == 0;
698 }
699
700 // Read the .gnu_debuglink and get the debug file name. If anything goes
701 // wrong, return an empty string.
ReadDebugLink(const uint8_t * debuglink,const size_t debuglink_size,const bool big_endian,const string & obj_file,const std::vector<string> & debug_dirs)702 string ReadDebugLink(const uint8_t* debuglink,
703 const size_t debuglink_size,
704 const bool big_endian,
705 const string& obj_file,
706 const std::vector<string>& debug_dirs) {
707 // Include '\0' + CRC32 (4 bytes).
708 size_t debuglink_len = strlen(reinterpret_cast<const char*>(debuglink)) + 5;
709 debuglink_len = 4 * ((debuglink_len + 3) / 4); // Round up to 4 bytes.
710
711 // Sanity check.
712 if (debuglink_len != debuglink_size) {
713 fprintf(stderr, "Mismatched .gnu_debuglink string / section size: "
714 "%zx %zx\n", debuglink_len, debuglink_size);
715 return string();
716 }
717
718 char obj_file_abspath[PATH_MAX];
719 if (!realpath(obj_file.c_str(), obj_file_abspath)) {
720 fprintf(stderr, "Cannot resolve absolute path for %s\n", obj_file.c_str());
721 return string();
722 }
723
724 std::vector<string> searched_paths;
725 string debuglink_path;
726 std::vector<string>::const_iterator it;
727 for (it = debug_dirs.begin(); it < debug_dirs.end(); ++it) {
728 const string& debug_dir = *it;
729 debuglink_path = debug_dir + "/" +
730 reinterpret_cast<const char*>(debuglink);
731
732 // There is the annoying case of /path/to/foo.so having foo.so as the
733 // debug link file name. Thus this may end up opening /path/to/foo.so again,
734 // and there is a small chance of the two files having the same CRC.
735 if (IsSameFile(obj_file_abspath, debuglink_path))
736 continue;
737
738 searched_paths.push_back(debug_dir);
739 int debuglink_fd = open(debuglink_path.c_str(), O_RDONLY);
740 if (debuglink_fd < 0)
741 continue;
742
743 FDWrapper debuglink_fd_wrapper(debuglink_fd);
744
745 // The CRC is the last 4 bytes in |debuglink|.
746 const google_breakpad::Endianness endianness = big_endian ?
747 google_breakpad::ENDIANNESS_BIG : google_breakpad::ENDIANNESS_LITTLE;
748 google_breakpad::ByteReader byte_reader(endianness);
749 uint32_t expected_crc =
750 byte_reader.ReadFourBytes(&debuglink[debuglink_size - 4]);
751
752 uint32_t actual_crc = 0;
753 while (true) {
754 const size_t kReadSize = 4096;
755 char buf[kReadSize];
756 ssize_t bytes_read = HANDLE_EINTR(read(debuglink_fd, &buf, kReadSize));
757 if (bytes_read < 0) {
758 fprintf(stderr, "Error reading debug ELF file %s.\n",
759 debuglink_path.c_str());
760 return string();
761 }
762 if (bytes_read == 0)
763 break;
764 actual_crc = google_breakpad::UpdateCrc32(actual_crc, buf, bytes_read);
765 }
766 if (actual_crc != expected_crc) {
767 fprintf(stderr, "Error reading debug ELF file - CRC32 mismatch: %s\n",
768 debuglink_path.c_str());
769 continue;
770 }
771
772 // Found debug file.
773 return debuglink_path;
774 }
775
776 // Not found case.
777 fprintf(stderr, "Failed to find debug ELF file for '%s' after trying:\n",
778 obj_file.c_str());
779 for (it = searched_paths.begin(); it < searched_paths.end(); ++it) {
780 const string& debug_dir = *it;
781 fprintf(stderr, " %s/%s\n", debug_dir.c_str(), debuglink);
782 }
783 return string();
784 }
785
786 //
787 // LoadSymbolsInfo
788 //
789 // Holds the state between the two calls to LoadSymbols() in case it's necessary
790 // to follow the .gnu_debuglink section and load debug information from a
791 // different file.
792 //
793 template<typename ElfClass>
794 class LoadSymbolsInfo {
795 public:
796 typedef typename ElfClass::Addr Addr;
797
LoadSymbolsInfo(const std::vector<string> & dbg_dirs)798 explicit LoadSymbolsInfo(const std::vector<string>& dbg_dirs) :
799 debug_dirs_(dbg_dirs),
800 has_loading_addr_(false) {}
801
802 // Keeps track of which sections have been loaded so sections don't
803 // accidentally get loaded twice from two different files.
LoadedSection(const string & section)804 void LoadedSection(const string& section) {
805 if (loaded_sections_.count(section) == 0) {
806 loaded_sections_.insert(section);
807 } else {
808 fprintf(stderr, "Section %s has already been loaded.\n",
809 section.c_str());
810 }
811 }
812
813 // The ELF file and linked debug file are expected to have the same preferred
814 // loading address.
set_loading_addr(Addr addr,const string & filename)815 void set_loading_addr(Addr addr, const string& filename) {
816 if (!has_loading_addr_) {
817 loading_addr_ = addr;
818 loaded_file_ = filename;
819 return;
820 }
821
822 if (addr != loading_addr_) {
823 fprintf(stderr,
824 "ELF file '%s' and debug ELF file '%s' "
825 "have different load addresses.\n",
826 loaded_file_.c_str(), filename.c_str());
827 assert(false);
828 }
829 }
830
831 // Setters and getters
debug_dirs() const832 const std::vector<string>& debug_dirs() const {
833 return debug_dirs_;
834 }
835
debuglink_file() const836 string debuglink_file() const {
837 return debuglink_file_;
838 }
set_debuglink_file(string file)839 void set_debuglink_file(string file) {
840 debuglink_file_ = file;
841 }
842
843 private:
844 const std::vector<string>& debug_dirs_; // Directories in which to
845 // search for the debug ELF file.
846
847 string debuglink_file_; // Full path to the debug ELF file.
848
849 bool has_loading_addr_; // Indicate if LOADING_ADDR_ is valid.
850
851 Addr loading_addr_; // Saves the preferred loading address from the
852 // first call to LoadSymbols().
853
854 string loaded_file_; // Name of the file loaded from the first call to
855 // LoadSymbols().
856
857 std::set<string> loaded_sections_; // Tracks the Loaded ELF sections
858 // between calls to LoadSymbols().
859 };
860
861 template<typename ElfClass>
LoadSymbols(const string & obj_file,const bool big_endian,const typename ElfClass::Ehdr * elf_header,const bool read_gnu_debug_link,LoadSymbolsInfo<ElfClass> * info,const DumpOptions & options,Module * module)862 bool LoadSymbols(const string& obj_file,
863 const bool big_endian,
864 const typename ElfClass::Ehdr* elf_header,
865 const bool read_gnu_debug_link,
866 LoadSymbolsInfo<ElfClass>* info,
867 const DumpOptions& options,
868 Module* module) {
869 typedef typename ElfClass::Addr Addr;
870 typedef typename ElfClass::Phdr Phdr;
871 typedef typename ElfClass::Shdr Shdr;
872
873 Addr loading_addr = GetLoadingAddress<ElfClass>(
874 GetOffset<ElfClass, Phdr>(elf_header, elf_header->e_phoff),
875 elf_header->e_phnum);
876 module->SetLoadAddress(loading_addr);
877 info->set_loading_addr(loading_addr, obj_file);
878
879 // Allow filtering of extraneous debug information in partitioned libraries.
880 // Such libraries contain debug information for all libraries extracted from
881 // the same combined library, implying extensive duplication.
882 vector<Module::Range> address_ranges = GetPtLoadSegmentRanges<ElfClass>(
883 GetOffset<ElfClass, Phdr>(elf_header, elf_header->e_phoff),
884 elf_header->e_phnum);
885 module->SetAddressRanges(address_ranges);
886
887 const Shdr* sections =
888 GetOffset<ElfClass, Shdr>(elf_header, elf_header->e_shoff);
889 const Shdr* section_names = sections + elf_header->e_shstrndx;
890 const char* names =
891 GetOffset<ElfClass, char>(elf_header, section_names->sh_offset);
892 const char* names_end = names + section_names->sh_size;
893 bool found_debug_info_section = false;
894 bool found_usable_info = false;
895
896 if ((options.symbol_data & SYMBOLS_AND_FILES) ||
897 (options.symbol_data & INLINES)) {
898 #ifndef NO_STABS_SUPPORT
899 // Look for STABS debugging information, and load it if present.
900 const Shdr* stab_section =
901 FindElfSectionByName<ElfClass>(".stab", SHT_PROGBITS,
902 sections, names, names_end,
903 elf_header->e_shnum);
904 if (stab_section) {
905 const Shdr* stabstr_section = stab_section->sh_link + sections;
906 if (stabstr_section) {
907 found_debug_info_section = true;
908 found_usable_info = true;
909 info->LoadedSection(".stab");
910 if (!LoadStabs<ElfClass>(elf_header, stab_section, stabstr_section,
911 big_endian, module)) {
912 fprintf(stderr, "%s: \".stab\" section found, but failed to load"
913 " STABS debugging information\n", obj_file.c_str());
914 }
915 }
916 }
917 #endif // NO_STABS_SUPPORT
918
919 // See if there are export symbols available.
920 const Shdr* symtab_section =
921 FindElfSectionByName<ElfClass>(".symtab", SHT_SYMTAB,
922 sections, names, names_end,
923 elf_header->e_shnum);
924 const Shdr* strtab_section =
925 FindElfSectionByName<ElfClass>(".strtab", SHT_STRTAB,
926 sections, names, names_end,
927 elf_header->e_shnum);
928 if (symtab_section && strtab_section) {
929 info->LoadedSection(".symtab");
930
931 const uint8_t* symtab =
932 GetOffset<ElfClass, uint8_t>(elf_header,
933 symtab_section->sh_offset);
934 const uint8_t* strtab =
935 GetOffset<ElfClass, uint8_t>(elf_header,
936 strtab_section->sh_offset);
937 bool result =
938 ELFSymbolsToModule(symtab,
939 symtab_section->sh_size,
940 strtab,
941 strtab_section->sh_size,
942 big_endian,
943 ElfClass::kAddrSize,
944 module);
945 found_usable_info = found_usable_info || result;
946 } else {
947 // Look in dynsym only if full symbol table was not available.
948 const Shdr* dynsym_section =
949 FindElfSectionByName<ElfClass>(".dynsym", SHT_DYNSYM,
950 sections, names, names_end,
951 elf_header->e_shnum);
952 const Shdr* dynstr_section =
953 FindElfSectionByName<ElfClass>(".dynstr", SHT_STRTAB,
954 sections, names, names_end,
955 elf_header->e_shnum);
956 if (dynsym_section && dynstr_section) {
957 info->LoadedSection(".dynsym");
958
959 const uint8_t* dynsyms =
960 GetOffset<ElfClass, uint8_t>(elf_header,
961 dynsym_section->sh_offset);
962 const uint8_t* dynstrs =
963 GetOffset<ElfClass, uint8_t>(elf_header,
964 dynstr_section->sh_offset);
965 bool result =
966 ELFSymbolsToModule(dynsyms,
967 dynsym_section->sh_size,
968 dynstrs,
969 dynstr_section->sh_size,
970 big_endian,
971 ElfClass::kAddrSize,
972 module);
973 found_usable_info = found_usable_info || result;
974 }
975 }
976
977 // Only Load .debug_info after loading symbol table to avoid duplicate
978 // PUBLIC records.
979 // Look for DWARF debugging information, and load it if present.
980 const Shdr* dwarf_section =
981 FindElfSectionByName<ElfClass>(".debug_info", SHT_PROGBITS,
982 sections, names, names_end,
983 elf_header->e_shnum);
984
985 // .debug_info section type is SHT_PROGBITS for mips on pnacl toolchains,
986 // but MIPS_DWARF for regular gnu toolchains, so both need to be checked
987 if (elf_header->e_machine == EM_MIPS && !dwarf_section) {
988 dwarf_section =
989 FindElfSectionByName<ElfClass>(".debug_info", SHT_MIPS_DWARF,
990 sections, names, names_end,
991 elf_header->e_shnum);
992 }
993
994 if (dwarf_section) {
995 found_debug_info_section = true;
996 found_usable_info = true;
997 info->LoadedSection(".debug_info");
998 if (!LoadDwarf<ElfClass>(obj_file, elf_header, big_endian,
999 options.handle_inter_cu_refs,
1000 options.symbol_data & INLINES, module)) {
1001 fprintf(stderr, "%s: \".debug_info\" section found, but failed to load "
1002 "DWARF debugging information\n", obj_file.c_str());
1003 }
1004 }
1005 }
1006
1007 if (options.symbol_data & CFI) {
1008 // Dwarf Call Frame Information (CFI) is actually independent from
1009 // the other DWARF debugging information, and can be used alone.
1010 const Shdr* dwarf_cfi_section =
1011 FindElfSectionByName<ElfClass>(".debug_frame", SHT_PROGBITS,
1012 sections, names, names_end,
1013 elf_header->e_shnum);
1014
1015 // .debug_frame section type is SHT_PROGBITS for mips on pnacl toolchains,
1016 // but MIPS_DWARF for regular gnu toolchains, so both need to be checked
1017 if (elf_header->e_machine == EM_MIPS && !dwarf_cfi_section) {
1018 dwarf_cfi_section =
1019 FindElfSectionByName<ElfClass>(".debug_frame", SHT_MIPS_DWARF,
1020 sections, names, names_end,
1021 elf_header->e_shnum);
1022 }
1023
1024 if (dwarf_cfi_section) {
1025 // Ignore the return value of this function; even without call frame
1026 // information, the other debugging information could be perfectly
1027 // useful.
1028 info->LoadedSection(".debug_frame");
1029 bool result =
1030 LoadDwarfCFI<ElfClass>(obj_file, elf_header, ".debug_frame",
1031 dwarf_cfi_section, false, 0, 0, big_endian,
1032 module);
1033 found_usable_info = found_usable_info || result;
1034 }
1035
1036 // Linux C++ exception handling information can also provide
1037 // unwinding data.
1038 const Shdr* eh_frame_section =
1039 FindElfSectionByName<ElfClass>(".eh_frame", SHT_PROGBITS,
1040 sections, names, names_end,
1041 elf_header->e_shnum);
1042 if (eh_frame_section) {
1043 // Pointers in .eh_frame data may be relative to the base addresses of
1044 // certain sections. Provide those sections if present.
1045 const Shdr* got_section =
1046 FindElfSectionByName<ElfClass>(".got", SHT_PROGBITS,
1047 sections, names, names_end,
1048 elf_header->e_shnum);
1049 const Shdr* text_section =
1050 FindElfSectionByName<ElfClass>(".text", SHT_PROGBITS,
1051 sections, names, names_end,
1052 elf_header->e_shnum);
1053 info->LoadedSection(".eh_frame");
1054 // As above, ignore the return value of this function.
1055 bool result =
1056 LoadDwarfCFI<ElfClass>(obj_file, elf_header, ".eh_frame",
1057 eh_frame_section, true,
1058 got_section, text_section, big_endian, module);
1059 found_usable_info = found_usable_info || result;
1060 }
1061 }
1062
1063 if (!found_debug_info_section) {
1064 fprintf(stderr, "%s: file contains no debugging information"
1065 " (no \".stab\" or \".debug_info\" sections)\n",
1066 obj_file.c_str());
1067
1068 // Failed, but maybe there's a .gnu_debuglink section?
1069 if (read_gnu_debug_link) {
1070 const Shdr* gnu_debuglink_section
1071 = FindElfSectionByName<ElfClass>(".gnu_debuglink", SHT_PROGBITS,
1072 sections, names,
1073 names_end, elf_header->e_shnum);
1074 if (gnu_debuglink_section) {
1075 if (!info->debug_dirs().empty()) {
1076 const uint8_t* debuglink_contents =
1077 GetOffset<ElfClass, uint8_t>(elf_header,
1078 gnu_debuglink_section->sh_offset);
1079 string debuglink_file =
1080 ReadDebugLink(debuglink_contents,
1081 gnu_debuglink_section->sh_size,
1082 big_endian,
1083 obj_file,
1084 info->debug_dirs());
1085 info->set_debuglink_file(debuglink_file);
1086 } else {
1087 fprintf(stderr, ".gnu_debuglink section found in '%s', "
1088 "but no debug path specified.\n", obj_file.c_str());
1089 }
1090 } else {
1091 fprintf(stderr, "%s does not contain a .gnu_debuglink section.\n",
1092 obj_file.c_str());
1093 }
1094 } else {
1095 // Return true if some usable information was found, since the caller
1096 // doesn't want to use .gnu_debuglink.
1097 return found_usable_info;
1098 }
1099
1100 // No debug info was found, let the user try again with .gnu_debuglink
1101 // if present.
1102 return false;
1103 }
1104
1105 return true;
1106 }
1107
1108 // Return the breakpad symbol file identifier for the architecture of
1109 // ELF_HEADER.
1110 template<typename ElfClass>
ElfArchitecture(const typename ElfClass::Ehdr * elf_header)1111 const char* ElfArchitecture(const typename ElfClass::Ehdr* elf_header) {
1112 typedef typename ElfClass::Half Half;
1113 Half arch = elf_header->e_machine;
1114 switch (arch) {
1115 case EM_386: return "x86";
1116 case EM_ARM: return "arm";
1117 case EM_AARCH64: return "arm64";
1118 case EM_MIPS: return "mips";
1119 case EM_PPC64: return "ppc64";
1120 case EM_PPC: return "ppc";
1121 case EM_S390: return "s390";
1122 case EM_SPARC: return "sparc";
1123 case EM_SPARCV9: return "sparcv9";
1124 case EM_X86_64: return "x86_64";
1125 case EM_RISCV: return "riscv";
1126 default: return NULL;
1127 }
1128 }
1129
1130 template<typename ElfClass>
SanitizeDebugFile(const typename ElfClass::Ehdr * debug_elf_header,const string & debuglink_file,const string & obj_filename,const char * obj_file_architecture,const bool obj_file_is_big_endian)1131 bool SanitizeDebugFile(const typename ElfClass::Ehdr* debug_elf_header,
1132 const string& debuglink_file,
1133 const string& obj_filename,
1134 const char* obj_file_architecture,
1135 const bool obj_file_is_big_endian) {
1136 const char* debug_architecture =
1137 ElfArchitecture<ElfClass>(debug_elf_header);
1138 if (!debug_architecture) {
1139 fprintf(stderr, "%s: unrecognized ELF machine architecture: %d\n",
1140 debuglink_file.c_str(), debug_elf_header->e_machine);
1141 return false;
1142 }
1143 if (strcmp(obj_file_architecture, debug_architecture)) {
1144 fprintf(stderr, "%s with ELF machine architecture %s does not match "
1145 "%s with ELF architecture %s\n",
1146 debuglink_file.c_str(), debug_architecture,
1147 obj_filename.c_str(), obj_file_architecture);
1148 return false;
1149 }
1150 bool debug_big_endian;
1151 if (!ElfEndianness<ElfClass>(debug_elf_header, &debug_big_endian))
1152 return false;
1153 if (debug_big_endian != obj_file_is_big_endian) {
1154 fprintf(stderr, "%s and %s does not match in endianness\n",
1155 obj_filename.c_str(), debuglink_file.c_str());
1156 return false;
1157 }
1158 return true;
1159 }
1160
1161 template<typename ElfClass>
InitModuleForElfClass(const typename ElfClass::Ehdr * elf_header,const string & obj_filename,const string & obj_os,scoped_ptr<Module> & module,bool enable_multiple_field)1162 bool InitModuleForElfClass(const typename ElfClass::Ehdr* elf_header,
1163 const string& obj_filename,
1164 const string& obj_os,
1165 scoped_ptr<Module>& module,
1166 bool enable_multiple_field) {
1167 PageAllocator allocator;
1168 wasteful_vector<uint8_t> identifier(&allocator, kDefaultBuildIdSize);
1169 if (!FileID::ElfFileIdentifierFromMappedFile(elf_header, identifier)) {
1170 fprintf(stderr, "%s: unable to generate file identifier\n",
1171 obj_filename.c_str());
1172 return false;
1173 }
1174
1175 const char* architecture = ElfArchitecture<ElfClass>(elf_header);
1176 if (!architecture) {
1177 fprintf(stderr, "%s: unrecognized ELF machine architecture: %d\n",
1178 obj_filename.c_str(), elf_header->e_machine);
1179 return false;
1180 }
1181
1182 char name_buf[NAME_MAX] = {};
1183 std::string name = google_breakpad::ElfFileSoNameFromMappedFile(
1184 elf_header, name_buf, sizeof(name_buf))
1185 ? name_buf
1186 : google_breakpad::BaseName(obj_filename);
1187
1188 // Add an extra "0" at the end. PDB files on Windows have an 'age'
1189 // number appended to the end of the file identifier; this isn't
1190 // really used or necessary on other platforms, but be consistent.
1191 string id = FileID::ConvertIdentifierToUUIDString(identifier) + "0";
1192 // This is just the raw Build ID in hex.
1193 string code_id = FileID::ConvertIdentifierToString(identifier);
1194
1195 module.reset(new Module(name, obj_os, architecture, id, code_id,
1196 enable_multiple_field));
1197
1198 return true;
1199 }
1200
1201 template<typename ElfClass>
ReadSymbolDataElfClass(const typename ElfClass::Ehdr * elf_header,const string & obj_filename,const string & obj_os,const std::vector<string> & debug_dirs,const DumpOptions & options,Module ** out_module)1202 bool ReadSymbolDataElfClass(const typename ElfClass::Ehdr* elf_header,
1203 const string& obj_filename,
1204 const string& obj_os,
1205 const std::vector<string>& debug_dirs,
1206 const DumpOptions& options,
1207 Module** out_module) {
1208 typedef typename ElfClass::Ehdr Ehdr;
1209
1210 *out_module = NULL;
1211
1212 scoped_ptr<Module> module;
1213 if (!InitModuleForElfClass<ElfClass>(elf_header, obj_filename, obj_os, module,
1214 options.enable_multiple_field)) {
1215 return false;
1216 }
1217
1218 // Figure out what endianness this file is.
1219 bool big_endian;
1220 if (!ElfEndianness<ElfClass>(elf_header, &big_endian))
1221 return false;
1222
1223 LoadSymbolsInfo<ElfClass> info(debug_dirs);
1224 if (!LoadSymbols<ElfClass>(obj_filename, big_endian, elf_header,
1225 !debug_dirs.empty(), &info,
1226 options, module.get())) {
1227 const string debuglink_file = info.debuglink_file();
1228 if (debuglink_file.empty())
1229 return false;
1230
1231 // Load debuglink ELF file.
1232 fprintf(stderr, "Found debugging info in %s\n", debuglink_file.c_str());
1233 MmapWrapper debug_map_wrapper;
1234 Ehdr* debug_elf_header = NULL;
1235 if (!LoadELF(debuglink_file, &debug_map_wrapper,
1236 reinterpret_cast<void**>(&debug_elf_header)) ||
1237 !SanitizeDebugFile<ElfClass>(debug_elf_header, debuglink_file,
1238 obj_filename,
1239 module->architecture().c_str(),
1240 big_endian)) {
1241 return false;
1242 }
1243
1244 if (!LoadSymbols<ElfClass>(debuglink_file, big_endian,
1245 debug_elf_header, false, &info,
1246 options, module.get())) {
1247 return false;
1248 }
1249 }
1250
1251 *out_module = module.release();
1252 return true;
1253 }
1254
1255 } // namespace
1256
1257 namespace google_breakpad {
1258
1259 // Not explicitly exported, but not static so it can be used in unit tests.
ReadSymbolDataInternal(const uint8_t * obj_file,const string & obj_filename,const string & obj_os,const std::vector<string> & debug_dirs,const DumpOptions & options,Module ** module)1260 bool ReadSymbolDataInternal(const uint8_t* obj_file,
1261 const string& obj_filename,
1262 const string& obj_os,
1263 const std::vector<string>& debug_dirs,
1264 const DumpOptions& options,
1265 Module** module) {
1266 if (!IsValidElf(obj_file)) {
1267 fprintf(stderr, "Not a valid ELF file: %s\n", obj_filename.c_str());
1268 return false;
1269 }
1270
1271 int elfclass = ElfClass(obj_file);
1272 if (elfclass == ELFCLASS32) {
1273 return ReadSymbolDataElfClass<ElfClass32>(
1274 reinterpret_cast<const Elf32_Ehdr*>(obj_file), obj_filename, obj_os,
1275 debug_dirs, options, module);
1276 }
1277 if (elfclass == ELFCLASS64) {
1278 return ReadSymbolDataElfClass<ElfClass64>(
1279 reinterpret_cast<const Elf64_Ehdr*>(obj_file), obj_filename, obj_os,
1280 debug_dirs, options, module);
1281 }
1282
1283 return false;
1284 }
1285
WriteSymbolFile(const string & load_path,const string & obj_file,const string & obj_os,const std::vector<string> & debug_dirs,const DumpOptions & options,std::ostream & sym_stream)1286 bool WriteSymbolFile(const string& load_path,
1287 const string& obj_file,
1288 const string& obj_os,
1289 const std::vector<string>& debug_dirs,
1290 const DumpOptions& options,
1291 std::ostream& sym_stream) {
1292 Module* module;
1293 if (!ReadSymbolData(load_path, obj_file, obj_os, debug_dirs, options,
1294 &module))
1295 return false;
1296
1297 bool result = module->Write(sym_stream, options.symbol_data);
1298 delete module;
1299 return result;
1300 }
1301
1302 // Read the selected object file's debugging information, and write out the
1303 // header only to |stream|. Return true on success; if an error occurs, report
1304 // it and return false.
WriteSymbolFileHeader(const string & load_path,const string & obj_file,const string & obj_os,std::ostream & sym_stream)1305 bool WriteSymbolFileHeader(const string& load_path,
1306 const string& obj_file,
1307 const string& obj_os,
1308 std::ostream& sym_stream) {
1309 MmapWrapper map_wrapper;
1310 void* elf_header = NULL;
1311 if (!LoadELF(load_path, &map_wrapper, &elf_header)) {
1312 fprintf(stderr, "Could not load ELF file: %s\n", obj_file.c_str());
1313 return false;
1314 }
1315
1316 if (!IsValidElf(elf_header)) {
1317 fprintf(stderr, "Not a valid ELF file: %s\n", obj_file.c_str());
1318 return false;
1319 }
1320
1321 int elfclass = ElfClass(elf_header);
1322 scoped_ptr<Module> module;
1323 if (elfclass == ELFCLASS32) {
1324 if (!InitModuleForElfClass<ElfClass32>(
1325 reinterpret_cast<const Elf32_Ehdr*>(elf_header), obj_file, obj_os,
1326 module, /*enable_multiple_field=*/false)) {
1327 fprintf(stderr, "Failed to load ELF module: %s\n", obj_file.c_str());
1328 return false;
1329 }
1330 } else if (elfclass == ELFCLASS64) {
1331 if (!InitModuleForElfClass<ElfClass64>(
1332 reinterpret_cast<const Elf64_Ehdr*>(elf_header), obj_file, obj_os,
1333 module, /*enable_multiple_field=*/false)) {
1334 fprintf(stderr, "Failed to load ELF module: %s\n", obj_file.c_str());
1335 return false;
1336 }
1337 } else {
1338 fprintf(stderr, "Unsupported module file: %s\n", obj_file.c_str());
1339 return false;
1340 }
1341
1342 return module->Write(sym_stream, ALL_SYMBOL_DATA);
1343 }
1344
ReadSymbolData(const string & load_path,const string & obj_file,const string & obj_os,const std::vector<string> & debug_dirs,const DumpOptions & options,Module ** module)1345 bool ReadSymbolData(const string& load_path,
1346 const string& obj_file,
1347 const string& obj_os,
1348 const std::vector<string>& debug_dirs,
1349 const DumpOptions& options,
1350 Module** module) {
1351 MmapWrapper map_wrapper;
1352 void* elf_header = NULL;
1353 if (!LoadELF(load_path, &map_wrapper, &elf_header))
1354 return false;
1355
1356 return ReadSymbolDataInternal(reinterpret_cast<uint8_t*>(elf_header),
1357 obj_file, obj_os, debug_dirs, options, module);
1358 }
1359
1360 } // namespace google_breakpad
1361