michael@0: // Copyright (c) 2011 Google Inc.
michael@0: // All rights reserved.
michael@0: //
michael@0: // Redistribution and use in source and binary forms, with or without
michael@0: // modification, are permitted provided that the following conditions are
michael@0: // met:
michael@0: //
michael@0: //     * Redistributions of source code must retain the above copyright
michael@0: // notice, this list of conditions and the following disclaimer.
michael@0: //     * Redistributions in binary form must reproduce the above
michael@0: // copyright notice, this list of conditions and the following disclaimer
michael@0: // in the documentation and/or other materials provided with the
michael@0: // distribution.
michael@0: //     * Neither the name of Google Inc. nor the names of its
michael@0: // contributors may be used to endorse or promote products derived from
michael@0: // this software without specific prior written permission.
michael@0: //
michael@0: // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
michael@0: // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
michael@0: // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
michael@0: // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
michael@0: // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
michael@0: // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
michael@0: // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
michael@0: // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
michael@0: // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
michael@0: // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
michael@0: // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
michael@0: 
michael@0: // Restructured in 2009 by: Jim Blandy <jimb@mozilla.com> <jimb@red-bean.com>
michael@0: 
michael@0: // dump_symbols.cc: implement google_breakpad::WriteSymbolFile:
michael@0: // Find all the debugging info in a file and dump it as a Breakpad symbol file.
michael@0: 
michael@0: #include "common/linux/dump_symbols.h"
michael@0: 
michael@0: #include <assert.h>
michael@0: #include <elf.h>
michael@0: #include <errno.h>
michael@0: #include <fcntl.h>
michael@0: #include <link.h>
michael@0: #include <stdio.h>
michael@0: #include <stdlib.h>
michael@0: #include <string.h>
michael@0: #include <sys/mman.h>
michael@0: #include <sys/stat.h>
michael@0: #include <unistd.h>
michael@0: 
michael@0: #include <iostream>
michael@0: #include <set>
michael@0: #include <string>
michael@0: #include <utility>
michael@0: #include <vector>
michael@0: 
michael@0: #include "common/arm_ex_reader.h"
michael@0: #include "common/dwarf/bytereader-inl.h"
michael@0: #include "common/dwarf/dwarf2diehandler.h"
michael@0: #include "common/dwarf_cfi_to_module.h"
michael@0: #include "common/dwarf_cu_to_module.h"
michael@0: #include "common/dwarf_line_to_module.h"
michael@0: #include "common/linux/elfutils.h"
michael@0: #include "common/linux/elfutils-inl.h"
michael@0: #include "common/linux/elf_symbols_to_module.h"
michael@0: #include "common/linux/file_id.h"
michael@0: #include "common/module.h"
michael@0: #include "common/scoped_ptr.h"
michael@0: #ifndef NO_STABS_SUPPORT
michael@0: #include "common/stabs_reader.h"
michael@0: #include "common/stabs_to_module.h"
michael@0: #endif
michael@0: #include "common/using_std_string.h"
michael@0: #include "common/logging.h"
michael@0: 
michael@0: #ifndef SHT_ARM_EXIDX
michael@0: // bionic and older glibc don't define it
michael@0: # define SHT_ARM_EXIDX (SHT_LOPROC + 1)
michael@0: #endif
michael@0: 
michael@0: // This namespace contains helper functions.
michael@0: namespace {
michael@0: 
michael@0: using google_breakpad::DwarfCFIToModule;
michael@0: using google_breakpad::DwarfCUToModule;
michael@0: using google_breakpad::DwarfLineToModule;
michael@0: using google_breakpad::ElfClass;
michael@0: using google_breakpad::ElfClass32;
michael@0: using google_breakpad::ElfClass64;
michael@0: using google_breakpad::FindElfSectionByName;
michael@0: using google_breakpad::GetOffset;
michael@0: using google_breakpad::IsValidElf;
michael@0: using google_breakpad::Module;
michael@0: #ifndef NO_STABS_SUPPORT
michael@0: using google_breakpad::StabsToModule;
michael@0: #endif
michael@0: using google_breakpad::UniqueString;
michael@0: using google_breakpad::scoped_ptr;
michael@0: 
michael@0: //
michael@0: // FDWrapper
michael@0: //
michael@0: // Wrapper class to make sure opened file is closed.
michael@0: //
michael@0: class FDWrapper {
michael@0:  public:
michael@0:   explicit FDWrapper(int fd) :
michael@0:     fd_(fd) {}
michael@0:   ~FDWrapper() {
michael@0:     if (fd_ != -1)
michael@0:       close(fd_);
michael@0:   }
michael@0:   int get() {
michael@0:     return fd_;
michael@0:   }
michael@0:   int release() {
michael@0:     int fd = fd_;
michael@0:     fd_ = -1;
michael@0:     return fd;
michael@0:   }
michael@0:  private:
michael@0:   int fd_;
michael@0: };
michael@0: 
michael@0: //
michael@0: // MmapWrapper
michael@0: //
michael@0: // Wrapper class to make sure mapped regions are unmapped.
michael@0: //
michael@0: class MmapWrapper {
michael@0:  public:
michael@0:   MmapWrapper() : is_set_(false) {}
michael@0:   ~MmapWrapper() {
michael@0:     if (is_set_ && base_ != NULL) {
michael@0:       assert(size_ > 0);
michael@0:       munmap(base_, size_);
michael@0:     }
michael@0:   }
michael@0:   void set(void *mapped_address, size_t mapped_size) {
michael@0:     is_set_ = true;
michael@0:     base_ = mapped_address;
michael@0:     size_ = mapped_size;
michael@0:   }
michael@0:   void release() {
michael@0:     assert(is_set_);
michael@0:     is_set_ = false;
michael@0:     base_ = NULL;
michael@0:     size_ = 0;
michael@0:   }
michael@0: 
michael@0:  private:
michael@0:   bool is_set_;
michael@0:   void *base_;
michael@0:   size_t size_;
michael@0: };
michael@0: 
michael@0: // Find the preferred loading address of the binary.
michael@0: template<typename ElfClass>
michael@0: typename ElfClass::Addr GetLoadingAddress(
michael@0:     const typename ElfClass::Phdr* program_headers,
michael@0:     int nheader) {
michael@0:   typedef typename ElfClass::Phdr Phdr;
michael@0: 
michael@0:   for (int i = 0; i < nheader; ++i) {
michael@0:     const Phdr& header = program_headers[i];
michael@0:     // For executable, it is the PT_LOAD segment with offset to zero.
michael@0:     if (header.p_type == PT_LOAD &&
michael@0:         header.p_offset == 0)
michael@0:       return header.p_vaddr;
michael@0:   }
michael@0:   // For other types of ELF, return 0.
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: #ifndef NO_STABS_SUPPORT
michael@0: template<typename ElfClass>
michael@0: bool LoadStabs(const typename ElfClass::Ehdr* elf_header,
michael@0:                const typename ElfClass::Shdr* stab_section,
michael@0:                const typename ElfClass::Shdr* stabstr_section,
michael@0:                const bool big_endian,
michael@0:                Module* module) {
michael@0:   // A callback object to handle data from the STABS reader.
michael@0:   StabsToModule handler(module);
michael@0:   // Find the addresses of the STABS data, and create a STABS reader object.
michael@0:   // On Linux, STABS entries always have 32-bit values, regardless of the
michael@0:   // address size of the architecture whose code they're describing, and
michael@0:   // the strings are always "unitized".
michael@0:   const uint8_t* stabs =
michael@0:       GetOffset<ElfClass, uint8_t>(elf_header, stab_section->sh_offset);
michael@0:   const uint8_t* stabstr =
michael@0:       GetOffset<ElfClass, uint8_t>(elf_header, stabstr_section->sh_offset);
michael@0:   google_breakpad::StabsReader reader(stabs, stab_section->sh_size,
michael@0:                                       stabstr, stabstr_section->sh_size,
michael@0:                                       big_endian, 4, true, &handler);
michael@0:   // Read the STABS data, and do post-processing.
michael@0:   if (!reader.Process())
michael@0:     return false;
michael@0:   handler.Finalize();
michael@0:   return true;
michael@0: }
michael@0: #endif  // NO_STABS_SUPPORT
michael@0: 
michael@0: // A line-to-module loader that accepts line number info parsed by
michael@0: // dwarf2reader::LineInfo and populates a Module and a line vector
michael@0: // with the results.
michael@0: class DumperLineToModule: public DwarfCUToModule::LineToModuleHandler {
michael@0:  public:
michael@0:   // Create a line-to-module converter using BYTE_READER.
michael@0:   explicit DumperLineToModule(dwarf2reader::ByteReader *byte_reader)
michael@0:       : byte_reader_(byte_reader) { }
michael@0:   void StartCompilationUnit(const string& compilation_dir) {
michael@0:     compilation_dir_ = compilation_dir;
michael@0:   }
michael@0:   void ReadProgram(const char *program, uint64 length,
michael@0:                    Module *module, std::vector<Module::Line> *lines) {
michael@0:     DwarfLineToModule handler(module, compilation_dir_, lines);
michael@0:     dwarf2reader::LineInfo parser(program, length, byte_reader_, &handler);
michael@0:     parser.Start();
michael@0:   }
michael@0:  private:
michael@0:   string compilation_dir_;
michael@0:   dwarf2reader::ByteReader *byte_reader_;
michael@0: };
michael@0: 
michael@0: template<typename ElfClass>
michael@0: bool LoadDwarf(const string& dwarf_filename,
michael@0:                const typename ElfClass::Ehdr* elf_header,
michael@0:                const bool big_endian,
michael@0:                Module* module) {
michael@0:   typedef typename ElfClass::Shdr Shdr;
michael@0: 
michael@0:   const dwarf2reader::Endianness endianness = big_endian ?
michael@0:       dwarf2reader::ENDIANNESS_BIG : dwarf2reader::ENDIANNESS_LITTLE;
michael@0:   dwarf2reader::ByteReader byte_reader(endianness);
michael@0: 
michael@0:   // Construct a context for this file.
michael@0:   DwarfCUToModule::FileContext file_context(dwarf_filename, module);
michael@0: 
michael@0:   // Build a map of the ELF file's sections.
michael@0:   const Shdr* sections =
michael@0:       GetOffset<ElfClass, Shdr>(elf_header, elf_header->e_shoff);
michael@0:   int num_sections = elf_header->e_shnum;
michael@0:   const Shdr* section_names = sections + elf_header->e_shstrndx;
michael@0:   for (int i = 0; i < num_sections; i++) {
michael@0:     const Shdr* section = &sections[i];
michael@0:     string name = GetOffset<ElfClass, char>(elf_header,
michael@0:                                             section_names->sh_offset) +
michael@0:                   section->sh_name;
michael@0:     const char* contents = GetOffset<ElfClass, char>(elf_header,
michael@0:                                                      section->sh_offset);
michael@0:     uint64 length = section->sh_size;
michael@0:     file_context.section_map[name] = std::make_pair(contents, length);
michael@0:   }
michael@0: 
michael@0:   // Parse all the compilation units in the .debug_info section.
michael@0:   DumperLineToModule line_to_module(&byte_reader);
michael@0:   std::pair<const char *, uint64> debug_info_section
michael@0:       = file_context.section_map[".debug_info"];
michael@0:   // This should never have been called if the file doesn't have a
michael@0:   // .debug_info section.
michael@0:   assert(debug_info_section.first);
michael@0:   uint64 debug_info_length = debug_info_section.second;
michael@0:   for (uint64 offset = 0; offset < debug_info_length;) {
michael@0:     // Make a handler for the root DIE that populates MODULE with the
michael@0:     // data that was found.
michael@0:     DwarfCUToModule::WarningReporter reporter(dwarf_filename, offset);
michael@0:     DwarfCUToModule root_handler(&file_context, &line_to_module, &reporter);
michael@0:     // Make a Dwarf2Handler that drives the DIEHandler.
michael@0:     dwarf2reader::DIEDispatcher die_dispatcher(&root_handler);
michael@0:     // Make a DWARF parser for the compilation unit at OFFSET.
michael@0:     dwarf2reader::CompilationUnit reader(file_context.section_map,
michael@0:                                          offset,
michael@0:                                          &byte_reader,
michael@0:                                          &die_dispatcher);
michael@0:     // Process the entire compilation unit; get the offset of the next.
michael@0:     offset += reader.Start();
michael@0:   }
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: // Fill REGISTER_NAMES with the register names appropriate to the
michael@0: // machine architecture given in HEADER, indexed by the register
michael@0: // numbers used in DWARF call frame information. Return true on
michael@0: // success, or false if HEADER's machine architecture is not
michael@0: // supported.
michael@0: template<typename ElfClass>
michael@0: bool DwarfCFIRegisterNames(const typename ElfClass::Ehdr* elf_header,
michael@0:                            std::vector<const UniqueString*>* register_names) {
michael@0:   switch (elf_header->e_machine) {
michael@0:     case EM_386:
michael@0:       *register_names = DwarfCFIToModule::RegisterNames::I386();
michael@0:       return true;
michael@0:     case EM_ARM:
michael@0:       *register_names = DwarfCFIToModule::RegisterNames::ARM();
michael@0:       return true;
michael@0:     case EM_X86_64:
michael@0:       *register_names = DwarfCFIToModule::RegisterNames::X86_64();
michael@0:       return true;
michael@0:     default:
michael@0:       return false;
michael@0:   }
michael@0: }
michael@0: 
michael@0: template<typename ElfClass>
michael@0: bool LoadDwarfCFI(const string& dwarf_filename,
michael@0:                   const typename ElfClass::Ehdr* elf_header,
michael@0:                   const char* section_name,
michael@0:                   const typename ElfClass::Shdr* section,
michael@0:                   const bool eh_frame,
michael@0:                   const typename ElfClass::Shdr* got_section,
michael@0:                   const typename ElfClass::Shdr* text_section,
michael@0:                   const bool big_endian,
michael@0:                   Module* module) {
michael@0:   // Find the appropriate set of register names for this file's
michael@0:   // architecture.
michael@0:   std::vector<const UniqueString*> register_names;
michael@0:   if (!DwarfCFIRegisterNames<ElfClass>(elf_header, &register_names)) {
michael@0:     fprintf(stderr, "%s: unrecognized ELF machine architecture '%d';"
michael@0:             " cannot convert DWARF call frame information\n",
michael@0:             dwarf_filename.c_str(), elf_header->e_machine);
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   const dwarf2reader::Endianness endianness = big_endian ?
michael@0:       dwarf2reader::ENDIANNESS_BIG : dwarf2reader::ENDIANNESS_LITTLE;
michael@0: 
michael@0:   // Find the call frame information and its size.
michael@0:   const char* cfi =
michael@0:       GetOffset<ElfClass, char>(elf_header, section->sh_offset);
michael@0:   size_t cfi_size = section->sh_size;
michael@0: 
michael@0:   // Plug together the parser, handler, and their entourages.
michael@0:   DwarfCFIToModule::Reporter module_reporter(dwarf_filename, section_name);
michael@0:   DwarfCFIToModule handler(module, register_names, &module_reporter);
michael@0:   dwarf2reader::ByteReader byte_reader(endianness);
michael@0: 
michael@0:   byte_reader.SetAddressSize(ElfClass::kAddrSize);
michael@0: 
michael@0:   // Provide the base addresses for .eh_frame encoded pointers, if
michael@0:   // possible.
michael@0:   byte_reader.SetCFIDataBase(section->sh_addr, cfi);
michael@0:   if (got_section)
michael@0:     byte_reader.SetDataBase(got_section->sh_addr);
michael@0:   if (text_section)
michael@0:     byte_reader.SetTextBase(text_section->sh_addr);
michael@0: 
michael@0:   dwarf2reader::CallFrameInfo::Reporter dwarf_reporter(dwarf_filename,
michael@0:                                                        section_name);
michael@0:   dwarf2reader::CallFrameInfo parser(cfi, cfi_size,
michael@0:                                      &byte_reader, &handler, &dwarf_reporter,
michael@0:                                      eh_frame);
michael@0:   parser.Start();
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: template<typename ElfClass>
michael@0: bool LoadARMexidx(const typename ElfClass::Ehdr* elf_header,
michael@0:                   const typename ElfClass::Shdr* exidx_section,
michael@0:                   const typename ElfClass::Shdr* extab_section,
michael@0:                   uint32_t loading_addr,
michael@0:                   Module* module) {
michael@0:   // To do this properly we need to know:
michael@0:   // * the bounds of the .ARM.exidx section in the mapped image
michael@0:   // * the bounds of the .ARM.extab section in the mapped image
michael@0:   // * the vma of the last byte in the text section associated with the .exidx
michael@0:   // The first two are easy.  The third is a bit tricky.  If we can't
michael@0:   // figure out what it is, just pass in zero.
michael@0:   const char *exidx_img
michael@0:     = GetOffset<ElfClass, char>(elf_header, exidx_section->sh_offset);
michael@0:   size_t exidx_size = exidx_section->sh_size;
michael@0:   const char *extab_img
michael@0:     = GetOffset<ElfClass, char>(elf_header, extab_section->sh_offset);
michael@0:   size_t extab_size = extab_section->sh_size;
michael@0: 
michael@0:   // The sh_link field of the exidx section gives the section number
michael@0:   // for the associated text section.
michael@0:   uint32_t exidx_text_last_svma = 0;
michael@0:   int exidx_text_sno = exidx_section->sh_link;
michael@0:   typedef typename ElfClass::Shdr Shdr;
michael@0:   // |sections| points to the section header table
michael@0:   const Shdr* sections
michael@0:     = GetOffset<ElfClass, Shdr>(elf_header, elf_header->e_shoff);
michael@0:   const int num_sections = elf_header->e_shnum;
michael@0:   if (exidx_text_sno >= 0 && exidx_text_sno < num_sections) {
michael@0:     const Shdr* exidx_text_shdr = &sections[exidx_text_sno];
michael@0:     if (exidx_text_shdr->sh_size > 0) {
michael@0:       exidx_text_last_svma
michael@0:         = exidx_text_shdr->sh_addr + exidx_text_shdr->sh_size - 1;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   arm_ex_to_module::ARMExToModule handler(module);
michael@0:   arm_ex_reader::ExceptionTableInfo
michael@0:     parser(exidx_img, exidx_size, extab_img, extab_size, exidx_text_last_svma,
michael@0:            &handler,
michael@0:            reinterpret_cast<const char*>(elf_header),
michael@0:            loading_addr);
michael@0:   parser.Start();
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: bool LoadELF(const string& obj_file, MmapWrapper* map_wrapper,
michael@0:              void** elf_header) {
michael@0:   int obj_fd = open(obj_file.c_str(), O_RDONLY);
michael@0:   if (obj_fd < 0) {
michael@0:     fprintf(stderr, "Failed to open ELF file '%s': %s\n",
michael@0:             obj_file.c_str(), strerror(errno));
michael@0:     return false;
michael@0:   }
michael@0:   FDWrapper obj_fd_wrapper(obj_fd);
michael@0:   struct stat st;
michael@0:   if (fstat(obj_fd, &st) != 0 && st.st_size <= 0) {
michael@0:     fprintf(stderr, "Unable to fstat ELF file '%s': %s\n",
michael@0:             obj_file.c_str(), strerror(errno));
michael@0:     return false;
michael@0:   }
michael@0:   void *obj_base = mmap(NULL, st.st_size,
michael@0:                         PROT_READ | PROT_WRITE, MAP_PRIVATE, obj_fd, 0);
michael@0:   if (obj_base == MAP_FAILED) {
michael@0:     fprintf(stderr, "Failed to mmap ELF file '%s': %s\n",
michael@0:             obj_file.c_str(), strerror(errno));
michael@0:     return false;
michael@0:   }
michael@0:   map_wrapper->set(obj_base, st.st_size);
michael@0:   *elf_header = obj_base;
michael@0:   if (!IsValidElf(*elf_header)) {
michael@0:     fprintf(stderr, "Not a valid ELF file: %s\n", obj_file.c_str());
michael@0:     return false;
michael@0:   }
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: // Get the endianness of ELF_HEADER. If it's invalid, return false.
michael@0: template<typename ElfClass>
michael@0: bool ElfEndianness(const typename ElfClass::Ehdr* elf_header,
michael@0:                    bool* big_endian) {
michael@0:   if (elf_header->e_ident[EI_DATA] == ELFDATA2LSB) {
michael@0:     *big_endian = false;
michael@0:     return true;
michael@0:   }
michael@0:   if (elf_header->e_ident[EI_DATA] == ELFDATA2MSB) {
michael@0:     *big_endian = true;
michael@0:     return true;
michael@0:   }
michael@0: 
michael@0:   fprintf(stderr, "bad data encoding in ELF header: %d\n",
michael@0:           elf_header->e_ident[EI_DATA]);
michael@0:   return false;
michael@0: }
michael@0: 
michael@0: // Read the .gnu_debuglink and get the debug file name. If anything goes
michael@0: // wrong, return an empty string.
michael@0: template<typename ElfClass>
michael@0: string ReadDebugLink(const char* debuglink,
michael@0:                      size_t debuglink_size,
michael@0:                      const string& obj_file,
michael@0:                      const std::vector<string>& debug_dirs) {
michael@0:   size_t debuglink_len = strlen(debuglink) + 5;  // '\0' + CRC32.
michael@0:   debuglink_len = 4 * ((debuglink_len + 3) / 4);  // Round to nearest 4 bytes.
michael@0: 
michael@0:   // Sanity check.
michael@0:   if (debuglink_len != debuglink_size) {
michael@0:     fprintf(stderr, "Mismatched .gnu_debuglink string / section size: "
michael@0:             "%zx %zx\n", debuglink_len, debuglink_size);
michael@0:     return "";
michael@0:   }
michael@0: 
michael@0:   bool found = false;
michael@0:   int debuglink_fd = -1;
michael@0:   string debuglink_path;
michael@0:   std::vector<string>::const_iterator it;
michael@0:   for (it = debug_dirs.begin(); it < debug_dirs.end(); ++it) {
michael@0:     const string& debug_dir = *it;
michael@0:     debuglink_path = debug_dir + "/" + debuglink;
michael@0:     debuglink_fd = open(debuglink_path.c_str(), O_RDONLY);
michael@0:     if (debuglink_fd >= 0) {
michael@0:       found = true;
michael@0:       break;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   if (!found) {
michael@0:     fprintf(stderr, "Failed to find debug ELF file for '%s' after trying:\n",
michael@0:             obj_file.c_str());
michael@0:     for (it = debug_dirs.begin(); it < debug_dirs.end(); ++it) {
michael@0:       const string debug_dir = *it;
michael@0:       fprintf(stderr, "  %s/%s\n", debug_dir.c_str(), debuglink);
michael@0:     }
michael@0:     return "";
michael@0:   }
michael@0: 
michael@0:   FDWrapper debuglink_fd_wrapper(debuglink_fd);
michael@0:   // TODO(thestig) check the CRC-32 at the end of the .gnu_debuglink
michael@0:   // section.
michael@0: 
michael@0:   return debuglink_path;
michael@0: }
michael@0: 
michael@0: //
michael@0: // LoadSymbolsInfo
michael@0: //
michael@0: // Holds the state between the two calls to LoadSymbols() in case it's necessary
michael@0: // to follow the .gnu_debuglink section and load debug information from a
michael@0: // different file.
michael@0: //
michael@0: template<typename ElfClass>
michael@0: class LoadSymbolsInfo {
michael@0:  public:
michael@0:   typedef typename ElfClass::Addr Addr;
michael@0: 
michael@0:   explicit LoadSymbolsInfo(const std::vector<string>& dbg_dirs) :
michael@0:     debug_dirs_(dbg_dirs),
michael@0:     has_loading_addr_(false) {}
michael@0: 
michael@0:   // Keeps track of which sections have been loaded so sections don't
michael@0:   // accidentally get loaded twice from two different files.
michael@0:   void LoadedSection(const string &section) {
michael@0:     if (loaded_sections_.count(section) == 0) {
michael@0:       loaded_sections_.insert(section);
michael@0:     } else {
michael@0:       fprintf(stderr, "Section %s has already been loaded.\n",
michael@0:               section.c_str());
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   // The ELF file and linked debug file are expected to have the same preferred
michael@0:   // loading address.
michael@0:   void set_loading_addr(Addr addr, const string &filename) {
michael@0:     if (!has_loading_addr_) {
michael@0:       loading_addr_ = addr;
michael@0:       loaded_file_ = filename;
michael@0:       return;
michael@0:     }
michael@0: 
michael@0:     if (addr != loading_addr_) {
michael@0:       fprintf(stderr,
michael@0:               "ELF file '%s' and debug ELF file '%s' "
michael@0:               "have different load addresses.\n",
michael@0:               loaded_file_.c_str(), filename.c_str());
michael@0:       assert(false);
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   // Setters and getters
michael@0:   const std::vector<string>& debug_dirs() const {
michael@0:     return debug_dirs_;
michael@0:   }
michael@0: 
michael@0:   string debuglink_file() const {
michael@0:     return debuglink_file_;
michael@0:   }
michael@0:   void set_debuglink_file(string file) {
michael@0:     debuglink_file_ = file;
michael@0:   }
michael@0: 
michael@0:  private:
michael@0:   const std::vector<string>& debug_dirs_; // Directories in which to
michael@0:                                           // search for the debug ELF file.
michael@0: 
michael@0:   string debuglink_file_;  // Full path to the debug ELF file.
michael@0: 
michael@0:   bool has_loading_addr_;  // Indicate if LOADING_ADDR_ is valid.
michael@0: 
michael@0:   Addr loading_addr_;  // Saves the preferred loading address from the
michael@0:                        // first call to LoadSymbols().
michael@0: 
michael@0:   string loaded_file_;  // Name of the file loaded from the first call to
michael@0:                         // LoadSymbols().
michael@0: 
michael@0:   std::set<string> loaded_sections_;  // Tracks the Loaded ELF sections
michael@0:                                       // between calls to LoadSymbols().
michael@0: };
michael@0: 
michael@0: template<typename ElfClass>
michael@0: bool LoadSymbols(const string& obj_file,
michael@0:                  const bool big_endian,
michael@0:                  const typename ElfClass::Ehdr* elf_header,
michael@0:                  const bool read_gnu_debug_link,
michael@0:                  LoadSymbolsInfo<ElfClass>* info,
michael@0:                  SymbolData symbol_data,
michael@0:                  Module* module) {
michael@0:   typedef typename ElfClass::Addr Addr;
michael@0:   typedef typename ElfClass::Phdr Phdr;
michael@0:   typedef typename ElfClass::Shdr Shdr;
michael@0: 
michael@0:   BPLOG(INFO) << "";
michael@0:   BPLOG(INFO) << "LoadSymbols: BEGIN   " << obj_file;
michael@0: 
michael@0:   Addr loading_addr = GetLoadingAddress<ElfClass>(
michael@0:       GetOffset<ElfClass, Phdr>(elf_header, elf_header->e_phoff),
michael@0:       elf_header->e_phnum);
michael@0:   module->SetLoadAddress(loading_addr);
michael@0:   info->set_loading_addr(loading_addr, obj_file);
michael@0: 
michael@0:   const Shdr* sections =
michael@0:       GetOffset<ElfClass, Shdr>(elf_header, elf_header->e_shoff);
michael@0:   const Shdr* section_names = sections + elf_header->e_shstrndx;
michael@0:   const char* names =
michael@0:       GetOffset<ElfClass, char>(elf_header, section_names->sh_offset);
michael@0:   const char *names_end = names + section_names->sh_size;
michael@0:   bool found_debug_info_section = false;
michael@0:   bool found_usable_info = false;
michael@0: 
michael@0:   if (symbol_data != ONLY_CFI) {
michael@0: #ifndef NO_STABS_SUPPORT
michael@0:     // Look for STABS debugging information, and load it if present.
michael@0:     const Shdr* stab_section =
michael@0:       FindElfSectionByName<ElfClass>(".stab", SHT_PROGBITS,
michael@0:                                      sections, names, names_end,
michael@0:                                      elf_header->e_shnum);
michael@0:     if (stab_section) {
michael@0:       const Shdr* stabstr_section = stab_section->sh_link + sections;
michael@0:       if (stabstr_section) {
michael@0:         found_debug_info_section = true;
michael@0:         found_usable_info = true;
michael@0:         info->LoadedSection(".stab");
michael@0:         if (!LoadStabs<ElfClass>(elf_header, stab_section, stabstr_section,
michael@0:                                  big_endian, module)) {
michael@0:           fprintf(stderr, "%s: \".stab\" section found, but failed to load"
michael@0:                   " STABS debugging information\n", obj_file.c_str());
michael@0:         }
michael@0:       }
michael@0:     }
michael@0: #endif  // NO_STABS_SUPPORT
michael@0: 
michael@0:     // Look for DWARF debugging information, and load it if present.
michael@0:     const Shdr* dwarf_section =
michael@0:       FindElfSectionByName<ElfClass>(".debug_info", SHT_PROGBITS,
michael@0:                                      sections, names, names_end,
michael@0:                                      elf_header->e_shnum);
michael@0:     if (dwarf_section) {
michael@0:       found_debug_info_section = true;
michael@0:       found_usable_info = true;
michael@0:       info->LoadedSection(".debug_info");
michael@0:       if (!LoadDwarf<ElfClass>(obj_file, elf_header, big_endian, module))
michael@0:         fprintf(stderr, "%s: \".debug_info\" section found, but failed to load "
michael@0:                 "DWARF debugging information\n", obj_file.c_str());
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   if (symbol_data != NO_CFI) {
michael@0:     // Dwarf Call Frame Information (CFI) is actually independent from
michael@0:     // the other DWARF debugging information, and can be used alone.
michael@0:     const Shdr* dwarf_cfi_section =
michael@0:         FindElfSectionByName<ElfClass>(".debug_frame", SHT_PROGBITS,
michael@0:                                        sections, names, names_end,
michael@0:                                        elf_header->e_shnum);
michael@0:     if (dwarf_cfi_section) {
michael@0:       // Ignore the return value of this function; even without call frame
michael@0:       // information, the other debugging information could be perfectly
michael@0:       // useful.
michael@0:       info->LoadedSection(".debug_frame");
michael@0:       bool result =
michael@0:           LoadDwarfCFI<ElfClass>(obj_file, elf_header, ".debug_frame",
michael@0:                                  dwarf_cfi_section, false, 0, 0, big_endian,
michael@0:                                  module);
michael@0:       found_usable_info = found_usable_info || result;
michael@0:       if (result)
michael@0:         BPLOG(INFO) << "LoadSymbols:   read CFI from .debug_frame";
michael@0:     }
michael@0: 
michael@0:     // Linux C++ exception handling information can also provide
michael@0:     // unwinding data.
michael@0:     const Shdr* eh_frame_section =
michael@0:         FindElfSectionByName<ElfClass>(".eh_frame", SHT_PROGBITS,
michael@0:                                        sections, names, names_end,
michael@0:                                        elf_header->e_shnum);
michael@0:     if (eh_frame_section) {
michael@0:       // Pointers in .eh_frame data may be relative to the base addresses of
michael@0:       // certain sections. Provide those sections if present.
michael@0:       const Shdr* got_section =
michael@0:           FindElfSectionByName<ElfClass>(".got", SHT_PROGBITS,
michael@0:                                          sections, names, names_end,
michael@0:                                          elf_header->e_shnum);
michael@0:       const Shdr* text_section =
michael@0:           FindElfSectionByName<ElfClass>(".text", SHT_PROGBITS,
michael@0:                                          sections, names, names_end,
michael@0:                                          elf_header->e_shnum);
michael@0:       info->LoadedSection(".eh_frame");
michael@0:       // As above, ignore the return value of this function.
michael@0:       bool result =
michael@0:           LoadDwarfCFI<ElfClass>(obj_file, elf_header, ".eh_frame",
michael@0:                                  eh_frame_section, true,
michael@0:                                  got_section, text_section, big_endian, module);
michael@0:       found_usable_info = found_usable_info || result;
michael@0:       if (result)
michael@0:         BPLOG(INFO) << "LoadSymbols:   read CFI from .eh_frame";
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   // ARM has special unwind tables that can be used.
michael@0:   const Shdr* arm_exidx_section =
michael@0:       FindElfSectionByName<ElfClass>(".ARM.exidx", SHT_ARM_EXIDX,
michael@0:                                      sections, names, names_end,
michael@0:                                      elf_header->e_shnum);
michael@0:   const Shdr* arm_extab_section =
michael@0:       FindElfSectionByName<ElfClass>(".ARM.extab", SHT_PROGBITS,
michael@0:                                      sections, names, names_end,
michael@0:                                      elf_header->e_shnum);
michael@0:   // Only load information from this section if there isn't a .debug_info
michael@0:   // section.
michael@0:   if (!found_debug_info_section
michael@0:       && arm_exidx_section && arm_extab_section && symbol_data != NO_CFI) {
michael@0:     info->LoadedSection(".ARM.exidx");
michael@0:     info->LoadedSection(".ARM.extab");
michael@0:     bool result = LoadARMexidx<ElfClass>(elf_header,
michael@0:                                          arm_exidx_section, arm_extab_section,
michael@0:                                          loading_addr, module);
michael@0:     found_usable_info = found_usable_info || result;
michael@0:     if (result)
michael@0:       BPLOG(INFO) << "LoadSymbols:   read EXIDX from .ARM.{exidx,extab}";
michael@0:   }
michael@0: 
michael@0:   if (!found_debug_info_section && symbol_data != ONLY_CFI) {
michael@0:     fprintf(stderr, "%s: file contains no debugging information"
michael@0:             " (no \".stab\" or \".debug_info\" sections)\n",
michael@0:             obj_file.c_str());
michael@0: 
michael@0:     // Failed, but maybe there's a .gnu_debuglink section?
michael@0:     if (read_gnu_debug_link) {
michael@0:       const Shdr* gnu_debuglink_section
michael@0:           = FindElfSectionByName<ElfClass>(".gnu_debuglink", SHT_PROGBITS,
michael@0:                                            sections, names,
michael@0:                                            names_end, elf_header->e_shnum);
michael@0:       if (gnu_debuglink_section) {
michael@0:         if (!info->debug_dirs().empty()) {
michael@0:           const char* debuglink_contents =
michael@0:               GetOffset<ElfClass, char>(elf_header,
michael@0:                                         gnu_debuglink_section->sh_offset);
michael@0:           string debuglink_file
michael@0:               = ReadDebugLink<ElfClass>(debuglink_contents,
michael@0:                                         gnu_debuglink_section->sh_size,
michael@0:                                         obj_file, info->debug_dirs());
michael@0:           info->set_debuglink_file(debuglink_file);
michael@0:         } else {
michael@0:           fprintf(stderr, ".gnu_debuglink section found in '%s', "
michael@0:                   "but no debug path specified.\n", obj_file.c_str());
michael@0:         }
michael@0:       } else {
michael@0:         fprintf(stderr, "%s does not contain a .gnu_debuglink section.\n",
michael@0:                 obj_file.c_str());
michael@0:       }
michael@0:     } else {
michael@0:       if (symbol_data != ONLY_CFI) {
michael@0:         // The caller doesn't want to consult .gnu_debuglink.
michael@0:         // See if there are export symbols available.
michael@0:         const Shdr* dynsym_section =
michael@0:           FindElfSectionByName<ElfClass>(".dynsym", SHT_DYNSYM,
michael@0:                                          sections, names, names_end,
michael@0:                                          elf_header->e_shnum);
michael@0:         const Shdr* dynstr_section =
michael@0:           FindElfSectionByName<ElfClass>(".dynstr", SHT_STRTAB,
michael@0:                                          sections, names, names_end,
michael@0:                                          elf_header->e_shnum);
michael@0:         if (dynsym_section && dynstr_section) {
michael@0:           info->LoadedSection(".dynsym");
michael@0: 
michael@0:           const uint8_t* dynsyms =
michael@0:               GetOffset<ElfClass, uint8_t>(elf_header,
michael@0:                                            dynsym_section->sh_offset);
michael@0:           const uint8_t* dynstrs =
michael@0:               GetOffset<ElfClass, uint8_t>(elf_header,
michael@0:                                            dynstr_section->sh_offset);
michael@0:           bool result =
michael@0:               ELFSymbolsToModule(dynsyms,
michael@0:                                  dynsym_section->sh_size,
michael@0:                                  dynstrs,
michael@0:                                  dynstr_section->sh_size,
michael@0:                                  big_endian,
michael@0:                                  ElfClass::kAddrSize,
michael@0:                                  module);
michael@0:           found_usable_info = found_usable_info || result;
michael@0:         }
michael@0:       }
michael@0: 
michael@0:       // Return true if some usable information was found, since
michael@0:       // the caller doesn't want to use .gnu_debuglink.
michael@0:       BPLOG(INFO) << "LoadSymbols: " 
michael@0:                   << (found_usable_info ? "SUCCESS " : "FAILURE ")
michael@0:                   << obj_file;
michael@0:       return found_usable_info;
michael@0:     }
michael@0: 
michael@0:     // No debug info was found, let the user try again with .gnu_debuglink
michael@0:     // if present.
michael@0:     BPLOG(INFO) << "LoadSymbols: FAILURE " << obj_file;
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   BPLOG(INFO) << "LoadSymbols: SUCCESS " << obj_file;
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: // Return the breakpad symbol file identifier for the architecture of
michael@0: // ELF_HEADER.
michael@0: template<typename ElfClass>
michael@0: const char* ElfArchitecture(const typename ElfClass::Ehdr* elf_header) {
michael@0:   typedef typename ElfClass::Half Half;
michael@0:   Half arch = elf_header->e_machine;
michael@0:   switch (arch) {
michael@0:     case EM_386:        return "x86";
michael@0:     case EM_ARM:        return "arm";
michael@0:     case EM_MIPS:       return "mips";
michael@0:     case EM_PPC64:      return "ppc64";
michael@0:     case EM_PPC:        return "ppc";
michael@0:     case EM_S390:       return "s390";
michael@0:     case EM_SPARC:      return "sparc";
michael@0:     case EM_SPARCV9:    return "sparcv9";
michael@0:     case EM_X86_64:     return "x86_64";
michael@0:     default: return NULL;
michael@0:   }
michael@0: }
michael@0: 
michael@0: // Format the Elf file identifier in IDENTIFIER as a UUID with the
michael@0: // dashes removed.
michael@0: string FormatIdentifier(unsigned char identifier[16]) {
michael@0:   char identifier_str[40];
michael@0:   google_breakpad::FileID::ConvertIdentifierToString(
michael@0:       identifier,
michael@0:       identifier_str,
michael@0:       sizeof(identifier_str));
michael@0:   string id_no_dash;
michael@0:   for (int i = 0; identifier_str[i] != '\0'; ++i)
michael@0:     if (identifier_str[i] != '-')
michael@0:       id_no_dash += identifier_str[i];
michael@0:   // Add an extra "0" by the end.  PDB files on Windows have an 'age'
michael@0:   // number appended to the end of the file identifier; this isn't
michael@0:   // really used or necessary on other platforms, but be consistent.
michael@0:   id_no_dash += '0';
michael@0:   return id_no_dash;
michael@0: }
michael@0: 
michael@0: // Return the non-directory portion of FILENAME: the portion after the
michael@0: // last slash, or the whole filename if there are no slashes.
michael@0: string BaseFileName(const string &filename) {
michael@0:   // Lots of copies!  basename's behavior is less than ideal.
michael@0:   char *c_filename = strdup(filename.c_str());
michael@0:   string base = basename(c_filename);
michael@0:   free(c_filename);
michael@0:   return base;
michael@0: }
michael@0: 
michael@0: template<typename ElfClass>
michael@0: bool ReadSymbolDataElfClass(const typename ElfClass::Ehdr* elf_header,
michael@0:                              const string& obj_filename,
michael@0:                              const std::vector<string>& debug_dirs,
michael@0:                              SymbolData symbol_data,
michael@0:                              Module** out_module) {
michael@0:   typedef typename ElfClass::Ehdr Ehdr;
michael@0:   typedef typename ElfClass::Shdr Shdr;
michael@0: 
michael@0:   *out_module = NULL;
michael@0: 
michael@0:   unsigned char identifier[16];
michael@0:   if (!google_breakpad::FileID::ElfFileIdentifierFromMappedFile(elf_header,
michael@0:                                                                 identifier)) {
michael@0:     fprintf(stderr, "%s: unable to generate file identifier\n",
michael@0:             obj_filename.c_str());
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   const char *architecture = ElfArchitecture<ElfClass>(elf_header);
michael@0:   if (!architecture) {
michael@0:     fprintf(stderr, "%s: unrecognized ELF machine architecture: %d\n",
michael@0:             obj_filename.c_str(), elf_header->e_machine);
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   // Figure out what endianness this file is.
michael@0:   bool big_endian;
michael@0:   if (!ElfEndianness<ElfClass>(elf_header, &big_endian))
michael@0:     return false;
michael@0: 
michael@0:   string name = BaseFileName(obj_filename);
michael@0:   string os = "Linux";
michael@0:   string id = FormatIdentifier(identifier);
michael@0: 
michael@0:   LoadSymbolsInfo<ElfClass> info(debug_dirs);
michael@0:   scoped_ptr<Module> module(new Module(name, os, architecture, id));
michael@0:   if (!LoadSymbols<ElfClass>(obj_filename, big_endian, elf_header,
michael@0:                              !debug_dirs.empty(), &info,
michael@0:                              symbol_data, module.get())) {
michael@0:     const string debuglink_file = info.debuglink_file();
michael@0:     if (debuglink_file.empty())
michael@0:       return false;
michael@0: 
michael@0:     // Load debuglink ELF file.
michael@0:     fprintf(stderr, "Found debugging info in %s\n", debuglink_file.c_str());
michael@0:     MmapWrapper debug_map_wrapper;
michael@0:     Ehdr* debug_elf_header = NULL;
michael@0:     if (!LoadELF(debuglink_file, &debug_map_wrapper,
michael@0:                  reinterpret_cast<void**>(&debug_elf_header)))
michael@0:       return false;
michael@0:     // Sanity checks to make sure everything matches up.
michael@0:     const char *debug_architecture =
michael@0:         ElfArchitecture<ElfClass>(debug_elf_header);
michael@0:     if (!debug_architecture) {
michael@0:       fprintf(stderr, "%s: unrecognized ELF machine architecture: %d\n",
michael@0:               debuglink_file.c_str(), debug_elf_header->e_machine);
michael@0:       return false;
michael@0:     }
michael@0:     if (strcmp(architecture, debug_architecture)) {
michael@0:       fprintf(stderr, "%s with ELF machine architecture %s does not match "
michael@0:               "%s with ELF architecture %s\n",
michael@0:               debuglink_file.c_str(), debug_architecture,
michael@0:               obj_filename.c_str(), architecture);
michael@0:       return false;
michael@0:     }
michael@0: 
michael@0:     bool debug_big_endian;
michael@0:     if (!ElfEndianness<ElfClass>(debug_elf_header, &debug_big_endian))
michael@0:       return false;
michael@0:     if (debug_big_endian != big_endian) {
michael@0:       fprintf(stderr, "%s and %s does not match in endianness\n",
michael@0:               obj_filename.c_str(), debuglink_file.c_str());
michael@0:       return false;
michael@0:     }
michael@0: 
michael@0:     if (!LoadSymbols<ElfClass>(debuglink_file, debug_big_endian,
michael@0:                                debug_elf_header, false, &info,
michael@0:                                symbol_data, module.get())) {
michael@0:       return false;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   *out_module = module.release();
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: }  // namespace
michael@0: 
michael@0: namespace google_breakpad {
michael@0: 
michael@0: // Not explicitly exported, but not static so it can be used in unit tests.
michael@0: bool ReadSymbolDataInternal(const uint8_t* obj_file,
michael@0:                             const string& obj_filename,
michael@0:                             const std::vector<string>& debug_dirs,
michael@0:                             SymbolData symbol_data,
michael@0:                             Module** module) {
michael@0: 
michael@0:   if (!IsValidElf(obj_file)) {
michael@0:     fprintf(stderr, "Not a valid ELF file: %s\n", obj_filename.c_str());
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   int elfclass = ElfClass(obj_file);
michael@0:   if (elfclass == ELFCLASS32) {
michael@0:     return ReadSymbolDataElfClass<ElfClass32>(
michael@0:         reinterpret_cast<const Elf32_Ehdr*>(obj_file), obj_filename, debug_dirs,
michael@0:         symbol_data, module);
michael@0:   }
michael@0:   if (elfclass == ELFCLASS64) {
michael@0:     return ReadSymbolDataElfClass<ElfClass64>(
michael@0:         reinterpret_cast<const Elf64_Ehdr*>(obj_file), obj_filename, debug_dirs,
michael@0:         symbol_data, module);
michael@0:   }
michael@0: 
michael@0:   return false;
michael@0: }
michael@0: 
michael@0: bool WriteSymbolFile(const string &obj_file,
michael@0:                      const std::vector<string>& debug_dirs,
michael@0:                      SymbolData symbol_data,
michael@0:                      std::ostream &sym_stream) {
michael@0:   Module* module;
michael@0:   if (!ReadSymbolData(obj_file, debug_dirs, symbol_data, &module))
michael@0:     return false;
michael@0: 
michael@0:   bool result = module->Write(sym_stream, symbol_data);
michael@0:   delete module;
michael@0:   return result;
michael@0: }
michael@0: 
michael@0: bool ReadSymbolData(const string& obj_file,
michael@0:                     const std::vector<string>& debug_dirs,
michael@0:                     SymbolData symbol_data,
michael@0:                     Module** module) {
michael@0:   MmapWrapper map_wrapper;
michael@0:   void* elf_header = NULL;
michael@0:   if (!LoadELF(obj_file, &map_wrapper, &elf_header))
michael@0:     return false;
michael@0: 
michael@0:   return ReadSymbolDataInternal(reinterpret_cast<uint8_t*>(elf_header),
michael@0:                                 obj_file, debug_dirs, symbol_data, module);
michael@0: }
michael@0: 
michael@0: }  // namespace google_breakpad