michael@0: /* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
michael@0: /* vim: set ts=8 sts=2 et sw=2 tw=80: */
michael@0: 
michael@0: // Copyright (c) 2006, 2011, 2012 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: // (derived from)
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: // dump_symbols.h: Read debugging information from an ELF file, and write
michael@0: // it out as a Breakpad symbol file.
michael@0: 
michael@0: // This file is derived from the following files in
michael@0: // toolkit/crashreporter/google-breakpad:
michael@0: //   src/common/linux/dump_symbols.cc
michael@0: //   src/common/linux/elfutils.cc
michael@0: //   src/common/linux/file_id.cc
michael@0: 
michael@0: #include <errno.h>
michael@0: #include <fcntl.h>
michael@0: #include <stdio.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: #include <arpa/inet.h>
michael@0: 
michael@0: #include <set>
michael@0: #include <string>
michael@0: #include <vector>
michael@0: 
michael@0: #include "mozilla/Assertions.h"
michael@0: 
michael@0: #include "LulPlatformMacros.h"
michael@0: #include "LulCommonExt.h"
michael@0: #include "LulDwarfExt.h"
michael@0: #if defined(LUL_PLAT_arm_android)
michael@0: # include "LulExidxExt.h"
michael@0: #endif
michael@0: #include "LulElfInt.h"
michael@0: #include "LulMainInt.h"
michael@0: 
michael@0: 
michael@0: #if defined(LUL_PLAT_arm_android) && !defined(SHT_ARM_EXIDX)
michael@0: // bionic and older glibsc don't define it
michael@0: # define SHT_ARM_EXIDX (SHT_LOPROC + 1)
michael@0: #endif
michael@0: 
michael@0: 
michael@0: // This namespace contains helper functions.
michael@0: namespace {
michael@0: 
michael@0: using lul::DwarfCFIToModule;
michael@0: using lul::FindElfSectionByName;
michael@0: using lul::GetOffset;
michael@0: using lul::IsValidElf;
michael@0: using lul::Module;
michael@0: using lul::UniqueString;
michael@0: using lul::scoped_ptr;
michael@0: using lul::Summariser;
michael@0: using std::string;
michael@0: using std::vector;
michael@0: using std::set;
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:       MOZ_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:     MOZ_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: 
michael@0: // Set NUM_DW_REGNAMES to be the number of Dwarf register names
michael@0: // appropriate to the machine architecture given in HEADER.  Return
michael@0: // true on 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:                            unsigned int* num_dw_regnames) {
michael@0:   switch (elf_header->e_machine) {
michael@0:     case EM_386:
michael@0:       *num_dw_regnames = DwarfCFIToModule::RegisterNames::I386();
michael@0:       return true;
michael@0:     case EM_ARM:
michael@0:       *num_dw_regnames = DwarfCFIToModule::RegisterNames::ARM();
michael@0:       return true;
michael@0:     case EM_X86_64:
michael@0:       *num_dw_regnames = DwarfCFIToModule::RegisterNames::X86_64();
michael@0:       return true;
michael@0:     default:
michael@0:       MOZ_ASSERT(0);
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:                   SecMap* smap,
michael@0:                   uintptr_t text_bias,
michael@0:                   void (*log)(const char*)) {
michael@0:   // Find the appropriate set of register names for this file's
michael@0:   // architecture.
michael@0:   unsigned int num_dw_regs = 0;
michael@0:   if (!DwarfCFIRegisterNames<ElfClass>(elf_header, &num_dw_regs)) {
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 lul::Endianness endianness
michael@0:     = big_endian ? lul::ENDIANNESS_BIG : lul::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: 
michael@0:   // Here's a summariser, which will receive the output of the
michael@0:   // parser, create summaries, and add them to |smap|.
michael@0:   Summariser* summ = new Summariser(smap, text_bias, log);
michael@0: 
michael@0:   DwarfCFIToModule::Reporter module_reporter(log, dwarf_filename, section_name);
michael@0:   DwarfCFIToModule handler(num_dw_regs, &module_reporter, summ);
michael@0:   lul::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:   lul::CallFrameInfo::Reporter dwarf_reporter(log, dwarf_filename,
michael@0:                                               section_name);
michael@0:   lul::CallFrameInfo parser(cfi, cfi_size,
michael@0:                             &byte_reader, &handler, &dwarf_reporter,
michael@0:                             eh_frame);
michael@0:   parser.Start();
michael@0: 
michael@0:   delete summ;
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: #if defined(LUL_PLAT_arm_android)
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:                   uintptr_t text_bias,
michael@0:                   SecMap* smap,
michael@0:                   void (*log)(const char*)) {
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:   // Note that we are reading EXIDX directly out of the mapped in
michael@0:   // executable image.  Unlike with the CFI reader, there is no
michael@0:   // auxiliary, temporary mapping used to read the unwind data.
michael@0:   //
michael@0:   // An .exidx section is always required, but the .extab section
michael@0:   // can be optionally omitted, provided that .exidx does not refer
michael@0:   // to it.  If the .exidx is erroneous and does refer to .extab even
michael@0:   // though .extab is missing, the range checks done by GET_EX_U32 in
michael@0:   // ExceptionTableInfo::ExtabEntryExtract should prevent any invalid
michael@0:   // memory accesses, and cause the .extab to be rejected as invalid.
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:     = extab_section
michael@0:         ? GetOffset<ElfClass, char>(elf_header, extab_section->sh_offset)
michael@0:         : nullptr;
michael@0:   size_t extab_size = extab_section ? extab_section->sh_size : 0;
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:   lul::ARMExToModule handler(smap, log);
michael@0:   lul::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, text_bias, log);
michael@0:   parser.Start();
michael@0:   return true;
michael@0: }
michael@0: #endif /* defined(LUL_PLAT_arm_android) */
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:   // Mapping it read-only is good enough.  In any case, mapping it
michael@0:   // read-write confuses Valgrind's debuginfo acquire/discard
michael@0:   // heuristics, making it hard to profile the profiler.
michael@0:   void *obj_base = mmap(nullptr, st.st_size,
michael@0:                         PROT_READ, 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: //
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 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:   string debuglink_file() const {
michael@0:     return debuglink_file_;
michael@0:   }
michael@0: 
michael@0:  private:
michael@0:   const 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:   set<string> loaded_sections_; // Tracks the Loaded ELF sections
michael@0:                                 // between calls to LoadSymbols().
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 non-PIC executables (e_type == ET_EXEC), the load address is
michael@0:   // the start address of the first PT_LOAD segment.  (ELF requires
michael@0:   // the segments to be sorted by load address.)  For PIC executables
michael@0:   // and dynamic libraries (e_type == ET_DYN), this address will
michael@0:   // normally be zero.
michael@0:   for (int i = 0; i < nheader; ++i) {
michael@0:     const Phdr& header = program_headers[i];
michael@0:     if (header.p_type == PT_LOAD)
michael@0:       return header.p_vaddr;
michael@0:   }
michael@0:   return 0;
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:                  SecMap* smap,
michael@0:                  void* rx_avma,
michael@0:                  void (*log)(const char*)) {
michael@0:   typedef typename ElfClass::Phdr Phdr;
michael@0:   typedef typename ElfClass::Shdr Shdr;
michael@0: 
michael@0:   char buf[500];
michael@0:   snprintf(buf, sizeof(buf), "LoadSymbols: BEGIN   %s\n", obj_file.c_str());
michael@0:   buf[sizeof(buf)-1] = 0;
michael@0:   log(buf);
michael@0: 
michael@0:   // This is how the text bias is calculated.
michael@0:   // BEGIN CALCULATE BIAS
michael@0:   uintptr_t loading_addr = GetLoadingAddress<ElfClass>(
michael@0:       GetOffset<ElfClass, Phdr>(elf_header, elf_header->e_phoff),
michael@0:       elf_header->e_phnum);
michael@0:   uintptr_t text_bias = ((uintptr_t)rx_avma) - loading_addr;
michael@0:   snprintf(buf, sizeof(buf),
michael@0:            "LoadSymbols:   rx_avma=%llx, text_bias=%llx",
michael@0:            (unsigned long long int)(uintptr_t)rx_avma,
michael@0:            (unsigned long long int)text_bias);
michael@0:   buf[sizeof(buf)-1] = 0;
michael@0:   log(buf);
michael@0:   // END CALCULATE BIAS
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_usable_info = false;
michael@0: 
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:                                smap, text_bias, log);
michael@0:     found_usable_info = found_usable_info || result;
michael@0:     if (result)
michael@0:       log("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,
michael@0:                                smap, text_bias, log);
michael@0:     found_usable_info = found_usable_info || result;
michael@0:     if (result)
michael@0:       log("LoadSymbols:   read CFI from .eh_frame");
michael@0:   }
michael@0: 
michael@0: # if defined(LUL_PLAT_arm_android)
michael@0:   // ARM has special unwind tables that can be used.  .exidx is
michael@0:   // always required, and .extab is normally required, but may
michael@0:   // be omitted if it is empty.  See comments on LoadARMexidx()
michael@0:   // for more details.
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:   const Shdr* debug_info_section =
michael@0:       FindElfSectionByName<ElfClass>(".debug_info", 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 (!debug_info_section && arm_exidx_section) {
michael@0:     info->LoadedSection(".ARM.exidx");
michael@0:     if (arm_extab_section)
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, text_bias, smap, log);
michael@0:     found_usable_info = found_usable_info || result;
michael@0:     if (result)
michael@0:       log("LoadSymbols:   read EXIDX from .ARM.{exidx,extab}");
michael@0:   }
michael@0: # endif /* defined(LUL_PLAT_arm_android) */
michael@0: 
michael@0:   snprintf(buf, sizeof(buf), "LoadSymbols: END     %s\n", obj_file.c_str());
michael@0:   buf[sizeof(buf)-1] = 0;
michael@0:   log(buf);
michael@0: 
michael@0:   return found_usable_info;
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:   lul::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 vector<string>& debug_dirs,
michael@0:                             SecMap* smap, void* rx_avma,
michael@0:                             void (*log)(const char*)) {
michael@0:   typedef typename ElfClass::Ehdr Ehdr;
michael@0: 
michael@0:   unsigned char identifier[16];
michael@0:   if (!lul
michael@0:       ::FileID::ElfFileIdentifierFromMappedFile(elf_header, 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:   if (!LoadSymbols<ElfClass>(obj_filename, big_endian, elf_header,
michael@0:                              !debug_dirs.empty(), &info,
michael@0:                              smap, rx_avma, log)) {
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:                                smap, rx_avma, log)) {
michael@0:       return false;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: }  // namespace (anon)
michael@0: 
michael@0: 
michael@0: namespace lul {
michael@0: 
michael@0: bool ReadSymbolDataInternal(const uint8_t* obj_file,
michael@0:                             const string& obj_filename,
michael@0:                             const vector<string>& debug_dirs,
michael@0:                             SecMap* smap, void* rx_avma,
michael@0:                             void (*log)(const char*)) {
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),
michael@0:         obj_filename, debug_dirs, smap, rx_avma, log);
michael@0:   }
michael@0:   if (elfclass == ELFCLASS64) {
michael@0:     return ReadSymbolDataElfClass<ElfClass64>(
michael@0:         reinterpret_cast<const Elf64_Ehdr*>(obj_file),
michael@0:         obj_filename, debug_dirs, smap, rx_avma, log);
michael@0:   }
michael@0: 
michael@0:   return false;
michael@0: }
michael@0: 
michael@0: bool ReadSymbolData(const string& obj_file,
michael@0:                     const vector<string>& debug_dirs,
michael@0:                     SecMap* smap, void* rx_avma,
michael@0:                     void (*log)(const char*)) {
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, smap, rx_avma, log);
michael@0: }
michael@0: 
michael@0: 
michael@0: namespace {
michael@0: 
michael@0: template<typename ElfClass>
michael@0: void FindElfClassSection(const char *elf_base,
michael@0:                          const char *section_name,
michael@0:                          typename ElfClass::Word section_type,
michael@0:                          const void **section_start,
michael@0:                          int *section_size) {
michael@0:   typedef typename ElfClass::Ehdr Ehdr;
michael@0:   typedef typename ElfClass::Shdr Shdr;
michael@0: 
michael@0:   MOZ_ASSERT(elf_base);
michael@0:   MOZ_ASSERT(section_start);
michael@0:   MOZ_ASSERT(section_size);
michael@0: 
michael@0:   MOZ_ASSERT(strncmp(elf_base, ELFMAG, SELFMAG) == 0);
michael@0: 
michael@0:   const Ehdr* elf_header = reinterpret_cast<const Ehdr*>(elf_base);
michael@0:   MOZ_ASSERT(elf_header->e_ident[EI_CLASS] == ElfClass::kClass);
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: 
michael@0:   const Shdr* section =
michael@0:     FindElfSectionByName<ElfClass>(section_name, section_type,
michael@0:                                    sections, names, names_end,
michael@0:                                    elf_header->e_shnum);
michael@0: 
michael@0:   if (section != NULL && section->sh_size > 0) {
michael@0:     *section_start = elf_base + section->sh_offset;
michael@0:     *section_size = section->sh_size;
michael@0:   }
michael@0: }
michael@0: 
michael@0: template<typename ElfClass>
michael@0: void FindElfClassSegment(const char *elf_base,
michael@0:                          typename ElfClass::Word segment_type,
michael@0:                          const void **segment_start,
michael@0:                          int *segment_size) {
michael@0:   typedef typename ElfClass::Ehdr Ehdr;
michael@0:   typedef typename ElfClass::Phdr Phdr;
michael@0: 
michael@0:   MOZ_ASSERT(elf_base);
michael@0:   MOZ_ASSERT(segment_start);
michael@0:   MOZ_ASSERT(segment_size);
michael@0: 
michael@0:   MOZ_ASSERT(strncmp(elf_base, ELFMAG, SELFMAG) == 0);
michael@0: 
michael@0:   const Ehdr* elf_header = reinterpret_cast<const Ehdr*>(elf_base);
michael@0:   MOZ_ASSERT(elf_header->e_ident[EI_CLASS] == ElfClass::kClass);
michael@0: 
michael@0:   const Phdr* phdrs =
michael@0:     GetOffset<ElfClass,Phdr>(elf_header, elf_header->e_phoff);
michael@0: 
michael@0:   for (int i = 0; i < elf_header->e_phnum; ++i) {
michael@0:     if (phdrs[i].p_type == segment_type) {
michael@0:       *segment_start = elf_base + phdrs[i].p_offset;
michael@0:       *segment_size = phdrs[i].p_filesz;
michael@0:       return;
michael@0:     }
michael@0:   }
michael@0: }
michael@0: 
michael@0: }  // namespace (anon)
michael@0: 
michael@0: bool IsValidElf(const void* elf_base) {
michael@0:   return strncmp(reinterpret_cast<const char*>(elf_base),
michael@0:                  ELFMAG, SELFMAG) == 0;
michael@0: }
michael@0: 
michael@0: int ElfClass(const void* elf_base) {
michael@0:   const ElfW(Ehdr)* elf_header =
michael@0:     reinterpret_cast<const ElfW(Ehdr)*>(elf_base);
michael@0: 
michael@0:   return elf_header->e_ident[EI_CLASS];
michael@0: }
michael@0: 
michael@0: bool FindElfSection(const void *elf_mapped_base,
michael@0:                     const char *section_name,
michael@0:                     uint32_t section_type,
michael@0:                     const void **section_start,
michael@0:                     int *section_size,
michael@0:                     int *elfclass) {
michael@0:   MOZ_ASSERT(elf_mapped_base);
michael@0:   MOZ_ASSERT(section_start);
michael@0:   MOZ_ASSERT(section_size);
michael@0: 
michael@0:   *section_start = NULL;
michael@0:   *section_size = 0;
michael@0: 
michael@0:   if (!IsValidElf(elf_mapped_base))
michael@0:     return false;
michael@0: 
michael@0:   int cls = ElfClass(elf_mapped_base);
michael@0:   if (elfclass) {
michael@0:     *elfclass = cls;
michael@0:   }
michael@0: 
michael@0:   const char* elf_base =
michael@0:     static_cast<const char*>(elf_mapped_base);
michael@0: 
michael@0:   if (cls == ELFCLASS32) {
michael@0:     FindElfClassSection<ElfClass32>(elf_base, section_name, section_type,
michael@0:                                     section_start, section_size);
michael@0:     return *section_start != NULL;
michael@0:   } else if (cls == ELFCLASS64) {
michael@0:     FindElfClassSection<ElfClass64>(elf_base, section_name, section_type,
michael@0:                                     section_start, section_size);
michael@0:     return *section_start != NULL;
michael@0:   }
michael@0: 
michael@0:   return false;
michael@0: }
michael@0: 
michael@0: bool FindElfSegment(const void *elf_mapped_base,
michael@0:                     uint32_t segment_type,
michael@0:                     const void **segment_start,
michael@0:                     int *segment_size,
michael@0:                     int *elfclass) {
michael@0:   MOZ_ASSERT(elf_mapped_base);
michael@0:   MOZ_ASSERT(segment_start);
michael@0:   MOZ_ASSERT(segment_size);
michael@0: 
michael@0:   *segment_start = NULL;
michael@0:   *segment_size = 0;
michael@0: 
michael@0:   if (!IsValidElf(elf_mapped_base))
michael@0:     return false;
michael@0: 
michael@0:   int cls = ElfClass(elf_mapped_base);
michael@0:   if (elfclass) {
michael@0:     *elfclass = cls;
michael@0:   }
michael@0: 
michael@0:   const char* elf_base =
michael@0:     static_cast<const char*>(elf_mapped_base);
michael@0: 
michael@0:   if (cls == ELFCLASS32) {
michael@0:     FindElfClassSegment<ElfClass32>(elf_base, segment_type,
michael@0:                                     segment_start, segment_size);
michael@0:     return *segment_start != NULL;
michael@0:   } else if (cls == ELFCLASS64) {
michael@0:     FindElfClassSegment<ElfClass64>(elf_base, segment_type,
michael@0:                                     segment_start, segment_size);
michael@0:     return *segment_start != NULL;
michael@0:   }
michael@0: 
michael@0:   return false;
michael@0: }
michael@0: 
michael@0: 
michael@0: // (derived from)
michael@0: // file_id.cc: Return a unique identifier for a file
michael@0: //
michael@0: // See file_id.h for documentation
michael@0: //
michael@0: 
michael@0: // ELF note name and desc are 32-bits word padded.
michael@0: #define NOTE_PADDING(a) ((a + 3) & ~3)
michael@0: 
michael@0: // These functions are also used inside the crashed process, so be safe
michael@0: // and use the syscall/libc wrappers instead of direct syscalls or libc.
michael@0: 
michael@0: template<typename ElfClass>
michael@0: static bool ElfClassBuildIDNoteIdentifier(const void *section, int length,
michael@0:                                           uint8_t identifier[kMDGUIDSize]) {
michael@0:   typedef typename ElfClass::Nhdr Nhdr;
michael@0: 
michael@0:   const void* section_end = reinterpret_cast<const char*>(section) + length;
michael@0:   const Nhdr* note_header = reinterpret_cast<const Nhdr*>(section);
michael@0:   while (reinterpret_cast<const void *>(note_header) < section_end) {
michael@0:     if (note_header->n_type == NT_GNU_BUILD_ID)
michael@0:       break;
michael@0:     note_header = reinterpret_cast<const Nhdr*>(
michael@0:                   reinterpret_cast<const char*>(note_header) + sizeof(Nhdr) +
michael@0:                   NOTE_PADDING(note_header->n_namesz) +
michael@0:                   NOTE_PADDING(note_header->n_descsz));
michael@0:   }
michael@0:   if (reinterpret_cast<const void *>(note_header) >= section_end ||
michael@0:       note_header->n_descsz == 0) {
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   const char* build_id = reinterpret_cast<const char*>(note_header) +
michael@0:     sizeof(Nhdr) + NOTE_PADDING(note_header->n_namesz);
michael@0:   // Copy as many bits of the build ID as will fit
michael@0:   // into the GUID space.
michael@0:   memset(identifier, 0, kMDGUIDSize);
michael@0:   memcpy(identifier, build_id,
michael@0:          std::min(kMDGUIDSize, (size_t)note_header->n_descsz));
michael@0: 
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: // Attempt to locate a .note.gnu.build-id section in an ELF binary
michael@0: // and copy as many bytes of it as will fit into |identifier|.
michael@0: static bool FindElfBuildIDNote(const void *elf_mapped_base,
michael@0:                                uint8_t identifier[kMDGUIDSize]) {
michael@0:   void* note_section;
michael@0:   int note_size, elfclass;
michael@0:   if ((!FindElfSegment(elf_mapped_base, PT_NOTE,
michael@0:                        (const void**)&note_section, &note_size, &elfclass) ||
michael@0:       note_size == 0)  &&
michael@0:       (!FindElfSection(elf_mapped_base, ".note.gnu.build-id", SHT_NOTE,
michael@0:                        (const void**)&note_section, &note_size, &elfclass) ||
michael@0:       note_size == 0)) {
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   if (elfclass == ELFCLASS32) {
michael@0:     return ElfClassBuildIDNoteIdentifier<ElfClass32>(note_section, note_size,
michael@0:                                                      identifier);
michael@0:   } else if (elfclass == ELFCLASS64) {
michael@0:     return ElfClassBuildIDNoteIdentifier<ElfClass64>(note_section, note_size,
michael@0:                                                      identifier);
michael@0:   }
michael@0: 
michael@0:   return false;
michael@0: }
michael@0: 
michael@0: // Attempt to locate the .text section of an ELF binary and generate
michael@0: // a simple hash by XORing the first page worth of bytes into |identifier|.
michael@0: static bool HashElfTextSection(const void *elf_mapped_base,
michael@0:                                uint8_t identifier[kMDGUIDSize]) {
michael@0:   void* text_section;
michael@0:   int text_size;
michael@0:   if (!FindElfSection(elf_mapped_base, ".text", SHT_PROGBITS,
michael@0:                       (const void**)&text_section, &text_size, NULL) ||
michael@0:       text_size == 0) {
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   memset(identifier, 0, kMDGUIDSize);
michael@0:   const uint8_t* ptr = reinterpret_cast<const uint8_t*>(text_section);
michael@0:   const uint8_t* ptr_end = ptr + std::min(text_size, 4096);
michael@0:   while (ptr < ptr_end) {
michael@0:     for (unsigned i = 0; i < kMDGUIDSize; i++)
michael@0:       identifier[i] ^= ptr[i];
michael@0:     ptr += kMDGUIDSize;
michael@0:   }
michael@0:   return true;
michael@0: }
michael@0: 
michael@0: // static
michael@0: bool FileID::ElfFileIdentifierFromMappedFile(const void* base,
michael@0:                                              uint8_t identifier[kMDGUIDSize]) {
michael@0:   // Look for a build id note first.
michael@0:   if (FindElfBuildIDNote(base, identifier))
michael@0:     return true;
michael@0: 
michael@0:   // Fall back on hashing the first page of the text section.
michael@0:   return HashElfTextSection(base, identifier);
michael@0: }
michael@0: 
michael@0: // static
michael@0: void FileID::ConvertIdentifierToString(const uint8_t identifier[kMDGUIDSize],
michael@0:                                        char* buffer, int buffer_length) {
michael@0:   uint8_t identifier_swapped[kMDGUIDSize];
michael@0: 
michael@0:   // Endian-ness swap to match dump processor expectation.
michael@0:   memcpy(identifier_swapped, identifier, kMDGUIDSize);
michael@0:   uint32_t* data1 = reinterpret_cast<uint32_t*>(identifier_swapped);
michael@0:   *data1 = htonl(*data1);
michael@0:   uint16_t* data2 = reinterpret_cast<uint16_t*>(identifier_swapped + 4);
michael@0:   *data2 = htons(*data2);
michael@0:   uint16_t* data3 = reinterpret_cast<uint16_t*>(identifier_swapped + 6);
michael@0:   *data3 = htons(*data3);
michael@0: 
michael@0:   int buffer_idx = 0;
michael@0:   for (unsigned int idx = 0;
michael@0:        (buffer_idx < buffer_length) && (idx < kMDGUIDSize);
michael@0:        ++idx) {
michael@0:     int hi = (identifier_swapped[idx] >> 4) & 0x0F;
michael@0:     int lo = (identifier_swapped[idx]) & 0x0F;
michael@0: 
michael@0:     if (idx == 4 || idx == 6 || idx == 8 || idx == 10)
michael@0:       buffer[buffer_idx++] = '-';
michael@0: 
michael@0:     buffer[buffer_idx++] = (hi >= 10) ? 'A' + hi - 10 : '0' + hi;
michael@0:     buffer[buffer_idx++] = (lo >= 10) ? 'A' + lo - 10 : '0' + lo;
michael@0:   }
michael@0: 
michael@0:   // NULL terminate
michael@0:   buffer[(buffer_idx < buffer_length) ? buffer_idx : buffer_idx - 1] = 0;
michael@0: }
michael@0: 
michael@0: }  // namespace lul