blob: b7788ca01a30eb7460716ead339c4fce6d7d98a0 [file] [log] [blame]
// Copyright (c) 2006, Google Inc.
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
// * Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following disclaimer
// in the documentation and/or other materials provided with the
// distribution.
// * Neither the name of Google Inc. nor the names of its
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "common/windows/pdb_source_line_writer.h"
#include <windows.h>
#include <winnt.h>
#include <atlbase.h>
#include <dia2.h>
#include <diacreate.h>
#include <ImageHlp.h>
#include <stdio.h>
#include <algorithm>
#include <limits>
#include <map>
#include <set>
#include <utility>
#include "common/windows/dia_util.h"
#include "common/windows/guid_string.h"
#include "common/windows/string_utils-inl.h"
// This constant may be missing from DbgHelp.h. See the documentation for
// IDiaSymbol::get_undecoratedNameEx.
#ifndef UNDNAME_NO_ECSU
#define UNDNAME_NO_ECSU 0x8000 // Suppresses enum/class/struct/union.
#endif // UNDNAME_NO_ECSU
/*
* Not defined in WinNT.h for some reason. Definitions taken from:
* http://uninformed.org/index.cgi?v=4&a=1&p=13
*
*/
typedef unsigned char UBYTE;
#if !defined(_WIN64)
#define UNW_FLAG_EHANDLER 0x01
#define UNW_FLAG_UHANDLER 0x02
#define UNW_FLAG_CHAININFO 0x04
#endif
union UnwindCode {
struct {
UBYTE offset_in_prolog;
UBYTE unwind_operation_code : 4;
UBYTE operation_info : 4;
};
USHORT frame_offset;
};
enum UnwindOperationCodes {
UWOP_PUSH_NONVOL = 0, /* info == register number */
UWOP_ALLOC_LARGE, /* no info, alloc size in next 2 slots */
UWOP_ALLOC_SMALL, /* info == size of allocation / 8 - 1 */
UWOP_SET_FPREG, /* no info, FP = RSP + UNWIND_INFO.FPRegOffset*16 */
UWOP_SAVE_NONVOL, /* info == register number, offset in next slot */
UWOP_SAVE_NONVOL_FAR, /* info == register number, offset in next 2 slots */
// XXX: these are missing from MSDN!
// See: http://www.osronline.com/ddkx/kmarch/64bitamd_4rs7.htm
UWOP_SAVE_XMM,
UWOP_SAVE_XMM_FAR,
UWOP_SAVE_XMM128, /* info == XMM reg number, offset in next slot */
UWOP_SAVE_XMM128_FAR, /* info == XMM reg number, offset in next 2 slots */
UWOP_PUSH_MACHFRAME /* info == 0: no error-code, 1: error-code */
};
// See: http://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
// Note: some fields removed as we don't use them.
struct UnwindInfo {
UBYTE version : 3;
UBYTE flags : 5;
UBYTE size_of_prolog;
UBYTE count_of_codes;
UBYTE frame_register : 4;
UBYTE frame_offset : 4;
UnwindCode unwind_code[1];
};
namespace google_breakpad {
namespace {
using std::vector;
// The symbol (among possibly many) selected to represent an rva.
struct SelectedSymbol {
SelectedSymbol(const CComPtr<IDiaSymbol>& symbol, bool is_public)
: symbol(symbol), is_public(is_public), is_multiple(false) {}
// The symbol to use for an rva.
CComPtr<IDiaSymbol> symbol;
// Whether this is a public or function symbol.
bool is_public;
// Whether the rva has multiple associated symbols. An rva will correspond to
// multiple symbols in the case of linker identical symbol folding.
bool is_multiple;
};
// Maps rva to the symbol to use for that address.
typedef std::map<DWORD, SelectedSymbol> SymbolMap;
// Record this in the map as the selected symbol for the rva if it satisfies the
// necessary conditions.
void MaybeRecordSymbol(DWORD rva,
const CComPtr<IDiaSymbol> symbol,
bool is_public,
SymbolMap* map) {
SymbolMap::iterator loc = map->find(rva);
if (loc == map->end()) {
map->insert(std::make_pair(rva, SelectedSymbol(symbol, is_public)));
return;
}
// Prefer function symbols to public symbols.
if (is_public && !loc->second.is_public) {
return;
}
loc->second.is_multiple = true;
// Take the 'least' symbol by lexicographical order of the decorated name. We
// use the decorated rather than undecorated name because computing the latter
// is expensive.
BSTR current_name, new_name;
loc->second.symbol->get_name(&current_name);
symbol->get_name(&new_name);
if (wcscmp(new_name, current_name) < 0) {
loc->second.symbol = symbol;
loc->second.is_public = is_public;
}
}
// A helper class to scope a PLOADED_IMAGE.
class AutoImage {
public:
explicit AutoImage(PLOADED_IMAGE img) : img_(img) {}
~AutoImage() {
if (img_)
ImageUnload(img_);
}
operator PLOADED_IMAGE() { return img_; }
PLOADED_IMAGE operator->() { return img_; }
private:
PLOADED_IMAGE img_;
};
bool SymbolsMatch(IDiaSymbol* a, IDiaSymbol* b) {
DWORD a_section, a_offset, b_section, b_offset;
if (FAILED(a->get_addressSection(&a_section)) ||
FAILED(a->get_addressOffset(&a_offset)) ||
FAILED(b->get_addressSection(&b_section)) ||
FAILED(b->get_addressOffset(&b_offset)))
return false;
return a_section == b_section && a_offset == b_offset;
}
bool CreateDiaDataSourceInstance(CComPtr<IDiaDataSource> &data_source) {
if (SUCCEEDED(data_source.CoCreateInstance(CLSID_DiaSource))) {
return true;
}
class DECLSPEC_UUID("B86AE24D-BF2F-4ac9-B5A2-34B14E4CE11D") DiaSource100;
class DECLSPEC_UUID("761D3BCD-1304-41D5-94E8-EAC54E4AC172") DiaSource110;
class DECLSPEC_UUID("3BFCEA48-620F-4B6B-81F7-B9AF75454C7D") DiaSource120;
class DECLSPEC_UUID("E6756135-1E65-4D17-8576-610761398C3C") DiaSource140;
// If the CoCreateInstance call above failed, msdia*.dll is not registered.
// We can try loading the DLL corresponding to the #included DIA SDK, but
// the DIA headers don't provide a version. Lets try to figure out which DIA
// version we're compiling against by comparing CLSIDs.
const wchar_t *msdia_dll = nullptr;
if (CLSID_DiaSource == _uuidof(DiaSource100)) {
msdia_dll = L"msdia100.dll";
} else if (CLSID_DiaSource == _uuidof(DiaSource110)) {
msdia_dll = L"msdia110.dll";
} else if (CLSID_DiaSource == _uuidof(DiaSource120)) {
msdia_dll = L"msdia120.dll";
} else if (CLSID_DiaSource == _uuidof(DiaSource140)) {
msdia_dll = L"msdia140.dll";
}
if (msdia_dll &&
SUCCEEDED(NoRegCoCreate(msdia_dll, CLSID_DiaSource, IID_IDiaDataSource,
reinterpret_cast<void **>(&data_source)))) {
return true;
}
return false;
}
} // namespace
PDBSourceLineWriter::PDBSourceLineWriter() : output_(NULL) {
}
PDBSourceLineWriter::~PDBSourceLineWriter() {
}
bool PDBSourceLineWriter::SetCodeFile(const wstring &exe_file) {
if (code_file_.empty()) {
code_file_ = exe_file;
return true;
}
// Setting a different code file path is an error. It is success only if the
// file paths are the same.
return exe_file == code_file_;
}
bool PDBSourceLineWriter::Open(const wstring &file, FileFormat format) {
Close();
code_file_.clear();
if (FAILED(CoInitialize(NULL))) {
fprintf(stderr, "CoInitialize failed\n");
return false;
}
CComPtr<IDiaDataSource> data_source;
if (!CreateDiaDataSourceInstance(data_source)) {
const int kGuidSize = 64;
wchar_t classid[kGuidSize] = {0};
StringFromGUID2(CLSID_DiaSource, classid, kGuidSize);
fprintf(stderr, "CoCreateInstance CLSID_DiaSource %S failed "
"(msdia*.dll unregistered?)\n", classid);
return false;
}
switch (format) {
case PDB_FILE:
if (FAILED(data_source->loadDataFromPdb(file.c_str()))) {
fprintf(stderr, "loadDataFromPdb failed for %ws\n", file.c_str());
return false;
}
break;
case EXE_FILE:
if (FAILED(data_source->loadDataForExe(file.c_str(), NULL, NULL))) {
fprintf(stderr, "loadDataForExe failed for %ws\n", file.c_str());
return false;
}
code_file_ = file;
break;
case ANY_FILE:
if (FAILED(data_source->loadDataFromPdb(file.c_str()))) {
if (FAILED(data_source->loadDataForExe(file.c_str(), NULL, NULL))) {
fprintf(stderr, "loadDataForPdb and loadDataFromExe failed for %ws\n",
file.c_str());
return false;
}
code_file_ = file;
}
break;
default:
fprintf(stderr, "Unknown file format\n");
return false;
}
if (FAILED(data_source->openSession(&session_))) {
fprintf(stderr, "openSession failed\n");
}
return true;
}
bool PDBSourceLineWriter::PrintLines(IDiaEnumLineNumbers *lines) {
// The line number format is:
// <rva> <line number> <source file id>
CComPtr<IDiaLineNumber> line;
ULONG count;
while (SUCCEEDED(lines->Next(1, &line, &count)) && count == 1) {
DWORD rva;
if (FAILED(line->get_relativeVirtualAddress(&rva))) {
fprintf(stderr, "failed to get line rva\n");
return false;
}
DWORD length;
if (FAILED(line->get_length(&length))) {
fprintf(stderr, "failed to get line code length\n");
return false;
}
DWORD dia_source_id;
if (FAILED(line->get_sourceFileId(&dia_source_id))) {
fprintf(stderr, "failed to get line source file id\n");
return false;
}
// duplicate file names are coalesced to share one ID
DWORD source_id = GetRealFileID(dia_source_id);
DWORD line_num;
if (FAILED(line->get_lineNumber(&line_num))) {
fprintf(stderr, "failed to get line number\n");
return false;
}
AddressRangeVector ranges;
MapAddressRange(image_map_, AddressRange(rva, length), &ranges);
for (size_t i = 0; i < ranges.size(); ++i) {
fprintf(output_, "%lx %lx %lu %lu\n", ranges[i].rva, ranges[i].length,
line_num, source_id);
}
line.Release();
}
return true;
}
bool PDBSourceLineWriter::PrintFunction(IDiaSymbol *function,
IDiaSymbol *block,
bool has_multiple_symbols) {
// The function format is:
// FUNC <address> <length> <param_stack_size> <function>
DWORD rva;
if (FAILED(block->get_relativeVirtualAddress(&rva))) {
fprintf(stderr, "couldn't get rva\n");
return false;
}
ULONGLONG length;
if (FAILED(block->get_length(&length))) {
fprintf(stderr, "failed to get function length\n");
return false;
}
if (length == 0) {
// Silently ignore zero-length functions, which can infrequently pop up.
return true;
}
CComBSTR name;
int stack_param_size;
if (!GetSymbolFunctionName(function, &name, &stack_param_size)) {
return false;
}
// If the decorated name didn't give the parameter size, try to
// calculate it.
if (stack_param_size < 0) {
stack_param_size = GetFunctionStackParamSize(function);
}
AddressRangeVector ranges;
MapAddressRange(image_map_, AddressRange(rva, static_cast<DWORD>(length)),
&ranges);
for (size_t i = 0; i < ranges.size(); ++i) {
const char* optional_multiple_field = has_multiple_symbols ? "m " : "";
fprintf(output_, "FUNC %s%lx %lx %x %ws\n", optional_multiple_field,
ranges[i].rva, ranges[i].length, stack_param_size, name.m_str);
}
CComPtr<IDiaEnumLineNumbers> lines;
if (FAILED(session_->findLinesByRVA(rva, DWORD(length), &lines))) {
return false;
}
if (!PrintLines(lines)) {
return false;
}
return true;
}
bool PDBSourceLineWriter::PrintSourceFiles() {
CComPtr<IDiaSymbol> global;
if (FAILED(session_->get_globalScope(&global))) {
fprintf(stderr, "get_globalScope failed\n");
return false;
}
CComPtr<IDiaEnumSymbols> compilands;
if (FAILED(global->findChildren(SymTagCompiland, NULL,
nsNone, &compilands))) {
fprintf(stderr, "findChildren failed\n");
return false;
}
CComPtr<IDiaSymbol> compiland;
ULONG count;
while (SUCCEEDED(compilands->Next(1, &compiland, &count)) && count == 1) {
CComPtr<IDiaEnumSourceFiles> source_files;
if (FAILED(session_->findFile(compiland, NULL, nsNone, &source_files))) {
return false;
}
CComPtr<IDiaSourceFile> file;
while (SUCCEEDED(source_files->Next(1, &file, &count)) && count == 1) {
DWORD file_id;
if (FAILED(file->get_uniqueId(&file_id))) {
return false;
}
CComBSTR file_name;
if (FAILED(file->get_fileName(&file_name))) {
return false;
}
wstring file_name_string(file_name);
if (!FileIDIsCached(file_name_string)) {
// this is a new file name, cache it and output a FILE line.
CacheFileID(file_name_string, file_id);
fwprintf(output_, L"FILE %d %ws\n", file_id, file_name_string.c_str());
} else {
// this file name has already been seen, just save this
// ID for later lookup.
StoreDuplicateFileID(file_name_string, file_id);
}
file.Release();
}
compiland.Release();
}
return true;
}
bool PDBSourceLineWriter::PrintFunctions() {
ULONG count = 0;
DWORD rva = 0;
CComPtr<IDiaSymbol> global;
HRESULT hr;
if (FAILED(session_->get_globalScope(&global))) {
fprintf(stderr, "get_globalScope failed\n");
return false;
}
CComPtr<IDiaEnumSymbols> symbols = NULL;
// Find all function symbols first.
SymbolMap rva_symbol;
hr = global->findChildren(SymTagFunction, NULL, nsNone, &symbols);
if (SUCCEEDED(hr)) {
CComPtr<IDiaSymbol> symbol = NULL;
while (SUCCEEDED(symbols->Next(1, &symbol, &count)) && count == 1) {
if (SUCCEEDED(symbol->get_relativeVirtualAddress(&rva))) {
// Potentially record this as the canonical symbol for this rva.
MaybeRecordSymbol(rva, symbol, false, &rva_symbol);
} else {
fprintf(stderr, "get_relativeVirtualAddress failed on the symbol\n");
return false;
}
symbol.Release();
}
symbols.Release();
}
// Find all public symbols and record public symbols that are not also private
// symbols.
hr = global->findChildren(SymTagPublicSymbol, NULL, nsNone, &symbols);
if (SUCCEEDED(hr)) {
CComPtr<IDiaSymbol> symbol = NULL;
while (SUCCEEDED(symbols->Next(1, &symbol, &count)) && count == 1) {
if (SUCCEEDED(symbol->get_relativeVirtualAddress(&rva))) {
// Potentially record this as the canonical symbol for this rva.
MaybeRecordSymbol(rva, symbol, true, &rva_symbol);
} else {
fprintf(stderr, "get_relativeVirtualAddress failed on the symbol\n");
return false;
}
symbol.Release();
}
symbols.Release();
}
// For each rva, dump the selected symbol at the address.
SymbolMap::iterator it;
for (it = rva_symbol.begin(); it != rva_symbol.end(); ++it) {
CComPtr<IDiaSymbol> symbol = it->second.symbol;
// Only print public symbols if there is no function symbol for the address.
if (!it->second.is_public) {
if (!PrintFunction(symbol, symbol, it->second.is_multiple))
return false;
} else {
if (!PrintCodePublicSymbol(symbol, it->second.is_multiple))
return false;
}
}
// When building with PGO, the compiler can split functions into
// "hot" and "cold" blocks, and move the "cold" blocks out to separate
// pages, so the function can be noncontiguous. To find these blocks,
// we have to iterate over all the compilands, and then find blocks
// that are children of them. We can then find the lexical parents
// of those blocks and print out an extra FUNC line for blocks
// that are not contained in their parent functions.
CComPtr<IDiaEnumSymbols> compilands;
if (FAILED(global->findChildren(SymTagCompiland, NULL,
nsNone, &compilands))) {
fprintf(stderr, "findChildren failed on the global\n");
return false;
}
CComPtr<IDiaSymbol> compiland;
while (SUCCEEDED(compilands->Next(1, &compiland, &count)) && count == 1) {
CComPtr<IDiaEnumSymbols> blocks;
if (FAILED(compiland->findChildren(SymTagBlock, NULL,
nsNone, &blocks))) {
fprintf(stderr, "findChildren failed on a compiland\n");
return false;
}
CComPtr<IDiaSymbol> block;
while (SUCCEEDED(blocks->Next(1, &block, &count)) && count == 1) {
// find this block's lexical parent function
CComPtr<IDiaSymbol> parent;
DWORD tag;
if (SUCCEEDED(block->get_lexicalParent(&parent)) &&
SUCCEEDED(parent->get_symTag(&tag)) &&
tag == SymTagFunction) {
// now get the block's offset and the function's offset and size,
// and determine if the block is outside of the function
DWORD func_rva, block_rva;
ULONGLONG func_length;
if (SUCCEEDED(block->get_relativeVirtualAddress(&block_rva)) &&
SUCCEEDED(parent->get_relativeVirtualAddress(&func_rva)) &&
SUCCEEDED(parent->get_length(&func_length))) {
if (block_rva < func_rva || block_rva > (func_rva + func_length)) {
if (!PrintFunction(parent, block, false)) {
return false;
}
}
}
}
parent.Release();
block.Release();
}
blocks.Release();
compiland.Release();
}
global.Release();
return true;
}
#undef max
bool PDBSourceLineWriter::PrintFrameDataUsingPDB() {
// It would be nice if it were possible to output frame data alongside the
// associated function, as is done with line numbers, but the DIA API
// doesn't make it possible to get the frame data in that way.
CComPtr<IDiaEnumFrameData> frame_data_enum;
if (!FindTable(session_, &frame_data_enum))
return false;
DWORD last_type = std::numeric_limits<DWORD>::max();
DWORD last_rva = std::numeric_limits<DWORD>::max();
DWORD last_code_size = 0;
DWORD last_prolog_size = std::numeric_limits<DWORD>::max();
CComPtr<IDiaFrameData> frame_data;
ULONG count = 0;
while (SUCCEEDED(frame_data_enum->Next(1, &frame_data, &count)) &&
count == 1) {
DWORD type;
if (FAILED(frame_data->get_type(&type)))
return false;
DWORD rva;
if (FAILED(frame_data->get_relativeVirtualAddress(&rva)))
return false;
DWORD code_size;
if (FAILED(frame_data->get_lengthBlock(&code_size)))
return false;
DWORD prolog_size;
if (FAILED(frame_data->get_lengthProlog(&prolog_size)))
return false;
// parameter_size is the size of parameters passed on the stack. If any
// parameters are not passed on the stack (such as in registers), their
// sizes will not be included in parameter_size.
DWORD parameter_size;
if (FAILED(frame_data->get_lengthParams(&parameter_size)))
return false;
DWORD saved_register_size;
if (FAILED(frame_data->get_lengthSavedRegisters(&saved_register_size)))
return false;
DWORD local_size;
if (FAILED(frame_data->get_lengthLocals(&local_size)))
return false;
// get_maxStack can return S_FALSE, just use 0 in that case.
DWORD max_stack_size = 0;
if (FAILED(frame_data->get_maxStack(&max_stack_size)))
return false;
// get_programString can return S_FALSE, indicating that there is no
// program string. In that case, check whether %ebp is used.
HRESULT program_string_result;
CComBSTR program_string;
if (FAILED(program_string_result = frame_data->get_program(
&program_string))) {
return false;
}
// get_allocatesBasePointer can return S_FALSE, treat that as though
// %ebp is not used.
BOOL allocates_base_pointer = FALSE;
if (program_string_result != S_OK) {
if (FAILED(frame_data->get_allocatesBasePointer(
&allocates_base_pointer))) {
return false;
}
}
// Only print out a line if type, rva, code_size, or prolog_size have
// changed from the last line. It is surprisingly common (especially in
// system library PDBs) for DIA to return a series of identical
// IDiaFrameData objects. For kernel32.pdb from Windows XP SP2 on x86,
// this check reduces the size of the dumped symbol file by a third.
if (type != last_type || rva != last_rva || code_size != last_code_size ||
prolog_size != last_prolog_size) {
// The prolog and the code portions of the frame have to be treated
// independently as they may have independently changed in size, or may
// even have been split.
// NOTE: If epilog size is ever non-zero, we have to do something
// similar with it.
// Figure out where the prolog bytes have landed.
AddressRangeVector prolog_ranges;
if (prolog_size > 0) {
MapAddressRange(image_map_, AddressRange(rva, prolog_size),
&prolog_ranges);
}
// And figure out where the code bytes have landed.
AddressRangeVector code_ranges;
MapAddressRange(image_map_,
AddressRange(rva + prolog_size,
code_size - prolog_size),
&code_ranges);
struct FrameInfo {
DWORD rva;
DWORD code_size;
DWORD prolog_size;
};
std::vector<FrameInfo> frame_infos;
// Special case: The prolog and the code bytes remain contiguous. This is
// only done for compactness of the symbol file, and we could actually
// be outputting independent frame info for the prolog and code portions.
if (prolog_ranges.size() == 1 && code_ranges.size() == 1 &&
prolog_ranges[0].end() == code_ranges[0].rva) {
FrameInfo fi = { prolog_ranges[0].rva,
prolog_ranges[0].length + code_ranges[0].length,
prolog_ranges[0].length };
frame_infos.push_back(fi);
} else {
// Otherwise we output the prolog and code frame info independently.
for (size_t i = 0; i < prolog_ranges.size(); ++i) {
FrameInfo fi = { prolog_ranges[i].rva,
prolog_ranges[i].length,
prolog_ranges[i].length };
frame_infos.push_back(fi);
}
for (size_t i = 0; i < code_ranges.size(); ++i) {
FrameInfo fi = { code_ranges[i].rva, code_ranges[i].length, 0 };
frame_infos.push_back(fi);
}
}
for (size_t i = 0; i < frame_infos.size(); ++i) {
const FrameInfo& fi(frame_infos[i]);
fprintf(output_, "STACK WIN %lx %lx %lx %lx %x %lx %lx %lx %lx %d ",
type, fi.rva, fi.code_size, fi.prolog_size,
0 /* epilog_size */, parameter_size, saved_register_size,
local_size, max_stack_size, program_string_result == S_OK);
if (program_string_result == S_OK) {
fprintf(output_, "%ws\n", program_string.m_str);
} else {
fprintf(output_, "%d\n", allocates_base_pointer);
}
}
last_type = type;
last_rva = rva;
last_code_size = code_size;
last_prolog_size = prolog_size;
}
frame_data.Release();
}
return true;
}
bool PDBSourceLineWriter::PrintFrameDataUsingEXE() {
if (code_file_.empty() && !FindPEFile()) {
fprintf(stderr, "Couldn't locate EXE or DLL file.\n");
return false;
}
// Convert wchar to native charset because ImageLoad only takes
// a PSTR as input.
string code_file;
if (!WindowsStringUtils::safe_wcstombs(code_file_, &code_file)) {
return false;
}
AutoImage img(ImageLoad((PSTR)code_file.c_str(), NULL));
if (!img) {
fprintf(stderr, "Failed to load %s\n", code_file.c_str());
return false;
}
PIMAGE_OPTIONAL_HEADER64 optional_header =
&(reinterpret_cast<PIMAGE_NT_HEADERS64>(img->FileHeader))->OptionalHeader;
if (optional_header->Magic != IMAGE_NT_OPTIONAL_HDR64_MAGIC) {
fprintf(stderr, "Not a PE32+ image\n");
return false;
}
// Read Exception Directory
DWORD exception_rva = optional_header->
DataDirectory[IMAGE_DIRECTORY_ENTRY_EXCEPTION].VirtualAddress;
DWORD exception_size = optional_header->
DataDirectory[IMAGE_DIRECTORY_ENTRY_EXCEPTION].Size;
PIMAGE_RUNTIME_FUNCTION_ENTRY funcs =
static_cast<PIMAGE_RUNTIME_FUNCTION_ENTRY>(
ImageRvaToVa(img->FileHeader,
img->MappedAddress,
exception_rva,
&img->LastRvaSection));
for (DWORD i = 0; i < exception_size / sizeof(*funcs); i++) {
DWORD unwind_rva = funcs[i].UnwindInfoAddress;
// handle chaining
while (unwind_rva & 0x1) {
unwind_rva ^= 0x1;
PIMAGE_RUNTIME_FUNCTION_ENTRY chained_func =
static_cast<PIMAGE_RUNTIME_FUNCTION_ENTRY>(
ImageRvaToVa(img->FileHeader,
img->MappedAddress,
unwind_rva,
&img->LastRvaSection));
unwind_rva = chained_func->UnwindInfoAddress;
}
UnwindInfo *unwind_info = static_cast<UnwindInfo *>(
ImageRvaToVa(img->FileHeader,
img->MappedAddress,
unwind_rva,
&img->LastRvaSection));
DWORD stack_size = 8; // minimal stack size is 8 for RIP
DWORD rip_offset = 8;
do {
for (UBYTE c = 0; c < unwind_info->count_of_codes; c++) {
UnwindCode *unwind_code = &unwind_info->unwind_code[c];
switch (unwind_code->unwind_operation_code) {
case UWOP_PUSH_NONVOL: {
stack_size += 8;
break;
}
case UWOP_ALLOC_LARGE: {
if (unwind_code->operation_info == 0) {
c++;
if (c < unwind_info->count_of_codes)
stack_size += (unwind_code + 1)->frame_offset * 8;
} else {
c += 2;
if (c < unwind_info->count_of_codes)
stack_size += (unwind_code + 1)->frame_offset |
((unwind_code + 2)->frame_offset << 16);
}
break;
}
case UWOP_ALLOC_SMALL: {
stack_size += unwind_code->operation_info * 8 + 8;
break;
}
case UWOP_SET_FPREG:
case UWOP_SAVE_XMM:
case UWOP_SAVE_XMM_FAR:
break;
case UWOP_SAVE_NONVOL:
case UWOP_SAVE_XMM128: {
c++; // skip slot with offset
break;
}
case UWOP_SAVE_NONVOL_FAR:
case UWOP_SAVE_XMM128_FAR: {
c += 2; // skip 2 slots with offset
break;
}
case UWOP_PUSH_MACHFRAME: {
if (unwind_code->operation_info) {
stack_size += 88;
} else {
stack_size += 80;
}
rip_offset += 80;
break;
}
}
}
if (unwind_info->flags & UNW_FLAG_CHAININFO) {
PIMAGE_RUNTIME_FUNCTION_ENTRY chained_func =
reinterpret_cast<PIMAGE_RUNTIME_FUNCTION_ENTRY>(
(unwind_info->unwind_code +
((unwind_info->count_of_codes + 1) & ~1)));
unwind_info = static_cast<UnwindInfo *>(
ImageRvaToVa(img->FileHeader,
img->MappedAddress,
chained_func->UnwindInfoAddress,
&img->LastRvaSection));
} else {
unwind_info = NULL;
}
} while (unwind_info);
fprintf(output_, "STACK CFI INIT %lx %lx .cfa: $rsp .ra: .cfa %lu - ^\n",
funcs[i].BeginAddress,
funcs[i].EndAddress - funcs[i].BeginAddress, rip_offset);
fprintf(output_, "STACK CFI %lx .cfa: $rsp %lu +\n",
funcs[i].BeginAddress, stack_size);
}
return true;
}
bool PDBSourceLineWriter::PrintFrameData() {
PDBModuleInfo info;
if (GetModuleInfo(&info) && info.cpu == L"x86_64") {
return PrintFrameDataUsingEXE();
} else {
return PrintFrameDataUsingPDB();
}
return false;
}
bool PDBSourceLineWriter::PrintCodePublicSymbol(IDiaSymbol *symbol,
bool has_multiple_symbols) {
BOOL is_code;
if (FAILED(symbol->get_code(&is_code))) {
return false;
}
if (!is_code) {
return true;
}
DWORD rva;
if (FAILED(symbol->get_relativeVirtualAddress(&rva))) {
return false;
}
CComBSTR name;
int stack_param_size;
if (!GetSymbolFunctionName(symbol, &name, &stack_param_size)) {
return false;
}
AddressRangeVector ranges;
MapAddressRange(image_map_, AddressRange(rva, 1), &ranges);
for (size_t i = 0; i < ranges.size(); ++i) {
const char* optional_multiple_field = has_multiple_symbols ? "m " : "";
fprintf(output_, "PUBLIC %s%lx %x %ws\n", optional_multiple_field,
ranges[i].rva, stack_param_size > 0 ? stack_param_size : 0,
name.m_str);
}
// Now walk the function in the original untranslated space, asking DIA
// what function is at that location, stepping through OMAP blocks. If
// we're still in the same function, emit another entry, because the
// symbol could have been split into multiple pieces. If we've gotten to
// another symbol in the original address space, then we're done for
// this symbol. See https://crbug.com/678874.
for (;;) {
// This steps to the next block in the original image. Simply doing
// rva++ would also be correct, but would emit tons of unnecessary
// entries.
rva = image_map_.subsequent_rva_block[rva];
if (rva == 0)
break;
CComPtr<IDiaSymbol> next_sym = NULL;
LONG displacement;
if (FAILED(session_->findSymbolByRVAEx(rva, SymTagPublicSymbol, &next_sym,
&displacement))) {
break;
}
if (!SymbolsMatch(symbol, next_sym))
break;
AddressRangeVector next_ranges;
MapAddressRange(image_map_, AddressRange(rva, 1), &next_ranges);
for (size_t i = 0; i < next_ranges.size(); ++i) {
fprintf(output_, "PUBLIC %lx %x %ws\n", next_ranges[i].rva,
stack_param_size > 0 ? stack_param_size : 0, name.m_str);
}
}
return true;
}
bool PDBSourceLineWriter::PrintPDBInfo() {
PDBModuleInfo info;
if (!GetModuleInfo(&info)) {
return false;
}
// Hard-code "windows" for the OS because that's the only thing that makes
// sense for PDB files. (This might not be strictly correct for Windows CE
// support, but we don't care about that at the moment.)
fprintf(output_, "MODULE windows %ws %ws %ws\n",
info.cpu.c_str(), info.debug_identifier.c_str(),
info.debug_file.c_str());
return true;
}
bool PDBSourceLineWriter::PrintPEInfo() {
PEModuleInfo info;
if (!GetPEInfo(&info)) {
return false;
}
fprintf(output_, "INFO CODE_ID %ws %ws\n",
info.code_identifier.c_str(),
info.code_file.c_str());
return true;
}
// wcstol_positive_strict is sort of like wcstol, but much stricter. string
// should be a buffer pointing to a null-terminated string containing only
// decimal digits. If the entire string can be converted to an integer
// without overflowing, and there are no non-digit characters before the
// result is set to the value and this function returns true. Otherwise,
// this function returns false. This is an alternative to the strtol, atoi,
// and scanf families, which are not as strict about input and in some cases
// don't provide a good way for the caller to determine if a conversion was
// successful.
static bool wcstol_positive_strict(wchar_t *string, int *result) {
int value = 0;
for (wchar_t *c = string; *c != '\0'; ++c) {
int last_value = value;
value *= 10;
// Detect overflow.
if (value / 10 != last_value || value < 0) {
return false;
}
if (*c < '0' || *c > '9') {
return false;
}
unsigned int c_value = *c - '0';
last_value = value;
value += c_value;
// Detect overflow.
if (value < last_value) {
return false;
}
// Forbid leading zeroes unless the string is just "0".
if (value == 0 && *(c+1) != '\0') {
return false;
}
}
*result = value;
return true;
}
bool PDBSourceLineWriter::FindPEFile() {
CComPtr<IDiaSymbol> global;
if (FAILED(session_->get_globalScope(&global))) {
fprintf(stderr, "get_globalScope failed\n");
return false;
}
CComBSTR symbols_file;
if (SUCCEEDED(global->get_symbolsFileName(&symbols_file))) {
wstring file(symbols_file);
// Look for an EXE or DLL file.
const wchar_t *extensions[] = { L"exe", L"dll" };
for (size_t i = 0; i < sizeof(extensions) / sizeof(extensions[0]); i++) {
size_t dot_pos = file.find_last_of(L".");
if (dot_pos != wstring::npos) {
file.replace(dot_pos + 1, wstring::npos, extensions[i]);
// Check if this file exists.
if (GetFileAttributesW(file.c_str()) != INVALID_FILE_ATTRIBUTES) {
code_file_ = file;
return true;
}
}
}
}
return false;
}
// static
bool PDBSourceLineWriter::GetSymbolFunctionName(IDiaSymbol *function,
BSTR *name,
int *stack_param_size) {
*stack_param_size = -1;
const DWORD undecorate_options = UNDNAME_NO_MS_KEYWORDS |
UNDNAME_NO_FUNCTION_RETURNS |
UNDNAME_NO_ALLOCATION_MODEL |
UNDNAME_NO_ALLOCATION_LANGUAGE |
UNDNAME_NO_THISTYPE |
UNDNAME_NO_ACCESS_SPECIFIERS |
UNDNAME_NO_THROW_SIGNATURES |
UNDNAME_NO_MEMBER_TYPE |
UNDNAME_NO_RETURN_UDT_MODEL |
UNDNAME_NO_ECSU;
// Use get_undecoratedNameEx to get readable C++ names with arguments.
if (function->get_undecoratedNameEx(undecorate_options, name) != S_OK) {
if (function->get_name(name) != S_OK) {
fprintf(stderr, "failed to get function name\n");
return false;
}
// It's possible for get_name to return an empty string, so
// special-case that.
if (wcscmp(*name, L"") == 0) {
SysFreeString(*name);
// dwarf_cu_to_module.cc uses "<name omitted>", so match that.
*name = SysAllocString(L"<name omitted>");
return true;
}
// If a name comes from get_name because no undecorated form existed,
// it's already formatted properly to be used as output. Don't do any
// additional processing.
//
// MSVC7's DIA seems to not undecorate names in as many cases as MSVC8's.
// This will result in calling get_name for some C++ symbols, so
// all of the parameter and return type information may not be included in
// the name string.
} else {
// C++ uses a bogus "void" argument for functions and methods that don't
// take any parameters. Take it out of the undecorated name because it's
// ugly and unnecessary.
const wchar_t *replace_string = L"(void)";
const size_t replace_length = wcslen(replace_string);
const wchar_t *replacement_string = L"()";
size_t length = wcslen(*name);
if (length >= replace_length) {
wchar_t *name_end = *name + length - replace_length;
if (wcscmp(name_end, replace_string) == 0) {
WindowsStringUtils::safe_wcscpy(name_end, replace_length,
replacement_string);
length = wcslen(*name);
}
}
// Undecorate names used for stdcall and fastcall. These names prefix
// the identifier with '_' (stdcall) or '@' (fastcall) and suffix it
// with '@' followed by the number of bytes of parameters, in decimal.
// If such a name is found, take note of the size and undecorate it.
// Only do this for names that aren't C++, which is determined based on
// whether the undecorated name contains any ':' or '(' characters.
if (!wcschr(*name, ':') && !wcschr(*name, '(') &&
(*name[0] == '_' || *name[0] == '@')) {
wchar_t *last_at = wcsrchr(*name + 1, '@');
if (last_at && wcstol_positive_strict(last_at + 1, stack_param_size)) {
// If this function adheres to the fastcall convention, it accepts up
// to the first 8 bytes of parameters in registers (%ecx and %edx).
// We're only interested in the stack space used for parameters, so
// so subtract 8 and don't let the size go below 0.
if (*name[0] == '@') {
if (*stack_param_size > 8) {
*stack_param_size -= 8;
} else {
*stack_param_size = 0;
}
}
// Undecorate the name by moving it one character to the left in its
// buffer, and terminating it where the last '@' had been.
WindowsStringUtils::safe_wcsncpy(*name, length,
*name + 1, last_at - *name - 1);
} else if (*name[0] == '_') {
// This symbol's name is encoded according to the cdecl rules. The
// name doesn't end in a '@' character followed by a decimal positive
// integer, so it's not a stdcall name. Strip off the leading
// underscore.
WindowsStringUtils::safe_wcsncpy(*name, length, *name + 1, length);
}
}
}
return true;
}
// static
int PDBSourceLineWriter::GetFunctionStackParamSize(IDiaSymbol *function) {
// This implementation is highly x86-specific.
// Gather the symbols corresponding to data.
CComPtr<IDiaEnumSymbols> data_children;
if (FAILED(function->findChildren(SymTagData, NULL, nsNone,
&data_children))) {
return 0;
}
// lowest_base is the lowest %ebp-relative byte offset used for a parameter.
// highest_end is one greater than the highest offset (i.e. base + length).
// Stack parameters are assumed to be contiguous, because in reality, they
// are.
int lowest_base = INT_MAX;
int highest_end = INT_MIN;
CComPtr<IDiaSymbol> child;
DWORD count;
while (SUCCEEDED(data_children->Next(1, &child, &count)) && count == 1) {
// If any operation fails at this point, just proceed to the next child.
// Use the next_child label instead of continue because child needs to
// be released before it's reused. Declare constructable/destructable
// types early to avoid gotos that cross initializations.
CComPtr<IDiaSymbol> child_type;
// DataIsObjectPtr is only used for |this|. Because |this| can be passed
// as a stack parameter, look for it in addition to traditional
// parameters.
DWORD child_kind;
if (FAILED(child->get_dataKind(&child_kind)) ||
(child_kind != DataIsParam && child_kind != DataIsObjectPtr)) {
goto next_child;
}
// Only concentrate on register-relative parameters. Parameters may also
// be enregistered (passed directly in a register), but those don't
// consume any stack space, so they're not of interest.
DWORD child_location_type;
if (FAILED(child->get_locationType(&child_location_type)) ||
child_location_type != LocIsRegRel) {
goto next_child;
}
// Of register-relative parameters, the only ones that make any sense are
// %ebp- or %esp-relative. Note that MSVC's debugging information always
// gives parameters as %ebp-relative even when a function doesn't use a
// traditional frame pointer and stack parameters are accessed relative to
// %esp, so just look for %ebp-relative parameters. If you wanted to
// access parameters, you'd probably want to treat these %ebp-relative
// offsets as if they were relative to %esp before a function's prolog
// executed.
DWORD child_register;
if (FAILED(child->get_registerId(&child_register)) ||
child_register != CV_REG_EBP) {
goto next_child;
}
LONG child_register_offset;
if (FAILED(child->get_offset(&child_register_offset))) {
goto next_child;
}
// IDiaSymbol::get_type can succeed but still pass back a NULL value.
if (FAILED(child->get_type(&child_type)) || !child_type) {
goto next_child;
}
ULONGLONG child_length;
if (FAILED(child_type->get_length(&child_length))) {
goto next_child;
}
// Extra scope to avoid goto jumping over variable initialization
{
int child_end = child_register_offset + static_cast<ULONG>(child_length);
if (child_register_offset < lowest_base) {
lowest_base = child_register_offset;
}
if (child_end > highest_end) {
highest_end = child_end;
}
}
next_child:
child.Release();
}
int param_size = 0;
// Make sure lowest_base isn't less than 4, because [%esp+4] is the lowest
// possible address to find a stack parameter before executing a function's
// prolog (see above). Some optimizations cause parameter offsets to be
// lower than 4, but we're not concerned with those because we're only
// looking for parameters contained in addresses higher than where the
// return address is stored.
if (lowest_base < 4) {
lowest_base = 4;
}
if (highest_end > lowest_base) {
// All stack parameters are pushed as at least 4-byte quantities. If the
// last type was narrower than 4 bytes, promote it. This assumes that all
// parameters' offsets are 4-byte-aligned, which is always the case. Only
// worry about the last type, because we're not summing the type sizes,
// just looking at the lowest and highest offsets.
int remainder = highest_end % 4;
if (remainder) {
highest_end += 4 - remainder;
}
param_size = highest_end - lowest_base;
}
return param_size;
}
bool PDBSourceLineWriter::WriteMap(FILE *map_file) {
output_ = map_file;
// Load the OMAP information, and disable auto-translation of addresses in
// preference of doing it ourselves.
OmapData omap_data;
if (!GetOmapDataAndDisableTranslation(session_, &omap_data))
return false;
BuildImageMap(omap_data, &image_map_);
bool ret = PrintPDBInfo();
// This is not a critical piece of the symbol file.
PrintPEInfo();
ret = ret &&
PrintSourceFiles() &&
PrintFunctions() &&
PrintFrameData();
output_ = NULL;
return ret;
}
void PDBSourceLineWriter::Close() {
session_.Release();
}
bool PDBSourceLineWriter::GetModuleInfo(PDBModuleInfo *info) {
if (!info) {
return false;
}
info->debug_file.clear();
info->debug_identifier.clear();
info->cpu.clear();
CComPtr<IDiaSymbol> global;
if (FAILED(session_->get_globalScope(&global))) {
return false;
}
DWORD machine_type;
// get_machineType can return S_FALSE.
if (global->get_machineType(&machine_type) == S_OK) {
// The documentation claims that get_machineType returns a value from
// the CV_CPU_TYPE_e enumeration, but that's not the case.
// Instead, it returns one of the IMAGE_FILE_MACHINE values as
// defined here:
// http://msdn.microsoft.com/en-us/library/ms680313%28VS.85%29.aspx
switch (machine_type) {
case IMAGE_FILE_MACHINE_I386:
info->cpu = L"x86";
break;
case IMAGE_FILE_MACHINE_AMD64:
info->cpu = L"x86_64";
break;
default:
info->cpu = L"unknown";
break;
}
} else {
// Unexpected, but handle gracefully.
info->cpu = L"unknown";
}
// DWORD* and int* are not compatible. This is clean and avoids a cast.
DWORD age;
if (FAILED(global->get_age(&age))) {
return false;
}
bool uses_guid;
if (!UsesGUID(&uses_guid)) {
return false;
}
if (uses_guid) {
GUID guid;
if (FAILED(global->get_guid(&guid))) {
return false;
}
// Use the same format that the MS symbol server uses in filesystem
// hierarchies.
wchar_t age_string[9];
swprintf(age_string, sizeof(age_string) / sizeof(age_string[0]),
L"%x", age);
// remove when VC++7.1 is no longer supported
age_string[sizeof(age_string) / sizeof(age_string[0]) - 1] = L'\0';
info->debug_identifier = GUIDString::GUIDToSymbolServerWString(&guid);
info->debug_identifier.append(age_string);
} else {
DWORD signature;
if (FAILED(global->get_signature(&signature))) {
return false;
}
// Use the same format that the MS symbol server uses in filesystem
// hierarchies.
wchar_t identifier_string[17];
swprintf(identifier_string,
sizeof(identifier_string) / sizeof(identifier_string[0]),
L"%08X%x", signature, age);
// remove when VC++7.1 is no longer supported
identifier_string[sizeof(identifier_string) /
sizeof(identifier_string[0]) - 1] = L'\0';
info->debug_identifier = identifier_string;
}
CComBSTR debug_file_string;
if (FAILED(global->get_symbolsFileName(&debug_file_string))) {
return false;
}
info->debug_file =
WindowsStringUtils::GetBaseName(wstring(debug_file_string));
return true;
}
bool PDBSourceLineWriter::GetPEInfo(PEModuleInfo *info) {
if (!info) {
return false;
}
if (code_file_.empty() && !FindPEFile()) {
fprintf(stderr, "Couldn't locate EXE or DLL file.\n");
return false;
}
// Convert wchar to native charset because ImageLoad only takes
// a PSTR as input.
string code_file;
if (!WindowsStringUtils::safe_wcstombs(code_file_, &code_file)) {
return false;
}
AutoImage img(ImageLoad((PSTR)code_file.c_str(), NULL));
if (!img) {
fprintf(stderr, "Failed to open PE file: %s\n", code_file.c_str());
return false;
}
info->code_file = WindowsStringUtils::GetBaseName(code_file_);
// The date and time that the file was created by the linker.
DWORD TimeDateStamp = img->FileHeader->FileHeader.TimeDateStamp;
// The size of the file in bytes, including all headers.
DWORD SizeOfImage = 0;
PIMAGE_OPTIONAL_HEADER64 opt =
&((PIMAGE_NT_HEADERS64)img->FileHeader)->OptionalHeader;
if (opt->Magic == IMAGE_NT_OPTIONAL_HDR64_MAGIC) {
// 64-bit PE file.
SizeOfImage = opt->SizeOfImage;
} else {
// 32-bit PE file.
SizeOfImage = img->FileHeader->OptionalHeader.SizeOfImage;
}
wchar_t code_identifier[32];
swprintf(code_identifier,
sizeof(code_identifier) / sizeof(code_identifier[0]),
L"%08X%X", TimeDateStamp, SizeOfImage);
info->code_identifier = code_identifier;
return true;
}
bool PDBSourceLineWriter::UsesGUID(bool *uses_guid) {
if (!uses_guid)
return false;
CComPtr<IDiaSymbol> global;
if (FAILED(session_->get_globalScope(&global)))
return false;
GUID guid;
if (FAILED(global->get_guid(&guid)))
return false;
DWORD signature;
if (FAILED(global->get_signature(&signature)))
return false;
// There are two possibilities for guid: either it's a real 128-bit GUID
// as identified in a code module by a new-style CodeView record, or it's
// a 32-bit signature (timestamp) as identified by an old-style record.
// See MDCVInfoPDB70 and MDCVInfoPDB20 in minidump_format.h.
//
// Because DIA doesn't provide a way to directly determine whether a module
// uses a GUID or a 32-bit signature, this code checks whether the first 32
// bits of guid are the same as the signature, and if the rest of guid is
// zero. If so, then with a pretty high degree of certainty, there's an
// old-style CodeView record in use. This method will only falsely find an
// an old-style CodeView record if a real 128-bit GUID has its first 32
// bits set the same as the module's signature (timestamp) and the rest of
// the GUID is set to 0. This is highly unlikely.
GUID signature_guid = {signature}; // 0-initializes other members
*uses_guid = !IsEqualGUID(guid, signature_guid);
return true;
}
} // namespace google_breakpad