mirror of
https://github.com/chromium/crashpad.git
synced 2025-03-10 06:36:02 +00:00
d7aeda9fca
* Switch to Chromium's clang package on Windows. Fuchsia's clang
package uses symlinks, which don't Just Work. See comment 3
on the linked bug for details.
* Also tweak the installed path of the clang package to use
win-amd64 instead of windows-amd64 to match gn's host_os
variable, which is needed by the mini_chromium change in the
roll mentioned below.
* Add `__declspec(noinline)` to sources of some end-to-end test
binaries where the test expects a certain stack layout that's
perturbed by inlining
* self_destroying_test_program.cc failed with Access Violation
instead of with Breakpoint. I'm not 100% sure what's happening
there as I couldn't reproduce it on my machine. The test uses
VirtualFree(..., MEM_DECOMMIT) to make parts of the stack unreadable
to make sure crashpad can handle that. My theory is that clang
optimized out the call to `_alloca()`, which results in not enough
of the stack being around for VirtualFree() to be able to decommit.
https://geidav.wordpress.com/tag/reserved-memory/#:~:text=How%20does%20Windows%20manage%20stack%20memory%3F
and https://godbolt.org/z/YdbYdKeMh suggest that this theory is
not completely off -- but the
`[[maybe_unused]] volatile void* do_not_optimize_away` bit feels
homeopathic. And yet, with it this seems to pass CQ (see try results
at patch sets 12 and 20) and without it it doesn't (see patch set 19).
Maybe this was just luck and the test still feels flakily!
(The linker's `/STACK` defaults to 1MB stack reserved and 4K
committed -- 4K kind of seems like it should be enough for that
VirtualFree call to succeed if that really is the problem.)
So this part is a bit experimental.
Anyways, nothing uses clang yet, so no behavior change at this point
(other than `gclient sync` downloading less stuff and not failing
by default on Windows).
Includes a one-rev mini_chromium roll:
12ef786772..08d490553b
$ git log 12ef78677..08d490553 --date=short --no-merges --format='%ad %ae %s'
2025-02-13 thakis Prepare mini_chromium for using clang on Windows
Bug: 384682775
Change-Id: I33b54acf18cb6eadfd2167c76c0b4a5824efa64d
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6242577
Reviewed-by: Mark Mentovai <mark@chromium.org>
Commit-Queue: Nico Weber <thakis@chromium.org>
102 lines
3.0 KiB
C++
102 lines
3.0 KiB
C++
// Copyright 2015 The Crashpad Authors
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
#include <malloc.h>
|
|
#include <stdlib.h>
|
|
#include <windows.h>
|
|
#include <winternl.h>
|
|
|
|
#include "base/logging.h"
|
|
#include "client/crashpad_client.h"
|
|
#include "snapshot/win/process_reader_win.h"
|
|
|
|
namespace crashpad {
|
|
namespace {
|
|
|
|
// Without this, clang optimizes away the _alloca below, which in turn
|
|
// makes the VirtualFree() crash with an access violation.
|
|
#if defined(__clang__)
|
|
#pragma clang optimize off
|
|
#endif
|
|
|
|
// We VirtualFree a region in ourselves (the stack) to confirm that the
|
|
// exception reporter captures as much as possible in the minidump and doesn't
|
|
// abort. __debugbreak() immediately after doing so because the process is
|
|
// clearly in a very broken state at this point.
|
|
__declspec(noinline) bool FreeOwnStackAndBreak() {
|
|
ProcessReaderWin process_reader;
|
|
if (!process_reader.Initialize(GetCurrentProcess(),
|
|
ProcessSuspensionState::kRunning)) {
|
|
LOG(ERROR) << "ProcessReaderWin Initialize";
|
|
return false;
|
|
}
|
|
|
|
const std::vector<ProcessReaderWin::Thread> threads =
|
|
process_reader.Threads();
|
|
if (threads.empty()) {
|
|
LOG(ERROR) << "no threads";
|
|
return false;
|
|
}
|
|
|
|
// Push the stack up a bit so that hopefully the crash handler can succeed,
|
|
// but won't be able to read the base of the stack.
|
|
[[maybe_unused]] volatile void* do_not_optimize_away = _alloca(16384);
|
|
|
|
// We can't succeed at MEM_RELEASEing this memory, but MEM_DECOMMIT is good
|
|
// enough to make it inaccessible.
|
|
if (!VirtualFree(reinterpret_cast<void*>(threads[0].stack_region_address),
|
|
100,
|
|
MEM_DECOMMIT)) {
|
|
PLOG(ERROR) << "VirtualFree";
|
|
return false;
|
|
}
|
|
|
|
// If the VirtualFree() succeeds, we may have already crashed. __debugbreak()
|
|
// just to be sure.
|
|
__debugbreak();
|
|
|
|
return true;
|
|
}
|
|
|
|
__declspec(noinline) int SelfDestroyingMain(int argc, wchar_t* argv[]) {
|
|
if (argc != 2) {
|
|
fprintf(stderr, "Usage: %ls <server_pipe_name>\n", argv[0]);
|
|
return EXIT_FAILURE;
|
|
}
|
|
|
|
CrashpadClient client;
|
|
if (!client.SetHandlerIPCPipe(argv[1])) {
|
|
LOG(ERROR) << "SetHandler";
|
|
return EXIT_FAILURE;
|
|
}
|
|
|
|
if (!FreeOwnStackAndBreak())
|
|
return EXIT_FAILURE;
|
|
|
|
// This will never be reached. On success, we'll have crashed above, or
|
|
// otherwise returned before here.
|
|
return EXIT_SUCCESS;
|
|
}
|
|
|
|
#if defined(__clang__)
|
|
#pragma clang optimize on
|
|
#endif
|
|
|
|
} // namespace
|
|
} // namespace crashpad
|
|
|
|
int wmain(int argc, wchar_t* argv[]) {
|
|
return crashpad::SelfDestroyingMain(argc, argv);
|
|
}
|