blob: 905d0a2e9b8871d6e72550c2c6df36c4d59f804a [file] [log] [blame]
// Copyright (c) 2012 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "base/sync_socket.h"
#include <limits.h>
#include <stddef.h>
#include "base/logging.h"
#include "base/macros.h"
#include "base/rand_util.h"
#include "base/threading/thread_restrictions.h"
#include "base/win/scoped_handle.h"
namespace base {
using win::ScopedHandle;
namespace {
// IMPORTANT: do not change how this name is generated because it will break
// in sandboxed scenarios as we might have by-name policies that allow pipe
// creation. Also keep the secure random number generation.
const wchar_t kPipeNameFormat[] = L"\\\\.\\pipe\\chrome.sync.%u.%u.%lu";
const size_t kPipePathMax = arraysize(kPipeNameFormat) + (3 * 10) + 1;
// To avoid users sending negative message lengths to Send/Receive
// we clamp message lengths, which are size_t, to no more than INT_MAX.
const size_t kMaxMessageLength = static_cast<size_t>(INT_MAX);
const int kOutBufferSize = 4096;
const int kInBufferSize = 4096;
const int kDefaultTimeoutMilliSeconds = 1000;
bool CreatePairImpl(HANDLE* socket_a, HANDLE* socket_b, bool overlapped) {
DCHECK_NE(socket_a, socket_b);
DCHECK_EQ(*socket_a, SyncSocket::kInvalidHandle);
DCHECK_EQ(*socket_b, SyncSocket::kInvalidHandle);
wchar_t name[kPipePathMax];
ScopedHandle handle_a;
DWORD flags = PIPE_ACCESS_DUPLEX | FILE_FLAG_FIRST_PIPE_INSTANCE;
if (overlapped)
flags |= FILE_FLAG_OVERLAPPED;
do {
unsigned long rnd_name;
RandBytes(&rnd_name, sizeof(rnd_name));
swprintf(name, kPipePathMax,
kPipeNameFormat,
GetCurrentProcessId(),
GetCurrentThreadId(),
rnd_name);
handle_a.Set(CreateNamedPipeW(
name,
flags,
PIPE_TYPE_BYTE | PIPE_READMODE_BYTE,
1,
kOutBufferSize,
kInBufferSize,
kDefaultTimeoutMilliSeconds,
NULL));
} while (!handle_a.IsValid() &&
(GetLastError() == ERROR_PIPE_BUSY));
if (!handle_a.IsValid()) {
NOTREACHED();
return false;
}
// The SECURITY_ANONYMOUS flag means that the server side (handle_a) cannot
// impersonate the client (handle_b). This allows us not to care which side
// ends up in which side of a privilege boundary.
flags = SECURITY_SQOS_PRESENT | SECURITY_ANONYMOUS;
if (overlapped)
flags |= FILE_FLAG_OVERLAPPED;
ScopedHandle handle_b(CreateFileW(name,
GENERIC_READ | GENERIC_WRITE,
0, // no sharing.
NULL, // default security attributes.
OPEN_EXISTING, // opens existing pipe.
flags,
NULL)); // no template file.
if (!handle_b.IsValid()) {
DPLOG(ERROR) << "CreateFileW failed";
return false;
}
if (!ConnectNamedPipe(handle_a.Get(), NULL)) {
DWORD error = GetLastError();
if (error != ERROR_PIPE_CONNECTED) {
DPLOG(ERROR) << "ConnectNamedPipe failed";
return false;
}
}
*socket_a = handle_a.Take();
*socket_b = handle_b.Take();
return true;
}
// Inline helper to avoid having the cast everywhere.
DWORD GetNextChunkSize(size_t current_pos, size_t max_size) {
// The following statement is for 64 bit portability.
return static_cast<DWORD>(((max_size - current_pos) <= UINT_MAX) ?
(max_size - current_pos) : UINT_MAX);
}
// Template function that supports calling ReadFile or WriteFile in an
// overlapped fashion and waits for IO completion. The function also waits
// on an event that can be used to cancel the operation. If the operation
// is cancelled, the function returns and closes the relevant socket object.
template <typename BufferType, typename Function>
size_t CancelableFileOperation(Function operation,
HANDLE file,
BufferType* buffer,
size_t length,
WaitableEvent* io_event,
WaitableEvent* cancel_event,
CancelableSyncSocket* socket,
DWORD timeout_in_ms) {
AssertBlockingAllowed();
// The buffer must be byte size or the length check won't make much sense.
static_assert(sizeof(buffer[0]) == sizeof(char), "incorrect buffer type");
DCHECK_GT(length, 0u);
DCHECK_LE(length, kMaxMessageLength);
DCHECK_NE(file, SyncSocket::kInvalidHandle);
// Track the finish time so we can calculate the timeout as data is read.
TimeTicks current_time, finish_time;
if (timeout_in_ms != INFINITE) {
current_time = TimeTicks::Now();
finish_time =
current_time + base::TimeDelta::FromMilliseconds(timeout_in_ms);
}
size_t count = 0;
do {
// The OVERLAPPED structure will be modified by ReadFile or WriteFile.
OVERLAPPED ol = { 0 };
ol.hEvent = io_event->handle();
const DWORD chunk = GetNextChunkSize(count, length);
// This is either the ReadFile or WriteFile call depending on whether
// we're receiving or sending data.
DWORD len = 0;
const BOOL operation_ok = operation(
file, static_cast<BufferType*>(buffer) + count, chunk, &len, &ol);
if (!operation_ok) {
if (::GetLastError() == ERROR_IO_PENDING) {
HANDLE events[] = { io_event->handle(), cancel_event->handle() };
const int wait_result = WaitForMultipleObjects(
arraysize(events), events, FALSE,
timeout_in_ms == INFINITE ?
timeout_in_ms :
static_cast<DWORD>(
(finish_time - current_time).InMilliseconds()));
if (wait_result != WAIT_OBJECT_0 + 0) {
// CancelIo() doesn't synchronously cancel outstanding IO, only marks
// outstanding IO for cancellation. We must call GetOverlappedResult()
// below to ensure in flight writes complete before returning.
CancelIo(file);
}
// We set the |bWait| parameter to TRUE for GetOverlappedResult() to
// ensure writes are complete before returning.
if (!GetOverlappedResult(file, &ol, &len, TRUE))
len = 0;
if (wait_result == WAIT_OBJECT_0 + 1) {
DVLOG(1) << "Shutdown was signaled. Closing socket.";
socket->Close();
return count;
}
// Timeouts will be handled by the while() condition below since
// GetOverlappedResult() may complete successfully after CancelIo().
DCHECK(wait_result == WAIT_OBJECT_0 + 0 || wait_result == WAIT_TIMEOUT);
} else {
break;
}
}
count += len;
// Quit the operation if we can't write/read anymore.
if (len != chunk)
break;
// Since TimeTicks::Now() is expensive, only bother updating the time if we
// have more work to do.
if (timeout_in_ms != INFINITE && count < length)
current_time = base::TimeTicks::Now();
} while (count < length &&
(timeout_in_ms == INFINITE || current_time < finish_time));
return count;
}
} // namespace
#if defined(COMPONENT_BUILD)
const SyncSocket::Handle SyncSocket::kInvalidHandle = INVALID_HANDLE_VALUE;
#endif
SyncSocket::SyncSocket() : handle_(kInvalidHandle) {}
SyncSocket::~SyncSocket() {
Close();
}
// static
bool SyncSocket::CreatePair(SyncSocket* socket_a, SyncSocket* socket_b) {
return CreatePairImpl(&socket_a->handle_, &socket_b->handle_, false);
}
// static
SyncSocket::Handle SyncSocket::UnwrapHandle(
const TransitDescriptor& descriptor) {
return descriptor;
}
bool SyncSocket::PrepareTransitDescriptor(ProcessHandle peer_process_handle,
TransitDescriptor* descriptor) {
DCHECK(descriptor);
if (!::DuplicateHandle(GetCurrentProcess(), handle(), peer_process_handle,
descriptor, 0, FALSE, DUPLICATE_SAME_ACCESS)) {
DPLOG(ERROR) << "Cannot duplicate socket handle for peer process.";
return false;
}
return true;
}
bool SyncSocket::Close() {
if (handle_ == kInvalidHandle)
return true;
const BOOL result = CloseHandle(handle_);
handle_ = kInvalidHandle;
return result == TRUE;
}
size_t SyncSocket::Send(const void* buffer, size_t length) {
AssertBlockingAllowed();
DCHECK_GT(length, 0u);
DCHECK_LE(length, kMaxMessageLength);
DCHECK_NE(handle_, kInvalidHandle);
size_t count = 0;
while (count < length) {
DWORD len;
DWORD chunk = GetNextChunkSize(count, length);
if (::WriteFile(handle_, static_cast<const char*>(buffer) + count, chunk,
&len, NULL) == FALSE) {
return count;
}
count += len;
}
return count;
}
size_t SyncSocket::ReceiveWithTimeout(void* buffer,
size_t length,
TimeDelta timeout) {
NOTIMPLEMENTED();
return 0;
}
size_t SyncSocket::Receive(void* buffer, size_t length) {
AssertBlockingAllowed();
DCHECK_GT(length, 0u);
DCHECK_LE(length, kMaxMessageLength);
DCHECK_NE(handle_, kInvalidHandle);
size_t count = 0;
while (count < length) {
DWORD len;
DWORD chunk = GetNextChunkSize(count, length);
if (::ReadFile(handle_, static_cast<char*>(buffer) + count, chunk, &len,
NULL) == FALSE) {
return count;
}
count += len;
}
return count;
}
size_t SyncSocket::Peek() {
DWORD available = 0;
PeekNamedPipe(handle_, NULL, 0, NULL, &available, NULL);
return available;
}
SyncSocket::Handle SyncSocket::Release() {
Handle r = handle_;
handle_ = kInvalidHandle;
return r;
}
CancelableSyncSocket::CancelableSyncSocket()
: shutdown_event_(base::WaitableEvent::ResetPolicy::MANUAL,
base::WaitableEvent::InitialState::NOT_SIGNALED),
file_operation_(base::WaitableEvent::ResetPolicy::MANUAL,
base::WaitableEvent::InitialState::NOT_SIGNALED) {}
CancelableSyncSocket::CancelableSyncSocket(Handle handle)
: SyncSocket(handle),
shutdown_event_(base::WaitableEvent::ResetPolicy::MANUAL,
base::WaitableEvent::InitialState::NOT_SIGNALED),
file_operation_(base::WaitableEvent::ResetPolicy::MANUAL,
base::WaitableEvent::InitialState::NOT_SIGNALED) {}
bool CancelableSyncSocket::Shutdown() {
// This doesn't shut down the pipe immediately, but subsequent Receive or Send
// methods will fail straight away.
shutdown_event_.Signal();
return true;
}
bool CancelableSyncSocket::Close() {
const bool result = SyncSocket::Close();
shutdown_event_.Reset();
return result;
}
size_t CancelableSyncSocket::Send(const void* buffer, size_t length) {
static const DWORD kWaitTimeOutInMs = 500;
return CancelableFileOperation(
&::WriteFile, handle_, reinterpret_cast<const char*>(buffer), length,
&file_operation_, &shutdown_event_, this, kWaitTimeOutInMs);
}
size_t CancelableSyncSocket::Receive(void* buffer, size_t length) {
return CancelableFileOperation(
&::ReadFile, handle_, reinterpret_cast<char*>(buffer), length,
&file_operation_, &shutdown_event_, this, INFINITE);
}
size_t CancelableSyncSocket::ReceiveWithTimeout(void* buffer,
size_t length,
TimeDelta timeout) {
return CancelableFileOperation(&::ReadFile, handle_,
reinterpret_cast<char*>(buffer), length,
&file_operation_, &shutdown_event_, this,
static_cast<DWORD>(timeout.InMilliseconds()));
}
// static
bool CancelableSyncSocket::CreatePair(CancelableSyncSocket* socket_a,
CancelableSyncSocket* socket_b) {
return CreatePairImpl(&socket_a->handle_, &socket_b->handle_, true);
}
} // namespace base