Add pipe implementation for OOP memory logging

chrome/profiling/memlog_receiver_pipe_win.cc

+// Copyright 2017 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 "chrome/profiling/memlog_receiver_pipe_win.h"
+
+#include "base/bind.h"
+#include "base/logging.h"
+#include "base/message_loop/message_loop.h"
+#include "base/strings/string16.h"
+#include "base/strings/utf_string_conversions.h"
+#include "base/threading/thread.h"
+#include "chrome/profiling/memlog_stream_receiver.h"
+
+namespace profiling {
+
+namespace {
+
+// Use a large buffer for our pipe. We don't want the sender to block
+// if at all possible since it will slow the app down quite a bit. Windows
+// seems to max out a 64K per read so we use that (larger would be better if it
+// worked).
+const int kReadBufferSize = 1024 * 64;
+
+}  // namespace
+
+MemlogReceiverPipe::CompletionThunk::CompletionThunk(HANDLE handle, Callback cb)
+    : handle_(handle), callback_(cb) {
+  base::MessageLoopForIO::current()->RegisterIOHandler(handle, this);
+  ZeroOverlapped();
+}
+
+MemlogReceiverPipe::CompletionThunk::~CompletionThunk() {
+  if (handle_ != INVALID_HANDLE_VALUE)
+    ::CloseHandle(handle_);
+}
+
+void MemlogReceiverPipe::CompletionThunk::ZeroOverlapped() {
+  memset(overlapped(), 0, sizeof(OVERLAPPED));
+}
+
+void MemlogReceiverPipe::CompletionThunk::OnIOCompleted(
+    base::MessagePumpForIO::IOContext* context,
+    DWORD bytes_transfered,
+    DWORD error) {
+  // Note: any crashes with this on the stack are likely a result of destroying
+  // a relevant class while there is I/O pending.
+  callback_.Run(static_cast<size_t>(bytes_transfered), error);
+}
+
+MemlogReceiverPipe::MemlogReceiverPipe(std::unique_ptr<CompletionThunk> thunk)
+    : thunk_(std::move(thunk)), read_buffer_(new char[kReadBufferSize]) {
+  // Need Unretained to avoid a reference cycle.
+  thunk_->set_callback(base::BindRepeating(&MemlogReceiverPipe::OnIOCompleted,
+                                           base::Unretained(this)));
+}
+
+MemlogReceiverPipe::~MemlogReceiverPipe() {}
+
+void MemlogReceiverPipe::StartReadingOnIOThread() {
+  ReadUntilBlocking();
+}
+
+int MemlogReceiverPipe::GetRemoteProcessID() {
+  ULONG id = 0;
+  ::GetNamedPipeClientProcessId(thunk_->handle(), &id);
+  return static_cast<int>(id);
+}
+
+void MemlogReceiverPipe::SetReceiver(
+    scoped_refptr<base::TaskRunner> task_runner,
+    scoped_refptr<MemlogStreamReceiver> receiver) {
+  receiver_task_runner_ = task_runner;
+  receiver_ = receiver;
+}
+
+void MemlogReceiverPipe::OnIOCompleted(size_t bytes_transfered, DWORD error) {
+  DCHECK(read_outstanding_);
+  read_outstanding_ = false;
+
+  // This will get called both for async completion of ConnectNamedPipe as well
+  // as async read completions.
+  if (bytes_transfered && receiver_) {
+    receiver_task_runner_->PostTask(
+        FROM_HERE,
+        base::BindOnce(&MemlogStreamReceiver::OnStreamData, receiver_,
+                       std::move(read_buffer_), bytes_transfered));
+    read_buffer_.reset(new char[kReadBufferSize]);
+  }
+  ReadUntilBlocking();
+}
+
+void MemlogReceiverPipe::ReadUntilBlocking() {
+  // TODO(brettw) note that the IO completion callback will always be issued,
+  // even for sync returns of ReadFile. If there is a lot of data ready to be
+  // read, it would be nice to process them all in this loop rather than having
+  // to go back to the message loop for each block, but that will require
+  // different IOContext structures for each one.
+  DWORD bytes_read = 0;
+  thunk_->ZeroOverlapped();
+
+  DCHECK(!read_outstanding_);
+  read_outstanding_ = true;
+  if (!::ReadFile(thunk_->handle(), read_buffer_.get(), kReadBufferSize,
+                  &bytes_read, thunk_->overlapped())) {
+    if (GetLastError() == ERROR_IO_PENDING) {
+      return;
+    } else {
+      if (receiver_) {
+        receiver_task_runner_->PostTask(
+            FROM_HERE,
+            base::BindOnce(&MemlogStreamReceiver::OnStreamComplete, receiver_));

[awong, 2017/06/15 21:37:24]

Why is this a BindOnce? I would have expected BindRepeating in the constructor?

[brettw, 2017/06/15 22:12:22]

I don't follow. What do you mean by the constructor? This is just posting a task
to a thread.

+      }
+      return;
+    }
+  }
+}
+
+}  // namespace profiling