Adds chrome.memoryBenchmarking.heapProfilerDump for the browser process.

content/common/memory_benchmark_messages.h

Index: content/common/memory_benchmark_messages.h
diff --git a/chrome/common/one_click_signin_messages.h b/content/common/memory_benchmark_messages.h
similarity index 53%
copy from chrome/common/one_click_signin_messages.h
copy to content/common/memory_benchmark_messages.h
index fcbb75a19c90726f05cb6cc4e6ad1d85e09146a2..e355adbd5fe0d1a00139d1dd77c172ae012c7d99 100644
--- a/chrome/common/one_click_signin_messages.h
+++ b/content/common/memory_benchmark_messages.h
@@ -2,11 +2,14 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
+// IPC messages for device motion.

[piman, 2013/05/10 18:13:52]

nit: fix description.

[bulach, 2013/05/10 18:46:03]

Done.

 // Multiply-included message file, hence no include guard.
 
-#include "content/public/common/password_form.h"
+#include <string>
+
 #include "ipc/ipc_message_macros.h"
 
-#define IPC_MESSAGE_START OneClickSigninMsgStart
+#define IPC_MESSAGE_START MemoryBenchmarkMsgStart
 
-IPC_MESSAGE_ROUTED1(OneClickSigninHostMsg_FormSubmitted, content::PasswordForm)
+IPC_SYNC_MESSAGE_ROUTED0_1(MemoryBenchmarkHostMsg_HeapProfilerDump,

[piman, 2013/05/10 18:13:52]

This message doesn't need to be ROUTED, it could just be CONTROL. That way you
don't need to find the right view in HeapProfilerDumpBrowser, you can directly
go to RenderThreadImpl::current()->Send()

[bulach, 2013/05/10 18:46:03]

ah, right.. much simpler, thanks!

+                           std::string /* dump file name */)

[piman, 2013/05/10 18:13:52]

nit: use FilePath in IPCs.

[palmer, 2013/05/10 18:42:17]

It's more than a nit. :) An arbitrary file pathname write IPC, for memory
benchmarking, cost us a Pwnium loss last year.

The correct pattern is: client process asks browser process for a file
descriptor/HANDLE to write to; browser chooses a pathname statically or
otherwise safely; browser opens it with correct flags; browser sends FD back to
client.

(Options: The client can close the FD when done; or the client can send a
message to the browser when it's time to close it and the browser actually calls
close; or the browser can unilaterally close it at a time of its choosing.)

Should this be guarded on #if defined(HAS_MEMORY_BENCHMARK) ? Part of the
previous bug was that the IPC endpoint was present, even though no callers
remained in a Release build.

[bulach, 2013/05/10 18:46:03]

Done.

[bulach, 2013/05/10 18:47:31]

sorry, I didn't address this comment in patchset #2, let me address this now...