Upstreaming RegisterProcessUtils()

chrome/browser/android/process_utils.cc

Index: chrome/browser/android/process_utils.cc
diff --git a/chrome/browser/android/process_utils.cc b/chrome/browser/android/process_utils.cc
index c0f52802273241fca6752799da785159567790df..e5cff54ad8d0efb6d9837dfb05e383ebb1434bb8 100644
--- a/chrome/browser/android/process_utils.cc
+++ b/chrome/browser/android/process_utils.cc
@@ -8,12 +8,37 @@
 
 #include "base/lazy_instance.h"
 #include "base/logging.h"
+#include "chrome/browser/browser_process.h"
+#include "chrome/browser/profiles/profile.h"
+#include "chrome/browser/profiles/profile_manager.h"
 #include "chrome/common/render_messages.h"
+#include "content/public/browser/browser_thread.h"
 #include "content/public/browser/render_process_host.h"
 #include "jni/ProcessUtils_jni.h"
+#include "net/http/http_network_session.h"
+#include "net/http/http_transaction_factory.h"
+#include "net/url_request/url_request_context.h"
+#include "net/url_request/url_request_context_getter.h"
 
 namespace {
 
+void CloseIdleConnectionsForProfile(
+    scoped_refptr<net::URLRequestContextGetter> context_getter) {
+  DCHECK(context_getter.get());
+  if (!content::BrowserThread::CurrentlyOn(content::BrowserThread::IO)) {
+    content::BrowserThread::PostTask(content::BrowserThread::IO, FROM_HERE,
+                           base::Bind(&CloseIdleConnectionsForProfile,
+                                      context_getter));
+    return;
+  }
+
+  net::URLRequestContext* context = context_getter->GetURLRequestContext();
+  DCHECK(context->http_transaction_factory());
+  net::HttpNetworkSession* session =
+      context->http_transaction_factory()->GetSession();
+  DCHECK(session);

[Philippe, 2012/12/03 13:17:46]

Are we missing 'session->CloseIdleConnections()' here?

I came across this function while investigating a Spdy crash (bug 163616 FYI).
It seems that this function does nothing more than getting the session.
I think the call to CloseIdleConnections() got unintentionally lost during the
review process (it was there in the first patch set).

[digit1, 2012/12/03 13:42:02]

Yes, this looks suspicious. What happens if you add the missing call? I.e. do
all our unit tests still pass? It might be worthwhile to try to add a regression
test if this doesn't turn out too long.

[Philippe, 2012/12/03 13:48:37]

Yes, I'm building/testing this right now. FYI, we are supposed to close idle
connections after 5min of inactivity to avoid consuming too much battery when
the app is in the background. This is triggered by the onPause() override on the
application side.
This has basically been a no-op upstream. I will send a patch fixing this soon.

+}
+
 // Only accessed from the JNI thread by ToggleWebKitSharedTimers() which is
 // implemented below.
 base::LazyInstance<std::vector<int /* process id */> > g_suspended_processes =
@@ -54,7 +79,23 @@ static void ToggleWebKitSharedTimers(JNIEnv* env,
   }
 }
 
-// TODO(pliard): http://crbug.com/137674
+static void CloseIdleConnections(JNIEnv* env, jclass obj) {
+  // Iterate through all loaded profiles (and their associated incognito
+  // profiles if created), and close the idle connections associated with each
+  // one.
+  std::vector<Profile*> profiles(
+      g_browser_process->profile_manager()->GetLoadedProfiles());
+  for (std::vector<Profile*>::iterator i = profiles.begin();
+      i != profiles.end(); i++) {
+    Profile* profile = *i;
+    CloseIdleConnectionsForProfile(profile->GetRequestContext());
+    if (profile->HasOffTheRecordProfile()) {
+      CloseIdleConnectionsForProfile(
+          profile->GetOffTheRecordProfile()->GetRequestContext());
+    }
+  }
+}
+
 bool RegisterProcessUtils(JNIEnv* env) {
   return RegisterNativesImpl(env);
 }