Add TCP socket mojo interfaces

mojo/services/public/interfaces/network/network_service.mojom

Index: mojo/services/public/interfaces/network/network_service.mojom
diff --git a/mojo/services/public/interfaces/network/network_service.mojom b/mojo/services/public/interfaces/network/network_service.mojom
index 79bfa6e828a20a7cb26271802350aa5d8198237e..78841448dccf895b4725ab2c5c117aa415a2202f 100644
--- a/mojo/services/public/interfaces/network/network_service.mojom
+++ b/mojo/services/public/interfaces/network/network_service.mojom
@@ -3,11 +3,18 @@
 // found in the LICENSE file.
 
 import "mojo/services/public/interfaces/network/cookie_store.mojom"
+import "mojo/services/public/interfaces/network/net_address.mojom"
+import "mojo/services/public/interfaces/network/network_error.mojom"
+import "mojo/services/public/interfaces/network/tcp_bound_socket.mojom"
+import "mojo/services/public/interfaces/network/tcp_client_socket.mojom"
 import "mojo/services/public/interfaces/network/url_loader.mojom"
 import "mojo/services/public/interfaces/network/web_socket.mojom"
 
 module mojo {
 
+// TODO Darin suggfests that this should probably be two classes. One for

[willchan no longer on Chromium, 2014/10/06 21:52:39]

s/suggfests/suggests/

And I agree with Darin's suggestion.

+// high-level origin-build requests like WebSockets and HTTP, and the other for

[wtc, 2014/10/03 22:37:43]

Typo: origin-build => origin-bound

+// non-origin-bound low-level stuff like DNS, UDP, and TCP.
 interface NetworkService {
   CreateURLLoader(URLLoader&? loader);
 
@@ -15,7 +22,34 @@ interface NetworkService {
 
   CreateWebSocket(WebSocket& socket);
 
-  // TODO(darin): Add other methods here.
+  // Creates a TCP socket bound to a given local address. This bound socket
+  // can be used for creating a client or server socket on that local address.
+  //
+  // If you want to create a client socket to connect to a server and are in
+  // the common case where you don't care about the local address it's bound
+  // to, use CreateTCPClientSocket.
+  //
+  // The local address can specify 0 for the port to specify that the OS should
+  // pick an available port for the given address, or it can pass 0 for the
+  // address and port for the OS to pick both the local address and port. In
+  // all success cases, the resulting local address will be passed to the
+  // callback as bound_to.
+  CreateTCPBoundSocket(NetAddress local_address,
+                       TCPBoundSocket& bound_socket)
+      => (NetworkError result, NetAddress? bound_to);
+
+  // Creates a client socket connected to the given remote address. A local
+  // address and port will be allocated for the connection and passed to the
+  // callback on success.
+  //
+  // If you want control over the local address and port, instead use
+  // CreateTCPBoundSocket.
+  CreateTCPClientSocket(NetAddress remote_address,
+                        handle<data_pipe_consumer> send_stream,
+                        handle<data_pipe_producer> receive_stream,
+                        TCPClientSocket& client_socket)
+      => (NetworkError result,
+          NetAddress? local_address);

[wtc, 2014/10/03 22:37:43]

1. Few applications are interested in the local address in this case, so
returning local_address is almost always wasted work (it requires a
getsockname() system call).

2. Nit: just curious why the local_address return value is put on its own line,
unlike line 39.

[brettw, 2014/10/03 22:54:44]

How expensive is this syscall? I did it this way because otherwise getting the
local address requires an async IPC which is very inconvenient for applications
that do want this value.

[willchan no longer on Chromium, 2014/10/06 21:52:39]

To my knowledge, it's a cheap syscall, but it's still got the syscall overhead.
But I agree with wtc@ that most applications don't care.

[wtc, 2014/10/07 20:47:45]

The getsockname system call should be a simple system call, so the
cost is probably dominated by the user space/kernel context switch.
Getting the local address of such a TCP client socket is an uncommon
operation (does any code in Chromium do this?), so it is fine for it
to be suboptimal.

 };
 
 }