1286636 - [FlyWeb] Use unbound UDP socket for mDNS queries

Status: RESOLVED FIXED

(Core :: Networking, defect)

Description

[Kannan Vijayan [:djvj]]

One of the implementation issues we ran into on Windows 10 is that win10 system processes open and hold UDP port 5353 on the mDNS multicast address.

Our current fallback implementation also uses a SO_REUSE UDP socket on port 5353, but is hampered by conflicts.  The main issue is that only broadcast packets sent to the multicast group are delivered to the gecko UDP socket.  Directed UDP messages to the machine on that port are not delivered, and presumably go to the system socket instead.

The workaround suggested is to use a single query socket opened on an arbitrary port.  The query socket does NOT join the mDNS multicast IP group, but DOES send packets to that group.  Since this socket is assigned its own unique port, query responses addressed to it will resolve properly.

Comment 1

[Kannan Vijayan [:djvj]]

Attachment: use-unbound-query-socket.patch

Tested this on Linux, Android, and my Win10 box.  Seems to work quite well.

Comment 2

[Justin D'Arcangelo [:justindarc]]

Comment on attachment 8770685 [details] [diff] [review]
use-unbound-query-socket.patch

Review of attachment 8770685 [details] [diff] [review]:
-----------------------------------------------------------------

LGTM. See my comment about removing the old block of commented-out code.

::: netwerk/dns/mdns/libmdns/fallback/MulticastDNS.jsm
@@ +343,5 @@
>        sockets.forEach((socket) => {
>          socket.send(MDNS_MULTICAST_GROUP, MDNS_PORT, data, data.length);
>        });
>      });
> +    */

We should probably just remove this block instead of commenting it out.

Comment 3

[Kannan Vijayan [:djvj]]

Attachment: discovery-fixes.patch

Updated patch.

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/048a09249037
Use dedicated custom-port query socket to do mDNS queries.  Respond to queries using per-interface sockets. r=justindarc

Comment 5

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/048a09249037

Comment 6

[Joel Maher ( :jmaher ) (UTC -8)]

this caused a small perf improvement on android (nexus 4.1: https://treeherder.mozilla.org/perf.html#/alerts?id=2167 ):
Summary of tests that improved:

  remote-blank summary android-4-2-armv7-api15 opt - 2.08% better


thanks for the win!