mbox series

[v3,net,0/4] wireguard fixes for 5.6-rc2

Message ID 20200214225723.63646-1-Jason@zx2c4.com
Headers show
Series wireguard fixes for 5.6-rc2 | expand

Message

Jason A. Donenfeld Feb. 14, 2020, 10:57 p.m. UTC 
Here are four fixes for wireguard collected since rc1:

1) Some small cleanups to the test suite to help massively parallel
   builds.

2) A change in how we reset our load calculation to avoid a more
   expensive comparison, suggested by Matt Dunwoodie.

3) I've been loading more and more of wireguard's surface into
   syzkaller, trying to get our coverage as complete as possible,
   leading in this case to a fix for mtu=0 devices.

4) A removal of superfluous code, pointed out by Eric Dumazet.

v2 fixes a logical problem in the patch for (3) pointed out by Eric Dumazet. v3
replaces some non-obvious bitmath in (3) with a more obvious expression, and
adds patch (4).

Jason A. Donenfeld (4):
  wireguard: selftests: reduce complexity and fix make races
  wireguard: receive: reset last_under_load to zero
  wireguard: send: account for mtu=0 devices
  wireguard: socket: remove extra call to synchronize_net

 drivers/net/wireguard/device.c                |  7 ++--
 drivers/net/wireguard/receive.c               |  7 +++-
 drivers/net/wireguard/send.c                  | 16 +++++---
 drivers/net/wireguard/socket.c                |  1 -
 .../testing/selftests/wireguard/qemu/Makefile | 38 +++++++------------
 5 files changed, 34 insertions(+), 35 deletions(-)

Comments

David Miller Feb. 17, 2020, 3:22 a.m. UTC | #1 
From: "Jason A. Donenfeld" <Jason@zx2c4.com>
Date: Fri, 14 Feb 2020 23:57:19 +0100

> Here are four fixes for wireguard collected since rc1:
> 
> 1) Some small cleanups to the test suite to help massively parallel
>    builds.
> 
> 2) A change in how we reset our load calculation to avoid a more
>    expensive comparison, suggested by Matt Dunwoodie.
> 
> 3) I've been loading more and more of wireguard's surface into
>    syzkaller, trying to get our coverage as complete as possible,
>    leading in this case to a fix for mtu=0 devices.
> 
> 4) A removal of superfluous code, pointed out by Eric Dumazet.
> 
> v2 fixes a logical problem in the patch for (3) pointed out by Eric Dumazet. v3
> replaces some non-obvious bitmath in (3) with a more obvious expression, and
> adds patch (4).

Series applied, thanks.