[net,0/4] wireguard fixes for 5.7-rc7

Hi Dave,

Hopefully these are the last fixes for 5.7:

1) A trivial bump in the selftest harness to support gcc-10.
   build.wireguard.com is still on gcc-9 but I'll probably switch to
   gcc-10 in the coming weeks.

2) A concurrency fix regarding userspace modifying the pre-shared key at
   the same time as packets are being processed, reported by Matt
   Dunwoodie.

3) We were previously clearing skb->hash on egress, which broke
   fq_codel, cake, and other things that actually make use of the flow
   hash for queueing, reported by Dave Taht and Toke Høiland-Jørgensen.

4) A fix for the increased memory usage caused by (3). This can be
   thought of as part of patch (3), but because of the separate
   reasoning and breadth of it I thought made it a bit cleaner to put in
   a standalone commit.

Fixes (2), (3), and (4) are -stable material.

Thanks,
Jason

Jason A. Donenfeld (4):
  wireguard: selftests: use newer iproute2 for gcc-10
  wireguard: noise: read preshared key while taking lock
  wireguard: queueing: preserve flow hash across packet scrubbing
  wireguard: noise: separate receive counter from send counter

 drivers/net/wireguard/messages.h              |  2 +-
 drivers/net/wireguard/noise.c                 | 22 ++++------
 drivers/net/wireguard/noise.h                 | 14 +++---
 drivers/net/wireguard/queueing.h              | 10 ++++-
 drivers/net/wireguard/receive.c               | 44 +++++++++----------
 drivers/net/wireguard/selftest/counter.c      | 17 ++++---
 drivers/net/wireguard/send.c                  | 19 ++++----
 .../testing/selftests/wireguard/qemu/Makefile |  2 +-
 8 files changed, 71 insertions(+), 59 deletions(-)

From: "Jason A. Donenfeld" <Jason@zx2c4.com>
Date: Tue, 19 May 2020 22:49:26 -0600

> Hopefully these are the last fixes for 5.7:
> 
> 1) A trivial bump in the selftest harness to support gcc-10.
>    build.wireguard.com is still on gcc-9 but I'll probably switch to
>    gcc-10 in the coming weeks.
> 
> 2) A concurrency fix regarding userspace modifying the pre-shared key at
>    the same time as packets are being processed, reported by Matt
>    Dunwoodie.
> 
> 3) We were previously clearing skb->hash on egress, which broke
>    fq_codel, cake, and other things that actually make use of the flow
>    hash for queueing, reported by Dave Taht and Toke Høiland-Jørgensen.
> 
> 4) A fix for the increased memory usage caused by (3). This can be
>    thought of as part of patch (3), but because of the separate
>    reasoning and breadth of it I thought made it a bit cleaner to put in
>    a standalone commit.

Series applied.

> Fixes (2), (3), and (4) are -stable material.

Queued up for -stable, thanks.