tun: Use netif_receive_skb instead of netif_rx
authorAndrey Konovalov <andreyknvl@google.com>
Thu, 1 Dec 2016 09:34:40 +0000 (10:34 +0100)
committerDavid S. Miller <davem@davemloft.net>
Wed, 7 Dec 2016 03:24:59 +0000 (22:24 -0500)
commitd4aea20d889e05575bb331a3dadf176176f7d631
treeb70dc69799ec8cf5db0b5d2ad24e52901e57b815
parentd10d0198708f3e17d10a83accb450f2ecba82883
tun: Use netif_receive_skb instead of netif_rx

This patch changes tun.c to call netif_receive_skb instead of netif_rx
when a packet is received (if CONFIG_4KSTACKS is not enabled to avoid
stack exhaustion). The difference between the two is that netif_rx queues
the packet into the backlog, and netif_receive_skb proccesses the packet
in the current context.

This patch is required for syzkaller [1] to collect coverage from packet
receive paths, when a packet being received through tun (syzkaller collects
coverage per process in the process context).

As mentioned by Eric this change also speeds up tun/tap. As measured by
Peter it speeds up his closed-loop single-stream tap/OVS benchmark by
about 23%, from 700k packets/second to 867k packets/second.

A similar patch was introduced back in 2010 [2, 3], but the author found
out that the patch doesn't help with the task he had in mind (for cgroups
to shape network traffic based on the original process) and decided not to
go further with it. The main concern back then was about possible stack
exhaustion with 4K stacks.

[1] https://github.com/google/syzkaller

[2] https://www.spinics.net/lists/netdev/thrd440.html#130570

[3] https://www.spinics.net/lists/netdev/msg130570.html

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/tun.c