udp: be less conservative with sock rmem accounting
authorPaolo Abeni <pabeni@redhat.com>
Fri, 2 Dec 2016 16:35:49 +0000 (17:35 +0100)
committerDavid S. Miller <davem@davemloft.net>
Sat, 3 Dec 2016 21:14:48 +0000 (16:14 -0500)
commit363dc73acacbbcdae98acf5612303e9770e04b1d
tree68077749dd6983143b8c58a8167ccbdb2448f7c7
parent12efa1fa43968fcb707e806dc119df499c17ac2c
udp: be less conservative with sock rmem accounting

Before commit 850cbaddb52d ("udp: use it's own memory accounting
schema"), the udp protocol allowed sk_rmem_alloc to grow beyond
the rcvbuf by the whole current packet's truesize. After said commit
we allow sk_rmem_alloc to exceed the rcvbuf only if the receive queue
is empty. As reported by Jesper this cause a performance regression
for some (small) values of rcvbuf.

This commit is intended to fix the regression restoring the old
handling of the rcvbuf limit.

Reported-by: Jesper Dangaard Brouer <brouer@redhat.com>
Fixes: 850cbaddb52d ("udp: use it's own memory accounting schema")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/udp.c