tcp: refine tcp_prune_ofo_queue() to not drop all packets
authorEric Dumazet <edumazet@google.com>
Wed, 17 Aug 2016 21:17:09 +0000 (14:17 -0700)
committerDavid S. Miller <davem@davemloft.net>
Fri, 19 Aug 2016 06:36:07 +0000 (23:36 -0700)
commit36a6503feddadbbad415fb3891e80f94c10a9b21
tree2c42e7d2101b14c87d5e4867f32de41ccd8e2ee6
parente2d8f646c79f26e094bfaf9b21be614d1e148a67
tcp: refine tcp_prune_ofo_queue() to not drop all packets

Over the years, TCP BDP has increased a lot, and is typically
in the order of ~10 Mbytes with help of clever Congestion Control
modules.

In presence of packet losses, TCP stores incoming packets into an out of
order queue, and number of skbs sitting there waiting for the missing
packets to be received can match the BDP (~10 Mbytes)

In some cases, TCP needs to make room for incoming skbs, and current
strategy can simply remove all skbs in the out of order queue as a last
resort, incurring a huge penalty, both for receiver and sender.

Unfortunately these 'last resort events' are quite frequent, forcing
sender to send all packets again, stalling the flow and wasting a lot of
resources.

This patch cleans only a part of the out of order queue in order
to meet the memory constraints.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Cc: C. Stephen Gun <csg@google.com>
Cc: Van Jacobson <vanj@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_input.c