diff options
| author | David Howells <dhowells@redhat.com> | 2022-11-11 16:47:35 +0300 |
|---|---|---|
| committer | David Howells <dhowells@redhat.com> | 2022-12-01 16:36:43 +0300 |
| commit | 5086d9a9dfec4866806da303115489b0606decb7 (patch) | |
| tree | 286a9b9c8cf80a3e853baa20cdf4bac9cc425718 /net/rxrpc/call_event.c | |
| parent | 32cf8edb079a6a687a2b5dba39a813a0bbd0ddf9 (diff) | |
| download | linux-5086d9a9dfec4866806da303115489b0606decb7.tar.xz | |
rxrpc: Move the cwnd degradation after transmitting packets
When we've gone for >1RTT without transmitting a packet, we should reduce
the ssthresh and cut the cwnd by half (as suggested in RFC2861 sec 3.1).
However, we may receive ACK packets in a batch and the first of these may
cut the cwnd, preventing further transmission, and each subsequent one cuts
the cwnd yet further, reducing it to the floor and killing performance.
Fix this by moving the cwnd reset to after doing the transmission and
resetting the base time such that we don't cut the cwnd by half again for
at least another RTT.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Diffstat (limited to 'net/rxrpc/call_event.c')
| -rw-r--r-- | net/rxrpc/call_event.c | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/net/rxrpc/call_event.c b/net/rxrpc/call_event.c index 9f1e490ab976..fd122e3726bd 100644 --- a/net/rxrpc/call_event.c +++ b/net/rxrpc/call_event.c @@ -427,6 +427,13 @@ void rxrpc_input_call_event(struct rxrpc_call *call, struct sk_buff *skb) rxrpc_transmit_some_data(call); + if (skb) { + struct rxrpc_skb_priv *sp = rxrpc_skb(skb); + + if (sp->hdr.type == RXRPC_PACKET_TYPE_ACK) + rxrpc_congestion_degrade(call); + } + if (test_and_clear_bit(RXRPC_CALL_EV_INITIAL_PING, &call->events)) rxrpc_send_initial_ping(call); |