From ba4e103848d3a2a28a0445e39f4a9564187efe54 Mon Sep 17 00:00:00 2001 From: David Howells Date: Fri, 3 May 2024 16:07:39 +0100 Subject: rxrpc: Fix congestion control algorithm Make the following fixes to the congestion control algorithm: (1) Don't vary the cwnd starting value by the size of RXRPC_TX_SMSS since that's currently held constant - set to the size of a jumbo subpacket payload so that we can create jumbo packets on the fly. The current code invariably picks 3 as the starting value. Further, the starting cwnd needs to be an even number because we ack every other packet, so set it to 4. (2) Don't cut ssthresh when we see an ACK come from the peer with a receive window (rwind) less than ssthresh. ssthresh keeps track of characteristics of the connection whereas rwind may be reduced by the peer for any reason - and may be reduced to 0. Fixes: 1fc4fa2ac93d ("rxrpc: Fix congestion management") Fixes: 0851115090a3 ("rxrpc: Reduce ssthresh to peer's receive window") Signed-off-by: David Howells Suggested-by: Simon Wilkinson cc: Marc Dionne cc: linux-afs@lists.infradead.org Reviewed-by: Jeffrey Altman > Link: https://lore.kernel.org/r/20240503150749.1001323-2-dhowells@redhat.com Signed-off-by: Jakub Kicinski --- net/rxrpc/ar-internal.h | 2 +- net/rxrpc/call_object.c | 7 +------ net/rxrpc/input.c | 3 --- 3 files changed, 2 insertions(+), 10 deletions(-) (limited to 'net/rxrpc') diff --git a/net/rxrpc/ar-internal.h b/net/rxrpc/ar-internal.h index 08c0a32db8c7..08de24658f4f 100644 --- a/net/rxrpc/ar-internal.h +++ b/net/rxrpc/ar-internal.h @@ -697,7 +697,7 @@ struct rxrpc_call { * packets) rather than bytes. */ #define RXRPC_TX_SMSS RXRPC_JUMBO_DATALEN -#define RXRPC_MIN_CWND (RXRPC_TX_SMSS > 2190 ? 2 : RXRPC_TX_SMSS > 1095 ? 3 : 4) +#define RXRPC_MIN_CWND 4 u8 cong_cwnd; /* Congestion window size */ u8 cong_extra; /* Extra to send for congestion management */ u8 cong_ssthresh; /* Slow-start threshold */ diff --git a/net/rxrpc/call_object.c b/net/rxrpc/call_object.c index 01fa71e8b1f7..f9e983a12c14 100644 --- a/net/rxrpc/call_object.c +++ b/net/rxrpc/call_object.c @@ -174,12 +174,7 @@ struct rxrpc_call *rxrpc_alloc_call(struct rxrpc_sock *rx, gfp_t gfp, call->rx_winsize = rxrpc_rx_window_size; call->tx_winsize = 16; - if (RXRPC_TX_SMSS > 2190) - call->cong_cwnd = 2; - else if (RXRPC_TX_SMSS > 1095) - call->cong_cwnd = 3; - else - call->cong_cwnd = 4; + call->cong_cwnd = RXRPC_MIN_CWND; call->cong_ssthresh = RXRPC_TX_MAX_WINDOW; call->rxnet = rxnet; diff --git a/net/rxrpc/input.c b/net/rxrpc/input.c index 3dedb8c0618c..860075f1290b 100644 --- a/net/rxrpc/input.c +++ b/net/rxrpc/input.c @@ -685,9 +685,6 @@ static void rxrpc_input_ack_trailer(struct rxrpc_call *call, struct sk_buff *skb call->tx_winsize = rwind; } - if (call->cong_ssthresh > rwind) - call->cong_ssthresh = rwind; - mtu = min(ntohl(trailer->maxMTU), ntohl(trailer->ifMTU)); peer = call->peer; -- cgit v1.2.3 From 012b7206918dcc5a4dcf1432b3e643114c95957e Mon Sep 17 00:00:00 2001 From: David Howells Date: Fri, 3 May 2024 16:07:40 +0100 Subject: rxrpc: Only transmit one ACK per jumbo packet received Only generate one ACK packet for all the subpackets in a jumbo packet. If we would like to generate more than one ACK, we prioritise them base on their reason code, in the order, highest first: OutOfSeq > NoSpace > ExceedsWin > Duplicate > Requested > Delay > Idle For the first four, we reference the lowest offending subpacket; for the last three, the highest. This reduces the number of ACKs we end up transmitting to one per UDP packet transmitted to reduce network loading and packet parsing. Fixes: 5d7edbc9231e ("rxrpc: Get rid of the Rx ring") Signed-off-by: David Howells cc: Marc Dionne cc: linux-afs@lists.infradead.org Reviewed-by: Jeffrey Altman > Link: https://lore.kernel.org/r/20240503150749.1001323-3-dhowells@redhat.com Signed-off-by: Jakub Kicinski --- net/rxrpc/input.c | 46 +++++++++++++++++++++++++++++++++++----------- 1 file changed, 35 insertions(+), 11 deletions(-) (limited to 'net/rxrpc') diff --git a/net/rxrpc/input.c b/net/rxrpc/input.c index 860075f1290b..16d49a861dbb 100644 --- a/net/rxrpc/input.c +++ b/net/rxrpc/input.c @@ -9,6 +9,17 @@ #include "ar-internal.h" +/* Override priority when generating ACKs for received DATA */ +static const u8 rxrpc_ack_priority[RXRPC_ACK__INVALID] = { + [RXRPC_ACK_IDLE] = 1, + [RXRPC_ACK_DELAY] = 2, + [RXRPC_ACK_REQUESTED] = 3, + [RXRPC_ACK_DUPLICATE] = 4, + [RXRPC_ACK_EXCEEDS_WINDOW] = 5, + [RXRPC_ACK_NOSPACE] = 6, + [RXRPC_ACK_OUT_OF_SEQUENCE] = 7, +}; + static void rxrpc_proto_abort(struct rxrpc_call *call, rxrpc_seq_t seq, enum rxrpc_abort_reason why) { @@ -365,7 +376,7 @@ static void rxrpc_input_queue_data(struct rxrpc_call *call, struct sk_buff *skb, * Process a DATA packet. */ static void rxrpc_input_data_one(struct rxrpc_call *call, struct sk_buff *skb, - bool *_notify) + bool *_notify, rxrpc_serial_t *_ack_serial, int *_ack_reason) { struct rxrpc_skb_priv *sp = rxrpc_skb(skb); struct sk_buff *oos; @@ -418,8 +429,6 @@ static void rxrpc_input_data_one(struct rxrpc_call *call, struct sk_buff *skb, /* Send an immediate ACK if we fill in a hole */ else if (!skb_queue_empty(&call->rx_oos_queue)) ack_reason = RXRPC_ACK_DELAY; - else - call->ackr_nr_unacked++; window++; if (after(window, wtop)) { @@ -497,12 +506,16 @@ static void rxrpc_input_data_one(struct rxrpc_call *call, struct sk_buff *skb, } send_ack: - if (ack_reason >= 0) - rxrpc_send_ACK(call, ack_reason, serial, - rxrpc_propose_ack_input_data); - else - rxrpc_propose_delay_ACK(call, serial, - rxrpc_propose_ack_input_data); + if (ack_reason >= 0) { + if (rxrpc_ack_priority[ack_reason] > rxrpc_ack_priority[*_ack_reason]) { + *_ack_serial = serial; + *_ack_reason = ack_reason; + } else if (rxrpc_ack_priority[ack_reason] == rxrpc_ack_priority[*_ack_reason] && + ack_reason == RXRPC_ACK_REQUESTED) { + *_ack_serial = serial; + *_ack_reason = ack_reason; + } + } } /* @@ -513,9 +526,11 @@ static bool rxrpc_input_split_jumbo(struct rxrpc_call *call, struct sk_buff *skb struct rxrpc_jumbo_header jhdr; struct rxrpc_skb_priv *sp = rxrpc_skb(skb), *jsp; struct sk_buff *jskb; + rxrpc_serial_t ack_serial = 0; unsigned int offset = sizeof(struct rxrpc_wire_header); unsigned int len = skb->len - offset; bool notify = false; + int ack_reason = 0; while (sp->hdr.flags & RXRPC_JUMBO_PACKET) { if (len < RXRPC_JUMBO_SUBPKTLEN) @@ -535,7 +550,7 @@ static bool rxrpc_input_split_jumbo(struct rxrpc_call *call, struct sk_buff *skb jsp = rxrpc_skb(jskb); jsp->offset = offset; jsp->len = RXRPC_JUMBO_DATALEN; - rxrpc_input_data_one(call, jskb, ¬ify); + rxrpc_input_data_one(call, jskb, ¬ify, &ack_serial, &ack_reason); rxrpc_free_skb(jskb, rxrpc_skb_put_jumbo_subpacket); sp->hdr.flags = jhdr.flags; @@ -548,7 +563,16 @@ static bool rxrpc_input_split_jumbo(struct rxrpc_call *call, struct sk_buff *skb sp->offset = offset; sp->len = len; - rxrpc_input_data_one(call, skb, ¬ify); + rxrpc_input_data_one(call, skb, ¬ify, &ack_serial, &ack_reason); + + if (ack_reason > 0) { + rxrpc_send_ACK(call, ack_reason, ack_serial, + rxrpc_propose_ack_input_data); + } else { + call->ackr_nr_unacked++; + rxrpc_propose_delay_ACK(call, sp->hdr.serial, + rxrpc_propose_ack_input_data); + } if (notify) { trace_rxrpc_notify_socket(call->debug_id, sp->hdr.serial); rxrpc_notify_socket(call); -- cgit v1.2.3