rxrpc: Don't use a ring buffer for call Tx queue

Change the way the Tx queueing works to make the following ends easier to
achieve:

 (1) The filling of packets, the encryption of packets and the transmission
     of packets can be handled in parallel by separate threads, rather than
     rxrpc_sendmsg() allocating, filling, encrypting and transmitting each
     packet before moving onto the next one.

 (2) Get rid of the fixed-size ring which sets a hard limit on the number
     of packets that can be retained in the ring.  This allows the number
     of packets to increase without having to allocate a very large ring or
     having variable-sized rings.

     [Note: the downside of this is that it's then less efficient to locate
     a packet for retransmission as we then have to step through a list and
     examine each buffer in the list.]

 (3) Allow the filler/encrypter to run ahead of the transmission window.

 (4) Make it easier to do zero copy UDP from the packet buffers.

 (5) Make it easier to do zero copy from userspace to the packet buffers -
     and thence to UDP (only if for unauthenticated connections).

To that end, the following changes are made:

 (1) Use the new rxrpc_txbuf struct instead of sk_buff for keeping packets
     to be transmitted in.  This allows them to be placed on multiple
     queues simultaneously.  An sk_buff isn't really necessary as it's
     never passed on to lower-level networking code.

 (2) Keep the transmissable packets in a linked list on the call struct
     rather than in a ring.  As a consequence, the annotation buffer isn't
     used either; rather a flag is set on the packet to indicate ackedness.

 (3) Use the RXRPC_CALL_TX_LAST flag to indicate that the last packet to be
     transmitted has been queued.  Add RXRPC_CALL_TX_ALL_ACKED to indicate
     that all packets up to and including the last got hard acked.

 (4) Wire headers are now stored in the txbuf rather than being concocted
     on the stack and they're stored immediately before the data, thereby
     allowing zerocopy of a single span.

 (5) Don't bother with instant-resend on transmission failure; rather,
     leave it for a timer or an ACK packet to trigger.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

1 files changed, 37 insertions, 74 deletions

diff --git a/net/rxrpc/call_event.c b/net/rxrpc/call_event.c
index 36f60ac1d95d..3c37b280eb20 100644
--- a/net/rxrpc/call_event.c
+++ b/net/rxrpc/call_event.c
@@ -148,62 +148,52 @@ static void rxrpc_congestion_timeout(struct rxrpc_call *call)
  */
 static void rxrpc_resend(struct rxrpc_call *call, unsigned long now_j)
 {
-	struct sk_buff *skb;
+	struct rxrpc_txbuf *txb;
 	unsigned long resend_at;
-	rxrpc_seq_t cursor, seq, top;
+	rxrpc_seq_t transmitted = READ_ONCE(call->tx_transmitted);
 	ktime_t now, max_age, oldest, ack_ts;
-	int ix;
-	u8 annotation, anno_type, retrans = 0, unacked = 0;
+	bool unacked = false;
+	LIST_HEAD(retrans_queue);
 
-	_enter("{%d,%d}", call->tx_hard_ack, call->tx_top);
+	_enter("{%d,%d}", call->acks_hard_ack, call->tx_top);
 
 	now = ktime_get_real();
 	max_age = ktime_sub_us(now, jiffies_to_usecs(call->peer->rto_j));
 
-	spin_lock_bh(&call->lock);
-
-	cursor = call->tx_hard_ack;
-	top = call->tx_top;
-	ASSERT(before_eq(cursor, top));
-	if (cursor == top)
-		goto out_unlock;
+	spin_lock(&call->tx_lock);
 
 	/* Scan the packet list without dropping the lock and decide which of
 	 * the packets in the Tx buffer we're going to resend and what the new
 	 * resend timeout will be.
 	 */
-	trace_rxrpc_resend(call, (cursor + 1) & RXRPC_RXTX_BUFF_MASK);
+	trace_rxrpc_resend(call);
 	oldest = now;
-	for (seq = cursor + 1; before_eq(seq, top); seq++) {
-		ix = seq & RXRPC_RXTX_BUFF_MASK;
-		annotation = call->rxtx_annotations[ix];
-		anno_type = annotation & RXRPC_TX_ANNO_MASK;
-		annotation &= ~RXRPC_TX_ANNO_MASK;
-		if (anno_type == RXRPC_TX_ANNO_ACK)
+	list_for_each_entry(txb, &call->tx_buffer, call_link) {
+		if (test_bit(RXRPC_TXBUF_ACKED, &txb->flags))
 			continue;
+		if (after(txb->seq, transmitted))
+			break;
 
-		skb = call->rxtx_buffer[ix];
-		rxrpc_see_skb(skb, rxrpc_skb_seen);
+		rxrpc_see_txbuf(txb, rxrpc_txbuf_see_unacked);
 
-		if (anno_type == RXRPC_TX_ANNO_UNACK) {
-			if (ktime_after(skb->tstamp, max_age)) {
-				if (ktime_before(skb->tstamp, oldest))
-					oldest = skb->tstamp;
+		if (test_bit(RXRPC_TXBUF_RESENT, &txb->flags)) {
+			if (ktime_after(txb->last_sent, max_age)) {
+				if (ktime_before(txb->last_sent, oldest))
+					oldest = txb->last_sent;
 				continue;
 			}
-			if (!(annotation & RXRPC_TX_ANNO_RESENT))
-				unacked++;
+			unacked = true;
 		}
 
-		/* Okay, we need to retransmit a packet. */
-		call->rxtx_annotations[ix] = RXRPC_TX_ANNO_RETRANS | annotation;
-		retrans++;
-		trace_rxrpc_retransmit(call, seq, annotation | anno_type,
-				       ktime_to_ns(ktime_sub(skb->tstamp, max_age)));
+		rxrpc_get_txbuf(txb, rxrpc_txbuf_get_retrans);
+		list_move_tail(&txb->tx_link, &retrans_queue);
 	}
 
+	spin_unlock(&call->tx_lock);
+
 	resend_at = nsecs_to_jiffies(ktime_to_ns(ktime_sub(now, oldest)));
-	resend_at += jiffies + rxrpc_get_rto_backoff(call->peer, retrans);
+	resend_at += jiffies + rxrpc_get_rto_backoff(call->peer,
+						     !list_empty(&retrans_queue));
 	WRITE_ONCE(call->resend_at, resend_at);
 
 	if (unacked)
@@ -213,7 +203,8 @@ static void rxrpc_resend(struct rxrpc_call *call, unsigned long now_j)
 	 * that an ACK got lost somewhere.  Send a ping to find out instead of
 	 * retransmitting data.
 	 */
-	if (!retrans) {
+	if (list_empty(&retrans_queue)) {
+		spin_lock_bh(&call->lock);
 		rxrpc_reduce_call_timer(call, resend_at, now_j,
 					rxrpc_timer_set_for_resend);
 		spin_unlock_bh(&call->lock);
@@ -225,50 +216,19 @@ static void rxrpc_resend(struct rxrpc_call *call, unsigned long now_j)
 		goto out;
 	}
 
-	/* Now go through the Tx window and perform the retransmissions.  We
-	 * have to drop the lock for each send.  If an ACK comes in whilst the
-	 * lock is dropped, it may clear some of the retransmission markers for
-	 * packets that it soft-ACKs.
-	 */
-	for (seq = cursor + 1; before_eq(seq, top); seq++) {
-		ix = seq & RXRPC_RXTX_BUFF_MASK;
-		annotation = call->rxtx_annotations[ix];
-		anno_type = annotation & RXRPC_TX_ANNO_MASK;
-		if (anno_type != RXRPC_TX_ANNO_RETRANS)
-			continue;
-
-		/* We need to reset the retransmission state, but we need to do
-		 * so before we drop the lock as a new ACK/NAK may come in and
-		 * confuse things
-		 */
-		annotation &= ~RXRPC_TX_ANNO_MASK;
-		annotation |= RXRPC_TX_ANNO_UNACK | RXRPC_TX_ANNO_RESENT;
-		call->rxtx_annotations[ix] = annotation;
-
-		skb = call->rxtx_buffer[ix];
-		if (!skb)
-			continue;
-
-		rxrpc_get_skb(skb, rxrpc_skb_got);
-		spin_unlock_bh(&call->lock);
-
+	while ((txb = list_first_entry_or_null(&retrans_queue,
+					       struct rxrpc_txbuf, tx_link))) {
+		list_del_init(&txb->tx_link);
+		set_bit(RXRPC_TXBUF_RESENT, &txb->flags);
 		rxrpc_inc_stat(call->rxnet, stat_tx_data_retrans);
-		if (rxrpc_send_data_packet(call, skb, true) < 0) {
-			rxrpc_free_skb(skb, rxrpc_skb_freed);
-			return;
-		}
+		rxrpc_send_data_packet(call, txb);
+		rxrpc_put_txbuf(txb, rxrpc_txbuf_put_trans);
 
-		if (rxrpc_is_client_call(call))
-			rxrpc_expose_client_call(call);
-
-		rxrpc_free_skb(skb, rxrpc_skb_freed);
-		spin_lock_bh(&call->lock);
-		if (after(call->tx_hard_ack, seq))
-			seq = call->tx_hard_ack;
+		trace_rxrpc_retransmit(call, txb->seq,
+				       ktime_to_ns(ktime_sub(txb->last_sent,
+							     max_age)));
 	}
 
-out_unlock:
-	spin_unlock_bh(&call->lock);
 out:
 	_leave("");
 }
@@ -301,6 +261,9 @@ recheck_state:
 		goto recheck_state;
 	}
 
+	if (READ_ONCE(call->acks_hard_ack) != call->tx_bottom)
+		rxrpc_shrink_call_tx_buffer(call);
+
 	if (call->state == RXRPC_CALL_COMPLETE) {
 		rxrpc_delete_call_timer(call);
 		goto out_put;