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, 40 insertions, 38 deletions

diff --git a/include/trace/events/rxrpc.h b/include/trace/events/rxrpc.h
index 284a1560b0a8..71ca74e40ec8 100644
--- a/include/trace/events/rxrpc.h
+++ b/include/trace/events/rxrpc.h
@@ -75,6 +75,7 @@
 	EM(rxrpc_call_got,			"GOT") \
 	EM(rxrpc_call_got_kernel,		"Gke") \
 	EM(rxrpc_call_got_timer,		"GTM") \
+	EM(rxrpc_call_got_tx,			"Gtx") \
 	EM(rxrpc_call_got_userid,		"Gus") \
 	EM(rxrpc_call_new_client,		"NWc") \
 	EM(rxrpc_call_new_service,		"NWs") \
@@ -83,20 +84,22 @@
 	EM(rxrpc_call_put_noqueue,		"PnQ") \
 	EM(rxrpc_call_put_notimer,		"PnT") \
 	EM(rxrpc_call_put_timer,		"PTM") \
+	EM(rxrpc_call_put_tx,			"Ptx") \
 	EM(rxrpc_call_put_userid,		"Pus") \
 	EM(rxrpc_call_queued,			"QUE") \
 	EM(rxrpc_call_queued_ref,		"QUR") \
 	EM(rxrpc_call_release,			"RLS") \
 	E_(rxrpc_call_seen,			"SEE")
 
-#define rxrpc_transmit_traces \
-	EM(rxrpc_transmit_await_reply,		"AWR") \
-	EM(rxrpc_transmit_end,			"END") \
-	EM(rxrpc_transmit_queue,		"QUE") \
-	EM(rxrpc_transmit_queue_last,		"QLS") \
-	EM(rxrpc_transmit_rotate,		"ROT") \
-	EM(rxrpc_transmit_rotate_last,		"RLS") \
-	E_(rxrpc_transmit_wait,			"WAI")
+#define rxrpc_txqueue_traces \
+	EM(rxrpc_txqueue_await_reply,		"AWR") \
+	EM(rxrpc_txqueue_dequeue,		"DEQ") \
+	EM(rxrpc_txqueue_end,			"END") \
+	EM(rxrpc_txqueue_queue,			"QUE") \
+	EM(rxrpc_txqueue_queue_last,		"QLS") \
+	EM(rxrpc_txqueue_rotate,		"ROT") \
+	EM(rxrpc_txqueue_rotate_last,		"RLS") \
+	E_(rxrpc_txqueue_wait,			"WAI")
 
 #define rxrpc_receive_traces \
 	EM(rxrpc_receive_end,			"END") \
@@ -259,6 +262,7 @@
 	EM(rxrpc_txbuf_alloc_ack,		"ALLOC ACK  ")	\
 	EM(rxrpc_txbuf_alloc_data,		"ALLOC DATA ")	\
 	EM(rxrpc_txbuf_free,			"FREE       ")	\
+	EM(rxrpc_txbuf_get_buffer,		"GET BUFFER ")	\
 	EM(rxrpc_txbuf_get_trans,		"GET TRANS  ")	\
 	EM(rxrpc_txbuf_get_retrans,		"GET RETRANS")	\
 	EM(rxrpc_txbuf_put_ack_tx,		"PUT ACK TX ")	\
@@ -266,6 +270,7 @@
 	EM(rxrpc_txbuf_put_nomem,		"PUT NOMEM  ")	\
 	EM(rxrpc_txbuf_put_rotated,		"PUT ROTATED")	\
 	EM(rxrpc_txbuf_put_send_aborted,	"PUT SEND-X ")	\
+	EM(rxrpc_txbuf_put_trans,		"PUT TRANS  ")	\
 	EM(rxrpc_txbuf_see_send_more,		"SEE SEND+  ")	\
 	E_(rxrpc_txbuf_see_unacked,		"SEE UNACKED")
 
@@ -295,9 +300,9 @@ enum rxrpc_rtt_rx_trace		{ rxrpc_rtt_rx_traces } __mode(byte);
 enum rxrpc_rtt_tx_trace		{ rxrpc_rtt_tx_traces } __mode(byte);
 enum rxrpc_skb_trace		{ rxrpc_skb_traces } __mode(byte);
 enum rxrpc_timer_trace		{ rxrpc_timer_traces } __mode(byte);
-enum rxrpc_transmit_trace	{ rxrpc_transmit_traces } __mode(byte);
 enum rxrpc_tx_point		{ rxrpc_tx_points } __mode(byte);
 enum rxrpc_txbuf_trace		{ rxrpc_txbuf_traces } __mode(byte);
+enum rxrpc_txqueue_trace	{ rxrpc_txqueue_traces } __mode(byte);
 
 #endif /* end __RXRPC_DECLARE_TRACE_ENUMS_ONCE_ONLY */
 
@@ -323,9 +328,9 @@ rxrpc_rtt_rx_traces;
 rxrpc_rtt_tx_traces;
 rxrpc_skb_traces;
 rxrpc_timer_traces;
-rxrpc_transmit_traces;
 rxrpc_tx_points;
 rxrpc_txbuf_traces;
+rxrpc_txqueue_traces;
 
 /*
  * Now redefine the EM() and E_() macros to map the enums to the strings that
@@ -605,15 +610,16 @@ TRACE_EVENT(rxrpc_call_complete,
 		      __entry->abort_code)
 	    );
 
-TRACE_EVENT(rxrpc_transmit,
-	    TP_PROTO(struct rxrpc_call *call, enum rxrpc_transmit_trace why),
+TRACE_EVENT(rxrpc_txqueue,
+	    TP_PROTO(struct rxrpc_call *call, enum rxrpc_txqueue_trace why),
 
 	    TP_ARGS(call, why),
 
 	    TP_STRUCT__entry(
 		    __field(unsigned int,		call		)
-		    __field(enum rxrpc_transmit_trace,	why		)
-		    __field(rxrpc_seq_t,		tx_hard_ack	)
+		    __field(enum rxrpc_txqueue_trace,	why		)
+		    __field(rxrpc_seq_t,		acks_hard_ack	)
+		    __field(rxrpc_seq_t,		tx_bottom	)
 		    __field(rxrpc_seq_t,		tx_top		)
 		    __field(int,			tx_winsize	)
 			     ),
@@ -621,16 +627,19 @@ TRACE_EVENT(rxrpc_transmit,
 	    TP_fast_assign(
 		    __entry->call = call->debug_id;
 		    __entry->why = why;
-		    __entry->tx_hard_ack = call->tx_hard_ack;
+		    __entry->acks_hard_ack = call->acks_hard_ack;
+		    __entry->tx_bottom = call->tx_bottom;
 		    __entry->tx_top = call->tx_top;
 		    __entry->tx_winsize = call->tx_winsize;
 			   ),
 
-	    TP_printk("c=%08x %s f=%08x n=%u/%u",
+	    TP_printk("c=%08x %s f=%08x h=%08x n=%u/%u/%u",
 		      __entry->call,
-		      __print_symbolic(__entry->why, rxrpc_transmit_traces),
-		      __entry->tx_hard_ack + 1,
-		      __entry->tx_top - __entry->tx_hard_ack,
+		      __print_symbolic(__entry->why, rxrpc_txqueue_traces),
+		      __entry->tx_bottom,
+		      __entry->acks_hard_ack,
+		      __entry->tx_top - __entry->tx_bottom,
+		      __entry->tx_top - __entry->acks_hard_ack,
 		      __entry->tx_winsize)
 	    );
 
@@ -1200,29 +1209,25 @@ TRACE_EVENT(rxrpc_drop_ack,
 	    );
 
 TRACE_EVENT(rxrpc_retransmit,
-	    TP_PROTO(struct rxrpc_call *call, rxrpc_seq_t seq, u8 annotation,
-		     s64 expiry),
+	    TP_PROTO(struct rxrpc_call *call, rxrpc_seq_t seq, s64 expiry),
 
-	    TP_ARGS(call, seq, annotation, expiry),
+	    TP_ARGS(call, seq, expiry),
 
 	    TP_STRUCT__entry(
 		    __field(unsigned int,		call		)
 		    __field(rxrpc_seq_t,		seq		)
-		    __field(u8,				annotation	)
 		    __field(s64,			expiry		)
 			     ),
 
 	    TP_fast_assign(
 		    __entry->call = call->debug_id;
 		    __entry->seq = seq;
-		    __entry->annotation = annotation;
 		    __entry->expiry = expiry;
 			   ),
 
-	    TP_printk("c=%08x q=%x a=%02x xp=%lld",
+	    TP_printk("c=%08x q=%x xp=%lld",
 		      __entry->call,
 		      __entry->seq,
-		      __entry->annotation,
 		      __entry->expiry)
 	    );
 
@@ -1245,14 +1250,14 @@ TRACE_EVENT(rxrpc_congest,
 	    TP_fast_assign(
 		    __entry->call	= call->debug_id;
 		    __entry->change	= change;
-		    __entry->hard_ack	= call->tx_hard_ack;
+		    __entry->hard_ack	= call->acks_hard_ack;
 		    __entry->top	= call->tx_top;
 		    __entry->lowest_nak	= call->acks_lowest_nak;
 		    __entry->ack_serial	= ack_serial;
 		    memcpy(&__entry->sum, summary, sizeof(__entry->sum));
 			   ),
 
-	    TP_printk("c=%08x r=%08x %s q=%08x %s cw=%u ss=%u nr=%u,%u nw=%u,%u r=%u b=%u u=%u d=%u l=%x%s%s%s",
+	    TP_printk("c=%08x r=%08x %s q=%08x %s cw=%u ss=%u nA=%u,%u+%u,%u r=%u b=%u u=%u d=%u l=%x%s%s%s",
 		      __entry->call,
 		      __entry->ack_serial,
 		      __print_symbolic(__entry->sum.ack_reason, rxrpc_ack_names),
@@ -1362,26 +1367,23 @@ TRACE_EVENT(rxrpc_connect_call,
 	    );
 
 TRACE_EVENT(rxrpc_resend,
-	    TP_PROTO(struct rxrpc_call *call, int ix),
+	    TP_PROTO(struct rxrpc_call *call),
 
-	    TP_ARGS(call, ix),
+	    TP_ARGS(call),
 
 	    TP_STRUCT__entry(
 		    __field(unsigned int,		call		)
-		    __field(int,			ix		)
-		    __array(u8,				anno, 64	)
+		    __field(rxrpc_seq_t,		seq		)
 			     ),
 
 	    TP_fast_assign(
 		    __entry->call = call->debug_id;
-		    __entry->ix = ix;
-		    memcpy(__entry->anno, call->rxtx_annotations, 64);
+		    __entry->seq = call->acks_hard_ack;
 			   ),
 
-	    TP_printk("c=%08x ix=%u a=%64phN",
+	    TP_printk("c=%08x q=%x",
 		      __entry->call,
-		      __entry->ix,
-		      __entry->anno)
+		      __entry->seq)
 	    );
 
 TRACE_EVENT(rxrpc_rx_icmp,
@@ -1461,7 +1463,7 @@ TRACE_EVENT(rxrpc_call_reset,
 		    __entry->call_id = call->call_id;
 		    __entry->call_serial = call->rx_serial;
 		    __entry->conn_serial = call->conn->hi_serial;
-		    __entry->tx_seq = call->tx_hard_ack;
+		    __entry->tx_seq = call->acks_hard_ack;
 		    __entry->rx_seq = call->rx_highest_seq;
 			   ),