1 /* Management of Tx window, Tx resend, ACKs and out-of-sequence reception
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 #include <linux/module.h>
13 #include <linux/circ_buf.h>
14 #include <linux/net.h>
15 #include <linux/skbuff.h>
16 #include <linux/udp.h>
18 #include <net/af_rxrpc.h>
19 #include "ar-internal.h"
21 static unsigned rxrpc_ack_defer
= 1;
23 static const char *rxrpc_acks
[] = {
24 "---", "REQ", "DUP", "OOS", "WIN", "MEM", "PNG", "PNR", "DLY", "IDL",
28 static const s8 rxrpc_ack_priority
[] = {
30 [RXRPC_ACK_DELAY
] = 1,
31 [RXRPC_ACK_REQUESTED
] = 2,
33 [RXRPC_ACK_PING_RESPONSE
] = 4,
34 [RXRPC_ACK_DUPLICATE
] = 5,
35 [RXRPC_ACK_OUT_OF_SEQUENCE
] = 6,
36 [RXRPC_ACK_EXCEEDS_WINDOW
] = 7,
37 [RXRPC_ACK_NOSPACE
] = 8,
41 * propose an ACK be sent
43 void __rxrpc_propose_ACK(struct rxrpc_call
*call
, uint8_t ack_reason
,
44 __be32 serial
, bool immediate
)
47 s8 prior
= rxrpc_ack_priority
[ack_reason
];
49 ASSERTCMP(prior
, >, 0);
51 _enter("{%d},%s,%%%x,%u",
52 call
->debug_id
, rxrpc_acks
[ack_reason
], ntohl(serial
),
55 if (prior
< rxrpc_ack_priority
[call
->ackr_reason
]) {
61 /* update DELAY, IDLE, REQUESTED and PING_RESPONSE ACK serial
63 if (prior
== rxrpc_ack_priority
[call
->ackr_reason
]) {
65 call
->ackr_serial
= serial
;
71 call
->ackr_reason
= ack_reason
;
72 call
->ackr_serial
= serial
;
76 _debug("run delay timer");
77 call
->ack_timer
.expires
= jiffies
+ rxrpc_ack_timeout
* HZ
;
78 add_timer(&call
->ack_timer
);
83 _debug("run defer timer");
89 case RXRPC_ACK_REQUESTED
:
92 if (!immediate
|| serial
== cpu_to_be32(1)) {
93 _debug("run defer timer");
94 expiry
= rxrpc_ack_defer
;
99 _debug("immediate ACK");
105 if (!timer_pending(&call
->ack_timer
) ||
106 time_after(call
->ack_timer
.expires
, expiry
))
107 mod_timer(&call
->ack_timer
, expiry
);
111 _debug("cancel timer %%%u", ntohl(serial
));
112 try_to_del_timer_sync(&call
->ack_timer
);
113 read_lock_bh(&call
->state_lock
);
114 if (call
->state
<= RXRPC_CALL_COMPLETE
&&
115 !test_and_set_bit(RXRPC_CALL_ACK
, &call
->events
))
116 rxrpc_queue_call(call
);
117 read_unlock_bh(&call
->state_lock
);
121 * propose an ACK be sent, locking the call structure
123 void rxrpc_propose_ACK(struct rxrpc_call
*call
, uint8_t ack_reason
,
124 __be32 serial
, bool immediate
)
126 s8 prior
= rxrpc_ack_priority
[ack_reason
];
128 if (prior
> rxrpc_ack_priority
[call
->ackr_reason
]) {
129 spin_lock_bh(&call
->lock
);
130 __rxrpc_propose_ACK(call
, ack_reason
, serial
, immediate
);
131 spin_unlock_bh(&call
->lock
);
136 * set the resend timer
138 static void rxrpc_set_resend(struct rxrpc_call
*call
, u8 resend
,
139 unsigned long resend_at
)
141 read_lock_bh(&call
->state_lock
);
142 if (call
->state
>= RXRPC_CALL_COMPLETE
)
146 _debug("SET RESEND");
147 set_bit(RXRPC_CALL_RESEND
, &call
->events
);
151 _debug("MODIFY RESEND TIMER");
152 set_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
153 mod_timer(&call
->resend_timer
, resend_at
);
155 _debug("KILL RESEND TIMER");
156 del_timer_sync(&call
->resend_timer
);
157 clear_bit(RXRPC_CALL_RESEND_TIMER
, &call
->events
);
158 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
160 read_unlock_bh(&call
->state_lock
);
166 static void rxrpc_resend(struct rxrpc_call
*call
)
168 struct rxrpc_skb_priv
*sp
;
169 struct rxrpc_header
*hdr
;
171 unsigned long *p_txb
, resend_at
;
175 _enter("{%d,%d,%d,%d},",
176 call
->acks_hard
, call
->acks_unacked
,
177 atomic_read(&call
->sequence
),
178 CIRC_CNT(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
));
184 for (loop
= call
->acks_tail
;
185 loop
!= call
->acks_head
|| stop
;
186 loop
= (loop
+ 1) & (call
->acks_winsz
- 1)
188 p_txb
= call
->acks_window
+ loop
;
189 smp_read_barrier_depends();
193 txb
= (struct sk_buff
*) *p_txb
;
196 if (sp
->need_resend
) {
199 /* each Tx packet has a new serial number */
201 htonl(atomic_inc_return(&call
->conn
->serial
));
203 hdr
= (struct rxrpc_header
*) txb
->head
;
204 hdr
->serial
= sp
->hdr
.serial
;
206 _proto("Tx DATA %%%u { #%d }",
207 ntohl(sp
->hdr
.serial
), ntohl(sp
->hdr
.seq
));
208 if (rxrpc_send_packet(call
->conn
->trans
, txb
) < 0) {
210 sp
->resend_at
= jiffies
+ 3;
213 jiffies
+ rxrpc_resend_timeout
* HZ
;
217 if (time_after_eq(jiffies
+ 1, sp
->resend_at
)) {
220 } else if (resend
& 2) {
221 if (time_before(sp
->resend_at
, resend_at
))
222 resend_at
= sp
->resend_at
;
224 resend_at
= sp
->resend_at
;
229 rxrpc_set_resend(call
, resend
, resend_at
);
234 * handle resend timer expiry
236 static void rxrpc_resend_timer(struct rxrpc_call
*call
)
238 struct rxrpc_skb_priv
*sp
;
240 unsigned long *p_txb
, resend_at
;
245 call
->acks_tail
, call
->acks_unacked
, call
->acks_head
);
250 for (loop
= call
->acks_unacked
;
251 loop
!= call
->acks_head
;
252 loop
= (loop
+ 1) & (call
->acks_winsz
- 1)
254 p_txb
= call
->acks_window
+ loop
;
255 smp_read_barrier_depends();
256 txb
= (struct sk_buff
*) (*p_txb
& ~1);
259 ASSERT(!(*p_txb
& 1));
261 if (sp
->need_resend
) {
263 } else if (time_after_eq(jiffies
+ 1, sp
->resend_at
)) {
266 } else if (resend
& 2) {
267 if (time_before(sp
->resend_at
, resend_at
))
268 resend_at
= sp
->resend_at
;
270 resend_at
= sp
->resend_at
;
275 rxrpc_set_resend(call
, resend
, resend_at
);
280 * process soft ACKs of our transmitted packets
281 * - these indicate packets the peer has or has not received, but hasn't yet
282 * given to the consumer, and so can still be discarded and re-requested
284 static int rxrpc_process_soft_ACKs(struct rxrpc_call
*call
,
285 struct rxrpc_ackpacket
*ack
,
288 struct rxrpc_skb_priv
*sp
;
290 unsigned long *p_txb
, resend_at
;
292 u8 sacks
[RXRPC_MAXACKS
], resend
;
294 _enter("{%d,%d},{%d},",
296 CIRC_CNT(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
),
299 if (skb_copy_bits(skb
, 0, sacks
, ack
->nAcks
) < 0)
304 for (loop
= 0; loop
< ack
->nAcks
; loop
++) {
305 p_txb
= call
->acks_window
;
306 p_txb
+= (call
->acks_tail
+ loop
) & (call
->acks_winsz
- 1);
307 smp_read_barrier_depends();
308 txb
= (struct sk_buff
*) (*p_txb
& ~1);
311 switch (sacks
[loop
]) {
312 case RXRPC_ACK_TYPE_ACK
:
316 case RXRPC_ACK_TYPE_NACK
:
322 _debug("Unsupported ACK type %d", sacks
[loop
]);
328 call
->acks_unacked
= (call
->acks_tail
+ loop
) & (call
->acks_winsz
- 1);
330 /* anything not explicitly ACK'd is implicitly NACK'd, but may just not
331 * have been received or processed yet by the far end */
332 for (loop
= call
->acks_unacked
;
333 loop
!= call
->acks_head
;
334 loop
= (loop
+ 1) & (call
->acks_winsz
- 1)
336 p_txb
= call
->acks_window
+ loop
;
337 smp_read_barrier_depends();
338 txb
= (struct sk_buff
*) (*p_txb
& ~1);
342 /* packet must have been discarded */
346 } else if (sp
->need_resend
) {
348 } else if (time_after_eq(jiffies
+ 1, sp
->resend_at
)) {
351 } else if (resend
& 2) {
352 if (time_before(sp
->resend_at
, resend_at
))
353 resend_at
= sp
->resend_at
;
355 resend_at
= sp
->resend_at
;
360 rxrpc_set_resend(call
, resend
, resend_at
);
365 _leave(" = -EPROTO");
370 * discard hard-ACK'd packets from the Tx window
372 static void rxrpc_rotate_tx_window(struct rxrpc_call
*call
, u32 hard
)
374 struct rxrpc_skb_priv
*sp
;
376 int tail
= call
->acks_tail
, old_tail
;
377 int win
= CIRC_CNT(call
->acks_head
, tail
, call
->acks_winsz
);
379 _enter("{%u,%u},%u", call
->acks_hard
, win
, hard
);
381 ASSERTCMP(hard
- call
->acks_hard
, <=, win
);
383 while (call
->acks_hard
< hard
) {
384 smp_read_barrier_depends();
385 _skb
= call
->acks_window
[tail
] & ~1;
386 sp
= rxrpc_skb((struct sk_buff
*) _skb
);
387 rxrpc_free_skb((struct sk_buff
*) _skb
);
389 tail
= (tail
+ 1) & (call
->acks_winsz
- 1);
390 call
->acks_tail
= tail
;
391 if (call
->acks_unacked
== old_tail
)
392 call
->acks_unacked
= tail
;
396 wake_up(&call
->tx_waitq
);
400 * clear the Tx window in the event of a failure
402 static void rxrpc_clear_tx_window(struct rxrpc_call
*call
)
404 rxrpc_rotate_tx_window(call
, atomic_read(&call
->sequence
));
408 * drain the out of sequence received packet queue into the packet Rx queue
410 static int rxrpc_drain_rx_oos_queue(struct rxrpc_call
*call
)
412 struct rxrpc_skb_priv
*sp
;
417 _enter("{%d,%d}", call
->rx_data_post
, call
->rx_first_oos
);
419 spin_lock_bh(&call
->lock
);
422 if (test_bit(RXRPC_CALL_RELEASED
, &call
->flags
))
423 goto socket_unavailable
;
425 skb
= skb_dequeue(&call
->rx_oos_queue
);
429 _debug("drain OOS packet %d [%d]",
430 ntohl(sp
->hdr
.seq
), call
->rx_first_oos
);
432 if (ntohl(sp
->hdr
.seq
) != call
->rx_first_oos
) {
433 skb_queue_head(&call
->rx_oos_queue
, skb
);
434 call
->rx_first_oos
= ntohl(rxrpc_skb(skb
)->hdr
.seq
);
435 _debug("requeue %p {%u}", skb
, call
->rx_first_oos
);
437 skb
->mark
= RXRPC_SKB_MARK_DATA
;
438 terminal
= ((sp
->hdr
.flags
& RXRPC_LAST_PACKET
) &&
439 !(sp
->hdr
.flags
& RXRPC_CLIENT_INITIATED
));
440 ret
= rxrpc_queue_rcv_skb(call
, skb
, true, terminal
);
442 _debug("drain #%u", call
->rx_data_post
);
443 call
->rx_data_post
++;
445 /* find out what the next packet is */
446 skb
= skb_peek(&call
->rx_oos_queue
);
449 ntohl(rxrpc_skb(skb
)->hdr
.seq
);
451 call
->rx_first_oos
= 0;
452 _debug("peek %p {%u}", skb
, call
->rx_first_oos
);
458 spin_unlock_bh(&call
->lock
);
459 _leave(" = %d", ret
);
464 * insert an out of sequence packet into the buffer
466 static void rxrpc_insert_oos_packet(struct rxrpc_call
*call
,
469 struct rxrpc_skb_priv
*sp
, *psp
;
474 seq
= ntohl(sp
->hdr
.seq
);
475 _enter(",,{%u}", seq
);
477 skb
->destructor
= rxrpc_packet_destructor
;
478 ASSERTCMP(sp
->call
, ==, NULL
);
480 rxrpc_get_call(call
);
482 /* insert into the buffer in sequence order */
483 spin_lock_bh(&call
->lock
);
485 skb_queue_walk(&call
->rx_oos_queue
, p
) {
487 if (ntohl(psp
->hdr
.seq
) > seq
) {
488 _debug("insert oos #%u before #%u",
489 seq
, ntohl(psp
->hdr
.seq
));
490 skb_insert(p
, skb
, &call
->rx_oos_queue
);
495 _debug("append oos #%u", seq
);
496 skb_queue_tail(&call
->rx_oos_queue
, skb
);
499 /* we might now have a new front to the queue */
500 if (call
->rx_first_oos
== 0 || seq
< call
->rx_first_oos
)
501 call
->rx_first_oos
= seq
;
503 read_lock(&call
->state_lock
);
504 if (call
->state
< RXRPC_CALL_COMPLETE
&&
505 call
->rx_data_post
== call
->rx_first_oos
) {
506 _debug("drain rx oos now");
507 set_bit(RXRPC_CALL_DRAIN_RX_OOS
, &call
->events
);
509 read_unlock(&call
->state_lock
);
511 spin_unlock_bh(&call
->lock
);
512 _leave(" [stored #%u]", call
->rx_first_oos
);
516 * clear the Tx window on final ACK reception
518 static void rxrpc_zap_tx_window(struct rxrpc_call
*call
)
520 struct rxrpc_skb_priv
*sp
;
522 unsigned long _skb
, *acks_window
;
523 uint8_t winsz
= call
->acks_winsz
;
526 acks_window
= call
->acks_window
;
527 call
->acks_window
= NULL
;
529 while (CIRC_CNT(call
->acks_head
, call
->acks_tail
, winsz
) > 0) {
530 tail
= call
->acks_tail
;
531 smp_read_barrier_depends();
532 _skb
= acks_window
[tail
] & ~1;
534 call
->acks_tail
= (call
->acks_tail
+ 1) & (winsz
- 1);
536 skb
= (struct sk_buff
*) _skb
;
538 _debug("+++ clear Tx %u", ntohl(sp
->hdr
.seq
));
546 * process the extra information that may be appended to an ACK packet
548 static void rxrpc_extract_ackinfo(struct rxrpc_call
*call
, struct sk_buff
*skb
,
549 unsigned latest
, int nAcks
)
551 struct rxrpc_ackinfo ackinfo
;
552 struct rxrpc_peer
*peer
;
555 if (skb_copy_bits(skb
, nAcks
+ 3, &ackinfo
, sizeof(ackinfo
)) < 0) {
556 _leave(" [no ackinfo]");
560 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
562 ntohl(ackinfo
.rxMTU
), ntohl(ackinfo
.maxMTU
),
563 ntohl(ackinfo
.rwind
), ntohl(ackinfo
.jumbo_max
));
565 mtu
= min(ntohl(ackinfo
.rxMTU
), ntohl(ackinfo
.maxMTU
));
567 peer
= call
->conn
->trans
->peer
;
568 if (mtu
< peer
->maxdata
) {
569 spin_lock_bh(&peer
->lock
);
571 peer
->mtu
= mtu
+ peer
->hdrsize
;
572 spin_unlock_bh(&peer
->lock
);
573 _net("Net MTU %u (maxdata %u)", peer
->mtu
, peer
->maxdata
);
578 * process packets in the reception queue
580 static int rxrpc_process_rx_queue(struct rxrpc_call
*call
,
583 struct rxrpc_ackpacket ack
;
584 struct rxrpc_skb_priv
*sp
;
593 skb
= skb_dequeue(&call
->rx_queue
);
597 _net("deferred skb %p", skb
);
601 _debug("process %s [st %d]", rxrpc_pkts
[sp
->hdr
.type
], call
->state
);
605 switch (sp
->hdr
.type
) {
606 /* data packets that wind up here have been received out of
607 * order, need security processing or are jumbo packets */
608 case RXRPC_PACKET_TYPE_DATA
:
609 _proto("OOSQ DATA %%%u { #%u }",
610 ntohl(sp
->hdr
.serial
), ntohl(sp
->hdr
.seq
));
612 /* secured packets must be verified and possibly decrypted */
613 if (rxrpc_verify_packet(call
, skb
, _abort_code
) < 0)
616 rxrpc_insert_oos_packet(call
, skb
);
617 goto process_further
;
619 /* partial ACK to process */
620 case RXRPC_PACKET_TYPE_ACK
:
621 if (skb_copy_bits(skb
, 0, &ack
, sizeof(ack
)) < 0) {
622 _debug("extraction failure");
625 if (!skb_pull(skb
, sizeof(ack
)))
628 latest
= ntohl(sp
->hdr
.serial
);
629 hard
= ntohl(ack
.firstPacket
);
630 tx
= atomic_read(&call
->sequence
);
632 _proto("Rx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
636 ntohl(ack
.previousPacket
),
638 rxrpc_acks
[ack
.reason
],
641 rxrpc_extract_ackinfo(call
, skb
, latest
, ack
.nAcks
);
643 if (ack
.reason
== RXRPC_ACK_PING
) {
644 _proto("Rx ACK %%%u PING Request", latest
);
645 rxrpc_propose_ACK(call
, RXRPC_ACK_PING_RESPONSE
,
646 sp
->hdr
.serial
, true);
649 /* discard any out-of-order or duplicate ACKs */
650 if (latest
- call
->acks_latest
<= 0) {
651 _debug("discard ACK %d <= %d",
652 latest
, call
->acks_latest
);
655 call
->acks_latest
= latest
;
657 if (call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
&&
658 call
->state
!= RXRPC_CALL_CLIENT_AWAIT_REPLY
&&
659 call
->state
!= RXRPC_CALL_SERVER_SEND_REPLY
&&
660 call
->state
!= RXRPC_CALL_SERVER_AWAIT_ACK
)
663 _debug("Tx=%d H=%u S=%d", tx
, call
->acks_hard
, call
->state
);
667 _debug("hard-ACK'd packet %d not transmitted"
673 if ((call
->state
== RXRPC_CALL_CLIENT_AWAIT_REPLY
||
674 call
->state
== RXRPC_CALL_SERVER_AWAIT_ACK
) &&
679 rxrpc_rotate_tx_window(call
, hard
- 1);
683 if (hard
- 1 + ack
.nAcks
> tx
) {
684 _debug("soft-ACK'd packet %d+%d not"
685 " transmitted (%d top)",
686 hard
- 1, ack
.nAcks
, tx
);
690 if (rxrpc_process_soft_ACKs(call
, &ack
, skb
) < 0)
695 /* complete ACK to process */
696 case RXRPC_PACKET_TYPE_ACKALL
:
699 /* abort and busy are handled elsewhere */
700 case RXRPC_PACKET_TYPE_BUSY
:
701 case RXRPC_PACKET_TYPE_ABORT
:
704 /* connection level events - also handled elsewhere */
705 case RXRPC_PACKET_TYPE_CHALLENGE
:
706 case RXRPC_PACKET_TYPE_RESPONSE
:
707 case RXRPC_PACKET_TYPE_DEBUG
:
711 /* if we've had a hard ACK that covers all the packets we've sent, then
712 * that ends that phase of the operation */
714 write_lock_bh(&call
->state_lock
);
715 _debug("ack all %d", call
->state
);
717 switch (call
->state
) {
718 case RXRPC_CALL_CLIENT_AWAIT_REPLY
:
719 call
->state
= RXRPC_CALL_CLIENT_RECV_REPLY
;
721 case RXRPC_CALL_SERVER_AWAIT_ACK
:
722 _debug("srv complete");
723 call
->state
= RXRPC_CALL_COMPLETE
;
726 case RXRPC_CALL_CLIENT_SEND_REQUEST
:
727 case RXRPC_CALL_SERVER_RECV_REQUEST
:
728 goto protocol_error_unlock
; /* can't occur yet */
730 write_unlock_bh(&call
->state_lock
);
731 goto discard
; /* assume packet left over from earlier phase */
734 write_unlock_bh(&call
->state_lock
);
736 /* if all the packets we sent are hard-ACK'd, then we can discard
737 * whatever we've got left */
738 _debug("clear Tx %d",
739 CIRC_CNT(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
));
741 del_timer_sync(&call
->resend_timer
);
742 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
743 clear_bit(RXRPC_CALL_RESEND_TIMER
, &call
->events
);
745 if (call
->acks_window
)
746 rxrpc_zap_tx_window(call
);
749 /* post the final ACK message for userspace to pick up */
751 skb
->mark
= RXRPC_SKB_MARK_FINAL_ACK
;
753 rxrpc_get_call(call
);
754 spin_lock_bh(&call
->lock
);
755 if (rxrpc_queue_rcv_skb(call
, skb
, true, true) < 0)
757 spin_unlock_bh(&call
->lock
);
758 goto process_further
;
763 goto process_further
;
765 protocol_error_unlock
:
766 write_unlock_bh(&call
->state_lock
);
769 _leave(" = -EPROTO");
774 * post a message to the socket Rx queue for recvmsg() to pick up
776 static int rxrpc_post_message(struct rxrpc_call
*call
, u32 mark
, u32 error
,
779 struct rxrpc_skb_priv
*sp
;
783 _enter("{%d,%lx},%u,%u,%d",
784 call
->debug_id
, call
->flags
, mark
, error
, fatal
);
786 /* remove timers and things for fatal messages */
788 del_timer_sync(&call
->resend_timer
);
789 del_timer_sync(&call
->ack_timer
);
790 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
793 if (mark
!= RXRPC_SKB_MARK_NEW_CALL
&&
794 !test_bit(RXRPC_CALL_HAS_USERID
, &call
->flags
)) {
795 _leave("[no userid]");
799 if (!test_bit(RXRPC_CALL_TERMINAL_MSG
, &call
->flags
)) {
800 skb
= alloc_skb(0, GFP_NOFS
);
809 memset(sp
, 0, sizeof(*sp
));
812 rxrpc_get_call(call
);
814 spin_lock_bh(&call
->lock
);
815 ret
= rxrpc_queue_rcv_skb(call
, skb
, true, fatal
);
816 spin_unlock_bh(&call
->lock
);
825 * handle background processing of incoming call packets and ACK / abort
828 void rxrpc_process_call(struct work_struct
*work
)
830 struct rxrpc_call
*call
=
831 container_of(work
, struct rxrpc_call
, processor
);
832 struct rxrpc_ackpacket ack
;
833 struct rxrpc_ackinfo ackinfo
;
834 struct rxrpc_header hdr
;
840 int genbit
, loop
, nbit
, ioc
, ret
, mtu
;
841 u32 abort_code
= RX_PROTOCOL_ERROR
;
844 //printk("\n--------------------\n");
845 _enter("{%d,%s,%lx} [%lu]",
846 call
->debug_id
, rxrpc_call_states
[call
->state
], call
->events
,
847 (jiffies
- call
->creation_jif
) / (HZ
/ 10));
849 if (test_and_set_bit(RXRPC_CALL_PROC_BUSY
, &call
->flags
)) {
850 _debug("XXXXXXXXXXXXX RUNNING ON MULTIPLE CPUS XXXXXXXXXXXXX");
854 /* there's a good chance we're going to have to send a message, so set
855 * one up in advance */
856 msg
.msg_name
= &call
->conn
->trans
->peer
->srx
.transport
.sin
;
857 msg
.msg_namelen
= sizeof(call
->conn
->trans
->peer
->srx
.transport
.sin
);
858 msg
.msg_control
= NULL
;
859 msg
.msg_controllen
= 0;
862 hdr
.epoch
= call
->conn
->epoch
;
864 hdr
.callNumber
= call
->call_id
;
866 hdr
.type
= RXRPC_PACKET_TYPE_ACK
;
867 hdr
.flags
= call
->conn
->out_clientflag
;
869 hdr
.securityIndex
= call
->conn
->security_ix
;
871 hdr
.serviceId
= call
->conn
->service_id
;
873 memset(iov
, 0, sizeof(iov
));
874 iov
[0].iov_base
= &hdr
;
875 iov
[0].iov_len
= sizeof(hdr
);
877 /* deal with events of a final nature */
878 if (test_bit(RXRPC_CALL_RELEASE
, &call
->events
)) {
879 rxrpc_release_call(call
);
880 clear_bit(RXRPC_CALL_RELEASE
, &call
->events
);
883 if (test_bit(RXRPC_CALL_RCVD_ERROR
, &call
->events
)) {
886 clear_bit(RXRPC_CALL_CONN_ABORT
, &call
->events
);
887 clear_bit(RXRPC_CALL_REJECT_BUSY
, &call
->events
);
888 clear_bit(RXRPC_CALL_ABORT
, &call
->events
);
890 error
= call
->conn
->trans
->peer
->net_error
;
891 _debug("post net error %d", error
);
893 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_NET_ERROR
,
896 clear_bit(RXRPC_CALL_RCVD_ERROR
, &call
->events
);
900 if (test_bit(RXRPC_CALL_CONN_ABORT
, &call
->events
)) {
901 ASSERTCMP(call
->state
, >, RXRPC_CALL_COMPLETE
);
903 clear_bit(RXRPC_CALL_REJECT_BUSY
, &call
->events
);
904 clear_bit(RXRPC_CALL_ABORT
, &call
->events
);
906 _debug("post conn abort");
908 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_LOCAL_ERROR
,
909 call
->conn
->error
, true) < 0)
911 clear_bit(RXRPC_CALL_CONN_ABORT
, &call
->events
);
915 if (test_bit(RXRPC_CALL_REJECT_BUSY
, &call
->events
)) {
916 hdr
.type
= RXRPC_PACKET_TYPE_BUSY
;
917 genbit
= RXRPC_CALL_REJECT_BUSY
;
921 if (test_bit(RXRPC_CALL_ABORT
, &call
->events
)) {
922 ASSERTCMP(call
->state
, >, RXRPC_CALL_COMPLETE
);
924 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_LOCAL_ERROR
,
925 ECONNABORTED
, true) < 0)
927 hdr
.type
= RXRPC_PACKET_TYPE_ABORT
;
928 data
= htonl(call
->abort_code
);
929 iov
[1].iov_base
= &data
;
930 iov
[1].iov_len
= sizeof(data
);
931 genbit
= RXRPC_CALL_ABORT
;
935 if (test_bit(RXRPC_CALL_ACK_FINAL
, &call
->events
)) {
936 genbit
= RXRPC_CALL_ACK_FINAL
;
938 ack
.bufferSpace
= htons(8);
941 ack
.reason
= RXRPC_ACK_IDLE
;
943 call
->ackr_reason
= 0;
945 spin_lock_bh(&call
->lock
);
946 ack
.serial
= call
->ackr_serial
;
947 ack
.previousPacket
= call
->ackr_prev_seq
;
948 ack
.firstPacket
= htonl(call
->rx_data_eaten
+ 1);
949 spin_unlock_bh(&call
->lock
);
953 iov
[1].iov_base
= &ack
;
954 iov
[1].iov_len
= sizeof(ack
);
955 iov
[2].iov_base
= &pad
;
957 iov
[3].iov_base
= &ackinfo
;
958 iov
[3].iov_len
= sizeof(ackinfo
);
962 if (call
->events
& ((1 << RXRPC_CALL_RCVD_BUSY
) |
963 (1 << RXRPC_CALL_RCVD_ABORT
))
967 if (test_bit(RXRPC_CALL_RCVD_ABORT
, &call
->events
))
968 mark
= RXRPC_SKB_MARK_REMOTE_ABORT
;
970 mark
= RXRPC_SKB_MARK_BUSY
;
972 _debug("post abort/busy");
973 rxrpc_clear_tx_window(call
);
974 if (rxrpc_post_message(call
, mark
, ECONNABORTED
, true) < 0)
977 clear_bit(RXRPC_CALL_RCVD_BUSY
, &call
->events
);
978 clear_bit(RXRPC_CALL_RCVD_ABORT
, &call
->events
);
982 if (test_and_clear_bit(RXRPC_CALL_RCVD_ACKALL
, &call
->events
)) {
983 _debug("do implicit ackall");
984 rxrpc_clear_tx_window(call
);
987 if (test_bit(RXRPC_CALL_LIFE_TIMER
, &call
->events
)) {
988 write_lock_bh(&call
->state_lock
);
989 if (call
->state
<= RXRPC_CALL_COMPLETE
) {
990 call
->state
= RXRPC_CALL_LOCALLY_ABORTED
;
991 call
->abort_code
= RX_CALL_TIMEOUT
;
992 set_bit(RXRPC_CALL_ABORT
, &call
->events
);
994 write_unlock_bh(&call
->state_lock
);
996 _debug("post timeout");
997 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_LOCAL_ERROR
,
1001 clear_bit(RXRPC_CALL_LIFE_TIMER
, &call
->events
);
1005 /* deal with assorted inbound messages */
1006 if (!skb_queue_empty(&call
->rx_queue
)) {
1007 switch (rxrpc_process_rx_queue(call
, &abort_code
)) {
1016 rxrpc_abort_call(call
, abort_code
);
1021 /* handle resending */
1022 if (test_and_clear_bit(RXRPC_CALL_RESEND_TIMER
, &call
->events
))
1023 rxrpc_resend_timer(call
);
1024 if (test_and_clear_bit(RXRPC_CALL_RESEND
, &call
->events
))
1027 /* consider sending an ordinary ACK */
1028 if (test_bit(RXRPC_CALL_ACK
, &call
->events
)) {
1029 _debug("send ACK: window: %d - %d { %lx }",
1030 call
->rx_data_eaten
, call
->ackr_win_top
,
1031 call
->ackr_window
[0]);
1033 if (call
->state
> RXRPC_CALL_SERVER_ACK_REQUEST
&&
1034 call
->ackr_reason
!= RXRPC_ACK_PING_RESPONSE
) {
1035 /* ACK by sending reply DATA packet in this state */
1036 clear_bit(RXRPC_CALL_ACK
, &call
->events
);
1037 goto maybe_reschedule
;
1040 genbit
= RXRPC_CALL_ACK
;
1042 acks
= kzalloc(call
->ackr_win_top
- call
->rx_data_eaten
,
1047 //hdr.flags = RXRPC_SLOW_START_OK;
1048 ack
.bufferSpace
= htons(8);
1053 spin_lock_bh(&call
->lock
);
1054 ack
.reason
= call
->ackr_reason
;
1055 ack
.serial
= call
->ackr_serial
;
1056 ack
.previousPacket
= call
->ackr_prev_seq
;
1057 ack
.firstPacket
= htonl(call
->rx_data_eaten
+ 1);
1060 for (loop
= 0; loop
< RXRPC_ACKR_WINDOW_ASZ
; loop
++) {
1061 nbit
= loop
* BITS_PER_LONG
;
1062 for (bits
= call
->ackr_window
[loop
]; bits
; bits
>>= 1
1064 _debug("- l=%d n=%d b=%lx", loop
, nbit
, bits
);
1066 acks
[nbit
] = RXRPC_ACK_TYPE_ACK
;
1067 ack
.nAcks
= nbit
+ 1;
1072 call
->ackr_reason
= 0;
1073 spin_unlock_bh(&call
->lock
);
1077 iov
[1].iov_base
= &ack
;
1078 iov
[1].iov_len
= sizeof(ack
);
1079 iov
[2].iov_base
= acks
;
1080 iov
[2].iov_len
= ack
.nAcks
;
1081 iov
[3].iov_base
= &pad
;
1083 iov
[4].iov_base
= &ackinfo
;
1084 iov
[4].iov_len
= sizeof(ackinfo
);
1086 switch (ack
.reason
) {
1087 case RXRPC_ACK_REQUESTED
:
1088 case RXRPC_ACK_DUPLICATE
:
1089 case RXRPC_ACK_OUT_OF_SEQUENCE
:
1090 case RXRPC_ACK_EXCEEDS_WINDOW
:
1091 case RXRPC_ACK_NOSPACE
:
1092 case RXRPC_ACK_PING
:
1093 case RXRPC_ACK_PING_RESPONSE
:
1094 goto send_ACK_with_skew
;
1095 case RXRPC_ACK_DELAY
:
1096 case RXRPC_ACK_IDLE
:
1101 /* handle completion of security negotiations on an incoming
1103 if (test_and_clear_bit(RXRPC_CALL_SECURED
, &call
->events
)) {
1105 spin_lock_bh(&call
->lock
);
1107 if (call
->state
== RXRPC_CALL_SERVER_SECURING
) {
1109 write_lock(&call
->conn
->lock
);
1110 if (!test_bit(RXRPC_CALL_RELEASED
, &call
->flags
) &&
1111 !test_bit(RXRPC_CALL_RELEASE
, &call
->events
)) {
1112 _debug("not released");
1113 call
->state
= RXRPC_CALL_SERVER_ACCEPTING
;
1114 list_move_tail(&call
->accept_link
,
1115 &call
->socket
->acceptq
);
1117 write_unlock(&call
->conn
->lock
);
1118 read_lock(&call
->state_lock
);
1119 if (call
->state
< RXRPC_CALL_COMPLETE
)
1120 set_bit(RXRPC_CALL_POST_ACCEPT
, &call
->events
);
1121 read_unlock(&call
->state_lock
);
1124 spin_unlock_bh(&call
->lock
);
1125 if (!test_bit(RXRPC_CALL_POST_ACCEPT
, &call
->events
))
1126 goto maybe_reschedule
;
1129 /* post a notification of an acceptable connection to the app */
1130 if (test_bit(RXRPC_CALL_POST_ACCEPT
, &call
->events
)) {
1131 _debug("post accept");
1132 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_NEW_CALL
,
1135 clear_bit(RXRPC_CALL_POST_ACCEPT
, &call
->events
);
1136 goto maybe_reschedule
;
1139 /* handle incoming call acceptance */
1140 if (test_and_clear_bit(RXRPC_CALL_ACCEPTED
, &call
->events
)) {
1142 ASSERTCMP(call
->rx_data_post
, ==, 0);
1143 call
->rx_data_post
= 1;
1144 read_lock_bh(&call
->state_lock
);
1145 if (call
->state
< RXRPC_CALL_COMPLETE
)
1146 set_bit(RXRPC_CALL_DRAIN_RX_OOS
, &call
->events
);
1147 read_unlock_bh(&call
->state_lock
);
1150 /* drain the out of sequence received packet queue into the packet Rx
1152 if (test_and_clear_bit(RXRPC_CALL_DRAIN_RX_OOS
, &call
->events
)) {
1153 while (call
->rx_data_post
== call
->rx_first_oos
)
1154 if (rxrpc_drain_rx_oos_queue(call
) < 0)
1156 goto maybe_reschedule
;
1159 /* other events may have been raised since we started checking */
1160 goto maybe_reschedule
;
1163 ack
.maxSkew
= htons(atomic_read(&call
->conn
->hi_serial
) -
1166 mtu
= call
->conn
->trans
->peer
->if_mtu
;
1167 mtu
-= call
->conn
->trans
->peer
->hdrsize
;
1168 ackinfo
.maxMTU
= htonl(mtu
);
1169 ackinfo
.rwind
= htonl(32);
1171 /* permit the peer to send us jumbo packets if it wants to */
1172 ackinfo
.rxMTU
= htonl(5692);
1173 ackinfo
.jumbo_max
= htonl(4);
1175 hdr
.serial
= htonl(atomic_inc_return(&call
->conn
->serial
));
1176 _proto("Tx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
1179 ntohl(ack
.firstPacket
),
1180 ntohl(ack
.previousPacket
),
1182 rxrpc_acks
[ack
.reason
],
1185 del_timer_sync(&call
->ack_timer
);
1187 set_bit(RXRPC_CALL_TX_SOFT_ACK
, &call
->flags
);
1188 goto send_message_2
;
1191 _debug("send message");
1193 hdr
.serial
= htonl(atomic_inc_return(&call
->conn
->serial
));
1194 _proto("Tx %s %%%u", rxrpc_pkts
[hdr
.type
], ntohl(hdr
.serial
));
1197 len
= iov
[0].iov_len
;
1199 if (iov
[4].iov_len
) {
1201 len
+= iov
[4].iov_len
;
1202 len
+= iov
[3].iov_len
;
1203 len
+= iov
[2].iov_len
;
1204 len
+= iov
[1].iov_len
;
1205 } else if (iov
[3].iov_len
) {
1207 len
+= iov
[3].iov_len
;
1208 len
+= iov
[2].iov_len
;
1209 len
+= iov
[1].iov_len
;
1210 } else if (iov
[2].iov_len
) {
1212 len
+= iov
[2].iov_len
;
1213 len
+= iov
[1].iov_len
;
1214 } else if (iov
[1].iov_len
) {
1216 len
+= iov
[1].iov_len
;
1219 ret
= kernel_sendmsg(call
->conn
->trans
->local
->socket
,
1220 &msg
, iov
, ioc
, len
);
1222 _debug("sendmsg failed: %d", ret
);
1223 read_lock_bh(&call
->state_lock
);
1224 if (call
->state
< RXRPC_CALL_DEAD
)
1225 rxrpc_queue_call(call
);
1226 read_unlock_bh(&call
->state_lock
);
1231 case RXRPC_CALL_ABORT
:
1232 clear_bit(genbit
, &call
->events
);
1233 clear_bit(RXRPC_CALL_RCVD_ABORT
, &call
->events
);
1236 case RXRPC_CALL_ACK_FINAL
:
1237 write_lock_bh(&call
->state_lock
);
1238 if (call
->state
== RXRPC_CALL_CLIENT_FINAL_ACK
)
1239 call
->state
= RXRPC_CALL_COMPLETE
;
1240 write_unlock_bh(&call
->state_lock
);
1244 clear_bit(genbit
, &call
->events
);
1245 switch (call
->state
) {
1246 case RXRPC_CALL_CLIENT_AWAIT_REPLY
:
1247 case RXRPC_CALL_CLIENT_RECV_REPLY
:
1248 case RXRPC_CALL_SERVER_RECV_REQUEST
:
1249 case RXRPC_CALL_SERVER_ACK_REQUEST
:
1250 _debug("start ACK timer");
1251 rxrpc_propose_ACK(call
, RXRPC_ACK_DELAY
,
1252 call
->ackr_serial
, false);
1256 goto maybe_reschedule
;
1260 del_timer_sync(&call
->ack_timer
);
1261 if (test_and_clear_bit(RXRPC_CALL_ACK_FINAL
, &call
->events
))
1262 rxrpc_put_call(call
);
1263 clear_bit(RXRPC_CALL_ACK
, &call
->events
);
1266 if (call
->events
|| !skb_queue_empty(&call
->rx_queue
)) {
1267 read_lock_bh(&call
->state_lock
);
1268 if (call
->state
< RXRPC_CALL_DEAD
)
1269 rxrpc_queue_call(call
);
1270 read_unlock_bh(&call
->state_lock
);
1273 /* don't leave aborted connections on the accept queue */
1274 if (call
->state
>= RXRPC_CALL_COMPLETE
&&
1275 !list_empty(&call
->accept_link
)) {
1276 _debug("X unlinking once-pending call %p { e=%lx f=%lx c=%x }",
1277 call
, call
->events
, call
->flags
,
1278 ntohl(call
->conn
->cid
));
1280 read_lock_bh(&call
->state_lock
);
1281 if (!test_bit(RXRPC_CALL_RELEASED
, &call
->flags
) &&
1282 !test_and_set_bit(RXRPC_CALL_RELEASE
, &call
->events
))
1283 rxrpc_queue_call(call
);
1284 read_unlock_bh(&call
->state_lock
);
1288 clear_bit(RXRPC_CALL_PROC_BUSY
, &call
->flags
);
1291 /* because we don't want two CPUs both processing the work item for one
1292 * call at the same time, we use a flag to note when it's busy; however
1293 * this means there's a race between clearing the flag and setting the
1294 * work pending bit and the work item being processed again */
1295 if (call
->events
&& !work_pending(&call
->processor
)) {
1296 _debug("jumpstart %x", ntohl(call
->conn
->cid
));
1297 rxrpc_queue_call(call
);
1304 _debug("out of memory");
1305 goto maybe_reschedule
;