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/slab.h>
17 #include <linux/udp.h>
19 #include <net/af_rxrpc.h>
20 #include "ar-internal.h"
22 static unsigned rxrpc_ack_defer
= 1;
24 static const char *const rxrpc_acks
[] = {
25 "---", "REQ", "DUP", "OOS", "WIN", "MEM", "PNG", "PNR", "DLY", "IDL",
29 static const s8 rxrpc_ack_priority
[] = {
31 [RXRPC_ACK_DELAY
] = 1,
32 [RXRPC_ACK_REQUESTED
] = 2,
34 [RXRPC_ACK_PING_RESPONSE
] = 4,
35 [RXRPC_ACK_DUPLICATE
] = 5,
36 [RXRPC_ACK_OUT_OF_SEQUENCE
] = 6,
37 [RXRPC_ACK_EXCEEDS_WINDOW
] = 7,
38 [RXRPC_ACK_NOSPACE
] = 8,
42 * propose an ACK be sent
44 void __rxrpc_propose_ACK(struct rxrpc_call
*call
, u8 ack_reason
,
45 __be32 serial
, bool immediate
)
48 s8 prior
= rxrpc_ack_priority
[ack_reason
];
50 ASSERTCMP(prior
, >, 0);
52 _enter("{%d},%s,%%%x,%u",
53 call
->debug_id
, rxrpc_acks
[ack_reason
], ntohl(serial
),
56 if (prior
< rxrpc_ack_priority
[call
->ackr_reason
]) {
62 /* update DELAY, IDLE, REQUESTED and PING_RESPONSE ACK serial
64 if (prior
== rxrpc_ack_priority
[call
->ackr_reason
]) {
66 call
->ackr_serial
= serial
;
72 call
->ackr_reason
= ack_reason
;
73 call
->ackr_serial
= serial
;
77 _debug("run delay timer");
78 call
->ack_timer
.expires
= jiffies
+ rxrpc_ack_timeout
* HZ
;
79 add_timer(&call
->ack_timer
);
84 _debug("run defer timer");
90 case RXRPC_ACK_REQUESTED
:
93 if (!immediate
|| serial
== cpu_to_be32(1)) {
94 _debug("run defer timer");
95 expiry
= rxrpc_ack_defer
;
100 _debug("immediate ACK");
106 if (!timer_pending(&call
->ack_timer
) ||
107 time_after(call
->ack_timer
.expires
, expiry
))
108 mod_timer(&call
->ack_timer
, expiry
);
112 _debug("cancel timer %%%u", ntohl(serial
));
113 try_to_del_timer_sync(&call
->ack_timer
);
114 read_lock_bh(&call
->state_lock
);
115 if (call
->state
<= RXRPC_CALL_COMPLETE
&&
116 !test_and_set_bit(RXRPC_CALL_ACK
, &call
->events
))
117 rxrpc_queue_call(call
);
118 read_unlock_bh(&call
->state_lock
);
122 * propose an ACK be sent, locking the call structure
124 void rxrpc_propose_ACK(struct rxrpc_call
*call
, u8 ack_reason
,
125 __be32 serial
, bool immediate
)
127 s8 prior
= rxrpc_ack_priority
[ack_reason
];
129 if (prior
> rxrpc_ack_priority
[call
->ackr_reason
]) {
130 spin_lock_bh(&call
->lock
);
131 __rxrpc_propose_ACK(call
, ack_reason
, serial
, immediate
);
132 spin_unlock_bh(&call
->lock
);
137 * set the resend timer
139 static void rxrpc_set_resend(struct rxrpc_call
*call
, u8 resend
,
140 unsigned long resend_at
)
142 read_lock_bh(&call
->state_lock
);
143 if (call
->state
>= RXRPC_CALL_COMPLETE
)
147 _debug("SET RESEND");
148 set_bit(RXRPC_CALL_RESEND
, &call
->events
);
152 _debug("MODIFY RESEND TIMER");
153 set_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
154 mod_timer(&call
->resend_timer
, resend_at
);
156 _debug("KILL RESEND TIMER");
157 del_timer_sync(&call
->resend_timer
);
158 clear_bit(RXRPC_CALL_RESEND_TIMER
, &call
->events
);
159 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
161 read_unlock_bh(&call
->state_lock
);
167 static void rxrpc_resend(struct rxrpc_call
*call
)
169 struct rxrpc_skb_priv
*sp
;
170 struct rxrpc_header
*hdr
;
172 unsigned long *p_txb
, resend_at
;
176 _enter("{%d,%d,%d,%d},",
177 call
->acks_hard
, call
->acks_unacked
,
178 atomic_read(&call
->sequence
),
179 CIRC_CNT(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
));
185 for (loop
= call
->acks_tail
;
186 loop
!= call
->acks_head
|| stop
;
187 loop
= (loop
+ 1) & (call
->acks_winsz
- 1)
189 p_txb
= call
->acks_window
+ loop
;
190 smp_read_barrier_depends();
194 txb
= (struct sk_buff
*) *p_txb
;
197 if (sp
->need_resend
) {
200 /* each Tx packet has a new serial number */
202 htonl(atomic_inc_return(&call
->conn
->serial
));
204 hdr
= (struct rxrpc_header
*) txb
->head
;
205 hdr
->serial
= sp
->hdr
.serial
;
207 _proto("Tx DATA %%%u { #%d }",
208 ntohl(sp
->hdr
.serial
), ntohl(sp
->hdr
.seq
));
209 if (rxrpc_send_packet(call
->conn
->trans
, txb
) < 0) {
211 sp
->resend_at
= jiffies
+ 3;
214 jiffies
+ rxrpc_resend_timeout
* HZ
;
218 if (time_after_eq(jiffies
+ 1, sp
->resend_at
)) {
221 } else if (resend
& 2) {
222 if (time_before(sp
->resend_at
, resend_at
))
223 resend_at
= sp
->resend_at
;
225 resend_at
= sp
->resend_at
;
230 rxrpc_set_resend(call
, resend
, resend_at
);
235 * handle resend timer expiry
237 static void rxrpc_resend_timer(struct rxrpc_call
*call
)
239 struct rxrpc_skb_priv
*sp
;
241 unsigned long *p_txb
, resend_at
;
246 call
->acks_tail
, call
->acks_unacked
, call
->acks_head
);
248 if (call
->state
>= RXRPC_CALL_COMPLETE
)
254 for (loop
= call
->acks_unacked
;
255 loop
!= call
->acks_head
;
256 loop
= (loop
+ 1) & (call
->acks_winsz
- 1)
258 p_txb
= call
->acks_window
+ loop
;
259 smp_read_barrier_depends();
260 txb
= (struct sk_buff
*) (*p_txb
& ~1);
263 ASSERT(!(*p_txb
& 1));
265 if (sp
->need_resend
) {
267 } else if (time_after_eq(jiffies
+ 1, sp
->resend_at
)) {
270 } else if (resend
& 2) {
271 if (time_before(sp
->resend_at
, resend_at
))
272 resend_at
= sp
->resend_at
;
274 resend_at
= sp
->resend_at
;
279 rxrpc_set_resend(call
, resend
, resend_at
);
284 * process soft ACKs of our transmitted packets
285 * - these indicate packets the peer has or has not received, but hasn't yet
286 * given to the consumer, and so can still be discarded and re-requested
288 static int rxrpc_process_soft_ACKs(struct rxrpc_call
*call
,
289 struct rxrpc_ackpacket
*ack
,
292 struct rxrpc_skb_priv
*sp
;
294 unsigned long *p_txb
, resend_at
;
296 u8 sacks
[RXRPC_MAXACKS
], resend
;
298 _enter("{%d,%d},{%d},",
300 CIRC_CNT(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
),
303 if (skb_copy_bits(skb
, 0, sacks
, ack
->nAcks
) < 0)
308 for (loop
= 0; loop
< ack
->nAcks
; loop
++) {
309 p_txb
= call
->acks_window
;
310 p_txb
+= (call
->acks_tail
+ loop
) & (call
->acks_winsz
- 1);
311 smp_read_barrier_depends();
312 txb
= (struct sk_buff
*) (*p_txb
& ~1);
315 switch (sacks
[loop
]) {
316 case RXRPC_ACK_TYPE_ACK
:
320 case RXRPC_ACK_TYPE_NACK
:
326 _debug("Unsupported ACK type %d", sacks
[loop
]);
332 call
->acks_unacked
= (call
->acks_tail
+ loop
) & (call
->acks_winsz
- 1);
334 /* anything not explicitly ACK'd is implicitly NACK'd, but may just not
335 * have been received or processed yet by the far end */
336 for (loop
= call
->acks_unacked
;
337 loop
!= call
->acks_head
;
338 loop
= (loop
+ 1) & (call
->acks_winsz
- 1)
340 p_txb
= call
->acks_window
+ loop
;
341 smp_read_barrier_depends();
342 txb
= (struct sk_buff
*) (*p_txb
& ~1);
346 /* packet must have been discarded */
350 } else if (sp
->need_resend
) {
352 } else if (time_after_eq(jiffies
+ 1, sp
->resend_at
)) {
355 } else if (resend
& 2) {
356 if (time_before(sp
->resend_at
, resend_at
))
357 resend_at
= sp
->resend_at
;
359 resend_at
= sp
->resend_at
;
364 rxrpc_set_resend(call
, resend
, resend_at
);
369 _leave(" = -EPROTO");
374 * discard hard-ACK'd packets from the Tx window
376 static void rxrpc_rotate_tx_window(struct rxrpc_call
*call
, u32 hard
)
379 int tail
= call
->acks_tail
, old_tail
;
380 int win
= CIRC_CNT(call
->acks_head
, tail
, call
->acks_winsz
);
382 _enter("{%u,%u},%u", call
->acks_hard
, win
, hard
);
384 ASSERTCMP(hard
- call
->acks_hard
, <=, win
);
386 while (call
->acks_hard
< hard
) {
387 smp_read_barrier_depends();
388 _skb
= call
->acks_window
[tail
] & ~1;
389 rxrpc_free_skb((struct sk_buff
*) _skb
);
391 tail
= (tail
+ 1) & (call
->acks_winsz
- 1);
392 call
->acks_tail
= tail
;
393 if (call
->acks_unacked
== old_tail
)
394 call
->acks_unacked
= tail
;
398 wake_up(&call
->tx_waitq
);
402 * clear the Tx window in the event of a failure
404 static void rxrpc_clear_tx_window(struct rxrpc_call
*call
)
406 rxrpc_rotate_tx_window(call
, atomic_read(&call
->sequence
));
410 * drain the out of sequence received packet queue into the packet Rx queue
412 static int rxrpc_drain_rx_oos_queue(struct rxrpc_call
*call
)
414 struct rxrpc_skb_priv
*sp
;
419 _enter("{%d,%d}", call
->rx_data_post
, call
->rx_first_oos
);
421 spin_lock_bh(&call
->lock
);
424 if (test_bit(RXRPC_CALL_RELEASED
, &call
->flags
))
425 goto socket_unavailable
;
427 skb
= skb_dequeue(&call
->rx_oos_queue
);
431 _debug("drain OOS packet %d [%d]",
432 ntohl(sp
->hdr
.seq
), call
->rx_first_oos
);
434 if (ntohl(sp
->hdr
.seq
) != call
->rx_first_oos
) {
435 skb_queue_head(&call
->rx_oos_queue
, skb
);
436 call
->rx_first_oos
= ntohl(rxrpc_skb(skb
)->hdr
.seq
);
437 _debug("requeue %p {%u}", skb
, call
->rx_first_oos
);
439 skb
->mark
= RXRPC_SKB_MARK_DATA
;
440 terminal
= ((sp
->hdr
.flags
& RXRPC_LAST_PACKET
) &&
441 !(sp
->hdr
.flags
& RXRPC_CLIENT_INITIATED
));
442 ret
= rxrpc_queue_rcv_skb(call
, skb
, true, terminal
);
444 _debug("drain #%u", call
->rx_data_post
);
445 call
->rx_data_post
++;
447 /* find out what the next packet is */
448 skb
= skb_peek(&call
->rx_oos_queue
);
451 ntohl(rxrpc_skb(skb
)->hdr
.seq
);
453 call
->rx_first_oos
= 0;
454 _debug("peek %p {%u}", skb
, call
->rx_first_oos
);
460 spin_unlock_bh(&call
->lock
);
461 _leave(" = %d", ret
);
466 * insert an out of sequence packet into the buffer
468 static void rxrpc_insert_oos_packet(struct rxrpc_call
*call
,
471 struct rxrpc_skb_priv
*sp
, *psp
;
476 seq
= ntohl(sp
->hdr
.seq
);
477 _enter(",,{%u}", seq
);
479 skb
->destructor
= rxrpc_packet_destructor
;
480 ASSERTCMP(sp
->call
, ==, NULL
);
482 rxrpc_get_call(call
);
484 /* insert into the buffer in sequence order */
485 spin_lock_bh(&call
->lock
);
487 skb_queue_walk(&call
->rx_oos_queue
, p
) {
489 if (ntohl(psp
->hdr
.seq
) > seq
) {
490 _debug("insert oos #%u before #%u",
491 seq
, ntohl(psp
->hdr
.seq
));
492 skb_insert(p
, skb
, &call
->rx_oos_queue
);
497 _debug("append oos #%u", seq
);
498 skb_queue_tail(&call
->rx_oos_queue
, skb
);
501 /* we might now have a new front to the queue */
502 if (call
->rx_first_oos
== 0 || seq
< call
->rx_first_oos
)
503 call
->rx_first_oos
= seq
;
505 read_lock(&call
->state_lock
);
506 if (call
->state
< RXRPC_CALL_COMPLETE
&&
507 call
->rx_data_post
== call
->rx_first_oos
) {
508 _debug("drain rx oos now");
509 set_bit(RXRPC_CALL_DRAIN_RX_OOS
, &call
->events
);
511 read_unlock(&call
->state_lock
);
513 spin_unlock_bh(&call
->lock
);
514 _leave(" [stored #%u]", call
->rx_first_oos
);
518 * clear the Tx window on final ACK reception
520 static void rxrpc_zap_tx_window(struct rxrpc_call
*call
)
522 struct rxrpc_skb_priv
*sp
;
524 unsigned long _skb
, *acks_window
;
525 u8 winsz
= call
->acks_winsz
;
528 acks_window
= call
->acks_window
;
529 call
->acks_window
= NULL
;
531 while (CIRC_CNT(call
->acks_head
, call
->acks_tail
, winsz
) > 0) {
532 tail
= call
->acks_tail
;
533 smp_read_barrier_depends();
534 _skb
= acks_window
[tail
] & ~1;
536 call
->acks_tail
= (call
->acks_tail
+ 1) & (winsz
- 1);
538 skb
= (struct sk_buff
*) _skb
;
540 _debug("+++ clear Tx %u", ntohl(sp
->hdr
.seq
));
548 * process the extra information that may be appended to an ACK packet
550 static void rxrpc_extract_ackinfo(struct rxrpc_call
*call
, struct sk_buff
*skb
,
551 unsigned latest
, int nAcks
)
553 struct rxrpc_ackinfo ackinfo
;
554 struct rxrpc_peer
*peer
;
557 if (skb_copy_bits(skb
, nAcks
+ 3, &ackinfo
, sizeof(ackinfo
)) < 0) {
558 _leave(" [no ackinfo]");
562 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
564 ntohl(ackinfo
.rxMTU
), ntohl(ackinfo
.maxMTU
),
565 ntohl(ackinfo
.rwind
), ntohl(ackinfo
.jumbo_max
));
567 mtu
= min(ntohl(ackinfo
.rxMTU
), ntohl(ackinfo
.maxMTU
));
569 peer
= call
->conn
->trans
->peer
;
570 if (mtu
< peer
->maxdata
) {
571 spin_lock_bh(&peer
->lock
);
573 peer
->mtu
= mtu
+ peer
->hdrsize
;
574 spin_unlock_bh(&peer
->lock
);
575 _net("Net MTU %u (maxdata %u)", peer
->mtu
, peer
->maxdata
);
580 * process packets in the reception queue
582 static int rxrpc_process_rx_queue(struct rxrpc_call
*call
,
585 struct rxrpc_ackpacket ack
;
586 struct rxrpc_skb_priv
*sp
;
595 skb
= skb_dequeue(&call
->rx_queue
);
599 _net("deferred skb %p", skb
);
603 _debug("process %s [st %d]", rxrpc_pkts
[sp
->hdr
.type
], call
->state
);
607 switch (sp
->hdr
.type
) {
608 /* data packets that wind up here have been received out of
609 * order, need security processing or are jumbo packets */
610 case RXRPC_PACKET_TYPE_DATA
:
611 _proto("OOSQ DATA %%%u { #%u }",
612 ntohl(sp
->hdr
.serial
), ntohl(sp
->hdr
.seq
));
614 /* secured packets must be verified and possibly decrypted */
615 if (rxrpc_verify_packet(call
, skb
, _abort_code
) < 0)
618 rxrpc_insert_oos_packet(call
, skb
);
619 goto process_further
;
621 /* partial ACK to process */
622 case RXRPC_PACKET_TYPE_ACK
:
623 if (skb_copy_bits(skb
, 0, &ack
, sizeof(ack
)) < 0) {
624 _debug("extraction failure");
627 if (!skb_pull(skb
, sizeof(ack
)))
630 latest
= ntohl(sp
->hdr
.serial
);
631 hard
= ntohl(ack
.firstPacket
);
632 tx
= atomic_read(&call
->sequence
);
634 _proto("Rx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
638 ntohl(ack
.previousPacket
),
640 rxrpc_acks
[ack
.reason
],
643 rxrpc_extract_ackinfo(call
, skb
, latest
, ack
.nAcks
);
645 if (ack
.reason
== RXRPC_ACK_PING
) {
646 _proto("Rx ACK %%%u PING Request", latest
);
647 rxrpc_propose_ACK(call
, RXRPC_ACK_PING_RESPONSE
,
648 sp
->hdr
.serial
, true);
651 /* discard any out-of-order or duplicate ACKs */
652 if (latest
- call
->acks_latest
<= 0) {
653 _debug("discard ACK %d <= %d",
654 latest
, call
->acks_latest
);
657 call
->acks_latest
= latest
;
659 if (call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
&&
660 call
->state
!= RXRPC_CALL_CLIENT_AWAIT_REPLY
&&
661 call
->state
!= RXRPC_CALL_SERVER_SEND_REPLY
&&
662 call
->state
!= RXRPC_CALL_SERVER_AWAIT_ACK
)
665 _debug("Tx=%d H=%u S=%d", tx
, call
->acks_hard
, call
->state
);
669 _debug("hard-ACK'd packet %d not transmitted"
675 if ((call
->state
== RXRPC_CALL_CLIENT_AWAIT_REPLY
||
676 call
->state
== RXRPC_CALL_SERVER_AWAIT_ACK
) &&
681 rxrpc_rotate_tx_window(call
, hard
- 1);
685 if (hard
- 1 + ack
.nAcks
> tx
) {
686 _debug("soft-ACK'd packet %d+%d not"
687 " transmitted (%d top)",
688 hard
- 1, ack
.nAcks
, tx
);
692 if (rxrpc_process_soft_ACKs(call
, &ack
, skb
) < 0)
697 /* complete ACK to process */
698 case RXRPC_PACKET_TYPE_ACKALL
:
701 /* abort and busy are handled elsewhere */
702 case RXRPC_PACKET_TYPE_BUSY
:
703 case RXRPC_PACKET_TYPE_ABORT
:
706 /* connection level events - also handled elsewhere */
707 case RXRPC_PACKET_TYPE_CHALLENGE
:
708 case RXRPC_PACKET_TYPE_RESPONSE
:
709 case RXRPC_PACKET_TYPE_DEBUG
:
713 /* if we've had a hard ACK that covers all the packets we've sent, then
714 * that ends that phase of the operation */
716 write_lock_bh(&call
->state_lock
);
717 _debug("ack all %d", call
->state
);
719 switch (call
->state
) {
720 case RXRPC_CALL_CLIENT_AWAIT_REPLY
:
721 call
->state
= RXRPC_CALL_CLIENT_RECV_REPLY
;
723 case RXRPC_CALL_SERVER_AWAIT_ACK
:
724 _debug("srv complete");
725 call
->state
= RXRPC_CALL_COMPLETE
;
728 case RXRPC_CALL_CLIENT_SEND_REQUEST
:
729 case RXRPC_CALL_SERVER_RECV_REQUEST
:
730 goto protocol_error_unlock
; /* can't occur yet */
732 write_unlock_bh(&call
->state_lock
);
733 goto discard
; /* assume packet left over from earlier phase */
736 write_unlock_bh(&call
->state_lock
);
738 /* if all the packets we sent are hard-ACK'd, then we can discard
739 * whatever we've got left */
740 _debug("clear Tx %d",
741 CIRC_CNT(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
));
743 del_timer_sync(&call
->resend_timer
);
744 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
745 clear_bit(RXRPC_CALL_RESEND_TIMER
, &call
->events
);
747 if (call
->acks_window
)
748 rxrpc_zap_tx_window(call
);
751 /* post the final ACK message for userspace to pick up */
753 skb
->mark
= RXRPC_SKB_MARK_FINAL_ACK
;
755 rxrpc_get_call(call
);
756 spin_lock_bh(&call
->lock
);
757 if (rxrpc_queue_rcv_skb(call
, skb
, true, true) < 0)
759 spin_unlock_bh(&call
->lock
);
760 goto process_further
;
765 goto process_further
;
767 protocol_error_unlock
:
768 write_unlock_bh(&call
->state_lock
);
771 _leave(" = -EPROTO");
776 * post a message to the socket Rx queue for recvmsg() to pick up
778 static int rxrpc_post_message(struct rxrpc_call
*call
, u32 mark
, u32 error
,
781 struct rxrpc_skb_priv
*sp
;
785 _enter("{%d,%lx},%u,%u,%d",
786 call
->debug_id
, call
->flags
, mark
, error
, fatal
);
788 /* remove timers and things for fatal messages */
790 del_timer_sync(&call
->resend_timer
);
791 del_timer_sync(&call
->ack_timer
);
792 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
795 if (mark
!= RXRPC_SKB_MARK_NEW_CALL
&&
796 !test_bit(RXRPC_CALL_HAS_USERID
, &call
->flags
)) {
797 _leave("[no userid]");
801 if (!test_bit(RXRPC_CALL_TERMINAL_MSG
, &call
->flags
)) {
802 skb
= alloc_skb(0, GFP_NOFS
);
811 memset(sp
, 0, sizeof(*sp
));
814 rxrpc_get_call(call
);
816 spin_lock_bh(&call
->lock
);
817 ret
= rxrpc_queue_rcv_skb(call
, skb
, true, fatal
);
818 spin_unlock_bh(&call
->lock
);
826 * handle background processing of incoming call packets and ACK / abort
829 void rxrpc_process_call(struct work_struct
*work
)
831 struct rxrpc_call
*call
=
832 container_of(work
, struct rxrpc_call
, processor
);
833 struct rxrpc_ackpacket ack
;
834 struct rxrpc_ackinfo ackinfo
;
835 struct rxrpc_header hdr
;
841 int genbit
, loop
, nbit
, ioc
, ret
, mtu
;
842 u32 abort_code
= RX_PROTOCOL_ERROR
;
845 //printk("\n--------------------\n");
846 _enter("{%d,%s,%lx} [%lu]",
847 call
->debug_id
, rxrpc_call_states
[call
->state
], call
->events
,
848 (jiffies
- call
->creation_jif
) / (HZ
/ 10));
850 if (test_and_set_bit(RXRPC_CALL_PROC_BUSY
, &call
->flags
)) {
851 _debug("XXXXXXXXXXXXX RUNNING ON MULTIPLE CPUS XXXXXXXXXXXXX");
855 /* there's a good chance we're going to have to send a message, so set
856 * one up in advance */
857 msg
.msg_name
= &call
->conn
->trans
->peer
->srx
.transport
.sin
;
858 msg
.msg_namelen
= sizeof(call
->conn
->trans
->peer
->srx
.transport
.sin
);
859 msg
.msg_control
= NULL
;
860 msg
.msg_controllen
= 0;
863 hdr
.epoch
= call
->conn
->epoch
;
865 hdr
.callNumber
= call
->call_id
;
867 hdr
.type
= RXRPC_PACKET_TYPE_ACK
;
868 hdr
.flags
= call
->conn
->out_clientflag
;
870 hdr
.securityIndex
= call
->conn
->security_ix
;
872 hdr
.serviceId
= call
->conn
->service_id
;
874 memset(iov
, 0, sizeof(iov
));
875 iov
[0].iov_base
= &hdr
;
876 iov
[0].iov_len
= sizeof(hdr
);
878 /* deal with events of a final nature */
879 if (test_bit(RXRPC_CALL_RELEASE
, &call
->events
)) {
880 rxrpc_release_call(call
);
881 clear_bit(RXRPC_CALL_RELEASE
, &call
->events
);
884 if (test_bit(RXRPC_CALL_RCVD_ERROR
, &call
->events
)) {
887 clear_bit(RXRPC_CALL_CONN_ABORT
, &call
->events
);
888 clear_bit(RXRPC_CALL_REJECT_BUSY
, &call
->events
);
889 clear_bit(RXRPC_CALL_ABORT
, &call
->events
);
891 error
= call
->conn
->trans
->peer
->net_error
;
892 _debug("post net error %d", error
);
894 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_NET_ERROR
,
897 clear_bit(RXRPC_CALL_RCVD_ERROR
, &call
->events
);
901 if (test_bit(RXRPC_CALL_CONN_ABORT
, &call
->events
)) {
902 ASSERTCMP(call
->state
, >, RXRPC_CALL_COMPLETE
);
904 clear_bit(RXRPC_CALL_REJECT_BUSY
, &call
->events
);
905 clear_bit(RXRPC_CALL_ABORT
, &call
->events
);
907 _debug("post conn abort");
909 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_LOCAL_ERROR
,
910 call
->conn
->error
, true) < 0)
912 clear_bit(RXRPC_CALL_CONN_ABORT
, &call
->events
);
916 if (test_bit(RXRPC_CALL_REJECT_BUSY
, &call
->events
)) {
917 hdr
.type
= RXRPC_PACKET_TYPE_BUSY
;
918 genbit
= RXRPC_CALL_REJECT_BUSY
;
922 if (test_bit(RXRPC_CALL_ABORT
, &call
->events
)) {
923 ASSERTCMP(call
->state
, >, RXRPC_CALL_COMPLETE
);
925 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_LOCAL_ERROR
,
926 ECONNABORTED
, true) < 0)
928 hdr
.type
= RXRPC_PACKET_TYPE_ABORT
;
929 data
= htonl(call
->abort_code
);
930 iov
[1].iov_base
= &data
;
931 iov
[1].iov_len
= sizeof(data
);
932 genbit
= RXRPC_CALL_ABORT
;
936 if (test_bit(RXRPC_CALL_ACK_FINAL
, &call
->events
)) {
937 genbit
= RXRPC_CALL_ACK_FINAL
;
939 ack
.bufferSpace
= htons(8);
942 ack
.reason
= RXRPC_ACK_IDLE
;
944 call
->ackr_reason
= 0;
946 spin_lock_bh(&call
->lock
);
947 ack
.serial
= call
->ackr_serial
;
948 ack
.previousPacket
= call
->ackr_prev_seq
;
949 ack
.firstPacket
= htonl(call
->rx_data_eaten
+ 1);
950 spin_unlock_bh(&call
->lock
);
954 iov
[1].iov_base
= &ack
;
955 iov
[1].iov_len
= sizeof(ack
);
956 iov
[2].iov_base
= &pad
;
958 iov
[3].iov_base
= &ackinfo
;
959 iov
[3].iov_len
= sizeof(ackinfo
);
963 if (call
->events
& ((1 << RXRPC_CALL_RCVD_BUSY
) |
964 (1 << RXRPC_CALL_RCVD_ABORT
))
968 if (test_bit(RXRPC_CALL_RCVD_ABORT
, &call
->events
))
969 mark
= RXRPC_SKB_MARK_REMOTE_ABORT
;
971 mark
= RXRPC_SKB_MARK_BUSY
;
973 _debug("post abort/busy");
974 rxrpc_clear_tx_window(call
);
975 if (rxrpc_post_message(call
, mark
, ECONNABORTED
, true) < 0)
978 clear_bit(RXRPC_CALL_RCVD_BUSY
, &call
->events
);
979 clear_bit(RXRPC_CALL_RCVD_ABORT
, &call
->events
);
983 if (test_and_clear_bit(RXRPC_CALL_RCVD_ACKALL
, &call
->events
)) {
984 _debug("do implicit ackall");
985 rxrpc_clear_tx_window(call
);
988 if (test_bit(RXRPC_CALL_LIFE_TIMER
, &call
->events
)) {
989 write_lock_bh(&call
->state_lock
);
990 if (call
->state
<= RXRPC_CALL_COMPLETE
) {
991 call
->state
= RXRPC_CALL_LOCALLY_ABORTED
;
992 call
->abort_code
= RX_CALL_TIMEOUT
;
993 set_bit(RXRPC_CALL_ABORT
, &call
->events
);
995 write_unlock_bh(&call
->state_lock
);
997 _debug("post timeout");
998 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_LOCAL_ERROR
,
1002 clear_bit(RXRPC_CALL_LIFE_TIMER
, &call
->events
);
1006 /* deal with assorted inbound messages */
1007 if (!skb_queue_empty(&call
->rx_queue
)) {
1008 switch (rxrpc_process_rx_queue(call
, &abort_code
)) {
1017 rxrpc_abort_call(call
, abort_code
);
1022 /* handle resending */
1023 if (test_and_clear_bit(RXRPC_CALL_RESEND_TIMER
, &call
->events
))
1024 rxrpc_resend_timer(call
);
1025 if (test_and_clear_bit(RXRPC_CALL_RESEND
, &call
->events
))
1028 /* consider sending an ordinary ACK */
1029 if (test_bit(RXRPC_CALL_ACK
, &call
->events
)) {
1030 _debug("send ACK: window: %d - %d { %lx }",
1031 call
->rx_data_eaten
, call
->ackr_win_top
,
1032 call
->ackr_window
[0]);
1034 if (call
->state
> RXRPC_CALL_SERVER_ACK_REQUEST
&&
1035 call
->ackr_reason
!= RXRPC_ACK_PING_RESPONSE
) {
1036 /* ACK by sending reply DATA packet in this state */
1037 clear_bit(RXRPC_CALL_ACK
, &call
->events
);
1038 goto maybe_reschedule
;
1041 genbit
= RXRPC_CALL_ACK
;
1043 acks
= kzalloc(call
->ackr_win_top
- call
->rx_data_eaten
,
1048 //hdr.flags = RXRPC_SLOW_START_OK;
1049 ack
.bufferSpace
= htons(8);
1054 spin_lock_bh(&call
->lock
);
1055 ack
.reason
= call
->ackr_reason
;
1056 ack
.serial
= call
->ackr_serial
;
1057 ack
.previousPacket
= call
->ackr_prev_seq
;
1058 ack
.firstPacket
= htonl(call
->rx_data_eaten
+ 1);
1061 for (loop
= 0; loop
< RXRPC_ACKR_WINDOW_ASZ
; loop
++) {
1062 nbit
= loop
* BITS_PER_LONG
;
1063 for (bits
= call
->ackr_window
[loop
]; bits
; bits
>>= 1
1065 _debug("- l=%d n=%d b=%lx", loop
, nbit
, bits
);
1067 acks
[nbit
] = RXRPC_ACK_TYPE_ACK
;
1068 ack
.nAcks
= nbit
+ 1;
1073 call
->ackr_reason
= 0;
1074 spin_unlock_bh(&call
->lock
);
1078 iov
[1].iov_base
= &ack
;
1079 iov
[1].iov_len
= sizeof(ack
);
1080 iov
[2].iov_base
= acks
;
1081 iov
[2].iov_len
= ack
.nAcks
;
1082 iov
[3].iov_base
= &pad
;
1084 iov
[4].iov_base
= &ackinfo
;
1085 iov
[4].iov_len
= sizeof(ackinfo
);
1087 switch (ack
.reason
) {
1088 case RXRPC_ACK_REQUESTED
:
1089 case RXRPC_ACK_DUPLICATE
:
1090 case RXRPC_ACK_OUT_OF_SEQUENCE
:
1091 case RXRPC_ACK_EXCEEDS_WINDOW
:
1092 case RXRPC_ACK_NOSPACE
:
1093 case RXRPC_ACK_PING
:
1094 case RXRPC_ACK_PING_RESPONSE
:
1095 goto send_ACK_with_skew
;
1096 case RXRPC_ACK_DELAY
:
1097 case RXRPC_ACK_IDLE
:
1102 /* handle completion of security negotiations on an incoming
1104 if (test_and_clear_bit(RXRPC_CALL_SECURED
, &call
->events
)) {
1106 spin_lock_bh(&call
->lock
);
1108 if (call
->state
== RXRPC_CALL_SERVER_SECURING
) {
1110 write_lock(&call
->conn
->lock
);
1111 if (!test_bit(RXRPC_CALL_RELEASED
, &call
->flags
) &&
1112 !test_bit(RXRPC_CALL_RELEASE
, &call
->events
)) {
1113 _debug("not released");
1114 call
->state
= RXRPC_CALL_SERVER_ACCEPTING
;
1115 list_move_tail(&call
->accept_link
,
1116 &call
->socket
->acceptq
);
1118 write_unlock(&call
->conn
->lock
);
1119 read_lock(&call
->state_lock
);
1120 if (call
->state
< RXRPC_CALL_COMPLETE
)
1121 set_bit(RXRPC_CALL_POST_ACCEPT
, &call
->events
);
1122 read_unlock(&call
->state_lock
);
1125 spin_unlock_bh(&call
->lock
);
1126 if (!test_bit(RXRPC_CALL_POST_ACCEPT
, &call
->events
))
1127 goto maybe_reschedule
;
1130 /* post a notification of an acceptable connection to the app */
1131 if (test_bit(RXRPC_CALL_POST_ACCEPT
, &call
->events
)) {
1132 _debug("post accept");
1133 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_NEW_CALL
,
1136 clear_bit(RXRPC_CALL_POST_ACCEPT
, &call
->events
);
1137 goto maybe_reschedule
;
1140 /* handle incoming call acceptance */
1141 if (test_and_clear_bit(RXRPC_CALL_ACCEPTED
, &call
->events
)) {
1143 ASSERTCMP(call
->rx_data_post
, ==, 0);
1144 call
->rx_data_post
= 1;
1145 read_lock_bh(&call
->state_lock
);
1146 if (call
->state
< RXRPC_CALL_COMPLETE
)
1147 set_bit(RXRPC_CALL_DRAIN_RX_OOS
, &call
->events
);
1148 read_unlock_bh(&call
->state_lock
);
1151 /* drain the out of sequence received packet queue into the packet Rx
1153 if (test_and_clear_bit(RXRPC_CALL_DRAIN_RX_OOS
, &call
->events
)) {
1154 while (call
->rx_data_post
== call
->rx_first_oos
)
1155 if (rxrpc_drain_rx_oos_queue(call
) < 0)
1157 goto maybe_reschedule
;
1160 /* other events may have been raised since we started checking */
1161 goto maybe_reschedule
;
1164 ack
.maxSkew
= htons(atomic_read(&call
->conn
->hi_serial
) -
1167 mtu
= call
->conn
->trans
->peer
->if_mtu
;
1168 mtu
-= call
->conn
->trans
->peer
->hdrsize
;
1169 ackinfo
.maxMTU
= htonl(mtu
);
1170 ackinfo
.rwind
= htonl(32);
1172 /* permit the peer to send us jumbo packets if it wants to */
1173 ackinfo
.rxMTU
= htonl(5692);
1174 ackinfo
.jumbo_max
= htonl(4);
1176 hdr
.serial
= htonl(atomic_inc_return(&call
->conn
->serial
));
1177 _proto("Tx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
1180 ntohl(ack
.firstPacket
),
1181 ntohl(ack
.previousPacket
),
1183 rxrpc_acks
[ack
.reason
],
1186 del_timer_sync(&call
->ack_timer
);
1188 set_bit(RXRPC_CALL_TX_SOFT_ACK
, &call
->flags
);
1189 goto send_message_2
;
1192 _debug("send message");
1194 hdr
.serial
= htonl(atomic_inc_return(&call
->conn
->serial
));
1195 _proto("Tx %s %%%u", rxrpc_pkts
[hdr
.type
], ntohl(hdr
.serial
));
1198 len
= iov
[0].iov_len
;
1200 if (iov
[4].iov_len
) {
1202 len
+= iov
[4].iov_len
;
1203 len
+= iov
[3].iov_len
;
1204 len
+= iov
[2].iov_len
;
1205 len
+= iov
[1].iov_len
;
1206 } else if (iov
[3].iov_len
) {
1208 len
+= iov
[3].iov_len
;
1209 len
+= iov
[2].iov_len
;
1210 len
+= iov
[1].iov_len
;
1211 } else if (iov
[2].iov_len
) {
1213 len
+= iov
[2].iov_len
;
1214 len
+= iov
[1].iov_len
;
1215 } else if (iov
[1].iov_len
) {
1217 len
+= iov
[1].iov_len
;
1220 ret
= kernel_sendmsg(call
->conn
->trans
->local
->socket
,
1221 &msg
, iov
, ioc
, len
);
1223 _debug("sendmsg failed: %d", ret
);
1224 read_lock_bh(&call
->state_lock
);
1225 if (call
->state
< RXRPC_CALL_DEAD
)
1226 rxrpc_queue_call(call
);
1227 read_unlock_bh(&call
->state_lock
);
1232 case RXRPC_CALL_ABORT
:
1233 clear_bit(genbit
, &call
->events
);
1234 clear_bit(RXRPC_CALL_RCVD_ABORT
, &call
->events
);
1237 case RXRPC_CALL_ACK_FINAL
:
1238 write_lock_bh(&call
->state_lock
);
1239 if (call
->state
== RXRPC_CALL_CLIENT_FINAL_ACK
)
1240 call
->state
= RXRPC_CALL_COMPLETE
;
1241 write_unlock_bh(&call
->state_lock
);
1245 clear_bit(genbit
, &call
->events
);
1246 switch (call
->state
) {
1247 case RXRPC_CALL_CLIENT_AWAIT_REPLY
:
1248 case RXRPC_CALL_CLIENT_RECV_REPLY
:
1249 case RXRPC_CALL_SERVER_RECV_REQUEST
:
1250 case RXRPC_CALL_SERVER_ACK_REQUEST
:
1251 _debug("start ACK timer");
1252 rxrpc_propose_ACK(call
, RXRPC_ACK_DELAY
,
1253 call
->ackr_serial
, false);
1257 goto maybe_reschedule
;
1261 del_timer_sync(&call
->ack_timer
);
1262 if (test_and_clear_bit(RXRPC_CALL_ACK_FINAL
, &call
->events
))
1263 rxrpc_put_call(call
);
1264 clear_bit(RXRPC_CALL_ACK
, &call
->events
);
1267 if (call
->events
|| !skb_queue_empty(&call
->rx_queue
)) {
1268 read_lock_bh(&call
->state_lock
);
1269 if (call
->state
< RXRPC_CALL_DEAD
)
1270 rxrpc_queue_call(call
);
1271 read_unlock_bh(&call
->state_lock
);
1274 /* don't leave aborted connections on the accept queue */
1275 if (call
->state
>= RXRPC_CALL_COMPLETE
&&
1276 !list_empty(&call
->accept_link
)) {
1277 _debug("X unlinking once-pending call %p { e=%lx f=%lx c=%x }",
1278 call
, call
->events
, call
->flags
,
1279 ntohl(call
->conn
->cid
));
1281 read_lock_bh(&call
->state_lock
);
1282 if (!test_bit(RXRPC_CALL_RELEASED
, &call
->flags
) &&
1283 !test_and_set_bit(RXRPC_CALL_RELEASE
, &call
->events
))
1284 rxrpc_queue_call(call
);
1285 read_unlock_bh(&call
->state_lock
);
1289 clear_bit(RXRPC_CALL_PROC_BUSY
, &call
->flags
);
1292 /* because we don't want two CPUs both processing the work item for one
1293 * call at the same time, we use a flag to note when it's busy; however
1294 * this means there's a race between clearing the flag and setting the
1295 * work pending bit and the work item being processed again */
1296 if (call
->events
&& !work_pending(&call
->processor
)) {
1297 _debug("jumpstart %x", ntohl(call
->conn
->cid
));
1298 rxrpc_queue_call(call
);
1305 _debug("out of memory");
1306 goto maybe_reschedule
;