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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/module.h>
15 #include <linux/circ_buf.h>
16 #include <linux/net.h>
17 #include <linux/skbuff.h>
18 #include <linux/slab.h>
19 #include <linux/udp.h>
21 #include <net/af_rxrpc.h>
22 #include "ar-internal.h"
25 * propose an ACK be sent
27 void __rxrpc_propose_ACK(struct rxrpc_call
*call
, u8 ack_reason
,
28 u32 serial
, bool immediate
)
31 s8 prior
= rxrpc_ack_priority
[ack_reason
];
33 ASSERTCMP(prior
, >, 0);
35 _enter("{%d},%s,%%%x,%u",
36 call
->debug_id
, rxrpc_acks(ack_reason
), serial
, immediate
);
38 if (prior
< rxrpc_ack_priority
[call
->ackr_reason
]) {
44 /* update DELAY, IDLE, REQUESTED and PING_RESPONSE ACK serial
46 if (prior
== rxrpc_ack_priority
[call
->ackr_reason
]) {
48 call
->ackr_serial
= serial
;
54 call
->ackr_reason
= ack_reason
;
55 call
->ackr_serial
= serial
;
59 _debug("run delay timer");
60 expiry
= rxrpc_soft_ack_delay
;
65 _debug("run defer timer");
66 expiry
= rxrpc_idle_ack_delay
;
71 case RXRPC_ACK_REQUESTED
:
72 expiry
= rxrpc_requested_ack_delay
;
75 if (!immediate
|| serial
== 1) {
76 _debug("run defer timer");
81 _debug("immediate ACK");
87 if (!timer_pending(&call
->ack_timer
) ||
88 time_after(call
->ack_timer
.expires
, expiry
))
89 mod_timer(&call
->ack_timer
, expiry
);
93 _debug("cancel timer %%%u", serial
);
94 try_to_del_timer_sync(&call
->ack_timer
);
95 read_lock_bh(&call
->state_lock
);
96 if (call
->state
<= RXRPC_CALL_COMPLETE
&&
97 !test_and_set_bit(RXRPC_CALL_EV_ACK
, &call
->events
))
98 rxrpc_queue_call(call
);
99 read_unlock_bh(&call
->state_lock
);
103 * propose an ACK be sent, locking the call structure
105 void rxrpc_propose_ACK(struct rxrpc_call
*call
, u8 ack_reason
,
106 u32 serial
, bool immediate
)
108 s8 prior
= rxrpc_ack_priority
[ack_reason
];
110 if (prior
> rxrpc_ack_priority
[call
->ackr_reason
]) {
111 spin_lock_bh(&call
->lock
);
112 __rxrpc_propose_ACK(call
, ack_reason
, serial
, immediate
);
113 spin_unlock_bh(&call
->lock
);
118 * set the resend timer
120 static void rxrpc_set_resend(struct rxrpc_call
*call
, u8 resend
,
121 unsigned long resend_at
)
123 read_lock_bh(&call
->state_lock
);
124 if (call
->state
>= RXRPC_CALL_COMPLETE
)
128 _debug("SET RESEND");
129 set_bit(RXRPC_CALL_EV_RESEND
, &call
->events
);
133 _debug("MODIFY RESEND TIMER");
134 set_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
135 mod_timer(&call
->resend_timer
, resend_at
);
137 _debug("KILL RESEND TIMER");
138 del_timer_sync(&call
->resend_timer
);
139 clear_bit(RXRPC_CALL_EV_RESEND_TIMER
, &call
->events
);
140 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
142 read_unlock_bh(&call
->state_lock
);
148 static void rxrpc_resend(struct rxrpc_call
*call
)
150 struct rxrpc_wire_header
*whdr
;
151 struct rxrpc_skb_priv
*sp
;
153 unsigned long *p_txb
, resend_at
;
158 _enter("{%d,%d,%d,%d},",
159 call
->acks_hard
, call
->acks_unacked
,
160 atomic_read(&call
->sequence
),
161 CIRC_CNT(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
));
167 for (loop
= call
->acks_tail
;
168 loop
!= call
->acks_head
|| stop
;
169 loop
= (loop
+ 1) & (call
->acks_winsz
- 1)
171 p_txb
= call
->acks_window
+ loop
;
172 smp_read_barrier_depends();
176 txb
= (struct sk_buff
*) *p_txb
;
179 if (sp
->need_resend
) {
180 sp
->need_resend
= false;
182 /* each Tx packet has a new serial number */
183 sp
->hdr
.serial
= atomic_inc_return(&call
->conn
->serial
);
185 whdr
= (struct rxrpc_wire_header
*)txb
->head
;
186 whdr
->serial
= htonl(sp
->hdr
.serial
);
188 _proto("Tx DATA %%%u { #%d }",
189 sp
->hdr
.serial
, sp
->hdr
.seq
);
190 if (rxrpc_send_data_packet(call
->conn
, txb
) < 0) {
192 sp
->resend_at
= jiffies
+ 3;
195 jiffies
+ rxrpc_resend_timeout
;
199 if (time_after_eq(jiffies
+ 1, sp
->resend_at
)) {
200 sp
->need_resend
= true;
202 } else if (resend
& 2) {
203 if (time_before(sp
->resend_at
, resend_at
))
204 resend_at
= sp
->resend_at
;
206 resend_at
= sp
->resend_at
;
211 rxrpc_set_resend(call
, resend
, resend_at
);
216 * handle resend timer expiry
218 static void rxrpc_resend_timer(struct rxrpc_call
*call
)
220 struct rxrpc_skb_priv
*sp
;
222 unsigned long *p_txb
, resend_at
;
227 call
->acks_tail
, call
->acks_unacked
, call
->acks_head
);
229 if (call
->state
>= RXRPC_CALL_COMPLETE
)
235 for (loop
= call
->acks_unacked
;
236 loop
!= call
->acks_head
;
237 loop
= (loop
+ 1) & (call
->acks_winsz
- 1)
239 p_txb
= call
->acks_window
+ loop
;
240 smp_read_barrier_depends();
241 txb
= (struct sk_buff
*) (*p_txb
& ~1);
244 ASSERT(!(*p_txb
& 1));
246 if (sp
->need_resend
) {
248 } else if (time_after_eq(jiffies
+ 1, sp
->resend_at
)) {
249 sp
->need_resend
= true;
251 } else if (resend
& 2) {
252 if (time_before(sp
->resend_at
, resend_at
))
253 resend_at
= sp
->resend_at
;
255 resend_at
= sp
->resend_at
;
260 rxrpc_set_resend(call
, resend
, resend_at
);
265 * process soft ACKs of our transmitted packets
266 * - these indicate packets the peer has or has not received, but hasn't yet
267 * given to the consumer, and so can still be discarded and re-requested
269 static int rxrpc_process_soft_ACKs(struct rxrpc_call
*call
,
270 struct rxrpc_ackpacket
*ack
,
273 struct rxrpc_skb_priv
*sp
;
275 unsigned long *p_txb
, resend_at
;
277 u8 sacks
[RXRPC_MAXACKS
], resend
;
279 _enter("{%d,%d},{%d},",
281 CIRC_CNT(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
),
284 if (skb_copy_bits(skb
, 0, sacks
, ack
->nAcks
) < 0)
289 for (loop
= 0; loop
< ack
->nAcks
; loop
++) {
290 p_txb
= call
->acks_window
;
291 p_txb
+= (call
->acks_tail
+ loop
) & (call
->acks_winsz
- 1);
292 smp_read_barrier_depends();
293 txb
= (struct sk_buff
*) (*p_txb
& ~1);
296 switch (sacks
[loop
]) {
297 case RXRPC_ACK_TYPE_ACK
:
298 sp
->need_resend
= false;
301 case RXRPC_ACK_TYPE_NACK
:
302 sp
->need_resend
= true;
307 _debug("Unsupported ACK type %d", sacks
[loop
]);
313 call
->acks_unacked
= (call
->acks_tail
+ loop
) & (call
->acks_winsz
- 1);
315 /* anything not explicitly ACK'd is implicitly NACK'd, but may just not
316 * have been received or processed yet by the far end */
317 for (loop
= call
->acks_unacked
;
318 loop
!= call
->acks_head
;
319 loop
= (loop
+ 1) & (call
->acks_winsz
- 1)
321 p_txb
= call
->acks_window
+ loop
;
322 smp_read_barrier_depends();
323 txb
= (struct sk_buff
*) (*p_txb
& ~1);
327 /* packet must have been discarded */
328 sp
->need_resend
= true;
331 } else if (sp
->need_resend
) {
333 } else if (time_after_eq(jiffies
+ 1, sp
->resend_at
)) {
334 sp
->need_resend
= true;
336 } else if (resend
& 2) {
337 if (time_before(sp
->resend_at
, resend_at
))
338 resend_at
= sp
->resend_at
;
340 resend_at
= sp
->resend_at
;
345 rxrpc_set_resend(call
, resend
, resend_at
);
350 _leave(" = -EPROTO");
355 * discard hard-ACK'd packets from the Tx window
357 static void rxrpc_rotate_tx_window(struct rxrpc_call
*call
, u32 hard
)
360 int tail
= call
->acks_tail
, old_tail
;
361 int win
= CIRC_CNT(call
->acks_head
, tail
, call
->acks_winsz
);
363 _enter("{%u,%u},%u", call
->acks_hard
, win
, hard
);
365 ASSERTCMP(hard
- call
->acks_hard
, <=, win
);
367 while (call
->acks_hard
< hard
) {
368 smp_read_barrier_depends();
369 _skb
= call
->acks_window
[tail
] & ~1;
370 rxrpc_free_skb((struct sk_buff
*) _skb
);
372 tail
= (tail
+ 1) & (call
->acks_winsz
- 1);
373 call
->acks_tail
= tail
;
374 if (call
->acks_unacked
== old_tail
)
375 call
->acks_unacked
= tail
;
379 wake_up(&call
->tx_waitq
);
383 * clear the Tx window in the event of a failure
385 static void rxrpc_clear_tx_window(struct rxrpc_call
*call
)
387 rxrpc_rotate_tx_window(call
, atomic_read(&call
->sequence
));
391 * drain the out of sequence received packet queue into the packet Rx queue
393 static int rxrpc_drain_rx_oos_queue(struct rxrpc_call
*call
)
395 struct rxrpc_skb_priv
*sp
;
400 _enter("{%d,%d}", call
->rx_data_post
, call
->rx_first_oos
);
402 spin_lock_bh(&call
->lock
);
405 if (test_bit(RXRPC_CALL_RELEASED
, &call
->flags
))
406 goto socket_unavailable
;
408 skb
= skb_dequeue(&call
->rx_oos_queue
);
412 _debug("drain OOS packet %d [%d]",
413 sp
->hdr
.seq
, call
->rx_first_oos
);
415 if (sp
->hdr
.seq
!= call
->rx_first_oos
) {
416 skb_queue_head(&call
->rx_oos_queue
, skb
);
417 call
->rx_first_oos
= rxrpc_skb(skb
)->hdr
.seq
;
418 _debug("requeue %p {%u}", skb
, call
->rx_first_oos
);
420 skb
->mark
= RXRPC_SKB_MARK_DATA
;
421 terminal
= ((sp
->hdr
.flags
& RXRPC_LAST_PACKET
) &&
422 !(sp
->hdr
.flags
& RXRPC_CLIENT_INITIATED
));
423 ret
= rxrpc_queue_rcv_skb(call
, skb
, true, terminal
);
425 _debug("drain #%u", call
->rx_data_post
);
426 call
->rx_data_post
++;
428 /* find out what the next packet is */
429 skb
= skb_peek(&call
->rx_oos_queue
);
431 call
->rx_first_oos
= rxrpc_skb(skb
)->hdr
.seq
;
433 call
->rx_first_oos
= 0;
434 _debug("peek %p {%u}", skb
, call
->rx_first_oos
);
440 spin_unlock_bh(&call
->lock
);
441 _leave(" = %d", ret
);
446 * insert an out of sequence packet into the buffer
448 static void rxrpc_insert_oos_packet(struct rxrpc_call
*call
,
451 struct rxrpc_skb_priv
*sp
, *psp
;
457 _enter(",,{%u}", seq
);
459 skb
->destructor
= rxrpc_packet_destructor
;
460 ASSERTCMP(sp
->call
, ==, NULL
);
462 rxrpc_get_call(call
);
463 atomic_inc(&call
->skb_count
);
465 /* insert into the buffer in sequence order */
466 spin_lock_bh(&call
->lock
);
468 skb_queue_walk(&call
->rx_oos_queue
, p
) {
470 if (psp
->hdr
.seq
> seq
) {
471 _debug("insert oos #%u before #%u", seq
, psp
->hdr
.seq
);
472 skb_insert(p
, skb
, &call
->rx_oos_queue
);
477 _debug("append oos #%u", seq
);
478 skb_queue_tail(&call
->rx_oos_queue
, skb
);
481 /* we might now have a new front to the queue */
482 if (call
->rx_first_oos
== 0 || seq
< call
->rx_first_oos
)
483 call
->rx_first_oos
= seq
;
485 read_lock(&call
->state_lock
);
486 if (call
->state
< RXRPC_CALL_COMPLETE
&&
487 call
->rx_data_post
== call
->rx_first_oos
) {
488 _debug("drain rx oos now");
489 set_bit(RXRPC_CALL_EV_DRAIN_RX_OOS
, &call
->events
);
491 read_unlock(&call
->state_lock
);
493 spin_unlock_bh(&call
->lock
);
494 _leave(" [stored #%u]", call
->rx_first_oos
);
498 * clear the Tx window on final ACK reception
500 static void rxrpc_zap_tx_window(struct rxrpc_call
*call
)
502 struct rxrpc_skb_priv
*sp
;
504 unsigned long _skb
, *acks_window
;
505 u8 winsz
= call
->acks_winsz
;
508 acks_window
= call
->acks_window
;
509 call
->acks_window
= NULL
;
511 while (CIRC_CNT(call
->acks_head
, call
->acks_tail
, winsz
) > 0) {
512 tail
= call
->acks_tail
;
513 smp_read_barrier_depends();
514 _skb
= acks_window
[tail
] & ~1;
516 call
->acks_tail
= (call
->acks_tail
+ 1) & (winsz
- 1);
518 skb
= (struct sk_buff
*) _skb
;
520 _debug("+++ clear Tx %u", sp
->hdr
.seq
);
528 * process the extra information that may be appended to an ACK packet
530 static void rxrpc_extract_ackinfo(struct rxrpc_call
*call
, struct sk_buff
*skb
,
531 unsigned int latest
, int nAcks
)
533 struct rxrpc_ackinfo ackinfo
;
534 struct rxrpc_peer
*peer
;
537 if (skb_copy_bits(skb
, nAcks
+ 3, &ackinfo
, sizeof(ackinfo
)) < 0) {
538 _leave(" [no ackinfo]");
542 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
544 ntohl(ackinfo
.rxMTU
), ntohl(ackinfo
.maxMTU
),
545 ntohl(ackinfo
.rwind
), ntohl(ackinfo
.jumbo_max
));
547 mtu
= min(ntohl(ackinfo
.rxMTU
), ntohl(ackinfo
.maxMTU
));
549 peer
= call
->conn
->params
.peer
;
550 if (mtu
< peer
->maxdata
) {
551 spin_lock_bh(&peer
->lock
);
553 peer
->mtu
= mtu
+ peer
->hdrsize
;
554 spin_unlock_bh(&peer
->lock
);
555 _net("Net MTU %u (maxdata %u)", peer
->mtu
, peer
->maxdata
);
560 * process packets in the reception queue
562 static int rxrpc_process_rx_queue(struct rxrpc_call
*call
,
565 struct rxrpc_ackpacket ack
;
566 struct rxrpc_skb_priv
*sp
;
575 skb
= skb_dequeue(&call
->rx_queue
);
579 _net("deferred skb %p", skb
);
583 _debug("process %s [st %d]", rxrpc_pkts
[sp
->hdr
.type
], call
->state
);
587 switch (sp
->hdr
.type
) {
588 /* data packets that wind up here have been received out of
589 * order, need security processing or are jumbo packets */
590 case RXRPC_PACKET_TYPE_DATA
:
591 _proto("OOSQ DATA %%%u { #%u }", sp
->hdr
.serial
, sp
->hdr
.seq
);
593 /* secured packets must be verified and possibly decrypted */
594 if (call
->conn
->security
->verify_packet(call
, skb
,
598 rxrpc_insert_oos_packet(call
, skb
);
599 goto process_further
;
601 /* partial ACK to process */
602 case RXRPC_PACKET_TYPE_ACK
:
603 if (skb_copy_bits(skb
, 0, &ack
, sizeof(ack
)) < 0) {
604 _debug("extraction failure");
607 if (!skb_pull(skb
, sizeof(ack
)))
610 latest
= sp
->hdr
.serial
;
611 hard
= ntohl(ack
.firstPacket
);
612 tx
= atomic_read(&call
->sequence
);
614 _proto("Rx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
618 ntohl(ack
.previousPacket
),
620 rxrpc_acks(ack
.reason
),
623 rxrpc_extract_ackinfo(call
, skb
, latest
, ack
.nAcks
);
625 if (ack
.reason
== RXRPC_ACK_PING
) {
626 _proto("Rx ACK %%%u PING Request", latest
);
627 rxrpc_propose_ACK(call
, RXRPC_ACK_PING_RESPONSE
,
628 sp
->hdr
.serial
, true);
631 /* discard any out-of-order or duplicate ACKs */
632 if (latest
- call
->acks_latest
<= 0) {
633 _debug("discard ACK %d <= %d",
634 latest
, call
->acks_latest
);
637 call
->acks_latest
= latest
;
639 if (call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
&&
640 call
->state
!= RXRPC_CALL_CLIENT_AWAIT_REPLY
&&
641 call
->state
!= RXRPC_CALL_SERVER_SEND_REPLY
&&
642 call
->state
!= RXRPC_CALL_SERVER_AWAIT_ACK
)
645 _debug("Tx=%d H=%u S=%d", tx
, call
->acks_hard
, call
->state
);
649 _debug("hard-ACK'd packet %d not transmitted"
655 if ((call
->state
== RXRPC_CALL_CLIENT_AWAIT_REPLY
||
656 call
->state
== RXRPC_CALL_SERVER_AWAIT_ACK
) &&
658 call
->acks_hard
= tx
;
663 rxrpc_rotate_tx_window(call
, hard
- 1);
667 if (hard
- 1 + ack
.nAcks
> tx
) {
668 _debug("soft-ACK'd packet %d+%d not"
669 " transmitted (%d top)",
670 hard
- 1, ack
.nAcks
, tx
);
674 if (rxrpc_process_soft_ACKs(call
, &ack
, skb
) < 0)
679 /* complete ACK to process */
680 case RXRPC_PACKET_TYPE_ACKALL
:
683 /* abort and busy are handled elsewhere */
684 case RXRPC_PACKET_TYPE_BUSY
:
685 case RXRPC_PACKET_TYPE_ABORT
:
688 /* connection level events - also handled elsewhere */
689 case RXRPC_PACKET_TYPE_CHALLENGE
:
690 case RXRPC_PACKET_TYPE_RESPONSE
:
691 case RXRPC_PACKET_TYPE_DEBUG
:
695 /* if we've had a hard ACK that covers all the packets we've sent, then
696 * that ends that phase of the operation */
698 write_lock_bh(&call
->state_lock
);
699 _debug("ack all %d", call
->state
);
701 switch (call
->state
) {
702 case RXRPC_CALL_CLIENT_AWAIT_REPLY
:
703 call
->state
= RXRPC_CALL_CLIENT_RECV_REPLY
;
705 case RXRPC_CALL_SERVER_AWAIT_ACK
:
706 _debug("srv complete");
707 call
->state
= RXRPC_CALL_COMPLETE
;
710 case RXRPC_CALL_CLIENT_SEND_REQUEST
:
711 case RXRPC_CALL_SERVER_RECV_REQUEST
:
712 goto protocol_error_unlock
; /* can't occur yet */
714 write_unlock_bh(&call
->state_lock
);
715 goto discard
; /* assume packet left over from earlier phase */
718 write_unlock_bh(&call
->state_lock
);
720 /* if all the packets we sent are hard-ACK'd, then we can discard
721 * whatever we've got left */
722 _debug("clear Tx %d",
723 CIRC_CNT(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
));
725 del_timer_sync(&call
->resend_timer
);
726 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
727 clear_bit(RXRPC_CALL_EV_RESEND_TIMER
, &call
->events
);
729 if (call
->acks_window
)
730 rxrpc_zap_tx_window(call
);
733 /* post the final ACK message for userspace to pick up */
735 skb
->mark
= RXRPC_SKB_MARK_FINAL_ACK
;
737 rxrpc_get_call(call
);
738 atomic_inc(&call
->skb_count
);
739 spin_lock_bh(&call
->lock
);
740 if (rxrpc_queue_rcv_skb(call
, skb
, true, true) < 0)
742 spin_unlock_bh(&call
->lock
);
743 goto process_further
;
748 goto process_further
;
750 protocol_error_unlock
:
751 write_unlock_bh(&call
->state_lock
);
754 _leave(" = -EPROTO");
759 * post a message to the socket Rx queue for recvmsg() to pick up
761 static int rxrpc_post_message(struct rxrpc_call
*call
, u32 mark
, u32 error
,
764 struct rxrpc_skb_priv
*sp
;
768 _enter("{%d,%lx},%u,%u,%d",
769 call
->debug_id
, call
->flags
, mark
, error
, fatal
);
771 /* remove timers and things for fatal messages */
773 del_timer_sync(&call
->resend_timer
);
774 del_timer_sync(&call
->ack_timer
);
775 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
778 if (mark
!= RXRPC_SKB_MARK_NEW_CALL
&&
779 !test_bit(RXRPC_CALL_HAS_USERID
, &call
->flags
)) {
780 _leave("[no userid]");
784 if (!test_bit(RXRPC_CALL_TERMINAL_MSG
, &call
->flags
)) {
785 skb
= alloc_skb(0, GFP_NOFS
);
794 memset(sp
, 0, sizeof(*sp
));
797 rxrpc_get_call(call
);
798 atomic_inc(&call
->skb_count
);
800 spin_lock_bh(&call
->lock
);
801 ret
= rxrpc_queue_rcv_skb(call
, skb
, true, fatal
);
802 spin_unlock_bh(&call
->lock
);
810 * handle background processing of incoming call packets and ACK / abort
813 void rxrpc_process_call(struct work_struct
*work
)
815 struct rxrpc_call
*call
=
816 container_of(work
, struct rxrpc_call
, processor
);
817 struct rxrpc_wire_header whdr
;
818 struct rxrpc_ackpacket ack
;
819 struct rxrpc_ackinfo ackinfo
;
822 enum rxrpc_call_event genbit
;
826 int loop
, nbit
, ioc
, ret
, mtu
;
827 u32 serial
, abort_code
= RX_PROTOCOL_ERROR
;
830 //printk("\n--------------------\n");
831 _enter("{%d,%s,%lx} [%lu]",
832 call
->debug_id
, rxrpc_call_states
[call
->state
], call
->events
,
833 (jiffies
- call
->creation_jif
) / (HZ
/ 10));
835 if (test_and_set_bit(RXRPC_CALL_PROC_BUSY
, &call
->flags
)) {
836 _debug("XXXXXXXXXXXXX RUNNING ON MULTIPLE CPUS XXXXXXXXXXXXX");
843 /* there's a good chance we're going to have to send a message, so set
844 * one up in advance */
845 msg
.msg_name
= &call
->conn
->params
.peer
->srx
.transport
;
846 msg
.msg_namelen
= call
->conn
->params
.peer
->srx
.transport_len
;
847 msg
.msg_control
= NULL
;
848 msg
.msg_controllen
= 0;
851 whdr
.epoch
= htonl(call
->conn
->proto
.epoch
);
852 whdr
.cid
= htonl(call
->cid
);
853 whdr
.callNumber
= htonl(call
->call_id
);
855 whdr
.type
= RXRPC_PACKET_TYPE_ACK
;
856 whdr
.flags
= call
->conn
->out_clientflag
;
858 whdr
.securityIndex
= call
->conn
->security_ix
;
860 whdr
.serviceId
= htons(call
->service_id
);
862 memset(iov
, 0, sizeof(iov
));
863 iov
[0].iov_base
= &whdr
;
864 iov
[0].iov_len
= sizeof(whdr
);
867 /* deal with events of a final nature */
868 if (test_bit(RXRPC_CALL_EV_RCVD_ERROR
, &call
->events
)) {
869 enum rxrpc_skb_mark mark
;
872 clear_bit(RXRPC_CALL_EV_CONN_ABORT
, &call
->events
);
873 clear_bit(RXRPC_CALL_EV_REJECT_BUSY
, &call
->events
);
874 clear_bit(RXRPC_CALL_EV_ABORT
, &call
->events
);
876 error
= call
->error_report
;
877 if (error
< RXRPC_LOCAL_ERROR_OFFSET
) {
878 mark
= RXRPC_SKB_MARK_NET_ERROR
;
879 _debug("post net error %d", error
);
881 mark
= RXRPC_SKB_MARK_LOCAL_ERROR
;
882 error
-= RXRPC_LOCAL_ERROR_OFFSET
;
883 _debug("post net local error %d", error
);
886 if (rxrpc_post_message(call
, mark
, error
, true) < 0)
888 clear_bit(RXRPC_CALL_EV_RCVD_ERROR
, &call
->events
);
892 if (test_bit(RXRPC_CALL_EV_CONN_ABORT
, &call
->events
)) {
893 ASSERTCMP(call
->state
, >, RXRPC_CALL_COMPLETE
);
895 clear_bit(RXRPC_CALL_EV_REJECT_BUSY
, &call
->events
);
896 clear_bit(RXRPC_CALL_EV_ABORT
, &call
->events
);
898 _debug("post conn abort");
900 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_LOCAL_ERROR
,
901 call
->conn
->error
, true) < 0)
903 clear_bit(RXRPC_CALL_EV_CONN_ABORT
, &call
->events
);
907 if (test_bit(RXRPC_CALL_EV_REJECT_BUSY
, &call
->events
)) {
908 whdr
.type
= RXRPC_PACKET_TYPE_BUSY
;
909 genbit
= RXRPC_CALL_EV_REJECT_BUSY
;
913 if (test_bit(RXRPC_CALL_EV_ABORT
, &call
->events
)) {
914 ASSERTCMP(call
->state
, >, RXRPC_CALL_COMPLETE
);
916 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_LOCAL_ERROR
,
917 ECONNABORTED
, true) < 0)
919 whdr
.type
= RXRPC_PACKET_TYPE_ABORT
;
920 data
= htonl(call
->local_abort
);
921 iov
[1].iov_base
= &data
;
922 iov
[1].iov_len
= sizeof(data
);
923 genbit
= RXRPC_CALL_EV_ABORT
;
927 if (test_bit(RXRPC_CALL_EV_ACK_FINAL
, &call
->events
)) {
928 genbit
= RXRPC_CALL_EV_ACK_FINAL
;
930 ack
.bufferSpace
= htons(8);
933 ack
.reason
= RXRPC_ACK_IDLE
;
935 call
->ackr_reason
= 0;
937 spin_lock_bh(&call
->lock
);
938 ack
.serial
= htonl(call
->ackr_serial
);
939 ack
.previousPacket
= htonl(call
->ackr_prev_seq
);
940 ack
.firstPacket
= htonl(call
->rx_data_eaten
+ 1);
941 spin_unlock_bh(&call
->lock
);
945 iov
[1].iov_base
= &ack
;
946 iov
[1].iov_len
= sizeof(ack
);
947 iov
[2].iov_base
= &pad
;
949 iov
[3].iov_base
= &ackinfo
;
950 iov
[3].iov_len
= sizeof(ackinfo
);
954 if (call
->events
& ((1 << RXRPC_CALL_EV_RCVD_BUSY
) |
955 (1 << RXRPC_CALL_EV_RCVD_ABORT
))
959 if (test_bit(RXRPC_CALL_EV_RCVD_ABORT
, &call
->events
))
960 mark
= RXRPC_SKB_MARK_REMOTE_ABORT
;
962 mark
= RXRPC_SKB_MARK_BUSY
;
964 _debug("post abort/busy");
965 rxrpc_clear_tx_window(call
);
966 if (rxrpc_post_message(call
, mark
, ECONNABORTED
, true) < 0)
969 clear_bit(RXRPC_CALL_EV_RCVD_BUSY
, &call
->events
);
970 clear_bit(RXRPC_CALL_EV_RCVD_ABORT
, &call
->events
);
974 if (test_and_clear_bit(RXRPC_CALL_EV_RCVD_ACKALL
, &call
->events
)) {
975 _debug("do implicit ackall");
976 rxrpc_clear_tx_window(call
);
979 if (test_bit(RXRPC_CALL_EV_LIFE_TIMER
, &call
->events
)) {
980 write_lock_bh(&call
->state_lock
);
981 if (call
->state
<= RXRPC_CALL_COMPLETE
) {
982 call
->state
= RXRPC_CALL_LOCALLY_ABORTED
;
983 call
->local_abort
= RX_CALL_TIMEOUT
;
984 set_bit(RXRPC_CALL_EV_ABORT
, &call
->events
);
986 write_unlock_bh(&call
->state_lock
);
988 _debug("post timeout");
989 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_LOCAL_ERROR
,
993 clear_bit(RXRPC_CALL_EV_LIFE_TIMER
, &call
->events
);
997 /* deal with assorted inbound messages */
998 if (!skb_queue_empty(&call
->rx_queue
)) {
999 switch (rxrpc_process_rx_queue(call
, &abort_code
)) {
1008 rxrpc_abort_call(call
, abort_code
);
1013 /* handle resending */
1014 if (test_and_clear_bit(RXRPC_CALL_EV_RESEND_TIMER
, &call
->events
))
1015 rxrpc_resend_timer(call
);
1016 if (test_and_clear_bit(RXRPC_CALL_EV_RESEND
, &call
->events
))
1019 /* consider sending an ordinary ACK */
1020 if (test_bit(RXRPC_CALL_EV_ACK
, &call
->events
)) {
1021 _debug("send ACK: window: %d - %d { %lx }",
1022 call
->rx_data_eaten
, call
->ackr_win_top
,
1023 call
->ackr_window
[0]);
1025 if (call
->state
> RXRPC_CALL_SERVER_ACK_REQUEST
&&
1026 call
->ackr_reason
!= RXRPC_ACK_PING_RESPONSE
) {
1027 /* ACK by sending reply DATA packet in this state */
1028 clear_bit(RXRPC_CALL_EV_ACK
, &call
->events
);
1029 goto maybe_reschedule
;
1032 genbit
= RXRPC_CALL_EV_ACK
;
1034 acks
= kzalloc(call
->ackr_win_top
- call
->rx_data_eaten
,
1039 //hdr.flags = RXRPC_SLOW_START_OK;
1040 ack
.bufferSpace
= htons(8);
1043 spin_lock_bh(&call
->lock
);
1044 ack
.reason
= call
->ackr_reason
;
1045 ack
.serial
= htonl(call
->ackr_serial
);
1046 ack
.previousPacket
= htonl(call
->ackr_prev_seq
);
1047 ack
.firstPacket
= htonl(call
->rx_data_eaten
+ 1);
1050 for (loop
= 0; loop
< RXRPC_ACKR_WINDOW_ASZ
; loop
++) {
1051 nbit
= loop
* BITS_PER_LONG
;
1052 for (bits
= call
->ackr_window
[loop
]; bits
; bits
>>= 1
1054 _debug("- l=%d n=%d b=%lx", loop
, nbit
, bits
);
1056 acks
[nbit
] = RXRPC_ACK_TYPE_ACK
;
1057 ack
.nAcks
= nbit
+ 1;
1062 call
->ackr_reason
= 0;
1063 spin_unlock_bh(&call
->lock
);
1067 iov
[1].iov_base
= &ack
;
1068 iov
[1].iov_len
= sizeof(ack
);
1069 iov
[2].iov_base
= acks
;
1070 iov
[2].iov_len
= ack
.nAcks
;
1071 iov
[3].iov_base
= &pad
;
1073 iov
[4].iov_base
= &ackinfo
;
1074 iov
[4].iov_len
= sizeof(ackinfo
);
1076 switch (ack
.reason
) {
1077 case RXRPC_ACK_REQUESTED
:
1078 case RXRPC_ACK_DUPLICATE
:
1079 case RXRPC_ACK_OUT_OF_SEQUENCE
:
1080 case RXRPC_ACK_EXCEEDS_WINDOW
:
1081 case RXRPC_ACK_NOSPACE
:
1082 case RXRPC_ACK_PING
:
1083 case RXRPC_ACK_PING_RESPONSE
:
1084 goto send_ACK_with_skew
;
1085 case RXRPC_ACK_DELAY
:
1086 case RXRPC_ACK_IDLE
:
1091 /* handle completion of security negotiations on an incoming
1093 if (test_and_clear_bit(RXRPC_CALL_EV_SECURED
, &call
->events
)) {
1095 spin_lock_bh(&call
->lock
);
1097 if (call
->state
== RXRPC_CALL_SERVER_SECURING
) {
1099 write_lock(&call
->socket
->call_lock
);
1100 if (!test_bit(RXRPC_CALL_RELEASED
, &call
->flags
) &&
1101 !test_bit(RXRPC_CALL_EV_RELEASE
, &call
->events
)) {
1102 _debug("not released");
1103 call
->state
= RXRPC_CALL_SERVER_ACCEPTING
;
1104 list_move_tail(&call
->accept_link
,
1105 &call
->socket
->acceptq
);
1107 write_unlock(&call
->socket
->call_lock
);
1108 read_lock(&call
->state_lock
);
1109 if (call
->state
< RXRPC_CALL_COMPLETE
)
1110 set_bit(RXRPC_CALL_EV_POST_ACCEPT
, &call
->events
);
1111 read_unlock(&call
->state_lock
);
1114 spin_unlock_bh(&call
->lock
);
1115 if (!test_bit(RXRPC_CALL_EV_POST_ACCEPT
, &call
->events
))
1116 goto maybe_reschedule
;
1119 /* post a notification of an acceptable connection to the app */
1120 if (test_bit(RXRPC_CALL_EV_POST_ACCEPT
, &call
->events
)) {
1121 _debug("post accept");
1122 if (rxrpc_post_message(call
, RXRPC_SKB_MARK_NEW_CALL
,
1125 clear_bit(RXRPC_CALL_EV_POST_ACCEPT
, &call
->events
);
1126 goto maybe_reschedule
;
1129 /* handle incoming call acceptance */
1130 if (test_and_clear_bit(RXRPC_CALL_EV_ACCEPTED
, &call
->events
)) {
1132 ASSERTCMP(call
->rx_data_post
, ==, 0);
1133 call
->rx_data_post
= 1;
1134 read_lock_bh(&call
->state_lock
);
1135 if (call
->state
< RXRPC_CALL_COMPLETE
)
1136 set_bit(RXRPC_CALL_EV_DRAIN_RX_OOS
, &call
->events
);
1137 read_unlock_bh(&call
->state_lock
);
1140 /* drain the out of sequence received packet queue into the packet Rx
1142 if (test_and_clear_bit(RXRPC_CALL_EV_DRAIN_RX_OOS
, &call
->events
)) {
1143 while (call
->rx_data_post
== call
->rx_first_oos
)
1144 if (rxrpc_drain_rx_oos_queue(call
) < 0)
1146 goto maybe_reschedule
;
1149 if (test_bit(RXRPC_CALL_EV_RELEASE
, &call
->events
)) {
1150 rxrpc_release_call(call
);
1151 clear_bit(RXRPC_CALL_EV_RELEASE
, &call
->events
);
1154 /* other events may have been raised since we started checking */
1155 goto maybe_reschedule
;
1158 ack
.maxSkew
= htons(atomic_read(&call
->conn
->hi_serial
) -
1161 mtu
= call
->conn
->params
.peer
->if_mtu
;
1162 mtu
-= call
->conn
->params
.peer
->hdrsize
;
1163 ackinfo
.maxMTU
= htonl(mtu
);
1164 ackinfo
.rwind
= htonl(rxrpc_rx_window_size
);
1166 /* permit the peer to send us jumbo packets if it wants to */
1167 ackinfo
.rxMTU
= htonl(rxrpc_rx_mtu
);
1168 ackinfo
.jumbo_max
= htonl(rxrpc_rx_jumbo_max
);
1170 serial
= atomic_inc_return(&call
->conn
->serial
);
1171 whdr
.serial
= htonl(serial
);
1172 _proto("Tx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
1175 ntohl(ack
.firstPacket
),
1176 ntohl(ack
.previousPacket
),
1178 rxrpc_acks(ack
.reason
),
1181 del_timer_sync(&call
->ack_timer
);
1183 set_bit(RXRPC_CALL_TX_SOFT_ACK
, &call
->flags
);
1184 goto send_message_2
;
1187 _debug("send message");
1189 serial
= atomic_inc_return(&call
->conn
->serial
);
1190 whdr
.serial
= htonl(serial
);
1191 _proto("Tx %s %%%u", rxrpc_pkts
[whdr
.type
], serial
);
1194 len
= iov
[0].iov_len
;
1196 if (iov
[4].iov_len
) {
1198 len
+= iov
[4].iov_len
;
1199 len
+= iov
[3].iov_len
;
1200 len
+= iov
[2].iov_len
;
1201 len
+= iov
[1].iov_len
;
1202 } else if (iov
[3].iov_len
) {
1204 len
+= iov
[3].iov_len
;
1205 len
+= iov
[2].iov_len
;
1206 len
+= iov
[1].iov_len
;
1207 } else if (iov
[2].iov_len
) {
1209 len
+= iov
[2].iov_len
;
1210 len
+= iov
[1].iov_len
;
1211 } else if (iov
[1].iov_len
) {
1213 len
+= iov
[1].iov_len
;
1216 ret
= kernel_sendmsg(call
->conn
->params
.local
->socket
,
1217 &msg
, iov
, ioc
, len
);
1219 _debug("sendmsg failed: %d", ret
);
1220 read_lock_bh(&call
->state_lock
);
1221 if (call
->state
< RXRPC_CALL_DEAD
)
1222 rxrpc_queue_call(call
);
1223 read_unlock_bh(&call
->state_lock
);
1228 case RXRPC_CALL_EV_ABORT
:
1229 clear_bit(genbit
, &call
->events
);
1230 clear_bit(RXRPC_CALL_EV_RCVD_ABORT
, &call
->events
);
1233 case RXRPC_CALL_EV_ACK_FINAL
:
1234 write_lock_bh(&call
->state_lock
);
1235 if (call
->state
== RXRPC_CALL_CLIENT_FINAL_ACK
)
1236 call
->state
= RXRPC_CALL_COMPLETE
;
1237 write_unlock_bh(&call
->state_lock
);
1241 clear_bit(genbit
, &call
->events
);
1242 switch (call
->state
) {
1243 case RXRPC_CALL_CLIENT_AWAIT_REPLY
:
1244 case RXRPC_CALL_CLIENT_RECV_REPLY
:
1245 case RXRPC_CALL_SERVER_RECV_REQUEST
:
1246 case RXRPC_CALL_SERVER_ACK_REQUEST
:
1247 _debug("start ACK timer");
1248 rxrpc_propose_ACK(call
, RXRPC_ACK_DELAY
,
1249 call
->ackr_serial
, false);
1253 goto maybe_reschedule
;
1257 del_timer_sync(&call
->ack_timer
);
1258 if (test_and_clear_bit(RXRPC_CALL_EV_ACK_FINAL
, &call
->events
))
1259 rxrpc_put_call(call
);
1260 clear_bit(RXRPC_CALL_EV_ACK
, &call
->events
);
1263 if (call
->events
|| !skb_queue_empty(&call
->rx_queue
)) {
1264 read_lock_bh(&call
->state_lock
);
1265 if (call
->state
< RXRPC_CALL_DEAD
)
1266 rxrpc_queue_call(call
);
1267 read_unlock_bh(&call
->state_lock
);
1270 /* don't leave aborted connections on the accept queue */
1271 if (call
->state
>= RXRPC_CALL_COMPLETE
&&
1272 !list_empty(&call
->accept_link
)) {
1273 _debug("X unlinking once-pending call %p { e=%lx f=%lx c=%x }",
1274 call
, call
->events
, call
->flags
, call
->conn
->proto
.cid
);
1276 read_lock_bh(&call
->state_lock
);
1277 if (!test_bit(RXRPC_CALL_RELEASED
, &call
->flags
) &&
1278 !test_and_set_bit(RXRPC_CALL_EV_RELEASE
, &call
->events
))
1279 rxrpc_queue_call(call
);
1280 read_unlock_bh(&call
->state_lock
);
1284 clear_bit(RXRPC_CALL_PROC_BUSY
, &call
->flags
);
1287 /* because we don't want two CPUs both processing the work item for one
1288 * call at the same time, we use a flag to note when it's busy; however
1289 * this means there's a race between clearing the flag and setting the
1290 * work pending bit and the work item being processed again */
1291 if (call
->events
&& !work_pending(&call
->processor
)) {
1292 _debug("jumpstart %x", call
->conn
->proto
.cid
);
1293 rxrpc_queue_call(call
);
1300 _debug("out of memory");
1301 goto maybe_reschedule
;