tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* Management of Tx window, Tx resend, ACKs and out-of-sequence reception
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
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.
10 */
11
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>
18#include <net/sock.h>
19#include <net/af_rxrpc.h>
20#include "ar-internal.h"
21
22/*
23 * propose an ACK be sent
24 */
25void __rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason,
26 u32 serial, bool immediate)
27{
28 unsigned long expiry;
29 s8 prior = rxrpc_ack_priority[ack_reason];
30
31 ASSERTCMP(prior, >, 0);
32
33 _enter("{%d},%s,%%%x,%u",
34 call->debug_id, rxrpc_acks(ack_reason), serial, immediate);
35
36 if (prior < rxrpc_ack_priority[call->ackr_reason]) {
37 if (immediate)
38 goto cancel_timer;
39 return;
40 }
41
42 /* update DELAY, IDLE, REQUESTED and PING_RESPONSE ACK serial
43 * numbers */
44 if (prior == rxrpc_ack_priority[call->ackr_reason]) {
45 if (prior <= 4)
46 call->ackr_serial = serial;
47 if (immediate)
48 goto cancel_timer;
49 return;
50 }
51
52 call->ackr_reason = ack_reason;
53 call->ackr_serial = serial;
54
55 switch (ack_reason) {
56 case RXRPC_ACK_DELAY:
57 _debug("run delay timer");
58 expiry = rxrpc_soft_ack_delay;
59 goto run_timer;
60
61 case RXRPC_ACK_IDLE:
62 if (!immediate) {
63 _debug("run defer timer");
64 expiry = rxrpc_idle_ack_delay;
65 goto run_timer;
66 }
67 goto cancel_timer;
68
69 case RXRPC_ACK_REQUESTED:
70 expiry = rxrpc_requested_ack_delay;
71 if (!expiry)
72 goto cancel_timer;
73 if (!immediate || serial == 1) {
74 _debug("run defer timer");
75 goto run_timer;
76 }
77
78 default:
79 _debug("immediate ACK");
80 goto cancel_timer;
81 }
82
83run_timer:
84 expiry += jiffies;
85 if (!timer_pending(&call->ack_timer) ||
86 time_after(call->ack_timer.expires, expiry))
87 mod_timer(&call->ack_timer, expiry);
88 return;
89
90cancel_timer:
91 _debug("cancel timer %%%u", serial);
92 try_to_del_timer_sync(&call->ack_timer);
93 read_lock_bh(&call->state_lock);
94 if (call->state <= RXRPC_CALL_COMPLETE &&
95 !test_and_set_bit(RXRPC_CALL_EV_ACK, &call->events))
96 rxrpc_queue_call(call);
97 read_unlock_bh(&call->state_lock);
98}
99
100/*
101 * propose an ACK be sent, locking the call structure
102 */
103void rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason,
104 u32 serial, bool immediate)
105{
106 s8 prior = rxrpc_ack_priority[ack_reason];
107
108 if (prior > rxrpc_ack_priority[call->ackr_reason]) {
109 spin_lock_bh(&call->lock);
110 __rxrpc_propose_ACK(call, ack_reason, serial, immediate);
111 spin_unlock_bh(&call->lock);
112 }
113}
114
115/*
116 * set the resend timer
117 */
118static void rxrpc_set_resend(struct rxrpc_call *call, u8 resend,
119 unsigned long resend_at)
120{
121 read_lock_bh(&call->state_lock);
122 if (call->state >= RXRPC_CALL_COMPLETE)
123 resend = 0;
124
125 if (resend & 1) {
126 _debug("SET RESEND");
127 set_bit(RXRPC_CALL_EV_RESEND, &call->events);
128 }
129
130 if (resend & 2) {
131 _debug("MODIFY RESEND TIMER");
132 set_bit(RXRPC_CALL_RUN_RTIMER, &call->flags);
133 mod_timer(&call->resend_timer, resend_at);
134 } else {
135 _debug("KILL RESEND TIMER");
136 del_timer_sync(&call->resend_timer);
137 clear_bit(RXRPC_CALL_EV_RESEND_TIMER, &call->events);
138 clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags);
139 }
140 read_unlock_bh(&call->state_lock);
141}
142
143/*
144 * resend packets
145 */
146static void rxrpc_resend(struct rxrpc_call *call)
147{
148 struct rxrpc_wire_header *whdr;
149 struct rxrpc_skb_priv *sp;
150 struct sk_buff *txb;
151 unsigned long *p_txb, resend_at;
152 bool stop;
153 int loop;
154 u8 resend;
155
156 _enter("{%d,%d,%d,%d},",
157 call->acks_hard, call->acks_unacked,
158 atomic_read(&call->sequence),
159 CIRC_CNT(call->acks_head, call->acks_tail, call->acks_winsz));
160
161 stop = false;
162 resend = 0;
163 resend_at = 0;
164
165 for (loop = call->acks_tail;
166 loop != call->acks_head || stop;
167 loop = (loop + 1) & (call->acks_winsz - 1)
168 ) {
169 p_txb = call->acks_window + loop;
170 smp_read_barrier_depends();
171 if (*p_txb & 1)
172 continue;
173
174 txb = (struct sk_buff *) *p_txb;
175 sp = rxrpc_skb(txb);
176
177 if (sp->need_resend) {
178 sp->need_resend = false;
179
180 /* each Tx packet has a new serial number */
181 sp->hdr.serial = atomic_inc_return(&call->conn->serial);
182
183 whdr = (struct rxrpc_wire_header *)txb->head;
184 whdr->serial = htonl(sp->hdr.serial);
185
186 _proto("Tx DATA %%%u { #%d }",
187 sp->hdr.serial, sp->hdr.seq);
188 if (rxrpc_send_packet(call->conn->trans, txb) < 0) {
189 stop = true;
190 sp->resend_at = jiffies + 3;
191 } else {
192 sp->resend_at =
193 jiffies + rxrpc_resend_timeout;
194 }
195 }
196
197 if (time_after_eq(jiffies + 1, sp->resend_at)) {
198 sp->need_resend = true;
199 resend |= 1;
200 } else if (resend & 2) {
201 if (time_before(sp->resend_at, resend_at))
202 resend_at = sp->resend_at;
203 } else {
204 resend_at = sp->resend_at;
205 resend |= 2;
206 }
207 }
208
209 rxrpc_set_resend(call, resend, resend_at);
210 _leave("");
211}
212
213/*
214 * handle resend timer expiry
215 */
216static void rxrpc_resend_timer(struct rxrpc_call *call)
217{
218 struct rxrpc_skb_priv *sp;
219 struct sk_buff *txb;
220 unsigned long *p_txb, resend_at;
221 int loop;
222 u8 resend;
223
224 _enter("%d,%d,%d",
225 call->acks_tail, call->acks_unacked, call->acks_head);
226
227 if (call->state >= RXRPC_CALL_COMPLETE)
228 return;
229
230 resend = 0;
231 resend_at = 0;
232
233 for (loop = call->acks_unacked;
234 loop != call->acks_head;
235 loop = (loop + 1) & (call->acks_winsz - 1)
236 ) {
237 p_txb = call->acks_window + loop;
238 smp_read_barrier_depends();
239 txb = (struct sk_buff *) (*p_txb & ~1);
240 sp = rxrpc_skb(txb);
241
242 ASSERT(!(*p_txb & 1));
243
244 if (sp->need_resend) {
245 ;
246 } else if (time_after_eq(jiffies + 1, sp->resend_at)) {
247 sp->need_resend = true;
248 resend |= 1;
249 } else if (resend & 2) {
250 if (time_before(sp->resend_at, resend_at))
251 resend_at = sp->resend_at;
252 } else {
253 resend_at = sp->resend_at;
254 resend |= 2;
255 }
256 }
257
258 rxrpc_set_resend(call, resend, resend_at);
259 _leave("");
260}
261
262/*
263 * process soft ACKs of our transmitted packets
264 * - these indicate packets the peer has or has not received, but hasn't yet
265 * given to the consumer, and so can still be discarded and re-requested
266 */
267static int rxrpc_process_soft_ACKs(struct rxrpc_call *call,
268 struct rxrpc_ackpacket *ack,
269 struct sk_buff *skb)
270{
271 struct rxrpc_skb_priv *sp;
272 struct sk_buff *txb;
273 unsigned long *p_txb, resend_at;
274 int loop;
275 u8 sacks[RXRPC_MAXACKS], resend;
276
277 _enter("{%d,%d},{%d},",
278 call->acks_hard,
279 CIRC_CNT(call->acks_head, call->acks_tail, call->acks_winsz),
280 ack->nAcks);
281
282 if (skb_copy_bits(skb, 0, sacks, ack->nAcks) < 0)
283 goto protocol_error;
284
285 resend = 0;
286 resend_at = 0;
287 for (loop = 0; loop < ack->nAcks; loop++) {
288 p_txb = call->acks_window;
289 p_txb += (call->acks_tail + loop) & (call->acks_winsz - 1);
290 smp_read_barrier_depends();
291 txb = (struct sk_buff *) (*p_txb & ~1);
292 sp = rxrpc_skb(txb);
293
294 switch (sacks[loop]) {
295 case RXRPC_ACK_TYPE_ACK:
296 sp->need_resend = false;
297 *p_txb |= 1;
298 break;
299 case RXRPC_ACK_TYPE_NACK:
300 sp->need_resend = true;
301 *p_txb &= ~1;
302 resend = 1;
303 break;
304 default:
305 _debug("Unsupported ACK type %d", sacks[loop]);
306 goto protocol_error;
307 }
308 }
309
310 smp_mb();
311 call->acks_unacked = (call->acks_tail + loop) & (call->acks_winsz - 1);
312
313 /* anything not explicitly ACK'd is implicitly NACK'd, but may just not
314 * have been received or processed yet by the far end */
315 for (loop = call->acks_unacked;
316 loop != call->acks_head;
317 loop = (loop + 1) & (call->acks_winsz - 1)
318 ) {
319 p_txb = call->acks_window + loop;
320 smp_read_barrier_depends();
321 txb = (struct sk_buff *) (*p_txb & ~1);
322 sp = rxrpc_skb(txb);
323
324 if (*p_txb & 1) {
325 /* packet must have been discarded */
326 sp->need_resend = true;
327 *p_txb &= ~1;
328 resend |= 1;
329 } else if (sp->need_resend) {
330 ;
331 } else if (time_after_eq(jiffies + 1, sp->resend_at)) {
332 sp->need_resend = true;
333 resend |= 1;
334 } else if (resend & 2) {
335 if (time_before(sp->resend_at, resend_at))
336 resend_at = sp->resend_at;
337 } else {
338 resend_at = sp->resend_at;
339 resend |= 2;
340 }
341 }
342
343 rxrpc_set_resend(call, resend, resend_at);
344 _leave(" = 0");
345 return 0;
346
347protocol_error:
348 _leave(" = -EPROTO");
349 return -EPROTO;
350}
351
352/*
353 * discard hard-ACK'd packets from the Tx window
354 */
355static void rxrpc_rotate_tx_window(struct rxrpc_call *call, u32 hard)
356{
357 unsigned long _skb;
358 int tail = call->acks_tail, old_tail;
359 int win = CIRC_CNT(call->acks_head, tail, call->acks_winsz);
360
361 _enter("{%u,%u},%u", call->acks_hard, win, hard);
362
363 ASSERTCMP(hard - call->acks_hard, <=, win);
364
365 while (call->acks_hard < hard) {
366 smp_read_barrier_depends();
367 _skb = call->acks_window[tail] & ~1;
368 rxrpc_free_skb((struct sk_buff *) _skb);
369 old_tail = tail;
370 tail = (tail + 1) & (call->acks_winsz - 1);
371 call->acks_tail = tail;
372 if (call->acks_unacked == old_tail)
373 call->acks_unacked = tail;
374 call->acks_hard++;
375 }
376
377 wake_up(&call->tx_waitq);
378}
379
380/*
381 * clear the Tx window in the event of a failure
382 */
383static void rxrpc_clear_tx_window(struct rxrpc_call *call)
384{
385 rxrpc_rotate_tx_window(call, atomic_read(&call->sequence));
386}
387
388/*
389 * drain the out of sequence received packet queue into the packet Rx queue
390 */
391static int rxrpc_drain_rx_oos_queue(struct rxrpc_call *call)
392{
393 struct rxrpc_skb_priv *sp;
394 struct sk_buff *skb;
395 bool terminal;
396 int ret;
397
398 _enter("{%d,%d}", call->rx_data_post, call->rx_first_oos);
399
400 spin_lock_bh(&call->lock);
401
402 ret = -ECONNRESET;
403 if (test_bit(RXRPC_CALL_RELEASED, &call->flags))
404 goto socket_unavailable;
405
406 skb = skb_dequeue(&call->rx_oos_queue);
407 if (skb) {
408 sp = rxrpc_skb(skb);
409
410 _debug("drain OOS packet %d [%d]",
411 sp->hdr.seq, call->rx_first_oos);
412
413 if (sp->hdr.seq != call->rx_first_oos) {
414 skb_queue_head(&call->rx_oos_queue, skb);
415 call->rx_first_oos = rxrpc_skb(skb)->hdr.seq;
416 _debug("requeue %p {%u}", skb, call->rx_first_oos);
417 } else {
418 skb->mark = RXRPC_SKB_MARK_DATA;
419 terminal = ((sp->hdr.flags & RXRPC_LAST_PACKET) &&
420 !(sp->hdr.flags & RXRPC_CLIENT_INITIATED));
421 ret = rxrpc_queue_rcv_skb(call, skb, true, terminal);
422 BUG_ON(ret < 0);
423 _debug("drain #%u", call->rx_data_post);
424 call->rx_data_post++;
425
426 /* find out what the next packet is */
427 skb = skb_peek(&call->rx_oos_queue);
428 if (skb)
429 call->rx_first_oos = rxrpc_skb(skb)->hdr.seq;
430 else
431 call->rx_first_oos = 0;
432 _debug("peek %p {%u}", skb, call->rx_first_oos);
433 }
434 }
435
436 ret = 0;
437socket_unavailable:
438 spin_unlock_bh(&call->lock);
439 _leave(" = %d", ret);
440 return ret;
441}
442
443/*
444 * insert an out of sequence packet into the buffer
445 */
446static void rxrpc_insert_oos_packet(struct rxrpc_call *call,
447 struct sk_buff *skb)
448{
449 struct rxrpc_skb_priv *sp, *psp;
450 struct sk_buff *p;
451 u32 seq;
452
453 sp = rxrpc_skb(skb);
454 seq = sp->hdr.seq;
455 _enter(",,{%u}", seq);
456
457 skb->destructor = rxrpc_packet_destructor;
458 ASSERTCMP(sp->call, ==, NULL);
459 sp->call = call;
460 rxrpc_get_call(call);
461
462 /* insert into the buffer in sequence order */
463 spin_lock_bh(&call->lock);
464
465 skb_queue_walk(&call->rx_oos_queue, p) {
466 psp = rxrpc_skb(p);
467 if (psp->hdr.seq > seq) {
468 _debug("insert oos #%u before #%u", seq, psp->hdr.seq);
469 skb_insert(p, skb, &call->rx_oos_queue);
470 goto inserted;
471 }
472 }
473
474 _debug("append oos #%u", seq);
475 skb_queue_tail(&call->rx_oos_queue, skb);
476inserted:
477
478 /* we might now have a new front to the queue */
479 if (call->rx_first_oos == 0 || seq < call->rx_first_oos)
480 call->rx_first_oos = seq;
481
482 read_lock(&call->state_lock);
483 if (call->state < RXRPC_CALL_COMPLETE &&
484 call->rx_data_post == call->rx_first_oos) {
485 _debug("drain rx oos now");
486 set_bit(RXRPC_CALL_EV_DRAIN_RX_OOS, &call->events);
487 }
488 read_unlock(&call->state_lock);
489
490 spin_unlock_bh(&call->lock);
491 _leave(" [stored #%u]", call->rx_first_oos);
492}
493
494/*
495 * clear the Tx window on final ACK reception
496 */
497static void rxrpc_zap_tx_window(struct rxrpc_call *call)
498{
499 struct rxrpc_skb_priv *sp;
500 struct sk_buff *skb;
501 unsigned long _skb, *acks_window;
502 u8 winsz = call->acks_winsz;
503 int tail;
504
505 acks_window = call->acks_window;
506 call->acks_window = NULL;
507
508 while (CIRC_CNT(call->acks_head, call->acks_tail, winsz) > 0) {
509 tail = call->acks_tail;
510 smp_read_barrier_depends();
511 _skb = acks_window[tail] & ~1;
512 smp_mb();
513 call->acks_tail = (call->acks_tail + 1) & (winsz - 1);
514
515 skb = (struct sk_buff *) _skb;
516 sp = rxrpc_skb(skb);
517 _debug("+++ clear Tx %u", sp->hdr.seq);
518 rxrpc_free_skb(skb);
519 }
520
521 kfree(acks_window);
522}
523
524/*
525 * process the extra information that may be appended to an ACK packet
526 */
527static void rxrpc_extract_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
528 unsigned int latest, int nAcks)
529{
530 struct rxrpc_ackinfo ackinfo;
531 struct rxrpc_peer *peer;
532 unsigned int mtu;
533
534 if (skb_copy_bits(skb, nAcks + 3, &ackinfo, sizeof(ackinfo)) < 0) {
535 _leave(" [no ackinfo]");
536 return;
537 }
538
539 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
540 latest,
541 ntohl(ackinfo.rxMTU), ntohl(ackinfo.maxMTU),
542 ntohl(ackinfo.rwind), ntohl(ackinfo.jumbo_max));
543
544 mtu = min(ntohl(ackinfo.rxMTU), ntohl(ackinfo.maxMTU));
545
546 peer = call->conn->trans->peer;
547 if (mtu < peer->maxdata) {
548 spin_lock_bh(&peer->lock);
549 peer->maxdata = mtu;
550 peer->mtu = mtu + peer->hdrsize;
551 spin_unlock_bh(&peer->lock);
552 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
553 }
554}
555
556/*
557 * process packets in the reception queue
558 */
559static int rxrpc_process_rx_queue(struct rxrpc_call *call,
560 u32 *_abort_code)
561{
562 struct rxrpc_ackpacket ack;
563 struct rxrpc_skb_priv *sp;
564 struct sk_buff *skb;
565 bool post_ACK;
566 int latest;
567 u32 hard, tx;
568
569 _enter("");
570
571process_further:
572 skb = skb_dequeue(&call->rx_queue);
573 if (!skb)
574 return -EAGAIN;
575
576 _net("deferred skb %p", skb);
577
578 sp = rxrpc_skb(skb);
579
580 _debug("process %s [st %d]", rxrpc_pkts[sp->hdr.type], call->state);
581
582 post_ACK = false;
583
584 switch (sp->hdr.type) {
585 /* data packets that wind up here have been received out of
586 * order, need security processing or are jumbo packets */
587 case RXRPC_PACKET_TYPE_DATA:
588 _proto("OOSQ DATA %%%u { #%u }", sp->hdr.serial, sp->hdr.seq);
589
590 /* secured packets must be verified and possibly decrypted */
591 if (call->conn->security->verify_packet(call, skb,
592 _abort_code) < 0)
593 goto protocol_error;
594
595 rxrpc_insert_oos_packet(call, skb);
596 goto process_further;
597
598 /* partial ACK to process */
599 case RXRPC_PACKET_TYPE_ACK:
600 if (skb_copy_bits(skb, 0, &ack, sizeof(ack)) < 0) {
601 _debug("extraction failure");
602 goto protocol_error;
603 }
604 if (!skb_pull(skb, sizeof(ack)))
605 BUG();
606
607 latest = sp->hdr.serial;
608 hard = ntohl(ack.firstPacket);
609 tx = atomic_read(&call->sequence);
610
611 _proto("Rx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
612 latest,
613 ntohs(ack.maxSkew),
614 hard,
615 ntohl(ack.previousPacket),
616 ntohl(ack.serial),
617 rxrpc_acks(ack.reason),
618 ack.nAcks);
619
620 rxrpc_extract_ackinfo(call, skb, latest, ack.nAcks);
621
622 if (ack.reason == RXRPC_ACK_PING) {
623 _proto("Rx ACK %%%u PING Request", latest);
624 rxrpc_propose_ACK(call, RXRPC_ACK_PING_RESPONSE,
625 sp->hdr.serial, true);
626 }
627
628 /* discard any out-of-order or duplicate ACKs */
629 if (latest - call->acks_latest <= 0) {
630 _debug("discard ACK %d <= %d",
631 latest, call->acks_latest);
632 goto discard;
633 }
634 call->acks_latest = latest;
635
636 if (call->state != RXRPC_CALL_CLIENT_SEND_REQUEST &&
637 call->state != RXRPC_CALL_CLIENT_AWAIT_REPLY &&
638 call->state != RXRPC_CALL_SERVER_SEND_REPLY &&
639 call->state != RXRPC_CALL_SERVER_AWAIT_ACK)
640 goto discard;
641
642 _debug("Tx=%d H=%u S=%d", tx, call->acks_hard, call->state);
643
644 if (hard > 0) {
645 if (hard - 1 > tx) {
646 _debug("hard-ACK'd packet %d not transmitted"
647 " (%d top)",
648 hard - 1, tx);
649 goto protocol_error;
650 }
651
652 if ((call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY ||
653 call->state == RXRPC_CALL_SERVER_AWAIT_ACK) &&
654 hard > tx) {
655 call->acks_hard = tx;
656 goto all_acked;
657 }
658
659 smp_rmb();
660 rxrpc_rotate_tx_window(call, hard - 1);
661 }
662
663 if (ack.nAcks > 0) {
664 if (hard - 1 + ack.nAcks > tx) {
665 _debug("soft-ACK'd packet %d+%d not"
666 " transmitted (%d top)",
667 hard - 1, ack.nAcks, tx);
668 goto protocol_error;
669 }
670
671 if (rxrpc_process_soft_ACKs(call, &ack, skb) < 0)
672 goto protocol_error;
673 }
674 goto discard;
675
676 /* complete ACK to process */
677 case RXRPC_PACKET_TYPE_ACKALL:
678 goto all_acked;
679
680 /* abort and busy are handled elsewhere */
681 case RXRPC_PACKET_TYPE_BUSY:
682 case RXRPC_PACKET_TYPE_ABORT:
683 BUG();
684
685 /* connection level events - also handled elsewhere */
686 case RXRPC_PACKET_TYPE_CHALLENGE:
687 case RXRPC_PACKET_TYPE_RESPONSE:
688 case RXRPC_PACKET_TYPE_DEBUG:
689 BUG();
690 }
691
692 /* if we've had a hard ACK that covers all the packets we've sent, then
693 * that ends that phase of the operation */
694all_acked:
695 write_lock_bh(&call->state_lock);
696 _debug("ack all %d", call->state);
697
698 switch (call->state) {
699 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
700 call->state = RXRPC_CALL_CLIENT_RECV_REPLY;
701 break;
702 case RXRPC_CALL_SERVER_AWAIT_ACK:
703 _debug("srv complete");
704 call->state = RXRPC_CALL_COMPLETE;
705 post_ACK = true;
706 break;
707 case RXRPC_CALL_CLIENT_SEND_REQUEST:
708 case RXRPC_CALL_SERVER_RECV_REQUEST:
709 goto protocol_error_unlock; /* can't occur yet */
710 default:
711 write_unlock_bh(&call->state_lock);
712 goto discard; /* assume packet left over from earlier phase */
713 }
714
715 write_unlock_bh(&call->state_lock);
716
717 /* if all the packets we sent are hard-ACK'd, then we can discard
718 * whatever we've got left */
719 _debug("clear Tx %d",
720 CIRC_CNT(call->acks_head, call->acks_tail, call->acks_winsz));
721
722 del_timer_sync(&call->resend_timer);
723 clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags);
724 clear_bit(RXRPC_CALL_EV_RESEND_TIMER, &call->events);
725
726 if (call->acks_window)
727 rxrpc_zap_tx_window(call);
728
729 if (post_ACK) {
730 /* post the final ACK message for userspace to pick up */
731 _debug("post ACK");
732 skb->mark = RXRPC_SKB_MARK_FINAL_ACK;
733 sp->call = call;
734 rxrpc_get_call(call);
735 spin_lock_bh(&call->lock);
736 if (rxrpc_queue_rcv_skb(call, skb, true, true) < 0)
737 BUG();
738 spin_unlock_bh(&call->lock);
739 goto process_further;
740 }
741
742discard:
743 rxrpc_free_skb(skb);
744 goto process_further;
745
746protocol_error_unlock:
747 write_unlock_bh(&call->state_lock);
748protocol_error:
749 rxrpc_free_skb(skb);
750 _leave(" = -EPROTO");
751 return -EPROTO;
752}
753
754/*
755 * post a message to the socket Rx queue for recvmsg() to pick up
756 */
757static int rxrpc_post_message(struct rxrpc_call *call, u32 mark, u32 error,
758 bool fatal)
759{
760 struct rxrpc_skb_priv *sp;
761 struct sk_buff *skb;
762 int ret;
763
764 _enter("{%d,%lx},%u,%u,%d",
765 call->debug_id, call->flags, mark, error, fatal);
766
767 /* remove timers and things for fatal messages */
768 if (fatal) {
769 del_timer_sync(&call->resend_timer);
770 del_timer_sync(&call->ack_timer);
771 clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags);
772 }
773
774 if (mark != RXRPC_SKB_MARK_NEW_CALL &&
775 !test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
776 _leave("[no userid]");
777 return 0;
778 }
779
780 if (!test_bit(RXRPC_CALL_TERMINAL_MSG, &call->flags)) {
781 skb = alloc_skb(0, GFP_NOFS);
782 if (!skb)
783 return -ENOMEM;
784
785 rxrpc_new_skb(skb);
786
787 skb->mark = mark;
788
789 sp = rxrpc_skb(skb);
790 memset(sp, 0, sizeof(*sp));
791 sp->error = error;
792 sp->call = call;
793 rxrpc_get_call(call);
794
795 spin_lock_bh(&call->lock);
796 ret = rxrpc_queue_rcv_skb(call, skb, true, fatal);
797 spin_unlock_bh(&call->lock);
798 BUG_ON(ret < 0);
799 }
800
801 return 0;
802}
803
804/*
805 * handle background processing of incoming call packets and ACK / abort
806 * generation
807 */
808void rxrpc_process_call(struct work_struct *work)
809{
810 struct rxrpc_call *call =
811 container_of(work, struct rxrpc_call, processor);
812 struct rxrpc_wire_header whdr;
813 struct rxrpc_ackpacket ack;
814 struct rxrpc_ackinfo ackinfo;
815 struct msghdr msg;
816 struct kvec iov[5];
817 enum rxrpc_call_event genbit;
818 unsigned long bits;
819 __be32 data, pad;
820 size_t len;
821 int loop, nbit, ioc, ret, mtu;
822 u32 serial, abort_code = RX_PROTOCOL_ERROR;
823 u8 *acks = NULL;
824
825 //printk("\n--------------------\n");
826 _enter("{%d,%s,%lx} [%lu]",
827 call->debug_id, rxrpc_call_states[call->state], call->events,
828 (jiffies - call->creation_jif) / (HZ / 10));
829
830 if (test_and_set_bit(RXRPC_CALL_PROC_BUSY, &call->flags)) {
831 _debug("XXXXXXXXXXXXX RUNNING ON MULTIPLE CPUS XXXXXXXXXXXXX");
832 return;
833 }
834
835 /* there's a good chance we're going to have to send a message, so set
836 * one up in advance */
837 msg.msg_name = &call->conn->trans->peer->srx.transport;
838 msg.msg_namelen = call->conn->trans->peer->srx.transport_len;
839 msg.msg_control = NULL;
840 msg.msg_controllen = 0;
841 msg.msg_flags = 0;
842
843 whdr.epoch = htonl(call->conn->epoch);
844 whdr.cid = htonl(call->cid);
845 whdr.callNumber = htonl(call->call_id);
846 whdr.seq = 0;
847 whdr.type = RXRPC_PACKET_TYPE_ACK;
848 whdr.flags = call->conn->out_clientflag;
849 whdr.userStatus = 0;
850 whdr.securityIndex = call->conn->security_ix;
851 whdr._rsvd = 0;
852 whdr.serviceId = htons(call->service_id);
853
854 memset(iov, 0, sizeof(iov));
855 iov[0].iov_base = &whdr;
856 iov[0].iov_len = sizeof(whdr);
857
858 /* deal with events of a final nature */
859 if (test_bit(RXRPC_CALL_EV_RELEASE, &call->events)) {
860 rxrpc_release_call(call);
861 clear_bit(RXRPC_CALL_EV_RELEASE, &call->events);
862 }
863
864 if (test_bit(RXRPC_CALL_EV_RCVD_ERROR, &call->events)) {
865 int error;
866
867 clear_bit(RXRPC_CALL_EV_CONN_ABORT, &call->events);
868 clear_bit(RXRPC_CALL_EV_REJECT_BUSY, &call->events);
869 clear_bit(RXRPC_CALL_EV_ABORT, &call->events);
870
871 error = call->conn->trans->peer->net_error;
872 _debug("post net error %d", error);
873
874 if (rxrpc_post_message(call, RXRPC_SKB_MARK_NET_ERROR,
875 error, true) < 0)
876 goto no_mem;
877 clear_bit(RXRPC_CALL_EV_RCVD_ERROR, &call->events);
878 goto kill_ACKs;
879 }
880
881 if (test_bit(RXRPC_CALL_EV_CONN_ABORT, &call->events)) {
882 ASSERTCMP(call->state, >, RXRPC_CALL_COMPLETE);
883
884 clear_bit(RXRPC_CALL_EV_REJECT_BUSY, &call->events);
885 clear_bit(RXRPC_CALL_EV_ABORT, &call->events);
886
887 _debug("post conn abort");
888
889 if (rxrpc_post_message(call, RXRPC_SKB_MARK_LOCAL_ERROR,
890 call->conn->error, true) < 0)
891 goto no_mem;
892 clear_bit(RXRPC_CALL_EV_CONN_ABORT, &call->events);
893 goto kill_ACKs;
894 }
895
896 if (test_bit(RXRPC_CALL_EV_REJECT_BUSY, &call->events)) {
897 whdr.type = RXRPC_PACKET_TYPE_BUSY;
898 genbit = RXRPC_CALL_EV_REJECT_BUSY;
899 goto send_message;
900 }
901
902 if (test_bit(RXRPC_CALL_EV_ABORT, &call->events)) {
903 ASSERTCMP(call->state, >, RXRPC_CALL_COMPLETE);
904
905 if (rxrpc_post_message(call, RXRPC_SKB_MARK_LOCAL_ERROR,
906 ECONNABORTED, true) < 0)
907 goto no_mem;
908 whdr.type = RXRPC_PACKET_TYPE_ABORT;
909 data = htonl(call->local_abort);
910 iov[1].iov_base = &data;
911 iov[1].iov_len = sizeof(data);
912 genbit = RXRPC_CALL_EV_ABORT;
913 goto send_message;
914 }
915
916 if (test_bit(RXRPC_CALL_EV_ACK_FINAL, &call->events)) {
917 genbit = RXRPC_CALL_EV_ACK_FINAL;
918
919 ack.bufferSpace = htons(8);
920 ack.maxSkew = 0;
921 ack.serial = 0;
922 ack.reason = RXRPC_ACK_IDLE;
923 ack.nAcks = 0;
924 call->ackr_reason = 0;
925
926 spin_lock_bh(&call->lock);
927 ack.serial = htonl(call->ackr_serial);
928 ack.previousPacket = htonl(call->ackr_prev_seq);
929 ack.firstPacket = htonl(call->rx_data_eaten + 1);
930 spin_unlock_bh(&call->lock);
931
932 pad = 0;
933
934 iov[1].iov_base = &ack;
935 iov[1].iov_len = sizeof(ack);
936 iov[2].iov_base = &pad;
937 iov[2].iov_len = 3;
938 iov[3].iov_base = &ackinfo;
939 iov[3].iov_len = sizeof(ackinfo);
940 goto send_ACK;
941 }
942
943 if (call->events & ((1 << RXRPC_CALL_EV_RCVD_BUSY) |
944 (1 << RXRPC_CALL_EV_RCVD_ABORT))
945 ) {
946 u32 mark;
947
948 if (test_bit(RXRPC_CALL_EV_RCVD_ABORT, &call->events))
949 mark = RXRPC_SKB_MARK_REMOTE_ABORT;
950 else
951 mark = RXRPC_SKB_MARK_BUSY;
952
953 _debug("post abort/busy");
954 rxrpc_clear_tx_window(call);
955 if (rxrpc_post_message(call, mark, ECONNABORTED, true) < 0)
956 goto no_mem;
957
958 clear_bit(RXRPC_CALL_EV_RCVD_BUSY, &call->events);
959 clear_bit(RXRPC_CALL_EV_RCVD_ABORT, &call->events);
960 goto kill_ACKs;
961 }
962
963 if (test_and_clear_bit(RXRPC_CALL_EV_RCVD_ACKALL, &call->events)) {
964 _debug("do implicit ackall");
965 rxrpc_clear_tx_window(call);
966 }
967
968 if (test_bit(RXRPC_CALL_EV_LIFE_TIMER, &call->events)) {
969 write_lock_bh(&call->state_lock);
970 if (call->state <= RXRPC_CALL_COMPLETE) {
971 call->state = RXRPC_CALL_LOCALLY_ABORTED;
972 call->local_abort = RX_CALL_TIMEOUT;
973 set_bit(RXRPC_CALL_EV_ABORT, &call->events);
974 }
975 write_unlock_bh(&call->state_lock);
976
977 _debug("post timeout");
978 if (rxrpc_post_message(call, RXRPC_SKB_MARK_LOCAL_ERROR,
979 ETIME, true) < 0)
980 goto no_mem;
981
982 clear_bit(RXRPC_CALL_EV_LIFE_TIMER, &call->events);
983 goto kill_ACKs;
984 }
985
986 /* deal with assorted inbound messages */
987 if (!skb_queue_empty(&call->rx_queue)) {
988 switch (rxrpc_process_rx_queue(call, &abort_code)) {
989 case 0:
990 case -EAGAIN:
991 break;
992 case -ENOMEM:
993 goto no_mem;
994 case -EKEYEXPIRED:
995 case -EKEYREJECTED:
996 case -EPROTO:
997 rxrpc_abort_call(call, abort_code);
998 goto kill_ACKs;
999 }
1000 }
1001
1002 /* handle resending */
1003 if (test_and_clear_bit(RXRPC_CALL_EV_RESEND_TIMER, &call->events))
1004 rxrpc_resend_timer(call);
1005 if (test_and_clear_bit(RXRPC_CALL_EV_RESEND, &call->events))
1006 rxrpc_resend(call);
1007
1008 /* consider sending an ordinary ACK */
1009 if (test_bit(RXRPC_CALL_EV_ACK, &call->events)) {
1010 _debug("send ACK: window: %d - %d { %lx }",
1011 call->rx_data_eaten, call->ackr_win_top,
1012 call->ackr_window[0]);
1013
1014 if (call->state > RXRPC_CALL_SERVER_ACK_REQUEST &&
1015 call->ackr_reason != RXRPC_ACK_PING_RESPONSE) {
1016 /* ACK by sending reply DATA packet in this state */
1017 clear_bit(RXRPC_CALL_EV_ACK, &call->events);
1018 goto maybe_reschedule;
1019 }
1020
1021 genbit = RXRPC_CALL_EV_ACK;
1022
1023 acks = kzalloc(call->ackr_win_top - call->rx_data_eaten,
1024 GFP_NOFS);
1025 if (!acks)
1026 goto no_mem;
1027
1028 //hdr.flags = RXRPC_SLOW_START_OK;
1029 ack.bufferSpace = htons(8);
1030 ack.maxSkew = 0;
1031
1032 spin_lock_bh(&call->lock);
1033 ack.reason = call->ackr_reason;
1034 ack.serial = htonl(call->ackr_serial);
1035 ack.previousPacket = htonl(call->ackr_prev_seq);
1036 ack.firstPacket = htonl(call->rx_data_eaten + 1);
1037
1038 ack.nAcks = 0;
1039 for (loop = 0; loop < RXRPC_ACKR_WINDOW_ASZ; loop++) {
1040 nbit = loop * BITS_PER_LONG;
1041 for (bits = call->ackr_window[loop]; bits; bits >>= 1
1042 ) {
1043 _debug("- l=%d n=%d b=%lx", loop, nbit, bits);
1044 if (bits & 1) {
1045 acks[nbit] = RXRPC_ACK_TYPE_ACK;
1046 ack.nAcks = nbit + 1;
1047 }
1048 nbit++;
1049 }
1050 }
1051 call->ackr_reason = 0;
1052 spin_unlock_bh(&call->lock);
1053
1054 pad = 0;
1055
1056 iov[1].iov_base = &ack;
1057 iov[1].iov_len = sizeof(ack);
1058 iov[2].iov_base = acks;
1059 iov[2].iov_len = ack.nAcks;
1060 iov[3].iov_base = &pad;
1061 iov[3].iov_len = 3;
1062 iov[4].iov_base = &ackinfo;
1063 iov[4].iov_len = sizeof(ackinfo);
1064
1065 switch (ack.reason) {
1066 case RXRPC_ACK_REQUESTED:
1067 case RXRPC_ACK_DUPLICATE:
1068 case RXRPC_ACK_OUT_OF_SEQUENCE:
1069 case RXRPC_ACK_EXCEEDS_WINDOW:
1070 case RXRPC_ACK_NOSPACE:
1071 case RXRPC_ACK_PING:
1072 case RXRPC_ACK_PING_RESPONSE:
1073 goto send_ACK_with_skew;
1074 case RXRPC_ACK_DELAY:
1075 case RXRPC_ACK_IDLE:
1076 goto send_ACK;
1077 }
1078 }
1079
1080 /* handle completion of security negotiations on an incoming
1081 * connection */
1082 if (test_and_clear_bit(RXRPC_CALL_EV_SECURED, &call->events)) {
1083 _debug("secured");
1084 spin_lock_bh(&call->lock);
1085
1086 if (call->state == RXRPC_CALL_SERVER_SECURING) {
1087 _debug("securing");
1088 write_lock(&call->conn->lock);
1089 if (!test_bit(RXRPC_CALL_RELEASED, &call->flags) &&
1090 !test_bit(RXRPC_CALL_EV_RELEASE, &call->events)) {
1091 _debug("not released");
1092 call->state = RXRPC_CALL_SERVER_ACCEPTING;
1093 list_move_tail(&call->accept_link,
1094 &call->socket->acceptq);
1095 }
1096 write_unlock(&call->conn->lock);
1097 read_lock(&call->state_lock);
1098 if (call->state < RXRPC_CALL_COMPLETE)
1099 set_bit(RXRPC_CALL_EV_POST_ACCEPT, &call->events);
1100 read_unlock(&call->state_lock);
1101 }
1102
1103 spin_unlock_bh(&call->lock);
1104 if (!test_bit(RXRPC_CALL_EV_POST_ACCEPT, &call->events))
1105 goto maybe_reschedule;
1106 }
1107
1108 /* post a notification of an acceptable connection to the app */
1109 if (test_bit(RXRPC_CALL_EV_POST_ACCEPT, &call->events)) {
1110 _debug("post accept");
1111 if (rxrpc_post_message(call, RXRPC_SKB_MARK_NEW_CALL,
1112 0, false) < 0)
1113 goto no_mem;
1114 clear_bit(RXRPC_CALL_EV_POST_ACCEPT, &call->events);
1115 goto maybe_reschedule;
1116 }
1117
1118 /* handle incoming call acceptance */
1119 if (test_and_clear_bit(RXRPC_CALL_EV_ACCEPTED, &call->events)) {
1120 _debug("accepted");
1121 ASSERTCMP(call->rx_data_post, ==, 0);
1122 call->rx_data_post = 1;
1123 read_lock_bh(&call->state_lock);
1124 if (call->state < RXRPC_CALL_COMPLETE)
1125 set_bit(RXRPC_CALL_EV_DRAIN_RX_OOS, &call->events);
1126 read_unlock_bh(&call->state_lock);
1127 }
1128
1129 /* drain the out of sequence received packet queue into the packet Rx
1130 * queue */
1131 if (test_and_clear_bit(RXRPC_CALL_EV_DRAIN_RX_OOS, &call->events)) {
1132 while (call->rx_data_post == call->rx_first_oos)
1133 if (rxrpc_drain_rx_oos_queue(call) < 0)
1134 break;
1135 goto maybe_reschedule;
1136 }
1137
1138 /* other events may have been raised since we started checking */
1139 goto maybe_reschedule;
1140
1141send_ACK_with_skew:
1142 ack.maxSkew = htons(atomic_read(&call->conn->hi_serial) -
1143 ntohl(ack.serial));
1144send_ACK:
1145 mtu = call->conn->trans->peer->if_mtu;
1146 mtu -= call->conn->trans->peer->hdrsize;
1147 ackinfo.maxMTU = htonl(mtu);
1148 ackinfo.rwind = htonl(rxrpc_rx_window_size);
1149
1150 /* permit the peer to send us jumbo packets if it wants to */
1151 ackinfo.rxMTU = htonl(rxrpc_rx_mtu);
1152 ackinfo.jumbo_max = htonl(rxrpc_rx_jumbo_max);
1153
1154 serial = atomic_inc_return(&call->conn->serial);
1155 whdr.serial = htonl(serial);
1156 _proto("Tx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
1157 serial,
1158 ntohs(ack.maxSkew),
1159 ntohl(ack.firstPacket),
1160 ntohl(ack.previousPacket),
1161 ntohl(ack.serial),
1162 rxrpc_acks(ack.reason),
1163 ack.nAcks);
1164
1165 del_timer_sync(&call->ack_timer);
1166 if (ack.nAcks > 0)
1167 set_bit(RXRPC_CALL_TX_SOFT_ACK, &call->flags);
1168 goto send_message_2;
1169
1170send_message:
1171 _debug("send message");
1172
1173 serial = atomic_inc_return(&call->conn->serial);
1174 whdr.serial = htonl(serial);
1175 _proto("Tx %s %%%u", rxrpc_pkts[whdr.type], serial);
1176send_message_2:
1177
1178 len = iov[0].iov_len;
1179 ioc = 1;
1180 if (iov[4].iov_len) {
1181 ioc = 5;
1182 len += iov[4].iov_len;
1183 len += iov[3].iov_len;
1184 len += iov[2].iov_len;
1185 len += iov[1].iov_len;
1186 } else if (iov[3].iov_len) {
1187 ioc = 4;
1188 len += iov[3].iov_len;
1189 len += iov[2].iov_len;
1190 len += iov[1].iov_len;
1191 } else if (iov[2].iov_len) {
1192 ioc = 3;
1193 len += iov[2].iov_len;
1194 len += iov[1].iov_len;
1195 } else if (iov[1].iov_len) {
1196 ioc = 2;
1197 len += iov[1].iov_len;
1198 }
1199
1200 ret = kernel_sendmsg(call->conn->trans->local->socket,
1201 &msg, iov, ioc, len);
1202 if (ret < 0) {
1203 _debug("sendmsg failed: %d", ret);
1204 read_lock_bh(&call->state_lock);
1205 if (call->state < RXRPC_CALL_DEAD)
1206 rxrpc_queue_call(call);
1207 read_unlock_bh(&call->state_lock);
1208 goto error;
1209 }
1210
1211 switch (genbit) {
1212 case RXRPC_CALL_EV_ABORT:
1213 clear_bit(genbit, &call->events);
1214 clear_bit(RXRPC_CALL_EV_RCVD_ABORT, &call->events);
1215 goto kill_ACKs;
1216
1217 case RXRPC_CALL_EV_ACK_FINAL:
1218 write_lock_bh(&call->state_lock);
1219 if (call->state == RXRPC_CALL_CLIENT_FINAL_ACK)
1220 call->state = RXRPC_CALL_COMPLETE;
1221 write_unlock_bh(&call->state_lock);
1222 goto kill_ACKs;
1223
1224 default:
1225 clear_bit(genbit, &call->events);
1226 switch (call->state) {
1227 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
1228 case RXRPC_CALL_CLIENT_RECV_REPLY:
1229 case RXRPC_CALL_SERVER_RECV_REQUEST:
1230 case RXRPC_CALL_SERVER_ACK_REQUEST:
1231 _debug("start ACK timer");
1232 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY,
1233 call->ackr_serial, false);
1234 default:
1235 break;
1236 }
1237 goto maybe_reschedule;
1238 }
1239
1240kill_ACKs:
1241 del_timer_sync(&call->ack_timer);
1242 if (test_and_clear_bit(RXRPC_CALL_EV_ACK_FINAL, &call->events))
1243 rxrpc_put_call(call);
1244 clear_bit(RXRPC_CALL_EV_ACK, &call->events);
1245
1246maybe_reschedule:
1247 if (call->events || !skb_queue_empty(&call->rx_queue)) {
1248 read_lock_bh(&call->state_lock);
1249 if (call->state < RXRPC_CALL_DEAD)
1250 rxrpc_queue_call(call);
1251 read_unlock_bh(&call->state_lock);
1252 }
1253
1254 /* don't leave aborted connections on the accept queue */
1255 if (call->state >= RXRPC_CALL_COMPLETE &&
1256 !list_empty(&call->accept_link)) {
1257 _debug("X unlinking once-pending call %p { e=%lx f=%lx c=%x }",
1258 call, call->events, call->flags, call->conn->cid);
1259
1260 read_lock_bh(&call->state_lock);
1261 if (!test_bit(RXRPC_CALL_RELEASED, &call->flags) &&
1262 !test_and_set_bit(RXRPC_CALL_EV_RELEASE, &call->events))
1263 rxrpc_queue_call(call);
1264 read_unlock_bh(&call->state_lock);
1265 }
1266
1267error:
1268 clear_bit(RXRPC_CALL_PROC_BUSY, &call->flags);
1269 kfree(acks);
1270
1271 /* because we don't want two CPUs both processing the work item for one
1272 * call at the same time, we use a flag to note when it's busy; however
1273 * this means there's a race between clearing the flag and setting the
1274 * work pending bit and the work item being processed again */
1275 if (call->events && !work_pending(&call->processor)) {
1276 _debug("jumpstart %x", call->conn->cid);
1277 rxrpc_queue_call(call);
1278 }
1279
1280 _leave("");
1281 return;
1282
1283no_mem:
1284 _debug("out of memory");
1285 goto maybe_reschedule;
1286}