?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6.git] / net / rxrpc / ar-ack.c
blobe4d9cbcff402ca09c7b0f09c536abcdc1857a4cb
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 static unsigned int rxrpc_ack_defer = 1;
23
24 static const char *const rxrpc_acks[] = {
25 "---", "REQ", "DUP", "OOS", "WIN", "MEM", "PNG", "PNR", "DLY", "IDL",
26 "-?-"
27 };
28
29 static const s8 rxrpc_ack_priority[] = {
30 [0] = 0,
31 [RXRPC_ACK_DELAY] = 1,
32 [RXRPC_ACK_REQUESTED] = 2,
33 [RXRPC_ACK_IDLE] = 3,
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,
39 };
40
41 /*
42 * propose an ACK be sent
43 */
44 void __rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason,
45 __be32 serial, bool immediate)
46 {
47 unsigned long expiry;
48 s8 prior = rxrpc_ack_priority[ack_reason];
49
50 ASSERTCMP(prior, >, 0);
51
52 _enter("{%d},%s,%%%x,%u",
53 call->debug_id, rxrpc_acks[ack_reason], ntohl(serial),
54 immediate);
55
56 if (prior < rxrpc_ack_priority[call->ackr_reason]) {
57 if (immediate)
58 goto cancel_timer;
59 return;
60 }
61
62 /* update DELAY, IDLE, REQUESTED and PING_RESPONSE ACK serial
63 * numbers */
64 if (prior == rxrpc_ack_priority[call->ackr_reason]) {
65 if (prior <= 4)
66 call->ackr_serial = serial;
67 if (immediate)
68 goto cancel_timer;
69 return;
70 }
71
72 call->ackr_reason = ack_reason;
73 call->ackr_serial = serial;
74
75 switch (ack_reason) {
76 case RXRPC_ACK_DELAY:
77 _debug("run delay timer");
78 call->ack_timer.expires = jiffies + rxrpc_ack_timeout * HZ;
79 add_timer(&call->ack_timer);
80 return;
81
82 case RXRPC_ACK_IDLE:
83 if (!immediate) {
84 _debug("run defer timer");
85 expiry = 1;
86 goto run_timer;
87 }
88 goto cancel_timer;
89
90 case RXRPC_ACK_REQUESTED:
91 if (!rxrpc_ack_defer)
92 goto cancel_timer;
93 if (!immediate || serial == cpu_to_be32(1)) {
94 _debug("run defer timer");
95 expiry = rxrpc_ack_defer;
96 goto run_timer;
97 }
98
99 default:
100 _debug("immediate ACK");
101 goto cancel_timer;
102 }
103
104 run_timer:
105 expiry += jiffies;
106 if (!timer_pending(&call->ack_timer) ||
107 time_after(call->ack_timer.expires, expiry))
108 mod_timer(&call->ack_timer, expiry);
109 return;
110
111 cancel_timer:
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);
119 }
120
121 /*
122 * propose an ACK be sent, locking the call structure
123 */
124 void rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason,
125 __be32 serial, bool immediate)
126 {
127 s8 prior = rxrpc_ack_priority[ack_reason];
128
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);
133 }
134 }
135
136 /*
137 * set the resend timer
138 */
139 static void rxrpc_set_resend(struct rxrpc_call *call, u8 resend,
140 unsigned long resend_at)
141 {
142 read_lock_bh(&call->state_lock);
143 if (call->state >= RXRPC_CALL_COMPLETE)
144 resend = 0;
145
146 if (resend & 1) {
147 _debug("SET RESEND");
148 set_bit(RXRPC_CALL_RESEND, &call->events);
149 }
150
151 if (resend & 2) {
152 _debug("MODIFY RESEND TIMER");
153 set_bit(RXRPC_CALL_RUN_RTIMER, &call->flags);
154 mod_timer(&call->resend_timer, resend_at);
155 } else {
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);
160 }
161 read_unlock_bh(&call->state_lock);
162 }
163
164 /*
165 * resend packets
166 */
167 static void rxrpc_resend(struct rxrpc_call *call)
168 {
169 struct rxrpc_skb_priv *sp;
170 struct rxrpc_header *hdr;
171 struct sk_buff *txb;
172 unsigned long *p_txb, resend_at;
173 int loop, stop;
174 u8 resend;
175
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));
180
181 stop = 0;
182 resend = 0;
183 resend_at = 0;
184
185 for (loop = call->acks_tail;
186 loop != call->acks_head || stop;
187 loop = (loop + 1) & (call->acks_winsz - 1)
188 ) {
189 p_txb = call->acks_window + loop;
190 smp_read_barrier_depends();
191 if (*p_txb & 1)
192 continue;
193
194 txb = (struct sk_buff *) *p_txb;
195 sp = rxrpc_skb(txb);
196
197 if (sp->need_resend) {
198 sp->need_resend = false;
199
200 /* each Tx packet has a new serial number */
201 sp->hdr.serial =
202 htonl(atomic_inc_return(&call->conn->serial));
203
204 hdr = (struct rxrpc_header *) txb->head;
205 hdr->serial = sp->hdr.serial;
206
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) {
210 stop = 0;
211 sp->resend_at = jiffies + 3;
212 } else {
213 sp->resend_at =
214 jiffies + rxrpc_resend_timeout * HZ;
215 }
216 }
217
218 if (time_after_eq(jiffies + 1, sp->resend_at)) {
219 sp->need_resend = true;
220 resend |= 1;
221 } else if (resend & 2) {
222 if (time_before(sp->resend_at, resend_at))
223 resend_at = sp->resend_at;
224 } else {
225 resend_at = sp->resend_at;
226 resend |= 2;
227 }
228 }
229
230 rxrpc_set_resend(call, resend, resend_at);
231 _leave("");
232 }
233
234 /*
235 * handle resend timer expiry
236 */
237 static void rxrpc_resend_timer(struct rxrpc_call *call)
238 {
239 struct rxrpc_skb_priv *sp;
240 struct sk_buff *txb;
241 unsigned long *p_txb, resend_at;
242 int loop;
243 u8 resend;
244
245 _enter("%d,%d,%d",
246 call->acks_tail, call->acks_unacked, call->acks_head);
247
248 if (call->state >= RXRPC_CALL_COMPLETE)
249 return;
250
251 resend = 0;
252 resend_at = 0;
253
254 for (loop = call->acks_unacked;
255 loop != call->acks_head;
256 loop = (loop + 1) & (call->acks_winsz - 1)
257 ) {
258 p_txb = call->acks_window + loop;
259 smp_read_barrier_depends();
260 txb = (struct sk_buff *) (*p_txb & ~1);
261 sp = rxrpc_skb(txb);
262
263 ASSERT(!(*p_txb & 1));
264
265 if (sp->need_resend) {
266 ;
267 } else if (time_after_eq(jiffies + 1, sp->resend_at)) {
268 sp->need_resend = true;
269 resend |= 1;
270 } else if (resend & 2) {
271 if (time_before(sp->resend_at, resend_at))
272 resend_at = sp->resend_at;
273 } else {
274 resend_at = sp->resend_at;
275 resend |= 2;
276 }
277 }
278
279 rxrpc_set_resend(call, resend, resend_at);
280 _leave("");
281 }
282
283 /*
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
287 */
288 static int rxrpc_process_soft_ACKs(struct rxrpc_call *call,
289 struct rxrpc_ackpacket *ack,
290 struct sk_buff *skb)
291 {
292 struct rxrpc_skb_priv *sp;
293 struct sk_buff *txb;
294 unsigned long *p_txb, resend_at;
295 int loop;
296 u8 sacks[RXRPC_MAXACKS], resend;
297
298 _enter("{%d,%d},{%d},",
299 call->acks_hard,
300 CIRC_CNT(call->acks_head, call->acks_tail, call->acks_winsz),
301 ack->nAcks);
302
303 if (skb_copy_bits(skb, 0, sacks, ack->nAcks) < 0)
304 goto protocol_error;
305
306 resend = 0;
307 resend_at = 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);
313 sp = rxrpc_skb(txb);
314
315 switch (sacks[loop]) {
316 case RXRPC_ACK_TYPE_ACK:
317 sp->need_resend = false;
318 *p_txb |= 1;
319 break;
320 case RXRPC_ACK_TYPE_NACK:
321 sp->need_resend = true;
322 *p_txb &= ~1;
323 resend = 1;
324 break;
325 default:
326 _debug("Unsupported ACK type %d", sacks[loop]);
327 goto protocol_error;
328 }
329 }
330
331 smp_mb();
332 call->acks_unacked = (call->acks_tail + loop) & (call->acks_winsz - 1);
333
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)
339 ) {
340 p_txb = call->acks_window + loop;
341 smp_read_barrier_depends();
342 txb = (struct sk_buff *) (*p_txb & ~1);
343 sp = rxrpc_skb(txb);
344
345 if (*p_txb & 1) {
346 /* packet must have been discarded */
347 sp->need_resend = true;
348 *p_txb &= ~1;
349 resend |= 1;
350 } else if (sp->need_resend) {
351 ;
352 } else if (time_after_eq(jiffies + 1, sp->resend_at)) {
353 sp->need_resend = true;
354 resend |= 1;
355 } else if (resend & 2) {
356 if (time_before(sp->resend_at, resend_at))
357 resend_at = sp->resend_at;
358 } else {
359 resend_at = sp->resend_at;
360 resend |= 2;
361 }
362 }
363
364 rxrpc_set_resend(call, resend, resend_at);
365 _leave(" = 0");
366 return 0;
367
368 protocol_error:
369 _leave(" = -EPROTO");
370 return -EPROTO;
371 }
372
373 /*
374 * discard hard-ACK'd packets from the Tx window
375 */
376 static void rxrpc_rotate_tx_window(struct rxrpc_call *call, u32 hard)
377 {
378 unsigned long _skb;
379 int tail = call->acks_tail, old_tail;
380 int win = CIRC_CNT(call->acks_head, tail, call->acks_winsz);
381
382 _enter("{%u,%u},%u", call->acks_hard, win, hard);
383
384 ASSERTCMP(hard - call->acks_hard, <=, win);
385
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);
390 old_tail = tail;
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;
395 call->acks_hard++;
396 }
397
398 wake_up(&call->tx_waitq);
399 }
400
401 /*
402 * clear the Tx window in the event of a failure
403 */
404 static void rxrpc_clear_tx_window(struct rxrpc_call *call)
405 {
406 rxrpc_rotate_tx_window(call, atomic_read(&call->sequence));
407 }
408
409 /*
410 * drain the out of sequence received packet queue into the packet Rx queue
411 */
412 static int rxrpc_drain_rx_oos_queue(struct rxrpc_call *call)
413 {
414 struct rxrpc_skb_priv *sp;
415 struct sk_buff *skb;
416 bool terminal;
417 int ret;
418
419 _enter("{%d,%d}", call->rx_data_post, call->rx_first_oos);
420
421 spin_lock_bh(&call->lock);
422
423 ret = -ECONNRESET;
424 if (test_bit(RXRPC_CALL_RELEASED, &call->flags))
425 goto socket_unavailable;
426
427 skb = skb_dequeue(&call->rx_oos_queue);
428 if (skb) {
429 sp = rxrpc_skb(skb);
430
431 _debug("drain OOS packet %d [%d]",
432 ntohl(sp->hdr.seq), call->rx_first_oos);
433
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);
438 } else {
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);
443 BUG_ON(ret < 0);
444 _debug("drain #%u", call->rx_data_post);
445 call->rx_data_post++;
446
447 /* find out what the next packet is */
448 skb = skb_peek(&call->rx_oos_queue);
449 if (skb)
450 call->rx_first_oos =
451 ntohl(rxrpc_skb(skb)->hdr.seq);
452 else
453 call->rx_first_oos = 0;
454 _debug("peek %p {%u}", skb, call->rx_first_oos);
455 }
456 }
457
458 ret = 0;
459 socket_unavailable:
460 spin_unlock_bh(&call->lock);
461 _leave(" = %d", ret);
462 return ret;
463 }
464
465 /*
466 * insert an out of sequence packet into the buffer
467 */
468 static void rxrpc_insert_oos_packet(struct rxrpc_call *call,
469 struct sk_buff *skb)
470 {
471 struct rxrpc_skb_priv *sp, *psp;
472 struct sk_buff *p;
473 u32 seq;
474
475 sp = rxrpc_skb(skb);
476 seq = ntohl(sp->hdr.seq);
477 _enter(",,{%u}", seq);
478
479 skb->destructor = rxrpc_packet_destructor;
480 ASSERTCMP(sp->call, ==, NULL);
481 sp->call = call;
482 rxrpc_get_call(call);
483
484 /* insert into the buffer in sequence order */
485 spin_lock_bh(&call->lock);
486
487 skb_queue_walk(&call->rx_oos_queue, p) {
488 psp = rxrpc_skb(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);
493 goto inserted;
494 }
495 }
496
497 _debug("append oos #%u", seq);
498 skb_queue_tail(&call->rx_oos_queue, skb);
499 inserted:
500
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;
504
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);
510 }
511 read_unlock(&call->state_lock);
512
513 spin_unlock_bh(&call->lock);
514 _leave(" [stored #%u]", call->rx_first_oos);
515 }
516
517 /*
518 * clear the Tx window on final ACK reception
519 */
520 static void rxrpc_zap_tx_window(struct rxrpc_call *call)
521 {
522 struct rxrpc_skb_priv *sp;
523 struct sk_buff *skb;
524 unsigned long _skb, *acks_window;
525 u8 winsz = call->acks_winsz;
526 int tail;
527
528 acks_window = call->acks_window;
529 call->acks_window = NULL;
530
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;
535 smp_mb();
536 call->acks_tail = (call->acks_tail + 1) & (winsz - 1);
537
538 skb = (struct sk_buff *) _skb;
539 sp = rxrpc_skb(skb);
540 _debug("+++ clear Tx %u", ntohl(sp->hdr.seq));
541 rxrpc_free_skb(skb);
542 }
543
544 kfree(acks_window);
545 }
546
547 /*
548 * process the extra information that may be appended to an ACK packet
549 */
550 static void rxrpc_extract_ackinfo(struct rxrpc_call *call, struct sk_buff *skb,
551 unsigned int latest, int nAcks)
552 {
553 struct rxrpc_ackinfo ackinfo;
554 struct rxrpc_peer *peer;
555 unsigned int mtu;
556
557 if (skb_copy_bits(skb, nAcks + 3, &ackinfo, sizeof(ackinfo)) < 0) {
558 _leave(" [no ackinfo]");
559 return;
560 }
561
562 _proto("Rx ACK %%%u Info { rx=%u max=%u rwin=%u jm=%u }",
563 latest,
564 ntohl(ackinfo.rxMTU), ntohl(ackinfo.maxMTU),
565 ntohl(ackinfo.rwind), ntohl(ackinfo.jumbo_max));
566
567 mtu = min(ntohl(ackinfo.rxMTU), ntohl(ackinfo.maxMTU));
568
569 peer = call->conn->trans->peer;
570 if (mtu < peer->maxdata) {
571 spin_lock_bh(&peer->lock);
572 peer->maxdata = mtu;
573 peer->mtu = mtu + peer->hdrsize;
574 spin_unlock_bh(&peer->lock);
575 _net("Net MTU %u (maxdata %u)", peer->mtu, peer->maxdata);
576 }
577 }
578
579 /*
580 * process packets in the reception queue
581 */
582 static int rxrpc_process_rx_queue(struct rxrpc_call *call,
583 u32 *_abort_code)
584 {
585 struct rxrpc_ackpacket ack;
586 struct rxrpc_skb_priv *sp;
587 struct sk_buff *skb;
588 bool post_ACK;
589 int latest;
590 u32 hard, tx;
591
592 _enter("");
593
594 process_further:
595 skb = skb_dequeue(&call->rx_queue);
596 if (!skb)
597 return -EAGAIN;
598
599 _net("deferred skb %p", skb);
600
601 sp = rxrpc_skb(skb);
602
603 _debug("process %s [st %d]", rxrpc_pkts[sp->hdr.type], call->state);
604
605 post_ACK = false;
606
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));
613
614 /* secured packets must be verified and possibly decrypted */
615 if (rxrpc_verify_packet(call, skb, _abort_code) < 0)
616 goto protocol_error;
617
618 rxrpc_insert_oos_packet(call, skb);
619 goto process_further;
620
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");
625 goto protocol_error;
626 }
627 if (!skb_pull(skb, sizeof(ack)))
628 BUG();
629
630 latest = ntohl(sp->hdr.serial);
631 hard = ntohl(ack.firstPacket);
632 tx = atomic_read(&call->sequence);
633
634 _proto("Rx ACK %%%u { m=%hu f=#%u p=#%u s=%%%u r=%s n=%u }",
635 latest,
636 ntohs(ack.maxSkew),
637 hard,
638 ntohl(ack.previousPacket),
639 ntohl(ack.serial),
640 rxrpc_acks[ack.reason],
641 ack.nAcks);
642
643 rxrpc_extract_ackinfo(call, skb, latest, ack.nAcks);
644
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);
649 }
650
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);
655 goto discard;
656 }
657 call->acks_latest = latest;
658
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)
663 goto discard;
664
665 _debug("Tx=%d H=%u S=%d", tx, call->acks_hard, call->state);
666
667 if (hard > 0) {
668 if (hard - 1 > tx) {
669 _debug("hard-ACK'd packet %d not transmitted"
670 " (%d top)",
671 hard - 1, tx);
672 goto protocol_error;
673 }
674
675 if ((call->state == RXRPC_CALL_CLIENT_AWAIT_REPLY ||
676 call->state == RXRPC_CALL_SERVER_AWAIT_ACK) &&
677 hard > tx)
678 goto all_acked;
679
680 smp_rmb();
681 rxrpc_rotate_tx_window(call, hard - 1);
682 }
683
684 if (ack.nAcks > 0) {
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);
689 goto protocol_error;
690 }
691
692 if (rxrpc_process_soft_ACKs(call, &ack, skb) < 0)
693 goto protocol_error;
694 }
695 goto discard;
696
697 /* complete ACK to process */
698 case RXRPC_PACKET_TYPE_ACKALL:
699 goto all_acked;
700
701 /* abort and busy are handled elsewhere */
702 case RXRPC_PACKET_TYPE_BUSY:
703 case RXRPC_PACKET_TYPE_ABORT:
704 BUG();
705
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:
710 BUG();
711 }
712
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 */
715 all_acked:
716 write_lock_bh(&call->state_lock);
717 _debug("ack all %d", call->state);
718
719 switch (call->state) {
720 case RXRPC_CALL_CLIENT_AWAIT_REPLY:
721 call->state = RXRPC_CALL_CLIENT_RECV_REPLY;
722 break;
723 case RXRPC_CALL_SERVER_AWAIT_ACK:
724 _debug("srv complete");
725 call->state = RXRPC_CALL_COMPLETE;
726 post_ACK = true;
727 break;
728 case RXRPC_CALL_CLIENT_SEND_REQUEST:
729 case RXRPC_CALL_SERVER_RECV_REQUEST:
730 goto protocol_error_unlock; /* can't occur yet */
731 default:
732 write_unlock_bh(&call->state_lock);
733 goto discard; /* assume packet left over from earlier phase */
734 }
735
736 write_unlock_bh(&call->state_lock);
737
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));
742
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);
746
747 if (call->acks_window)
748 rxrpc_zap_tx_window(call);
749
750 if (post_ACK) {
751 /* post the final ACK message for userspace to pick up */
752 _debug("post ACK");
753 skb->mark = RXRPC_SKB_MARK_FINAL_ACK;
754 sp->call = call;
755 rxrpc_get_call(call);
756 spin_lock_bh(&call->lock);
757 if (rxrpc_queue_rcv_skb(call, skb, true, true) < 0)
758 BUG();
759 spin_unlock_bh(&call->lock);
760 goto process_further;
761 }
762
763 discard:
764 rxrpc_free_skb(skb);
765 goto process_further;
766
767 protocol_error_unlock:
768 write_unlock_bh(&call->state_lock);
769 protocol_error:
770 rxrpc_free_skb(skb);
771 _leave(" = -EPROTO");
772 return -EPROTO;
773 }
774
775 /*
776 * post a message to the socket Rx queue for recvmsg() to pick up
777 */
778 static int rxrpc_post_message(struct rxrpc_call *call, u32 mark, u32 error,
779 bool fatal)
780 {
781 struct rxrpc_skb_priv *sp;
782 struct sk_buff *skb;
783 int ret;
784
785 _enter("{%d,%lx},%u,%u,%d",
786 call->debug_id, call->flags, mark, error, fatal);
787
788 /* remove timers and things for fatal messages */
789 if (fatal) {
790 del_timer_sync(&call->resend_timer);
791 del_timer_sync(&call->ack_timer);
792 clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags);
793 }
794
795 if (mark != RXRPC_SKB_MARK_NEW_CALL &&
796 !test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
797 _leave("[no userid]");
798 return 0;
799 }
800
801 if (!test_bit(RXRPC_CALL_TERMINAL_MSG, &call->flags)) {
802 skb = alloc_skb(0, GFP_NOFS);
803 if (!skb)
804 return -ENOMEM;
805
806 rxrpc_new_skb(skb);
807
808 skb->mark = mark;
809
810 sp = rxrpc_skb(skb);
811 memset(sp, 0, sizeof(*sp));
812 sp->error = error;
813 sp->call = call;
814 rxrpc_get_call(call);
815
816 spin_lock_bh(&call->lock);
817 ret = rxrpc_queue_rcv_skb(call, skb, true, fatal);
818 spin_unlock_bh(&call->lock);
819 BUG_ON(ret < 0);
820 }
821
822 return 0;
823 }
824
825 /*
826 * handle background processing of incoming call packets and ACK / abort
827 * generation
828 */
829 void rxrpc_process_call(struct work_struct *work)
830 {
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;
836 struct msghdr msg;
837 struct kvec iov[5];
838 unsigned long bits;
839 __be32 data, pad;
840 size_t len;
841 int genbit, loop, nbit, ioc, ret, mtu;
842 u32 abort_code = RX_PROTOCOL_ERROR;
843 u8 *acks = NULL;
844
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));
849
850 if (test_and_set_bit(RXRPC_CALL_PROC_BUSY, &call->flags)) {
851 _debug("XXXXXXXXXXXXX RUNNING ON MULTIPLE CPUS XXXXXXXXXXXXX");
852 return;
853 }
854
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;
861 msg.msg_flags = 0;
862
863 hdr.epoch = call->conn->epoch;
864 hdr.cid = call->cid;
865 hdr.callNumber = call->call_id;
866 hdr.seq = 0;
867 hdr.type = RXRPC_PACKET_TYPE_ACK;
868 hdr.flags = call->conn->out_clientflag;
869 hdr.userStatus = 0;
870 hdr.securityIndex = call->conn->security_ix;
871 hdr._rsvd = 0;
872 hdr.serviceId = call->conn->service_id;
873
874 memset(iov, 0, sizeof(iov));
875 iov[0].iov_base = &hdr;
876 iov[0].iov_len = sizeof(hdr);
877
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);
882 }
883
884 if (test_bit(RXRPC_CALL_RCVD_ERROR, &call->events)) {
885 int error;
886
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);
890
891 error = call->conn->trans->peer->net_error;
892 _debug("post net error %d", error);
893
894 if (rxrpc_post_message(call, RXRPC_SKB_MARK_NET_ERROR,
895 error, true) < 0)
896 goto no_mem;
897 clear_bit(RXRPC_CALL_RCVD_ERROR, &call->events);
898 goto kill_ACKs;
899 }
900
901 if (test_bit(RXRPC_CALL_CONN_ABORT, &call->events)) {
902 ASSERTCMP(call->state, >, RXRPC_CALL_COMPLETE);
903
904 clear_bit(RXRPC_CALL_REJECT_BUSY, &call->events);
905 clear_bit(RXRPC_CALL_ABORT, &call->events);
906
907 _debug("post conn abort");
908
909 if (rxrpc_post_message(call, RXRPC_SKB_MARK_LOCAL_ERROR,
910 call->conn->error, true) < 0)
911 goto no_mem;
912 clear_bit(RXRPC_CALL_CONN_ABORT, &call->events);
913 goto kill_ACKs;
914 }
915
916 if (test_bit(RXRPC_CALL_REJECT_BUSY, &call->events)) {
917 hdr.type = RXRPC_PACKET_TYPE_BUSY;
918 genbit = RXRPC_CALL_REJECT_BUSY;
919 goto send_message;
920 }
921
922 if (test_bit(RXRPC_CALL_ABORT, &call->events)) {
923 ASSERTCMP(call->state, >, RXRPC_CALL_COMPLETE);
924
925 if (rxrpc_post_message(call, RXRPC_SKB_MARK_LOCAL_ERROR,
926 ECONNABORTED, true) < 0)
927 goto no_mem;
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;
933 goto send_message;
934 }
935
936 if (test_bit(RXRPC_CALL_ACK_FINAL, &call->events)) {
937 genbit = RXRPC_CALL_ACK_FINAL;
938
939 ack.bufferSpace = htons(8);
940 ack.maxSkew = 0;
941 ack.serial = 0;
942 ack.reason = RXRPC_ACK_IDLE;
943 ack.nAcks = 0;
944 call->ackr_reason = 0;
945
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);
951
952 pad = 0;
953
954 iov[1].iov_base = &ack;
955 iov[1].iov_len = sizeof(ack);
956 iov[2].iov_base = &pad;
957 iov[2].iov_len = 3;
958 iov[3].iov_base = &ackinfo;
959 iov[3].iov_len = sizeof(ackinfo);
960 goto send_ACK;
961 }
962
963 if (call->events & ((1 << RXRPC_CALL_RCVD_BUSY) |
964 (1 << RXRPC_CALL_RCVD_ABORT))
965 ) {
966 u32 mark;
967
968 if (test_bit(RXRPC_CALL_RCVD_ABORT, &call->events))
969 mark = RXRPC_SKB_MARK_REMOTE_ABORT;
970 else
971 mark = RXRPC_SKB_MARK_BUSY;
972
973 _debug("post abort/busy");
974 rxrpc_clear_tx_window(call);
975 if (rxrpc_post_message(call, mark, ECONNABORTED, true) < 0)
976 goto no_mem;
977
978 clear_bit(RXRPC_CALL_RCVD_BUSY, &call->events);
979 clear_bit(RXRPC_CALL_RCVD_ABORT, &call->events);
980 goto kill_ACKs;
981 }
982
983 if (test_and_clear_bit(RXRPC_CALL_RCVD_ACKALL, &call->events)) {
984 _debug("do implicit ackall");
985 rxrpc_clear_tx_window(call);
986 }
987
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);
994 }
995 write_unlock_bh(&call->state_lock);
996
997 _debug("post timeout");
998 if (rxrpc_post_message(call, RXRPC_SKB_MARK_LOCAL_ERROR,
999 ETIME, true) < 0)
1000 goto no_mem;
1001
1002 clear_bit(RXRPC_CALL_LIFE_TIMER, &call->events);
1003 goto kill_ACKs;
1004 }
1005
1006 /* deal with assorted inbound messages */
1007 if (!skb_queue_empty(&call->rx_queue)) {
1008 switch (rxrpc_process_rx_queue(call, &abort_code)) {
1009 case 0:
1010 case -EAGAIN:
1011 break;
1012 case -ENOMEM:
1013 goto no_mem;
1014 case -EKEYEXPIRED:
1015 case -EKEYREJECTED:
1016 case -EPROTO:
1017 rxrpc_abort_call(call, abort_code);
1018 goto kill_ACKs;
1019 }
1020 }
1021
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))
1026 rxrpc_resend(call);
1027
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]);
1033
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;
1039 }
1040
1041 genbit = RXRPC_CALL_ACK;
1042
1043 acks = kzalloc(call->ackr_win_top - call->rx_data_eaten,
1044 GFP_NOFS);
1045 if (!acks)
1046 goto no_mem;
1047
1048 //hdr.flags = RXRPC_SLOW_START_OK;
1049 ack.bufferSpace = htons(8);
1050 ack.maxSkew = 0;
1051 ack.serial = 0;
1052 ack.reason = 0;
1053
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);
1059
1060 ack.nAcks = 0;
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
1064 ) {
1065 _debug("- l=%d n=%d b=%lx", loop, nbit, bits);
1066 if (bits & 1) {
1067 acks[nbit] = RXRPC_ACK_TYPE_ACK;
1068 ack.nAcks = nbit + 1;
1069 }
1070 nbit++;
1071 }
1072 }
1073 call->ackr_reason = 0;
1074 spin_unlock_bh(&call->lock);
1075
1076 pad = 0;
1077
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;
1083 iov[3].iov_len = 3;
1084 iov[4].iov_base = &ackinfo;
1085 iov[4].iov_len = sizeof(ackinfo);
1086
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:
1098 goto send_ACK;
1099 }
1100 }
1101
1102 /* handle completion of security negotiations on an incoming
1103 * connection */
1104 if (test_and_clear_bit(RXRPC_CALL_SECURED, &call->events)) {
1105 _debug("secured");
1106 spin_lock_bh(&call->lock);
1107
1108 if (call->state == RXRPC_CALL_SERVER_SECURING) {
1109 _debug("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);
1117 }
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);
1123 }
1124
1125 spin_unlock_bh(&call->lock);
1126 if (!test_bit(RXRPC_CALL_POST_ACCEPT, &call->events))
1127 goto maybe_reschedule;
1128 }
1129
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,
1134 0, false) < 0)
1135 goto no_mem;
1136 clear_bit(RXRPC_CALL_POST_ACCEPT, &call->events);
1137 goto maybe_reschedule;
1138 }
1139
1140 /* handle incoming call acceptance */
1141 if (test_and_clear_bit(RXRPC_CALL_ACCEPTED, &call->events)) {
1142 _debug("accepted");
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);
1149 }
1150
1151 /* drain the out of sequence received packet queue into the packet Rx
1152 * queue */
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)
1156 break;
1157 goto maybe_reschedule;
1158 }
1159
1160 /* other events may have been raised since we started checking */
1161 goto maybe_reschedule;
1162
1163 send_ACK_with_skew:
1164 ack.maxSkew = htons(atomic_read(&call->conn->hi_serial) -
1165 ntohl(ack.serial));
1166 send_ACK:
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);
1171
1172 /* permit the peer to send us jumbo packets if it wants to */
1173 ackinfo.rxMTU = htonl(5692);
1174 ackinfo.jumbo_max = htonl(4);
1175
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 }",
1178 ntohl(hdr.serial),
1179 ntohs(ack.maxSkew),
1180 ntohl(ack.firstPacket),
1181 ntohl(ack.previousPacket),
1182 ntohl(ack.serial),
1183 rxrpc_acks[ack.reason],
1184 ack.nAcks);
1185
1186 del_timer_sync(&call->ack_timer);
1187 if (ack.nAcks > 0)
1188 set_bit(RXRPC_CALL_TX_SOFT_ACK, &call->flags);
1189 goto send_message_2;
1190
1191 send_message:
1192 _debug("send message");
1193
1194 hdr.serial = htonl(atomic_inc_return(&call->conn->serial));
1195 _proto("Tx %s %%%u", rxrpc_pkts[hdr.type], ntohl(hdr.serial));
1196 send_message_2:
1197
1198 len = iov[0].iov_len;
1199 ioc = 1;
1200 if (iov[4].iov_len) {
1201 ioc = 5;
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) {
1207 ioc = 4;
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) {
1212 ioc = 3;
1213 len += iov[2].iov_len;
1214 len += iov[1].iov_len;
1215 } else if (iov[1].iov_len) {
1216 ioc = 2;
1217 len += iov[1].iov_len;
1218 }
1219
1220 ret = kernel_sendmsg(call->conn->trans->local->socket,
1221 &msg, iov, ioc, len);
1222 if (ret < 0) {
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);
1228 goto error;
1229 }
1230
1231 switch (genbit) {
1232 case RXRPC_CALL_ABORT:
1233 clear_bit(genbit, &call->events);
1234 clear_bit(RXRPC_CALL_RCVD_ABORT, &call->events);
1235 goto kill_ACKs;
1236
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);
1242 goto kill_ACKs;
1243
1244 default:
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);
1254 default:
1255 break;
1256 }
1257 goto maybe_reschedule;
1258 }
1259
1260 kill_ACKs:
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);
1265
1266 maybe_reschedule:
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);
1272 }
1273
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));
1280
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);
1286 }
1287
1288 error:
1289 clear_bit(RXRPC_CALL_PROC_BUSY, &call->flags);
1290 kfree(acks);
1291
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);
1299 }
1300
1301 _leave("");
1302 return;
1303
1304 no_mem:
1305 _debug("out of memory");
1306 goto maybe_reschedule;
1307 }
Linus' kernel tree