fs/hfsplus: use bool instead of int for is_known_namespace() return value
[linux-2.6-block.git] / net / sctp / ulpqueue.c
CommitLineData
60c778b2 1/* SCTP kernel implementation
1da177e4
LT
2 * (C) Copyright IBM Corp. 2001, 2004
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
5 * Copyright (c) 2001 Intel Corp.
6 * Copyright (c) 2001 Nokia, Inc.
7 * Copyright (c) 2001 La Monte H.P. Yarroll
8 *
9 * This abstraction carries sctp events to the ULP (sockets).
10 *
60c778b2 11 * This SCTP implementation is free software;
1da177e4
LT
12 * you can redistribute it and/or modify it under the terms of
13 * the GNU General Public License as published by
14 * the Free Software Foundation; either version 2, or (at your option)
15 * any later version.
16 *
60c778b2 17 * This SCTP implementation is distributed in the hope that it
1da177e4
LT
18 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
19 * ************************
20 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
21 * See the GNU General Public License for more details.
22 *
23 * You should have received a copy of the GNU General Public License
4b2f13a2
JK
24 * along with GNU CC; see the file COPYING. If not, see
25 * <http://www.gnu.org/licenses/>.
1da177e4
LT
26 *
27 * Please send any bug reports or fixes you make to the
28 * email address(es):
91705c61 29 * lksctp developers <linux-sctp@vger.kernel.org>
1da177e4 30 *
1da177e4
LT
31 * Written or modified by:
32 * Jon Grimm <jgrimm@us.ibm.com>
33 * La Monte H.P. Yarroll <piggy@acm.org>
34 * Sridhar Samudrala <sri@us.ibm.com>
1da177e4
LT
35 */
36
5a0e3ad6 37#include <linux/slab.h>
1da177e4
LT
38#include <linux/types.h>
39#include <linux/skbuff.h>
40#include <net/sock.h>
8465a5fc 41#include <net/busy_poll.h>
1da177e4
LT
42#include <net/sctp/structs.h>
43#include <net/sctp/sctp.h>
44#include <net/sctp/sm.h>
45
46/* Forward declarations for internal helpers. */
26ac8e5f 47static struct sctp_ulpevent *sctp_ulpq_reasm(struct sctp_ulpq *ulpq,
8728b834 48 struct sctp_ulpevent *);
26ac8e5f 49static struct sctp_ulpevent *sctp_ulpq_order(struct sctp_ulpq *,
8728b834 50 struct sctp_ulpevent *);
ef5d4cf2 51static void sctp_ulpq_reasm_drain(struct sctp_ulpq *ulpq);
1da177e4
LT
52
53/* 1st Level Abstractions */
54
55/* Initialize a ULP queue from a block of memory. */
56struct sctp_ulpq *sctp_ulpq_init(struct sctp_ulpq *ulpq,
57 struct sctp_association *asoc)
58{
59 memset(ulpq, 0, sizeof(struct sctp_ulpq));
60
61 ulpq->asoc = asoc;
62 skb_queue_head_init(&ulpq->reasm);
63 skb_queue_head_init(&ulpq->lobby);
64 ulpq->pd_mode = 0;
1da177e4
LT
65
66 return ulpq;
67}
68
69
70/* Flush the reassembly and ordering queues. */
0b58a811 71void sctp_ulpq_flush(struct sctp_ulpq *ulpq)
1da177e4
LT
72{
73 struct sk_buff *skb;
74 struct sctp_ulpevent *event;
75
76 while ((skb = __skb_dequeue(&ulpq->lobby)) != NULL) {
77 event = sctp_skb2event(skb);
78 sctp_ulpevent_free(event);
79 }
80
81 while ((skb = __skb_dequeue(&ulpq->reasm)) != NULL) {
82 event = sctp_skb2event(skb);
83 sctp_ulpevent_free(event);
84 }
85
86}
87
88/* Dispose of a ulpqueue. */
89void sctp_ulpq_free(struct sctp_ulpq *ulpq)
90{
91 sctp_ulpq_flush(ulpq);
1da177e4
LT
92}
93
94/* Process an incoming DATA chunk. */
95int sctp_ulpq_tail_data(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk,
dd0fc66f 96 gfp_t gfp)
1da177e4
LT
97{
98 struct sk_buff_head temp;
1da177e4 99 struct sctp_ulpevent *event;
d003b41b 100 int event_eor = 0;
1da177e4 101
1da177e4
LT
102 /* Create an event from the incoming chunk. */
103 event = sctp_ulpevent_make_rcvmsg(chunk->asoc, chunk, gfp);
104 if (!event)
105 return -ENOMEM;
106
107 /* Do reassembly if needed. */
108 event = sctp_ulpq_reasm(ulpq, event);
109
110 /* Do ordering if needed. */
cb3f837b 111 if ((event) && (event->msg_flags & MSG_EOR)) {
1da177e4
LT
112 /* Create a temporary list to collect chunks on. */
113 skb_queue_head_init(&temp);
114 __skb_queue_tail(&temp, sctp_event2skb(event));
115
116 event = sctp_ulpq_order(ulpq, event);
117 }
118
8728b834
DM
119 /* Send event to the ULP. 'event' is the sctp_ulpevent for
120 * very first SKB on the 'temp' list.
121 */
d003b41b
LR
122 if (event) {
123 event_eor = (event->msg_flags & MSG_EOR) ? 1 : 0;
1da177e4 124 sctp_ulpq_tail_event(ulpq, event);
d003b41b 125 }
1da177e4 126
d003b41b 127 return event_eor;
1da177e4
LT
128}
129
130/* Add a new event for propagation to the ULP. */
131/* Clear the partial delivery mode for this socket. Note: This
132 * assumes that no association is currently in partial delivery mode.
133 */
b6e1331f 134int sctp_clear_pd(struct sock *sk, struct sctp_association *asoc)
1da177e4
LT
135{
136 struct sctp_sock *sp = sctp_sk(sk);
137
b6e1331f
VY
138 if (atomic_dec_and_test(&sp->pd_mode)) {
139 /* This means there are no other associations in PD, so
140 * we can go ahead and clear out the lobby in one shot
141 */
142 if (!skb_queue_empty(&sp->pd_lobby)) {
143 struct list_head *list;
144 sctp_skb_list_tail(&sp->pd_lobby, &sk->sk_receive_queue);
145 list = (struct list_head *)&sctp_sk(sk)->pd_lobby;
146 INIT_LIST_HEAD(list);
147 return 1;
148 }
149 } else {
150 /* There are other associations in PD, so we only need to
151 * pull stuff out of the lobby that belongs to the
152 * associations that is exiting PD (all of its notifications
153 * are posted here).
154 */
155 if (!skb_queue_empty(&sp->pd_lobby) && asoc) {
156 struct sk_buff *skb, *tmp;
157 struct sctp_ulpevent *event;
158
159 sctp_skb_for_each(skb, &sp->pd_lobby, tmp) {
160 event = sctp_skb2event(skb);
161 if (event->asoc == asoc) {
162 __skb_unlink(skb, &sp->pd_lobby);
163 __skb_queue_tail(&sk->sk_receive_queue,
164 skb);
165 }
166 }
167 }
1da177e4 168 }
b6e1331f 169
1da177e4
LT
170 return 0;
171}
172
d49d91d7
VY
173/* Set the pd_mode on the socket and ulpq */
174static void sctp_ulpq_set_pd(struct sctp_ulpq *ulpq)
175{
176 struct sctp_sock *sp = sctp_sk(ulpq->asoc->base.sk);
177
178 atomic_inc(&sp->pd_mode);
179 ulpq->pd_mode = 1;
180}
181
1da177e4
LT
182/* Clear the pd_mode and restart any pending messages waiting for delivery. */
183static int sctp_ulpq_clear_pd(struct sctp_ulpq *ulpq)
184{
185 ulpq->pd_mode = 0;
ef5d4cf2 186 sctp_ulpq_reasm_drain(ulpq);
b6e1331f 187 return sctp_clear_pd(ulpq->asoc->base.sk, ulpq->asoc);
1da177e4
LT
188}
189
8728b834
DM
190/* If the SKB of 'event' is on a list, it is the first such member
191 * of that list.
192 */
1da177e4
LT
193int sctp_ulpq_tail_event(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event)
194{
195 struct sock *sk = ulpq->asoc->base.sk;
8728b834
DM
196 struct sk_buff_head *queue, *skb_list;
197 struct sk_buff *skb = sctp_event2skb(event);
1da177e4
LT
198 int clear_pd = 0;
199
8728b834
DM
200 skb_list = (struct sk_buff_head *) skb->prev;
201
1da177e4
LT
202 /* If the socket is just going to throw this away, do not
203 * even try to deliver it.
204 */
205 if (sock_flag(sk, SOCK_DEAD) || (sk->sk_shutdown & RCV_SHUTDOWN))
206 goto out_free;
207
2c8c56e1 208 if (!sctp_ulpevent_is_notification(event)) {
8465a5fc 209 sk_mark_napi_id(sk, skb);
2c8c56e1
ED
210 sk_incoming_cpu_update(sk);
211 }
1da177e4
LT
212 /* Check if the user wishes to receive this event. */
213 if (!sctp_ulpevent_is_enabled(event, &sctp_sk(sk)->subscribe))
214 goto out_free;
215
216 /* If we are in partial delivery mode, post to the lobby until
217 * partial delivery is cleared, unless, of course _this_ is
218 * the association the cause of the partial delivery.
219 */
220
b6e1331f 221 if (atomic_read(&sctp_sk(sk)->pd_mode) == 0) {
1da177e4 222 queue = &sk->sk_receive_queue;
b6e1331f
VY
223 } else {
224 if (ulpq->pd_mode) {
225 /* If the association is in partial delivery, we
226 * need to finish delivering the partially processed
227 * packet before passing any other data. This is
228 * because we don't truly support stream interleaving.
229 */
230 if ((event->msg_flags & MSG_NOTIFICATION) ||
231 (SCTP_DATA_NOT_FRAG ==
232 (event->msg_flags & SCTP_DATA_FRAG_MASK)))
233 queue = &sctp_sk(sk)->pd_lobby;
234 else {
235 clear_pd = event->msg_flags & MSG_EOR;
236 queue = &sk->sk_receive_queue;
237 }
238 } else {
239 /*
240 * If fragment interleave is enabled, we
25985edc 241 * can queue this to the receive queue instead
b6e1331f
VY
242 * of the lobby.
243 */
244 if (sctp_sk(sk)->frag_interleave)
245 queue = &sk->sk_receive_queue;
246 else
247 queue = &sctp_sk(sk)->pd_lobby;
1da177e4 248 }
b6e1331f 249 }
1da177e4
LT
250
251 /* If we are harvesting multiple skbs they will be
252 * collected on a list.
253 */
8728b834
DM
254 if (skb_list)
255 sctp_skb_list_tail(skb_list, queue);
1da177e4 256 else
8728b834 257 __skb_queue_tail(queue, skb);
1da177e4
LT
258
259 /* Did we just complete partial delivery and need to get
260 * rolling again? Move pending data to the receive
261 * queue.
262 */
263 if (clear_pd)
264 sctp_ulpq_clear_pd(ulpq);
265
266 if (queue == &sk->sk_receive_queue)
676d2369 267 sk->sk_data_ready(sk);
1da177e4
LT
268 return 1;
269
270out_free:
8728b834
DM
271 if (skb_list)
272 sctp_queue_purge_ulpevents(skb_list);
1da177e4
LT
273 else
274 sctp_ulpevent_free(event);
8728b834 275
1da177e4
LT
276 return 0;
277}
278
279/* 2nd Level Abstractions */
280
281/* Helper function to store chunks that need to be reassembled. */
01f2d384 282static void sctp_ulpq_store_reasm(struct sctp_ulpq *ulpq,
1da177e4
LT
283 struct sctp_ulpevent *event)
284{
285 struct sk_buff *pos;
286 struct sctp_ulpevent *cevent;
287 __u32 tsn, ctsn;
288
289 tsn = event->tsn;
290
291 /* See if it belongs at the end. */
292 pos = skb_peek_tail(&ulpq->reasm);
293 if (!pos) {
294 __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event));
295 return;
296 }
297
298 /* Short circuit just dropping it at the end. */
299 cevent = sctp_skb2event(pos);
300 ctsn = cevent->tsn;
301 if (TSN_lt(ctsn, tsn)) {
302 __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event));
303 return;
304 }
305
306 /* Find the right place in this list. We store them by TSN. */
307 skb_queue_walk(&ulpq->reasm, pos) {
308 cevent = sctp_skb2event(pos);
309 ctsn = cevent->tsn;
310
311 if (TSN_lt(tsn, ctsn))
312 break;
313 }
314
315 /* Insert before pos. */
43f59c89 316 __skb_queue_before(&ulpq->reasm, pos, sctp_event2skb(event));
1da177e4
LT
317
318}
319
320/* Helper function to return an event corresponding to the reassembled
321 * datagram.
322 * This routine creates a re-assembled skb given the first and last skb's
323 * as stored in the reassembly queue. The skb's may be non-linear if the sctp
324 * payload was fragmented on the way and ip had to reassemble them.
325 * We add the rest of skb's to the first skb's fraglist.
326 */
b01a2407
EB
327static struct sctp_ulpevent *sctp_make_reassembled_event(struct net *net,
328 struct sk_buff_head *queue, struct sk_buff *f_frag,
329 struct sk_buff *l_frag)
1da177e4
LT
330{
331 struct sk_buff *pos;
672e7cca 332 struct sk_buff *new = NULL;
1da177e4
LT
333 struct sctp_ulpevent *event;
334 struct sk_buff *pnext, *last;
335 struct sk_buff *list = skb_shinfo(f_frag)->frag_list;
336
337 /* Store the pointer to the 2nd skb */
338 if (f_frag == l_frag)
339 pos = NULL;
340 else
341 pos = f_frag->next;
342
343 /* Get the last skb in the f_frag's frag_list if present. */
8d72651d 344 for (last = list; list; last = list, list = list->next)
345 ;
1da177e4
LT
346
347 /* Add the list of remaining fragments to the first fragments
348 * frag_list.
349 */
350 if (last)
351 last->next = pos;
d808ad9a
YH
352 else {
353 if (skb_cloned(f_frag)) {
354 /* This is a cloned skb, we can't just modify
355 * the frag_list. We need a new skb to do that.
356 * Instead of calling skb_unshare(), we'll do it
357 * ourselves since we need to delay the free.
358 */
359 new = skb_copy(f_frag, GFP_ATOMIC);
360 if (!new)
361 return NULL; /* try again later */
362
363 sctp_skb_set_owner_r(new, f_frag->sk);
364
365 skb_shinfo(new)->frag_list = pos;
366 } else
367 skb_shinfo(f_frag)->frag_list = pos;
368 }
1da177e4
LT
369
370 /* Remove the first fragment from the reassembly queue. */
8728b834 371 __skb_unlink(f_frag, queue);
672e7cca 372
d808ad9a
YH
373 /* if we did unshare, then free the old skb and re-assign */
374 if (new) {
375 kfree_skb(f_frag);
376 f_frag = new;
377 }
672e7cca 378
1da177e4
LT
379 while (pos) {
380
381 pnext = pos->next;
382
383 /* Update the len and data_len fields of the first fragment. */
384 f_frag->len += pos->len;
385 f_frag->data_len += pos->len;
386
387 /* Remove the fragment from the reassembly queue. */
8728b834 388 __skb_unlink(pos, queue);
d808ad9a 389
1da177e4
LT
390 /* Break if we have reached the last fragment. */
391 if (pos == l_frag)
392 break;
393 pos->next = pnext;
394 pos = pnext;
3ff50b79 395 }
1da177e4
LT
396
397 event = sctp_skb2event(f_frag);
b01a2407 398 SCTP_INC_STATS(net, SCTP_MIB_REASMUSRMSGS);
1da177e4
LT
399
400 return event;
401}
402
403
404/* Helper function to check if an incoming chunk has filled up the last
405 * missing fragment in a SCTP datagram and return the corresponding event.
406 */
01f2d384 407static struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_ulpq *ulpq)
1da177e4
LT
408{
409 struct sk_buff *pos;
410 struct sctp_ulpevent *cevent;
411 struct sk_buff *first_frag = NULL;
412 __u32 ctsn, next_tsn;
413 struct sctp_ulpevent *retval = NULL;
d49d91d7
VY
414 struct sk_buff *pd_first = NULL;
415 struct sk_buff *pd_last = NULL;
416 size_t pd_len = 0;
417 struct sctp_association *asoc;
418 u32 pd_point;
1da177e4
LT
419
420 /* Initialized to 0 just to avoid compiler warning message. Will
421 * never be used with this value. It is referenced only after it
422 * is set when we find the first fragment of a message.
423 */
424 next_tsn = 0;
425
426 /* The chunks are held in the reasm queue sorted by TSN.
427 * Walk through the queue sequentially and look for a sequence of
428 * fragmented chunks that complete a datagram.
429 * 'first_frag' and next_tsn are reset when we find a chunk which
430 * is the first fragment of a datagram. Once these 2 fields are set
431 * we expect to find the remaining middle fragments and the last
432 * fragment in order. If not, first_frag is reset to NULL and we
433 * start the next pass when we find another first fragment.
d49d91d7
VY
434 *
435 * There is a potential to do partial delivery if user sets
436 * SCTP_PARTIAL_DELIVERY_POINT option. Lets count some things here
437 * to see if can do PD.
1da177e4
LT
438 */
439 skb_queue_walk(&ulpq->reasm, pos) {
440 cevent = sctp_skb2event(pos);
441 ctsn = cevent->tsn;
442
443 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
444 case SCTP_DATA_FIRST_FRAG:
d49d91d7
VY
445 /* If this "FIRST_FRAG" is the first
446 * element in the queue, then count it towards
447 * possible PD.
448 */
449 if (pos == ulpq->reasm.next) {
450 pd_first = pos;
451 pd_last = pos;
452 pd_len = pos->len;
453 } else {
454 pd_first = NULL;
455 pd_last = NULL;
456 pd_len = 0;
457 }
458
1da177e4
LT
459 first_frag = pos;
460 next_tsn = ctsn + 1;
461 break;
462
463 case SCTP_DATA_MIDDLE_FRAG:
d49d91d7 464 if ((first_frag) && (ctsn == next_tsn)) {
1da177e4 465 next_tsn++;
d49d91d7
VY
466 if (pd_first) {
467 pd_last = pos;
468 pd_len += pos->len;
469 }
470 } else
1da177e4
LT
471 first_frag = NULL;
472 break;
473
474 case SCTP_DATA_LAST_FRAG:
475 if (first_frag && (ctsn == next_tsn))
476 goto found;
477 else
478 first_frag = NULL;
479 break;
3ff50b79 480 }
d49d91d7
VY
481 }
482
483 asoc = ulpq->asoc;
484 if (pd_first) {
485 /* Make sure we can enter partial deliver.
486 * We can trigger partial delivery only if framgent
487 * interleave is set, or the socket is not already
488 * in partial delivery.
489 */
490 if (!sctp_sk(asoc->base.sk)->frag_interleave &&
491 atomic_read(&sctp_sk(asoc->base.sk)->pd_mode))
492 goto done;
1da177e4 493
d49d91d7
VY
494 cevent = sctp_skb2event(pd_first);
495 pd_point = sctp_sk(asoc->base.sk)->pd_point;
496 if (pd_point && pd_point <= pd_len) {
b01a2407
EB
497 retval = sctp_make_reassembled_event(sock_net(asoc->base.sk),
498 &ulpq->reasm,
d49d91d7
VY
499 pd_first,
500 pd_last);
501 if (retval)
502 sctp_ulpq_set_pd(ulpq);
503 }
1da177e4
LT
504 }
505done:
506 return retval;
507found:
b01a2407
EB
508 retval = sctp_make_reassembled_event(sock_net(ulpq->asoc->base.sk),
509 &ulpq->reasm, first_frag, pos);
1da177e4
LT
510 if (retval)
511 retval->msg_flags |= MSG_EOR;
512 goto done;
513}
514
515/* Retrieve the next set of fragments of a partial message. */
01f2d384 516static struct sctp_ulpevent *sctp_ulpq_retrieve_partial(struct sctp_ulpq *ulpq)
1da177e4
LT
517{
518 struct sk_buff *pos, *last_frag, *first_frag;
519 struct sctp_ulpevent *cevent;
520 __u32 ctsn, next_tsn;
521 int is_last;
522 struct sctp_ulpevent *retval;
523
524 /* The chunks are held in the reasm queue sorted by TSN.
525 * Walk through the queue sequentially and look for the first
526 * sequence of fragmented chunks.
527 */
528
529 if (skb_queue_empty(&ulpq->reasm))
530 return NULL;
531
532 last_frag = first_frag = NULL;
533 retval = NULL;
534 next_tsn = 0;
535 is_last = 0;
536
537 skb_queue_walk(&ulpq->reasm, pos) {
538 cevent = sctp_skb2event(pos);
539 ctsn = cevent->tsn;
540
541 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
d003b41b
LR
542 case SCTP_DATA_FIRST_FRAG:
543 if (!first_frag)
544 return NULL;
545 goto done;
1da177e4
LT
546 case SCTP_DATA_MIDDLE_FRAG:
547 if (!first_frag) {
548 first_frag = pos;
549 next_tsn = ctsn + 1;
550 last_frag = pos;
d003b41b 551 } else if (next_tsn == ctsn) {
1da177e4 552 next_tsn++;
d003b41b
LR
553 last_frag = pos;
554 } else
1da177e4
LT
555 goto done;
556 break;
557 case SCTP_DATA_LAST_FRAG:
558 if (!first_frag)
559 first_frag = pos;
560 else if (ctsn != next_tsn)
561 goto done;
562 last_frag = pos;
563 is_last = 1;
564 goto done;
565 default:
566 return NULL;
3ff50b79 567 }
1da177e4
LT
568 }
569
570 /* We have the reassembled event. There is no need to look
571 * further.
572 */
573done:
b01a2407
EB
574 retval = sctp_make_reassembled_event(sock_net(ulpq->asoc->base.sk),
575 &ulpq->reasm, first_frag, last_frag);
1da177e4
LT
576 if (retval && is_last)
577 retval->msg_flags |= MSG_EOR;
578
579 return retval;
580}
581
582
583/* Helper function to reassemble chunks. Hold chunks on the reasm queue that
584 * need reassembling.
585 */
586static struct sctp_ulpevent *sctp_ulpq_reasm(struct sctp_ulpq *ulpq,
587 struct sctp_ulpevent *event)
588{
589 struct sctp_ulpevent *retval = NULL;
590
591 /* Check if this is part of a fragmented message. */
592 if (SCTP_DATA_NOT_FRAG == (event->msg_flags & SCTP_DATA_FRAG_MASK)) {
593 event->msg_flags |= MSG_EOR;
594 return event;
595 }
596
597 sctp_ulpq_store_reasm(ulpq, event);
598 if (!ulpq->pd_mode)
599 retval = sctp_ulpq_retrieve_reassembled(ulpq);
600 else {
601 __u32 ctsn, ctsnap;
602
603 /* Do not even bother unless this is the next tsn to
604 * be delivered.
605 */
606 ctsn = event->tsn;
607 ctsnap = sctp_tsnmap_get_ctsn(&ulpq->asoc->peer.tsn_map);
608 if (TSN_lte(ctsn, ctsnap))
609 retval = sctp_ulpq_retrieve_partial(ulpq);
610 }
611
612 return retval;
613}
614
615/* Retrieve the first part (sequential fragments) for partial delivery. */
01f2d384 616static struct sctp_ulpevent *sctp_ulpq_retrieve_first(struct sctp_ulpq *ulpq)
1da177e4
LT
617{
618 struct sk_buff *pos, *last_frag, *first_frag;
619 struct sctp_ulpevent *cevent;
620 __u32 ctsn, next_tsn;
621 struct sctp_ulpevent *retval;
622
623 /* The chunks are held in the reasm queue sorted by TSN.
624 * Walk through the queue sequentially and look for a sequence of
625 * fragmented chunks that start a datagram.
626 */
627
628 if (skb_queue_empty(&ulpq->reasm))
629 return NULL;
630
631 last_frag = first_frag = NULL;
632 retval = NULL;
633 next_tsn = 0;
634
635 skb_queue_walk(&ulpq->reasm, pos) {
636 cevent = sctp_skb2event(pos);
637 ctsn = cevent->tsn;
638
639 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
640 case SCTP_DATA_FIRST_FRAG:
641 if (!first_frag) {
642 first_frag = pos;
643 next_tsn = ctsn + 1;
644 last_frag = pos;
645 } else
646 goto done;
647 break;
648
649 case SCTP_DATA_MIDDLE_FRAG:
650 if (!first_frag)
651 return NULL;
652 if (ctsn == next_tsn) {
653 next_tsn++;
654 last_frag = pos;
655 } else
656 goto done;
657 break;
d003b41b
LR
658
659 case SCTP_DATA_LAST_FRAG:
660 if (!first_frag)
661 return NULL;
662 else
663 goto done;
664 break;
665
1da177e4
LT
666 default:
667 return NULL;
3ff50b79 668 }
1da177e4
LT
669 }
670
671 /* We have the reassembled event. There is no need to look
672 * further.
673 */
674done:
b01a2407
EB
675 retval = sctp_make_reassembled_event(sock_net(ulpq->asoc->base.sk),
676 &ulpq->reasm, first_frag, last_frag);
1da177e4
LT
677 return retval;
678}
679
ea2dfb37
VY
680/*
681 * Flush out stale fragments from the reassembly queue when processing
682 * a Forward TSN.
683 *
684 * RFC 3758, Section 3.6
685 *
686 * After receiving and processing a FORWARD TSN, the data receiver MUST
687 * take cautions in updating its re-assembly queue. The receiver MUST
688 * remove any partially reassembled message, which is still missing one
689 * or more TSNs earlier than or equal to the new cumulative TSN point.
690 * In the event that the receiver has invoked the partial delivery API,
691 * a notification SHOULD also be generated to inform the upper layer API
692 * that the message being partially delivered will NOT be completed.
693 */
694void sctp_ulpq_reasm_flushtsn(struct sctp_ulpq *ulpq, __u32 fwd_tsn)
695{
696 struct sk_buff *pos, *tmp;
697 struct sctp_ulpevent *event;
698 __u32 tsn;
699
700 if (skb_queue_empty(&ulpq->reasm))
701 return;
702
703 skb_queue_walk_safe(&ulpq->reasm, pos, tmp) {
704 event = sctp_skb2event(pos);
705 tsn = event->tsn;
706
707 /* Since the entire message must be abandoned by the
708 * sender (item A3 in Section 3.5, RFC 3758), we can
709 * free all fragments on the list that are less then
710 * or equal to ctsn_point
711 */
712 if (TSN_lte(tsn, fwd_tsn)) {
713 __skb_unlink(pos, &ulpq->reasm);
714 sctp_ulpevent_free(event);
715 } else
716 break;
717 }
718}
719
ef5d4cf2
VY
720/*
721 * Drain the reassembly queue. If we just cleared parted delivery, it
722 * is possible that the reassembly queue will contain already reassembled
723 * messages. Retrieve any such messages and give them to the user.
724 */
725static void sctp_ulpq_reasm_drain(struct sctp_ulpq *ulpq)
726{
727 struct sctp_ulpevent *event = NULL;
728 struct sk_buff_head temp;
729
730 if (skb_queue_empty(&ulpq->reasm))
731 return;
732
733 while ((event = sctp_ulpq_retrieve_reassembled(ulpq)) != NULL) {
734 /* Do ordering if needed. */
cb3f837b 735 if ((event) && (event->msg_flags & MSG_EOR)) {
ef5d4cf2
VY
736 skb_queue_head_init(&temp);
737 __skb_queue_tail(&temp, sctp_event2skb(event));
738
739 event = sctp_ulpq_order(ulpq, event);
740 }
741
742 /* Send event to the ULP. 'event' is the
743 * sctp_ulpevent for very first SKB on the temp' list.
744 */
745 if (event)
746 sctp_ulpq_tail_event(ulpq, event);
747 }
748}
749
750
1da177e4
LT
751/* Helper function to gather skbs that have possibly become
752 * ordered by an an incoming chunk.
753 */
01f2d384 754static void sctp_ulpq_retrieve_ordered(struct sctp_ulpq *ulpq,
1da177e4
LT
755 struct sctp_ulpevent *event)
756{
8728b834 757 struct sk_buff_head *event_list;
1da177e4
LT
758 struct sk_buff *pos, *tmp;
759 struct sctp_ulpevent *cevent;
760 struct sctp_stream *in;
efea2c6b 761 __u16 sid, csid, cssn;
1da177e4
LT
762
763 sid = event->stream;
1da177e4
LT
764 in = &ulpq->asoc->ssnmap->in;
765
8728b834
DM
766 event_list = (struct sk_buff_head *) sctp_event2skb(event)->prev;
767
1da177e4
LT
768 /* We are holding the chunks by stream, by SSN. */
769 sctp_skb_for_each(pos, &ulpq->lobby, tmp) {
770 cevent = (struct sctp_ulpevent *) pos->cb;
771 csid = cevent->stream;
772 cssn = cevent->ssn;
773
774 /* Have we gone too far? */
775 if (csid > sid)
776 break;
777
778 /* Have we not gone far enough? */
779 if (csid < sid)
780 continue;
781
782 if (cssn != sctp_ssn_peek(in, sid))
783 break;
784
785 /* Found it, so mark in the ssnmap. */
786 sctp_ssn_next(in, sid);
787
8728b834 788 __skb_unlink(pos, &ulpq->lobby);
1da177e4
LT
789
790 /* Attach all gathered skbs to the event. */
8728b834 791 __skb_queue_tail(event_list, pos);
1da177e4
LT
792 }
793}
794
795/* Helper function to store chunks needing ordering. */
01f2d384 796static void sctp_ulpq_store_ordered(struct sctp_ulpq *ulpq,
1da177e4
LT
797 struct sctp_ulpevent *event)
798{
799 struct sk_buff *pos;
800 struct sctp_ulpevent *cevent;
801 __u16 sid, csid;
802 __u16 ssn, cssn;
803
804 pos = skb_peek_tail(&ulpq->lobby);
805 if (!pos) {
806 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
807 return;
808 }
809
810 sid = event->stream;
811 ssn = event->ssn;
d808ad9a 812
1da177e4
LT
813 cevent = (struct sctp_ulpevent *) pos->cb;
814 csid = cevent->stream;
815 cssn = cevent->ssn;
816 if (sid > csid) {
817 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
818 return;
819 }
820
821 if ((sid == csid) && SSN_lt(cssn, ssn)) {
822 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
823 return;
824 }
825
826 /* Find the right place in this list. We store them by
827 * stream ID and then by SSN.
828 */
829 skb_queue_walk(&ulpq->lobby, pos) {
830 cevent = (struct sctp_ulpevent *) pos->cb;
831 csid = cevent->stream;
832 cssn = cevent->ssn;
833
834 if (csid > sid)
835 break;
836 if (csid == sid && SSN_lt(ssn, cssn))
837 break;
838 }
839
840
841 /* Insert before pos. */
43f59c89 842 __skb_queue_before(&ulpq->lobby, pos, sctp_event2skb(event));
1da177e4
LT
843}
844
845static struct sctp_ulpevent *sctp_ulpq_order(struct sctp_ulpq *ulpq,
8728b834 846 struct sctp_ulpevent *event)
1da177e4
LT
847{
848 __u16 sid, ssn;
849 struct sctp_stream *in;
850
851 /* Check if this message needs ordering. */
852 if (SCTP_DATA_UNORDERED & event->msg_flags)
853 return event;
854
855 /* Note: The stream ID must be verified before this routine. */
856 sid = event->stream;
857 ssn = event->ssn;
858 in = &ulpq->asoc->ssnmap->in;
859
860 /* Is this the expected SSN for this stream ID? */
861 if (ssn != sctp_ssn_peek(in, sid)) {
862 /* We've received something out of order, so find where it
863 * needs to be placed. We order by stream and then by SSN.
864 */
865 sctp_ulpq_store_ordered(ulpq, event);
866 return NULL;
867 }
868
869 /* Mark that the next chunk has been found. */
870 sctp_ssn_next(in, sid);
871
872 /* Go find any other chunks that were waiting for
873 * ordering.
874 */
875 sctp_ulpq_retrieve_ordered(ulpq, event);
876
877 return event;
878}
879
880/* Helper function to gather skbs that have possibly become
881 * ordered by forward tsn skipping their dependencies.
882 */
01f2d384 883static void sctp_ulpq_reap_ordered(struct sctp_ulpq *ulpq, __u16 sid)
1da177e4
LT
884{
885 struct sk_buff *pos, *tmp;
886 struct sctp_ulpevent *cevent;
8728b834 887 struct sctp_ulpevent *event;
1da177e4
LT
888 struct sctp_stream *in;
889 struct sk_buff_head temp;
c068be54 890 struct sk_buff_head *lobby = &ulpq->lobby;
1da177e4
LT
891 __u16 csid, cssn;
892
893 in = &ulpq->asoc->ssnmap->in;
894
895 /* We are holding the chunks by stream, by SSN. */
8728b834
DM
896 skb_queue_head_init(&temp);
897 event = NULL;
c068be54 898 sctp_skb_for_each(pos, lobby, tmp) {
1da177e4
LT
899 cevent = (struct sctp_ulpevent *) pos->cb;
900 csid = cevent->stream;
901 cssn = cevent->ssn;
902
ea2dfb37
VY
903 /* Have we gone too far? */
904 if (csid > sid)
1da177e4
LT
905 break;
906
ea2dfb37
VY
907 /* Have we not gone far enough? */
908 if (csid < sid)
909 continue;
910
911 /* see if this ssn has been marked by skipping */
c068be54 912 if (!SSN_lt(cssn, sctp_ssn_peek(in, csid)))
ea2dfb37 913 break;
1da177e4 914
c068be54 915 __skb_unlink(pos, lobby);
ea2dfb37 916 if (!event)
1da177e4
LT
917 /* Create a temporary list to collect chunks on. */
918 event = sctp_skb2event(pos);
ea2dfb37
VY
919
920 /* Attach all gathered skbs to the event. */
921 __skb_queue_tail(&temp, pos);
1da177e4
LT
922 }
923
c068be54
VY
924 /* If we didn't reap any data, see if the next expected SSN
925 * is next on the queue and if so, use that.
926 */
927 if (event == NULL && pos != (struct sk_buff *)lobby) {
928 cevent = (struct sctp_ulpevent *) pos->cb;
929 csid = cevent->stream;
930 cssn = cevent->ssn;
931
932 if (csid == sid && cssn == sctp_ssn_peek(in, csid)) {
933 sctp_ssn_next(in, csid);
934 __skb_unlink(pos, lobby);
935 __skb_queue_tail(&temp, pos);
936 event = sctp_skb2event(pos);
937 }
938 }
939
8728b834
DM
940 /* Send event to the ULP. 'event' is the sctp_ulpevent for
941 * very first SKB on the 'temp' list.
942 */
ea2dfb37
VY
943 if (event) {
944 /* see if we have more ordered that we can deliver */
945 sctp_ulpq_retrieve_ordered(ulpq, event);
1da177e4 946 sctp_ulpq_tail_event(ulpq, event);
ea2dfb37 947 }
1da177e4
LT
948}
949
ea2dfb37
VY
950/* Skip over an SSN. This is used during the processing of
951 * Forwared TSN chunk to skip over the abandoned ordered data
952 */
1da177e4
LT
953void sctp_ulpq_skip(struct sctp_ulpq *ulpq, __u16 sid, __u16 ssn)
954{
955 struct sctp_stream *in;
956
957 /* Note: The stream ID must be verified before this routine. */
958 in = &ulpq->asoc->ssnmap->in;
959
960 /* Is this an old SSN? If so ignore. */
961 if (SSN_lt(ssn, sctp_ssn_peek(in, sid)))
962 return;
963
964 /* Mark that we are no longer expecting this SSN or lower. */
965 sctp_ssn_skip(in, sid, ssn);
966
967 /* Go find any other chunks that were waiting for
d808ad9a 968 * ordering and deliver them if needed.
1da177e4 969 */
ea2dfb37 970 sctp_ulpq_reap_ordered(ulpq, sid);
1da177e4
LT
971}
972
16d14ef9
PE
973static __u16 sctp_ulpq_renege_list(struct sctp_ulpq *ulpq,
974 struct sk_buff_head *list, __u16 needed)
1da177e4
LT
975{
976 __u16 freed = 0;
95ac7b85
LR
977 __u32 tsn, last_tsn;
978 struct sk_buff *skb, *flist, *last;
1da177e4
LT
979 struct sctp_ulpevent *event;
980 struct sctp_tsnmap *tsnmap;
981
982 tsnmap = &ulpq->asoc->peer.tsn_map;
983
e67f85ec 984 while ((skb = skb_peek_tail(list)) != NULL) {
1da177e4
LT
985 event = sctp_skb2event(skb);
986 tsn = event->tsn;
987
e67f85ec
LR
988 /* Don't renege below the Cumulative TSN ACK Point. */
989 if (TSN_lte(tsn, sctp_tsnmap_get_ctsn(tsnmap)))
990 break;
991
95ac7b85
LR
992 /* Events in ordering queue may have multiple fragments
993 * corresponding to additional TSNs. Sum the total
994 * freed space; find the last TSN.
995 */
e67f85ec 996 freed += skb_headlen(skb);
95ac7b85
LR
997 flist = skb_shinfo(skb)->frag_list;
998 for (last = flist; flist; flist = flist->next) {
999 last = flist;
1000 freed += skb_headlen(last);
1001 }
1002 if (last)
1003 last_tsn = sctp_skb2event(last)->tsn;
1004 else
1005 last_tsn = tsn;
1006
1007 /* Unlink the event, then renege all applicable TSNs. */
1008 __skb_unlink(skb, list);
1da177e4 1009 sctp_ulpevent_free(event);
95ac7b85
LR
1010 while (TSN_lte(tsn, last_tsn)) {
1011 sctp_tsnmap_renege(tsnmap, tsn);
1012 tsn++;
1013 }
1da177e4
LT
1014 if (freed >= needed)
1015 return freed;
1016 }
1017
1018 return freed;
1019}
1020
16d14ef9
PE
1021/* Renege 'needed' bytes from the ordering queue. */
1022static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed)
1023{
1024 return sctp_ulpq_renege_list(ulpq, &ulpq->lobby, needed);
1025}
1026
1da177e4
LT
1027/* Renege 'needed' bytes from the reassembly queue. */
1028static __u16 sctp_ulpq_renege_frags(struct sctp_ulpq *ulpq, __u16 needed)
1029{
16d14ef9 1030 return sctp_ulpq_renege_list(ulpq, &ulpq->reasm, needed);
1da177e4
LT
1031}
1032
1033/* Partial deliver the first message as there is pressure on rwnd. */
1034void sctp_ulpq_partial_delivery(struct sctp_ulpq *ulpq,
dd0fc66f 1035 gfp_t gfp)
1da177e4
LT
1036{
1037 struct sctp_ulpevent *event;
1038 struct sctp_association *asoc;
b6e1331f 1039 struct sctp_sock *sp;
d003b41b
LR
1040 __u32 ctsn;
1041 struct sk_buff *skb;
1da177e4
LT
1042
1043 asoc = ulpq->asoc;
b6e1331f 1044 sp = sctp_sk(asoc->base.sk);
1da177e4 1045
b6e1331f 1046 /* If the association is already in Partial Delivery mode
d003b41b 1047 * we have nothing to do.
b6e1331f
VY
1048 */
1049 if (ulpq->pd_mode)
1050 return;
1da177e4 1051
d003b41b
LR
1052 /* Data must be at or below the Cumulative TSN ACK Point to
1053 * start partial delivery.
1054 */
1055 skb = skb_peek(&asoc->ulpq.reasm);
1056 if (skb != NULL) {
1057 ctsn = sctp_skb2event(skb)->tsn;
1058 if (!TSN_lte(ctsn, sctp_tsnmap_get_ctsn(&asoc->peer.tsn_map)))
1059 return;
1060 }
1061
b6e1331f
VY
1062 /* If the user enabled fragment interleave socket option,
1063 * multiple associations can enter partial delivery.
1064 * Otherwise, we can only enter partial delivery if the
1065 * socket is not in partial deliver mode.
1066 */
1067 if (sp->frag_interleave || atomic_read(&sp->pd_mode) == 0) {
1da177e4
LT
1068 /* Is partial delivery possible? */
1069 event = sctp_ulpq_retrieve_first(ulpq);
1070 /* Send event to the ULP. */
1071 if (event) {
1072 sctp_ulpq_tail_event(ulpq, event);
d49d91d7 1073 sctp_ulpq_set_pd(ulpq);
1da177e4
LT
1074 return;
1075 }
1076 }
1077}
1078
1079/* Renege some packets to make room for an incoming chunk. */
1080void sctp_ulpq_renege(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk,
dd0fc66f 1081 gfp_t gfp)
1da177e4
LT
1082{
1083 struct sctp_association *asoc;
1084 __u16 needed, freed;
1085
1086 asoc = ulpq->asoc;
1087
1088 if (chunk) {
1089 needed = ntohs(chunk->chunk_hdr->length);
1090 needed -= sizeof(sctp_data_chunk_t);
d808ad9a 1091 } else
1da177e4
LT
1092 needed = SCTP_DEFAULT_MAXWINDOW;
1093
1094 freed = 0;
1095
1096 if (skb_queue_empty(&asoc->base.sk->sk_receive_queue)) {
1097 freed = sctp_ulpq_renege_order(ulpq, needed);
1098 if (freed < needed) {
1099 freed += sctp_ulpq_renege_frags(ulpq, needed - freed);
1100 }
1101 }
1102 /* If able to free enough room, accept this chunk. */
1103 if (chunk && (freed >= needed)) {
d003b41b
LR
1104 int retval;
1105 retval = sctp_ulpq_tail_data(ulpq, chunk, gfp);
1106 /*
1107 * Enter partial delivery if chunk has not been
1108 * delivered; otherwise, drain the reassembly queue.
1109 */
1110 if (retval <= 0)
1111 sctp_ulpq_partial_delivery(ulpq, gfp);
1112 else if (retval == 1)
1113 sctp_ulpq_reasm_drain(ulpq);
1da177e4
LT
1114 }
1115
3ab224be 1116 sk_mem_reclaim(asoc->base.sk);
1da177e4
LT
1117}
1118
1119
1120
1121/* Notify the application if an association is aborted and in
1122 * partial delivery mode. Send up any pending received messages.
1123 */
dd0fc66f 1124void sctp_ulpq_abort_pd(struct sctp_ulpq *ulpq, gfp_t gfp)
1da177e4
LT
1125{
1126 struct sctp_ulpevent *ev = NULL;
1127 struct sock *sk;
1128
1129 if (!ulpq->pd_mode)
1130 return;
1131
1132 sk = ulpq->asoc->base.sk;
1133 if (sctp_ulpevent_type_enabled(SCTP_PARTIAL_DELIVERY_EVENT,
1134 &sctp_sk(sk)->subscribe))
1135 ev = sctp_ulpevent_make_pdapi(ulpq->asoc,
1136 SCTP_PARTIAL_DELIVERY_ABORTED,
1137 gfp);
1138 if (ev)
1139 __skb_queue_tail(&sk->sk_receive_queue, sctp_event2skb(ev));
1140
1141 /* If there is data waiting, send it up the socket now. */
1142 if (sctp_ulpq_clear_pd(ulpq) || ev)
676d2369 1143 sk->sk_data_ready(sk);
1da177e4 1144}