blob: ac80c34f6c2c1572cadaea0d09298f5e421f243b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* SCTP kernel reference Implementation
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 *
11 * The SCTP reference implementation is free software;
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 *
17 * The SCTP reference implementation is distributed in the hope that it
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
24 * along with GNU CC; see the file COPYING. If not, write to
25 * the Free Software Foundation, 59 Temple Place - Suite 330,
26 * Boston, MA 02111-1307, USA.
27 *
28 * Please send any bug reports or fixes you make to the
29 * email address(es):
30 * lksctp developers <lksctp-developers@lists.sourceforge.net>
31 *
32 * Or submit a bug report through the following website:
33 * http://www.sf.net/projects/lksctp
34 *
35 * Written or modified by:
36 * Jon Grimm <jgrimm@us.ibm.com>
37 * La Monte H.P. Yarroll <piggy@acm.org>
38 * Sridhar Samudrala <sri@us.ibm.com>
39 *
40 * Any bugs reported given to us we will try to fix... any fixes shared will
41 * be incorporated into the next SCTP release.
42 */
43
44#include <linux/types.h>
45#include <linux/skbuff.h>
46#include <net/sock.h>
47#include <net/sctp/structs.h>
48#include <net/sctp/sctp.h>
49#include <net/sctp/sm.h>
50
51/* Forward declarations for internal helpers. */
52static struct sctp_ulpevent * sctp_ulpq_reasm(struct sctp_ulpq *ulpq,
David S. Miller8728b832005-08-09 19:25:21 -070053 struct sctp_ulpevent *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070054static struct sctp_ulpevent * sctp_ulpq_order(struct sctp_ulpq *,
David S. Miller8728b832005-08-09 19:25:21 -070055 struct sctp_ulpevent *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
57/* 1st Level Abstractions */
58
59/* Initialize a ULP queue from a block of memory. */
60struct sctp_ulpq *sctp_ulpq_init(struct sctp_ulpq *ulpq,
61 struct sctp_association *asoc)
62{
63 memset(ulpq, 0, sizeof(struct sctp_ulpq));
64
65 ulpq->asoc = asoc;
66 skb_queue_head_init(&ulpq->reasm);
67 skb_queue_head_init(&ulpq->lobby);
68 ulpq->pd_mode = 0;
69 ulpq->malloced = 0;
70
71 return ulpq;
72}
73
74
75/* Flush the reassembly and ordering queues. */
Vlad Yasevich0b58a812007-03-19 17:01:17 -070076void sctp_ulpq_flush(struct sctp_ulpq *ulpq)
Linus Torvalds1da177e2005-04-16 15:20:36 -070077{
78 struct sk_buff *skb;
79 struct sctp_ulpevent *event;
80
81 while ((skb = __skb_dequeue(&ulpq->lobby)) != NULL) {
82 event = sctp_skb2event(skb);
83 sctp_ulpevent_free(event);
84 }
85
86 while ((skb = __skb_dequeue(&ulpq->reasm)) != NULL) {
87 event = sctp_skb2event(skb);
88 sctp_ulpevent_free(event);
89 }
90
91}
92
93/* Dispose of a ulpqueue. */
94void sctp_ulpq_free(struct sctp_ulpq *ulpq)
95{
96 sctp_ulpq_flush(ulpq);
97 if (ulpq->malloced)
98 kfree(ulpq);
99}
100
101/* Process an incoming DATA chunk. */
102int sctp_ulpq_tail_data(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk,
Al Virodd0fc662005-10-07 07:46:04 +0100103 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
105 struct sk_buff_head temp;
106 sctp_data_chunk_t *hdr;
107 struct sctp_ulpevent *event;
108
109 hdr = (sctp_data_chunk_t *) chunk->chunk_hdr;
110
111 /* Create an event from the incoming chunk. */
112 event = sctp_ulpevent_make_rcvmsg(chunk->asoc, chunk, gfp);
113 if (!event)
114 return -ENOMEM;
115
116 /* Do reassembly if needed. */
117 event = sctp_ulpq_reasm(ulpq, event);
118
119 /* Do ordering if needed. */
120 if ((event) && (event->msg_flags & MSG_EOR)){
121 /* Create a temporary list to collect chunks on. */
122 skb_queue_head_init(&temp);
123 __skb_queue_tail(&temp, sctp_event2skb(event));
124
125 event = sctp_ulpq_order(ulpq, event);
126 }
127
David S. Miller8728b832005-08-09 19:25:21 -0700128 /* Send event to the ULP. 'event' is the sctp_ulpevent for
129 * very first SKB on the 'temp' list.
130 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 if (event)
132 sctp_ulpq_tail_event(ulpq, event);
133
134 return 0;
135}
136
137/* Add a new event for propagation to the ULP. */
138/* Clear the partial delivery mode for this socket. Note: This
139 * assumes that no association is currently in partial delivery mode.
140 */
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700141int sctp_clear_pd(struct sock *sk, struct sctp_association *asoc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142{
143 struct sctp_sock *sp = sctp_sk(sk);
144
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700145 if (atomic_dec_and_test(&sp->pd_mode)) {
146 /* This means there are no other associations in PD, so
147 * we can go ahead and clear out the lobby in one shot
148 */
149 if (!skb_queue_empty(&sp->pd_lobby)) {
150 struct list_head *list;
151 sctp_skb_list_tail(&sp->pd_lobby, &sk->sk_receive_queue);
152 list = (struct list_head *)&sctp_sk(sk)->pd_lobby;
153 INIT_LIST_HEAD(list);
154 return 1;
155 }
156 } else {
157 /* There are other associations in PD, so we only need to
158 * pull stuff out of the lobby that belongs to the
159 * associations that is exiting PD (all of its notifications
160 * are posted here).
161 */
162 if (!skb_queue_empty(&sp->pd_lobby) && asoc) {
163 struct sk_buff *skb, *tmp;
164 struct sctp_ulpevent *event;
165
166 sctp_skb_for_each(skb, &sp->pd_lobby, tmp) {
167 event = sctp_skb2event(skb);
168 if (event->asoc == asoc) {
169 __skb_unlink(skb, &sp->pd_lobby);
170 __skb_queue_tail(&sk->sk_receive_queue,
171 skb);
172 }
173 }
174 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 }
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 return 0;
178}
179
180/* Clear the pd_mode and restart any pending messages waiting for delivery. */
181static int sctp_ulpq_clear_pd(struct sctp_ulpq *ulpq)
182{
183 ulpq->pd_mode = 0;
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700184 return sctp_clear_pd(ulpq->asoc->base.sk, ulpq->asoc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185}
186
David S. Miller8728b832005-08-09 19:25:21 -0700187/* If the SKB of 'event' is on a list, it is the first such member
188 * of that list.
189 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190int sctp_ulpq_tail_event(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event)
191{
192 struct sock *sk = ulpq->asoc->base.sk;
David S. Miller8728b832005-08-09 19:25:21 -0700193 struct sk_buff_head *queue, *skb_list;
194 struct sk_buff *skb = sctp_event2skb(event);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 int clear_pd = 0;
196
David S. Miller8728b832005-08-09 19:25:21 -0700197 skb_list = (struct sk_buff_head *) skb->prev;
198
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 /* If the socket is just going to throw this away, do not
200 * even try to deliver it.
201 */
202 if (sock_flag(sk, SOCK_DEAD) || (sk->sk_shutdown & RCV_SHUTDOWN))
203 goto out_free;
204
205 /* Check if the user wishes to receive this event. */
206 if (!sctp_ulpevent_is_enabled(event, &sctp_sk(sk)->subscribe))
207 goto out_free;
208
209 /* If we are in partial delivery mode, post to the lobby until
210 * partial delivery is cleared, unless, of course _this_ is
211 * the association the cause of the partial delivery.
212 */
213
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700214 if (atomic_read(&sctp_sk(sk)->pd_mode) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 queue = &sk->sk_receive_queue;
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700216 } else {
217 if (ulpq->pd_mode) {
218 /* If the association is in partial delivery, we
219 * need to finish delivering the partially processed
220 * packet before passing any other data. This is
221 * because we don't truly support stream interleaving.
222 */
223 if ((event->msg_flags & MSG_NOTIFICATION) ||
224 (SCTP_DATA_NOT_FRAG ==
225 (event->msg_flags & SCTP_DATA_FRAG_MASK)))
226 queue = &sctp_sk(sk)->pd_lobby;
227 else {
228 clear_pd = event->msg_flags & MSG_EOR;
229 queue = &sk->sk_receive_queue;
230 }
231 } else {
232 /*
233 * If fragment interleave is enabled, we
234 * can queue this to the recieve queue instead
235 * of the lobby.
236 */
237 if (sctp_sk(sk)->frag_interleave)
238 queue = &sk->sk_receive_queue;
239 else
240 queue = &sctp_sk(sk)->pd_lobby;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 }
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
244 /* If we are harvesting multiple skbs they will be
245 * collected on a list.
246 */
David S. Miller8728b832005-08-09 19:25:21 -0700247 if (skb_list)
248 sctp_skb_list_tail(skb_list, queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 else
David S. Miller8728b832005-08-09 19:25:21 -0700250 __skb_queue_tail(queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252 /* Did we just complete partial delivery and need to get
253 * rolling again? Move pending data to the receive
254 * queue.
255 */
256 if (clear_pd)
257 sctp_ulpq_clear_pd(ulpq);
258
259 if (queue == &sk->sk_receive_queue)
260 sk->sk_data_ready(sk, 0);
261 return 1;
262
263out_free:
David S. Miller8728b832005-08-09 19:25:21 -0700264 if (skb_list)
265 sctp_queue_purge_ulpevents(skb_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 else
267 sctp_ulpevent_free(event);
David S. Miller8728b832005-08-09 19:25:21 -0700268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 return 0;
270}
271
272/* 2nd Level Abstractions */
273
274/* Helper function to store chunks that need to be reassembled. */
275static inline void sctp_ulpq_store_reasm(struct sctp_ulpq *ulpq,
276 struct sctp_ulpevent *event)
277{
278 struct sk_buff *pos;
279 struct sctp_ulpevent *cevent;
280 __u32 tsn, ctsn;
281
282 tsn = event->tsn;
283
284 /* See if it belongs at the end. */
285 pos = skb_peek_tail(&ulpq->reasm);
286 if (!pos) {
287 __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event));
288 return;
289 }
290
291 /* Short circuit just dropping it at the end. */
292 cevent = sctp_skb2event(pos);
293 ctsn = cevent->tsn;
294 if (TSN_lt(ctsn, tsn)) {
295 __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event));
296 return;
297 }
298
299 /* Find the right place in this list. We store them by TSN. */
300 skb_queue_walk(&ulpq->reasm, pos) {
301 cevent = sctp_skb2event(pos);
302 ctsn = cevent->tsn;
303
304 if (TSN_lt(tsn, ctsn))
305 break;
306 }
307
308 /* Insert before pos. */
309 __skb_insert(sctp_event2skb(event), pos->prev, pos, &ulpq->reasm);
310
311}
312
313/* Helper function to return an event corresponding to the reassembled
314 * datagram.
315 * This routine creates a re-assembled skb given the first and last skb's
316 * as stored in the reassembly queue. The skb's may be non-linear if the sctp
317 * payload was fragmented on the way and ip had to reassemble them.
318 * We add the rest of skb's to the first skb's fraglist.
319 */
David S. Miller8728b832005-08-09 19:25:21 -0700320static struct sctp_ulpevent *sctp_make_reassembled_event(struct sk_buff_head *queue, struct sk_buff *f_frag, struct sk_buff *l_frag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
322 struct sk_buff *pos;
Vladislav Yasevich672e7cc2006-05-05 17:03:49 -0700323 struct sk_buff *new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 struct sctp_ulpevent *event;
325 struct sk_buff *pnext, *last;
326 struct sk_buff *list = skb_shinfo(f_frag)->frag_list;
327
328 /* Store the pointer to the 2nd skb */
329 if (f_frag == l_frag)
330 pos = NULL;
331 else
332 pos = f_frag->next;
333
334 /* Get the last skb in the f_frag's frag_list if present. */
335 for (last = list; list; last = list, list = list->next);
336
337 /* Add the list of remaining fragments to the first fragments
338 * frag_list.
339 */
340 if (last)
341 last->next = pos;
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900342 else {
343 if (skb_cloned(f_frag)) {
344 /* This is a cloned skb, we can't just modify
345 * the frag_list. We need a new skb to do that.
346 * Instead of calling skb_unshare(), we'll do it
347 * ourselves since we need to delay the free.
348 */
349 new = skb_copy(f_frag, GFP_ATOMIC);
350 if (!new)
351 return NULL; /* try again later */
Vladislav Yasevich672e7cc2006-05-05 17:03:49 -0700352
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900353 sctp_skb_set_owner_r(new, f_frag->sk);
Vladislav Yasevich672e7cc2006-05-05 17:03:49 -0700354
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900355 skb_shinfo(new)->frag_list = pos;
356 } else
357 skb_shinfo(f_frag)->frag_list = pos;
358 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
360 /* Remove the first fragment from the reassembly queue. */
David S. Miller8728b832005-08-09 19:25:21 -0700361 __skb_unlink(f_frag, queue);
Vladislav Yasevich672e7cc2006-05-05 17:03:49 -0700362
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900363 /* if we did unshare, then free the old skb and re-assign */
364 if (new) {
365 kfree_skb(f_frag);
366 f_frag = new;
367 }
Vladislav Yasevich672e7cc2006-05-05 17:03:49 -0700368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 while (pos) {
370
371 pnext = pos->next;
372
373 /* Update the len and data_len fields of the first fragment. */
374 f_frag->len += pos->len;
375 f_frag->data_len += pos->len;
376
377 /* Remove the fragment from the reassembly queue. */
David S. Miller8728b832005-08-09 19:25:21 -0700378 __skb_unlink(pos, queue);
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900379
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 /* Break if we have reached the last fragment. */
381 if (pos == l_frag)
382 break;
383 pos->next = pnext;
384 pos = pnext;
385 };
386
387 event = sctp_skb2event(f_frag);
388 SCTP_INC_STATS(SCTP_MIB_REASMUSRMSGS);
389
390 return event;
391}
392
393
394/* Helper function to check if an incoming chunk has filled up the last
395 * missing fragment in a SCTP datagram and return the corresponding event.
396 */
397static inline struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_ulpq *ulpq)
398{
399 struct sk_buff *pos;
400 struct sctp_ulpevent *cevent;
401 struct sk_buff *first_frag = NULL;
402 __u32 ctsn, next_tsn;
403 struct sctp_ulpevent *retval = NULL;
404
405 /* Initialized to 0 just to avoid compiler warning message. Will
406 * never be used with this value. It is referenced only after it
407 * is set when we find the first fragment of a message.
408 */
409 next_tsn = 0;
410
411 /* The chunks are held in the reasm queue sorted by TSN.
412 * Walk through the queue sequentially and look for a sequence of
413 * fragmented chunks that complete a datagram.
414 * 'first_frag' and next_tsn are reset when we find a chunk which
415 * is the first fragment of a datagram. Once these 2 fields are set
416 * we expect to find the remaining middle fragments and the last
417 * fragment in order. If not, first_frag is reset to NULL and we
418 * start the next pass when we find another first fragment.
419 */
420 skb_queue_walk(&ulpq->reasm, pos) {
421 cevent = sctp_skb2event(pos);
422 ctsn = cevent->tsn;
423
424 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
425 case SCTP_DATA_FIRST_FRAG:
426 first_frag = pos;
427 next_tsn = ctsn + 1;
428 break;
429
430 case SCTP_DATA_MIDDLE_FRAG:
431 if ((first_frag) && (ctsn == next_tsn))
432 next_tsn++;
433 else
434 first_frag = NULL;
435 break;
436
437 case SCTP_DATA_LAST_FRAG:
438 if (first_frag && (ctsn == next_tsn))
439 goto found;
440 else
441 first_frag = NULL;
442 break;
443 };
444
445 }
446done:
447 return retval;
448found:
David S. Miller8728b832005-08-09 19:25:21 -0700449 retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 if (retval)
451 retval->msg_flags |= MSG_EOR;
452 goto done;
453}
454
455/* Retrieve the next set of fragments of a partial message. */
456static inline struct sctp_ulpevent *sctp_ulpq_retrieve_partial(struct sctp_ulpq *ulpq)
457{
458 struct sk_buff *pos, *last_frag, *first_frag;
459 struct sctp_ulpevent *cevent;
460 __u32 ctsn, next_tsn;
461 int is_last;
462 struct sctp_ulpevent *retval;
463
464 /* The chunks are held in the reasm queue sorted by TSN.
465 * Walk through the queue sequentially and look for the first
466 * sequence of fragmented chunks.
467 */
468
469 if (skb_queue_empty(&ulpq->reasm))
470 return NULL;
471
472 last_frag = first_frag = NULL;
473 retval = NULL;
474 next_tsn = 0;
475 is_last = 0;
476
477 skb_queue_walk(&ulpq->reasm, pos) {
478 cevent = sctp_skb2event(pos);
479 ctsn = cevent->tsn;
480
481 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
482 case SCTP_DATA_MIDDLE_FRAG:
483 if (!first_frag) {
484 first_frag = pos;
485 next_tsn = ctsn + 1;
486 last_frag = pos;
487 } else if (next_tsn == ctsn)
488 next_tsn++;
489 else
490 goto done;
491 break;
492 case SCTP_DATA_LAST_FRAG:
493 if (!first_frag)
494 first_frag = pos;
495 else if (ctsn != next_tsn)
496 goto done;
497 last_frag = pos;
498 is_last = 1;
499 goto done;
500 default:
501 return NULL;
502 };
503 }
504
505 /* We have the reassembled event. There is no need to look
506 * further.
507 */
508done:
David S. Miller8728b832005-08-09 19:25:21 -0700509 retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, last_frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 if (retval && is_last)
511 retval->msg_flags |= MSG_EOR;
512
513 return retval;
514}
515
516
517/* Helper function to reassemble chunks. Hold chunks on the reasm queue that
518 * need reassembling.
519 */
520static struct sctp_ulpevent *sctp_ulpq_reasm(struct sctp_ulpq *ulpq,
521 struct sctp_ulpevent *event)
522{
523 struct sctp_ulpevent *retval = NULL;
524
525 /* Check if this is part of a fragmented message. */
526 if (SCTP_DATA_NOT_FRAG == (event->msg_flags & SCTP_DATA_FRAG_MASK)) {
527 event->msg_flags |= MSG_EOR;
528 return event;
529 }
530
531 sctp_ulpq_store_reasm(ulpq, event);
532 if (!ulpq->pd_mode)
533 retval = sctp_ulpq_retrieve_reassembled(ulpq);
534 else {
535 __u32 ctsn, ctsnap;
536
537 /* Do not even bother unless this is the next tsn to
538 * be delivered.
539 */
540 ctsn = event->tsn;
541 ctsnap = sctp_tsnmap_get_ctsn(&ulpq->asoc->peer.tsn_map);
542 if (TSN_lte(ctsn, ctsnap))
543 retval = sctp_ulpq_retrieve_partial(ulpq);
544 }
545
546 return retval;
547}
548
549/* Retrieve the first part (sequential fragments) for partial delivery. */
550static inline struct sctp_ulpevent *sctp_ulpq_retrieve_first(struct sctp_ulpq *ulpq)
551{
552 struct sk_buff *pos, *last_frag, *first_frag;
553 struct sctp_ulpevent *cevent;
554 __u32 ctsn, next_tsn;
555 struct sctp_ulpevent *retval;
556
557 /* The chunks are held in the reasm queue sorted by TSN.
558 * Walk through the queue sequentially and look for a sequence of
559 * fragmented chunks that start a datagram.
560 */
561
562 if (skb_queue_empty(&ulpq->reasm))
563 return NULL;
564
565 last_frag = first_frag = NULL;
566 retval = NULL;
567 next_tsn = 0;
568
569 skb_queue_walk(&ulpq->reasm, pos) {
570 cevent = sctp_skb2event(pos);
571 ctsn = cevent->tsn;
572
573 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
574 case SCTP_DATA_FIRST_FRAG:
575 if (!first_frag) {
576 first_frag = pos;
577 next_tsn = ctsn + 1;
578 last_frag = pos;
579 } else
580 goto done;
581 break;
582
583 case SCTP_DATA_MIDDLE_FRAG:
584 if (!first_frag)
585 return NULL;
586 if (ctsn == next_tsn) {
587 next_tsn++;
588 last_frag = pos;
589 } else
590 goto done;
591 break;
592 default:
593 return NULL;
594 };
595 }
596
597 /* We have the reassembled event. There is no need to look
598 * further.
599 */
600done:
David S. Miller8728b832005-08-09 19:25:21 -0700601 retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, last_frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 return retval;
603}
604
605/* Helper function to gather skbs that have possibly become
606 * ordered by an an incoming chunk.
607 */
608static inline void sctp_ulpq_retrieve_ordered(struct sctp_ulpq *ulpq,
609 struct sctp_ulpevent *event)
610{
David S. Miller8728b832005-08-09 19:25:21 -0700611 struct sk_buff_head *event_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 struct sk_buff *pos, *tmp;
613 struct sctp_ulpevent *cevent;
614 struct sctp_stream *in;
615 __u16 sid, csid;
616 __u16 ssn, cssn;
617
618 sid = event->stream;
619 ssn = event->ssn;
620 in = &ulpq->asoc->ssnmap->in;
621
David S. Miller8728b832005-08-09 19:25:21 -0700622 event_list = (struct sk_buff_head *) sctp_event2skb(event)->prev;
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 /* We are holding the chunks by stream, by SSN. */
625 sctp_skb_for_each(pos, &ulpq->lobby, tmp) {
626 cevent = (struct sctp_ulpevent *) pos->cb;
627 csid = cevent->stream;
628 cssn = cevent->ssn;
629
630 /* Have we gone too far? */
631 if (csid > sid)
632 break;
633
634 /* Have we not gone far enough? */
635 if (csid < sid)
636 continue;
637
638 if (cssn != sctp_ssn_peek(in, sid))
639 break;
640
641 /* Found it, so mark in the ssnmap. */
642 sctp_ssn_next(in, sid);
643
David S. Miller8728b832005-08-09 19:25:21 -0700644 __skb_unlink(pos, &ulpq->lobby);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
646 /* Attach all gathered skbs to the event. */
David S. Miller8728b832005-08-09 19:25:21 -0700647 __skb_queue_tail(event_list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 }
649}
650
651/* Helper function to store chunks needing ordering. */
652static inline void sctp_ulpq_store_ordered(struct sctp_ulpq *ulpq,
653 struct sctp_ulpevent *event)
654{
655 struct sk_buff *pos;
656 struct sctp_ulpevent *cevent;
657 __u16 sid, csid;
658 __u16 ssn, cssn;
659
660 pos = skb_peek_tail(&ulpq->lobby);
661 if (!pos) {
662 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
663 return;
664 }
665
666 sid = event->stream;
667 ssn = event->ssn;
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 cevent = (struct sctp_ulpevent *) pos->cb;
670 csid = cevent->stream;
671 cssn = cevent->ssn;
672 if (sid > csid) {
673 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
674 return;
675 }
676
677 if ((sid == csid) && SSN_lt(cssn, ssn)) {
678 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
679 return;
680 }
681
682 /* Find the right place in this list. We store them by
683 * stream ID and then by SSN.
684 */
685 skb_queue_walk(&ulpq->lobby, pos) {
686 cevent = (struct sctp_ulpevent *) pos->cb;
687 csid = cevent->stream;
688 cssn = cevent->ssn;
689
690 if (csid > sid)
691 break;
692 if (csid == sid && SSN_lt(ssn, cssn))
693 break;
694 }
695
696
697 /* Insert before pos. */
698 __skb_insert(sctp_event2skb(event), pos->prev, pos, &ulpq->lobby);
699
700}
701
702static struct sctp_ulpevent *sctp_ulpq_order(struct sctp_ulpq *ulpq,
David S. Miller8728b832005-08-09 19:25:21 -0700703 struct sctp_ulpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704{
705 __u16 sid, ssn;
706 struct sctp_stream *in;
707
708 /* Check if this message needs ordering. */
709 if (SCTP_DATA_UNORDERED & event->msg_flags)
710 return event;
711
712 /* Note: The stream ID must be verified before this routine. */
713 sid = event->stream;
714 ssn = event->ssn;
715 in = &ulpq->asoc->ssnmap->in;
716
717 /* Is this the expected SSN for this stream ID? */
718 if (ssn != sctp_ssn_peek(in, sid)) {
719 /* We've received something out of order, so find where it
720 * needs to be placed. We order by stream and then by SSN.
721 */
722 sctp_ulpq_store_ordered(ulpq, event);
723 return NULL;
724 }
725
726 /* Mark that the next chunk has been found. */
727 sctp_ssn_next(in, sid);
728
729 /* Go find any other chunks that were waiting for
730 * ordering.
731 */
732 sctp_ulpq_retrieve_ordered(ulpq, event);
733
734 return event;
735}
736
737/* Helper function to gather skbs that have possibly become
738 * ordered by forward tsn skipping their dependencies.
739 */
740static inline void sctp_ulpq_reap_ordered(struct sctp_ulpq *ulpq)
741{
742 struct sk_buff *pos, *tmp;
743 struct sctp_ulpevent *cevent;
David S. Miller8728b832005-08-09 19:25:21 -0700744 struct sctp_ulpevent *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 struct sctp_stream *in;
746 struct sk_buff_head temp;
747 __u16 csid, cssn;
748
749 in = &ulpq->asoc->ssnmap->in;
750
751 /* We are holding the chunks by stream, by SSN. */
David S. Miller8728b832005-08-09 19:25:21 -0700752 skb_queue_head_init(&temp);
753 event = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 sctp_skb_for_each(pos, &ulpq->lobby, tmp) {
755 cevent = (struct sctp_ulpevent *) pos->cb;
756 csid = cevent->stream;
757 cssn = cevent->ssn;
758
759 if (cssn != sctp_ssn_peek(in, csid))
760 break;
761
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900762 /* Found it, so mark in the ssnmap. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 sctp_ssn_next(in, csid);
764
David S. Miller8728b832005-08-09 19:25:21 -0700765 __skb_unlink(pos, &ulpq->lobby);
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900766 if (!event) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 /* Create a temporary list to collect chunks on. */
768 event = sctp_skb2event(pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 __skb_queue_tail(&temp, sctp_event2skb(event));
770 } else {
771 /* Attach all gathered skbs to the event. */
David S. Miller8728b832005-08-09 19:25:21 -0700772 __skb_queue_tail(&temp, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 }
774 }
775
David S. Miller8728b832005-08-09 19:25:21 -0700776 /* Send event to the ULP. 'event' is the sctp_ulpevent for
777 * very first SKB on the 'temp' list.
778 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 if (event)
780 sctp_ulpq_tail_event(ulpq, event);
781}
782
783/* Skip over an SSN. */
784void sctp_ulpq_skip(struct sctp_ulpq *ulpq, __u16 sid, __u16 ssn)
785{
786 struct sctp_stream *in;
787
788 /* Note: The stream ID must be verified before this routine. */
789 in = &ulpq->asoc->ssnmap->in;
790
791 /* Is this an old SSN? If so ignore. */
792 if (SSN_lt(ssn, sctp_ssn_peek(in, sid)))
793 return;
794
795 /* Mark that we are no longer expecting this SSN or lower. */
796 sctp_ssn_skip(in, sid, ssn);
797
798 /* Go find any other chunks that were waiting for
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900799 * ordering and deliver them if needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 */
801 sctp_ulpq_reap_ordered(ulpq);
802 return;
803}
804
805/* Renege 'needed' bytes from the ordering queue. */
806static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed)
807{
808 __u16 freed = 0;
809 __u32 tsn;
810 struct sk_buff *skb;
811 struct sctp_ulpevent *event;
812 struct sctp_tsnmap *tsnmap;
813
814 tsnmap = &ulpq->asoc->peer.tsn_map;
815
816 while ((skb = __skb_dequeue_tail(&ulpq->lobby)) != NULL) {
817 freed += skb_headlen(skb);
818 event = sctp_skb2event(skb);
819 tsn = event->tsn;
820
821 sctp_ulpevent_free(event);
822 sctp_tsnmap_renege(tsnmap, tsn);
823 if (freed >= needed)
824 return freed;
825 }
826
827 return freed;
828}
829
830/* Renege 'needed' bytes from the reassembly queue. */
831static __u16 sctp_ulpq_renege_frags(struct sctp_ulpq *ulpq, __u16 needed)
832{
833 __u16 freed = 0;
834 __u32 tsn;
835 struct sk_buff *skb;
836 struct sctp_ulpevent *event;
837 struct sctp_tsnmap *tsnmap;
838
839 tsnmap = &ulpq->asoc->peer.tsn_map;
840
841 /* Walk backwards through the list, reneges the newest tsns. */
842 while ((skb = __skb_dequeue_tail(&ulpq->reasm)) != NULL) {
843 freed += skb_headlen(skb);
844 event = sctp_skb2event(skb);
845 tsn = event->tsn;
846
847 sctp_ulpevent_free(event);
848 sctp_tsnmap_renege(tsnmap, tsn);
849 if (freed >= needed)
850 return freed;
851 }
852
853 return freed;
854}
855
856/* Partial deliver the first message as there is pressure on rwnd. */
857void sctp_ulpq_partial_delivery(struct sctp_ulpq *ulpq,
Alexey Dobriyan3182cd82005-07-11 20:57:47 -0700858 struct sctp_chunk *chunk,
Al Virodd0fc662005-10-07 07:46:04 +0100859 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860{
861 struct sctp_ulpevent *event;
862 struct sctp_association *asoc;
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700863 struct sctp_sock *sp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
865 asoc = ulpq->asoc;
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700866 sp = sctp_sk(asoc->base.sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700868 /* If the association is already in Partial Delivery mode
869 * we have noting to do.
870 */
871 if (ulpq->pd_mode)
872 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700874 /* If the user enabled fragment interleave socket option,
875 * multiple associations can enter partial delivery.
876 * Otherwise, we can only enter partial delivery if the
877 * socket is not in partial deliver mode.
878 */
879 if (sp->frag_interleave || atomic_read(&sp->pd_mode) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 /* Is partial delivery possible? */
881 event = sctp_ulpq_retrieve_first(ulpq);
882 /* Send event to the ULP. */
883 if (event) {
884 sctp_ulpq_tail_event(ulpq, event);
Vlad Yasevichb6e13312007-04-20 12:23:15 -0700885 atomic_inc(&sp->pd_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 ulpq->pd_mode = 1;
887 return;
888 }
889 }
890}
891
892/* Renege some packets to make room for an incoming chunk. */
893void sctp_ulpq_renege(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk,
Al Virodd0fc662005-10-07 07:46:04 +0100894 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895{
896 struct sctp_association *asoc;
897 __u16 needed, freed;
898
899 asoc = ulpq->asoc;
900
901 if (chunk) {
902 needed = ntohs(chunk->chunk_hdr->length);
903 needed -= sizeof(sctp_data_chunk_t);
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900904 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 needed = SCTP_DEFAULT_MAXWINDOW;
906
907 freed = 0;
908
909 if (skb_queue_empty(&asoc->base.sk->sk_receive_queue)) {
910 freed = sctp_ulpq_renege_order(ulpq, needed);
911 if (freed < needed) {
912 freed += sctp_ulpq_renege_frags(ulpq, needed - freed);
913 }
914 }
915 /* If able to free enough room, accept this chunk. */
916 if (chunk && (freed >= needed)) {
917 __u32 tsn;
918 tsn = ntohl(chunk->subh.data_hdr->tsn);
919 sctp_tsnmap_mark(&asoc->peer.tsn_map, tsn);
920 sctp_ulpq_tail_data(ulpq, chunk, gfp);
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900921
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 sctp_ulpq_partial_delivery(ulpq, chunk, gfp);
923 }
924
925 return;
926}
927
928
929
930/* Notify the application if an association is aborted and in
931 * partial delivery mode. Send up any pending received messages.
932 */
Al Virodd0fc662005-10-07 07:46:04 +0100933void sctp_ulpq_abort_pd(struct sctp_ulpq *ulpq, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
935 struct sctp_ulpevent *ev = NULL;
936 struct sock *sk;
937
938 if (!ulpq->pd_mode)
939 return;
940
941 sk = ulpq->asoc->base.sk;
942 if (sctp_ulpevent_type_enabled(SCTP_PARTIAL_DELIVERY_EVENT,
943 &sctp_sk(sk)->subscribe))
944 ev = sctp_ulpevent_make_pdapi(ulpq->asoc,
945 SCTP_PARTIAL_DELIVERY_ABORTED,
946 gfp);
947 if (ev)
948 __skb_queue_tail(&sk->sk_receive_queue, sctp_event2skb(ev));
949
950 /* If there is data waiting, send it up the socket now. */
951 if (sctp_ulpq_clear_pd(ulpq) || ev)
952 sk->sk_data_ready(sk, 0);
953}