Root/
1 | /* SCTP kernel 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 | * This SCTP 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 | * This SCTP 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, see |
25 | * <http://www.gnu.org/licenses/>. |
26 | * |
27 | * Please send any bug reports or fixes you make to the |
28 | * email address(es): |
29 | * lksctp developers <linux-sctp@vger.kernel.org> |
30 | * |
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> |
35 | */ |
36 | |
37 | #include <linux/slab.h> |
38 | #include <linux/types.h> |
39 | #include <linux/skbuff.h> |
40 | #include <net/sock.h> |
41 | #include <net/sctp/structs.h> |
42 | #include <net/sctp/sctp.h> |
43 | #include <net/sctp/sm.h> |
44 | |
45 | /* Forward declarations for internal helpers. */ |
46 | static struct sctp_ulpevent *sctp_ulpq_reasm(struct sctp_ulpq *ulpq, |
47 | struct sctp_ulpevent *); |
48 | static struct sctp_ulpevent *sctp_ulpq_order(struct sctp_ulpq *, |
49 | struct sctp_ulpevent *); |
50 | static void sctp_ulpq_reasm_drain(struct sctp_ulpq *ulpq); |
51 | |
52 | /* 1st Level Abstractions */ |
53 | |
54 | /* Initialize a ULP queue from a block of memory. */ |
55 | struct sctp_ulpq *sctp_ulpq_init(struct sctp_ulpq *ulpq, |
56 | struct sctp_association *asoc) |
57 | { |
58 | memset(ulpq, 0, sizeof(struct sctp_ulpq)); |
59 | |
60 | ulpq->asoc = asoc; |
61 | skb_queue_head_init(&ulpq->reasm); |
62 | skb_queue_head_init(&ulpq->lobby); |
63 | ulpq->pd_mode = 0; |
64 | |
65 | return ulpq; |
66 | } |
67 | |
68 | |
69 | /* Flush the reassembly and ordering queues. */ |
70 | void sctp_ulpq_flush(struct sctp_ulpq *ulpq) |
71 | { |
72 | struct sk_buff *skb; |
73 | struct sctp_ulpevent *event; |
74 | |
75 | while ((skb = __skb_dequeue(&ulpq->lobby)) != NULL) { |
76 | event = sctp_skb2event(skb); |
77 | sctp_ulpevent_free(event); |
78 | } |
79 | |
80 | while ((skb = __skb_dequeue(&ulpq->reasm)) != NULL) { |
81 | event = sctp_skb2event(skb); |
82 | sctp_ulpevent_free(event); |
83 | } |
84 | |
85 | } |
86 | |
87 | /* Dispose of a ulpqueue. */ |
88 | void sctp_ulpq_free(struct sctp_ulpq *ulpq) |
89 | { |
90 | sctp_ulpq_flush(ulpq); |
91 | } |
92 | |
93 | /* Process an incoming DATA chunk. */ |
94 | int sctp_ulpq_tail_data(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk, |
95 | gfp_t gfp) |
96 | { |
97 | struct sk_buff_head temp; |
98 | struct sctp_ulpevent *event; |
99 | int event_eor = 0; |
100 | |
101 | /* Create an event from the incoming chunk. */ |
102 | event = sctp_ulpevent_make_rcvmsg(chunk->asoc, chunk, gfp); |
103 | if (!event) |
104 | return -ENOMEM; |
105 | |
106 | /* Do reassembly if needed. */ |
107 | event = sctp_ulpq_reasm(ulpq, event); |
108 | |
109 | /* Do ordering if needed. */ |
110 | if ((event) && (event->msg_flags & MSG_EOR)) { |
111 | /* Create a temporary list to collect chunks on. */ |
112 | skb_queue_head_init(&temp); |
113 | __skb_queue_tail(&temp, sctp_event2skb(event)); |
114 | |
115 | event = sctp_ulpq_order(ulpq, event); |
116 | } |
117 | |
118 | /* Send event to the ULP. 'event' is the sctp_ulpevent for |
119 | * very first SKB on the 'temp' list. |
120 | */ |
121 | if (event) { |
122 | event_eor = (event->msg_flags & MSG_EOR) ? 1 : 0; |
123 | sctp_ulpq_tail_event(ulpq, event); |
124 | } |
125 | |
126 | return event_eor; |
127 | } |
128 | |
129 | /* Add a new event for propagation to the ULP. */ |
130 | /* Clear the partial delivery mode for this socket. Note: This |
131 | * assumes that no association is currently in partial delivery mode. |
132 | */ |
133 | int sctp_clear_pd(struct sock *sk, struct sctp_association *asoc) |
134 | { |
135 | struct sctp_sock *sp = sctp_sk(sk); |
136 | |
137 | if (atomic_dec_and_test(&sp->pd_mode)) { |
138 | /* This means there are no other associations in PD, so |
139 | * we can go ahead and clear out the lobby in one shot |
140 | */ |
141 | if (!skb_queue_empty(&sp->pd_lobby)) { |
142 | struct list_head *list; |
143 | sctp_skb_list_tail(&sp->pd_lobby, &sk->sk_receive_queue); |
144 | list = (struct list_head *)&sctp_sk(sk)->pd_lobby; |
145 | INIT_LIST_HEAD(list); |
146 | return 1; |
147 | } |
148 | } else { |
149 | /* There are other associations in PD, so we only need to |
150 | * pull stuff out of the lobby that belongs to the |
151 | * associations that is exiting PD (all of its notifications |
152 | * are posted here). |
153 | */ |
154 | if (!skb_queue_empty(&sp->pd_lobby) && asoc) { |
155 | struct sk_buff *skb, *tmp; |
156 | struct sctp_ulpevent *event; |
157 | |
158 | sctp_skb_for_each(skb, &sp->pd_lobby, tmp) { |
159 | event = sctp_skb2event(skb); |
160 | if (event->asoc == asoc) { |
161 | __skb_unlink(skb, &sp->pd_lobby); |
162 | __skb_queue_tail(&sk->sk_receive_queue, |
163 | skb); |
164 | } |
165 | } |
166 | } |
167 | } |
168 | |
169 | return 0; |
170 | } |
171 | |
172 | /* Set the pd_mode on the socket and ulpq */ |
173 | static void sctp_ulpq_set_pd(struct sctp_ulpq *ulpq) |
174 | { |
175 | struct sctp_sock *sp = sctp_sk(ulpq->asoc->base.sk); |
176 | |
177 | atomic_inc(&sp->pd_mode); |
178 | ulpq->pd_mode = 1; |
179 | } |
180 | |
181 | /* Clear the pd_mode and restart any pending messages waiting for delivery. */ |
182 | static int sctp_ulpq_clear_pd(struct sctp_ulpq *ulpq) |
183 | { |
184 | ulpq->pd_mode = 0; |
185 | sctp_ulpq_reasm_drain(ulpq); |
186 | return sctp_clear_pd(ulpq->asoc->base.sk, ulpq->asoc); |
187 | } |
188 | |
189 | /* If the SKB of 'event' is on a list, it is the first such member |
190 | * of that list. |
191 | */ |
192 | int sctp_ulpq_tail_event(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event) |
193 | { |
194 | struct sock *sk = ulpq->asoc->base.sk; |
195 | struct sk_buff_head *queue, *skb_list; |
196 | struct sk_buff *skb = sctp_event2skb(event); |
197 | int clear_pd = 0; |
198 | |
199 | skb_list = (struct sk_buff_head *) skb->prev; |
200 | |
201 | /* If the socket is just going to throw this away, do not |
202 | * even try to deliver it. |
203 | */ |
204 | if (sock_flag(sk, SOCK_DEAD) || (sk->sk_shutdown & RCV_SHUTDOWN)) |
205 | goto out_free; |
206 | |
207 | /* Check if the user wishes to receive this event. */ |
208 | if (!sctp_ulpevent_is_enabled(event, &sctp_sk(sk)->subscribe)) |
209 | goto out_free; |
210 | |
211 | /* If we are in partial delivery mode, post to the lobby until |
212 | * partial delivery is cleared, unless, of course _this_ is |
213 | * the association the cause of the partial delivery. |
214 | */ |
215 | |
216 | if (atomic_read(&sctp_sk(sk)->pd_mode) == 0) { |
217 | queue = &sk->sk_receive_queue; |
218 | } else { |
219 | if (ulpq->pd_mode) { |
220 | /* If the association is in partial delivery, we |
221 | * need to finish delivering the partially processed |
222 | * packet before passing any other data. This is |
223 | * because we don't truly support stream interleaving. |
224 | */ |
225 | if ((event->msg_flags & MSG_NOTIFICATION) || |
226 | (SCTP_DATA_NOT_FRAG == |
227 | (event->msg_flags & SCTP_DATA_FRAG_MASK))) |
228 | queue = &sctp_sk(sk)->pd_lobby; |
229 | else { |
230 | clear_pd = event->msg_flags & MSG_EOR; |
231 | queue = &sk->sk_receive_queue; |
232 | } |
233 | } else { |
234 | /* |
235 | * If fragment interleave is enabled, we |
236 | * can queue this to the receive queue instead |
237 | * of the lobby. |
238 | */ |
239 | if (sctp_sk(sk)->frag_interleave) |
240 | queue = &sk->sk_receive_queue; |
241 | else |
242 | queue = &sctp_sk(sk)->pd_lobby; |
243 | } |
244 | } |
245 | |
246 | /* If we are harvesting multiple skbs they will be |
247 | * collected on a list. |
248 | */ |
249 | if (skb_list) |
250 | sctp_skb_list_tail(skb_list, queue); |
251 | else |
252 | __skb_queue_tail(queue, skb); |
253 | |
254 | /* Did we just complete partial delivery and need to get |
255 | * rolling again? Move pending data to the receive |
256 | * queue. |
257 | */ |
258 | if (clear_pd) |
259 | sctp_ulpq_clear_pd(ulpq); |
260 | |
261 | if (queue == &sk->sk_receive_queue) |
262 | sk->sk_data_ready(sk); |
263 | return 1; |
264 | |
265 | out_free: |
266 | if (skb_list) |
267 | sctp_queue_purge_ulpevents(skb_list); |
268 | else |
269 | sctp_ulpevent_free(event); |
270 | |
271 | return 0; |
272 | } |
273 | |
274 | /* 2nd Level Abstractions */ |
275 | |
276 | /* Helper function to store chunks that need to be reassembled. */ |
277 | static void sctp_ulpq_store_reasm(struct sctp_ulpq *ulpq, |
278 | struct sctp_ulpevent *event) |
279 | { |
280 | struct sk_buff *pos; |
281 | struct sctp_ulpevent *cevent; |
282 | __u32 tsn, ctsn; |
283 | |
284 | tsn = event->tsn; |
285 | |
286 | /* See if it belongs at the end. */ |
287 | pos = skb_peek_tail(&ulpq->reasm); |
288 | if (!pos) { |
289 | __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event)); |
290 | return; |
291 | } |
292 | |
293 | /* Short circuit just dropping it at the end. */ |
294 | cevent = sctp_skb2event(pos); |
295 | ctsn = cevent->tsn; |
296 | if (TSN_lt(ctsn, tsn)) { |
297 | __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event)); |
298 | return; |
299 | } |
300 | |
301 | /* Find the right place in this list. We store them by TSN. */ |
302 | skb_queue_walk(&ulpq->reasm, pos) { |
303 | cevent = sctp_skb2event(pos); |
304 | ctsn = cevent->tsn; |
305 | |
306 | if (TSN_lt(tsn, ctsn)) |
307 | break; |
308 | } |
309 | |
310 | /* Insert before pos. */ |
311 | __skb_queue_before(&ulpq->reasm, pos, sctp_event2skb(event)); |
312 | |
313 | } |
314 | |
315 | /* Helper function to return an event corresponding to the reassembled |
316 | * datagram. |
317 | * This routine creates a re-assembled skb given the first and last skb's |
318 | * as stored in the reassembly queue. The skb's may be non-linear if the sctp |
319 | * payload was fragmented on the way and ip had to reassemble them. |
320 | * We add the rest of skb's to the first skb's fraglist. |
321 | */ |
322 | static struct sctp_ulpevent *sctp_make_reassembled_event(struct net *net, |
323 | struct sk_buff_head *queue, struct sk_buff *f_frag, |
324 | struct sk_buff *l_frag) |
325 | { |
326 | struct sk_buff *pos; |
327 | struct sk_buff *new = NULL; |
328 | struct sctp_ulpevent *event; |
329 | struct sk_buff *pnext, *last; |
330 | struct sk_buff *list = skb_shinfo(f_frag)->frag_list; |
331 | |
332 | /* Store the pointer to the 2nd skb */ |
333 | if (f_frag == l_frag) |
334 | pos = NULL; |
335 | else |
336 | pos = f_frag->next; |
337 | |
338 | /* Get the last skb in the f_frag's frag_list if present. */ |
339 | for (last = list; list; last = list, list = list->next) |
340 | ; |
341 | |
342 | /* Add the list of remaining fragments to the first fragments |
343 | * frag_list. |
344 | */ |
345 | if (last) |
346 | last->next = pos; |
347 | else { |
348 | if (skb_cloned(f_frag)) { |
349 | /* This is a cloned skb, we can't just modify |
350 | * the frag_list. We need a new skb to do that. |
351 | * Instead of calling skb_unshare(), we'll do it |
352 | * ourselves since we need to delay the free. |
353 | */ |
354 | new = skb_copy(f_frag, GFP_ATOMIC); |
355 | if (!new) |
356 | return NULL; /* try again later */ |
357 | |
358 | sctp_skb_set_owner_r(new, f_frag->sk); |
359 | |
360 | skb_shinfo(new)->frag_list = pos; |
361 | } else |
362 | skb_shinfo(f_frag)->frag_list = pos; |
363 | } |
364 | |
365 | /* Remove the first fragment from the reassembly queue. */ |
366 | __skb_unlink(f_frag, queue); |
367 | |
368 | /* if we did unshare, then free the old skb and re-assign */ |
369 | if (new) { |
370 | kfree_skb(f_frag); |
371 | f_frag = new; |
372 | } |
373 | |
374 | while (pos) { |
375 | |
376 | pnext = pos->next; |
377 | |
378 | /* Update the len and data_len fields of the first fragment. */ |
379 | f_frag->len += pos->len; |
380 | f_frag->data_len += pos->len; |
381 | |
382 | /* Remove the fragment from the reassembly queue. */ |
383 | __skb_unlink(pos, queue); |
384 | |
385 | /* Break if we have reached the last fragment. */ |
386 | if (pos == l_frag) |
387 | break; |
388 | pos->next = pnext; |
389 | pos = pnext; |
390 | } |
391 | |
392 | event = sctp_skb2event(f_frag); |
393 | SCTP_INC_STATS(net, SCTP_MIB_REASMUSRMSGS); |
394 | |
395 | return event; |
396 | } |
397 | |
398 | |
399 | /* Helper function to check if an incoming chunk has filled up the last |
400 | * missing fragment in a SCTP datagram and return the corresponding event. |
401 | */ |
402 | static struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_ulpq *ulpq) |
403 | { |
404 | struct sk_buff *pos; |
405 | struct sctp_ulpevent *cevent; |
406 | struct sk_buff *first_frag = NULL; |
407 | __u32 ctsn, next_tsn; |
408 | struct sctp_ulpevent *retval = NULL; |
409 | struct sk_buff *pd_first = NULL; |
410 | struct sk_buff *pd_last = NULL; |
411 | size_t pd_len = 0; |
412 | struct sctp_association *asoc; |
413 | u32 pd_point; |
414 | |
415 | /* Initialized to 0 just to avoid compiler warning message. Will |
416 | * never be used with this value. It is referenced only after it |
417 | * is set when we find the first fragment of a message. |
418 | */ |
419 | next_tsn = 0; |
420 | |
421 | /* The chunks are held in the reasm queue sorted by TSN. |
422 | * Walk through the queue sequentially and look for a sequence of |
423 | * fragmented chunks that complete a datagram. |
424 | * 'first_frag' and next_tsn are reset when we find a chunk which |
425 | * is the first fragment of a datagram. Once these 2 fields are set |
426 | * we expect to find the remaining middle fragments and the last |
427 | * fragment in order. If not, first_frag is reset to NULL and we |
428 | * start the next pass when we find another first fragment. |
429 | * |
430 | * There is a potential to do partial delivery if user sets |
431 | * SCTP_PARTIAL_DELIVERY_POINT option. Lets count some things here |
432 | * to see if can do PD. |
433 | */ |
434 | skb_queue_walk(&ulpq->reasm, pos) { |
435 | cevent = sctp_skb2event(pos); |
436 | ctsn = cevent->tsn; |
437 | |
438 | switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) { |
439 | case SCTP_DATA_FIRST_FRAG: |
440 | /* If this "FIRST_FRAG" is the first |
441 | * element in the queue, then count it towards |
442 | * possible PD. |
443 | */ |
444 | if (pos == ulpq->reasm.next) { |
445 | pd_first = pos; |
446 | pd_last = pos; |
447 | pd_len = pos->len; |
448 | } else { |
449 | pd_first = NULL; |
450 | pd_last = NULL; |
451 | pd_len = 0; |
452 | } |
453 | |
454 | first_frag = pos; |
455 | next_tsn = ctsn + 1; |
456 | break; |
457 | |
458 | case SCTP_DATA_MIDDLE_FRAG: |
459 | if ((first_frag) && (ctsn == next_tsn)) { |
460 | next_tsn++; |
461 | if (pd_first) { |
462 | pd_last = pos; |
463 | pd_len += pos->len; |
464 | } |
465 | } else |
466 | first_frag = NULL; |
467 | break; |
468 | |
469 | case SCTP_DATA_LAST_FRAG: |
470 | if (first_frag && (ctsn == next_tsn)) |
471 | goto found; |
472 | else |
473 | first_frag = NULL; |
474 | break; |
475 | } |
476 | } |
477 | |
478 | asoc = ulpq->asoc; |
479 | if (pd_first) { |
480 | /* Make sure we can enter partial deliver. |
481 | * We can trigger partial delivery only if framgent |
482 | * interleave is set, or the socket is not already |
483 | * in partial delivery. |
484 | */ |
485 | if (!sctp_sk(asoc->base.sk)->frag_interleave && |
486 | atomic_read(&sctp_sk(asoc->base.sk)->pd_mode)) |
487 | goto done; |
488 | |
489 | cevent = sctp_skb2event(pd_first); |
490 | pd_point = sctp_sk(asoc->base.sk)->pd_point; |
491 | if (pd_point && pd_point <= pd_len) { |
492 | retval = sctp_make_reassembled_event(sock_net(asoc->base.sk), |
493 | &ulpq->reasm, |
494 | pd_first, |
495 | pd_last); |
496 | if (retval) |
497 | sctp_ulpq_set_pd(ulpq); |
498 | } |
499 | } |
500 | done: |
501 | return retval; |
502 | found: |
503 | retval = sctp_make_reassembled_event(sock_net(ulpq->asoc->base.sk), |
504 | &ulpq->reasm, first_frag, pos); |
505 | if (retval) |
506 | retval->msg_flags |= MSG_EOR; |
507 | goto done; |
508 | } |
509 | |
510 | /* Retrieve the next set of fragments of a partial message. */ |
511 | static struct sctp_ulpevent *sctp_ulpq_retrieve_partial(struct sctp_ulpq *ulpq) |
512 | { |
513 | struct sk_buff *pos, *last_frag, *first_frag; |
514 | struct sctp_ulpevent *cevent; |
515 | __u32 ctsn, next_tsn; |
516 | int is_last; |
517 | struct sctp_ulpevent *retval; |
518 | |
519 | /* The chunks are held in the reasm queue sorted by TSN. |
520 | * Walk through the queue sequentially and look for the first |
521 | * sequence of fragmented chunks. |
522 | */ |
523 | |
524 | if (skb_queue_empty(&ulpq->reasm)) |
525 | return NULL; |
526 | |
527 | last_frag = first_frag = NULL; |
528 | retval = NULL; |
529 | next_tsn = 0; |
530 | is_last = 0; |
531 | |
532 | skb_queue_walk(&ulpq->reasm, pos) { |
533 | cevent = sctp_skb2event(pos); |
534 | ctsn = cevent->tsn; |
535 | |
536 | switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) { |
537 | case SCTP_DATA_FIRST_FRAG: |
538 | if (!first_frag) |
539 | return NULL; |
540 | goto done; |
541 | case SCTP_DATA_MIDDLE_FRAG: |
542 | if (!first_frag) { |
543 | first_frag = pos; |
544 | next_tsn = ctsn + 1; |
545 | last_frag = pos; |
546 | } else if (next_tsn == ctsn) { |
547 | next_tsn++; |
548 | last_frag = pos; |
549 | } else |
550 | goto done; |
551 | break; |
552 | case SCTP_DATA_LAST_FRAG: |
553 | if (!first_frag) |
554 | first_frag = pos; |
555 | else if (ctsn != next_tsn) |
556 | goto done; |
557 | last_frag = pos; |
558 | is_last = 1; |
559 | goto done; |
560 | default: |
561 | return NULL; |
562 | } |
563 | } |
564 | |
565 | /* We have the reassembled event. There is no need to look |
566 | * further. |
567 | */ |
568 | done: |
569 | retval = sctp_make_reassembled_event(sock_net(ulpq->asoc->base.sk), |
570 | &ulpq->reasm, first_frag, last_frag); |
571 | if (retval && is_last) |
572 | retval->msg_flags |= MSG_EOR; |
573 | |
574 | return retval; |
575 | } |
576 | |
577 | |
578 | /* Helper function to reassemble chunks. Hold chunks on the reasm queue that |
579 | * need reassembling. |
580 | */ |
581 | static struct sctp_ulpevent *sctp_ulpq_reasm(struct sctp_ulpq *ulpq, |
582 | struct sctp_ulpevent *event) |
583 | { |
584 | struct sctp_ulpevent *retval = NULL; |
585 | |
586 | /* Check if this is part of a fragmented message. */ |
587 | if (SCTP_DATA_NOT_FRAG == (event->msg_flags & SCTP_DATA_FRAG_MASK)) { |
588 | event->msg_flags |= MSG_EOR; |
589 | return event; |
590 | } |
591 | |
592 | sctp_ulpq_store_reasm(ulpq, event); |
593 | if (!ulpq->pd_mode) |
594 | retval = sctp_ulpq_retrieve_reassembled(ulpq); |
595 | else { |
596 | __u32 ctsn, ctsnap; |
597 | |
598 | /* Do not even bother unless this is the next tsn to |
599 | * be delivered. |
600 | */ |
601 | ctsn = event->tsn; |
602 | ctsnap = sctp_tsnmap_get_ctsn(&ulpq->asoc->peer.tsn_map); |
603 | if (TSN_lte(ctsn, ctsnap)) |
604 | retval = sctp_ulpq_retrieve_partial(ulpq); |
605 | } |
606 | |
607 | return retval; |
608 | } |
609 | |
610 | /* Retrieve the first part (sequential fragments) for partial delivery. */ |
611 | static struct sctp_ulpevent *sctp_ulpq_retrieve_first(struct sctp_ulpq *ulpq) |
612 | { |
613 | struct sk_buff *pos, *last_frag, *first_frag; |
614 | struct sctp_ulpevent *cevent; |
615 | __u32 ctsn, next_tsn; |
616 | struct sctp_ulpevent *retval; |
617 | |
618 | /* The chunks are held in the reasm queue sorted by TSN. |
619 | * Walk through the queue sequentially and look for a sequence of |
620 | * fragmented chunks that start a datagram. |
621 | */ |
622 | |
623 | if (skb_queue_empty(&ulpq->reasm)) |
624 | return NULL; |
625 | |
626 | last_frag = first_frag = NULL; |
627 | retval = NULL; |
628 | next_tsn = 0; |
629 | |
630 | skb_queue_walk(&ulpq->reasm, pos) { |
631 | cevent = sctp_skb2event(pos); |
632 | ctsn = cevent->tsn; |
633 | |
634 | switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) { |
635 | case SCTP_DATA_FIRST_FRAG: |
636 | if (!first_frag) { |
637 | first_frag = pos; |
638 | next_tsn = ctsn + 1; |
639 | last_frag = pos; |
640 | } else |
641 | goto done; |
642 | break; |
643 | |
644 | case SCTP_DATA_MIDDLE_FRAG: |
645 | if (!first_frag) |
646 | return NULL; |
647 | if (ctsn == next_tsn) { |
648 | next_tsn++; |
649 | last_frag = pos; |
650 | } else |
651 | goto done; |
652 | break; |
653 | |
654 | case SCTP_DATA_LAST_FRAG: |
655 | if (!first_frag) |
656 | return NULL; |
657 | else |
658 | goto done; |
659 | break; |
660 | |
661 | default: |
662 | return NULL; |
663 | } |
664 | } |
665 | |
666 | /* We have the reassembled event. There is no need to look |
667 | * further. |
668 | */ |
669 | done: |
670 | retval = sctp_make_reassembled_event(sock_net(ulpq->asoc->base.sk), |
671 | &ulpq->reasm, first_frag, last_frag); |
672 | return retval; |
673 | } |
674 | |
675 | /* |
676 | * Flush out stale fragments from the reassembly queue when processing |
677 | * a Forward TSN. |
678 | * |
679 | * RFC 3758, Section 3.6 |
680 | * |
681 | * After receiving and processing a FORWARD TSN, the data receiver MUST |
682 | * take cautions in updating its re-assembly queue. The receiver MUST |
683 | * remove any partially reassembled message, which is still missing one |
684 | * or more TSNs earlier than or equal to the new cumulative TSN point. |
685 | * In the event that the receiver has invoked the partial delivery API, |
686 | * a notification SHOULD also be generated to inform the upper layer API |
687 | * that the message being partially delivered will NOT be completed. |
688 | */ |
689 | void sctp_ulpq_reasm_flushtsn(struct sctp_ulpq *ulpq, __u32 fwd_tsn) |
690 | { |
691 | struct sk_buff *pos, *tmp; |
692 | struct sctp_ulpevent *event; |
693 | __u32 tsn; |
694 | |
695 | if (skb_queue_empty(&ulpq->reasm)) |
696 | return; |
697 | |
698 | skb_queue_walk_safe(&ulpq->reasm, pos, tmp) { |
699 | event = sctp_skb2event(pos); |
700 | tsn = event->tsn; |
701 | |
702 | /* Since the entire message must be abandoned by the |
703 | * sender (item A3 in Section 3.5, RFC 3758), we can |
704 | * free all fragments on the list that are less then |
705 | * or equal to ctsn_point |
706 | */ |
707 | if (TSN_lte(tsn, fwd_tsn)) { |
708 | __skb_unlink(pos, &ulpq->reasm); |
709 | sctp_ulpevent_free(event); |
710 | } else |
711 | break; |
712 | } |
713 | } |
714 | |
715 | /* |
716 | * Drain the reassembly queue. If we just cleared parted delivery, it |
717 | * is possible that the reassembly queue will contain already reassembled |
718 | * messages. Retrieve any such messages and give them to the user. |
719 | */ |
720 | static void sctp_ulpq_reasm_drain(struct sctp_ulpq *ulpq) |
721 | { |
722 | struct sctp_ulpevent *event = NULL; |
723 | struct sk_buff_head temp; |
724 | |
725 | if (skb_queue_empty(&ulpq->reasm)) |
726 | return; |
727 | |
728 | while ((event = sctp_ulpq_retrieve_reassembled(ulpq)) != NULL) { |
729 | /* Do ordering if needed. */ |
730 | if ((event) && (event->msg_flags & MSG_EOR)) { |
731 | skb_queue_head_init(&temp); |
732 | __skb_queue_tail(&temp, sctp_event2skb(event)); |
733 | |
734 | event = sctp_ulpq_order(ulpq, event); |
735 | } |
736 | |
737 | /* Send event to the ULP. 'event' is the |
738 | * sctp_ulpevent for very first SKB on the temp' list. |
739 | */ |
740 | if (event) |
741 | sctp_ulpq_tail_event(ulpq, event); |
742 | } |
743 | } |
744 | |
745 | |
746 | /* Helper function to gather skbs that have possibly become |
747 | * ordered by an an incoming chunk. |
748 | */ |
749 | static void sctp_ulpq_retrieve_ordered(struct sctp_ulpq *ulpq, |
750 | struct sctp_ulpevent *event) |
751 | { |
752 | struct sk_buff_head *event_list; |
753 | struct sk_buff *pos, *tmp; |
754 | struct sctp_ulpevent *cevent; |
755 | struct sctp_stream *in; |
756 | __u16 sid, csid, cssn; |
757 | |
758 | sid = event->stream; |
759 | in = &ulpq->asoc->ssnmap->in; |
760 | |
761 | event_list = (struct sk_buff_head *) sctp_event2skb(event)->prev; |
762 | |
763 | /* We are holding the chunks by stream, by SSN. */ |
764 | sctp_skb_for_each(pos, &ulpq->lobby, tmp) { |
765 | cevent = (struct sctp_ulpevent *) pos->cb; |
766 | csid = cevent->stream; |
767 | cssn = cevent->ssn; |
768 | |
769 | /* Have we gone too far? */ |
770 | if (csid > sid) |
771 | break; |
772 | |
773 | /* Have we not gone far enough? */ |
774 | if (csid < sid) |
775 | continue; |
776 | |
777 | if (cssn != sctp_ssn_peek(in, sid)) |
778 | break; |
779 | |
780 | /* Found it, so mark in the ssnmap. */ |
781 | sctp_ssn_next(in, sid); |
782 | |
783 | __skb_unlink(pos, &ulpq->lobby); |
784 | |
785 | /* Attach all gathered skbs to the event. */ |
786 | __skb_queue_tail(event_list, pos); |
787 | } |
788 | } |
789 | |
790 | /* Helper function to store chunks needing ordering. */ |
791 | static void sctp_ulpq_store_ordered(struct sctp_ulpq *ulpq, |
792 | struct sctp_ulpevent *event) |
793 | { |
794 | struct sk_buff *pos; |
795 | struct sctp_ulpevent *cevent; |
796 | __u16 sid, csid; |
797 | __u16 ssn, cssn; |
798 | |
799 | pos = skb_peek_tail(&ulpq->lobby); |
800 | if (!pos) { |
801 | __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event)); |
802 | return; |
803 | } |
804 | |
805 | sid = event->stream; |
806 | ssn = event->ssn; |
807 | |
808 | cevent = (struct sctp_ulpevent *) pos->cb; |
809 | csid = cevent->stream; |
810 | cssn = cevent->ssn; |
811 | if (sid > csid) { |
812 | __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event)); |
813 | return; |
814 | } |
815 | |
816 | if ((sid == csid) && SSN_lt(cssn, ssn)) { |
817 | __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event)); |
818 | return; |
819 | } |
820 | |
821 | /* Find the right place in this list. We store them by |
822 | * stream ID and then by SSN. |
823 | */ |
824 | skb_queue_walk(&ulpq->lobby, pos) { |
825 | cevent = (struct sctp_ulpevent *) pos->cb; |
826 | csid = cevent->stream; |
827 | cssn = cevent->ssn; |
828 | |
829 | if (csid > sid) |
830 | break; |
831 | if (csid == sid && SSN_lt(ssn, cssn)) |
832 | break; |
833 | } |
834 | |
835 | |
836 | /* Insert before pos. */ |
837 | __skb_queue_before(&ulpq->lobby, pos, sctp_event2skb(event)); |
838 | } |
839 | |
840 | static struct sctp_ulpevent *sctp_ulpq_order(struct sctp_ulpq *ulpq, |
841 | struct sctp_ulpevent *event) |
842 | { |
843 | __u16 sid, ssn; |
844 | struct sctp_stream *in; |
845 | |
846 | /* Check if this message needs ordering. */ |
847 | if (SCTP_DATA_UNORDERED & event->msg_flags) |
848 | return event; |
849 | |
850 | /* Note: The stream ID must be verified before this routine. */ |
851 | sid = event->stream; |
852 | ssn = event->ssn; |
853 | in = &ulpq->asoc->ssnmap->in; |
854 | |
855 | /* Is this the expected SSN for this stream ID? */ |
856 | if (ssn != sctp_ssn_peek(in, sid)) { |
857 | /* We've received something out of order, so find where it |
858 | * needs to be placed. We order by stream and then by SSN. |
859 | */ |
860 | sctp_ulpq_store_ordered(ulpq, event); |
861 | return NULL; |
862 | } |
863 | |
864 | /* Mark that the next chunk has been found. */ |
865 | sctp_ssn_next(in, sid); |
866 | |
867 | /* Go find any other chunks that were waiting for |
868 | * ordering. |
869 | */ |
870 | sctp_ulpq_retrieve_ordered(ulpq, event); |
871 | |
872 | return event; |
873 | } |
874 | |
875 | /* Helper function to gather skbs that have possibly become |
876 | * ordered by forward tsn skipping their dependencies. |
877 | */ |
878 | static void sctp_ulpq_reap_ordered(struct sctp_ulpq *ulpq, __u16 sid) |
879 | { |
880 | struct sk_buff *pos, *tmp; |
881 | struct sctp_ulpevent *cevent; |
882 | struct sctp_ulpevent *event; |
883 | struct sctp_stream *in; |
884 | struct sk_buff_head temp; |
885 | struct sk_buff_head *lobby = &ulpq->lobby; |
886 | __u16 csid, cssn; |
887 | |
888 | in = &ulpq->asoc->ssnmap->in; |
889 | |
890 | /* We are holding the chunks by stream, by SSN. */ |
891 | skb_queue_head_init(&temp); |
892 | event = NULL; |
893 | sctp_skb_for_each(pos, lobby, tmp) { |
894 | cevent = (struct sctp_ulpevent *) pos->cb; |
895 | csid = cevent->stream; |
896 | cssn = cevent->ssn; |
897 | |
898 | /* Have we gone too far? */ |
899 | if (csid > sid) |
900 | break; |
901 | |
902 | /* Have we not gone far enough? */ |
903 | if (csid < sid) |
904 | continue; |
905 | |
906 | /* see if this ssn has been marked by skipping */ |
907 | if (!SSN_lt(cssn, sctp_ssn_peek(in, csid))) |
908 | break; |
909 | |
910 | __skb_unlink(pos, lobby); |
911 | if (!event) |
912 | /* Create a temporary list to collect chunks on. */ |
913 | event = sctp_skb2event(pos); |
914 | |
915 | /* Attach all gathered skbs to the event. */ |
916 | __skb_queue_tail(&temp, pos); |
917 | } |
918 | |
919 | /* If we didn't reap any data, see if the next expected SSN |
920 | * is next on the queue and if so, use that. |
921 | */ |
922 | if (event == NULL && pos != (struct sk_buff *)lobby) { |
923 | cevent = (struct sctp_ulpevent *) pos->cb; |
924 | csid = cevent->stream; |
925 | cssn = cevent->ssn; |
926 | |
927 | if (csid == sid && cssn == sctp_ssn_peek(in, csid)) { |
928 | sctp_ssn_next(in, csid); |
929 | __skb_unlink(pos, lobby); |
930 | __skb_queue_tail(&temp, pos); |
931 | event = sctp_skb2event(pos); |
932 | } |
933 | } |
934 | |
935 | /* Send event to the ULP. 'event' is the sctp_ulpevent for |
936 | * very first SKB on the 'temp' list. |
937 | */ |
938 | if (event) { |
939 | /* see if we have more ordered that we can deliver */ |
940 | sctp_ulpq_retrieve_ordered(ulpq, event); |
941 | sctp_ulpq_tail_event(ulpq, event); |
942 | } |
943 | } |
944 | |
945 | /* Skip over an SSN. This is used during the processing of |
946 | * Forwared TSN chunk to skip over the abandoned ordered data |
947 | */ |
948 | void sctp_ulpq_skip(struct sctp_ulpq *ulpq, __u16 sid, __u16 ssn) |
949 | { |
950 | struct sctp_stream *in; |
951 | |
952 | /* Note: The stream ID must be verified before this routine. */ |
953 | in = &ulpq->asoc->ssnmap->in; |
954 | |
955 | /* Is this an old SSN? If so ignore. */ |
956 | if (SSN_lt(ssn, sctp_ssn_peek(in, sid))) |
957 | return; |
958 | |
959 | /* Mark that we are no longer expecting this SSN or lower. */ |
960 | sctp_ssn_skip(in, sid, ssn); |
961 | |
962 | /* Go find any other chunks that were waiting for |
963 | * ordering and deliver them if needed. |
964 | */ |
965 | sctp_ulpq_reap_ordered(ulpq, sid); |
966 | } |
967 | |
968 | static __u16 sctp_ulpq_renege_list(struct sctp_ulpq *ulpq, |
969 | struct sk_buff_head *list, __u16 needed) |
970 | { |
971 | __u16 freed = 0; |
972 | __u32 tsn, last_tsn; |
973 | struct sk_buff *skb, *flist, *last; |
974 | struct sctp_ulpevent *event; |
975 | struct sctp_tsnmap *tsnmap; |
976 | |
977 | tsnmap = &ulpq->asoc->peer.tsn_map; |
978 | |
979 | while ((skb = skb_peek_tail(list)) != NULL) { |
980 | event = sctp_skb2event(skb); |
981 | tsn = event->tsn; |
982 | |
983 | /* Don't renege below the Cumulative TSN ACK Point. */ |
984 | if (TSN_lte(tsn, sctp_tsnmap_get_ctsn(tsnmap))) |
985 | break; |
986 | |
987 | /* Events in ordering queue may have multiple fragments |
988 | * corresponding to additional TSNs. Sum the total |
989 | * freed space; find the last TSN. |
990 | */ |
991 | freed += skb_headlen(skb); |
992 | flist = skb_shinfo(skb)->frag_list; |
993 | for (last = flist; flist; flist = flist->next) { |
994 | last = flist; |
995 | freed += skb_headlen(last); |
996 | } |
997 | if (last) |
998 | last_tsn = sctp_skb2event(last)->tsn; |
999 | else |
1000 | last_tsn = tsn; |
1001 | |
1002 | /* Unlink the event, then renege all applicable TSNs. */ |
1003 | __skb_unlink(skb, list); |
1004 | sctp_ulpevent_free(event); |
1005 | while (TSN_lte(tsn, last_tsn)) { |
1006 | sctp_tsnmap_renege(tsnmap, tsn); |
1007 | tsn++; |
1008 | } |
1009 | if (freed >= needed) |
1010 | return freed; |
1011 | } |
1012 | |
1013 | return freed; |
1014 | } |
1015 | |
1016 | /* Renege 'needed' bytes from the ordering queue. */ |
1017 | static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed) |
1018 | { |
1019 | return sctp_ulpq_renege_list(ulpq, &ulpq->lobby, needed); |
1020 | } |
1021 | |
1022 | /* Renege 'needed' bytes from the reassembly queue. */ |
1023 | static __u16 sctp_ulpq_renege_frags(struct sctp_ulpq *ulpq, __u16 needed) |
1024 | { |
1025 | return sctp_ulpq_renege_list(ulpq, &ulpq->reasm, needed); |
1026 | } |
1027 | |
1028 | /* Partial deliver the first message as there is pressure on rwnd. */ |
1029 | void sctp_ulpq_partial_delivery(struct sctp_ulpq *ulpq, |
1030 | gfp_t gfp) |
1031 | { |
1032 | struct sctp_ulpevent *event; |
1033 | struct sctp_association *asoc; |
1034 | struct sctp_sock *sp; |
1035 | __u32 ctsn; |
1036 | struct sk_buff *skb; |
1037 | |
1038 | asoc = ulpq->asoc; |
1039 | sp = sctp_sk(asoc->base.sk); |
1040 | |
1041 | /* If the association is already in Partial Delivery mode |
1042 | * we have nothing to do. |
1043 | */ |
1044 | if (ulpq->pd_mode) |
1045 | return; |
1046 | |
1047 | /* Data must be at or below the Cumulative TSN ACK Point to |
1048 | * start partial delivery. |
1049 | */ |
1050 | skb = skb_peek(&asoc->ulpq.reasm); |
1051 | if (skb != NULL) { |
1052 | ctsn = sctp_skb2event(skb)->tsn; |
1053 | if (!TSN_lte(ctsn, sctp_tsnmap_get_ctsn(&asoc->peer.tsn_map))) |
1054 | return; |
1055 | } |
1056 | |
1057 | /* If the user enabled fragment interleave socket option, |
1058 | * multiple associations can enter partial delivery. |
1059 | * Otherwise, we can only enter partial delivery if the |
1060 | * socket is not in partial deliver mode. |
1061 | */ |
1062 | if (sp->frag_interleave || atomic_read(&sp->pd_mode) == 0) { |
1063 | /* Is partial delivery possible? */ |
1064 | event = sctp_ulpq_retrieve_first(ulpq); |
1065 | /* Send event to the ULP. */ |
1066 | if (event) { |
1067 | sctp_ulpq_tail_event(ulpq, event); |
1068 | sctp_ulpq_set_pd(ulpq); |
1069 | return; |
1070 | } |
1071 | } |
1072 | } |
1073 | |
1074 | /* Renege some packets to make room for an incoming chunk. */ |
1075 | void sctp_ulpq_renege(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk, |
1076 | gfp_t gfp) |
1077 | { |
1078 | struct sctp_association *asoc; |
1079 | __u16 needed, freed; |
1080 | |
1081 | asoc = ulpq->asoc; |
1082 | |
1083 | if (chunk) { |
1084 | needed = ntohs(chunk->chunk_hdr->length); |
1085 | needed -= sizeof(sctp_data_chunk_t); |
1086 | } else |
1087 | needed = SCTP_DEFAULT_MAXWINDOW; |
1088 | |
1089 | freed = 0; |
1090 | |
1091 | if (skb_queue_empty(&asoc->base.sk->sk_receive_queue)) { |
1092 | freed = sctp_ulpq_renege_order(ulpq, needed); |
1093 | if (freed < needed) { |
1094 | freed += sctp_ulpq_renege_frags(ulpq, needed - freed); |
1095 | } |
1096 | } |
1097 | /* If able to free enough room, accept this chunk. */ |
1098 | if (chunk && (freed >= needed)) { |
1099 | int retval; |
1100 | retval = sctp_ulpq_tail_data(ulpq, chunk, gfp); |
1101 | /* |
1102 | * Enter partial delivery if chunk has not been |
1103 | * delivered; otherwise, drain the reassembly queue. |
1104 | */ |
1105 | if (retval <= 0) |
1106 | sctp_ulpq_partial_delivery(ulpq, gfp); |
1107 | else if (retval == 1) |
1108 | sctp_ulpq_reasm_drain(ulpq); |
1109 | } |
1110 | |
1111 | sk_mem_reclaim(asoc->base.sk); |
1112 | } |
1113 | |
1114 | |
1115 | |
1116 | /* Notify the application if an association is aborted and in |
1117 | * partial delivery mode. Send up any pending received messages. |
1118 | */ |
1119 | void sctp_ulpq_abort_pd(struct sctp_ulpq *ulpq, gfp_t gfp) |
1120 | { |
1121 | struct sctp_ulpevent *ev = NULL; |
1122 | struct sock *sk; |
1123 | |
1124 | if (!ulpq->pd_mode) |
1125 | return; |
1126 | |
1127 | sk = ulpq->asoc->base.sk; |
1128 | if (sctp_ulpevent_type_enabled(SCTP_PARTIAL_DELIVERY_EVENT, |
1129 | &sctp_sk(sk)->subscribe)) |
1130 | ev = sctp_ulpevent_make_pdapi(ulpq->asoc, |
1131 | SCTP_PARTIAL_DELIVERY_ABORTED, |
1132 | gfp); |
1133 | if (ev) |
1134 | __skb_queue_tail(&sk->sk_receive_queue, sctp_event2skb(ev)); |
1135 | |
1136 | /* If there is data waiting, send it up the socket now. */ |
1137 | if (sctp_ulpq_clear_pd(ulpq) || ev) |
1138 | sk->sk_data_ready(sk); |
1139 | } |
1140 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9