Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* 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. */ | |
52 | static struct sctp_ulpevent * sctp_ulpq_reasm(struct sctp_ulpq *ulpq, | |
8728b834 | 53 | struct sctp_ulpevent *); |
1da177e4 | 54 | static struct sctp_ulpevent * sctp_ulpq_order(struct sctp_ulpq *, |
8728b834 | 55 | struct sctp_ulpevent *); |
1da177e4 LT |
56 | |
57 | /* 1st Level Abstractions */ | |
58 | ||
59 | /* Initialize a ULP queue from a block of memory. */ | |
60 | struct 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. */ | |
0b58a811 | 76 | void sctp_ulpq_flush(struct sctp_ulpq *ulpq) |
1da177e4 LT |
77 | { |
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. */ | |
94 | void 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. */ | |
102 | int sctp_ulpq_tail_data(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk, | |
dd0fc66f | 103 | gfp_t gfp) |
1da177e4 LT |
104 | { |
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 | ||
8728b834 DM |
128 | /* Send event to the ULP. 'event' is the sctp_ulpevent for |
129 | * very first SKB on the 'temp' list. | |
130 | */ | |
1da177e4 LT |
131 | 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 | */ | |
b6e1331f | 141 | int sctp_clear_pd(struct sock *sk, struct sctp_association *asoc) |
1da177e4 LT |
142 | { |
143 | struct sctp_sock *sp = sctp_sk(sk); | |
144 | ||
b6e1331f VY |
145 | 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 | } | |
1da177e4 | 175 | } |
b6e1331f | 176 | |
1da177e4 LT |
177 | return 0; |
178 | } | |
179 | ||
d49d91d7 VY |
180 | /* Set the pd_mode on the socket and ulpq */ |
181 | static void sctp_ulpq_set_pd(struct sctp_ulpq *ulpq) | |
182 | { | |
183 | struct sctp_sock *sp = sctp_sk(ulpq->asoc->base.sk); | |
184 | ||
185 | atomic_inc(&sp->pd_mode); | |
186 | ulpq->pd_mode = 1; | |
187 | } | |
188 | ||
1da177e4 LT |
189 | /* Clear the pd_mode and restart any pending messages waiting for delivery. */ |
190 | static int sctp_ulpq_clear_pd(struct sctp_ulpq *ulpq) | |
191 | { | |
192 | ulpq->pd_mode = 0; | |
b6e1331f | 193 | return sctp_clear_pd(ulpq->asoc->base.sk, ulpq->asoc); |
1da177e4 LT |
194 | } |
195 | ||
8728b834 DM |
196 | /* If the SKB of 'event' is on a list, it is the first such member |
197 | * of that list. | |
198 | */ | |
1da177e4 LT |
199 | int sctp_ulpq_tail_event(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event) |
200 | { | |
201 | struct sock *sk = ulpq->asoc->base.sk; | |
8728b834 DM |
202 | struct sk_buff_head *queue, *skb_list; |
203 | struct sk_buff *skb = sctp_event2skb(event); | |
1da177e4 LT |
204 | int clear_pd = 0; |
205 | ||
8728b834 DM |
206 | skb_list = (struct sk_buff_head *) skb->prev; |
207 | ||
1da177e4 LT |
208 | /* If the socket is just going to throw this away, do not |
209 | * even try to deliver it. | |
210 | */ | |
211 | if (sock_flag(sk, SOCK_DEAD) || (sk->sk_shutdown & RCV_SHUTDOWN)) | |
212 | goto out_free; | |
213 | ||
214 | /* Check if the user wishes to receive this event. */ | |
215 | if (!sctp_ulpevent_is_enabled(event, &sctp_sk(sk)->subscribe)) | |
216 | goto out_free; | |
217 | ||
218 | /* If we are in partial delivery mode, post to the lobby until | |
219 | * partial delivery is cleared, unless, of course _this_ is | |
220 | * the association the cause of the partial delivery. | |
221 | */ | |
222 | ||
b6e1331f | 223 | if (atomic_read(&sctp_sk(sk)->pd_mode) == 0) { |
1da177e4 | 224 | queue = &sk->sk_receive_queue; |
b6e1331f VY |
225 | } else { |
226 | if (ulpq->pd_mode) { | |
227 | /* If the association is in partial delivery, we | |
228 | * need to finish delivering the partially processed | |
229 | * packet before passing any other data. This is | |
230 | * because we don't truly support stream interleaving. | |
231 | */ | |
232 | if ((event->msg_flags & MSG_NOTIFICATION) || | |
233 | (SCTP_DATA_NOT_FRAG == | |
234 | (event->msg_flags & SCTP_DATA_FRAG_MASK))) | |
235 | queue = &sctp_sk(sk)->pd_lobby; | |
236 | else { | |
237 | clear_pd = event->msg_flags & MSG_EOR; | |
238 | queue = &sk->sk_receive_queue; | |
239 | } | |
240 | } else { | |
241 | /* | |
242 | * If fragment interleave is enabled, we | |
243 | * can queue this to the recieve queue instead | |
244 | * of the lobby. | |
245 | */ | |
246 | if (sctp_sk(sk)->frag_interleave) | |
247 | queue = &sk->sk_receive_queue; | |
248 | else | |
249 | queue = &sctp_sk(sk)->pd_lobby; | |
1da177e4 | 250 | } |
b6e1331f | 251 | } |
1da177e4 LT |
252 | |
253 | /* If we are harvesting multiple skbs they will be | |
254 | * collected on a list. | |
255 | */ | |
8728b834 DM |
256 | if (skb_list) |
257 | sctp_skb_list_tail(skb_list, queue); | |
1da177e4 | 258 | else |
8728b834 | 259 | __skb_queue_tail(queue, skb); |
1da177e4 LT |
260 | |
261 | /* Did we just complete partial delivery and need to get | |
262 | * rolling again? Move pending data to the receive | |
263 | * queue. | |
264 | */ | |
265 | if (clear_pd) | |
266 | sctp_ulpq_clear_pd(ulpq); | |
267 | ||
268 | if (queue == &sk->sk_receive_queue) | |
269 | sk->sk_data_ready(sk, 0); | |
270 | return 1; | |
271 | ||
272 | out_free: | |
8728b834 DM |
273 | if (skb_list) |
274 | sctp_queue_purge_ulpevents(skb_list); | |
1da177e4 LT |
275 | else |
276 | sctp_ulpevent_free(event); | |
8728b834 | 277 | |
1da177e4 LT |
278 | return 0; |
279 | } | |
280 | ||
281 | /* 2nd Level Abstractions */ | |
282 | ||
283 | /* Helper function to store chunks that need to be reassembled. */ | |
284 | static inline void sctp_ulpq_store_reasm(struct sctp_ulpq *ulpq, | |
285 | struct sctp_ulpevent *event) | |
286 | { | |
287 | struct sk_buff *pos; | |
288 | struct sctp_ulpevent *cevent; | |
289 | __u32 tsn, ctsn; | |
290 | ||
291 | tsn = event->tsn; | |
292 | ||
293 | /* See if it belongs at the end. */ | |
294 | pos = skb_peek_tail(&ulpq->reasm); | |
295 | if (!pos) { | |
296 | __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event)); | |
297 | return; | |
298 | } | |
299 | ||
300 | /* Short circuit just dropping it at the end. */ | |
301 | cevent = sctp_skb2event(pos); | |
302 | ctsn = cevent->tsn; | |
303 | if (TSN_lt(ctsn, tsn)) { | |
304 | __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event)); | |
305 | return; | |
306 | } | |
307 | ||
308 | /* Find the right place in this list. We store them by TSN. */ | |
309 | skb_queue_walk(&ulpq->reasm, pos) { | |
310 | cevent = sctp_skb2event(pos); | |
311 | ctsn = cevent->tsn; | |
312 | ||
313 | if (TSN_lt(tsn, ctsn)) | |
314 | break; | |
315 | } | |
316 | ||
317 | /* Insert before pos. */ | |
318 | __skb_insert(sctp_event2skb(event), pos->prev, pos, &ulpq->reasm); | |
319 | ||
320 | } | |
321 | ||
322 | /* Helper function to return an event corresponding to the reassembled | |
323 | * datagram. | |
324 | * This routine creates a re-assembled skb given the first and last skb's | |
325 | * as stored in the reassembly queue. The skb's may be non-linear if the sctp | |
326 | * payload was fragmented on the way and ip had to reassemble them. | |
327 | * We add the rest of skb's to the first skb's fraglist. | |
328 | */ | |
8728b834 | 329 | static struct sctp_ulpevent *sctp_make_reassembled_event(struct sk_buff_head *queue, struct sk_buff *f_frag, 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. */ | |
344 | for (last = list; list; last = list, list = list->next); | |
345 | ||
346 | /* Add the list of remaining fragments to the first fragments | |
347 | * frag_list. | |
348 | */ | |
349 | if (last) | |
350 | last->next = pos; | |
d808ad9a YH |
351 | else { |
352 | if (skb_cloned(f_frag)) { | |
353 | /* This is a cloned skb, we can't just modify | |
354 | * the frag_list. We need a new skb to do that. | |
355 | * Instead of calling skb_unshare(), we'll do it | |
356 | * ourselves since we need to delay the free. | |
357 | */ | |
358 | new = skb_copy(f_frag, GFP_ATOMIC); | |
359 | if (!new) | |
360 | return NULL; /* try again later */ | |
361 | ||
362 | sctp_skb_set_owner_r(new, f_frag->sk); | |
363 | ||
364 | skb_shinfo(new)->frag_list = pos; | |
365 | } else | |
366 | skb_shinfo(f_frag)->frag_list = pos; | |
367 | } | |
1da177e4 LT |
368 | |
369 | /* Remove the first fragment from the reassembly queue. */ | |
8728b834 | 370 | __skb_unlink(f_frag, queue); |
672e7cca | 371 | |
d808ad9a YH |
372 | /* if we did unshare, then free the old skb and re-assign */ |
373 | if (new) { | |
374 | kfree_skb(f_frag); | |
375 | f_frag = new; | |
376 | } | |
672e7cca | 377 | |
1da177e4 LT |
378 | while (pos) { |
379 | ||
380 | pnext = pos->next; | |
381 | ||
382 | /* Update the len and data_len fields of the first fragment. */ | |
383 | f_frag->len += pos->len; | |
384 | f_frag->data_len += pos->len; | |
385 | ||
386 | /* Remove the fragment from the reassembly queue. */ | |
8728b834 | 387 | __skb_unlink(pos, queue); |
d808ad9a | 388 | |
1da177e4 LT |
389 | /* Break if we have reached the last fragment. */ |
390 | if (pos == l_frag) | |
391 | break; | |
392 | pos->next = pnext; | |
393 | pos = pnext; | |
3ff50b79 | 394 | } |
1da177e4 LT |
395 | |
396 | event = sctp_skb2event(f_frag); | |
397 | SCTP_INC_STATS(SCTP_MIB_REASMUSRMSGS); | |
398 | ||
399 | return event; | |
400 | } | |
401 | ||
402 | ||
403 | /* Helper function to check if an incoming chunk has filled up the last | |
404 | * missing fragment in a SCTP datagram and return the corresponding event. | |
405 | */ | |
406 | static inline struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_ulpq *ulpq) | |
407 | { | |
408 | struct sk_buff *pos; | |
409 | struct sctp_ulpevent *cevent; | |
410 | struct sk_buff *first_frag = NULL; | |
411 | __u32 ctsn, next_tsn; | |
412 | struct sctp_ulpevent *retval = NULL; | |
d49d91d7 VY |
413 | struct sk_buff *pd_first = NULL; |
414 | struct sk_buff *pd_last = NULL; | |
415 | size_t pd_len = 0; | |
416 | struct sctp_association *asoc; | |
417 | u32 pd_point; | |
1da177e4 LT |
418 | |
419 | /* Initialized to 0 just to avoid compiler warning message. Will | |
420 | * never be used with this value. It is referenced only after it | |
421 | * is set when we find the first fragment of a message. | |
422 | */ | |
423 | next_tsn = 0; | |
424 | ||
425 | /* The chunks are held in the reasm queue sorted by TSN. | |
426 | * Walk through the queue sequentially and look for a sequence of | |
427 | * fragmented chunks that complete a datagram. | |
428 | * 'first_frag' and next_tsn are reset when we find a chunk which | |
429 | * is the first fragment of a datagram. Once these 2 fields are set | |
430 | * we expect to find the remaining middle fragments and the last | |
431 | * fragment in order. If not, first_frag is reset to NULL and we | |
432 | * start the next pass when we find another first fragment. | |
d49d91d7 VY |
433 | * |
434 | * There is a potential to do partial delivery if user sets | |
435 | * SCTP_PARTIAL_DELIVERY_POINT option. Lets count some things here | |
436 | * to see if can do PD. | |
1da177e4 LT |
437 | */ |
438 | skb_queue_walk(&ulpq->reasm, pos) { | |
439 | cevent = sctp_skb2event(pos); | |
440 | ctsn = cevent->tsn; | |
441 | ||
442 | switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) { | |
443 | case SCTP_DATA_FIRST_FRAG: | |
d49d91d7 VY |
444 | /* If this "FIRST_FRAG" is the first |
445 | * element in the queue, then count it towards | |
446 | * possible PD. | |
447 | */ | |
448 | if (pos == ulpq->reasm.next) { | |
449 | pd_first = pos; | |
450 | pd_last = pos; | |
451 | pd_len = pos->len; | |
452 | } else { | |
453 | pd_first = NULL; | |
454 | pd_last = NULL; | |
455 | pd_len = 0; | |
456 | } | |
457 | ||
1da177e4 LT |
458 | first_frag = pos; |
459 | next_tsn = ctsn + 1; | |
460 | break; | |
461 | ||
462 | case SCTP_DATA_MIDDLE_FRAG: | |
d49d91d7 | 463 | if ((first_frag) && (ctsn == next_tsn)) { |
1da177e4 | 464 | next_tsn++; |
d49d91d7 VY |
465 | if (pd_first) { |
466 | pd_last = pos; | |
467 | pd_len += pos->len; | |
468 | } | |
469 | } else | |
1da177e4 LT |
470 | first_frag = NULL; |
471 | break; | |
472 | ||
473 | case SCTP_DATA_LAST_FRAG: | |
474 | if (first_frag && (ctsn == next_tsn)) | |
475 | goto found; | |
476 | else | |
477 | first_frag = NULL; | |
478 | break; | |
3ff50b79 | 479 | } |
d49d91d7 VY |
480 | } |
481 | ||
482 | asoc = ulpq->asoc; | |
483 | if (pd_first) { | |
484 | /* Make sure we can enter partial deliver. | |
485 | * We can trigger partial delivery only if framgent | |
486 | * interleave is set, or the socket is not already | |
487 | * in partial delivery. | |
488 | */ | |
489 | if (!sctp_sk(asoc->base.sk)->frag_interleave && | |
490 | atomic_read(&sctp_sk(asoc->base.sk)->pd_mode)) | |
491 | goto done; | |
1da177e4 | 492 | |
d49d91d7 VY |
493 | cevent = sctp_skb2event(pd_first); |
494 | pd_point = sctp_sk(asoc->base.sk)->pd_point; | |
495 | if (pd_point && pd_point <= pd_len) { | |
496 | retval = sctp_make_reassembled_event(&ulpq->reasm, | |
497 | pd_first, | |
498 | pd_last); | |
499 | if (retval) | |
500 | sctp_ulpq_set_pd(ulpq); | |
501 | } | |
1da177e4 LT |
502 | } |
503 | done: | |
504 | return retval; | |
505 | found: | |
8728b834 | 506 | retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, pos); |
1da177e4 LT |
507 | if (retval) |
508 | retval->msg_flags |= MSG_EOR; | |
509 | goto done; | |
510 | } | |
511 | ||
512 | /* Retrieve the next set of fragments of a partial message. */ | |
513 | static inline struct sctp_ulpevent *sctp_ulpq_retrieve_partial(struct sctp_ulpq *ulpq) | |
514 | { | |
515 | struct sk_buff *pos, *last_frag, *first_frag; | |
516 | struct sctp_ulpevent *cevent; | |
517 | __u32 ctsn, next_tsn; | |
518 | int is_last; | |
519 | struct sctp_ulpevent *retval; | |
520 | ||
521 | /* The chunks are held in the reasm queue sorted by TSN. | |
522 | * Walk through the queue sequentially and look for the first | |
523 | * sequence of fragmented chunks. | |
524 | */ | |
525 | ||
526 | if (skb_queue_empty(&ulpq->reasm)) | |
527 | return NULL; | |
528 | ||
529 | last_frag = first_frag = NULL; | |
530 | retval = NULL; | |
531 | next_tsn = 0; | |
532 | is_last = 0; | |
533 | ||
534 | skb_queue_walk(&ulpq->reasm, pos) { | |
535 | cevent = sctp_skb2event(pos); | |
536 | ctsn = cevent->tsn; | |
537 | ||
538 | switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) { | |
539 | case SCTP_DATA_MIDDLE_FRAG: | |
540 | if (!first_frag) { | |
541 | first_frag = pos; | |
542 | next_tsn = ctsn + 1; | |
543 | last_frag = pos; | |
544 | } else if (next_tsn == ctsn) | |
545 | next_tsn++; | |
546 | else | |
547 | goto done; | |
548 | break; | |
549 | case SCTP_DATA_LAST_FRAG: | |
550 | if (!first_frag) | |
551 | first_frag = pos; | |
552 | else if (ctsn != next_tsn) | |
553 | goto done; | |
554 | last_frag = pos; | |
555 | is_last = 1; | |
556 | goto done; | |
557 | default: | |
558 | return NULL; | |
3ff50b79 | 559 | } |
1da177e4 LT |
560 | } |
561 | ||
562 | /* We have the reassembled event. There is no need to look | |
563 | * further. | |
564 | */ | |
565 | done: | |
8728b834 | 566 | retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, last_frag); |
1da177e4 LT |
567 | if (retval && is_last) |
568 | retval->msg_flags |= MSG_EOR; | |
569 | ||
570 | return retval; | |
571 | } | |
572 | ||
573 | ||
574 | /* Helper function to reassemble chunks. Hold chunks on the reasm queue that | |
575 | * need reassembling. | |
576 | */ | |
577 | static struct sctp_ulpevent *sctp_ulpq_reasm(struct sctp_ulpq *ulpq, | |
578 | struct sctp_ulpevent *event) | |
579 | { | |
580 | struct sctp_ulpevent *retval = NULL; | |
581 | ||
582 | /* Check if this is part of a fragmented message. */ | |
583 | if (SCTP_DATA_NOT_FRAG == (event->msg_flags & SCTP_DATA_FRAG_MASK)) { | |
584 | event->msg_flags |= MSG_EOR; | |
585 | return event; | |
586 | } | |
587 | ||
588 | sctp_ulpq_store_reasm(ulpq, event); | |
589 | if (!ulpq->pd_mode) | |
590 | retval = sctp_ulpq_retrieve_reassembled(ulpq); | |
591 | else { | |
592 | __u32 ctsn, ctsnap; | |
593 | ||
594 | /* Do not even bother unless this is the next tsn to | |
595 | * be delivered. | |
596 | */ | |
597 | ctsn = event->tsn; | |
598 | ctsnap = sctp_tsnmap_get_ctsn(&ulpq->asoc->peer.tsn_map); | |
599 | if (TSN_lte(ctsn, ctsnap)) | |
600 | retval = sctp_ulpq_retrieve_partial(ulpq); | |
601 | } | |
602 | ||
603 | return retval; | |
604 | } | |
605 | ||
606 | /* Retrieve the first part (sequential fragments) for partial delivery. */ | |
607 | static inline struct sctp_ulpevent *sctp_ulpq_retrieve_first(struct sctp_ulpq *ulpq) | |
608 | { | |
609 | struct sk_buff *pos, *last_frag, *first_frag; | |
610 | struct sctp_ulpevent *cevent; | |
611 | __u32 ctsn, next_tsn; | |
612 | struct sctp_ulpevent *retval; | |
613 | ||
614 | /* The chunks are held in the reasm queue sorted by TSN. | |
615 | * Walk through the queue sequentially and look for a sequence of | |
616 | * fragmented chunks that start a datagram. | |
617 | */ | |
618 | ||
619 | if (skb_queue_empty(&ulpq->reasm)) | |
620 | return NULL; | |
621 | ||
622 | last_frag = first_frag = NULL; | |
623 | retval = NULL; | |
624 | next_tsn = 0; | |
625 | ||
626 | skb_queue_walk(&ulpq->reasm, pos) { | |
627 | cevent = sctp_skb2event(pos); | |
628 | ctsn = cevent->tsn; | |
629 | ||
630 | switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) { | |
631 | case SCTP_DATA_FIRST_FRAG: | |
632 | if (!first_frag) { | |
633 | first_frag = pos; | |
634 | next_tsn = ctsn + 1; | |
635 | last_frag = pos; | |
636 | } else | |
637 | goto done; | |
638 | break; | |
639 | ||
640 | case SCTP_DATA_MIDDLE_FRAG: | |
641 | if (!first_frag) | |
642 | return NULL; | |
643 | if (ctsn == next_tsn) { | |
644 | next_tsn++; | |
645 | last_frag = pos; | |
646 | } else | |
647 | goto done; | |
648 | break; | |
649 | default: | |
650 | return NULL; | |
3ff50b79 | 651 | } |
1da177e4 LT |
652 | } |
653 | ||
654 | /* We have the reassembled event. There is no need to look | |
655 | * further. | |
656 | */ | |
657 | done: | |
8728b834 | 658 | retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, last_frag); |
1da177e4 LT |
659 | return retval; |
660 | } | |
661 | ||
ea2dfb37 VY |
662 | /* |
663 | * Flush out stale fragments from the reassembly queue when processing | |
664 | * a Forward TSN. | |
665 | * | |
666 | * RFC 3758, Section 3.6 | |
667 | * | |
668 | * After receiving and processing a FORWARD TSN, the data receiver MUST | |
669 | * take cautions in updating its re-assembly queue. The receiver MUST | |
670 | * remove any partially reassembled message, which is still missing one | |
671 | * or more TSNs earlier than or equal to the new cumulative TSN point. | |
672 | * In the event that the receiver has invoked the partial delivery API, | |
673 | * a notification SHOULD also be generated to inform the upper layer API | |
674 | * that the message being partially delivered will NOT be completed. | |
675 | */ | |
676 | void sctp_ulpq_reasm_flushtsn(struct sctp_ulpq *ulpq, __u32 fwd_tsn) | |
677 | { | |
678 | struct sk_buff *pos, *tmp; | |
679 | struct sctp_ulpevent *event; | |
680 | __u32 tsn; | |
681 | ||
682 | if (skb_queue_empty(&ulpq->reasm)) | |
683 | return; | |
684 | ||
685 | skb_queue_walk_safe(&ulpq->reasm, pos, tmp) { | |
686 | event = sctp_skb2event(pos); | |
687 | tsn = event->tsn; | |
688 | ||
689 | /* Since the entire message must be abandoned by the | |
690 | * sender (item A3 in Section 3.5, RFC 3758), we can | |
691 | * free all fragments on the list that are less then | |
692 | * or equal to ctsn_point | |
693 | */ | |
694 | if (TSN_lte(tsn, fwd_tsn)) { | |
695 | __skb_unlink(pos, &ulpq->reasm); | |
696 | sctp_ulpevent_free(event); | |
697 | } else | |
698 | break; | |
699 | } | |
700 | } | |
701 | ||
1da177e4 LT |
702 | /* Helper function to gather skbs that have possibly become |
703 | * ordered by an an incoming chunk. | |
704 | */ | |
705 | static inline void sctp_ulpq_retrieve_ordered(struct sctp_ulpq *ulpq, | |
706 | struct sctp_ulpevent *event) | |
707 | { | |
8728b834 | 708 | struct sk_buff_head *event_list; |
1da177e4 LT |
709 | struct sk_buff *pos, *tmp; |
710 | struct sctp_ulpevent *cevent; | |
711 | struct sctp_stream *in; | |
712 | __u16 sid, csid; | |
713 | __u16 ssn, cssn; | |
714 | ||
715 | sid = event->stream; | |
716 | ssn = event->ssn; | |
717 | in = &ulpq->asoc->ssnmap->in; | |
718 | ||
8728b834 DM |
719 | event_list = (struct sk_buff_head *) sctp_event2skb(event)->prev; |
720 | ||
1da177e4 LT |
721 | /* We are holding the chunks by stream, by SSN. */ |
722 | sctp_skb_for_each(pos, &ulpq->lobby, tmp) { | |
723 | cevent = (struct sctp_ulpevent *) pos->cb; | |
724 | csid = cevent->stream; | |
725 | cssn = cevent->ssn; | |
726 | ||
727 | /* Have we gone too far? */ | |
728 | if (csid > sid) | |
729 | break; | |
730 | ||
731 | /* Have we not gone far enough? */ | |
732 | if (csid < sid) | |
733 | continue; | |
734 | ||
735 | if (cssn != sctp_ssn_peek(in, sid)) | |
736 | break; | |
737 | ||
738 | /* Found it, so mark in the ssnmap. */ | |
739 | sctp_ssn_next(in, sid); | |
740 | ||
8728b834 | 741 | __skb_unlink(pos, &ulpq->lobby); |
1da177e4 LT |
742 | |
743 | /* Attach all gathered skbs to the event. */ | |
8728b834 | 744 | __skb_queue_tail(event_list, pos); |
1da177e4 LT |
745 | } |
746 | } | |
747 | ||
748 | /* Helper function to store chunks needing ordering. */ | |
749 | static inline void sctp_ulpq_store_ordered(struct sctp_ulpq *ulpq, | |
750 | struct sctp_ulpevent *event) | |
751 | { | |
752 | struct sk_buff *pos; | |
753 | struct sctp_ulpevent *cevent; | |
754 | __u16 sid, csid; | |
755 | __u16 ssn, cssn; | |
756 | ||
757 | pos = skb_peek_tail(&ulpq->lobby); | |
758 | if (!pos) { | |
759 | __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event)); | |
760 | return; | |
761 | } | |
762 | ||
763 | sid = event->stream; | |
764 | ssn = event->ssn; | |
d808ad9a | 765 | |
1da177e4 LT |
766 | cevent = (struct sctp_ulpevent *) pos->cb; |
767 | csid = cevent->stream; | |
768 | cssn = cevent->ssn; | |
769 | if (sid > csid) { | |
770 | __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event)); | |
771 | return; | |
772 | } | |
773 | ||
774 | if ((sid == csid) && SSN_lt(cssn, ssn)) { | |
775 | __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event)); | |
776 | return; | |
777 | } | |
778 | ||
779 | /* Find the right place in this list. We store them by | |
780 | * stream ID and then by SSN. | |
781 | */ | |
782 | skb_queue_walk(&ulpq->lobby, pos) { | |
783 | cevent = (struct sctp_ulpevent *) pos->cb; | |
784 | csid = cevent->stream; | |
785 | cssn = cevent->ssn; | |
786 | ||
787 | if (csid > sid) | |
788 | break; | |
789 | if (csid == sid && SSN_lt(ssn, cssn)) | |
790 | break; | |
791 | } | |
792 | ||
793 | ||
794 | /* Insert before pos. */ | |
795 | __skb_insert(sctp_event2skb(event), pos->prev, pos, &ulpq->lobby); | |
796 | ||
797 | } | |
798 | ||
799 | static struct sctp_ulpevent *sctp_ulpq_order(struct sctp_ulpq *ulpq, | |
8728b834 | 800 | struct sctp_ulpevent *event) |
1da177e4 LT |
801 | { |
802 | __u16 sid, ssn; | |
803 | struct sctp_stream *in; | |
804 | ||
805 | /* Check if this message needs ordering. */ | |
806 | if (SCTP_DATA_UNORDERED & event->msg_flags) | |
807 | return event; | |
808 | ||
809 | /* Note: The stream ID must be verified before this routine. */ | |
810 | sid = event->stream; | |
811 | ssn = event->ssn; | |
812 | in = &ulpq->asoc->ssnmap->in; | |
813 | ||
814 | /* Is this the expected SSN for this stream ID? */ | |
815 | if (ssn != sctp_ssn_peek(in, sid)) { | |
816 | /* We've received something out of order, so find where it | |
817 | * needs to be placed. We order by stream and then by SSN. | |
818 | */ | |
819 | sctp_ulpq_store_ordered(ulpq, event); | |
820 | return NULL; | |
821 | } | |
822 | ||
823 | /* Mark that the next chunk has been found. */ | |
824 | sctp_ssn_next(in, sid); | |
825 | ||
826 | /* Go find any other chunks that were waiting for | |
827 | * ordering. | |
828 | */ | |
829 | sctp_ulpq_retrieve_ordered(ulpq, event); | |
830 | ||
831 | return event; | |
832 | } | |
833 | ||
834 | /* Helper function to gather skbs that have possibly become | |
835 | * ordered by forward tsn skipping their dependencies. | |
836 | */ | |
ea2dfb37 | 837 | static inline void sctp_ulpq_reap_ordered(struct sctp_ulpq *ulpq, __u16 sid) |
1da177e4 LT |
838 | { |
839 | struct sk_buff *pos, *tmp; | |
840 | struct sctp_ulpevent *cevent; | |
8728b834 | 841 | struct sctp_ulpevent *event; |
1da177e4 LT |
842 | struct sctp_stream *in; |
843 | struct sk_buff_head temp; | |
844 | __u16 csid, cssn; | |
845 | ||
846 | in = &ulpq->asoc->ssnmap->in; | |
847 | ||
848 | /* We are holding the chunks by stream, by SSN. */ | |
8728b834 DM |
849 | skb_queue_head_init(&temp); |
850 | event = NULL; | |
1da177e4 LT |
851 | sctp_skb_for_each(pos, &ulpq->lobby, tmp) { |
852 | cevent = (struct sctp_ulpevent *) pos->cb; | |
853 | csid = cevent->stream; | |
854 | cssn = cevent->ssn; | |
855 | ||
ea2dfb37 VY |
856 | /* Have we gone too far? */ |
857 | if (csid > sid) | |
1da177e4 LT |
858 | break; |
859 | ||
ea2dfb37 VY |
860 | /* Have we not gone far enough? */ |
861 | if (csid < sid) | |
862 | continue; | |
863 | ||
864 | /* see if this ssn has been marked by skipping */ | |
865 | if (!SSN_lt(cssn, sctp_ssn_peek(in, csid))) | |
866 | break; | |
1da177e4 | 867 | |
8728b834 | 868 | __skb_unlink(pos, &ulpq->lobby); |
ea2dfb37 | 869 | if (!event) |
1da177e4 LT |
870 | /* Create a temporary list to collect chunks on. */ |
871 | event = sctp_skb2event(pos); | |
ea2dfb37 VY |
872 | |
873 | /* Attach all gathered skbs to the event. */ | |
874 | __skb_queue_tail(&temp, pos); | |
1da177e4 LT |
875 | } |
876 | ||
8728b834 DM |
877 | /* Send event to the ULP. 'event' is the sctp_ulpevent for |
878 | * very first SKB on the 'temp' list. | |
879 | */ | |
ea2dfb37 VY |
880 | if (event) { |
881 | /* see if we have more ordered that we can deliver */ | |
882 | sctp_ulpq_retrieve_ordered(ulpq, event); | |
1da177e4 | 883 | sctp_ulpq_tail_event(ulpq, event); |
ea2dfb37 | 884 | } |
1da177e4 LT |
885 | } |
886 | ||
ea2dfb37 VY |
887 | /* Skip over an SSN. This is used during the processing of |
888 | * Forwared TSN chunk to skip over the abandoned ordered data | |
889 | */ | |
1da177e4 LT |
890 | void sctp_ulpq_skip(struct sctp_ulpq *ulpq, __u16 sid, __u16 ssn) |
891 | { | |
892 | struct sctp_stream *in; | |
893 | ||
894 | /* Note: The stream ID must be verified before this routine. */ | |
895 | in = &ulpq->asoc->ssnmap->in; | |
896 | ||
897 | /* Is this an old SSN? If so ignore. */ | |
898 | if (SSN_lt(ssn, sctp_ssn_peek(in, sid))) | |
899 | return; | |
900 | ||
901 | /* Mark that we are no longer expecting this SSN or lower. */ | |
902 | sctp_ssn_skip(in, sid, ssn); | |
903 | ||
904 | /* Go find any other chunks that were waiting for | |
d808ad9a | 905 | * ordering and deliver them if needed. |
1da177e4 | 906 | */ |
ea2dfb37 | 907 | sctp_ulpq_reap_ordered(ulpq, sid); |
1da177e4 LT |
908 | return; |
909 | } | |
910 | ||
911 | /* Renege 'needed' bytes from the ordering queue. */ | |
912 | static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed) | |
913 | { | |
914 | __u16 freed = 0; | |
915 | __u32 tsn; | |
916 | struct sk_buff *skb; | |
917 | struct sctp_ulpevent *event; | |
918 | struct sctp_tsnmap *tsnmap; | |
919 | ||
920 | tsnmap = &ulpq->asoc->peer.tsn_map; | |
921 | ||
922 | while ((skb = __skb_dequeue_tail(&ulpq->lobby)) != NULL) { | |
923 | freed += skb_headlen(skb); | |
924 | event = sctp_skb2event(skb); | |
925 | tsn = event->tsn; | |
926 | ||
927 | sctp_ulpevent_free(event); | |
928 | sctp_tsnmap_renege(tsnmap, tsn); | |
929 | if (freed >= needed) | |
930 | return freed; | |
931 | } | |
932 | ||
933 | return freed; | |
934 | } | |
935 | ||
936 | /* Renege 'needed' bytes from the reassembly queue. */ | |
937 | static __u16 sctp_ulpq_renege_frags(struct sctp_ulpq *ulpq, __u16 needed) | |
938 | { | |
939 | __u16 freed = 0; | |
940 | __u32 tsn; | |
941 | struct sk_buff *skb; | |
942 | struct sctp_ulpevent *event; | |
943 | struct sctp_tsnmap *tsnmap; | |
944 | ||
945 | tsnmap = &ulpq->asoc->peer.tsn_map; | |
946 | ||
947 | /* Walk backwards through the list, reneges the newest tsns. */ | |
948 | while ((skb = __skb_dequeue_tail(&ulpq->reasm)) != NULL) { | |
949 | freed += skb_headlen(skb); | |
950 | event = sctp_skb2event(skb); | |
951 | tsn = event->tsn; | |
952 | ||
953 | sctp_ulpevent_free(event); | |
954 | sctp_tsnmap_renege(tsnmap, tsn); | |
955 | if (freed >= needed) | |
956 | return freed; | |
957 | } | |
958 | ||
959 | return freed; | |
960 | } | |
961 | ||
962 | /* Partial deliver the first message as there is pressure on rwnd. */ | |
963 | void sctp_ulpq_partial_delivery(struct sctp_ulpq *ulpq, | |
3182cd84 | 964 | struct sctp_chunk *chunk, |
dd0fc66f | 965 | gfp_t gfp) |
1da177e4 LT |
966 | { |
967 | struct sctp_ulpevent *event; | |
968 | struct sctp_association *asoc; | |
b6e1331f | 969 | struct sctp_sock *sp; |
1da177e4 LT |
970 | |
971 | asoc = ulpq->asoc; | |
b6e1331f | 972 | sp = sctp_sk(asoc->base.sk); |
1da177e4 | 973 | |
b6e1331f VY |
974 | /* If the association is already in Partial Delivery mode |
975 | * we have noting to do. | |
976 | */ | |
977 | if (ulpq->pd_mode) | |
978 | return; | |
1da177e4 | 979 | |
b6e1331f VY |
980 | /* If the user enabled fragment interleave socket option, |
981 | * multiple associations can enter partial delivery. | |
982 | * Otherwise, we can only enter partial delivery if the | |
983 | * socket is not in partial deliver mode. | |
984 | */ | |
985 | if (sp->frag_interleave || atomic_read(&sp->pd_mode) == 0) { | |
1da177e4 LT |
986 | /* Is partial delivery possible? */ |
987 | event = sctp_ulpq_retrieve_first(ulpq); | |
988 | /* Send event to the ULP. */ | |
989 | if (event) { | |
990 | sctp_ulpq_tail_event(ulpq, event); | |
d49d91d7 | 991 | sctp_ulpq_set_pd(ulpq); |
1da177e4 LT |
992 | return; |
993 | } | |
994 | } | |
995 | } | |
996 | ||
997 | /* Renege some packets to make room for an incoming chunk. */ | |
998 | void sctp_ulpq_renege(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk, | |
dd0fc66f | 999 | gfp_t gfp) |
1da177e4 LT |
1000 | { |
1001 | struct sctp_association *asoc; | |
1002 | __u16 needed, freed; | |
1003 | ||
1004 | asoc = ulpq->asoc; | |
1005 | ||
1006 | if (chunk) { | |
1007 | needed = ntohs(chunk->chunk_hdr->length); | |
1008 | needed -= sizeof(sctp_data_chunk_t); | |
d808ad9a | 1009 | } else |
1da177e4 LT |
1010 | needed = SCTP_DEFAULT_MAXWINDOW; |
1011 | ||
1012 | freed = 0; | |
1013 | ||
1014 | if (skb_queue_empty(&asoc->base.sk->sk_receive_queue)) { | |
1015 | freed = sctp_ulpq_renege_order(ulpq, needed); | |
1016 | if (freed < needed) { | |
1017 | freed += sctp_ulpq_renege_frags(ulpq, needed - freed); | |
1018 | } | |
1019 | } | |
1020 | /* If able to free enough room, accept this chunk. */ | |
1021 | if (chunk && (freed >= needed)) { | |
1022 | __u32 tsn; | |
1023 | tsn = ntohl(chunk->subh.data_hdr->tsn); | |
1024 | sctp_tsnmap_mark(&asoc->peer.tsn_map, tsn); | |
1025 | sctp_ulpq_tail_data(ulpq, chunk, gfp); | |
d808ad9a | 1026 | |
1da177e4 LT |
1027 | sctp_ulpq_partial_delivery(ulpq, chunk, gfp); |
1028 | } | |
1029 | ||
4d93df0a | 1030 | sk_stream_mem_reclaim(asoc->base.sk); |
1da177e4 LT |
1031 | return; |
1032 | } | |
1033 | ||
1034 | ||
1035 | ||
1036 | /* Notify the application if an association is aborted and in | |
1037 | * partial delivery mode. Send up any pending received messages. | |
1038 | */ | |
dd0fc66f | 1039 | void sctp_ulpq_abort_pd(struct sctp_ulpq *ulpq, gfp_t gfp) |
1da177e4 LT |
1040 | { |
1041 | struct sctp_ulpevent *ev = NULL; | |
1042 | struct sock *sk; | |
1043 | ||
1044 | if (!ulpq->pd_mode) | |
1045 | return; | |
1046 | ||
1047 | sk = ulpq->asoc->base.sk; | |
1048 | if (sctp_ulpevent_type_enabled(SCTP_PARTIAL_DELIVERY_EVENT, | |
1049 | &sctp_sk(sk)->subscribe)) | |
1050 | ev = sctp_ulpevent_make_pdapi(ulpq->asoc, | |
1051 | SCTP_PARTIAL_DELIVERY_ABORTED, | |
1052 | gfp); | |
1053 | if (ev) | |
1054 | __skb_queue_tail(&sk->sk_receive_queue, sctp_event2skb(ev)); | |
1055 | ||
1056 | /* If there is data waiting, send it up the socket now. */ | |
1057 | if (sctp_ulpq_clear_pd(ulpq) || ev) | |
1058 | sk->sk_data_ready(sk, 0); | |
1059 | } |