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
9 * This abstraction carries sctp events to the ULP (sockets).
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)
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.
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.
28 * Please send any bug reports or fixes you make to the
30 * lksctp developers <lksctp-developers@lists.sourceforge.net>
32 * Or submit a bug report through the following website:
33 * http://www.sf.net/projects/lksctp
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>
40 * Any bugs reported given to us we will try to fix... any fixes shared will
41 * be incorporated into the next SCTP release.
44 #include <linux/types.h>
45 #include <linux/skbuff.h>
47 #include <net/sctp/structs.h>
48 #include <net/sctp/sctp.h>
49 #include <net/sctp/sm.h>
51 /* Forward declarations for internal helpers. */
52 static struct sctp_ulpevent * sctp_ulpq_reasm(struct sctp_ulpq *ulpq,
53 struct sctp_ulpevent *);
54 static struct sctp_ulpevent * sctp_ulpq_order(struct sctp_ulpq *,
55 struct sctp_ulpevent *);
57 /* 1st Level Abstractions */
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)
63 memset(ulpq, 0, sizeof(struct sctp_ulpq));
66 skb_queue_head_init(&ulpq->reasm);
67 skb_queue_head_init(&ulpq->lobby);
75 /* Flush the reassembly and ordering queues. */
76 static void sctp_ulpq_flush(struct sctp_ulpq *ulpq)
79 struct sctp_ulpevent *event;
81 while ((skb = __skb_dequeue(&ulpq->lobby)) != NULL) {
82 event = sctp_skb2event(skb);
83 sctp_ulpevent_free(event);
86 while ((skb = __skb_dequeue(&ulpq->reasm)) != NULL) {
87 event = sctp_skb2event(skb);
88 sctp_ulpevent_free(event);
93 /* Dispose of a ulpqueue. */
94 void sctp_ulpq_free(struct sctp_ulpq *ulpq)
96 sctp_ulpq_flush(ulpq);
101 /* Process an incoming DATA chunk. */
102 int sctp_ulpq_tail_data(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk,
105 struct sk_buff_head temp;
106 sctp_data_chunk_t *hdr;
107 struct sctp_ulpevent *event;
109 hdr = (sctp_data_chunk_t *) chunk->chunk_hdr;
111 /* Create an event from the incoming chunk. */
112 event = sctp_ulpevent_make_rcvmsg(chunk->asoc, chunk, gfp);
116 /* Do reassembly if needed. */
117 event = sctp_ulpq_reasm(ulpq, event);
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));
125 event = sctp_ulpq_order(ulpq, event);
128 /* Send event to the ULP. 'event' is the sctp_ulpevent for
129 * very first SKB on the 'temp' list.
132 sctp_ulpq_tail_event(ulpq, event);
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.
141 int sctp_clear_pd(struct sock *sk)
143 struct sctp_sock *sp = sctp_sk(sk);
146 if (!skb_queue_empty(&sp->pd_lobby)) {
147 struct list_head *list;
148 sctp_skb_list_tail(&sp->pd_lobby, &sk->sk_receive_queue);
149 list = (struct list_head *)&sctp_sk(sk)->pd_lobby;
150 INIT_LIST_HEAD(list);
156 /* Clear the pd_mode and restart any pending messages waiting for delivery. */
157 static int sctp_ulpq_clear_pd(struct sctp_ulpq *ulpq)
160 return sctp_clear_pd(ulpq->asoc->base.sk);
163 /* If the SKB of 'event' is on a list, it is the first such member
166 int sctp_ulpq_tail_event(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event)
168 struct sock *sk = ulpq->asoc->base.sk;
169 struct sk_buff_head *queue, *skb_list;
170 struct sk_buff *skb = sctp_event2skb(event);
173 skb_list = (struct sk_buff_head *) skb->prev;
175 /* If the socket is just going to throw this away, do not
176 * even try to deliver it.
178 if (sock_flag(sk, SOCK_DEAD) || (sk->sk_shutdown & RCV_SHUTDOWN))
181 /* Check if the user wishes to receive this event. */
182 if (!sctp_ulpevent_is_enabled(event, &sctp_sk(sk)->subscribe))
185 /* If we are in partial delivery mode, post to the lobby until
186 * partial delivery is cleared, unless, of course _this_ is
187 * the association the cause of the partial delivery.
190 if (!sctp_sk(sk)->pd_mode) {
191 queue = &sk->sk_receive_queue;
192 } else if (ulpq->pd_mode) {
193 if (event->msg_flags & MSG_NOTIFICATION)
194 queue = &sctp_sk(sk)->pd_lobby;
196 clear_pd = event->msg_flags & MSG_EOR;
197 queue = &sk->sk_receive_queue;
200 queue = &sctp_sk(sk)->pd_lobby;
203 /* If we are harvesting multiple skbs they will be
204 * collected on a list.
207 sctp_skb_list_tail(skb_list, queue);
209 __skb_queue_tail(queue, skb);
211 /* Did we just complete partial delivery and need to get
212 * rolling again? Move pending data to the receive
216 sctp_ulpq_clear_pd(ulpq);
218 if (queue == &sk->sk_receive_queue)
219 sk->sk_data_ready(sk, 0);
224 sctp_queue_purge_ulpevents(skb_list);
226 sctp_ulpevent_free(event);
231 /* 2nd Level Abstractions */
233 /* Helper function to store chunks that need to be reassembled. */
234 static inline void sctp_ulpq_store_reasm(struct sctp_ulpq *ulpq,
235 struct sctp_ulpevent *event)
238 struct sctp_ulpevent *cevent;
243 /* See if it belongs at the end. */
244 pos = skb_peek_tail(&ulpq->reasm);
246 __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event));
250 /* Short circuit just dropping it at the end. */
251 cevent = sctp_skb2event(pos);
253 if (TSN_lt(ctsn, tsn)) {
254 __skb_queue_tail(&ulpq->reasm, sctp_event2skb(event));
258 /* Find the right place in this list. We store them by TSN. */
259 skb_queue_walk(&ulpq->reasm, pos) {
260 cevent = sctp_skb2event(pos);
263 if (TSN_lt(tsn, ctsn))
267 /* Insert before pos. */
268 __skb_insert(sctp_event2skb(event), pos->prev, pos, &ulpq->reasm);
272 /* Helper function to return an event corresponding to the reassembled
274 * This routine creates a re-assembled skb given the first and last skb's
275 * as stored in the reassembly queue. The skb's may be non-linear if the sctp
276 * payload was fragmented on the way and ip had to reassemble them.
277 * We add the rest of skb's to the first skb's fraglist.
279 static struct sctp_ulpevent *sctp_make_reassembled_event(struct sk_buff_head *queue, struct sk_buff *f_frag, struct sk_buff *l_frag)
282 struct sk_buff *new = NULL;
283 struct sctp_ulpevent *event;
284 struct sk_buff *pnext, *last;
285 struct sk_buff *list = skb_shinfo(f_frag)->frag_list;
287 /* Store the pointer to the 2nd skb */
288 if (f_frag == l_frag)
293 /* Get the last skb in the f_frag's frag_list if present. */
294 for (last = list; list; last = list, list = list->next);
296 /* Add the list of remaining fragments to the first fragments
302 if (skb_cloned(f_frag)) {
303 /* This is a cloned skb, we can't just modify
304 * the frag_list. We need a new skb to do that.
305 * Instead of calling skb_unshare(), we'll do it
306 * ourselves since we need to delay the free.
308 new = skb_copy(f_frag, GFP_ATOMIC);
310 return NULL; /* try again later */
312 sctp_skb_set_owner_r(new, f_frag->sk);
314 skb_shinfo(new)->frag_list = pos;
316 skb_shinfo(f_frag)->frag_list = pos;
319 /* Remove the first fragment from the reassembly queue. */
320 __skb_unlink(f_frag, queue);
322 /* if we did unshare, then free the old skb and re-assign */
332 /* Update the len and data_len fields of the first fragment. */
333 f_frag->len += pos->len;
334 f_frag->data_len += pos->len;
336 /* Remove the fragment from the reassembly queue. */
337 __skb_unlink(pos, queue);
339 /* Break if we have reached the last fragment. */
346 event = sctp_skb2event(f_frag);
347 SCTP_INC_STATS(SCTP_MIB_REASMUSRMSGS);
353 /* Helper function to check if an incoming chunk has filled up the last
354 * missing fragment in a SCTP datagram and return the corresponding event.
356 static inline struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_ulpq *ulpq)
359 struct sctp_ulpevent *cevent;
360 struct sk_buff *first_frag = NULL;
361 __u32 ctsn, next_tsn;
362 struct sctp_ulpevent *retval = NULL;
364 /* Initialized to 0 just to avoid compiler warning message. Will
365 * never be used with this value. It is referenced only after it
366 * is set when we find the first fragment of a message.
370 /* The chunks are held in the reasm queue sorted by TSN.
371 * Walk through the queue sequentially and look for a sequence of
372 * fragmented chunks that complete a datagram.
373 * 'first_frag' and next_tsn are reset when we find a chunk which
374 * is the first fragment of a datagram. Once these 2 fields are set
375 * we expect to find the remaining middle fragments and the last
376 * fragment in order. If not, first_frag is reset to NULL and we
377 * start the next pass when we find another first fragment.
379 skb_queue_walk(&ulpq->reasm, pos) {
380 cevent = sctp_skb2event(pos);
383 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
384 case SCTP_DATA_FIRST_FRAG:
389 case SCTP_DATA_MIDDLE_FRAG:
390 if ((first_frag) && (ctsn == next_tsn))
396 case SCTP_DATA_LAST_FRAG:
397 if (first_frag && (ctsn == next_tsn))
408 retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, pos);
410 retval->msg_flags |= MSG_EOR;
414 /* Retrieve the next set of fragments of a partial message. */
415 static inline struct sctp_ulpevent *sctp_ulpq_retrieve_partial(struct sctp_ulpq *ulpq)
417 struct sk_buff *pos, *last_frag, *first_frag;
418 struct sctp_ulpevent *cevent;
419 __u32 ctsn, next_tsn;
421 struct sctp_ulpevent *retval;
423 /* The chunks are held in the reasm queue sorted by TSN.
424 * Walk through the queue sequentially and look for the first
425 * sequence of fragmented chunks.
428 if (skb_queue_empty(&ulpq->reasm))
431 last_frag = first_frag = NULL;
436 skb_queue_walk(&ulpq->reasm, pos) {
437 cevent = sctp_skb2event(pos);
440 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
441 case SCTP_DATA_MIDDLE_FRAG:
446 } else if (next_tsn == ctsn)
451 case SCTP_DATA_LAST_FRAG:
454 else if (ctsn != next_tsn)
464 /* We have the reassembled event. There is no need to look
468 retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, last_frag);
469 if (retval && is_last)
470 retval->msg_flags |= MSG_EOR;
476 /* Helper function to reassemble chunks. Hold chunks on the reasm queue that
479 static struct sctp_ulpevent *sctp_ulpq_reasm(struct sctp_ulpq *ulpq,
480 struct sctp_ulpevent *event)
482 struct sctp_ulpevent *retval = NULL;
484 /* Check if this is part of a fragmented message. */
485 if (SCTP_DATA_NOT_FRAG == (event->msg_flags & SCTP_DATA_FRAG_MASK)) {
486 event->msg_flags |= MSG_EOR;
490 sctp_ulpq_store_reasm(ulpq, event);
492 retval = sctp_ulpq_retrieve_reassembled(ulpq);
496 /* Do not even bother unless this is the next tsn to
500 ctsnap = sctp_tsnmap_get_ctsn(&ulpq->asoc->peer.tsn_map);
501 if (TSN_lte(ctsn, ctsnap))
502 retval = sctp_ulpq_retrieve_partial(ulpq);
508 /* Retrieve the first part (sequential fragments) for partial delivery. */
509 static inline struct sctp_ulpevent *sctp_ulpq_retrieve_first(struct sctp_ulpq *ulpq)
511 struct sk_buff *pos, *last_frag, *first_frag;
512 struct sctp_ulpevent *cevent;
513 __u32 ctsn, next_tsn;
514 struct sctp_ulpevent *retval;
516 /* The chunks are held in the reasm queue sorted by TSN.
517 * Walk through the queue sequentially and look for a sequence of
518 * fragmented chunks that start a datagram.
521 if (skb_queue_empty(&ulpq->reasm))
524 last_frag = first_frag = NULL;
528 skb_queue_walk(&ulpq->reasm, pos) {
529 cevent = sctp_skb2event(pos);
532 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
533 case SCTP_DATA_FIRST_FRAG:
542 case SCTP_DATA_MIDDLE_FRAG:
545 if (ctsn == next_tsn) {
556 /* We have the reassembled event. There is no need to look
560 retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, last_frag);
564 /* Helper function to gather skbs that have possibly become
565 * ordered by an an incoming chunk.
567 static inline void sctp_ulpq_retrieve_ordered(struct sctp_ulpq *ulpq,
568 struct sctp_ulpevent *event)
570 struct sk_buff_head *event_list;
571 struct sk_buff *pos, *tmp;
572 struct sctp_ulpevent *cevent;
573 struct sctp_stream *in;
579 in = &ulpq->asoc->ssnmap->in;
581 event_list = (struct sk_buff_head *) sctp_event2skb(event)->prev;
583 /* We are holding the chunks by stream, by SSN. */
584 sctp_skb_for_each(pos, &ulpq->lobby, tmp) {
585 cevent = (struct sctp_ulpevent *) pos->cb;
586 csid = cevent->stream;
589 /* Have we gone too far? */
593 /* Have we not gone far enough? */
597 if (cssn != sctp_ssn_peek(in, sid))
600 /* Found it, so mark in the ssnmap. */
601 sctp_ssn_next(in, sid);
603 __skb_unlink(pos, &ulpq->lobby);
605 /* Attach all gathered skbs to the event. */
606 __skb_queue_tail(event_list, pos);
610 /* Helper function to store chunks needing ordering. */
611 static inline void sctp_ulpq_store_ordered(struct sctp_ulpq *ulpq,
612 struct sctp_ulpevent *event)
615 struct sctp_ulpevent *cevent;
619 pos = skb_peek_tail(&ulpq->lobby);
621 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
628 cevent = (struct sctp_ulpevent *) pos->cb;
629 csid = cevent->stream;
632 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
636 if ((sid == csid) && SSN_lt(cssn, ssn)) {
637 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
641 /* Find the right place in this list. We store them by
642 * stream ID and then by SSN.
644 skb_queue_walk(&ulpq->lobby, pos) {
645 cevent = (struct sctp_ulpevent *) pos->cb;
646 csid = cevent->stream;
651 if (csid == sid && SSN_lt(ssn, cssn))
656 /* Insert before pos. */
657 __skb_insert(sctp_event2skb(event), pos->prev, pos, &ulpq->lobby);
661 static struct sctp_ulpevent *sctp_ulpq_order(struct sctp_ulpq *ulpq,
662 struct sctp_ulpevent *event)
665 struct sctp_stream *in;
667 /* Check if this message needs ordering. */
668 if (SCTP_DATA_UNORDERED & event->msg_flags)
671 /* Note: The stream ID must be verified before this routine. */
674 in = &ulpq->asoc->ssnmap->in;
676 /* Is this the expected SSN for this stream ID? */
677 if (ssn != sctp_ssn_peek(in, sid)) {
678 /* We've received something out of order, so find where it
679 * needs to be placed. We order by stream and then by SSN.
681 sctp_ulpq_store_ordered(ulpq, event);
685 /* Mark that the next chunk has been found. */
686 sctp_ssn_next(in, sid);
688 /* Go find any other chunks that were waiting for
691 sctp_ulpq_retrieve_ordered(ulpq, event);
696 /* Helper function to gather skbs that have possibly become
697 * ordered by forward tsn skipping their dependencies.
699 static inline void sctp_ulpq_reap_ordered(struct sctp_ulpq *ulpq)
701 struct sk_buff *pos, *tmp;
702 struct sctp_ulpevent *cevent;
703 struct sctp_ulpevent *event;
704 struct sctp_stream *in;
705 struct sk_buff_head temp;
708 in = &ulpq->asoc->ssnmap->in;
710 /* We are holding the chunks by stream, by SSN. */
711 skb_queue_head_init(&temp);
713 sctp_skb_for_each(pos, &ulpq->lobby, tmp) {
714 cevent = (struct sctp_ulpevent *) pos->cb;
715 csid = cevent->stream;
718 if (cssn != sctp_ssn_peek(in, csid))
721 /* Found it, so mark in the ssnmap. */
722 sctp_ssn_next(in, csid);
724 __skb_unlink(pos, &ulpq->lobby);
726 /* Create a temporary list to collect chunks on. */
727 event = sctp_skb2event(pos);
728 __skb_queue_tail(&temp, sctp_event2skb(event));
730 /* Attach all gathered skbs to the event. */
731 __skb_queue_tail(&temp, pos);
735 /* Send event to the ULP. 'event' is the sctp_ulpevent for
736 * very first SKB on the 'temp' list.
739 sctp_ulpq_tail_event(ulpq, event);
742 /* Skip over an SSN. */
743 void sctp_ulpq_skip(struct sctp_ulpq *ulpq, __u16 sid, __u16 ssn)
745 struct sctp_stream *in;
747 /* Note: The stream ID must be verified before this routine. */
748 in = &ulpq->asoc->ssnmap->in;
750 /* Is this an old SSN? If so ignore. */
751 if (SSN_lt(ssn, sctp_ssn_peek(in, sid)))
754 /* Mark that we are no longer expecting this SSN or lower. */
755 sctp_ssn_skip(in, sid, ssn);
757 /* Go find any other chunks that were waiting for
758 * ordering and deliver them if needed.
760 sctp_ulpq_reap_ordered(ulpq);
764 /* Renege 'needed' bytes from the ordering queue. */
765 static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed)
770 struct sctp_ulpevent *event;
771 struct sctp_tsnmap *tsnmap;
773 tsnmap = &ulpq->asoc->peer.tsn_map;
775 while ((skb = __skb_dequeue_tail(&ulpq->lobby)) != NULL) {
776 freed += skb_headlen(skb);
777 event = sctp_skb2event(skb);
780 sctp_ulpevent_free(event);
781 sctp_tsnmap_renege(tsnmap, tsn);
789 /* Renege 'needed' bytes from the reassembly queue. */
790 static __u16 sctp_ulpq_renege_frags(struct sctp_ulpq *ulpq, __u16 needed)
795 struct sctp_ulpevent *event;
796 struct sctp_tsnmap *tsnmap;
798 tsnmap = &ulpq->asoc->peer.tsn_map;
800 /* Walk backwards through the list, reneges the newest tsns. */
801 while ((skb = __skb_dequeue_tail(&ulpq->reasm)) != NULL) {
802 freed += skb_headlen(skb);
803 event = sctp_skb2event(skb);
806 sctp_ulpevent_free(event);
807 sctp_tsnmap_renege(tsnmap, tsn);
815 /* Partial deliver the first message as there is pressure on rwnd. */
816 void sctp_ulpq_partial_delivery(struct sctp_ulpq *ulpq,
817 struct sctp_chunk *chunk,
820 struct sctp_ulpevent *event;
821 struct sctp_association *asoc;
825 /* Are we already in partial delivery mode? */
826 if (!sctp_sk(asoc->base.sk)->pd_mode) {
828 /* Is partial delivery possible? */
829 event = sctp_ulpq_retrieve_first(ulpq);
830 /* Send event to the ULP. */
832 sctp_ulpq_tail_event(ulpq, event);
833 sctp_sk(asoc->base.sk)->pd_mode = 1;
840 /* Renege some packets to make room for an incoming chunk. */
841 void sctp_ulpq_renege(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk,
844 struct sctp_association *asoc;
850 needed = ntohs(chunk->chunk_hdr->length);
851 needed -= sizeof(sctp_data_chunk_t);
853 needed = SCTP_DEFAULT_MAXWINDOW;
857 if (skb_queue_empty(&asoc->base.sk->sk_receive_queue)) {
858 freed = sctp_ulpq_renege_order(ulpq, needed);
859 if (freed < needed) {
860 freed += sctp_ulpq_renege_frags(ulpq, needed - freed);
863 /* If able to free enough room, accept this chunk. */
864 if (chunk && (freed >= needed)) {
866 tsn = ntohl(chunk->subh.data_hdr->tsn);
867 sctp_tsnmap_mark(&asoc->peer.tsn_map, tsn);
868 sctp_ulpq_tail_data(ulpq, chunk, gfp);
870 sctp_ulpq_partial_delivery(ulpq, chunk, gfp);
878 /* Notify the application if an association is aborted and in
879 * partial delivery mode. Send up any pending received messages.
881 void sctp_ulpq_abort_pd(struct sctp_ulpq *ulpq, gfp_t gfp)
883 struct sctp_ulpevent *ev = NULL;
889 sk = ulpq->asoc->base.sk;
890 if (sctp_ulpevent_type_enabled(SCTP_PARTIAL_DELIVERY_EVENT,
891 &sctp_sk(sk)->subscribe))
892 ev = sctp_ulpevent_make_pdapi(ulpq->asoc,
893 SCTP_PARTIAL_DELIVERY_ABORTED,
896 __skb_queue_tail(&sk->sk_receive_queue, sctp_event2skb(ev));
898 /* If there is data waiting, send it up the socket now. */
899 if (sctp_ulpq_clear_pd(ulpq) || ev)
900 sk->sk_data_ready(sk, 0);