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,
103 unsigned int __nocast gfp)
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 sctp_ulpevent *event;
283 struct sk_buff *pnext, *last;
284 struct sk_buff *list = skb_shinfo(f_frag)->frag_list;
286 /* Store the pointer to the 2nd skb */
287 if (f_frag == l_frag)
292 /* Get the last skb in the f_frag's frag_list if present. */
293 for (last = list; list; last = list, list = list->next);
295 /* Add the list of remaining fragments to the first fragments
301 skb_shinfo(f_frag)->frag_list = pos;
303 /* Remove the first fragment from the reassembly queue. */
304 __skb_unlink(f_frag, queue);
309 /* Update the len and data_len fields of the first fragment. */
310 f_frag->len += pos->len;
311 f_frag->data_len += pos->len;
313 /* Remove the fragment from the reassembly queue. */
314 __skb_unlink(pos, queue);
316 /* Break if we have reached the last fragment. */
323 event = sctp_skb2event(f_frag);
324 SCTP_INC_STATS(SCTP_MIB_REASMUSRMSGS);
330 /* Helper function to check if an incoming chunk has filled up the last
331 * missing fragment in a SCTP datagram and return the corresponding event.
333 static inline struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_ulpq *ulpq)
336 struct sctp_ulpevent *cevent;
337 struct sk_buff *first_frag = NULL;
338 __u32 ctsn, next_tsn;
339 struct sctp_ulpevent *retval = NULL;
341 /* Initialized to 0 just to avoid compiler warning message. Will
342 * never be used with this value. It is referenced only after it
343 * is set when we find the first fragment of a message.
347 /* The chunks are held in the reasm queue sorted by TSN.
348 * Walk through the queue sequentially and look for a sequence of
349 * fragmented chunks that complete a datagram.
350 * 'first_frag' and next_tsn are reset when we find a chunk which
351 * is the first fragment of a datagram. Once these 2 fields are set
352 * we expect to find the remaining middle fragments and the last
353 * fragment in order. If not, first_frag is reset to NULL and we
354 * start the next pass when we find another first fragment.
356 skb_queue_walk(&ulpq->reasm, pos) {
357 cevent = sctp_skb2event(pos);
360 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
361 case SCTP_DATA_FIRST_FRAG:
366 case SCTP_DATA_MIDDLE_FRAG:
367 if ((first_frag) && (ctsn == next_tsn))
373 case SCTP_DATA_LAST_FRAG:
374 if (first_frag && (ctsn == next_tsn))
385 retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, pos);
387 retval->msg_flags |= MSG_EOR;
391 /* Retrieve the next set of fragments of a partial message. */
392 static inline struct sctp_ulpevent *sctp_ulpq_retrieve_partial(struct sctp_ulpq *ulpq)
394 struct sk_buff *pos, *last_frag, *first_frag;
395 struct sctp_ulpevent *cevent;
396 __u32 ctsn, next_tsn;
398 struct sctp_ulpevent *retval;
400 /* The chunks are held in the reasm queue sorted by TSN.
401 * Walk through the queue sequentially and look for the first
402 * sequence of fragmented chunks.
405 if (skb_queue_empty(&ulpq->reasm))
408 last_frag = first_frag = NULL;
413 skb_queue_walk(&ulpq->reasm, pos) {
414 cevent = sctp_skb2event(pos);
417 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
418 case SCTP_DATA_MIDDLE_FRAG:
423 } else if (next_tsn == ctsn)
428 case SCTP_DATA_LAST_FRAG:
431 else if (ctsn != next_tsn)
441 /* We have the reassembled event. There is no need to look
445 retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, last_frag);
446 if (retval && is_last)
447 retval->msg_flags |= MSG_EOR;
453 /* Helper function to reassemble chunks. Hold chunks on the reasm queue that
456 static struct sctp_ulpevent *sctp_ulpq_reasm(struct sctp_ulpq *ulpq,
457 struct sctp_ulpevent *event)
459 struct sctp_ulpevent *retval = NULL;
461 /* Check if this is part of a fragmented message. */
462 if (SCTP_DATA_NOT_FRAG == (event->msg_flags & SCTP_DATA_FRAG_MASK)) {
463 event->msg_flags |= MSG_EOR;
467 sctp_ulpq_store_reasm(ulpq, event);
469 retval = sctp_ulpq_retrieve_reassembled(ulpq);
473 /* Do not even bother unless this is the next tsn to
477 ctsnap = sctp_tsnmap_get_ctsn(&ulpq->asoc->peer.tsn_map);
478 if (TSN_lte(ctsn, ctsnap))
479 retval = sctp_ulpq_retrieve_partial(ulpq);
485 /* Retrieve the first part (sequential fragments) for partial delivery. */
486 static inline struct sctp_ulpevent *sctp_ulpq_retrieve_first(struct sctp_ulpq *ulpq)
488 struct sk_buff *pos, *last_frag, *first_frag;
489 struct sctp_ulpevent *cevent;
490 __u32 ctsn, next_tsn;
491 struct sctp_ulpevent *retval;
493 /* The chunks are held in the reasm queue sorted by TSN.
494 * Walk through the queue sequentially and look for a sequence of
495 * fragmented chunks that start a datagram.
498 if (skb_queue_empty(&ulpq->reasm))
501 last_frag = first_frag = NULL;
505 skb_queue_walk(&ulpq->reasm, pos) {
506 cevent = sctp_skb2event(pos);
509 switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) {
510 case SCTP_DATA_FIRST_FRAG:
519 case SCTP_DATA_MIDDLE_FRAG:
522 if (ctsn == next_tsn) {
533 /* We have the reassembled event. There is no need to look
537 retval = sctp_make_reassembled_event(&ulpq->reasm, first_frag, last_frag);
541 /* Helper function to gather skbs that have possibly become
542 * ordered by an an incoming chunk.
544 static inline void sctp_ulpq_retrieve_ordered(struct sctp_ulpq *ulpq,
545 struct sctp_ulpevent *event)
547 struct sk_buff_head *event_list;
548 struct sk_buff *pos, *tmp;
549 struct sctp_ulpevent *cevent;
550 struct sctp_stream *in;
556 in = &ulpq->asoc->ssnmap->in;
558 event_list = (struct sk_buff_head *) sctp_event2skb(event)->prev;
560 /* We are holding the chunks by stream, by SSN. */
561 sctp_skb_for_each(pos, &ulpq->lobby, tmp) {
562 cevent = (struct sctp_ulpevent *) pos->cb;
563 csid = cevent->stream;
566 /* Have we gone too far? */
570 /* Have we not gone far enough? */
574 if (cssn != sctp_ssn_peek(in, sid))
577 /* Found it, so mark in the ssnmap. */
578 sctp_ssn_next(in, sid);
580 __skb_unlink(pos, &ulpq->lobby);
582 /* Attach all gathered skbs to the event. */
583 __skb_queue_tail(event_list, pos);
587 /* Helper function to store chunks needing ordering. */
588 static inline void sctp_ulpq_store_ordered(struct sctp_ulpq *ulpq,
589 struct sctp_ulpevent *event)
592 struct sctp_ulpevent *cevent;
596 pos = skb_peek_tail(&ulpq->lobby);
598 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
605 cevent = (struct sctp_ulpevent *) pos->cb;
606 csid = cevent->stream;
609 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
613 if ((sid == csid) && SSN_lt(cssn, ssn)) {
614 __skb_queue_tail(&ulpq->lobby, sctp_event2skb(event));
618 /* Find the right place in this list. We store them by
619 * stream ID and then by SSN.
621 skb_queue_walk(&ulpq->lobby, pos) {
622 cevent = (struct sctp_ulpevent *) pos->cb;
623 csid = cevent->stream;
628 if (csid == sid && SSN_lt(ssn, cssn))
633 /* Insert before pos. */
634 __skb_insert(sctp_event2skb(event), pos->prev, pos, &ulpq->lobby);
638 static struct sctp_ulpevent *sctp_ulpq_order(struct sctp_ulpq *ulpq,
639 struct sctp_ulpevent *event)
642 struct sctp_stream *in;
644 /* Check if this message needs ordering. */
645 if (SCTP_DATA_UNORDERED & event->msg_flags)
648 /* Note: The stream ID must be verified before this routine. */
651 in = &ulpq->asoc->ssnmap->in;
653 /* Is this the expected SSN for this stream ID? */
654 if (ssn != sctp_ssn_peek(in, sid)) {
655 /* We've received something out of order, so find where it
656 * needs to be placed. We order by stream and then by SSN.
658 sctp_ulpq_store_ordered(ulpq, event);
662 /* Mark that the next chunk has been found. */
663 sctp_ssn_next(in, sid);
665 /* Go find any other chunks that were waiting for
668 sctp_ulpq_retrieve_ordered(ulpq, event);
673 /* Helper function to gather skbs that have possibly become
674 * ordered by forward tsn skipping their dependencies.
676 static inline void sctp_ulpq_reap_ordered(struct sctp_ulpq *ulpq)
678 struct sk_buff *pos, *tmp;
679 struct sctp_ulpevent *cevent;
680 struct sctp_ulpevent *event;
681 struct sctp_stream *in;
682 struct sk_buff_head temp;
685 in = &ulpq->asoc->ssnmap->in;
687 /* We are holding the chunks by stream, by SSN. */
688 skb_queue_head_init(&temp);
690 sctp_skb_for_each(pos, &ulpq->lobby, tmp) {
691 cevent = (struct sctp_ulpevent *) pos->cb;
692 csid = cevent->stream;
695 if (cssn != sctp_ssn_peek(in, csid))
698 /* Found it, so mark in the ssnmap. */
699 sctp_ssn_next(in, csid);
701 __skb_unlink(pos, &ulpq->lobby);
703 /* Create a temporary list to collect chunks on. */
704 event = sctp_skb2event(pos);
705 __skb_queue_tail(&temp, sctp_event2skb(event));
707 /* Attach all gathered skbs to the event. */
708 __skb_queue_tail(&temp, pos);
712 /* Send event to the ULP. 'event' is the sctp_ulpevent for
713 * very first SKB on the 'temp' list.
716 sctp_ulpq_tail_event(ulpq, event);
719 /* Skip over an SSN. */
720 void sctp_ulpq_skip(struct sctp_ulpq *ulpq, __u16 sid, __u16 ssn)
722 struct sctp_stream *in;
724 /* Note: The stream ID must be verified before this routine. */
725 in = &ulpq->asoc->ssnmap->in;
727 /* Is this an old SSN? If so ignore. */
728 if (SSN_lt(ssn, sctp_ssn_peek(in, sid)))
731 /* Mark that we are no longer expecting this SSN or lower. */
732 sctp_ssn_skip(in, sid, ssn);
734 /* Go find any other chunks that were waiting for
735 * ordering and deliver them if needed.
737 sctp_ulpq_reap_ordered(ulpq);
741 /* Renege 'needed' bytes from the ordering queue. */
742 static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed)
747 struct sctp_ulpevent *event;
748 struct sctp_tsnmap *tsnmap;
750 tsnmap = &ulpq->asoc->peer.tsn_map;
752 while ((skb = __skb_dequeue_tail(&ulpq->lobby)) != NULL) {
753 freed += skb_headlen(skb);
754 event = sctp_skb2event(skb);
757 sctp_ulpevent_free(event);
758 sctp_tsnmap_renege(tsnmap, tsn);
766 /* Renege 'needed' bytes from the reassembly queue. */
767 static __u16 sctp_ulpq_renege_frags(struct sctp_ulpq *ulpq, __u16 needed)
772 struct sctp_ulpevent *event;
773 struct sctp_tsnmap *tsnmap;
775 tsnmap = &ulpq->asoc->peer.tsn_map;
777 /* Walk backwards through the list, reneges the newest tsns. */
778 while ((skb = __skb_dequeue_tail(&ulpq->reasm)) != NULL) {
779 freed += skb_headlen(skb);
780 event = sctp_skb2event(skb);
783 sctp_ulpevent_free(event);
784 sctp_tsnmap_renege(tsnmap, tsn);
792 /* Partial deliver the first message as there is pressure on rwnd. */
793 void sctp_ulpq_partial_delivery(struct sctp_ulpq *ulpq,
794 struct sctp_chunk *chunk,
795 unsigned int __nocast gfp)
797 struct sctp_ulpevent *event;
798 struct sctp_association *asoc;
802 /* Are we already in partial delivery mode? */
803 if (!sctp_sk(asoc->base.sk)->pd_mode) {
805 /* Is partial delivery possible? */
806 event = sctp_ulpq_retrieve_first(ulpq);
807 /* Send event to the ULP. */
809 sctp_ulpq_tail_event(ulpq, event);
810 sctp_sk(asoc->base.sk)->pd_mode = 1;
817 /* Renege some packets to make room for an incoming chunk. */
818 void sctp_ulpq_renege(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk,
819 unsigned int __nocast gfp)
821 struct sctp_association *asoc;
827 needed = ntohs(chunk->chunk_hdr->length);
828 needed -= sizeof(sctp_data_chunk_t);
830 needed = SCTP_DEFAULT_MAXWINDOW;
834 if (skb_queue_empty(&asoc->base.sk->sk_receive_queue)) {
835 freed = sctp_ulpq_renege_order(ulpq, needed);
836 if (freed < needed) {
837 freed += sctp_ulpq_renege_frags(ulpq, needed - freed);
840 /* If able to free enough room, accept this chunk. */
841 if (chunk && (freed >= needed)) {
843 tsn = ntohl(chunk->subh.data_hdr->tsn);
844 sctp_tsnmap_mark(&asoc->peer.tsn_map, tsn);
845 sctp_ulpq_tail_data(ulpq, chunk, gfp);
847 sctp_ulpq_partial_delivery(ulpq, chunk, gfp);
855 /* Notify the application if an association is aborted and in
856 * partial delivery mode. Send up any pending received messages.
858 void sctp_ulpq_abort_pd(struct sctp_ulpq *ulpq, unsigned int __nocast gfp)
860 struct sctp_ulpevent *ev = NULL;
866 sk = ulpq->asoc->base.sk;
867 if (sctp_ulpevent_type_enabled(SCTP_PARTIAL_DELIVERY_EVENT,
868 &sctp_sk(sk)->subscribe))
869 ev = sctp_ulpevent_make_pdapi(ulpq->asoc,
870 SCTP_PARTIAL_DELIVERY_ABORTED,
873 __skb_queue_tail(&sk->sk_receive_queue, sctp_event2skb(ev));
875 /* If there is data waiting, send it up the socket now. */
876 if (sctp_ulpq_clear_pd(ulpq) || ev)
877 sk->sk_data_ready(sk, 0);