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 La Monte H.P. Yarroll
8 * This file is part of the SCTP kernel reference Implementation
10 * This module provides the abstraction for an SCTP association.
12 * The SCTP reference implementation is free software;
13 * you can redistribute it and/or modify it under the terms of
14 * the GNU General Public License as published by
15 * the Free Software Foundation; either version 2, or (at your option)
18 * The SCTP reference implementation is distributed in the hope that it
19 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
20 * ************************
21 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
22 * See the GNU General Public License for more details.
24 * You should have received a copy of the GNU General Public License
25 * along with GNU CC; see the file COPYING. If not, write to
26 * the Free Software Foundation, 59 Temple Place - Suite 330,
27 * Boston, MA 02111-1307, USA.
29 * Please send any bug reports or fixes you make to the
31 * lksctp developers <lksctp-developers@lists.sourceforge.net>
33 * Or submit a bug report through the following website:
34 * http://www.sf.net/projects/lksctp
36 * Written or modified by:
37 * La Monte H.P. Yarroll <piggy@acm.org>
38 * Karl Knutson <karl@athena.chicago.il.us>
39 * Jon Grimm <jgrimm@us.ibm.com>
40 * Xingang Guo <xingang.guo@intel.com>
41 * Hui Huang <hui.huang@nokia.com>
42 * Sridhar Samudrala <sri@us.ibm.com>
43 * Daisy Chang <daisyc@us.ibm.com>
44 * Ryan Layer <rmlayer@us.ibm.com>
45 * Kevin Gao <kevin.gao@intel.com>
47 * Any bugs reported given to us we will try to fix... any fixes shared will
48 * be incorporated into the next SCTP release.
51 #include <linux/types.h>
52 #include <linux/fcntl.h>
53 #include <linux/poll.h>
54 #include <linux/init.h>
56 #include <linux/slab.h>
59 #include <net/sctp/sctp.h>
60 #include <net/sctp/sm.h>
62 /* Forward declarations for internal functions. */
63 static void sctp_assoc_bh_rcv(struct work_struct *work);
66 /* 1st Level Abstractions. */
68 /* Initialize a new association from provided memory. */
69 static struct sctp_association *sctp_association_init(struct sctp_association *asoc,
70 const struct sctp_endpoint *ep,
71 const struct sock *sk,
80 /* Retrieve the SCTP per socket area. */
81 sp = sctp_sk((struct sock *)sk);
83 /* Init all variables to a known value. */
84 memset(asoc, 0, sizeof(struct sctp_association));
86 /* Discarding const is appropriate here. */
87 asoc->ep = (struct sctp_endpoint *)ep;
88 sctp_endpoint_hold(asoc->ep);
91 asoc->base.sk = (struct sock *)sk;
92 sock_hold(asoc->base.sk);
94 /* Initialize the common base substructure. */
95 asoc->base.type = SCTP_EP_TYPE_ASSOCIATION;
97 /* Initialize the object handling fields. */
98 atomic_set(&asoc->base.refcnt, 1);
100 asoc->base.malloced = 0;
102 /* Initialize the bind addr area. */
103 sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port);
105 asoc->state = SCTP_STATE_CLOSED;
107 /* Set these values from the socket values, a conversion between
108 * millsecons to seconds/microseconds must also be done.
110 asoc->cookie_life.tv_sec = sp->assocparams.sasoc_cookie_life / 1000;
111 asoc->cookie_life.tv_usec = (sp->assocparams.sasoc_cookie_life % 1000)
113 asoc->frag_point = 0;
115 /* Set the association max_retrans and RTO values from the
118 asoc->max_retrans = sp->assocparams.sasoc_asocmaxrxt;
119 asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial);
120 asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max);
121 asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min);
123 asoc->overall_error_count = 0;
125 /* Initialize the association's heartbeat interval based on the
126 * sock configured value.
128 asoc->hbinterval = msecs_to_jiffies(sp->hbinterval);
130 /* Initialize path max retrans value. */
131 asoc->pathmaxrxt = sp->pathmaxrxt;
133 /* Initialize default path MTU. */
134 asoc->pathmtu = sp->pathmtu;
136 /* Set association default SACK delay */
137 asoc->sackdelay = msecs_to_jiffies(sp->sackdelay);
139 /* Set the association default flags controlling
140 * Heartbeat, SACK delay, and Path MTU Discovery.
142 asoc->param_flags = sp->param_flags;
144 /* Initialize the maximum mumber of new data packets that can be sent
147 asoc->max_burst = sp->max_burst;
149 /* initialize association timers */
150 asoc->timeouts[SCTP_EVENT_TIMEOUT_NONE] = 0;
151 asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial;
152 asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial;
153 asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial;
154 asoc->timeouts[SCTP_EVENT_TIMEOUT_T3_RTX] = 0;
155 asoc->timeouts[SCTP_EVENT_TIMEOUT_T4_RTO] = 0;
157 /* sctpimpguide Section 2.12.2
158 * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the
159 * recommended value of 5 times 'RTO.Max'.
161 asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
164 asoc->timeouts[SCTP_EVENT_TIMEOUT_HEARTBEAT] = 0;
165 asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
166 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] =
169 /* Initilizes the timers */
170 for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
171 init_timer(&asoc->timers[i]);
172 asoc->timers[i].function = sctp_timer_events[i];
173 asoc->timers[i].data = (unsigned long) asoc;
176 /* Pull default initialization values from the sock options.
177 * Note: This assumes that the values have already been
178 * validated in the sock.
180 asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams;
181 asoc->c.sinit_num_ostreams = sp->initmsg.sinit_num_ostreams;
182 asoc->max_init_attempts = sp->initmsg.sinit_max_attempts;
184 asoc->max_init_timeo =
185 msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo);
187 /* Allocate storage for the ssnmap after the inbound and outbound
188 * streams have been negotiated during Init.
192 /* Set the local window size for receive.
193 * This is also the rcvbuf space per association.
194 * RFC 6 - A SCTP receiver MUST be able to receive a minimum of
195 * 1500 bytes in one SCTP packet.
197 if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
198 asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
200 asoc->rwnd = sk->sk_rcvbuf/2;
202 asoc->a_rwnd = asoc->rwnd;
206 /* Use my own max window until I learn something better. */
207 asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW;
209 /* Set the sndbuf size for transmit. */
210 asoc->sndbuf_used = 0;
212 /* Initialize the receive memory counter */
213 atomic_set(&asoc->rmem_alloc, 0);
215 init_waitqueue_head(&asoc->wait);
217 asoc->c.my_vtag = sctp_generate_tag(ep);
218 asoc->peer.i.init_tag = 0; /* INIT needs a vtag of 0. */
219 asoc->c.peer_vtag = 0;
221 asoc->c.peer_ttag = 0;
222 asoc->c.my_port = ep->base.bind_addr.port;
224 asoc->c.initial_tsn = sctp_generate_tsn(ep);
226 asoc->next_tsn = asoc->c.initial_tsn;
228 asoc->ctsn_ack_point = asoc->next_tsn - 1;
229 asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
230 asoc->highest_sacked = asoc->ctsn_ack_point;
231 asoc->last_cwr_tsn = asoc->ctsn_ack_point;
232 asoc->unack_data = 0;
234 /* ADDIP Section 4.1 Asconf Chunk Procedures
236 * When an endpoint has an ASCONF signaled change to be sent to the
237 * remote endpoint it should do the following:
239 * A2) a serial number should be assigned to the chunk. The serial
240 * number SHOULD be a monotonically increasing number. The serial
241 * numbers SHOULD be initialized at the start of the
242 * association to the same value as the initial TSN.
244 asoc->addip_serial = asoc->c.initial_tsn;
246 INIT_LIST_HEAD(&asoc->addip_chunk_list);
248 /* Make an empty list of remote transport addresses. */
249 INIT_LIST_HEAD(&asoc->peer.transport_addr_list);
250 asoc->peer.transport_count = 0;
252 /* RFC 2960 5.1 Normal Establishment of an Association
254 * After the reception of the first data chunk in an
255 * association the endpoint must immediately respond with a
256 * sack to acknowledge the data chunk. Subsequent
257 * acknowledgements should be done as described in Section
260 * [We implement this by telling a new association that it
261 * already received one packet.]
263 asoc->peer.sack_needed = 1;
265 /* Assume that the peer recongizes ASCONF until reported otherwise
266 * via an ERROR chunk.
268 asoc->peer.asconf_capable = 1;
270 /* Create an input queue. */
271 sctp_inq_init(&asoc->base.inqueue);
272 sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv);
274 /* Create an output queue. */
275 sctp_outq_init(asoc, &asoc->outqueue);
277 if (!sctp_ulpq_init(&asoc->ulpq, asoc))
280 /* Set up the tsn tracking. */
281 sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_SIZE, 0);
287 /* Assume that peer would support both address types unless we are
290 asoc->peer.ipv4_address = 1;
291 asoc->peer.ipv6_address = 1;
292 INIT_LIST_HEAD(&asoc->asocs);
294 asoc->autoclose = sp->autoclose;
296 asoc->default_stream = sp->default_stream;
297 asoc->default_ppid = sp->default_ppid;
298 asoc->default_flags = sp->default_flags;
299 asoc->default_context = sp->default_context;
300 asoc->default_timetolive = sp->default_timetolive;
301 asoc->default_rcv_context = sp->default_rcv_context;
303 /* AUTH related initializations */
304 INIT_LIST_HEAD(&asoc->endpoint_shared_keys);
305 err = sctp_auth_asoc_copy_shkeys(ep, asoc, gfp);
309 asoc->active_key_id = ep->active_key_id;
310 asoc->asoc_shared_key = NULL;
312 asoc->default_hmac_id = 0;
313 /* Save the hmacs and chunks list into this association */
314 if (ep->auth_hmacs_list)
315 memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
316 ntohs(ep->auth_hmacs_list->param_hdr.length));
317 if (ep->auth_chunk_list)
318 memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
319 ntohs(ep->auth_chunk_list->param_hdr.length));
321 /* Get the AUTH random number for this association */
322 p = (sctp_paramhdr_t *)asoc->c.auth_random;
323 p->type = SCTP_PARAM_RANDOM;
324 p->length = htons(sizeof(sctp_paramhdr_t) + SCTP_AUTH_RANDOM_LENGTH);
325 get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
330 sctp_endpoint_put(asoc->ep);
331 sock_put(asoc->base.sk);
335 /* Allocate and initialize a new association */
336 struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
337 const struct sock *sk,
341 struct sctp_association *asoc;
343 asoc = t_new(struct sctp_association, gfp);
347 if (!sctp_association_init(asoc, ep, sk, scope, gfp))
350 asoc->base.malloced = 1;
351 SCTP_DBG_OBJCNT_INC(assoc);
352 SCTP_DEBUG_PRINTK("Created asoc %p\n", asoc);
362 /* Free this association if possible. There may still be users, so
363 * the actual deallocation may be delayed.
365 void sctp_association_free(struct sctp_association *asoc)
367 struct sock *sk = asoc->base.sk;
368 struct sctp_transport *transport;
369 struct list_head *pos, *temp;
372 /* Only real associations count against the endpoint, so
373 * don't bother for if this is a temporary association.
376 list_del(&asoc->asocs);
378 /* Decrement the backlog value for a TCP-style listening
381 if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
382 sk->sk_ack_backlog--;
385 /* Mark as dead, so other users can know this structure is
390 /* Dispose of any data lying around in the outqueue. */
391 sctp_outq_free(&asoc->outqueue);
393 /* Dispose of any pending messages for the upper layer. */
394 sctp_ulpq_free(&asoc->ulpq);
396 /* Dispose of any pending chunks on the inqueue. */
397 sctp_inq_free(&asoc->base.inqueue);
399 /* Free ssnmap storage. */
400 sctp_ssnmap_free(asoc->ssnmap);
402 /* Clean up the bound address list. */
403 sctp_bind_addr_free(&asoc->base.bind_addr);
405 /* Do we need to go through all of our timers and
406 * delete them? To be safe we will try to delete all, but we
407 * should be able to go through and make a guess based
410 for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
411 if (timer_pending(&asoc->timers[i]) &&
412 del_timer(&asoc->timers[i]))
413 sctp_association_put(asoc);
416 /* Free peer's cached cookie. */
417 kfree(asoc->peer.cookie);
419 /* Release the transport structures. */
420 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
421 transport = list_entry(pos, struct sctp_transport, transports);
423 sctp_transport_free(transport);
426 asoc->peer.transport_count = 0;
428 /* Free any cached ASCONF_ACK chunk. */
429 if (asoc->addip_last_asconf_ack)
430 sctp_chunk_free(asoc->addip_last_asconf_ack);
432 /* Free any cached ASCONF chunk. */
433 if (asoc->addip_last_asconf)
434 sctp_chunk_free(asoc->addip_last_asconf);
436 /* AUTH - Free the endpoint shared keys */
437 sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
439 /* AUTH - Free the association shared key */
440 sctp_auth_key_put(asoc->asoc_shared_key);
442 sctp_association_put(asoc);
445 /* Cleanup and free up an association. */
446 static void sctp_association_destroy(struct sctp_association *asoc)
448 SCTP_ASSERT(asoc->base.dead, "Assoc is not dead", return);
450 sctp_endpoint_put(asoc->ep);
451 sock_put(asoc->base.sk);
453 if (asoc->assoc_id != 0) {
454 spin_lock_bh(&sctp_assocs_id_lock);
455 idr_remove(&sctp_assocs_id, asoc->assoc_id);
456 spin_unlock_bh(&sctp_assocs_id_lock);
459 BUG_TRAP(!atomic_read(&asoc->rmem_alloc));
461 if (asoc->base.malloced) {
463 SCTP_DBG_OBJCNT_DEC(assoc);
467 /* Change the primary destination address for the peer. */
468 void sctp_assoc_set_primary(struct sctp_association *asoc,
469 struct sctp_transport *transport)
471 asoc->peer.primary_path = transport;
473 /* Set a default msg_name for events. */
474 memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
475 sizeof(union sctp_addr));
477 /* If the primary path is changing, assume that the
478 * user wants to use this new path.
480 if ((transport->state == SCTP_ACTIVE) ||
481 (transport->state == SCTP_UNKNOWN))
482 asoc->peer.active_path = transport;
485 * SFR-CACC algorithm:
486 * Upon the receipt of a request to change the primary
487 * destination address, on the data structure for the new
488 * primary destination, the sender MUST do the following:
490 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
491 * to this destination address earlier. The sender MUST set
492 * CYCLING_CHANGEOVER to indicate that this switch is a
493 * double switch to the same destination address.
495 if (transport->cacc.changeover_active)
496 transport->cacc.cycling_changeover = 1;
498 /* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
499 * a changeover has occurred.
501 transport->cacc.changeover_active = 1;
503 /* 3) The sender MUST store the next TSN to be sent in
504 * next_tsn_at_change.
506 transport->cacc.next_tsn_at_change = asoc->next_tsn;
509 /* Remove a transport from an association. */
510 void sctp_assoc_rm_peer(struct sctp_association *asoc,
511 struct sctp_transport *peer)
513 struct list_head *pos;
514 struct sctp_transport *transport;
516 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_rm_peer:association %p addr: ",
520 ntohs(peer->ipaddr.v4.sin_port));
522 /* If we are to remove the current retran_path, update it
523 * to the next peer before removing this peer from the list.
525 if (asoc->peer.retran_path == peer)
526 sctp_assoc_update_retran_path(asoc);
528 /* Remove this peer from the list. */
529 list_del(&peer->transports);
531 /* Get the first transport of asoc. */
532 pos = asoc->peer.transport_addr_list.next;
533 transport = list_entry(pos, struct sctp_transport, transports);
535 /* Update any entries that match the peer to be deleted. */
536 if (asoc->peer.primary_path == peer)
537 sctp_assoc_set_primary(asoc, transport);
538 if (asoc->peer.active_path == peer)
539 asoc->peer.active_path = transport;
540 if (asoc->peer.last_data_from == peer)
541 asoc->peer.last_data_from = transport;
543 /* If we remove the transport an INIT was last sent to, set it to
544 * NULL. Combined with the update of the retran path above, this
545 * will cause the next INIT to be sent to the next available
546 * transport, maintaining the cycle.
548 if (asoc->init_last_sent_to == peer)
549 asoc->init_last_sent_to = NULL;
551 asoc->peer.transport_count--;
553 sctp_transport_free(peer);
556 /* Add a transport address to an association. */
557 struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
558 const union sctp_addr *addr,
560 const int peer_state)
562 struct sctp_transport *peer;
563 struct sctp_sock *sp;
566 sp = sctp_sk(asoc->base.sk);
568 /* AF_INET and AF_INET6 share common port field. */
569 port = ntohs(addr->v4.sin_port);
571 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_add_peer:association %p addr: ",
572 " port: %d state:%d\n",
578 /* Set the port if it has not been set yet. */
579 if (0 == asoc->peer.port)
580 asoc->peer.port = port;
582 /* Check to see if this is a duplicate. */
583 peer = sctp_assoc_lookup_paddr(asoc, addr);
585 if (peer->state == SCTP_UNKNOWN) {
586 if (peer_state == SCTP_ACTIVE)
587 peer->state = SCTP_ACTIVE;
588 if (peer_state == SCTP_UNCONFIRMED)
589 peer->state = SCTP_UNCONFIRMED;
594 peer = sctp_transport_new(addr, gfp);
598 sctp_transport_set_owner(peer, asoc);
600 /* Initialize the peer's heartbeat interval based on the
601 * association configured value.
603 peer->hbinterval = asoc->hbinterval;
605 /* Set the path max_retrans. */
606 peer->pathmaxrxt = asoc->pathmaxrxt;
608 /* Initialize the peer's SACK delay timeout based on the
609 * association configured value.
611 peer->sackdelay = asoc->sackdelay;
613 /* Enable/disable heartbeat, SACK delay, and path MTU discovery
614 * based on association setting.
616 peer->param_flags = asoc->param_flags;
618 /* Initialize the pmtu of the transport. */
619 if (peer->param_flags & SPP_PMTUD_ENABLE)
620 sctp_transport_pmtu(peer);
621 else if (asoc->pathmtu)
622 peer->pathmtu = asoc->pathmtu;
624 peer->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
626 /* If this is the first transport addr on this association,
627 * initialize the association PMTU to the peer's PMTU.
628 * If not and the current association PMTU is higher than the new
629 * peer's PMTU, reset the association PMTU to the new peer's PMTU.
632 asoc->pathmtu = min_t(int, peer->pathmtu, asoc->pathmtu);
634 asoc->pathmtu = peer->pathmtu;
636 SCTP_DEBUG_PRINTK("sctp_assoc_add_peer:association %p PMTU set to "
637 "%d\n", asoc, asoc->pathmtu);
639 asoc->frag_point = sctp_frag_point(sp, asoc->pathmtu);
641 /* The asoc->peer.port might not be meaningful yet, but
642 * initialize the packet structure anyway.
644 sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
649 * o The initial cwnd before DATA transmission or after a sufficiently
650 * long idle period MUST be set to
651 * min(4*MTU, max(2*MTU, 4380 bytes))
653 * o The initial value of ssthresh MAY be arbitrarily high
654 * (for example, implementations MAY use the size of the
655 * receiver advertised window).
657 peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
659 /* At this point, we may not have the receiver's advertised window,
660 * so initialize ssthresh to the default value and it will be set
661 * later when we process the INIT.
663 peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
665 peer->partial_bytes_acked = 0;
666 peer->flight_size = 0;
668 /* Set the transport's RTO.initial value */
669 peer->rto = asoc->rto_initial;
671 /* Set the peer's active state. */
672 peer->state = peer_state;
674 /* Attach the remote transport to our asoc. */
675 list_add_tail(&peer->transports, &asoc->peer.transport_addr_list);
676 asoc->peer.transport_count++;
678 /* If we do not yet have a primary path, set one. */
679 if (!asoc->peer.primary_path) {
680 sctp_assoc_set_primary(asoc, peer);
681 asoc->peer.retran_path = peer;
684 if (asoc->peer.active_path == asoc->peer.retran_path) {
685 asoc->peer.retran_path = peer;
691 /* Delete a transport address from an association. */
692 void sctp_assoc_del_peer(struct sctp_association *asoc,
693 const union sctp_addr *addr)
695 struct list_head *pos;
696 struct list_head *temp;
697 struct sctp_transport *transport;
699 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
700 transport = list_entry(pos, struct sctp_transport, transports);
701 if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
702 /* Do book keeping for removing the peer and free it. */
703 sctp_assoc_rm_peer(asoc, transport);
709 /* Lookup a transport by address. */
710 struct sctp_transport *sctp_assoc_lookup_paddr(
711 const struct sctp_association *asoc,
712 const union sctp_addr *address)
714 struct sctp_transport *t;
715 struct list_head *pos;
717 /* Cycle through all transports searching for a peer address. */
719 list_for_each(pos, &asoc->peer.transport_addr_list) {
720 t = list_entry(pos, struct sctp_transport, transports);
721 if (sctp_cmp_addr_exact(address, &t->ipaddr))
728 /* Engage in transport control operations.
729 * Mark the transport up or down and send a notification to the user.
730 * Select and update the new active and retran paths.
732 void sctp_assoc_control_transport(struct sctp_association *asoc,
733 struct sctp_transport *transport,
734 sctp_transport_cmd_t command,
735 sctp_sn_error_t error)
737 struct sctp_transport *t = NULL;
738 struct sctp_transport *first;
739 struct sctp_transport *second;
740 struct sctp_ulpevent *event;
741 struct sockaddr_storage addr;
742 struct list_head *pos;
745 /* Record the transition on the transport. */
747 case SCTP_TRANSPORT_UP:
748 /* If we are moving from UNCONFIRMED state due
749 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
750 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
752 if (SCTP_UNCONFIRMED == transport->state &&
753 SCTP_HEARTBEAT_SUCCESS == error)
754 spc_state = SCTP_ADDR_CONFIRMED;
756 spc_state = SCTP_ADDR_AVAILABLE;
757 transport->state = SCTP_ACTIVE;
760 case SCTP_TRANSPORT_DOWN:
761 /* if the transort was never confirmed, do not transition it
764 if (transport->state != SCTP_UNCONFIRMED)
765 transport->state = SCTP_INACTIVE;
767 spc_state = SCTP_ADDR_UNREACHABLE;
774 /* Generate and send a SCTP_PEER_ADDR_CHANGE notification to the
777 memset(&addr, 0, sizeof(struct sockaddr_storage));
778 memcpy(&addr, &transport->ipaddr, transport->af_specific->sockaddr_len);
779 event = sctp_ulpevent_make_peer_addr_change(asoc, &addr,
780 0, spc_state, error, GFP_ATOMIC);
782 sctp_ulpq_tail_event(&asoc->ulpq, event);
784 /* Select new active and retran paths. */
786 /* Look for the two most recently used active transports.
788 * This code produces the wrong ordering whenever jiffies
789 * rolls over, but we still get usable transports, so we don't
792 first = NULL; second = NULL;
794 list_for_each(pos, &asoc->peer.transport_addr_list) {
795 t = list_entry(pos, struct sctp_transport, transports);
797 if ((t->state == SCTP_INACTIVE) ||
798 (t->state == SCTP_UNCONFIRMED))
800 if (!first || t->last_time_heard > first->last_time_heard) {
804 if (!second || t->last_time_heard > second->last_time_heard)
808 /* RFC 2960 6.4 Multi-Homed SCTP Endpoints
810 * By default, an endpoint should always transmit to the
811 * primary path, unless the SCTP user explicitly specifies the
812 * destination transport address (and possibly source
813 * transport address) to use.
815 * [If the primary is active but not most recent, bump the most
816 * recently used transport.]
818 if (((asoc->peer.primary_path->state == SCTP_ACTIVE) ||
819 (asoc->peer.primary_path->state == SCTP_UNKNOWN)) &&
820 first != asoc->peer.primary_path) {
822 first = asoc->peer.primary_path;
825 /* If we failed to find a usable transport, just camp on the
826 * primary, even if it is inactive.
829 first = asoc->peer.primary_path;
830 second = asoc->peer.primary_path;
833 /* Set the active and retran transports. */
834 asoc->peer.active_path = first;
835 asoc->peer.retran_path = second;
838 /* Hold a reference to an association. */
839 void sctp_association_hold(struct sctp_association *asoc)
841 atomic_inc(&asoc->base.refcnt);
844 /* Release a reference to an association and cleanup
845 * if there are no more references.
847 void sctp_association_put(struct sctp_association *asoc)
849 if (atomic_dec_and_test(&asoc->base.refcnt))
850 sctp_association_destroy(asoc);
853 /* Allocate the next TSN, Transmission Sequence Number, for the given
856 __u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
858 /* From Section 1.6 Serial Number Arithmetic:
859 * Transmission Sequence Numbers wrap around when they reach
860 * 2**32 - 1. That is, the next TSN a DATA chunk MUST use
861 * after transmitting TSN = 2*32 - 1 is TSN = 0.
863 __u32 retval = asoc->next_tsn;
870 /* Compare two addresses to see if they match. Wildcard addresses
871 * only match themselves.
873 int sctp_cmp_addr_exact(const union sctp_addr *ss1,
874 const union sctp_addr *ss2)
878 af = sctp_get_af_specific(ss1->sa.sa_family);
882 return af->cmp_addr(ss1, ss2);
885 /* Return an ecne chunk to get prepended to a packet.
886 * Note: We are sly and return a shared, prealloced chunk. FIXME:
887 * No we don't, but we could/should.
889 struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
891 struct sctp_chunk *chunk;
893 /* Send ECNE if needed.
894 * Not being able to allocate a chunk here is not deadly.
897 chunk = sctp_make_ecne(asoc, asoc->last_ecne_tsn);
905 * Find which transport this TSN was sent on.
907 struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
910 struct sctp_transport *active;
911 struct sctp_transport *match;
912 struct list_head *entry, *pos;
913 struct sctp_transport *transport;
914 struct sctp_chunk *chunk;
915 __be32 key = htonl(tsn);
920 * FIXME: In general, find a more efficient data structure for
925 * The general strategy is to search each transport's transmitted
926 * list. Return which transport this TSN lives on.
928 * Let's be hopeful and check the active_path first.
929 * Another optimization would be to know if there is only one
930 * outbound path and not have to look for the TSN at all.
934 active = asoc->peer.active_path;
936 list_for_each(entry, &active->transmitted) {
937 chunk = list_entry(entry, struct sctp_chunk, transmitted_list);
939 if (key == chunk->subh.data_hdr->tsn) {
945 /* If not found, go search all the other transports. */
946 list_for_each(pos, &asoc->peer.transport_addr_list) {
947 transport = list_entry(pos, struct sctp_transport, transports);
949 if (transport == active)
951 list_for_each(entry, &transport->transmitted) {
952 chunk = list_entry(entry, struct sctp_chunk,
954 if (key == chunk->subh.data_hdr->tsn) {
964 /* Is this the association we are looking for? */
965 struct sctp_transport *sctp_assoc_is_match(struct sctp_association *asoc,
966 const union sctp_addr *laddr,
967 const union sctp_addr *paddr)
969 struct sctp_transport *transport;
971 if ((htons(asoc->base.bind_addr.port) == laddr->v4.sin_port) &&
972 (htons(asoc->peer.port) == paddr->v4.sin_port)) {
973 transport = sctp_assoc_lookup_paddr(asoc, paddr);
977 if (sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
978 sctp_sk(asoc->base.sk)))
987 /* Do delayed input processing. This is scheduled by sctp_rcv(). */
988 static void sctp_assoc_bh_rcv(struct work_struct *work)
990 struct sctp_association *asoc =
991 container_of(work, struct sctp_association,
992 base.inqueue.immediate);
993 struct sctp_endpoint *ep;
994 struct sctp_chunk *chunk;
996 struct sctp_inq *inqueue;
998 sctp_subtype_t subtype;
1001 /* The association should be held so we should be safe. */
1005 inqueue = &asoc->base.inqueue;
1006 sctp_association_hold(asoc);
1007 while (NULL != (chunk = sctp_inq_pop(inqueue))) {
1008 state = asoc->state;
1009 subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
1011 /* Remember where the last DATA chunk came from so we
1012 * know where to send the SACK.
1014 if (sctp_chunk_is_data(chunk))
1015 asoc->peer.last_data_from = chunk->transport;
1017 SCTP_INC_STATS(SCTP_MIB_INCTRLCHUNKS);
1019 if (chunk->transport)
1020 chunk->transport->last_time_heard = jiffies;
1022 /* Run through the state machine. */
1023 error = sctp_do_sm(SCTP_EVENT_T_CHUNK, subtype,
1024 state, ep, asoc, chunk, GFP_ATOMIC);
1026 /* Check to see if the association is freed in response to
1027 * the incoming chunk. If so, get out of the while loop.
1029 if (asoc->base.dead)
1032 /* If there is an error on chunk, discard this packet. */
1034 chunk->pdiscard = 1;
1036 sctp_association_put(asoc);
1039 /* This routine moves an association from its old sk to a new sk. */
1040 void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1042 struct sctp_sock *newsp = sctp_sk(newsk);
1043 struct sock *oldsk = assoc->base.sk;
1045 /* Delete the association from the old endpoint's list of
1048 list_del_init(&assoc->asocs);
1050 /* Decrement the backlog value for a TCP-style socket. */
1051 if (sctp_style(oldsk, TCP))
1052 oldsk->sk_ack_backlog--;
1054 /* Release references to the old endpoint and the sock. */
1055 sctp_endpoint_put(assoc->ep);
1056 sock_put(assoc->base.sk);
1058 /* Get a reference to the new endpoint. */
1059 assoc->ep = newsp->ep;
1060 sctp_endpoint_hold(assoc->ep);
1062 /* Get a reference to the new sock. */
1063 assoc->base.sk = newsk;
1064 sock_hold(assoc->base.sk);
1066 /* Add the association to the new endpoint's list of associations. */
1067 sctp_endpoint_add_asoc(newsp->ep, assoc);
1070 /* Update an association (possibly from unexpected COOKIE-ECHO processing). */
1071 void sctp_assoc_update(struct sctp_association *asoc,
1072 struct sctp_association *new)
1074 struct sctp_transport *trans;
1075 struct list_head *pos, *temp;
1077 /* Copy in new parameters of peer. */
1079 asoc->peer.rwnd = new->peer.rwnd;
1080 asoc->peer.sack_needed = new->peer.sack_needed;
1081 asoc->peer.i = new->peer.i;
1082 sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_SIZE,
1083 asoc->peer.i.initial_tsn);
1085 /* Remove any peer addresses not present in the new association. */
1086 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1087 trans = list_entry(pos, struct sctp_transport, transports);
1088 if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr))
1089 sctp_assoc_del_peer(asoc, &trans->ipaddr);
1091 if (asoc->state >= SCTP_STATE_ESTABLISHED)
1092 sctp_transport_reset(trans);
1095 /* If the case is A (association restart), use
1096 * initial_tsn as next_tsn. If the case is B, use
1097 * current next_tsn in case data sent to peer
1098 * has been discarded and needs retransmission.
1100 if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1101 asoc->next_tsn = new->next_tsn;
1102 asoc->ctsn_ack_point = new->ctsn_ack_point;
1103 asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1105 /* Reinitialize SSN for both local streams
1106 * and peer's streams.
1108 sctp_ssnmap_clear(asoc->ssnmap);
1110 /* Flush the ULP reassembly and ordered queue.
1111 * Any data there will now be stale and will
1114 sctp_ulpq_flush(&asoc->ulpq);
1116 /* reset the overall association error count so
1117 * that the restarted association doesn't get torn
1118 * down on the next retransmission timer.
1120 asoc->overall_error_count = 0;
1123 /* Add any peer addresses from the new association. */
1124 list_for_each(pos, &new->peer.transport_addr_list) {
1125 trans = list_entry(pos, struct sctp_transport,
1127 if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr))
1128 sctp_assoc_add_peer(asoc, &trans->ipaddr,
1129 GFP_ATOMIC, trans->state);
1132 asoc->ctsn_ack_point = asoc->next_tsn - 1;
1133 asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1134 if (!asoc->ssnmap) {
1135 /* Move the ssnmap. */
1136 asoc->ssnmap = new->ssnmap;
1140 if (!asoc->assoc_id) {
1141 /* get a new association id since we don't have one
1144 sctp_assoc_set_id(asoc, GFP_ATOMIC);
1148 /* SCTP-AUTH: XXX something needs to be done here*/
1151 /* Update the retran path for sending a retransmitted packet.
1152 * Round-robin through the active transports, else round-robin
1153 * through the inactive transports as this is the next best thing
1156 void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1158 struct sctp_transport *t, *next;
1159 struct list_head *head = &asoc->peer.transport_addr_list;
1160 struct list_head *pos;
1162 /* Find the next transport in a round-robin fashion. */
1163 t = asoc->peer.retran_path;
1164 pos = &t->transports;
1168 /* Skip the head. */
1169 if (pos->next == head)
1174 t = list_entry(pos, struct sctp_transport, transports);
1176 /* Try to find an active transport. */
1178 if ((t->state == SCTP_ACTIVE) ||
1179 (t->state == SCTP_UNKNOWN)) {
1182 /* Keep track of the next transport in case
1183 * we don't find any active transport.
1189 /* We have exhausted the list, but didn't find any
1190 * other active transports. If so, use the next
1193 if (t == asoc->peer.retran_path) {
1199 asoc->peer.retran_path = t;
1201 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_update_retran_path:association"
1206 ntohs(t->ipaddr.v4.sin_port));
1209 /* Choose the transport for sending a INIT packet. */
1210 struct sctp_transport *sctp_assoc_choose_init_transport(
1211 struct sctp_association *asoc)
1213 struct sctp_transport *t;
1215 /* Use the retran path. If the last INIT was sent over the
1216 * retran path, update the retran path and use it.
1218 if (!asoc->init_last_sent_to) {
1219 t = asoc->peer.active_path;
1221 if (asoc->init_last_sent_to == asoc->peer.retran_path)
1222 sctp_assoc_update_retran_path(asoc);
1223 t = asoc->peer.retran_path;
1226 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_update_retran_path:association"
1231 ntohs(t->ipaddr.v4.sin_port));
1236 /* Choose the transport for sending a SHUTDOWN packet. */
1237 struct sctp_transport *sctp_assoc_choose_shutdown_transport(
1238 struct sctp_association *asoc)
1240 /* If this is the first time SHUTDOWN is sent, use the active path,
1241 * else use the retran path. If the last SHUTDOWN was sent over the
1242 * retran path, update the retran path and use it.
1244 if (!asoc->shutdown_last_sent_to)
1245 return asoc->peer.active_path;
1247 if (asoc->shutdown_last_sent_to == asoc->peer.retran_path)
1248 sctp_assoc_update_retran_path(asoc);
1249 return asoc->peer.retran_path;
1254 /* Update the association's pmtu and frag_point by going through all the
1255 * transports. This routine is called when a transport's PMTU has changed.
1257 void sctp_assoc_sync_pmtu(struct sctp_association *asoc)
1259 struct sctp_transport *t;
1260 struct list_head *pos;
1266 /* Get the lowest pmtu of all the transports. */
1267 list_for_each(pos, &asoc->peer.transport_addr_list) {
1268 t = list_entry(pos, struct sctp_transport, transports);
1269 if (t->pmtu_pending && t->dst) {
1270 sctp_transport_update_pmtu(t, dst_mtu(t->dst));
1271 t->pmtu_pending = 0;
1273 if (!pmtu || (t->pathmtu < pmtu))
1278 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
1279 asoc->pathmtu = pmtu;
1280 asoc->frag_point = sctp_frag_point(sp, pmtu);
1283 SCTP_DEBUG_PRINTK("%s: asoc:%p, pmtu:%d, frag_point:%d\n",
1284 __FUNCTION__, asoc, asoc->pathmtu, asoc->frag_point);
1287 /* Should we send a SACK to update our peer? */
1288 static inline int sctp_peer_needs_update(struct sctp_association *asoc)
1290 switch (asoc->state) {
1291 case SCTP_STATE_ESTABLISHED:
1292 case SCTP_STATE_SHUTDOWN_PENDING:
1293 case SCTP_STATE_SHUTDOWN_RECEIVED:
1294 case SCTP_STATE_SHUTDOWN_SENT:
1295 if ((asoc->rwnd > asoc->a_rwnd) &&
1296 ((asoc->rwnd - asoc->a_rwnd) >=
1297 min_t(__u32, (asoc->base.sk->sk_rcvbuf >> 1), asoc->pathmtu)))
1306 /* Increase asoc's rwnd by len and send any window update SACK if needed. */
1307 void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned len)
1309 struct sctp_chunk *sack;
1310 struct timer_list *timer;
1312 if (asoc->rwnd_over) {
1313 if (asoc->rwnd_over >= len) {
1314 asoc->rwnd_over -= len;
1316 asoc->rwnd += (len - asoc->rwnd_over);
1317 asoc->rwnd_over = 0;
1323 SCTP_DEBUG_PRINTK("%s: asoc %p rwnd increased by %d to (%u, %u) "
1324 "- %u\n", __FUNCTION__, asoc, len, asoc->rwnd,
1325 asoc->rwnd_over, asoc->a_rwnd);
1327 /* Send a window update SACK if the rwnd has increased by at least the
1328 * minimum of the association's PMTU and half of the receive buffer.
1329 * The algorithm used is similar to the one described in
1330 * Section 4.2.3.3 of RFC 1122.
1332 if (sctp_peer_needs_update(asoc)) {
1333 asoc->a_rwnd = asoc->rwnd;
1334 SCTP_DEBUG_PRINTK("%s: Sending window update SACK- asoc: %p "
1335 "rwnd: %u a_rwnd: %u\n", __FUNCTION__,
1336 asoc, asoc->rwnd, asoc->a_rwnd);
1337 sack = sctp_make_sack(asoc);
1341 asoc->peer.sack_needed = 0;
1343 sctp_outq_tail(&asoc->outqueue, sack);
1345 /* Stop the SACK timer. */
1346 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1347 if (timer_pending(timer) && del_timer(timer))
1348 sctp_association_put(asoc);
1352 /* Decrease asoc's rwnd by len. */
1353 void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned len)
1355 SCTP_ASSERT(asoc->rwnd, "rwnd zero", return);
1356 SCTP_ASSERT(!asoc->rwnd_over, "rwnd_over not zero", return);
1357 if (asoc->rwnd >= len) {
1360 asoc->rwnd_over = len - asoc->rwnd;
1363 SCTP_DEBUG_PRINTK("%s: asoc %p rwnd decreased by %d to (%u, %u)\n",
1364 __FUNCTION__, asoc, len, asoc->rwnd,
1368 /* Build the bind address list for the association based on info from the
1369 * local endpoint and the remote peer.
1371 int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1377 /* Use scoping rules to determine the subset of addresses from
1380 scope = sctp_scope(&asoc->peer.active_path->ipaddr);
1381 flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1382 if (asoc->peer.ipv4_address)
1383 flags |= SCTP_ADDR4_PEERSUPP;
1384 if (asoc->peer.ipv6_address)
1385 flags |= SCTP_ADDR6_PEERSUPP;
1387 return sctp_bind_addr_copy(&asoc->base.bind_addr,
1388 &asoc->ep->base.bind_addr,
1392 /* Build the association's bind address list from the cookie. */
1393 int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1394 struct sctp_cookie *cookie,
1397 int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length);
1398 int var_size3 = cookie->raw_addr_list_len;
1399 __u8 *raw = (__u8 *)cookie->peer_init + var_size2;
1401 return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1402 asoc->ep->base.bind_addr.port, gfp);
1405 /* Lookup laddr in the bind address list of an association. */
1406 int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1407 const union sctp_addr *laddr)
1411 if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1412 sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1413 sctp_sk(asoc->base.sk)))
1419 /* Set an association id for a given association */
1420 int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1425 if (unlikely(!idr_pre_get(&sctp_assocs_id, gfp)))
1428 spin_lock_bh(&sctp_assocs_id_lock);
1429 error = idr_get_new_above(&sctp_assocs_id, (void *)asoc,
1431 spin_unlock_bh(&sctp_assocs_id_lock);
1432 if (error == -EAGAIN)
1437 asoc->assoc_id = (sctp_assoc_t) assoc_id;