ia64/linux-2.6.18-xen.hg

view net/sctp/associola.c @ 871:9cbcc9008446

xen/x86: don't initialize cpu_data[]'s apicid field on generic code

Afaict, this is not only redundant with the intialization done in
drivers/xen/core/smpboot.c, but actually results - at least for
secondary CPUs - in the Xen-specific value written to be later
overwritten with whatever the generic code determines (with no
guarantee that the two values are identical).

Signed-off-by: Jan Beulich <jbeulich@novell.com>
author Keir Fraser <keir.fraser@citrix.com>
date Thu May 14 10:09:15 2009 +0100 (2009-05-14)
parents 831230e53067
children
line source
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
7 *
8 * This file is part of the SCTP kernel reference Implementation
9 *
10 * This module provides the abstraction for an SCTP association.
11 *
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)
16 * any later version.
17 *
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.
23 *
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.
28 *
29 * Please send any bug reports or fixes you make to the
30 * email address(es):
31 * lksctp developers <lksctp-developers@lists.sourceforge.net>
32 *
33 * Or submit a bug report through the following website:
34 * http://www.sf.net/projects/lksctp
35 *
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>
46 *
47 * Any bugs reported given to us we will try to fix... any fixes shared will
48 * be incorporated into the next SCTP release.
49 */
51 #include <linux/types.h>
52 #include <linux/fcntl.h>
53 #include <linux/poll.h>
54 #include <linux/init.h>
55 #include <linux/sched.h>
57 #include <linux/slab.h>
58 #include <linux/in.h>
59 #include <net/ipv6.h>
60 #include <net/sctp/sctp.h>
61 #include <net/sctp/sm.h>
63 /* Forward declarations for internal functions. */
64 static void sctp_assoc_bh_rcv(struct sctp_association *asoc);
67 /* 1st Level Abstractions. */
69 /* Initialize a new association from provided memory. */
70 static struct sctp_association *sctp_association_init(struct sctp_association *asoc,
71 const struct sctp_endpoint *ep,
72 const struct sock *sk,
73 sctp_scope_t scope,
74 gfp_t gfp)
75 {
76 struct sctp_sock *sp;
77 int i;
79 /* Retrieve the SCTP per socket area. */
80 sp = sctp_sk((struct sock *)sk);
82 /* Init all variables to a known value. */
83 memset(asoc, 0, sizeof(struct sctp_association));
85 /* Discarding const is appropriate here. */
86 asoc->ep = (struct sctp_endpoint *)ep;
87 sctp_endpoint_hold(asoc->ep);
89 /* Hold the sock. */
90 asoc->base.sk = (struct sock *)sk;
91 sock_hold(asoc->base.sk);
93 /* Initialize the common base substructure. */
94 asoc->base.type = SCTP_EP_TYPE_ASSOCIATION;
96 /* Initialize the object handling fields. */
97 atomic_set(&asoc->base.refcnt, 1);
98 asoc->base.dead = 0;
99 asoc->base.malloced = 0;
101 /* Initialize the bind addr area. */
102 sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port);
103 rwlock_init(&asoc->base.addr_lock);
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.
109 */
110 asoc->cookie_life.tv_sec = sp->assocparams.sasoc_cookie_life / 1000;
111 asoc->cookie_life.tv_usec = (sp->assocparams.sasoc_cookie_life % 1000)
112 * 1000;
113 asoc->frag_point = 0;
115 /* Set the association max_retrans and RTO values from the
116 * socket values.
117 */
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.
127 */
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.
141 */
142 asoc->param_flags = sp->param_flags;
144 /* Initialize the maximum mumber of new data packets that can be sent
145 * in a burst.
146 */
147 asoc->max_burst = sctp_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'.
160 */
161 asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
162 = 5 * asoc->rto_max;
164 asoc->timeouts[SCTP_EVENT_TIMEOUT_HEARTBEAT] = 0;
165 asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
166 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] =
167 sp->autoclose * HZ;
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;
174 }
176 /* Pull default initialization values from the sock options.
177 * Note: This assumes that the values have already been
178 * validated in the sock.
179 */
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.
189 */
190 asoc->ssnmap = NULL;
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.
196 */
197 if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
198 asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
199 else
200 asoc->rwnd = sk->sk_rcvbuf/2;
202 asoc->a_rwnd = asoc->rwnd;
204 asoc->rwnd_over = 0;
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;
220 asoc->c.my_ttag = 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
235 *
236 * When an endpoint has an ASCONF signaled change to be sent to the
237 * remote endpoint it should do the following:
238 * ...
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.
243 */
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
253 *
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
258 * 6.2.
259 *
260 * [We implement this by telling a new association that it
261 * already received one packet.]
262 */
263 asoc->peer.sack_needed = 1;
265 /* Assume that the peer recongizes ASCONF until reported otherwise
266 * via an ERROR chunk.
267 */
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,
273 (void (*)(void *))sctp_assoc_bh_rcv,
274 asoc);
276 /* Create an output queue. */
277 sctp_outq_init(asoc, &asoc->outqueue);
279 if (!sctp_ulpq_init(&asoc->ulpq, asoc))
280 goto fail_init;
282 /* Set up the tsn tracking. */
283 sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_SIZE, 0);
285 asoc->need_ecne = 0;
287 asoc->assoc_id = 0;
289 /* Assume that peer would support both address types unless we are
290 * told otherwise.
291 */
292 asoc->peer.ipv4_address = 1;
293 asoc->peer.ipv6_address = 1;
294 INIT_LIST_HEAD(&asoc->asocs);
296 asoc->autoclose = sp->autoclose;
298 asoc->default_stream = sp->default_stream;
299 asoc->default_ppid = sp->default_ppid;
300 asoc->default_flags = sp->default_flags;
301 asoc->default_context = sp->default_context;
302 asoc->default_timetolive = sp->default_timetolive;
304 return asoc;
306 fail_init:
307 sctp_endpoint_put(asoc->ep);
308 sock_put(asoc->base.sk);
309 return NULL;
310 }
312 /* Allocate and initialize a new association */
313 struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
314 const struct sock *sk,
315 sctp_scope_t scope,
316 gfp_t gfp)
317 {
318 struct sctp_association *asoc;
320 asoc = t_new(struct sctp_association, gfp);
321 if (!asoc)
322 goto fail;
324 if (!sctp_association_init(asoc, ep, sk, scope, gfp))
325 goto fail_init;
327 asoc->base.malloced = 1;
328 SCTP_DBG_OBJCNT_INC(assoc);
329 SCTP_DEBUG_PRINTK("Created asoc %p\n", asoc);
331 return asoc;
333 fail_init:
334 kfree(asoc);
335 fail:
336 return NULL;
337 }
339 /* Free this association if possible. There may still be users, so
340 * the actual deallocation may be delayed.
341 */
342 void sctp_association_free(struct sctp_association *asoc)
343 {
344 struct sock *sk = asoc->base.sk;
345 struct sctp_transport *transport;
346 struct list_head *pos, *temp;
347 int i;
349 list_del(&asoc->asocs);
351 /* Decrement the backlog value for a TCP-style listening socket. */
352 if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
353 sk->sk_ack_backlog--;
355 /* Mark as dead, so other users can know this structure is
356 * going away.
357 */
358 asoc->base.dead = 1;
360 /* Dispose of any data lying around in the outqueue. */
361 sctp_outq_free(&asoc->outqueue);
363 /* Dispose of any pending messages for the upper layer. */
364 sctp_ulpq_free(&asoc->ulpq);
366 /* Dispose of any pending chunks on the inqueue. */
367 sctp_inq_free(&asoc->base.inqueue);
369 /* Free ssnmap storage. */
370 sctp_ssnmap_free(asoc->ssnmap);
372 /* Clean up the bound address list. */
373 sctp_bind_addr_free(&asoc->base.bind_addr);
375 /* Do we need to go through all of our timers and
376 * delete them? To be safe we will try to delete all, but we
377 * should be able to go through and make a guess based
378 * on our state.
379 */
380 for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
381 if (timer_pending(&asoc->timers[i]) &&
382 del_timer(&asoc->timers[i]))
383 sctp_association_put(asoc);
384 }
386 /* Free peer's cached cookie. */
387 kfree(asoc->peer.cookie);
389 /* Release the transport structures. */
390 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
391 transport = list_entry(pos, struct sctp_transport, transports);
392 list_del(pos);
393 sctp_transport_free(transport);
394 }
396 asoc->peer.transport_count = 0;
398 /* Free any cached ASCONF_ACK chunk. */
399 if (asoc->addip_last_asconf_ack)
400 sctp_chunk_free(asoc->addip_last_asconf_ack);
402 /* Free any cached ASCONF chunk. */
403 if (asoc->addip_last_asconf)
404 sctp_chunk_free(asoc->addip_last_asconf);
406 sctp_association_put(asoc);
407 }
409 /* Cleanup and free up an association. */
410 static void sctp_association_destroy(struct sctp_association *asoc)
411 {
412 SCTP_ASSERT(asoc->base.dead, "Assoc is not dead", return);
414 sctp_endpoint_put(asoc->ep);
415 sock_put(asoc->base.sk);
417 if (asoc->assoc_id != 0) {
418 spin_lock_bh(&sctp_assocs_id_lock);
419 idr_remove(&sctp_assocs_id, asoc->assoc_id);
420 spin_unlock_bh(&sctp_assocs_id_lock);
421 }
423 BUG_TRAP(!atomic_read(&asoc->rmem_alloc));
425 if (asoc->base.malloced) {
426 kfree(asoc);
427 SCTP_DBG_OBJCNT_DEC(assoc);
428 }
429 }
431 /* Change the primary destination address for the peer. */
432 void sctp_assoc_set_primary(struct sctp_association *asoc,
433 struct sctp_transport *transport)
434 {
435 asoc->peer.primary_path = transport;
437 /* Set a default msg_name for events. */
438 memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
439 sizeof(union sctp_addr));
441 /* If the primary path is changing, assume that the
442 * user wants to use this new path.
443 */
444 if ((transport->state == SCTP_ACTIVE) ||
445 (transport->state == SCTP_UNKNOWN))
446 asoc->peer.active_path = transport;
448 /*
449 * SFR-CACC algorithm:
450 * Upon the receipt of a request to change the primary
451 * destination address, on the data structure for the new
452 * primary destination, the sender MUST do the following:
453 *
454 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
455 * to this destination address earlier. The sender MUST set
456 * CYCLING_CHANGEOVER to indicate that this switch is a
457 * double switch to the same destination address.
458 */
459 if (transport->cacc.changeover_active)
460 transport->cacc.cycling_changeover = 1;
462 /* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
463 * a changeover has occurred.
464 */
465 transport->cacc.changeover_active = 1;
467 /* 3) The sender MUST store the next TSN to be sent in
468 * next_tsn_at_change.
469 */
470 transport->cacc.next_tsn_at_change = asoc->next_tsn;
471 }
473 /* Remove a transport from an association. */
474 void sctp_assoc_rm_peer(struct sctp_association *asoc,
475 struct sctp_transport *peer)
476 {
477 struct list_head *pos;
478 struct sctp_transport *transport;
480 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_rm_peer:association %p addr: ",
481 " port: %d\n",
482 asoc,
483 (&peer->ipaddr),
484 peer->ipaddr.v4.sin_port);
486 /* If we are to remove the current retran_path, update it
487 * to the next peer before removing this peer from the list.
488 */
489 if (asoc->peer.retran_path == peer)
490 sctp_assoc_update_retran_path(asoc);
492 /* Remove this peer from the list. */
493 list_del(&peer->transports);
495 /* Get the first transport of asoc. */
496 pos = asoc->peer.transport_addr_list.next;
497 transport = list_entry(pos, struct sctp_transport, transports);
499 /* Update any entries that match the peer to be deleted. */
500 if (asoc->peer.primary_path == peer)
501 sctp_assoc_set_primary(asoc, transport);
502 if (asoc->peer.active_path == peer)
503 asoc->peer.active_path = transport;
504 if (asoc->peer.last_data_from == peer)
505 asoc->peer.last_data_from = transport;
507 /* If we remove the transport an INIT was last sent to, set it to
508 * NULL. Combined with the update of the retran path above, this
509 * will cause the next INIT to be sent to the next available
510 * transport, maintaining the cycle.
511 */
512 if (asoc->init_last_sent_to == peer)
513 asoc->init_last_sent_to = NULL;
515 asoc->peer.transport_count--;
517 sctp_transport_free(peer);
518 }
520 /* Add a transport address to an association. */
521 struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
522 const union sctp_addr *addr,
523 const gfp_t gfp,
524 const int peer_state)
525 {
526 struct sctp_transport *peer;
527 struct sctp_sock *sp;
528 unsigned short port;
530 sp = sctp_sk(asoc->base.sk);
532 /* AF_INET and AF_INET6 share common port field. */
533 port = addr->v4.sin_port;
535 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_add_peer:association %p addr: ",
536 " port: %d state:%d\n",
537 asoc,
538 addr,
539 addr->v4.sin_port,
540 peer_state);
542 /* Set the port if it has not been set yet. */
543 if (0 == asoc->peer.port)
544 asoc->peer.port = port;
546 /* Check to see if this is a duplicate. */
547 peer = sctp_assoc_lookup_paddr(asoc, addr);
548 if (peer) {
549 if (peer->state == SCTP_UNKNOWN) {
550 if (peer_state == SCTP_ACTIVE)
551 peer->state = SCTP_ACTIVE;
552 if (peer_state == SCTP_UNCONFIRMED)
553 peer->state = SCTP_UNCONFIRMED;
554 }
555 return peer;
556 }
558 peer = sctp_transport_new(addr, gfp);
559 if (!peer)
560 return NULL;
562 sctp_transport_set_owner(peer, asoc);
564 /* Initialize the peer's heartbeat interval based on the
565 * association configured value.
566 */
567 peer->hbinterval = asoc->hbinterval;
569 /* Set the path max_retrans. */
570 peer->pathmaxrxt = asoc->pathmaxrxt;
572 /* Initialize the peer's SACK delay timeout based on the
573 * association configured value.
574 */
575 peer->sackdelay = asoc->sackdelay;
577 /* Enable/disable heartbeat, SACK delay, and path MTU discovery
578 * based on association setting.
579 */
580 peer->param_flags = asoc->param_flags;
582 /* Initialize the pmtu of the transport. */
583 if (peer->param_flags & SPP_PMTUD_ENABLE)
584 sctp_transport_pmtu(peer);
585 else if (asoc->pathmtu)
586 peer->pathmtu = asoc->pathmtu;
587 else
588 peer->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
590 /* If this is the first transport addr on this association,
591 * initialize the association PMTU to the peer's PMTU.
592 * If not and the current association PMTU is higher than the new
593 * peer's PMTU, reset the association PMTU to the new peer's PMTU.
594 */
595 if (asoc->pathmtu)
596 asoc->pathmtu = min_t(int, peer->pathmtu, asoc->pathmtu);
597 else
598 asoc->pathmtu = peer->pathmtu;
600 SCTP_DEBUG_PRINTK("sctp_assoc_add_peer:association %p PMTU set to "
601 "%d\n", asoc, asoc->pathmtu);
603 asoc->frag_point = sctp_frag_point(sp, asoc->pathmtu);
605 /* The asoc->peer.port might not be meaningful yet, but
606 * initialize the packet structure anyway.
607 */
608 sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
609 asoc->peer.port);
611 /* 7.2.1 Slow-Start
612 *
613 * o The initial cwnd before DATA transmission or after a sufficiently
614 * long idle period MUST be set to
615 * min(4*MTU, max(2*MTU, 4380 bytes))
616 *
617 * o The initial value of ssthresh MAY be arbitrarily high
618 * (for example, implementations MAY use the size of the
619 * receiver advertised window).
620 */
621 peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
623 /* At this point, we may not have the receiver's advertised window,
624 * so initialize ssthresh to the default value and it will be set
625 * later when we process the INIT.
626 */
627 peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
629 peer->partial_bytes_acked = 0;
630 peer->flight_size = 0;
632 /* Set the transport's RTO.initial value */
633 peer->rto = asoc->rto_initial;
635 /* Set the peer's active state. */
636 peer->state = peer_state;
638 /* Attach the remote transport to our asoc. */
639 list_add_tail(&peer->transports, &asoc->peer.transport_addr_list);
640 asoc->peer.transport_count++;
642 /* If we do not yet have a primary path, set one. */
643 if (!asoc->peer.primary_path) {
644 sctp_assoc_set_primary(asoc, peer);
645 asoc->peer.retran_path = peer;
646 }
648 if (asoc->peer.active_path == asoc->peer.retran_path) {
649 asoc->peer.retran_path = peer;
650 }
652 return peer;
653 }
655 /* Delete a transport address from an association. */
656 void sctp_assoc_del_peer(struct sctp_association *asoc,
657 const union sctp_addr *addr)
658 {
659 struct list_head *pos;
660 struct list_head *temp;
661 struct sctp_transport *transport;
663 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
664 transport = list_entry(pos, struct sctp_transport, transports);
665 if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
666 /* Do book keeping for removing the peer and free it. */
667 sctp_assoc_rm_peer(asoc, transport);
668 break;
669 }
670 }
671 }
673 /* Lookup a transport by address. */
674 struct sctp_transport *sctp_assoc_lookup_paddr(
675 const struct sctp_association *asoc,
676 const union sctp_addr *address)
677 {
678 struct sctp_transport *t;
679 struct list_head *pos;
681 /* Cycle through all transports searching for a peer address. */
683 list_for_each(pos, &asoc->peer.transport_addr_list) {
684 t = list_entry(pos, struct sctp_transport, transports);
685 if (sctp_cmp_addr_exact(address, &t->ipaddr))
686 return t;
687 }
689 return NULL;
690 }
692 /* Engage in transport control operations.
693 * Mark the transport up or down and send a notification to the user.
694 * Select and update the new active and retran paths.
695 */
696 void sctp_assoc_control_transport(struct sctp_association *asoc,
697 struct sctp_transport *transport,
698 sctp_transport_cmd_t command,
699 sctp_sn_error_t error)
700 {
701 struct sctp_transport *t = NULL;
702 struct sctp_transport *first;
703 struct sctp_transport *second;
704 struct sctp_ulpevent *event;
705 struct list_head *pos;
706 int spc_state = 0;
708 /* Record the transition on the transport. */
709 switch (command) {
710 case SCTP_TRANSPORT_UP:
711 transport->state = SCTP_ACTIVE;
712 spc_state = SCTP_ADDR_AVAILABLE;
713 break;
715 case SCTP_TRANSPORT_DOWN:
716 transport->state = SCTP_INACTIVE;
717 spc_state = SCTP_ADDR_UNREACHABLE;
718 break;
720 default:
721 return;
722 };
724 /* Generate and send a SCTP_PEER_ADDR_CHANGE notification to the
725 * user.
726 */
727 event = sctp_ulpevent_make_peer_addr_change(asoc,
728 (struct sockaddr_storage *) &transport->ipaddr,
729 0, spc_state, error, GFP_ATOMIC);
730 if (event)
731 sctp_ulpq_tail_event(&asoc->ulpq, event);
733 /* Select new active and retran paths. */
735 /* Look for the two most recently used active transports.
736 *
737 * This code produces the wrong ordering whenever jiffies
738 * rolls over, but we still get usable transports, so we don't
739 * worry about it.
740 */
741 first = NULL; second = NULL;
743 list_for_each(pos, &asoc->peer.transport_addr_list) {
744 t = list_entry(pos, struct sctp_transport, transports);
746 if ((t->state == SCTP_INACTIVE) ||
747 (t->state == SCTP_UNCONFIRMED))
748 continue;
749 if (!first || t->last_time_heard > first->last_time_heard) {
750 second = first;
751 first = t;
752 }
753 if (!second || t->last_time_heard > second->last_time_heard)
754 second = t;
755 }
757 /* RFC 2960 6.4 Multi-Homed SCTP Endpoints
758 *
759 * By default, an endpoint should always transmit to the
760 * primary path, unless the SCTP user explicitly specifies the
761 * destination transport address (and possibly source
762 * transport address) to use.
763 *
764 * [If the primary is active but not most recent, bump the most
765 * recently used transport.]
766 */
767 if (((asoc->peer.primary_path->state == SCTP_ACTIVE) ||
768 (asoc->peer.primary_path->state == SCTP_UNKNOWN)) &&
769 first != asoc->peer.primary_path) {
770 second = first;
771 first = asoc->peer.primary_path;
772 }
774 /* If we failed to find a usable transport, just camp on the
775 * primary, even if it is inactive.
776 */
777 if (!first) {
778 first = asoc->peer.primary_path;
779 second = asoc->peer.primary_path;
780 }
782 /* Set the active and retran transports. */
783 asoc->peer.active_path = first;
784 asoc->peer.retran_path = second;
785 }
787 /* Hold a reference to an association. */
788 void sctp_association_hold(struct sctp_association *asoc)
789 {
790 atomic_inc(&asoc->base.refcnt);
791 }
793 /* Release a reference to an association and cleanup
794 * if there are no more references.
795 */
796 void sctp_association_put(struct sctp_association *asoc)
797 {
798 if (atomic_dec_and_test(&asoc->base.refcnt))
799 sctp_association_destroy(asoc);
800 }
802 /* Allocate the next TSN, Transmission Sequence Number, for the given
803 * association.
804 */
805 __u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
806 {
807 /* From Section 1.6 Serial Number Arithmetic:
808 * Transmission Sequence Numbers wrap around when they reach
809 * 2**32 - 1. That is, the next TSN a DATA chunk MUST use
810 * after transmitting TSN = 2*32 - 1 is TSN = 0.
811 */
812 __u32 retval = asoc->next_tsn;
813 asoc->next_tsn++;
814 asoc->unack_data++;
816 return retval;
817 }
819 /* Compare two addresses to see if they match. Wildcard addresses
820 * only match themselves.
821 */
822 int sctp_cmp_addr_exact(const union sctp_addr *ss1,
823 const union sctp_addr *ss2)
824 {
825 struct sctp_af *af;
827 af = sctp_get_af_specific(ss1->sa.sa_family);
828 if (unlikely(!af))
829 return 0;
831 return af->cmp_addr(ss1, ss2);
832 }
834 /* Return an ecne chunk to get prepended to a packet.
835 * Note: We are sly and return a shared, prealloced chunk. FIXME:
836 * No we don't, but we could/should.
837 */
838 struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
839 {
840 struct sctp_chunk *chunk;
842 /* Send ECNE if needed.
843 * Not being able to allocate a chunk here is not deadly.
844 */
845 if (asoc->need_ecne)
846 chunk = sctp_make_ecne(asoc, asoc->last_ecne_tsn);
847 else
848 chunk = NULL;
850 return chunk;
851 }
853 /*
854 * Find which transport this TSN was sent on.
855 */
856 struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
857 __u32 tsn)
858 {
859 struct sctp_transport *active;
860 struct sctp_transport *match;
861 struct list_head *entry, *pos;
862 struct sctp_transport *transport;
863 struct sctp_chunk *chunk;
864 __u32 key = htonl(tsn);
866 match = NULL;
868 /*
869 * FIXME: In general, find a more efficient data structure for
870 * searching.
871 */
873 /*
874 * The general strategy is to search each transport's transmitted
875 * list. Return which transport this TSN lives on.
876 *
877 * Let's be hopeful and check the active_path first.
878 * Another optimization would be to know if there is only one
879 * outbound path and not have to look for the TSN at all.
880 *
881 */
883 active = asoc->peer.active_path;
885 list_for_each(entry, &active->transmitted) {
886 chunk = list_entry(entry, struct sctp_chunk, transmitted_list);
888 if (key == chunk->subh.data_hdr->tsn) {
889 match = active;
890 goto out;
891 }
892 }
894 /* If not found, go search all the other transports. */
895 list_for_each(pos, &asoc->peer.transport_addr_list) {
896 transport = list_entry(pos, struct sctp_transport, transports);
898 if (transport == active)
899 break;
900 list_for_each(entry, &transport->transmitted) {
901 chunk = list_entry(entry, struct sctp_chunk,
902 transmitted_list);
903 if (key == chunk->subh.data_hdr->tsn) {
904 match = transport;
905 goto out;
906 }
907 }
908 }
909 out:
910 return match;
911 }
913 /* Is this the association we are looking for? */
914 struct sctp_transport *sctp_assoc_is_match(struct sctp_association *asoc,
915 const union sctp_addr *laddr,
916 const union sctp_addr *paddr)
917 {
918 struct sctp_transport *transport;
920 sctp_read_lock(&asoc->base.addr_lock);
922 if ((asoc->base.bind_addr.port == laddr->v4.sin_port) &&
923 (asoc->peer.port == paddr->v4.sin_port)) {
924 transport = sctp_assoc_lookup_paddr(asoc, paddr);
925 if (!transport)
926 goto out;
928 if (sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
929 sctp_sk(asoc->base.sk)))
930 goto out;
931 }
932 transport = NULL;
934 out:
935 sctp_read_unlock(&asoc->base.addr_lock);
936 return transport;
937 }
939 /* Do delayed input processing. This is scheduled by sctp_rcv(). */
940 static void sctp_assoc_bh_rcv(struct sctp_association *asoc)
941 {
942 struct sctp_endpoint *ep;
943 struct sctp_chunk *chunk;
944 struct sock *sk;
945 struct sctp_inq *inqueue;
946 int state;
947 sctp_subtype_t subtype;
948 int error = 0;
950 /* The association should be held so we should be safe. */
951 ep = asoc->ep;
952 sk = asoc->base.sk;
954 inqueue = &asoc->base.inqueue;
955 sctp_association_hold(asoc);
956 while (NULL != (chunk = sctp_inq_pop(inqueue))) {
957 state = asoc->state;
958 subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
960 /* Remember where the last DATA chunk came from so we
961 * know where to send the SACK.
962 */
963 if (sctp_chunk_is_data(chunk))
964 asoc->peer.last_data_from = chunk->transport;
965 else
966 SCTP_INC_STATS(SCTP_MIB_INCTRLCHUNKS);
968 if (chunk->transport)
969 chunk->transport->last_time_heard = jiffies;
971 /* Run through the state machine. */
972 error = sctp_do_sm(SCTP_EVENT_T_CHUNK, subtype,
973 state, ep, asoc, chunk, GFP_ATOMIC);
975 /* Check to see if the association is freed in response to
976 * the incoming chunk. If so, get out of the while loop.
977 */
978 if (asoc->base.dead)
979 break;
981 /* If there is an error on chunk, discard this packet. */
982 if (error && chunk)
983 chunk->pdiscard = 1;
984 }
985 sctp_association_put(asoc);
986 }
988 /* This routine moves an association from its old sk to a new sk. */
989 void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
990 {
991 struct sctp_sock *newsp = sctp_sk(newsk);
992 struct sock *oldsk = assoc->base.sk;
994 /* Delete the association from the old endpoint's list of
995 * associations.
996 */
997 list_del_init(&assoc->asocs);
999 /* Decrement the backlog value for a TCP-style socket. */
1000 if (sctp_style(oldsk, TCP))
1001 oldsk->sk_ack_backlog--;
1003 /* Release references to the old endpoint and the sock. */
1004 sctp_endpoint_put(assoc->ep);
1005 sock_put(assoc->base.sk);
1007 /* Get a reference to the new endpoint. */
1008 assoc->ep = newsp->ep;
1009 sctp_endpoint_hold(assoc->ep);
1011 /* Get a reference to the new sock. */
1012 assoc->base.sk = newsk;
1013 sock_hold(assoc->base.sk);
1015 /* Add the association to the new endpoint's list of associations. */
1016 sctp_endpoint_add_asoc(newsp->ep, assoc);
1019 /* Update an association (possibly from unexpected COOKIE-ECHO processing). */
1020 void sctp_assoc_update(struct sctp_association *asoc,
1021 struct sctp_association *new)
1023 struct sctp_transport *trans;
1024 struct list_head *pos, *temp;
1026 /* Copy in new parameters of peer. */
1027 asoc->c = new->c;
1028 asoc->peer.rwnd = new->peer.rwnd;
1029 asoc->peer.sack_needed = new->peer.sack_needed;
1030 asoc->peer.i = new->peer.i;
1031 sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_SIZE,
1032 asoc->peer.i.initial_tsn);
1034 /* Remove any peer addresses not present in the new association. */
1035 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1036 trans = list_entry(pos, struct sctp_transport, transports);
1037 if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr))
1038 sctp_assoc_del_peer(asoc, &trans->ipaddr);
1041 /* If the case is A (association restart), use
1042 * initial_tsn as next_tsn. If the case is B, use
1043 * current next_tsn in case data sent to peer
1044 * has been discarded and needs retransmission.
1045 */
1046 if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1047 asoc->next_tsn = new->next_tsn;
1048 asoc->ctsn_ack_point = new->ctsn_ack_point;
1049 asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1051 /* Reinitialize SSN for both local streams
1052 * and peer's streams.
1053 */
1054 sctp_ssnmap_clear(asoc->ssnmap);
1056 } else {
1057 /* Add any peer addresses from the new association. */
1058 list_for_each(pos, &new->peer.transport_addr_list) {
1059 trans = list_entry(pos, struct sctp_transport,
1060 transports);
1061 if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr))
1062 sctp_assoc_add_peer(asoc, &trans->ipaddr,
1063 GFP_ATOMIC, trans->state);
1066 asoc->ctsn_ack_point = asoc->next_tsn - 1;
1067 asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1068 if (!asoc->ssnmap) {
1069 /* Move the ssnmap. */
1070 asoc->ssnmap = new->ssnmap;
1071 new->ssnmap = NULL;
1076 /* Update the retran path for sending a retransmitted packet.
1077 * Round-robin through the active transports, else round-robin
1078 * through the inactive transports as this is the next best thing
1079 * we can try.
1080 */
1081 void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1083 struct sctp_transport *t, *next;
1084 struct list_head *head = &asoc->peer.transport_addr_list;
1085 struct list_head *pos;
1087 /* Find the next transport in a round-robin fashion. */
1088 t = asoc->peer.retran_path;
1089 pos = &t->transports;
1090 next = NULL;
1092 while (1) {
1093 /* Skip the head. */
1094 if (pos->next == head)
1095 pos = head->next;
1096 else
1097 pos = pos->next;
1099 t = list_entry(pos, struct sctp_transport, transports);
1101 /* Try to find an active transport. */
1103 if ((t->state == SCTP_ACTIVE) ||
1104 (t->state == SCTP_UNKNOWN)) {
1105 break;
1106 } else {
1107 /* Keep track of the next transport in case
1108 * we don't find any active transport.
1109 */
1110 if (!next)
1111 next = t;
1114 /* We have exhausted the list, but didn't find any
1115 * other active transports. If so, use the next
1116 * transport.
1117 */
1118 if (t == asoc->peer.retran_path) {
1119 t = next;
1120 break;
1124 asoc->peer.retran_path = t;
1126 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_update_retran_path:association"
1127 " %p addr: ",
1128 " port: %d\n",
1129 asoc,
1130 (&t->ipaddr),
1131 t->ipaddr.v4.sin_port);
1134 /* Choose the transport for sending a INIT packet. */
1135 struct sctp_transport *sctp_assoc_choose_init_transport(
1136 struct sctp_association *asoc)
1138 struct sctp_transport *t;
1140 /* Use the retran path. If the last INIT was sent over the
1141 * retran path, update the retran path and use it.
1142 */
1143 if (!asoc->init_last_sent_to) {
1144 t = asoc->peer.active_path;
1145 } else {
1146 if (asoc->init_last_sent_to == asoc->peer.retran_path)
1147 sctp_assoc_update_retran_path(asoc);
1148 t = asoc->peer.retran_path;
1151 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_update_retran_path:association"
1152 " %p addr: ",
1153 " port: %d\n",
1154 asoc,
1155 (&t->ipaddr),
1156 t->ipaddr.v4.sin_port);
1158 return t;
1161 /* Choose the transport for sending a SHUTDOWN packet. */
1162 struct sctp_transport *sctp_assoc_choose_shutdown_transport(
1163 struct sctp_association *asoc)
1165 /* If this is the first time SHUTDOWN is sent, use the active path,
1166 * else use the retran path. If the last SHUTDOWN was sent over the
1167 * retran path, update the retran path and use it.
1168 */
1169 if (!asoc->shutdown_last_sent_to)
1170 return asoc->peer.active_path;
1171 else {
1172 if (asoc->shutdown_last_sent_to == asoc->peer.retran_path)
1173 sctp_assoc_update_retran_path(asoc);
1174 return asoc->peer.retran_path;
1179 /* Update the association's pmtu and frag_point by going through all the
1180 * transports. This routine is called when a transport's PMTU has changed.
1181 */
1182 void sctp_assoc_sync_pmtu(struct sctp_association *asoc)
1184 struct sctp_transport *t;
1185 struct list_head *pos;
1186 __u32 pmtu = 0;
1188 if (!asoc)
1189 return;
1191 /* Get the lowest pmtu of all the transports. */
1192 list_for_each(pos, &asoc->peer.transport_addr_list) {
1193 t = list_entry(pos, struct sctp_transport, transports);
1194 if (!pmtu || (t->pathmtu < pmtu))
1195 pmtu = t->pathmtu;
1198 if (pmtu) {
1199 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
1200 asoc->pathmtu = pmtu;
1201 asoc->frag_point = sctp_frag_point(sp, pmtu);
1204 SCTP_DEBUG_PRINTK("%s: asoc:%p, pmtu:%d, frag_point:%d\n",
1205 __FUNCTION__, asoc, asoc->pathmtu, asoc->frag_point);
1208 /* Should we send a SACK to update our peer? */
1209 static inline int sctp_peer_needs_update(struct sctp_association *asoc)
1211 switch (asoc->state) {
1212 case SCTP_STATE_ESTABLISHED:
1213 case SCTP_STATE_SHUTDOWN_PENDING:
1214 case SCTP_STATE_SHUTDOWN_RECEIVED:
1215 case SCTP_STATE_SHUTDOWN_SENT:
1216 if ((asoc->rwnd > asoc->a_rwnd) &&
1217 ((asoc->rwnd - asoc->a_rwnd) >=
1218 min_t(__u32, (asoc->base.sk->sk_rcvbuf >> 1), asoc->pathmtu)))
1219 return 1;
1220 break;
1221 default:
1222 break;
1224 return 0;
1227 /* Increase asoc's rwnd by len and send any window update SACK if needed. */
1228 void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned len)
1230 struct sctp_chunk *sack;
1231 struct timer_list *timer;
1233 if (asoc->rwnd_over) {
1234 if (asoc->rwnd_over >= len) {
1235 asoc->rwnd_over -= len;
1236 } else {
1237 asoc->rwnd += (len - asoc->rwnd_over);
1238 asoc->rwnd_over = 0;
1240 } else {
1241 asoc->rwnd += len;
1244 SCTP_DEBUG_PRINTK("%s: asoc %p rwnd increased by %d to (%u, %u) "
1245 "- %u\n", __FUNCTION__, asoc, len, asoc->rwnd,
1246 asoc->rwnd_over, asoc->a_rwnd);
1248 /* Send a window update SACK if the rwnd has increased by at least the
1249 * minimum of the association's PMTU and half of the receive buffer.
1250 * The algorithm used is similar to the one described in
1251 * Section 4.2.3.3 of RFC 1122.
1252 */
1253 if (sctp_peer_needs_update(asoc)) {
1254 asoc->a_rwnd = asoc->rwnd;
1255 SCTP_DEBUG_PRINTK("%s: Sending window update SACK- asoc: %p "
1256 "rwnd: %u a_rwnd: %u\n", __FUNCTION__,
1257 asoc, asoc->rwnd, asoc->a_rwnd);
1258 sack = sctp_make_sack(asoc);
1259 if (!sack)
1260 return;
1262 asoc->peer.sack_needed = 0;
1264 sctp_outq_tail(&asoc->outqueue, sack);
1266 /* Stop the SACK timer. */
1267 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1268 if (timer_pending(timer) && del_timer(timer))
1269 sctp_association_put(asoc);
1273 /* Decrease asoc's rwnd by len. */
1274 void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned len)
1276 SCTP_ASSERT(asoc->rwnd, "rwnd zero", return);
1277 SCTP_ASSERT(!asoc->rwnd_over, "rwnd_over not zero", return);
1278 if (asoc->rwnd >= len) {
1279 asoc->rwnd -= len;
1280 } else {
1281 asoc->rwnd_over = len - asoc->rwnd;
1282 asoc->rwnd = 0;
1284 SCTP_DEBUG_PRINTK("%s: asoc %p rwnd decreased by %d to (%u, %u)\n",
1285 __FUNCTION__, asoc, len, asoc->rwnd,
1286 asoc->rwnd_over);
1289 /* Build the bind address list for the association based on info from the
1290 * local endpoint and the remote peer.
1291 */
1292 int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1293 gfp_t gfp)
1295 sctp_scope_t scope;
1296 int flags;
1298 /* Use scoping rules to determine the subset of addresses from
1299 * the endpoint.
1300 */
1301 scope = sctp_scope(&asoc->peer.active_path->ipaddr);
1302 flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1303 if (asoc->peer.ipv4_address)
1304 flags |= SCTP_ADDR4_PEERSUPP;
1305 if (asoc->peer.ipv6_address)
1306 flags |= SCTP_ADDR6_PEERSUPP;
1308 return sctp_bind_addr_copy(&asoc->base.bind_addr,
1309 &asoc->ep->base.bind_addr,
1310 scope, gfp, flags);
1313 /* Build the association's bind address list from the cookie. */
1314 int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1315 struct sctp_cookie *cookie,
1316 gfp_t gfp)
1318 int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length);
1319 int var_size3 = cookie->raw_addr_list_len;
1320 __u8 *raw = (__u8 *)cookie->peer_init + var_size2;
1322 return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1323 asoc->ep->base.bind_addr.port, gfp);
1326 /* Lookup laddr in the bind address list of an association. */
1327 int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1328 const union sctp_addr *laddr)
1330 int found;
1332 sctp_read_lock(&asoc->base.addr_lock);
1333 if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1334 sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1335 sctp_sk(asoc->base.sk))) {
1336 found = 1;
1337 goto out;
1340 found = 0;
1341 out:
1342 sctp_read_unlock(&asoc->base.addr_lock);
1343 return found;