1 /* SCTP kernel implementation
2 * Copyright (c) 1999-2000 Cisco, Inc.
3 * Copyright (c) 1999-2001 Motorola, Inc.
4 * Copyright (c) 2001-2003 International Business Machines Corp.
5 * Copyright (c) 2001 Intel Corp.
6 * Copyright (c) 2001 La Monte H.P. Yarroll
8 * This file is part of the SCTP kernel implementation
10 * This module provides the abstraction for an SCTP tranport representing
11 * a remote transport address. For local transport addresses, we just use
14 * This SCTP implementation is free software;
15 * you can redistribute it and/or modify it under the terms of
16 * the GNU General Public License as published by
17 * the Free Software Foundation; either version 2, or (at your option)
20 * This SCTP implementation is distributed in the hope that it
21 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
22 * ************************
23 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
24 * See the GNU General Public License for more details.
26 * You should have received a copy of the GNU General Public License
27 * along with GNU CC; see the file COPYING. If not, see
28 * <http://www.gnu.org/licenses/>.
30 * Please send any bug reports or fixes you make to the
32 * lksctp developers <linux-sctp@vger.kernel.org>
34 * Written or modified by:
35 * La Monte H.P. Yarroll <piggy@acm.org>
36 * Karl Knutson <karl@athena.chicago.il.us>
37 * Jon Grimm <jgrimm@us.ibm.com>
38 * Xingang Guo <xingang.guo@intel.com>
39 * Hui Huang <hui.huang@nokia.com>
40 * Sridhar Samudrala <sri@us.ibm.com>
41 * Ardelle Fan <ardelle.fan@intel.com>
44 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
46 #include <linux/slab.h>
47 #include <linux/types.h>
48 #include <linux/random.h>
49 #include <net/sctp/sctp.h>
50 #include <net/sctp/sm.h>
52 /* 1st Level Abstractions. */
54 /* Initialize a new transport from provided memory. */
55 static struct sctp_transport
*sctp_transport_init(struct net
*net
,
56 struct sctp_transport
*peer
,
57 const union sctp_addr
*addr
,
60 /* Copy in the address. */
62 peer
->af_specific
= sctp_get_af_specific(addr
->sa
.sa_family
);
63 memset(&peer
->saddr
, 0, sizeof(union sctp_addr
));
65 peer
->sack_generation
= 0;
67 /* From 6.3.1 RTO Calculation:
69 * C1) Until an RTT measurement has been made for a packet sent to the
70 * given destination transport address, set RTO to the protocol
71 * parameter 'RTO.Initial'.
73 peer
->rto
= msecs_to_jiffies(net
->sctp
.rto_initial
);
75 peer
->last_time_heard
= ktime_get();
76 peer
->last_time_ecne_reduced
= jiffies
;
78 peer
->param_flags
= SPP_HB_DISABLE
|
82 /* Initialize the default path max_retrans. */
83 peer
->pathmaxrxt
= net
->sctp
.max_retrans_path
;
84 peer
->pf_retrans
= net
->sctp
.pf_retrans
;
86 INIT_LIST_HEAD(&peer
->transmitted
);
87 INIT_LIST_HEAD(&peer
->send_ready
);
88 INIT_LIST_HEAD(&peer
->transports
);
90 setup_timer(&peer
->T3_rtx_timer
, sctp_generate_t3_rtx_event
,
92 setup_timer(&peer
->hb_timer
, sctp_generate_heartbeat_event
,
94 setup_timer(&peer
->proto_unreach_timer
,
95 sctp_generate_proto_unreach_event
, (unsigned long)peer
);
97 /* Initialize the 64-bit random nonce sent with heartbeat. */
98 get_random_bytes(&peer
->hb_nonce
, sizeof(peer
->hb_nonce
));
100 atomic_set(&peer
->refcnt
, 1);
105 /* Allocate and initialize a new transport. */
106 struct sctp_transport
*sctp_transport_new(struct net
*net
,
107 const union sctp_addr
*addr
,
110 struct sctp_transport
*transport
;
112 transport
= kzalloc(sizeof(*transport
), gfp
);
116 if (!sctp_transport_init(net
, transport
, addr
, gfp
))
119 SCTP_DBG_OBJCNT_INC(transport
);
130 /* This transport is no longer needed. Free up if possible, or
131 * delay until it last reference count.
133 void sctp_transport_free(struct sctp_transport
*transport
)
135 /* Try to delete the heartbeat timer. */
136 if (del_timer(&transport
->hb_timer
))
137 sctp_transport_put(transport
);
139 /* Delete the T3_rtx timer if it's active.
140 * There is no point in not doing this now and letting
141 * structure hang around in memory since we know
142 * the tranport is going away.
144 if (del_timer(&transport
->T3_rtx_timer
))
145 sctp_transport_put(transport
);
147 /* Delete the ICMP proto unreachable timer if it's active. */
148 if (del_timer(&transport
->proto_unreach_timer
))
149 sctp_association_put(transport
->asoc
);
151 sctp_transport_put(transport
);
154 static void sctp_transport_destroy_rcu(struct rcu_head
*head
)
156 struct sctp_transport
*transport
;
158 transport
= container_of(head
, struct sctp_transport
, rcu
);
160 dst_release(transport
->dst
);
162 SCTP_DBG_OBJCNT_DEC(transport
);
165 /* Destroy the transport data structure.
166 * Assumes there are no more users of this structure.
168 static void sctp_transport_destroy(struct sctp_transport
*transport
)
170 if (unlikely(atomic_read(&transport
->refcnt
))) {
171 WARN(1, "Attempt to destroy undead transport %p!\n", transport
);
175 sctp_packet_free(&transport
->packet
);
178 sctp_association_put(transport
->asoc
);
180 call_rcu(&transport
->rcu
, sctp_transport_destroy_rcu
);
183 /* Start T3_rtx timer if it is not already running and update the heartbeat
184 * timer. This routine is called every time a DATA chunk is sent.
186 void sctp_transport_reset_timers(struct sctp_transport
*transport
)
188 /* RFC 2960 6.3.2 Retransmission Timer Rules
190 * R1) Every time a DATA chunk is sent to any address(including a
191 * retransmission), if the T3-rtx timer of that address is not running
192 * start it running so that it will expire after the RTO of that
196 if (!timer_pending(&transport
->T3_rtx_timer
))
197 if (!mod_timer(&transport
->T3_rtx_timer
,
198 jiffies
+ transport
->rto
))
199 sctp_transport_hold(transport
);
201 /* When a data chunk is sent, reset the heartbeat interval. */
202 if (!mod_timer(&transport
->hb_timer
,
203 sctp_transport_timeout(transport
)))
204 sctp_transport_hold(transport
);
207 /* This transport has been assigned to an association.
208 * Initialize fields from the association or from the sock itself.
209 * Register the reference count in the association.
211 void sctp_transport_set_owner(struct sctp_transport
*transport
,
212 struct sctp_association
*asoc
)
214 transport
->asoc
= asoc
;
215 sctp_association_hold(asoc
);
218 /* Initialize the pmtu of a transport. */
219 void sctp_transport_pmtu(struct sctp_transport
*transport
, struct sock
*sk
)
221 /* If we don't have a fresh route, look one up */
222 if (!transport
->dst
|| transport
->dst
->obsolete
) {
223 dst_release(transport
->dst
);
224 transport
->af_specific
->get_dst(transport
, &transport
->saddr
,
228 if (transport
->dst
) {
229 transport
->pathmtu
= dst_mtu(transport
->dst
);
231 transport
->pathmtu
= SCTP_DEFAULT_MAXSEGMENT
;
234 void sctp_transport_update_pmtu(struct sock
*sk
, struct sctp_transport
*t
, u32 pmtu
)
236 struct dst_entry
*dst
;
238 if (unlikely(pmtu
< SCTP_DEFAULT_MINSEGMENT
)) {
239 pr_warn("%s: Reported pmtu %d too low, using default minimum of %d\n",
241 SCTP_DEFAULT_MINSEGMENT
);
242 /* Use default minimum segment size and disable
243 * pmtu discovery on this transport.
245 t
->pathmtu
= SCTP_DEFAULT_MINSEGMENT
;
250 dst
= sctp_transport_dst_check(t
);
252 t
->af_specific
->get_dst(t
, &t
->saddr
, &t
->fl
, sk
);
255 dst
->ops
->update_pmtu(dst
, sk
, NULL
, pmtu
);
257 dst
= sctp_transport_dst_check(t
);
259 t
->af_specific
->get_dst(t
, &t
->saddr
, &t
->fl
, sk
);
263 /* Caches the dst entry and source address for a transport's destination
266 void sctp_transport_route(struct sctp_transport
*transport
,
267 union sctp_addr
*saddr
, struct sctp_sock
*opt
)
269 struct sctp_association
*asoc
= transport
->asoc
;
270 struct sctp_af
*af
= transport
->af_specific
;
272 af
->get_dst(transport
, saddr
, &transport
->fl
, sctp_opt2sk(opt
));
275 memcpy(&transport
->saddr
, saddr
, sizeof(union sctp_addr
));
277 af
->get_saddr(opt
, transport
, &transport
->fl
);
279 if ((transport
->param_flags
& SPP_PMTUD_DISABLE
) && transport
->pathmtu
) {
282 if (transport
->dst
) {
283 transport
->pathmtu
= dst_mtu(transport
->dst
);
285 /* Initialize sk->sk_rcv_saddr, if the transport is the
286 * association's active path for getsockname().
288 if (asoc
&& (!asoc
->peer
.primary_path
||
289 (transport
== asoc
->peer
.active_path
)))
290 opt
->pf
->to_sk_saddr(&transport
->saddr
,
293 transport
->pathmtu
= SCTP_DEFAULT_MAXSEGMENT
;
296 /* Hold a reference to a transport. */
297 int sctp_transport_hold(struct sctp_transport
*transport
)
299 return atomic_add_unless(&transport
->refcnt
, 1, 0);
302 /* Release a reference to a transport and clean up
303 * if there are no more references.
305 void sctp_transport_put(struct sctp_transport
*transport
)
307 if (atomic_dec_and_test(&transport
->refcnt
))
308 sctp_transport_destroy(transport
);
311 /* Update transport's RTO based on the newly calculated RTT. */
312 void sctp_transport_update_rto(struct sctp_transport
*tp
, __u32 rtt
)
314 if (unlikely(!tp
->rto_pending
))
315 /* We should not be doing any RTO updates unless rto_pending is set. */
316 pr_debug("%s: rto_pending not set on transport %p!\n", __func__
, tp
);
318 if (tp
->rttvar
|| tp
->srtt
) {
319 struct net
*net
= sock_net(tp
->asoc
->base
.sk
);
320 /* 6.3.1 C3) When a new RTT measurement R' is made, set
321 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
322 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
325 /* Note: The above algorithm has been rewritten to
326 * express rto_beta and rto_alpha as inverse powers
328 * For example, assuming the default value of RTO.Alpha of
329 * 1/8, rto_alpha would be expressed as 3.
331 tp
->rttvar
= tp
->rttvar
- (tp
->rttvar
>> net
->sctp
.rto_beta
)
332 + (((__u32
)abs((__s64
)tp
->srtt
- (__s64
)rtt
)) >> net
->sctp
.rto_beta
);
333 tp
->srtt
= tp
->srtt
- (tp
->srtt
>> net
->sctp
.rto_alpha
)
334 + (rtt
>> net
->sctp
.rto_alpha
);
336 /* 6.3.1 C2) When the first RTT measurement R is made, set
337 * SRTT <- R, RTTVAR <- R/2.
340 tp
->rttvar
= rtt
>> 1;
343 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
344 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
347 tp
->rttvar
= SCTP_CLOCK_GRANULARITY
;
349 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
350 tp
->rto
= tp
->srtt
+ (tp
->rttvar
<< 2);
352 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
353 * seconds then it is rounded up to RTO.Min seconds.
355 if (tp
->rto
< tp
->asoc
->rto_min
)
356 tp
->rto
= tp
->asoc
->rto_min
;
358 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
359 * at least RTO.max seconds.
361 if (tp
->rto
> tp
->asoc
->rto_max
)
362 tp
->rto
= tp
->asoc
->rto_max
;
364 sctp_max_rto(tp
->asoc
, tp
);
367 /* Reset rto_pending so that a new RTT measurement is started when a
368 * new data chunk is sent.
372 pr_debug("%s: transport:%p, rtt:%d, srtt:%d rttvar:%d, rto:%ld\n",
373 __func__
, tp
, rtt
, tp
->srtt
, tp
->rttvar
, tp
->rto
);
376 /* This routine updates the transport's cwnd and partial_bytes_acked
377 * parameters based on the bytes acked in the received SACK.
379 void sctp_transport_raise_cwnd(struct sctp_transport
*transport
,
380 __u32 sack_ctsn
, __u32 bytes_acked
)
382 struct sctp_association
*asoc
= transport
->asoc
;
383 __u32 cwnd
, ssthresh
, flight_size
, pba
, pmtu
;
385 cwnd
= transport
->cwnd
;
386 flight_size
= transport
->flight_size
;
388 /* See if we need to exit Fast Recovery first */
389 if (asoc
->fast_recovery
&&
390 TSN_lte(asoc
->fast_recovery_exit
, sack_ctsn
))
391 asoc
->fast_recovery
= 0;
393 /* The appropriate cwnd increase algorithm is performed if, and only
394 * if the cumulative TSN whould advanced and the congestion window is
395 * being fully utilized.
397 if (TSN_lte(sack_ctsn
, transport
->asoc
->ctsn_ack_point
) ||
398 (flight_size
< cwnd
))
401 ssthresh
= transport
->ssthresh
;
402 pba
= transport
->partial_bytes_acked
;
403 pmtu
= transport
->asoc
->pathmtu
;
405 if (cwnd
<= ssthresh
) {
407 * o When cwnd is less than or equal to ssthresh, an SCTP
408 * endpoint MUST use the slow-start algorithm to increase
409 * cwnd only if the current congestion window is being fully
410 * utilized, an incoming SACK advances the Cumulative TSN
411 * Ack Point, and the data sender is not in Fast Recovery.
412 * Only when these three conditions are met can the cwnd be
413 * increased; otherwise, the cwnd MUST not be increased.
414 * If these conditions are met, then cwnd MUST be increased
415 * by, at most, the lesser of 1) the total size of the
416 * previously outstanding DATA chunk(s) acknowledged, and
417 * 2) the destination's path MTU. This upper bound protects
418 * against the ACK-Splitting attack outlined in [SAVAGE99].
420 if (asoc
->fast_recovery
)
423 if (bytes_acked
> pmtu
)
428 pr_debug("%s: slow start: transport:%p, bytes_acked:%d, "
429 "cwnd:%d, ssthresh:%d, flight_size:%d, pba:%d\n",
430 __func__
, transport
, bytes_acked
, cwnd
, ssthresh
,
433 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
434 * upon each SACK arrival that advances the Cumulative TSN Ack
435 * Point, increase partial_bytes_acked by the total number of
436 * bytes of all new chunks acknowledged in that SACK including
437 * chunks acknowledged by the new Cumulative TSN Ack and by
440 * When partial_bytes_acked is equal to or greater than cwnd
441 * and before the arrival of the SACK the sender had cwnd or
442 * more bytes of data outstanding (i.e., before arrival of the
443 * SACK, flightsize was greater than or equal to cwnd),
444 * increase cwnd by MTU, and reset partial_bytes_acked to
445 * (partial_bytes_acked - cwnd).
450 pba
= ((cwnd
< pba
) ? (pba
- cwnd
) : 0);
453 pr_debug("%s: congestion avoidance: transport:%p, "
454 "bytes_acked:%d, cwnd:%d, ssthresh:%d, "
455 "flight_size:%d, pba:%d\n", __func__
,
456 transport
, bytes_acked
, cwnd
, ssthresh
,
460 transport
->cwnd
= cwnd
;
461 transport
->partial_bytes_acked
= pba
;
464 /* This routine is used to lower the transport's cwnd when congestion is
467 void sctp_transport_lower_cwnd(struct sctp_transport
*transport
,
468 sctp_lower_cwnd_t reason
)
470 struct sctp_association
*asoc
= transport
->asoc
;
473 case SCTP_LOWER_CWND_T3_RTX
:
474 /* RFC 2960 Section 7.2.3, sctpimpguide
475 * When the T3-rtx timer expires on an address, SCTP should
476 * perform slow start by:
477 * ssthresh = max(cwnd/2, 4*MTU)
479 * partial_bytes_acked = 0
481 transport
->ssthresh
= max(transport
->cwnd
/2,
483 transport
->cwnd
= asoc
->pathmtu
;
485 /* T3-rtx also clears fast recovery */
486 asoc
->fast_recovery
= 0;
489 case SCTP_LOWER_CWND_FAST_RTX
:
490 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
491 * destination address(es) to which the missing DATA chunks
492 * were last sent, according to the formula described in
495 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
496 * losses from SACK (see Section 7.2.4), An endpoint
497 * should do the following:
498 * ssthresh = max(cwnd/2, 4*MTU)
500 * partial_bytes_acked = 0
502 if (asoc
->fast_recovery
)
505 /* Mark Fast recovery */
506 asoc
->fast_recovery
= 1;
507 asoc
->fast_recovery_exit
= asoc
->next_tsn
- 1;
509 transport
->ssthresh
= max(transport
->cwnd
/2,
511 transport
->cwnd
= transport
->ssthresh
;
514 case SCTP_LOWER_CWND_ECNE
:
515 /* RFC 2481 Section 6.1.2.
516 * If the sender receives an ECN-Echo ACK packet
517 * then the sender knows that congestion was encountered in the
518 * network on the path from the sender to the receiver. The
519 * indication of congestion should be treated just as a
520 * congestion loss in non-ECN Capable TCP. That is, the TCP
521 * source halves the congestion window "cwnd" and reduces the
522 * slow start threshold "ssthresh".
523 * A critical condition is that TCP does not react to
524 * congestion indications more than once every window of
525 * data (or more loosely more than once every round-trip time).
527 if (time_after(jiffies
, transport
->last_time_ecne_reduced
+
529 transport
->ssthresh
= max(transport
->cwnd
/2,
531 transport
->cwnd
= transport
->ssthresh
;
532 transport
->last_time_ecne_reduced
= jiffies
;
536 case SCTP_LOWER_CWND_INACTIVE
:
537 /* RFC 2960 Section 7.2.1, sctpimpguide
538 * When the endpoint does not transmit data on a given
539 * transport address, the cwnd of the transport address
540 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
541 * NOTE: Although the draft recommends that this check needs
542 * to be done every RTO interval, we do it every hearbeat
545 transport
->cwnd
= max(transport
->cwnd
/2,
550 transport
->partial_bytes_acked
= 0;
552 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d\n",
553 __func__
, transport
, reason
, transport
->cwnd
,
554 transport
->ssthresh
);
557 /* Apply Max.Burst limit to the congestion window:
558 * sctpimpguide-05 2.14.2
559 * D) When the time comes for the sender to
560 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
561 * first be applied to limit how many new DATA chunks may be sent.
562 * The limit is applied by adjusting cwnd as follows:
563 * if ((flightsize+ Max.Burst * MTU) < cwnd)
564 * cwnd = flightsize + Max.Burst * MTU
567 void sctp_transport_burst_limited(struct sctp_transport
*t
)
569 struct sctp_association
*asoc
= t
->asoc
;
570 u32 old_cwnd
= t
->cwnd
;
573 if (t
->burst_limited
|| asoc
->max_burst
== 0)
576 max_burst_bytes
= t
->flight_size
+ (asoc
->max_burst
* asoc
->pathmtu
);
577 if (max_burst_bytes
< old_cwnd
) {
578 t
->cwnd
= max_burst_bytes
;
579 t
->burst_limited
= old_cwnd
;
583 /* Restore the old cwnd congestion window, after the burst had it's
586 void sctp_transport_burst_reset(struct sctp_transport
*t
)
588 if (t
->burst_limited
) {
589 t
->cwnd
= t
->burst_limited
;
590 t
->burst_limited
= 0;
594 /* What is the next timeout value for this transport? */
595 unsigned long sctp_transport_timeout(struct sctp_transport
*trans
)
597 /* RTO + timer slack +/- 50% of RTO */
598 unsigned long timeout
= (trans
->rto
>> 1) + prandom_u32_max(trans
->rto
);
600 if (trans
->state
!= SCTP_UNCONFIRMED
&&
601 trans
->state
!= SCTP_PF
)
602 timeout
+= trans
->hbinterval
;
604 return timeout
+ jiffies
;
607 /* Reset transport variables to their initial values */
608 void sctp_transport_reset(struct sctp_transport
*t
)
610 struct sctp_association
*asoc
= t
->asoc
;
612 /* RFC 2960 (bis), Section 5.2.4
613 * All the congestion control parameters (e.g., cwnd, ssthresh)
614 * related to this peer MUST be reset to their initial values
615 * (see Section 6.2.1)
617 t
->cwnd
= min(4*asoc
->pathmtu
, max_t(__u32
, 2*asoc
->pathmtu
, 4380));
618 t
->burst_limited
= 0;
619 t
->ssthresh
= asoc
->peer
.i
.a_rwnd
;
620 t
->rto
= asoc
->rto_initial
;
621 sctp_max_rto(asoc
, t
);
626 /* Reset these additional varibles so that we have a clean
629 t
->partial_bytes_acked
= 0;
635 /* Initialize the state information for SFR-CACC */
636 t
->cacc
.changeover_active
= 0;
637 t
->cacc
.cycling_changeover
= 0;
638 t
->cacc
.next_tsn_at_change
= 0;
639 t
->cacc
.cacc_saw_newack
= 0;
642 /* Schedule retransmission on the given transport */
643 void sctp_transport_immediate_rtx(struct sctp_transport
*t
)
645 /* Stop pending T3_rtx_timer */
646 if (del_timer(&t
->T3_rtx_timer
))
647 sctp_transport_put(t
);
649 sctp_retransmit(&t
->asoc
->outqueue
, t
, SCTP_RTXR_T3_RTX
);
650 if (!timer_pending(&t
->T3_rtx_timer
)) {
651 if (!mod_timer(&t
->T3_rtx_timer
, jiffies
+ t
->rto
))
652 sctp_transport_hold(t
);