1 /* SCTP kernel implementation
2 * (C) Copyright IBM Corp. 2001, 2004
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
6 * This file is part of the SCTP kernel implementation
8 * These functions handle output processing.
10 * This SCTP implementation is free software;
11 * you can redistribute it and/or modify it under the terms of
12 * the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
16 * This SCTP implementation is distributed in the hope that it
17 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
18 * ************************
19 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
20 * See the GNU General Public License for more details.
22 * You should have received a copy of the GNU General Public License
23 * along with GNU CC; see the file COPYING. If not, write to
24 * the Free Software Foundation, 59 Temple Place - Suite 330,
25 * Boston, MA 02111-1307, USA.
27 * Please send any bug reports or fixes you make to the
29 * lksctp developers <lksctp-developers@lists.sourceforge.net>
31 * Or submit a bug report through the following website:
32 * http://www.sf.net/projects/lksctp
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@austin.ibm.com>
38 * Sridhar Samudrala <sri@us.ibm.com>
40 * Any bugs reported given to us we will try to fix... any fixes shared will
41 * be incorporated into the next SCTP release.
44 #include <linux/types.h>
45 #include <linux/kernel.h>
46 #include <linux/wait.h>
47 #include <linux/time.h>
49 #include <linux/ipv6.h>
50 #include <linux/init.h>
51 #include <net/inet_ecn.h>
53 #include <net/net_namespace.h>
57 #endif /* TEST_FRAME (not defined) */
59 #include <linux/socket.h> /* for sa_family_t */
62 #include <net/sctp/sctp.h>
63 #include <net/sctp/sm.h>
64 #include <net/sctp/checksum.h>
66 /* Forward declarations for private helpers. */
67 static sctp_xmit_t
sctp_packet_append_data(struct sctp_packet
*packet
,
68 struct sctp_chunk
*chunk
);
71 * This appears to be a followup set of initializations.
73 struct sctp_packet
*sctp_packet_config(struct sctp_packet
*packet
,
74 __u32 vtag
, int ecn_capable
)
76 struct sctp_chunk
*chunk
= NULL
;
78 SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __func__
,
82 packet
->has_cookie_echo
= 0;
89 if (ecn_capable
&& sctp_packet_empty(packet
)) {
90 chunk
= sctp_get_ecne_prepend(packet
->transport
->asoc
);
92 /* If there a is a prepend chunk stick it on the list before
93 * any other chunks get appended.
96 sctp_packet_append_chunk(packet
, chunk
);
102 /* Initialize the packet structure. */
103 struct sctp_packet
*sctp_packet_init(struct sctp_packet
*packet
,
104 struct sctp_transport
*transport
,
105 __u16 sport
, __u16 dport
)
107 struct sctp_association
*asoc
= transport
->asoc
;
110 SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __func__
,
113 packet
->transport
= transport
;
114 packet
->source_port
= sport
;
115 packet
->destination_port
= dport
;
116 INIT_LIST_HEAD(&packet
->chunk_list
);
118 struct sctp_sock
*sp
= sctp_sk(asoc
->base
.sk
);
119 overhead
= sp
->pf
->af
->net_header_len
;
121 overhead
= sizeof(struct ipv6hdr
);
123 overhead
+= sizeof(struct sctphdr
);
124 packet
->overhead
= overhead
;
125 packet
->size
= overhead
;
127 packet
->has_cookie_echo
= 0;
128 packet
->has_sack
= 0;
129 packet
->has_auth
= 0;
130 packet
->has_data
= 0;
131 packet
->ipfragok
= 0;
132 packet
->malloced
= 0;
138 void sctp_packet_free(struct sctp_packet
*packet
)
140 struct sctp_chunk
*chunk
, *tmp
;
142 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__
, packet
);
144 list_for_each_entry_safe(chunk
, tmp
, &packet
->chunk_list
, list
) {
145 list_del_init(&chunk
->list
);
146 sctp_chunk_free(chunk
);
149 if (packet
->malloced
)
153 /* This routine tries to append the chunk to the offered packet. If adding
154 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
155 * is not present in the packet, it transmits the input packet.
156 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
157 * as it can fit in the packet, but any more data that does not fit in this
158 * packet can be sent only after receiving the COOKIE_ACK.
160 sctp_xmit_t
sctp_packet_transmit_chunk(struct sctp_packet
*packet
,
161 struct sctp_chunk
*chunk
,
167 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__
,
170 switch ((retval
= (sctp_packet_append_chunk(packet
, chunk
)))) {
171 case SCTP_XMIT_PMTU_FULL
:
172 if (!packet
->has_cookie_echo
) {
173 error
= sctp_packet_transmit(packet
);
175 chunk
->skb
->sk
->sk_err
= -error
;
177 /* If we have an empty packet, then we can NOT ever
181 retval
= sctp_packet_append_chunk(packet
,
186 case SCTP_XMIT_RWND_FULL
:
188 case SCTP_XMIT_NAGLE_DELAY
:
195 /* Try to bundle an auth chunk into the packet. */
196 static sctp_xmit_t
sctp_packet_bundle_auth(struct sctp_packet
*pkt
,
197 struct sctp_chunk
*chunk
)
199 struct sctp_association
*asoc
= pkt
->transport
->asoc
;
200 struct sctp_chunk
*auth
;
201 sctp_xmit_t retval
= SCTP_XMIT_OK
;
203 /* if we don't have an association, we can't do authentication */
207 /* See if this is an auth chunk we are bundling or if
208 * auth is already bundled.
210 if (chunk
->chunk_hdr
->type
== SCTP_CID_AUTH
|| pkt
->auth
)
213 /* if the peer did not request this chunk to be authenticated,
219 auth
= sctp_make_auth(asoc
);
223 retval
= sctp_packet_append_chunk(pkt
, auth
);
228 /* Try to bundle a SACK with the packet. */
229 static sctp_xmit_t
sctp_packet_bundle_sack(struct sctp_packet
*pkt
,
230 struct sctp_chunk
*chunk
)
232 sctp_xmit_t retval
= SCTP_XMIT_OK
;
234 /* If sending DATA and haven't aleady bundled a SACK, try to
235 * bundle one in to the packet.
237 if (sctp_chunk_is_data(chunk
) && !pkt
->has_sack
&&
238 !pkt
->has_cookie_echo
) {
239 struct sctp_association
*asoc
;
240 asoc
= pkt
->transport
->asoc
;
242 if (asoc
->a_rwnd
> asoc
->rwnd
) {
243 struct sctp_chunk
*sack
;
244 asoc
->a_rwnd
= asoc
->rwnd
;
245 sack
= sctp_make_sack(asoc
);
247 struct timer_list
*timer
;
248 retval
= sctp_packet_append_chunk(pkt
, sack
);
249 asoc
->peer
.sack_needed
= 0;
250 timer
= &asoc
->timers
[SCTP_EVENT_TIMEOUT_SACK
];
251 if (timer_pending(timer
) && del_timer(timer
))
252 sctp_association_put(asoc
);
259 /* Append a chunk to the offered packet reporting back any inability to do
262 sctp_xmit_t
sctp_packet_append_chunk(struct sctp_packet
*packet
,
263 struct sctp_chunk
*chunk
)
265 sctp_xmit_t retval
= SCTP_XMIT_OK
;
266 __u16 chunk_len
= WORD_ROUND(ntohs(chunk
->chunk_hdr
->length
));
271 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__
, packet
,
274 /* Try to bundle AUTH chunk */
275 retval
= sctp_packet_bundle_auth(packet
, chunk
);
276 if (retval
!= SCTP_XMIT_OK
)
279 /* Try to bundle SACK chunk */
280 retval
= sctp_packet_bundle_sack(packet
, chunk
);
281 if (retval
!= SCTP_XMIT_OK
)
284 psize
= packet
->size
;
285 pmtu
= ((packet
->transport
->asoc
) ?
286 (packet
->transport
->asoc
->pathmtu
) :
287 (packet
->transport
->pathmtu
));
289 too_big
= (psize
+ chunk_len
> pmtu
);
291 /* Decide if we need to fragment or resubmit later. */
293 /* It's OK to fragmet at IP level if any one of the following
295 * 1. The packet is empty (meaning this chunk is greater
297 * 2. The chunk we are adding is a control chunk
298 * 3. The packet doesn't have any data in it yet and data
299 * requires authentication.
301 if (sctp_packet_empty(packet
) || !sctp_chunk_is_data(chunk
) ||
302 (!packet
->has_data
&& chunk
->auth
)) {
303 /* We no longer do re-fragmentation.
304 * Just fragment at the IP layer, if we
305 * actually hit this condition
307 packet
->ipfragok
= 1;
311 retval
= SCTP_XMIT_PMTU_FULL
;
317 /* We believe that this chunk is OK to add to the packet (as
318 * long as we have the cwnd for it).
321 /* DATA is a special case since we must examine both rwnd and cwnd
322 * before we send DATA.
324 switch (chunk
->chunk_hdr
->type
) {
326 retval
= sctp_packet_append_data(packet
, chunk
);
327 if (SCTP_XMIT_OK
!= retval
)
329 /* Disallow SACK bundling after DATA. */
330 packet
->has_sack
= 1;
331 /* Disallow AUTH bundling after DATA */
332 packet
->has_auth
= 1;
333 /* Let it be knows that packet has DATA in it */
334 packet
->has_data
= 1;
335 /* timestamp the chunk for rtx purposes */
336 chunk
->sent_at
= jiffies
;
338 case SCTP_CID_COOKIE_ECHO
:
339 packet
->has_cookie_echo
= 1;
343 packet
->has_sack
= 1;
347 packet
->has_auth
= 1;
348 packet
->auth
= chunk
;
352 /* It is OK to send this chunk. */
353 list_add_tail(&chunk
->list
, &packet
->chunk_list
);
354 packet
->size
+= chunk_len
;
355 chunk
->transport
= packet
->transport
;
360 /* All packets are sent to the network through this function from
363 * The return value is a normal kernel error return value.
365 int sctp_packet_transmit(struct sctp_packet
*packet
)
367 struct sctp_transport
*tp
= packet
->transport
;
368 struct sctp_association
*asoc
= tp
->asoc
;
370 __be32 crc32
= __constant_cpu_to_be32(0);
371 struct sk_buff
*nskb
;
372 struct sctp_chunk
*chunk
, *tmp
;
375 int padding
; /* How much padding do we need? */
377 struct dst_entry
*dst
= tp
->dst
;
378 unsigned char *auth
= NULL
; /* pointer to auth in skb data */
379 __u32 cksum_buf_len
= sizeof(struct sctphdr
);
381 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__
, packet
);
383 /* Do NOT generate a chunkless packet. */
384 if (list_empty(&packet
->chunk_list
))
387 /* Set up convenience variables... */
388 chunk
= list_entry(packet
->chunk_list
.next
, struct sctp_chunk
, list
);
391 /* Allocate the new skb. */
392 nskb
= alloc_skb(packet
->size
+ LL_MAX_HEADER
, GFP_ATOMIC
);
396 /* Make sure the outbound skb has enough header room reserved. */
397 skb_reserve(nskb
, packet
->overhead
+ LL_MAX_HEADER
);
399 /* Set the owning socket so that we know where to get the
400 * destination IP address.
402 skb_set_owner_w(nskb
, sk
);
404 /* The 'obsolete' field of dst is set to 2 when a dst is freed. */
405 if (!dst
|| (dst
->obsolete
> 1)) {
407 sctp_transport_route(tp
, NULL
, sctp_sk(sk
));
408 if (asoc
&& (asoc
->param_flags
& SPP_PMTUD_ENABLE
)) {
409 sctp_assoc_sync_pmtu(asoc
);
412 nskb
->dst
= dst_clone(tp
->dst
);
417 /* Build the SCTP header. */
418 sh
= (struct sctphdr
*)skb_push(nskb
, sizeof(struct sctphdr
));
419 sh
->source
= htons(packet
->source_port
);
420 sh
->dest
= htons(packet
->destination_port
);
422 /* From 6.8 Adler-32 Checksum Calculation:
423 * After the packet is constructed (containing the SCTP common
424 * header and one or more control or DATA chunks), the
427 * 1) Fill in the proper Verification Tag in the SCTP common
428 * header and initialize the checksum field to 0's.
430 sh
->vtag
= htonl(packet
->vtag
);
436 * An endpoint bundles chunks by simply including multiple
437 * chunks in one outbound SCTP packet. ...
441 * 3.2 Chunk Field Descriptions
443 * The total length of a chunk (including Type, Length and
444 * Value fields) MUST be a multiple of 4 bytes. If the length
445 * of the chunk is not a multiple of 4 bytes, the sender MUST
446 * pad the chunk with all zero bytes and this padding is not
447 * included in the chunk length field. The sender should
448 * never pad with more than 3 bytes.
450 * [This whole comment explains WORD_ROUND() below.]
452 SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
453 list_for_each_entry_safe(chunk
, tmp
, &packet
->chunk_list
, list
) {
454 list_del_init(&chunk
->list
);
455 if (sctp_chunk_is_data(chunk
)) {
457 if (!chunk
->has_tsn
) {
458 sctp_chunk_assign_ssn(chunk
);
459 sctp_chunk_assign_tsn(chunk
);
461 /* 6.3.1 C4) When data is in flight and when allowed
462 * by rule C5, a new RTT measurement MUST be made each
463 * round trip. Furthermore, new RTT measurements
464 * SHOULD be made no more than once per round-trip
465 * for a given destination transport address.
468 if (!tp
->rto_pending
) {
469 chunk
->rtt_in_progress
= 1;
478 padding
= WORD_ROUND(chunk
->skb
->len
) - chunk
->skb
->len
;
480 memset(skb_put(chunk
->skb
, padding
), 0, padding
);
482 /* if this is the auth chunk that we are adding,
483 * store pointer where it will be added and put
484 * the auth into the packet.
486 if (chunk
== packet
->auth
)
487 auth
= skb_tail_pointer(nskb
);
489 cksum_buf_len
+= chunk
->skb
->len
;
490 memcpy(skb_put(nskb
, chunk
->skb
->len
),
491 chunk
->skb
->data
, chunk
->skb
->len
);
493 SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n",
495 sctp_cname(SCTP_ST_CHUNK(
496 chunk
->chunk_hdr
->type
)),
497 chunk
->has_tsn
? "TSN" : "No TSN",
499 ntohl(chunk
->subh
.data_hdr
->tsn
) : 0,
500 "length", ntohs(chunk
->chunk_hdr
->length
),
501 "chunk->skb->len", chunk
->skb
->len
,
502 "rtt_in_progress", chunk
->rtt_in_progress
);
505 * If this is a control chunk, this is our last
506 * reference. Free data chunks after they've been
507 * acknowledged or have failed.
509 if (!sctp_chunk_is_data(chunk
))
510 sctp_chunk_free(chunk
);
513 /* SCTP-AUTH, Section 6.2
514 * The sender MUST calculate the MAC as described in RFC2104 [2]
515 * using the hash function H as described by the MAC Identifier and
516 * the shared association key K based on the endpoint pair shared key
517 * described by the shared key identifier. The 'data' used for the
518 * computation of the AUTH-chunk is given by the AUTH chunk with its
519 * HMAC field set to zero (as shown in Figure 6) followed by all
520 * chunks that are placed after the AUTH chunk in the SCTP packet.
523 sctp_auth_calculate_hmac(asoc
, nskb
,
524 (struct sctp_auth_chunk
*)auth
,
527 /* 2) Calculate the Adler-32 checksum of the whole packet,
528 * including the SCTP common header and all the
531 * Note: Adler-32 is no longer applicable, as has been replaced
532 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
534 if (!(dst
->dev
->features
& NETIF_F_NO_CSUM
)) {
535 crc32
= sctp_start_cksum((__u8
*)sh
, cksum_buf_len
);
536 crc32
= sctp_end_cksum(crc32
);
538 nskb
->ip_summed
= CHECKSUM_UNNECESSARY
;
540 /* 3) Put the resultant value into the checksum field in the
541 * common header, and leave the rest of the bits unchanged.
543 sh
->checksum
= crc32
;
545 /* IP layer ECN support
547 * "The ECN-Capable Transport (ECT) bit would be set by the
548 * data sender to indicate that the end-points of the
549 * transport protocol are ECN-capable."
551 * Now setting the ECT bit all the time, as it should not cause
552 * any problems protocol-wise even if our peer ignores it.
554 * Note: The works for IPv6 layer checks this bit too later
555 * in transmission. See IP6_ECN_flow_xmit().
557 (*tp
->af_specific
->ecn_capable
)(nskb
->sk
);
559 /* Set up the IP options. */
560 /* BUG: not implemented
561 * For v4 this all lives somewhere in sk->sk_opt...
564 /* Dump that on IP! */
565 if (asoc
&& asoc
->peer
.last_sent_to
!= tp
) {
566 /* Considering the multiple CPU scenario, this is a
567 * "correcter" place for last_sent_to. --xguo
569 asoc
->peer
.last_sent_to
= tp
;
573 struct timer_list
*timer
;
574 unsigned long timeout
;
576 tp
->last_time_used
= jiffies
;
578 /* Restart the AUTOCLOSE timer when sending data. */
579 if (sctp_state(asoc
, ESTABLISHED
) && asoc
->autoclose
) {
580 timer
= &asoc
->timers
[SCTP_EVENT_TIMEOUT_AUTOCLOSE
];
581 timeout
= asoc
->timeouts
[SCTP_EVENT_TIMEOUT_AUTOCLOSE
];
583 if (!mod_timer(timer
, jiffies
+ timeout
))
584 sctp_association_hold(asoc
);
588 SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n",
591 nskb
->local_df
= packet
->ipfragok
;
592 (*tp
->af_specific
->sctp_xmit
)(nskb
, tp
);
595 packet
->size
= packet
->overhead
;
599 IP_INC_STATS_BH(&init_net
, IPSTATS_MIB_OUTNOROUTES
);
601 /* FIXME: Returning the 'err' will effect all the associations
602 * associated with a socket, although only one of the paths of the
603 * association is unreachable.
604 * The real failure of a transport or association can be passed on
605 * to the user via notifications. So setting this error may not be
608 /* err = -EHOSTUNREACH; */
610 /* Control chunks are unreliable so just drop them. DATA chunks
611 * will get resent or dropped later.
614 list_for_each_entry_safe(chunk
, tmp
, &packet
->chunk_list
, list
) {
615 list_del_init(&chunk
->list
);
616 if (!sctp_chunk_is_data(chunk
))
617 sctp_chunk_free(chunk
);
625 /********************************************************************
626 * 2nd Level Abstractions
627 ********************************************************************/
629 /* This private function handles the specifics of appending DATA chunks. */
630 static sctp_xmit_t
sctp_packet_append_data(struct sctp_packet
*packet
,
631 struct sctp_chunk
*chunk
)
633 sctp_xmit_t retval
= SCTP_XMIT_OK
;
634 size_t datasize
, rwnd
, inflight
;
635 struct sctp_transport
*transport
= packet
->transport
;
636 __u32 max_burst_bytes
;
637 struct sctp_association
*asoc
= transport
->asoc
;
638 struct sctp_sock
*sp
= sctp_sk(asoc
->base
.sk
);
639 struct sctp_outq
*q
= &asoc
->outqueue
;
641 /* RFC 2960 6.1 Transmission of DATA Chunks
643 * A) At any given time, the data sender MUST NOT transmit new data to
644 * any destination transport address if its peer's rwnd indicates
645 * that the peer has no buffer space (i.e. rwnd is 0, see Section
646 * 6.2.1). However, regardless of the value of rwnd (including if it
647 * is 0), the data sender can always have one DATA chunk in flight to
648 * the receiver if allowed by cwnd (see rule B below). This rule
649 * allows the sender to probe for a change in rwnd that the sender
650 * missed due to the SACK having been lost in transit from the data
651 * receiver to the data sender.
654 rwnd
= asoc
->peer
.rwnd
;
655 inflight
= asoc
->outqueue
.outstanding_bytes
;
657 datasize
= sctp_data_size(chunk
);
659 if (datasize
> rwnd
) {
661 /* We have (at least) one data chunk in flight,
662 * so we can't fall back to rule 6.1 B).
664 retval
= SCTP_XMIT_RWND_FULL
;
669 /* sctpimpguide-05 2.14.2
670 * D) When the time comes for the sender to
671 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
672 * first be applied to limit how many new DATA chunks may be sent.
673 * The limit is applied by adjusting cwnd as follows:
674 * if ((flightsize + Max.Burst * MTU) < cwnd)
675 * cwnd = flightsize + Max.Burst * MTU
677 max_burst_bytes
= asoc
->max_burst
* asoc
->pathmtu
;
678 if ((transport
->flight_size
+ max_burst_bytes
) < transport
->cwnd
) {
679 transport
->cwnd
= transport
->flight_size
+ max_burst_bytes
;
680 SCTP_DEBUG_PRINTK("%s: cwnd limited by max_burst: "
681 "transport: %p, cwnd: %d, "
682 "ssthresh: %d, flight_size: %d, "
687 transport
->flight_size
,
688 transport
->partial_bytes_acked
);
691 /* RFC 2960 6.1 Transmission of DATA Chunks
693 * B) At any given time, the sender MUST NOT transmit new data
694 * to a given transport address if it has cwnd or more bytes
695 * of data outstanding to that transport address.
697 /* RFC 7.2.4 & the Implementers Guide 2.8.
700 * When a Fast Retransmit is being performed the sender SHOULD
701 * ignore the value of cwnd and SHOULD NOT delay retransmission.
703 if (chunk
->fast_retransmit
!= SCTP_NEED_FRTX
)
704 if (transport
->flight_size
>= transport
->cwnd
) {
705 retval
= SCTP_XMIT_RWND_FULL
;
709 /* Nagle's algorithm to solve small-packet problem:
710 * Inhibit the sending of new chunks when new outgoing data arrives
711 * if any previously transmitted data on the connection remains
714 if (!sp
->nodelay
&& sctp_packet_empty(packet
) &&
715 q
->outstanding_bytes
&& sctp_state(asoc
, ESTABLISHED
)) {
716 unsigned len
= datasize
+ q
->out_qlen
;
718 /* Check whether this chunk and all the rest of pending
719 * data will fit or delay in hopes of bundling a full
722 if (len
< asoc
->frag_point
) {
723 retval
= SCTP_XMIT_NAGLE_DELAY
;
728 /* Keep track of how many bytes are in flight over this transport. */
729 transport
->flight_size
+= datasize
;
731 /* Keep track of how many bytes are in flight to the receiver. */
732 asoc
->outqueue
.outstanding_bytes
+= datasize
;
734 /* Update our view of the receiver's rwnd. Include sk_buff overhead
735 * while updating peer.rwnd so that it reduces the chances of a
736 * receiver running out of receive buffer space even when receive
737 * window is still open. This can happen when a sender is sending
738 * sending small messages.
740 datasize
+= sizeof(struct sk_buff
);
746 asoc
->peer
.rwnd
= rwnd
;
747 /* Has been accepted for transmission. */
748 if (!asoc
->peer
.prsctp_capable
)
749 chunk
->msg
->can_abandon
= 0;