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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
46 #include <linux/types.h>
47 #include <linux/kernel.h>
48 #include <linux/wait.h>
49 #include <linux/time.h>
51 #include <linux/ipv6.h>
52 #include <linux/init.h>
53 #include <linux/slab.h>
54 #include <net/inet_ecn.h>
57 #include <net/net_namespace.h>
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_chunk(struct sctp_packet *packet,
68 struct sctp_chunk *chunk);
69 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
70 struct sctp_chunk *chunk);
71 static void sctp_packet_append_data(struct sctp_packet *packet,
72 struct sctp_chunk *chunk);
73 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
74 struct sctp_chunk *chunk,
77 static void sctp_packet_reset(struct sctp_packet *packet)
79 packet->size = packet->overhead;
80 packet->has_cookie_echo = 0;
89 * This appears to be a followup set of initializations.
91 struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
92 __u32 vtag, int ecn_capable)
94 struct sctp_chunk *chunk = NULL;
96 pr_debug("%s: packet:%p vtag:0x%x\n", __func__, packet, vtag);
100 if (ecn_capable && sctp_packet_empty(packet)) {
101 chunk = sctp_get_ecne_prepend(packet->transport->asoc);
103 /* If there a is a prepend chunk stick it on the list before
104 * any other chunks get appended.
107 sctp_packet_append_chunk(packet, chunk);
113 /* Initialize the packet structure. */
114 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
115 struct sctp_transport *transport,
116 __u16 sport, __u16 dport)
118 struct sctp_association *asoc = transport->asoc;
121 pr_debug("%s: packet:%p transport:%p\n", __func__, packet, transport);
123 packet->transport = transport;
124 packet->source_port = sport;
125 packet->destination_port = dport;
126 INIT_LIST_HEAD(&packet->chunk_list);
128 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
129 overhead = sp->pf->af->net_header_len;
131 overhead = sizeof(struct ipv6hdr);
133 overhead += sizeof(struct sctphdr);
134 packet->overhead = overhead;
135 sctp_packet_reset(packet);
142 void sctp_packet_free(struct sctp_packet *packet)
144 struct sctp_chunk *chunk, *tmp;
146 pr_debug("%s: packet:%p\n", __func__, packet);
148 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
149 list_del_init(&chunk->list);
150 sctp_chunk_free(chunk);
154 /* This routine tries to append the chunk to the offered packet. If adding
155 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
156 * is not present in the packet, it transmits the input packet.
157 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
158 * as it can fit in the packet, but any more data that does not fit in this
159 * packet can be sent only after receiving the COOKIE_ACK.
161 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
162 struct sctp_chunk *chunk,
168 pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
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->has_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);
225 if (retval != SCTP_XMIT_OK)
226 sctp_chunk_free(auth);
231 /* Try to bundle a SACK with the packet. */
232 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
233 struct sctp_chunk *chunk)
235 sctp_xmit_t retval = SCTP_XMIT_OK;
237 /* If sending DATA and haven't aleady bundled a SACK, try to
238 * bundle one in to the packet.
240 if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
241 !pkt->has_cookie_echo) {
242 struct sctp_association *asoc;
243 struct timer_list *timer;
244 asoc = pkt->transport->asoc;
245 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
247 /* If the SACK timer is running, we have a pending SACK */
248 if (timer_pending(timer)) {
249 struct sctp_chunk *sack;
251 if (pkt->transport->sack_generation !=
252 pkt->transport->asoc->peer.sack_generation)
255 asoc->a_rwnd = asoc->rwnd;
256 sack = sctp_make_sack(asoc);
258 retval = __sctp_packet_append_chunk(pkt, sack);
259 if (retval != SCTP_XMIT_OK) {
260 sctp_chunk_free(sack);
263 asoc->peer.sack_needed = 0;
264 if (del_timer(timer))
265 sctp_association_put(asoc);
274 /* Append a chunk to the offered packet reporting back any inability to do
277 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
278 struct sctp_chunk *chunk)
280 sctp_xmit_t retval = SCTP_XMIT_OK;
281 __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
283 /* Check to see if this chunk will fit into the packet */
284 retval = sctp_packet_will_fit(packet, chunk, chunk_len);
285 if (retval != SCTP_XMIT_OK)
288 /* We believe that this chunk is OK to add to the packet */
289 switch (chunk->chunk_hdr->type) {
291 /* Account for the data being in the packet */
292 sctp_packet_append_data(packet, chunk);
293 /* Disallow SACK bundling after DATA. */
294 packet->has_sack = 1;
295 /* Disallow AUTH bundling after DATA */
296 packet->has_auth = 1;
297 /* Let it be knows that packet has DATA in it */
298 packet->has_data = 1;
299 /* timestamp the chunk for rtx purposes */
300 chunk->sent_at = jiffies;
302 case SCTP_CID_COOKIE_ECHO:
303 packet->has_cookie_echo = 1;
307 packet->has_sack = 1;
309 chunk->asoc->stats.osacks++;
313 packet->has_auth = 1;
314 packet->auth = chunk;
318 /* It is OK to send this chunk. */
319 list_add_tail(&chunk->list, &packet->chunk_list);
320 packet->size += chunk_len;
321 chunk->transport = packet->transport;
326 /* Append a chunk to the offered packet reporting back any inability to do
329 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
330 struct sctp_chunk *chunk)
332 sctp_xmit_t retval = SCTP_XMIT_OK;
334 pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
336 /* Data chunks are special. Before seeing what else we can
337 * bundle into this packet, check to see if we are allowed to
340 if (sctp_chunk_is_data(chunk)) {
341 retval = sctp_packet_can_append_data(packet, chunk);
342 if (retval != SCTP_XMIT_OK)
346 /* Try to bundle AUTH chunk */
347 retval = sctp_packet_bundle_auth(packet, chunk);
348 if (retval != SCTP_XMIT_OK)
351 /* Try to bundle SACK chunk */
352 retval = sctp_packet_bundle_sack(packet, chunk);
353 if (retval != SCTP_XMIT_OK)
356 retval = __sctp_packet_append_chunk(packet, chunk);
362 static void sctp_packet_release_owner(struct sk_buff *skb)
367 static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
371 skb->destructor = sctp_packet_release_owner;
374 * The data chunks have already been accounted for in sctp_sendmsg(),
375 * therefore only reserve a single byte to keep socket around until
376 * the packet has been transmitted.
378 atomic_inc(&sk->sk_wmem_alloc);
381 /* All packets are sent to the network through this function from
384 * The return value is a normal kernel error return value.
386 int sctp_packet_transmit(struct sctp_packet *packet)
388 struct sctp_transport *tp = packet->transport;
389 struct sctp_association *asoc = tp->asoc;
391 struct sk_buff *nskb;
392 struct sctp_chunk *chunk, *tmp;
395 int padding; /* How much padding do we need? */
397 struct dst_entry *dst = tp->dst;
398 unsigned char *auth = NULL; /* pointer to auth in skb data */
399 __u32 cksum_buf_len = sizeof(struct sctphdr);
401 pr_debug("%s: packet:%p\n", __func__, packet);
403 /* Do NOT generate a chunkless packet. */
404 if (list_empty(&packet->chunk_list))
407 /* Set up convenience variables... */
408 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
411 /* Allocate the new skb. */
412 nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
416 /* Make sure the outbound skb has enough header room reserved. */
417 skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
419 /* Set the owning socket so that we know where to get the
420 * destination IP address.
422 sctp_packet_set_owner_w(nskb, sk);
424 if (!sctp_transport_dst_check(tp)) {
425 sctp_transport_route(tp, NULL, sctp_sk(sk));
426 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
427 sctp_assoc_sync_pmtu(sk, asoc);
430 dst = dst_clone(tp->dst);
431 skb_dst_set(nskb, dst);
435 /* Build the SCTP header. */
436 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
437 skb_reset_transport_header(nskb);
438 sh->source = htons(packet->source_port);
439 sh->dest = htons(packet->destination_port);
441 /* From 6.8 Adler-32 Checksum Calculation:
442 * After the packet is constructed (containing the SCTP common
443 * header and one or more control or DATA chunks), the
446 * 1) Fill in the proper Verification Tag in the SCTP common
447 * header and initialize the checksum field to 0's.
449 sh->vtag = htonl(packet->vtag);
455 * An endpoint bundles chunks by simply including multiple
456 * chunks in one outbound SCTP packet. ...
460 * 3.2 Chunk Field Descriptions
462 * The total length of a chunk (including Type, Length and
463 * Value fields) MUST be a multiple of 4 bytes. If the length
464 * of the chunk is not a multiple of 4 bytes, the sender MUST
465 * pad the chunk with all zero bytes and this padding is not
466 * included in the chunk length field. The sender should
467 * never pad with more than 3 bytes.
469 * [This whole comment explains WORD_ROUND() below.]
472 pr_debug("***sctp_transmit_packet***\n");
474 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
475 list_del_init(&chunk->list);
476 if (sctp_chunk_is_data(chunk)) {
477 /* 6.3.1 C4) When data is in flight and when allowed
478 * by rule C5, a new RTT measurement MUST be made each
479 * round trip. Furthermore, new RTT measurements
480 * SHOULD be made no more than once per round-trip
481 * for a given destination transport address.
484 if (!tp->rto_pending) {
485 chunk->rtt_in_progress = 1;
491 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
493 memset(skb_put(chunk->skb, padding), 0, padding);
495 /* if this is the auth chunk that we are adding,
496 * store pointer where it will be added and put
497 * the auth into the packet.
499 if (chunk == packet->auth)
500 auth = skb_tail_pointer(nskb);
502 cksum_buf_len += chunk->skb->len;
503 memcpy(skb_put(nskb, chunk->skb->len),
504 chunk->skb->data, chunk->skb->len);
506 pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, "
507 "rtt_in_progress:%d\n", chunk,
508 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
509 chunk->has_tsn ? "TSN" : "No TSN",
510 chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
511 ntohs(chunk->chunk_hdr->length), chunk->skb->len,
512 chunk->rtt_in_progress);
515 * If this is a control chunk, this is our last
516 * reference. Free data chunks after they've been
517 * acknowledged or have failed.
519 if (!sctp_chunk_is_data(chunk))
520 sctp_chunk_free(chunk);
523 /* SCTP-AUTH, Section 6.2
524 * The sender MUST calculate the MAC as described in RFC2104 [2]
525 * using the hash function H as described by the MAC Identifier and
526 * the shared association key K based on the endpoint pair shared key
527 * described by the shared key identifier. The 'data' used for the
528 * computation of the AUTH-chunk is given by the AUTH chunk with its
529 * HMAC field set to zero (as shown in Figure 6) followed by all
530 * chunks that are placed after the AUTH chunk in the SCTP packet.
533 sctp_auth_calculate_hmac(asoc, nskb,
534 (struct sctp_auth_chunk *)auth,
537 /* 2) Calculate the Adler-32 checksum of the whole packet,
538 * including the SCTP common header and all the
541 * Note: Adler-32 is no longer applicable, as has been replaced
542 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
544 if (!sctp_checksum_disable) {
545 if (!(dst->dev->features & NETIF_F_SCTP_CSUM)) {
546 __u32 crc32 = sctp_start_cksum((__u8 *)sh, cksum_buf_len);
548 /* 3) Put the resultant value into the checksum field in the
549 * common header, and leave the rest of the bits unchanged.
551 sh->checksum = sctp_end_cksum(crc32);
553 /* no need to seed pseudo checksum for SCTP */
554 nskb->ip_summed = CHECKSUM_PARTIAL;
555 nskb->csum_start = (skb_transport_header(nskb) -
557 nskb->csum_offset = offsetof(struct sctphdr, checksum);
561 /* IP layer ECN support
563 * "The ECN-Capable Transport (ECT) bit would be set by the
564 * data sender to indicate that the end-points of the
565 * transport protocol are ECN-capable."
567 * Now setting the ECT bit all the time, as it should not cause
568 * any problems protocol-wise even if our peer ignores it.
570 * Note: The works for IPv6 layer checks this bit too later
571 * in transmission. See IP6_ECN_flow_xmit().
573 (*tp->af_specific->ecn_capable)(nskb->sk);
575 /* Set up the IP options. */
576 /* BUG: not implemented
577 * For v4 this all lives somewhere in sk->sk_opt...
580 /* Dump that on IP! */
582 asoc->stats.opackets++;
583 if (asoc->peer.last_sent_to != tp)
584 /* Considering the multiple CPU scenario, this is a
585 * "correcter" place for last_sent_to. --xguo
587 asoc->peer.last_sent_to = tp;
591 struct timer_list *timer;
592 unsigned long timeout;
594 /* Restart the AUTOCLOSE timer when sending data. */
595 if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) {
596 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
597 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
599 if (!mod_timer(timer, jiffies + timeout))
600 sctp_association_hold(asoc);
604 pr_debug("***sctp_transmit_packet*** skb->len:%d\n", nskb->len);
606 nskb->local_df = packet->ipfragok;
607 (*tp->af_specific->sctp_xmit)(nskb, tp);
610 sctp_packet_reset(packet);
614 IP_INC_STATS_BH(sock_net(asoc->base.sk), IPSTATS_MIB_OUTNOROUTES);
616 /* FIXME: Returning the 'err' will effect all the associations
617 * associated with a socket, although only one of the paths of the
618 * association is unreachable.
619 * The real failure of a transport or association can be passed on
620 * to the user via notifications. So setting this error may not be
623 /* err = -EHOSTUNREACH; */
625 /* Control chunks are unreliable so just drop them. DATA chunks
626 * will get resent or dropped later.
629 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
630 list_del_init(&chunk->list);
631 if (!sctp_chunk_is_data(chunk))
632 sctp_chunk_free(chunk);
640 /********************************************************************
641 * 2nd Level Abstractions
642 ********************************************************************/
644 /* This private function check to see if a chunk can be added */
645 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
646 struct sctp_chunk *chunk)
648 sctp_xmit_t retval = SCTP_XMIT_OK;
649 size_t datasize, rwnd, inflight, flight_size;
650 struct sctp_transport *transport = packet->transport;
651 struct sctp_association *asoc = transport->asoc;
652 struct sctp_outq *q = &asoc->outqueue;
654 /* RFC 2960 6.1 Transmission of DATA Chunks
656 * A) At any given time, the data sender MUST NOT transmit new data to
657 * any destination transport address if its peer's rwnd indicates
658 * that the peer has no buffer space (i.e. rwnd is 0, see Section
659 * 6.2.1). However, regardless of the value of rwnd (including if it
660 * is 0), the data sender can always have one DATA chunk in flight to
661 * the receiver if allowed by cwnd (see rule B below). This rule
662 * allows the sender to probe for a change in rwnd that the sender
663 * missed due to the SACK having been lost in transit from the data
664 * receiver to the data sender.
667 rwnd = asoc->peer.rwnd;
668 inflight = q->outstanding_bytes;
669 flight_size = transport->flight_size;
671 datasize = sctp_data_size(chunk);
673 if (datasize > rwnd) {
675 /* We have (at least) one data chunk in flight,
676 * so we can't fall back to rule 6.1 B).
678 retval = SCTP_XMIT_RWND_FULL;
683 /* RFC 2960 6.1 Transmission of DATA Chunks
685 * B) At any given time, the sender MUST NOT transmit new data
686 * to a given transport address if it has cwnd or more bytes
687 * of data outstanding to that transport address.
689 /* RFC 7.2.4 & the Implementers Guide 2.8.
692 * When a Fast Retransmit is being performed the sender SHOULD
693 * ignore the value of cwnd and SHOULD NOT delay retransmission.
695 if (chunk->fast_retransmit != SCTP_NEED_FRTX)
696 if (flight_size >= transport->cwnd) {
697 retval = SCTP_XMIT_RWND_FULL;
701 /* Nagle's algorithm to solve small-packet problem:
702 * Inhibit the sending of new chunks when new outgoing data arrives
703 * if any previously transmitted data on the connection remains
706 if (!sctp_sk(asoc->base.sk)->nodelay && sctp_packet_empty(packet) &&
707 inflight && sctp_state(asoc, ESTABLISHED)) {
708 unsigned int max = transport->pathmtu - packet->overhead;
709 unsigned int len = chunk->skb->len + q->out_qlen;
711 /* Check whether this chunk and all the rest of pending
712 * data will fit or delay in hopes of bundling a full
714 * Don't delay large message writes that may have been
715 * fragmeneted into small peices.
717 if ((len < max) && chunk->msg->can_delay) {
718 retval = SCTP_XMIT_NAGLE_DELAY;
727 /* This private function does management things when adding DATA chunk */
728 static void sctp_packet_append_data(struct sctp_packet *packet,
729 struct sctp_chunk *chunk)
731 struct sctp_transport *transport = packet->transport;
732 size_t datasize = sctp_data_size(chunk);
733 struct sctp_association *asoc = transport->asoc;
734 u32 rwnd = asoc->peer.rwnd;
736 /* Keep track of how many bytes are in flight over this transport. */
737 transport->flight_size += datasize;
739 /* Keep track of how many bytes are in flight to the receiver. */
740 asoc->outqueue.outstanding_bytes += datasize;
742 /* Update our view of the receiver's rwnd. */
748 asoc->peer.rwnd = rwnd;
749 /* Has been accepted for transmission. */
750 if (!asoc->peer.prsctp_capable)
751 chunk->msg->can_abandon = 0;
752 sctp_chunk_assign_tsn(chunk);
753 sctp_chunk_assign_ssn(chunk);
756 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
757 struct sctp_chunk *chunk,
763 sctp_xmit_t retval = SCTP_XMIT_OK;
765 psize = packet->size;
766 pmtu = ((packet->transport->asoc) ?
767 (packet->transport->asoc->pathmtu) :
768 (packet->transport->pathmtu));
770 too_big = (psize + chunk_len > pmtu);
772 /* Decide if we need to fragment or resubmit later. */
774 /* It's OK to fragmet at IP level if any one of the following
776 * 1. The packet is empty (meaning this chunk is greater
778 * 2. The chunk we are adding is a control chunk
779 * 3. The packet doesn't have any data in it yet and data
780 * requires authentication.
782 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
783 (!packet->has_data && chunk->auth)) {
784 /* We no longer do re-fragmentation.
785 * Just fragment at the IP layer, if we
786 * actually hit this condition
788 packet->ipfragok = 1;
790 retval = SCTP_XMIT_PMTU_FULL;