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, see
24 * <http://www.gnu.org/licenses/>.
26 * Please send any bug reports or fixes you make to the
28 * lksctp developers <linux-sctp@vger.kernel.org>
30 * Written or modified by:
31 * La Monte H.P. Yarroll <piggy@acm.org>
32 * Karl Knutson <karl@athena.chicago.il.us>
33 * Jon Grimm <jgrimm@austin.ibm.com>
34 * Sridhar Samudrala <sri@us.ibm.com>
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #include <linux/types.h>
40 #include <linux/kernel.h>
41 #include <linux/wait.h>
42 #include <linux/time.h>
44 #include <linux/ipv6.h>
45 #include <linux/init.h>
46 #include <linux/slab.h>
47 #include <net/inet_ecn.h>
50 #include <net/net_namespace.h>
52 #include <linux/socket.h> /* for sa_family_t */
55 #include <net/sctp/sctp.h>
56 #include <net/sctp/sm.h>
57 #include <net/sctp/checksum.h>
59 /* Forward declarations for private helpers. */
60 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
61 struct sctp_chunk *chunk);
62 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
63 struct sctp_chunk *chunk);
64 static void sctp_packet_append_data(struct sctp_packet *packet,
65 struct sctp_chunk *chunk);
66 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
67 struct sctp_chunk *chunk,
70 static void sctp_packet_reset(struct sctp_packet *packet)
72 packet->size = packet->overhead;
73 packet->has_cookie_echo = 0;
82 * This appears to be a followup set of initializations.
84 struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
85 __u32 vtag, int ecn_capable)
87 struct sctp_chunk *chunk = NULL;
89 pr_debug("%s: packet:%p vtag:0x%x\n", __func__, packet, vtag);
93 if (ecn_capable && sctp_packet_empty(packet)) {
94 chunk = sctp_get_ecne_prepend(packet->transport->asoc);
96 /* If there a is a prepend chunk stick it on the list before
97 * any other chunks get appended.
100 sctp_packet_append_chunk(packet, chunk);
106 /* Initialize the packet structure. */
107 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
108 struct sctp_transport *transport,
109 __u16 sport, __u16 dport)
111 struct sctp_association *asoc = transport->asoc;
114 pr_debug("%s: packet:%p transport:%p\n", __func__, packet, transport);
116 packet->transport = transport;
117 packet->source_port = sport;
118 packet->destination_port = dport;
119 INIT_LIST_HEAD(&packet->chunk_list);
121 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
122 overhead = sp->pf->af->net_header_len;
124 overhead = sizeof(struct ipv6hdr);
126 overhead += sizeof(struct sctphdr);
127 packet->overhead = overhead;
128 sctp_packet_reset(packet);
135 void sctp_packet_free(struct sctp_packet *packet)
137 struct sctp_chunk *chunk, *tmp;
139 pr_debug("%s: packet:%p\n", __func__, packet);
141 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
142 list_del_init(&chunk->list);
143 sctp_chunk_free(chunk);
147 /* This routine tries to append the chunk to the offered packet. If adding
148 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
149 * is not present in the packet, it transmits the input packet.
150 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
151 * as it can fit in the packet, but any more data that does not fit in this
152 * packet can be sent only after receiving the COOKIE_ACK.
154 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
155 struct sctp_chunk *chunk,
161 pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
163 switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
164 case SCTP_XMIT_PMTU_FULL:
165 if (!packet->has_cookie_echo) {
166 error = sctp_packet_transmit(packet);
168 chunk->skb->sk->sk_err = -error;
170 /* If we have an empty packet, then we can NOT ever
174 retval = sctp_packet_append_chunk(packet,
179 case SCTP_XMIT_RWND_FULL:
181 case SCTP_XMIT_DELAY:
188 /* Try to bundle an auth chunk into the packet. */
189 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
190 struct sctp_chunk *chunk)
192 struct sctp_association *asoc = pkt->transport->asoc;
193 struct sctp_chunk *auth;
194 sctp_xmit_t retval = SCTP_XMIT_OK;
196 /* if we don't have an association, we can't do authentication */
200 /* See if this is an auth chunk we are bundling or if
201 * auth is already bundled.
203 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
206 /* if the peer did not request this chunk to be authenticated,
212 auth = sctp_make_auth(asoc);
216 retval = __sctp_packet_append_chunk(pkt, auth);
218 if (retval != SCTP_XMIT_OK)
219 sctp_chunk_free(auth);
224 /* Try to bundle a SACK with the packet. */
225 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
226 struct sctp_chunk *chunk)
228 sctp_xmit_t retval = SCTP_XMIT_OK;
230 /* If sending DATA and haven't aleady bundled a SACK, try to
231 * bundle one in to the packet.
233 if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
234 !pkt->has_cookie_echo) {
235 struct sctp_association *asoc;
236 struct timer_list *timer;
237 asoc = pkt->transport->asoc;
238 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
240 /* If the SACK timer is running, we have a pending SACK */
241 if (timer_pending(timer)) {
242 struct sctp_chunk *sack;
244 if (pkt->transport->sack_generation !=
245 pkt->transport->asoc->peer.sack_generation)
248 asoc->a_rwnd = asoc->rwnd;
249 sack = sctp_make_sack(asoc);
251 retval = __sctp_packet_append_chunk(pkt, sack);
252 if (retval != SCTP_XMIT_OK) {
253 sctp_chunk_free(sack);
256 asoc->peer.sack_needed = 0;
257 if (del_timer(timer))
258 sctp_association_put(asoc);
267 /* Append a chunk to the offered packet reporting back any inability to do
270 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
271 struct sctp_chunk *chunk)
273 sctp_xmit_t retval = SCTP_XMIT_OK;
274 __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
276 /* Check to see if this chunk will fit into the packet */
277 retval = sctp_packet_will_fit(packet, chunk, chunk_len);
278 if (retval != SCTP_XMIT_OK)
281 /* We believe that this chunk is OK to add to the packet */
282 switch (chunk->chunk_hdr->type) {
284 /* Account for the data being in the packet */
285 sctp_packet_append_data(packet, chunk);
286 /* Disallow SACK bundling after DATA. */
287 packet->has_sack = 1;
288 /* Disallow AUTH bundling after DATA */
289 packet->has_auth = 1;
290 /* Let it be knows that packet has DATA in it */
291 packet->has_data = 1;
292 /* timestamp the chunk for rtx purposes */
293 chunk->sent_at = jiffies;
295 case SCTP_CID_COOKIE_ECHO:
296 packet->has_cookie_echo = 1;
300 packet->has_sack = 1;
302 chunk->asoc->stats.osacks++;
306 packet->has_auth = 1;
307 packet->auth = chunk;
311 /* It is OK to send this chunk. */
312 list_add_tail(&chunk->list, &packet->chunk_list);
313 packet->size += chunk_len;
314 chunk->transport = packet->transport;
319 /* Append a chunk to the offered packet reporting back any inability to do
322 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
323 struct sctp_chunk *chunk)
325 sctp_xmit_t retval = SCTP_XMIT_OK;
327 pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
329 /* Data chunks are special. Before seeing what else we can
330 * bundle into this packet, check to see if we are allowed to
333 if (sctp_chunk_is_data(chunk)) {
334 retval = sctp_packet_can_append_data(packet, chunk);
335 if (retval != SCTP_XMIT_OK)
339 /* Try to bundle AUTH chunk */
340 retval = sctp_packet_bundle_auth(packet, chunk);
341 if (retval != SCTP_XMIT_OK)
344 /* Try to bundle SACK chunk */
345 retval = sctp_packet_bundle_sack(packet, chunk);
346 if (retval != SCTP_XMIT_OK)
349 retval = __sctp_packet_append_chunk(packet, chunk);
355 static void sctp_packet_release_owner(struct sk_buff *skb)
360 static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
364 skb->destructor = sctp_packet_release_owner;
367 * The data chunks have already been accounted for in sctp_sendmsg(),
368 * therefore only reserve a single byte to keep socket around until
369 * the packet has been transmitted.
371 atomic_inc(&sk->sk_wmem_alloc);
374 /* All packets are sent to the network through this function from
377 * The return value is a normal kernel error return value.
379 int sctp_packet_transmit(struct sctp_packet *packet)
381 struct sctp_transport *tp = packet->transport;
382 struct sctp_association *asoc = tp->asoc;
384 struct sk_buff *nskb;
385 struct sctp_chunk *chunk, *tmp;
388 int padding; /* How much padding do we need? */
390 struct dst_entry *dst;
391 unsigned char *auth = NULL; /* pointer to auth in skb data */
393 pr_debug("%s: packet:%p\n", __func__, packet);
395 /* Do NOT generate a chunkless packet. */
396 if (list_empty(&packet->chunk_list))
399 /* Set up convenience variables... */
400 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
403 /* Allocate the new skb. */
404 nskb = alloc_skb(packet->size + MAX_HEADER, GFP_ATOMIC);
408 /* Make sure the outbound skb has enough header room reserved. */
409 skb_reserve(nskb, packet->overhead + MAX_HEADER);
411 /* Set the owning socket so that we know where to get the
412 * destination IP address.
414 sctp_packet_set_owner_w(nskb, sk);
416 if (!sctp_transport_dst_check(tp)) {
417 sctp_transport_route(tp, NULL, sctp_sk(sk));
418 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
419 sctp_assoc_sync_pmtu(sk, asoc);
422 dst = dst_clone(tp->dst);
425 skb_dst_set(nskb, dst);
427 /* Build the SCTP header. */
428 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
429 skb_reset_transport_header(nskb);
430 sh->source = htons(packet->source_port);
431 sh->dest = htons(packet->destination_port);
433 /* From 6.8 Adler-32 Checksum Calculation:
434 * After the packet is constructed (containing the SCTP common
435 * header and one or more control or DATA chunks), the
438 * 1) Fill in the proper Verification Tag in the SCTP common
439 * header and initialize the checksum field to 0's.
441 sh->vtag = htonl(packet->vtag);
447 * An endpoint bundles chunks by simply including multiple
448 * chunks in one outbound SCTP packet. ...
452 * 3.2 Chunk Field Descriptions
454 * The total length of a chunk (including Type, Length and
455 * Value fields) MUST be a multiple of 4 bytes. If the length
456 * of the chunk is not a multiple of 4 bytes, the sender MUST
457 * pad the chunk with all zero bytes and this padding is not
458 * included in the chunk length field. The sender should
459 * never pad with more than 3 bytes.
461 * [This whole comment explains WORD_ROUND() below.]
464 pr_debug("***sctp_transmit_packet***\n");
466 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
467 list_del_init(&chunk->list);
468 if (sctp_chunk_is_data(chunk)) {
469 /* 6.3.1 C4) When data is in flight and when allowed
470 * by rule C5, a new RTT measurement MUST be made each
471 * round trip. Furthermore, new RTT measurements
472 * SHOULD be made no more than once per round-trip
473 * for a given destination transport address.
476 if (!chunk->resent && !tp->rto_pending) {
477 chunk->rtt_in_progress = 1;
484 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
486 memset(skb_put(chunk->skb, padding), 0, padding);
488 /* if this is the auth chunk that we are adding,
489 * store pointer where it will be added and put
490 * the auth into the packet.
492 if (chunk == packet->auth)
493 auth = skb_tail_pointer(nskb);
495 memcpy(skb_put(nskb, chunk->skb->len),
496 chunk->skb->data, chunk->skb->len);
498 pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, "
499 "rtt_in_progress:%d\n", chunk,
500 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
501 chunk->has_tsn ? "TSN" : "No TSN",
502 chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
503 ntohs(chunk->chunk_hdr->length), chunk->skb->len,
504 chunk->rtt_in_progress);
507 * If this is a control chunk, this is our last
508 * reference. Free data chunks after they've been
509 * acknowledged or have failed.
511 if (!sctp_chunk_is_data(chunk))
512 sctp_chunk_free(chunk);
515 /* SCTP-AUTH, Section 6.2
516 * The sender MUST calculate the MAC as described in RFC2104 [2]
517 * using the hash function H as described by the MAC Identifier and
518 * the shared association key K based on the endpoint pair shared key
519 * described by the shared key identifier. The 'data' used for the
520 * computation of the AUTH-chunk is given by the AUTH chunk with its
521 * HMAC field set to zero (as shown in Figure 6) followed by all
522 * chunks that are placed after the AUTH chunk in the SCTP packet.
525 sctp_auth_calculate_hmac(asoc, nskb,
526 (struct sctp_auth_chunk *)auth,
529 /* 2) Calculate the Adler-32 checksum of the whole packet,
530 * including the SCTP common header and all the
533 * Note: Adler-32 is no longer applicable, as has been replaced
534 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
536 if (!sctp_checksum_disable) {
537 if (!(dst->dev->features & NETIF_F_SCTP_CSUM) ||
538 (dst_xfrm(dst) != NULL) || packet->ipfragok) {
539 sh->checksum = sctp_compute_cksum(nskb, 0);
541 /* no need to seed pseudo checksum for SCTP */
542 nskb->ip_summed = CHECKSUM_PARTIAL;
543 nskb->csum_start = skb_transport_header(nskb) - nskb->head;
544 nskb->csum_offset = offsetof(struct sctphdr, checksum);
548 /* IP layer ECN support
550 * "The ECN-Capable Transport (ECT) bit would be set by the
551 * data sender to indicate that the end-points of the
552 * transport protocol are ECN-capable."
554 * Now setting the ECT bit all the time, as it should not cause
555 * any problems protocol-wise even if our peer ignores it.
557 * Note: The works for IPv6 layer checks this bit too later
558 * in transmission. See IP6_ECN_flow_xmit().
560 tp->af_specific->ecn_capable(nskb->sk);
562 /* Set up the IP options. */
563 /* BUG: not implemented
564 * For v4 this all lives somewhere in sk->sk_opt...
567 /* Dump that on IP! */
569 asoc->stats.opackets++;
570 if (asoc->peer.last_sent_to != tp)
571 /* Considering the multiple CPU scenario, this is a
572 * "correcter" place for last_sent_to. --xguo
574 asoc->peer.last_sent_to = tp;
578 struct timer_list *timer;
579 unsigned long timeout;
581 /* Restart the AUTOCLOSE timer when sending data. */
582 if (sctp_state(asoc, ESTABLISHED) &&
583 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) {
584 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
585 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
587 if (!mod_timer(timer, jiffies + timeout))
588 sctp_association_hold(asoc);
592 pr_debug("***sctp_transmit_packet*** skb->len:%d\n", nskb->len);
594 nskb->ignore_df = packet->ipfragok;
595 tp->af_specific->sctp_xmit(nskb, tp);
598 sctp_packet_reset(packet);
604 IP_INC_STATS(sock_net(asoc->base.sk), IPSTATS_MIB_OUTNOROUTES);
606 /* FIXME: Returning the 'err' will effect all the associations
607 * associated with a socket, although only one of the paths of the
608 * association is unreachable.
609 * The real failure of a transport or association can be passed on
610 * to the user via notifications. So setting this error may not be
613 /* err = -EHOSTUNREACH; */
615 /* Control chunks are unreliable so just drop them. DATA chunks
616 * will get resent or dropped later.
619 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
620 list_del_init(&chunk->list);
621 if (!sctp_chunk_is_data(chunk))
622 sctp_chunk_free(chunk);
630 /********************************************************************
631 * 2nd Level Abstractions
632 ********************************************************************/
634 /* This private function check to see if a chunk can be added */
635 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
636 struct sctp_chunk *chunk)
638 size_t datasize, rwnd, inflight, flight_size;
639 struct sctp_transport *transport = packet->transport;
640 struct sctp_association *asoc = transport->asoc;
641 struct sctp_outq *q = &asoc->outqueue;
643 /* RFC 2960 6.1 Transmission of DATA Chunks
645 * A) At any given time, the data sender MUST NOT transmit new data to
646 * any destination transport address if its peer's rwnd indicates
647 * that the peer has no buffer space (i.e. rwnd is 0, see Section
648 * 6.2.1). However, regardless of the value of rwnd (including if it
649 * is 0), the data sender can always have one DATA chunk in flight to
650 * the receiver if allowed by cwnd (see rule B below). This rule
651 * allows the sender to probe for a change in rwnd that the sender
652 * missed due to the SACK having been lost in transit from the data
653 * receiver to the data sender.
656 rwnd = asoc->peer.rwnd;
657 inflight = q->outstanding_bytes;
658 flight_size = transport->flight_size;
660 datasize = sctp_data_size(chunk);
662 if (datasize > rwnd && inflight > 0)
663 /* We have (at least) one data chunk in flight,
664 * so we can't fall back to rule 6.1 B).
666 return SCTP_XMIT_RWND_FULL;
668 /* RFC 2960 6.1 Transmission of DATA Chunks
670 * B) At any given time, the sender MUST NOT transmit new data
671 * to a given transport address if it has cwnd or more bytes
672 * of data outstanding to that transport address.
674 /* RFC 7.2.4 & the Implementers Guide 2.8.
677 * When a Fast Retransmit is being performed the sender SHOULD
678 * ignore the value of cwnd and SHOULD NOT delay retransmission.
680 if (chunk->fast_retransmit != SCTP_NEED_FRTX &&
681 flight_size >= transport->cwnd)
682 return SCTP_XMIT_RWND_FULL;
684 /* Nagle's algorithm to solve small-packet problem:
685 * Inhibit the sending of new chunks when new outgoing data arrives
686 * if any previously transmitted data on the connection remains
690 if (sctp_sk(asoc->base.sk)->nodelay)
694 if (!sctp_packet_empty(packet))
695 /* Append to packet */
699 /* Nothing unacked */
702 if (!sctp_state(asoc, ESTABLISHED))
705 /* Check whether this chunk and all the rest of pending data will fit
706 * or delay in hopes of bundling a full sized packet.
708 if (chunk->skb->len + q->out_qlen >= transport->pathmtu - packet->overhead)
709 /* Enough data queued to fill a packet */
712 /* Don't delay large message writes that may have been fragmented */
713 if (!chunk->msg->can_delay)
716 /* Defer until all data acked or packet full */
717 return SCTP_XMIT_DELAY;
720 /* This private function does management things when adding DATA chunk */
721 static void sctp_packet_append_data(struct sctp_packet *packet,
722 struct sctp_chunk *chunk)
724 struct sctp_transport *transport = packet->transport;
725 size_t datasize = sctp_data_size(chunk);
726 struct sctp_association *asoc = transport->asoc;
727 u32 rwnd = asoc->peer.rwnd;
729 /* Keep track of how many bytes are in flight over this transport. */
730 transport->flight_size += datasize;
732 /* Keep track of how many bytes are in flight to the receiver. */
733 asoc->outqueue.outstanding_bytes += datasize;
735 /* Update our view of the receiver's rwnd. */
741 asoc->peer.rwnd = rwnd;
742 /* Has been accepted for transmission. */
743 if (!asoc->peer.prsctp_capable)
744 chunk->msg->can_abandon = 0;
745 sctp_chunk_assign_tsn(chunk);
746 sctp_chunk_assign_ssn(chunk);
749 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
750 struct sctp_chunk *chunk,
756 sctp_xmit_t retval = SCTP_XMIT_OK;
758 psize = packet->size;
759 pmtu = ((packet->transport->asoc) ?
760 (packet->transport->asoc->pathmtu) :
761 (packet->transport->pathmtu));
763 too_big = (psize + chunk_len > pmtu);
765 /* Decide if we need to fragment or resubmit later. */
767 /* It's OK to fragmet at IP level if any one of the following
769 * 1. The packet is empty (meaning this chunk is greater
771 * 2. The chunk we are adding is a control chunk
772 * 3. The packet doesn't have any data in it yet and data
773 * requires authentication.
775 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
776 (!packet->has_data && chunk->auth)) {
777 /* We no longer do re-fragmentation.
778 * Just fragment at the IP layer, if we
779 * actually hit this condition
781 packet->ipfragok = 1;
783 retval = SCTP_XMIT_PMTU_FULL;