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 <linux-sctp@vger.kernel.org>
31 * Written or modified by:
32 * La Monte H.P. Yarroll <piggy@acm.org>
33 * Karl Knutson <karl@athena.chicago.il.us>
34 * Jon Grimm <jgrimm@austin.ibm.com>
35 * Sridhar Samudrala <sri@us.ibm.com>
38 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
40 #include <linux/types.h>
41 #include <linux/kernel.h>
42 #include <linux/wait.h>
43 #include <linux/time.h>
45 #include <linux/ipv6.h>
46 #include <linux/init.h>
47 #include <linux/slab.h>
48 #include <net/inet_ecn.h>
51 #include <net/net_namespace.h>
53 #include <linux/socket.h> /* for sa_family_t */
56 #include <net/sctp/sctp.h>
57 #include <net/sctp/sm.h>
58 #include <net/sctp/checksum.h>
60 /* Forward declarations for private helpers. */
61 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
62 struct sctp_chunk *chunk);
63 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
64 struct sctp_chunk *chunk);
65 static void sctp_packet_append_data(struct sctp_packet *packet,
66 struct sctp_chunk *chunk);
67 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
68 struct sctp_chunk *chunk,
71 static void sctp_packet_reset(struct sctp_packet *packet)
73 packet->size = packet->overhead;
74 packet->has_cookie_echo = 0;
83 * This appears to be a followup set of initializations.
85 struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
86 __u32 vtag, int ecn_capable)
88 struct sctp_chunk *chunk = NULL;
90 pr_debug("%s: packet:%p vtag:0x%x\n", __func__, packet, vtag);
94 if (ecn_capable && sctp_packet_empty(packet)) {
95 chunk = sctp_get_ecne_prepend(packet->transport->asoc);
97 /* If there a is a prepend chunk stick it on the list before
98 * any other chunks get appended.
101 sctp_packet_append_chunk(packet, chunk);
107 /* Initialize the packet structure. */
108 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
109 struct sctp_transport *transport,
110 __u16 sport, __u16 dport)
112 struct sctp_association *asoc = transport->asoc;
115 pr_debug("%s: packet:%p transport:%p\n", __func__, packet, transport);
117 packet->transport = transport;
118 packet->source_port = sport;
119 packet->destination_port = dport;
120 INIT_LIST_HEAD(&packet->chunk_list);
122 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
123 overhead = sp->pf->af->net_header_len;
125 overhead = sizeof(struct ipv6hdr);
127 overhead += sizeof(struct sctphdr);
128 packet->overhead = overhead;
129 sctp_packet_reset(packet);
136 void sctp_packet_free(struct sctp_packet *packet)
138 struct sctp_chunk *chunk, *tmp;
140 pr_debug("%s: packet:%p\n", __func__, packet);
142 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
143 list_del_init(&chunk->list);
144 sctp_chunk_free(chunk);
148 /* This routine tries to append the chunk to the offered packet. If adding
149 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
150 * is not present in the packet, it transmits the input packet.
151 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
152 * as it can fit in the packet, but any more data that does not fit in this
153 * packet can be sent only after receiving the COOKIE_ACK.
155 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
156 struct sctp_chunk *chunk,
162 pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
164 switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
165 case SCTP_XMIT_PMTU_FULL:
166 if (!packet->has_cookie_echo) {
167 error = sctp_packet_transmit(packet);
169 chunk->skb->sk->sk_err = -error;
171 /* If we have an empty packet, then we can NOT ever
175 retval = sctp_packet_append_chunk(packet,
180 case SCTP_XMIT_RWND_FULL:
182 case SCTP_XMIT_NAGLE_DELAY:
189 /* Try to bundle an auth chunk into the packet. */
190 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
191 struct sctp_chunk *chunk)
193 struct sctp_association *asoc = pkt->transport->asoc;
194 struct sctp_chunk *auth;
195 sctp_xmit_t retval = SCTP_XMIT_OK;
197 /* if we don't have an association, we can't do authentication */
201 /* See if this is an auth chunk we are bundling or if
202 * auth is already bundled.
204 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
207 /* if the peer did not request this chunk to be authenticated,
213 auth = sctp_make_auth(asoc);
217 retval = __sctp_packet_append_chunk(pkt, auth);
219 if (retval != SCTP_XMIT_OK)
220 sctp_chunk_free(auth);
225 /* Try to bundle a SACK with the packet. */
226 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
227 struct sctp_chunk *chunk)
229 sctp_xmit_t retval = SCTP_XMIT_OK;
231 /* If sending DATA and haven't aleady bundled a SACK, try to
232 * bundle one in to the packet.
234 if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
235 !pkt->has_cookie_echo) {
236 struct sctp_association *asoc;
237 struct timer_list *timer;
238 asoc = pkt->transport->asoc;
239 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
241 /* If the SACK timer is running, we have a pending SACK */
242 if (timer_pending(timer)) {
243 struct sctp_chunk *sack;
245 if (pkt->transport->sack_generation !=
246 pkt->transport->asoc->peer.sack_generation)
249 asoc->a_rwnd = asoc->rwnd;
250 sack = sctp_make_sack(asoc);
252 retval = __sctp_packet_append_chunk(pkt, sack);
253 if (retval != SCTP_XMIT_OK) {
254 sctp_chunk_free(sack);
257 asoc->peer.sack_needed = 0;
258 if (del_timer(timer))
259 sctp_association_put(asoc);
268 /* Append a chunk to the offered packet reporting back any inability to do
271 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
272 struct sctp_chunk *chunk)
274 sctp_xmit_t retval = SCTP_XMIT_OK;
275 __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
277 /* Check to see if this chunk will fit into the packet */
278 retval = sctp_packet_will_fit(packet, chunk, chunk_len);
279 if (retval != SCTP_XMIT_OK)
282 /* We believe that this chunk is OK to add to the packet */
283 switch (chunk->chunk_hdr->type) {
285 /* Account for the data being in the packet */
286 sctp_packet_append_data(packet, chunk);
287 /* Disallow SACK bundling after DATA. */
288 packet->has_sack = 1;
289 /* Disallow AUTH bundling after DATA */
290 packet->has_auth = 1;
291 /* Let it be knows that packet has DATA in it */
292 packet->has_data = 1;
293 /* timestamp the chunk for rtx purposes */
294 chunk->sent_at = jiffies;
296 case SCTP_CID_COOKIE_ECHO:
297 packet->has_cookie_echo = 1;
301 packet->has_sack = 1;
303 chunk->asoc->stats.osacks++;
307 packet->has_auth = 1;
308 packet->auth = chunk;
312 /* It is OK to send this chunk. */
313 list_add_tail(&chunk->list, &packet->chunk_list);
314 packet->size += chunk_len;
315 chunk->transport = packet->transport;
320 /* Append a chunk to the offered packet reporting back any inability to do
323 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
324 struct sctp_chunk *chunk)
326 sctp_xmit_t retval = SCTP_XMIT_OK;
328 pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
330 /* Data chunks are special. Before seeing what else we can
331 * bundle into this packet, check to see if we are allowed to
334 if (sctp_chunk_is_data(chunk)) {
335 retval = sctp_packet_can_append_data(packet, chunk);
336 if (retval != SCTP_XMIT_OK)
340 /* Try to bundle AUTH chunk */
341 retval = sctp_packet_bundle_auth(packet, chunk);
342 if (retval != SCTP_XMIT_OK)
345 /* Try to bundle SACK chunk */
346 retval = sctp_packet_bundle_sack(packet, chunk);
347 if (retval != SCTP_XMIT_OK)
350 retval = __sctp_packet_append_chunk(packet, chunk);
356 static void sctp_packet_release_owner(struct sk_buff *skb)
361 static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
365 skb->destructor = sctp_packet_release_owner;
368 * The data chunks have already been accounted for in sctp_sendmsg(),
369 * therefore only reserve a single byte to keep socket around until
370 * the packet has been transmitted.
372 atomic_inc(&sk->sk_wmem_alloc);
375 /* All packets are sent to the network through this function from
378 * The return value is a normal kernel error return value.
380 int sctp_packet_transmit(struct sctp_packet *packet)
382 struct sctp_transport *tp = packet->transport;
383 struct sctp_association *asoc = tp->asoc;
385 struct sk_buff *nskb;
386 struct sctp_chunk *chunk, *tmp;
389 int padding; /* How much padding do we need? */
391 struct dst_entry *dst = tp->dst;
392 unsigned char *auth = NULL; /* pointer to auth in skb data */
394 pr_debug("%s: packet:%p\n", __func__, packet);
396 /* Do NOT generate a chunkless packet. */
397 if (list_empty(&packet->chunk_list))
400 /* Set up convenience variables... */
401 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
404 /* Allocate the new skb. */
405 nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
409 /* Make sure the outbound skb has enough header room reserved. */
410 skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
412 /* Set the owning socket so that we know where to get the
413 * destination IP address.
415 sctp_packet_set_owner_w(nskb, sk);
417 if (!sctp_transport_dst_check(tp)) {
418 sctp_transport_route(tp, NULL, sctp_sk(sk));
419 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
420 sctp_assoc_sync_pmtu(sk, asoc);
423 dst = dst_clone(tp->dst);
424 skb_dst_set(nskb, dst);
428 /* Build the SCTP header. */
429 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
430 skb_reset_transport_header(nskb);
431 sh->source = htons(packet->source_port);
432 sh->dest = htons(packet->destination_port);
434 /* From 6.8 Adler-32 Checksum Calculation:
435 * After the packet is constructed (containing the SCTP common
436 * header and one or more control or DATA chunks), the
439 * 1) Fill in the proper Verification Tag in the SCTP common
440 * header and initialize the checksum field to 0's.
442 sh->vtag = htonl(packet->vtag);
448 * An endpoint bundles chunks by simply including multiple
449 * chunks in one outbound SCTP packet. ...
453 * 3.2 Chunk Field Descriptions
455 * The total length of a chunk (including Type, Length and
456 * Value fields) MUST be a multiple of 4 bytes. If the length
457 * of the chunk is not a multiple of 4 bytes, the sender MUST
458 * pad the chunk with all zero bytes and this padding is not
459 * included in the chunk length field. The sender should
460 * never pad with more than 3 bytes.
462 * [This whole comment explains WORD_ROUND() below.]
465 pr_debug("***sctp_transmit_packet***\n");
467 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
468 list_del_init(&chunk->list);
469 if (sctp_chunk_is_data(chunk)) {
470 /* 6.3.1 C4) When data is in flight and when allowed
471 * by rule C5, a new RTT measurement MUST be made each
472 * round trip. Furthermore, new RTT measurements
473 * SHOULD be made no more than once per round-trip
474 * for a given destination transport address.
477 if (!chunk->resent && !tp->rto_pending) {
478 chunk->rtt_in_progress = 1;
485 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
487 memset(skb_put(chunk->skb, padding), 0, padding);
489 /* if this is the auth chunk that we are adding,
490 * store pointer where it will be added and put
491 * the auth into the packet.
493 if (chunk == packet->auth)
494 auth = skb_tail_pointer(nskb);
496 memcpy(skb_put(nskb, chunk->skb->len),
497 chunk->skb->data, chunk->skb->len);
499 pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, "
500 "rtt_in_progress:%d\n", chunk,
501 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
502 chunk->has_tsn ? "TSN" : "No TSN",
503 chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
504 ntohs(chunk->chunk_hdr->length), chunk->skb->len,
505 chunk->rtt_in_progress);
508 * If this is a control chunk, this is our last
509 * reference. Free data chunks after they've been
510 * acknowledged or have failed.
512 if (!sctp_chunk_is_data(chunk))
513 sctp_chunk_free(chunk);
516 /* SCTP-AUTH, Section 6.2
517 * The sender MUST calculate the MAC as described in RFC2104 [2]
518 * using the hash function H as described by the MAC Identifier and
519 * the shared association key K based on the endpoint pair shared key
520 * described by the shared key identifier. The 'data' used for the
521 * computation of the AUTH-chunk is given by the AUTH chunk with its
522 * HMAC field set to zero (as shown in Figure 6) followed by all
523 * chunks that are placed after the AUTH chunk in the SCTP packet.
526 sctp_auth_calculate_hmac(asoc, nskb,
527 (struct sctp_auth_chunk *)auth,
530 /* 2) Calculate the Adler-32 checksum of the whole packet,
531 * including the SCTP common header and all the
534 * Note: Adler-32 is no longer applicable, as has been replaced
535 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
537 if (!sctp_checksum_disable) {
538 if (!(dst->dev->features & NETIF_F_SCTP_CSUM) ||
539 (dst_xfrm(dst) != NULL) || packet->ipfragok) {
540 sh->checksum = sctp_compute_cksum(nskb, 0);
542 /* no need to seed pseudo checksum for SCTP */
543 nskb->ip_summed = CHECKSUM_PARTIAL;
544 nskb->csum_start = (skb_transport_header(nskb) -
546 nskb->csum_offset = offsetof(struct sctphdr, checksum);
550 /* IP layer ECN support
552 * "The ECN-Capable Transport (ECT) bit would be set by the
553 * data sender to indicate that the end-points of the
554 * transport protocol are ECN-capable."
556 * Now setting the ECT bit all the time, as it should not cause
557 * any problems protocol-wise even if our peer ignores it.
559 * Note: The works for IPv6 layer checks this bit too later
560 * in transmission. See IP6_ECN_flow_xmit().
562 (*tp->af_specific->ecn_capable)(nskb->sk);
564 /* Set up the IP options. */
565 /* BUG: not implemented
566 * For v4 this all lives somewhere in sk->sk_opt...
569 /* Dump that on IP! */
571 asoc->stats.opackets++;
572 if (asoc->peer.last_sent_to != tp)
573 /* Considering the multiple CPU scenario, this is a
574 * "correcter" place for last_sent_to. --xguo
576 asoc->peer.last_sent_to = tp;
580 struct timer_list *timer;
581 unsigned long timeout;
583 /* Restart the AUTOCLOSE timer when sending data. */
584 if (sctp_state(asoc, ESTABLISHED) &&
585 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) {
586 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
587 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
589 if (!mod_timer(timer, jiffies + timeout))
590 sctp_association_hold(asoc);
594 pr_debug("***sctp_transmit_packet*** skb->len:%d\n", nskb->len);
596 nskb->local_df = packet->ipfragok;
597 (*tp->af_specific->sctp_xmit)(nskb, tp);
600 sctp_packet_reset(packet);
604 IP_INC_STATS_BH(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 sctp_xmit_t retval = SCTP_XMIT_OK;
639 size_t datasize, rwnd, inflight, flight_size;
640 struct sctp_transport *transport = packet->transport;
641 struct sctp_association *asoc = transport->asoc;
642 struct sctp_outq *q = &asoc->outqueue;
644 /* RFC 2960 6.1 Transmission of DATA Chunks
646 * A) At any given time, the data sender MUST NOT transmit new data to
647 * any destination transport address if its peer's rwnd indicates
648 * that the peer has no buffer space (i.e. rwnd is 0, see Section
649 * 6.2.1). However, regardless of the value of rwnd (including if it
650 * is 0), the data sender can always have one DATA chunk in flight to
651 * the receiver if allowed by cwnd (see rule B below). This rule
652 * allows the sender to probe for a change in rwnd that the sender
653 * missed due to the SACK having been lost in transit from the data
654 * receiver to the data sender.
657 rwnd = asoc->peer.rwnd;
658 inflight = q->outstanding_bytes;
659 flight_size = transport->flight_size;
661 datasize = sctp_data_size(chunk);
663 if (datasize > rwnd) {
665 /* We have (at least) one data chunk in flight,
666 * so we can't fall back to rule 6.1 B).
668 retval = SCTP_XMIT_RWND_FULL;
673 /* RFC 2960 6.1 Transmission of DATA Chunks
675 * B) At any given time, the sender MUST NOT transmit new data
676 * to a given transport address if it has cwnd or more bytes
677 * of data outstanding to that transport address.
679 /* RFC 7.2.4 & the Implementers Guide 2.8.
682 * When a Fast Retransmit is being performed the sender SHOULD
683 * ignore the value of cwnd and SHOULD NOT delay retransmission.
685 if (chunk->fast_retransmit != SCTP_NEED_FRTX)
686 if (flight_size >= transport->cwnd) {
687 retval = SCTP_XMIT_RWND_FULL;
691 /* Nagle's algorithm to solve small-packet problem:
692 * Inhibit the sending of new chunks when new outgoing data arrives
693 * if any previously transmitted data on the connection remains
696 if (!sctp_sk(asoc->base.sk)->nodelay && sctp_packet_empty(packet) &&
697 inflight && sctp_state(asoc, ESTABLISHED)) {
698 unsigned int max = transport->pathmtu - packet->overhead;
699 unsigned int len = chunk->skb->len + q->out_qlen;
701 /* Check whether this chunk and all the rest of pending
702 * data will fit or delay in hopes of bundling a full
704 * Don't delay large message writes that may have been
705 * fragmeneted into small peices.
707 if ((len < max) && chunk->msg->can_delay) {
708 retval = SCTP_XMIT_NAGLE_DELAY;
717 /* This private function does management things when adding DATA chunk */
718 static void sctp_packet_append_data(struct sctp_packet *packet,
719 struct sctp_chunk *chunk)
721 struct sctp_transport *transport = packet->transport;
722 size_t datasize = sctp_data_size(chunk);
723 struct sctp_association *asoc = transport->asoc;
724 u32 rwnd = asoc->peer.rwnd;
726 /* Keep track of how many bytes are in flight over this transport. */
727 transport->flight_size += datasize;
729 /* Keep track of how many bytes are in flight to the receiver. */
730 asoc->outqueue.outstanding_bytes += datasize;
732 /* Update our view of the receiver's rwnd. */
738 asoc->peer.rwnd = rwnd;
739 /* Has been accepted for transmission. */
740 if (!asoc->peer.prsctp_capable)
741 chunk->msg->can_abandon = 0;
742 sctp_chunk_assign_tsn(chunk);
743 sctp_chunk_assign_ssn(chunk);
746 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
747 struct sctp_chunk *chunk,
753 sctp_xmit_t retval = SCTP_XMIT_OK;
755 psize = packet->size;
756 pmtu = ((packet->transport->asoc) ?
757 (packet->transport->asoc->pathmtu) :
758 (packet->transport->pathmtu));
760 too_big = (psize + chunk_len > pmtu);
762 /* Decide if we need to fragment or resubmit later. */
764 /* It's OK to fragmet at IP level if any one of the following
766 * 1. The packet is empty (meaning this chunk is greater
768 * 2. The chunk we are adding is a control chunk
769 * 3. The packet doesn't have any data in it yet and data
770 * requires authentication.
772 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
773 (!packet->has_data && chunk->auth)) {
774 /* We no longer do re-fragmentation.
775 * Just fragment at the IP layer, if we
776 * actually hit this condition
778 packet->ipfragok = 1;
780 retval = SCTP_XMIT_PMTU_FULL;