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>
54 #include <net/net_namespace.h>
56 #include <linux/socket.h> /* for sa_family_t */
59 #include <net/sctp/sctp.h>
60 #include <net/sctp/sm.h>
61 #include <net/sctp/checksum.h>
63 /* Forward declarations for private helpers. */
64 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
65 struct sctp_chunk *chunk);
66 static void sctp_packet_append_data(struct sctp_packet *packet,
67 struct sctp_chunk *chunk);
68 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
69 struct sctp_chunk *chunk,
72 static void sctp_packet_reset(struct sctp_packet *packet)
74 packet->size = packet->overhead;
75 packet->has_cookie_echo = 0;
84 * This appears to be a followup set of initializations.
86 struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
87 __u32 vtag, int ecn_capable)
89 struct sctp_chunk *chunk = NULL;
91 SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __func__,
94 sctp_packet_reset(packet);
97 if (ecn_capable && sctp_packet_empty(packet)) {
98 chunk = sctp_get_ecne_prepend(packet->transport->asoc);
100 /* If there a is a prepend chunk stick it on the list before
101 * any other chunks get appended.
104 sctp_packet_append_chunk(packet, chunk);
110 /* Initialize the packet structure. */
111 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
112 struct sctp_transport *transport,
113 __u16 sport, __u16 dport)
115 struct sctp_association *asoc = transport->asoc;
118 SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __func__,
121 packet->transport = transport;
122 packet->source_port = sport;
123 packet->destination_port = dport;
124 INIT_LIST_HEAD(&packet->chunk_list);
126 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
127 overhead = sp->pf->af->net_header_len;
129 overhead = sizeof(struct ipv6hdr);
131 overhead += sizeof(struct sctphdr);
132 packet->overhead = overhead;
133 sctp_packet_reset(packet);
135 packet->malloced = 0;
140 void sctp_packet_free(struct sctp_packet *packet)
142 struct sctp_chunk *chunk, *tmp;
144 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
146 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
147 list_del_init(&chunk->list);
148 sctp_chunk_free(chunk);
151 if (packet->malloced)
155 /* This routine tries to append the chunk to the offered packet. If adding
156 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
157 * is not present in the packet, it transmits the input packet.
158 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
159 * as it can fit in the packet, but any more data that does not fit in this
160 * packet can be sent only after receiving the COOKIE_ACK.
162 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
163 struct sctp_chunk *chunk,
169 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__,
172 switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
173 case SCTP_XMIT_PMTU_FULL:
174 if (!packet->has_cookie_echo) {
175 error = sctp_packet_transmit(packet);
177 chunk->skb->sk->sk_err = -error;
179 /* If we have an empty packet, then we can NOT ever
183 retval = sctp_packet_append_chunk(packet,
188 case SCTP_XMIT_RWND_FULL:
190 case SCTP_XMIT_NAGLE_DELAY:
197 /* Try to bundle an auth chunk into the packet. */
198 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
199 struct sctp_chunk *chunk)
201 struct sctp_association *asoc = pkt->transport->asoc;
202 struct sctp_chunk *auth;
203 sctp_xmit_t retval = SCTP_XMIT_OK;
205 /* if we don't have an association, we can't do authentication */
209 /* See if this is an auth chunk we are bundling or if
210 * auth is already bundled.
212 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
215 /* if the peer did not request this chunk to be authenticated,
221 auth = sctp_make_auth(asoc);
225 retval = sctp_packet_append_chunk(pkt, auth);
230 /* Try to bundle a SACK with the packet. */
231 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
232 struct sctp_chunk *chunk)
234 sctp_xmit_t retval = SCTP_XMIT_OK;
236 /* If sending DATA and haven't aleady bundled a SACK, try to
237 * bundle one in to the packet.
239 if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
240 !pkt->has_cookie_echo) {
241 struct sctp_association *asoc;
242 struct timer_list *timer;
243 asoc = pkt->transport->asoc;
244 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
246 /* If the SACK timer is running, we have a pending SACK */
247 if (timer_pending(timer)) {
248 struct sctp_chunk *sack;
249 asoc->a_rwnd = asoc->rwnd;
250 sack = sctp_make_sack(asoc);
252 retval = sctp_packet_append_chunk(pkt, sack);
253 asoc->peer.sack_needed = 0;
254 if (del_timer(timer))
255 sctp_association_put(asoc);
262 /* Append a chunk to the offered packet reporting back any inability to do
265 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
266 struct sctp_chunk *chunk)
268 sctp_xmit_t retval = SCTP_XMIT_OK;
269 __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
271 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__, packet,
274 /* Data chunks are special. Before seeing what else we can
275 * bundle into this packet, check to see if we are allowed to
278 if (sctp_chunk_is_data(chunk)) {
279 retval = sctp_packet_can_append_data(packet, chunk);
280 if (retval != SCTP_XMIT_OK)
284 /* Try to bundle AUTH chunk */
285 retval = sctp_packet_bundle_auth(packet, chunk);
286 if (retval != SCTP_XMIT_OK)
289 /* Try to bundle SACK chunk */
290 retval = sctp_packet_bundle_sack(packet, chunk);
291 if (retval != SCTP_XMIT_OK)
294 /* Check to see if this chunk will fit into the packet */
295 retval = sctp_packet_will_fit(packet, chunk, chunk_len);
296 if (retval != SCTP_XMIT_OK)
299 /* We believe that this chunk is OK to add to the packet */
300 switch (chunk->chunk_hdr->type) {
302 /* Account for the data being in the packet */
303 sctp_packet_append_data(packet, chunk);
304 /* Disallow SACK bundling after DATA. */
305 packet->has_sack = 1;
306 /* Disallow AUTH bundling after DATA */
307 packet->has_auth = 1;
308 /* Let it be knows that packet has DATA in it */
309 packet->has_data = 1;
310 /* timestamp the chunk for rtx purposes */
311 chunk->sent_at = jiffies;
313 case SCTP_CID_COOKIE_ECHO:
314 packet->has_cookie_echo = 1;
318 packet->has_sack = 1;
322 packet->has_auth = 1;
323 packet->auth = chunk;
327 /* It is OK to send this chunk. */
328 list_add_tail(&chunk->list, &packet->chunk_list);
329 packet->size += chunk_len;
330 chunk->transport = packet->transport;
335 /* All packets are sent to the network through this function from
338 * The return value is a normal kernel error return value.
340 int sctp_packet_transmit(struct sctp_packet *packet)
342 struct sctp_transport *tp = packet->transport;
343 struct sctp_association *asoc = tp->asoc;
345 struct sk_buff *nskb;
346 struct sctp_chunk *chunk, *tmp;
349 int padding; /* How much padding do we need? */
351 struct dst_entry *dst = tp->dst;
352 unsigned char *auth = NULL; /* pointer to auth in skb data */
353 __u32 cksum_buf_len = sizeof(struct sctphdr);
355 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
357 /* Do NOT generate a chunkless packet. */
358 if (list_empty(&packet->chunk_list))
361 /* Set up convenience variables... */
362 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
365 /* Allocate the new skb. */
366 nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
370 /* Make sure the outbound skb has enough header room reserved. */
371 skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
373 /* Set the owning socket so that we know where to get the
374 * destination IP address.
376 skb_set_owner_w(nskb, sk);
378 /* The 'obsolete' field of dst is set to 2 when a dst is freed. */
379 if (!dst || (dst->obsolete > 1)) {
381 sctp_transport_route(tp, NULL, sctp_sk(sk));
382 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
383 sctp_assoc_sync_pmtu(asoc);
386 dst = dst_clone(tp->dst);
387 skb_dst_set(nskb, dst);
391 /* Build the SCTP header. */
392 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
393 skb_reset_transport_header(nskb);
394 sh->source = htons(packet->source_port);
395 sh->dest = htons(packet->destination_port);
397 /* From 6.8 Adler-32 Checksum Calculation:
398 * After the packet is constructed (containing the SCTP common
399 * header and one or more control or DATA chunks), the
402 * 1) Fill in the proper Verification Tag in the SCTP common
403 * header and initialize the checksum field to 0's.
405 sh->vtag = htonl(packet->vtag);
411 * An endpoint bundles chunks by simply including multiple
412 * chunks in one outbound SCTP packet. ...
416 * 3.2 Chunk Field Descriptions
418 * The total length of a chunk (including Type, Length and
419 * Value fields) MUST be a multiple of 4 bytes. If the length
420 * of the chunk is not a multiple of 4 bytes, the sender MUST
421 * pad the chunk with all zero bytes and this padding is not
422 * included in the chunk length field. The sender should
423 * never pad with more than 3 bytes.
425 * [This whole comment explains WORD_ROUND() below.]
427 SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
428 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
429 list_del_init(&chunk->list);
430 if (sctp_chunk_is_data(chunk)) {
432 if (!chunk->has_tsn) {
433 sctp_chunk_assign_ssn(chunk);
434 sctp_chunk_assign_tsn(chunk);
436 /* 6.3.1 C4) When data is in flight and when allowed
437 * by rule C5, a new RTT measurement MUST be made each
438 * round trip. Furthermore, new RTT measurements
439 * SHOULD be made no more than once per round-trip
440 * for a given destination transport address.
443 if (!tp->rto_pending) {
444 chunk->rtt_in_progress = 1;
453 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
455 memset(skb_put(chunk->skb, padding), 0, padding);
457 /* if this is the auth chunk that we are adding,
458 * store pointer where it will be added and put
459 * the auth into the packet.
461 if (chunk == packet->auth)
462 auth = skb_tail_pointer(nskb);
464 cksum_buf_len += chunk->skb->len;
465 memcpy(skb_put(nskb, chunk->skb->len),
466 chunk->skb->data, chunk->skb->len);
468 SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n",
470 sctp_cname(SCTP_ST_CHUNK(
471 chunk->chunk_hdr->type)),
472 chunk->has_tsn ? "TSN" : "No TSN",
474 ntohl(chunk->subh.data_hdr->tsn) : 0,
475 "length", ntohs(chunk->chunk_hdr->length),
476 "chunk->skb->len", chunk->skb->len,
477 "rtt_in_progress", chunk->rtt_in_progress);
480 * If this is a control chunk, this is our last
481 * reference. Free data chunks after they've been
482 * acknowledged or have failed.
484 if (!sctp_chunk_is_data(chunk))
485 sctp_chunk_free(chunk);
488 /* SCTP-AUTH, Section 6.2
489 * The sender MUST calculate the MAC as described in RFC2104 [2]
490 * using the hash function H as described by the MAC Identifier and
491 * the shared association key K based on the endpoint pair shared key
492 * described by the shared key identifier. The 'data' used for the
493 * computation of the AUTH-chunk is given by the AUTH chunk with its
494 * HMAC field set to zero (as shown in Figure 6) followed by all
495 * chunks that are placed after the AUTH chunk in the SCTP packet.
498 sctp_auth_calculate_hmac(asoc, nskb,
499 (struct sctp_auth_chunk *)auth,
502 /* 2) Calculate the Adler-32 checksum of the whole packet,
503 * including the SCTP common header and all the
506 * Note: Adler-32 is no longer applicable, as has been replaced
507 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
509 if (!sctp_checksum_disable &&
510 !(dst->dev->features & (NETIF_F_NO_CSUM | NETIF_F_SCTP_CSUM))) {
511 __u32 crc32 = sctp_start_cksum((__u8 *)sh, cksum_buf_len);
513 /* 3) Put the resultant value into the checksum field in the
514 * common header, and leave the rest of the bits unchanged.
516 sh->checksum = sctp_end_cksum(crc32);
518 if (dst->dev->features & NETIF_F_SCTP_CSUM) {
519 /* no need to seed psuedo checksum for SCTP */
520 nskb->ip_summed = CHECKSUM_PARTIAL;
521 nskb->csum_start = (skb_transport_header(nskb) -
523 nskb->csum_offset = offsetof(struct sctphdr, checksum);
525 nskb->ip_summed = CHECKSUM_UNNECESSARY;
529 /* IP layer ECN support
531 * "The ECN-Capable Transport (ECT) bit would be set by the
532 * data sender to indicate that the end-points of the
533 * transport protocol are ECN-capable."
535 * Now setting the ECT bit all the time, as it should not cause
536 * any problems protocol-wise even if our peer ignores it.
538 * Note: The works for IPv6 layer checks this bit too later
539 * in transmission. See IP6_ECN_flow_xmit().
541 (*tp->af_specific->ecn_capable)(nskb->sk);
543 /* Set up the IP options. */
544 /* BUG: not implemented
545 * For v4 this all lives somewhere in sk->sk_opt...
548 /* Dump that on IP! */
549 if (asoc && asoc->peer.last_sent_to != tp) {
550 /* Considering the multiple CPU scenario, this is a
551 * "correcter" place for last_sent_to. --xguo
553 asoc->peer.last_sent_to = tp;
557 struct timer_list *timer;
558 unsigned long timeout;
560 tp->last_time_used = jiffies;
562 /* Restart the AUTOCLOSE timer when sending data. */
563 if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) {
564 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
565 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
567 if (!mod_timer(timer, jiffies + timeout))
568 sctp_association_hold(asoc);
572 SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n",
575 nskb->local_df = packet->ipfragok;
576 (*tp->af_specific->sctp_xmit)(nskb, tp);
579 sctp_packet_reset(packet);
583 IP_INC_STATS_BH(&init_net, IPSTATS_MIB_OUTNOROUTES);
585 /* FIXME: Returning the 'err' will effect all the associations
586 * associated with a socket, although only one of the paths of the
587 * association is unreachable.
588 * The real failure of a transport or association can be passed on
589 * to the user via notifications. So setting this error may not be
592 /* err = -EHOSTUNREACH; */
594 /* Control chunks are unreliable so just drop them. DATA chunks
595 * will get resent or dropped later.
598 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
599 list_del_init(&chunk->list);
600 if (!sctp_chunk_is_data(chunk))
601 sctp_chunk_free(chunk);
609 /********************************************************************
610 * 2nd Level Abstractions
611 ********************************************************************/
613 /* This private function check to see if a chunk can be added */
614 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
615 struct sctp_chunk *chunk)
617 sctp_xmit_t retval = SCTP_XMIT_OK;
618 size_t datasize, rwnd, inflight, flight_size;
619 struct sctp_transport *transport = packet->transport;
620 __u32 max_burst_bytes;
621 struct sctp_association *asoc = transport->asoc;
622 struct sctp_outq *q = &asoc->outqueue;
624 /* RFC 2960 6.1 Transmission of DATA Chunks
626 * A) At any given time, the data sender MUST NOT transmit new data to
627 * any destination transport address if its peer's rwnd indicates
628 * that the peer has no buffer space (i.e. rwnd is 0, see Section
629 * 6.2.1). However, regardless of the value of rwnd (including if it
630 * is 0), the data sender can always have one DATA chunk in flight to
631 * the receiver if allowed by cwnd (see rule B below). This rule
632 * allows the sender to probe for a change in rwnd that the sender
633 * missed due to the SACK having been lost in transit from the data
634 * receiver to the data sender.
637 rwnd = asoc->peer.rwnd;
638 inflight = q->outstanding_bytes;
639 flight_size = transport->flight_size;
641 datasize = sctp_data_size(chunk);
643 if (datasize > rwnd) {
645 /* We have (at least) one data chunk in flight,
646 * so we can't fall back to rule 6.1 B).
648 retval = SCTP_XMIT_RWND_FULL;
653 /* sctpimpguide-05 2.14.2
654 * D) When the time comes for the sender to
655 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
656 * first be applied to limit how many new DATA chunks may be sent.
657 * The limit is applied by adjusting cwnd as follows:
658 * if ((flightsize + Max.Burst * MTU) < cwnd)
659 * cwnd = flightsize + Max.Burst * MTU
661 max_burst_bytes = asoc->max_burst * asoc->pathmtu;
662 if ((flight_size + max_burst_bytes) < transport->cwnd) {
663 transport->cwnd = flight_size + max_burst_bytes;
664 SCTP_DEBUG_PRINTK("%s: cwnd limited by max_burst: "
665 "transport: %p, cwnd: %d, "
666 "ssthresh: %d, flight_size: %d, "
671 transport->flight_size,
672 transport->partial_bytes_acked);
675 /* RFC 2960 6.1 Transmission of DATA Chunks
677 * B) At any given time, the sender MUST NOT transmit new data
678 * to a given transport address if it has cwnd or more bytes
679 * of data outstanding to that transport address.
681 /* RFC 7.2.4 & the Implementers Guide 2.8.
684 * When a Fast Retransmit is being performed the sender SHOULD
685 * ignore the value of cwnd and SHOULD NOT delay retransmission.
687 if (chunk->fast_retransmit != SCTP_NEED_FRTX)
688 if (flight_size >= transport->cwnd) {
689 retval = SCTP_XMIT_RWND_FULL;
693 /* Nagle's algorithm to solve small-packet problem:
694 * Inhibit the sending of new chunks when new outgoing data arrives
695 * if any previously transmitted data on the connection remains
698 if (!sctp_sk(asoc->base.sk)->nodelay && sctp_packet_empty(packet) &&
699 inflight && sctp_state(asoc, ESTABLISHED)) {
700 unsigned max = transport->pathmtu - packet->overhead;
701 unsigned len = chunk->skb->len + q->out_qlen;
703 /* Check whether this chunk and all the rest of pending
704 * data will fit or delay in hopes of bundling a full
706 * Don't delay large message writes that may have been
707 * fragmeneted into small peices.
709 if ((len < max) && (chunk->msg->msg_size < max)) {
710 retval = SCTP_XMIT_NAGLE_DELAY;
719 /* This private function does management things when adding DATA chunk */
720 static void sctp_packet_append_data(struct sctp_packet *packet,
721 struct sctp_chunk *chunk)
723 struct sctp_transport *transport = packet->transport;
724 size_t datasize = sctp_data_size(chunk);
725 struct sctp_association *asoc = transport->asoc;
726 u32 rwnd = asoc->peer.rwnd;
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;
752 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
753 struct sctp_chunk *chunk,
759 sctp_xmit_t retval = SCTP_XMIT_OK;
761 psize = packet->size;
762 pmtu = ((packet->transport->asoc) ?
763 (packet->transport->asoc->pathmtu) :
764 (packet->transport->pathmtu));
766 too_big = (psize + chunk_len > pmtu);
768 /* Decide if we need to fragment or resubmit later. */
770 /* It's OK to fragmet at IP level if any one of the following
772 * 1. The packet is empty (meaning this chunk is greater
774 * 2. The chunk we are adding is a control chunk
775 * 3. The packet doesn't have any data in it yet and data
776 * requires authentication.
778 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
779 (!packet->has_data && chunk->auth)) {
780 /* We no longer do re-fragmentation.
781 * Just fragment at the IP layer, if we
782 * actually hit this condition
784 packet->ipfragok = 1;
786 retval = SCTP_XMIT_PMTU_FULL;