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_append_data(struct sctp_packet *packet,
65 struct sctp_chunk *chunk);
68 * This appears to be a followup set of initializations.
70 struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
71 __u32 vtag, int ecn_capable)
73 struct sctp_chunk *chunk = NULL;
75 SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __func__,
79 packet->has_cookie_echo = 0;
86 if (ecn_capable && sctp_packet_empty(packet)) {
87 chunk = sctp_get_ecne_prepend(packet->transport->asoc);
89 /* If there a is a prepend chunk stick it on the list before
90 * any other chunks get appended.
93 sctp_packet_append_chunk(packet, chunk);
99 /* Initialize the packet structure. */
100 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
101 struct sctp_transport *transport,
102 __u16 sport, __u16 dport)
104 struct sctp_association *asoc = transport->asoc;
107 SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __func__,
110 packet->transport = transport;
111 packet->source_port = sport;
112 packet->destination_port = dport;
113 INIT_LIST_HEAD(&packet->chunk_list);
115 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
116 overhead = sp->pf->af->net_header_len;
118 overhead = sizeof(struct ipv6hdr);
120 overhead += sizeof(struct sctphdr);
121 packet->overhead = overhead;
122 packet->size = overhead;
124 packet->has_cookie_echo = 0;
125 packet->has_sack = 0;
126 packet->has_auth = 0;
127 packet->has_data = 0;
128 packet->ipfragok = 0;
129 packet->malloced = 0;
135 void sctp_packet_free(struct sctp_packet *packet)
137 struct sctp_chunk *chunk, *tmp;
139 SCTP_DEBUG_PRINTK("%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);
146 if (packet->malloced)
150 /* This routine tries to append the chunk to the offered packet. If adding
151 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
152 * is not present in the packet, it transmits the input packet.
153 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
154 * as it can fit in the packet, but any more data that does not fit in this
155 * packet can be sent only after receiving the COOKIE_ACK.
157 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
158 struct sctp_chunk *chunk,
164 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__,
167 switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
168 case SCTP_XMIT_PMTU_FULL:
169 if (!packet->has_cookie_echo) {
170 error = sctp_packet_transmit(packet);
172 chunk->skb->sk->sk_err = -error;
174 /* If we have an empty packet, then we can NOT ever
178 retval = sctp_packet_append_chunk(packet,
183 case SCTP_XMIT_RWND_FULL:
185 case SCTP_XMIT_NAGLE_DELAY:
192 /* Try to bundle an auth chunk into the packet. */
193 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
194 struct sctp_chunk *chunk)
196 struct sctp_association *asoc = pkt->transport->asoc;
197 struct sctp_chunk *auth;
198 sctp_xmit_t retval = SCTP_XMIT_OK;
200 /* if we don't have an association, we can't do authentication */
204 /* See if this is an auth chunk we are bundling or if
205 * auth is already bundled.
207 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->auth)
210 /* if the peer did not request this chunk to be authenticated,
216 auth = sctp_make_auth(asoc);
220 retval = sctp_packet_append_chunk(pkt, 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 asoc = pkt->transport->asoc;
239 if (asoc->a_rwnd > asoc->rwnd) {
240 struct sctp_chunk *sack;
241 asoc->a_rwnd = asoc->rwnd;
242 sack = sctp_make_sack(asoc);
244 struct timer_list *timer;
245 retval = sctp_packet_append_chunk(pkt, sack);
246 asoc->peer.sack_needed = 0;
247 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
248 if (timer_pending(timer) && del_timer(timer))
249 sctp_association_put(asoc);
256 /* Append a chunk to the offered packet reporting back any inability to do
259 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
260 struct sctp_chunk *chunk)
262 sctp_xmit_t retval = SCTP_XMIT_OK;
263 __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
268 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__, packet,
271 /* Try to bundle AUTH chunk */
272 retval = sctp_packet_bundle_auth(packet, chunk);
273 if (retval != SCTP_XMIT_OK)
276 /* Try to bundle SACK chunk */
277 retval = sctp_packet_bundle_sack(packet, chunk);
278 if (retval != SCTP_XMIT_OK)
281 psize = packet->size;
282 pmtu = ((packet->transport->asoc) ?
283 (packet->transport->asoc->pathmtu) :
284 (packet->transport->pathmtu));
286 too_big = (psize + chunk_len > pmtu);
288 /* Decide if we need to fragment or resubmit later. */
290 /* It's OK to fragmet at IP level if any one of the following
292 * 1. The packet is empty (meaning this chunk is greater
294 * 2. The chunk we are adding is a control chunk
295 * 3. The packet doesn't have any data in it yet and data
296 * requires authentication.
298 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
299 (!packet->has_data && chunk->auth)) {
300 /* We no longer do re-fragmentation.
301 * Just fragment at the IP layer, if we
302 * actually hit this condition
304 packet->ipfragok = 1;
308 retval = SCTP_XMIT_PMTU_FULL;
314 /* We believe that this chunk is OK to add to the packet (as
315 * long as we have the cwnd for it).
318 /* DATA is a special case since we must examine both rwnd and cwnd
319 * before we send DATA.
321 switch (chunk->chunk_hdr->type) {
323 retval = sctp_packet_append_data(packet, chunk);
324 if (SCTP_XMIT_OK != retval)
326 /* Disallow SACK bundling after DATA. */
327 packet->has_sack = 1;
328 /* Disallow AUTH bundling after DATA */
329 packet->has_auth = 1;
330 /* Let it be knows that packet has DATA in it */
331 packet->has_data = 1;
332 /* timestamp the chunk for rtx purposes */
333 chunk->sent_at = jiffies;
335 case SCTP_CID_COOKIE_ECHO:
336 packet->has_cookie_echo = 1;
340 packet->has_sack = 1;
344 packet->has_auth = 1;
345 packet->auth = chunk;
349 /* It is OK to send this chunk. */
350 list_add_tail(&chunk->list, &packet->chunk_list);
351 packet->size += chunk_len;
352 chunk->transport = packet->transport;
357 /* All packets are sent to the network through this function from
360 * The return value is a normal kernel error return value.
362 int sctp_packet_transmit(struct sctp_packet *packet)
364 struct sctp_transport *tp = packet->transport;
365 struct sctp_association *asoc = tp->asoc;
367 struct sk_buff *nskb;
368 struct sctp_chunk *chunk, *tmp;
371 int padding; /* How much padding do we need? */
373 struct dst_entry *dst = tp->dst;
374 unsigned char *auth = NULL; /* pointer to auth in skb data */
375 __u32 cksum_buf_len = sizeof(struct sctphdr);
377 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
379 /* Do NOT generate a chunkless packet. */
380 if (list_empty(&packet->chunk_list))
383 /* Set up convenience variables... */
384 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
387 /* Allocate the new skb. */
388 nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
392 /* Make sure the outbound skb has enough header room reserved. */
393 skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
395 /* Set the owning socket so that we know where to get the
396 * destination IP address.
398 skb_set_owner_w(nskb, sk);
400 /* The 'obsolete' field of dst is set to 2 when a dst is freed. */
401 if (!dst || (dst->obsolete > 1)) {
403 sctp_transport_route(tp, NULL, sctp_sk(sk));
404 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
405 sctp_assoc_sync_pmtu(asoc);
408 nskb->dst = dst_clone(tp->dst);
413 /* Build the SCTP header. */
414 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
415 sh->source = htons(packet->source_port);
416 sh->dest = htons(packet->destination_port);
418 /* From 6.8 Adler-32 Checksum Calculation:
419 * After the packet is constructed (containing the SCTP common
420 * header and one or more control or DATA chunks), the
423 * 1) Fill in the proper Verification Tag in the SCTP common
424 * header and initialize the checksum field to 0's.
426 sh->vtag = htonl(packet->vtag);
432 * An endpoint bundles chunks by simply including multiple
433 * chunks in one outbound SCTP packet. ...
437 * 3.2 Chunk Field Descriptions
439 * The total length of a chunk (including Type, Length and
440 * Value fields) MUST be a multiple of 4 bytes. If the length
441 * of the chunk is not a multiple of 4 bytes, the sender MUST
442 * pad the chunk with all zero bytes and this padding is not
443 * included in the chunk length field. The sender should
444 * never pad with more than 3 bytes.
446 * [This whole comment explains WORD_ROUND() below.]
448 SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
449 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
450 list_del_init(&chunk->list);
451 if (sctp_chunk_is_data(chunk)) {
453 if (!chunk->has_tsn) {
454 sctp_chunk_assign_ssn(chunk);
455 sctp_chunk_assign_tsn(chunk);
457 /* 6.3.1 C4) When data is in flight and when allowed
458 * by rule C5, a new RTT measurement MUST be made each
459 * round trip. Furthermore, new RTT measurements
460 * SHOULD be made no more than once per round-trip
461 * for a given destination transport address.
464 if (!tp->rto_pending) {
465 chunk->rtt_in_progress = 1;
474 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
476 memset(skb_put(chunk->skb, padding), 0, padding);
478 /* if this is the auth chunk that we are adding,
479 * store pointer where it will be added and put
480 * the auth into the packet.
482 if (chunk == packet->auth)
483 auth = skb_tail_pointer(nskb);
485 cksum_buf_len += chunk->skb->len;
486 memcpy(skb_put(nskb, chunk->skb->len),
487 chunk->skb->data, chunk->skb->len);
489 SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n",
491 sctp_cname(SCTP_ST_CHUNK(
492 chunk->chunk_hdr->type)),
493 chunk->has_tsn ? "TSN" : "No TSN",
495 ntohl(chunk->subh.data_hdr->tsn) : 0,
496 "length", ntohs(chunk->chunk_hdr->length),
497 "chunk->skb->len", chunk->skb->len,
498 "rtt_in_progress", chunk->rtt_in_progress);
501 * If this is a control chunk, this is our last
502 * reference. Free data chunks after they've been
503 * acknowledged or have failed.
505 if (!sctp_chunk_is_data(chunk))
506 sctp_chunk_free(chunk);
509 /* SCTP-AUTH, Section 6.2
510 * The sender MUST calculate the MAC as described in RFC2104 [2]
511 * using the hash function H as described by the MAC Identifier and
512 * the shared association key K based on the endpoint pair shared key
513 * described by the shared key identifier. The 'data' used for the
514 * computation of the AUTH-chunk is given by the AUTH chunk with its
515 * HMAC field set to zero (as shown in Figure 6) followed by all
516 * chunks that are placed after the AUTH chunk in the SCTP packet.
519 sctp_auth_calculate_hmac(asoc, nskb,
520 (struct sctp_auth_chunk *)auth,
523 /* 2) Calculate the Adler-32 checksum of the whole packet,
524 * including the SCTP common header and all the
527 * Note: Adler-32 is no longer applicable, as has been replaced
528 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
530 if (!sctp_checksum_disable && !(dst->dev->features & NETIF_F_NO_CSUM)) {
531 __u32 crc32 = sctp_start_cksum((__u8 *)sh, cksum_buf_len);
533 /* 3) Put the resultant value into the checksum field in the
534 * common header, and leave the rest of the bits unchanged.
536 sh->checksum = sctp_end_cksum(crc32);
538 nskb->ip_summed = CHECKSUM_UNNECESSARY;
540 /* IP layer ECN support
542 * "The ECN-Capable Transport (ECT) bit would be set by the
543 * data sender to indicate that the end-points of the
544 * transport protocol are ECN-capable."
546 * Now setting the ECT bit all the time, as it should not cause
547 * any problems protocol-wise even if our peer ignores it.
549 * Note: The works for IPv6 layer checks this bit too later
550 * in transmission. See IP6_ECN_flow_xmit().
552 (*tp->af_specific->ecn_capable)(nskb->sk);
554 /* Set up the IP options. */
555 /* BUG: not implemented
556 * For v4 this all lives somewhere in sk->sk_opt...
559 /* Dump that on IP! */
560 if (asoc && asoc->peer.last_sent_to != tp) {
561 /* Considering the multiple CPU scenario, this is a
562 * "correcter" place for last_sent_to. --xguo
564 asoc->peer.last_sent_to = tp;
568 struct timer_list *timer;
569 unsigned long timeout;
571 tp->last_time_used = jiffies;
573 /* Restart the AUTOCLOSE timer when sending data. */
574 if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) {
575 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
576 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
578 if (!mod_timer(timer, jiffies + timeout))
579 sctp_association_hold(asoc);
583 SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n",
586 nskb->local_df = packet->ipfragok;
587 (*tp->af_specific->sctp_xmit)(nskb, tp);
590 packet->size = packet->overhead;
594 IP_INC_STATS_BH(&init_net, IPSTATS_MIB_OUTNOROUTES);
596 /* FIXME: Returning the 'err' will effect all the associations
597 * associated with a socket, although only one of the paths of the
598 * association is unreachable.
599 * The real failure of a transport or association can be passed on
600 * to the user via notifications. So setting this error may not be
603 /* err = -EHOSTUNREACH; */
605 /* Control chunks are unreliable so just drop them. DATA chunks
606 * will get resent or dropped later.
609 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
610 list_del_init(&chunk->list);
611 if (!sctp_chunk_is_data(chunk))
612 sctp_chunk_free(chunk);
620 /********************************************************************
621 * 2nd Level Abstractions
622 ********************************************************************/
624 /* This private function handles the specifics of appending DATA chunks. */
625 static sctp_xmit_t sctp_packet_append_data(struct sctp_packet *packet,
626 struct sctp_chunk *chunk)
628 sctp_xmit_t retval = SCTP_XMIT_OK;
629 size_t datasize, rwnd, inflight;
630 struct sctp_transport *transport = packet->transport;
631 __u32 max_burst_bytes;
632 struct sctp_association *asoc = transport->asoc;
633 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
634 struct sctp_outq *q = &asoc->outqueue;
636 /* RFC 2960 6.1 Transmission of DATA Chunks
638 * A) At any given time, the data sender MUST NOT transmit new data to
639 * any destination transport address if its peer's rwnd indicates
640 * that the peer has no buffer space (i.e. rwnd is 0, see Section
641 * 6.2.1). However, regardless of the value of rwnd (including if it
642 * is 0), the data sender can always have one DATA chunk in flight to
643 * the receiver if allowed by cwnd (see rule B below). This rule
644 * allows the sender to probe for a change in rwnd that the sender
645 * missed due to the SACK having been lost in transit from the data
646 * receiver to the data sender.
649 rwnd = asoc->peer.rwnd;
650 inflight = asoc->outqueue.outstanding_bytes;
652 datasize = sctp_data_size(chunk);
654 if (datasize > rwnd) {
656 /* We have (at least) one data chunk in flight,
657 * so we can't fall back to rule 6.1 B).
659 retval = SCTP_XMIT_RWND_FULL;
664 /* sctpimpguide-05 2.14.2
665 * D) When the time comes for the sender to
666 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
667 * first be applied to limit how many new DATA chunks may be sent.
668 * The limit is applied by adjusting cwnd as follows:
669 * if ((flightsize + Max.Burst * MTU) < cwnd)
670 * cwnd = flightsize + Max.Burst * MTU
672 max_burst_bytes = asoc->max_burst * asoc->pathmtu;
673 if ((transport->flight_size + max_burst_bytes) < transport->cwnd) {
674 transport->cwnd = transport->flight_size + max_burst_bytes;
675 SCTP_DEBUG_PRINTK("%s: cwnd limited by max_burst: "
676 "transport: %p, cwnd: %d, "
677 "ssthresh: %d, flight_size: %d, "
682 transport->flight_size,
683 transport->partial_bytes_acked);
686 /* RFC 2960 6.1 Transmission of DATA Chunks
688 * B) At any given time, the sender MUST NOT transmit new data
689 * to a given transport address if it has cwnd or more bytes
690 * of data outstanding to that transport address.
692 /* RFC 7.2.4 & the Implementers Guide 2.8.
695 * When a Fast Retransmit is being performed the sender SHOULD
696 * ignore the value of cwnd and SHOULD NOT delay retransmission.
698 if (chunk->fast_retransmit != SCTP_NEED_FRTX)
699 if (transport->flight_size >= transport->cwnd) {
700 retval = SCTP_XMIT_RWND_FULL;
704 /* Nagle's algorithm to solve small-packet problem:
705 * Inhibit the sending of new chunks when new outgoing data arrives
706 * if any previously transmitted data on the connection remains
709 if (!sp->nodelay && sctp_packet_empty(packet) &&
710 q->outstanding_bytes && sctp_state(asoc, ESTABLISHED)) {
711 unsigned len = datasize + q->out_qlen;
713 /* Check whether this chunk and all the rest of pending
714 * data will fit or delay in hopes of bundling a full
717 if (len < asoc->frag_point) {
718 retval = SCTP_XMIT_NAGLE_DELAY;
723 /* Keep track of how many bytes are in flight over this transport. */
724 transport->flight_size += datasize;
726 /* Keep track of how many bytes are in flight to the receiver. */
727 asoc->outqueue.outstanding_bytes += datasize;
729 /* Update our view of the receiver's rwnd. Include sk_buff overhead
730 * while updating peer.rwnd so that it reduces the chances of a
731 * receiver running out of receive buffer space even when receive
732 * window is still open. This can happen when a sender is sending
733 * sending small messages.
735 datasize += sizeof(struct sk_buff);
741 asoc->peer.rwnd = rwnd;
742 /* Has been accepted for transmission. */
743 if (!asoc->peer.prsctp_capable)
744 chunk->msg->can_abandon = 0;