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
30 * Written or modified by:
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_transport *tp = packet->transport;
88 struct sctp_association *asoc = tp->asoc;
90 pr_debug("%s: packet:%p vtag:0x%x\n", __func__, packet, vtag);
94 if (asoc && tp->dst) {
95 struct sock *sk = asoc->base.sk;
98 if (__sk_dst_get(sk) != tp->dst) {
100 sk_setup_caps(sk, tp->dst);
103 if (sk_can_gso(sk)) {
104 struct net_device *dev = tp->dst->dev;
106 packet->max_size = dev->gso_max_size;
108 packet->max_size = asoc->pathmtu;
113 packet->max_size = tp->pathmtu;
116 if (ecn_capable && sctp_packet_empty(packet)) {
117 struct sctp_chunk *chunk;
119 /* If there a is a prepend chunk stick it on the list before
120 * any other chunks get appended.
122 chunk = sctp_get_ecne_prepend(asoc);
124 sctp_packet_append_chunk(packet, chunk);
130 /* Initialize the packet structure. */
131 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
132 struct sctp_transport *transport,
133 __u16 sport, __u16 dport)
135 struct sctp_association *asoc = transport->asoc;
138 pr_debug("%s: packet:%p transport:%p\n", __func__, packet, transport);
140 packet->transport = transport;
141 packet->source_port = sport;
142 packet->destination_port = dport;
143 INIT_LIST_HEAD(&packet->chunk_list);
145 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
146 overhead = sp->pf->af->net_header_len;
148 overhead = sizeof(struct ipv6hdr);
150 overhead += sizeof(struct sctphdr);
151 packet->overhead = overhead;
152 sctp_packet_reset(packet);
159 void sctp_packet_free(struct sctp_packet *packet)
161 struct sctp_chunk *chunk, *tmp;
163 pr_debug("%s: packet:%p\n", __func__, packet);
165 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
166 list_del_init(&chunk->list);
167 sctp_chunk_free(chunk);
171 /* This routine tries to append the chunk to the offered packet. If adding
172 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
173 * is not present in the packet, it transmits the input packet.
174 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
175 * as it can fit in the packet, but any more data that does not fit in this
176 * packet can be sent only after receiving the COOKIE_ACK.
178 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
179 struct sctp_chunk *chunk,
180 int one_packet, gfp_t gfp)
184 pr_debug("%s: packet:%p size:%Zu chunk:%p size:%d\n", __func__,
185 packet, packet->size, chunk, chunk->skb ? chunk->skb->len : -1);
187 switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
188 case SCTP_XMIT_PMTU_FULL:
189 if (!packet->has_cookie_echo) {
192 error = sctp_packet_transmit(packet, gfp);
194 chunk->skb->sk->sk_err = -error;
196 /* If we have an empty packet, then we can NOT ever
200 retval = sctp_packet_append_chunk(packet,
205 case SCTP_XMIT_RWND_FULL:
207 case SCTP_XMIT_DELAY:
214 /* Try to bundle an auth chunk into the packet. */
215 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
216 struct sctp_chunk *chunk)
218 struct sctp_association *asoc = pkt->transport->asoc;
219 struct sctp_chunk *auth;
220 sctp_xmit_t retval = SCTP_XMIT_OK;
222 /* if we don't have an association, we can't do authentication */
226 /* See if this is an auth chunk we are bundling or if
227 * auth is already bundled.
229 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
232 /* if the peer did not request this chunk to be authenticated,
238 auth = sctp_make_auth(asoc);
242 retval = __sctp_packet_append_chunk(pkt, auth);
244 if (retval != SCTP_XMIT_OK)
245 sctp_chunk_free(auth);
250 /* Try to bundle a SACK with the packet. */
251 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
252 struct sctp_chunk *chunk)
254 sctp_xmit_t retval = SCTP_XMIT_OK;
256 /* If sending DATA and haven't aleady bundled a SACK, try to
257 * bundle one in to the packet.
259 if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
260 !pkt->has_cookie_echo) {
261 struct sctp_association *asoc;
262 struct timer_list *timer;
263 asoc = pkt->transport->asoc;
264 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
266 /* If the SACK timer is running, we have a pending SACK */
267 if (timer_pending(timer)) {
268 struct sctp_chunk *sack;
270 if (pkt->transport->sack_generation !=
271 pkt->transport->asoc->peer.sack_generation)
274 asoc->a_rwnd = asoc->rwnd;
275 sack = sctp_make_sack(asoc);
277 retval = __sctp_packet_append_chunk(pkt, sack);
278 if (retval != SCTP_XMIT_OK) {
279 sctp_chunk_free(sack);
282 asoc->peer.sack_needed = 0;
283 if (del_timer(timer))
284 sctp_association_put(asoc);
293 /* Append a chunk to the offered packet reporting back any inability to do
296 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
297 struct sctp_chunk *chunk)
299 sctp_xmit_t retval = SCTP_XMIT_OK;
300 __u16 chunk_len = SCTP_PAD4(ntohs(chunk->chunk_hdr->length));
302 /* Check to see if this chunk will fit into the packet */
303 retval = sctp_packet_will_fit(packet, chunk, chunk_len);
304 if (retval != SCTP_XMIT_OK)
307 /* We believe that this chunk is OK to add to the packet */
308 switch (chunk->chunk_hdr->type) {
310 /* Account for the data being in the packet */
311 sctp_packet_append_data(packet, chunk);
312 /* Disallow SACK bundling after DATA. */
313 packet->has_sack = 1;
314 /* Disallow AUTH bundling after DATA */
315 packet->has_auth = 1;
316 /* Let it be knows that packet has DATA in it */
317 packet->has_data = 1;
318 /* timestamp the chunk for rtx purposes */
319 chunk->sent_at = jiffies;
320 /* Mainly used for prsctp RTX policy */
323 case SCTP_CID_COOKIE_ECHO:
324 packet->has_cookie_echo = 1;
328 packet->has_sack = 1;
330 chunk->asoc->stats.osacks++;
334 packet->has_auth = 1;
335 packet->auth = chunk;
339 /* It is OK to send this chunk. */
340 list_add_tail(&chunk->list, &packet->chunk_list);
341 packet->size += chunk_len;
342 chunk->transport = packet->transport;
347 /* Append a chunk to the offered packet reporting back any inability to do
350 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
351 struct sctp_chunk *chunk)
353 sctp_xmit_t retval = SCTP_XMIT_OK;
355 pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
357 /* Data chunks are special. Before seeing what else we can
358 * bundle into this packet, check to see if we are allowed to
361 if (sctp_chunk_is_data(chunk)) {
362 retval = sctp_packet_can_append_data(packet, chunk);
363 if (retval != SCTP_XMIT_OK)
367 /* Try to bundle AUTH chunk */
368 retval = sctp_packet_bundle_auth(packet, chunk);
369 if (retval != SCTP_XMIT_OK)
372 /* Try to bundle SACK chunk */
373 retval = sctp_packet_bundle_sack(packet, chunk);
374 if (retval != SCTP_XMIT_OK)
377 retval = __sctp_packet_append_chunk(packet, chunk);
383 static void sctp_packet_release_owner(struct sk_buff *skb)
388 static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
392 skb->destructor = sctp_packet_release_owner;
395 * The data chunks have already been accounted for in sctp_sendmsg(),
396 * therefore only reserve a single byte to keep socket around until
397 * the packet has been transmitted.
399 atomic_inc(&sk->sk_wmem_alloc);
402 static int sctp_packet_pack(struct sctp_packet *packet,
403 struct sk_buff *head, int gso, gfp_t gfp)
405 struct sctp_transport *tp = packet->transport;
406 struct sctp_auth_chunk *auth = NULL;
407 struct sctp_chunk *chunk, *tmp;
408 int pkt_count = 0, pkt_size;
409 struct sock *sk = head->sk;
410 struct sk_buff *nskb;
414 skb_shinfo(head)->gso_type = sk->sk_gso_type;
415 NAPI_GRO_CB(head)->last = head;
418 pkt_size = packet->size;
423 /* calculate the pkt_size and alloc nskb */
424 pkt_size = packet->overhead;
425 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list,
427 int padded = SCTP_PAD4(chunk->skb->len);
429 if (chunk == packet->auth)
431 else if (auth_len + padded + packet->overhead >
434 else if (pkt_size + padded > tp->pathmtu)
438 nskb = alloc_skb(pkt_size + MAX_HEADER, gfp);
441 skb_reserve(nskb, packet->overhead + MAX_HEADER);
444 /* merge chunks into nskb and append nskb into head list */
445 pkt_size -= packet->overhead;
446 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
449 list_del_init(&chunk->list);
450 if (sctp_chunk_is_data(chunk)) {
451 if (!sctp_chunk_retransmitted(chunk) &&
453 chunk->rtt_in_progress = 1;
458 padding = SCTP_PAD4(chunk->skb->len) - chunk->skb->len;
460 memset(skb_put(chunk->skb, padding), 0, padding);
462 if (chunk == packet->auth)
463 auth = (struct sctp_auth_chunk *)
464 skb_tail_pointer(nskb);
466 memcpy(skb_put(nskb, chunk->skb->len), chunk->skb->data,
469 pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, rtt_in_progress:%d\n",
471 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
472 chunk->has_tsn ? "TSN" : "No TSN",
473 chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
474 ntohs(chunk->chunk_hdr->length), chunk->skb->len,
475 chunk->rtt_in_progress);
477 pkt_size -= SCTP_PAD4(chunk->skb->len);
479 if (!sctp_chunk_is_data(chunk) && chunk != packet->auth)
480 sctp_chunk_free(chunk);
487 sctp_auth_calculate_hmac(tp->asoc, nskb, auth, gfp);
488 /* free auth if no more chunks, or add it back */
489 if (list_empty(&packet->chunk_list))
490 sctp_chunk_free(packet->auth);
492 list_add(&packet->auth->list,
493 &packet->chunk_list);
497 if (skb_gro_receive(&head, nskb)) {
501 if (WARN_ON_ONCE(skb_shinfo(head)->gso_segs >=
502 sk->sk_gso_max_segs))
507 } while (!list_empty(&packet->chunk_list));
510 memset(head->cb, 0, max(sizeof(struct inet_skb_parm),
511 sizeof(struct inet6_skb_parm)));
512 skb_shinfo(head)->gso_segs = pkt_count;
513 skb_shinfo(head)->gso_size = GSO_BY_FRAGS;
515 if (skb_dst(head) != tp->dst) {
517 sk_setup_caps(sk, tp->dst);
523 if (sctp_checksum_disable)
526 if (!(skb_dst(head)->dev->features & NETIF_F_SCTP_CRC) ||
527 dst_xfrm(skb_dst(head)) || packet->ipfragok) {
529 (struct sctphdr *)skb_transport_header(head);
531 sh->checksum = sctp_compute_cksum(head, 0);
534 head->ip_summed = CHECKSUM_PARTIAL;
535 head->csum_start = skb_transport_header(head) - head->head;
536 head->csum_offset = offsetof(struct sctphdr, checksum);
542 /* All packets are sent to the network through this function from
545 * The return value is always 0 for now.
547 int sctp_packet_transmit(struct sctp_packet *packet, gfp_t gfp)
549 struct sctp_transport *tp = packet->transport;
550 struct sctp_association *asoc = tp->asoc;
551 struct sctp_chunk *chunk, *tmp;
552 int pkt_count, gso = 0;
553 struct dst_entry *dst;
554 struct sk_buff *head;
558 pr_debug("%s: packet:%p\n", __func__, packet);
559 if (list_empty(&packet->chunk_list))
561 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
565 if (packet->size > tp->pathmtu && !packet->ipfragok) {
566 if (!sk_can_gso(sk)) {
567 pr_err_once("Trying to GSO but underlying device doesn't support it.");
574 head = alloc_skb((gso ? packet->overhead : packet->size) +
578 skb_reserve(head, packet->overhead + MAX_HEADER);
579 sctp_packet_set_owner_w(head, sk);
581 /* set sctp header */
582 sh = (struct sctphdr *)skb_push(head, sizeof(struct sctphdr));
583 skb_reset_transport_header(head);
584 sh->source = htons(packet->source_port);
585 sh->dest = htons(packet->destination_port);
586 sh->vtag = htonl(packet->vtag);
589 /* update dst if in need */
590 if (!sctp_transport_dst_check(tp)) {
591 sctp_transport_route(tp, NULL, sctp_sk(sk));
592 if (asoc && asoc->param_flags & SPP_PMTUD_ENABLE)
593 sctp_assoc_sync_pmtu(sk, asoc);
595 dst = dst_clone(tp->dst);
597 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
601 skb_dst_set(head, dst);
604 pkt_count = sctp_packet_pack(packet, head, gso, gfp);
609 pr_debug("***sctp_transmit_packet*** skb->len:%d\n", head->len);
611 /* start autoclose timer */
612 if (packet->has_data && sctp_state(asoc, ESTABLISHED) &&
613 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) {
614 struct timer_list *timer =
615 &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
616 unsigned long timeout =
617 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
619 if (!mod_timer(timer, jiffies + timeout))
620 sctp_association_hold(asoc);
624 tp->af_specific->ecn_capable(sk);
626 asoc->stats.opackets += pkt_count;
627 if (asoc->peer.last_sent_to != tp)
628 asoc->peer.last_sent_to = tp;
630 head->ignore_df = packet->ipfragok;
631 tp->af_specific->sctp_xmit(head, tp);
634 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
635 list_del_init(&chunk->list);
636 if (!sctp_chunk_is_data(chunk))
637 sctp_chunk_free(chunk);
639 sctp_packet_reset(packet);
643 /********************************************************************
644 * 2nd Level Abstractions
645 ********************************************************************/
647 /* This private function check to see if a chunk can be added */
648 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
649 struct sctp_chunk *chunk)
651 size_t datasize, rwnd, inflight, flight_size;
652 struct sctp_transport *transport = packet->transport;
653 struct sctp_association *asoc = transport->asoc;
654 struct sctp_outq *q = &asoc->outqueue;
656 /* RFC 2960 6.1 Transmission of DATA Chunks
658 * A) At any given time, the data sender MUST NOT transmit new data to
659 * any destination transport address if its peer's rwnd indicates
660 * that the peer has no buffer space (i.e. rwnd is 0, see Section
661 * 6.2.1). However, regardless of the value of rwnd (including if it
662 * is 0), the data sender can always have one DATA chunk in flight to
663 * the receiver if allowed by cwnd (see rule B below). This rule
664 * allows the sender to probe for a change in rwnd that the sender
665 * missed due to the SACK having been lost in transit from the data
666 * receiver to the data sender.
669 rwnd = asoc->peer.rwnd;
670 inflight = q->outstanding_bytes;
671 flight_size = transport->flight_size;
673 datasize = sctp_data_size(chunk);
675 if (datasize > rwnd && inflight > 0)
676 /* We have (at least) one data chunk in flight,
677 * so we can't fall back to rule 6.1 B).
679 return SCTP_XMIT_RWND_FULL;
681 /* RFC 2960 6.1 Transmission of DATA Chunks
683 * B) At any given time, the sender MUST NOT transmit new data
684 * to a given transport address if it has cwnd or more bytes
685 * of data outstanding to that transport address.
687 /* RFC 7.2.4 & the Implementers Guide 2.8.
690 * When a Fast Retransmit is being performed the sender SHOULD
691 * ignore the value of cwnd and SHOULD NOT delay retransmission.
693 if (chunk->fast_retransmit != SCTP_NEED_FRTX &&
694 flight_size >= transport->cwnd)
695 return SCTP_XMIT_RWND_FULL;
697 /* Nagle's algorithm to solve small-packet problem:
698 * Inhibit the sending of new chunks when new outgoing data arrives
699 * if any previously transmitted data on the connection remains
703 if (sctp_sk(asoc->base.sk)->nodelay)
707 if (!sctp_packet_empty(packet))
708 /* Append to packet */
712 /* Nothing unacked */
715 if (!sctp_state(asoc, ESTABLISHED))
718 /* Check whether this chunk and all the rest of pending data will fit
719 * or delay in hopes of bundling a full sized packet.
721 if (chunk->skb->len + q->out_qlen >
722 transport->pathmtu - packet->overhead - sizeof(sctp_data_chunk_t) - 4)
723 /* Enough data queued to fill a packet */
726 /* Don't delay large message writes that may have been fragmented */
727 if (!chunk->msg->can_delay)
730 /* Defer until all data acked or packet full */
731 return SCTP_XMIT_DELAY;
734 /* This private function does management things when adding DATA chunk */
735 static void sctp_packet_append_data(struct sctp_packet *packet,
736 struct sctp_chunk *chunk)
738 struct sctp_transport *transport = packet->transport;
739 size_t datasize = sctp_data_size(chunk);
740 struct sctp_association *asoc = transport->asoc;
741 u32 rwnd = asoc->peer.rwnd;
743 /* Keep track of how many bytes are in flight over this transport. */
744 transport->flight_size += datasize;
746 /* Keep track of how many bytes are in flight to the receiver. */
747 asoc->outqueue.outstanding_bytes += datasize;
749 /* Update our view of the receiver's rwnd. */
755 asoc->peer.rwnd = rwnd;
756 sctp_chunk_assign_tsn(chunk);
757 sctp_chunk_assign_ssn(chunk);
760 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
761 struct sctp_chunk *chunk,
764 size_t psize, pmtu, maxsize;
765 sctp_xmit_t retval = SCTP_XMIT_OK;
767 psize = packet->size;
768 if (packet->transport->asoc)
769 pmtu = packet->transport->asoc->pathmtu;
771 pmtu = packet->transport->pathmtu;
773 /* Decide if we need to fragment or resubmit later. */
774 if (psize + chunk_len > pmtu) {
775 /* It's OK to fragment at IP level if any one of the following
777 * 1. The packet is empty (meaning this chunk is greater
779 * 2. The packet doesn't have any data in it yet and data
780 * requires authentication.
782 if (sctp_packet_empty(packet) ||
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;
792 /* Similarly, if this chunk was built before a PMTU
793 * reduction, we have to fragment it at IP level now. So
794 * if the packet already contains something, we need to
797 maxsize = pmtu - packet->overhead;
799 maxsize -= SCTP_PAD4(packet->auth->skb->len);
800 if (chunk_len > maxsize)
801 retval = SCTP_XMIT_PMTU_FULL;
803 /* It is also okay to fragment if the chunk we are
804 * adding is a control chunk, but only if current packet
805 * is not a GSO one otherwise it causes fragmentation of
806 * a large frame. So in this case we allow the
807 * fragmentation by forcing it to be in a new packet.
809 if (!sctp_chunk_is_data(chunk) && packet->has_data)
810 retval = SCTP_XMIT_PMTU_FULL;
812 if (psize + chunk_len > packet->max_size)
813 /* Hit GSO/PMTU limit, gotta flush */
814 retval = SCTP_XMIT_PMTU_FULL;
816 if (!packet->transport->burst_limited &&
817 psize + chunk_len > (packet->transport->cwnd >> 1))
818 /* Do not allow a single GSO packet to use more
821 retval = SCTP_XMIT_PMTU_FULL;
823 if (packet->transport->burst_limited &&
824 psize + chunk_len > (packet->transport->burst_limited >> 1))
825 /* Do not allow a single GSO packet to use more
826 * than half of original cwnd.
828 retval = SCTP_XMIT_PMTU_FULL;
829 /* Otherwise it will fit in the GSO packet */