]> git.karo-electronics.de Git - karo-tx-linux.git/blob - net/sctp/output.c
Merge tag 'openrisc-for-linus' of git://github.com/openrisc/linux
[karo-tx-linux.git] / net / sctp / output.c
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.
5  *
6  * This file is part of the SCTP kernel implementation
7  *
8  * These functions handle output processing.
9  *
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)
14  * any later version.
15  *
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.
21  *
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/>.
25  *
26  * Please send any bug reports or fixes you make to the
27  * email address(es):
28  *    lksctp developers <linux-sctp@vger.kernel.org>
29  *
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>
35  */
36
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39 #include <linux/types.h>
40 #include <linux/kernel.h>
41 #include <linux/wait.h>
42 #include <linux/time.h>
43 #include <linux/ip.h>
44 #include <linux/ipv6.h>
45 #include <linux/init.h>
46 #include <linux/slab.h>
47 #include <net/inet_ecn.h>
48 #include <net/ip.h>
49 #include <net/icmp.h>
50 #include <net/net_namespace.h>
51
52 #include <linux/socket.h> /* for sa_family_t */
53 #include <net/sock.h>
54
55 #include <net/sctp/sctp.h>
56 #include <net/sctp/sm.h>
57 #include <net/sctp/checksum.h>
58
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,
68                                         u16 chunk_len);
69
70 static void sctp_packet_reset(struct sctp_packet *packet)
71 {
72         packet->size = packet->overhead;
73         packet->has_cookie_echo = 0;
74         packet->has_sack = 0;
75         packet->has_data = 0;
76         packet->has_auth = 0;
77         packet->ipfragok = 0;
78         packet->auth = NULL;
79 }
80
81 /* Config a packet.
82  * This appears to be a followup set of initializations.
83  */
84 void sctp_packet_config(struct sctp_packet *packet, __u32 vtag,
85                         int ecn_capable)
86 {
87         struct sctp_transport *tp = packet->transport;
88         struct sctp_association *asoc = tp->asoc;
89
90         pr_debug("%s: packet:%p vtag:0x%x\n", __func__, packet, vtag);
91
92         packet->vtag = vtag;
93
94         if (asoc && tp->dst) {
95                 struct sock *sk = asoc->base.sk;
96
97                 rcu_read_lock();
98                 if (__sk_dst_get(sk) != tp->dst) {
99                         dst_hold(tp->dst);
100                         sk_setup_caps(sk, tp->dst);
101                 }
102
103                 if (sk_can_gso(sk)) {
104                         struct net_device *dev = tp->dst->dev;
105
106                         packet->max_size = dev->gso_max_size;
107                 } else {
108                         packet->max_size = asoc->pathmtu;
109                 }
110                 rcu_read_unlock();
111
112         } else {
113                 packet->max_size = tp->pathmtu;
114         }
115
116         if (ecn_capable && sctp_packet_empty(packet)) {
117                 struct sctp_chunk *chunk;
118
119                 /* If there a is a prepend chunk stick it on the list before
120                  * any other chunks get appended.
121                  */
122                 chunk = sctp_get_ecne_prepend(asoc);
123                 if (chunk)
124                         sctp_packet_append_chunk(packet, chunk);
125         }
126 }
127
128 /* Initialize the packet structure. */
129 void sctp_packet_init(struct sctp_packet *packet,
130                       struct sctp_transport *transport,
131                       __u16 sport, __u16 dport)
132 {
133         struct sctp_association *asoc = transport->asoc;
134         size_t overhead;
135
136         pr_debug("%s: packet:%p transport:%p\n", __func__, packet, transport);
137
138         packet->transport = transport;
139         packet->source_port = sport;
140         packet->destination_port = dport;
141         INIT_LIST_HEAD(&packet->chunk_list);
142         if (asoc) {
143                 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
144                 overhead = sp->pf->af->net_header_len;
145         } else {
146                 overhead = sizeof(struct ipv6hdr);
147         }
148         overhead += sizeof(struct sctphdr);
149         packet->overhead = overhead;
150         sctp_packet_reset(packet);
151         packet->vtag = 0;
152 }
153
154 /* Free a packet.  */
155 void sctp_packet_free(struct sctp_packet *packet)
156 {
157         struct sctp_chunk *chunk, *tmp;
158
159         pr_debug("%s: packet:%p\n", __func__, packet);
160
161         list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
162                 list_del_init(&chunk->list);
163                 sctp_chunk_free(chunk);
164         }
165 }
166
167 /* This routine tries to append the chunk to the offered packet. If adding
168  * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
169  * is not present in the packet, it transmits the input packet.
170  * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
171  * as it can fit in the packet, but any more data that does not fit in this
172  * packet can be sent only after receiving the COOKIE_ACK.
173  */
174 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
175                                        struct sctp_chunk *chunk,
176                                        int one_packet, gfp_t gfp)
177 {
178         sctp_xmit_t retval;
179
180         pr_debug("%s: packet:%p size:%Zu chunk:%p size:%d\n", __func__,
181                  packet, packet->size, chunk, chunk->skb ? chunk->skb->len : -1);
182
183         switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
184         case SCTP_XMIT_PMTU_FULL:
185                 if (!packet->has_cookie_echo) {
186                         int error = 0;
187
188                         error = sctp_packet_transmit(packet, gfp);
189                         if (error < 0)
190                                 chunk->skb->sk->sk_err = -error;
191
192                         /* If we have an empty packet, then we can NOT ever
193                          * return PMTU_FULL.
194                          */
195                         if (!one_packet)
196                                 retval = sctp_packet_append_chunk(packet,
197                                                                   chunk);
198                 }
199                 break;
200
201         case SCTP_XMIT_RWND_FULL:
202         case SCTP_XMIT_OK:
203         case SCTP_XMIT_DELAY:
204                 break;
205         }
206
207         return retval;
208 }
209
210 /* Try to bundle an auth chunk into the packet. */
211 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
212                                            struct sctp_chunk *chunk)
213 {
214         struct sctp_association *asoc = pkt->transport->asoc;
215         struct sctp_chunk *auth;
216         sctp_xmit_t retval = SCTP_XMIT_OK;
217
218         /* if we don't have an association, we can't do authentication */
219         if (!asoc)
220                 return retval;
221
222         /* See if this is an auth chunk we are bundling or if
223          * auth is already bundled.
224          */
225         if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
226                 return retval;
227
228         /* if the peer did not request this chunk to be authenticated,
229          * don't do it
230          */
231         if (!chunk->auth)
232                 return retval;
233
234         auth = sctp_make_auth(asoc);
235         if (!auth)
236                 return retval;
237
238         retval = __sctp_packet_append_chunk(pkt, auth);
239
240         if (retval != SCTP_XMIT_OK)
241                 sctp_chunk_free(auth);
242
243         return retval;
244 }
245
246 /* Try to bundle a SACK with the packet. */
247 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
248                                            struct sctp_chunk *chunk)
249 {
250         sctp_xmit_t retval = SCTP_XMIT_OK;
251
252         /* If sending DATA and haven't aleady bundled a SACK, try to
253          * bundle one in to the packet.
254          */
255         if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
256             !pkt->has_cookie_echo) {
257                 struct sctp_association *asoc;
258                 struct timer_list *timer;
259                 asoc = pkt->transport->asoc;
260                 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
261
262                 /* If the SACK timer is running, we have a pending SACK */
263                 if (timer_pending(timer)) {
264                         struct sctp_chunk *sack;
265
266                         if (pkt->transport->sack_generation !=
267                             pkt->transport->asoc->peer.sack_generation)
268                                 return retval;
269
270                         asoc->a_rwnd = asoc->rwnd;
271                         sack = sctp_make_sack(asoc);
272                         if (sack) {
273                                 retval = __sctp_packet_append_chunk(pkt, sack);
274                                 if (retval != SCTP_XMIT_OK) {
275                                         sctp_chunk_free(sack);
276                                         goto out;
277                                 }
278                                 asoc->peer.sack_needed = 0;
279                                 if (del_timer(timer))
280                                         sctp_association_put(asoc);
281                         }
282                 }
283         }
284 out:
285         return retval;
286 }
287
288
289 /* Append a chunk to the offered packet reporting back any inability to do
290  * so.
291  */
292 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
293                                               struct sctp_chunk *chunk)
294 {
295         sctp_xmit_t retval = SCTP_XMIT_OK;
296         __u16 chunk_len = SCTP_PAD4(ntohs(chunk->chunk_hdr->length));
297
298         /* Check to see if this chunk will fit into the packet */
299         retval = sctp_packet_will_fit(packet, chunk, chunk_len);
300         if (retval != SCTP_XMIT_OK)
301                 goto finish;
302
303         /* We believe that this chunk is OK to add to the packet */
304         switch (chunk->chunk_hdr->type) {
305         case SCTP_CID_DATA:
306                 /* Account for the data being in the packet */
307                 sctp_packet_append_data(packet, chunk);
308                 /* Disallow SACK bundling after DATA. */
309                 packet->has_sack = 1;
310                 /* Disallow AUTH bundling after DATA */
311                 packet->has_auth = 1;
312                 /* Let it be knows that packet has DATA in it */
313                 packet->has_data = 1;
314                 /* timestamp the chunk for rtx purposes */
315                 chunk->sent_at = jiffies;
316                 /* Mainly used for prsctp RTX policy */
317                 chunk->sent_count++;
318                 break;
319         case SCTP_CID_COOKIE_ECHO:
320                 packet->has_cookie_echo = 1;
321                 break;
322
323         case SCTP_CID_SACK:
324                 packet->has_sack = 1;
325                 if (chunk->asoc)
326                         chunk->asoc->stats.osacks++;
327                 break;
328
329         case SCTP_CID_AUTH:
330                 packet->has_auth = 1;
331                 packet->auth = chunk;
332                 break;
333         }
334
335         /* It is OK to send this chunk.  */
336         list_add_tail(&chunk->list, &packet->chunk_list);
337         packet->size += chunk_len;
338         chunk->transport = packet->transport;
339 finish:
340         return retval;
341 }
342
343 /* Append a chunk to the offered packet reporting back any inability to do
344  * so.
345  */
346 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
347                                      struct sctp_chunk *chunk)
348 {
349         sctp_xmit_t retval = SCTP_XMIT_OK;
350
351         pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
352
353         /* Data chunks are special.  Before seeing what else we can
354          * bundle into this packet, check to see if we are allowed to
355          * send this DATA.
356          */
357         if (sctp_chunk_is_data(chunk)) {
358                 retval = sctp_packet_can_append_data(packet, chunk);
359                 if (retval != SCTP_XMIT_OK)
360                         goto finish;
361         }
362
363         /* Try to bundle AUTH chunk */
364         retval = sctp_packet_bundle_auth(packet, chunk);
365         if (retval != SCTP_XMIT_OK)
366                 goto finish;
367
368         /* Try to bundle SACK chunk */
369         retval = sctp_packet_bundle_sack(packet, chunk);
370         if (retval != SCTP_XMIT_OK)
371                 goto finish;
372
373         retval = __sctp_packet_append_chunk(packet, chunk);
374
375 finish:
376         return retval;
377 }
378
379 static void sctp_packet_release_owner(struct sk_buff *skb)
380 {
381         sk_free(skb->sk);
382 }
383
384 static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
385 {
386         skb_orphan(skb);
387         skb->sk = sk;
388         skb->destructor = sctp_packet_release_owner;
389
390         /*
391          * The data chunks have already been accounted for in sctp_sendmsg(),
392          * therefore only reserve a single byte to keep socket around until
393          * the packet has been transmitted.
394          */
395         atomic_inc(&sk->sk_wmem_alloc);
396 }
397
398 static int sctp_packet_pack(struct sctp_packet *packet,
399                             struct sk_buff *head, int gso, gfp_t gfp)
400 {
401         struct sctp_transport *tp = packet->transport;
402         struct sctp_auth_chunk *auth = NULL;
403         struct sctp_chunk *chunk, *tmp;
404         int pkt_count = 0, pkt_size;
405         struct sock *sk = head->sk;
406         struct sk_buff *nskb;
407         int auth_len = 0;
408
409         if (gso) {
410                 skb_shinfo(head)->gso_type = sk->sk_gso_type;
411                 NAPI_GRO_CB(head)->last = head;
412         } else {
413                 nskb = head;
414                 pkt_size = packet->size;
415                 goto merge;
416         }
417
418         do {
419                 /* calculate the pkt_size and alloc nskb */
420                 pkt_size = packet->overhead;
421                 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list,
422                                          list) {
423                         int padded = SCTP_PAD4(chunk->skb->len);
424
425                         if (chunk == packet->auth)
426                                 auth_len = padded;
427                         else if (auth_len + padded + packet->overhead >
428                                  tp->pathmtu)
429                                 return 0;
430                         else if (pkt_size + padded > tp->pathmtu)
431                                 break;
432                         pkt_size += padded;
433                 }
434                 nskb = alloc_skb(pkt_size + MAX_HEADER, gfp);
435                 if (!nskb)
436                         return 0;
437                 skb_reserve(nskb, packet->overhead + MAX_HEADER);
438
439 merge:
440                 /* merge chunks into nskb and append nskb into head list */
441                 pkt_size -= packet->overhead;
442                 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
443                         int padding;
444
445                         list_del_init(&chunk->list);
446                         if (sctp_chunk_is_data(chunk)) {
447                                 if (!sctp_chunk_retransmitted(chunk) &&
448                                     !tp->rto_pending) {
449                                         chunk->rtt_in_progress = 1;
450                                         tp->rto_pending = 1;
451                                 }
452                         }
453
454                         padding = SCTP_PAD4(chunk->skb->len) - chunk->skb->len;
455                         if (padding)
456                                 memset(skb_put(chunk->skb, padding), 0, padding);
457
458                         if (chunk == packet->auth)
459                                 auth = (struct sctp_auth_chunk *)
460                                                         skb_tail_pointer(nskb);
461
462                         memcpy(skb_put(nskb, chunk->skb->len), chunk->skb->data,
463                                chunk->skb->len);
464
465                         pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, rtt_in_progress:%d\n",
466                                  chunk,
467                                  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
468                                  chunk->has_tsn ? "TSN" : "No TSN",
469                                  chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
470                                  ntohs(chunk->chunk_hdr->length), chunk->skb->len,
471                                  chunk->rtt_in_progress);
472
473                         pkt_size -= SCTP_PAD4(chunk->skb->len);
474
475                         if (!sctp_chunk_is_data(chunk) && chunk != packet->auth)
476                                 sctp_chunk_free(chunk);
477
478                         if (!pkt_size)
479                                 break;
480                 }
481
482                 if (auth) {
483                         sctp_auth_calculate_hmac(tp->asoc, nskb, auth, gfp);
484                         /* free auth if no more chunks, or add it back */
485                         if (list_empty(&packet->chunk_list))
486                                 sctp_chunk_free(packet->auth);
487                         else
488                                 list_add(&packet->auth->list,
489                                          &packet->chunk_list);
490                 }
491
492                 if (gso) {
493                         if (skb_gro_receive(&head, nskb)) {
494                                 kfree_skb(nskb);
495                                 return 0;
496                         }
497                         if (WARN_ON_ONCE(skb_shinfo(head)->gso_segs >=
498                                          sk->sk_gso_max_segs))
499                                 return 0;
500                 }
501
502                 pkt_count++;
503         } while (!list_empty(&packet->chunk_list));
504
505         if (gso) {
506                 memset(head->cb, 0, max(sizeof(struct inet_skb_parm),
507                                         sizeof(struct inet6_skb_parm)));
508                 skb_shinfo(head)->gso_segs = pkt_count;
509                 skb_shinfo(head)->gso_size = GSO_BY_FRAGS;
510                 rcu_read_lock();
511                 if (skb_dst(head) != tp->dst) {
512                         dst_hold(tp->dst);
513                         sk_setup_caps(sk, tp->dst);
514                 }
515                 rcu_read_unlock();
516                 goto chksum;
517         }
518
519         if (sctp_checksum_disable)
520                 return 1;
521
522         if (!(skb_dst(head)->dev->features & NETIF_F_SCTP_CRC) ||
523             dst_xfrm(skb_dst(head)) || packet->ipfragok) {
524                 struct sctphdr *sh =
525                         (struct sctphdr *)skb_transport_header(head);
526
527                 sh->checksum = sctp_compute_cksum(head, 0);
528         } else {
529 chksum:
530                 head->ip_summed = CHECKSUM_PARTIAL;
531                 head->csum_start = skb_transport_header(head) - head->head;
532                 head->csum_offset = offsetof(struct sctphdr, checksum);
533         }
534
535         return pkt_count;
536 }
537
538 /* All packets are sent to the network through this function from
539  * sctp_outq_tail().
540  *
541  * The return value is always 0 for now.
542  */
543 int sctp_packet_transmit(struct sctp_packet *packet, gfp_t gfp)
544 {
545         struct sctp_transport *tp = packet->transport;
546         struct sctp_association *asoc = tp->asoc;
547         struct sctp_chunk *chunk, *tmp;
548         int pkt_count, gso = 0;
549         int confirm;
550         struct dst_entry *dst;
551         struct sk_buff *head;
552         struct sctphdr *sh;
553         struct sock *sk;
554
555         pr_debug("%s: packet:%p\n", __func__, packet);
556         if (list_empty(&packet->chunk_list))
557                 return 0;
558         chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
559         sk = chunk->skb->sk;
560
561         /* check gso */
562         if (packet->size > tp->pathmtu && !packet->ipfragok) {
563                 if (!sk_can_gso(sk)) {
564                         pr_err_once("Trying to GSO but underlying device doesn't support it.");
565                         goto out;
566                 }
567                 gso = 1;
568         }
569
570         /* alloc head skb */
571         head = alloc_skb((gso ? packet->overhead : packet->size) +
572                          MAX_HEADER, gfp);
573         if (!head)
574                 goto out;
575         skb_reserve(head, packet->overhead + MAX_HEADER);
576         sctp_packet_set_owner_w(head, sk);
577
578         /* set sctp header */
579         sh = (struct sctphdr *)skb_push(head, sizeof(struct sctphdr));
580         skb_reset_transport_header(head);
581         sh->source = htons(packet->source_port);
582         sh->dest = htons(packet->destination_port);
583         sh->vtag = htonl(packet->vtag);
584         sh->checksum = 0;
585
586         /* update dst if in need */
587         if (!sctp_transport_dst_check(tp)) {
588                 sctp_transport_route(tp, NULL, sctp_sk(sk));
589                 if (asoc && asoc->param_flags & SPP_PMTUD_ENABLE)
590                         sctp_assoc_sync_pmtu(sk, asoc);
591         }
592         dst = dst_clone(tp->dst);
593         if (!dst) {
594                 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
595                 kfree_skb(head);
596                 goto out;
597         }
598         skb_dst_set(head, dst);
599
600         /* pack up chunks */
601         pkt_count = sctp_packet_pack(packet, head, gso, gfp);
602         if (!pkt_count) {
603                 kfree_skb(head);
604                 goto out;
605         }
606         pr_debug("***sctp_transmit_packet*** skb->len:%d\n", head->len);
607
608         /* start autoclose timer */
609         if (packet->has_data && sctp_state(asoc, ESTABLISHED) &&
610             asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) {
611                 struct timer_list *timer =
612                         &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
613                 unsigned long timeout =
614                         asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
615
616                 if (!mod_timer(timer, jiffies + timeout))
617                         sctp_association_hold(asoc);
618         }
619
620         /* sctp xmit */
621         tp->af_specific->ecn_capable(sk);
622         if (asoc) {
623                 asoc->stats.opackets += pkt_count;
624                 if (asoc->peer.last_sent_to != tp)
625                         asoc->peer.last_sent_to = tp;
626         }
627         head->ignore_df = packet->ipfragok;
628         confirm = tp->dst_pending_confirm;
629         if (confirm)
630                 skb_set_dst_pending_confirm(head, 1);
631         /* neighbour should be confirmed on successful transmission or
632          * positive error
633          */
634         if (tp->af_specific->sctp_xmit(head, tp) >= 0 && confirm)
635                 tp->dst_pending_confirm = 0;
636
637 out:
638         list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
639                 list_del_init(&chunk->list);
640                 if (!sctp_chunk_is_data(chunk))
641                         sctp_chunk_free(chunk);
642         }
643         sctp_packet_reset(packet);
644         return 0;
645 }
646
647 /********************************************************************
648  * 2nd Level Abstractions
649  ********************************************************************/
650
651 /* This private function check to see if a chunk can be added */
652 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
653                                            struct sctp_chunk *chunk)
654 {
655         size_t datasize, rwnd, inflight, flight_size;
656         struct sctp_transport *transport = packet->transport;
657         struct sctp_association *asoc = transport->asoc;
658         struct sctp_outq *q = &asoc->outqueue;
659
660         /* RFC 2960 6.1  Transmission of DATA Chunks
661          *
662          * A) At any given time, the data sender MUST NOT transmit new data to
663          * any destination transport address if its peer's rwnd indicates
664          * that the peer has no buffer space (i.e. rwnd is 0, see Section
665          * 6.2.1).  However, regardless of the value of rwnd (including if it
666          * is 0), the data sender can always have one DATA chunk in flight to
667          * the receiver if allowed by cwnd (see rule B below).  This rule
668          * allows the sender to probe for a change in rwnd that the sender
669          * missed due to the SACK having been lost in transit from the data
670          * receiver to the data sender.
671          */
672
673         rwnd = asoc->peer.rwnd;
674         inflight = q->outstanding_bytes;
675         flight_size = transport->flight_size;
676
677         datasize = sctp_data_size(chunk);
678
679         if (datasize > rwnd && inflight > 0)
680                 /* We have (at least) one data chunk in flight,
681                  * so we can't fall back to rule 6.1 B).
682                  */
683                 return SCTP_XMIT_RWND_FULL;
684
685         /* RFC 2960 6.1  Transmission of DATA Chunks
686          *
687          * B) At any given time, the sender MUST NOT transmit new data
688          * to a given transport address if it has cwnd or more bytes
689          * of data outstanding to that transport address.
690          */
691         /* RFC 7.2.4 & the Implementers Guide 2.8.
692          *
693          * 3) ...
694          *    When a Fast Retransmit is being performed the sender SHOULD
695          *    ignore the value of cwnd and SHOULD NOT delay retransmission.
696          */
697         if (chunk->fast_retransmit != SCTP_NEED_FRTX &&
698             flight_size >= transport->cwnd)
699                 return SCTP_XMIT_RWND_FULL;
700
701         /* Nagle's algorithm to solve small-packet problem:
702          * Inhibit the sending of new chunks when new outgoing data arrives
703          * if any previously transmitted data on the connection remains
704          * unacknowledged.
705          */
706
707         if ((sctp_sk(asoc->base.sk)->nodelay || inflight == 0) &&
708             !chunk->msg->force_delay)
709                 /* Nothing unacked */
710                 return SCTP_XMIT_OK;
711
712         if (!sctp_packet_empty(packet))
713                 /* Append to packet */
714                 return SCTP_XMIT_OK;
715
716         if (!sctp_state(asoc, ESTABLISHED))
717                 return SCTP_XMIT_OK;
718
719         /* Check whether this chunk and all the rest of pending data will fit
720          * or delay in hopes of bundling a full sized packet.
721          */
722         if (chunk->skb->len + q->out_qlen >
723                 transport->pathmtu - packet->overhead - sizeof(sctp_data_chunk_t) - 4)
724                 /* Enough data queued to fill a packet */
725                 return SCTP_XMIT_OK;
726
727         /* Don't delay large message writes that may have been fragmented */
728         if (!chunk->msg->can_delay)
729                 return SCTP_XMIT_OK;
730
731         /* Defer until all data acked or packet full */
732         return SCTP_XMIT_DELAY;
733 }
734
735 /* This private function does management things when adding DATA chunk */
736 static void sctp_packet_append_data(struct sctp_packet *packet,
737                                 struct sctp_chunk *chunk)
738 {
739         struct sctp_transport *transport = packet->transport;
740         size_t datasize = sctp_data_size(chunk);
741         struct sctp_association *asoc = transport->asoc;
742         u32 rwnd = asoc->peer.rwnd;
743
744         /* Keep track of how many bytes are in flight over this transport. */
745         transport->flight_size += datasize;
746
747         /* Keep track of how many bytes are in flight to the receiver. */
748         asoc->outqueue.outstanding_bytes += datasize;
749
750         /* Update our view of the receiver's rwnd. */
751         if (datasize < rwnd)
752                 rwnd -= datasize;
753         else
754                 rwnd = 0;
755
756         asoc->peer.rwnd = rwnd;
757         sctp_chunk_assign_tsn(chunk);
758         sctp_chunk_assign_ssn(chunk);
759 }
760
761 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
762                                         struct sctp_chunk *chunk,
763                                         u16 chunk_len)
764 {
765         size_t psize, pmtu, maxsize;
766         sctp_xmit_t retval = SCTP_XMIT_OK;
767
768         psize = packet->size;
769         if (packet->transport->asoc)
770                 pmtu = packet->transport->asoc->pathmtu;
771         else
772                 pmtu = packet->transport->pathmtu;
773
774         /* Decide if we need to fragment or resubmit later. */
775         if (psize + chunk_len > pmtu) {
776                 /* It's OK to fragment at IP level if any one of the following
777                  * is true:
778                  *      1. The packet is empty (meaning this chunk is greater
779                  *         the MTU)
780                  *      2. The packet doesn't have any data in it yet and data
781                  *         requires authentication.
782                  */
783                 if (sctp_packet_empty(packet) ||
784                     (!packet->has_data && chunk->auth)) {
785                         /* We no longer do re-fragmentation.
786                          * Just fragment at the IP layer, if we
787                          * actually hit this condition
788                          */
789                         packet->ipfragok = 1;
790                         goto out;
791                 }
792
793                 /* Similarly, if this chunk was built before a PMTU
794                  * reduction, we have to fragment it at IP level now. So
795                  * if the packet already contains something, we need to
796                  * flush.
797                  */
798                 maxsize = pmtu - packet->overhead;
799                 if (packet->auth)
800                         maxsize -= SCTP_PAD4(packet->auth->skb->len);
801                 if (chunk_len > maxsize)
802                         retval = SCTP_XMIT_PMTU_FULL;
803
804                 /* It is also okay to fragment if the chunk we are
805                  * adding is a control chunk, but only if current packet
806                  * is not a GSO one otherwise it causes fragmentation of
807                  * a large frame. So in this case we allow the
808                  * fragmentation by forcing it to be in a new packet.
809                  */
810                 if (!sctp_chunk_is_data(chunk) && packet->has_data)
811                         retval = SCTP_XMIT_PMTU_FULL;
812
813                 if (psize + chunk_len > packet->max_size)
814                         /* Hit GSO/PMTU limit, gotta flush */
815                         retval = SCTP_XMIT_PMTU_FULL;
816
817                 if (!packet->transport->burst_limited &&
818                     psize + chunk_len > (packet->transport->cwnd >> 1))
819                         /* Do not allow a single GSO packet to use more
820                          * than half of cwnd.
821                          */
822                         retval = SCTP_XMIT_PMTU_FULL;
823
824                 if (packet->transport->burst_limited &&
825                     psize + chunk_len > (packet->transport->burst_limited >> 1))
826                         /* Do not allow a single GSO packet to use more
827                          * than half of original cwnd.
828                          */
829                         retval = SCTP_XMIT_PMTU_FULL;
830                 /* Otherwise it will fit in the GSO packet */
831         }
832
833 out:
834         return retval;
835 }