1 | // SPDX-License-Identifier: GPL-2.0-or-later |
2 | /* SCTP kernel implementation |
3 | * (C) Copyright IBM Corp. 2001, 2004 |
4 | * Copyright (c) 1999-2000 Cisco, Inc. |
5 | * Copyright (c) 1999-2001 Motorola, Inc. |
6 | * |
7 | * This file is part of the SCTP kernel implementation |
8 | * |
9 | * These functions handle output processing. |
10 | * |
11 | * Please send any bug reports or fixes you make to the |
12 | * email address(es): |
13 | * lksctp developers <linux-sctp@vger.kernel.org> |
14 | * |
15 | * Written or modified by: |
16 | * La Monte H.P. Yarroll <piggy@acm.org> |
17 | * Karl Knutson <karl@athena.chicago.il.us> |
18 | * Jon Grimm <jgrimm@austin.ibm.com> |
19 | * Sridhar Samudrala <sri@us.ibm.com> |
20 | */ |
21 | |
22 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
23 | |
24 | #include <linux/types.h> |
25 | #include <linux/kernel.h> |
26 | #include <linux/wait.h> |
27 | #include <linux/time.h> |
28 | #include <linux/ip.h> |
29 | #include <linux/ipv6.h> |
30 | #include <linux/init.h> |
31 | #include <linux/slab.h> |
32 | #include <net/inet_ecn.h> |
33 | #include <net/ip.h> |
34 | #include <net/icmp.h> |
35 | #include <net/net_namespace.h> |
36 | |
37 | #include <linux/socket.h> /* for sa_family_t */ |
38 | #include <net/sock.h> |
39 | |
40 | #include <net/sctp/sctp.h> |
41 | #include <net/sctp/sm.h> |
42 | #include <net/sctp/checksum.h> |
43 | |
44 | /* Forward declarations for private helpers. */ |
45 | static enum sctp_xmit __sctp_packet_append_chunk(struct sctp_packet *packet, |
46 | struct sctp_chunk *chunk); |
47 | static enum sctp_xmit sctp_packet_can_append_data(struct sctp_packet *packet, |
48 | struct sctp_chunk *chunk); |
49 | static void sctp_packet_append_data(struct sctp_packet *packet, |
50 | struct sctp_chunk *chunk); |
51 | static enum sctp_xmit sctp_packet_will_fit(struct sctp_packet *packet, |
52 | struct sctp_chunk *chunk, |
53 | u16 chunk_len); |
54 | |
55 | static void sctp_packet_reset(struct sctp_packet *packet) |
56 | { |
57 | /* sctp_packet_transmit() relies on this to reset size to the |
58 | * current overhead after sending packets. |
59 | */ |
60 | packet->size = packet->overhead; |
61 | |
62 | packet->has_cookie_echo = 0; |
63 | packet->has_sack = 0; |
64 | packet->has_data = 0; |
65 | packet->has_auth = 0; |
66 | packet->ipfragok = 0; |
67 | packet->auth = NULL; |
68 | } |
69 | |
70 | /* Config a packet. |
71 | * This appears to be a followup set of initializations. |
72 | */ |
73 | void sctp_packet_config(struct sctp_packet *packet, __u32 vtag, |
74 | int ecn_capable) |
75 | { |
76 | struct sctp_transport *tp = packet->transport; |
77 | struct sctp_association *asoc = tp->asoc; |
78 | struct sctp_sock *sp = NULL; |
79 | struct sock *sk; |
80 | |
81 | pr_debug("%s: packet:%p vtag:0x%x\n" , __func__, packet, vtag); |
82 | packet->vtag = vtag; |
83 | |
84 | /* do the following jobs only once for a flush schedule */ |
85 | if (!sctp_packet_empty(packet)) |
86 | return; |
87 | |
88 | /* set packet max_size with pathmtu, then calculate overhead */ |
89 | packet->max_size = tp->pathmtu; |
90 | |
91 | if (asoc) { |
92 | sk = asoc->base.sk; |
93 | sp = sctp_sk(sk); |
94 | } |
95 | packet->overhead = sctp_mtu_payload(sp, mtu: 0, extra: 0); |
96 | packet->size = packet->overhead; |
97 | |
98 | if (!asoc) |
99 | return; |
100 | |
101 | /* update dst or transport pathmtu if in need */ |
102 | if (!sctp_transport_dst_check(t: tp)) { |
103 | sctp_transport_route(tp, NULL, sp); |
104 | if (asoc->param_flags & SPP_PMTUD_ENABLE) |
105 | sctp_assoc_sync_pmtu(asoc); |
106 | } else if (!sctp_transport_pl_enabled(t: tp) && |
107 | asoc->param_flags & SPP_PMTUD_ENABLE) { |
108 | if (!sctp_transport_pmtu_check(t: tp)) |
109 | sctp_assoc_sync_pmtu(asoc); |
110 | } |
111 | |
112 | if (asoc->pmtu_pending) { |
113 | if (asoc->param_flags & SPP_PMTUD_ENABLE) |
114 | sctp_assoc_sync_pmtu(asoc); |
115 | asoc->pmtu_pending = 0; |
116 | } |
117 | |
118 | /* If there a is a prepend chunk stick it on the list before |
119 | * any other chunks get appended. |
120 | */ |
121 | if (ecn_capable) { |
122 | struct sctp_chunk *chunk = sctp_get_ecne_prepend(asoc); |
123 | |
124 | if (chunk) |
125 | sctp_packet_append_chunk(packet, chunk); |
126 | } |
127 | |
128 | if (!tp->dst) |
129 | return; |
130 | |
131 | /* set packet max_size with gso_max_size if gso is enabled*/ |
132 | rcu_read_lock(); |
133 | if (__sk_dst_get(sk) != tp->dst) { |
134 | dst_hold(dst: tp->dst); |
135 | sk_setup_caps(sk, dst: tp->dst); |
136 | } |
137 | packet->max_size = sk_can_gso(sk) ? min(READ_ONCE(tp->dst->dev->gso_max_size), |
138 | GSO_LEGACY_MAX_SIZE) |
139 | : asoc->pathmtu; |
140 | rcu_read_unlock(); |
141 | } |
142 | |
143 | /* Initialize the packet structure. */ |
144 | void sctp_packet_init(struct sctp_packet *packet, |
145 | struct sctp_transport *transport, |
146 | __u16 sport, __u16 dport) |
147 | { |
148 | pr_debug("%s: packet:%p transport:%p\n" , __func__, packet, transport); |
149 | |
150 | packet->transport = transport; |
151 | packet->source_port = sport; |
152 | packet->destination_port = dport; |
153 | INIT_LIST_HEAD(list: &packet->chunk_list); |
154 | /* The overhead will be calculated by sctp_packet_config() */ |
155 | packet->overhead = 0; |
156 | sctp_packet_reset(packet); |
157 | packet->vtag = 0; |
158 | } |
159 | |
160 | /* Free a packet. */ |
161 | void sctp_packet_free(struct sctp_packet *packet) |
162 | { |
163 | struct sctp_chunk *chunk, *tmp; |
164 | |
165 | pr_debug("%s: packet:%p\n" , __func__, packet); |
166 | |
167 | list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) { |
168 | list_del_init(entry: &chunk->list); |
169 | sctp_chunk_free(chunk); |
170 | } |
171 | } |
172 | |
173 | /* This routine tries to append the chunk to the offered packet. If adding |
174 | * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk |
175 | * is not present in the packet, it transmits the input packet. |
176 | * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long |
177 | * as it can fit in the packet, but any more data that does not fit in this |
178 | * packet can be sent only after receiving the COOKIE_ACK. |
179 | */ |
180 | enum sctp_xmit sctp_packet_transmit_chunk(struct sctp_packet *packet, |
181 | struct sctp_chunk *chunk, |
182 | int one_packet, gfp_t gfp) |
183 | { |
184 | enum sctp_xmit retval; |
185 | |
186 | pr_debug("%s: packet:%p size:%zu chunk:%p size:%d\n" , __func__, |
187 | packet, packet->size, chunk, chunk->skb ? chunk->skb->len : -1); |
188 | |
189 | switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) { |
190 | case SCTP_XMIT_PMTU_FULL: |
191 | if (!packet->has_cookie_echo) { |
192 | int error = 0; |
193 | |
194 | error = sctp_packet_transmit(packet, gfp); |
195 | if (error < 0) |
196 | chunk->skb->sk->sk_err = -error; |
197 | |
198 | /* If we have an empty packet, then we can NOT ever |
199 | * return PMTU_FULL. |
200 | */ |
201 | if (!one_packet) |
202 | retval = sctp_packet_append_chunk(packet, |
203 | chunk); |
204 | } |
205 | break; |
206 | |
207 | case SCTP_XMIT_RWND_FULL: |
208 | case SCTP_XMIT_OK: |
209 | case SCTP_XMIT_DELAY: |
210 | break; |
211 | } |
212 | |
213 | return retval; |
214 | } |
215 | |
216 | /* Try to bundle a pad chunk into a packet with a heartbeat chunk for PLPMTUTD probe */ |
217 | static enum sctp_xmit sctp_packet_bundle_pad(struct sctp_packet *pkt, struct sctp_chunk *chunk) |
218 | { |
219 | struct sctp_transport *t = pkt->transport; |
220 | struct sctp_chunk *pad; |
221 | int overhead = 0; |
222 | |
223 | if (!chunk->pmtu_probe) |
224 | return SCTP_XMIT_OK; |
225 | |
226 | /* calculate the Padding Data size for the pad chunk */ |
227 | overhead += sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr); |
228 | overhead += sizeof(struct sctp_sender_hb_info) + sizeof(struct sctp_pad_chunk); |
229 | pad = sctp_make_pad(asoc: t->asoc, len: t->pl.probe_size - overhead); |
230 | if (!pad) |
231 | return SCTP_XMIT_DELAY; |
232 | |
233 | list_add_tail(new: &pad->list, head: &pkt->chunk_list); |
234 | pkt->size += SCTP_PAD4(ntohs(pad->chunk_hdr->length)); |
235 | chunk->transport = t; |
236 | |
237 | return SCTP_XMIT_OK; |
238 | } |
239 | |
240 | /* Try to bundle an auth chunk into the packet. */ |
241 | static enum sctp_xmit sctp_packet_bundle_auth(struct sctp_packet *pkt, |
242 | struct sctp_chunk *chunk) |
243 | { |
244 | struct sctp_association *asoc = pkt->transport->asoc; |
245 | enum sctp_xmit retval = SCTP_XMIT_OK; |
246 | struct sctp_chunk *auth; |
247 | |
248 | /* if we don't have an association, we can't do authentication */ |
249 | if (!asoc) |
250 | return retval; |
251 | |
252 | /* See if this is an auth chunk we are bundling or if |
253 | * auth is already bundled. |
254 | */ |
255 | if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth) |
256 | return retval; |
257 | |
258 | /* if the peer did not request this chunk to be authenticated, |
259 | * don't do it |
260 | */ |
261 | if (!chunk->auth) |
262 | return retval; |
263 | |
264 | auth = sctp_make_auth(asoc, key_id: chunk->shkey->key_id); |
265 | if (!auth) |
266 | return retval; |
267 | |
268 | auth->shkey = chunk->shkey; |
269 | sctp_auth_shkey_hold(sh_key: auth->shkey); |
270 | |
271 | retval = __sctp_packet_append_chunk(packet: pkt, chunk: auth); |
272 | |
273 | if (retval != SCTP_XMIT_OK) |
274 | sctp_chunk_free(auth); |
275 | |
276 | return retval; |
277 | } |
278 | |
279 | /* Try to bundle a SACK with the packet. */ |
280 | static enum sctp_xmit sctp_packet_bundle_sack(struct sctp_packet *pkt, |
281 | struct sctp_chunk *chunk) |
282 | { |
283 | enum sctp_xmit retval = SCTP_XMIT_OK; |
284 | |
285 | /* If sending DATA and haven't aleady bundled a SACK, try to |
286 | * bundle one in to the packet. |
287 | */ |
288 | if (sctp_chunk_is_data(chunk) && !pkt->has_sack && |
289 | !pkt->has_cookie_echo) { |
290 | struct sctp_association *asoc; |
291 | struct timer_list *timer; |
292 | asoc = pkt->transport->asoc; |
293 | timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK]; |
294 | |
295 | /* If the SACK timer is running, we have a pending SACK */ |
296 | if (timer_pending(timer)) { |
297 | struct sctp_chunk *sack; |
298 | |
299 | if (pkt->transport->sack_generation != |
300 | pkt->transport->asoc->peer.sack_generation) |
301 | return retval; |
302 | |
303 | asoc->a_rwnd = asoc->rwnd; |
304 | sack = sctp_make_sack(asoc); |
305 | if (sack) { |
306 | retval = __sctp_packet_append_chunk(packet: pkt, chunk: sack); |
307 | if (retval != SCTP_XMIT_OK) { |
308 | sctp_chunk_free(sack); |
309 | goto out; |
310 | } |
311 | SCTP_INC_STATS(asoc->base.net, |
312 | SCTP_MIB_OUTCTRLCHUNKS); |
313 | asoc->stats.octrlchunks++; |
314 | asoc->peer.sack_needed = 0; |
315 | if (del_timer(timer)) |
316 | sctp_association_put(asoc); |
317 | } |
318 | } |
319 | } |
320 | out: |
321 | return retval; |
322 | } |
323 | |
324 | |
325 | /* Append a chunk to the offered packet reporting back any inability to do |
326 | * so. |
327 | */ |
328 | static enum sctp_xmit __sctp_packet_append_chunk(struct sctp_packet *packet, |
329 | struct sctp_chunk *chunk) |
330 | { |
331 | __u16 chunk_len = SCTP_PAD4(ntohs(chunk->chunk_hdr->length)); |
332 | enum sctp_xmit retval = SCTP_XMIT_OK; |
333 | |
334 | /* Check to see if this chunk will fit into the packet */ |
335 | retval = sctp_packet_will_fit(packet, chunk, chunk_len); |
336 | if (retval != SCTP_XMIT_OK) |
337 | goto finish; |
338 | |
339 | /* We believe that this chunk is OK to add to the packet */ |
340 | switch (chunk->chunk_hdr->type) { |
341 | case SCTP_CID_DATA: |
342 | case SCTP_CID_I_DATA: |
343 | /* Account for the data being in the packet */ |
344 | sctp_packet_append_data(packet, chunk); |
345 | /* Disallow SACK bundling after DATA. */ |
346 | packet->has_sack = 1; |
347 | /* Disallow AUTH bundling after DATA */ |
348 | packet->has_auth = 1; |
349 | /* Let it be knows that packet has DATA in it */ |
350 | packet->has_data = 1; |
351 | /* timestamp the chunk for rtx purposes */ |
352 | chunk->sent_at = jiffies; |
353 | /* Mainly used for prsctp RTX policy */ |
354 | chunk->sent_count++; |
355 | break; |
356 | case SCTP_CID_COOKIE_ECHO: |
357 | packet->has_cookie_echo = 1; |
358 | break; |
359 | |
360 | case SCTP_CID_SACK: |
361 | packet->has_sack = 1; |
362 | if (chunk->asoc) |
363 | chunk->asoc->stats.osacks++; |
364 | break; |
365 | |
366 | case SCTP_CID_AUTH: |
367 | packet->has_auth = 1; |
368 | packet->auth = chunk; |
369 | break; |
370 | } |
371 | |
372 | /* It is OK to send this chunk. */ |
373 | list_add_tail(new: &chunk->list, head: &packet->chunk_list); |
374 | packet->size += chunk_len; |
375 | chunk->transport = packet->transport; |
376 | finish: |
377 | return retval; |
378 | } |
379 | |
380 | /* Append a chunk to the offered packet reporting back any inability to do |
381 | * so. |
382 | */ |
383 | enum sctp_xmit sctp_packet_append_chunk(struct sctp_packet *packet, |
384 | struct sctp_chunk *chunk) |
385 | { |
386 | enum sctp_xmit retval = SCTP_XMIT_OK; |
387 | |
388 | pr_debug("%s: packet:%p chunk:%p\n" , __func__, packet, chunk); |
389 | |
390 | /* Data chunks are special. Before seeing what else we can |
391 | * bundle into this packet, check to see if we are allowed to |
392 | * send this DATA. |
393 | */ |
394 | if (sctp_chunk_is_data(chunk)) { |
395 | retval = sctp_packet_can_append_data(packet, chunk); |
396 | if (retval != SCTP_XMIT_OK) |
397 | goto finish; |
398 | } |
399 | |
400 | /* Try to bundle AUTH chunk */ |
401 | retval = sctp_packet_bundle_auth(pkt: packet, chunk); |
402 | if (retval != SCTP_XMIT_OK) |
403 | goto finish; |
404 | |
405 | /* Try to bundle SACK chunk */ |
406 | retval = sctp_packet_bundle_sack(pkt: packet, chunk); |
407 | if (retval != SCTP_XMIT_OK) |
408 | goto finish; |
409 | |
410 | retval = __sctp_packet_append_chunk(packet, chunk); |
411 | if (retval != SCTP_XMIT_OK) |
412 | goto finish; |
413 | |
414 | retval = sctp_packet_bundle_pad(pkt: packet, chunk); |
415 | |
416 | finish: |
417 | return retval; |
418 | } |
419 | |
420 | static void sctp_packet_gso_append(struct sk_buff *head, struct sk_buff *skb) |
421 | { |
422 | if (SCTP_OUTPUT_CB(head)->last == head) |
423 | skb_shinfo(head)->frag_list = skb; |
424 | else |
425 | SCTP_OUTPUT_CB(head)->last->next = skb; |
426 | SCTP_OUTPUT_CB(head)->last = skb; |
427 | |
428 | head->truesize += skb->truesize; |
429 | head->data_len += skb->len; |
430 | head->len += skb->len; |
431 | refcount_add(i: skb->truesize, r: &head->sk->sk_wmem_alloc); |
432 | |
433 | __skb_header_release(skb); |
434 | } |
435 | |
436 | static int sctp_packet_pack(struct sctp_packet *packet, |
437 | struct sk_buff *head, int gso, gfp_t gfp) |
438 | { |
439 | struct sctp_transport *tp = packet->transport; |
440 | struct sctp_auth_chunk *auth = NULL; |
441 | struct sctp_chunk *chunk, *tmp; |
442 | int pkt_count = 0, pkt_size; |
443 | struct sock *sk = head->sk; |
444 | struct sk_buff *nskb; |
445 | int auth_len = 0; |
446 | |
447 | if (gso) { |
448 | skb_shinfo(head)->gso_type = sk->sk_gso_type; |
449 | SCTP_OUTPUT_CB(head)->last = head; |
450 | } else { |
451 | nskb = head; |
452 | pkt_size = packet->size; |
453 | goto merge; |
454 | } |
455 | |
456 | do { |
457 | /* calculate the pkt_size and alloc nskb */ |
458 | pkt_size = packet->overhead; |
459 | list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, |
460 | list) { |
461 | int padded = SCTP_PAD4(chunk->skb->len); |
462 | |
463 | if (chunk == packet->auth) |
464 | auth_len = padded; |
465 | else if (auth_len + padded + packet->overhead > |
466 | tp->pathmtu) |
467 | return 0; |
468 | else if (pkt_size + padded > tp->pathmtu) |
469 | break; |
470 | pkt_size += padded; |
471 | } |
472 | nskb = alloc_skb(size: pkt_size + MAX_HEADER, priority: gfp); |
473 | if (!nskb) |
474 | return 0; |
475 | skb_reserve(skb: nskb, len: packet->overhead + MAX_HEADER); |
476 | |
477 | merge: |
478 | /* merge chunks into nskb and append nskb into head list */ |
479 | pkt_size -= packet->overhead; |
480 | list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) { |
481 | int padding; |
482 | |
483 | list_del_init(entry: &chunk->list); |
484 | if (sctp_chunk_is_data(chunk)) { |
485 | if (!sctp_chunk_retransmitted(chunk) && |
486 | !tp->rto_pending) { |
487 | chunk->rtt_in_progress = 1; |
488 | tp->rto_pending = 1; |
489 | } |
490 | } |
491 | |
492 | padding = SCTP_PAD4(chunk->skb->len) - chunk->skb->len; |
493 | if (padding) |
494 | skb_put_zero(skb: chunk->skb, len: padding); |
495 | |
496 | if (chunk == packet->auth) |
497 | auth = (struct sctp_auth_chunk *) |
498 | skb_tail_pointer(skb: nskb); |
499 | |
500 | skb_put_data(skb: nskb, data: chunk->skb->data, len: chunk->skb->len); |
501 | |
502 | pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, rtt_in_progress:%d\n" , |
503 | chunk, |
504 | sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)), |
505 | chunk->has_tsn ? "TSN" : "No TSN" , |
506 | chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0, |
507 | ntohs(chunk->chunk_hdr->length), chunk->skb->len, |
508 | chunk->rtt_in_progress); |
509 | |
510 | pkt_size -= SCTP_PAD4(chunk->skb->len); |
511 | |
512 | if (!sctp_chunk_is_data(chunk) && chunk != packet->auth) |
513 | sctp_chunk_free(chunk); |
514 | |
515 | if (!pkt_size) |
516 | break; |
517 | } |
518 | |
519 | if (auth) { |
520 | sctp_auth_calculate_hmac(asoc: tp->asoc, skb: nskb, auth, |
521 | ep_key: packet->auth->shkey, gfp); |
522 | /* free auth if no more chunks, or add it back */ |
523 | if (list_empty(head: &packet->chunk_list)) |
524 | sctp_chunk_free(packet->auth); |
525 | else |
526 | list_add(new: &packet->auth->list, |
527 | head: &packet->chunk_list); |
528 | } |
529 | |
530 | if (gso) |
531 | sctp_packet_gso_append(head, skb: nskb); |
532 | |
533 | pkt_count++; |
534 | } while (!list_empty(head: &packet->chunk_list)); |
535 | |
536 | if (gso) { |
537 | memset(head->cb, 0, max(sizeof(struct inet_skb_parm), |
538 | sizeof(struct inet6_skb_parm))); |
539 | skb_shinfo(head)->gso_segs = pkt_count; |
540 | skb_shinfo(head)->gso_size = GSO_BY_FRAGS; |
541 | goto chksum; |
542 | } |
543 | |
544 | if (sctp_checksum_disable) |
545 | return 1; |
546 | |
547 | if (!(tp->dst->dev->features & NETIF_F_SCTP_CRC) || |
548 | dst_xfrm(dst: tp->dst) || packet->ipfragok || tp->encap_port) { |
549 | struct sctphdr *sh = |
550 | (struct sctphdr *)skb_transport_header(skb: head); |
551 | |
552 | sh->checksum = sctp_compute_cksum(skb: head, offset: 0); |
553 | } else { |
554 | chksum: |
555 | head->ip_summed = CHECKSUM_PARTIAL; |
556 | head->csum_not_inet = 1; |
557 | head->csum_start = skb_transport_header(skb: head) - head->head; |
558 | head->csum_offset = offsetof(struct sctphdr, checksum); |
559 | } |
560 | |
561 | return pkt_count; |
562 | } |
563 | |
564 | /* All packets are sent to the network through this function from |
565 | * sctp_outq_tail(). |
566 | * |
567 | * The return value is always 0 for now. |
568 | */ |
569 | int sctp_packet_transmit(struct sctp_packet *packet, gfp_t gfp) |
570 | { |
571 | struct sctp_transport *tp = packet->transport; |
572 | struct sctp_association *asoc = tp->asoc; |
573 | struct sctp_chunk *chunk, *tmp; |
574 | int pkt_count, gso = 0; |
575 | struct sk_buff *head; |
576 | struct sctphdr *sh; |
577 | struct sock *sk; |
578 | |
579 | pr_debug("%s: packet:%p\n" , __func__, packet); |
580 | if (list_empty(head: &packet->chunk_list)) |
581 | return 0; |
582 | chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list); |
583 | sk = chunk->skb->sk; |
584 | |
585 | if (packet->size > tp->pathmtu && !packet->ipfragok && !chunk->pmtu_probe) { |
586 | if (tp->pl.state == SCTP_PL_ERROR) { /* do IP fragmentation if in Error state */ |
587 | packet->ipfragok = 1; |
588 | } else { |
589 | if (!sk_can_gso(sk)) { /* check gso */ |
590 | pr_err_once("Trying to GSO but underlying device doesn't support it." ); |
591 | goto out; |
592 | } |
593 | gso = 1; |
594 | } |
595 | } |
596 | |
597 | /* alloc head skb */ |
598 | head = alloc_skb(size: (gso ? packet->overhead : packet->size) + |
599 | MAX_HEADER, priority: gfp); |
600 | if (!head) |
601 | goto out; |
602 | skb_reserve(skb: head, len: packet->overhead + MAX_HEADER); |
603 | skb_set_owner_w(skb: head, sk); |
604 | |
605 | /* set sctp header */ |
606 | sh = skb_push(skb: head, len: sizeof(struct sctphdr)); |
607 | skb_reset_transport_header(skb: head); |
608 | sh->source = htons(packet->source_port); |
609 | sh->dest = htons(packet->destination_port); |
610 | sh->vtag = htonl(packet->vtag); |
611 | sh->checksum = 0; |
612 | |
613 | /* drop packet if no dst */ |
614 | if (!tp->dst) { |
615 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
616 | kfree_skb(skb: head); |
617 | goto out; |
618 | } |
619 | |
620 | /* pack up chunks */ |
621 | pkt_count = sctp_packet_pack(packet, head, gso, gfp); |
622 | if (!pkt_count) { |
623 | kfree_skb(skb: head); |
624 | goto out; |
625 | } |
626 | pr_debug("***sctp_transmit_packet*** skb->len:%d\n" , head->len); |
627 | |
628 | /* start autoclose timer */ |
629 | if (packet->has_data && sctp_state(asoc, ESTABLISHED) && |
630 | asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) { |
631 | struct timer_list *timer = |
632 | &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE]; |
633 | unsigned long timeout = |
634 | asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]; |
635 | |
636 | if (!mod_timer(timer, expires: jiffies + timeout)) |
637 | sctp_association_hold(asoc); |
638 | } |
639 | |
640 | /* sctp xmit */ |
641 | tp->af_specific->ecn_capable(sk); |
642 | if (asoc) { |
643 | asoc->stats.opackets += pkt_count; |
644 | if (asoc->peer.last_sent_to != tp) |
645 | asoc->peer.last_sent_to = tp; |
646 | } |
647 | head->ignore_df = packet->ipfragok; |
648 | if (tp->dst_pending_confirm) |
649 | skb_set_dst_pending_confirm(skb: head, val: 1); |
650 | /* neighbour should be confirmed on successful transmission or |
651 | * positive error |
652 | */ |
653 | if (tp->af_specific->sctp_xmit(head, tp) >= 0 && |
654 | tp->dst_pending_confirm) |
655 | tp->dst_pending_confirm = 0; |
656 | |
657 | out: |
658 | list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) { |
659 | list_del_init(entry: &chunk->list); |
660 | if (!sctp_chunk_is_data(chunk)) |
661 | sctp_chunk_free(chunk); |
662 | } |
663 | sctp_packet_reset(packet); |
664 | return 0; |
665 | } |
666 | |
667 | /******************************************************************** |
668 | * 2nd Level Abstractions |
669 | ********************************************************************/ |
670 | |
671 | /* This private function check to see if a chunk can be added */ |
672 | static enum sctp_xmit sctp_packet_can_append_data(struct sctp_packet *packet, |
673 | struct sctp_chunk *chunk) |
674 | { |
675 | size_t datasize, rwnd, inflight, flight_size; |
676 | struct sctp_transport *transport = packet->transport; |
677 | struct sctp_association *asoc = transport->asoc; |
678 | struct sctp_outq *q = &asoc->outqueue; |
679 | |
680 | /* RFC 2960 6.1 Transmission of DATA Chunks |
681 | * |
682 | * A) At any given time, the data sender MUST NOT transmit new data to |
683 | * any destination transport address if its peer's rwnd indicates |
684 | * that the peer has no buffer space (i.e. rwnd is 0, see Section |
685 | * 6.2.1). However, regardless of the value of rwnd (including if it |
686 | * is 0), the data sender can always have one DATA chunk in flight to |
687 | * the receiver if allowed by cwnd (see rule B below). This rule |
688 | * allows the sender to probe for a change in rwnd that the sender |
689 | * missed due to the SACK having been lost in transit from the data |
690 | * receiver to the data sender. |
691 | */ |
692 | |
693 | rwnd = asoc->peer.rwnd; |
694 | inflight = q->outstanding_bytes; |
695 | flight_size = transport->flight_size; |
696 | |
697 | datasize = sctp_data_size(chunk); |
698 | |
699 | if (datasize > rwnd && inflight > 0) |
700 | /* We have (at least) one data chunk in flight, |
701 | * so we can't fall back to rule 6.1 B). |
702 | */ |
703 | return SCTP_XMIT_RWND_FULL; |
704 | |
705 | /* RFC 2960 6.1 Transmission of DATA Chunks |
706 | * |
707 | * B) At any given time, the sender MUST NOT transmit new data |
708 | * to a given transport address if it has cwnd or more bytes |
709 | * of data outstanding to that transport address. |
710 | */ |
711 | /* RFC 7.2.4 & the Implementers Guide 2.8. |
712 | * |
713 | * 3) ... |
714 | * When a Fast Retransmit is being performed the sender SHOULD |
715 | * ignore the value of cwnd and SHOULD NOT delay retransmission. |
716 | */ |
717 | if (chunk->fast_retransmit != SCTP_NEED_FRTX && |
718 | flight_size >= transport->cwnd) |
719 | return SCTP_XMIT_RWND_FULL; |
720 | |
721 | /* Nagle's algorithm to solve small-packet problem: |
722 | * Inhibit the sending of new chunks when new outgoing data arrives |
723 | * if any previously transmitted data on the connection remains |
724 | * unacknowledged. |
725 | */ |
726 | |
727 | if ((sctp_sk(sk: asoc->base.sk)->nodelay || inflight == 0) && |
728 | !asoc->force_delay) |
729 | /* Nothing unacked */ |
730 | return SCTP_XMIT_OK; |
731 | |
732 | if (!sctp_packet_empty(packet)) |
733 | /* Append to packet */ |
734 | return SCTP_XMIT_OK; |
735 | |
736 | if (!sctp_state(asoc, ESTABLISHED)) |
737 | return SCTP_XMIT_OK; |
738 | |
739 | /* Check whether this chunk and all the rest of pending data will fit |
740 | * or delay in hopes of bundling a full sized packet. |
741 | */ |
742 | if (chunk->skb->len + q->out_qlen > transport->pathmtu - |
743 | packet->overhead - sctp_datachk_len(stream: &chunk->asoc->stream) - 4) |
744 | /* Enough data queued to fill a packet */ |
745 | return SCTP_XMIT_OK; |
746 | |
747 | /* Don't delay large message writes that may have been fragmented */ |
748 | if (!chunk->msg->can_delay) |
749 | return SCTP_XMIT_OK; |
750 | |
751 | /* Defer until all data acked or packet full */ |
752 | return SCTP_XMIT_DELAY; |
753 | } |
754 | |
755 | /* This private function does management things when adding DATA chunk */ |
756 | static void sctp_packet_append_data(struct sctp_packet *packet, |
757 | struct sctp_chunk *chunk) |
758 | { |
759 | struct sctp_transport *transport = packet->transport; |
760 | size_t datasize = sctp_data_size(chunk); |
761 | struct sctp_association *asoc = transport->asoc; |
762 | u32 rwnd = asoc->peer.rwnd; |
763 | |
764 | /* Keep track of how many bytes are in flight over this transport. */ |
765 | transport->flight_size += datasize; |
766 | |
767 | /* Keep track of how many bytes are in flight to the receiver. */ |
768 | asoc->outqueue.outstanding_bytes += datasize; |
769 | |
770 | /* Update our view of the receiver's rwnd. */ |
771 | if (datasize < rwnd) |
772 | rwnd -= datasize; |
773 | else |
774 | rwnd = 0; |
775 | |
776 | asoc->peer.rwnd = rwnd; |
777 | sctp_chunk_assign_tsn(chunk); |
778 | asoc->stream.si->assign_number(chunk); |
779 | } |
780 | |
781 | static enum sctp_xmit sctp_packet_will_fit(struct sctp_packet *packet, |
782 | struct sctp_chunk *chunk, |
783 | u16 chunk_len) |
784 | { |
785 | enum sctp_xmit retval = SCTP_XMIT_OK; |
786 | size_t psize, pmtu, maxsize; |
787 | |
788 | /* Don't bundle in this packet if this chunk's auth key doesn't |
789 | * match other chunks already enqueued on this packet. Also, |
790 | * don't bundle the chunk with auth key if other chunks in this |
791 | * packet don't have auth key. |
792 | */ |
793 | if ((packet->auth && chunk->shkey != packet->auth->shkey) || |
794 | (!packet->auth && chunk->shkey && |
795 | chunk->chunk_hdr->type != SCTP_CID_AUTH)) |
796 | return SCTP_XMIT_PMTU_FULL; |
797 | |
798 | psize = packet->size; |
799 | if (packet->transport->asoc) |
800 | pmtu = packet->transport->asoc->pathmtu; |
801 | else |
802 | pmtu = packet->transport->pathmtu; |
803 | |
804 | /* Decide if we need to fragment or resubmit later. */ |
805 | if (psize + chunk_len > pmtu) { |
806 | /* It's OK to fragment at IP level if any one of the following |
807 | * is true: |
808 | * 1. The packet is empty (meaning this chunk is greater |
809 | * the MTU) |
810 | * 2. The packet doesn't have any data in it yet and data |
811 | * requires authentication. |
812 | */ |
813 | if (sctp_packet_empty(packet) || |
814 | (!packet->has_data && chunk->auth)) { |
815 | /* We no longer do re-fragmentation. |
816 | * Just fragment at the IP layer, if we |
817 | * actually hit this condition |
818 | */ |
819 | packet->ipfragok = 1; |
820 | goto out; |
821 | } |
822 | |
823 | /* Similarly, if this chunk was built before a PMTU |
824 | * reduction, we have to fragment it at IP level now. So |
825 | * if the packet already contains something, we need to |
826 | * flush. |
827 | */ |
828 | maxsize = pmtu - packet->overhead; |
829 | if (packet->auth) |
830 | maxsize -= SCTP_PAD4(packet->auth->skb->len); |
831 | if (chunk_len > maxsize) |
832 | retval = SCTP_XMIT_PMTU_FULL; |
833 | |
834 | /* It is also okay to fragment if the chunk we are |
835 | * adding is a control chunk, but only if current packet |
836 | * is not a GSO one otherwise it causes fragmentation of |
837 | * a large frame. So in this case we allow the |
838 | * fragmentation by forcing it to be in a new packet. |
839 | */ |
840 | if (!sctp_chunk_is_data(chunk) && packet->has_data) |
841 | retval = SCTP_XMIT_PMTU_FULL; |
842 | |
843 | if (psize + chunk_len > packet->max_size) |
844 | /* Hit GSO/PMTU limit, gotta flush */ |
845 | retval = SCTP_XMIT_PMTU_FULL; |
846 | |
847 | if (!packet->transport->burst_limited && |
848 | psize + chunk_len > (packet->transport->cwnd >> 1)) |
849 | /* Do not allow a single GSO packet to use more |
850 | * than half of cwnd. |
851 | */ |
852 | retval = SCTP_XMIT_PMTU_FULL; |
853 | |
854 | if (packet->transport->burst_limited && |
855 | psize + chunk_len > (packet->transport->burst_limited >> 1)) |
856 | /* Do not allow a single GSO packet to use more |
857 | * than half of original cwnd. |
858 | */ |
859 | retval = SCTP_XMIT_PMTU_FULL; |
860 | /* Otherwise it will fit in the GSO packet */ |
861 | } |
862 | |
863 | out: |
864 | return retval; |
865 | } |
866 | |