Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* SCTP kernel implementation
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
5 * Copyright (c) 2001-2003 International Business Machines, Corp.
6 * Copyright (c) 2001 Intel Corp.
7 * Copyright (c) 2001 Nokia, Inc.
8 * Copyright (c) 2001 La Monte H.P. Yarroll
9 *
10 * This file is part of the SCTP kernel implementation
11 *
12 * These functions handle all input from the IP layer into SCTP.
13 *
14 * Please send any bug reports or fixes you make to the
15 * email address(es):
16 * lksctp developers <linux-sctp@vger.kernel.org>
17 *
18 * Written or modified by:
19 * La Monte H.P. Yarroll <piggy@acm.org>
20 * Karl Knutson <karl@athena.chicago.il.us>
21 * Xingang Guo <xingang.guo@intel.com>
22 * Jon Grimm <jgrimm@us.ibm.com>
23 * Hui Huang <hui.huang@nokia.com>
24 * Daisy Chang <daisyc@us.ibm.com>
25 * Sridhar Samudrala <sri@us.ibm.com>
26 * Ardelle Fan <ardelle.fan@intel.com>
27 */
28
29#include <linux/types.h>
30#include <linux/list.h> /* For struct list_head */
31#include <linux/socket.h>
32#include <linux/ip.h>
33#include <linux/time.h> /* For struct timeval */
34#include <linux/slab.h>
35#include <net/ip.h>
36#include <net/icmp.h>
37#include <net/snmp.h>
38#include <net/sock.h>
39#include <net/xfrm.h>
40#include <net/sctp/sctp.h>
41#include <net/sctp/sm.h>
42#include <net/sctp/checksum.h>
43#include <net/net_namespace.h>
44#include <linux/rhashtable.h>
45#include <net/sock_reuseport.h>
46
47/* Forward declarations for internal helpers. */
48static int sctp_rcv_ootb(struct sk_buff *);
49static struct sctp_association *__sctp_rcv_lookup(struct net *net,
50 struct sk_buff *skb,
51 const union sctp_addr *paddr,
52 const union sctp_addr *laddr,
53 struct sctp_transport **transportp);
54static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(
55 struct net *net, struct sk_buff *skb,
56 const union sctp_addr *laddr,
57 const union sctp_addr *daddr);
58static struct sctp_association *__sctp_lookup_association(
59 struct net *net,
60 const union sctp_addr *local,
61 const union sctp_addr *peer,
62 struct sctp_transport **pt);
63
64static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb);
65
66
67/* Calculate the SCTP checksum of an SCTP packet. */
68static inline int sctp_rcv_checksum(struct net *net, struct sk_buff *skb)
69{
70 struct sctphdr *sh = sctp_hdr(skb);
71 __le32 cmp = sh->checksum;
72 __le32 val = sctp_compute_cksum(skb, 0);
73
74 if (val != cmp) {
75 /* CRC failure, dump it. */
76 __SCTP_INC_STATS(net, SCTP_MIB_CHECKSUMERRORS);
77 return -1;
78 }
79 return 0;
80}
81
82/*
83 * This is the routine which IP calls when receiving an SCTP packet.
84 */
85int sctp_rcv(struct sk_buff *skb)
86{
87 struct sock *sk;
88 struct sctp_association *asoc;
89 struct sctp_endpoint *ep = NULL;
90 struct sctp_ep_common *rcvr;
91 struct sctp_transport *transport = NULL;
92 struct sctp_chunk *chunk;
93 union sctp_addr src;
94 union sctp_addr dest;
95 int family;
96 struct sctp_af *af;
97 struct net *net = dev_net(skb->dev);
98 bool is_gso = skb_is_gso(skb) && skb_is_gso_sctp(skb);
99
100 if (skb->pkt_type != PACKET_HOST)
101 goto discard_it;
102
103 __SCTP_INC_STATS(net, SCTP_MIB_INSCTPPACKS);
104
105 /* If packet is too small to contain a single chunk, let's not
106 * waste time on it anymore.
107 */
108 if (skb->len < sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) +
109 skb_transport_offset(skb))
110 goto discard_it;
111
112 /* If the packet is fragmented and we need to do crc checking,
113 * it's better to just linearize it otherwise crc computing
114 * takes longer.
115 */
116 if ((!is_gso && skb_linearize(skb)) ||
117 !pskb_may_pull(skb, sizeof(struct sctphdr)))
118 goto discard_it;
119
120 /* Pull up the IP header. */
121 __skb_pull(skb, skb_transport_offset(skb));
122
123 skb->csum_valid = 0; /* Previous value not applicable */
124 if (skb_csum_unnecessary(skb))
125 __skb_decr_checksum_unnecessary(skb);
126 else if (!sctp_checksum_disable &&
127 !is_gso &&
128 sctp_rcv_checksum(net, skb) < 0)
129 goto discard_it;
130 skb->csum_valid = 1;
131
132 __skb_pull(skb, sizeof(struct sctphdr));
133
134 family = ipver2af(ip_hdr(skb)->version);
135 af = sctp_get_af_specific(family);
136 if (unlikely(!af))
137 goto discard_it;
138 SCTP_INPUT_CB(skb)->af = af;
139
140 /* Initialize local addresses for lookups. */
141 af->from_skb(&src, skb, 1);
142 af->from_skb(&dest, skb, 0);
143
144 /* If the packet is to or from a non-unicast address,
145 * silently discard the packet.
146 *
147 * This is not clearly defined in the RFC except in section
148 * 8.4 - OOTB handling. However, based on the book "Stream Control
149 * Transmission Protocol" 2.1, "It is important to note that the
150 * IP address of an SCTP transport address must be a routable
151 * unicast address. In other words, IP multicast addresses and
152 * IP broadcast addresses cannot be used in an SCTP transport
153 * address."
154 */
155 if (!af->addr_valid(&src, NULL, skb) ||
156 !af->addr_valid(&dest, NULL, skb))
157 goto discard_it;
158
159 asoc = __sctp_rcv_lookup(net, skb, &src, &dest, &transport);
160
161 if (!asoc)
162 ep = __sctp_rcv_lookup_endpoint(net, skb, &dest, &src);
163
164 /* Retrieve the common input handling substructure. */
165 rcvr = asoc ? &asoc->base : &ep->base;
166 sk = rcvr->sk;
167
168 /*
169 * If a frame arrives on an interface and the receiving socket is
170 * bound to another interface, via SO_BINDTODEVICE, treat it as OOTB
171 */
172 if (sk->sk_bound_dev_if && (sk->sk_bound_dev_if != af->skb_iif(skb))) {
173 if (transport) {
174 sctp_transport_put(transport);
175 asoc = NULL;
176 transport = NULL;
177 } else {
178 sctp_endpoint_put(ep);
179 ep = NULL;
180 }
181 sk = net->sctp.ctl_sock;
182 ep = sctp_sk(sk)->ep;
183 sctp_endpoint_hold(ep);
184 rcvr = &ep->base;
185 }
186
187 /*
188 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
189 * An SCTP packet is called an "out of the blue" (OOTB)
190 * packet if it is correctly formed, i.e., passed the
191 * receiver's checksum check, but the receiver is not
192 * able to identify the association to which this
193 * packet belongs.
194 */
195 if (!asoc) {
196 if (sctp_rcv_ootb(skb)) {
197 __SCTP_INC_STATS(net, SCTP_MIB_OUTOFBLUES);
198 goto discard_release;
199 }
200 }
201
202 if (!xfrm_policy_check(sk, XFRM_POLICY_IN, skb, family))
203 goto discard_release;
204 nf_reset_ct(skb);
205
206 if (sk_filter(sk, skb))
207 goto discard_release;
208
209 /* Create an SCTP packet structure. */
210 chunk = sctp_chunkify(skb, asoc, sk, GFP_ATOMIC);
211 if (!chunk)
212 goto discard_release;
213 SCTP_INPUT_CB(skb)->chunk = chunk;
214
215 /* Remember what endpoint is to handle this packet. */
216 chunk->rcvr = rcvr;
217
218 /* Remember the SCTP header. */
219 chunk->sctp_hdr = sctp_hdr(skb);
220
221 /* Set the source and destination addresses of the incoming chunk. */
222 sctp_init_addrs(chunk, &src, &dest);
223
224 /* Remember where we came from. */
225 chunk->transport = transport;
226
227 /* Acquire access to the sock lock. Note: We are safe from other
228 * bottom halves on this lock, but a user may be in the lock too,
229 * so check if it is busy.
230 */
231 bh_lock_sock(sk);
232
233 if (sk != rcvr->sk) {
234 /* Our cached sk is different from the rcvr->sk. This is
235 * because migrate()/accept() may have moved the association
236 * to a new socket and released all the sockets. So now we
237 * are holding a lock on the old socket while the user may
238 * be doing something with the new socket. Switch our veiw
239 * of the current sk.
240 */
241 bh_unlock_sock(sk);
242 sk = rcvr->sk;
243 bh_lock_sock(sk);
244 }
245
246 if (sock_owned_by_user(sk) || !sctp_newsk_ready(sk)) {
247 if (sctp_add_backlog(sk, skb)) {
248 bh_unlock_sock(sk);
249 sctp_chunk_free(chunk);
250 skb = NULL; /* sctp_chunk_free already freed the skb */
251 goto discard_release;
252 }
253 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_BACKLOG);
254 } else {
255 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_SOFTIRQ);
256 sctp_inq_push(&chunk->rcvr->inqueue, chunk);
257 }
258
259 bh_unlock_sock(sk);
260
261 /* Release the asoc/ep ref we took in the lookup calls. */
262 if (transport)
263 sctp_transport_put(transport);
264 else
265 sctp_endpoint_put(ep);
266
267 return 0;
268
269discard_it:
270 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_DISCARDS);
271 kfree_skb(skb);
272 return 0;
273
274discard_release:
275 /* Release the asoc/ep ref we took in the lookup calls. */
276 if (transport)
277 sctp_transport_put(transport);
278 else
279 sctp_endpoint_put(ep);
280
281 goto discard_it;
282}
283
284/* Process the backlog queue of the socket. Every skb on
285 * the backlog holds a ref on an association or endpoint.
286 * We hold this ref throughout the state machine to make
287 * sure that the structure we need is still around.
288 */
289int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb)
290{
291 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
292 struct sctp_inq *inqueue = &chunk->rcvr->inqueue;
293 struct sctp_transport *t = chunk->transport;
294 struct sctp_ep_common *rcvr = NULL;
295 int backloged = 0;
296
297 rcvr = chunk->rcvr;
298
299 /* If the rcvr is dead then the association or endpoint
300 * has been deleted and we can safely drop the chunk
301 * and refs that we are holding.
302 */
303 if (rcvr->dead) {
304 sctp_chunk_free(chunk);
305 goto done;
306 }
307
308 if (unlikely(rcvr->sk != sk)) {
309 /* In this case, the association moved from one socket to
310 * another. We are currently sitting on the backlog of the
311 * old socket, so we need to move.
312 * However, since we are here in the process context we
313 * need to take make sure that the user doesn't own
314 * the new socket when we process the packet.
315 * If the new socket is user-owned, queue the chunk to the
316 * backlog of the new socket without dropping any refs.
317 * Otherwise, we can safely push the chunk on the inqueue.
318 */
319
320 sk = rcvr->sk;
321 local_bh_disable();
322 bh_lock_sock(sk);
323
324 if (sock_owned_by_user(sk) || !sctp_newsk_ready(sk)) {
325 if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))
326 sctp_chunk_free(chunk);
327 else
328 backloged = 1;
329 } else
330 sctp_inq_push(inqueue, chunk);
331
332 bh_unlock_sock(sk);
333 local_bh_enable();
334
335 /* If the chunk was backloged again, don't drop refs */
336 if (backloged)
337 return 0;
338 } else {
339 if (!sctp_newsk_ready(sk)) {
340 if (!sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))
341 return 0;
342 sctp_chunk_free(chunk);
343 } else {
344 sctp_inq_push(inqueue, chunk);
345 }
346 }
347
348done:
349 /* Release the refs we took in sctp_add_backlog */
350 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
351 sctp_transport_put(t);
352 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
353 sctp_endpoint_put(sctp_ep(rcvr));
354 else
355 BUG();
356
357 return 0;
358}
359
360static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb)
361{
362 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
363 struct sctp_transport *t = chunk->transport;
364 struct sctp_ep_common *rcvr = chunk->rcvr;
365 int ret;
366
367 ret = sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf));
368 if (!ret) {
369 /* Hold the assoc/ep while hanging on the backlog queue.
370 * This way, we know structures we need will not disappear
371 * from us
372 */
373 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
374 sctp_transport_hold(t);
375 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
376 sctp_endpoint_hold(sctp_ep(rcvr));
377 else
378 BUG();
379 }
380 return ret;
381
382}
383
384/* Handle icmp frag needed error. */
385void sctp_icmp_frag_needed(struct sock *sk, struct sctp_association *asoc,
386 struct sctp_transport *t, __u32 pmtu)
387{
388 if (!t || (t->pathmtu <= pmtu))
389 return;
390
391 if (sock_owned_by_user(sk)) {
392 atomic_set(&t->mtu_info, pmtu);
393 asoc->pmtu_pending = 1;
394 t->pmtu_pending = 1;
395 return;
396 }
397
398 if (!(t->param_flags & SPP_PMTUD_ENABLE))
399 /* We can't allow retransmitting in such case, as the
400 * retransmission would be sized just as before, and thus we
401 * would get another icmp, and retransmit again.
402 */
403 return;
404
405 /* Update transports view of the MTU. Return if no update was needed.
406 * If an update wasn't needed/possible, it also doesn't make sense to
407 * try to retransmit now.
408 */
409 if (!sctp_transport_update_pmtu(t, pmtu))
410 return;
411
412 /* Update association pmtu. */
413 sctp_assoc_sync_pmtu(asoc);
414
415 /* Retransmit with the new pmtu setting. */
416 sctp_retransmit(&asoc->outqueue, t, SCTP_RTXR_PMTUD);
417}
418
419void sctp_icmp_redirect(struct sock *sk, struct sctp_transport *t,
420 struct sk_buff *skb)
421{
422 struct dst_entry *dst;
423
424 if (sock_owned_by_user(sk) || !t)
425 return;
426 dst = sctp_transport_dst_check(t);
427 if (dst)
428 dst->ops->redirect(dst, sk, skb);
429}
430
431/*
432 * SCTP Implementer's Guide, 2.37 ICMP handling procedures
433 *
434 * ICMP8) If the ICMP code is a "Unrecognized next header type encountered"
435 * or a "Protocol Unreachable" treat this message as an abort
436 * with the T bit set.
437 *
438 * This function sends an event to the state machine, which will abort the
439 * association.
440 *
441 */
442void sctp_icmp_proto_unreachable(struct sock *sk,
443 struct sctp_association *asoc,
444 struct sctp_transport *t)
445{
446 if (sock_owned_by_user(sk)) {
447 if (timer_pending(&t->proto_unreach_timer))
448 return;
449 else {
450 if (!mod_timer(&t->proto_unreach_timer,
451 jiffies + (HZ/20)))
452 sctp_association_hold(asoc);
453 }
454 } else {
455 struct net *net = sock_net(sk);
456
457 pr_debug("%s: unrecognized next header type "
458 "encountered!\n", __func__);
459
460 if (del_timer(&t->proto_unreach_timer))
461 sctp_association_put(asoc);
462
463 sctp_do_sm(net, SCTP_EVENT_T_OTHER,
464 SCTP_ST_OTHER(SCTP_EVENT_ICMP_PROTO_UNREACH),
465 asoc->state, asoc->ep, asoc, t,
466 GFP_ATOMIC);
467 }
468}
469
470/* Common lookup code for icmp/icmpv6 error handler. */
471struct sock *sctp_err_lookup(struct net *net, int family, struct sk_buff *skb,
472 struct sctphdr *sctphdr,
473 struct sctp_association **app,
474 struct sctp_transport **tpp)
475{
476 struct sctp_init_chunk *chunkhdr, _chunkhdr;
477 union sctp_addr saddr;
478 union sctp_addr daddr;
479 struct sctp_af *af;
480 struct sock *sk = NULL;
481 struct sctp_association *asoc;
482 struct sctp_transport *transport = NULL;
483 __u32 vtag = ntohl(sctphdr->vtag);
484
485 *app = NULL; *tpp = NULL;
486
487 af = sctp_get_af_specific(family);
488 if (unlikely(!af)) {
489 return NULL;
490 }
491
492 /* Initialize local addresses for lookups. */
493 af->from_skb(&saddr, skb, 1);
494 af->from_skb(&daddr, skb, 0);
495
496 /* Look for an association that matches the incoming ICMP error
497 * packet.
498 */
499 asoc = __sctp_lookup_association(net, &saddr, &daddr, &transport);
500 if (!asoc)
501 return NULL;
502
503 sk = asoc->base.sk;
504
505 /* RFC 4960, Appendix C. ICMP Handling
506 *
507 * ICMP6) An implementation MUST validate that the Verification Tag
508 * contained in the ICMP message matches the Verification Tag of
509 * the peer. If the Verification Tag is not 0 and does NOT
510 * match, discard the ICMP message. If it is 0 and the ICMP
511 * message contains enough bytes to verify that the chunk type is
512 * an INIT chunk and that the Initiate Tag matches the tag of the
513 * peer, continue with ICMP7. If the ICMP message is too short
514 * or the chunk type or the Initiate Tag does not match, silently
515 * discard the packet.
516 */
517 if (vtag == 0) {
518 /* chunk header + first 4 octects of init header */
519 chunkhdr = skb_header_pointer(skb, skb_transport_offset(skb) +
520 sizeof(struct sctphdr),
521 sizeof(struct sctp_chunkhdr) +
522 sizeof(__be32), &_chunkhdr);
523 if (!chunkhdr ||
524 chunkhdr->chunk_hdr.type != SCTP_CID_INIT ||
525 ntohl(chunkhdr->init_hdr.init_tag) != asoc->c.my_vtag)
526 goto out;
527
528 } else if (vtag != asoc->c.peer_vtag) {
529 goto out;
530 }
531
532 bh_lock_sock(sk);
533
534 /* If too many ICMPs get dropped on busy
535 * servers this needs to be solved differently.
536 */
537 if (sock_owned_by_user(sk))
538 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
539
540 *app = asoc;
541 *tpp = transport;
542 return sk;
543
544out:
545 sctp_transport_put(transport);
546 return NULL;
547}
548
549/* Common cleanup code for icmp/icmpv6 error handler. */
550void sctp_err_finish(struct sock *sk, struct sctp_transport *t)
551 __releases(&((__sk)->sk_lock.slock))
552{
553 bh_unlock_sock(sk);
554 sctp_transport_put(t);
555}
556
557/*
558 * This routine is called by the ICMP module when it gets some
559 * sort of error condition. If err < 0 then the socket should
560 * be closed and the error returned to the user. If err > 0
561 * it's just the icmp type << 8 | icmp code. After adjustment
562 * header points to the first 8 bytes of the sctp header. We need
563 * to find the appropriate port.
564 *
565 * The locking strategy used here is very "optimistic". When
566 * someone else accesses the socket the ICMP is just dropped
567 * and for some paths there is no check at all.
568 * A more general error queue to queue errors for later handling
569 * is probably better.
570 *
571 */
572int sctp_v4_err(struct sk_buff *skb, __u32 info)
573{
574 const struct iphdr *iph = (const struct iphdr *)skb->data;
575 const int ihlen = iph->ihl * 4;
576 const int type = icmp_hdr(skb)->type;
577 const int code = icmp_hdr(skb)->code;
578 struct sock *sk;
579 struct sctp_association *asoc = NULL;
580 struct sctp_transport *transport;
581 struct inet_sock *inet;
582 __u16 saveip, savesctp;
583 int err;
584 struct net *net = dev_net(skb->dev);
585
586 /* Fix up skb to look at the embedded net header. */
587 saveip = skb->network_header;
588 savesctp = skb->transport_header;
589 skb_reset_network_header(skb);
590 skb_set_transport_header(skb, ihlen);
591 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &transport);
592 /* Put back, the original values. */
593 skb->network_header = saveip;
594 skb->transport_header = savesctp;
595 if (!sk) {
596 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
597 return -ENOENT;
598 }
599 /* Warning: The sock lock is held. Remember to call
600 * sctp_err_finish!
601 */
602
603 switch (type) {
604 case ICMP_PARAMETERPROB:
605 err = EPROTO;
606 break;
607 case ICMP_DEST_UNREACH:
608 if (code > NR_ICMP_UNREACH)
609 goto out_unlock;
610
611 /* PMTU discovery (RFC1191) */
612 if (ICMP_FRAG_NEEDED == code) {
613 sctp_icmp_frag_needed(sk, asoc, transport,
614 SCTP_TRUNC4(info));
615 goto out_unlock;
616 } else {
617 if (ICMP_PROT_UNREACH == code) {
618 sctp_icmp_proto_unreachable(sk, asoc,
619 transport);
620 goto out_unlock;
621 }
622 }
623 err = icmp_err_convert[code].errno;
624 break;
625 case ICMP_TIME_EXCEEDED:
626 /* Ignore any time exceeded errors due to fragment reassembly
627 * timeouts.
628 */
629 if (ICMP_EXC_FRAGTIME == code)
630 goto out_unlock;
631
632 err = EHOSTUNREACH;
633 break;
634 case ICMP_REDIRECT:
635 sctp_icmp_redirect(sk, transport, skb);
636 /* Fall through to out_unlock. */
637 default:
638 goto out_unlock;
639 }
640
641 inet = inet_sk(sk);
642 if (!sock_owned_by_user(sk) && inet->recverr) {
643 sk->sk_err = err;
644 sk->sk_error_report(sk);
645 } else { /* Only an error on timeout */
646 sk->sk_err_soft = err;
647 }
648
649out_unlock:
650 sctp_err_finish(sk, transport);
651 return 0;
652}
653
654/*
655 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
656 *
657 * This function scans all the chunks in the OOTB packet to determine if
658 * the packet should be discarded right away. If a response might be needed
659 * for this packet, or, if further processing is possible, the packet will
660 * be queued to a proper inqueue for the next phase of handling.
661 *
662 * Output:
663 * Return 0 - If further processing is needed.
664 * Return 1 - If the packet can be discarded right away.
665 */
666static int sctp_rcv_ootb(struct sk_buff *skb)
667{
668 struct sctp_chunkhdr *ch, _ch;
669 int ch_end, offset = 0;
670
671 /* Scan through all the chunks in the packet. */
672 do {
673 /* Make sure we have at least the header there */
674 if (offset + sizeof(_ch) > skb->len)
675 break;
676
677 ch = skb_header_pointer(skb, offset, sizeof(*ch), &_ch);
678
679 /* Break out if chunk length is less then minimal. */
680 if (ntohs(ch->length) < sizeof(_ch))
681 break;
682
683 ch_end = offset + SCTP_PAD4(ntohs(ch->length));
684 if (ch_end > skb->len)
685 break;
686
687 /* RFC 8.4, 2) If the OOTB packet contains an ABORT chunk, the
688 * receiver MUST silently discard the OOTB packet and take no
689 * further action.
690 */
691 if (SCTP_CID_ABORT == ch->type)
692 goto discard;
693
694 /* RFC 8.4, 6) If the packet contains a SHUTDOWN COMPLETE
695 * chunk, the receiver should silently discard the packet
696 * and take no further action.
697 */
698 if (SCTP_CID_SHUTDOWN_COMPLETE == ch->type)
699 goto discard;
700
701 /* RFC 4460, 2.11.2
702 * This will discard packets with INIT chunk bundled as
703 * subsequent chunks in the packet. When INIT is first,
704 * the normal INIT processing will discard the chunk.
705 */
706 if (SCTP_CID_INIT == ch->type && (void *)ch != skb->data)
707 goto discard;
708
709 offset = ch_end;
710 } while (ch_end < skb->len);
711
712 return 0;
713
714discard:
715 return 1;
716}
717
718/* Insert endpoint into the hash table. */
719static int __sctp_hash_endpoint(struct sctp_endpoint *ep)
720{
721 struct sock *sk = ep->base.sk;
722 struct net *net = sock_net(sk);
723 struct sctp_hashbucket *head;
724 struct sctp_ep_common *epb;
725
726 epb = &ep->base;
727 epb->hashent = sctp_ep_hashfn(net, epb->bind_addr.port);
728 head = &sctp_ep_hashtable[epb->hashent];
729
730 if (sk->sk_reuseport) {
731 bool any = sctp_is_ep_boundall(sk);
732 struct sctp_ep_common *epb2;
733 struct list_head *list;
734 int cnt = 0, err = 1;
735
736 list_for_each(list, &ep->base.bind_addr.address_list)
737 cnt++;
738
739 sctp_for_each_hentry(epb2, &head->chain) {
740 struct sock *sk2 = epb2->sk;
741
742 if (!net_eq(sock_net(sk2), net) || sk2 == sk ||
743 !uid_eq(sock_i_uid(sk2), sock_i_uid(sk)) ||
744 !sk2->sk_reuseport)
745 continue;
746
747 err = sctp_bind_addrs_check(sctp_sk(sk2),
748 sctp_sk(sk), cnt);
749 if (!err) {
750 err = reuseport_add_sock(sk, sk2, any);
751 if (err)
752 return err;
753 break;
754 } else if (err < 0) {
755 return err;
756 }
757 }
758
759 if (err) {
760 err = reuseport_alloc(sk, any);
761 if (err)
762 return err;
763 }
764 }
765
766 write_lock(&head->lock);
767 hlist_add_head(&epb->node, &head->chain);
768 write_unlock(&head->lock);
769 return 0;
770}
771
772/* Add an endpoint to the hash. Local BH-safe. */
773int sctp_hash_endpoint(struct sctp_endpoint *ep)
774{
775 int err;
776
777 local_bh_disable();
778 err = __sctp_hash_endpoint(ep);
779 local_bh_enable();
780
781 return err;
782}
783
784/* Remove endpoint from the hash table. */
785static void __sctp_unhash_endpoint(struct sctp_endpoint *ep)
786{
787 struct sock *sk = ep->base.sk;
788 struct sctp_hashbucket *head;
789 struct sctp_ep_common *epb;
790
791 epb = &ep->base;
792
793 epb->hashent = sctp_ep_hashfn(sock_net(sk), epb->bind_addr.port);
794
795 head = &sctp_ep_hashtable[epb->hashent];
796
797 if (rcu_access_pointer(sk->sk_reuseport_cb))
798 reuseport_detach_sock(sk);
799
800 write_lock(&head->lock);
801 hlist_del_init(&epb->node);
802 write_unlock(&head->lock);
803}
804
805/* Remove endpoint from the hash. Local BH-safe. */
806void sctp_unhash_endpoint(struct sctp_endpoint *ep)
807{
808 local_bh_disable();
809 __sctp_unhash_endpoint(ep);
810 local_bh_enable();
811}
812
813static inline __u32 sctp_hashfn(const struct net *net, __be16 lport,
814 const union sctp_addr *paddr, __u32 seed)
815{
816 __u32 addr;
817
818 if (paddr->sa.sa_family == AF_INET6)
819 addr = jhash(&paddr->v6.sin6_addr, 16, seed);
820 else
821 addr = (__force __u32)paddr->v4.sin_addr.s_addr;
822
823 return jhash_3words(addr, ((__force __u32)paddr->v4.sin_port) << 16 |
824 (__force __u32)lport, net_hash_mix(net), seed);
825}
826
827/* Look up an endpoint. */
828static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(
829 struct net *net, struct sk_buff *skb,
830 const union sctp_addr *laddr,
831 const union sctp_addr *paddr)
832{
833 struct sctp_hashbucket *head;
834 struct sctp_ep_common *epb;
835 struct sctp_endpoint *ep;
836 struct sock *sk;
837 __be16 lport;
838 int hash;
839
840 lport = laddr->v4.sin_port;
841 hash = sctp_ep_hashfn(net, ntohs(lport));
842 head = &sctp_ep_hashtable[hash];
843 read_lock(&head->lock);
844 sctp_for_each_hentry(epb, &head->chain) {
845 ep = sctp_ep(epb);
846 if (sctp_endpoint_is_match(ep, net, laddr))
847 goto hit;
848 }
849
850 ep = sctp_sk(net->sctp.ctl_sock)->ep;
851
852hit:
853 sk = ep->base.sk;
854 if (sk->sk_reuseport) {
855 __u32 phash = sctp_hashfn(net, lport, paddr, 0);
856
857 sk = reuseport_select_sock(sk, phash, skb,
858 sizeof(struct sctphdr));
859 if (sk)
860 ep = sctp_sk(sk)->ep;
861 }
862 sctp_endpoint_hold(ep);
863 read_unlock(&head->lock);
864 return ep;
865}
866
867/* rhashtable for transport */
868struct sctp_hash_cmp_arg {
869 const union sctp_addr *paddr;
870 const struct net *net;
871 __be16 lport;
872};
873
874static inline int sctp_hash_cmp(struct rhashtable_compare_arg *arg,
875 const void *ptr)
876{
877 struct sctp_transport *t = (struct sctp_transport *)ptr;
878 const struct sctp_hash_cmp_arg *x = arg->key;
879 int err = 1;
880
881 if (!sctp_cmp_addr_exact(&t->ipaddr, x->paddr))
882 return err;
883 if (!sctp_transport_hold(t))
884 return err;
885
886 if (!net_eq(t->asoc->base.net, x->net))
887 goto out;
888 if (x->lport != htons(t->asoc->base.bind_addr.port))
889 goto out;
890
891 err = 0;
892out:
893 sctp_transport_put(t);
894 return err;
895}
896
897static inline __u32 sctp_hash_obj(const void *data, u32 len, u32 seed)
898{
899 const struct sctp_transport *t = data;
900
901 return sctp_hashfn(t->asoc->base.net,
902 htons(t->asoc->base.bind_addr.port),
903 &t->ipaddr, seed);
904}
905
906static inline __u32 sctp_hash_key(const void *data, u32 len, u32 seed)
907{
908 const struct sctp_hash_cmp_arg *x = data;
909
910 return sctp_hashfn(x->net, x->lport, x->paddr, seed);
911}
912
913static const struct rhashtable_params sctp_hash_params = {
914 .head_offset = offsetof(struct sctp_transport, node),
915 .hashfn = sctp_hash_key,
916 .obj_hashfn = sctp_hash_obj,
917 .obj_cmpfn = sctp_hash_cmp,
918 .automatic_shrinking = true,
919};
920
921int sctp_transport_hashtable_init(void)
922{
923 return rhltable_init(&sctp_transport_hashtable, &sctp_hash_params);
924}
925
926void sctp_transport_hashtable_destroy(void)
927{
928 rhltable_destroy(&sctp_transport_hashtable);
929}
930
931int sctp_hash_transport(struct sctp_transport *t)
932{
933 struct sctp_transport *transport;
934 struct rhlist_head *tmp, *list;
935 struct sctp_hash_cmp_arg arg;
936 int err;
937
938 if (t->asoc->temp)
939 return 0;
940
941 arg.net = t->asoc->base.net;
942 arg.paddr = &t->ipaddr;
943 arg.lport = htons(t->asoc->base.bind_addr.port);
944
945 rcu_read_lock();
946 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
947 sctp_hash_params);
948
949 rhl_for_each_entry_rcu(transport, tmp, list, node)
950 if (transport->asoc->ep == t->asoc->ep) {
951 rcu_read_unlock();
952 return -EEXIST;
953 }
954 rcu_read_unlock();
955
956 err = rhltable_insert_key(&sctp_transport_hashtable, &arg,
957 &t->node, sctp_hash_params);
958 if (err)
959 pr_err_once("insert transport fail, errno %d\n", err);
960
961 return err;
962}
963
964void sctp_unhash_transport(struct sctp_transport *t)
965{
966 if (t->asoc->temp)
967 return;
968
969 rhltable_remove(&sctp_transport_hashtable, &t->node,
970 sctp_hash_params);
971}
972
973/* return a transport with holding it */
974struct sctp_transport *sctp_addrs_lookup_transport(
975 struct net *net,
976 const union sctp_addr *laddr,
977 const union sctp_addr *paddr)
978{
979 struct rhlist_head *tmp, *list;
980 struct sctp_transport *t;
981 struct sctp_hash_cmp_arg arg = {
982 .paddr = paddr,
983 .net = net,
984 .lport = laddr->v4.sin_port,
985 };
986
987 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
988 sctp_hash_params);
989
990 rhl_for_each_entry_rcu(t, tmp, list, node) {
991 if (!sctp_transport_hold(t))
992 continue;
993
994 if (sctp_bind_addr_match(&t->asoc->base.bind_addr,
995 laddr, sctp_sk(t->asoc->base.sk)))
996 return t;
997 sctp_transport_put(t);
998 }
999
1000 return NULL;
1001}
1002
1003/* return a transport without holding it, as it's only used under sock lock */
1004struct sctp_transport *sctp_epaddr_lookup_transport(
1005 const struct sctp_endpoint *ep,
1006 const union sctp_addr *paddr)
1007{
1008 struct rhlist_head *tmp, *list;
1009 struct sctp_transport *t;
1010 struct sctp_hash_cmp_arg arg = {
1011 .paddr = paddr,
1012 .net = ep->base.net,
1013 .lport = htons(ep->base.bind_addr.port),
1014 };
1015
1016 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
1017 sctp_hash_params);
1018
1019 rhl_for_each_entry_rcu(t, tmp, list, node)
1020 if (ep == t->asoc->ep)
1021 return t;
1022
1023 return NULL;
1024}
1025
1026/* Look up an association. */
1027static struct sctp_association *__sctp_lookup_association(
1028 struct net *net,
1029 const union sctp_addr *local,
1030 const union sctp_addr *peer,
1031 struct sctp_transport **pt)
1032{
1033 struct sctp_transport *t;
1034 struct sctp_association *asoc = NULL;
1035
1036 t = sctp_addrs_lookup_transport(net, local, peer);
1037 if (!t)
1038 goto out;
1039
1040 asoc = t->asoc;
1041 *pt = t;
1042
1043out:
1044 return asoc;
1045}
1046
1047/* Look up an association. protected by RCU read lock */
1048static
1049struct sctp_association *sctp_lookup_association(struct net *net,
1050 const union sctp_addr *laddr,
1051 const union sctp_addr *paddr,
1052 struct sctp_transport **transportp)
1053{
1054 struct sctp_association *asoc;
1055
1056 rcu_read_lock();
1057 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1058 rcu_read_unlock();
1059
1060 return asoc;
1061}
1062
1063/* Is there an association matching the given local and peer addresses? */
1064bool sctp_has_association(struct net *net,
1065 const union sctp_addr *laddr,
1066 const union sctp_addr *paddr)
1067{
1068 struct sctp_transport *transport;
1069
1070 if (sctp_lookup_association(net, laddr, paddr, &transport)) {
1071 sctp_transport_put(transport);
1072 return true;
1073 }
1074
1075 return false;
1076}
1077
1078/*
1079 * SCTP Implementors Guide, 2.18 Handling of address
1080 * parameters within the INIT or INIT-ACK.
1081 *
1082 * D) When searching for a matching TCB upon reception of an INIT
1083 * or INIT-ACK chunk the receiver SHOULD use not only the
1084 * source address of the packet (containing the INIT or
1085 * INIT-ACK) but the receiver SHOULD also use all valid
1086 * address parameters contained within the chunk.
1087 *
1088 * 2.18.3 Solution description
1089 *
1090 * This new text clearly specifies to an implementor the need
1091 * to look within the INIT or INIT-ACK. Any implementation that
1092 * does not do this, may not be able to establish associations
1093 * in certain circumstances.
1094 *
1095 */
1096static struct sctp_association *__sctp_rcv_init_lookup(struct net *net,
1097 struct sk_buff *skb,
1098 const union sctp_addr *laddr, struct sctp_transport **transportp)
1099{
1100 struct sctp_association *asoc;
1101 union sctp_addr addr;
1102 union sctp_addr *paddr = &addr;
1103 struct sctphdr *sh = sctp_hdr(skb);
1104 union sctp_params params;
1105 struct sctp_init_chunk *init;
1106 struct sctp_af *af;
1107
1108 /*
1109 * This code will NOT touch anything inside the chunk--it is
1110 * strictly READ-ONLY.
1111 *
1112 * RFC 2960 3 SCTP packet Format
1113 *
1114 * Multiple chunks can be bundled into one SCTP packet up to
1115 * the MTU size, except for the INIT, INIT ACK, and SHUTDOWN
1116 * COMPLETE chunks. These chunks MUST NOT be bundled with any
1117 * other chunk in a packet. See Section 6.10 for more details
1118 * on chunk bundling.
1119 */
1120
1121 /* Find the start of the TLVs and the end of the chunk. This is
1122 * the region we search for address parameters.
1123 */
1124 init = (struct sctp_init_chunk *)skb->data;
1125
1126 /* Walk the parameters looking for embedded addresses. */
1127 sctp_walk_params(params, init, init_hdr.params) {
1128
1129 /* Note: Ignoring hostname addresses. */
1130 af = sctp_get_af_specific(param_type2af(params.p->type));
1131 if (!af)
1132 continue;
1133
1134 af->from_addr_param(paddr, params.addr, sh->source, 0);
1135
1136 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1137 if (asoc)
1138 return asoc;
1139 }
1140
1141 return NULL;
1142}
1143
1144/* ADD-IP, Section 5.2
1145 * When an endpoint receives an ASCONF Chunk from the remote peer
1146 * special procedures may be needed to identify the association the
1147 * ASCONF Chunk is associated with. To properly find the association
1148 * the following procedures SHOULD be followed:
1149 *
1150 * D2) If the association is not found, use the address found in the
1151 * Address Parameter TLV combined with the port number found in the
1152 * SCTP common header. If found proceed to rule D4.
1153 *
1154 * D2-ext) If more than one ASCONF Chunks are packed together, use the
1155 * address found in the ASCONF Address Parameter TLV of each of the
1156 * subsequent ASCONF Chunks. If found, proceed to rule D4.
1157 */
1158static struct sctp_association *__sctp_rcv_asconf_lookup(
1159 struct net *net,
1160 struct sctp_chunkhdr *ch,
1161 const union sctp_addr *laddr,
1162 __be16 peer_port,
1163 struct sctp_transport **transportp)
1164{
1165 struct sctp_addip_chunk *asconf = (struct sctp_addip_chunk *)ch;
1166 struct sctp_af *af;
1167 union sctp_addr_param *param;
1168 union sctp_addr paddr;
1169
1170 /* Skip over the ADDIP header and find the Address parameter */
1171 param = (union sctp_addr_param *)(asconf + 1);
1172
1173 af = sctp_get_af_specific(param_type2af(param->p.type));
1174 if (unlikely(!af))
1175 return NULL;
1176
1177 af->from_addr_param(&paddr, param, peer_port, 0);
1178
1179 return __sctp_lookup_association(net, laddr, &paddr, transportp);
1180}
1181
1182
1183/* SCTP-AUTH, Section 6.3:
1184* If the receiver does not find a STCB for a packet containing an AUTH
1185* chunk as the first chunk and not a COOKIE-ECHO chunk as the second
1186* chunk, it MUST use the chunks after the AUTH chunk to look up an existing
1187* association.
1188*
1189* This means that any chunks that can help us identify the association need
1190* to be looked at to find this association.
1191*/
1192static struct sctp_association *__sctp_rcv_walk_lookup(struct net *net,
1193 struct sk_buff *skb,
1194 const union sctp_addr *laddr,
1195 struct sctp_transport **transportp)
1196{
1197 struct sctp_association *asoc = NULL;
1198 struct sctp_chunkhdr *ch;
1199 int have_auth = 0;
1200 unsigned int chunk_num = 1;
1201 __u8 *ch_end;
1202
1203 /* Walk through the chunks looking for AUTH or ASCONF chunks
1204 * to help us find the association.
1205 */
1206 ch = (struct sctp_chunkhdr *)skb->data;
1207 do {
1208 /* Break out if chunk length is less then minimal. */
1209 if (ntohs(ch->length) < sizeof(*ch))
1210 break;
1211
1212 ch_end = ((__u8 *)ch) + SCTP_PAD4(ntohs(ch->length));
1213 if (ch_end > skb_tail_pointer(skb))
1214 break;
1215
1216 switch (ch->type) {
1217 case SCTP_CID_AUTH:
1218 have_auth = chunk_num;
1219 break;
1220
1221 case SCTP_CID_COOKIE_ECHO:
1222 /* If a packet arrives containing an AUTH chunk as
1223 * a first chunk, a COOKIE-ECHO chunk as the second
1224 * chunk, and possibly more chunks after them, and
1225 * the receiver does not have an STCB for that
1226 * packet, then authentication is based on
1227 * the contents of the COOKIE- ECHO chunk.
1228 */
1229 if (have_auth == 1 && chunk_num == 2)
1230 return NULL;
1231 break;
1232
1233 case SCTP_CID_ASCONF:
1234 if (have_auth || net->sctp.addip_noauth)
1235 asoc = __sctp_rcv_asconf_lookup(
1236 net, ch, laddr,
1237 sctp_hdr(skb)->source,
1238 transportp);
1239 default:
1240 break;
1241 }
1242
1243 if (asoc)
1244 break;
1245
1246 ch = (struct sctp_chunkhdr *)ch_end;
1247 chunk_num++;
1248 } while (ch_end < skb_tail_pointer(skb));
1249
1250 return asoc;
1251}
1252
1253/*
1254 * There are circumstances when we need to look inside the SCTP packet
1255 * for information to help us find the association. Examples
1256 * include looking inside of INIT/INIT-ACK chunks or after the AUTH
1257 * chunks.
1258 */
1259static struct sctp_association *__sctp_rcv_lookup_harder(struct net *net,
1260 struct sk_buff *skb,
1261 const union sctp_addr *laddr,
1262 struct sctp_transport **transportp)
1263{
1264 struct sctp_chunkhdr *ch;
1265
1266 /* We do not allow GSO frames here as we need to linearize and
1267 * then cannot guarantee frame boundaries. This shouldn't be an
1268 * issue as packets hitting this are mostly INIT or INIT-ACK and
1269 * those cannot be on GSO-style anyway.
1270 */
1271 if (skb_is_gso(skb) && skb_is_gso_sctp(skb))
1272 return NULL;
1273
1274 ch = (struct sctp_chunkhdr *)skb->data;
1275
1276 /* The code below will attempt to walk the chunk and extract
1277 * parameter information. Before we do that, we need to verify
1278 * that the chunk length doesn't cause overflow. Otherwise, we'll
1279 * walk off the end.
1280 */
1281 if (SCTP_PAD4(ntohs(ch->length)) > skb->len)
1282 return NULL;
1283
1284 /* If this is INIT/INIT-ACK look inside the chunk too. */
1285 if (ch->type == SCTP_CID_INIT || ch->type == SCTP_CID_INIT_ACK)
1286 return __sctp_rcv_init_lookup(net, skb, laddr, transportp);
1287
1288 return __sctp_rcv_walk_lookup(net, skb, laddr, transportp);
1289}
1290
1291/* Lookup an association for an inbound skb. */
1292static struct sctp_association *__sctp_rcv_lookup(struct net *net,
1293 struct sk_buff *skb,
1294 const union sctp_addr *paddr,
1295 const union sctp_addr *laddr,
1296 struct sctp_transport **transportp)
1297{
1298 struct sctp_association *asoc;
1299
1300 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1301 if (asoc)
1302 goto out;
1303
1304 /* Further lookup for INIT/INIT-ACK packets.
1305 * SCTP Implementors Guide, 2.18 Handling of address
1306 * parameters within the INIT or INIT-ACK.
1307 */
1308 asoc = __sctp_rcv_lookup_harder(net, skb, laddr, transportp);
1309 if (asoc)
1310 goto out;
1311
1312 if (paddr->sa.sa_family == AF_INET)
1313 pr_debug("sctp: asoc not found for src:%pI4:%d dst:%pI4:%d\n",
1314 &laddr->v4.sin_addr, ntohs(laddr->v4.sin_port),
1315 &paddr->v4.sin_addr, ntohs(paddr->v4.sin_port));
1316 else
1317 pr_debug("sctp: asoc not found for src:%pI6:%d dst:%pI6:%d\n",
1318 &laddr->v6.sin6_addr, ntohs(laddr->v6.sin6_port),
1319 &paddr->v6.sin6_addr, ntohs(paddr->v6.sin6_port));
1320
1321out:
1322 return asoc;
1323}
1/* SCTP kernel implementation
2 * Copyright (c) 1999-2000 Cisco, Inc.
3 * Copyright (c) 1999-2001 Motorola, Inc.
4 * Copyright (c) 2001-2003 International Business Machines, Corp.
5 * Copyright (c) 2001 Intel Corp.
6 * Copyright (c) 2001 Nokia, Inc.
7 * Copyright (c) 2001 La Monte H.P. Yarroll
8 *
9 * This file is part of the SCTP kernel implementation
10 *
11 * These functions handle all input from the IP layer into SCTP.
12 *
13 * This SCTP implementation is free software;
14 * you can redistribute it and/or modify it under the terms of
15 * the GNU General Public License as published by
16 * the Free Software Foundation; either version 2, or (at your option)
17 * any later version.
18 *
19 * This SCTP implementation is distributed in the hope that it
20 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
21 * ************************
22 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
23 * See the GNU General Public License for more details.
24 *
25 * You should have received a copy of the GNU General Public License
26 * along with GNU CC; see the file COPYING. If not, see
27 * <http://www.gnu.org/licenses/>.
28 *
29 * Please send any bug reports or fixes you make to the
30 * email address(es):
31 * lksctp developers <linux-sctp@vger.kernel.org>
32 *
33 * Written or modified by:
34 * La Monte H.P. Yarroll <piggy@acm.org>
35 * Karl Knutson <karl@athena.chicago.il.us>
36 * Xingang Guo <xingang.guo@intel.com>
37 * Jon Grimm <jgrimm@us.ibm.com>
38 * Hui Huang <hui.huang@nokia.com>
39 * Daisy Chang <daisyc@us.ibm.com>
40 * Sridhar Samudrala <sri@us.ibm.com>
41 * Ardelle Fan <ardelle.fan@intel.com>
42 */
43
44#include <linux/types.h>
45#include <linux/list.h> /* For struct list_head */
46#include <linux/socket.h>
47#include <linux/ip.h>
48#include <linux/time.h> /* For struct timeval */
49#include <linux/slab.h>
50#include <net/ip.h>
51#include <net/icmp.h>
52#include <net/snmp.h>
53#include <net/sock.h>
54#include <net/xfrm.h>
55#include <net/sctp/sctp.h>
56#include <net/sctp/sm.h>
57#include <net/sctp/checksum.h>
58#include <net/net_namespace.h>
59
60/* Forward declarations for internal helpers. */
61static int sctp_rcv_ootb(struct sk_buff *);
62static struct sctp_association *__sctp_rcv_lookup(struct net *net,
63 struct sk_buff *skb,
64 const union sctp_addr *paddr,
65 const union sctp_addr *laddr,
66 struct sctp_transport **transportp);
67static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(struct net *net,
68 const union sctp_addr *laddr);
69static struct sctp_association *__sctp_lookup_association(
70 struct net *net,
71 const union sctp_addr *local,
72 const union sctp_addr *peer,
73 struct sctp_transport **pt);
74
75static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb);
76
77
78/* Calculate the SCTP checksum of an SCTP packet. */
79static inline int sctp_rcv_checksum(struct net *net, struct sk_buff *skb)
80{
81 struct sctphdr *sh = sctp_hdr(skb);
82 __le32 cmp = sh->checksum;
83 __le32 val = sctp_compute_cksum(skb, 0);
84
85 if (val != cmp) {
86 /* CRC failure, dump it. */
87 __SCTP_INC_STATS(net, SCTP_MIB_CHECKSUMERRORS);
88 return -1;
89 }
90 return 0;
91}
92
93/*
94 * This is the routine which IP calls when receiving an SCTP packet.
95 */
96int sctp_rcv(struct sk_buff *skb)
97{
98 struct sock *sk;
99 struct sctp_association *asoc;
100 struct sctp_endpoint *ep = NULL;
101 struct sctp_ep_common *rcvr;
102 struct sctp_transport *transport = NULL;
103 struct sctp_chunk *chunk;
104 union sctp_addr src;
105 union sctp_addr dest;
106 int family;
107 struct sctp_af *af;
108 struct net *net = dev_net(skb->dev);
109
110 if (skb->pkt_type != PACKET_HOST)
111 goto discard_it;
112
113 __SCTP_INC_STATS(net, SCTP_MIB_INSCTPPACKS);
114
115 /* If packet is too small to contain a single chunk, let's not
116 * waste time on it anymore.
117 */
118 if (skb->len < sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) +
119 skb_transport_offset(skb))
120 goto discard_it;
121
122 /* If the packet is fragmented and we need to do crc checking,
123 * it's better to just linearize it otherwise crc computing
124 * takes longer.
125 */
126 if ((!(skb_shinfo(skb)->gso_type & SKB_GSO_SCTP) &&
127 skb_linearize(skb)) ||
128 !pskb_may_pull(skb, sizeof(struct sctphdr)))
129 goto discard_it;
130
131 /* Pull up the IP header. */
132 __skb_pull(skb, skb_transport_offset(skb));
133
134 skb->csum_valid = 0; /* Previous value not applicable */
135 if (skb_csum_unnecessary(skb))
136 __skb_decr_checksum_unnecessary(skb);
137 else if (!sctp_checksum_disable &&
138 !(skb_shinfo(skb)->gso_type & SKB_GSO_SCTP) &&
139 sctp_rcv_checksum(net, skb) < 0)
140 goto discard_it;
141 skb->csum_valid = 1;
142
143 __skb_pull(skb, sizeof(struct sctphdr));
144
145 family = ipver2af(ip_hdr(skb)->version);
146 af = sctp_get_af_specific(family);
147 if (unlikely(!af))
148 goto discard_it;
149 SCTP_INPUT_CB(skb)->af = af;
150
151 /* Initialize local addresses for lookups. */
152 af->from_skb(&src, skb, 1);
153 af->from_skb(&dest, skb, 0);
154
155 /* If the packet is to or from a non-unicast address,
156 * silently discard the packet.
157 *
158 * This is not clearly defined in the RFC except in section
159 * 8.4 - OOTB handling. However, based on the book "Stream Control
160 * Transmission Protocol" 2.1, "It is important to note that the
161 * IP address of an SCTP transport address must be a routable
162 * unicast address. In other words, IP multicast addresses and
163 * IP broadcast addresses cannot be used in an SCTP transport
164 * address."
165 */
166 if (!af->addr_valid(&src, NULL, skb) ||
167 !af->addr_valid(&dest, NULL, skb))
168 goto discard_it;
169
170 asoc = __sctp_rcv_lookup(net, skb, &src, &dest, &transport);
171
172 if (!asoc)
173 ep = __sctp_rcv_lookup_endpoint(net, &dest);
174
175 /* Retrieve the common input handling substructure. */
176 rcvr = asoc ? &asoc->base : &ep->base;
177 sk = rcvr->sk;
178
179 /*
180 * If a frame arrives on an interface and the receiving socket is
181 * bound to another interface, via SO_BINDTODEVICE, treat it as OOTB
182 */
183 if (sk->sk_bound_dev_if && (sk->sk_bound_dev_if != af->skb_iif(skb))) {
184 if (transport) {
185 sctp_transport_put(transport);
186 asoc = NULL;
187 transport = NULL;
188 } else {
189 sctp_endpoint_put(ep);
190 ep = NULL;
191 }
192 sk = net->sctp.ctl_sock;
193 ep = sctp_sk(sk)->ep;
194 sctp_endpoint_hold(ep);
195 rcvr = &ep->base;
196 }
197
198 /*
199 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
200 * An SCTP packet is called an "out of the blue" (OOTB)
201 * packet if it is correctly formed, i.e., passed the
202 * receiver's checksum check, but the receiver is not
203 * able to identify the association to which this
204 * packet belongs.
205 */
206 if (!asoc) {
207 if (sctp_rcv_ootb(skb)) {
208 __SCTP_INC_STATS(net, SCTP_MIB_OUTOFBLUES);
209 goto discard_release;
210 }
211 }
212
213 if (!xfrm_policy_check(sk, XFRM_POLICY_IN, skb, family))
214 goto discard_release;
215 nf_reset(skb);
216
217 if (sk_filter(sk, skb))
218 goto discard_release;
219
220 /* Create an SCTP packet structure. */
221 chunk = sctp_chunkify(skb, asoc, sk, GFP_ATOMIC);
222 if (!chunk)
223 goto discard_release;
224 SCTP_INPUT_CB(skb)->chunk = chunk;
225
226 /* Remember what endpoint is to handle this packet. */
227 chunk->rcvr = rcvr;
228
229 /* Remember the SCTP header. */
230 chunk->sctp_hdr = sctp_hdr(skb);
231
232 /* Set the source and destination addresses of the incoming chunk. */
233 sctp_init_addrs(chunk, &src, &dest);
234
235 /* Remember where we came from. */
236 chunk->transport = transport;
237
238 /* Acquire access to the sock lock. Note: We are safe from other
239 * bottom halves on this lock, but a user may be in the lock too,
240 * so check if it is busy.
241 */
242 bh_lock_sock(sk);
243
244 if (sk != rcvr->sk) {
245 /* Our cached sk is different from the rcvr->sk. This is
246 * because migrate()/accept() may have moved the association
247 * to a new socket and released all the sockets. So now we
248 * are holding a lock on the old socket while the user may
249 * be doing something with the new socket. Switch our veiw
250 * of the current sk.
251 */
252 bh_unlock_sock(sk);
253 sk = rcvr->sk;
254 bh_lock_sock(sk);
255 }
256
257 if (sock_owned_by_user(sk)) {
258 if (sctp_add_backlog(sk, skb)) {
259 bh_unlock_sock(sk);
260 sctp_chunk_free(chunk);
261 skb = NULL; /* sctp_chunk_free already freed the skb */
262 goto discard_release;
263 }
264 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_BACKLOG);
265 } else {
266 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_SOFTIRQ);
267 sctp_inq_push(&chunk->rcvr->inqueue, chunk);
268 }
269
270 bh_unlock_sock(sk);
271
272 /* Release the asoc/ep ref we took in the lookup calls. */
273 if (transport)
274 sctp_transport_put(transport);
275 else
276 sctp_endpoint_put(ep);
277
278 return 0;
279
280discard_it:
281 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_DISCARDS);
282 kfree_skb(skb);
283 return 0;
284
285discard_release:
286 /* Release the asoc/ep ref we took in the lookup calls. */
287 if (transport)
288 sctp_transport_put(transport);
289 else
290 sctp_endpoint_put(ep);
291
292 goto discard_it;
293}
294
295/* Process the backlog queue of the socket. Every skb on
296 * the backlog holds a ref on an association or endpoint.
297 * We hold this ref throughout the state machine to make
298 * sure that the structure we need is still around.
299 */
300int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb)
301{
302 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
303 struct sctp_inq *inqueue = &chunk->rcvr->inqueue;
304 struct sctp_transport *t = chunk->transport;
305 struct sctp_ep_common *rcvr = NULL;
306 int backloged = 0;
307
308 rcvr = chunk->rcvr;
309
310 /* If the rcvr is dead then the association or endpoint
311 * has been deleted and we can safely drop the chunk
312 * and refs that we are holding.
313 */
314 if (rcvr->dead) {
315 sctp_chunk_free(chunk);
316 goto done;
317 }
318
319 if (unlikely(rcvr->sk != sk)) {
320 /* In this case, the association moved from one socket to
321 * another. We are currently sitting on the backlog of the
322 * old socket, so we need to move.
323 * However, since we are here in the process context we
324 * need to take make sure that the user doesn't own
325 * the new socket when we process the packet.
326 * If the new socket is user-owned, queue the chunk to the
327 * backlog of the new socket without dropping any refs.
328 * Otherwise, we can safely push the chunk on the inqueue.
329 */
330
331 sk = rcvr->sk;
332 local_bh_disable();
333 bh_lock_sock(sk);
334
335 if (sock_owned_by_user(sk)) {
336 if (sk_add_backlog(sk, skb, sk->sk_rcvbuf))
337 sctp_chunk_free(chunk);
338 else
339 backloged = 1;
340 } else
341 sctp_inq_push(inqueue, chunk);
342
343 bh_unlock_sock(sk);
344 local_bh_enable();
345
346 /* If the chunk was backloged again, don't drop refs */
347 if (backloged)
348 return 0;
349 } else {
350 sctp_inq_push(inqueue, chunk);
351 }
352
353done:
354 /* Release the refs we took in sctp_add_backlog */
355 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
356 sctp_transport_put(t);
357 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
358 sctp_endpoint_put(sctp_ep(rcvr));
359 else
360 BUG();
361
362 return 0;
363}
364
365static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb)
366{
367 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
368 struct sctp_transport *t = chunk->transport;
369 struct sctp_ep_common *rcvr = chunk->rcvr;
370 int ret;
371
372 ret = sk_add_backlog(sk, skb, sk->sk_rcvbuf);
373 if (!ret) {
374 /* Hold the assoc/ep while hanging on the backlog queue.
375 * This way, we know structures we need will not disappear
376 * from us
377 */
378 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
379 sctp_transport_hold(t);
380 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
381 sctp_endpoint_hold(sctp_ep(rcvr));
382 else
383 BUG();
384 }
385 return ret;
386
387}
388
389/* Handle icmp frag needed error. */
390void sctp_icmp_frag_needed(struct sock *sk, struct sctp_association *asoc,
391 struct sctp_transport *t, __u32 pmtu)
392{
393 if (!t || (t->pathmtu <= pmtu))
394 return;
395
396 if (sock_owned_by_user(sk)) {
397 asoc->pmtu_pending = 1;
398 t->pmtu_pending = 1;
399 return;
400 }
401
402 if (t->param_flags & SPP_PMTUD_ENABLE) {
403 /* Update transports view of the MTU */
404 sctp_transport_update_pmtu(sk, t, pmtu);
405
406 /* Update association pmtu. */
407 sctp_assoc_sync_pmtu(sk, asoc);
408 }
409
410 /* Retransmit with the new pmtu setting.
411 * Normally, if PMTU discovery is disabled, an ICMP Fragmentation
412 * Needed will never be sent, but if a message was sent before
413 * PMTU discovery was disabled that was larger than the PMTU, it
414 * would not be fragmented, so it must be re-transmitted fragmented.
415 */
416 sctp_retransmit(&asoc->outqueue, t, SCTP_RTXR_PMTUD);
417}
418
419void sctp_icmp_redirect(struct sock *sk, struct sctp_transport *t,
420 struct sk_buff *skb)
421{
422 struct dst_entry *dst;
423
424 if (!t)
425 return;
426 dst = sctp_transport_dst_check(t);
427 if (dst)
428 dst->ops->redirect(dst, sk, skb);
429}
430
431/*
432 * SCTP Implementer's Guide, 2.37 ICMP handling procedures
433 *
434 * ICMP8) If the ICMP code is a "Unrecognized next header type encountered"
435 * or a "Protocol Unreachable" treat this message as an abort
436 * with the T bit set.
437 *
438 * This function sends an event to the state machine, which will abort the
439 * association.
440 *
441 */
442void sctp_icmp_proto_unreachable(struct sock *sk,
443 struct sctp_association *asoc,
444 struct sctp_transport *t)
445{
446 if (sock_owned_by_user(sk)) {
447 if (timer_pending(&t->proto_unreach_timer))
448 return;
449 else {
450 if (!mod_timer(&t->proto_unreach_timer,
451 jiffies + (HZ/20)))
452 sctp_association_hold(asoc);
453 }
454 } else {
455 struct net *net = sock_net(sk);
456
457 pr_debug("%s: unrecognized next header type "
458 "encountered!\n", __func__);
459
460 if (del_timer(&t->proto_unreach_timer))
461 sctp_association_put(asoc);
462
463 sctp_do_sm(net, SCTP_EVENT_T_OTHER,
464 SCTP_ST_OTHER(SCTP_EVENT_ICMP_PROTO_UNREACH),
465 asoc->state, asoc->ep, asoc, t,
466 GFP_ATOMIC);
467 }
468}
469
470/* Common lookup code for icmp/icmpv6 error handler. */
471struct sock *sctp_err_lookup(struct net *net, int family, struct sk_buff *skb,
472 struct sctphdr *sctphdr,
473 struct sctp_association **app,
474 struct sctp_transport **tpp)
475{
476 union sctp_addr saddr;
477 union sctp_addr daddr;
478 struct sctp_af *af;
479 struct sock *sk = NULL;
480 struct sctp_association *asoc;
481 struct sctp_transport *transport = NULL;
482 struct sctp_init_chunk *chunkhdr;
483 __u32 vtag = ntohl(sctphdr->vtag);
484 int len = skb->len - ((void *)sctphdr - (void *)skb->data);
485
486 *app = NULL; *tpp = NULL;
487
488 af = sctp_get_af_specific(family);
489 if (unlikely(!af)) {
490 return NULL;
491 }
492
493 /* Initialize local addresses for lookups. */
494 af->from_skb(&saddr, skb, 1);
495 af->from_skb(&daddr, skb, 0);
496
497 /* Look for an association that matches the incoming ICMP error
498 * packet.
499 */
500 asoc = __sctp_lookup_association(net, &saddr, &daddr, &transport);
501 if (!asoc)
502 return NULL;
503
504 sk = asoc->base.sk;
505
506 /* RFC 4960, Appendix C. ICMP Handling
507 *
508 * ICMP6) An implementation MUST validate that the Verification Tag
509 * contained in the ICMP message matches the Verification Tag of
510 * the peer. If the Verification Tag is not 0 and does NOT
511 * match, discard the ICMP message. If it is 0 and the ICMP
512 * message contains enough bytes to verify that the chunk type is
513 * an INIT chunk and that the Initiate Tag matches the tag of the
514 * peer, continue with ICMP7. If the ICMP message is too short
515 * or the chunk type or the Initiate Tag does not match, silently
516 * discard the packet.
517 */
518 if (vtag == 0) {
519 chunkhdr = (void *)sctphdr + sizeof(struct sctphdr);
520 if (len < sizeof(struct sctphdr) + sizeof(sctp_chunkhdr_t)
521 + sizeof(__be32) ||
522 chunkhdr->chunk_hdr.type != SCTP_CID_INIT ||
523 ntohl(chunkhdr->init_hdr.init_tag) != asoc->c.my_vtag) {
524 goto out;
525 }
526 } else if (vtag != asoc->c.peer_vtag) {
527 goto out;
528 }
529
530 bh_lock_sock(sk);
531
532 /* If too many ICMPs get dropped on busy
533 * servers this needs to be solved differently.
534 */
535 if (sock_owned_by_user(sk))
536 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
537
538 *app = asoc;
539 *tpp = transport;
540 return sk;
541
542out:
543 sctp_transport_put(transport);
544 return NULL;
545}
546
547/* Common cleanup code for icmp/icmpv6 error handler. */
548void sctp_err_finish(struct sock *sk, struct sctp_transport *t)
549{
550 bh_unlock_sock(sk);
551 sctp_transport_put(t);
552}
553
554/*
555 * This routine is called by the ICMP module when it gets some
556 * sort of error condition. If err < 0 then the socket should
557 * be closed and the error returned to the user. If err > 0
558 * it's just the icmp type << 8 | icmp code. After adjustment
559 * header points to the first 8 bytes of the sctp header. We need
560 * to find the appropriate port.
561 *
562 * The locking strategy used here is very "optimistic". When
563 * someone else accesses the socket the ICMP is just dropped
564 * and for some paths there is no check at all.
565 * A more general error queue to queue errors for later handling
566 * is probably better.
567 *
568 */
569void sctp_v4_err(struct sk_buff *skb, __u32 info)
570{
571 const struct iphdr *iph = (const struct iphdr *)skb->data;
572 const int ihlen = iph->ihl * 4;
573 const int type = icmp_hdr(skb)->type;
574 const int code = icmp_hdr(skb)->code;
575 struct sock *sk;
576 struct sctp_association *asoc = NULL;
577 struct sctp_transport *transport;
578 struct inet_sock *inet;
579 __u16 saveip, savesctp;
580 int err;
581 struct net *net = dev_net(skb->dev);
582
583 /* Fix up skb to look at the embedded net header. */
584 saveip = skb->network_header;
585 savesctp = skb->transport_header;
586 skb_reset_network_header(skb);
587 skb_set_transport_header(skb, ihlen);
588 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &transport);
589 /* Put back, the original values. */
590 skb->network_header = saveip;
591 skb->transport_header = savesctp;
592 if (!sk) {
593 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
594 return;
595 }
596 /* Warning: The sock lock is held. Remember to call
597 * sctp_err_finish!
598 */
599
600 switch (type) {
601 case ICMP_PARAMETERPROB:
602 err = EPROTO;
603 break;
604 case ICMP_DEST_UNREACH:
605 if (code > NR_ICMP_UNREACH)
606 goto out_unlock;
607
608 /* PMTU discovery (RFC1191) */
609 if (ICMP_FRAG_NEEDED == code) {
610 sctp_icmp_frag_needed(sk, asoc, transport,
611 SCTP_TRUNC4(info));
612 goto out_unlock;
613 } else {
614 if (ICMP_PROT_UNREACH == code) {
615 sctp_icmp_proto_unreachable(sk, asoc,
616 transport);
617 goto out_unlock;
618 }
619 }
620 err = icmp_err_convert[code].errno;
621 break;
622 case ICMP_TIME_EXCEEDED:
623 /* Ignore any time exceeded errors due to fragment reassembly
624 * timeouts.
625 */
626 if (ICMP_EXC_FRAGTIME == code)
627 goto out_unlock;
628
629 err = EHOSTUNREACH;
630 break;
631 case ICMP_REDIRECT:
632 sctp_icmp_redirect(sk, transport, skb);
633 /* Fall through to out_unlock. */
634 default:
635 goto out_unlock;
636 }
637
638 inet = inet_sk(sk);
639 if (!sock_owned_by_user(sk) && inet->recverr) {
640 sk->sk_err = err;
641 sk->sk_error_report(sk);
642 } else { /* Only an error on timeout */
643 sk->sk_err_soft = err;
644 }
645
646out_unlock:
647 sctp_err_finish(sk, transport);
648}
649
650/*
651 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
652 *
653 * This function scans all the chunks in the OOTB packet to determine if
654 * the packet should be discarded right away. If a response might be needed
655 * for this packet, or, if further processing is possible, the packet will
656 * be queued to a proper inqueue for the next phase of handling.
657 *
658 * Output:
659 * Return 0 - If further processing is needed.
660 * Return 1 - If the packet can be discarded right away.
661 */
662static int sctp_rcv_ootb(struct sk_buff *skb)
663{
664 sctp_chunkhdr_t *ch, _ch;
665 int ch_end, offset = 0;
666
667 /* Scan through all the chunks in the packet. */
668 do {
669 /* Make sure we have at least the header there */
670 if (offset + sizeof(sctp_chunkhdr_t) > skb->len)
671 break;
672
673 ch = skb_header_pointer(skb, offset, sizeof(*ch), &_ch);
674
675 /* Break out if chunk length is less then minimal. */
676 if (ntohs(ch->length) < sizeof(sctp_chunkhdr_t))
677 break;
678
679 ch_end = offset + SCTP_PAD4(ntohs(ch->length));
680 if (ch_end > skb->len)
681 break;
682
683 /* RFC 8.4, 2) If the OOTB packet contains an ABORT chunk, the
684 * receiver MUST silently discard the OOTB packet and take no
685 * further action.
686 */
687 if (SCTP_CID_ABORT == ch->type)
688 goto discard;
689
690 /* RFC 8.4, 6) If the packet contains a SHUTDOWN COMPLETE
691 * chunk, the receiver should silently discard the packet
692 * and take no further action.
693 */
694 if (SCTP_CID_SHUTDOWN_COMPLETE == ch->type)
695 goto discard;
696
697 /* RFC 4460, 2.11.2
698 * This will discard packets with INIT chunk bundled as
699 * subsequent chunks in the packet. When INIT is first,
700 * the normal INIT processing will discard the chunk.
701 */
702 if (SCTP_CID_INIT == ch->type && (void *)ch != skb->data)
703 goto discard;
704
705 offset = ch_end;
706 } while (ch_end < skb->len);
707
708 return 0;
709
710discard:
711 return 1;
712}
713
714/* Insert endpoint into the hash table. */
715static void __sctp_hash_endpoint(struct sctp_endpoint *ep)
716{
717 struct net *net = sock_net(ep->base.sk);
718 struct sctp_ep_common *epb;
719 struct sctp_hashbucket *head;
720
721 epb = &ep->base;
722
723 epb->hashent = sctp_ep_hashfn(net, epb->bind_addr.port);
724 head = &sctp_ep_hashtable[epb->hashent];
725
726 write_lock(&head->lock);
727 hlist_add_head(&epb->node, &head->chain);
728 write_unlock(&head->lock);
729}
730
731/* Add an endpoint to the hash. Local BH-safe. */
732void sctp_hash_endpoint(struct sctp_endpoint *ep)
733{
734 local_bh_disable();
735 __sctp_hash_endpoint(ep);
736 local_bh_enable();
737}
738
739/* Remove endpoint from the hash table. */
740static void __sctp_unhash_endpoint(struct sctp_endpoint *ep)
741{
742 struct net *net = sock_net(ep->base.sk);
743 struct sctp_hashbucket *head;
744 struct sctp_ep_common *epb;
745
746 epb = &ep->base;
747
748 epb->hashent = sctp_ep_hashfn(net, epb->bind_addr.port);
749
750 head = &sctp_ep_hashtable[epb->hashent];
751
752 write_lock(&head->lock);
753 hlist_del_init(&epb->node);
754 write_unlock(&head->lock);
755}
756
757/* Remove endpoint from the hash. Local BH-safe. */
758void sctp_unhash_endpoint(struct sctp_endpoint *ep)
759{
760 local_bh_disable();
761 __sctp_unhash_endpoint(ep);
762 local_bh_enable();
763}
764
765/* Look up an endpoint. */
766static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(struct net *net,
767 const union sctp_addr *laddr)
768{
769 struct sctp_hashbucket *head;
770 struct sctp_ep_common *epb;
771 struct sctp_endpoint *ep;
772 int hash;
773
774 hash = sctp_ep_hashfn(net, ntohs(laddr->v4.sin_port));
775 head = &sctp_ep_hashtable[hash];
776 read_lock(&head->lock);
777 sctp_for_each_hentry(epb, &head->chain) {
778 ep = sctp_ep(epb);
779 if (sctp_endpoint_is_match(ep, net, laddr))
780 goto hit;
781 }
782
783 ep = sctp_sk(net->sctp.ctl_sock)->ep;
784
785hit:
786 sctp_endpoint_hold(ep);
787 read_unlock(&head->lock);
788 return ep;
789}
790
791/* rhashtable for transport */
792struct sctp_hash_cmp_arg {
793 const union sctp_addr *paddr;
794 const struct net *net;
795 u16 lport;
796};
797
798static inline int sctp_hash_cmp(struct rhashtable_compare_arg *arg,
799 const void *ptr)
800{
801 struct sctp_transport *t = (struct sctp_transport *)ptr;
802 const struct sctp_hash_cmp_arg *x = arg->key;
803 int err = 1;
804
805 if (!sctp_cmp_addr_exact(&t->ipaddr, x->paddr))
806 return err;
807 if (!sctp_transport_hold(t))
808 return err;
809
810 if (!net_eq(sock_net(t->asoc->base.sk), x->net))
811 goto out;
812 if (x->lport != htons(t->asoc->base.bind_addr.port))
813 goto out;
814
815 err = 0;
816out:
817 sctp_transport_put(t);
818 return err;
819}
820
821static inline u32 sctp_hash_obj(const void *data, u32 len, u32 seed)
822{
823 const struct sctp_transport *t = data;
824 const union sctp_addr *paddr = &t->ipaddr;
825 const struct net *net = sock_net(t->asoc->base.sk);
826 u16 lport = htons(t->asoc->base.bind_addr.port);
827 u32 addr;
828
829 if (paddr->sa.sa_family == AF_INET6)
830 addr = jhash(&paddr->v6.sin6_addr, 16, seed);
831 else
832 addr = paddr->v4.sin_addr.s_addr;
833
834 return jhash_3words(addr, ((__u32)paddr->v4.sin_port) << 16 |
835 (__force __u32)lport, net_hash_mix(net), seed);
836}
837
838static inline u32 sctp_hash_key(const void *data, u32 len, u32 seed)
839{
840 const struct sctp_hash_cmp_arg *x = data;
841 const union sctp_addr *paddr = x->paddr;
842 const struct net *net = x->net;
843 u16 lport = x->lport;
844 u32 addr;
845
846 if (paddr->sa.sa_family == AF_INET6)
847 addr = jhash(&paddr->v6.sin6_addr, 16, seed);
848 else
849 addr = paddr->v4.sin_addr.s_addr;
850
851 return jhash_3words(addr, ((__u32)paddr->v4.sin_port) << 16 |
852 (__force __u32)lport, net_hash_mix(net), seed);
853}
854
855static const struct rhashtable_params sctp_hash_params = {
856 .head_offset = offsetof(struct sctp_transport, node),
857 .hashfn = sctp_hash_key,
858 .obj_hashfn = sctp_hash_obj,
859 .obj_cmpfn = sctp_hash_cmp,
860 .automatic_shrinking = true,
861};
862
863int sctp_transport_hashtable_init(void)
864{
865 return rhltable_init(&sctp_transport_hashtable, &sctp_hash_params);
866}
867
868void sctp_transport_hashtable_destroy(void)
869{
870 rhltable_destroy(&sctp_transport_hashtable);
871}
872
873int sctp_hash_transport(struct sctp_transport *t)
874{
875 struct sctp_hash_cmp_arg arg;
876 int err;
877
878 if (t->asoc->temp)
879 return 0;
880
881 arg.net = sock_net(t->asoc->base.sk);
882 arg.paddr = &t->ipaddr;
883 arg.lport = htons(t->asoc->base.bind_addr.port);
884
885 err = rhltable_insert_key(&sctp_transport_hashtable, &arg,
886 &t->node, sctp_hash_params);
887 if (err)
888 pr_err_once("insert transport fail, errno %d\n", err);
889
890 return err;
891}
892
893void sctp_unhash_transport(struct sctp_transport *t)
894{
895 if (t->asoc->temp)
896 return;
897
898 rhltable_remove(&sctp_transport_hashtable, &t->node,
899 sctp_hash_params);
900}
901
902/* return a transport with holding it */
903struct sctp_transport *sctp_addrs_lookup_transport(
904 struct net *net,
905 const union sctp_addr *laddr,
906 const union sctp_addr *paddr)
907{
908 struct rhlist_head *tmp, *list;
909 struct sctp_transport *t;
910 struct sctp_hash_cmp_arg arg = {
911 .paddr = paddr,
912 .net = net,
913 .lport = laddr->v4.sin_port,
914 };
915
916 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
917 sctp_hash_params);
918
919 rhl_for_each_entry_rcu(t, tmp, list, node) {
920 if (!sctp_transport_hold(t))
921 continue;
922
923 if (sctp_bind_addr_match(&t->asoc->base.bind_addr,
924 laddr, sctp_sk(t->asoc->base.sk)))
925 return t;
926 sctp_transport_put(t);
927 }
928
929 return NULL;
930}
931
932/* return a transport without holding it, as it's only used under sock lock */
933struct sctp_transport *sctp_epaddr_lookup_transport(
934 const struct sctp_endpoint *ep,
935 const union sctp_addr *paddr)
936{
937 struct net *net = sock_net(ep->base.sk);
938 struct rhlist_head *tmp, *list;
939 struct sctp_transport *t;
940 struct sctp_hash_cmp_arg arg = {
941 .paddr = paddr,
942 .net = net,
943 .lport = htons(ep->base.bind_addr.port),
944 };
945
946 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
947 sctp_hash_params);
948
949 rhl_for_each_entry_rcu(t, tmp, list, node)
950 if (ep == t->asoc->ep)
951 return t;
952
953 return NULL;
954}
955
956/* Look up an association. */
957static struct sctp_association *__sctp_lookup_association(
958 struct net *net,
959 const union sctp_addr *local,
960 const union sctp_addr *peer,
961 struct sctp_transport **pt)
962{
963 struct sctp_transport *t;
964 struct sctp_association *asoc = NULL;
965
966 t = sctp_addrs_lookup_transport(net, local, peer);
967 if (!t)
968 goto out;
969
970 asoc = t->asoc;
971 *pt = t;
972
973out:
974 return asoc;
975}
976
977/* Look up an association. protected by RCU read lock */
978static
979struct sctp_association *sctp_lookup_association(struct net *net,
980 const union sctp_addr *laddr,
981 const union sctp_addr *paddr,
982 struct sctp_transport **transportp)
983{
984 struct sctp_association *asoc;
985
986 rcu_read_lock();
987 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
988 rcu_read_unlock();
989
990 return asoc;
991}
992
993/* Is there an association matching the given local and peer addresses? */
994int sctp_has_association(struct net *net,
995 const union sctp_addr *laddr,
996 const union sctp_addr *paddr)
997{
998 struct sctp_association *asoc;
999 struct sctp_transport *transport;
1000
1001 if ((asoc = sctp_lookup_association(net, laddr, paddr, &transport))) {
1002 sctp_transport_put(transport);
1003 return 1;
1004 }
1005
1006 return 0;
1007}
1008
1009/*
1010 * SCTP Implementors Guide, 2.18 Handling of address
1011 * parameters within the INIT or INIT-ACK.
1012 *
1013 * D) When searching for a matching TCB upon reception of an INIT
1014 * or INIT-ACK chunk the receiver SHOULD use not only the
1015 * source address of the packet (containing the INIT or
1016 * INIT-ACK) but the receiver SHOULD also use all valid
1017 * address parameters contained within the chunk.
1018 *
1019 * 2.18.3 Solution description
1020 *
1021 * This new text clearly specifies to an implementor the need
1022 * to look within the INIT or INIT-ACK. Any implementation that
1023 * does not do this, may not be able to establish associations
1024 * in certain circumstances.
1025 *
1026 */
1027static struct sctp_association *__sctp_rcv_init_lookup(struct net *net,
1028 struct sk_buff *skb,
1029 const union sctp_addr *laddr, struct sctp_transport **transportp)
1030{
1031 struct sctp_association *asoc;
1032 union sctp_addr addr;
1033 union sctp_addr *paddr = &addr;
1034 struct sctphdr *sh = sctp_hdr(skb);
1035 union sctp_params params;
1036 sctp_init_chunk_t *init;
1037 struct sctp_af *af;
1038
1039 /*
1040 * This code will NOT touch anything inside the chunk--it is
1041 * strictly READ-ONLY.
1042 *
1043 * RFC 2960 3 SCTP packet Format
1044 *
1045 * Multiple chunks can be bundled into one SCTP packet up to
1046 * the MTU size, except for the INIT, INIT ACK, and SHUTDOWN
1047 * COMPLETE chunks. These chunks MUST NOT be bundled with any
1048 * other chunk in a packet. See Section 6.10 for more details
1049 * on chunk bundling.
1050 */
1051
1052 /* Find the start of the TLVs and the end of the chunk. This is
1053 * the region we search for address parameters.
1054 */
1055 init = (sctp_init_chunk_t *)skb->data;
1056
1057 /* Walk the parameters looking for embedded addresses. */
1058 sctp_walk_params(params, init, init_hdr.params) {
1059
1060 /* Note: Ignoring hostname addresses. */
1061 af = sctp_get_af_specific(param_type2af(params.p->type));
1062 if (!af)
1063 continue;
1064
1065 af->from_addr_param(paddr, params.addr, sh->source, 0);
1066
1067 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1068 if (asoc)
1069 return asoc;
1070 }
1071
1072 return NULL;
1073}
1074
1075/* ADD-IP, Section 5.2
1076 * When an endpoint receives an ASCONF Chunk from the remote peer
1077 * special procedures may be needed to identify the association the
1078 * ASCONF Chunk is associated with. To properly find the association
1079 * the following procedures SHOULD be followed:
1080 *
1081 * D2) If the association is not found, use the address found in the
1082 * Address Parameter TLV combined with the port number found in the
1083 * SCTP common header. If found proceed to rule D4.
1084 *
1085 * D2-ext) If more than one ASCONF Chunks are packed together, use the
1086 * address found in the ASCONF Address Parameter TLV of each of the
1087 * subsequent ASCONF Chunks. If found, proceed to rule D4.
1088 */
1089static struct sctp_association *__sctp_rcv_asconf_lookup(
1090 struct net *net,
1091 sctp_chunkhdr_t *ch,
1092 const union sctp_addr *laddr,
1093 __be16 peer_port,
1094 struct sctp_transport **transportp)
1095{
1096 sctp_addip_chunk_t *asconf = (struct sctp_addip_chunk *)ch;
1097 struct sctp_af *af;
1098 union sctp_addr_param *param;
1099 union sctp_addr paddr;
1100
1101 /* Skip over the ADDIP header and find the Address parameter */
1102 param = (union sctp_addr_param *)(asconf + 1);
1103
1104 af = sctp_get_af_specific(param_type2af(param->p.type));
1105 if (unlikely(!af))
1106 return NULL;
1107
1108 af->from_addr_param(&paddr, param, peer_port, 0);
1109
1110 return __sctp_lookup_association(net, laddr, &paddr, transportp);
1111}
1112
1113
1114/* SCTP-AUTH, Section 6.3:
1115* If the receiver does not find a STCB for a packet containing an AUTH
1116* chunk as the first chunk and not a COOKIE-ECHO chunk as the second
1117* chunk, it MUST use the chunks after the AUTH chunk to look up an existing
1118* association.
1119*
1120* This means that any chunks that can help us identify the association need
1121* to be looked at to find this association.
1122*/
1123static struct sctp_association *__sctp_rcv_walk_lookup(struct net *net,
1124 struct sk_buff *skb,
1125 const union sctp_addr *laddr,
1126 struct sctp_transport **transportp)
1127{
1128 struct sctp_association *asoc = NULL;
1129 sctp_chunkhdr_t *ch;
1130 int have_auth = 0;
1131 unsigned int chunk_num = 1;
1132 __u8 *ch_end;
1133
1134 /* Walk through the chunks looking for AUTH or ASCONF chunks
1135 * to help us find the association.
1136 */
1137 ch = (sctp_chunkhdr_t *) skb->data;
1138 do {
1139 /* Break out if chunk length is less then minimal. */
1140 if (ntohs(ch->length) < sizeof(sctp_chunkhdr_t))
1141 break;
1142
1143 ch_end = ((__u8 *)ch) + SCTP_PAD4(ntohs(ch->length));
1144 if (ch_end > skb_tail_pointer(skb))
1145 break;
1146
1147 switch (ch->type) {
1148 case SCTP_CID_AUTH:
1149 have_auth = chunk_num;
1150 break;
1151
1152 case SCTP_CID_COOKIE_ECHO:
1153 /* If a packet arrives containing an AUTH chunk as
1154 * a first chunk, a COOKIE-ECHO chunk as the second
1155 * chunk, and possibly more chunks after them, and
1156 * the receiver does not have an STCB for that
1157 * packet, then authentication is based on
1158 * the contents of the COOKIE- ECHO chunk.
1159 */
1160 if (have_auth == 1 && chunk_num == 2)
1161 return NULL;
1162 break;
1163
1164 case SCTP_CID_ASCONF:
1165 if (have_auth || net->sctp.addip_noauth)
1166 asoc = __sctp_rcv_asconf_lookup(
1167 net, ch, laddr,
1168 sctp_hdr(skb)->source,
1169 transportp);
1170 default:
1171 break;
1172 }
1173
1174 if (asoc)
1175 break;
1176
1177 ch = (sctp_chunkhdr_t *) ch_end;
1178 chunk_num++;
1179 } while (ch_end < skb_tail_pointer(skb));
1180
1181 return asoc;
1182}
1183
1184/*
1185 * There are circumstances when we need to look inside the SCTP packet
1186 * for information to help us find the association. Examples
1187 * include looking inside of INIT/INIT-ACK chunks or after the AUTH
1188 * chunks.
1189 */
1190static struct sctp_association *__sctp_rcv_lookup_harder(struct net *net,
1191 struct sk_buff *skb,
1192 const union sctp_addr *laddr,
1193 struct sctp_transport **transportp)
1194{
1195 sctp_chunkhdr_t *ch;
1196
1197 /* We do not allow GSO frames here as we need to linearize and
1198 * then cannot guarantee frame boundaries. This shouldn't be an
1199 * issue as packets hitting this are mostly INIT or INIT-ACK and
1200 * those cannot be on GSO-style anyway.
1201 */
1202 if ((skb_shinfo(skb)->gso_type & SKB_GSO_SCTP) == SKB_GSO_SCTP)
1203 return NULL;
1204
1205 ch = (sctp_chunkhdr_t *) skb->data;
1206
1207 /* The code below will attempt to walk the chunk and extract
1208 * parameter information. Before we do that, we need to verify
1209 * that the chunk length doesn't cause overflow. Otherwise, we'll
1210 * walk off the end.
1211 */
1212 if (SCTP_PAD4(ntohs(ch->length)) > skb->len)
1213 return NULL;
1214
1215 /* If this is INIT/INIT-ACK look inside the chunk too. */
1216 if (ch->type == SCTP_CID_INIT || ch->type == SCTP_CID_INIT_ACK)
1217 return __sctp_rcv_init_lookup(net, skb, laddr, transportp);
1218
1219 return __sctp_rcv_walk_lookup(net, skb, laddr, transportp);
1220}
1221
1222/* Lookup an association for an inbound skb. */
1223static struct sctp_association *__sctp_rcv_lookup(struct net *net,
1224 struct sk_buff *skb,
1225 const union sctp_addr *paddr,
1226 const union sctp_addr *laddr,
1227 struct sctp_transport **transportp)
1228{
1229 struct sctp_association *asoc;
1230
1231 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1232
1233 /* Further lookup for INIT/INIT-ACK packets.
1234 * SCTP Implementors Guide, 2.18 Handling of address
1235 * parameters within the INIT or INIT-ACK.
1236 */
1237 if (!asoc)
1238 asoc = __sctp_rcv_lookup_harder(net, skb, laddr, transportp);
1239
1240 return asoc;
1241}