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,
54 int dif, int sdif);
55static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(
56 struct net *net, struct sk_buff *skb,
57 const union sctp_addr *laddr,
58 const union sctp_addr *daddr,
59 int dif, int sdif);
60static struct sctp_association *__sctp_lookup_association(
61 struct net *net,
62 const union sctp_addr *local,
63 const union sctp_addr *peer,
64 struct sctp_transport **pt,
65 int dif, int sdif);
66
67static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb);
68
69
70/* Calculate the SCTP checksum of an SCTP packet. */
71static inline int sctp_rcv_checksum(struct net *net, struct sk_buff *skb)
72{
73 struct sctphdr *sh = sctp_hdr(skb);
74 __le32 cmp = sh->checksum;
75 __le32 val = sctp_compute_cksum(skb, 0);
76
77 if (val != cmp) {
78 /* CRC failure, dump it. */
79 __SCTP_INC_STATS(net, SCTP_MIB_CHECKSUMERRORS);
80 return -1;
81 }
82 return 0;
83}
84
85/*
86 * This is the routine which IP calls when receiving an SCTP packet.
87 */
88int sctp_rcv(struct sk_buff *skb)
89{
90 struct sock *sk;
91 struct sctp_association *asoc;
92 struct sctp_endpoint *ep = NULL;
93 struct sctp_ep_common *rcvr;
94 struct sctp_transport *transport = NULL;
95 struct sctp_chunk *chunk;
96 union sctp_addr src;
97 union sctp_addr dest;
98 int family;
99 struct sctp_af *af;
100 struct net *net = dev_net(skb->dev);
101 bool is_gso = skb_is_gso(skb) && skb_is_gso_sctp(skb);
102 int dif, sdif;
103
104 if (skb->pkt_type != PACKET_HOST)
105 goto discard_it;
106
107 __SCTP_INC_STATS(net, SCTP_MIB_INSCTPPACKS);
108
109 /* If packet is too small to contain a single chunk, let's not
110 * waste time on it anymore.
111 */
112 if (skb->len < sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) +
113 skb_transport_offset(skb))
114 goto discard_it;
115
116 /* If the packet is fragmented and we need to do crc checking,
117 * it's better to just linearize it otherwise crc computing
118 * takes longer.
119 */
120 if ((!is_gso && skb_linearize(skb)) ||
121 !pskb_may_pull(skb, sizeof(struct sctphdr)))
122 goto discard_it;
123
124 /* Pull up the IP header. */
125 __skb_pull(skb, skb_transport_offset(skb));
126
127 skb->csum_valid = 0; /* Previous value not applicable */
128 if (skb_csum_unnecessary(skb))
129 __skb_decr_checksum_unnecessary(skb);
130 else if (!sctp_checksum_disable &&
131 !is_gso &&
132 sctp_rcv_checksum(net, skb) < 0)
133 goto discard_it;
134 skb->csum_valid = 1;
135
136 __skb_pull(skb, sizeof(struct sctphdr));
137
138 family = ipver2af(ip_hdr(skb)->version);
139 af = sctp_get_af_specific(family);
140 if (unlikely(!af))
141 goto discard_it;
142 SCTP_INPUT_CB(skb)->af = af;
143
144 /* Initialize local addresses for lookups. */
145 af->from_skb(&src, skb, 1);
146 af->from_skb(&dest, skb, 0);
147 dif = af->skb_iif(skb);
148 sdif = af->skb_sdif(skb);
149
150 /* If the packet is to or from a non-unicast address,
151 * silently discard the packet.
152 *
153 * This is not clearly defined in the RFC except in section
154 * 8.4 - OOTB handling. However, based on the book "Stream Control
155 * Transmission Protocol" 2.1, "It is important to note that the
156 * IP address of an SCTP transport address must be a routable
157 * unicast address. In other words, IP multicast addresses and
158 * IP broadcast addresses cannot be used in an SCTP transport
159 * address."
160 */
161 if (!af->addr_valid(&src, NULL, skb) ||
162 !af->addr_valid(&dest, NULL, skb))
163 goto discard_it;
164
165 asoc = __sctp_rcv_lookup(net, skb, &src, &dest, &transport, dif, sdif);
166
167 if (!asoc)
168 ep = __sctp_rcv_lookup_endpoint(net, skb, &dest, &src, dif, sdif);
169
170 /* Retrieve the common input handling substructure. */
171 rcvr = asoc ? &asoc->base : &ep->base;
172 sk = rcvr->sk;
173
174 /*
175 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
176 * An SCTP packet is called an "out of the blue" (OOTB)
177 * packet if it is correctly formed, i.e., passed the
178 * receiver's checksum check, but the receiver is not
179 * able to identify the association to which this
180 * packet belongs.
181 */
182 if (!asoc) {
183 if (sctp_rcv_ootb(skb)) {
184 __SCTP_INC_STATS(net, SCTP_MIB_OUTOFBLUES);
185 goto discard_release;
186 }
187 }
188
189 if (!xfrm_policy_check(sk, XFRM_POLICY_IN, skb, family))
190 goto discard_release;
191 nf_reset_ct(skb);
192
193 if (sk_filter(sk, skb))
194 goto discard_release;
195
196 /* Create an SCTP packet structure. */
197 chunk = sctp_chunkify(skb, asoc, sk, GFP_ATOMIC);
198 if (!chunk)
199 goto discard_release;
200 SCTP_INPUT_CB(skb)->chunk = chunk;
201
202 /* Remember what endpoint is to handle this packet. */
203 chunk->rcvr = rcvr;
204
205 /* Remember the SCTP header. */
206 chunk->sctp_hdr = sctp_hdr(skb);
207
208 /* Set the source and destination addresses of the incoming chunk. */
209 sctp_init_addrs(chunk, &src, &dest);
210
211 /* Remember where we came from. */
212 chunk->transport = transport;
213
214 /* Acquire access to the sock lock. Note: We are safe from other
215 * bottom halves on this lock, but a user may be in the lock too,
216 * so check if it is busy.
217 */
218 bh_lock_sock(sk);
219
220 if (sk != rcvr->sk) {
221 /* Our cached sk is different from the rcvr->sk. This is
222 * because migrate()/accept() may have moved the association
223 * to a new socket and released all the sockets. So now we
224 * are holding a lock on the old socket while the user may
225 * be doing something with the new socket. Switch our veiw
226 * of the current sk.
227 */
228 bh_unlock_sock(sk);
229 sk = rcvr->sk;
230 bh_lock_sock(sk);
231 }
232
233 if (sock_owned_by_user(sk) || !sctp_newsk_ready(sk)) {
234 if (sctp_add_backlog(sk, skb)) {
235 bh_unlock_sock(sk);
236 sctp_chunk_free(chunk);
237 skb = NULL; /* sctp_chunk_free already freed the skb */
238 goto discard_release;
239 }
240 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_BACKLOG);
241 } else {
242 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_SOFTIRQ);
243 sctp_inq_push(&chunk->rcvr->inqueue, chunk);
244 }
245
246 bh_unlock_sock(sk);
247
248 /* Release the asoc/ep ref we took in the lookup calls. */
249 if (transport)
250 sctp_transport_put(transport);
251 else
252 sctp_endpoint_put(ep);
253
254 return 0;
255
256discard_it:
257 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_DISCARDS);
258 kfree_skb(skb);
259 return 0;
260
261discard_release:
262 /* Release the asoc/ep ref we took in the lookup calls. */
263 if (transport)
264 sctp_transport_put(transport);
265 else
266 sctp_endpoint_put(ep);
267
268 goto discard_it;
269}
270
271/* Process the backlog queue of the socket. Every skb on
272 * the backlog holds a ref on an association or endpoint.
273 * We hold this ref throughout the state machine to make
274 * sure that the structure we need is still around.
275 */
276int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb)
277{
278 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
279 struct sctp_inq *inqueue = &chunk->rcvr->inqueue;
280 struct sctp_transport *t = chunk->transport;
281 struct sctp_ep_common *rcvr = NULL;
282 int backloged = 0;
283
284 rcvr = chunk->rcvr;
285
286 /* If the rcvr is dead then the association or endpoint
287 * has been deleted and we can safely drop the chunk
288 * and refs that we are holding.
289 */
290 if (rcvr->dead) {
291 sctp_chunk_free(chunk);
292 goto done;
293 }
294
295 if (unlikely(rcvr->sk != sk)) {
296 /* In this case, the association moved from one socket to
297 * another. We are currently sitting on the backlog of the
298 * old socket, so we need to move.
299 * However, since we are here in the process context we
300 * need to take make sure that the user doesn't own
301 * the new socket when we process the packet.
302 * If the new socket is user-owned, queue the chunk to the
303 * backlog of the new socket without dropping any refs.
304 * Otherwise, we can safely push the chunk on the inqueue.
305 */
306
307 sk = rcvr->sk;
308 local_bh_disable();
309 bh_lock_sock(sk);
310
311 if (sock_owned_by_user(sk) || !sctp_newsk_ready(sk)) {
312 if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))
313 sctp_chunk_free(chunk);
314 else
315 backloged = 1;
316 } else
317 sctp_inq_push(inqueue, chunk);
318
319 bh_unlock_sock(sk);
320 local_bh_enable();
321
322 /* If the chunk was backloged again, don't drop refs */
323 if (backloged)
324 return 0;
325 } else {
326 if (!sctp_newsk_ready(sk)) {
327 if (!sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))
328 return 0;
329 sctp_chunk_free(chunk);
330 } else {
331 sctp_inq_push(inqueue, chunk);
332 }
333 }
334
335done:
336 /* Release the refs we took in sctp_add_backlog */
337 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
338 sctp_transport_put(t);
339 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
340 sctp_endpoint_put(sctp_ep(rcvr));
341 else
342 BUG();
343
344 return 0;
345}
346
347static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb)
348{
349 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
350 struct sctp_transport *t = chunk->transport;
351 struct sctp_ep_common *rcvr = chunk->rcvr;
352 int ret;
353
354 ret = sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf));
355 if (!ret) {
356 /* Hold the assoc/ep while hanging on the backlog queue.
357 * This way, we know structures we need will not disappear
358 * from us
359 */
360 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
361 sctp_transport_hold(t);
362 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
363 sctp_endpoint_hold(sctp_ep(rcvr));
364 else
365 BUG();
366 }
367 return ret;
368
369}
370
371/* Handle icmp frag needed error. */
372void sctp_icmp_frag_needed(struct sock *sk, struct sctp_association *asoc,
373 struct sctp_transport *t, __u32 pmtu)
374{
375 if (!t ||
376 (t->pathmtu <= pmtu &&
377 t->pl.probe_size + sctp_transport_pl_hlen(t) <= pmtu))
378 return;
379
380 if (sock_owned_by_user(sk)) {
381 atomic_set(&t->mtu_info, pmtu);
382 asoc->pmtu_pending = 1;
383 t->pmtu_pending = 1;
384 return;
385 }
386
387 if (!(t->param_flags & SPP_PMTUD_ENABLE))
388 /* We can't allow retransmitting in such case, as the
389 * retransmission would be sized just as before, and thus we
390 * would get another icmp, and retransmit again.
391 */
392 return;
393
394 /* Update transports view of the MTU. Return if no update was needed.
395 * If an update wasn't needed/possible, it also doesn't make sense to
396 * try to retransmit now.
397 */
398 if (!sctp_transport_update_pmtu(t, pmtu))
399 return;
400
401 /* Update association pmtu. */
402 sctp_assoc_sync_pmtu(asoc);
403
404 /* Retransmit with the new pmtu setting. */
405 sctp_retransmit(&asoc->outqueue, t, SCTP_RTXR_PMTUD);
406}
407
408void sctp_icmp_redirect(struct sock *sk, struct sctp_transport *t,
409 struct sk_buff *skb)
410{
411 struct dst_entry *dst;
412
413 if (sock_owned_by_user(sk) || !t)
414 return;
415 dst = sctp_transport_dst_check(t);
416 if (dst)
417 dst->ops->redirect(dst, sk, skb);
418}
419
420/*
421 * SCTP Implementer's Guide, 2.37 ICMP handling procedures
422 *
423 * ICMP8) If the ICMP code is a "Unrecognized next header type encountered"
424 * or a "Protocol Unreachable" treat this message as an abort
425 * with the T bit set.
426 *
427 * This function sends an event to the state machine, which will abort the
428 * association.
429 *
430 */
431void sctp_icmp_proto_unreachable(struct sock *sk,
432 struct sctp_association *asoc,
433 struct sctp_transport *t)
434{
435 if (sock_owned_by_user(sk)) {
436 if (timer_pending(&t->proto_unreach_timer))
437 return;
438 else {
439 if (!mod_timer(&t->proto_unreach_timer,
440 jiffies + (HZ/20)))
441 sctp_transport_hold(t);
442 }
443 } else {
444 struct net *net = sock_net(sk);
445
446 pr_debug("%s: unrecognized next header type "
447 "encountered!\n", __func__);
448
449 if (del_timer(&t->proto_unreach_timer))
450 sctp_transport_put(t);
451
452 sctp_do_sm(net, SCTP_EVENT_T_OTHER,
453 SCTP_ST_OTHER(SCTP_EVENT_ICMP_PROTO_UNREACH),
454 asoc->state, asoc->ep, asoc, t,
455 GFP_ATOMIC);
456 }
457}
458
459/* Common lookup code for icmp/icmpv6 error handler. */
460struct sock *sctp_err_lookup(struct net *net, int family, struct sk_buff *skb,
461 struct sctphdr *sctphdr,
462 struct sctp_association **app,
463 struct sctp_transport **tpp)
464{
465 struct sctp_init_chunk *chunkhdr, _chunkhdr;
466 union sctp_addr saddr;
467 union sctp_addr daddr;
468 struct sctp_af *af;
469 struct sock *sk = NULL;
470 struct sctp_association *asoc;
471 struct sctp_transport *transport = NULL;
472 __u32 vtag = ntohl(sctphdr->vtag);
473 int sdif = inet_sdif(skb);
474 int dif = inet_iif(skb);
475
476 *app = NULL; *tpp = NULL;
477
478 af = sctp_get_af_specific(family);
479 if (unlikely(!af)) {
480 return NULL;
481 }
482
483 /* Initialize local addresses for lookups. */
484 af->from_skb(&saddr, skb, 1);
485 af->from_skb(&daddr, skb, 0);
486
487 /* Look for an association that matches the incoming ICMP error
488 * packet.
489 */
490 asoc = __sctp_lookup_association(net, &saddr, &daddr, &transport, dif, sdif);
491 if (!asoc)
492 return NULL;
493
494 sk = asoc->base.sk;
495
496 /* RFC 4960, Appendix C. ICMP Handling
497 *
498 * ICMP6) An implementation MUST validate that the Verification Tag
499 * contained in the ICMP message matches the Verification Tag of
500 * the peer. If the Verification Tag is not 0 and does NOT
501 * match, discard the ICMP message. If it is 0 and the ICMP
502 * message contains enough bytes to verify that the chunk type is
503 * an INIT chunk and that the Initiate Tag matches the tag of the
504 * peer, continue with ICMP7. If the ICMP message is too short
505 * or the chunk type or the Initiate Tag does not match, silently
506 * discard the packet.
507 */
508 if (vtag == 0) {
509 /* chunk header + first 4 octects of init header */
510 chunkhdr = skb_header_pointer(skb, skb_transport_offset(skb) +
511 sizeof(struct sctphdr),
512 sizeof(struct sctp_chunkhdr) +
513 sizeof(__be32), &_chunkhdr);
514 if (!chunkhdr ||
515 chunkhdr->chunk_hdr.type != SCTP_CID_INIT ||
516 ntohl(chunkhdr->init_hdr.init_tag) != asoc->c.my_vtag)
517 goto out;
518
519 } else if (vtag != asoc->c.peer_vtag) {
520 goto out;
521 }
522
523 bh_lock_sock(sk);
524
525 /* If too many ICMPs get dropped on busy
526 * servers this needs to be solved differently.
527 */
528 if (sock_owned_by_user(sk))
529 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
530
531 *app = asoc;
532 *tpp = transport;
533 return sk;
534
535out:
536 sctp_transport_put(transport);
537 return NULL;
538}
539
540/* Common cleanup code for icmp/icmpv6 error handler. */
541void sctp_err_finish(struct sock *sk, struct sctp_transport *t)
542 __releases(&((__sk)->sk_lock.slock))
543{
544 bh_unlock_sock(sk);
545 sctp_transport_put(t);
546}
547
548static void sctp_v4_err_handle(struct sctp_transport *t, struct sk_buff *skb,
549 __u8 type, __u8 code, __u32 info)
550{
551 struct sctp_association *asoc = t->asoc;
552 struct sock *sk = asoc->base.sk;
553 int err = 0;
554
555 switch (type) {
556 case ICMP_PARAMETERPROB:
557 err = EPROTO;
558 break;
559 case ICMP_DEST_UNREACH:
560 if (code > NR_ICMP_UNREACH)
561 return;
562 if (code == ICMP_FRAG_NEEDED) {
563 sctp_icmp_frag_needed(sk, asoc, t, SCTP_TRUNC4(info));
564 return;
565 }
566 if (code == ICMP_PROT_UNREACH) {
567 sctp_icmp_proto_unreachable(sk, asoc, t);
568 return;
569 }
570 err = icmp_err_convert[code].errno;
571 break;
572 case ICMP_TIME_EXCEEDED:
573 if (code == ICMP_EXC_FRAGTIME)
574 return;
575
576 err = EHOSTUNREACH;
577 break;
578 case ICMP_REDIRECT:
579 sctp_icmp_redirect(sk, t, skb);
580 return;
581 default:
582 return;
583 }
584 if (!sock_owned_by_user(sk) && inet_test_bit(RECVERR, sk)) {
585 sk->sk_err = err;
586 sk_error_report(sk);
587 } else { /* Only an error on timeout */
588 WRITE_ONCE(sk->sk_err_soft, err);
589 }
590}
591
592/*
593 * This routine is called by the ICMP module when it gets some
594 * sort of error condition. If err < 0 then the socket should
595 * be closed and the error returned to the user. If err > 0
596 * it's just the icmp type << 8 | icmp code. After adjustment
597 * header points to the first 8 bytes of the sctp header. We need
598 * to find the appropriate port.
599 *
600 * The locking strategy used here is very "optimistic". When
601 * someone else accesses the socket the ICMP is just dropped
602 * and for some paths there is no check at all.
603 * A more general error queue to queue errors for later handling
604 * is probably better.
605 *
606 */
607int sctp_v4_err(struct sk_buff *skb, __u32 info)
608{
609 const struct iphdr *iph = (const struct iphdr *)skb->data;
610 const int type = icmp_hdr(skb)->type;
611 const int code = icmp_hdr(skb)->code;
612 struct net *net = dev_net(skb->dev);
613 struct sctp_transport *transport;
614 struct sctp_association *asoc;
615 __u16 saveip, savesctp;
616 struct sock *sk;
617
618 /* Fix up skb to look at the embedded net header. */
619 saveip = skb->network_header;
620 savesctp = skb->transport_header;
621 skb_reset_network_header(skb);
622 skb_set_transport_header(skb, iph->ihl * 4);
623 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &transport);
624 /* Put back, the original values. */
625 skb->network_header = saveip;
626 skb->transport_header = savesctp;
627 if (!sk) {
628 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
629 return -ENOENT;
630 }
631
632 sctp_v4_err_handle(transport, skb, type, code, info);
633 sctp_err_finish(sk, transport);
634
635 return 0;
636}
637
638int sctp_udp_v4_err(struct sock *sk, struct sk_buff *skb)
639{
640 struct net *net = dev_net(skb->dev);
641 struct sctp_association *asoc;
642 struct sctp_transport *t;
643 struct icmphdr *hdr;
644 __u32 info = 0;
645
646 skb->transport_header += sizeof(struct udphdr);
647 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &t);
648 if (!sk) {
649 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
650 return -ENOENT;
651 }
652
653 skb->transport_header -= sizeof(struct udphdr);
654 hdr = (struct icmphdr *)(skb_network_header(skb) - sizeof(struct icmphdr));
655 if (hdr->type == ICMP_REDIRECT) {
656 /* can't be handled without outer iphdr known, leave it to udp_err */
657 sctp_err_finish(sk, t);
658 return 0;
659 }
660 if (hdr->type == ICMP_DEST_UNREACH && hdr->code == ICMP_FRAG_NEEDED)
661 info = ntohs(hdr->un.frag.mtu);
662 sctp_v4_err_handle(t, skb, hdr->type, hdr->code, info);
663
664 sctp_err_finish(sk, t);
665 return 1;
666}
667
668/*
669 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
670 *
671 * This function scans all the chunks in the OOTB packet to determine if
672 * the packet should be discarded right away. If a response might be needed
673 * for this packet, or, if further processing is possible, the packet will
674 * be queued to a proper inqueue for the next phase of handling.
675 *
676 * Output:
677 * Return 0 - If further processing is needed.
678 * Return 1 - If the packet can be discarded right away.
679 */
680static int sctp_rcv_ootb(struct sk_buff *skb)
681{
682 struct sctp_chunkhdr *ch, _ch;
683 int ch_end, offset = 0;
684
685 /* Scan through all the chunks in the packet. */
686 do {
687 /* Make sure we have at least the header there */
688 if (offset + sizeof(_ch) > skb->len)
689 break;
690
691 ch = skb_header_pointer(skb, offset, sizeof(*ch), &_ch);
692
693 /* Break out if chunk length is less then minimal. */
694 if (!ch || ntohs(ch->length) < sizeof(_ch))
695 break;
696
697 ch_end = offset + SCTP_PAD4(ntohs(ch->length));
698 if (ch_end > skb->len)
699 break;
700
701 /* RFC 8.4, 2) If the OOTB packet contains an ABORT chunk, the
702 * receiver MUST silently discard the OOTB packet and take no
703 * further action.
704 */
705 if (SCTP_CID_ABORT == ch->type)
706 goto discard;
707
708 /* RFC 8.4, 6) If the packet contains a SHUTDOWN COMPLETE
709 * chunk, the receiver should silently discard the packet
710 * and take no further action.
711 */
712 if (SCTP_CID_SHUTDOWN_COMPLETE == ch->type)
713 goto discard;
714
715 /* RFC 4460, 2.11.2
716 * This will discard packets with INIT chunk bundled as
717 * subsequent chunks in the packet. When INIT is first,
718 * the normal INIT processing will discard the chunk.
719 */
720 if (SCTP_CID_INIT == ch->type && (void *)ch != skb->data)
721 goto discard;
722
723 offset = ch_end;
724 } while (ch_end < skb->len);
725
726 return 0;
727
728discard:
729 return 1;
730}
731
732/* Insert endpoint into the hash table. */
733static int __sctp_hash_endpoint(struct sctp_endpoint *ep)
734{
735 struct sock *sk = ep->base.sk;
736 struct net *net = sock_net(sk);
737 struct sctp_hashbucket *head;
738
739 ep->hashent = sctp_ep_hashfn(net, ep->base.bind_addr.port);
740 head = &sctp_ep_hashtable[ep->hashent];
741
742 if (sk->sk_reuseport) {
743 bool any = sctp_is_ep_boundall(sk);
744 struct sctp_endpoint *ep2;
745 struct list_head *list;
746 int cnt = 0, err = 1;
747
748 list_for_each(list, &ep->base.bind_addr.address_list)
749 cnt++;
750
751 sctp_for_each_hentry(ep2, &head->chain) {
752 struct sock *sk2 = ep2->base.sk;
753
754 if (!net_eq(sock_net(sk2), net) || sk2 == sk ||
755 !uid_eq(sock_i_uid(sk2), sock_i_uid(sk)) ||
756 !sk2->sk_reuseport)
757 continue;
758
759 err = sctp_bind_addrs_check(sctp_sk(sk2),
760 sctp_sk(sk), cnt);
761 if (!err) {
762 err = reuseport_add_sock(sk, sk2, any);
763 if (err)
764 return err;
765 break;
766 } else if (err < 0) {
767 return err;
768 }
769 }
770
771 if (err) {
772 err = reuseport_alloc(sk, any);
773 if (err)
774 return err;
775 }
776 }
777
778 write_lock(&head->lock);
779 hlist_add_head(&ep->node, &head->chain);
780 write_unlock(&head->lock);
781 return 0;
782}
783
784/* Add an endpoint to the hash. Local BH-safe. */
785int sctp_hash_endpoint(struct sctp_endpoint *ep)
786{
787 int err;
788
789 local_bh_disable();
790 err = __sctp_hash_endpoint(ep);
791 local_bh_enable();
792
793 return err;
794}
795
796/* Remove endpoint from the hash table. */
797static void __sctp_unhash_endpoint(struct sctp_endpoint *ep)
798{
799 struct sock *sk = ep->base.sk;
800 struct sctp_hashbucket *head;
801
802 ep->hashent = sctp_ep_hashfn(sock_net(sk), ep->base.bind_addr.port);
803
804 head = &sctp_ep_hashtable[ep->hashent];
805
806 if (rcu_access_pointer(sk->sk_reuseport_cb))
807 reuseport_detach_sock(sk);
808
809 write_lock(&head->lock);
810 hlist_del_init(&ep->node);
811 write_unlock(&head->lock);
812}
813
814/* Remove endpoint from the hash. Local BH-safe. */
815void sctp_unhash_endpoint(struct sctp_endpoint *ep)
816{
817 local_bh_disable();
818 __sctp_unhash_endpoint(ep);
819 local_bh_enable();
820}
821
822static inline __u32 sctp_hashfn(const struct net *net, __be16 lport,
823 const union sctp_addr *paddr, __u32 seed)
824{
825 __u32 addr;
826
827 if (paddr->sa.sa_family == AF_INET6)
828 addr = jhash(&paddr->v6.sin6_addr, 16, seed);
829 else
830 addr = (__force __u32)paddr->v4.sin_addr.s_addr;
831
832 return jhash_3words(addr, ((__force __u32)paddr->v4.sin_port) << 16 |
833 (__force __u32)lport, net_hash_mix(net), seed);
834}
835
836/* Look up an endpoint. */
837static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(
838 struct net *net, struct sk_buff *skb,
839 const union sctp_addr *laddr,
840 const union sctp_addr *paddr,
841 int dif, int sdif)
842{
843 struct sctp_hashbucket *head;
844 struct sctp_endpoint *ep;
845 struct sock *sk;
846 __be16 lport;
847 int hash;
848
849 lport = laddr->v4.sin_port;
850 hash = sctp_ep_hashfn(net, ntohs(lport));
851 head = &sctp_ep_hashtable[hash];
852 read_lock(&head->lock);
853 sctp_for_each_hentry(ep, &head->chain) {
854 if (sctp_endpoint_is_match(ep, net, laddr, dif, sdif))
855 goto hit;
856 }
857
858 ep = sctp_sk(net->sctp.ctl_sock)->ep;
859
860hit:
861 sk = ep->base.sk;
862 if (sk->sk_reuseport) {
863 __u32 phash = sctp_hashfn(net, lport, paddr, 0);
864
865 sk = reuseport_select_sock(sk, phash, skb,
866 sizeof(struct sctphdr));
867 if (sk)
868 ep = sctp_sk(sk)->ep;
869 }
870 sctp_endpoint_hold(ep);
871 read_unlock(&head->lock);
872 return ep;
873}
874
875/* rhashtable for transport */
876struct sctp_hash_cmp_arg {
877 const union sctp_addr *paddr;
878 const struct net *net;
879 __be16 lport;
880};
881
882static inline int sctp_hash_cmp(struct rhashtable_compare_arg *arg,
883 const void *ptr)
884{
885 struct sctp_transport *t = (struct sctp_transport *)ptr;
886 const struct sctp_hash_cmp_arg *x = arg->key;
887 int err = 1;
888
889 if (!sctp_cmp_addr_exact(&t->ipaddr, x->paddr))
890 return err;
891 if (!sctp_transport_hold(t))
892 return err;
893
894 if (!net_eq(t->asoc->base.net, x->net))
895 goto out;
896 if (x->lport != htons(t->asoc->base.bind_addr.port))
897 goto out;
898
899 err = 0;
900out:
901 sctp_transport_put(t);
902 return err;
903}
904
905static inline __u32 sctp_hash_obj(const void *data, u32 len, u32 seed)
906{
907 const struct sctp_transport *t = data;
908
909 return sctp_hashfn(t->asoc->base.net,
910 htons(t->asoc->base.bind_addr.port),
911 &t->ipaddr, seed);
912}
913
914static inline __u32 sctp_hash_key(const void *data, u32 len, u32 seed)
915{
916 const struct sctp_hash_cmp_arg *x = data;
917
918 return sctp_hashfn(x->net, x->lport, x->paddr, seed);
919}
920
921static const struct rhashtable_params sctp_hash_params = {
922 .head_offset = offsetof(struct sctp_transport, node),
923 .hashfn = sctp_hash_key,
924 .obj_hashfn = sctp_hash_obj,
925 .obj_cmpfn = sctp_hash_cmp,
926 .automatic_shrinking = true,
927};
928
929int sctp_transport_hashtable_init(void)
930{
931 return rhltable_init(&sctp_transport_hashtable, &sctp_hash_params);
932}
933
934void sctp_transport_hashtable_destroy(void)
935{
936 rhltable_destroy(&sctp_transport_hashtable);
937}
938
939int sctp_hash_transport(struct sctp_transport *t)
940{
941 struct sctp_transport *transport;
942 struct rhlist_head *tmp, *list;
943 struct sctp_hash_cmp_arg arg;
944 int err;
945
946 if (t->asoc->temp)
947 return 0;
948
949 arg.net = t->asoc->base.net;
950 arg.paddr = &t->ipaddr;
951 arg.lport = htons(t->asoc->base.bind_addr.port);
952
953 rcu_read_lock();
954 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
955 sctp_hash_params);
956
957 rhl_for_each_entry_rcu(transport, tmp, list, node)
958 if (transport->asoc->ep == t->asoc->ep) {
959 rcu_read_unlock();
960 return -EEXIST;
961 }
962 rcu_read_unlock();
963
964 err = rhltable_insert_key(&sctp_transport_hashtable, &arg,
965 &t->node, sctp_hash_params);
966 if (err)
967 pr_err_once("insert transport fail, errno %d\n", err);
968
969 return err;
970}
971
972void sctp_unhash_transport(struct sctp_transport *t)
973{
974 if (t->asoc->temp)
975 return;
976
977 rhltable_remove(&sctp_transport_hashtable, &t->node,
978 sctp_hash_params);
979}
980
981bool sctp_sk_bound_dev_eq(struct net *net, int bound_dev_if, int dif, int sdif)
982{
983 bool l3mdev_accept = true;
984
985#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
986 l3mdev_accept = !!READ_ONCE(net->sctp.l3mdev_accept);
987#endif
988 return inet_bound_dev_eq(l3mdev_accept, bound_dev_if, dif, sdif);
989}
990
991/* return a transport with holding it */
992struct sctp_transport *sctp_addrs_lookup_transport(
993 struct net *net,
994 const union sctp_addr *laddr,
995 const union sctp_addr *paddr,
996 int dif, int sdif)
997{
998 struct rhlist_head *tmp, *list;
999 struct sctp_transport *t;
1000 int bound_dev_if;
1001 struct sctp_hash_cmp_arg arg = {
1002 .paddr = paddr,
1003 .net = net,
1004 .lport = laddr->v4.sin_port,
1005 };
1006
1007 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
1008 sctp_hash_params);
1009
1010 rhl_for_each_entry_rcu(t, tmp, list, node) {
1011 if (!sctp_transport_hold(t))
1012 continue;
1013
1014 bound_dev_if = READ_ONCE(t->asoc->base.sk->sk_bound_dev_if);
1015 if (sctp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif) &&
1016 sctp_bind_addr_match(&t->asoc->base.bind_addr,
1017 laddr, sctp_sk(t->asoc->base.sk)))
1018 return t;
1019 sctp_transport_put(t);
1020 }
1021
1022 return NULL;
1023}
1024
1025/* return a transport without holding it, as it's only used under sock lock */
1026struct sctp_transport *sctp_epaddr_lookup_transport(
1027 const struct sctp_endpoint *ep,
1028 const union sctp_addr *paddr)
1029{
1030 struct rhlist_head *tmp, *list;
1031 struct sctp_transport *t;
1032 struct sctp_hash_cmp_arg arg = {
1033 .paddr = paddr,
1034 .net = ep->base.net,
1035 .lport = htons(ep->base.bind_addr.port),
1036 };
1037
1038 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
1039 sctp_hash_params);
1040
1041 rhl_for_each_entry_rcu(t, tmp, list, node)
1042 if (ep == t->asoc->ep)
1043 return t;
1044
1045 return NULL;
1046}
1047
1048/* Look up an association. */
1049static struct sctp_association *__sctp_lookup_association(
1050 struct net *net,
1051 const union sctp_addr *local,
1052 const union sctp_addr *peer,
1053 struct sctp_transport **pt,
1054 int dif, int sdif)
1055{
1056 struct sctp_transport *t;
1057 struct sctp_association *asoc = NULL;
1058
1059 t = sctp_addrs_lookup_transport(net, local, peer, dif, sdif);
1060 if (!t)
1061 goto out;
1062
1063 asoc = t->asoc;
1064 *pt = t;
1065
1066out:
1067 return asoc;
1068}
1069
1070/* Look up an association. protected by RCU read lock */
1071static
1072struct sctp_association *sctp_lookup_association(struct net *net,
1073 const union sctp_addr *laddr,
1074 const union sctp_addr *paddr,
1075 struct sctp_transport **transportp,
1076 int dif, int sdif)
1077{
1078 struct sctp_association *asoc;
1079
1080 rcu_read_lock();
1081 asoc = __sctp_lookup_association(net, laddr, paddr, transportp, dif, sdif);
1082 rcu_read_unlock();
1083
1084 return asoc;
1085}
1086
1087/* Is there an association matching the given local and peer addresses? */
1088bool sctp_has_association(struct net *net,
1089 const union sctp_addr *laddr,
1090 const union sctp_addr *paddr,
1091 int dif, int sdif)
1092{
1093 struct sctp_transport *transport;
1094
1095 if (sctp_lookup_association(net, laddr, paddr, &transport, dif, sdif)) {
1096 sctp_transport_put(transport);
1097 return true;
1098 }
1099
1100 return false;
1101}
1102
1103/*
1104 * SCTP Implementors Guide, 2.18 Handling of address
1105 * parameters within the INIT or INIT-ACK.
1106 *
1107 * D) When searching for a matching TCB upon reception of an INIT
1108 * or INIT-ACK chunk the receiver SHOULD use not only the
1109 * source address of the packet (containing the INIT or
1110 * INIT-ACK) but the receiver SHOULD also use all valid
1111 * address parameters contained within the chunk.
1112 *
1113 * 2.18.3 Solution description
1114 *
1115 * This new text clearly specifies to an implementor the need
1116 * to look within the INIT or INIT-ACK. Any implementation that
1117 * does not do this, may not be able to establish associations
1118 * in certain circumstances.
1119 *
1120 */
1121static struct sctp_association *__sctp_rcv_init_lookup(struct net *net,
1122 struct sk_buff *skb,
1123 const union sctp_addr *laddr, struct sctp_transport **transportp,
1124 int dif, int sdif)
1125{
1126 struct sctp_association *asoc;
1127 union sctp_addr addr;
1128 union sctp_addr *paddr = &addr;
1129 struct sctphdr *sh = sctp_hdr(skb);
1130 union sctp_params params;
1131 struct sctp_init_chunk *init;
1132 struct sctp_af *af;
1133
1134 /*
1135 * This code will NOT touch anything inside the chunk--it is
1136 * strictly READ-ONLY.
1137 *
1138 * RFC 2960 3 SCTP packet Format
1139 *
1140 * Multiple chunks can be bundled into one SCTP packet up to
1141 * the MTU size, except for the INIT, INIT ACK, and SHUTDOWN
1142 * COMPLETE chunks. These chunks MUST NOT be bundled with any
1143 * other chunk in a packet. See Section 6.10 for more details
1144 * on chunk bundling.
1145 */
1146
1147 /* Find the start of the TLVs and the end of the chunk. This is
1148 * the region we search for address parameters.
1149 */
1150 init = (struct sctp_init_chunk *)skb->data;
1151
1152 /* Walk the parameters looking for embedded addresses. */
1153 sctp_walk_params(params, init) {
1154
1155 /* Note: Ignoring hostname addresses. */
1156 af = sctp_get_af_specific(param_type2af(params.p->type));
1157 if (!af)
1158 continue;
1159
1160 if (!af->from_addr_param(paddr, params.addr, sh->source, 0))
1161 continue;
1162
1163 asoc = __sctp_lookup_association(net, laddr, paddr, transportp, dif, sdif);
1164 if (asoc)
1165 return asoc;
1166 }
1167
1168 return NULL;
1169}
1170
1171/* ADD-IP, Section 5.2
1172 * When an endpoint receives an ASCONF Chunk from the remote peer
1173 * special procedures may be needed to identify the association the
1174 * ASCONF Chunk is associated with. To properly find the association
1175 * the following procedures SHOULD be followed:
1176 *
1177 * D2) If the association is not found, use the address found in the
1178 * Address Parameter TLV combined with the port number found in the
1179 * SCTP common header. If found proceed to rule D4.
1180 *
1181 * D2-ext) If more than one ASCONF Chunks are packed together, use the
1182 * address found in the ASCONF Address Parameter TLV of each of the
1183 * subsequent ASCONF Chunks. If found, proceed to rule D4.
1184 */
1185static struct sctp_association *__sctp_rcv_asconf_lookup(
1186 struct net *net,
1187 struct sctp_chunkhdr *ch,
1188 const union sctp_addr *laddr,
1189 __be16 peer_port,
1190 struct sctp_transport **transportp,
1191 int dif, int sdif)
1192{
1193 struct sctp_addip_chunk *asconf = (struct sctp_addip_chunk *)ch;
1194 struct sctp_af *af;
1195 union sctp_addr_param *param;
1196 union sctp_addr paddr;
1197
1198 if (ntohs(ch->length) < sizeof(*asconf) + sizeof(struct sctp_paramhdr))
1199 return NULL;
1200
1201 /* Skip over the ADDIP header and find the Address parameter */
1202 param = (union sctp_addr_param *)(asconf + 1);
1203
1204 af = sctp_get_af_specific(param_type2af(param->p.type));
1205 if (unlikely(!af))
1206 return NULL;
1207
1208 if (!af->from_addr_param(&paddr, param, peer_port, 0))
1209 return NULL;
1210
1211 return __sctp_lookup_association(net, laddr, &paddr, transportp, dif, sdif);
1212}
1213
1214
1215/* SCTP-AUTH, Section 6.3:
1216* If the receiver does not find a STCB for a packet containing an AUTH
1217* chunk as the first chunk and not a COOKIE-ECHO chunk as the second
1218* chunk, it MUST use the chunks after the AUTH chunk to look up an existing
1219* association.
1220*
1221* This means that any chunks that can help us identify the association need
1222* to be looked at to find this association.
1223*/
1224static struct sctp_association *__sctp_rcv_walk_lookup(struct net *net,
1225 struct sk_buff *skb,
1226 const union sctp_addr *laddr,
1227 struct sctp_transport **transportp,
1228 int dif, int sdif)
1229{
1230 struct sctp_association *asoc = NULL;
1231 struct sctp_chunkhdr *ch;
1232 int have_auth = 0;
1233 unsigned int chunk_num = 1;
1234 __u8 *ch_end;
1235
1236 /* Walk through the chunks looking for AUTH or ASCONF chunks
1237 * to help us find the association.
1238 */
1239 ch = (struct sctp_chunkhdr *)skb->data;
1240 do {
1241 /* Break out if chunk length is less then minimal. */
1242 if (ntohs(ch->length) < sizeof(*ch))
1243 break;
1244
1245 ch_end = ((__u8 *)ch) + SCTP_PAD4(ntohs(ch->length));
1246 if (ch_end > skb_tail_pointer(skb))
1247 break;
1248
1249 switch (ch->type) {
1250 case SCTP_CID_AUTH:
1251 have_auth = chunk_num;
1252 break;
1253
1254 case SCTP_CID_COOKIE_ECHO:
1255 /* If a packet arrives containing an AUTH chunk as
1256 * a first chunk, a COOKIE-ECHO chunk as the second
1257 * chunk, and possibly more chunks after them, and
1258 * the receiver does not have an STCB for that
1259 * packet, then authentication is based on
1260 * the contents of the COOKIE- ECHO chunk.
1261 */
1262 if (have_auth == 1 && chunk_num == 2)
1263 return NULL;
1264 break;
1265
1266 case SCTP_CID_ASCONF:
1267 if (have_auth || net->sctp.addip_noauth)
1268 asoc = __sctp_rcv_asconf_lookup(
1269 net, ch, laddr,
1270 sctp_hdr(skb)->source,
1271 transportp, dif, sdif);
1272 break;
1273 default:
1274 break;
1275 }
1276
1277 if (asoc)
1278 break;
1279
1280 ch = (struct sctp_chunkhdr *)ch_end;
1281 chunk_num++;
1282 } while (ch_end + sizeof(*ch) < skb_tail_pointer(skb));
1283
1284 return asoc;
1285}
1286
1287/*
1288 * There are circumstances when we need to look inside the SCTP packet
1289 * for information to help us find the association. Examples
1290 * include looking inside of INIT/INIT-ACK chunks or after the AUTH
1291 * chunks.
1292 */
1293static struct sctp_association *__sctp_rcv_lookup_harder(struct net *net,
1294 struct sk_buff *skb,
1295 const union sctp_addr *laddr,
1296 struct sctp_transport **transportp,
1297 int dif, int sdif)
1298{
1299 struct sctp_chunkhdr *ch;
1300
1301 /* We do not allow GSO frames here as we need to linearize and
1302 * then cannot guarantee frame boundaries. This shouldn't be an
1303 * issue as packets hitting this are mostly INIT or INIT-ACK and
1304 * those cannot be on GSO-style anyway.
1305 */
1306 if (skb_is_gso(skb) && skb_is_gso_sctp(skb))
1307 return NULL;
1308
1309 ch = (struct sctp_chunkhdr *)skb->data;
1310
1311 /* The code below will attempt to walk the chunk and extract
1312 * parameter information. Before we do that, we need to verify
1313 * that the chunk length doesn't cause overflow. Otherwise, we'll
1314 * walk off the end.
1315 */
1316 if (SCTP_PAD4(ntohs(ch->length)) > skb->len)
1317 return NULL;
1318
1319 /* If this is INIT/INIT-ACK look inside the chunk too. */
1320 if (ch->type == SCTP_CID_INIT || ch->type == SCTP_CID_INIT_ACK)
1321 return __sctp_rcv_init_lookup(net, skb, laddr, transportp, dif, sdif);
1322
1323 return __sctp_rcv_walk_lookup(net, skb, laddr, transportp, dif, sdif);
1324}
1325
1326/* Lookup an association for an inbound skb. */
1327static struct sctp_association *__sctp_rcv_lookup(struct net *net,
1328 struct sk_buff *skb,
1329 const union sctp_addr *paddr,
1330 const union sctp_addr *laddr,
1331 struct sctp_transport **transportp,
1332 int dif, int sdif)
1333{
1334 struct sctp_association *asoc;
1335
1336 asoc = __sctp_lookup_association(net, laddr, paddr, transportp, dif, sdif);
1337 if (asoc)
1338 goto out;
1339
1340 /* Further lookup for INIT/INIT-ACK packets.
1341 * SCTP Implementors Guide, 2.18 Handling of address
1342 * parameters within the INIT or INIT-ACK.
1343 */
1344 asoc = __sctp_rcv_lookup_harder(net, skb, laddr, transportp, dif, sdif);
1345 if (asoc)
1346 goto out;
1347
1348 if (paddr->sa.sa_family == AF_INET)
1349 pr_debug("sctp: asoc not found for src:%pI4:%d dst:%pI4:%d\n",
1350 &laddr->v4.sin_addr, ntohs(laddr->v4.sin_port),
1351 &paddr->v4.sin_addr, ntohs(paddr->v4.sin_port));
1352 else
1353 pr_debug("sctp: asoc not found for src:%pI6:%d dst:%pI6:%d\n",
1354 &laddr->v6.sin6_addr, ntohs(laddr->v6.sin6_port),
1355 &paddr->v6.sin6_addr, ntohs(paddr->v6.sin6_port));
1356
1357out:
1358 return asoc;
1359}
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 bool is_gso = skb_is_gso(skb) && skb_is_gso_sctp(skb);
110
111 if (skb->pkt_type != PACKET_HOST)
112 goto discard_it;
113
114 __SCTP_INC_STATS(net, SCTP_MIB_INSCTPPACKS);
115
116 /* If packet is too small to contain a single chunk, let's not
117 * waste time on it anymore.
118 */
119 if (skb->len < sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) +
120 skb_transport_offset(skb))
121 goto discard_it;
122
123 /* If the packet is fragmented and we need to do crc checking,
124 * it's better to just linearize it otherwise crc computing
125 * takes longer.
126 */
127 if ((!is_gso && 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 !is_gso &&
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 /* We can't allow retransmitting in such case, as the
404 * retransmission would be sized just as before, and thus we
405 * would get another icmp, and retransmit again.
406 */
407 return;
408
409 /* Update transports view of the MTU. Return if no update was needed.
410 * If an update wasn't needed/possible, it also doesn't make sense to
411 * try to retransmit now.
412 */
413 if (!sctp_transport_update_pmtu(t, pmtu))
414 return;
415
416 /* Update association pmtu. */
417 sctp_assoc_sync_pmtu(asoc);
418
419 /* Retransmit with the new pmtu setting. */
420 sctp_retransmit(&asoc->outqueue, t, SCTP_RTXR_PMTUD);
421}
422
423void sctp_icmp_redirect(struct sock *sk, struct sctp_transport *t,
424 struct sk_buff *skb)
425{
426 struct dst_entry *dst;
427
428 if (sock_owned_by_user(sk) || !t)
429 return;
430 dst = sctp_transport_dst_check(t);
431 if (dst)
432 dst->ops->redirect(dst, sk, skb);
433}
434
435/*
436 * SCTP Implementer's Guide, 2.37 ICMP handling procedures
437 *
438 * ICMP8) If the ICMP code is a "Unrecognized next header type encountered"
439 * or a "Protocol Unreachable" treat this message as an abort
440 * with the T bit set.
441 *
442 * This function sends an event to the state machine, which will abort the
443 * association.
444 *
445 */
446void sctp_icmp_proto_unreachable(struct sock *sk,
447 struct sctp_association *asoc,
448 struct sctp_transport *t)
449{
450 if (sock_owned_by_user(sk)) {
451 if (timer_pending(&t->proto_unreach_timer))
452 return;
453 else {
454 if (!mod_timer(&t->proto_unreach_timer,
455 jiffies + (HZ/20)))
456 sctp_association_hold(asoc);
457 }
458 } else {
459 struct net *net = sock_net(sk);
460
461 pr_debug("%s: unrecognized next header type "
462 "encountered!\n", __func__);
463
464 if (del_timer(&t->proto_unreach_timer))
465 sctp_association_put(asoc);
466
467 sctp_do_sm(net, SCTP_EVENT_T_OTHER,
468 SCTP_ST_OTHER(SCTP_EVENT_ICMP_PROTO_UNREACH),
469 asoc->state, asoc->ep, asoc, t,
470 GFP_ATOMIC);
471 }
472}
473
474/* Common lookup code for icmp/icmpv6 error handler. */
475struct sock *sctp_err_lookup(struct net *net, int family, struct sk_buff *skb,
476 struct sctphdr *sctphdr,
477 struct sctp_association **app,
478 struct sctp_transport **tpp)
479{
480 struct sctp_init_chunk *chunkhdr, _chunkhdr;
481 union sctp_addr saddr;
482 union sctp_addr daddr;
483 struct sctp_af *af;
484 struct sock *sk = NULL;
485 struct sctp_association *asoc;
486 struct sctp_transport *transport = NULL;
487 __u32 vtag = ntohl(sctphdr->vtag);
488
489 *app = NULL; *tpp = NULL;
490
491 af = sctp_get_af_specific(family);
492 if (unlikely(!af)) {
493 return NULL;
494 }
495
496 /* Initialize local addresses for lookups. */
497 af->from_skb(&saddr, skb, 1);
498 af->from_skb(&daddr, skb, 0);
499
500 /* Look for an association that matches the incoming ICMP error
501 * packet.
502 */
503 asoc = __sctp_lookup_association(net, &saddr, &daddr, &transport);
504 if (!asoc)
505 return NULL;
506
507 sk = asoc->base.sk;
508
509 /* RFC 4960, Appendix C. ICMP Handling
510 *
511 * ICMP6) An implementation MUST validate that the Verification Tag
512 * contained in the ICMP message matches the Verification Tag of
513 * the peer. If the Verification Tag is not 0 and does NOT
514 * match, discard the ICMP message. If it is 0 and the ICMP
515 * message contains enough bytes to verify that the chunk type is
516 * an INIT chunk and that the Initiate Tag matches the tag of the
517 * peer, continue with ICMP7. If the ICMP message is too short
518 * or the chunk type or the Initiate Tag does not match, silently
519 * discard the packet.
520 */
521 if (vtag == 0) {
522 /* chunk header + first 4 octects of init header */
523 chunkhdr = skb_header_pointer(skb, skb_transport_offset(skb) +
524 sizeof(struct sctphdr),
525 sizeof(struct sctp_chunkhdr) +
526 sizeof(__be32), &_chunkhdr);
527 if (!chunkhdr ||
528 chunkhdr->chunk_hdr.type != SCTP_CID_INIT ||
529 ntohl(chunkhdr->init_hdr.init_tag) != asoc->c.my_vtag)
530 goto out;
531
532 } else if (vtag != asoc->c.peer_vtag) {
533 goto out;
534 }
535
536 bh_lock_sock(sk);
537
538 /* If too many ICMPs get dropped on busy
539 * servers this needs to be solved differently.
540 */
541 if (sock_owned_by_user(sk))
542 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
543
544 *app = asoc;
545 *tpp = transport;
546 return sk;
547
548out:
549 sctp_transport_put(transport);
550 return NULL;
551}
552
553/* Common cleanup code for icmp/icmpv6 error handler. */
554void sctp_err_finish(struct sock *sk, struct sctp_transport *t)
555{
556 bh_unlock_sock(sk);
557 sctp_transport_put(t);
558}
559
560/*
561 * This routine is called by the ICMP module when it gets some
562 * sort of error condition. If err < 0 then the socket should
563 * be closed and the error returned to the user. If err > 0
564 * it's just the icmp type << 8 | icmp code. After adjustment
565 * header points to the first 8 bytes of the sctp header. We need
566 * to find the appropriate port.
567 *
568 * The locking strategy used here is very "optimistic". When
569 * someone else accesses the socket the ICMP is just dropped
570 * and for some paths there is no check at all.
571 * A more general error queue to queue errors for later handling
572 * is probably better.
573 *
574 */
575void sctp_v4_err(struct sk_buff *skb, __u32 info)
576{
577 const struct iphdr *iph = (const struct iphdr *)skb->data;
578 const int ihlen = iph->ihl * 4;
579 const int type = icmp_hdr(skb)->type;
580 const int code = icmp_hdr(skb)->code;
581 struct sock *sk;
582 struct sctp_association *asoc = NULL;
583 struct sctp_transport *transport;
584 struct inet_sock *inet;
585 __u16 saveip, savesctp;
586 int err;
587 struct net *net = dev_net(skb->dev);
588
589 /* Fix up skb to look at the embedded net header. */
590 saveip = skb->network_header;
591 savesctp = skb->transport_header;
592 skb_reset_network_header(skb);
593 skb_set_transport_header(skb, ihlen);
594 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &transport);
595 /* Put back, the original values. */
596 skb->network_header = saveip;
597 skb->transport_header = savesctp;
598 if (!sk) {
599 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
600 return;
601 }
602 /* Warning: The sock lock is held. Remember to call
603 * sctp_err_finish!
604 */
605
606 switch (type) {
607 case ICMP_PARAMETERPROB:
608 err = EPROTO;
609 break;
610 case ICMP_DEST_UNREACH:
611 if (code > NR_ICMP_UNREACH)
612 goto out_unlock;
613
614 /* PMTU discovery (RFC1191) */
615 if (ICMP_FRAG_NEEDED == code) {
616 sctp_icmp_frag_needed(sk, asoc, transport,
617 SCTP_TRUNC4(info));
618 goto out_unlock;
619 } else {
620 if (ICMP_PROT_UNREACH == code) {
621 sctp_icmp_proto_unreachable(sk, asoc,
622 transport);
623 goto out_unlock;
624 }
625 }
626 err = icmp_err_convert[code].errno;
627 break;
628 case ICMP_TIME_EXCEEDED:
629 /* Ignore any time exceeded errors due to fragment reassembly
630 * timeouts.
631 */
632 if (ICMP_EXC_FRAGTIME == code)
633 goto out_unlock;
634
635 err = EHOSTUNREACH;
636 break;
637 case ICMP_REDIRECT:
638 sctp_icmp_redirect(sk, transport, skb);
639 /* Fall through to out_unlock. */
640 default:
641 goto out_unlock;
642 }
643
644 inet = inet_sk(sk);
645 if (!sock_owned_by_user(sk) && inet->recverr) {
646 sk->sk_err = err;
647 sk->sk_error_report(sk);
648 } else { /* Only an error on timeout */
649 sk->sk_err_soft = err;
650 }
651
652out_unlock:
653 sctp_err_finish(sk, transport);
654}
655
656/*
657 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
658 *
659 * This function scans all the chunks in the OOTB packet to determine if
660 * the packet should be discarded right away. If a response might be needed
661 * for this packet, or, if further processing is possible, the packet will
662 * be queued to a proper inqueue for the next phase of handling.
663 *
664 * Output:
665 * Return 0 - If further processing is needed.
666 * Return 1 - If the packet can be discarded right away.
667 */
668static int sctp_rcv_ootb(struct sk_buff *skb)
669{
670 struct sctp_chunkhdr *ch, _ch;
671 int ch_end, offset = 0;
672
673 /* Scan through all the chunks in the packet. */
674 do {
675 /* Make sure we have at least the header there */
676 if (offset + sizeof(_ch) > skb->len)
677 break;
678
679 ch = skb_header_pointer(skb, offset, sizeof(*ch), &_ch);
680
681 /* Break out if chunk length is less then minimal. */
682 if (ntohs(ch->length) < sizeof(_ch))
683 break;
684
685 ch_end = offset + SCTP_PAD4(ntohs(ch->length));
686 if (ch_end > skb->len)
687 break;
688
689 /* RFC 8.4, 2) If the OOTB packet contains an ABORT chunk, the
690 * receiver MUST silently discard the OOTB packet and take no
691 * further action.
692 */
693 if (SCTP_CID_ABORT == ch->type)
694 goto discard;
695
696 /* RFC 8.4, 6) If the packet contains a SHUTDOWN COMPLETE
697 * chunk, the receiver should silently discard the packet
698 * and take no further action.
699 */
700 if (SCTP_CID_SHUTDOWN_COMPLETE == ch->type)
701 goto discard;
702
703 /* RFC 4460, 2.11.2
704 * This will discard packets with INIT chunk bundled as
705 * subsequent chunks in the packet. When INIT is first,
706 * the normal INIT processing will discard the chunk.
707 */
708 if (SCTP_CID_INIT == ch->type && (void *)ch != skb->data)
709 goto discard;
710
711 offset = ch_end;
712 } while (ch_end < skb->len);
713
714 return 0;
715
716discard:
717 return 1;
718}
719
720/* Insert endpoint into the hash table. */
721static void __sctp_hash_endpoint(struct sctp_endpoint *ep)
722{
723 struct net *net = sock_net(ep->base.sk);
724 struct sctp_ep_common *epb;
725 struct sctp_hashbucket *head;
726
727 epb = &ep->base;
728
729 epb->hashent = sctp_ep_hashfn(net, epb->bind_addr.port);
730 head = &sctp_ep_hashtable[epb->hashent];
731
732 write_lock(&head->lock);
733 hlist_add_head(&epb->node, &head->chain);
734 write_unlock(&head->lock);
735}
736
737/* Add an endpoint to the hash. Local BH-safe. */
738void sctp_hash_endpoint(struct sctp_endpoint *ep)
739{
740 local_bh_disable();
741 __sctp_hash_endpoint(ep);
742 local_bh_enable();
743}
744
745/* Remove endpoint from the hash table. */
746static void __sctp_unhash_endpoint(struct sctp_endpoint *ep)
747{
748 struct net *net = sock_net(ep->base.sk);
749 struct sctp_hashbucket *head;
750 struct sctp_ep_common *epb;
751
752 epb = &ep->base;
753
754 epb->hashent = sctp_ep_hashfn(net, epb->bind_addr.port);
755
756 head = &sctp_ep_hashtable[epb->hashent];
757
758 write_lock(&head->lock);
759 hlist_del_init(&epb->node);
760 write_unlock(&head->lock);
761}
762
763/* Remove endpoint from the hash. Local BH-safe. */
764void sctp_unhash_endpoint(struct sctp_endpoint *ep)
765{
766 local_bh_disable();
767 __sctp_unhash_endpoint(ep);
768 local_bh_enable();
769}
770
771/* Look up an endpoint. */
772static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(struct net *net,
773 const union sctp_addr *laddr)
774{
775 struct sctp_hashbucket *head;
776 struct sctp_ep_common *epb;
777 struct sctp_endpoint *ep;
778 int hash;
779
780 hash = sctp_ep_hashfn(net, ntohs(laddr->v4.sin_port));
781 head = &sctp_ep_hashtable[hash];
782 read_lock(&head->lock);
783 sctp_for_each_hentry(epb, &head->chain) {
784 ep = sctp_ep(epb);
785 if (sctp_endpoint_is_match(ep, net, laddr))
786 goto hit;
787 }
788
789 ep = sctp_sk(net->sctp.ctl_sock)->ep;
790
791hit:
792 sctp_endpoint_hold(ep);
793 read_unlock(&head->lock);
794 return ep;
795}
796
797/* rhashtable for transport */
798struct sctp_hash_cmp_arg {
799 const union sctp_addr *paddr;
800 const struct net *net;
801 __be16 lport;
802};
803
804static inline int sctp_hash_cmp(struct rhashtable_compare_arg *arg,
805 const void *ptr)
806{
807 struct sctp_transport *t = (struct sctp_transport *)ptr;
808 const struct sctp_hash_cmp_arg *x = arg->key;
809 int err = 1;
810
811 if (!sctp_cmp_addr_exact(&t->ipaddr, x->paddr))
812 return err;
813 if (!sctp_transport_hold(t))
814 return err;
815
816 if (!net_eq(sock_net(t->asoc->base.sk), x->net))
817 goto out;
818 if (x->lport != htons(t->asoc->base.bind_addr.port))
819 goto out;
820
821 err = 0;
822out:
823 sctp_transport_put(t);
824 return err;
825}
826
827static inline __u32 sctp_hash_obj(const void *data, u32 len, u32 seed)
828{
829 const struct sctp_transport *t = data;
830 const union sctp_addr *paddr = &t->ipaddr;
831 const struct net *net = sock_net(t->asoc->base.sk);
832 __be16 lport = htons(t->asoc->base.bind_addr.port);
833 __u32 addr;
834
835 if (paddr->sa.sa_family == AF_INET6)
836 addr = jhash(&paddr->v6.sin6_addr, 16, seed);
837 else
838 addr = (__force __u32)paddr->v4.sin_addr.s_addr;
839
840 return jhash_3words(addr, ((__force __u32)paddr->v4.sin_port) << 16 |
841 (__force __u32)lport, net_hash_mix(net), seed);
842}
843
844static inline __u32 sctp_hash_key(const void *data, u32 len, u32 seed)
845{
846 const struct sctp_hash_cmp_arg *x = data;
847 const union sctp_addr *paddr = x->paddr;
848 const struct net *net = x->net;
849 __be16 lport = x->lport;
850 __u32 addr;
851
852 if (paddr->sa.sa_family == AF_INET6)
853 addr = jhash(&paddr->v6.sin6_addr, 16, seed);
854 else
855 addr = (__force __u32)paddr->v4.sin_addr.s_addr;
856
857 return jhash_3words(addr, ((__force __u32)paddr->v4.sin_port) << 16 |
858 (__force __u32)lport, net_hash_mix(net), seed);
859}
860
861static const struct rhashtable_params sctp_hash_params = {
862 .head_offset = offsetof(struct sctp_transport, node),
863 .hashfn = sctp_hash_key,
864 .obj_hashfn = sctp_hash_obj,
865 .obj_cmpfn = sctp_hash_cmp,
866 .automatic_shrinking = true,
867};
868
869int sctp_transport_hashtable_init(void)
870{
871 return rhltable_init(&sctp_transport_hashtable, &sctp_hash_params);
872}
873
874void sctp_transport_hashtable_destroy(void)
875{
876 rhltable_destroy(&sctp_transport_hashtable);
877}
878
879int sctp_hash_transport(struct sctp_transport *t)
880{
881 struct sctp_transport *transport;
882 struct rhlist_head *tmp, *list;
883 struct sctp_hash_cmp_arg arg;
884 int err;
885
886 if (t->asoc->temp)
887 return 0;
888
889 arg.net = sock_net(t->asoc->base.sk);
890 arg.paddr = &t->ipaddr;
891 arg.lport = htons(t->asoc->base.bind_addr.port);
892
893 rcu_read_lock();
894 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
895 sctp_hash_params);
896
897 rhl_for_each_entry_rcu(transport, tmp, list, node)
898 if (transport->asoc->ep == t->asoc->ep) {
899 rcu_read_unlock();
900 return -EEXIST;
901 }
902 rcu_read_unlock();
903
904 err = rhltable_insert_key(&sctp_transport_hashtable, &arg,
905 &t->node, sctp_hash_params);
906 if (err)
907 pr_err_once("insert transport fail, errno %d\n", err);
908
909 return err;
910}
911
912void sctp_unhash_transport(struct sctp_transport *t)
913{
914 if (t->asoc->temp)
915 return;
916
917 rhltable_remove(&sctp_transport_hashtable, &t->node,
918 sctp_hash_params);
919}
920
921/* return a transport with holding it */
922struct sctp_transport *sctp_addrs_lookup_transport(
923 struct net *net,
924 const union sctp_addr *laddr,
925 const union sctp_addr *paddr)
926{
927 struct rhlist_head *tmp, *list;
928 struct sctp_transport *t;
929 struct sctp_hash_cmp_arg arg = {
930 .paddr = paddr,
931 .net = net,
932 .lport = laddr->v4.sin_port,
933 };
934
935 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
936 sctp_hash_params);
937
938 rhl_for_each_entry_rcu(t, tmp, list, node) {
939 if (!sctp_transport_hold(t))
940 continue;
941
942 if (sctp_bind_addr_match(&t->asoc->base.bind_addr,
943 laddr, sctp_sk(t->asoc->base.sk)))
944 return t;
945 sctp_transport_put(t);
946 }
947
948 return NULL;
949}
950
951/* return a transport without holding it, as it's only used under sock lock */
952struct sctp_transport *sctp_epaddr_lookup_transport(
953 const struct sctp_endpoint *ep,
954 const union sctp_addr *paddr)
955{
956 struct net *net = sock_net(ep->base.sk);
957 struct rhlist_head *tmp, *list;
958 struct sctp_transport *t;
959 struct sctp_hash_cmp_arg arg = {
960 .paddr = paddr,
961 .net = net,
962 .lport = htons(ep->base.bind_addr.port),
963 };
964
965 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
966 sctp_hash_params);
967
968 rhl_for_each_entry_rcu(t, tmp, list, node)
969 if (ep == t->asoc->ep)
970 return t;
971
972 return NULL;
973}
974
975/* Look up an association. */
976static struct sctp_association *__sctp_lookup_association(
977 struct net *net,
978 const union sctp_addr *local,
979 const union sctp_addr *peer,
980 struct sctp_transport **pt)
981{
982 struct sctp_transport *t;
983 struct sctp_association *asoc = NULL;
984
985 t = sctp_addrs_lookup_transport(net, local, peer);
986 if (!t)
987 goto out;
988
989 asoc = t->asoc;
990 *pt = t;
991
992out:
993 return asoc;
994}
995
996/* Look up an association. protected by RCU read lock */
997static
998struct sctp_association *sctp_lookup_association(struct net *net,
999 const union sctp_addr *laddr,
1000 const union sctp_addr *paddr,
1001 struct sctp_transport **transportp)
1002{
1003 struct sctp_association *asoc;
1004
1005 rcu_read_lock();
1006 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1007 rcu_read_unlock();
1008
1009 return asoc;
1010}
1011
1012/* Is there an association matching the given local and peer addresses? */
1013bool sctp_has_association(struct net *net,
1014 const union sctp_addr *laddr,
1015 const union sctp_addr *paddr)
1016{
1017 struct sctp_transport *transport;
1018
1019 if (sctp_lookup_association(net, laddr, paddr, &transport)) {
1020 sctp_transport_put(transport);
1021 return true;
1022 }
1023
1024 return false;
1025}
1026
1027/*
1028 * SCTP Implementors Guide, 2.18 Handling of address
1029 * parameters within the INIT or INIT-ACK.
1030 *
1031 * D) When searching for a matching TCB upon reception of an INIT
1032 * or INIT-ACK chunk the receiver SHOULD use not only the
1033 * source address of the packet (containing the INIT or
1034 * INIT-ACK) but the receiver SHOULD also use all valid
1035 * address parameters contained within the chunk.
1036 *
1037 * 2.18.3 Solution description
1038 *
1039 * This new text clearly specifies to an implementor the need
1040 * to look within the INIT or INIT-ACK. Any implementation that
1041 * does not do this, may not be able to establish associations
1042 * in certain circumstances.
1043 *
1044 */
1045static struct sctp_association *__sctp_rcv_init_lookup(struct net *net,
1046 struct sk_buff *skb,
1047 const union sctp_addr *laddr, struct sctp_transport **transportp)
1048{
1049 struct sctp_association *asoc;
1050 union sctp_addr addr;
1051 union sctp_addr *paddr = &addr;
1052 struct sctphdr *sh = sctp_hdr(skb);
1053 union sctp_params params;
1054 struct sctp_init_chunk *init;
1055 struct sctp_af *af;
1056
1057 /*
1058 * This code will NOT touch anything inside the chunk--it is
1059 * strictly READ-ONLY.
1060 *
1061 * RFC 2960 3 SCTP packet Format
1062 *
1063 * Multiple chunks can be bundled into one SCTP packet up to
1064 * the MTU size, except for the INIT, INIT ACK, and SHUTDOWN
1065 * COMPLETE chunks. These chunks MUST NOT be bundled with any
1066 * other chunk in a packet. See Section 6.10 for more details
1067 * on chunk bundling.
1068 */
1069
1070 /* Find the start of the TLVs and the end of the chunk. This is
1071 * the region we search for address parameters.
1072 */
1073 init = (struct sctp_init_chunk *)skb->data;
1074
1075 /* Walk the parameters looking for embedded addresses. */
1076 sctp_walk_params(params, init, init_hdr.params) {
1077
1078 /* Note: Ignoring hostname addresses. */
1079 af = sctp_get_af_specific(param_type2af(params.p->type));
1080 if (!af)
1081 continue;
1082
1083 af->from_addr_param(paddr, params.addr, sh->source, 0);
1084
1085 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1086 if (asoc)
1087 return asoc;
1088 }
1089
1090 return NULL;
1091}
1092
1093/* ADD-IP, Section 5.2
1094 * When an endpoint receives an ASCONF Chunk from the remote peer
1095 * special procedures may be needed to identify the association the
1096 * ASCONF Chunk is associated with. To properly find the association
1097 * the following procedures SHOULD be followed:
1098 *
1099 * D2) If the association is not found, use the address found in the
1100 * Address Parameter TLV combined with the port number found in the
1101 * SCTP common header. If found proceed to rule D4.
1102 *
1103 * D2-ext) If more than one ASCONF Chunks are packed together, use the
1104 * address found in the ASCONF Address Parameter TLV of each of the
1105 * subsequent ASCONF Chunks. If found, proceed to rule D4.
1106 */
1107static struct sctp_association *__sctp_rcv_asconf_lookup(
1108 struct net *net,
1109 struct sctp_chunkhdr *ch,
1110 const union sctp_addr *laddr,
1111 __be16 peer_port,
1112 struct sctp_transport **transportp)
1113{
1114 struct sctp_addip_chunk *asconf = (struct sctp_addip_chunk *)ch;
1115 struct sctp_af *af;
1116 union sctp_addr_param *param;
1117 union sctp_addr paddr;
1118
1119 /* Skip over the ADDIP header and find the Address parameter */
1120 param = (union sctp_addr_param *)(asconf + 1);
1121
1122 af = sctp_get_af_specific(param_type2af(param->p.type));
1123 if (unlikely(!af))
1124 return NULL;
1125
1126 af->from_addr_param(&paddr, param, peer_port, 0);
1127
1128 return __sctp_lookup_association(net, laddr, &paddr, transportp);
1129}
1130
1131
1132/* SCTP-AUTH, Section 6.3:
1133* If the receiver does not find a STCB for a packet containing an AUTH
1134* chunk as the first chunk and not a COOKIE-ECHO chunk as the second
1135* chunk, it MUST use the chunks after the AUTH chunk to look up an existing
1136* association.
1137*
1138* This means that any chunks that can help us identify the association need
1139* to be looked at to find this association.
1140*/
1141static struct sctp_association *__sctp_rcv_walk_lookup(struct net *net,
1142 struct sk_buff *skb,
1143 const union sctp_addr *laddr,
1144 struct sctp_transport **transportp)
1145{
1146 struct sctp_association *asoc = NULL;
1147 struct sctp_chunkhdr *ch;
1148 int have_auth = 0;
1149 unsigned int chunk_num = 1;
1150 __u8 *ch_end;
1151
1152 /* Walk through the chunks looking for AUTH or ASCONF chunks
1153 * to help us find the association.
1154 */
1155 ch = (struct sctp_chunkhdr *)skb->data;
1156 do {
1157 /* Break out if chunk length is less then minimal. */
1158 if (ntohs(ch->length) < sizeof(*ch))
1159 break;
1160
1161 ch_end = ((__u8 *)ch) + SCTP_PAD4(ntohs(ch->length));
1162 if (ch_end > skb_tail_pointer(skb))
1163 break;
1164
1165 switch (ch->type) {
1166 case SCTP_CID_AUTH:
1167 have_auth = chunk_num;
1168 break;
1169
1170 case SCTP_CID_COOKIE_ECHO:
1171 /* If a packet arrives containing an AUTH chunk as
1172 * a first chunk, a COOKIE-ECHO chunk as the second
1173 * chunk, and possibly more chunks after them, and
1174 * the receiver does not have an STCB for that
1175 * packet, then authentication is based on
1176 * the contents of the COOKIE- ECHO chunk.
1177 */
1178 if (have_auth == 1 && chunk_num == 2)
1179 return NULL;
1180 break;
1181
1182 case SCTP_CID_ASCONF:
1183 if (have_auth || net->sctp.addip_noauth)
1184 asoc = __sctp_rcv_asconf_lookup(
1185 net, ch, laddr,
1186 sctp_hdr(skb)->source,
1187 transportp);
1188 default:
1189 break;
1190 }
1191
1192 if (asoc)
1193 break;
1194
1195 ch = (struct sctp_chunkhdr *)ch_end;
1196 chunk_num++;
1197 } while (ch_end < skb_tail_pointer(skb));
1198
1199 return asoc;
1200}
1201
1202/*
1203 * There are circumstances when we need to look inside the SCTP packet
1204 * for information to help us find the association. Examples
1205 * include looking inside of INIT/INIT-ACK chunks or after the AUTH
1206 * chunks.
1207 */
1208static struct sctp_association *__sctp_rcv_lookup_harder(struct net *net,
1209 struct sk_buff *skb,
1210 const union sctp_addr *laddr,
1211 struct sctp_transport **transportp)
1212{
1213 struct sctp_chunkhdr *ch;
1214
1215 /* We do not allow GSO frames here as we need to linearize and
1216 * then cannot guarantee frame boundaries. This shouldn't be an
1217 * issue as packets hitting this are mostly INIT or INIT-ACK and
1218 * those cannot be on GSO-style anyway.
1219 */
1220 if (skb_is_gso(skb) && skb_is_gso_sctp(skb))
1221 return NULL;
1222
1223 ch = (struct sctp_chunkhdr *)skb->data;
1224
1225 /* The code below will attempt to walk the chunk and extract
1226 * parameter information. Before we do that, we need to verify
1227 * that the chunk length doesn't cause overflow. Otherwise, we'll
1228 * walk off the end.
1229 */
1230 if (SCTP_PAD4(ntohs(ch->length)) > skb->len)
1231 return NULL;
1232
1233 /* If this is INIT/INIT-ACK look inside the chunk too. */
1234 if (ch->type == SCTP_CID_INIT || ch->type == SCTP_CID_INIT_ACK)
1235 return __sctp_rcv_init_lookup(net, skb, laddr, transportp);
1236
1237 return __sctp_rcv_walk_lookup(net, skb, laddr, transportp);
1238}
1239
1240/* Lookup an association for an inbound skb. */
1241static struct sctp_association *__sctp_rcv_lookup(struct net *net,
1242 struct sk_buff *skb,
1243 const union sctp_addr *paddr,
1244 const union sctp_addr *laddr,
1245 struct sctp_transport **transportp)
1246{
1247 struct sctp_association *asoc;
1248
1249 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1250 if (asoc)
1251 goto out;
1252
1253 /* Further lookup for INIT/INIT-ACK packets.
1254 * SCTP Implementors Guide, 2.18 Handling of address
1255 * parameters within the INIT or INIT-ACK.
1256 */
1257 asoc = __sctp_rcv_lookup_harder(net, skb, laddr, transportp);
1258 if (asoc)
1259 goto out;
1260
1261 if (paddr->sa.sa_family == AF_INET)
1262 pr_debug("sctp: asoc not found for src:%pI4:%d dst:%pI4:%d\n",
1263 &laddr->v4.sin_addr, ntohs(laddr->v4.sin_port),
1264 &paddr->v4.sin_addr, ntohs(paddr->v4.sin_port));
1265 else
1266 pr_debug("sctp: asoc not found for src:%pI6:%d dst:%pI6:%d\n",
1267 &laddr->v6.sin6_addr, ntohs(laddr->v6.sin6_port),
1268 &paddr->v6.sin6_addr, ntohs(paddr->v6.sin6_port));
1269
1270out:
1271 return asoc;
1272}