Loading...
1/*
2 * TCP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on:
9 * linux/net/ipv4/tcp.c
10 * linux/net/ipv4/tcp_input.c
11 * linux/net/ipv4/tcp_output.c
12 *
13 * Fixes:
14 * Hideaki YOSHIFUJI : sin6_scope_id support
15 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
16 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
17 * a single port at the same time.
18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
19 *
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
24 */
25
26#include <linux/bottom_half.h>
27#include <linux/module.h>
28#include <linux/errno.h>
29#include <linux/types.h>
30#include <linux/socket.h>
31#include <linux/sockios.h>
32#include <linux/net.h>
33#include <linux/jiffies.h>
34#include <linux/in.h>
35#include <linux/in6.h>
36#include <linux/netdevice.h>
37#include <linux/init.h>
38#include <linux/jhash.h>
39#include <linux/ipsec.h>
40#include <linux/times.h>
41#include <linux/slab.h>
42#include <linux/uaccess.h>
43#include <linux/ipv6.h>
44#include <linux/icmpv6.h>
45#include <linux/random.h>
46
47#include <net/tcp.h>
48#include <net/ndisc.h>
49#include <net/inet6_hashtables.h>
50#include <net/inet6_connection_sock.h>
51#include <net/ipv6.h>
52#include <net/transp_v6.h>
53#include <net/addrconf.h>
54#include <net/ip6_route.h>
55#include <net/ip6_checksum.h>
56#include <net/inet_ecn.h>
57#include <net/protocol.h>
58#include <net/xfrm.h>
59#include <net/snmp.h>
60#include <net/dsfield.h>
61#include <net/timewait_sock.h>
62#include <net/inet_common.h>
63#include <net/secure_seq.h>
64#include <net/busy_poll.h>
65
66#include <linux/proc_fs.h>
67#include <linux/seq_file.h>
68
69#include <crypto/hash.h>
70#include <linux/scatterlist.h>
71
72static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
73static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
74 struct request_sock *req);
75
76static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
77
78static const struct inet_connection_sock_af_ops ipv6_mapped;
79static const struct inet_connection_sock_af_ops ipv6_specific;
80#ifdef CONFIG_TCP_MD5SIG
81static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
82static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
83#else
84static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
85 const struct in6_addr *addr)
86{
87 return NULL;
88}
89#endif
90
91static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
92{
93 struct dst_entry *dst = skb_dst(skb);
94
95 if (dst && dst_hold_safe(dst)) {
96 const struct rt6_info *rt = (const struct rt6_info *)dst;
97
98 sk->sk_rx_dst = dst;
99 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
100 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
101 }
102}
103
104static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
105{
106 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
107 ipv6_hdr(skb)->saddr.s6_addr32,
108 tcp_hdr(skb)->dest,
109 tcp_hdr(skb)->source);
110}
111
112static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
113 int addr_len)
114{
115 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
116 struct inet_sock *inet = inet_sk(sk);
117 struct inet_connection_sock *icsk = inet_csk(sk);
118 struct ipv6_pinfo *np = inet6_sk(sk);
119 struct tcp_sock *tp = tcp_sk(sk);
120 struct in6_addr *saddr = NULL, *final_p, final;
121 struct ipv6_txoptions *opt;
122 struct flowi6 fl6;
123 struct dst_entry *dst;
124 int addr_type;
125 int err;
126
127 if (addr_len < SIN6_LEN_RFC2133)
128 return -EINVAL;
129
130 if (usin->sin6_family != AF_INET6)
131 return -EAFNOSUPPORT;
132
133 memset(&fl6, 0, sizeof(fl6));
134
135 if (np->sndflow) {
136 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
137 IP6_ECN_flow_init(fl6.flowlabel);
138 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
139 struct ip6_flowlabel *flowlabel;
140 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
141 if (!flowlabel)
142 return -EINVAL;
143 fl6_sock_release(flowlabel);
144 }
145 }
146
147 /*
148 * connect() to INADDR_ANY means loopback (BSD'ism).
149 */
150
151 if (ipv6_addr_any(&usin->sin6_addr))
152 usin->sin6_addr.s6_addr[15] = 0x1;
153
154 addr_type = ipv6_addr_type(&usin->sin6_addr);
155
156 if (addr_type & IPV6_ADDR_MULTICAST)
157 return -ENETUNREACH;
158
159 if (addr_type&IPV6_ADDR_LINKLOCAL) {
160 if (addr_len >= sizeof(struct sockaddr_in6) &&
161 usin->sin6_scope_id) {
162 /* If interface is set while binding, indices
163 * must coincide.
164 */
165 if (sk->sk_bound_dev_if &&
166 sk->sk_bound_dev_if != usin->sin6_scope_id)
167 return -EINVAL;
168
169 sk->sk_bound_dev_if = usin->sin6_scope_id;
170 }
171
172 /* Connect to link-local address requires an interface */
173 if (!sk->sk_bound_dev_if)
174 return -EINVAL;
175 }
176
177 if (tp->rx_opt.ts_recent_stamp &&
178 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
179 tp->rx_opt.ts_recent = 0;
180 tp->rx_opt.ts_recent_stamp = 0;
181 tp->write_seq = 0;
182 }
183
184 sk->sk_v6_daddr = usin->sin6_addr;
185 np->flow_label = fl6.flowlabel;
186
187 /*
188 * TCP over IPv4
189 */
190
191 if (addr_type == IPV6_ADDR_MAPPED) {
192 u32 exthdrlen = icsk->icsk_ext_hdr_len;
193 struct sockaddr_in sin;
194
195 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
196
197 if (__ipv6_only_sock(sk))
198 return -ENETUNREACH;
199
200 sin.sin_family = AF_INET;
201 sin.sin_port = usin->sin6_port;
202 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
203
204 icsk->icsk_af_ops = &ipv6_mapped;
205 sk->sk_backlog_rcv = tcp_v4_do_rcv;
206#ifdef CONFIG_TCP_MD5SIG
207 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
208#endif
209
210 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
211
212 if (err) {
213 icsk->icsk_ext_hdr_len = exthdrlen;
214 icsk->icsk_af_ops = &ipv6_specific;
215 sk->sk_backlog_rcv = tcp_v6_do_rcv;
216#ifdef CONFIG_TCP_MD5SIG
217 tp->af_specific = &tcp_sock_ipv6_specific;
218#endif
219 goto failure;
220 }
221 np->saddr = sk->sk_v6_rcv_saddr;
222
223 return err;
224 }
225
226 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
227 saddr = &sk->sk_v6_rcv_saddr;
228
229 fl6.flowi6_proto = IPPROTO_TCP;
230 fl6.daddr = sk->sk_v6_daddr;
231 fl6.saddr = saddr ? *saddr : np->saddr;
232 fl6.flowi6_oif = sk->sk_bound_dev_if;
233 fl6.flowi6_mark = sk->sk_mark;
234 fl6.fl6_dport = usin->sin6_port;
235 fl6.fl6_sport = inet->inet_sport;
236
237 opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
238 final_p = fl6_update_dst(&fl6, opt, &final);
239
240 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
241
242 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
243 if (IS_ERR(dst)) {
244 err = PTR_ERR(dst);
245 goto failure;
246 }
247
248 if (!saddr) {
249 saddr = &fl6.saddr;
250 sk->sk_v6_rcv_saddr = *saddr;
251 }
252
253 /* set the source address */
254 np->saddr = *saddr;
255 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
256
257 sk->sk_gso_type = SKB_GSO_TCPV6;
258 ip6_dst_store(sk, dst, NULL, NULL);
259
260 if (tcp_death_row.sysctl_tw_recycle &&
261 !tp->rx_opt.ts_recent_stamp &&
262 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
263 tcp_fetch_timewait_stamp(sk, dst);
264
265 icsk->icsk_ext_hdr_len = 0;
266 if (opt)
267 icsk->icsk_ext_hdr_len = opt->opt_flen +
268 opt->opt_nflen;
269
270 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
271
272 inet->inet_dport = usin->sin6_port;
273
274 tcp_set_state(sk, TCP_SYN_SENT);
275 err = inet6_hash_connect(&tcp_death_row, sk);
276 if (err)
277 goto late_failure;
278
279 sk_set_txhash(sk);
280
281 if (!tp->write_seq && likely(!tp->repair))
282 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
283 sk->sk_v6_daddr.s6_addr32,
284 inet->inet_sport,
285 inet->inet_dport);
286
287 err = tcp_connect(sk);
288 if (err)
289 goto late_failure;
290
291 return 0;
292
293late_failure:
294 tcp_set_state(sk, TCP_CLOSE);
295 __sk_dst_reset(sk);
296failure:
297 inet->inet_dport = 0;
298 sk->sk_route_caps = 0;
299 return err;
300}
301
302static void tcp_v6_mtu_reduced(struct sock *sk)
303{
304 struct dst_entry *dst;
305
306 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
307 return;
308
309 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
310 if (!dst)
311 return;
312
313 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
314 tcp_sync_mss(sk, dst_mtu(dst));
315 tcp_simple_retransmit(sk);
316 }
317}
318
319static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
320 u8 type, u8 code, int offset, __be32 info)
321{
322 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
323 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
324 struct net *net = dev_net(skb->dev);
325 struct request_sock *fastopen;
326 struct ipv6_pinfo *np;
327 struct tcp_sock *tp;
328 __u32 seq, snd_una;
329 struct sock *sk;
330 bool fatal;
331 int err;
332
333 sk = __inet6_lookup_established(net, &tcp_hashinfo,
334 &hdr->daddr, th->dest,
335 &hdr->saddr, ntohs(th->source),
336 skb->dev->ifindex);
337
338 if (!sk) {
339 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
340 ICMP6_MIB_INERRORS);
341 return;
342 }
343
344 if (sk->sk_state == TCP_TIME_WAIT) {
345 inet_twsk_put(inet_twsk(sk));
346 return;
347 }
348 seq = ntohl(th->seq);
349 fatal = icmpv6_err_convert(type, code, &err);
350 if (sk->sk_state == TCP_NEW_SYN_RECV)
351 return tcp_req_err(sk, seq, fatal);
352
353 bh_lock_sock(sk);
354 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
355 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
356
357 if (sk->sk_state == TCP_CLOSE)
358 goto out;
359
360 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
361 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
362 goto out;
363 }
364
365 tp = tcp_sk(sk);
366 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
367 fastopen = tp->fastopen_rsk;
368 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
369 if (sk->sk_state != TCP_LISTEN &&
370 !between(seq, snd_una, tp->snd_nxt)) {
371 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
372 goto out;
373 }
374
375 np = inet6_sk(sk);
376
377 if (type == NDISC_REDIRECT) {
378 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
379
380 if (dst)
381 dst->ops->redirect(dst, sk, skb);
382 goto out;
383 }
384
385 if (type == ICMPV6_PKT_TOOBIG) {
386 /* We are not interested in TCP_LISTEN and open_requests
387 * (SYN-ACKs send out by Linux are always <576bytes so
388 * they should go through unfragmented).
389 */
390 if (sk->sk_state == TCP_LISTEN)
391 goto out;
392
393 if (!ip6_sk_accept_pmtu(sk))
394 goto out;
395
396 tp->mtu_info = ntohl(info);
397 if (!sock_owned_by_user(sk))
398 tcp_v6_mtu_reduced(sk);
399 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
400 &tp->tsq_flags))
401 sock_hold(sk);
402 goto out;
403 }
404
405
406 /* Might be for an request_sock */
407 switch (sk->sk_state) {
408 case TCP_SYN_SENT:
409 case TCP_SYN_RECV:
410 /* Only in fast or simultaneous open. If a fast open socket is
411 * is already accepted it is treated as a connected one below.
412 */
413 if (fastopen && !fastopen->sk)
414 break;
415
416 if (!sock_owned_by_user(sk)) {
417 sk->sk_err = err;
418 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
419
420 tcp_done(sk);
421 } else
422 sk->sk_err_soft = err;
423 goto out;
424 }
425
426 if (!sock_owned_by_user(sk) && np->recverr) {
427 sk->sk_err = err;
428 sk->sk_error_report(sk);
429 } else
430 sk->sk_err_soft = err;
431
432out:
433 bh_unlock_sock(sk);
434 sock_put(sk);
435}
436
437
438static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
439 struct flowi *fl,
440 struct request_sock *req,
441 struct tcp_fastopen_cookie *foc,
442 bool attach_req)
443{
444 struct inet_request_sock *ireq = inet_rsk(req);
445 struct ipv6_pinfo *np = inet6_sk(sk);
446 struct flowi6 *fl6 = &fl->u.ip6;
447 struct sk_buff *skb;
448 int err = -ENOMEM;
449
450 /* First, grab a route. */
451 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
452 IPPROTO_TCP)) == NULL)
453 goto done;
454
455 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
456
457 if (skb) {
458 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
459 &ireq->ir_v6_rmt_addr);
460
461 fl6->daddr = ireq->ir_v6_rmt_addr;
462 if (np->repflow && ireq->pktopts)
463 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
464
465 rcu_read_lock();
466 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
467 np->tclass);
468 rcu_read_unlock();
469 err = net_xmit_eval(err);
470 }
471
472done:
473 return err;
474}
475
476
477static void tcp_v6_reqsk_destructor(struct request_sock *req)
478{
479 kfree_skb(inet_rsk(req)->pktopts);
480}
481
482#ifdef CONFIG_TCP_MD5SIG
483static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
484 const struct in6_addr *addr)
485{
486 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
487}
488
489static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
490 const struct sock *addr_sk)
491{
492 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
493}
494
495static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
496 int optlen)
497{
498 struct tcp_md5sig cmd;
499 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
500
501 if (optlen < sizeof(cmd))
502 return -EINVAL;
503
504 if (copy_from_user(&cmd, optval, sizeof(cmd)))
505 return -EFAULT;
506
507 if (sin6->sin6_family != AF_INET6)
508 return -EINVAL;
509
510 if (!cmd.tcpm_keylen) {
511 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
512 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
513 AF_INET);
514 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
515 AF_INET6);
516 }
517
518 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
519 return -EINVAL;
520
521 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
522 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
523 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
524
525 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
526 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
527}
528
529static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
530 const struct in6_addr *daddr,
531 const struct in6_addr *saddr, int nbytes)
532{
533 struct tcp6_pseudohdr *bp;
534 struct scatterlist sg;
535
536 bp = &hp->md5_blk.ip6;
537 /* 1. TCP pseudo-header (RFC2460) */
538 bp->saddr = *saddr;
539 bp->daddr = *daddr;
540 bp->protocol = cpu_to_be32(IPPROTO_TCP);
541 bp->len = cpu_to_be32(nbytes);
542
543 sg_init_one(&sg, bp, sizeof(*bp));
544 ahash_request_set_crypt(hp->md5_req, &sg, NULL, sizeof(*bp));
545 return crypto_ahash_update(hp->md5_req);
546}
547
548static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
549 const struct in6_addr *daddr, struct in6_addr *saddr,
550 const struct tcphdr *th)
551{
552 struct tcp_md5sig_pool *hp;
553 struct ahash_request *req;
554
555 hp = tcp_get_md5sig_pool();
556 if (!hp)
557 goto clear_hash_noput;
558 req = hp->md5_req;
559
560 if (crypto_ahash_init(req))
561 goto clear_hash;
562 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
563 goto clear_hash;
564 if (tcp_md5_hash_header(hp, th))
565 goto clear_hash;
566 if (tcp_md5_hash_key(hp, key))
567 goto clear_hash;
568 ahash_request_set_crypt(req, NULL, md5_hash, 0);
569 if (crypto_ahash_final(req))
570 goto clear_hash;
571
572 tcp_put_md5sig_pool();
573 return 0;
574
575clear_hash:
576 tcp_put_md5sig_pool();
577clear_hash_noput:
578 memset(md5_hash, 0, 16);
579 return 1;
580}
581
582static int tcp_v6_md5_hash_skb(char *md5_hash,
583 const struct tcp_md5sig_key *key,
584 const struct sock *sk,
585 const struct sk_buff *skb)
586{
587 const struct in6_addr *saddr, *daddr;
588 struct tcp_md5sig_pool *hp;
589 struct ahash_request *req;
590 const struct tcphdr *th = tcp_hdr(skb);
591
592 if (sk) { /* valid for establish/request sockets */
593 saddr = &sk->sk_v6_rcv_saddr;
594 daddr = &sk->sk_v6_daddr;
595 } else {
596 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
597 saddr = &ip6h->saddr;
598 daddr = &ip6h->daddr;
599 }
600
601 hp = tcp_get_md5sig_pool();
602 if (!hp)
603 goto clear_hash_noput;
604 req = hp->md5_req;
605
606 if (crypto_ahash_init(req))
607 goto clear_hash;
608
609 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
610 goto clear_hash;
611 if (tcp_md5_hash_header(hp, th))
612 goto clear_hash;
613 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
614 goto clear_hash;
615 if (tcp_md5_hash_key(hp, key))
616 goto clear_hash;
617 ahash_request_set_crypt(req, NULL, md5_hash, 0);
618 if (crypto_ahash_final(req))
619 goto clear_hash;
620
621 tcp_put_md5sig_pool();
622 return 0;
623
624clear_hash:
625 tcp_put_md5sig_pool();
626clear_hash_noput:
627 memset(md5_hash, 0, 16);
628 return 1;
629}
630
631#endif
632
633static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
634 const struct sk_buff *skb)
635{
636#ifdef CONFIG_TCP_MD5SIG
637 const __u8 *hash_location = NULL;
638 struct tcp_md5sig_key *hash_expected;
639 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
640 const struct tcphdr *th = tcp_hdr(skb);
641 int genhash;
642 u8 newhash[16];
643
644 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
645 hash_location = tcp_parse_md5sig_option(th);
646
647 /* We've parsed the options - do we have a hash? */
648 if (!hash_expected && !hash_location)
649 return false;
650
651 if (hash_expected && !hash_location) {
652 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
653 return true;
654 }
655
656 if (!hash_expected && hash_location) {
657 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
658 return true;
659 }
660
661 /* check the signature */
662 genhash = tcp_v6_md5_hash_skb(newhash,
663 hash_expected,
664 NULL, skb);
665
666 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
667 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
668 genhash ? "failed" : "mismatch",
669 &ip6h->saddr, ntohs(th->source),
670 &ip6h->daddr, ntohs(th->dest));
671 return true;
672 }
673#endif
674 return false;
675}
676
677static void tcp_v6_init_req(struct request_sock *req,
678 const struct sock *sk_listener,
679 struct sk_buff *skb)
680{
681 struct inet_request_sock *ireq = inet_rsk(req);
682 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
683
684 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
685 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
686
687 /* So that link locals have meaning */
688 if (!sk_listener->sk_bound_dev_if &&
689 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
690 ireq->ir_iif = tcp_v6_iif(skb);
691
692 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
693 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
694 np->rxopt.bits.rxinfo ||
695 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
696 np->rxopt.bits.rxohlim || np->repflow)) {
697 atomic_inc(&skb->users);
698 ireq->pktopts = skb;
699 }
700}
701
702static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
703 struct flowi *fl,
704 const struct request_sock *req,
705 bool *strict)
706{
707 if (strict)
708 *strict = true;
709 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
710}
711
712struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
713 .family = AF_INET6,
714 .obj_size = sizeof(struct tcp6_request_sock),
715 .rtx_syn_ack = tcp_rtx_synack,
716 .send_ack = tcp_v6_reqsk_send_ack,
717 .destructor = tcp_v6_reqsk_destructor,
718 .send_reset = tcp_v6_send_reset,
719 .syn_ack_timeout = tcp_syn_ack_timeout,
720};
721
722static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
723 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
724 sizeof(struct ipv6hdr),
725#ifdef CONFIG_TCP_MD5SIG
726 .req_md5_lookup = tcp_v6_md5_lookup,
727 .calc_md5_hash = tcp_v6_md5_hash_skb,
728#endif
729 .init_req = tcp_v6_init_req,
730#ifdef CONFIG_SYN_COOKIES
731 .cookie_init_seq = cookie_v6_init_sequence,
732#endif
733 .route_req = tcp_v6_route_req,
734 .init_seq = tcp_v6_init_sequence,
735 .send_synack = tcp_v6_send_synack,
736};
737
738static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
739 u32 ack, u32 win, u32 tsval, u32 tsecr,
740 int oif, struct tcp_md5sig_key *key, int rst,
741 u8 tclass, u32 label)
742{
743 const struct tcphdr *th = tcp_hdr(skb);
744 struct tcphdr *t1;
745 struct sk_buff *buff;
746 struct flowi6 fl6;
747 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
748 struct sock *ctl_sk = net->ipv6.tcp_sk;
749 unsigned int tot_len = sizeof(struct tcphdr);
750 struct dst_entry *dst;
751 __be32 *topt;
752
753 if (tsecr)
754 tot_len += TCPOLEN_TSTAMP_ALIGNED;
755#ifdef CONFIG_TCP_MD5SIG
756 if (key)
757 tot_len += TCPOLEN_MD5SIG_ALIGNED;
758#endif
759
760 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
761 GFP_ATOMIC);
762 if (!buff)
763 return;
764
765 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
766
767 t1 = (struct tcphdr *) skb_push(buff, tot_len);
768 skb_reset_transport_header(buff);
769
770 /* Swap the send and the receive. */
771 memset(t1, 0, sizeof(*t1));
772 t1->dest = th->source;
773 t1->source = th->dest;
774 t1->doff = tot_len / 4;
775 t1->seq = htonl(seq);
776 t1->ack_seq = htonl(ack);
777 t1->ack = !rst || !th->ack;
778 t1->rst = rst;
779 t1->window = htons(win);
780
781 topt = (__be32 *)(t1 + 1);
782
783 if (tsecr) {
784 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
785 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
786 *topt++ = htonl(tsval);
787 *topt++ = htonl(tsecr);
788 }
789
790#ifdef CONFIG_TCP_MD5SIG
791 if (key) {
792 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
793 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
794 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
795 &ipv6_hdr(skb)->saddr,
796 &ipv6_hdr(skb)->daddr, t1);
797 }
798#endif
799
800 memset(&fl6, 0, sizeof(fl6));
801 fl6.daddr = ipv6_hdr(skb)->saddr;
802 fl6.saddr = ipv6_hdr(skb)->daddr;
803 fl6.flowlabel = label;
804
805 buff->ip_summed = CHECKSUM_PARTIAL;
806 buff->csum = 0;
807
808 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
809
810 fl6.flowi6_proto = IPPROTO_TCP;
811 if (rt6_need_strict(&fl6.daddr) && !oif)
812 fl6.flowi6_oif = tcp_v6_iif(skb);
813 else {
814 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
815 oif = skb->skb_iif;
816
817 fl6.flowi6_oif = oif;
818 }
819
820 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
821 fl6.fl6_dport = t1->dest;
822 fl6.fl6_sport = t1->source;
823 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
824
825 /* Pass a socket to ip6_dst_lookup either it is for RST
826 * Underlying function will use this to retrieve the network
827 * namespace
828 */
829 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
830 if (!IS_ERR(dst)) {
831 skb_dst_set(buff, dst);
832 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
833 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
834 if (rst)
835 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
836 return;
837 }
838
839 kfree_skb(buff);
840}
841
842static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
843{
844 const struct tcphdr *th = tcp_hdr(skb);
845 u32 seq = 0, ack_seq = 0;
846 struct tcp_md5sig_key *key = NULL;
847#ifdef CONFIG_TCP_MD5SIG
848 const __u8 *hash_location = NULL;
849 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
850 unsigned char newhash[16];
851 int genhash;
852 struct sock *sk1 = NULL;
853#endif
854 int oif;
855
856 if (th->rst)
857 return;
858
859 /* If sk not NULL, it means we did a successful lookup and incoming
860 * route had to be correct. prequeue might have dropped our dst.
861 */
862 if (!sk && !ipv6_unicast_destination(skb))
863 return;
864
865#ifdef CONFIG_TCP_MD5SIG
866 hash_location = tcp_parse_md5sig_option(th);
867 if (sk && sk_fullsock(sk)) {
868 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
869 } else if (hash_location) {
870 /*
871 * active side is lost. Try to find listening socket through
872 * source port, and then find md5 key through listening socket.
873 * we are not loose security here:
874 * Incoming packet is checked with md5 hash with finding key,
875 * no RST generated if md5 hash doesn't match.
876 */
877 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
878 &tcp_hashinfo, NULL, 0,
879 &ipv6h->saddr,
880 th->source, &ipv6h->daddr,
881 ntohs(th->source), tcp_v6_iif(skb));
882 if (!sk1)
883 return;
884
885 rcu_read_lock();
886 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
887 if (!key)
888 goto release_sk1;
889
890 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
891 if (genhash || memcmp(hash_location, newhash, 16) != 0)
892 goto release_sk1;
893 }
894#endif
895
896 if (th->ack)
897 seq = ntohl(th->ack_seq);
898 else
899 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
900 (th->doff << 2);
901
902 oif = sk ? sk->sk_bound_dev_if : 0;
903 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
904
905#ifdef CONFIG_TCP_MD5SIG
906release_sk1:
907 if (sk1) {
908 rcu_read_unlock();
909 sock_put(sk1);
910 }
911#endif
912}
913
914static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
915 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
916 struct tcp_md5sig_key *key, u8 tclass,
917 u32 label)
918{
919 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
920 tclass, label);
921}
922
923static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
924{
925 struct inet_timewait_sock *tw = inet_twsk(sk);
926 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
927
928 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
929 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
930 tcp_time_stamp + tcptw->tw_ts_offset,
931 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
932 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
933
934 inet_twsk_put(tw);
935}
936
937static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
938 struct request_sock *req)
939{
940 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
941 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
942 */
943 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
944 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
945 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
946 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
947 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
948 0, 0);
949}
950
951
952static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
953{
954#ifdef CONFIG_SYN_COOKIES
955 const struct tcphdr *th = tcp_hdr(skb);
956
957 if (!th->syn)
958 sk = cookie_v6_check(sk, skb);
959#endif
960 return sk;
961}
962
963static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
964{
965 if (skb->protocol == htons(ETH_P_IP))
966 return tcp_v4_conn_request(sk, skb);
967
968 if (!ipv6_unicast_destination(skb))
969 goto drop;
970
971 return tcp_conn_request(&tcp6_request_sock_ops,
972 &tcp_request_sock_ipv6_ops, sk, skb);
973
974drop:
975 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
976 return 0; /* don't send reset */
977}
978
979static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
980 struct request_sock *req,
981 struct dst_entry *dst,
982 struct request_sock *req_unhash,
983 bool *own_req)
984{
985 struct inet_request_sock *ireq;
986 struct ipv6_pinfo *newnp;
987 const struct ipv6_pinfo *np = inet6_sk(sk);
988 struct ipv6_txoptions *opt;
989 struct tcp6_sock *newtcp6sk;
990 struct inet_sock *newinet;
991 struct tcp_sock *newtp;
992 struct sock *newsk;
993#ifdef CONFIG_TCP_MD5SIG
994 struct tcp_md5sig_key *key;
995#endif
996 struct flowi6 fl6;
997
998 if (skb->protocol == htons(ETH_P_IP)) {
999 /*
1000 * v6 mapped
1001 */
1002
1003 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1004 req_unhash, own_req);
1005
1006 if (!newsk)
1007 return NULL;
1008
1009 newtcp6sk = (struct tcp6_sock *)newsk;
1010 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1011
1012 newinet = inet_sk(newsk);
1013 newnp = inet6_sk(newsk);
1014 newtp = tcp_sk(newsk);
1015
1016 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1017
1018 newnp->saddr = newsk->sk_v6_rcv_saddr;
1019
1020 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1021 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1022#ifdef CONFIG_TCP_MD5SIG
1023 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1024#endif
1025
1026 newnp->ipv6_ac_list = NULL;
1027 newnp->ipv6_fl_list = NULL;
1028 newnp->pktoptions = NULL;
1029 newnp->opt = NULL;
1030 newnp->mcast_oif = tcp_v6_iif(skb);
1031 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1032 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1033 if (np->repflow)
1034 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1035
1036 /*
1037 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1038 * here, tcp_create_openreq_child now does this for us, see the comment in
1039 * that function for the gory details. -acme
1040 */
1041
1042 /* It is tricky place. Until this moment IPv4 tcp
1043 worked with IPv6 icsk.icsk_af_ops.
1044 Sync it now.
1045 */
1046 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1047
1048 return newsk;
1049 }
1050
1051 ireq = inet_rsk(req);
1052
1053 if (sk_acceptq_is_full(sk))
1054 goto out_overflow;
1055
1056 if (!dst) {
1057 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1058 if (!dst)
1059 goto out;
1060 }
1061
1062 newsk = tcp_create_openreq_child(sk, req, skb);
1063 if (!newsk)
1064 goto out_nonewsk;
1065
1066 /*
1067 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1068 * count here, tcp_create_openreq_child now does this for us, see the
1069 * comment in that function for the gory details. -acme
1070 */
1071
1072 newsk->sk_gso_type = SKB_GSO_TCPV6;
1073 ip6_dst_store(newsk, dst, NULL, NULL);
1074 inet6_sk_rx_dst_set(newsk, skb);
1075
1076 newtcp6sk = (struct tcp6_sock *)newsk;
1077 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1078
1079 newtp = tcp_sk(newsk);
1080 newinet = inet_sk(newsk);
1081 newnp = inet6_sk(newsk);
1082
1083 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1084
1085 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1086 newnp->saddr = ireq->ir_v6_loc_addr;
1087 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1088 newsk->sk_bound_dev_if = ireq->ir_iif;
1089
1090 /* Now IPv6 options...
1091
1092 First: no IPv4 options.
1093 */
1094 newinet->inet_opt = NULL;
1095 newnp->ipv6_ac_list = NULL;
1096 newnp->ipv6_fl_list = NULL;
1097
1098 /* Clone RX bits */
1099 newnp->rxopt.all = np->rxopt.all;
1100
1101 newnp->pktoptions = NULL;
1102 newnp->opt = NULL;
1103 newnp->mcast_oif = tcp_v6_iif(skb);
1104 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1105 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1106 if (np->repflow)
1107 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1108
1109 /* Clone native IPv6 options from listening socket (if any)
1110
1111 Yes, keeping reference count would be much more clever,
1112 but we make one more one thing there: reattach optmem
1113 to newsk.
1114 */
1115 opt = rcu_dereference(np->opt);
1116 if (opt) {
1117 opt = ipv6_dup_options(newsk, opt);
1118 RCU_INIT_POINTER(newnp->opt, opt);
1119 }
1120 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1121 if (opt)
1122 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1123 opt->opt_flen;
1124
1125 tcp_ca_openreq_child(newsk, dst);
1126
1127 tcp_sync_mss(newsk, dst_mtu(dst));
1128 newtp->advmss = dst_metric_advmss(dst);
1129 if (tcp_sk(sk)->rx_opt.user_mss &&
1130 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1131 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1132
1133 tcp_initialize_rcv_mss(newsk);
1134
1135 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1136 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1137
1138#ifdef CONFIG_TCP_MD5SIG
1139 /* Copy over the MD5 key from the original socket */
1140 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1141 if (key) {
1142 /* We're using one, so create a matching key
1143 * on the newsk structure. If we fail to get
1144 * memory, then we end up not copying the key
1145 * across. Shucks.
1146 */
1147 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1148 AF_INET6, key->key, key->keylen,
1149 sk_gfp_mask(sk, GFP_ATOMIC));
1150 }
1151#endif
1152
1153 if (__inet_inherit_port(sk, newsk) < 0) {
1154 inet_csk_prepare_forced_close(newsk);
1155 tcp_done(newsk);
1156 goto out;
1157 }
1158 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1159 if (*own_req) {
1160 tcp_move_syn(newtp, req);
1161
1162 /* Clone pktoptions received with SYN, if we own the req */
1163 if (ireq->pktopts) {
1164 newnp->pktoptions = skb_clone(ireq->pktopts,
1165 sk_gfp_mask(sk, GFP_ATOMIC));
1166 consume_skb(ireq->pktopts);
1167 ireq->pktopts = NULL;
1168 if (newnp->pktoptions)
1169 skb_set_owner_r(newnp->pktoptions, newsk);
1170 }
1171 }
1172
1173 return newsk;
1174
1175out_overflow:
1176 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1177out_nonewsk:
1178 dst_release(dst);
1179out:
1180 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1181 return NULL;
1182}
1183
1184/* The socket must have it's spinlock held when we get
1185 * here, unless it is a TCP_LISTEN socket.
1186 *
1187 * We have a potential double-lock case here, so even when
1188 * doing backlog processing we use the BH locking scheme.
1189 * This is because we cannot sleep with the original spinlock
1190 * held.
1191 */
1192static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1193{
1194 struct ipv6_pinfo *np = inet6_sk(sk);
1195 struct tcp_sock *tp;
1196 struct sk_buff *opt_skb = NULL;
1197
1198 /* Imagine: socket is IPv6. IPv4 packet arrives,
1199 goes to IPv4 receive handler and backlogged.
1200 From backlog it always goes here. Kerboom...
1201 Fortunately, tcp_rcv_established and rcv_established
1202 handle them correctly, but it is not case with
1203 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1204 */
1205
1206 if (skb->protocol == htons(ETH_P_IP))
1207 return tcp_v4_do_rcv(sk, skb);
1208
1209 if (sk_filter(sk, skb))
1210 goto discard;
1211
1212 /*
1213 * socket locking is here for SMP purposes as backlog rcv
1214 * is currently called with bh processing disabled.
1215 */
1216
1217 /* Do Stevens' IPV6_PKTOPTIONS.
1218
1219 Yes, guys, it is the only place in our code, where we
1220 may make it not affecting IPv4.
1221 The rest of code is protocol independent,
1222 and I do not like idea to uglify IPv4.
1223
1224 Actually, all the idea behind IPV6_PKTOPTIONS
1225 looks not very well thought. For now we latch
1226 options, received in the last packet, enqueued
1227 by tcp. Feel free to propose better solution.
1228 --ANK (980728)
1229 */
1230 if (np->rxopt.all)
1231 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1232
1233 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1234 struct dst_entry *dst = sk->sk_rx_dst;
1235
1236 sock_rps_save_rxhash(sk, skb);
1237 sk_mark_napi_id(sk, skb);
1238 if (dst) {
1239 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1240 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1241 dst_release(dst);
1242 sk->sk_rx_dst = NULL;
1243 }
1244 }
1245
1246 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1247 if (opt_skb)
1248 goto ipv6_pktoptions;
1249 return 0;
1250 }
1251
1252 if (tcp_checksum_complete(skb))
1253 goto csum_err;
1254
1255 if (sk->sk_state == TCP_LISTEN) {
1256 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1257
1258 if (!nsk)
1259 goto discard;
1260
1261 if (nsk != sk) {
1262 sock_rps_save_rxhash(nsk, skb);
1263 sk_mark_napi_id(nsk, skb);
1264 if (tcp_child_process(sk, nsk, skb))
1265 goto reset;
1266 if (opt_skb)
1267 __kfree_skb(opt_skb);
1268 return 0;
1269 }
1270 } else
1271 sock_rps_save_rxhash(sk, skb);
1272
1273 if (tcp_rcv_state_process(sk, skb))
1274 goto reset;
1275 if (opt_skb)
1276 goto ipv6_pktoptions;
1277 return 0;
1278
1279reset:
1280 tcp_v6_send_reset(sk, skb);
1281discard:
1282 if (opt_skb)
1283 __kfree_skb(opt_skb);
1284 kfree_skb(skb);
1285 return 0;
1286csum_err:
1287 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1288 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1289 goto discard;
1290
1291
1292ipv6_pktoptions:
1293 /* Do you ask, what is it?
1294
1295 1. skb was enqueued by tcp.
1296 2. skb is added to tail of read queue, rather than out of order.
1297 3. socket is not in passive state.
1298 4. Finally, it really contains options, which user wants to receive.
1299 */
1300 tp = tcp_sk(sk);
1301 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1302 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1303 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1304 np->mcast_oif = tcp_v6_iif(opt_skb);
1305 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1306 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1307 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1308 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1309 if (np->repflow)
1310 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1311 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1312 skb_set_owner_r(opt_skb, sk);
1313 opt_skb = xchg(&np->pktoptions, opt_skb);
1314 } else {
1315 __kfree_skb(opt_skb);
1316 opt_skb = xchg(&np->pktoptions, NULL);
1317 }
1318 }
1319
1320 kfree_skb(opt_skb);
1321 return 0;
1322}
1323
1324static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1325 const struct tcphdr *th)
1326{
1327 /* This is tricky: we move IP6CB at its correct location into
1328 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1329 * _decode_session6() uses IP6CB().
1330 * barrier() makes sure compiler won't play aliasing games.
1331 */
1332 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1333 sizeof(struct inet6_skb_parm));
1334 barrier();
1335
1336 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1337 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1338 skb->len - th->doff*4);
1339 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1340 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1341 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1342 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1343 TCP_SKB_CB(skb)->sacked = 0;
1344}
1345
1346static void tcp_v6_restore_cb(struct sk_buff *skb)
1347{
1348 /* We need to move header back to the beginning if xfrm6_policy_check()
1349 * and tcp_v6_fill_cb() are going to be called again.
1350 */
1351 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1352 sizeof(struct inet6_skb_parm));
1353}
1354
1355static int tcp_v6_rcv(struct sk_buff *skb)
1356{
1357 const struct tcphdr *th;
1358 const struct ipv6hdr *hdr;
1359 struct sock *sk;
1360 int ret;
1361 struct net *net = dev_net(skb->dev);
1362
1363 if (skb->pkt_type != PACKET_HOST)
1364 goto discard_it;
1365
1366 /*
1367 * Count it even if it's bad.
1368 */
1369 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1370
1371 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1372 goto discard_it;
1373
1374 th = tcp_hdr(skb);
1375
1376 if (th->doff < sizeof(struct tcphdr)/4)
1377 goto bad_packet;
1378 if (!pskb_may_pull(skb, th->doff*4))
1379 goto discard_it;
1380
1381 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1382 goto csum_error;
1383
1384 th = tcp_hdr(skb);
1385 hdr = ipv6_hdr(skb);
1386
1387lookup:
1388 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1389 th->source, th->dest, inet6_iif(skb));
1390 if (!sk)
1391 goto no_tcp_socket;
1392
1393process:
1394 if (sk->sk_state == TCP_TIME_WAIT)
1395 goto do_time_wait;
1396
1397 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1398 struct request_sock *req = inet_reqsk(sk);
1399 struct sock *nsk;
1400
1401 sk = req->rsk_listener;
1402 tcp_v6_fill_cb(skb, hdr, th);
1403 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1404 reqsk_put(req);
1405 goto discard_it;
1406 }
1407 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1408 inet_csk_reqsk_queue_drop_and_put(sk, req);
1409 goto lookup;
1410 }
1411 sock_hold(sk);
1412 nsk = tcp_check_req(sk, skb, req, false);
1413 if (!nsk) {
1414 reqsk_put(req);
1415 goto discard_and_relse;
1416 }
1417 if (nsk == sk) {
1418 reqsk_put(req);
1419 tcp_v6_restore_cb(skb);
1420 } else if (tcp_child_process(sk, nsk, skb)) {
1421 tcp_v6_send_reset(nsk, skb);
1422 goto discard_and_relse;
1423 } else {
1424 sock_put(sk);
1425 return 0;
1426 }
1427 }
1428 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1429 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1430 goto discard_and_relse;
1431 }
1432
1433 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1434 goto discard_and_relse;
1435
1436 tcp_v6_fill_cb(skb, hdr, th);
1437
1438 if (tcp_v6_inbound_md5_hash(sk, skb))
1439 goto discard_and_relse;
1440
1441 if (sk_filter(sk, skb))
1442 goto discard_and_relse;
1443
1444 skb->dev = NULL;
1445
1446 if (sk->sk_state == TCP_LISTEN) {
1447 ret = tcp_v6_do_rcv(sk, skb);
1448 goto put_and_return;
1449 }
1450
1451 sk_incoming_cpu_update(sk);
1452
1453 bh_lock_sock_nested(sk);
1454 tcp_segs_in(tcp_sk(sk), skb);
1455 ret = 0;
1456 if (!sock_owned_by_user(sk)) {
1457 if (!tcp_prequeue(sk, skb))
1458 ret = tcp_v6_do_rcv(sk, skb);
1459 } else if (unlikely(sk_add_backlog(sk, skb,
1460 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1461 bh_unlock_sock(sk);
1462 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1463 goto discard_and_relse;
1464 }
1465 bh_unlock_sock(sk);
1466
1467put_and_return:
1468 sock_put(sk);
1469 return ret ? -1 : 0;
1470
1471no_tcp_socket:
1472 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1473 goto discard_it;
1474
1475 tcp_v6_fill_cb(skb, hdr, th);
1476
1477 if (tcp_checksum_complete(skb)) {
1478csum_error:
1479 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1480bad_packet:
1481 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1482 } else {
1483 tcp_v6_send_reset(NULL, skb);
1484 }
1485
1486discard_it:
1487 kfree_skb(skb);
1488 return 0;
1489
1490discard_and_relse:
1491 sock_put(sk);
1492 goto discard_it;
1493
1494do_time_wait:
1495 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1496 inet_twsk_put(inet_twsk(sk));
1497 goto discard_it;
1498 }
1499
1500 tcp_v6_fill_cb(skb, hdr, th);
1501
1502 if (tcp_checksum_complete(skb)) {
1503 inet_twsk_put(inet_twsk(sk));
1504 goto csum_error;
1505 }
1506
1507 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1508 case TCP_TW_SYN:
1509 {
1510 struct sock *sk2;
1511
1512 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1513 skb, __tcp_hdrlen(th),
1514 &ipv6_hdr(skb)->saddr, th->source,
1515 &ipv6_hdr(skb)->daddr,
1516 ntohs(th->dest), tcp_v6_iif(skb));
1517 if (sk2) {
1518 struct inet_timewait_sock *tw = inet_twsk(sk);
1519 inet_twsk_deschedule_put(tw);
1520 sk = sk2;
1521 tcp_v6_restore_cb(skb);
1522 goto process;
1523 }
1524 /* Fall through to ACK */
1525 }
1526 case TCP_TW_ACK:
1527 tcp_v6_timewait_ack(sk, skb);
1528 break;
1529 case TCP_TW_RST:
1530 tcp_v6_restore_cb(skb);
1531 tcp_v6_send_reset(sk, skb);
1532 inet_twsk_deschedule_put(inet_twsk(sk));
1533 goto discard_it;
1534 case TCP_TW_SUCCESS:
1535 ;
1536 }
1537 goto discard_it;
1538}
1539
1540static void tcp_v6_early_demux(struct sk_buff *skb)
1541{
1542 const struct ipv6hdr *hdr;
1543 const struct tcphdr *th;
1544 struct sock *sk;
1545
1546 if (skb->pkt_type != PACKET_HOST)
1547 return;
1548
1549 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1550 return;
1551
1552 hdr = ipv6_hdr(skb);
1553 th = tcp_hdr(skb);
1554
1555 if (th->doff < sizeof(struct tcphdr) / 4)
1556 return;
1557
1558 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1559 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1560 &hdr->saddr, th->source,
1561 &hdr->daddr, ntohs(th->dest),
1562 inet6_iif(skb));
1563 if (sk) {
1564 skb->sk = sk;
1565 skb->destructor = sock_edemux;
1566 if (sk_fullsock(sk)) {
1567 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1568
1569 if (dst)
1570 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1571 if (dst &&
1572 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1573 skb_dst_set_noref(skb, dst);
1574 }
1575 }
1576}
1577
1578static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1579 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1580 .twsk_unique = tcp_twsk_unique,
1581 .twsk_destructor = tcp_twsk_destructor,
1582};
1583
1584static const struct inet_connection_sock_af_ops ipv6_specific = {
1585 .queue_xmit = inet6_csk_xmit,
1586 .send_check = tcp_v6_send_check,
1587 .rebuild_header = inet6_sk_rebuild_header,
1588 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1589 .conn_request = tcp_v6_conn_request,
1590 .syn_recv_sock = tcp_v6_syn_recv_sock,
1591 .net_header_len = sizeof(struct ipv6hdr),
1592 .net_frag_header_len = sizeof(struct frag_hdr),
1593 .setsockopt = ipv6_setsockopt,
1594 .getsockopt = ipv6_getsockopt,
1595 .addr2sockaddr = inet6_csk_addr2sockaddr,
1596 .sockaddr_len = sizeof(struct sockaddr_in6),
1597 .bind_conflict = inet6_csk_bind_conflict,
1598#ifdef CONFIG_COMPAT
1599 .compat_setsockopt = compat_ipv6_setsockopt,
1600 .compat_getsockopt = compat_ipv6_getsockopt,
1601#endif
1602 .mtu_reduced = tcp_v6_mtu_reduced,
1603};
1604
1605#ifdef CONFIG_TCP_MD5SIG
1606static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1607 .md5_lookup = tcp_v6_md5_lookup,
1608 .calc_md5_hash = tcp_v6_md5_hash_skb,
1609 .md5_parse = tcp_v6_parse_md5_keys,
1610};
1611#endif
1612
1613/*
1614 * TCP over IPv4 via INET6 API
1615 */
1616static const struct inet_connection_sock_af_ops ipv6_mapped = {
1617 .queue_xmit = ip_queue_xmit,
1618 .send_check = tcp_v4_send_check,
1619 .rebuild_header = inet_sk_rebuild_header,
1620 .sk_rx_dst_set = inet_sk_rx_dst_set,
1621 .conn_request = tcp_v6_conn_request,
1622 .syn_recv_sock = tcp_v6_syn_recv_sock,
1623 .net_header_len = sizeof(struct iphdr),
1624 .setsockopt = ipv6_setsockopt,
1625 .getsockopt = ipv6_getsockopt,
1626 .addr2sockaddr = inet6_csk_addr2sockaddr,
1627 .sockaddr_len = sizeof(struct sockaddr_in6),
1628 .bind_conflict = inet6_csk_bind_conflict,
1629#ifdef CONFIG_COMPAT
1630 .compat_setsockopt = compat_ipv6_setsockopt,
1631 .compat_getsockopt = compat_ipv6_getsockopt,
1632#endif
1633 .mtu_reduced = tcp_v4_mtu_reduced,
1634};
1635
1636#ifdef CONFIG_TCP_MD5SIG
1637static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1638 .md5_lookup = tcp_v4_md5_lookup,
1639 .calc_md5_hash = tcp_v4_md5_hash_skb,
1640 .md5_parse = tcp_v6_parse_md5_keys,
1641};
1642#endif
1643
1644/* NOTE: A lot of things set to zero explicitly by call to
1645 * sk_alloc() so need not be done here.
1646 */
1647static int tcp_v6_init_sock(struct sock *sk)
1648{
1649 struct inet_connection_sock *icsk = inet_csk(sk);
1650
1651 tcp_init_sock(sk);
1652
1653 icsk->icsk_af_ops = &ipv6_specific;
1654
1655#ifdef CONFIG_TCP_MD5SIG
1656 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1657#endif
1658
1659 return 0;
1660}
1661
1662static void tcp_v6_destroy_sock(struct sock *sk)
1663{
1664 tcp_v4_destroy_sock(sk);
1665 inet6_destroy_sock(sk);
1666}
1667
1668#ifdef CONFIG_PROC_FS
1669/* Proc filesystem TCPv6 sock list dumping. */
1670static void get_openreq6(struct seq_file *seq,
1671 const struct request_sock *req, int i)
1672{
1673 long ttd = req->rsk_timer.expires - jiffies;
1674 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1675 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1676
1677 if (ttd < 0)
1678 ttd = 0;
1679
1680 seq_printf(seq,
1681 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1682 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1683 i,
1684 src->s6_addr32[0], src->s6_addr32[1],
1685 src->s6_addr32[2], src->s6_addr32[3],
1686 inet_rsk(req)->ir_num,
1687 dest->s6_addr32[0], dest->s6_addr32[1],
1688 dest->s6_addr32[2], dest->s6_addr32[3],
1689 ntohs(inet_rsk(req)->ir_rmt_port),
1690 TCP_SYN_RECV,
1691 0, 0, /* could print option size, but that is af dependent. */
1692 1, /* timers active (only the expire timer) */
1693 jiffies_to_clock_t(ttd),
1694 req->num_timeout,
1695 from_kuid_munged(seq_user_ns(seq),
1696 sock_i_uid(req->rsk_listener)),
1697 0, /* non standard timer */
1698 0, /* open_requests have no inode */
1699 0, req);
1700}
1701
1702static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1703{
1704 const struct in6_addr *dest, *src;
1705 __u16 destp, srcp;
1706 int timer_active;
1707 unsigned long timer_expires;
1708 const struct inet_sock *inet = inet_sk(sp);
1709 const struct tcp_sock *tp = tcp_sk(sp);
1710 const struct inet_connection_sock *icsk = inet_csk(sp);
1711 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1712 int rx_queue;
1713 int state;
1714
1715 dest = &sp->sk_v6_daddr;
1716 src = &sp->sk_v6_rcv_saddr;
1717 destp = ntohs(inet->inet_dport);
1718 srcp = ntohs(inet->inet_sport);
1719
1720 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1721 timer_active = 1;
1722 timer_expires = icsk->icsk_timeout;
1723 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1724 timer_active = 4;
1725 timer_expires = icsk->icsk_timeout;
1726 } else if (timer_pending(&sp->sk_timer)) {
1727 timer_active = 2;
1728 timer_expires = sp->sk_timer.expires;
1729 } else {
1730 timer_active = 0;
1731 timer_expires = jiffies;
1732 }
1733
1734 state = sk_state_load(sp);
1735 if (state == TCP_LISTEN)
1736 rx_queue = sp->sk_ack_backlog;
1737 else
1738 /* Because we don't lock the socket,
1739 * we might find a transient negative value.
1740 */
1741 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1742
1743 seq_printf(seq,
1744 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1745 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1746 i,
1747 src->s6_addr32[0], src->s6_addr32[1],
1748 src->s6_addr32[2], src->s6_addr32[3], srcp,
1749 dest->s6_addr32[0], dest->s6_addr32[1],
1750 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1751 state,
1752 tp->write_seq - tp->snd_una,
1753 rx_queue,
1754 timer_active,
1755 jiffies_delta_to_clock_t(timer_expires - jiffies),
1756 icsk->icsk_retransmits,
1757 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1758 icsk->icsk_probes_out,
1759 sock_i_ino(sp),
1760 atomic_read(&sp->sk_refcnt), sp,
1761 jiffies_to_clock_t(icsk->icsk_rto),
1762 jiffies_to_clock_t(icsk->icsk_ack.ato),
1763 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1764 tp->snd_cwnd,
1765 state == TCP_LISTEN ?
1766 fastopenq->max_qlen :
1767 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1768 );
1769}
1770
1771static void get_timewait6_sock(struct seq_file *seq,
1772 struct inet_timewait_sock *tw, int i)
1773{
1774 long delta = tw->tw_timer.expires - jiffies;
1775 const struct in6_addr *dest, *src;
1776 __u16 destp, srcp;
1777
1778 dest = &tw->tw_v6_daddr;
1779 src = &tw->tw_v6_rcv_saddr;
1780 destp = ntohs(tw->tw_dport);
1781 srcp = ntohs(tw->tw_sport);
1782
1783 seq_printf(seq,
1784 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1785 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1786 i,
1787 src->s6_addr32[0], src->s6_addr32[1],
1788 src->s6_addr32[2], src->s6_addr32[3], srcp,
1789 dest->s6_addr32[0], dest->s6_addr32[1],
1790 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1791 tw->tw_substate, 0, 0,
1792 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1793 atomic_read(&tw->tw_refcnt), tw);
1794}
1795
1796static int tcp6_seq_show(struct seq_file *seq, void *v)
1797{
1798 struct tcp_iter_state *st;
1799 struct sock *sk = v;
1800
1801 if (v == SEQ_START_TOKEN) {
1802 seq_puts(seq,
1803 " sl "
1804 "local_address "
1805 "remote_address "
1806 "st tx_queue rx_queue tr tm->when retrnsmt"
1807 " uid timeout inode\n");
1808 goto out;
1809 }
1810 st = seq->private;
1811
1812 if (sk->sk_state == TCP_TIME_WAIT)
1813 get_timewait6_sock(seq, v, st->num);
1814 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1815 get_openreq6(seq, v, st->num);
1816 else
1817 get_tcp6_sock(seq, v, st->num);
1818out:
1819 return 0;
1820}
1821
1822static const struct file_operations tcp6_afinfo_seq_fops = {
1823 .owner = THIS_MODULE,
1824 .open = tcp_seq_open,
1825 .read = seq_read,
1826 .llseek = seq_lseek,
1827 .release = seq_release_net
1828};
1829
1830static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1831 .name = "tcp6",
1832 .family = AF_INET6,
1833 .seq_fops = &tcp6_afinfo_seq_fops,
1834 .seq_ops = {
1835 .show = tcp6_seq_show,
1836 },
1837};
1838
1839int __net_init tcp6_proc_init(struct net *net)
1840{
1841 return tcp_proc_register(net, &tcp6_seq_afinfo);
1842}
1843
1844void tcp6_proc_exit(struct net *net)
1845{
1846 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1847}
1848#endif
1849
1850static void tcp_v6_clear_sk(struct sock *sk, int size)
1851{
1852 struct inet_sock *inet = inet_sk(sk);
1853
1854 /* we do not want to clear pinet6 field, because of RCU lookups */
1855 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1856
1857 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1858 memset(&inet->pinet6 + 1, 0, size);
1859}
1860
1861struct proto tcpv6_prot = {
1862 .name = "TCPv6",
1863 .owner = THIS_MODULE,
1864 .close = tcp_close,
1865 .connect = tcp_v6_connect,
1866 .disconnect = tcp_disconnect,
1867 .accept = inet_csk_accept,
1868 .ioctl = tcp_ioctl,
1869 .init = tcp_v6_init_sock,
1870 .destroy = tcp_v6_destroy_sock,
1871 .shutdown = tcp_shutdown,
1872 .setsockopt = tcp_setsockopt,
1873 .getsockopt = tcp_getsockopt,
1874 .recvmsg = tcp_recvmsg,
1875 .sendmsg = tcp_sendmsg,
1876 .sendpage = tcp_sendpage,
1877 .backlog_rcv = tcp_v6_do_rcv,
1878 .release_cb = tcp_release_cb,
1879 .hash = inet6_hash,
1880 .unhash = inet_unhash,
1881 .get_port = inet_csk_get_port,
1882 .enter_memory_pressure = tcp_enter_memory_pressure,
1883 .stream_memory_free = tcp_stream_memory_free,
1884 .sockets_allocated = &tcp_sockets_allocated,
1885 .memory_allocated = &tcp_memory_allocated,
1886 .memory_pressure = &tcp_memory_pressure,
1887 .orphan_count = &tcp_orphan_count,
1888 .sysctl_mem = sysctl_tcp_mem,
1889 .sysctl_wmem = sysctl_tcp_wmem,
1890 .sysctl_rmem = sysctl_tcp_rmem,
1891 .max_header = MAX_TCP_HEADER,
1892 .obj_size = sizeof(struct tcp6_sock),
1893 .slab_flags = SLAB_DESTROY_BY_RCU,
1894 .twsk_prot = &tcp6_timewait_sock_ops,
1895 .rsk_prot = &tcp6_request_sock_ops,
1896 .h.hashinfo = &tcp_hashinfo,
1897 .no_autobind = true,
1898#ifdef CONFIG_COMPAT
1899 .compat_setsockopt = compat_tcp_setsockopt,
1900 .compat_getsockopt = compat_tcp_getsockopt,
1901#endif
1902 .clear_sk = tcp_v6_clear_sk,
1903 .diag_destroy = tcp_abort,
1904};
1905
1906static const struct inet6_protocol tcpv6_protocol = {
1907 .early_demux = tcp_v6_early_demux,
1908 .handler = tcp_v6_rcv,
1909 .err_handler = tcp_v6_err,
1910 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1911};
1912
1913static struct inet_protosw tcpv6_protosw = {
1914 .type = SOCK_STREAM,
1915 .protocol = IPPROTO_TCP,
1916 .prot = &tcpv6_prot,
1917 .ops = &inet6_stream_ops,
1918 .flags = INET_PROTOSW_PERMANENT |
1919 INET_PROTOSW_ICSK,
1920};
1921
1922static int __net_init tcpv6_net_init(struct net *net)
1923{
1924 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1925 SOCK_RAW, IPPROTO_TCP, net);
1926}
1927
1928static void __net_exit tcpv6_net_exit(struct net *net)
1929{
1930 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1931}
1932
1933static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1934{
1935 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1936}
1937
1938static struct pernet_operations tcpv6_net_ops = {
1939 .init = tcpv6_net_init,
1940 .exit = tcpv6_net_exit,
1941 .exit_batch = tcpv6_net_exit_batch,
1942};
1943
1944int __init tcpv6_init(void)
1945{
1946 int ret;
1947
1948 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1949 if (ret)
1950 goto out;
1951
1952 /* register inet6 protocol */
1953 ret = inet6_register_protosw(&tcpv6_protosw);
1954 if (ret)
1955 goto out_tcpv6_protocol;
1956
1957 ret = register_pernet_subsys(&tcpv6_net_ops);
1958 if (ret)
1959 goto out_tcpv6_protosw;
1960out:
1961 return ret;
1962
1963out_tcpv6_protosw:
1964 inet6_unregister_protosw(&tcpv6_protosw);
1965out_tcpv6_protocol:
1966 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1967 goto out;
1968}
1969
1970void tcpv6_exit(void)
1971{
1972 unregister_pernet_subsys(&tcpv6_net_ops);
1973 inet6_unregister_protosw(&tcpv6_protosw);
1974 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1975}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * TCP over IPv6
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
8 *
9 * Based on:
10 * linux/net/ipv4/tcp.c
11 * linux/net/ipv4/tcp_input.c
12 * linux/net/ipv4/tcp_output.c
13 *
14 * Fixes:
15 * Hideaki YOSHIFUJI : sin6_scope_id support
16 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
17 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
18 * a single port at the same time.
19 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
20 */
21
22#include <linux/bottom_half.h>
23#include <linux/module.h>
24#include <linux/errno.h>
25#include <linux/types.h>
26#include <linux/socket.h>
27#include <linux/sockios.h>
28#include <linux/net.h>
29#include <linux/jiffies.h>
30#include <linux/in.h>
31#include <linux/in6.h>
32#include <linux/netdevice.h>
33#include <linux/init.h>
34#include <linux/jhash.h>
35#include <linux/ipsec.h>
36#include <linux/times.h>
37#include <linux/slab.h>
38#include <linux/uaccess.h>
39#include <linux/ipv6.h>
40#include <linux/icmpv6.h>
41#include <linux/random.h>
42#include <linux/indirect_call_wrapper.h>
43
44#include <net/tcp.h>
45#include <net/ndisc.h>
46#include <net/inet6_hashtables.h>
47#include <net/inet6_connection_sock.h>
48#include <net/ipv6.h>
49#include <net/transp_v6.h>
50#include <net/addrconf.h>
51#include <net/ip6_route.h>
52#include <net/ip6_checksum.h>
53#include <net/inet_ecn.h>
54#include <net/protocol.h>
55#include <net/xfrm.h>
56#include <net/snmp.h>
57#include <net/dsfield.h>
58#include <net/timewait_sock.h>
59#include <net/inet_common.h>
60#include <net/secure_seq.h>
61#include <net/busy_poll.h>
62
63#include <linux/proc_fs.h>
64#include <linux/seq_file.h>
65
66#include <crypto/hash.h>
67#include <linux/scatterlist.h>
68
69#include <trace/events/tcp.h>
70
71static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
72static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
73 struct request_sock *req);
74
75static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
76
77static const struct inet_connection_sock_af_ops ipv6_mapped;
78const struct inet_connection_sock_af_ops ipv6_specific;
79#ifdef CONFIG_TCP_MD5SIG
80static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
81static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
82#else
83static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
84 const struct in6_addr *addr,
85 int l3index)
86{
87 return NULL;
88}
89#endif
90
91/* Helper returning the inet6 address from a given tcp socket.
92 * It can be used in TCP stack instead of inet6_sk(sk).
93 * This avoids a dereference and allow compiler optimizations.
94 * It is a specialized version of inet6_sk_generic().
95 */
96static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk)
97{
98 unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo);
99
100 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
101}
102
103static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
104{
105 struct dst_entry *dst = skb_dst(skb);
106
107 if (dst && dst_hold_safe(dst)) {
108 const struct rt6_info *rt = (const struct rt6_info *)dst;
109
110 sk->sk_rx_dst = dst;
111 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
112 tcp_inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
113 }
114}
115
116static u32 tcp_v6_init_seq(const struct sk_buff *skb)
117{
118 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
119 ipv6_hdr(skb)->saddr.s6_addr32,
120 tcp_hdr(skb)->dest,
121 tcp_hdr(skb)->source);
122}
123
124static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
125{
126 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
127 ipv6_hdr(skb)->saddr.s6_addr32);
128}
129
130static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
131 int addr_len)
132{
133 /* This check is replicated from tcp_v6_connect() and intended to
134 * prevent BPF program called below from accessing bytes that are out
135 * of the bound specified by user in addr_len.
136 */
137 if (addr_len < SIN6_LEN_RFC2133)
138 return -EINVAL;
139
140 sock_owned_by_me(sk);
141
142 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr);
143}
144
145static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
146 int addr_len)
147{
148 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
149 struct inet_sock *inet = inet_sk(sk);
150 struct inet_connection_sock *icsk = inet_csk(sk);
151 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
152 struct tcp_sock *tp = tcp_sk(sk);
153 struct in6_addr *saddr = NULL, *final_p, final;
154 struct ipv6_txoptions *opt;
155 struct flowi6 fl6;
156 struct dst_entry *dst;
157 int addr_type;
158 int err;
159 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
160
161 if (addr_len < SIN6_LEN_RFC2133)
162 return -EINVAL;
163
164 if (usin->sin6_family != AF_INET6)
165 return -EAFNOSUPPORT;
166
167 memset(&fl6, 0, sizeof(fl6));
168
169 if (np->sndflow) {
170 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
171 IP6_ECN_flow_init(fl6.flowlabel);
172 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
173 struct ip6_flowlabel *flowlabel;
174 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
175 if (IS_ERR(flowlabel))
176 return -EINVAL;
177 fl6_sock_release(flowlabel);
178 }
179 }
180
181 /*
182 * connect() to INADDR_ANY means loopback (BSD'ism).
183 */
184
185 if (ipv6_addr_any(&usin->sin6_addr)) {
186 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
187 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
188 &usin->sin6_addr);
189 else
190 usin->sin6_addr = in6addr_loopback;
191 }
192
193 addr_type = ipv6_addr_type(&usin->sin6_addr);
194
195 if (addr_type & IPV6_ADDR_MULTICAST)
196 return -ENETUNREACH;
197
198 if (addr_type&IPV6_ADDR_LINKLOCAL) {
199 if (addr_len >= sizeof(struct sockaddr_in6) &&
200 usin->sin6_scope_id) {
201 /* If interface is set while binding, indices
202 * must coincide.
203 */
204 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
205 return -EINVAL;
206
207 sk->sk_bound_dev_if = usin->sin6_scope_id;
208 }
209
210 /* Connect to link-local address requires an interface */
211 if (!sk->sk_bound_dev_if)
212 return -EINVAL;
213 }
214
215 if (tp->rx_opt.ts_recent_stamp &&
216 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
217 tp->rx_opt.ts_recent = 0;
218 tp->rx_opt.ts_recent_stamp = 0;
219 WRITE_ONCE(tp->write_seq, 0);
220 }
221
222 sk->sk_v6_daddr = usin->sin6_addr;
223 np->flow_label = fl6.flowlabel;
224
225 /*
226 * TCP over IPv4
227 */
228
229 if (addr_type & IPV6_ADDR_MAPPED) {
230 u32 exthdrlen = icsk->icsk_ext_hdr_len;
231 struct sockaddr_in sin;
232
233 if (__ipv6_only_sock(sk))
234 return -ENETUNREACH;
235
236 sin.sin_family = AF_INET;
237 sin.sin_port = usin->sin6_port;
238 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
239
240 icsk->icsk_af_ops = &ipv6_mapped;
241 if (sk_is_mptcp(sk))
242 mptcpv6_handle_mapped(sk, true);
243 sk->sk_backlog_rcv = tcp_v4_do_rcv;
244#ifdef CONFIG_TCP_MD5SIG
245 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
246#endif
247
248 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
249
250 if (err) {
251 icsk->icsk_ext_hdr_len = exthdrlen;
252 icsk->icsk_af_ops = &ipv6_specific;
253 if (sk_is_mptcp(sk))
254 mptcpv6_handle_mapped(sk, false);
255 sk->sk_backlog_rcv = tcp_v6_do_rcv;
256#ifdef CONFIG_TCP_MD5SIG
257 tp->af_specific = &tcp_sock_ipv6_specific;
258#endif
259 goto failure;
260 }
261 np->saddr = sk->sk_v6_rcv_saddr;
262
263 return err;
264 }
265
266 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
267 saddr = &sk->sk_v6_rcv_saddr;
268
269 fl6.flowi6_proto = IPPROTO_TCP;
270 fl6.daddr = sk->sk_v6_daddr;
271 fl6.saddr = saddr ? *saddr : np->saddr;
272 fl6.flowi6_oif = sk->sk_bound_dev_if;
273 fl6.flowi6_mark = sk->sk_mark;
274 fl6.fl6_dport = usin->sin6_port;
275 fl6.fl6_sport = inet->inet_sport;
276 fl6.flowi6_uid = sk->sk_uid;
277
278 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
279 final_p = fl6_update_dst(&fl6, opt, &final);
280
281 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
282
283 dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
284 if (IS_ERR(dst)) {
285 err = PTR_ERR(dst);
286 goto failure;
287 }
288
289 if (!saddr) {
290 saddr = &fl6.saddr;
291 sk->sk_v6_rcv_saddr = *saddr;
292 }
293
294 /* set the source address */
295 np->saddr = *saddr;
296 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
297
298 sk->sk_gso_type = SKB_GSO_TCPV6;
299 ip6_dst_store(sk, dst, NULL, NULL);
300
301 icsk->icsk_ext_hdr_len = 0;
302 if (opt)
303 icsk->icsk_ext_hdr_len = opt->opt_flen +
304 opt->opt_nflen;
305
306 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
307
308 inet->inet_dport = usin->sin6_port;
309
310 tcp_set_state(sk, TCP_SYN_SENT);
311 err = inet6_hash_connect(tcp_death_row, sk);
312 if (err)
313 goto late_failure;
314
315 sk_set_txhash(sk);
316
317 if (likely(!tp->repair)) {
318 if (!tp->write_seq)
319 WRITE_ONCE(tp->write_seq,
320 secure_tcpv6_seq(np->saddr.s6_addr32,
321 sk->sk_v6_daddr.s6_addr32,
322 inet->inet_sport,
323 inet->inet_dport));
324 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
325 np->saddr.s6_addr32,
326 sk->sk_v6_daddr.s6_addr32);
327 }
328
329 if (tcp_fastopen_defer_connect(sk, &err))
330 return err;
331 if (err)
332 goto late_failure;
333
334 err = tcp_connect(sk);
335 if (err)
336 goto late_failure;
337
338 return 0;
339
340late_failure:
341 tcp_set_state(sk, TCP_CLOSE);
342failure:
343 inet->inet_dport = 0;
344 sk->sk_route_caps = 0;
345 return err;
346}
347
348static void tcp_v6_mtu_reduced(struct sock *sk)
349{
350 struct dst_entry *dst;
351
352 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
353 return;
354
355 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
356 if (!dst)
357 return;
358
359 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
360 tcp_sync_mss(sk, dst_mtu(dst));
361 tcp_simple_retransmit(sk);
362 }
363}
364
365static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
366 u8 type, u8 code, int offset, __be32 info)
367{
368 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
369 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
370 struct net *net = dev_net(skb->dev);
371 struct request_sock *fastopen;
372 struct ipv6_pinfo *np;
373 struct tcp_sock *tp;
374 __u32 seq, snd_una;
375 struct sock *sk;
376 bool fatal;
377 int err;
378
379 sk = __inet6_lookup_established(net, &tcp_hashinfo,
380 &hdr->daddr, th->dest,
381 &hdr->saddr, ntohs(th->source),
382 skb->dev->ifindex, inet6_sdif(skb));
383
384 if (!sk) {
385 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
386 ICMP6_MIB_INERRORS);
387 return -ENOENT;
388 }
389
390 if (sk->sk_state == TCP_TIME_WAIT) {
391 inet_twsk_put(inet_twsk(sk));
392 return 0;
393 }
394 seq = ntohl(th->seq);
395 fatal = icmpv6_err_convert(type, code, &err);
396 if (sk->sk_state == TCP_NEW_SYN_RECV) {
397 tcp_req_err(sk, seq, fatal);
398 return 0;
399 }
400
401 bh_lock_sock(sk);
402 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
403 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
404
405 if (sk->sk_state == TCP_CLOSE)
406 goto out;
407
408 if (ipv6_hdr(skb)->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
409 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
410 goto out;
411 }
412
413 tp = tcp_sk(sk);
414 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
415 fastopen = rcu_dereference(tp->fastopen_rsk);
416 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
417 if (sk->sk_state != TCP_LISTEN &&
418 !between(seq, snd_una, tp->snd_nxt)) {
419 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
420 goto out;
421 }
422
423 np = tcp_inet6_sk(sk);
424
425 if (type == NDISC_REDIRECT) {
426 if (!sock_owned_by_user(sk)) {
427 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
428
429 if (dst)
430 dst->ops->redirect(dst, sk, skb);
431 }
432 goto out;
433 }
434
435 if (type == ICMPV6_PKT_TOOBIG) {
436 /* We are not interested in TCP_LISTEN and open_requests
437 * (SYN-ACKs send out by Linux are always <576bytes so
438 * they should go through unfragmented).
439 */
440 if (sk->sk_state == TCP_LISTEN)
441 goto out;
442
443 if (!ip6_sk_accept_pmtu(sk))
444 goto out;
445
446 tp->mtu_info = ntohl(info);
447 if (!sock_owned_by_user(sk))
448 tcp_v6_mtu_reduced(sk);
449 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
450 &sk->sk_tsq_flags))
451 sock_hold(sk);
452 goto out;
453 }
454
455
456 /* Might be for an request_sock */
457 switch (sk->sk_state) {
458 case TCP_SYN_SENT:
459 case TCP_SYN_RECV:
460 /* Only in fast or simultaneous open. If a fast open socket is
461 * is already accepted it is treated as a connected one below.
462 */
463 if (fastopen && !fastopen->sk)
464 break;
465
466 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
467
468 if (!sock_owned_by_user(sk)) {
469 sk->sk_err = err;
470 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
471
472 tcp_done(sk);
473 } else
474 sk->sk_err_soft = err;
475 goto out;
476 case TCP_LISTEN:
477 break;
478 default:
479 /* check if this ICMP message allows revert of backoff.
480 * (see RFC 6069)
481 */
482 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
483 code == ICMPV6_NOROUTE)
484 tcp_ld_RTO_revert(sk, seq);
485 }
486
487 if (!sock_owned_by_user(sk) && np->recverr) {
488 sk->sk_err = err;
489 sk->sk_error_report(sk);
490 } else
491 sk->sk_err_soft = err;
492
493out:
494 bh_unlock_sock(sk);
495 sock_put(sk);
496 return 0;
497}
498
499
500static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
501 struct flowi *fl,
502 struct request_sock *req,
503 struct tcp_fastopen_cookie *foc,
504 enum tcp_synack_type synack_type)
505{
506 struct inet_request_sock *ireq = inet_rsk(req);
507 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
508 struct ipv6_txoptions *opt;
509 struct flowi6 *fl6 = &fl->u.ip6;
510 struct sk_buff *skb;
511 int err = -ENOMEM;
512
513 /* First, grab a route. */
514 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
515 IPPROTO_TCP)) == NULL)
516 goto done;
517
518 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
519
520 if (skb) {
521 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
522 &ireq->ir_v6_rmt_addr);
523
524 fl6->daddr = ireq->ir_v6_rmt_addr;
525 if (np->repflow && ireq->pktopts)
526 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
527
528 rcu_read_lock();
529 opt = ireq->ipv6_opt;
530 if (!opt)
531 opt = rcu_dereference(np->opt);
532 err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass,
533 sk->sk_priority);
534 rcu_read_unlock();
535 err = net_xmit_eval(err);
536 }
537
538done:
539 return err;
540}
541
542
543static void tcp_v6_reqsk_destructor(struct request_sock *req)
544{
545 kfree(inet_rsk(req)->ipv6_opt);
546 kfree_skb(inet_rsk(req)->pktopts);
547}
548
549#ifdef CONFIG_TCP_MD5SIG
550static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
551 const struct in6_addr *addr,
552 int l3index)
553{
554 return tcp_md5_do_lookup(sk, l3index,
555 (union tcp_md5_addr *)addr, AF_INET6);
556}
557
558static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
559 const struct sock *addr_sk)
560{
561 int l3index;
562
563 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
564 addr_sk->sk_bound_dev_if);
565 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
566 l3index);
567}
568
569static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
570 sockptr_t optval, int optlen)
571{
572 struct tcp_md5sig cmd;
573 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
574 int l3index = 0;
575 u8 prefixlen;
576
577 if (optlen < sizeof(cmd))
578 return -EINVAL;
579
580 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
581 return -EFAULT;
582
583 if (sin6->sin6_family != AF_INET6)
584 return -EINVAL;
585
586 if (optname == TCP_MD5SIG_EXT &&
587 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
588 prefixlen = cmd.tcpm_prefixlen;
589 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
590 prefixlen > 32))
591 return -EINVAL;
592 } else {
593 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
594 }
595
596 if (optname == TCP_MD5SIG_EXT &&
597 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
598 struct net_device *dev;
599
600 rcu_read_lock();
601 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
602 if (dev && netif_is_l3_master(dev))
603 l3index = dev->ifindex;
604 rcu_read_unlock();
605
606 /* ok to reference set/not set outside of rcu;
607 * right now device MUST be an L3 master
608 */
609 if (!dev || !l3index)
610 return -EINVAL;
611 }
612
613 if (!cmd.tcpm_keylen) {
614 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
615 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
616 AF_INET, prefixlen,
617 l3index);
618 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
619 AF_INET6, prefixlen, l3index);
620 }
621
622 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
623 return -EINVAL;
624
625 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
626 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
627 AF_INET, prefixlen, l3index,
628 cmd.tcpm_key, cmd.tcpm_keylen,
629 GFP_KERNEL);
630
631 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
632 AF_INET6, prefixlen, l3index,
633 cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
634}
635
636static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
637 const struct in6_addr *daddr,
638 const struct in6_addr *saddr,
639 const struct tcphdr *th, int nbytes)
640{
641 struct tcp6_pseudohdr *bp;
642 struct scatterlist sg;
643 struct tcphdr *_th;
644
645 bp = hp->scratch;
646 /* 1. TCP pseudo-header (RFC2460) */
647 bp->saddr = *saddr;
648 bp->daddr = *daddr;
649 bp->protocol = cpu_to_be32(IPPROTO_TCP);
650 bp->len = cpu_to_be32(nbytes);
651
652 _th = (struct tcphdr *)(bp + 1);
653 memcpy(_th, th, sizeof(*th));
654 _th->check = 0;
655
656 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
657 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
658 sizeof(*bp) + sizeof(*th));
659 return crypto_ahash_update(hp->md5_req);
660}
661
662static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
663 const struct in6_addr *daddr, struct in6_addr *saddr,
664 const struct tcphdr *th)
665{
666 struct tcp_md5sig_pool *hp;
667 struct ahash_request *req;
668
669 hp = tcp_get_md5sig_pool();
670 if (!hp)
671 goto clear_hash_noput;
672 req = hp->md5_req;
673
674 if (crypto_ahash_init(req))
675 goto clear_hash;
676 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
677 goto clear_hash;
678 if (tcp_md5_hash_key(hp, key))
679 goto clear_hash;
680 ahash_request_set_crypt(req, NULL, md5_hash, 0);
681 if (crypto_ahash_final(req))
682 goto clear_hash;
683
684 tcp_put_md5sig_pool();
685 return 0;
686
687clear_hash:
688 tcp_put_md5sig_pool();
689clear_hash_noput:
690 memset(md5_hash, 0, 16);
691 return 1;
692}
693
694static int tcp_v6_md5_hash_skb(char *md5_hash,
695 const struct tcp_md5sig_key *key,
696 const struct sock *sk,
697 const struct sk_buff *skb)
698{
699 const struct in6_addr *saddr, *daddr;
700 struct tcp_md5sig_pool *hp;
701 struct ahash_request *req;
702 const struct tcphdr *th = tcp_hdr(skb);
703
704 if (sk) { /* valid for establish/request sockets */
705 saddr = &sk->sk_v6_rcv_saddr;
706 daddr = &sk->sk_v6_daddr;
707 } else {
708 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
709 saddr = &ip6h->saddr;
710 daddr = &ip6h->daddr;
711 }
712
713 hp = tcp_get_md5sig_pool();
714 if (!hp)
715 goto clear_hash_noput;
716 req = hp->md5_req;
717
718 if (crypto_ahash_init(req))
719 goto clear_hash;
720
721 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
722 goto clear_hash;
723 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
724 goto clear_hash;
725 if (tcp_md5_hash_key(hp, key))
726 goto clear_hash;
727 ahash_request_set_crypt(req, NULL, md5_hash, 0);
728 if (crypto_ahash_final(req))
729 goto clear_hash;
730
731 tcp_put_md5sig_pool();
732 return 0;
733
734clear_hash:
735 tcp_put_md5sig_pool();
736clear_hash_noput:
737 memset(md5_hash, 0, 16);
738 return 1;
739}
740
741#endif
742
743static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
744 const struct sk_buff *skb,
745 int dif, int sdif)
746{
747#ifdef CONFIG_TCP_MD5SIG
748 const __u8 *hash_location = NULL;
749 struct tcp_md5sig_key *hash_expected;
750 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
751 const struct tcphdr *th = tcp_hdr(skb);
752 int genhash, l3index;
753 u8 newhash[16];
754
755 /* sdif set, means packet ingressed via a device
756 * in an L3 domain and dif is set to the l3mdev
757 */
758 l3index = sdif ? dif : 0;
759
760 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr, l3index);
761 hash_location = tcp_parse_md5sig_option(th);
762
763 /* We've parsed the options - do we have a hash? */
764 if (!hash_expected && !hash_location)
765 return false;
766
767 if (hash_expected && !hash_location) {
768 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
769 return true;
770 }
771
772 if (!hash_expected && hash_location) {
773 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
774 return true;
775 }
776
777 /* check the signature */
778 genhash = tcp_v6_md5_hash_skb(newhash,
779 hash_expected,
780 NULL, skb);
781
782 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
783 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
784 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u L3 index %d\n",
785 genhash ? "failed" : "mismatch",
786 &ip6h->saddr, ntohs(th->source),
787 &ip6h->daddr, ntohs(th->dest), l3index);
788 return true;
789 }
790#endif
791 return false;
792}
793
794static void tcp_v6_init_req(struct request_sock *req,
795 const struct sock *sk_listener,
796 struct sk_buff *skb)
797{
798 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
799 struct inet_request_sock *ireq = inet_rsk(req);
800 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
801
802 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
803 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
804
805 /* So that link locals have meaning */
806 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
807 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
808 ireq->ir_iif = tcp_v6_iif(skb);
809
810 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
811 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
812 np->rxopt.bits.rxinfo ||
813 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
814 np->rxopt.bits.rxohlim || np->repflow)) {
815 refcount_inc(&skb->users);
816 ireq->pktopts = skb;
817 }
818}
819
820static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
821 struct flowi *fl,
822 const struct request_sock *req)
823{
824 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
825}
826
827struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
828 .family = AF_INET6,
829 .obj_size = sizeof(struct tcp6_request_sock),
830 .rtx_syn_ack = tcp_rtx_synack,
831 .send_ack = tcp_v6_reqsk_send_ack,
832 .destructor = tcp_v6_reqsk_destructor,
833 .send_reset = tcp_v6_send_reset,
834 .syn_ack_timeout = tcp_syn_ack_timeout,
835};
836
837const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
838 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
839 sizeof(struct ipv6hdr),
840#ifdef CONFIG_TCP_MD5SIG
841 .req_md5_lookup = tcp_v6_md5_lookup,
842 .calc_md5_hash = tcp_v6_md5_hash_skb,
843#endif
844 .init_req = tcp_v6_init_req,
845#ifdef CONFIG_SYN_COOKIES
846 .cookie_init_seq = cookie_v6_init_sequence,
847#endif
848 .route_req = tcp_v6_route_req,
849 .init_seq = tcp_v6_init_seq,
850 .init_ts_off = tcp_v6_init_ts_off,
851 .send_synack = tcp_v6_send_synack,
852};
853
854static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
855 u32 ack, u32 win, u32 tsval, u32 tsecr,
856 int oif, struct tcp_md5sig_key *key, int rst,
857 u8 tclass, __be32 label, u32 priority)
858{
859 const struct tcphdr *th = tcp_hdr(skb);
860 struct tcphdr *t1;
861 struct sk_buff *buff;
862 struct flowi6 fl6;
863 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
864 struct sock *ctl_sk = net->ipv6.tcp_sk;
865 unsigned int tot_len = sizeof(struct tcphdr);
866 struct dst_entry *dst;
867 __be32 *topt;
868 __u32 mark = 0;
869
870 if (tsecr)
871 tot_len += TCPOLEN_TSTAMP_ALIGNED;
872#ifdef CONFIG_TCP_MD5SIG
873 if (key)
874 tot_len += TCPOLEN_MD5SIG_ALIGNED;
875#endif
876
877 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
878 GFP_ATOMIC);
879 if (!buff)
880 return;
881
882 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
883
884 t1 = skb_push(buff, tot_len);
885 skb_reset_transport_header(buff);
886
887 /* Swap the send and the receive. */
888 memset(t1, 0, sizeof(*t1));
889 t1->dest = th->source;
890 t1->source = th->dest;
891 t1->doff = tot_len / 4;
892 t1->seq = htonl(seq);
893 t1->ack_seq = htonl(ack);
894 t1->ack = !rst || !th->ack;
895 t1->rst = rst;
896 t1->window = htons(win);
897
898 topt = (__be32 *)(t1 + 1);
899
900 if (tsecr) {
901 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
902 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
903 *topt++ = htonl(tsval);
904 *topt++ = htonl(tsecr);
905 }
906
907#ifdef CONFIG_TCP_MD5SIG
908 if (key) {
909 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
910 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
911 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
912 &ipv6_hdr(skb)->saddr,
913 &ipv6_hdr(skb)->daddr, t1);
914 }
915#endif
916
917 memset(&fl6, 0, sizeof(fl6));
918 fl6.daddr = ipv6_hdr(skb)->saddr;
919 fl6.saddr = ipv6_hdr(skb)->daddr;
920 fl6.flowlabel = label;
921
922 buff->ip_summed = CHECKSUM_PARTIAL;
923 buff->csum = 0;
924
925 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
926
927 fl6.flowi6_proto = IPPROTO_TCP;
928 if (rt6_need_strict(&fl6.daddr) && !oif)
929 fl6.flowi6_oif = tcp_v6_iif(skb);
930 else {
931 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
932 oif = skb->skb_iif;
933
934 fl6.flowi6_oif = oif;
935 }
936
937 if (sk) {
938 if (sk->sk_state == TCP_TIME_WAIT) {
939 mark = inet_twsk(sk)->tw_mark;
940 /* autoflowlabel relies on buff->hash */
941 skb_set_hash(buff, inet_twsk(sk)->tw_txhash,
942 PKT_HASH_TYPE_L4);
943 } else {
944 mark = sk->sk_mark;
945 }
946 buff->tstamp = tcp_transmit_time(sk);
947 }
948 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
949 fl6.fl6_dport = t1->dest;
950 fl6.fl6_sport = t1->source;
951 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
952 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
953
954 /* Pass a socket to ip6_dst_lookup either it is for RST
955 * Underlying function will use this to retrieve the network
956 * namespace
957 */
958 dst = ip6_dst_lookup_flow(sock_net(ctl_sk), ctl_sk, &fl6, NULL);
959 if (!IS_ERR(dst)) {
960 skb_dst_set(buff, dst);
961 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, tclass,
962 priority);
963 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
964 if (rst)
965 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
966 return;
967 }
968
969 kfree_skb(buff);
970}
971
972static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
973{
974 const struct tcphdr *th = tcp_hdr(skb);
975 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
976 u32 seq = 0, ack_seq = 0;
977 struct tcp_md5sig_key *key = NULL;
978#ifdef CONFIG_TCP_MD5SIG
979 const __u8 *hash_location = NULL;
980 unsigned char newhash[16];
981 int genhash;
982 struct sock *sk1 = NULL;
983#endif
984 __be32 label = 0;
985 u32 priority = 0;
986 struct net *net;
987 int oif = 0;
988
989 if (th->rst)
990 return;
991
992 /* If sk not NULL, it means we did a successful lookup and incoming
993 * route had to be correct. prequeue might have dropped our dst.
994 */
995 if (!sk && !ipv6_unicast_destination(skb))
996 return;
997
998 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
999#ifdef CONFIG_TCP_MD5SIG
1000 rcu_read_lock();
1001 hash_location = tcp_parse_md5sig_option(th);
1002 if (sk && sk_fullsock(sk)) {
1003 int l3index;
1004
1005 /* sdif set, means packet ingressed via a device
1006 * in an L3 domain and inet_iif is set to it.
1007 */
1008 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1009 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1010 } else if (hash_location) {
1011 int dif = tcp_v6_iif_l3_slave(skb);
1012 int sdif = tcp_v6_sdif(skb);
1013 int l3index;
1014
1015 /*
1016 * active side is lost. Try to find listening socket through
1017 * source port, and then find md5 key through listening socket.
1018 * we are not loose security here:
1019 * Incoming packet is checked with md5 hash with finding key,
1020 * no RST generated if md5 hash doesn't match.
1021 */
1022 sk1 = inet6_lookup_listener(net,
1023 &tcp_hashinfo, NULL, 0,
1024 &ipv6h->saddr,
1025 th->source, &ipv6h->daddr,
1026 ntohs(th->source), dif, sdif);
1027 if (!sk1)
1028 goto out;
1029
1030 /* sdif set, means packet ingressed via a device
1031 * in an L3 domain and dif is set to it.
1032 */
1033 l3index = tcp_v6_sdif(skb) ? dif : 0;
1034
1035 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1036 if (!key)
1037 goto out;
1038
1039 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
1040 if (genhash || memcmp(hash_location, newhash, 16) != 0)
1041 goto out;
1042 }
1043#endif
1044
1045 if (th->ack)
1046 seq = ntohl(th->ack_seq);
1047 else
1048 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1049 (th->doff << 2);
1050
1051 if (sk) {
1052 oif = sk->sk_bound_dev_if;
1053 if (sk_fullsock(sk)) {
1054 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1055
1056 trace_tcp_send_reset(sk, skb);
1057 if (np->repflow)
1058 label = ip6_flowlabel(ipv6h);
1059 priority = sk->sk_priority;
1060 }
1061 if (sk->sk_state == TCP_TIME_WAIT) {
1062 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1063 priority = inet_twsk(sk)->tw_priority;
1064 }
1065 } else {
1066 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1067 label = ip6_flowlabel(ipv6h);
1068 }
1069
1070 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0,
1071 label, priority);
1072
1073#ifdef CONFIG_TCP_MD5SIG
1074out:
1075 rcu_read_unlock();
1076#endif
1077}
1078
1079static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1080 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1081 struct tcp_md5sig_key *key, u8 tclass,
1082 __be32 label, u32 priority)
1083{
1084 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1085 tclass, label, priority);
1086}
1087
1088static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1089{
1090 struct inet_timewait_sock *tw = inet_twsk(sk);
1091 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1092
1093 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1094 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1095 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1096 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1097 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority);
1098
1099 inet_twsk_put(tw);
1100}
1101
1102static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1103 struct request_sock *req)
1104{
1105 int l3index;
1106
1107 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1108
1109 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1110 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1111 */
1112 /* RFC 7323 2.3
1113 * The window field (SEG.WND) of every outgoing segment, with the
1114 * exception of <SYN> segments, MUST be right-shifted by
1115 * Rcv.Wind.Shift bits:
1116 */
1117 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1118 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1119 tcp_rsk(req)->rcv_nxt,
1120 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1121 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1122 req->ts_recent, sk->sk_bound_dev_if,
1123 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
1124 0, 0, sk->sk_priority);
1125}
1126
1127
1128static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1129{
1130#ifdef CONFIG_SYN_COOKIES
1131 const struct tcphdr *th = tcp_hdr(skb);
1132
1133 if (!th->syn)
1134 sk = cookie_v6_check(sk, skb);
1135#endif
1136 return sk;
1137}
1138
1139u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1140 struct tcphdr *th, u32 *cookie)
1141{
1142 u16 mss = 0;
1143#ifdef CONFIG_SYN_COOKIES
1144 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1145 &tcp_request_sock_ipv6_ops, sk, th);
1146 if (mss) {
1147 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1148 tcp_synq_overflow(sk);
1149 }
1150#endif
1151 return mss;
1152}
1153
1154static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1155{
1156 if (skb->protocol == htons(ETH_P_IP))
1157 return tcp_v4_conn_request(sk, skb);
1158
1159 if (!ipv6_unicast_destination(skb))
1160 goto drop;
1161
1162 return tcp_conn_request(&tcp6_request_sock_ops,
1163 &tcp_request_sock_ipv6_ops, sk, skb);
1164
1165drop:
1166 tcp_listendrop(sk);
1167 return 0; /* don't send reset */
1168}
1169
1170static void tcp_v6_restore_cb(struct sk_buff *skb)
1171{
1172 /* We need to move header back to the beginning if xfrm6_policy_check()
1173 * and tcp_v6_fill_cb() are going to be called again.
1174 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1175 */
1176 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1177 sizeof(struct inet6_skb_parm));
1178}
1179
1180static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1181 struct request_sock *req,
1182 struct dst_entry *dst,
1183 struct request_sock *req_unhash,
1184 bool *own_req)
1185{
1186 struct inet_request_sock *ireq;
1187 struct ipv6_pinfo *newnp;
1188 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1189 struct ipv6_txoptions *opt;
1190 struct inet_sock *newinet;
1191 struct tcp_sock *newtp;
1192 struct sock *newsk;
1193#ifdef CONFIG_TCP_MD5SIG
1194 struct tcp_md5sig_key *key;
1195 int l3index;
1196#endif
1197 struct flowi6 fl6;
1198
1199 if (skb->protocol == htons(ETH_P_IP)) {
1200 /*
1201 * v6 mapped
1202 */
1203
1204 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1205 req_unhash, own_req);
1206
1207 if (!newsk)
1208 return NULL;
1209
1210 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1211
1212 newinet = inet_sk(newsk);
1213 newnp = tcp_inet6_sk(newsk);
1214 newtp = tcp_sk(newsk);
1215
1216 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1217
1218 newnp->saddr = newsk->sk_v6_rcv_saddr;
1219
1220 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1221 if (sk_is_mptcp(newsk))
1222 mptcpv6_handle_mapped(newsk, true);
1223 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1224#ifdef CONFIG_TCP_MD5SIG
1225 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1226#endif
1227
1228 newnp->ipv6_mc_list = NULL;
1229 newnp->ipv6_ac_list = NULL;
1230 newnp->ipv6_fl_list = NULL;
1231 newnp->pktoptions = NULL;
1232 newnp->opt = NULL;
1233 newnp->mcast_oif = inet_iif(skb);
1234 newnp->mcast_hops = ip_hdr(skb)->ttl;
1235 newnp->rcv_flowinfo = 0;
1236 if (np->repflow)
1237 newnp->flow_label = 0;
1238
1239 /*
1240 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1241 * here, tcp_create_openreq_child now does this for us, see the comment in
1242 * that function for the gory details. -acme
1243 */
1244
1245 /* It is tricky place. Until this moment IPv4 tcp
1246 worked with IPv6 icsk.icsk_af_ops.
1247 Sync it now.
1248 */
1249 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1250
1251 return newsk;
1252 }
1253
1254 ireq = inet_rsk(req);
1255
1256 if (sk_acceptq_is_full(sk))
1257 goto out_overflow;
1258
1259 if (!dst) {
1260 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1261 if (!dst)
1262 goto out;
1263 }
1264
1265 newsk = tcp_create_openreq_child(sk, req, skb);
1266 if (!newsk)
1267 goto out_nonewsk;
1268
1269 /*
1270 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1271 * count here, tcp_create_openreq_child now does this for us, see the
1272 * comment in that function for the gory details. -acme
1273 */
1274
1275 newsk->sk_gso_type = SKB_GSO_TCPV6;
1276 ip6_dst_store(newsk, dst, NULL, NULL);
1277 inet6_sk_rx_dst_set(newsk, skb);
1278
1279 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1280
1281 newtp = tcp_sk(newsk);
1282 newinet = inet_sk(newsk);
1283 newnp = tcp_inet6_sk(newsk);
1284
1285 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1286
1287 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1288 newnp->saddr = ireq->ir_v6_loc_addr;
1289 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1290 newsk->sk_bound_dev_if = ireq->ir_iif;
1291
1292 /* Now IPv6 options...
1293
1294 First: no IPv4 options.
1295 */
1296 newinet->inet_opt = NULL;
1297 newnp->ipv6_mc_list = NULL;
1298 newnp->ipv6_ac_list = NULL;
1299 newnp->ipv6_fl_list = NULL;
1300
1301 /* Clone RX bits */
1302 newnp->rxopt.all = np->rxopt.all;
1303
1304 newnp->pktoptions = NULL;
1305 newnp->opt = NULL;
1306 newnp->mcast_oif = tcp_v6_iif(skb);
1307 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1308 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1309 if (np->repflow)
1310 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1311
1312 /* Clone native IPv6 options from listening socket (if any)
1313
1314 Yes, keeping reference count would be much more clever,
1315 but we make one more one thing there: reattach optmem
1316 to newsk.
1317 */
1318 opt = ireq->ipv6_opt;
1319 if (!opt)
1320 opt = rcu_dereference(np->opt);
1321 if (opt) {
1322 opt = ipv6_dup_options(newsk, opt);
1323 RCU_INIT_POINTER(newnp->opt, opt);
1324 }
1325 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1326 if (opt)
1327 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1328 opt->opt_flen;
1329
1330 tcp_ca_openreq_child(newsk, dst);
1331
1332 tcp_sync_mss(newsk, dst_mtu(dst));
1333 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1334
1335 tcp_initialize_rcv_mss(newsk);
1336
1337 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1338 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1339
1340#ifdef CONFIG_TCP_MD5SIG
1341 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1342
1343 /* Copy over the MD5 key from the original socket */
1344 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1345 if (key) {
1346 /* We're using one, so create a matching key
1347 * on the newsk structure. If we fail to get
1348 * memory, then we end up not copying the key
1349 * across. Shucks.
1350 */
1351 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1352 AF_INET6, 128, l3index, key->key, key->keylen,
1353 sk_gfp_mask(sk, GFP_ATOMIC));
1354 }
1355#endif
1356
1357 if (__inet_inherit_port(sk, newsk) < 0) {
1358 inet_csk_prepare_forced_close(newsk);
1359 tcp_done(newsk);
1360 goto out;
1361 }
1362 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1363 if (*own_req) {
1364 tcp_move_syn(newtp, req);
1365
1366 /* Clone pktoptions received with SYN, if we own the req */
1367 if (ireq->pktopts) {
1368 newnp->pktoptions = skb_clone(ireq->pktopts,
1369 sk_gfp_mask(sk, GFP_ATOMIC));
1370 consume_skb(ireq->pktopts);
1371 ireq->pktopts = NULL;
1372 if (newnp->pktoptions) {
1373 tcp_v6_restore_cb(newnp->pktoptions);
1374 skb_set_owner_r(newnp->pktoptions, newsk);
1375 }
1376 }
1377 }
1378
1379 return newsk;
1380
1381out_overflow:
1382 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1383out_nonewsk:
1384 dst_release(dst);
1385out:
1386 tcp_listendrop(sk);
1387 return NULL;
1388}
1389
1390/* The socket must have it's spinlock held when we get
1391 * here, unless it is a TCP_LISTEN socket.
1392 *
1393 * We have a potential double-lock case here, so even when
1394 * doing backlog processing we use the BH locking scheme.
1395 * This is because we cannot sleep with the original spinlock
1396 * held.
1397 */
1398static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1399{
1400 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1401 struct sk_buff *opt_skb = NULL;
1402 struct tcp_sock *tp;
1403
1404 /* Imagine: socket is IPv6. IPv4 packet arrives,
1405 goes to IPv4 receive handler and backlogged.
1406 From backlog it always goes here. Kerboom...
1407 Fortunately, tcp_rcv_established and rcv_established
1408 handle them correctly, but it is not case with
1409 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1410 */
1411
1412 if (skb->protocol == htons(ETH_P_IP))
1413 return tcp_v4_do_rcv(sk, skb);
1414
1415 /*
1416 * socket locking is here for SMP purposes as backlog rcv
1417 * is currently called with bh processing disabled.
1418 */
1419
1420 /* Do Stevens' IPV6_PKTOPTIONS.
1421
1422 Yes, guys, it is the only place in our code, where we
1423 may make it not affecting IPv4.
1424 The rest of code is protocol independent,
1425 and I do not like idea to uglify IPv4.
1426
1427 Actually, all the idea behind IPV6_PKTOPTIONS
1428 looks not very well thought. For now we latch
1429 options, received in the last packet, enqueued
1430 by tcp. Feel free to propose better solution.
1431 --ANK (980728)
1432 */
1433 if (np->rxopt.all)
1434 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1435
1436 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1437 struct dst_entry *dst = sk->sk_rx_dst;
1438
1439 sock_rps_save_rxhash(sk, skb);
1440 sk_mark_napi_id(sk, skb);
1441 if (dst) {
1442 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1443 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1444 dst_release(dst);
1445 sk->sk_rx_dst = NULL;
1446 }
1447 }
1448
1449 tcp_rcv_established(sk, skb);
1450 if (opt_skb)
1451 goto ipv6_pktoptions;
1452 return 0;
1453 }
1454
1455 if (tcp_checksum_complete(skb))
1456 goto csum_err;
1457
1458 if (sk->sk_state == TCP_LISTEN) {
1459 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1460
1461 if (!nsk)
1462 goto discard;
1463
1464 if (nsk != sk) {
1465 if (tcp_child_process(sk, nsk, skb))
1466 goto reset;
1467 if (opt_skb)
1468 __kfree_skb(opt_skb);
1469 return 0;
1470 }
1471 } else
1472 sock_rps_save_rxhash(sk, skb);
1473
1474 if (tcp_rcv_state_process(sk, skb))
1475 goto reset;
1476 if (opt_skb)
1477 goto ipv6_pktoptions;
1478 return 0;
1479
1480reset:
1481 tcp_v6_send_reset(sk, skb);
1482discard:
1483 if (opt_skb)
1484 __kfree_skb(opt_skb);
1485 kfree_skb(skb);
1486 return 0;
1487csum_err:
1488 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1489 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1490 goto discard;
1491
1492
1493ipv6_pktoptions:
1494 /* Do you ask, what is it?
1495
1496 1. skb was enqueued by tcp.
1497 2. skb is added to tail of read queue, rather than out of order.
1498 3. socket is not in passive state.
1499 4. Finally, it really contains options, which user wants to receive.
1500 */
1501 tp = tcp_sk(sk);
1502 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1503 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1504 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1505 np->mcast_oif = tcp_v6_iif(opt_skb);
1506 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1507 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1508 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1509 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1510 if (np->repflow)
1511 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1512 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1513 skb_set_owner_r(opt_skb, sk);
1514 tcp_v6_restore_cb(opt_skb);
1515 opt_skb = xchg(&np->pktoptions, opt_skb);
1516 } else {
1517 __kfree_skb(opt_skb);
1518 opt_skb = xchg(&np->pktoptions, NULL);
1519 }
1520 }
1521
1522 kfree_skb(opt_skb);
1523 return 0;
1524}
1525
1526static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1527 const struct tcphdr *th)
1528{
1529 /* This is tricky: we move IP6CB at its correct location into
1530 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1531 * _decode_session6() uses IP6CB().
1532 * barrier() makes sure compiler won't play aliasing games.
1533 */
1534 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1535 sizeof(struct inet6_skb_parm));
1536 barrier();
1537
1538 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1539 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1540 skb->len - th->doff*4);
1541 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1542 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1543 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1544 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1545 TCP_SKB_CB(skb)->sacked = 0;
1546 TCP_SKB_CB(skb)->has_rxtstamp =
1547 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1548}
1549
1550INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1551{
1552 struct sk_buff *skb_to_free;
1553 int sdif = inet6_sdif(skb);
1554 int dif = inet6_iif(skb);
1555 const struct tcphdr *th;
1556 const struct ipv6hdr *hdr;
1557 bool refcounted;
1558 struct sock *sk;
1559 int ret;
1560 struct net *net = dev_net(skb->dev);
1561
1562 if (skb->pkt_type != PACKET_HOST)
1563 goto discard_it;
1564
1565 /*
1566 * Count it even if it's bad.
1567 */
1568 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1569
1570 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1571 goto discard_it;
1572
1573 th = (const struct tcphdr *)skb->data;
1574
1575 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1576 goto bad_packet;
1577 if (!pskb_may_pull(skb, th->doff*4))
1578 goto discard_it;
1579
1580 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1581 goto csum_error;
1582
1583 th = (const struct tcphdr *)skb->data;
1584 hdr = ipv6_hdr(skb);
1585
1586lookup:
1587 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1588 th->source, th->dest, inet6_iif(skb), sdif,
1589 &refcounted);
1590 if (!sk)
1591 goto no_tcp_socket;
1592
1593process:
1594 if (sk->sk_state == TCP_TIME_WAIT)
1595 goto do_time_wait;
1596
1597 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1598 struct request_sock *req = inet_reqsk(sk);
1599 bool req_stolen = false;
1600 struct sock *nsk;
1601
1602 sk = req->rsk_listener;
1603 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif)) {
1604 sk_drops_add(sk, skb);
1605 reqsk_put(req);
1606 goto discard_it;
1607 }
1608 if (tcp_checksum_complete(skb)) {
1609 reqsk_put(req);
1610 goto csum_error;
1611 }
1612 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1613 inet_csk_reqsk_queue_drop_and_put(sk, req);
1614 goto lookup;
1615 }
1616 sock_hold(sk);
1617 refcounted = true;
1618 nsk = NULL;
1619 if (!tcp_filter(sk, skb)) {
1620 th = (const struct tcphdr *)skb->data;
1621 hdr = ipv6_hdr(skb);
1622 tcp_v6_fill_cb(skb, hdr, th);
1623 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1624 }
1625 if (!nsk) {
1626 reqsk_put(req);
1627 if (req_stolen) {
1628 /* Another cpu got exclusive access to req
1629 * and created a full blown socket.
1630 * Try to feed this packet to this socket
1631 * instead of discarding it.
1632 */
1633 tcp_v6_restore_cb(skb);
1634 sock_put(sk);
1635 goto lookup;
1636 }
1637 goto discard_and_relse;
1638 }
1639 if (nsk == sk) {
1640 reqsk_put(req);
1641 tcp_v6_restore_cb(skb);
1642 } else if (tcp_child_process(sk, nsk, skb)) {
1643 tcp_v6_send_reset(nsk, skb);
1644 goto discard_and_relse;
1645 } else {
1646 sock_put(sk);
1647 return 0;
1648 }
1649 }
1650 if (hdr->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
1651 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1652 goto discard_and_relse;
1653 }
1654
1655 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1656 goto discard_and_relse;
1657
1658 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif))
1659 goto discard_and_relse;
1660
1661 if (tcp_filter(sk, skb))
1662 goto discard_and_relse;
1663 th = (const struct tcphdr *)skb->data;
1664 hdr = ipv6_hdr(skb);
1665 tcp_v6_fill_cb(skb, hdr, th);
1666
1667 skb->dev = NULL;
1668
1669 if (sk->sk_state == TCP_LISTEN) {
1670 ret = tcp_v6_do_rcv(sk, skb);
1671 goto put_and_return;
1672 }
1673
1674 sk_incoming_cpu_update(sk);
1675
1676 bh_lock_sock_nested(sk);
1677 tcp_segs_in(tcp_sk(sk), skb);
1678 ret = 0;
1679 if (!sock_owned_by_user(sk)) {
1680 skb_to_free = sk->sk_rx_skb_cache;
1681 sk->sk_rx_skb_cache = NULL;
1682 ret = tcp_v6_do_rcv(sk, skb);
1683 } else {
1684 if (tcp_add_backlog(sk, skb))
1685 goto discard_and_relse;
1686 skb_to_free = NULL;
1687 }
1688 bh_unlock_sock(sk);
1689 if (skb_to_free)
1690 __kfree_skb(skb_to_free);
1691put_and_return:
1692 if (refcounted)
1693 sock_put(sk);
1694 return ret ? -1 : 0;
1695
1696no_tcp_socket:
1697 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1698 goto discard_it;
1699
1700 tcp_v6_fill_cb(skb, hdr, th);
1701
1702 if (tcp_checksum_complete(skb)) {
1703csum_error:
1704 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1705bad_packet:
1706 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1707 } else {
1708 tcp_v6_send_reset(NULL, skb);
1709 }
1710
1711discard_it:
1712 kfree_skb(skb);
1713 return 0;
1714
1715discard_and_relse:
1716 sk_drops_add(sk, skb);
1717 if (refcounted)
1718 sock_put(sk);
1719 goto discard_it;
1720
1721do_time_wait:
1722 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1723 inet_twsk_put(inet_twsk(sk));
1724 goto discard_it;
1725 }
1726
1727 tcp_v6_fill_cb(skb, hdr, th);
1728
1729 if (tcp_checksum_complete(skb)) {
1730 inet_twsk_put(inet_twsk(sk));
1731 goto csum_error;
1732 }
1733
1734 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1735 case TCP_TW_SYN:
1736 {
1737 struct sock *sk2;
1738
1739 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1740 skb, __tcp_hdrlen(th),
1741 &ipv6_hdr(skb)->saddr, th->source,
1742 &ipv6_hdr(skb)->daddr,
1743 ntohs(th->dest),
1744 tcp_v6_iif_l3_slave(skb),
1745 sdif);
1746 if (sk2) {
1747 struct inet_timewait_sock *tw = inet_twsk(sk);
1748 inet_twsk_deschedule_put(tw);
1749 sk = sk2;
1750 tcp_v6_restore_cb(skb);
1751 refcounted = false;
1752 goto process;
1753 }
1754 }
1755 /* to ACK */
1756 fallthrough;
1757 case TCP_TW_ACK:
1758 tcp_v6_timewait_ack(sk, skb);
1759 break;
1760 case TCP_TW_RST:
1761 tcp_v6_send_reset(sk, skb);
1762 inet_twsk_deschedule_put(inet_twsk(sk));
1763 goto discard_it;
1764 case TCP_TW_SUCCESS:
1765 ;
1766 }
1767 goto discard_it;
1768}
1769
1770INDIRECT_CALLABLE_SCOPE void tcp_v6_early_demux(struct sk_buff *skb)
1771{
1772 const struct ipv6hdr *hdr;
1773 const struct tcphdr *th;
1774 struct sock *sk;
1775
1776 if (skb->pkt_type != PACKET_HOST)
1777 return;
1778
1779 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1780 return;
1781
1782 hdr = ipv6_hdr(skb);
1783 th = tcp_hdr(skb);
1784
1785 if (th->doff < sizeof(struct tcphdr) / 4)
1786 return;
1787
1788 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1789 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1790 &hdr->saddr, th->source,
1791 &hdr->daddr, ntohs(th->dest),
1792 inet6_iif(skb), inet6_sdif(skb));
1793 if (sk) {
1794 skb->sk = sk;
1795 skb->destructor = sock_edemux;
1796 if (sk_fullsock(sk)) {
1797 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1798
1799 if (dst)
1800 dst = dst_check(dst, tcp_inet6_sk(sk)->rx_dst_cookie);
1801 if (dst &&
1802 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1803 skb_dst_set_noref(skb, dst);
1804 }
1805 }
1806}
1807
1808static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1809 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1810 .twsk_unique = tcp_twsk_unique,
1811 .twsk_destructor = tcp_twsk_destructor,
1812};
1813
1814INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
1815{
1816 struct ipv6_pinfo *np = inet6_sk(sk);
1817
1818 __tcp_v6_send_check(skb, &np->saddr, &sk->sk_v6_daddr);
1819}
1820
1821const struct inet_connection_sock_af_ops ipv6_specific = {
1822 .queue_xmit = inet6_csk_xmit,
1823 .send_check = tcp_v6_send_check,
1824 .rebuild_header = inet6_sk_rebuild_header,
1825 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1826 .conn_request = tcp_v6_conn_request,
1827 .syn_recv_sock = tcp_v6_syn_recv_sock,
1828 .net_header_len = sizeof(struct ipv6hdr),
1829 .net_frag_header_len = sizeof(struct frag_hdr),
1830 .setsockopt = ipv6_setsockopt,
1831 .getsockopt = ipv6_getsockopt,
1832 .addr2sockaddr = inet6_csk_addr2sockaddr,
1833 .sockaddr_len = sizeof(struct sockaddr_in6),
1834 .mtu_reduced = tcp_v6_mtu_reduced,
1835};
1836
1837#ifdef CONFIG_TCP_MD5SIG
1838static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1839 .md5_lookup = tcp_v6_md5_lookup,
1840 .calc_md5_hash = tcp_v6_md5_hash_skb,
1841 .md5_parse = tcp_v6_parse_md5_keys,
1842};
1843#endif
1844
1845/*
1846 * TCP over IPv4 via INET6 API
1847 */
1848static const struct inet_connection_sock_af_ops ipv6_mapped = {
1849 .queue_xmit = ip_queue_xmit,
1850 .send_check = tcp_v4_send_check,
1851 .rebuild_header = inet_sk_rebuild_header,
1852 .sk_rx_dst_set = inet_sk_rx_dst_set,
1853 .conn_request = tcp_v6_conn_request,
1854 .syn_recv_sock = tcp_v6_syn_recv_sock,
1855 .net_header_len = sizeof(struct iphdr),
1856 .setsockopt = ipv6_setsockopt,
1857 .getsockopt = ipv6_getsockopt,
1858 .addr2sockaddr = inet6_csk_addr2sockaddr,
1859 .sockaddr_len = sizeof(struct sockaddr_in6),
1860 .mtu_reduced = tcp_v4_mtu_reduced,
1861};
1862
1863#ifdef CONFIG_TCP_MD5SIG
1864static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1865 .md5_lookup = tcp_v4_md5_lookup,
1866 .calc_md5_hash = tcp_v4_md5_hash_skb,
1867 .md5_parse = tcp_v6_parse_md5_keys,
1868};
1869#endif
1870
1871/* NOTE: A lot of things set to zero explicitly by call to
1872 * sk_alloc() so need not be done here.
1873 */
1874static int tcp_v6_init_sock(struct sock *sk)
1875{
1876 struct inet_connection_sock *icsk = inet_csk(sk);
1877
1878 tcp_init_sock(sk);
1879
1880 icsk->icsk_af_ops = &ipv6_specific;
1881
1882#ifdef CONFIG_TCP_MD5SIG
1883 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1884#endif
1885
1886 return 0;
1887}
1888
1889static void tcp_v6_destroy_sock(struct sock *sk)
1890{
1891 tcp_v4_destroy_sock(sk);
1892 inet6_destroy_sock(sk);
1893}
1894
1895#ifdef CONFIG_PROC_FS
1896/* Proc filesystem TCPv6 sock list dumping. */
1897static void get_openreq6(struct seq_file *seq,
1898 const struct request_sock *req, int i)
1899{
1900 long ttd = req->rsk_timer.expires - jiffies;
1901 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1902 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1903
1904 if (ttd < 0)
1905 ttd = 0;
1906
1907 seq_printf(seq,
1908 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1909 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1910 i,
1911 src->s6_addr32[0], src->s6_addr32[1],
1912 src->s6_addr32[2], src->s6_addr32[3],
1913 inet_rsk(req)->ir_num,
1914 dest->s6_addr32[0], dest->s6_addr32[1],
1915 dest->s6_addr32[2], dest->s6_addr32[3],
1916 ntohs(inet_rsk(req)->ir_rmt_port),
1917 TCP_SYN_RECV,
1918 0, 0, /* could print option size, but that is af dependent. */
1919 1, /* timers active (only the expire timer) */
1920 jiffies_to_clock_t(ttd),
1921 req->num_timeout,
1922 from_kuid_munged(seq_user_ns(seq),
1923 sock_i_uid(req->rsk_listener)),
1924 0, /* non standard timer */
1925 0, /* open_requests have no inode */
1926 0, req);
1927}
1928
1929static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1930{
1931 const struct in6_addr *dest, *src;
1932 __u16 destp, srcp;
1933 int timer_active;
1934 unsigned long timer_expires;
1935 const struct inet_sock *inet = inet_sk(sp);
1936 const struct tcp_sock *tp = tcp_sk(sp);
1937 const struct inet_connection_sock *icsk = inet_csk(sp);
1938 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1939 int rx_queue;
1940 int state;
1941
1942 dest = &sp->sk_v6_daddr;
1943 src = &sp->sk_v6_rcv_saddr;
1944 destp = ntohs(inet->inet_dport);
1945 srcp = ntohs(inet->inet_sport);
1946
1947 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1948 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
1949 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1950 timer_active = 1;
1951 timer_expires = icsk->icsk_timeout;
1952 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1953 timer_active = 4;
1954 timer_expires = icsk->icsk_timeout;
1955 } else if (timer_pending(&sp->sk_timer)) {
1956 timer_active = 2;
1957 timer_expires = sp->sk_timer.expires;
1958 } else {
1959 timer_active = 0;
1960 timer_expires = jiffies;
1961 }
1962
1963 state = inet_sk_state_load(sp);
1964 if (state == TCP_LISTEN)
1965 rx_queue = READ_ONCE(sp->sk_ack_backlog);
1966 else
1967 /* Because we don't lock the socket,
1968 * we might find a transient negative value.
1969 */
1970 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
1971 READ_ONCE(tp->copied_seq), 0);
1972
1973 seq_printf(seq,
1974 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1975 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1976 i,
1977 src->s6_addr32[0], src->s6_addr32[1],
1978 src->s6_addr32[2], src->s6_addr32[3], srcp,
1979 dest->s6_addr32[0], dest->s6_addr32[1],
1980 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1981 state,
1982 READ_ONCE(tp->write_seq) - tp->snd_una,
1983 rx_queue,
1984 timer_active,
1985 jiffies_delta_to_clock_t(timer_expires - jiffies),
1986 icsk->icsk_retransmits,
1987 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1988 icsk->icsk_probes_out,
1989 sock_i_ino(sp),
1990 refcount_read(&sp->sk_refcnt), sp,
1991 jiffies_to_clock_t(icsk->icsk_rto),
1992 jiffies_to_clock_t(icsk->icsk_ack.ato),
1993 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
1994 tp->snd_cwnd,
1995 state == TCP_LISTEN ?
1996 fastopenq->max_qlen :
1997 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1998 );
1999}
2000
2001static void get_timewait6_sock(struct seq_file *seq,
2002 struct inet_timewait_sock *tw, int i)
2003{
2004 long delta = tw->tw_timer.expires - jiffies;
2005 const struct in6_addr *dest, *src;
2006 __u16 destp, srcp;
2007
2008 dest = &tw->tw_v6_daddr;
2009 src = &tw->tw_v6_rcv_saddr;
2010 destp = ntohs(tw->tw_dport);
2011 srcp = ntohs(tw->tw_sport);
2012
2013 seq_printf(seq,
2014 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2015 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2016 i,
2017 src->s6_addr32[0], src->s6_addr32[1],
2018 src->s6_addr32[2], src->s6_addr32[3], srcp,
2019 dest->s6_addr32[0], dest->s6_addr32[1],
2020 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2021 tw->tw_substate, 0, 0,
2022 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2023 refcount_read(&tw->tw_refcnt), tw);
2024}
2025
2026static int tcp6_seq_show(struct seq_file *seq, void *v)
2027{
2028 struct tcp_iter_state *st;
2029 struct sock *sk = v;
2030
2031 if (v == SEQ_START_TOKEN) {
2032 seq_puts(seq,
2033 " sl "
2034 "local_address "
2035 "remote_address "
2036 "st tx_queue rx_queue tr tm->when retrnsmt"
2037 " uid timeout inode\n");
2038 goto out;
2039 }
2040 st = seq->private;
2041
2042 if (sk->sk_state == TCP_TIME_WAIT)
2043 get_timewait6_sock(seq, v, st->num);
2044 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2045 get_openreq6(seq, v, st->num);
2046 else
2047 get_tcp6_sock(seq, v, st->num);
2048out:
2049 return 0;
2050}
2051
2052static const struct seq_operations tcp6_seq_ops = {
2053 .show = tcp6_seq_show,
2054 .start = tcp_seq_start,
2055 .next = tcp_seq_next,
2056 .stop = tcp_seq_stop,
2057};
2058
2059static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2060 .family = AF_INET6,
2061};
2062
2063int __net_init tcp6_proc_init(struct net *net)
2064{
2065 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2066 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2067 return -ENOMEM;
2068 return 0;
2069}
2070
2071void tcp6_proc_exit(struct net *net)
2072{
2073 remove_proc_entry("tcp6", net->proc_net);
2074}
2075#endif
2076
2077struct proto tcpv6_prot = {
2078 .name = "TCPv6",
2079 .owner = THIS_MODULE,
2080 .close = tcp_close,
2081 .pre_connect = tcp_v6_pre_connect,
2082 .connect = tcp_v6_connect,
2083 .disconnect = tcp_disconnect,
2084 .accept = inet_csk_accept,
2085 .ioctl = tcp_ioctl,
2086 .init = tcp_v6_init_sock,
2087 .destroy = tcp_v6_destroy_sock,
2088 .shutdown = tcp_shutdown,
2089 .setsockopt = tcp_setsockopt,
2090 .getsockopt = tcp_getsockopt,
2091 .keepalive = tcp_set_keepalive,
2092 .recvmsg = tcp_recvmsg,
2093 .sendmsg = tcp_sendmsg,
2094 .sendpage = tcp_sendpage,
2095 .backlog_rcv = tcp_v6_do_rcv,
2096 .release_cb = tcp_release_cb,
2097 .hash = inet6_hash,
2098 .unhash = inet_unhash,
2099 .get_port = inet_csk_get_port,
2100 .enter_memory_pressure = tcp_enter_memory_pressure,
2101 .leave_memory_pressure = tcp_leave_memory_pressure,
2102 .stream_memory_free = tcp_stream_memory_free,
2103 .sockets_allocated = &tcp_sockets_allocated,
2104 .memory_allocated = &tcp_memory_allocated,
2105 .memory_pressure = &tcp_memory_pressure,
2106 .orphan_count = &tcp_orphan_count,
2107 .sysctl_mem = sysctl_tcp_mem,
2108 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2109 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2110 .max_header = MAX_TCP_HEADER,
2111 .obj_size = sizeof(struct tcp6_sock),
2112 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2113 .twsk_prot = &tcp6_timewait_sock_ops,
2114 .rsk_prot = &tcp6_request_sock_ops,
2115 .h.hashinfo = &tcp_hashinfo,
2116 .no_autobind = true,
2117 .diag_destroy = tcp_abort,
2118};
2119EXPORT_SYMBOL_GPL(tcpv6_prot);
2120
2121/* thinking of making this const? Don't.
2122 * early_demux can change based on sysctl.
2123 */
2124static struct inet6_protocol tcpv6_protocol = {
2125 .early_demux = tcp_v6_early_demux,
2126 .early_demux_handler = tcp_v6_early_demux,
2127 .handler = tcp_v6_rcv,
2128 .err_handler = tcp_v6_err,
2129 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2130};
2131
2132static struct inet_protosw tcpv6_protosw = {
2133 .type = SOCK_STREAM,
2134 .protocol = IPPROTO_TCP,
2135 .prot = &tcpv6_prot,
2136 .ops = &inet6_stream_ops,
2137 .flags = INET_PROTOSW_PERMANENT |
2138 INET_PROTOSW_ICSK,
2139};
2140
2141static int __net_init tcpv6_net_init(struct net *net)
2142{
2143 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2144 SOCK_RAW, IPPROTO_TCP, net);
2145}
2146
2147static void __net_exit tcpv6_net_exit(struct net *net)
2148{
2149 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2150}
2151
2152static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2153{
2154 inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2155}
2156
2157static struct pernet_operations tcpv6_net_ops = {
2158 .init = tcpv6_net_init,
2159 .exit = tcpv6_net_exit,
2160 .exit_batch = tcpv6_net_exit_batch,
2161};
2162
2163int __init tcpv6_init(void)
2164{
2165 int ret;
2166
2167 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2168 if (ret)
2169 goto out;
2170
2171 /* register inet6 protocol */
2172 ret = inet6_register_protosw(&tcpv6_protosw);
2173 if (ret)
2174 goto out_tcpv6_protocol;
2175
2176 ret = register_pernet_subsys(&tcpv6_net_ops);
2177 if (ret)
2178 goto out_tcpv6_protosw;
2179
2180 ret = mptcpv6_init();
2181 if (ret)
2182 goto out_tcpv6_pernet_subsys;
2183
2184out:
2185 return ret;
2186
2187out_tcpv6_pernet_subsys:
2188 unregister_pernet_subsys(&tcpv6_net_ops);
2189out_tcpv6_protosw:
2190 inet6_unregister_protosw(&tcpv6_protosw);
2191out_tcpv6_protocol:
2192 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2193 goto out;
2194}
2195
2196void tcpv6_exit(void)
2197{
2198 unregister_pernet_subsys(&tcpv6_net_ops);
2199 inet6_unregister_protosw(&tcpv6_protosw);
2200 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2201}