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
75INDIRECT_CALLABLE_SCOPE 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#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
80static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
81static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
82#endif
83
84/* Helper returning the inet6 address from a given tcp socket.
85 * It can be used in TCP stack instead of inet6_sk(sk).
86 * This avoids a dereference and allow compiler optimizations.
87 * It is a specialized version of inet6_sk_generic().
88 */
89#define tcp_inet6_sk(sk) (&container_of_const(tcp_sk(sk), \
90 struct tcp6_sock, tcp)->inet6)
91
92static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
93{
94 struct dst_entry *dst = skb_dst(skb);
95
96 if (dst && dst_hold_safe(dst)) {
97 const struct rt6_info *rt = (const struct rt6_info *)dst;
98
99 rcu_assign_pointer(sk->sk_rx_dst, dst);
100 sk->sk_rx_dst_ifindex = skb->skb_iif;
101 sk->sk_rx_dst_cookie = rt6_get_cookie(rt);
102 }
103}
104
105static u32 tcp_v6_init_seq(const struct sk_buff *skb)
106{
107 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
108 ipv6_hdr(skb)->saddr.s6_addr32,
109 tcp_hdr(skb)->dest,
110 tcp_hdr(skb)->source);
111}
112
113static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
114{
115 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
116 ipv6_hdr(skb)->saddr.s6_addr32);
117}
118
119static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
120 int addr_len)
121{
122 /* This check is replicated from tcp_v6_connect() and intended to
123 * prevent BPF program called below from accessing bytes that are out
124 * of the bound specified by user in addr_len.
125 */
126 if (addr_len < SIN6_LEN_RFC2133)
127 return -EINVAL;
128
129 sock_owned_by_me(sk);
130
131 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr, &addr_len);
132}
133
134static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
135 int addr_len)
136{
137 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
138 struct inet_connection_sock *icsk = inet_csk(sk);
139 struct in6_addr *saddr = NULL, *final_p, final;
140 struct inet_timewait_death_row *tcp_death_row;
141 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
142 struct inet_sock *inet = inet_sk(sk);
143 struct tcp_sock *tp = tcp_sk(sk);
144 struct net *net = sock_net(sk);
145 struct ipv6_txoptions *opt;
146 struct dst_entry *dst;
147 struct flowi6 fl6;
148 int addr_type;
149 int err;
150
151 if (addr_len < SIN6_LEN_RFC2133)
152 return -EINVAL;
153
154 if (usin->sin6_family != AF_INET6)
155 return -EAFNOSUPPORT;
156
157 memset(&fl6, 0, sizeof(fl6));
158
159 if (inet6_test_bit(SNDFLOW, sk)) {
160 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
161 IP6_ECN_flow_init(fl6.flowlabel);
162 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
163 struct ip6_flowlabel *flowlabel;
164 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
165 if (IS_ERR(flowlabel))
166 return -EINVAL;
167 fl6_sock_release(flowlabel);
168 }
169 }
170
171 /*
172 * connect() to INADDR_ANY means loopback (BSD'ism).
173 */
174
175 if (ipv6_addr_any(&usin->sin6_addr)) {
176 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
177 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
178 &usin->sin6_addr);
179 else
180 usin->sin6_addr = in6addr_loopback;
181 }
182
183 addr_type = ipv6_addr_type(&usin->sin6_addr);
184
185 if (addr_type & IPV6_ADDR_MULTICAST)
186 return -ENETUNREACH;
187
188 if (addr_type&IPV6_ADDR_LINKLOCAL) {
189 if (addr_len >= sizeof(struct sockaddr_in6) &&
190 usin->sin6_scope_id) {
191 /* If interface is set while binding, indices
192 * must coincide.
193 */
194 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
195 return -EINVAL;
196
197 sk->sk_bound_dev_if = usin->sin6_scope_id;
198 }
199
200 /* Connect to link-local address requires an interface */
201 if (!sk->sk_bound_dev_if)
202 return -EINVAL;
203 }
204
205 if (tp->rx_opt.ts_recent_stamp &&
206 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
207 tp->rx_opt.ts_recent = 0;
208 tp->rx_opt.ts_recent_stamp = 0;
209 WRITE_ONCE(tp->write_seq, 0);
210 }
211
212 sk->sk_v6_daddr = usin->sin6_addr;
213 np->flow_label = fl6.flowlabel;
214
215 /*
216 * TCP over IPv4
217 */
218
219 if (addr_type & IPV6_ADDR_MAPPED) {
220 u32 exthdrlen = icsk->icsk_ext_hdr_len;
221 struct sockaddr_in sin;
222
223 if (ipv6_only_sock(sk))
224 return -ENETUNREACH;
225
226 sin.sin_family = AF_INET;
227 sin.sin_port = usin->sin6_port;
228 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
229
230 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
231 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_mapped);
232 if (sk_is_mptcp(sk))
233 mptcpv6_handle_mapped(sk, true);
234 sk->sk_backlog_rcv = tcp_v4_do_rcv;
235#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
236 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
237#endif
238
239 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
240
241 if (err) {
242 icsk->icsk_ext_hdr_len = exthdrlen;
243 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
244 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_specific);
245 if (sk_is_mptcp(sk))
246 mptcpv6_handle_mapped(sk, false);
247 sk->sk_backlog_rcv = tcp_v6_do_rcv;
248#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
249 tp->af_specific = &tcp_sock_ipv6_specific;
250#endif
251 goto failure;
252 }
253 np->saddr = sk->sk_v6_rcv_saddr;
254
255 return err;
256 }
257
258 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
259 saddr = &sk->sk_v6_rcv_saddr;
260
261 fl6.flowi6_proto = IPPROTO_TCP;
262 fl6.daddr = sk->sk_v6_daddr;
263 fl6.saddr = saddr ? *saddr : np->saddr;
264 fl6.flowlabel = ip6_make_flowinfo(np->tclass, np->flow_label);
265 fl6.flowi6_oif = sk->sk_bound_dev_if;
266 fl6.flowi6_mark = sk->sk_mark;
267 fl6.fl6_dport = usin->sin6_port;
268 fl6.fl6_sport = inet->inet_sport;
269 fl6.flowi6_uid = sk->sk_uid;
270
271 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
272 final_p = fl6_update_dst(&fl6, opt, &final);
273
274 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
275
276 dst = ip6_dst_lookup_flow(net, sk, &fl6, final_p);
277 if (IS_ERR(dst)) {
278 err = PTR_ERR(dst);
279 goto failure;
280 }
281
282 tp->tcp_usec_ts = dst_tcp_usec_ts(dst);
283 tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
284
285 if (!saddr) {
286 saddr = &fl6.saddr;
287
288 err = inet_bhash2_update_saddr(sk, saddr, AF_INET6);
289 if (err)
290 goto failure;
291 }
292
293 /* set the source address */
294 np->saddr = *saddr;
295 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
296
297 sk->sk_gso_type = SKB_GSO_TCPV6;
298 ip6_dst_store(sk, dst, NULL, NULL);
299
300 icsk->icsk_ext_hdr_len = 0;
301 if (opt)
302 icsk->icsk_ext_hdr_len = opt->opt_flen +
303 opt->opt_nflen;
304
305 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
306
307 inet->inet_dport = usin->sin6_port;
308
309 tcp_set_state(sk, TCP_SYN_SENT);
310 err = inet6_hash_connect(tcp_death_row, sk);
311 if (err)
312 goto late_failure;
313
314 sk_set_txhash(sk);
315
316 if (likely(!tp->repair)) {
317 if (!tp->write_seq)
318 WRITE_ONCE(tp->write_seq,
319 secure_tcpv6_seq(np->saddr.s6_addr32,
320 sk->sk_v6_daddr.s6_addr32,
321 inet->inet_sport,
322 inet->inet_dport));
323 tp->tsoffset = secure_tcpv6_ts_off(net, np->saddr.s6_addr32,
324 sk->sk_v6_daddr.s6_addr32);
325 }
326
327 if (tcp_fastopen_defer_connect(sk, &err))
328 return err;
329 if (err)
330 goto late_failure;
331
332 err = tcp_connect(sk);
333 if (err)
334 goto late_failure;
335
336 return 0;
337
338late_failure:
339 tcp_set_state(sk, TCP_CLOSE);
340 inet_bhash2_reset_saddr(sk);
341failure:
342 inet->inet_dport = 0;
343 sk->sk_route_caps = 0;
344 return err;
345}
346
347static void tcp_v6_mtu_reduced(struct sock *sk)
348{
349 struct dst_entry *dst;
350 u32 mtu;
351
352 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
353 return;
354
355 mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
356
357 /* Drop requests trying to increase our current mss.
358 * Check done in __ip6_rt_update_pmtu() is too late.
359 */
360 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache)
361 return;
362
363 dst = inet6_csk_update_pmtu(sk, mtu);
364 if (!dst)
365 return;
366
367 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
368 tcp_sync_mss(sk, dst_mtu(dst));
369 tcp_simple_retransmit(sk);
370 }
371}
372
373static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
374 u8 type, u8 code, int offset, __be32 info)
375{
376 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
377 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
378 struct net *net = dev_net(skb->dev);
379 struct request_sock *fastopen;
380 struct ipv6_pinfo *np;
381 struct tcp_sock *tp;
382 __u32 seq, snd_una;
383 struct sock *sk;
384 bool fatal;
385 int err;
386
387 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
388 &hdr->daddr, th->dest,
389 &hdr->saddr, ntohs(th->source),
390 skb->dev->ifindex, inet6_sdif(skb));
391
392 if (!sk) {
393 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
394 ICMP6_MIB_INERRORS);
395 return -ENOENT;
396 }
397
398 if (sk->sk_state == TCP_TIME_WAIT) {
399 /* To increase the counter of ignored icmps for TCP-AO */
400 tcp_ao_ignore_icmp(sk, AF_INET6, type, code);
401 inet_twsk_put(inet_twsk(sk));
402 return 0;
403 }
404 seq = ntohl(th->seq);
405 fatal = icmpv6_err_convert(type, code, &err);
406 if (sk->sk_state == TCP_NEW_SYN_RECV) {
407 tcp_req_err(sk, seq, fatal);
408 return 0;
409 }
410
411 if (tcp_ao_ignore_icmp(sk, AF_INET6, type, code)) {
412 sock_put(sk);
413 return 0;
414 }
415
416 bh_lock_sock(sk);
417 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
418 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
419
420 if (sk->sk_state == TCP_CLOSE)
421 goto out;
422
423 if (static_branch_unlikely(&ip6_min_hopcount)) {
424 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
425 if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
426 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
427 goto out;
428 }
429 }
430
431 tp = tcp_sk(sk);
432 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
433 fastopen = rcu_dereference(tp->fastopen_rsk);
434 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
435 if (sk->sk_state != TCP_LISTEN &&
436 !between(seq, snd_una, tp->snd_nxt)) {
437 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
438 goto out;
439 }
440
441 np = tcp_inet6_sk(sk);
442
443 if (type == NDISC_REDIRECT) {
444 if (!sock_owned_by_user(sk)) {
445 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
446
447 if (dst)
448 dst->ops->redirect(dst, sk, skb);
449 }
450 goto out;
451 }
452
453 if (type == ICMPV6_PKT_TOOBIG) {
454 u32 mtu = ntohl(info);
455
456 /* We are not interested in TCP_LISTEN and open_requests
457 * (SYN-ACKs send out by Linux are always <576bytes so
458 * they should go through unfragmented).
459 */
460 if (sk->sk_state == TCP_LISTEN)
461 goto out;
462
463 if (!ip6_sk_accept_pmtu(sk))
464 goto out;
465
466 if (mtu < IPV6_MIN_MTU)
467 goto out;
468
469 WRITE_ONCE(tp->mtu_info, mtu);
470
471 if (!sock_owned_by_user(sk))
472 tcp_v6_mtu_reduced(sk);
473 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
474 &sk->sk_tsq_flags))
475 sock_hold(sk);
476 goto out;
477 }
478
479
480 /* Might be for an request_sock */
481 switch (sk->sk_state) {
482 case TCP_SYN_SENT:
483 case TCP_SYN_RECV:
484 /* Only in fast or simultaneous open. If a fast open socket is
485 * already accepted it is treated as a connected one below.
486 */
487 if (fastopen && !fastopen->sk)
488 break;
489
490 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
491
492 if (!sock_owned_by_user(sk)) {
493 WRITE_ONCE(sk->sk_err, err);
494 sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
495
496 tcp_done(sk);
497 } else {
498 WRITE_ONCE(sk->sk_err_soft, err);
499 }
500 goto out;
501 case TCP_LISTEN:
502 break;
503 default:
504 /* check if this ICMP message allows revert of backoff.
505 * (see RFC 6069)
506 */
507 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
508 code == ICMPV6_NOROUTE)
509 tcp_ld_RTO_revert(sk, seq);
510 }
511
512 if (!sock_owned_by_user(sk) && inet6_test_bit(RECVERR6, sk)) {
513 WRITE_ONCE(sk->sk_err, err);
514 sk_error_report(sk);
515 } else {
516 WRITE_ONCE(sk->sk_err_soft, err);
517 }
518out:
519 bh_unlock_sock(sk);
520 sock_put(sk);
521 return 0;
522}
523
524
525static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
526 struct flowi *fl,
527 struct request_sock *req,
528 struct tcp_fastopen_cookie *foc,
529 enum tcp_synack_type synack_type,
530 struct sk_buff *syn_skb)
531{
532 struct inet_request_sock *ireq = inet_rsk(req);
533 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
534 struct ipv6_txoptions *opt;
535 struct flowi6 *fl6 = &fl->u.ip6;
536 struct sk_buff *skb;
537 int err = -ENOMEM;
538 u8 tclass;
539
540 /* First, grab a route. */
541 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
542 IPPROTO_TCP)) == NULL)
543 goto done;
544
545 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
546
547 if (skb) {
548 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
549 &ireq->ir_v6_rmt_addr);
550
551 fl6->daddr = ireq->ir_v6_rmt_addr;
552 if (inet6_test_bit(REPFLOW, sk) && ireq->pktopts)
553 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
554
555 tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ?
556 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
557 (np->tclass & INET_ECN_MASK) :
558 np->tclass;
559
560 if (!INET_ECN_is_capable(tclass) &&
561 tcp_bpf_ca_needs_ecn((struct sock *)req))
562 tclass |= INET_ECN_ECT_0;
563
564 rcu_read_lock();
565 opt = ireq->ipv6_opt;
566 if (!opt)
567 opt = rcu_dereference(np->opt);
568 err = ip6_xmit(sk, skb, fl6, skb->mark ? : READ_ONCE(sk->sk_mark),
569 opt, tclass, READ_ONCE(sk->sk_priority));
570 rcu_read_unlock();
571 err = net_xmit_eval(err);
572 }
573
574done:
575 return err;
576}
577
578
579static void tcp_v6_reqsk_destructor(struct request_sock *req)
580{
581 kfree(inet_rsk(req)->ipv6_opt);
582 consume_skb(inet_rsk(req)->pktopts);
583}
584
585#ifdef CONFIG_TCP_MD5SIG
586static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
587 const struct in6_addr *addr,
588 int l3index)
589{
590 return tcp_md5_do_lookup(sk, l3index,
591 (union tcp_md5_addr *)addr, AF_INET6);
592}
593
594static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
595 const struct sock *addr_sk)
596{
597 int l3index;
598
599 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
600 addr_sk->sk_bound_dev_if);
601 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
602 l3index);
603}
604
605static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
606 sockptr_t optval, int optlen)
607{
608 struct tcp_md5sig cmd;
609 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
610 union tcp_ao_addr *addr;
611 int l3index = 0;
612 u8 prefixlen;
613 bool l3flag;
614 u8 flags;
615
616 if (optlen < sizeof(cmd))
617 return -EINVAL;
618
619 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
620 return -EFAULT;
621
622 if (sin6->sin6_family != AF_INET6)
623 return -EINVAL;
624
625 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
626 l3flag = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
627
628 if (optname == TCP_MD5SIG_EXT &&
629 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
630 prefixlen = cmd.tcpm_prefixlen;
631 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
632 prefixlen > 32))
633 return -EINVAL;
634 } else {
635 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
636 }
637
638 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
639 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
640 struct net_device *dev;
641
642 rcu_read_lock();
643 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
644 if (dev && netif_is_l3_master(dev))
645 l3index = dev->ifindex;
646 rcu_read_unlock();
647
648 /* ok to reference set/not set outside of rcu;
649 * right now device MUST be an L3 master
650 */
651 if (!dev || !l3index)
652 return -EINVAL;
653 }
654
655 if (!cmd.tcpm_keylen) {
656 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
657 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
658 AF_INET, prefixlen,
659 l3index, flags);
660 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
661 AF_INET6, prefixlen, l3index, flags);
662 }
663
664 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
665 return -EINVAL;
666
667 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
668 addr = (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3];
669
670 /* Don't allow keys for peers that have a matching TCP-AO key.
671 * See the comment in tcp_ao_add_cmd()
672 */
673 if (tcp_ao_required(sk, addr, AF_INET,
674 l3flag ? l3index : -1, false))
675 return -EKEYREJECTED;
676 return tcp_md5_do_add(sk, addr,
677 AF_INET, prefixlen, l3index, flags,
678 cmd.tcpm_key, cmd.tcpm_keylen);
679 }
680
681 addr = (union tcp_md5_addr *)&sin6->sin6_addr;
682
683 /* Don't allow keys for peers that have a matching TCP-AO key.
684 * See the comment in tcp_ao_add_cmd()
685 */
686 if (tcp_ao_required(sk, addr, AF_INET6, l3flag ? l3index : -1, false))
687 return -EKEYREJECTED;
688
689 return tcp_md5_do_add(sk, addr, AF_INET6, prefixlen, l3index, flags,
690 cmd.tcpm_key, cmd.tcpm_keylen);
691}
692
693static int tcp_v6_md5_hash_headers(struct tcp_sigpool *hp,
694 const struct in6_addr *daddr,
695 const struct in6_addr *saddr,
696 const struct tcphdr *th, int nbytes)
697{
698 struct tcp6_pseudohdr *bp;
699 struct scatterlist sg;
700 struct tcphdr *_th;
701
702 bp = hp->scratch;
703 /* 1. TCP pseudo-header (RFC2460) */
704 bp->saddr = *saddr;
705 bp->daddr = *daddr;
706 bp->protocol = cpu_to_be32(IPPROTO_TCP);
707 bp->len = cpu_to_be32(nbytes);
708
709 _th = (struct tcphdr *)(bp + 1);
710 memcpy(_th, th, sizeof(*th));
711 _th->check = 0;
712
713 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
714 ahash_request_set_crypt(hp->req, &sg, NULL,
715 sizeof(*bp) + sizeof(*th));
716 return crypto_ahash_update(hp->req);
717}
718
719static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
720 const struct in6_addr *daddr, struct in6_addr *saddr,
721 const struct tcphdr *th)
722{
723 struct tcp_sigpool hp;
724
725 if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
726 goto clear_hash_nostart;
727
728 if (crypto_ahash_init(hp.req))
729 goto clear_hash;
730 if (tcp_v6_md5_hash_headers(&hp, daddr, saddr, th, th->doff << 2))
731 goto clear_hash;
732 if (tcp_md5_hash_key(&hp, key))
733 goto clear_hash;
734 ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
735 if (crypto_ahash_final(hp.req))
736 goto clear_hash;
737
738 tcp_sigpool_end(&hp);
739 return 0;
740
741clear_hash:
742 tcp_sigpool_end(&hp);
743clear_hash_nostart:
744 memset(md5_hash, 0, 16);
745 return 1;
746}
747
748static int tcp_v6_md5_hash_skb(char *md5_hash,
749 const struct tcp_md5sig_key *key,
750 const struct sock *sk,
751 const struct sk_buff *skb)
752{
753 const struct tcphdr *th = tcp_hdr(skb);
754 const struct in6_addr *saddr, *daddr;
755 struct tcp_sigpool hp;
756
757 if (sk) { /* valid for establish/request sockets */
758 saddr = &sk->sk_v6_rcv_saddr;
759 daddr = &sk->sk_v6_daddr;
760 } else {
761 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
762 saddr = &ip6h->saddr;
763 daddr = &ip6h->daddr;
764 }
765
766 if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
767 goto clear_hash_nostart;
768
769 if (crypto_ahash_init(hp.req))
770 goto clear_hash;
771
772 if (tcp_v6_md5_hash_headers(&hp, daddr, saddr, th, skb->len))
773 goto clear_hash;
774 if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2))
775 goto clear_hash;
776 if (tcp_md5_hash_key(&hp, key))
777 goto clear_hash;
778 ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
779 if (crypto_ahash_final(hp.req))
780 goto clear_hash;
781
782 tcp_sigpool_end(&hp);
783 return 0;
784
785clear_hash:
786 tcp_sigpool_end(&hp);
787clear_hash_nostart:
788 memset(md5_hash, 0, 16);
789 return 1;
790}
791#endif
792
793static void tcp_v6_init_req(struct request_sock *req,
794 const struct sock *sk_listener,
795 struct sk_buff *skb)
796{
797 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
798 struct inet_request_sock *ireq = inet_rsk(req);
799 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
800
801 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
802 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
803
804 /* So that link locals have meaning */
805 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
806 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
807 ireq->ir_iif = tcp_v6_iif(skb);
808
809 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
810 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
811 np->rxopt.bits.rxinfo ||
812 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
813 np->rxopt.bits.rxohlim || inet6_test_bit(REPFLOW, sk_listener))) {
814 refcount_inc(&skb->users);
815 ireq->pktopts = skb;
816 }
817}
818
819static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
820 struct sk_buff *skb,
821 struct flowi *fl,
822 struct request_sock *req)
823{
824 tcp_v6_init_req(req, sk, skb);
825
826 if (security_inet_conn_request(sk, skb, req))
827 return NULL;
828
829 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
830}
831
832struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
833 .family = AF_INET6,
834 .obj_size = sizeof(struct tcp6_request_sock),
835 .rtx_syn_ack = tcp_rtx_synack,
836 .send_ack = tcp_v6_reqsk_send_ack,
837 .destructor = tcp_v6_reqsk_destructor,
838 .send_reset = tcp_v6_send_reset,
839 .syn_ack_timeout = tcp_syn_ack_timeout,
840};
841
842const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
843 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
844 sizeof(struct ipv6hdr),
845#ifdef CONFIG_TCP_MD5SIG
846 .req_md5_lookup = tcp_v6_md5_lookup,
847 .calc_md5_hash = tcp_v6_md5_hash_skb,
848#endif
849#ifdef CONFIG_TCP_AO
850 .ao_lookup = tcp_v6_ao_lookup_rsk,
851 .ao_calc_key = tcp_v6_ao_calc_key_rsk,
852 .ao_synack_hash = tcp_v6_ao_synack_hash,
853#endif
854#ifdef CONFIG_SYN_COOKIES
855 .cookie_init_seq = cookie_v6_init_sequence,
856#endif
857 .route_req = tcp_v6_route_req,
858 .init_seq = tcp_v6_init_seq,
859 .init_ts_off = tcp_v6_init_ts_off,
860 .send_synack = tcp_v6_send_synack,
861};
862
863static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
864 u32 ack, u32 win, u32 tsval, u32 tsecr,
865 int oif, int rst, u8 tclass, __be32 label,
866 u32 priority, u32 txhash, struct tcp_key *key)
867{
868 const struct tcphdr *th = tcp_hdr(skb);
869 struct tcphdr *t1;
870 struct sk_buff *buff;
871 struct flowi6 fl6;
872 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
873 struct sock *ctl_sk = net->ipv6.tcp_sk;
874 unsigned int tot_len = sizeof(struct tcphdr);
875 __be32 mrst = 0, *topt;
876 struct dst_entry *dst;
877 __u32 mark = 0;
878
879 if (tsecr)
880 tot_len += TCPOLEN_TSTAMP_ALIGNED;
881 if (tcp_key_is_md5(key))
882 tot_len += TCPOLEN_MD5SIG_ALIGNED;
883 if (tcp_key_is_ao(key))
884 tot_len += tcp_ao_len_aligned(key->ao_key);
885
886#ifdef CONFIG_MPTCP
887 if (rst && !tcp_key_is_md5(key)) {
888 mrst = mptcp_reset_option(skb);
889
890 if (mrst)
891 tot_len += sizeof(__be32);
892 }
893#endif
894
895 buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
896 if (!buff)
897 return;
898
899 skb_reserve(buff, MAX_TCP_HEADER);
900
901 t1 = skb_push(buff, tot_len);
902 skb_reset_transport_header(buff);
903
904 /* Swap the send and the receive. */
905 memset(t1, 0, sizeof(*t1));
906 t1->dest = th->source;
907 t1->source = th->dest;
908 t1->doff = tot_len / 4;
909 t1->seq = htonl(seq);
910 t1->ack_seq = htonl(ack);
911 t1->ack = !rst || !th->ack;
912 t1->rst = rst;
913 t1->window = htons(win);
914
915 topt = (__be32 *)(t1 + 1);
916
917 if (tsecr) {
918 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
919 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
920 *topt++ = htonl(tsval);
921 *topt++ = htonl(tsecr);
922 }
923
924 if (mrst)
925 *topt++ = mrst;
926
927#ifdef CONFIG_TCP_MD5SIG
928 if (tcp_key_is_md5(key)) {
929 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
930 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
931 tcp_v6_md5_hash_hdr((__u8 *)topt, key->md5_key,
932 &ipv6_hdr(skb)->saddr,
933 &ipv6_hdr(skb)->daddr, t1);
934 }
935#endif
936#ifdef CONFIG_TCP_AO
937 if (tcp_key_is_ao(key)) {
938 *topt++ = htonl((TCPOPT_AO << 24) |
939 (tcp_ao_len(key->ao_key) << 16) |
940 (key->ao_key->sndid << 8) |
941 (key->rcv_next));
942
943 tcp_ao_hash_hdr(AF_INET6, (char *)topt, key->ao_key,
944 key->traffic_key,
945 (union tcp_ao_addr *)&ipv6_hdr(skb)->saddr,
946 (union tcp_ao_addr *)&ipv6_hdr(skb)->daddr,
947 t1, key->sne);
948 }
949#endif
950
951 memset(&fl6, 0, sizeof(fl6));
952 fl6.daddr = ipv6_hdr(skb)->saddr;
953 fl6.saddr = ipv6_hdr(skb)->daddr;
954 fl6.flowlabel = label;
955
956 buff->ip_summed = CHECKSUM_PARTIAL;
957
958 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
959
960 fl6.flowi6_proto = IPPROTO_TCP;
961 if (rt6_need_strict(&fl6.daddr) && !oif)
962 fl6.flowi6_oif = tcp_v6_iif(skb);
963 else {
964 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
965 oif = skb->skb_iif;
966
967 fl6.flowi6_oif = oif;
968 }
969
970 if (sk) {
971 if (sk->sk_state == TCP_TIME_WAIT)
972 mark = inet_twsk(sk)->tw_mark;
973 else
974 mark = READ_ONCE(sk->sk_mark);
975 skb_set_delivery_time(buff, tcp_transmit_time(sk), true);
976 }
977 if (txhash) {
978 /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */
979 skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4);
980 }
981 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
982 fl6.fl6_dport = t1->dest;
983 fl6.fl6_sport = t1->source;
984 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
985 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
986
987 /* Pass a socket to ip6_dst_lookup either it is for RST
988 * Underlying function will use this to retrieve the network
989 * namespace
990 */
991 if (sk && sk->sk_state != TCP_TIME_WAIT)
992 dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/
993 else
994 dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL);
995 if (!IS_ERR(dst)) {
996 skb_dst_set(buff, dst);
997 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
998 tclass & ~INET_ECN_MASK, priority);
999 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
1000 if (rst)
1001 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
1002 return;
1003 }
1004
1005 kfree_skb(buff);
1006}
1007
1008static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
1009{
1010 const struct tcphdr *th = tcp_hdr(skb);
1011 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
1012 const __u8 *md5_hash_location = NULL;
1013#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1014 bool allocated_traffic_key = false;
1015#endif
1016 const struct tcp_ao_hdr *aoh;
1017 struct tcp_key key = {};
1018 u32 seq = 0, ack_seq = 0;
1019 __be32 label = 0;
1020 u32 priority = 0;
1021 struct net *net;
1022 u32 txhash = 0;
1023 int oif = 0;
1024#ifdef CONFIG_TCP_MD5SIG
1025 unsigned char newhash[16];
1026 int genhash;
1027 struct sock *sk1 = NULL;
1028#endif
1029
1030 if (th->rst)
1031 return;
1032
1033 /* If sk not NULL, it means we did a successful lookup and incoming
1034 * route had to be correct. prequeue might have dropped our dst.
1035 */
1036 if (!sk && !ipv6_unicast_destination(skb))
1037 return;
1038
1039 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
1040 /* Invalid TCP option size or twice included auth */
1041 if (tcp_parse_auth_options(th, &md5_hash_location, &aoh))
1042 return;
1043#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1044 rcu_read_lock();
1045#endif
1046#ifdef CONFIG_TCP_MD5SIG
1047 if (sk && sk_fullsock(sk)) {
1048 int l3index;
1049
1050 /* sdif set, means packet ingressed via a device
1051 * in an L3 domain and inet_iif is set to it.
1052 */
1053 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1054 key.md5_key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1055 if (key.md5_key)
1056 key.type = TCP_KEY_MD5;
1057 } else if (md5_hash_location) {
1058 int dif = tcp_v6_iif_l3_slave(skb);
1059 int sdif = tcp_v6_sdif(skb);
1060 int l3index;
1061
1062 /*
1063 * active side is lost. Try to find listening socket through
1064 * source port, and then find md5 key through listening socket.
1065 * we are not loose security here:
1066 * Incoming packet is checked with md5 hash with finding key,
1067 * no RST generated if md5 hash doesn't match.
1068 */
1069 sk1 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1070 NULL, 0, &ipv6h->saddr, th->source,
1071 &ipv6h->daddr, ntohs(th->source),
1072 dif, sdif);
1073 if (!sk1)
1074 goto out;
1075
1076 /* sdif set, means packet ingressed via a device
1077 * in an L3 domain and dif is set to it.
1078 */
1079 l3index = tcp_v6_sdif(skb) ? dif : 0;
1080
1081 key.md5_key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1082 if (!key.md5_key)
1083 goto out;
1084 key.type = TCP_KEY_MD5;
1085
1086 genhash = tcp_v6_md5_hash_skb(newhash, key.md5_key, NULL, skb);
1087 if (genhash || memcmp(md5_hash_location, newhash, 16) != 0)
1088 goto out;
1089 }
1090#endif
1091
1092 if (th->ack)
1093 seq = ntohl(th->ack_seq);
1094 else
1095 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1096 (th->doff << 2);
1097
1098#ifdef CONFIG_TCP_AO
1099 if (aoh) {
1100 int l3index;
1101
1102 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1103 if (tcp_ao_prepare_reset(sk, skb, aoh, l3index, seq,
1104 &key.ao_key, &key.traffic_key,
1105 &allocated_traffic_key,
1106 &key.rcv_next, &key.sne))
1107 goto out;
1108 key.type = TCP_KEY_AO;
1109 }
1110#endif
1111
1112 if (sk) {
1113 oif = sk->sk_bound_dev_if;
1114 if (sk_fullsock(sk)) {
1115 trace_tcp_send_reset(sk, skb);
1116 if (inet6_test_bit(REPFLOW, sk))
1117 label = ip6_flowlabel(ipv6h);
1118 priority = READ_ONCE(sk->sk_priority);
1119 txhash = sk->sk_txhash;
1120 }
1121 if (sk->sk_state == TCP_TIME_WAIT) {
1122 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1123 priority = inet_twsk(sk)->tw_priority;
1124 txhash = inet_twsk(sk)->tw_txhash;
1125 }
1126 } else {
1127 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1128 label = ip6_flowlabel(ipv6h);
1129 }
1130
1131 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, 1,
1132 ipv6_get_dsfield(ipv6h), label, priority, txhash,
1133 &key);
1134
1135#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1136out:
1137 if (allocated_traffic_key)
1138 kfree(key.traffic_key);
1139 rcu_read_unlock();
1140#endif
1141}
1142
1143static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1144 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1145 struct tcp_key *key, u8 tclass,
1146 __be32 label, u32 priority, u32 txhash)
1147{
1148 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, 0,
1149 tclass, label, priority, txhash, key);
1150}
1151
1152static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1153{
1154 struct inet_timewait_sock *tw = inet_twsk(sk);
1155 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1156 struct tcp_key key = {};
1157#ifdef CONFIG_TCP_AO
1158 struct tcp_ao_info *ao_info;
1159
1160 if (static_branch_unlikely(&tcp_ao_needed.key)) {
1161
1162 /* FIXME: the segment to-be-acked is not verified yet */
1163 ao_info = rcu_dereference(tcptw->ao_info);
1164 if (ao_info) {
1165 const struct tcp_ao_hdr *aoh;
1166
1167 /* Invalid TCP option size or twice included auth */
1168 if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1169 goto out;
1170 if (aoh)
1171 key.ao_key = tcp_ao_established_key(ao_info,
1172 aoh->rnext_keyid, -1);
1173 }
1174 }
1175 if (key.ao_key) {
1176 struct tcp_ao_key *rnext_key;
1177
1178 key.traffic_key = snd_other_key(key.ao_key);
1179 /* rcv_next switches to our rcv_next */
1180 rnext_key = READ_ONCE(ao_info->rnext_key);
1181 key.rcv_next = rnext_key->rcvid;
1182 key.sne = READ_ONCE(ao_info->snd_sne);
1183 key.type = TCP_KEY_AO;
1184#else
1185 if (0) {
1186#endif
1187#ifdef CONFIG_TCP_MD5SIG
1188 } else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1189 key.md5_key = tcp_twsk_md5_key(tcptw);
1190 if (key.md5_key)
1191 key.type = TCP_KEY_MD5;
1192#endif
1193 }
1194
1195 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1196 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1197 tcp_tw_tsval(tcptw),
1198 tcptw->tw_ts_recent, tw->tw_bound_dev_if, &key,
1199 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority,
1200 tw->tw_txhash);
1201
1202#ifdef CONFIG_TCP_AO
1203out:
1204#endif
1205 inet_twsk_put(tw);
1206}
1207
1208static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1209 struct request_sock *req)
1210{
1211 struct tcp_key key = {};
1212
1213#ifdef CONFIG_TCP_AO
1214 if (static_branch_unlikely(&tcp_ao_needed.key) &&
1215 tcp_rsk_used_ao(req)) {
1216 const struct in6_addr *addr = &ipv6_hdr(skb)->saddr;
1217 const struct tcp_ao_hdr *aoh;
1218 int l3index;
1219
1220 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1221 /* Invalid TCP option size or twice included auth */
1222 if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1223 return;
1224 if (!aoh)
1225 return;
1226 key.ao_key = tcp_ao_do_lookup(sk, l3index,
1227 (union tcp_ao_addr *)addr,
1228 AF_INET6, aoh->rnext_keyid, -1);
1229 if (unlikely(!key.ao_key)) {
1230 /* Send ACK with any matching MKT for the peer */
1231 key.ao_key = tcp_ao_do_lookup(sk, l3index,
1232 (union tcp_ao_addr *)addr,
1233 AF_INET6, -1, -1);
1234 /* Matching key disappeared (user removed the key?)
1235 * let the handshake timeout.
1236 */
1237 if (!key.ao_key) {
1238 net_info_ratelimited("TCP-AO key for (%pI6, %d)->(%pI6, %d) suddenly disappeared, won't ACK new connection\n",
1239 addr,
1240 ntohs(tcp_hdr(skb)->source),
1241 &ipv6_hdr(skb)->daddr,
1242 ntohs(tcp_hdr(skb)->dest));
1243 return;
1244 }
1245 }
1246 key.traffic_key = kmalloc(tcp_ao_digest_size(key.ao_key), GFP_ATOMIC);
1247 if (!key.traffic_key)
1248 return;
1249
1250 key.type = TCP_KEY_AO;
1251 key.rcv_next = aoh->keyid;
1252 tcp_v6_ao_calc_key_rsk(key.ao_key, key.traffic_key, req);
1253#else
1254 if (0) {
1255#endif
1256#ifdef CONFIG_TCP_MD5SIG
1257 } else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1258 int l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1259
1260 key.md5_key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr,
1261 l3index);
1262 if (key.md5_key)
1263 key.type = TCP_KEY_MD5;
1264#endif
1265 }
1266
1267 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1268 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1269 */
1270 /* RFC 7323 2.3
1271 * The window field (SEG.WND) of every outgoing segment, with the
1272 * exception of <SYN> segments, MUST be right-shifted by
1273 * Rcv.Wind.Shift bits:
1274 */
1275 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1276 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1277 tcp_rsk(req)->rcv_nxt,
1278 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1279 tcp_rsk_tsval(tcp_rsk(req)),
1280 READ_ONCE(req->ts_recent), sk->sk_bound_dev_if,
1281 &key, ipv6_get_dsfield(ipv6_hdr(skb)), 0,
1282 READ_ONCE(sk->sk_priority),
1283 READ_ONCE(tcp_rsk(req)->txhash));
1284 if (tcp_key_is_ao(&key))
1285 kfree(key.traffic_key);
1286}
1287
1288
1289static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1290{
1291#ifdef CONFIG_SYN_COOKIES
1292 const struct tcphdr *th = tcp_hdr(skb);
1293
1294 if (!th->syn)
1295 sk = cookie_v6_check(sk, skb);
1296#endif
1297 return sk;
1298}
1299
1300u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1301 struct tcphdr *th, u32 *cookie)
1302{
1303 u16 mss = 0;
1304#ifdef CONFIG_SYN_COOKIES
1305 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1306 &tcp_request_sock_ipv6_ops, sk, th);
1307 if (mss) {
1308 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1309 tcp_synq_overflow(sk);
1310 }
1311#endif
1312 return mss;
1313}
1314
1315static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1316{
1317 if (skb->protocol == htons(ETH_P_IP))
1318 return tcp_v4_conn_request(sk, skb);
1319
1320 if (!ipv6_unicast_destination(skb))
1321 goto drop;
1322
1323 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1324 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1325 return 0;
1326 }
1327
1328 return tcp_conn_request(&tcp6_request_sock_ops,
1329 &tcp_request_sock_ipv6_ops, sk, skb);
1330
1331drop:
1332 tcp_listendrop(sk);
1333 return 0; /* don't send reset */
1334}
1335
1336static void tcp_v6_restore_cb(struct sk_buff *skb)
1337{
1338 /* We need to move header back to the beginning if xfrm6_policy_check()
1339 * and tcp_v6_fill_cb() are going to be called again.
1340 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1341 */
1342 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1343 sizeof(struct inet6_skb_parm));
1344}
1345
1346static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1347 struct request_sock *req,
1348 struct dst_entry *dst,
1349 struct request_sock *req_unhash,
1350 bool *own_req)
1351{
1352 struct inet_request_sock *ireq;
1353 struct ipv6_pinfo *newnp;
1354 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1355 struct ipv6_txoptions *opt;
1356 struct inet_sock *newinet;
1357 bool found_dup_sk = false;
1358 struct tcp_sock *newtp;
1359 struct sock *newsk;
1360#ifdef CONFIG_TCP_MD5SIG
1361 struct tcp_md5sig_key *key;
1362 int l3index;
1363#endif
1364 struct flowi6 fl6;
1365
1366 if (skb->protocol == htons(ETH_P_IP)) {
1367 /*
1368 * v6 mapped
1369 */
1370
1371 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1372 req_unhash, own_req);
1373
1374 if (!newsk)
1375 return NULL;
1376
1377 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1378
1379 newnp = tcp_inet6_sk(newsk);
1380 newtp = tcp_sk(newsk);
1381
1382 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1383
1384 newnp->saddr = newsk->sk_v6_rcv_saddr;
1385
1386 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1387 if (sk_is_mptcp(newsk))
1388 mptcpv6_handle_mapped(newsk, true);
1389 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1390#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1391 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1392#endif
1393
1394 newnp->ipv6_mc_list = NULL;
1395 newnp->ipv6_ac_list = NULL;
1396 newnp->ipv6_fl_list = NULL;
1397 newnp->pktoptions = NULL;
1398 newnp->opt = NULL;
1399 newnp->mcast_oif = inet_iif(skb);
1400 newnp->mcast_hops = ip_hdr(skb)->ttl;
1401 newnp->rcv_flowinfo = 0;
1402 if (inet6_test_bit(REPFLOW, sk))
1403 newnp->flow_label = 0;
1404
1405 /*
1406 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1407 * here, tcp_create_openreq_child now does this for us, see the comment in
1408 * that function for the gory details. -acme
1409 */
1410
1411 /* It is tricky place. Until this moment IPv4 tcp
1412 worked with IPv6 icsk.icsk_af_ops.
1413 Sync it now.
1414 */
1415 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1416
1417 return newsk;
1418 }
1419
1420 ireq = inet_rsk(req);
1421
1422 if (sk_acceptq_is_full(sk))
1423 goto out_overflow;
1424
1425 if (!dst) {
1426 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1427 if (!dst)
1428 goto out;
1429 }
1430
1431 newsk = tcp_create_openreq_child(sk, req, skb);
1432 if (!newsk)
1433 goto out_nonewsk;
1434
1435 /*
1436 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1437 * count here, tcp_create_openreq_child now does this for us, see the
1438 * comment in that function for the gory details. -acme
1439 */
1440
1441 newsk->sk_gso_type = SKB_GSO_TCPV6;
1442 ip6_dst_store(newsk, dst, NULL, NULL);
1443 inet6_sk_rx_dst_set(newsk, skb);
1444
1445 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1446
1447 newtp = tcp_sk(newsk);
1448 newinet = inet_sk(newsk);
1449 newnp = tcp_inet6_sk(newsk);
1450
1451 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1452
1453 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1454 newnp->saddr = ireq->ir_v6_loc_addr;
1455 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1456 newsk->sk_bound_dev_if = ireq->ir_iif;
1457
1458 /* Now IPv6 options...
1459
1460 First: no IPv4 options.
1461 */
1462 newinet->inet_opt = NULL;
1463 newnp->ipv6_mc_list = NULL;
1464 newnp->ipv6_ac_list = NULL;
1465 newnp->ipv6_fl_list = NULL;
1466
1467 /* Clone RX bits */
1468 newnp->rxopt.all = np->rxopt.all;
1469
1470 newnp->pktoptions = NULL;
1471 newnp->opt = NULL;
1472 newnp->mcast_oif = tcp_v6_iif(skb);
1473 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1474 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1475 if (inet6_test_bit(REPFLOW, sk))
1476 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1477
1478 /* Set ToS of the new socket based upon the value of incoming SYN.
1479 * ECT bits are set later in tcp_init_transfer().
1480 */
1481 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1482 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1483
1484 /* Clone native IPv6 options from listening socket (if any)
1485
1486 Yes, keeping reference count would be much more clever,
1487 but we make one more one thing there: reattach optmem
1488 to newsk.
1489 */
1490 opt = ireq->ipv6_opt;
1491 if (!opt)
1492 opt = rcu_dereference(np->opt);
1493 if (opt) {
1494 opt = ipv6_dup_options(newsk, opt);
1495 RCU_INIT_POINTER(newnp->opt, opt);
1496 }
1497 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1498 if (opt)
1499 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1500 opt->opt_flen;
1501
1502 tcp_ca_openreq_child(newsk, dst);
1503
1504 tcp_sync_mss(newsk, dst_mtu(dst));
1505 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1506
1507 tcp_initialize_rcv_mss(newsk);
1508
1509 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1510 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1511
1512#ifdef CONFIG_TCP_MD5SIG
1513 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1514
1515 if (!tcp_rsk_used_ao(req)) {
1516 /* Copy over the MD5 key from the original socket */
1517 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1518 if (key) {
1519 const union tcp_md5_addr *addr;
1520
1521 addr = (union tcp_md5_addr *)&newsk->sk_v6_daddr;
1522 if (tcp_md5_key_copy(newsk, addr, AF_INET6, 128, l3index, key)) {
1523 inet_csk_prepare_forced_close(newsk);
1524 tcp_done(newsk);
1525 goto out;
1526 }
1527 }
1528 }
1529#endif
1530#ifdef CONFIG_TCP_AO
1531 /* Copy over tcp_ao_info if any */
1532 if (tcp_ao_copy_all_matching(sk, newsk, req, skb, AF_INET6))
1533 goto out; /* OOM */
1534#endif
1535
1536 if (__inet_inherit_port(sk, newsk) < 0) {
1537 inet_csk_prepare_forced_close(newsk);
1538 tcp_done(newsk);
1539 goto out;
1540 }
1541 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1542 &found_dup_sk);
1543 if (*own_req) {
1544 tcp_move_syn(newtp, req);
1545
1546 /* Clone pktoptions received with SYN, if we own the req */
1547 if (ireq->pktopts) {
1548 newnp->pktoptions = skb_clone_and_charge_r(ireq->pktopts, newsk);
1549 consume_skb(ireq->pktopts);
1550 ireq->pktopts = NULL;
1551 if (newnp->pktoptions)
1552 tcp_v6_restore_cb(newnp->pktoptions);
1553 }
1554 } else {
1555 if (!req_unhash && found_dup_sk) {
1556 /* This code path should only be executed in the
1557 * syncookie case only
1558 */
1559 bh_unlock_sock(newsk);
1560 sock_put(newsk);
1561 newsk = NULL;
1562 }
1563 }
1564
1565 return newsk;
1566
1567out_overflow:
1568 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1569out_nonewsk:
1570 dst_release(dst);
1571out:
1572 tcp_listendrop(sk);
1573 return NULL;
1574}
1575
1576INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1577 u32));
1578/* The socket must have it's spinlock held when we get
1579 * here, unless it is a TCP_LISTEN socket.
1580 *
1581 * We have a potential double-lock case here, so even when
1582 * doing backlog processing we use the BH locking scheme.
1583 * This is because we cannot sleep with the original spinlock
1584 * held.
1585 */
1586INDIRECT_CALLABLE_SCOPE
1587int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1588{
1589 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1590 struct sk_buff *opt_skb = NULL;
1591 enum skb_drop_reason reason;
1592 struct tcp_sock *tp;
1593
1594 /* Imagine: socket is IPv6. IPv4 packet arrives,
1595 goes to IPv4 receive handler and backlogged.
1596 From backlog it always goes here. Kerboom...
1597 Fortunately, tcp_rcv_established and rcv_established
1598 handle them correctly, but it is not case with
1599 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1600 */
1601
1602 if (skb->protocol == htons(ETH_P_IP))
1603 return tcp_v4_do_rcv(sk, skb);
1604
1605 /*
1606 * socket locking is here for SMP purposes as backlog rcv
1607 * is currently called with bh processing disabled.
1608 */
1609
1610 /* Do Stevens' IPV6_PKTOPTIONS.
1611
1612 Yes, guys, it is the only place in our code, where we
1613 may make it not affecting IPv4.
1614 The rest of code is protocol independent,
1615 and I do not like idea to uglify IPv4.
1616
1617 Actually, all the idea behind IPV6_PKTOPTIONS
1618 looks not very well thought. For now we latch
1619 options, received in the last packet, enqueued
1620 by tcp. Feel free to propose better solution.
1621 --ANK (980728)
1622 */
1623 if (np->rxopt.all)
1624 opt_skb = skb_clone_and_charge_r(skb, sk);
1625
1626 reason = SKB_DROP_REASON_NOT_SPECIFIED;
1627 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1628 struct dst_entry *dst;
1629
1630 dst = rcu_dereference_protected(sk->sk_rx_dst,
1631 lockdep_sock_is_held(sk));
1632
1633 sock_rps_save_rxhash(sk, skb);
1634 sk_mark_napi_id(sk, skb);
1635 if (dst) {
1636 if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1637 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1638 dst, sk->sk_rx_dst_cookie) == NULL) {
1639 RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1640 dst_release(dst);
1641 }
1642 }
1643
1644 tcp_rcv_established(sk, skb);
1645 if (opt_skb)
1646 goto ipv6_pktoptions;
1647 return 0;
1648 }
1649
1650 if (tcp_checksum_complete(skb))
1651 goto csum_err;
1652
1653 if (sk->sk_state == TCP_LISTEN) {
1654 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1655
1656 if (!nsk)
1657 goto discard;
1658
1659 if (nsk != sk) {
1660 if (tcp_child_process(sk, nsk, skb))
1661 goto reset;
1662 if (opt_skb)
1663 __kfree_skb(opt_skb);
1664 return 0;
1665 }
1666 } else
1667 sock_rps_save_rxhash(sk, skb);
1668
1669 if (tcp_rcv_state_process(sk, skb))
1670 goto reset;
1671 if (opt_skb)
1672 goto ipv6_pktoptions;
1673 return 0;
1674
1675reset:
1676 tcp_v6_send_reset(sk, skb);
1677discard:
1678 if (opt_skb)
1679 __kfree_skb(opt_skb);
1680 kfree_skb_reason(skb, reason);
1681 return 0;
1682csum_err:
1683 reason = SKB_DROP_REASON_TCP_CSUM;
1684 trace_tcp_bad_csum(skb);
1685 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1686 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1687 goto discard;
1688
1689
1690ipv6_pktoptions:
1691 /* Do you ask, what is it?
1692
1693 1. skb was enqueued by tcp.
1694 2. skb is added to tail of read queue, rather than out of order.
1695 3. socket is not in passive state.
1696 4. Finally, it really contains options, which user wants to receive.
1697 */
1698 tp = tcp_sk(sk);
1699 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1700 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1701 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1702 WRITE_ONCE(np->mcast_oif, tcp_v6_iif(opt_skb));
1703 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1704 WRITE_ONCE(np->mcast_hops,
1705 ipv6_hdr(opt_skb)->hop_limit);
1706 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1707 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1708 if (inet6_test_bit(REPFLOW, sk))
1709 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1710 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1711 tcp_v6_restore_cb(opt_skb);
1712 opt_skb = xchg(&np->pktoptions, opt_skb);
1713 } else {
1714 __kfree_skb(opt_skb);
1715 opt_skb = xchg(&np->pktoptions, NULL);
1716 }
1717 }
1718
1719 consume_skb(opt_skb);
1720 return 0;
1721}
1722
1723static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1724 const struct tcphdr *th)
1725{
1726 /* This is tricky: we move IP6CB at its correct location into
1727 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1728 * _decode_session6() uses IP6CB().
1729 * barrier() makes sure compiler won't play aliasing games.
1730 */
1731 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1732 sizeof(struct inet6_skb_parm));
1733 barrier();
1734
1735 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1736 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1737 skb->len - th->doff*4);
1738 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1739 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1740 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1741 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1742 TCP_SKB_CB(skb)->sacked = 0;
1743 TCP_SKB_CB(skb)->has_rxtstamp =
1744 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1745}
1746
1747INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1748{
1749 enum skb_drop_reason drop_reason;
1750 int sdif = inet6_sdif(skb);
1751 int dif = inet6_iif(skb);
1752 const struct tcphdr *th;
1753 const struct ipv6hdr *hdr;
1754 bool refcounted;
1755 struct sock *sk;
1756 int ret;
1757 struct net *net = dev_net(skb->dev);
1758
1759 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1760 if (skb->pkt_type != PACKET_HOST)
1761 goto discard_it;
1762
1763 /*
1764 * Count it even if it's bad.
1765 */
1766 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1767
1768 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1769 goto discard_it;
1770
1771 th = (const struct tcphdr *)skb->data;
1772
1773 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
1774 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1775 goto bad_packet;
1776 }
1777 if (!pskb_may_pull(skb, th->doff*4))
1778 goto discard_it;
1779
1780 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1781 goto csum_error;
1782
1783 th = (const struct tcphdr *)skb->data;
1784 hdr = ipv6_hdr(skb);
1785
1786lookup:
1787 sk = __inet6_lookup_skb(net->ipv4.tcp_death_row.hashinfo, skb, __tcp_hdrlen(th),
1788 th->source, th->dest, inet6_iif(skb), sdif,
1789 &refcounted);
1790 if (!sk)
1791 goto no_tcp_socket;
1792
1793process:
1794 if (sk->sk_state == TCP_TIME_WAIT)
1795 goto do_time_wait;
1796
1797 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1798 struct request_sock *req = inet_reqsk(sk);
1799 bool req_stolen = false;
1800 struct sock *nsk;
1801
1802 sk = req->rsk_listener;
1803 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1804 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1805 else
1806 drop_reason = tcp_inbound_hash(sk, req, skb,
1807 &hdr->saddr, &hdr->daddr,
1808 AF_INET6, dif, sdif);
1809 if (drop_reason) {
1810 sk_drops_add(sk, skb);
1811 reqsk_put(req);
1812 goto discard_it;
1813 }
1814 if (tcp_checksum_complete(skb)) {
1815 reqsk_put(req);
1816 goto csum_error;
1817 }
1818 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1819 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
1820 if (!nsk) {
1821 inet_csk_reqsk_queue_drop_and_put(sk, req);
1822 goto lookup;
1823 }
1824 sk = nsk;
1825 /* reuseport_migrate_sock() has already held one sk_refcnt
1826 * before returning.
1827 */
1828 } else {
1829 sock_hold(sk);
1830 }
1831 refcounted = true;
1832 nsk = NULL;
1833 if (!tcp_filter(sk, skb)) {
1834 th = (const struct tcphdr *)skb->data;
1835 hdr = ipv6_hdr(skb);
1836 tcp_v6_fill_cb(skb, hdr, th);
1837 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1838 } else {
1839 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1840 }
1841 if (!nsk) {
1842 reqsk_put(req);
1843 if (req_stolen) {
1844 /* Another cpu got exclusive access to req
1845 * and created a full blown socket.
1846 * Try to feed this packet to this socket
1847 * instead of discarding it.
1848 */
1849 tcp_v6_restore_cb(skb);
1850 sock_put(sk);
1851 goto lookup;
1852 }
1853 goto discard_and_relse;
1854 }
1855 nf_reset_ct(skb);
1856 if (nsk == sk) {
1857 reqsk_put(req);
1858 tcp_v6_restore_cb(skb);
1859 } else if (tcp_child_process(sk, nsk, skb)) {
1860 tcp_v6_send_reset(nsk, skb);
1861 goto discard_and_relse;
1862 } else {
1863 sock_put(sk);
1864 return 0;
1865 }
1866 }
1867
1868 if (static_branch_unlikely(&ip6_min_hopcount)) {
1869 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
1870 if (unlikely(hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount))) {
1871 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1872 drop_reason = SKB_DROP_REASON_TCP_MINTTL;
1873 goto discard_and_relse;
1874 }
1875 }
1876
1877 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
1878 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1879 goto discard_and_relse;
1880 }
1881
1882 drop_reason = tcp_inbound_hash(sk, NULL, skb, &hdr->saddr, &hdr->daddr,
1883 AF_INET6, dif, sdif);
1884 if (drop_reason)
1885 goto discard_and_relse;
1886
1887 nf_reset_ct(skb);
1888
1889 if (tcp_filter(sk, skb)) {
1890 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1891 goto discard_and_relse;
1892 }
1893 th = (const struct tcphdr *)skb->data;
1894 hdr = ipv6_hdr(skb);
1895 tcp_v6_fill_cb(skb, hdr, th);
1896
1897 skb->dev = NULL;
1898
1899 if (sk->sk_state == TCP_LISTEN) {
1900 ret = tcp_v6_do_rcv(sk, skb);
1901 goto put_and_return;
1902 }
1903
1904 sk_incoming_cpu_update(sk);
1905
1906 bh_lock_sock_nested(sk);
1907 tcp_segs_in(tcp_sk(sk), skb);
1908 ret = 0;
1909 if (!sock_owned_by_user(sk)) {
1910 ret = tcp_v6_do_rcv(sk, skb);
1911 } else {
1912 if (tcp_add_backlog(sk, skb, &drop_reason))
1913 goto discard_and_relse;
1914 }
1915 bh_unlock_sock(sk);
1916put_and_return:
1917 if (refcounted)
1918 sock_put(sk);
1919 return ret ? -1 : 0;
1920
1921no_tcp_socket:
1922 drop_reason = SKB_DROP_REASON_NO_SOCKET;
1923 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1924 goto discard_it;
1925
1926 tcp_v6_fill_cb(skb, hdr, th);
1927
1928 if (tcp_checksum_complete(skb)) {
1929csum_error:
1930 drop_reason = SKB_DROP_REASON_TCP_CSUM;
1931 trace_tcp_bad_csum(skb);
1932 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1933bad_packet:
1934 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1935 } else {
1936 tcp_v6_send_reset(NULL, skb);
1937 }
1938
1939discard_it:
1940 SKB_DR_OR(drop_reason, NOT_SPECIFIED);
1941 kfree_skb_reason(skb, drop_reason);
1942 return 0;
1943
1944discard_and_relse:
1945 sk_drops_add(sk, skb);
1946 if (refcounted)
1947 sock_put(sk);
1948 goto discard_it;
1949
1950do_time_wait:
1951 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1952 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1953 inet_twsk_put(inet_twsk(sk));
1954 goto discard_it;
1955 }
1956
1957 tcp_v6_fill_cb(skb, hdr, th);
1958
1959 if (tcp_checksum_complete(skb)) {
1960 inet_twsk_put(inet_twsk(sk));
1961 goto csum_error;
1962 }
1963
1964 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1965 case TCP_TW_SYN:
1966 {
1967 struct sock *sk2;
1968
1969 sk2 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1970 skb, __tcp_hdrlen(th),
1971 &ipv6_hdr(skb)->saddr, th->source,
1972 &ipv6_hdr(skb)->daddr,
1973 ntohs(th->dest),
1974 tcp_v6_iif_l3_slave(skb),
1975 sdif);
1976 if (sk2) {
1977 struct inet_timewait_sock *tw = inet_twsk(sk);
1978 inet_twsk_deschedule_put(tw);
1979 sk = sk2;
1980 tcp_v6_restore_cb(skb);
1981 refcounted = false;
1982 goto process;
1983 }
1984 }
1985 /* to ACK */
1986 fallthrough;
1987 case TCP_TW_ACK:
1988 tcp_v6_timewait_ack(sk, skb);
1989 break;
1990 case TCP_TW_RST:
1991 tcp_v6_send_reset(sk, skb);
1992 inet_twsk_deschedule_put(inet_twsk(sk));
1993 goto discard_it;
1994 case TCP_TW_SUCCESS:
1995 ;
1996 }
1997 goto discard_it;
1998}
1999
2000void tcp_v6_early_demux(struct sk_buff *skb)
2001{
2002 struct net *net = dev_net(skb->dev);
2003 const struct ipv6hdr *hdr;
2004 const struct tcphdr *th;
2005 struct sock *sk;
2006
2007 if (skb->pkt_type != PACKET_HOST)
2008 return;
2009
2010 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
2011 return;
2012
2013 hdr = ipv6_hdr(skb);
2014 th = tcp_hdr(skb);
2015
2016 if (th->doff < sizeof(struct tcphdr) / 4)
2017 return;
2018
2019 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
2020 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
2021 &hdr->saddr, th->source,
2022 &hdr->daddr, ntohs(th->dest),
2023 inet6_iif(skb), inet6_sdif(skb));
2024 if (sk) {
2025 skb->sk = sk;
2026 skb->destructor = sock_edemux;
2027 if (sk_fullsock(sk)) {
2028 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
2029
2030 if (dst)
2031 dst = dst_check(dst, sk->sk_rx_dst_cookie);
2032 if (dst &&
2033 sk->sk_rx_dst_ifindex == skb->skb_iif)
2034 skb_dst_set_noref(skb, dst);
2035 }
2036 }
2037}
2038
2039static struct timewait_sock_ops tcp6_timewait_sock_ops = {
2040 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
2041 .twsk_unique = tcp_twsk_unique,
2042 .twsk_destructor = tcp_twsk_destructor,
2043};
2044
2045INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
2046{
2047 __tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr);
2048}
2049
2050const struct inet_connection_sock_af_ops ipv6_specific = {
2051 .queue_xmit = inet6_csk_xmit,
2052 .send_check = tcp_v6_send_check,
2053 .rebuild_header = inet6_sk_rebuild_header,
2054 .sk_rx_dst_set = inet6_sk_rx_dst_set,
2055 .conn_request = tcp_v6_conn_request,
2056 .syn_recv_sock = tcp_v6_syn_recv_sock,
2057 .net_header_len = sizeof(struct ipv6hdr),
2058 .setsockopt = ipv6_setsockopt,
2059 .getsockopt = ipv6_getsockopt,
2060 .addr2sockaddr = inet6_csk_addr2sockaddr,
2061 .sockaddr_len = sizeof(struct sockaddr_in6),
2062 .mtu_reduced = tcp_v6_mtu_reduced,
2063};
2064
2065#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2066static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
2067#ifdef CONFIG_TCP_MD5SIG
2068 .md5_lookup = tcp_v6_md5_lookup,
2069 .calc_md5_hash = tcp_v6_md5_hash_skb,
2070 .md5_parse = tcp_v6_parse_md5_keys,
2071#endif
2072#ifdef CONFIG_TCP_AO
2073 .ao_lookup = tcp_v6_ao_lookup,
2074 .calc_ao_hash = tcp_v6_ao_hash_skb,
2075 .ao_parse = tcp_v6_parse_ao,
2076 .ao_calc_key_sk = tcp_v6_ao_calc_key_sk,
2077#endif
2078};
2079#endif
2080
2081/*
2082 * TCP over IPv4 via INET6 API
2083 */
2084static const struct inet_connection_sock_af_ops ipv6_mapped = {
2085 .queue_xmit = ip_queue_xmit,
2086 .send_check = tcp_v4_send_check,
2087 .rebuild_header = inet_sk_rebuild_header,
2088 .sk_rx_dst_set = inet_sk_rx_dst_set,
2089 .conn_request = tcp_v6_conn_request,
2090 .syn_recv_sock = tcp_v6_syn_recv_sock,
2091 .net_header_len = sizeof(struct iphdr),
2092 .setsockopt = ipv6_setsockopt,
2093 .getsockopt = ipv6_getsockopt,
2094 .addr2sockaddr = inet6_csk_addr2sockaddr,
2095 .sockaddr_len = sizeof(struct sockaddr_in6),
2096 .mtu_reduced = tcp_v4_mtu_reduced,
2097};
2098
2099#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2100static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
2101#ifdef CONFIG_TCP_MD5SIG
2102 .md5_lookup = tcp_v4_md5_lookup,
2103 .calc_md5_hash = tcp_v4_md5_hash_skb,
2104 .md5_parse = tcp_v6_parse_md5_keys,
2105#endif
2106#ifdef CONFIG_TCP_AO
2107 .ao_lookup = tcp_v6_ao_lookup,
2108 .calc_ao_hash = tcp_v4_ao_hash_skb,
2109 .ao_parse = tcp_v6_parse_ao,
2110 .ao_calc_key_sk = tcp_v4_ao_calc_key_sk,
2111#endif
2112};
2113#endif
2114
2115/* NOTE: A lot of things set to zero explicitly by call to
2116 * sk_alloc() so need not be done here.
2117 */
2118static int tcp_v6_init_sock(struct sock *sk)
2119{
2120 struct inet_connection_sock *icsk = inet_csk(sk);
2121
2122 tcp_init_sock(sk);
2123
2124 icsk->icsk_af_ops = &ipv6_specific;
2125
2126#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2127 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
2128#endif
2129
2130 return 0;
2131}
2132
2133#ifdef CONFIG_PROC_FS
2134/* Proc filesystem TCPv6 sock list dumping. */
2135static void get_openreq6(struct seq_file *seq,
2136 const struct request_sock *req, int i)
2137{
2138 long ttd = req->rsk_timer.expires - jiffies;
2139 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
2140 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
2141
2142 if (ttd < 0)
2143 ttd = 0;
2144
2145 seq_printf(seq,
2146 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2147 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
2148 i,
2149 src->s6_addr32[0], src->s6_addr32[1],
2150 src->s6_addr32[2], src->s6_addr32[3],
2151 inet_rsk(req)->ir_num,
2152 dest->s6_addr32[0], dest->s6_addr32[1],
2153 dest->s6_addr32[2], dest->s6_addr32[3],
2154 ntohs(inet_rsk(req)->ir_rmt_port),
2155 TCP_SYN_RECV,
2156 0, 0, /* could print option size, but that is af dependent. */
2157 1, /* timers active (only the expire timer) */
2158 jiffies_to_clock_t(ttd),
2159 req->num_timeout,
2160 from_kuid_munged(seq_user_ns(seq),
2161 sock_i_uid(req->rsk_listener)),
2162 0, /* non standard timer */
2163 0, /* open_requests have no inode */
2164 0, req);
2165}
2166
2167static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2168{
2169 const struct in6_addr *dest, *src;
2170 __u16 destp, srcp;
2171 int timer_active;
2172 unsigned long timer_expires;
2173 const struct inet_sock *inet = inet_sk(sp);
2174 const struct tcp_sock *tp = tcp_sk(sp);
2175 const struct inet_connection_sock *icsk = inet_csk(sp);
2176 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2177 int rx_queue;
2178 int state;
2179
2180 dest = &sp->sk_v6_daddr;
2181 src = &sp->sk_v6_rcv_saddr;
2182 destp = ntohs(inet->inet_dport);
2183 srcp = ntohs(inet->inet_sport);
2184
2185 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2186 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2187 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2188 timer_active = 1;
2189 timer_expires = icsk->icsk_timeout;
2190 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2191 timer_active = 4;
2192 timer_expires = icsk->icsk_timeout;
2193 } else if (timer_pending(&sp->sk_timer)) {
2194 timer_active = 2;
2195 timer_expires = sp->sk_timer.expires;
2196 } else {
2197 timer_active = 0;
2198 timer_expires = jiffies;
2199 }
2200
2201 state = inet_sk_state_load(sp);
2202 if (state == TCP_LISTEN)
2203 rx_queue = READ_ONCE(sp->sk_ack_backlog);
2204 else
2205 /* Because we don't lock the socket,
2206 * we might find a transient negative value.
2207 */
2208 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2209 READ_ONCE(tp->copied_seq), 0);
2210
2211 seq_printf(seq,
2212 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2213 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2214 i,
2215 src->s6_addr32[0], src->s6_addr32[1],
2216 src->s6_addr32[2], src->s6_addr32[3], srcp,
2217 dest->s6_addr32[0], dest->s6_addr32[1],
2218 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2219 state,
2220 READ_ONCE(tp->write_seq) - tp->snd_una,
2221 rx_queue,
2222 timer_active,
2223 jiffies_delta_to_clock_t(timer_expires - jiffies),
2224 icsk->icsk_retransmits,
2225 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
2226 icsk->icsk_probes_out,
2227 sock_i_ino(sp),
2228 refcount_read(&sp->sk_refcnt), sp,
2229 jiffies_to_clock_t(icsk->icsk_rto),
2230 jiffies_to_clock_t(icsk->icsk_ack.ato),
2231 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2232 tcp_snd_cwnd(tp),
2233 state == TCP_LISTEN ?
2234 fastopenq->max_qlen :
2235 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2236 );
2237}
2238
2239static void get_timewait6_sock(struct seq_file *seq,
2240 struct inet_timewait_sock *tw, int i)
2241{
2242 long delta = tw->tw_timer.expires - jiffies;
2243 const struct in6_addr *dest, *src;
2244 __u16 destp, srcp;
2245
2246 dest = &tw->tw_v6_daddr;
2247 src = &tw->tw_v6_rcv_saddr;
2248 destp = ntohs(tw->tw_dport);
2249 srcp = ntohs(tw->tw_sport);
2250
2251 seq_printf(seq,
2252 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2253 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2254 i,
2255 src->s6_addr32[0], src->s6_addr32[1],
2256 src->s6_addr32[2], src->s6_addr32[3], srcp,
2257 dest->s6_addr32[0], dest->s6_addr32[1],
2258 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2259 tw->tw_substate, 0, 0,
2260 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2261 refcount_read(&tw->tw_refcnt), tw);
2262}
2263
2264static int tcp6_seq_show(struct seq_file *seq, void *v)
2265{
2266 struct tcp_iter_state *st;
2267 struct sock *sk = v;
2268
2269 if (v == SEQ_START_TOKEN) {
2270 seq_puts(seq,
2271 " sl "
2272 "local_address "
2273 "remote_address "
2274 "st tx_queue rx_queue tr tm->when retrnsmt"
2275 " uid timeout inode\n");
2276 goto out;
2277 }
2278 st = seq->private;
2279
2280 if (sk->sk_state == TCP_TIME_WAIT)
2281 get_timewait6_sock(seq, v, st->num);
2282 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2283 get_openreq6(seq, v, st->num);
2284 else
2285 get_tcp6_sock(seq, v, st->num);
2286out:
2287 return 0;
2288}
2289
2290static const struct seq_operations tcp6_seq_ops = {
2291 .show = tcp6_seq_show,
2292 .start = tcp_seq_start,
2293 .next = tcp_seq_next,
2294 .stop = tcp_seq_stop,
2295};
2296
2297static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2298 .family = AF_INET6,
2299};
2300
2301int __net_init tcp6_proc_init(struct net *net)
2302{
2303 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2304 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2305 return -ENOMEM;
2306 return 0;
2307}
2308
2309void tcp6_proc_exit(struct net *net)
2310{
2311 remove_proc_entry("tcp6", net->proc_net);
2312}
2313#endif
2314
2315struct proto tcpv6_prot = {
2316 .name = "TCPv6",
2317 .owner = THIS_MODULE,
2318 .close = tcp_close,
2319 .pre_connect = tcp_v6_pre_connect,
2320 .connect = tcp_v6_connect,
2321 .disconnect = tcp_disconnect,
2322 .accept = inet_csk_accept,
2323 .ioctl = tcp_ioctl,
2324 .init = tcp_v6_init_sock,
2325 .destroy = tcp_v4_destroy_sock,
2326 .shutdown = tcp_shutdown,
2327 .setsockopt = tcp_setsockopt,
2328 .getsockopt = tcp_getsockopt,
2329 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
2330 .keepalive = tcp_set_keepalive,
2331 .recvmsg = tcp_recvmsg,
2332 .sendmsg = tcp_sendmsg,
2333 .splice_eof = tcp_splice_eof,
2334 .backlog_rcv = tcp_v6_do_rcv,
2335 .release_cb = tcp_release_cb,
2336 .hash = inet6_hash,
2337 .unhash = inet_unhash,
2338 .get_port = inet_csk_get_port,
2339 .put_port = inet_put_port,
2340#ifdef CONFIG_BPF_SYSCALL
2341 .psock_update_sk_prot = tcp_bpf_update_proto,
2342#endif
2343 .enter_memory_pressure = tcp_enter_memory_pressure,
2344 .leave_memory_pressure = tcp_leave_memory_pressure,
2345 .stream_memory_free = tcp_stream_memory_free,
2346 .sockets_allocated = &tcp_sockets_allocated,
2347
2348 .memory_allocated = &tcp_memory_allocated,
2349 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
2350
2351 .memory_pressure = &tcp_memory_pressure,
2352 .orphan_count = &tcp_orphan_count,
2353 .sysctl_mem = sysctl_tcp_mem,
2354 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2355 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2356 .max_header = MAX_TCP_HEADER,
2357 .obj_size = sizeof(struct tcp6_sock),
2358 .ipv6_pinfo_offset = offsetof(struct tcp6_sock, inet6),
2359 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2360 .twsk_prot = &tcp6_timewait_sock_ops,
2361 .rsk_prot = &tcp6_request_sock_ops,
2362 .h.hashinfo = NULL,
2363 .no_autobind = true,
2364 .diag_destroy = tcp_abort,
2365};
2366EXPORT_SYMBOL_GPL(tcpv6_prot);
2367
2368static const struct inet6_protocol tcpv6_protocol = {
2369 .handler = tcp_v6_rcv,
2370 .err_handler = tcp_v6_err,
2371 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2372};
2373
2374static struct inet_protosw tcpv6_protosw = {
2375 .type = SOCK_STREAM,
2376 .protocol = IPPROTO_TCP,
2377 .prot = &tcpv6_prot,
2378 .ops = &inet6_stream_ops,
2379 .flags = INET_PROTOSW_PERMANENT |
2380 INET_PROTOSW_ICSK,
2381};
2382
2383static int __net_init tcpv6_net_init(struct net *net)
2384{
2385 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2386 SOCK_RAW, IPPROTO_TCP, net);
2387}
2388
2389static void __net_exit tcpv6_net_exit(struct net *net)
2390{
2391 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2392}
2393
2394static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2395{
2396 tcp_twsk_purge(net_exit_list, AF_INET6);
2397}
2398
2399static struct pernet_operations tcpv6_net_ops = {
2400 .init = tcpv6_net_init,
2401 .exit = tcpv6_net_exit,
2402 .exit_batch = tcpv6_net_exit_batch,
2403};
2404
2405int __init tcpv6_init(void)
2406{
2407 int ret;
2408
2409 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2410 if (ret)
2411 goto out;
2412
2413 /* register inet6 protocol */
2414 ret = inet6_register_protosw(&tcpv6_protosw);
2415 if (ret)
2416 goto out_tcpv6_protocol;
2417
2418 ret = register_pernet_subsys(&tcpv6_net_ops);
2419 if (ret)
2420 goto out_tcpv6_protosw;
2421
2422 ret = mptcpv6_init();
2423 if (ret)
2424 goto out_tcpv6_pernet_subsys;
2425
2426out:
2427 return ret;
2428
2429out_tcpv6_pernet_subsys:
2430 unregister_pernet_subsys(&tcpv6_net_ops);
2431out_tcpv6_protosw:
2432 inet6_unregister_protosw(&tcpv6_protosw);
2433out_tcpv6_protocol:
2434 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2435 goto out;
2436}
2437
2438void tcpv6_exit(void)
2439{
2440 unregister_pernet_subsys(&tcpv6_net_ops);
2441 inet6_unregister_protosw(&tcpv6_protosw);
2442 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2443}