Loading...
1/*
2 * TCP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on:
9 * linux/net/ipv4/tcp.c
10 * linux/net/ipv4/tcp_input.c
11 * linux/net/ipv4/tcp_output.c
12 *
13 * Fixes:
14 * Hideaki YOSHIFUJI : sin6_scope_id support
15 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
16 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
17 * a single port at the same time.
18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
19 *
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
24 */
25
26#include <linux/bottom_half.h>
27#include <linux/module.h>
28#include <linux/errno.h>
29#include <linux/types.h>
30#include <linux/socket.h>
31#include <linux/sockios.h>
32#include <linux/net.h>
33#include <linux/jiffies.h>
34#include <linux/in.h>
35#include <linux/in6.h>
36#include <linux/netdevice.h>
37#include <linux/init.h>
38#include <linux/jhash.h>
39#include <linux/ipsec.h>
40#include <linux/times.h>
41#include <linux/slab.h>
42#include <linux/uaccess.h>
43#include <linux/ipv6.h>
44#include <linux/icmpv6.h>
45#include <linux/random.h>
46
47#include <net/tcp.h>
48#include <net/ndisc.h>
49#include <net/inet6_hashtables.h>
50#include <net/inet6_connection_sock.h>
51#include <net/ipv6.h>
52#include <net/transp_v6.h>
53#include <net/addrconf.h>
54#include <net/ip6_route.h>
55#include <net/ip6_checksum.h>
56#include <net/inet_ecn.h>
57#include <net/protocol.h>
58#include <net/xfrm.h>
59#include <net/snmp.h>
60#include <net/dsfield.h>
61#include <net/timewait_sock.h>
62#include <net/inet_common.h>
63#include <net/secure_seq.h>
64#include <net/busy_poll.h>
65
66#include <linux/proc_fs.h>
67#include <linux/seq_file.h>
68
69#include <crypto/hash.h>
70#include <linux/scatterlist.h>
71
72static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
73static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
74 struct request_sock *req);
75
76static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
77
78static const struct inet_connection_sock_af_ops ipv6_mapped;
79static const struct inet_connection_sock_af_ops ipv6_specific;
80#ifdef CONFIG_TCP_MD5SIG
81static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
82static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
83#else
84static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
85 const struct in6_addr *addr)
86{
87 return NULL;
88}
89#endif
90
91static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
92{
93 struct dst_entry *dst = skb_dst(skb);
94
95 if (dst && dst_hold_safe(dst)) {
96 const struct rt6_info *rt = (const struct rt6_info *)dst;
97
98 sk->sk_rx_dst = dst;
99 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
100 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
101 }
102}
103
104static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
105{
106 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
107 ipv6_hdr(skb)->saddr.s6_addr32,
108 tcp_hdr(skb)->dest,
109 tcp_hdr(skb)->source);
110}
111
112static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
113 int addr_len)
114{
115 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
116 struct inet_sock *inet = inet_sk(sk);
117 struct inet_connection_sock *icsk = inet_csk(sk);
118 struct ipv6_pinfo *np = inet6_sk(sk);
119 struct tcp_sock *tp = tcp_sk(sk);
120 struct in6_addr *saddr = NULL, *final_p, final;
121 struct ipv6_txoptions *opt;
122 struct flowi6 fl6;
123 struct dst_entry *dst;
124 int addr_type;
125 int err;
126
127 if (addr_len < SIN6_LEN_RFC2133)
128 return -EINVAL;
129
130 if (usin->sin6_family != AF_INET6)
131 return -EAFNOSUPPORT;
132
133 memset(&fl6, 0, sizeof(fl6));
134
135 if (np->sndflow) {
136 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
137 IP6_ECN_flow_init(fl6.flowlabel);
138 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
139 struct ip6_flowlabel *flowlabel;
140 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
141 if (!flowlabel)
142 return -EINVAL;
143 fl6_sock_release(flowlabel);
144 }
145 }
146
147 /*
148 * connect() to INADDR_ANY means loopback (BSD'ism).
149 */
150
151 if (ipv6_addr_any(&usin->sin6_addr))
152 usin->sin6_addr.s6_addr[15] = 0x1;
153
154 addr_type = ipv6_addr_type(&usin->sin6_addr);
155
156 if (addr_type & IPV6_ADDR_MULTICAST)
157 return -ENETUNREACH;
158
159 if (addr_type&IPV6_ADDR_LINKLOCAL) {
160 if (addr_len >= sizeof(struct sockaddr_in6) &&
161 usin->sin6_scope_id) {
162 /* If interface is set while binding, indices
163 * must coincide.
164 */
165 if (sk->sk_bound_dev_if &&
166 sk->sk_bound_dev_if != usin->sin6_scope_id)
167 return -EINVAL;
168
169 sk->sk_bound_dev_if = usin->sin6_scope_id;
170 }
171
172 /* Connect to link-local address requires an interface */
173 if (!sk->sk_bound_dev_if)
174 return -EINVAL;
175 }
176
177 if (tp->rx_opt.ts_recent_stamp &&
178 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
179 tp->rx_opt.ts_recent = 0;
180 tp->rx_opt.ts_recent_stamp = 0;
181 tp->write_seq = 0;
182 }
183
184 sk->sk_v6_daddr = usin->sin6_addr;
185 np->flow_label = fl6.flowlabel;
186
187 /*
188 * TCP over IPv4
189 */
190
191 if (addr_type == IPV6_ADDR_MAPPED) {
192 u32 exthdrlen = icsk->icsk_ext_hdr_len;
193 struct sockaddr_in sin;
194
195 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
196
197 if (__ipv6_only_sock(sk))
198 return -ENETUNREACH;
199
200 sin.sin_family = AF_INET;
201 sin.sin_port = usin->sin6_port;
202 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
203
204 icsk->icsk_af_ops = &ipv6_mapped;
205 sk->sk_backlog_rcv = tcp_v4_do_rcv;
206#ifdef CONFIG_TCP_MD5SIG
207 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
208#endif
209
210 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
211
212 if (err) {
213 icsk->icsk_ext_hdr_len = exthdrlen;
214 icsk->icsk_af_ops = &ipv6_specific;
215 sk->sk_backlog_rcv = tcp_v6_do_rcv;
216#ifdef CONFIG_TCP_MD5SIG
217 tp->af_specific = &tcp_sock_ipv6_specific;
218#endif
219 goto failure;
220 }
221 np->saddr = sk->sk_v6_rcv_saddr;
222
223 return err;
224 }
225
226 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
227 saddr = &sk->sk_v6_rcv_saddr;
228
229 fl6.flowi6_proto = IPPROTO_TCP;
230 fl6.daddr = sk->sk_v6_daddr;
231 fl6.saddr = saddr ? *saddr : np->saddr;
232 fl6.flowi6_oif = sk->sk_bound_dev_if;
233 fl6.flowi6_mark = sk->sk_mark;
234 fl6.fl6_dport = usin->sin6_port;
235 fl6.fl6_sport = inet->inet_sport;
236
237 opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
238 final_p = fl6_update_dst(&fl6, opt, &final);
239
240 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
241
242 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
243 if (IS_ERR(dst)) {
244 err = PTR_ERR(dst);
245 goto failure;
246 }
247
248 if (!saddr) {
249 saddr = &fl6.saddr;
250 sk->sk_v6_rcv_saddr = *saddr;
251 }
252
253 /* set the source address */
254 np->saddr = *saddr;
255 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
256
257 sk->sk_gso_type = SKB_GSO_TCPV6;
258 ip6_dst_store(sk, dst, NULL, NULL);
259
260 if (tcp_death_row.sysctl_tw_recycle &&
261 !tp->rx_opt.ts_recent_stamp &&
262 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
263 tcp_fetch_timewait_stamp(sk, dst);
264
265 icsk->icsk_ext_hdr_len = 0;
266 if (opt)
267 icsk->icsk_ext_hdr_len = opt->opt_flen +
268 opt->opt_nflen;
269
270 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
271
272 inet->inet_dport = usin->sin6_port;
273
274 tcp_set_state(sk, TCP_SYN_SENT);
275 err = inet6_hash_connect(&tcp_death_row, sk);
276 if (err)
277 goto late_failure;
278
279 sk_set_txhash(sk);
280
281 if (!tp->write_seq && likely(!tp->repair))
282 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
283 sk->sk_v6_daddr.s6_addr32,
284 inet->inet_sport,
285 inet->inet_dport);
286
287 err = tcp_connect(sk);
288 if (err)
289 goto late_failure;
290
291 return 0;
292
293late_failure:
294 tcp_set_state(sk, TCP_CLOSE);
295 __sk_dst_reset(sk);
296failure:
297 inet->inet_dport = 0;
298 sk->sk_route_caps = 0;
299 return err;
300}
301
302static void tcp_v6_mtu_reduced(struct sock *sk)
303{
304 struct dst_entry *dst;
305
306 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
307 return;
308
309 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
310 if (!dst)
311 return;
312
313 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
314 tcp_sync_mss(sk, dst_mtu(dst));
315 tcp_simple_retransmit(sk);
316 }
317}
318
319static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
320 u8 type, u8 code, int offset, __be32 info)
321{
322 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
323 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
324 struct net *net = dev_net(skb->dev);
325 struct request_sock *fastopen;
326 struct ipv6_pinfo *np;
327 struct tcp_sock *tp;
328 __u32 seq, snd_una;
329 struct sock *sk;
330 bool fatal;
331 int err;
332
333 sk = __inet6_lookup_established(net, &tcp_hashinfo,
334 &hdr->daddr, th->dest,
335 &hdr->saddr, ntohs(th->source),
336 skb->dev->ifindex);
337
338 if (!sk) {
339 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
340 ICMP6_MIB_INERRORS);
341 return;
342 }
343
344 if (sk->sk_state == TCP_TIME_WAIT) {
345 inet_twsk_put(inet_twsk(sk));
346 return;
347 }
348 seq = ntohl(th->seq);
349 fatal = icmpv6_err_convert(type, code, &err);
350 if (sk->sk_state == TCP_NEW_SYN_RECV)
351 return tcp_req_err(sk, seq, fatal);
352
353 bh_lock_sock(sk);
354 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
355 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
356
357 if (sk->sk_state == TCP_CLOSE)
358 goto out;
359
360 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
361 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
362 goto out;
363 }
364
365 tp = tcp_sk(sk);
366 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
367 fastopen = tp->fastopen_rsk;
368 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
369 if (sk->sk_state != TCP_LISTEN &&
370 !between(seq, snd_una, tp->snd_nxt)) {
371 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
372 goto out;
373 }
374
375 np = inet6_sk(sk);
376
377 if (type == NDISC_REDIRECT) {
378 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
379
380 if (dst)
381 dst->ops->redirect(dst, sk, skb);
382 goto out;
383 }
384
385 if (type == ICMPV6_PKT_TOOBIG) {
386 /* We are not interested in TCP_LISTEN and open_requests
387 * (SYN-ACKs send out by Linux are always <576bytes so
388 * they should go through unfragmented).
389 */
390 if (sk->sk_state == TCP_LISTEN)
391 goto out;
392
393 if (!ip6_sk_accept_pmtu(sk))
394 goto out;
395
396 tp->mtu_info = ntohl(info);
397 if (!sock_owned_by_user(sk))
398 tcp_v6_mtu_reduced(sk);
399 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
400 &tp->tsq_flags))
401 sock_hold(sk);
402 goto out;
403 }
404
405
406 /* Might be for an request_sock */
407 switch (sk->sk_state) {
408 case TCP_SYN_SENT:
409 case TCP_SYN_RECV:
410 /* Only in fast or simultaneous open. If a fast open socket is
411 * is already accepted it is treated as a connected one below.
412 */
413 if (fastopen && !fastopen->sk)
414 break;
415
416 if (!sock_owned_by_user(sk)) {
417 sk->sk_err = err;
418 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
419
420 tcp_done(sk);
421 } else
422 sk->sk_err_soft = err;
423 goto out;
424 }
425
426 if (!sock_owned_by_user(sk) && np->recverr) {
427 sk->sk_err = err;
428 sk->sk_error_report(sk);
429 } else
430 sk->sk_err_soft = err;
431
432out:
433 bh_unlock_sock(sk);
434 sock_put(sk);
435}
436
437
438static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
439 struct flowi *fl,
440 struct request_sock *req,
441 struct tcp_fastopen_cookie *foc,
442 bool attach_req)
443{
444 struct inet_request_sock *ireq = inet_rsk(req);
445 struct ipv6_pinfo *np = inet6_sk(sk);
446 struct flowi6 *fl6 = &fl->u.ip6;
447 struct sk_buff *skb;
448 int err = -ENOMEM;
449
450 /* First, grab a route. */
451 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
452 IPPROTO_TCP)) == NULL)
453 goto done;
454
455 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
456
457 if (skb) {
458 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
459 &ireq->ir_v6_rmt_addr);
460
461 fl6->daddr = ireq->ir_v6_rmt_addr;
462 if (np->repflow && ireq->pktopts)
463 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
464
465 rcu_read_lock();
466 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
467 np->tclass);
468 rcu_read_unlock();
469 err = net_xmit_eval(err);
470 }
471
472done:
473 return err;
474}
475
476
477static void tcp_v6_reqsk_destructor(struct request_sock *req)
478{
479 kfree_skb(inet_rsk(req)->pktopts);
480}
481
482#ifdef CONFIG_TCP_MD5SIG
483static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
484 const struct in6_addr *addr)
485{
486 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
487}
488
489static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
490 const struct sock *addr_sk)
491{
492 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
493}
494
495static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
496 int optlen)
497{
498 struct tcp_md5sig cmd;
499 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
500
501 if (optlen < sizeof(cmd))
502 return -EINVAL;
503
504 if (copy_from_user(&cmd, optval, sizeof(cmd)))
505 return -EFAULT;
506
507 if (sin6->sin6_family != AF_INET6)
508 return -EINVAL;
509
510 if (!cmd.tcpm_keylen) {
511 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
512 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
513 AF_INET);
514 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
515 AF_INET6);
516 }
517
518 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
519 return -EINVAL;
520
521 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
522 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
523 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
524
525 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
526 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
527}
528
529static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
530 const struct in6_addr *daddr,
531 const struct in6_addr *saddr, int nbytes)
532{
533 struct tcp6_pseudohdr *bp;
534 struct scatterlist sg;
535
536 bp = &hp->md5_blk.ip6;
537 /* 1. TCP pseudo-header (RFC2460) */
538 bp->saddr = *saddr;
539 bp->daddr = *daddr;
540 bp->protocol = cpu_to_be32(IPPROTO_TCP);
541 bp->len = cpu_to_be32(nbytes);
542
543 sg_init_one(&sg, bp, sizeof(*bp));
544 ahash_request_set_crypt(hp->md5_req, &sg, NULL, sizeof(*bp));
545 return crypto_ahash_update(hp->md5_req);
546}
547
548static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
549 const struct in6_addr *daddr, struct in6_addr *saddr,
550 const struct tcphdr *th)
551{
552 struct tcp_md5sig_pool *hp;
553 struct ahash_request *req;
554
555 hp = tcp_get_md5sig_pool();
556 if (!hp)
557 goto clear_hash_noput;
558 req = hp->md5_req;
559
560 if (crypto_ahash_init(req))
561 goto clear_hash;
562 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
563 goto clear_hash;
564 if (tcp_md5_hash_header(hp, th))
565 goto clear_hash;
566 if (tcp_md5_hash_key(hp, key))
567 goto clear_hash;
568 ahash_request_set_crypt(req, NULL, md5_hash, 0);
569 if (crypto_ahash_final(req))
570 goto clear_hash;
571
572 tcp_put_md5sig_pool();
573 return 0;
574
575clear_hash:
576 tcp_put_md5sig_pool();
577clear_hash_noput:
578 memset(md5_hash, 0, 16);
579 return 1;
580}
581
582static int tcp_v6_md5_hash_skb(char *md5_hash,
583 const struct tcp_md5sig_key *key,
584 const struct sock *sk,
585 const struct sk_buff *skb)
586{
587 const struct in6_addr *saddr, *daddr;
588 struct tcp_md5sig_pool *hp;
589 struct ahash_request *req;
590 const struct tcphdr *th = tcp_hdr(skb);
591
592 if (sk) { /* valid for establish/request sockets */
593 saddr = &sk->sk_v6_rcv_saddr;
594 daddr = &sk->sk_v6_daddr;
595 } else {
596 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
597 saddr = &ip6h->saddr;
598 daddr = &ip6h->daddr;
599 }
600
601 hp = tcp_get_md5sig_pool();
602 if (!hp)
603 goto clear_hash_noput;
604 req = hp->md5_req;
605
606 if (crypto_ahash_init(req))
607 goto clear_hash;
608
609 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
610 goto clear_hash;
611 if (tcp_md5_hash_header(hp, th))
612 goto clear_hash;
613 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
614 goto clear_hash;
615 if (tcp_md5_hash_key(hp, key))
616 goto clear_hash;
617 ahash_request_set_crypt(req, NULL, md5_hash, 0);
618 if (crypto_ahash_final(req))
619 goto clear_hash;
620
621 tcp_put_md5sig_pool();
622 return 0;
623
624clear_hash:
625 tcp_put_md5sig_pool();
626clear_hash_noput:
627 memset(md5_hash, 0, 16);
628 return 1;
629}
630
631#endif
632
633static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
634 const struct sk_buff *skb)
635{
636#ifdef CONFIG_TCP_MD5SIG
637 const __u8 *hash_location = NULL;
638 struct tcp_md5sig_key *hash_expected;
639 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
640 const struct tcphdr *th = tcp_hdr(skb);
641 int genhash;
642 u8 newhash[16];
643
644 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
645 hash_location = tcp_parse_md5sig_option(th);
646
647 /* We've parsed the options - do we have a hash? */
648 if (!hash_expected && !hash_location)
649 return false;
650
651 if (hash_expected && !hash_location) {
652 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
653 return true;
654 }
655
656 if (!hash_expected && hash_location) {
657 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
658 return true;
659 }
660
661 /* check the signature */
662 genhash = tcp_v6_md5_hash_skb(newhash,
663 hash_expected,
664 NULL, skb);
665
666 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
667 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
668 genhash ? "failed" : "mismatch",
669 &ip6h->saddr, ntohs(th->source),
670 &ip6h->daddr, ntohs(th->dest));
671 return true;
672 }
673#endif
674 return false;
675}
676
677static void tcp_v6_init_req(struct request_sock *req,
678 const struct sock *sk_listener,
679 struct sk_buff *skb)
680{
681 struct inet_request_sock *ireq = inet_rsk(req);
682 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
683
684 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
685 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
686
687 /* So that link locals have meaning */
688 if (!sk_listener->sk_bound_dev_if &&
689 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
690 ireq->ir_iif = tcp_v6_iif(skb);
691
692 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
693 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
694 np->rxopt.bits.rxinfo ||
695 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
696 np->rxopt.bits.rxohlim || np->repflow)) {
697 atomic_inc(&skb->users);
698 ireq->pktopts = skb;
699 }
700}
701
702static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
703 struct flowi *fl,
704 const struct request_sock *req,
705 bool *strict)
706{
707 if (strict)
708 *strict = true;
709 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
710}
711
712struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
713 .family = AF_INET6,
714 .obj_size = sizeof(struct tcp6_request_sock),
715 .rtx_syn_ack = tcp_rtx_synack,
716 .send_ack = tcp_v6_reqsk_send_ack,
717 .destructor = tcp_v6_reqsk_destructor,
718 .send_reset = tcp_v6_send_reset,
719 .syn_ack_timeout = tcp_syn_ack_timeout,
720};
721
722static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
723 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
724 sizeof(struct ipv6hdr),
725#ifdef CONFIG_TCP_MD5SIG
726 .req_md5_lookup = tcp_v6_md5_lookup,
727 .calc_md5_hash = tcp_v6_md5_hash_skb,
728#endif
729 .init_req = tcp_v6_init_req,
730#ifdef CONFIG_SYN_COOKIES
731 .cookie_init_seq = cookie_v6_init_sequence,
732#endif
733 .route_req = tcp_v6_route_req,
734 .init_seq = tcp_v6_init_sequence,
735 .send_synack = tcp_v6_send_synack,
736};
737
738static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
739 u32 ack, u32 win, u32 tsval, u32 tsecr,
740 int oif, struct tcp_md5sig_key *key, int rst,
741 u8 tclass, u32 label)
742{
743 const struct tcphdr *th = tcp_hdr(skb);
744 struct tcphdr *t1;
745 struct sk_buff *buff;
746 struct flowi6 fl6;
747 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
748 struct sock *ctl_sk = net->ipv6.tcp_sk;
749 unsigned int tot_len = sizeof(struct tcphdr);
750 struct dst_entry *dst;
751 __be32 *topt;
752
753 if (tsecr)
754 tot_len += TCPOLEN_TSTAMP_ALIGNED;
755#ifdef CONFIG_TCP_MD5SIG
756 if (key)
757 tot_len += TCPOLEN_MD5SIG_ALIGNED;
758#endif
759
760 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
761 GFP_ATOMIC);
762 if (!buff)
763 return;
764
765 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
766
767 t1 = (struct tcphdr *) skb_push(buff, tot_len);
768 skb_reset_transport_header(buff);
769
770 /* Swap the send and the receive. */
771 memset(t1, 0, sizeof(*t1));
772 t1->dest = th->source;
773 t1->source = th->dest;
774 t1->doff = tot_len / 4;
775 t1->seq = htonl(seq);
776 t1->ack_seq = htonl(ack);
777 t1->ack = !rst || !th->ack;
778 t1->rst = rst;
779 t1->window = htons(win);
780
781 topt = (__be32 *)(t1 + 1);
782
783 if (tsecr) {
784 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
785 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
786 *topt++ = htonl(tsval);
787 *topt++ = htonl(tsecr);
788 }
789
790#ifdef CONFIG_TCP_MD5SIG
791 if (key) {
792 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
793 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
794 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
795 &ipv6_hdr(skb)->saddr,
796 &ipv6_hdr(skb)->daddr, t1);
797 }
798#endif
799
800 memset(&fl6, 0, sizeof(fl6));
801 fl6.daddr = ipv6_hdr(skb)->saddr;
802 fl6.saddr = ipv6_hdr(skb)->daddr;
803 fl6.flowlabel = label;
804
805 buff->ip_summed = CHECKSUM_PARTIAL;
806 buff->csum = 0;
807
808 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
809
810 fl6.flowi6_proto = IPPROTO_TCP;
811 if (rt6_need_strict(&fl6.daddr) && !oif)
812 fl6.flowi6_oif = tcp_v6_iif(skb);
813 else {
814 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
815 oif = skb->skb_iif;
816
817 fl6.flowi6_oif = oif;
818 }
819
820 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
821 fl6.fl6_dport = t1->dest;
822 fl6.fl6_sport = t1->source;
823 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
824
825 /* Pass a socket to ip6_dst_lookup either it is for RST
826 * Underlying function will use this to retrieve the network
827 * namespace
828 */
829 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
830 if (!IS_ERR(dst)) {
831 skb_dst_set(buff, dst);
832 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
833 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
834 if (rst)
835 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
836 return;
837 }
838
839 kfree_skb(buff);
840}
841
842static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
843{
844 const struct tcphdr *th = tcp_hdr(skb);
845 u32 seq = 0, ack_seq = 0;
846 struct tcp_md5sig_key *key = NULL;
847#ifdef CONFIG_TCP_MD5SIG
848 const __u8 *hash_location = NULL;
849 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
850 unsigned char newhash[16];
851 int genhash;
852 struct sock *sk1 = NULL;
853#endif
854 int oif;
855
856 if (th->rst)
857 return;
858
859 /* If sk not NULL, it means we did a successful lookup and incoming
860 * route had to be correct. prequeue might have dropped our dst.
861 */
862 if (!sk && !ipv6_unicast_destination(skb))
863 return;
864
865#ifdef CONFIG_TCP_MD5SIG
866 hash_location = tcp_parse_md5sig_option(th);
867 if (sk && sk_fullsock(sk)) {
868 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
869 } else if (hash_location) {
870 /*
871 * active side is lost. Try to find listening socket through
872 * source port, and then find md5 key through listening socket.
873 * we are not loose security here:
874 * Incoming packet is checked with md5 hash with finding key,
875 * no RST generated if md5 hash doesn't match.
876 */
877 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
878 &tcp_hashinfo, NULL, 0,
879 &ipv6h->saddr,
880 th->source, &ipv6h->daddr,
881 ntohs(th->source), tcp_v6_iif(skb));
882 if (!sk1)
883 return;
884
885 rcu_read_lock();
886 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
887 if (!key)
888 goto release_sk1;
889
890 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
891 if (genhash || memcmp(hash_location, newhash, 16) != 0)
892 goto release_sk1;
893 }
894#endif
895
896 if (th->ack)
897 seq = ntohl(th->ack_seq);
898 else
899 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
900 (th->doff << 2);
901
902 oif = sk ? sk->sk_bound_dev_if : 0;
903 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
904
905#ifdef CONFIG_TCP_MD5SIG
906release_sk1:
907 if (sk1) {
908 rcu_read_unlock();
909 sock_put(sk1);
910 }
911#endif
912}
913
914static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
915 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
916 struct tcp_md5sig_key *key, u8 tclass,
917 u32 label)
918{
919 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
920 tclass, label);
921}
922
923static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
924{
925 struct inet_timewait_sock *tw = inet_twsk(sk);
926 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
927
928 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
929 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
930 tcp_time_stamp + tcptw->tw_ts_offset,
931 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
932 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
933
934 inet_twsk_put(tw);
935}
936
937static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
938 struct request_sock *req)
939{
940 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
941 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
942 */
943 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
944 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
945 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
946 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
947 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
948 0, 0);
949}
950
951
952static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
953{
954#ifdef CONFIG_SYN_COOKIES
955 const struct tcphdr *th = tcp_hdr(skb);
956
957 if (!th->syn)
958 sk = cookie_v6_check(sk, skb);
959#endif
960 return sk;
961}
962
963static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
964{
965 if (skb->protocol == htons(ETH_P_IP))
966 return tcp_v4_conn_request(sk, skb);
967
968 if (!ipv6_unicast_destination(skb))
969 goto drop;
970
971 return tcp_conn_request(&tcp6_request_sock_ops,
972 &tcp_request_sock_ipv6_ops, sk, skb);
973
974drop:
975 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
976 return 0; /* don't send reset */
977}
978
979static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
980 struct request_sock *req,
981 struct dst_entry *dst,
982 struct request_sock *req_unhash,
983 bool *own_req)
984{
985 struct inet_request_sock *ireq;
986 struct ipv6_pinfo *newnp;
987 const struct ipv6_pinfo *np = inet6_sk(sk);
988 struct ipv6_txoptions *opt;
989 struct tcp6_sock *newtcp6sk;
990 struct inet_sock *newinet;
991 struct tcp_sock *newtp;
992 struct sock *newsk;
993#ifdef CONFIG_TCP_MD5SIG
994 struct tcp_md5sig_key *key;
995#endif
996 struct flowi6 fl6;
997
998 if (skb->protocol == htons(ETH_P_IP)) {
999 /*
1000 * v6 mapped
1001 */
1002
1003 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1004 req_unhash, own_req);
1005
1006 if (!newsk)
1007 return NULL;
1008
1009 newtcp6sk = (struct tcp6_sock *)newsk;
1010 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1011
1012 newinet = inet_sk(newsk);
1013 newnp = inet6_sk(newsk);
1014 newtp = tcp_sk(newsk);
1015
1016 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1017
1018 newnp->saddr = newsk->sk_v6_rcv_saddr;
1019
1020 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1021 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1022#ifdef CONFIG_TCP_MD5SIG
1023 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1024#endif
1025
1026 newnp->ipv6_ac_list = NULL;
1027 newnp->ipv6_fl_list = NULL;
1028 newnp->pktoptions = NULL;
1029 newnp->opt = NULL;
1030 newnp->mcast_oif = tcp_v6_iif(skb);
1031 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1032 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1033 if (np->repflow)
1034 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1035
1036 /*
1037 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1038 * here, tcp_create_openreq_child now does this for us, see the comment in
1039 * that function for the gory details. -acme
1040 */
1041
1042 /* It is tricky place. Until this moment IPv4 tcp
1043 worked with IPv6 icsk.icsk_af_ops.
1044 Sync it now.
1045 */
1046 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1047
1048 return newsk;
1049 }
1050
1051 ireq = inet_rsk(req);
1052
1053 if (sk_acceptq_is_full(sk))
1054 goto out_overflow;
1055
1056 if (!dst) {
1057 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1058 if (!dst)
1059 goto out;
1060 }
1061
1062 newsk = tcp_create_openreq_child(sk, req, skb);
1063 if (!newsk)
1064 goto out_nonewsk;
1065
1066 /*
1067 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1068 * count here, tcp_create_openreq_child now does this for us, see the
1069 * comment in that function for the gory details. -acme
1070 */
1071
1072 newsk->sk_gso_type = SKB_GSO_TCPV6;
1073 ip6_dst_store(newsk, dst, NULL, NULL);
1074 inet6_sk_rx_dst_set(newsk, skb);
1075
1076 newtcp6sk = (struct tcp6_sock *)newsk;
1077 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1078
1079 newtp = tcp_sk(newsk);
1080 newinet = inet_sk(newsk);
1081 newnp = inet6_sk(newsk);
1082
1083 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1084
1085 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1086 newnp->saddr = ireq->ir_v6_loc_addr;
1087 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1088 newsk->sk_bound_dev_if = ireq->ir_iif;
1089
1090 /* Now IPv6 options...
1091
1092 First: no IPv4 options.
1093 */
1094 newinet->inet_opt = NULL;
1095 newnp->ipv6_ac_list = NULL;
1096 newnp->ipv6_fl_list = NULL;
1097
1098 /* Clone RX bits */
1099 newnp->rxopt.all = np->rxopt.all;
1100
1101 newnp->pktoptions = NULL;
1102 newnp->opt = NULL;
1103 newnp->mcast_oif = tcp_v6_iif(skb);
1104 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1105 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1106 if (np->repflow)
1107 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1108
1109 /* Clone native IPv6 options from listening socket (if any)
1110
1111 Yes, keeping reference count would be much more clever,
1112 but we make one more one thing there: reattach optmem
1113 to newsk.
1114 */
1115 opt = rcu_dereference(np->opt);
1116 if (opt) {
1117 opt = ipv6_dup_options(newsk, opt);
1118 RCU_INIT_POINTER(newnp->opt, opt);
1119 }
1120 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1121 if (opt)
1122 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1123 opt->opt_flen;
1124
1125 tcp_ca_openreq_child(newsk, dst);
1126
1127 tcp_sync_mss(newsk, dst_mtu(dst));
1128 newtp->advmss = dst_metric_advmss(dst);
1129 if (tcp_sk(sk)->rx_opt.user_mss &&
1130 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1131 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1132
1133 tcp_initialize_rcv_mss(newsk);
1134
1135 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1136 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1137
1138#ifdef CONFIG_TCP_MD5SIG
1139 /* Copy over the MD5 key from the original socket */
1140 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1141 if (key) {
1142 /* We're using one, so create a matching key
1143 * on the newsk structure. If we fail to get
1144 * memory, then we end up not copying the key
1145 * across. Shucks.
1146 */
1147 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1148 AF_INET6, key->key, key->keylen,
1149 sk_gfp_mask(sk, GFP_ATOMIC));
1150 }
1151#endif
1152
1153 if (__inet_inherit_port(sk, newsk) < 0) {
1154 inet_csk_prepare_forced_close(newsk);
1155 tcp_done(newsk);
1156 goto out;
1157 }
1158 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1159 if (*own_req) {
1160 tcp_move_syn(newtp, req);
1161
1162 /* Clone pktoptions received with SYN, if we own the req */
1163 if (ireq->pktopts) {
1164 newnp->pktoptions = skb_clone(ireq->pktopts,
1165 sk_gfp_mask(sk, GFP_ATOMIC));
1166 consume_skb(ireq->pktopts);
1167 ireq->pktopts = NULL;
1168 if (newnp->pktoptions)
1169 skb_set_owner_r(newnp->pktoptions, newsk);
1170 }
1171 }
1172
1173 return newsk;
1174
1175out_overflow:
1176 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1177out_nonewsk:
1178 dst_release(dst);
1179out:
1180 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1181 return NULL;
1182}
1183
1184/* The socket must have it's spinlock held when we get
1185 * here, unless it is a TCP_LISTEN socket.
1186 *
1187 * We have a potential double-lock case here, so even when
1188 * doing backlog processing we use the BH locking scheme.
1189 * This is because we cannot sleep with the original spinlock
1190 * held.
1191 */
1192static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1193{
1194 struct ipv6_pinfo *np = inet6_sk(sk);
1195 struct tcp_sock *tp;
1196 struct sk_buff *opt_skb = NULL;
1197
1198 /* Imagine: socket is IPv6. IPv4 packet arrives,
1199 goes to IPv4 receive handler and backlogged.
1200 From backlog it always goes here. Kerboom...
1201 Fortunately, tcp_rcv_established and rcv_established
1202 handle them correctly, but it is not case with
1203 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1204 */
1205
1206 if (skb->protocol == htons(ETH_P_IP))
1207 return tcp_v4_do_rcv(sk, skb);
1208
1209 if (sk_filter(sk, skb))
1210 goto discard;
1211
1212 /*
1213 * socket locking is here for SMP purposes as backlog rcv
1214 * is currently called with bh processing disabled.
1215 */
1216
1217 /* Do Stevens' IPV6_PKTOPTIONS.
1218
1219 Yes, guys, it is the only place in our code, where we
1220 may make it not affecting IPv4.
1221 The rest of code is protocol independent,
1222 and I do not like idea to uglify IPv4.
1223
1224 Actually, all the idea behind IPV6_PKTOPTIONS
1225 looks not very well thought. For now we latch
1226 options, received in the last packet, enqueued
1227 by tcp. Feel free to propose better solution.
1228 --ANK (980728)
1229 */
1230 if (np->rxopt.all)
1231 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1232
1233 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1234 struct dst_entry *dst = sk->sk_rx_dst;
1235
1236 sock_rps_save_rxhash(sk, skb);
1237 sk_mark_napi_id(sk, skb);
1238 if (dst) {
1239 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1240 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1241 dst_release(dst);
1242 sk->sk_rx_dst = NULL;
1243 }
1244 }
1245
1246 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1247 if (opt_skb)
1248 goto ipv6_pktoptions;
1249 return 0;
1250 }
1251
1252 if (tcp_checksum_complete(skb))
1253 goto csum_err;
1254
1255 if (sk->sk_state == TCP_LISTEN) {
1256 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1257
1258 if (!nsk)
1259 goto discard;
1260
1261 if (nsk != sk) {
1262 sock_rps_save_rxhash(nsk, skb);
1263 sk_mark_napi_id(nsk, skb);
1264 if (tcp_child_process(sk, nsk, skb))
1265 goto reset;
1266 if (opt_skb)
1267 __kfree_skb(opt_skb);
1268 return 0;
1269 }
1270 } else
1271 sock_rps_save_rxhash(sk, skb);
1272
1273 if (tcp_rcv_state_process(sk, skb))
1274 goto reset;
1275 if (opt_skb)
1276 goto ipv6_pktoptions;
1277 return 0;
1278
1279reset:
1280 tcp_v6_send_reset(sk, skb);
1281discard:
1282 if (opt_skb)
1283 __kfree_skb(opt_skb);
1284 kfree_skb(skb);
1285 return 0;
1286csum_err:
1287 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1288 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1289 goto discard;
1290
1291
1292ipv6_pktoptions:
1293 /* Do you ask, what is it?
1294
1295 1. skb was enqueued by tcp.
1296 2. skb is added to tail of read queue, rather than out of order.
1297 3. socket is not in passive state.
1298 4. Finally, it really contains options, which user wants to receive.
1299 */
1300 tp = tcp_sk(sk);
1301 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1302 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1303 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1304 np->mcast_oif = tcp_v6_iif(opt_skb);
1305 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1306 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1307 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1308 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1309 if (np->repflow)
1310 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1311 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1312 skb_set_owner_r(opt_skb, sk);
1313 opt_skb = xchg(&np->pktoptions, opt_skb);
1314 } else {
1315 __kfree_skb(opt_skb);
1316 opt_skb = xchg(&np->pktoptions, NULL);
1317 }
1318 }
1319
1320 kfree_skb(opt_skb);
1321 return 0;
1322}
1323
1324static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1325 const struct tcphdr *th)
1326{
1327 /* This is tricky: we move IP6CB at its correct location into
1328 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1329 * _decode_session6() uses IP6CB().
1330 * barrier() makes sure compiler won't play aliasing games.
1331 */
1332 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1333 sizeof(struct inet6_skb_parm));
1334 barrier();
1335
1336 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1337 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1338 skb->len - th->doff*4);
1339 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1340 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1341 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1342 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1343 TCP_SKB_CB(skb)->sacked = 0;
1344}
1345
1346static void tcp_v6_restore_cb(struct sk_buff *skb)
1347{
1348 /* We need to move header back to the beginning if xfrm6_policy_check()
1349 * and tcp_v6_fill_cb() are going to be called again.
1350 */
1351 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1352 sizeof(struct inet6_skb_parm));
1353}
1354
1355static int tcp_v6_rcv(struct sk_buff *skb)
1356{
1357 const struct tcphdr *th;
1358 const struct ipv6hdr *hdr;
1359 struct sock *sk;
1360 int ret;
1361 struct net *net = dev_net(skb->dev);
1362
1363 if (skb->pkt_type != PACKET_HOST)
1364 goto discard_it;
1365
1366 /*
1367 * Count it even if it's bad.
1368 */
1369 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1370
1371 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1372 goto discard_it;
1373
1374 th = tcp_hdr(skb);
1375
1376 if (th->doff < sizeof(struct tcphdr)/4)
1377 goto bad_packet;
1378 if (!pskb_may_pull(skb, th->doff*4))
1379 goto discard_it;
1380
1381 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1382 goto csum_error;
1383
1384 th = tcp_hdr(skb);
1385 hdr = ipv6_hdr(skb);
1386
1387lookup:
1388 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1389 th->source, th->dest, inet6_iif(skb));
1390 if (!sk)
1391 goto no_tcp_socket;
1392
1393process:
1394 if (sk->sk_state == TCP_TIME_WAIT)
1395 goto do_time_wait;
1396
1397 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1398 struct request_sock *req = inet_reqsk(sk);
1399 struct sock *nsk;
1400
1401 sk = req->rsk_listener;
1402 tcp_v6_fill_cb(skb, hdr, th);
1403 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1404 reqsk_put(req);
1405 goto discard_it;
1406 }
1407 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1408 inet_csk_reqsk_queue_drop_and_put(sk, req);
1409 goto lookup;
1410 }
1411 sock_hold(sk);
1412 nsk = tcp_check_req(sk, skb, req, false);
1413 if (!nsk) {
1414 reqsk_put(req);
1415 goto discard_and_relse;
1416 }
1417 if (nsk == sk) {
1418 reqsk_put(req);
1419 tcp_v6_restore_cb(skb);
1420 } else if (tcp_child_process(sk, nsk, skb)) {
1421 tcp_v6_send_reset(nsk, skb);
1422 goto discard_and_relse;
1423 } else {
1424 sock_put(sk);
1425 return 0;
1426 }
1427 }
1428 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1429 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1430 goto discard_and_relse;
1431 }
1432
1433 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1434 goto discard_and_relse;
1435
1436 tcp_v6_fill_cb(skb, hdr, th);
1437
1438 if (tcp_v6_inbound_md5_hash(sk, skb))
1439 goto discard_and_relse;
1440
1441 if (sk_filter(sk, skb))
1442 goto discard_and_relse;
1443
1444 skb->dev = NULL;
1445
1446 if (sk->sk_state == TCP_LISTEN) {
1447 ret = tcp_v6_do_rcv(sk, skb);
1448 goto put_and_return;
1449 }
1450
1451 sk_incoming_cpu_update(sk);
1452
1453 bh_lock_sock_nested(sk);
1454 tcp_segs_in(tcp_sk(sk), skb);
1455 ret = 0;
1456 if (!sock_owned_by_user(sk)) {
1457 if (!tcp_prequeue(sk, skb))
1458 ret = tcp_v6_do_rcv(sk, skb);
1459 } else if (unlikely(sk_add_backlog(sk, skb,
1460 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1461 bh_unlock_sock(sk);
1462 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1463 goto discard_and_relse;
1464 }
1465 bh_unlock_sock(sk);
1466
1467put_and_return:
1468 sock_put(sk);
1469 return ret ? -1 : 0;
1470
1471no_tcp_socket:
1472 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1473 goto discard_it;
1474
1475 tcp_v6_fill_cb(skb, hdr, th);
1476
1477 if (tcp_checksum_complete(skb)) {
1478csum_error:
1479 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1480bad_packet:
1481 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1482 } else {
1483 tcp_v6_send_reset(NULL, skb);
1484 }
1485
1486discard_it:
1487 kfree_skb(skb);
1488 return 0;
1489
1490discard_and_relse:
1491 sock_put(sk);
1492 goto discard_it;
1493
1494do_time_wait:
1495 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1496 inet_twsk_put(inet_twsk(sk));
1497 goto discard_it;
1498 }
1499
1500 tcp_v6_fill_cb(skb, hdr, th);
1501
1502 if (tcp_checksum_complete(skb)) {
1503 inet_twsk_put(inet_twsk(sk));
1504 goto csum_error;
1505 }
1506
1507 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1508 case TCP_TW_SYN:
1509 {
1510 struct sock *sk2;
1511
1512 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1513 skb, __tcp_hdrlen(th),
1514 &ipv6_hdr(skb)->saddr, th->source,
1515 &ipv6_hdr(skb)->daddr,
1516 ntohs(th->dest), tcp_v6_iif(skb));
1517 if (sk2) {
1518 struct inet_timewait_sock *tw = inet_twsk(sk);
1519 inet_twsk_deschedule_put(tw);
1520 sk = sk2;
1521 tcp_v6_restore_cb(skb);
1522 goto process;
1523 }
1524 /* Fall through to ACK */
1525 }
1526 case TCP_TW_ACK:
1527 tcp_v6_timewait_ack(sk, skb);
1528 break;
1529 case TCP_TW_RST:
1530 tcp_v6_restore_cb(skb);
1531 tcp_v6_send_reset(sk, skb);
1532 inet_twsk_deschedule_put(inet_twsk(sk));
1533 goto discard_it;
1534 case TCP_TW_SUCCESS:
1535 ;
1536 }
1537 goto discard_it;
1538}
1539
1540static void tcp_v6_early_demux(struct sk_buff *skb)
1541{
1542 const struct ipv6hdr *hdr;
1543 const struct tcphdr *th;
1544 struct sock *sk;
1545
1546 if (skb->pkt_type != PACKET_HOST)
1547 return;
1548
1549 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1550 return;
1551
1552 hdr = ipv6_hdr(skb);
1553 th = tcp_hdr(skb);
1554
1555 if (th->doff < sizeof(struct tcphdr) / 4)
1556 return;
1557
1558 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1559 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1560 &hdr->saddr, th->source,
1561 &hdr->daddr, ntohs(th->dest),
1562 inet6_iif(skb));
1563 if (sk) {
1564 skb->sk = sk;
1565 skb->destructor = sock_edemux;
1566 if (sk_fullsock(sk)) {
1567 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1568
1569 if (dst)
1570 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1571 if (dst &&
1572 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1573 skb_dst_set_noref(skb, dst);
1574 }
1575 }
1576}
1577
1578static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1579 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1580 .twsk_unique = tcp_twsk_unique,
1581 .twsk_destructor = tcp_twsk_destructor,
1582};
1583
1584static const struct inet_connection_sock_af_ops ipv6_specific = {
1585 .queue_xmit = inet6_csk_xmit,
1586 .send_check = tcp_v6_send_check,
1587 .rebuild_header = inet6_sk_rebuild_header,
1588 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1589 .conn_request = tcp_v6_conn_request,
1590 .syn_recv_sock = tcp_v6_syn_recv_sock,
1591 .net_header_len = sizeof(struct ipv6hdr),
1592 .net_frag_header_len = sizeof(struct frag_hdr),
1593 .setsockopt = ipv6_setsockopt,
1594 .getsockopt = ipv6_getsockopt,
1595 .addr2sockaddr = inet6_csk_addr2sockaddr,
1596 .sockaddr_len = sizeof(struct sockaddr_in6),
1597 .bind_conflict = inet6_csk_bind_conflict,
1598#ifdef CONFIG_COMPAT
1599 .compat_setsockopt = compat_ipv6_setsockopt,
1600 .compat_getsockopt = compat_ipv6_getsockopt,
1601#endif
1602 .mtu_reduced = tcp_v6_mtu_reduced,
1603};
1604
1605#ifdef CONFIG_TCP_MD5SIG
1606static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1607 .md5_lookup = tcp_v6_md5_lookup,
1608 .calc_md5_hash = tcp_v6_md5_hash_skb,
1609 .md5_parse = tcp_v6_parse_md5_keys,
1610};
1611#endif
1612
1613/*
1614 * TCP over IPv4 via INET6 API
1615 */
1616static const struct inet_connection_sock_af_ops ipv6_mapped = {
1617 .queue_xmit = ip_queue_xmit,
1618 .send_check = tcp_v4_send_check,
1619 .rebuild_header = inet_sk_rebuild_header,
1620 .sk_rx_dst_set = inet_sk_rx_dst_set,
1621 .conn_request = tcp_v6_conn_request,
1622 .syn_recv_sock = tcp_v6_syn_recv_sock,
1623 .net_header_len = sizeof(struct iphdr),
1624 .setsockopt = ipv6_setsockopt,
1625 .getsockopt = ipv6_getsockopt,
1626 .addr2sockaddr = inet6_csk_addr2sockaddr,
1627 .sockaddr_len = sizeof(struct sockaddr_in6),
1628 .bind_conflict = inet6_csk_bind_conflict,
1629#ifdef CONFIG_COMPAT
1630 .compat_setsockopt = compat_ipv6_setsockopt,
1631 .compat_getsockopt = compat_ipv6_getsockopt,
1632#endif
1633 .mtu_reduced = tcp_v4_mtu_reduced,
1634};
1635
1636#ifdef CONFIG_TCP_MD5SIG
1637static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1638 .md5_lookup = tcp_v4_md5_lookup,
1639 .calc_md5_hash = tcp_v4_md5_hash_skb,
1640 .md5_parse = tcp_v6_parse_md5_keys,
1641};
1642#endif
1643
1644/* NOTE: A lot of things set to zero explicitly by call to
1645 * sk_alloc() so need not be done here.
1646 */
1647static int tcp_v6_init_sock(struct sock *sk)
1648{
1649 struct inet_connection_sock *icsk = inet_csk(sk);
1650
1651 tcp_init_sock(sk);
1652
1653 icsk->icsk_af_ops = &ipv6_specific;
1654
1655#ifdef CONFIG_TCP_MD5SIG
1656 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1657#endif
1658
1659 return 0;
1660}
1661
1662static void tcp_v6_destroy_sock(struct sock *sk)
1663{
1664 tcp_v4_destroy_sock(sk);
1665 inet6_destroy_sock(sk);
1666}
1667
1668#ifdef CONFIG_PROC_FS
1669/* Proc filesystem TCPv6 sock list dumping. */
1670static void get_openreq6(struct seq_file *seq,
1671 const struct request_sock *req, int i)
1672{
1673 long ttd = req->rsk_timer.expires - jiffies;
1674 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1675 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1676
1677 if (ttd < 0)
1678 ttd = 0;
1679
1680 seq_printf(seq,
1681 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1682 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1683 i,
1684 src->s6_addr32[0], src->s6_addr32[1],
1685 src->s6_addr32[2], src->s6_addr32[3],
1686 inet_rsk(req)->ir_num,
1687 dest->s6_addr32[0], dest->s6_addr32[1],
1688 dest->s6_addr32[2], dest->s6_addr32[3],
1689 ntohs(inet_rsk(req)->ir_rmt_port),
1690 TCP_SYN_RECV,
1691 0, 0, /* could print option size, but that is af dependent. */
1692 1, /* timers active (only the expire timer) */
1693 jiffies_to_clock_t(ttd),
1694 req->num_timeout,
1695 from_kuid_munged(seq_user_ns(seq),
1696 sock_i_uid(req->rsk_listener)),
1697 0, /* non standard timer */
1698 0, /* open_requests have no inode */
1699 0, req);
1700}
1701
1702static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1703{
1704 const struct in6_addr *dest, *src;
1705 __u16 destp, srcp;
1706 int timer_active;
1707 unsigned long timer_expires;
1708 const struct inet_sock *inet = inet_sk(sp);
1709 const struct tcp_sock *tp = tcp_sk(sp);
1710 const struct inet_connection_sock *icsk = inet_csk(sp);
1711 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1712 int rx_queue;
1713 int state;
1714
1715 dest = &sp->sk_v6_daddr;
1716 src = &sp->sk_v6_rcv_saddr;
1717 destp = ntohs(inet->inet_dport);
1718 srcp = ntohs(inet->inet_sport);
1719
1720 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1721 timer_active = 1;
1722 timer_expires = icsk->icsk_timeout;
1723 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1724 timer_active = 4;
1725 timer_expires = icsk->icsk_timeout;
1726 } else if (timer_pending(&sp->sk_timer)) {
1727 timer_active = 2;
1728 timer_expires = sp->sk_timer.expires;
1729 } else {
1730 timer_active = 0;
1731 timer_expires = jiffies;
1732 }
1733
1734 state = sk_state_load(sp);
1735 if (state == TCP_LISTEN)
1736 rx_queue = sp->sk_ack_backlog;
1737 else
1738 /* Because we don't lock the socket,
1739 * we might find a transient negative value.
1740 */
1741 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1742
1743 seq_printf(seq,
1744 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1745 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1746 i,
1747 src->s6_addr32[0], src->s6_addr32[1],
1748 src->s6_addr32[2], src->s6_addr32[3], srcp,
1749 dest->s6_addr32[0], dest->s6_addr32[1],
1750 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1751 state,
1752 tp->write_seq - tp->snd_una,
1753 rx_queue,
1754 timer_active,
1755 jiffies_delta_to_clock_t(timer_expires - jiffies),
1756 icsk->icsk_retransmits,
1757 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1758 icsk->icsk_probes_out,
1759 sock_i_ino(sp),
1760 atomic_read(&sp->sk_refcnt), sp,
1761 jiffies_to_clock_t(icsk->icsk_rto),
1762 jiffies_to_clock_t(icsk->icsk_ack.ato),
1763 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1764 tp->snd_cwnd,
1765 state == TCP_LISTEN ?
1766 fastopenq->max_qlen :
1767 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1768 );
1769}
1770
1771static void get_timewait6_sock(struct seq_file *seq,
1772 struct inet_timewait_sock *tw, int i)
1773{
1774 long delta = tw->tw_timer.expires - jiffies;
1775 const struct in6_addr *dest, *src;
1776 __u16 destp, srcp;
1777
1778 dest = &tw->tw_v6_daddr;
1779 src = &tw->tw_v6_rcv_saddr;
1780 destp = ntohs(tw->tw_dport);
1781 srcp = ntohs(tw->tw_sport);
1782
1783 seq_printf(seq,
1784 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1785 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1786 i,
1787 src->s6_addr32[0], src->s6_addr32[1],
1788 src->s6_addr32[2], src->s6_addr32[3], srcp,
1789 dest->s6_addr32[0], dest->s6_addr32[1],
1790 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1791 tw->tw_substate, 0, 0,
1792 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1793 atomic_read(&tw->tw_refcnt), tw);
1794}
1795
1796static int tcp6_seq_show(struct seq_file *seq, void *v)
1797{
1798 struct tcp_iter_state *st;
1799 struct sock *sk = v;
1800
1801 if (v == SEQ_START_TOKEN) {
1802 seq_puts(seq,
1803 " sl "
1804 "local_address "
1805 "remote_address "
1806 "st tx_queue rx_queue tr tm->when retrnsmt"
1807 " uid timeout inode\n");
1808 goto out;
1809 }
1810 st = seq->private;
1811
1812 if (sk->sk_state == TCP_TIME_WAIT)
1813 get_timewait6_sock(seq, v, st->num);
1814 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1815 get_openreq6(seq, v, st->num);
1816 else
1817 get_tcp6_sock(seq, v, st->num);
1818out:
1819 return 0;
1820}
1821
1822static const struct file_operations tcp6_afinfo_seq_fops = {
1823 .owner = THIS_MODULE,
1824 .open = tcp_seq_open,
1825 .read = seq_read,
1826 .llseek = seq_lseek,
1827 .release = seq_release_net
1828};
1829
1830static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1831 .name = "tcp6",
1832 .family = AF_INET6,
1833 .seq_fops = &tcp6_afinfo_seq_fops,
1834 .seq_ops = {
1835 .show = tcp6_seq_show,
1836 },
1837};
1838
1839int __net_init tcp6_proc_init(struct net *net)
1840{
1841 return tcp_proc_register(net, &tcp6_seq_afinfo);
1842}
1843
1844void tcp6_proc_exit(struct net *net)
1845{
1846 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1847}
1848#endif
1849
1850static void tcp_v6_clear_sk(struct sock *sk, int size)
1851{
1852 struct inet_sock *inet = inet_sk(sk);
1853
1854 /* we do not want to clear pinet6 field, because of RCU lookups */
1855 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1856
1857 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1858 memset(&inet->pinet6 + 1, 0, size);
1859}
1860
1861struct proto tcpv6_prot = {
1862 .name = "TCPv6",
1863 .owner = THIS_MODULE,
1864 .close = tcp_close,
1865 .connect = tcp_v6_connect,
1866 .disconnect = tcp_disconnect,
1867 .accept = inet_csk_accept,
1868 .ioctl = tcp_ioctl,
1869 .init = tcp_v6_init_sock,
1870 .destroy = tcp_v6_destroy_sock,
1871 .shutdown = tcp_shutdown,
1872 .setsockopt = tcp_setsockopt,
1873 .getsockopt = tcp_getsockopt,
1874 .recvmsg = tcp_recvmsg,
1875 .sendmsg = tcp_sendmsg,
1876 .sendpage = tcp_sendpage,
1877 .backlog_rcv = tcp_v6_do_rcv,
1878 .release_cb = tcp_release_cb,
1879 .hash = inet6_hash,
1880 .unhash = inet_unhash,
1881 .get_port = inet_csk_get_port,
1882 .enter_memory_pressure = tcp_enter_memory_pressure,
1883 .stream_memory_free = tcp_stream_memory_free,
1884 .sockets_allocated = &tcp_sockets_allocated,
1885 .memory_allocated = &tcp_memory_allocated,
1886 .memory_pressure = &tcp_memory_pressure,
1887 .orphan_count = &tcp_orphan_count,
1888 .sysctl_mem = sysctl_tcp_mem,
1889 .sysctl_wmem = sysctl_tcp_wmem,
1890 .sysctl_rmem = sysctl_tcp_rmem,
1891 .max_header = MAX_TCP_HEADER,
1892 .obj_size = sizeof(struct tcp6_sock),
1893 .slab_flags = SLAB_DESTROY_BY_RCU,
1894 .twsk_prot = &tcp6_timewait_sock_ops,
1895 .rsk_prot = &tcp6_request_sock_ops,
1896 .h.hashinfo = &tcp_hashinfo,
1897 .no_autobind = true,
1898#ifdef CONFIG_COMPAT
1899 .compat_setsockopt = compat_tcp_setsockopt,
1900 .compat_getsockopt = compat_tcp_getsockopt,
1901#endif
1902 .clear_sk = tcp_v6_clear_sk,
1903 .diag_destroy = tcp_abort,
1904};
1905
1906static const struct inet6_protocol tcpv6_protocol = {
1907 .early_demux = tcp_v6_early_demux,
1908 .handler = tcp_v6_rcv,
1909 .err_handler = tcp_v6_err,
1910 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1911};
1912
1913static struct inet_protosw tcpv6_protosw = {
1914 .type = SOCK_STREAM,
1915 .protocol = IPPROTO_TCP,
1916 .prot = &tcpv6_prot,
1917 .ops = &inet6_stream_ops,
1918 .flags = INET_PROTOSW_PERMANENT |
1919 INET_PROTOSW_ICSK,
1920};
1921
1922static int __net_init tcpv6_net_init(struct net *net)
1923{
1924 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1925 SOCK_RAW, IPPROTO_TCP, net);
1926}
1927
1928static void __net_exit tcpv6_net_exit(struct net *net)
1929{
1930 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1931}
1932
1933static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1934{
1935 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1936}
1937
1938static struct pernet_operations tcpv6_net_ops = {
1939 .init = tcpv6_net_init,
1940 .exit = tcpv6_net_exit,
1941 .exit_batch = tcpv6_net_exit_batch,
1942};
1943
1944int __init tcpv6_init(void)
1945{
1946 int ret;
1947
1948 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1949 if (ret)
1950 goto out;
1951
1952 /* register inet6 protocol */
1953 ret = inet6_register_protosw(&tcpv6_protosw);
1954 if (ret)
1955 goto out_tcpv6_protocol;
1956
1957 ret = register_pernet_subsys(&tcpv6_net_ops);
1958 if (ret)
1959 goto out_tcpv6_protosw;
1960out:
1961 return ret;
1962
1963out_tcpv6_protosw:
1964 inet6_unregister_protosw(&tcpv6_protosw);
1965out_tcpv6_protocol:
1966 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1967 goto out;
1968}
1969
1970void tcpv6_exit(void)
1971{
1972 unregister_pernet_subsys(&tcpv6_net_ops);
1973 inet6_unregister_protosw(&tcpv6_protosw);
1974 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1975}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * TCP over IPv6
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
8 *
9 * Based on:
10 * linux/net/ipv4/tcp.c
11 * linux/net/ipv4/tcp_input.c
12 * linux/net/ipv4/tcp_output.c
13 *
14 * Fixes:
15 * Hideaki YOSHIFUJI : sin6_scope_id support
16 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
17 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
18 * a single port at the same time.
19 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
20 */
21
22#include <linux/bottom_half.h>
23#include <linux/module.h>
24#include <linux/errno.h>
25#include <linux/types.h>
26#include <linux/socket.h>
27#include <linux/sockios.h>
28#include <linux/net.h>
29#include <linux/jiffies.h>
30#include <linux/in.h>
31#include <linux/in6.h>
32#include <linux/netdevice.h>
33#include <linux/init.h>
34#include <linux/jhash.h>
35#include <linux/ipsec.h>
36#include <linux/times.h>
37#include <linux/slab.h>
38#include <linux/uaccess.h>
39#include <linux/ipv6.h>
40#include <linux/icmpv6.h>
41#include <linux/random.h>
42#include <linux/indirect_call_wrapper.h>
43
44#include <net/tcp.h>
45#include <net/ndisc.h>
46#include <net/inet6_hashtables.h>
47#include <net/inet6_connection_sock.h>
48#include <net/ipv6.h>
49#include <net/transp_v6.h>
50#include <net/addrconf.h>
51#include <net/ip6_route.h>
52#include <net/ip6_checksum.h>
53#include <net/inet_ecn.h>
54#include <net/protocol.h>
55#include <net/xfrm.h>
56#include <net/snmp.h>
57#include <net/dsfield.h>
58#include <net/timewait_sock.h>
59#include <net/inet_common.h>
60#include <net/secure_seq.h>
61#include <net/busy_poll.h>
62
63#include <linux/proc_fs.h>
64#include <linux/seq_file.h>
65
66#include <crypto/hash.h>
67#include <linux/scatterlist.h>
68
69#include <trace/events/tcp.h>
70
71static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
72static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
73 struct request_sock *req);
74
75static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
76
77static const struct inet_connection_sock_af_ops ipv6_mapped;
78const struct inet_connection_sock_af_ops ipv6_specific;
79#ifdef CONFIG_TCP_MD5SIG
80static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
81static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
82#else
83static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
84 const struct in6_addr *addr,
85 int l3index)
86{
87 return NULL;
88}
89#endif
90
91/* Helper returning the inet6 address from a given tcp socket.
92 * It can be used in TCP stack instead of inet6_sk(sk).
93 * This avoids a dereference and allow compiler optimizations.
94 * It is a specialized version of inet6_sk_generic().
95 */
96static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk)
97{
98 unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo);
99
100 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
101}
102
103static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
104{
105 struct dst_entry *dst = skb_dst(skb);
106
107 if (dst && dst_hold_safe(dst)) {
108 const struct rt6_info *rt = (const struct rt6_info *)dst;
109
110 sk->sk_rx_dst = dst;
111 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
112 tcp_inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
113 }
114}
115
116static u32 tcp_v6_init_seq(const struct sk_buff *skb)
117{
118 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
119 ipv6_hdr(skb)->saddr.s6_addr32,
120 tcp_hdr(skb)->dest,
121 tcp_hdr(skb)->source);
122}
123
124static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
125{
126 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
127 ipv6_hdr(skb)->saddr.s6_addr32);
128}
129
130static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
131 int addr_len)
132{
133 /* This check is replicated from tcp_v6_connect() and intended to
134 * prevent BPF program called below from accessing bytes that are out
135 * of the bound specified by user in addr_len.
136 */
137 if (addr_len < SIN6_LEN_RFC2133)
138 return -EINVAL;
139
140 sock_owned_by_me(sk);
141
142 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr);
143}
144
145static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
146 int addr_len)
147{
148 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
149 struct inet_sock *inet = inet_sk(sk);
150 struct inet_connection_sock *icsk = inet_csk(sk);
151 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
152 struct tcp_sock *tp = tcp_sk(sk);
153 struct in6_addr *saddr = NULL, *final_p, final;
154 struct ipv6_txoptions *opt;
155 struct flowi6 fl6;
156 struct dst_entry *dst;
157 int addr_type;
158 int err;
159 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
160
161 if (addr_len < SIN6_LEN_RFC2133)
162 return -EINVAL;
163
164 if (usin->sin6_family != AF_INET6)
165 return -EAFNOSUPPORT;
166
167 memset(&fl6, 0, sizeof(fl6));
168
169 if (np->sndflow) {
170 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
171 IP6_ECN_flow_init(fl6.flowlabel);
172 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
173 struct ip6_flowlabel *flowlabel;
174 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
175 if (IS_ERR(flowlabel))
176 return -EINVAL;
177 fl6_sock_release(flowlabel);
178 }
179 }
180
181 /*
182 * connect() to INADDR_ANY means loopback (BSD'ism).
183 */
184
185 if (ipv6_addr_any(&usin->sin6_addr)) {
186 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
187 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
188 &usin->sin6_addr);
189 else
190 usin->sin6_addr = in6addr_loopback;
191 }
192
193 addr_type = ipv6_addr_type(&usin->sin6_addr);
194
195 if (addr_type & IPV6_ADDR_MULTICAST)
196 return -ENETUNREACH;
197
198 if (addr_type&IPV6_ADDR_LINKLOCAL) {
199 if (addr_len >= sizeof(struct sockaddr_in6) &&
200 usin->sin6_scope_id) {
201 /* If interface is set while binding, indices
202 * must coincide.
203 */
204 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
205 return -EINVAL;
206
207 sk->sk_bound_dev_if = usin->sin6_scope_id;
208 }
209
210 /* Connect to link-local address requires an interface */
211 if (!sk->sk_bound_dev_if)
212 return -EINVAL;
213 }
214
215 if (tp->rx_opt.ts_recent_stamp &&
216 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
217 tp->rx_opt.ts_recent = 0;
218 tp->rx_opt.ts_recent_stamp = 0;
219 WRITE_ONCE(tp->write_seq, 0);
220 }
221
222 sk->sk_v6_daddr = usin->sin6_addr;
223 np->flow_label = fl6.flowlabel;
224
225 /*
226 * TCP over IPv4
227 */
228
229 if (addr_type & IPV6_ADDR_MAPPED) {
230 u32 exthdrlen = icsk->icsk_ext_hdr_len;
231 struct sockaddr_in sin;
232
233 if (__ipv6_only_sock(sk))
234 return -ENETUNREACH;
235
236 sin.sin_family = AF_INET;
237 sin.sin_port = usin->sin6_port;
238 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
239
240 icsk->icsk_af_ops = &ipv6_mapped;
241 if (sk_is_mptcp(sk))
242 mptcpv6_handle_mapped(sk, true);
243 sk->sk_backlog_rcv = tcp_v4_do_rcv;
244#ifdef CONFIG_TCP_MD5SIG
245 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
246#endif
247
248 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
249
250 if (err) {
251 icsk->icsk_ext_hdr_len = exthdrlen;
252 icsk->icsk_af_ops = &ipv6_specific;
253 if (sk_is_mptcp(sk))
254 mptcpv6_handle_mapped(sk, false);
255 sk->sk_backlog_rcv = tcp_v6_do_rcv;
256#ifdef CONFIG_TCP_MD5SIG
257 tp->af_specific = &tcp_sock_ipv6_specific;
258#endif
259 goto failure;
260 }
261 np->saddr = sk->sk_v6_rcv_saddr;
262
263 return err;
264 }
265
266 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
267 saddr = &sk->sk_v6_rcv_saddr;
268
269 fl6.flowi6_proto = IPPROTO_TCP;
270 fl6.daddr = sk->sk_v6_daddr;
271 fl6.saddr = saddr ? *saddr : np->saddr;
272 fl6.flowi6_oif = sk->sk_bound_dev_if;
273 fl6.flowi6_mark = sk->sk_mark;
274 fl6.fl6_dport = usin->sin6_port;
275 fl6.fl6_sport = inet->inet_sport;
276 fl6.flowi6_uid = sk->sk_uid;
277
278 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
279 final_p = fl6_update_dst(&fl6, opt, &final);
280
281 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
282
283 dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
284 if (IS_ERR(dst)) {
285 err = PTR_ERR(dst);
286 goto failure;
287 }
288
289 if (!saddr) {
290 saddr = &fl6.saddr;
291 sk->sk_v6_rcv_saddr = *saddr;
292 }
293
294 /* set the source address */
295 np->saddr = *saddr;
296 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
297
298 sk->sk_gso_type = SKB_GSO_TCPV6;
299 ip6_dst_store(sk, dst, NULL, NULL);
300
301 icsk->icsk_ext_hdr_len = 0;
302 if (opt)
303 icsk->icsk_ext_hdr_len = opt->opt_flen +
304 opt->opt_nflen;
305
306 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
307
308 inet->inet_dport = usin->sin6_port;
309
310 tcp_set_state(sk, TCP_SYN_SENT);
311 err = inet6_hash_connect(tcp_death_row, sk);
312 if (err)
313 goto late_failure;
314
315 sk_set_txhash(sk);
316
317 if (likely(!tp->repair)) {
318 if (!tp->write_seq)
319 WRITE_ONCE(tp->write_seq,
320 secure_tcpv6_seq(np->saddr.s6_addr32,
321 sk->sk_v6_daddr.s6_addr32,
322 inet->inet_sport,
323 inet->inet_dport));
324 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
325 np->saddr.s6_addr32,
326 sk->sk_v6_daddr.s6_addr32);
327 }
328
329 if (tcp_fastopen_defer_connect(sk, &err))
330 return err;
331 if (err)
332 goto late_failure;
333
334 err = tcp_connect(sk);
335 if (err)
336 goto late_failure;
337
338 return 0;
339
340late_failure:
341 tcp_set_state(sk, TCP_CLOSE);
342failure:
343 inet->inet_dport = 0;
344 sk->sk_route_caps = 0;
345 return err;
346}
347
348static void tcp_v6_mtu_reduced(struct sock *sk)
349{
350 struct dst_entry *dst;
351 u32 mtu;
352
353 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
354 return;
355
356 mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
357
358 /* Drop requests trying to increase our current mss.
359 * Check done in __ip6_rt_update_pmtu() is too late.
360 */
361 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache)
362 return;
363
364 dst = inet6_csk_update_pmtu(sk, mtu);
365 if (!dst)
366 return;
367
368 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
369 tcp_sync_mss(sk, dst_mtu(dst));
370 tcp_simple_retransmit(sk);
371 }
372}
373
374static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
375 u8 type, u8 code, int offset, __be32 info)
376{
377 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
378 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
379 struct net *net = dev_net(skb->dev);
380 struct request_sock *fastopen;
381 struct ipv6_pinfo *np;
382 struct tcp_sock *tp;
383 __u32 seq, snd_una;
384 struct sock *sk;
385 bool fatal;
386 int err;
387
388 sk = __inet6_lookup_established(net, &tcp_hashinfo,
389 &hdr->daddr, th->dest,
390 &hdr->saddr, ntohs(th->source),
391 skb->dev->ifindex, inet6_sdif(skb));
392
393 if (!sk) {
394 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
395 ICMP6_MIB_INERRORS);
396 return -ENOENT;
397 }
398
399 if (sk->sk_state == TCP_TIME_WAIT) {
400 inet_twsk_put(inet_twsk(sk));
401 return 0;
402 }
403 seq = ntohl(th->seq);
404 fatal = icmpv6_err_convert(type, code, &err);
405 if (sk->sk_state == TCP_NEW_SYN_RECV) {
406 tcp_req_err(sk, seq, fatal);
407 return 0;
408 }
409
410 bh_lock_sock(sk);
411 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
412 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
413
414 if (sk->sk_state == TCP_CLOSE)
415 goto out;
416
417 if (ipv6_hdr(skb)->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
418 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
419 goto out;
420 }
421
422 tp = tcp_sk(sk);
423 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
424 fastopen = rcu_dereference(tp->fastopen_rsk);
425 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
426 if (sk->sk_state != TCP_LISTEN &&
427 !between(seq, snd_una, tp->snd_nxt)) {
428 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
429 goto out;
430 }
431
432 np = tcp_inet6_sk(sk);
433
434 if (type == NDISC_REDIRECT) {
435 if (!sock_owned_by_user(sk)) {
436 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
437
438 if (dst)
439 dst->ops->redirect(dst, sk, skb);
440 }
441 goto out;
442 }
443
444 if (type == ICMPV6_PKT_TOOBIG) {
445 u32 mtu = ntohl(info);
446
447 /* We are not interested in TCP_LISTEN and open_requests
448 * (SYN-ACKs send out by Linux are always <576bytes so
449 * they should go through unfragmented).
450 */
451 if (sk->sk_state == TCP_LISTEN)
452 goto out;
453
454 if (!ip6_sk_accept_pmtu(sk))
455 goto out;
456
457 if (mtu < IPV6_MIN_MTU)
458 goto out;
459
460 WRITE_ONCE(tp->mtu_info, mtu);
461
462 if (!sock_owned_by_user(sk))
463 tcp_v6_mtu_reduced(sk);
464 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
465 &sk->sk_tsq_flags))
466 sock_hold(sk);
467 goto out;
468 }
469
470
471 /* Might be for an request_sock */
472 switch (sk->sk_state) {
473 case TCP_SYN_SENT:
474 case TCP_SYN_RECV:
475 /* Only in fast or simultaneous open. If a fast open socket is
476 * already accepted it is treated as a connected one below.
477 */
478 if (fastopen && !fastopen->sk)
479 break;
480
481 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
482
483 if (!sock_owned_by_user(sk)) {
484 sk->sk_err = err;
485 sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
486
487 tcp_done(sk);
488 } else
489 sk->sk_err_soft = err;
490 goto out;
491 case TCP_LISTEN:
492 break;
493 default:
494 /* check if this ICMP message allows revert of backoff.
495 * (see RFC 6069)
496 */
497 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
498 code == ICMPV6_NOROUTE)
499 tcp_ld_RTO_revert(sk, seq);
500 }
501
502 if (!sock_owned_by_user(sk) && np->recverr) {
503 sk->sk_err = err;
504 sk_error_report(sk);
505 } else
506 sk->sk_err_soft = err;
507
508out:
509 bh_unlock_sock(sk);
510 sock_put(sk);
511 return 0;
512}
513
514
515static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
516 struct flowi *fl,
517 struct request_sock *req,
518 struct tcp_fastopen_cookie *foc,
519 enum tcp_synack_type synack_type,
520 struct sk_buff *syn_skb)
521{
522 struct inet_request_sock *ireq = inet_rsk(req);
523 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
524 struct ipv6_txoptions *opt;
525 struct flowi6 *fl6 = &fl->u.ip6;
526 struct sk_buff *skb;
527 int err = -ENOMEM;
528 u8 tclass;
529
530 /* First, grab a route. */
531 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
532 IPPROTO_TCP)) == NULL)
533 goto done;
534
535 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
536
537 if (skb) {
538 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
539 &ireq->ir_v6_rmt_addr);
540
541 fl6->daddr = ireq->ir_v6_rmt_addr;
542 if (np->repflow && ireq->pktopts)
543 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
544
545 tclass = sock_net(sk)->ipv4.sysctl_tcp_reflect_tos ?
546 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
547 (np->tclass & INET_ECN_MASK) :
548 np->tclass;
549
550 if (!INET_ECN_is_capable(tclass) &&
551 tcp_bpf_ca_needs_ecn((struct sock *)req))
552 tclass |= INET_ECN_ECT_0;
553
554 rcu_read_lock();
555 opt = ireq->ipv6_opt;
556 if (!opt)
557 opt = rcu_dereference(np->opt);
558 err = ip6_xmit(sk, skb, fl6, skb->mark ? : sk->sk_mark, opt,
559 tclass, sk->sk_priority);
560 rcu_read_unlock();
561 err = net_xmit_eval(err);
562 }
563
564done:
565 return err;
566}
567
568
569static void tcp_v6_reqsk_destructor(struct request_sock *req)
570{
571 kfree(inet_rsk(req)->ipv6_opt);
572 kfree_skb(inet_rsk(req)->pktopts);
573}
574
575#ifdef CONFIG_TCP_MD5SIG
576static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
577 const struct in6_addr *addr,
578 int l3index)
579{
580 return tcp_md5_do_lookup(sk, l3index,
581 (union tcp_md5_addr *)addr, AF_INET6);
582}
583
584static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
585 const struct sock *addr_sk)
586{
587 int l3index;
588
589 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
590 addr_sk->sk_bound_dev_if);
591 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
592 l3index);
593}
594
595static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
596 sockptr_t optval, int optlen)
597{
598 struct tcp_md5sig cmd;
599 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
600 int l3index = 0;
601 u8 prefixlen;
602
603 if (optlen < sizeof(cmd))
604 return -EINVAL;
605
606 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
607 return -EFAULT;
608
609 if (sin6->sin6_family != AF_INET6)
610 return -EINVAL;
611
612 if (optname == TCP_MD5SIG_EXT &&
613 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
614 prefixlen = cmd.tcpm_prefixlen;
615 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
616 prefixlen > 32))
617 return -EINVAL;
618 } else {
619 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
620 }
621
622 if (optname == TCP_MD5SIG_EXT &&
623 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
624 struct net_device *dev;
625
626 rcu_read_lock();
627 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
628 if (dev && netif_is_l3_master(dev))
629 l3index = dev->ifindex;
630 rcu_read_unlock();
631
632 /* ok to reference set/not set outside of rcu;
633 * right now device MUST be an L3 master
634 */
635 if (!dev || !l3index)
636 return -EINVAL;
637 }
638
639 if (!cmd.tcpm_keylen) {
640 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
641 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
642 AF_INET, prefixlen,
643 l3index);
644 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
645 AF_INET6, prefixlen, l3index);
646 }
647
648 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
649 return -EINVAL;
650
651 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
652 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
653 AF_INET, prefixlen, l3index,
654 cmd.tcpm_key, cmd.tcpm_keylen,
655 GFP_KERNEL);
656
657 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
658 AF_INET6, prefixlen, l3index,
659 cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
660}
661
662static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
663 const struct in6_addr *daddr,
664 const struct in6_addr *saddr,
665 const struct tcphdr *th, int nbytes)
666{
667 struct tcp6_pseudohdr *bp;
668 struct scatterlist sg;
669 struct tcphdr *_th;
670
671 bp = hp->scratch;
672 /* 1. TCP pseudo-header (RFC2460) */
673 bp->saddr = *saddr;
674 bp->daddr = *daddr;
675 bp->protocol = cpu_to_be32(IPPROTO_TCP);
676 bp->len = cpu_to_be32(nbytes);
677
678 _th = (struct tcphdr *)(bp + 1);
679 memcpy(_th, th, sizeof(*th));
680 _th->check = 0;
681
682 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
683 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
684 sizeof(*bp) + sizeof(*th));
685 return crypto_ahash_update(hp->md5_req);
686}
687
688static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
689 const struct in6_addr *daddr, struct in6_addr *saddr,
690 const struct tcphdr *th)
691{
692 struct tcp_md5sig_pool *hp;
693 struct ahash_request *req;
694
695 hp = tcp_get_md5sig_pool();
696 if (!hp)
697 goto clear_hash_noput;
698 req = hp->md5_req;
699
700 if (crypto_ahash_init(req))
701 goto clear_hash;
702 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
703 goto clear_hash;
704 if (tcp_md5_hash_key(hp, key))
705 goto clear_hash;
706 ahash_request_set_crypt(req, NULL, md5_hash, 0);
707 if (crypto_ahash_final(req))
708 goto clear_hash;
709
710 tcp_put_md5sig_pool();
711 return 0;
712
713clear_hash:
714 tcp_put_md5sig_pool();
715clear_hash_noput:
716 memset(md5_hash, 0, 16);
717 return 1;
718}
719
720static int tcp_v6_md5_hash_skb(char *md5_hash,
721 const struct tcp_md5sig_key *key,
722 const struct sock *sk,
723 const struct sk_buff *skb)
724{
725 const struct in6_addr *saddr, *daddr;
726 struct tcp_md5sig_pool *hp;
727 struct ahash_request *req;
728 const struct tcphdr *th = tcp_hdr(skb);
729
730 if (sk) { /* valid for establish/request sockets */
731 saddr = &sk->sk_v6_rcv_saddr;
732 daddr = &sk->sk_v6_daddr;
733 } else {
734 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
735 saddr = &ip6h->saddr;
736 daddr = &ip6h->daddr;
737 }
738
739 hp = tcp_get_md5sig_pool();
740 if (!hp)
741 goto clear_hash_noput;
742 req = hp->md5_req;
743
744 if (crypto_ahash_init(req))
745 goto clear_hash;
746
747 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
748 goto clear_hash;
749 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
750 goto clear_hash;
751 if (tcp_md5_hash_key(hp, key))
752 goto clear_hash;
753 ahash_request_set_crypt(req, NULL, md5_hash, 0);
754 if (crypto_ahash_final(req))
755 goto clear_hash;
756
757 tcp_put_md5sig_pool();
758 return 0;
759
760clear_hash:
761 tcp_put_md5sig_pool();
762clear_hash_noput:
763 memset(md5_hash, 0, 16);
764 return 1;
765}
766
767#endif
768
769static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
770 const struct sk_buff *skb,
771 int dif, int sdif)
772{
773#ifdef CONFIG_TCP_MD5SIG
774 const __u8 *hash_location = NULL;
775 struct tcp_md5sig_key *hash_expected;
776 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
777 const struct tcphdr *th = tcp_hdr(skb);
778 int genhash, l3index;
779 u8 newhash[16];
780
781 /* sdif set, means packet ingressed via a device
782 * in an L3 domain and dif is set to the l3mdev
783 */
784 l3index = sdif ? dif : 0;
785
786 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr, l3index);
787 hash_location = tcp_parse_md5sig_option(th);
788
789 /* We've parsed the options - do we have a hash? */
790 if (!hash_expected && !hash_location)
791 return false;
792
793 if (hash_expected && !hash_location) {
794 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
795 return true;
796 }
797
798 if (!hash_expected && hash_location) {
799 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
800 return true;
801 }
802
803 /* check the signature */
804 genhash = tcp_v6_md5_hash_skb(newhash,
805 hash_expected,
806 NULL, skb);
807
808 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
809 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
810 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u L3 index %d\n",
811 genhash ? "failed" : "mismatch",
812 &ip6h->saddr, ntohs(th->source),
813 &ip6h->daddr, ntohs(th->dest), l3index);
814 return true;
815 }
816#endif
817 return false;
818}
819
820static void tcp_v6_init_req(struct request_sock *req,
821 const struct sock *sk_listener,
822 struct sk_buff *skb)
823{
824 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
825 struct inet_request_sock *ireq = inet_rsk(req);
826 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
827
828 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
829 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
830
831 /* So that link locals have meaning */
832 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
833 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
834 ireq->ir_iif = tcp_v6_iif(skb);
835
836 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
837 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
838 np->rxopt.bits.rxinfo ||
839 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
840 np->rxopt.bits.rxohlim || np->repflow)) {
841 refcount_inc(&skb->users);
842 ireq->pktopts = skb;
843 }
844}
845
846static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
847 struct sk_buff *skb,
848 struct flowi *fl,
849 struct request_sock *req)
850{
851 tcp_v6_init_req(req, sk, skb);
852
853 if (security_inet_conn_request(sk, skb, req))
854 return NULL;
855
856 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
857}
858
859struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
860 .family = AF_INET6,
861 .obj_size = sizeof(struct tcp6_request_sock),
862 .rtx_syn_ack = tcp_rtx_synack,
863 .send_ack = tcp_v6_reqsk_send_ack,
864 .destructor = tcp_v6_reqsk_destructor,
865 .send_reset = tcp_v6_send_reset,
866 .syn_ack_timeout = tcp_syn_ack_timeout,
867};
868
869const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
870 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
871 sizeof(struct ipv6hdr),
872#ifdef CONFIG_TCP_MD5SIG
873 .req_md5_lookup = tcp_v6_md5_lookup,
874 .calc_md5_hash = tcp_v6_md5_hash_skb,
875#endif
876#ifdef CONFIG_SYN_COOKIES
877 .cookie_init_seq = cookie_v6_init_sequence,
878#endif
879 .route_req = tcp_v6_route_req,
880 .init_seq = tcp_v6_init_seq,
881 .init_ts_off = tcp_v6_init_ts_off,
882 .send_synack = tcp_v6_send_synack,
883};
884
885static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
886 u32 ack, u32 win, u32 tsval, u32 tsecr,
887 int oif, struct tcp_md5sig_key *key, int rst,
888 u8 tclass, __be32 label, u32 priority)
889{
890 const struct tcphdr *th = tcp_hdr(skb);
891 struct tcphdr *t1;
892 struct sk_buff *buff;
893 struct flowi6 fl6;
894 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
895 struct sock *ctl_sk = net->ipv6.tcp_sk;
896 unsigned int tot_len = sizeof(struct tcphdr);
897 __be32 mrst = 0, *topt;
898 struct dst_entry *dst;
899 __u32 mark = 0;
900
901 if (tsecr)
902 tot_len += TCPOLEN_TSTAMP_ALIGNED;
903#ifdef CONFIG_TCP_MD5SIG
904 if (key)
905 tot_len += TCPOLEN_MD5SIG_ALIGNED;
906#endif
907
908#ifdef CONFIG_MPTCP
909 if (rst && !key) {
910 mrst = mptcp_reset_option(skb);
911
912 if (mrst)
913 tot_len += sizeof(__be32);
914 }
915#endif
916
917 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
918 GFP_ATOMIC);
919 if (!buff)
920 return;
921
922 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
923
924 t1 = skb_push(buff, tot_len);
925 skb_reset_transport_header(buff);
926
927 /* Swap the send and the receive. */
928 memset(t1, 0, sizeof(*t1));
929 t1->dest = th->source;
930 t1->source = th->dest;
931 t1->doff = tot_len / 4;
932 t1->seq = htonl(seq);
933 t1->ack_seq = htonl(ack);
934 t1->ack = !rst || !th->ack;
935 t1->rst = rst;
936 t1->window = htons(win);
937
938 topt = (__be32 *)(t1 + 1);
939
940 if (tsecr) {
941 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
942 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
943 *topt++ = htonl(tsval);
944 *topt++ = htonl(tsecr);
945 }
946
947 if (mrst)
948 *topt++ = mrst;
949
950#ifdef CONFIG_TCP_MD5SIG
951 if (key) {
952 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
953 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
954 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
955 &ipv6_hdr(skb)->saddr,
956 &ipv6_hdr(skb)->daddr, t1);
957 }
958#endif
959
960 memset(&fl6, 0, sizeof(fl6));
961 fl6.daddr = ipv6_hdr(skb)->saddr;
962 fl6.saddr = ipv6_hdr(skb)->daddr;
963 fl6.flowlabel = label;
964
965 buff->ip_summed = CHECKSUM_PARTIAL;
966 buff->csum = 0;
967
968 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
969
970 fl6.flowi6_proto = IPPROTO_TCP;
971 if (rt6_need_strict(&fl6.daddr) && !oif)
972 fl6.flowi6_oif = tcp_v6_iif(skb);
973 else {
974 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
975 oif = skb->skb_iif;
976
977 fl6.flowi6_oif = oif;
978 }
979
980 if (sk) {
981 if (sk->sk_state == TCP_TIME_WAIT) {
982 mark = inet_twsk(sk)->tw_mark;
983 /* autoflowlabel relies on buff->hash */
984 skb_set_hash(buff, inet_twsk(sk)->tw_txhash,
985 PKT_HASH_TYPE_L4);
986 } else {
987 mark = sk->sk_mark;
988 }
989 buff->tstamp = tcp_transmit_time(sk);
990 }
991 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
992 fl6.fl6_dport = t1->dest;
993 fl6.fl6_sport = t1->source;
994 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
995 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
996
997 /* Pass a socket to ip6_dst_lookup either it is for RST
998 * Underlying function will use this to retrieve the network
999 * namespace
1000 */
1001 dst = ip6_dst_lookup_flow(sock_net(ctl_sk), ctl_sk, &fl6, NULL);
1002 if (!IS_ERR(dst)) {
1003 skb_dst_set(buff, dst);
1004 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
1005 tclass & ~INET_ECN_MASK, priority);
1006 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
1007 if (rst)
1008 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
1009 return;
1010 }
1011
1012 kfree_skb(buff);
1013}
1014
1015static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
1016{
1017 const struct tcphdr *th = tcp_hdr(skb);
1018 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
1019 u32 seq = 0, ack_seq = 0;
1020 struct tcp_md5sig_key *key = NULL;
1021#ifdef CONFIG_TCP_MD5SIG
1022 const __u8 *hash_location = NULL;
1023 unsigned char newhash[16];
1024 int genhash;
1025 struct sock *sk1 = NULL;
1026#endif
1027 __be32 label = 0;
1028 u32 priority = 0;
1029 struct net *net;
1030 int oif = 0;
1031
1032 if (th->rst)
1033 return;
1034
1035 /* If sk not NULL, it means we did a successful lookup and incoming
1036 * route had to be correct. prequeue might have dropped our dst.
1037 */
1038 if (!sk && !ipv6_unicast_destination(skb))
1039 return;
1040
1041 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
1042#ifdef CONFIG_TCP_MD5SIG
1043 rcu_read_lock();
1044 hash_location = tcp_parse_md5sig_option(th);
1045 if (sk && sk_fullsock(sk)) {
1046 int l3index;
1047
1048 /* sdif set, means packet ingressed via a device
1049 * in an L3 domain and inet_iif is set to it.
1050 */
1051 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1052 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1053 } else if (hash_location) {
1054 int dif = tcp_v6_iif_l3_slave(skb);
1055 int sdif = tcp_v6_sdif(skb);
1056 int l3index;
1057
1058 /*
1059 * active side is lost. Try to find listening socket through
1060 * source port, and then find md5 key through listening socket.
1061 * we are not loose security here:
1062 * Incoming packet is checked with md5 hash with finding key,
1063 * no RST generated if md5 hash doesn't match.
1064 */
1065 sk1 = inet6_lookup_listener(net,
1066 &tcp_hashinfo, NULL, 0,
1067 &ipv6h->saddr,
1068 th->source, &ipv6h->daddr,
1069 ntohs(th->source), dif, sdif);
1070 if (!sk1)
1071 goto out;
1072
1073 /* sdif set, means packet ingressed via a device
1074 * in an L3 domain and dif is set to it.
1075 */
1076 l3index = tcp_v6_sdif(skb) ? dif : 0;
1077
1078 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1079 if (!key)
1080 goto out;
1081
1082 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
1083 if (genhash || memcmp(hash_location, newhash, 16) != 0)
1084 goto out;
1085 }
1086#endif
1087
1088 if (th->ack)
1089 seq = ntohl(th->ack_seq);
1090 else
1091 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1092 (th->doff << 2);
1093
1094 if (sk) {
1095 oif = sk->sk_bound_dev_if;
1096 if (sk_fullsock(sk)) {
1097 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1098
1099 trace_tcp_send_reset(sk, skb);
1100 if (np->repflow)
1101 label = ip6_flowlabel(ipv6h);
1102 priority = sk->sk_priority;
1103 }
1104 if (sk->sk_state == TCP_TIME_WAIT) {
1105 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1106 priority = inet_twsk(sk)->tw_priority;
1107 }
1108 } else {
1109 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1110 label = ip6_flowlabel(ipv6h);
1111 }
1112
1113 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1,
1114 ipv6_get_dsfield(ipv6h), label, priority);
1115
1116#ifdef CONFIG_TCP_MD5SIG
1117out:
1118 rcu_read_unlock();
1119#endif
1120}
1121
1122static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1123 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1124 struct tcp_md5sig_key *key, u8 tclass,
1125 __be32 label, u32 priority)
1126{
1127 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1128 tclass, label, priority);
1129}
1130
1131static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1132{
1133 struct inet_timewait_sock *tw = inet_twsk(sk);
1134 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1135
1136 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1137 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1138 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1139 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1140 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority);
1141
1142 inet_twsk_put(tw);
1143}
1144
1145static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1146 struct request_sock *req)
1147{
1148 int l3index;
1149
1150 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1151
1152 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1153 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1154 */
1155 /* RFC 7323 2.3
1156 * The window field (SEG.WND) of every outgoing segment, with the
1157 * exception of <SYN> segments, MUST be right-shifted by
1158 * Rcv.Wind.Shift bits:
1159 */
1160 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1161 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1162 tcp_rsk(req)->rcv_nxt,
1163 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1164 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1165 req->ts_recent, sk->sk_bound_dev_if,
1166 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
1167 ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority);
1168}
1169
1170
1171static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1172{
1173#ifdef CONFIG_SYN_COOKIES
1174 const struct tcphdr *th = tcp_hdr(skb);
1175
1176 if (!th->syn)
1177 sk = cookie_v6_check(sk, skb);
1178#endif
1179 return sk;
1180}
1181
1182u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1183 struct tcphdr *th, u32 *cookie)
1184{
1185 u16 mss = 0;
1186#ifdef CONFIG_SYN_COOKIES
1187 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1188 &tcp_request_sock_ipv6_ops, sk, th);
1189 if (mss) {
1190 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1191 tcp_synq_overflow(sk);
1192 }
1193#endif
1194 return mss;
1195}
1196
1197static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1198{
1199 if (skb->protocol == htons(ETH_P_IP))
1200 return tcp_v4_conn_request(sk, skb);
1201
1202 if (!ipv6_unicast_destination(skb))
1203 goto drop;
1204
1205 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1206 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1207 return 0;
1208 }
1209
1210 return tcp_conn_request(&tcp6_request_sock_ops,
1211 &tcp_request_sock_ipv6_ops, sk, skb);
1212
1213drop:
1214 tcp_listendrop(sk);
1215 return 0; /* don't send reset */
1216}
1217
1218static void tcp_v6_restore_cb(struct sk_buff *skb)
1219{
1220 /* We need to move header back to the beginning if xfrm6_policy_check()
1221 * and tcp_v6_fill_cb() are going to be called again.
1222 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1223 */
1224 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1225 sizeof(struct inet6_skb_parm));
1226}
1227
1228static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1229 struct request_sock *req,
1230 struct dst_entry *dst,
1231 struct request_sock *req_unhash,
1232 bool *own_req)
1233{
1234 struct inet_request_sock *ireq;
1235 struct ipv6_pinfo *newnp;
1236 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1237 struct ipv6_txoptions *opt;
1238 struct inet_sock *newinet;
1239 bool found_dup_sk = false;
1240 struct tcp_sock *newtp;
1241 struct sock *newsk;
1242#ifdef CONFIG_TCP_MD5SIG
1243 struct tcp_md5sig_key *key;
1244 int l3index;
1245#endif
1246 struct flowi6 fl6;
1247
1248 if (skb->protocol == htons(ETH_P_IP)) {
1249 /*
1250 * v6 mapped
1251 */
1252
1253 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1254 req_unhash, own_req);
1255
1256 if (!newsk)
1257 return NULL;
1258
1259 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1260
1261 newinet = inet_sk(newsk);
1262 newnp = tcp_inet6_sk(newsk);
1263 newtp = tcp_sk(newsk);
1264
1265 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1266
1267 newnp->saddr = newsk->sk_v6_rcv_saddr;
1268
1269 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1270 if (sk_is_mptcp(newsk))
1271 mptcpv6_handle_mapped(newsk, true);
1272 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1273#ifdef CONFIG_TCP_MD5SIG
1274 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1275#endif
1276
1277 newnp->ipv6_mc_list = NULL;
1278 newnp->ipv6_ac_list = NULL;
1279 newnp->ipv6_fl_list = NULL;
1280 newnp->pktoptions = NULL;
1281 newnp->opt = NULL;
1282 newnp->mcast_oif = inet_iif(skb);
1283 newnp->mcast_hops = ip_hdr(skb)->ttl;
1284 newnp->rcv_flowinfo = 0;
1285 if (np->repflow)
1286 newnp->flow_label = 0;
1287
1288 /*
1289 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1290 * here, tcp_create_openreq_child now does this for us, see the comment in
1291 * that function for the gory details. -acme
1292 */
1293
1294 /* It is tricky place. Until this moment IPv4 tcp
1295 worked with IPv6 icsk.icsk_af_ops.
1296 Sync it now.
1297 */
1298 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1299
1300 return newsk;
1301 }
1302
1303 ireq = inet_rsk(req);
1304
1305 if (sk_acceptq_is_full(sk))
1306 goto out_overflow;
1307
1308 if (!dst) {
1309 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1310 if (!dst)
1311 goto out;
1312 }
1313
1314 newsk = tcp_create_openreq_child(sk, req, skb);
1315 if (!newsk)
1316 goto out_nonewsk;
1317
1318 /*
1319 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1320 * count here, tcp_create_openreq_child now does this for us, see the
1321 * comment in that function for the gory details. -acme
1322 */
1323
1324 newsk->sk_gso_type = SKB_GSO_TCPV6;
1325 ip6_dst_store(newsk, dst, NULL, NULL);
1326 inet6_sk_rx_dst_set(newsk, skb);
1327
1328 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1329
1330 newtp = tcp_sk(newsk);
1331 newinet = inet_sk(newsk);
1332 newnp = tcp_inet6_sk(newsk);
1333
1334 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1335
1336 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1337 newnp->saddr = ireq->ir_v6_loc_addr;
1338 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1339 newsk->sk_bound_dev_if = ireq->ir_iif;
1340
1341 /* Now IPv6 options...
1342
1343 First: no IPv4 options.
1344 */
1345 newinet->inet_opt = NULL;
1346 newnp->ipv6_mc_list = NULL;
1347 newnp->ipv6_ac_list = NULL;
1348 newnp->ipv6_fl_list = NULL;
1349
1350 /* Clone RX bits */
1351 newnp->rxopt.all = np->rxopt.all;
1352
1353 newnp->pktoptions = NULL;
1354 newnp->opt = NULL;
1355 newnp->mcast_oif = tcp_v6_iif(skb);
1356 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1357 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1358 if (np->repflow)
1359 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1360
1361 /* Set ToS of the new socket based upon the value of incoming SYN.
1362 * ECT bits are set later in tcp_init_transfer().
1363 */
1364 if (sock_net(sk)->ipv4.sysctl_tcp_reflect_tos)
1365 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1366
1367 /* Clone native IPv6 options from listening socket (if any)
1368
1369 Yes, keeping reference count would be much more clever,
1370 but we make one more one thing there: reattach optmem
1371 to newsk.
1372 */
1373 opt = ireq->ipv6_opt;
1374 if (!opt)
1375 opt = rcu_dereference(np->opt);
1376 if (opt) {
1377 opt = ipv6_dup_options(newsk, opt);
1378 RCU_INIT_POINTER(newnp->opt, opt);
1379 }
1380 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1381 if (opt)
1382 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1383 opt->opt_flen;
1384
1385 tcp_ca_openreq_child(newsk, dst);
1386
1387 tcp_sync_mss(newsk, dst_mtu(dst));
1388 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1389
1390 tcp_initialize_rcv_mss(newsk);
1391
1392 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1393 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1394
1395#ifdef CONFIG_TCP_MD5SIG
1396 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1397
1398 /* Copy over the MD5 key from the original socket */
1399 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1400 if (key) {
1401 /* We're using one, so create a matching key
1402 * on the newsk structure. If we fail to get
1403 * memory, then we end up not copying the key
1404 * across. Shucks.
1405 */
1406 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1407 AF_INET6, 128, l3index, key->key, key->keylen,
1408 sk_gfp_mask(sk, GFP_ATOMIC));
1409 }
1410#endif
1411
1412 if (__inet_inherit_port(sk, newsk) < 0) {
1413 inet_csk_prepare_forced_close(newsk);
1414 tcp_done(newsk);
1415 goto out;
1416 }
1417 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1418 &found_dup_sk);
1419 if (*own_req) {
1420 tcp_move_syn(newtp, req);
1421
1422 /* Clone pktoptions received with SYN, if we own the req */
1423 if (ireq->pktopts) {
1424 newnp->pktoptions = skb_clone(ireq->pktopts,
1425 sk_gfp_mask(sk, GFP_ATOMIC));
1426 consume_skb(ireq->pktopts);
1427 ireq->pktopts = NULL;
1428 if (newnp->pktoptions) {
1429 tcp_v6_restore_cb(newnp->pktoptions);
1430 skb_set_owner_r(newnp->pktoptions, newsk);
1431 }
1432 }
1433 } else {
1434 if (!req_unhash && found_dup_sk) {
1435 /* This code path should only be executed in the
1436 * syncookie case only
1437 */
1438 bh_unlock_sock(newsk);
1439 sock_put(newsk);
1440 newsk = NULL;
1441 }
1442 }
1443
1444 return newsk;
1445
1446out_overflow:
1447 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1448out_nonewsk:
1449 dst_release(dst);
1450out:
1451 tcp_listendrop(sk);
1452 return NULL;
1453}
1454
1455INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1456 u32));
1457/* The socket must have it's spinlock held when we get
1458 * here, unless it is a TCP_LISTEN socket.
1459 *
1460 * We have a potential double-lock case here, so even when
1461 * doing backlog processing we use the BH locking scheme.
1462 * This is because we cannot sleep with the original spinlock
1463 * held.
1464 */
1465static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1466{
1467 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1468 struct sk_buff *opt_skb = NULL;
1469 struct tcp_sock *tp;
1470
1471 /* Imagine: socket is IPv6. IPv4 packet arrives,
1472 goes to IPv4 receive handler and backlogged.
1473 From backlog it always goes here. Kerboom...
1474 Fortunately, tcp_rcv_established and rcv_established
1475 handle them correctly, but it is not case with
1476 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1477 */
1478
1479 if (skb->protocol == htons(ETH_P_IP))
1480 return tcp_v4_do_rcv(sk, skb);
1481
1482 /*
1483 * socket locking is here for SMP purposes as backlog rcv
1484 * is currently called with bh processing disabled.
1485 */
1486
1487 /* Do Stevens' IPV6_PKTOPTIONS.
1488
1489 Yes, guys, it is the only place in our code, where we
1490 may make it not affecting IPv4.
1491 The rest of code is protocol independent,
1492 and I do not like idea to uglify IPv4.
1493
1494 Actually, all the idea behind IPV6_PKTOPTIONS
1495 looks not very well thought. For now we latch
1496 options, received in the last packet, enqueued
1497 by tcp. Feel free to propose better solution.
1498 --ANK (980728)
1499 */
1500 if (np->rxopt.all)
1501 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1502
1503 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1504 struct dst_entry *dst = sk->sk_rx_dst;
1505
1506 sock_rps_save_rxhash(sk, skb);
1507 sk_mark_napi_id(sk, skb);
1508 if (dst) {
1509 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1510 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1511 dst, np->rx_dst_cookie) == NULL) {
1512 dst_release(dst);
1513 sk->sk_rx_dst = NULL;
1514 }
1515 }
1516
1517 tcp_rcv_established(sk, skb);
1518 if (opt_skb)
1519 goto ipv6_pktoptions;
1520 return 0;
1521 }
1522
1523 if (tcp_checksum_complete(skb))
1524 goto csum_err;
1525
1526 if (sk->sk_state == TCP_LISTEN) {
1527 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1528
1529 if (!nsk)
1530 goto discard;
1531
1532 if (nsk != sk) {
1533 if (tcp_child_process(sk, nsk, skb))
1534 goto reset;
1535 if (opt_skb)
1536 __kfree_skb(opt_skb);
1537 return 0;
1538 }
1539 } else
1540 sock_rps_save_rxhash(sk, skb);
1541
1542 if (tcp_rcv_state_process(sk, skb))
1543 goto reset;
1544 if (opt_skb)
1545 goto ipv6_pktoptions;
1546 return 0;
1547
1548reset:
1549 tcp_v6_send_reset(sk, skb);
1550discard:
1551 if (opt_skb)
1552 __kfree_skb(opt_skb);
1553 kfree_skb(skb);
1554 return 0;
1555csum_err:
1556 trace_tcp_bad_csum(skb);
1557 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1558 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1559 goto discard;
1560
1561
1562ipv6_pktoptions:
1563 /* Do you ask, what is it?
1564
1565 1. skb was enqueued by tcp.
1566 2. skb is added to tail of read queue, rather than out of order.
1567 3. socket is not in passive state.
1568 4. Finally, it really contains options, which user wants to receive.
1569 */
1570 tp = tcp_sk(sk);
1571 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1572 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1573 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1574 np->mcast_oif = tcp_v6_iif(opt_skb);
1575 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1576 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1577 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1578 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1579 if (np->repflow)
1580 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1581 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1582 skb_set_owner_r(opt_skb, sk);
1583 tcp_v6_restore_cb(opt_skb);
1584 opt_skb = xchg(&np->pktoptions, opt_skb);
1585 } else {
1586 __kfree_skb(opt_skb);
1587 opt_skb = xchg(&np->pktoptions, NULL);
1588 }
1589 }
1590
1591 kfree_skb(opt_skb);
1592 return 0;
1593}
1594
1595static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1596 const struct tcphdr *th)
1597{
1598 /* This is tricky: we move IP6CB at its correct location into
1599 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1600 * _decode_session6() uses IP6CB().
1601 * barrier() makes sure compiler won't play aliasing games.
1602 */
1603 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1604 sizeof(struct inet6_skb_parm));
1605 barrier();
1606
1607 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1608 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1609 skb->len - th->doff*4);
1610 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1611 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1612 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1613 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1614 TCP_SKB_CB(skb)->sacked = 0;
1615 TCP_SKB_CB(skb)->has_rxtstamp =
1616 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1617}
1618
1619INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1620{
1621 struct sk_buff *skb_to_free;
1622 int sdif = inet6_sdif(skb);
1623 int dif = inet6_iif(skb);
1624 const struct tcphdr *th;
1625 const struct ipv6hdr *hdr;
1626 bool refcounted;
1627 struct sock *sk;
1628 int ret;
1629 struct net *net = dev_net(skb->dev);
1630
1631 if (skb->pkt_type != PACKET_HOST)
1632 goto discard_it;
1633
1634 /*
1635 * Count it even if it's bad.
1636 */
1637 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1638
1639 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1640 goto discard_it;
1641
1642 th = (const struct tcphdr *)skb->data;
1643
1644 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1645 goto bad_packet;
1646 if (!pskb_may_pull(skb, th->doff*4))
1647 goto discard_it;
1648
1649 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1650 goto csum_error;
1651
1652 th = (const struct tcphdr *)skb->data;
1653 hdr = ipv6_hdr(skb);
1654
1655lookup:
1656 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1657 th->source, th->dest, inet6_iif(skb), sdif,
1658 &refcounted);
1659 if (!sk)
1660 goto no_tcp_socket;
1661
1662process:
1663 if (sk->sk_state == TCP_TIME_WAIT)
1664 goto do_time_wait;
1665
1666 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1667 struct request_sock *req = inet_reqsk(sk);
1668 bool req_stolen = false;
1669 struct sock *nsk;
1670
1671 sk = req->rsk_listener;
1672 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif)) {
1673 sk_drops_add(sk, skb);
1674 reqsk_put(req);
1675 goto discard_it;
1676 }
1677 if (tcp_checksum_complete(skb)) {
1678 reqsk_put(req);
1679 goto csum_error;
1680 }
1681 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1682 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
1683 if (!nsk) {
1684 inet_csk_reqsk_queue_drop_and_put(sk, req);
1685 goto lookup;
1686 }
1687 sk = nsk;
1688 /* reuseport_migrate_sock() has already held one sk_refcnt
1689 * before returning.
1690 */
1691 } else {
1692 sock_hold(sk);
1693 }
1694 refcounted = true;
1695 nsk = NULL;
1696 if (!tcp_filter(sk, skb)) {
1697 th = (const struct tcphdr *)skb->data;
1698 hdr = ipv6_hdr(skb);
1699 tcp_v6_fill_cb(skb, hdr, th);
1700 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1701 }
1702 if (!nsk) {
1703 reqsk_put(req);
1704 if (req_stolen) {
1705 /* Another cpu got exclusive access to req
1706 * and created a full blown socket.
1707 * Try to feed this packet to this socket
1708 * instead of discarding it.
1709 */
1710 tcp_v6_restore_cb(skb);
1711 sock_put(sk);
1712 goto lookup;
1713 }
1714 goto discard_and_relse;
1715 }
1716 if (nsk == sk) {
1717 reqsk_put(req);
1718 tcp_v6_restore_cb(skb);
1719 } else if (tcp_child_process(sk, nsk, skb)) {
1720 tcp_v6_send_reset(nsk, skb);
1721 goto discard_and_relse;
1722 } else {
1723 sock_put(sk);
1724 return 0;
1725 }
1726 }
1727 if (hdr->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
1728 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1729 goto discard_and_relse;
1730 }
1731
1732 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1733 goto discard_and_relse;
1734
1735 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif))
1736 goto discard_and_relse;
1737
1738 if (tcp_filter(sk, skb))
1739 goto discard_and_relse;
1740 th = (const struct tcphdr *)skb->data;
1741 hdr = ipv6_hdr(skb);
1742 tcp_v6_fill_cb(skb, hdr, th);
1743
1744 skb->dev = NULL;
1745
1746 if (sk->sk_state == TCP_LISTEN) {
1747 ret = tcp_v6_do_rcv(sk, skb);
1748 goto put_and_return;
1749 }
1750
1751 sk_incoming_cpu_update(sk);
1752
1753 bh_lock_sock_nested(sk);
1754 tcp_segs_in(tcp_sk(sk), skb);
1755 ret = 0;
1756 if (!sock_owned_by_user(sk)) {
1757 skb_to_free = sk->sk_rx_skb_cache;
1758 sk->sk_rx_skb_cache = NULL;
1759 ret = tcp_v6_do_rcv(sk, skb);
1760 } else {
1761 if (tcp_add_backlog(sk, skb))
1762 goto discard_and_relse;
1763 skb_to_free = NULL;
1764 }
1765 bh_unlock_sock(sk);
1766 if (skb_to_free)
1767 __kfree_skb(skb_to_free);
1768put_and_return:
1769 if (refcounted)
1770 sock_put(sk);
1771 return ret ? -1 : 0;
1772
1773no_tcp_socket:
1774 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1775 goto discard_it;
1776
1777 tcp_v6_fill_cb(skb, hdr, th);
1778
1779 if (tcp_checksum_complete(skb)) {
1780csum_error:
1781 trace_tcp_bad_csum(skb);
1782 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1783bad_packet:
1784 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1785 } else {
1786 tcp_v6_send_reset(NULL, skb);
1787 }
1788
1789discard_it:
1790 kfree_skb(skb);
1791 return 0;
1792
1793discard_and_relse:
1794 sk_drops_add(sk, skb);
1795 if (refcounted)
1796 sock_put(sk);
1797 goto discard_it;
1798
1799do_time_wait:
1800 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1801 inet_twsk_put(inet_twsk(sk));
1802 goto discard_it;
1803 }
1804
1805 tcp_v6_fill_cb(skb, hdr, th);
1806
1807 if (tcp_checksum_complete(skb)) {
1808 inet_twsk_put(inet_twsk(sk));
1809 goto csum_error;
1810 }
1811
1812 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1813 case TCP_TW_SYN:
1814 {
1815 struct sock *sk2;
1816
1817 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1818 skb, __tcp_hdrlen(th),
1819 &ipv6_hdr(skb)->saddr, th->source,
1820 &ipv6_hdr(skb)->daddr,
1821 ntohs(th->dest),
1822 tcp_v6_iif_l3_slave(skb),
1823 sdif);
1824 if (sk2) {
1825 struct inet_timewait_sock *tw = inet_twsk(sk);
1826 inet_twsk_deschedule_put(tw);
1827 sk = sk2;
1828 tcp_v6_restore_cb(skb);
1829 refcounted = false;
1830 goto process;
1831 }
1832 }
1833 /* to ACK */
1834 fallthrough;
1835 case TCP_TW_ACK:
1836 tcp_v6_timewait_ack(sk, skb);
1837 break;
1838 case TCP_TW_RST:
1839 tcp_v6_send_reset(sk, skb);
1840 inet_twsk_deschedule_put(inet_twsk(sk));
1841 goto discard_it;
1842 case TCP_TW_SUCCESS:
1843 ;
1844 }
1845 goto discard_it;
1846}
1847
1848INDIRECT_CALLABLE_SCOPE void tcp_v6_early_demux(struct sk_buff *skb)
1849{
1850 const struct ipv6hdr *hdr;
1851 const struct tcphdr *th;
1852 struct sock *sk;
1853
1854 if (skb->pkt_type != PACKET_HOST)
1855 return;
1856
1857 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1858 return;
1859
1860 hdr = ipv6_hdr(skb);
1861 th = tcp_hdr(skb);
1862
1863 if (th->doff < sizeof(struct tcphdr) / 4)
1864 return;
1865
1866 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1867 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1868 &hdr->saddr, th->source,
1869 &hdr->daddr, ntohs(th->dest),
1870 inet6_iif(skb), inet6_sdif(skb));
1871 if (sk) {
1872 skb->sk = sk;
1873 skb->destructor = sock_edemux;
1874 if (sk_fullsock(sk)) {
1875 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1876
1877 if (dst)
1878 dst = dst_check(dst, tcp_inet6_sk(sk)->rx_dst_cookie);
1879 if (dst &&
1880 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1881 skb_dst_set_noref(skb, dst);
1882 }
1883 }
1884}
1885
1886static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1887 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1888 .twsk_unique = tcp_twsk_unique,
1889 .twsk_destructor = tcp_twsk_destructor,
1890};
1891
1892INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
1893{
1894 struct ipv6_pinfo *np = inet6_sk(sk);
1895
1896 __tcp_v6_send_check(skb, &np->saddr, &sk->sk_v6_daddr);
1897}
1898
1899const struct inet_connection_sock_af_ops ipv6_specific = {
1900 .queue_xmit = inet6_csk_xmit,
1901 .send_check = tcp_v6_send_check,
1902 .rebuild_header = inet6_sk_rebuild_header,
1903 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1904 .conn_request = tcp_v6_conn_request,
1905 .syn_recv_sock = tcp_v6_syn_recv_sock,
1906 .net_header_len = sizeof(struct ipv6hdr),
1907 .net_frag_header_len = sizeof(struct frag_hdr),
1908 .setsockopt = ipv6_setsockopt,
1909 .getsockopt = ipv6_getsockopt,
1910 .addr2sockaddr = inet6_csk_addr2sockaddr,
1911 .sockaddr_len = sizeof(struct sockaddr_in6),
1912 .mtu_reduced = tcp_v6_mtu_reduced,
1913};
1914
1915#ifdef CONFIG_TCP_MD5SIG
1916static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1917 .md5_lookup = tcp_v6_md5_lookup,
1918 .calc_md5_hash = tcp_v6_md5_hash_skb,
1919 .md5_parse = tcp_v6_parse_md5_keys,
1920};
1921#endif
1922
1923/*
1924 * TCP over IPv4 via INET6 API
1925 */
1926static const struct inet_connection_sock_af_ops ipv6_mapped = {
1927 .queue_xmit = ip_queue_xmit,
1928 .send_check = tcp_v4_send_check,
1929 .rebuild_header = inet_sk_rebuild_header,
1930 .sk_rx_dst_set = inet_sk_rx_dst_set,
1931 .conn_request = tcp_v6_conn_request,
1932 .syn_recv_sock = tcp_v6_syn_recv_sock,
1933 .net_header_len = sizeof(struct iphdr),
1934 .setsockopt = ipv6_setsockopt,
1935 .getsockopt = ipv6_getsockopt,
1936 .addr2sockaddr = inet6_csk_addr2sockaddr,
1937 .sockaddr_len = sizeof(struct sockaddr_in6),
1938 .mtu_reduced = tcp_v4_mtu_reduced,
1939};
1940
1941#ifdef CONFIG_TCP_MD5SIG
1942static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1943 .md5_lookup = tcp_v4_md5_lookup,
1944 .calc_md5_hash = tcp_v4_md5_hash_skb,
1945 .md5_parse = tcp_v6_parse_md5_keys,
1946};
1947#endif
1948
1949/* NOTE: A lot of things set to zero explicitly by call to
1950 * sk_alloc() so need not be done here.
1951 */
1952static int tcp_v6_init_sock(struct sock *sk)
1953{
1954 struct inet_connection_sock *icsk = inet_csk(sk);
1955
1956 tcp_init_sock(sk);
1957
1958 icsk->icsk_af_ops = &ipv6_specific;
1959
1960#ifdef CONFIG_TCP_MD5SIG
1961 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1962#endif
1963
1964 return 0;
1965}
1966
1967static void tcp_v6_destroy_sock(struct sock *sk)
1968{
1969 tcp_v4_destroy_sock(sk);
1970 inet6_destroy_sock(sk);
1971}
1972
1973#ifdef CONFIG_PROC_FS
1974/* Proc filesystem TCPv6 sock list dumping. */
1975static void get_openreq6(struct seq_file *seq,
1976 const struct request_sock *req, int i)
1977{
1978 long ttd = req->rsk_timer.expires - jiffies;
1979 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1980 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1981
1982 if (ttd < 0)
1983 ttd = 0;
1984
1985 seq_printf(seq,
1986 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1987 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1988 i,
1989 src->s6_addr32[0], src->s6_addr32[1],
1990 src->s6_addr32[2], src->s6_addr32[3],
1991 inet_rsk(req)->ir_num,
1992 dest->s6_addr32[0], dest->s6_addr32[1],
1993 dest->s6_addr32[2], dest->s6_addr32[3],
1994 ntohs(inet_rsk(req)->ir_rmt_port),
1995 TCP_SYN_RECV,
1996 0, 0, /* could print option size, but that is af dependent. */
1997 1, /* timers active (only the expire timer) */
1998 jiffies_to_clock_t(ttd),
1999 req->num_timeout,
2000 from_kuid_munged(seq_user_ns(seq),
2001 sock_i_uid(req->rsk_listener)),
2002 0, /* non standard timer */
2003 0, /* open_requests have no inode */
2004 0, req);
2005}
2006
2007static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2008{
2009 const struct in6_addr *dest, *src;
2010 __u16 destp, srcp;
2011 int timer_active;
2012 unsigned long timer_expires;
2013 const struct inet_sock *inet = inet_sk(sp);
2014 const struct tcp_sock *tp = tcp_sk(sp);
2015 const struct inet_connection_sock *icsk = inet_csk(sp);
2016 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2017 int rx_queue;
2018 int state;
2019
2020 dest = &sp->sk_v6_daddr;
2021 src = &sp->sk_v6_rcv_saddr;
2022 destp = ntohs(inet->inet_dport);
2023 srcp = ntohs(inet->inet_sport);
2024
2025 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2026 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2027 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2028 timer_active = 1;
2029 timer_expires = icsk->icsk_timeout;
2030 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2031 timer_active = 4;
2032 timer_expires = icsk->icsk_timeout;
2033 } else if (timer_pending(&sp->sk_timer)) {
2034 timer_active = 2;
2035 timer_expires = sp->sk_timer.expires;
2036 } else {
2037 timer_active = 0;
2038 timer_expires = jiffies;
2039 }
2040
2041 state = inet_sk_state_load(sp);
2042 if (state == TCP_LISTEN)
2043 rx_queue = READ_ONCE(sp->sk_ack_backlog);
2044 else
2045 /* Because we don't lock the socket,
2046 * we might find a transient negative value.
2047 */
2048 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2049 READ_ONCE(tp->copied_seq), 0);
2050
2051 seq_printf(seq,
2052 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2053 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2054 i,
2055 src->s6_addr32[0], src->s6_addr32[1],
2056 src->s6_addr32[2], src->s6_addr32[3], srcp,
2057 dest->s6_addr32[0], dest->s6_addr32[1],
2058 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2059 state,
2060 READ_ONCE(tp->write_seq) - tp->snd_una,
2061 rx_queue,
2062 timer_active,
2063 jiffies_delta_to_clock_t(timer_expires - jiffies),
2064 icsk->icsk_retransmits,
2065 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
2066 icsk->icsk_probes_out,
2067 sock_i_ino(sp),
2068 refcount_read(&sp->sk_refcnt), sp,
2069 jiffies_to_clock_t(icsk->icsk_rto),
2070 jiffies_to_clock_t(icsk->icsk_ack.ato),
2071 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2072 tp->snd_cwnd,
2073 state == TCP_LISTEN ?
2074 fastopenq->max_qlen :
2075 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2076 );
2077}
2078
2079static void get_timewait6_sock(struct seq_file *seq,
2080 struct inet_timewait_sock *tw, int i)
2081{
2082 long delta = tw->tw_timer.expires - jiffies;
2083 const struct in6_addr *dest, *src;
2084 __u16 destp, srcp;
2085
2086 dest = &tw->tw_v6_daddr;
2087 src = &tw->tw_v6_rcv_saddr;
2088 destp = ntohs(tw->tw_dport);
2089 srcp = ntohs(tw->tw_sport);
2090
2091 seq_printf(seq,
2092 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2093 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2094 i,
2095 src->s6_addr32[0], src->s6_addr32[1],
2096 src->s6_addr32[2], src->s6_addr32[3], srcp,
2097 dest->s6_addr32[0], dest->s6_addr32[1],
2098 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2099 tw->tw_substate, 0, 0,
2100 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2101 refcount_read(&tw->tw_refcnt), tw);
2102}
2103
2104static int tcp6_seq_show(struct seq_file *seq, void *v)
2105{
2106 struct tcp_iter_state *st;
2107 struct sock *sk = v;
2108
2109 if (v == SEQ_START_TOKEN) {
2110 seq_puts(seq,
2111 " sl "
2112 "local_address "
2113 "remote_address "
2114 "st tx_queue rx_queue tr tm->when retrnsmt"
2115 " uid timeout inode\n");
2116 goto out;
2117 }
2118 st = seq->private;
2119
2120 if (sk->sk_state == TCP_TIME_WAIT)
2121 get_timewait6_sock(seq, v, st->num);
2122 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2123 get_openreq6(seq, v, st->num);
2124 else
2125 get_tcp6_sock(seq, v, st->num);
2126out:
2127 return 0;
2128}
2129
2130static const struct seq_operations tcp6_seq_ops = {
2131 .show = tcp6_seq_show,
2132 .start = tcp_seq_start,
2133 .next = tcp_seq_next,
2134 .stop = tcp_seq_stop,
2135};
2136
2137static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2138 .family = AF_INET6,
2139};
2140
2141int __net_init tcp6_proc_init(struct net *net)
2142{
2143 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2144 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2145 return -ENOMEM;
2146 return 0;
2147}
2148
2149void tcp6_proc_exit(struct net *net)
2150{
2151 remove_proc_entry("tcp6", net->proc_net);
2152}
2153#endif
2154
2155struct proto tcpv6_prot = {
2156 .name = "TCPv6",
2157 .owner = THIS_MODULE,
2158 .close = tcp_close,
2159 .pre_connect = tcp_v6_pre_connect,
2160 .connect = tcp_v6_connect,
2161 .disconnect = tcp_disconnect,
2162 .accept = inet_csk_accept,
2163 .ioctl = tcp_ioctl,
2164 .init = tcp_v6_init_sock,
2165 .destroy = tcp_v6_destroy_sock,
2166 .shutdown = tcp_shutdown,
2167 .setsockopt = tcp_setsockopt,
2168 .getsockopt = tcp_getsockopt,
2169 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
2170 .keepalive = tcp_set_keepalive,
2171 .recvmsg = tcp_recvmsg,
2172 .sendmsg = tcp_sendmsg,
2173 .sendpage = tcp_sendpage,
2174 .backlog_rcv = tcp_v6_do_rcv,
2175 .release_cb = tcp_release_cb,
2176 .hash = inet6_hash,
2177 .unhash = inet_unhash,
2178 .get_port = inet_csk_get_port,
2179#ifdef CONFIG_BPF_SYSCALL
2180 .psock_update_sk_prot = tcp_bpf_update_proto,
2181#endif
2182 .enter_memory_pressure = tcp_enter_memory_pressure,
2183 .leave_memory_pressure = tcp_leave_memory_pressure,
2184 .stream_memory_free = tcp_stream_memory_free,
2185 .sockets_allocated = &tcp_sockets_allocated,
2186 .memory_allocated = &tcp_memory_allocated,
2187 .memory_pressure = &tcp_memory_pressure,
2188 .orphan_count = &tcp_orphan_count,
2189 .sysctl_mem = sysctl_tcp_mem,
2190 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2191 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2192 .max_header = MAX_TCP_HEADER,
2193 .obj_size = sizeof(struct tcp6_sock),
2194 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2195 .twsk_prot = &tcp6_timewait_sock_ops,
2196 .rsk_prot = &tcp6_request_sock_ops,
2197 .h.hashinfo = &tcp_hashinfo,
2198 .no_autobind = true,
2199 .diag_destroy = tcp_abort,
2200};
2201EXPORT_SYMBOL_GPL(tcpv6_prot);
2202
2203/* thinking of making this const? Don't.
2204 * early_demux can change based on sysctl.
2205 */
2206static struct inet6_protocol tcpv6_protocol = {
2207 .early_demux = tcp_v6_early_demux,
2208 .early_demux_handler = tcp_v6_early_demux,
2209 .handler = tcp_v6_rcv,
2210 .err_handler = tcp_v6_err,
2211 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2212};
2213
2214static struct inet_protosw tcpv6_protosw = {
2215 .type = SOCK_STREAM,
2216 .protocol = IPPROTO_TCP,
2217 .prot = &tcpv6_prot,
2218 .ops = &inet6_stream_ops,
2219 .flags = INET_PROTOSW_PERMANENT |
2220 INET_PROTOSW_ICSK,
2221};
2222
2223static int __net_init tcpv6_net_init(struct net *net)
2224{
2225 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2226 SOCK_RAW, IPPROTO_TCP, net);
2227}
2228
2229static void __net_exit tcpv6_net_exit(struct net *net)
2230{
2231 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2232}
2233
2234static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2235{
2236 inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2237}
2238
2239static struct pernet_operations tcpv6_net_ops = {
2240 .init = tcpv6_net_init,
2241 .exit = tcpv6_net_exit,
2242 .exit_batch = tcpv6_net_exit_batch,
2243};
2244
2245int __init tcpv6_init(void)
2246{
2247 int ret;
2248
2249 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2250 if (ret)
2251 goto out;
2252
2253 /* register inet6 protocol */
2254 ret = inet6_register_protosw(&tcpv6_protosw);
2255 if (ret)
2256 goto out_tcpv6_protocol;
2257
2258 ret = register_pernet_subsys(&tcpv6_net_ops);
2259 if (ret)
2260 goto out_tcpv6_protosw;
2261
2262 ret = mptcpv6_init();
2263 if (ret)
2264 goto out_tcpv6_pernet_subsys;
2265
2266out:
2267 return ret;
2268
2269out_tcpv6_pernet_subsys:
2270 unregister_pernet_subsys(&tcpv6_net_ops);
2271out_tcpv6_protosw:
2272 inet6_unregister_protosw(&tcpv6_protosw);
2273out_tcpv6_protocol:
2274 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2275 goto out;
2276}
2277
2278void tcpv6_exit(void)
2279{
2280 unregister_pernet_subsys(&tcpv6_net_ops);
2281 inet6_unregister_protosw(&tcpv6_protosw);
2282 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2283}