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/*
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/netdma.h>
63#include <net/inet_common.h>
64#include <net/secure_seq.h>
65#include <net/tcp_memcontrol.h>
66#include <net/busy_poll.h>
67
68#include <linux/proc_fs.h>
69#include <linux/seq_file.h>
70
71#include <linux/crypto.h>
72#include <linux/scatterlist.h>
73
74static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
75static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
76 struct request_sock *req);
77
78static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
79
80static const struct inet_connection_sock_af_ops ipv6_mapped;
81static const struct inet_connection_sock_af_ops ipv6_specific;
82#ifdef CONFIG_TCP_MD5SIG
83static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
84static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
85#else
86static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
87 const struct in6_addr *addr)
88{
89 return NULL;
90}
91#endif
92
93static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
94{
95 struct dst_entry *dst = skb_dst(skb);
96 const struct rt6_info *rt = (const struct rt6_info *)dst;
97
98 dst_hold(dst);
99 sk->sk_rx_dst = dst;
100 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
101 if (rt->rt6i_node)
102 inet6_sk(sk)->rx_dst_cookie = rt->rt6i_node->fn_sernum;
103}
104
105static void tcp_v6_hash(struct sock *sk)
106{
107 if (sk->sk_state != TCP_CLOSE) {
108 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
109 tcp_prot.hash(sk);
110 return;
111 }
112 local_bh_disable();
113 __inet6_hash(sk, NULL);
114 local_bh_enable();
115 }
116}
117
118static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
119{
120 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
121 ipv6_hdr(skb)->saddr.s6_addr32,
122 tcp_hdr(skb)->dest,
123 tcp_hdr(skb)->source);
124}
125
126static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
127 int addr_len)
128{
129 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
130 struct inet_sock *inet = inet_sk(sk);
131 struct inet_connection_sock *icsk = inet_csk(sk);
132 struct ipv6_pinfo *np = inet6_sk(sk);
133 struct tcp_sock *tp = tcp_sk(sk);
134 struct in6_addr *saddr = NULL, *final_p, final;
135 struct rt6_info *rt;
136 struct flowi6 fl6;
137 struct dst_entry *dst;
138 int addr_type;
139 int err;
140
141 if (addr_len < SIN6_LEN_RFC2133)
142 return -EINVAL;
143
144 if (usin->sin6_family != AF_INET6)
145 return -EAFNOSUPPORT;
146
147 memset(&fl6, 0, sizeof(fl6));
148
149 if (np->sndflow) {
150 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
151 IP6_ECN_flow_init(fl6.flowlabel);
152 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
153 struct ip6_flowlabel *flowlabel;
154 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
155 if (flowlabel == NULL)
156 return -EINVAL;
157 fl6_sock_release(flowlabel);
158 }
159 }
160
161 /*
162 * connect() to INADDR_ANY means loopback (BSD'ism).
163 */
164
165 if (ipv6_addr_any(&usin->sin6_addr))
166 usin->sin6_addr.s6_addr[15] = 0x1;
167
168 addr_type = ipv6_addr_type(&usin->sin6_addr);
169
170 if (addr_type & IPV6_ADDR_MULTICAST)
171 return -ENETUNREACH;
172
173 if (addr_type&IPV6_ADDR_LINKLOCAL) {
174 if (addr_len >= sizeof(struct sockaddr_in6) &&
175 usin->sin6_scope_id) {
176 /* If interface is set while binding, indices
177 * must coincide.
178 */
179 if (sk->sk_bound_dev_if &&
180 sk->sk_bound_dev_if != usin->sin6_scope_id)
181 return -EINVAL;
182
183 sk->sk_bound_dev_if = usin->sin6_scope_id;
184 }
185
186 /* Connect to link-local address requires an interface */
187 if (!sk->sk_bound_dev_if)
188 return -EINVAL;
189 }
190
191 if (tp->rx_opt.ts_recent_stamp &&
192 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
193 tp->rx_opt.ts_recent = 0;
194 tp->rx_opt.ts_recent_stamp = 0;
195 tp->write_seq = 0;
196 }
197
198 sk->sk_v6_daddr = usin->sin6_addr;
199 np->flow_label = fl6.flowlabel;
200
201 /*
202 * TCP over IPv4
203 */
204
205 if (addr_type == IPV6_ADDR_MAPPED) {
206 u32 exthdrlen = icsk->icsk_ext_hdr_len;
207 struct sockaddr_in sin;
208
209 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
210
211 if (__ipv6_only_sock(sk))
212 return -ENETUNREACH;
213
214 sin.sin_family = AF_INET;
215 sin.sin_port = usin->sin6_port;
216 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
217
218 icsk->icsk_af_ops = &ipv6_mapped;
219 sk->sk_backlog_rcv = tcp_v4_do_rcv;
220#ifdef CONFIG_TCP_MD5SIG
221 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
222#endif
223
224 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
225
226 if (err) {
227 icsk->icsk_ext_hdr_len = exthdrlen;
228 icsk->icsk_af_ops = &ipv6_specific;
229 sk->sk_backlog_rcv = tcp_v6_do_rcv;
230#ifdef CONFIG_TCP_MD5SIG
231 tp->af_specific = &tcp_sock_ipv6_specific;
232#endif
233 goto failure;
234 } else {
235 ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
236 ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
237 &sk->sk_v6_rcv_saddr);
238 }
239
240 return err;
241 }
242
243 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
244 saddr = &sk->sk_v6_rcv_saddr;
245
246 fl6.flowi6_proto = IPPROTO_TCP;
247 fl6.daddr = sk->sk_v6_daddr;
248 fl6.saddr = saddr ? *saddr : np->saddr;
249 fl6.flowi6_oif = sk->sk_bound_dev_if;
250 fl6.flowi6_mark = sk->sk_mark;
251 fl6.fl6_dport = usin->sin6_port;
252 fl6.fl6_sport = inet->inet_sport;
253
254 final_p = fl6_update_dst(&fl6, np->opt, &final);
255
256 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
257
258 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
259 if (IS_ERR(dst)) {
260 err = PTR_ERR(dst);
261 goto failure;
262 }
263
264 if (saddr == NULL) {
265 saddr = &fl6.saddr;
266 sk->sk_v6_rcv_saddr = *saddr;
267 }
268
269 /* set the source address */
270 np->saddr = *saddr;
271 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
272
273 sk->sk_gso_type = SKB_GSO_TCPV6;
274 __ip6_dst_store(sk, dst, NULL, NULL);
275
276 rt = (struct rt6_info *) dst;
277 if (tcp_death_row.sysctl_tw_recycle &&
278 !tp->rx_opt.ts_recent_stamp &&
279 ipv6_addr_equal(&rt->rt6i_dst.addr, &sk->sk_v6_daddr))
280 tcp_fetch_timewait_stamp(sk, dst);
281
282 icsk->icsk_ext_hdr_len = 0;
283 if (np->opt)
284 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
285 np->opt->opt_nflen);
286
287 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
288
289 inet->inet_dport = usin->sin6_port;
290
291 tcp_set_state(sk, TCP_SYN_SENT);
292 err = inet6_hash_connect(&tcp_death_row, sk);
293 if (err)
294 goto late_failure;
295
296 if (!tp->write_seq && likely(!tp->repair))
297 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
298 sk->sk_v6_daddr.s6_addr32,
299 inet->inet_sport,
300 inet->inet_dport);
301
302 err = tcp_connect(sk);
303 if (err)
304 goto late_failure;
305
306 return 0;
307
308late_failure:
309 tcp_set_state(sk, TCP_CLOSE);
310 __sk_dst_reset(sk);
311failure:
312 inet->inet_dport = 0;
313 sk->sk_route_caps = 0;
314 return err;
315}
316
317static void tcp_v6_mtu_reduced(struct sock *sk)
318{
319 struct dst_entry *dst;
320
321 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
322 return;
323
324 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
325 if (!dst)
326 return;
327
328 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
329 tcp_sync_mss(sk, dst_mtu(dst));
330 tcp_simple_retransmit(sk);
331 }
332}
333
334static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
335 u8 type, u8 code, int offset, __be32 info)
336{
337 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
338 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
339 struct ipv6_pinfo *np;
340 struct sock *sk;
341 int err;
342 struct tcp_sock *tp;
343 __u32 seq;
344 struct net *net = dev_net(skb->dev);
345
346 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
347 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
348
349 if (sk == NULL) {
350 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
351 ICMP6_MIB_INERRORS);
352 return;
353 }
354
355 if (sk->sk_state == TCP_TIME_WAIT) {
356 inet_twsk_put(inet_twsk(sk));
357 return;
358 }
359
360 bh_lock_sock(sk);
361 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
362 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
363
364 if (sk->sk_state == TCP_CLOSE)
365 goto out;
366
367 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
368 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
369 goto out;
370 }
371
372 tp = tcp_sk(sk);
373 seq = ntohl(th->seq);
374 if (sk->sk_state != TCP_LISTEN &&
375 !between(seq, tp->snd_una, tp->snd_nxt)) {
376 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
377 goto out;
378 }
379
380 np = inet6_sk(sk);
381
382 if (type == NDISC_REDIRECT) {
383 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
384
385 if (dst)
386 dst->ops->redirect(dst, sk, skb);
387 goto out;
388 }
389
390 if (type == ICMPV6_PKT_TOOBIG) {
391 /* We are not interested in TCP_LISTEN and open_requests
392 * (SYN-ACKs send out by Linux are always <576bytes so
393 * they should go through unfragmented).
394 */
395 if (sk->sk_state == TCP_LISTEN)
396 goto out;
397
398 if (!ip6_sk_accept_pmtu(sk))
399 goto out;
400
401 tp->mtu_info = ntohl(info);
402 if (!sock_owned_by_user(sk))
403 tcp_v6_mtu_reduced(sk);
404 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
405 &tp->tsq_flags))
406 sock_hold(sk);
407 goto out;
408 }
409
410 icmpv6_err_convert(type, code, &err);
411
412 /* Might be for an request_sock */
413 switch (sk->sk_state) {
414 struct request_sock *req, **prev;
415 case TCP_LISTEN:
416 if (sock_owned_by_user(sk))
417 goto out;
418
419 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
420 &hdr->saddr, inet6_iif(skb));
421 if (!req)
422 goto out;
423
424 /* ICMPs are not backlogged, hence we cannot get
425 * an established socket here.
426 */
427 WARN_ON(req->sk != NULL);
428
429 if (seq != tcp_rsk(req)->snt_isn) {
430 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
431 goto out;
432 }
433
434 inet_csk_reqsk_queue_drop(sk, req, prev);
435 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
436 goto out;
437
438 case TCP_SYN_SENT:
439 case TCP_SYN_RECV: /* Cannot happen.
440 It can, it SYNs are crossed. --ANK */
441 if (!sock_owned_by_user(sk)) {
442 sk->sk_err = err;
443 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
444
445 tcp_done(sk);
446 } else
447 sk->sk_err_soft = err;
448 goto out;
449 }
450
451 if (!sock_owned_by_user(sk) && np->recverr) {
452 sk->sk_err = err;
453 sk->sk_error_report(sk);
454 } else
455 sk->sk_err_soft = err;
456
457out:
458 bh_unlock_sock(sk);
459 sock_put(sk);
460}
461
462
463static int tcp_v6_send_synack(struct sock *sk, struct dst_entry *dst,
464 struct flowi6 *fl6,
465 struct request_sock *req,
466 u16 queue_mapping)
467{
468 struct inet_request_sock *ireq = inet_rsk(req);
469 struct ipv6_pinfo *np = inet6_sk(sk);
470 struct sk_buff *skb;
471 int err = -ENOMEM;
472
473 /* First, grab a route. */
474 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
475 goto done;
476
477 skb = tcp_make_synack(sk, dst, req, NULL);
478
479 if (skb) {
480 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
481 &ireq->ir_v6_rmt_addr);
482
483 fl6->daddr = ireq->ir_v6_rmt_addr;
484 if (np->repflow && (ireq->pktopts != NULL))
485 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
486
487 skb_set_queue_mapping(skb, queue_mapping);
488 err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass);
489 err = net_xmit_eval(err);
490 }
491
492done:
493 return err;
494}
495
496static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req)
497{
498 struct flowi6 fl6;
499 int res;
500
501 res = tcp_v6_send_synack(sk, NULL, &fl6, req, 0);
502 if (!res) {
503 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
504 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
505 }
506 return res;
507}
508
509static void tcp_v6_reqsk_destructor(struct request_sock *req)
510{
511 kfree_skb(inet_rsk(req)->pktopts);
512}
513
514#ifdef CONFIG_TCP_MD5SIG
515static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
516 const struct in6_addr *addr)
517{
518 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
519}
520
521static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
522 struct sock *addr_sk)
523{
524 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
525}
526
527static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
528 struct request_sock *req)
529{
530 return tcp_v6_md5_do_lookup(sk, &inet_rsk(req)->ir_v6_rmt_addr);
531}
532
533static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
534 int optlen)
535{
536 struct tcp_md5sig cmd;
537 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
538
539 if (optlen < sizeof(cmd))
540 return -EINVAL;
541
542 if (copy_from_user(&cmd, optval, sizeof(cmd)))
543 return -EFAULT;
544
545 if (sin6->sin6_family != AF_INET6)
546 return -EINVAL;
547
548 if (!cmd.tcpm_keylen) {
549 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
550 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
551 AF_INET);
552 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
553 AF_INET6);
554 }
555
556 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
557 return -EINVAL;
558
559 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
560 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
561 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
562
563 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
564 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
565}
566
567static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
568 const struct in6_addr *daddr,
569 const struct in6_addr *saddr, int nbytes)
570{
571 struct tcp6_pseudohdr *bp;
572 struct scatterlist sg;
573
574 bp = &hp->md5_blk.ip6;
575 /* 1. TCP pseudo-header (RFC2460) */
576 bp->saddr = *saddr;
577 bp->daddr = *daddr;
578 bp->protocol = cpu_to_be32(IPPROTO_TCP);
579 bp->len = cpu_to_be32(nbytes);
580
581 sg_init_one(&sg, bp, sizeof(*bp));
582 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
583}
584
585static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
586 const struct in6_addr *daddr, struct in6_addr *saddr,
587 const struct tcphdr *th)
588{
589 struct tcp_md5sig_pool *hp;
590 struct hash_desc *desc;
591
592 hp = tcp_get_md5sig_pool();
593 if (!hp)
594 goto clear_hash_noput;
595 desc = &hp->md5_desc;
596
597 if (crypto_hash_init(desc))
598 goto clear_hash;
599 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
600 goto clear_hash;
601 if (tcp_md5_hash_header(hp, th))
602 goto clear_hash;
603 if (tcp_md5_hash_key(hp, key))
604 goto clear_hash;
605 if (crypto_hash_final(desc, md5_hash))
606 goto clear_hash;
607
608 tcp_put_md5sig_pool();
609 return 0;
610
611clear_hash:
612 tcp_put_md5sig_pool();
613clear_hash_noput:
614 memset(md5_hash, 0, 16);
615 return 1;
616}
617
618static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
619 const struct sock *sk,
620 const struct request_sock *req,
621 const struct sk_buff *skb)
622{
623 const struct in6_addr *saddr, *daddr;
624 struct tcp_md5sig_pool *hp;
625 struct hash_desc *desc;
626 const struct tcphdr *th = tcp_hdr(skb);
627
628 if (sk) {
629 saddr = &inet6_sk(sk)->saddr;
630 daddr = &sk->sk_v6_daddr;
631 } else if (req) {
632 saddr = &inet_rsk(req)->ir_v6_loc_addr;
633 daddr = &inet_rsk(req)->ir_v6_rmt_addr;
634 } else {
635 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
636 saddr = &ip6h->saddr;
637 daddr = &ip6h->daddr;
638 }
639
640 hp = tcp_get_md5sig_pool();
641 if (!hp)
642 goto clear_hash_noput;
643 desc = &hp->md5_desc;
644
645 if (crypto_hash_init(desc))
646 goto clear_hash;
647
648 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
649 goto clear_hash;
650 if (tcp_md5_hash_header(hp, th))
651 goto clear_hash;
652 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
653 goto clear_hash;
654 if (tcp_md5_hash_key(hp, key))
655 goto clear_hash;
656 if (crypto_hash_final(desc, md5_hash))
657 goto clear_hash;
658
659 tcp_put_md5sig_pool();
660 return 0;
661
662clear_hash:
663 tcp_put_md5sig_pool();
664clear_hash_noput:
665 memset(md5_hash, 0, 16);
666 return 1;
667}
668
669static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
670{
671 const __u8 *hash_location = NULL;
672 struct tcp_md5sig_key *hash_expected;
673 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
674 const struct tcphdr *th = tcp_hdr(skb);
675 int genhash;
676 u8 newhash[16];
677
678 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
679 hash_location = tcp_parse_md5sig_option(th);
680
681 /* We've parsed the options - do we have a hash? */
682 if (!hash_expected && !hash_location)
683 return 0;
684
685 if (hash_expected && !hash_location) {
686 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
687 return 1;
688 }
689
690 if (!hash_expected && hash_location) {
691 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
692 return 1;
693 }
694
695 /* check the signature */
696 genhash = tcp_v6_md5_hash_skb(newhash,
697 hash_expected,
698 NULL, NULL, skb);
699
700 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
701 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
702 genhash ? "failed" : "mismatch",
703 &ip6h->saddr, ntohs(th->source),
704 &ip6h->daddr, ntohs(th->dest));
705 return 1;
706 }
707 return 0;
708}
709#endif
710
711struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
712 .family = AF_INET6,
713 .obj_size = sizeof(struct tcp6_request_sock),
714 .rtx_syn_ack = tcp_v6_rtx_synack,
715 .send_ack = tcp_v6_reqsk_send_ack,
716 .destructor = tcp_v6_reqsk_destructor,
717 .send_reset = tcp_v6_send_reset,
718 .syn_ack_timeout = tcp_syn_ack_timeout,
719};
720
721#ifdef CONFIG_TCP_MD5SIG
722static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
723 .md5_lookup = tcp_v6_reqsk_md5_lookup,
724 .calc_md5_hash = tcp_v6_md5_hash_skb,
725};
726#endif
727
728static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
729 u32 tsval, u32 tsecr, int oif,
730 struct tcp_md5sig_key *key, int rst, u8 tclass,
731 u32 label)
732{
733 const struct tcphdr *th = tcp_hdr(skb);
734 struct tcphdr *t1;
735 struct sk_buff *buff;
736 struct flowi6 fl6;
737 struct net *net = dev_net(skb_dst(skb)->dev);
738 struct sock *ctl_sk = net->ipv6.tcp_sk;
739 unsigned int tot_len = sizeof(struct tcphdr);
740 struct dst_entry *dst;
741 __be32 *topt;
742
743 if (tsecr)
744 tot_len += TCPOLEN_TSTAMP_ALIGNED;
745#ifdef CONFIG_TCP_MD5SIG
746 if (key)
747 tot_len += TCPOLEN_MD5SIG_ALIGNED;
748#endif
749
750 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
751 GFP_ATOMIC);
752 if (buff == NULL)
753 return;
754
755 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
756
757 t1 = (struct tcphdr *) skb_push(buff, tot_len);
758 skb_reset_transport_header(buff);
759
760 /* Swap the send and the receive. */
761 memset(t1, 0, sizeof(*t1));
762 t1->dest = th->source;
763 t1->source = th->dest;
764 t1->doff = tot_len / 4;
765 t1->seq = htonl(seq);
766 t1->ack_seq = htonl(ack);
767 t1->ack = !rst || !th->ack;
768 t1->rst = rst;
769 t1->window = htons(win);
770
771 topt = (__be32 *)(t1 + 1);
772
773 if (tsecr) {
774 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
775 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
776 *topt++ = htonl(tsval);
777 *topt++ = htonl(tsecr);
778 }
779
780#ifdef CONFIG_TCP_MD5SIG
781 if (key) {
782 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
783 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
784 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
785 &ipv6_hdr(skb)->saddr,
786 &ipv6_hdr(skb)->daddr, t1);
787 }
788#endif
789
790 memset(&fl6, 0, sizeof(fl6));
791 fl6.daddr = ipv6_hdr(skb)->saddr;
792 fl6.saddr = ipv6_hdr(skb)->daddr;
793 fl6.flowlabel = label;
794
795 buff->ip_summed = CHECKSUM_PARTIAL;
796 buff->csum = 0;
797
798 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
799
800 fl6.flowi6_proto = IPPROTO_TCP;
801 if (rt6_need_strict(&fl6.daddr) && !oif)
802 fl6.flowi6_oif = inet6_iif(skb);
803 else
804 fl6.flowi6_oif = oif;
805 fl6.fl6_dport = t1->dest;
806 fl6.fl6_sport = t1->source;
807 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
808
809 /* Pass a socket to ip6_dst_lookup either it is for RST
810 * Underlying function will use this to retrieve the network
811 * namespace
812 */
813 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
814 if (!IS_ERR(dst)) {
815 skb_dst_set(buff, dst);
816 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
817 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
818 if (rst)
819 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
820 return;
821 }
822
823 kfree_skb(buff);
824}
825
826static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
827{
828 const struct tcphdr *th = tcp_hdr(skb);
829 u32 seq = 0, ack_seq = 0;
830 struct tcp_md5sig_key *key = NULL;
831#ifdef CONFIG_TCP_MD5SIG
832 const __u8 *hash_location = NULL;
833 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
834 unsigned char newhash[16];
835 int genhash;
836 struct sock *sk1 = NULL;
837#endif
838 int oif;
839
840 if (th->rst)
841 return;
842
843 if (!ipv6_unicast_destination(skb))
844 return;
845
846#ifdef CONFIG_TCP_MD5SIG
847 hash_location = tcp_parse_md5sig_option(th);
848 if (!sk && hash_location) {
849 /*
850 * active side is lost. Try to find listening socket through
851 * source port, and then find md5 key through listening socket.
852 * we are not loose security here:
853 * Incoming packet is checked with md5 hash with finding key,
854 * no RST generated if md5 hash doesn't match.
855 */
856 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
857 &tcp_hashinfo, &ipv6h->saddr,
858 th->source, &ipv6h->daddr,
859 ntohs(th->source), inet6_iif(skb));
860 if (!sk1)
861 return;
862
863 rcu_read_lock();
864 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
865 if (!key)
866 goto release_sk1;
867
868 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
869 if (genhash || memcmp(hash_location, newhash, 16) != 0)
870 goto release_sk1;
871 } else {
872 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
873 }
874#endif
875
876 if (th->ack)
877 seq = ntohl(th->ack_seq);
878 else
879 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
880 (th->doff << 2);
881
882 oif = sk ? sk->sk_bound_dev_if : 0;
883 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
884
885#ifdef CONFIG_TCP_MD5SIG
886release_sk1:
887 if (sk1) {
888 rcu_read_unlock();
889 sock_put(sk1);
890 }
891#endif
892}
893
894static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
895 u32 win, u32 tsval, u32 tsecr, int oif,
896 struct tcp_md5sig_key *key, u8 tclass,
897 u32 label)
898{
899 tcp_v6_send_response(skb, seq, ack, win, tsval, tsecr, oif, key, 0, tclass,
900 label);
901}
902
903static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
904{
905 struct inet_timewait_sock *tw = inet_twsk(sk);
906 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
907
908 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
909 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
910 tcp_time_stamp + tcptw->tw_ts_offset,
911 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
912 tw->tw_tclass, (tw->tw_flowlabel << 12));
913
914 inet_twsk_put(tw);
915}
916
917static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
918 struct request_sock *req)
919{
920 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1,
921 req->rcv_wnd, tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
922 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
923 0, 0);
924}
925
926
927static struct sock *tcp_v6_hnd_req(struct sock *sk, struct sk_buff *skb)
928{
929 struct request_sock *req, **prev;
930 const struct tcphdr *th = tcp_hdr(skb);
931 struct sock *nsk;
932
933 /* Find possible connection requests. */
934 req = inet6_csk_search_req(sk, &prev, th->source,
935 &ipv6_hdr(skb)->saddr,
936 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
937 if (req)
938 return tcp_check_req(sk, skb, req, prev, false);
939
940 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
941 &ipv6_hdr(skb)->saddr, th->source,
942 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
943
944 if (nsk) {
945 if (nsk->sk_state != TCP_TIME_WAIT) {
946 bh_lock_sock(nsk);
947 return nsk;
948 }
949 inet_twsk_put(inet_twsk(nsk));
950 return NULL;
951 }
952
953#ifdef CONFIG_SYN_COOKIES
954 if (!th->syn)
955 sk = cookie_v6_check(sk, skb);
956#endif
957 return sk;
958}
959
960/* FIXME: this is substantially similar to the ipv4 code.
961 * Can some kind of merge be done? -- erics
962 */
963static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
964{
965 struct tcp_options_received tmp_opt;
966 struct request_sock *req;
967 struct inet_request_sock *ireq;
968 struct ipv6_pinfo *np = inet6_sk(sk);
969 struct tcp_sock *tp = tcp_sk(sk);
970 __u32 isn = TCP_SKB_CB(skb)->when;
971 struct dst_entry *dst = NULL;
972 struct flowi6 fl6;
973 bool want_cookie = false;
974
975 if (skb->protocol == htons(ETH_P_IP))
976 return tcp_v4_conn_request(sk, skb);
977
978 if (!ipv6_unicast_destination(skb))
979 goto drop;
980
981 if ((sysctl_tcp_syncookies == 2 ||
982 inet_csk_reqsk_queue_is_full(sk)) && !isn) {
983 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
984 if (!want_cookie)
985 goto drop;
986 }
987
988 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
989 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
990 goto drop;
991 }
992
993 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
994 if (req == NULL)
995 goto drop;
996
997#ifdef CONFIG_TCP_MD5SIG
998 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
999#endif
1000
1001 tcp_clear_options(&tmp_opt);
1002 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1003 tmp_opt.user_mss = tp->rx_opt.user_mss;
1004 tcp_parse_options(skb, &tmp_opt, 0, NULL);
1005
1006 if (want_cookie && !tmp_opt.saw_tstamp)
1007 tcp_clear_options(&tmp_opt);
1008
1009 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1010 tcp_openreq_init(req, &tmp_opt, skb);
1011
1012 ireq = inet_rsk(req);
1013 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
1014 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
1015 if (!want_cookie || tmp_opt.tstamp_ok)
1016 TCP_ECN_create_request(req, skb, sock_net(sk));
1017
1018 ireq->ir_iif = sk->sk_bound_dev_if;
1019
1020 /* So that link locals have meaning */
1021 if (!sk->sk_bound_dev_if &&
1022 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
1023 ireq->ir_iif = inet6_iif(skb);
1024
1025 if (!isn) {
1026 if (ipv6_opt_accepted(sk, skb) ||
1027 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1028 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim ||
1029 np->repflow) {
1030 atomic_inc(&skb->users);
1031 ireq->pktopts = skb;
1032 }
1033
1034 if (want_cookie) {
1035 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1036 req->cookie_ts = tmp_opt.tstamp_ok;
1037 goto have_isn;
1038 }
1039
1040 /* VJ's idea. We save last timestamp seen
1041 * from the destination in peer table, when entering
1042 * state TIME-WAIT, and check against it before
1043 * accepting new connection request.
1044 *
1045 * If "isn" is not zero, this request hit alive
1046 * timewait bucket, so that all the necessary checks
1047 * are made in the function processing timewait state.
1048 */
1049 if (tmp_opt.saw_tstamp &&
1050 tcp_death_row.sysctl_tw_recycle &&
1051 (dst = inet6_csk_route_req(sk, &fl6, req)) != NULL) {
1052 if (!tcp_peer_is_proven(req, dst, true)) {
1053 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1054 goto drop_and_release;
1055 }
1056 }
1057 /* Kill the following clause, if you dislike this way. */
1058 else if (!sysctl_tcp_syncookies &&
1059 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1060 (sysctl_max_syn_backlog >> 2)) &&
1061 !tcp_peer_is_proven(req, dst, false)) {
1062 /* Without syncookies last quarter of
1063 * backlog is filled with destinations,
1064 * proven to be alive.
1065 * It means that we continue to communicate
1066 * to destinations, already remembered
1067 * to the moment of synflood.
1068 */
1069 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1070 &ireq->ir_v6_rmt_addr, ntohs(tcp_hdr(skb)->source));
1071 goto drop_and_release;
1072 }
1073
1074 isn = tcp_v6_init_sequence(skb);
1075 }
1076have_isn:
1077 tcp_rsk(req)->snt_isn = isn;
1078
1079 if (security_inet_conn_request(sk, skb, req))
1080 goto drop_and_release;
1081
1082 if (tcp_v6_send_synack(sk, dst, &fl6, req,
1083 skb_get_queue_mapping(skb)) ||
1084 want_cookie)
1085 goto drop_and_free;
1086
1087 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1088 tcp_rsk(req)->listener = NULL;
1089 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1090 return 0;
1091
1092drop_and_release:
1093 dst_release(dst);
1094drop_and_free:
1095 reqsk_free(req);
1096drop:
1097 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1098 return 0; /* don't send reset */
1099}
1100
1101static struct sock *tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1102 struct request_sock *req,
1103 struct dst_entry *dst)
1104{
1105 struct inet_request_sock *ireq;
1106 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1107 struct tcp6_sock *newtcp6sk;
1108 struct inet_sock *newinet;
1109 struct tcp_sock *newtp;
1110 struct sock *newsk;
1111#ifdef CONFIG_TCP_MD5SIG
1112 struct tcp_md5sig_key *key;
1113#endif
1114 struct flowi6 fl6;
1115
1116 if (skb->protocol == htons(ETH_P_IP)) {
1117 /*
1118 * v6 mapped
1119 */
1120
1121 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1122
1123 if (newsk == NULL)
1124 return NULL;
1125
1126 newtcp6sk = (struct tcp6_sock *)newsk;
1127 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1128
1129 newinet = inet_sk(newsk);
1130 newnp = inet6_sk(newsk);
1131 newtp = tcp_sk(newsk);
1132
1133 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1134
1135 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newsk->sk_v6_daddr);
1136
1137 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1138
1139 newsk->sk_v6_rcv_saddr = newnp->saddr;
1140
1141 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1142 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1143#ifdef CONFIG_TCP_MD5SIG
1144 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1145#endif
1146
1147 newnp->ipv6_ac_list = NULL;
1148 newnp->ipv6_fl_list = NULL;
1149 newnp->pktoptions = NULL;
1150 newnp->opt = NULL;
1151 newnp->mcast_oif = inet6_iif(skb);
1152 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1153 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1154 if (np->repflow)
1155 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1156
1157 /*
1158 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1159 * here, tcp_create_openreq_child now does this for us, see the comment in
1160 * that function for the gory details. -acme
1161 */
1162
1163 /* It is tricky place. Until this moment IPv4 tcp
1164 worked with IPv6 icsk.icsk_af_ops.
1165 Sync it now.
1166 */
1167 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1168
1169 return newsk;
1170 }
1171
1172 ireq = inet_rsk(req);
1173
1174 if (sk_acceptq_is_full(sk))
1175 goto out_overflow;
1176
1177 if (!dst) {
1178 dst = inet6_csk_route_req(sk, &fl6, req);
1179 if (!dst)
1180 goto out;
1181 }
1182
1183 newsk = tcp_create_openreq_child(sk, req, skb);
1184 if (newsk == NULL)
1185 goto out_nonewsk;
1186
1187 /*
1188 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1189 * count here, tcp_create_openreq_child now does this for us, see the
1190 * comment in that function for the gory details. -acme
1191 */
1192
1193 newsk->sk_gso_type = SKB_GSO_TCPV6;
1194 __ip6_dst_store(newsk, dst, NULL, NULL);
1195 inet6_sk_rx_dst_set(newsk, skb);
1196
1197 newtcp6sk = (struct tcp6_sock *)newsk;
1198 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1199
1200 newtp = tcp_sk(newsk);
1201 newinet = inet_sk(newsk);
1202 newnp = inet6_sk(newsk);
1203
1204 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1205
1206 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1207 newnp->saddr = ireq->ir_v6_loc_addr;
1208 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1209 newsk->sk_bound_dev_if = ireq->ir_iif;
1210
1211 /* Now IPv6 options...
1212
1213 First: no IPv4 options.
1214 */
1215 newinet->inet_opt = NULL;
1216 newnp->ipv6_ac_list = NULL;
1217 newnp->ipv6_fl_list = NULL;
1218
1219 /* Clone RX bits */
1220 newnp->rxopt.all = np->rxopt.all;
1221
1222 /* Clone pktoptions received with SYN */
1223 newnp->pktoptions = NULL;
1224 if (ireq->pktopts != NULL) {
1225 newnp->pktoptions = skb_clone(ireq->pktopts,
1226 sk_gfp_atomic(sk, GFP_ATOMIC));
1227 consume_skb(ireq->pktopts);
1228 ireq->pktopts = NULL;
1229 if (newnp->pktoptions)
1230 skb_set_owner_r(newnp->pktoptions, newsk);
1231 }
1232 newnp->opt = NULL;
1233 newnp->mcast_oif = inet6_iif(skb);
1234 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1235 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1236 if (np->repflow)
1237 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1238
1239 /* Clone native IPv6 options from listening socket (if any)
1240
1241 Yes, keeping reference count would be much more clever,
1242 but we make one more one thing there: reattach optmem
1243 to newsk.
1244 */
1245 if (np->opt)
1246 newnp->opt = ipv6_dup_options(newsk, np->opt);
1247
1248 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1249 if (newnp->opt)
1250 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1251 newnp->opt->opt_flen);
1252
1253 tcp_sync_mss(newsk, dst_mtu(dst));
1254 newtp->advmss = dst_metric_advmss(dst);
1255 if (tcp_sk(sk)->rx_opt.user_mss &&
1256 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1257 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1258
1259 tcp_initialize_rcv_mss(newsk);
1260
1261 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1262 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1263
1264#ifdef CONFIG_TCP_MD5SIG
1265 /* Copy over the MD5 key from the original socket */
1266 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1267 if (key != NULL) {
1268 /* We're using one, so create a matching key
1269 * on the newsk structure. If we fail to get
1270 * memory, then we end up not copying the key
1271 * across. Shucks.
1272 */
1273 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1274 AF_INET6, key->key, key->keylen,
1275 sk_gfp_atomic(sk, GFP_ATOMIC));
1276 }
1277#endif
1278
1279 if (__inet_inherit_port(sk, newsk) < 0) {
1280 inet_csk_prepare_forced_close(newsk);
1281 tcp_done(newsk);
1282 goto out;
1283 }
1284 __inet6_hash(newsk, NULL);
1285
1286 return newsk;
1287
1288out_overflow:
1289 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1290out_nonewsk:
1291 dst_release(dst);
1292out:
1293 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1294 return NULL;
1295}
1296
1297static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1298{
1299 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1300 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1301 &ipv6_hdr(skb)->daddr, skb->csum)) {
1302 skb->ip_summed = CHECKSUM_UNNECESSARY;
1303 return 0;
1304 }
1305 }
1306
1307 skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1308 &ipv6_hdr(skb)->saddr,
1309 &ipv6_hdr(skb)->daddr, 0));
1310
1311 if (skb->len <= 76)
1312 return __skb_checksum_complete(skb);
1313 return 0;
1314}
1315
1316/* The socket must have it's spinlock held when we get
1317 * here.
1318 *
1319 * We have a potential double-lock case here, so even when
1320 * doing backlog processing we use the BH locking scheme.
1321 * This is because we cannot sleep with the original spinlock
1322 * held.
1323 */
1324static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1325{
1326 struct ipv6_pinfo *np = inet6_sk(sk);
1327 struct tcp_sock *tp;
1328 struct sk_buff *opt_skb = NULL;
1329
1330 /* Imagine: socket is IPv6. IPv4 packet arrives,
1331 goes to IPv4 receive handler and backlogged.
1332 From backlog it always goes here. Kerboom...
1333 Fortunately, tcp_rcv_established and rcv_established
1334 handle them correctly, but it is not case with
1335 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1336 */
1337
1338 if (skb->protocol == htons(ETH_P_IP))
1339 return tcp_v4_do_rcv(sk, skb);
1340
1341#ifdef CONFIG_TCP_MD5SIG
1342 if (tcp_v6_inbound_md5_hash(sk, skb))
1343 goto discard;
1344#endif
1345
1346 if (sk_filter(sk, skb))
1347 goto discard;
1348
1349 /*
1350 * socket locking is here for SMP purposes as backlog rcv
1351 * is currently called with bh processing disabled.
1352 */
1353
1354 /* Do Stevens' IPV6_PKTOPTIONS.
1355
1356 Yes, guys, it is the only place in our code, where we
1357 may make it not affecting IPv4.
1358 The rest of code is protocol independent,
1359 and I do not like idea to uglify IPv4.
1360
1361 Actually, all the idea behind IPV6_PKTOPTIONS
1362 looks not very well thought. For now we latch
1363 options, received in the last packet, enqueued
1364 by tcp. Feel free to propose better solution.
1365 --ANK (980728)
1366 */
1367 if (np->rxopt.all)
1368 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1369
1370 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1371 struct dst_entry *dst = sk->sk_rx_dst;
1372
1373 sock_rps_save_rxhash(sk, skb);
1374 if (dst) {
1375 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1376 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1377 dst_release(dst);
1378 sk->sk_rx_dst = NULL;
1379 }
1380 }
1381
1382 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1383 if (opt_skb)
1384 goto ipv6_pktoptions;
1385 return 0;
1386 }
1387
1388 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1389 goto csum_err;
1390
1391 if (sk->sk_state == TCP_LISTEN) {
1392 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1393 if (!nsk)
1394 goto discard;
1395
1396 /*
1397 * Queue it on the new socket if the new socket is active,
1398 * otherwise we just shortcircuit this and continue with
1399 * the new socket..
1400 */
1401 if (nsk != sk) {
1402 sock_rps_save_rxhash(nsk, skb);
1403 if (tcp_child_process(sk, nsk, skb))
1404 goto reset;
1405 if (opt_skb)
1406 __kfree_skb(opt_skb);
1407 return 0;
1408 }
1409 } else
1410 sock_rps_save_rxhash(sk, skb);
1411
1412 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1413 goto reset;
1414 if (opt_skb)
1415 goto ipv6_pktoptions;
1416 return 0;
1417
1418reset:
1419 tcp_v6_send_reset(sk, skb);
1420discard:
1421 if (opt_skb)
1422 __kfree_skb(opt_skb);
1423 kfree_skb(skb);
1424 return 0;
1425csum_err:
1426 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1427 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1428 goto discard;
1429
1430
1431ipv6_pktoptions:
1432 /* Do you ask, what is it?
1433
1434 1. skb was enqueued by tcp.
1435 2. skb is added to tail of read queue, rather than out of order.
1436 3. socket is not in passive state.
1437 4. Finally, it really contains options, which user wants to receive.
1438 */
1439 tp = tcp_sk(sk);
1440 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1441 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1442 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1443 np->mcast_oif = inet6_iif(opt_skb);
1444 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1445 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1446 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1447 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1448 if (np->repflow)
1449 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1450 if (ipv6_opt_accepted(sk, opt_skb)) {
1451 skb_set_owner_r(opt_skb, sk);
1452 opt_skb = xchg(&np->pktoptions, opt_skb);
1453 } else {
1454 __kfree_skb(opt_skb);
1455 opt_skb = xchg(&np->pktoptions, NULL);
1456 }
1457 }
1458
1459 kfree_skb(opt_skb);
1460 return 0;
1461}
1462
1463static int tcp_v6_rcv(struct sk_buff *skb)
1464{
1465 const struct tcphdr *th;
1466 const struct ipv6hdr *hdr;
1467 struct sock *sk;
1468 int ret;
1469 struct net *net = dev_net(skb->dev);
1470
1471 if (skb->pkt_type != PACKET_HOST)
1472 goto discard_it;
1473
1474 /*
1475 * Count it even if it's bad.
1476 */
1477 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1478
1479 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1480 goto discard_it;
1481
1482 th = tcp_hdr(skb);
1483
1484 if (th->doff < sizeof(struct tcphdr)/4)
1485 goto bad_packet;
1486 if (!pskb_may_pull(skb, th->doff*4))
1487 goto discard_it;
1488
1489 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1490 goto csum_error;
1491
1492 th = tcp_hdr(skb);
1493 hdr = ipv6_hdr(skb);
1494 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1495 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1496 skb->len - th->doff*4);
1497 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1498 TCP_SKB_CB(skb)->when = 0;
1499 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1500 TCP_SKB_CB(skb)->sacked = 0;
1501
1502 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1503 if (!sk)
1504 goto no_tcp_socket;
1505
1506process:
1507 if (sk->sk_state == TCP_TIME_WAIT)
1508 goto do_time_wait;
1509
1510 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1511 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1512 goto discard_and_relse;
1513 }
1514
1515 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1516 goto discard_and_relse;
1517
1518 if (sk_filter(sk, skb))
1519 goto discard_and_relse;
1520
1521 sk_mark_napi_id(sk, skb);
1522 skb->dev = NULL;
1523
1524 bh_lock_sock_nested(sk);
1525 ret = 0;
1526 if (!sock_owned_by_user(sk)) {
1527#ifdef CONFIG_NET_DMA
1528 struct tcp_sock *tp = tcp_sk(sk);
1529 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1530 tp->ucopy.dma_chan = net_dma_find_channel();
1531 if (tp->ucopy.dma_chan)
1532 ret = tcp_v6_do_rcv(sk, skb);
1533 else
1534#endif
1535 {
1536 if (!tcp_prequeue(sk, skb))
1537 ret = tcp_v6_do_rcv(sk, skb);
1538 }
1539 } else if (unlikely(sk_add_backlog(sk, skb,
1540 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1541 bh_unlock_sock(sk);
1542 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1543 goto discard_and_relse;
1544 }
1545 bh_unlock_sock(sk);
1546
1547 sock_put(sk);
1548 return ret ? -1 : 0;
1549
1550no_tcp_socket:
1551 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1552 goto discard_it;
1553
1554 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1555csum_error:
1556 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1557bad_packet:
1558 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1559 } else {
1560 tcp_v6_send_reset(NULL, skb);
1561 }
1562
1563discard_it:
1564 kfree_skb(skb);
1565 return 0;
1566
1567discard_and_relse:
1568 sock_put(sk);
1569 goto discard_it;
1570
1571do_time_wait:
1572 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1573 inet_twsk_put(inet_twsk(sk));
1574 goto discard_it;
1575 }
1576
1577 if (skb->len < (th->doff<<2)) {
1578 inet_twsk_put(inet_twsk(sk));
1579 goto bad_packet;
1580 }
1581 if (tcp_checksum_complete(skb)) {
1582 inet_twsk_put(inet_twsk(sk));
1583 goto csum_error;
1584 }
1585
1586 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1587 case TCP_TW_SYN:
1588 {
1589 struct sock *sk2;
1590
1591 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1592 &ipv6_hdr(skb)->saddr, th->source,
1593 &ipv6_hdr(skb)->daddr,
1594 ntohs(th->dest), inet6_iif(skb));
1595 if (sk2 != NULL) {
1596 struct inet_timewait_sock *tw = inet_twsk(sk);
1597 inet_twsk_deschedule(tw, &tcp_death_row);
1598 inet_twsk_put(tw);
1599 sk = sk2;
1600 goto process;
1601 }
1602 /* Fall through to ACK */
1603 }
1604 case TCP_TW_ACK:
1605 tcp_v6_timewait_ack(sk, skb);
1606 break;
1607 case TCP_TW_RST:
1608 goto no_tcp_socket;
1609 case TCP_TW_SUCCESS:
1610 ;
1611 }
1612 goto discard_it;
1613}
1614
1615static void tcp_v6_early_demux(struct sk_buff *skb)
1616{
1617 const struct ipv6hdr *hdr;
1618 const struct tcphdr *th;
1619 struct sock *sk;
1620
1621 if (skb->pkt_type != PACKET_HOST)
1622 return;
1623
1624 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1625 return;
1626
1627 hdr = ipv6_hdr(skb);
1628 th = tcp_hdr(skb);
1629
1630 if (th->doff < sizeof(struct tcphdr) / 4)
1631 return;
1632
1633 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1634 &hdr->saddr, th->source,
1635 &hdr->daddr, ntohs(th->dest),
1636 inet6_iif(skb));
1637 if (sk) {
1638 skb->sk = sk;
1639 skb->destructor = sock_edemux;
1640 if (sk->sk_state != TCP_TIME_WAIT) {
1641 struct dst_entry *dst = sk->sk_rx_dst;
1642
1643 if (dst)
1644 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1645 if (dst &&
1646 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1647 skb_dst_set_noref(skb, dst);
1648 }
1649 }
1650}
1651
1652static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1653 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1654 .twsk_unique = tcp_twsk_unique,
1655 .twsk_destructor = tcp_twsk_destructor,
1656};
1657
1658static const struct inet_connection_sock_af_ops ipv6_specific = {
1659 .queue_xmit = inet6_csk_xmit,
1660 .send_check = tcp_v6_send_check,
1661 .rebuild_header = inet6_sk_rebuild_header,
1662 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1663 .conn_request = tcp_v6_conn_request,
1664 .syn_recv_sock = tcp_v6_syn_recv_sock,
1665 .net_header_len = sizeof(struct ipv6hdr),
1666 .net_frag_header_len = sizeof(struct frag_hdr),
1667 .setsockopt = ipv6_setsockopt,
1668 .getsockopt = ipv6_getsockopt,
1669 .addr2sockaddr = inet6_csk_addr2sockaddr,
1670 .sockaddr_len = sizeof(struct sockaddr_in6),
1671 .bind_conflict = inet6_csk_bind_conflict,
1672#ifdef CONFIG_COMPAT
1673 .compat_setsockopt = compat_ipv6_setsockopt,
1674 .compat_getsockopt = compat_ipv6_getsockopt,
1675#endif
1676};
1677
1678#ifdef CONFIG_TCP_MD5SIG
1679static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1680 .md5_lookup = tcp_v6_md5_lookup,
1681 .calc_md5_hash = tcp_v6_md5_hash_skb,
1682 .md5_parse = tcp_v6_parse_md5_keys,
1683};
1684#endif
1685
1686/*
1687 * TCP over IPv4 via INET6 API
1688 */
1689static const struct inet_connection_sock_af_ops ipv6_mapped = {
1690 .queue_xmit = ip_queue_xmit,
1691 .send_check = tcp_v4_send_check,
1692 .rebuild_header = inet_sk_rebuild_header,
1693 .sk_rx_dst_set = inet_sk_rx_dst_set,
1694 .conn_request = tcp_v6_conn_request,
1695 .syn_recv_sock = tcp_v6_syn_recv_sock,
1696 .net_header_len = sizeof(struct iphdr),
1697 .setsockopt = ipv6_setsockopt,
1698 .getsockopt = ipv6_getsockopt,
1699 .addr2sockaddr = inet6_csk_addr2sockaddr,
1700 .sockaddr_len = sizeof(struct sockaddr_in6),
1701 .bind_conflict = inet6_csk_bind_conflict,
1702#ifdef CONFIG_COMPAT
1703 .compat_setsockopt = compat_ipv6_setsockopt,
1704 .compat_getsockopt = compat_ipv6_getsockopt,
1705#endif
1706};
1707
1708#ifdef CONFIG_TCP_MD5SIG
1709static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1710 .md5_lookup = tcp_v4_md5_lookup,
1711 .calc_md5_hash = tcp_v4_md5_hash_skb,
1712 .md5_parse = tcp_v6_parse_md5_keys,
1713};
1714#endif
1715
1716/* NOTE: A lot of things set to zero explicitly by call to
1717 * sk_alloc() so need not be done here.
1718 */
1719static int tcp_v6_init_sock(struct sock *sk)
1720{
1721 struct inet_connection_sock *icsk = inet_csk(sk);
1722
1723 tcp_init_sock(sk);
1724
1725 icsk->icsk_af_ops = &ipv6_specific;
1726
1727#ifdef CONFIG_TCP_MD5SIG
1728 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1729#endif
1730
1731 return 0;
1732}
1733
1734static void tcp_v6_destroy_sock(struct sock *sk)
1735{
1736 tcp_v4_destroy_sock(sk);
1737 inet6_destroy_sock(sk);
1738}
1739
1740#ifdef CONFIG_PROC_FS
1741/* Proc filesystem TCPv6 sock list dumping. */
1742static void get_openreq6(struct seq_file *seq,
1743 const struct sock *sk, struct request_sock *req, int i, kuid_t uid)
1744{
1745 int ttd = req->expires - jiffies;
1746 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1747 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1748
1749 if (ttd < 0)
1750 ttd = 0;
1751
1752 seq_printf(seq,
1753 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1754 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1755 i,
1756 src->s6_addr32[0], src->s6_addr32[1],
1757 src->s6_addr32[2], src->s6_addr32[3],
1758 inet_rsk(req)->ir_num,
1759 dest->s6_addr32[0], dest->s6_addr32[1],
1760 dest->s6_addr32[2], dest->s6_addr32[3],
1761 ntohs(inet_rsk(req)->ir_rmt_port),
1762 TCP_SYN_RECV,
1763 0, 0, /* could print option size, but that is af dependent. */
1764 1, /* timers active (only the expire timer) */
1765 jiffies_to_clock_t(ttd),
1766 req->num_timeout,
1767 from_kuid_munged(seq_user_ns(seq), uid),
1768 0, /* non standard timer */
1769 0, /* open_requests have no inode */
1770 0, req);
1771}
1772
1773static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1774{
1775 const struct in6_addr *dest, *src;
1776 __u16 destp, srcp;
1777 int timer_active;
1778 unsigned long timer_expires;
1779 const struct inet_sock *inet = inet_sk(sp);
1780 const struct tcp_sock *tp = tcp_sk(sp);
1781 const struct inet_connection_sock *icsk = inet_csk(sp);
1782
1783 dest = &sp->sk_v6_daddr;
1784 src = &sp->sk_v6_rcv_saddr;
1785 destp = ntohs(inet->inet_dport);
1786 srcp = ntohs(inet->inet_sport);
1787
1788 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1789 timer_active = 1;
1790 timer_expires = icsk->icsk_timeout;
1791 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1792 timer_active = 4;
1793 timer_expires = icsk->icsk_timeout;
1794 } else if (timer_pending(&sp->sk_timer)) {
1795 timer_active = 2;
1796 timer_expires = sp->sk_timer.expires;
1797 } else {
1798 timer_active = 0;
1799 timer_expires = jiffies;
1800 }
1801
1802 seq_printf(seq,
1803 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1804 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1805 i,
1806 src->s6_addr32[0], src->s6_addr32[1],
1807 src->s6_addr32[2], src->s6_addr32[3], srcp,
1808 dest->s6_addr32[0], dest->s6_addr32[1],
1809 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1810 sp->sk_state,
1811 tp->write_seq-tp->snd_una,
1812 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1813 timer_active,
1814 jiffies_delta_to_clock_t(timer_expires - jiffies),
1815 icsk->icsk_retransmits,
1816 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1817 icsk->icsk_probes_out,
1818 sock_i_ino(sp),
1819 atomic_read(&sp->sk_refcnt), sp,
1820 jiffies_to_clock_t(icsk->icsk_rto),
1821 jiffies_to_clock_t(icsk->icsk_ack.ato),
1822 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1823 tp->snd_cwnd,
1824 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
1825 );
1826}
1827
1828static void get_timewait6_sock(struct seq_file *seq,
1829 struct inet_timewait_sock *tw, int i)
1830{
1831 const struct in6_addr *dest, *src;
1832 __u16 destp, srcp;
1833 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
1834
1835 dest = &tw->tw_v6_daddr;
1836 src = &tw->tw_v6_rcv_saddr;
1837 destp = ntohs(tw->tw_dport);
1838 srcp = ntohs(tw->tw_sport);
1839
1840 seq_printf(seq,
1841 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1842 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1843 i,
1844 src->s6_addr32[0], src->s6_addr32[1],
1845 src->s6_addr32[2], src->s6_addr32[3], srcp,
1846 dest->s6_addr32[0], dest->s6_addr32[1],
1847 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1848 tw->tw_substate, 0, 0,
1849 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1850 atomic_read(&tw->tw_refcnt), tw);
1851}
1852
1853static int tcp6_seq_show(struct seq_file *seq, void *v)
1854{
1855 struct tcp_iter_state *st;
1856 struct sock *sk = v;
1857
1858 if (v == SEQ_START_TOKEN) {
1859 seq_puts(seq,
1860 " sl "
1861 "local_address "
1862 "remote_address "
1863 "st tx_queue rx_queue tr tm->when retrnsmt"
1864 " uid timeout inode\n");
1865 goto out;
1866 }
1867 st = seq->private;
1868
1869 switch (st->state) {
1870 case TCP_SEQ_STATE_LISTENING:
1871 case TCP_SEQ_STATE_ESTABLISHED:
1872 if (sk->sk_state == TCP_TIME_WAIT)
1873 get_timewait6_sock(seq, v, st->num);
1874 else
1875 get_tcp6_sock(seq, v, st->num);
1876 break;
1877 case TCP_SEQ_STATE_OPENREQ:
1878 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1879 break;
1880 }
1881out:
1882 return 0;
1883}
1884
1885static const struct file_operations tcp6_afinfo_seq_fops = {
1886 .owner = THIS_MODULE,
1887 .open = tcp_seq_open,
1888 .read = seq_read,
1889 .llseek = seq_lseek,
1890 .release = seq_release_net
1891};
1892
1893static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1894 .name = "tcp6",
1895 .family = AF_INET6,
1896 .seq_fops = &tcp6_afinfo_seq_fops,
1897 .seq_ops = {
1898 .show = tcp6_seq_show,
1899 },
1900};
1901
1902int __net_init tcp6_proc_init(struct net *net)
1903{
1904 return tcp_proc_register(net, &tcp6_seq_afinfo);
1905}
1906
1907void tcp6_proc_exit(struct net *net)
1908{
1909 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1910}
1911#endif
1912
1913static void tcp_v6_clear_sk(struct sock *sk, int size)
1914{
1915 struct inet_sock *inet = inet_sk(sk);
1916
1917 /* we do not want to clear pinet6 field, because of RCU lookups */
1918 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1919
1920 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1921 memset(&inet->pinet6 + 1, 0, size);
1922}
1923
1924struct proto tcpv6_prot = {
1925 .name = "TCPv6",
1926 .owner = THIS_MODULE,
1927 .close = tcp_close,
1928 .connect = tcp_v6_connect,
1929 .disconnect = tcp_disconnect,
1930 .accept = inet_csk_accept,
1931 .ioctl = tcp_ioctl,
1932 .init = tcp_v6_init_sock,
1933 .destroy = tcp_v6_destroy_sock,
1934 .shutdown = tcp_shutdown,
1935 .setsockopt = tcp_setsockopt,
1936 .getsockopt = tcp_getsockopt,
1937 .recvmsg = tcp_recvmsg,
1938 .sendmsg = tcp_sendmsg,
1939 .sendpage = tcp_sendpage,
1940 .backlog_rcv = tcp_v6_do_rcv,
1941 .release_cb = tcp_release_cb,
1942 .mtu_reduced = tcp_v6_mtu_reduced,
1943 .hash = tcp_v6_hash,
1944 .unhash = inet_unhash,
1945 .get_port = inet_csk_get_port,
1946 .enter_memory_pressure = tcp_enter_memory_pressure,
1947 .stream_memory_free = tcp_stream_memory_free,
1948 .sockets_allocated = &tcp_sockets_allocated,
1949 .memory_allocated = &tcp_memory_allocated,
1950 .memory_pressure = &tcp_memory_pressure,
1951 .orphan_count = &tcp_orphan_count,
1952 .sysctl_mem = sysctl_tcp_mem,
1953 .sysctl_wmem = sysctl_tcp_wmem,
1954 .sysctl_rmem = sysctl_tcp_rmem,
1955 .max_header = MAX_TCP_HEADER,
1956 .obj_size = sizeof(struct tcp6_sock),
1957 .slab_flags = SLAB_DESTROY_BY_RCU,
1958 .twsk_prot = &tcp6_timewait_sock_ops,
1959 .rsk_prot = &tcp6_request_sock_ops,
1960 .h.hashinfo = &tcp_hashinfo,
1961 .no_autobind = true,
1962#ifdef CONFIG_COMPAT
1963 .compat_setsockopt = compat_tcp_setsockopt,
1964 .compat_getsockopt = compat_tcp_getsockopt,
1965#endif
1966#ifdef CONFIG_MEMCG_KMEM
1967 .proto_cgroup = tcp_proto_cgroup,
1968#endif
1969 .clear_sk = tcp_v6_clear_sk,
1970};
1971
1972static const struct inet6_protocol tcpv6_protocol = {
1973 .early_demux = tcp_v6_early_demux,
1974 .handler = tcp_v6_rcv,
1975 .err_handler = tcp_v6_err,
1976 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1977};
1978
1979static struct inet_protosw tcpv6_protosw = {
1980 .type = SOCK_STREAM,
1981 .protocol = IPPROTO_TCP,
1982 .prot = &tcpv6_prot,
1983 .ops = &inet6_stream_ops,
1984 .no_check = 0,
1985 .flags = INET_PROTOSW_PERMANENT |
1986 INET_PROTOSW_ICSK,
1987};
1988
1989static int __net_init tcpv6_net_init(struct net *net)
1990{
1991 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1992 SOCK_RAW, IPPROTO_TCP, net);
1993}
1994
1995static void __net_exit tcpv6_net_exit(struct net *net)
1996{
1997 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1998}
1999
2000static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2001{
2002 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2003}
2004
2005static struct pernet_operations tcpv6_net_ops = {
2006 .init = tcpv6_net_init,
2007 .exit = tcpv6_net_exit,
2008 .exit_batch = tcpv6_net_exit_batch,
2009};
2010
2011int __init tcpv6_init(void)
2012{
2013 int ret;
2014
2015 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2016 if (ret)
2017 goto out;
2018
2019 /* register inet6 protocol */
2020 ret = inet6_register_protosw(&tcpv6_protosw);
2021 if (ret)
2022 goto out_tcpv6_protocol;
2023
2024 ret = register_pernet_subsys(&tcpv6_net_ops);
2025 if (ret)
2026 goto out_tcpv6_protosw;
2027out:
2028 return ret;
2029
2030out_tcpv6_protosw:
2031 inet6_unregister_protosw(&tcpv6_protosw);
2032out_tcpv6_protocol:
2033 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2034 goto out;
2035}
2036
2037void tcpv6_exit(void)
2038{
2039 unregister_pernet_subsys(&tcpv6_net_ops);
2040 inet6_unregister_protosw(&tcpv6_protosw);
2041 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2042}