net: vrf: Flip IPv6 output path from FIB lookup hook to out hook
[cascardo/linux.git] / net / ipv6 / tcp_ipv6.c
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
72 static void     tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
73 static void     tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
74                                       struct request_sock *req);
75
76 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
77
78 static const struct inet_connection_sock_af_ops ipv6_mapped;
79 static const struct inet_connection_sock_af_ops ipv6_specific;
80 #ifdef CONFIG_TCP_MD5SIG
81 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
82 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
83 #else
84 static 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
91 static 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
104 static __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
112 static 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, lockdep_sock_is_held(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
293 late_failure:
294         tcp_set_state(sk, TCP_CLOSE);
295         __sk_dst_reset(sk);
296 failure:
297         inet->inet_dport = 0;
298         sk->sk_route_caps = 0;
299         return err;
300 }
301
302 static 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
319 static 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(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(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(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(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
432 out:
433         bh_unlock_sock(sk);
434         sock_put(sk);
435 }
436
437
438 static 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                               enum tcp_synack_type synack_type)
443 {
444         struct inet_request_sock *ireq = inet_rsk(req);
445         struct ipv6_pinfo *np = inet6_sk(sk);
446         struct ipv6_txoptions *opt;
447         struct flowi6 *fl6 = &fl->u.ip6;
448         struct sk_buff *skb;
449         int err = -ENOMEM;
450
451         /* First, grab a route. */
452         if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
453                                                IPPROTO_TCP)) == NULL)
454                 goto done;
455
456         skb = tcp_make_synack(sk, dst, req, foc, synack_type);
457
458         if (skb) {
459                 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
460                                     &ireq->ir_v6_rmt_addr);
461
462                 fl6->daddr = ireq->ir_v6_rmt_addr;
463                 if (np->repflow && ireq->pktopts)
464                         fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
465
466                 rcu_read_lock();
467                 opt = ireq->ipv6_opt;
468                 if (!opt)
469                         opt = rcu_dereference(np->opt);
470                 err = ip6_xmit(sk, skb, fl6, opt, np->tclass);
471                 rcu_read_unlock();
472                 err = net_xmit_eval(err);
473         }
474
475 done:
476         return err;
477 }
478
479
480 static void tcp_v6_reqsk_destructor(struct request_sock *req)
481 {
482         kfree(inet_rsk(req)->ipv6_opt);
483         kfree_skb(inet_rsk(req)->pktopts);
484 }
485
486 #ifdef CONFIG_TCP_MD5SIG
487 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
488                                                    const struct in6_addr *addr)
489 {
490         return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
491 }
492
493 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
494                                                 const struct sock *addr_sk)
495 {
496         return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
497 }
498
499 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
500                                  int optlen)
501 {
502         struct tcp_md5sig cmd;
503         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
504
505         if (optlen < sizeof(cmd))
506                 return -EINVAL;
507
508         if (copy_from_user(&cmd, optval, sizeof(cmd)))
509                 return -EFAULT;
510
511         if (sin6->sin6_family != AF_INET6)
512                 return -EINVAL;
513
514         if (!cmd.tcpm_keylen) {
515                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
516                         return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
517                                               AF_INET);
518                 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
519                                       AF_INET6);
520         }
521
522         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
523                 return -EINVAL;
524
525         if (ipv6_addr_v4mapped(&sin6->sin6_addr))
526                 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
527                                       AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
528
529         return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
530                               AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
531 }
532
533 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
534                                    const struct in6_addr *daddr,
535                                    const struct in6_addr *saddr,
536                                    const struct tcphdr *th, int nbytes)
537 {
538         struct tcp6_pseudohdr *bp;
539         struct scatterlist sg;
540         struct tcphdr *_th;
541
542         bp = hp->scratch;
543         /* 1. TCP pseudo-header (RFC2460) */
544         bp->saddr = *saddr;
545         bp->daddr = *daddr;
546         bp->protocol = cpu_to_be32(IPPROTO_TCP);
547         bp->len = cpu_to_be32(nbytes);
548
549         _th = (struct tcphdr *)(bp + 1);
550         memcpy(_th, th, sizeof(*th));
551         _th->check = 0;
552
553         sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
554         ahash_request_set_crypt(hp->md5_req, &sg, NULL,
555                                 sizeof(*bp) + sizeof(*th));
556         return crypto_ahash_update(hp->md5_req);
557 }
558
559 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
560                                const struct in6_addr *daddr, struct in6_addr *saddr,
561                                const struct tcphdr *th)
562 {
563         struct tcp_md5sig_pool *hp;
564         struct ahash_request *req;
565
566         hp = tcp_get_md5sig_pool();
567         if (!hp)
568                 goto clear_hash_noput;
569         req = hp->md5_req;
570
571         if (crypto_ahash_init(req))
572                 goto clear_hash;
573         if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
574                 goto clear_hash;
575         if (tcp_md5_hash_key(hp, key))
576                 goto clear_hash;
577         ahash_request_set_crypt(req, NULL, md5_hash, 0);
578         if (crypto_ahash_final(req))
579                 goto clear_hash;
580
581         tcp_put_md5sig_pool();
582         return 0;
583
584 clear_hash:
585         tcp_put_md5sig_pool();
586 clear_hash_noput:
587         memset(md5_hash, 0, 16);
588         return 1;
589 }
590
591 static int tcp_v6_md5_hash_skb(char *md5_hash,
592                                const struct tcp_md5sig_key *key,
593                                const struct sock *sk,
594                                const struct sk_buff *skb)
595 {
596         const struct in6_addr *saddr, *daddr;
597         struct tcp_md5sig_pool *hp;
598         struct ahash_request *req;
599         const struct tcphdr *th = tcp_hdr(skb);
600
601         if (sk) { /* valid for establish/request sockets */
602                 saddr = &sk->sk_v6_rcv_saddr;
603                 daddr = &sk->sk_v6_daddr;
604         } else {
605                 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
606                 saddr = &ip6h->saddr;
607                 daddr = &ip6h->daddr;
608         }
609
610         hp = tcp_get_md5sig_pool();
611         if (!hp)
612                 goto clear_hash_noput;
613         req = hp->md5_req;
614
615         if (crypto_ahash_init(req))
616                 goto clear_hash;
617
618         if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
619                 goto clear_hash;
620         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
621                 goto clear_hash;
622         if (tcp_md5_hash_key(hp, key))
623                 goto clear_hash;
624         ahash_request_set_crypt(req, NULL, md5_hash, 0);
625         if (crypto_ahash_final(req))
626                 goto clear_hash;
627
628         tcp_put_md5sig_pool();
629         return 0;
630
631 clear_hash:
632         tcp_put_md5sig_pool();
633 clear_hash_noput:
634         memset(md5_hash, 0, 16);
635         return 1;
636 }
637
638 #endif
639
640 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
641                                     const struct sk_buff *skb)
642 {
643 #ifdef CONFIG_TCP_MD5SIG
644         const __u8 *hash_location = NULL;
645         struct tcp_md5sig_key *hash_expected;
646         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
647         const struct tcphdr *th = tcp_hdr(skb);
648         int genhash;
649         u8 newhash[16];
650
651         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
652         hash_location = tcp_parse_md5sig_option(th);
653
654         /* We've parsed the options - do we have a hash? */
655         if (!hash_expected && !hash_location)
656                 return false;
657
658         if (hash_expected && !hash_location) {
659                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
660                 return true;
661         }
662
663         if (!hash_expected && hash_location) {
664                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
665                 return true;
666         }
667
668         /* check the signature */
669         genhash = tcp_v6_md5_hash_skb(newhash,
670                                       hash_expected,
671                                       NULL, skb);
672
673         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
674                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
675                 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
676                                      genhash ? "failed" : "mismatch",
677                                      &ip6h->saddr, ntohs(th->source),
678                                      &ip6h->daddr, ntohs(th->dest));
679                 return true;
680         }
681 #endif
682         return false;
683 }
684
685 static void tcp_v6_init_req(struct request_sock *req,
686                             const struct sock *sk_listener,
687                             struct sk_buff *skb)
688 {
689         struct inet_request_sock *ireq = inet_rsk(req);
690         const struct ipv6_pinfo *np = inet6_sk(sk_listener);
691
692         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
693         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
694
695         /* So that link locals have meaning */
696         if (!sk_listener->sk_bound_dev_if &&
697             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
698                 ireq->ir_iif = tcp_v6_iif(skb);
699
700         if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
701             (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
702              np->rxopt.bits.rxinfo ||
703              np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
704              np->rxopt.bits.rxohlim || np->repflow)) {
705                 atomic_inc(&skb->users);
706                 ireq->pktopts = skb;
707         }
708 }
709
710 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
711                                           struct flowi *fl,
712                                           const struct request_sock *req,
713                                           bool *strict)
714 {
715         if (strict)
716                 *strict = true;
717         return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
718 }
719
720 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
721         .family         =       AF_INET6,
722         .obj_size       =       sizeof(struct tcp6_request_sock),
723         .rtx_syn_ack    =       tcp_rtx_synack,
724         .send_ack       =       tcp_v6_reqsk_send_ack,
725         .destructor     =       tcp_v6_reqsk_destructor,
726         .send_reset     =       tcp_v6_send_reset,
727         .syn_ack_timeout =      tcp_syn_ack_timeout,
728 };
729
730 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
731         .mss_clamp      =       IPV6_MIN_MTU - sizeof(struct tcphdr) -
732                                 sizeof(struct ipv6hdr),
733 #ifdef CONFIG_TCP_MD5SIG
734         .req_md5_lookup =       tcp_v6_md5_lookup,
735         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
736 #endif
737         .init_req       =       tcp_v6_init_req,
738 #ifdef CONFIG_SYN_COOKIES
739         .cookie_init_seq =      cookie_v6_init_sequence,
740 #endif
741         .route_req      =       tcp_v6_route_req,
742         .init_seq       =       tcp_v6_init_sequence,
743         .send_synack    =       tcp_v6_send_synack,
744 };
745
746 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
747                                  u32 ack, u32 win, u32 tsval, u32 tsecr,
748                                  int oif, struct tcp_md5sig_key *key, int rst,
749                                  u8 tclass, __be32 label)
750 {
751         const struct tcphdr *th = tcp_hdr(skb);
752         struct tcphdr *t1;
753         struct sk_buff *buff;
754         struct flowi6 fl6;
755         struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
756         struct sock *ctl_sk = net->ipv6.tcp_sk;
757         unsigned int tot_len = sizeof(struct tcphdr);
758         struct dst_entry *dst;
759         __be32 *topt;
760
761         if (tsecr)
762                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
763 #ifdef CONFIG_TCP_MD5SIG
764         if (key)
765                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
766 #endif
767
768         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
769                          GFP_ATOMIC);
770         if (!buff)
771                 return;
772
773         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
774
775         t1 = (struct tcphdr *) skb_push(buff, tot_len);
776         skb_reset_transport_header(buff);
777
778         /* Swap the send and the receive. */
779         memset(t1, 0, sizeof(*t1));
780         t1->dest = th->source;
781         t1->source = th->dest;
782         t1->doff = tot_len / 4;
783         t1->seq = htonl(seq);
784         t1->ack_seq = htonl(ack);
785         t1->ack = !rst || !th->ack;
786         t1->rst = rst;
787         t1->window = htons(win);
788
789         topt = (__be32 *)(t1 + 1);
790
791         if (tsecr) {
792                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
793                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
794                 *topt++ = htonl(tsval);
795                 *topt++ = htonl(tsecr);
796         }
797
798 #ifdef CONFIG_TCP_MD5SIG
799         if (key) {
800                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
801                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
802                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
803                                     &ipv6_hdr(skb)->saddr,
804                                     &ipv6_hdr(skb)->daddr, t1);
805         }
806 #endif
807
808         memset(&fl6, 0, sizeof(fl6));
809         fl6.daddr = ipv6_hdr(skb)->saddr;
810         fl6.saddr = ipv6_hdr(skb)->daddr;
811         fl6.flowlabel = label;
812
813         buff->ip_summed = CHECKSUM_PARTIAL;
814         buff->csum = 0;
815
816         __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
817
818         fl6.flowi6_proto = IPPROTO_TCP;
819         if (rt6_need_strict(&fl6.daddr) && !oif)
820                 fl6.flowi6_oif = tcp_v6_iif(skb);
821         else {
822                 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
823                         oif = skb->skb_iif;
824
825                 fl6.flowi6_oif = oif;
826         }
827
828         fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
829         fl6.fl6_dport = t1->dest;
830         fl6.fl6_sport = t1->source;
831         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
832
833         /* Pass a socket to ip6_dst_lookup either it is for RST
834          * Underlying function will use this to retrieve the network
835          * namespace
836          */
837         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
838         if (!IS_ERR(dst)) {
839                 skb_dst_set(buff, dst);
840                 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
841                 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
842                 if (rst)
843                         TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
844                 return;
845         }
846
847         kfree_skb(buff);
848 }
849
850 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
851 {
852         const struct tcphdr *th = tcp_hdr(skb);
853         u32 seq = 0, ack_seq = 0;
854         struct tcp_md5sig_key *key = NULL;
855 #ifdef CONFIG_TCP_MD5SIG
856         const __u8 *hash_location = NULL;
857         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
858         unsigned char newhash[16];
859         int genhash;
860         struct sock *sk1 = NULL;
861 #endif
862         int oif;
863
864         if (th->rst)
865                 return;
866
867         /* If sk not NULL, it means we did a successful lookup and incoming
868          * route had to be correct. prequeue might have dropped our dst.
869          */
870         if (!sk && !ipv6_unicast_destination(skb))
871                 return;
872
873 #ifdef CONFIG_TCP_MD5SIG
874         rcu_read_lock();
875         hash_location = tcp_parse_md5sig_option(th);
876         if (sk && sk_fullsock(sk)) {
877                 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
878         } else if (hash_location) {
879                 /*
880                  * active side is lost. Try to find listening socket through
881                  * source port, and then find md5 key through listening socket.
882                  * we are not loose security here:
883                  * Incoming packet is checked with md5 hash with finding key,
884                  * no RST generated if md5 hash doesn't match.
885                  */
886                 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
887                                            &tcp_hashinfo, NULL, 0,
888                                            &ipv6h->saddr,
889                                            th->source, &ipv6h->daddr,
890                                            ntohs(th->source), tcp_v6_iif(skb));
891                 if (!sk1)
892                         goto out;
893
894                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
895                 if (!key)
896                         goto out;
897
898                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
899                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
900                         goto out;
901         }
902 #endif
903
904         if (th->ack)
905                 seq = ntohl(th->ack_seq);
906         else
907                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
908                           (th->doff << 2);
909
910         oif = sk ? sk->sk_bound_dev_if : 0;
911         tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
912
913 #ifdef CONFIG_TCP_MD5SIG
914 out:
915         rcu_read_unlock();
916 #endif
917 }
918
919 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
920                             u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
921                             struct tcp_md5sig_key *key, u8 tclass,
922                             __be32 label)
923 {
924         tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
925                              tclass, label);
926 }
927
928 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
929 {
930         struct inet_timewait_sock *tw = inet_twsk(sk);
931         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
932
933         tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
934                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
935                         tcp_time_stamp + tcptw->tw_ts_offset,
936                         tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
937                         tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
938
939         inet_twsk_put(tw);
940 }
941
942 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
943                                   struct request_sock *req)
944 {
945         /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
946          * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
947          */
948         /* RFC 7323 2.3
949          * The window field (SEG.WND) of every outgoing segment, with the
950          * exception of <SYN> segments, MUST be right-shifted by
951          * Rcv.Wind.Shift bits:
952          */
953         tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
954                         tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
955                         tcp_rsk(req)->rcv_nxt,
956                         req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
957                         tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
958                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
959                         0, 0);
960 }
961
962
963 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
964 {
965 #ifdef CONFIG_SYN_COOKIES
966         const struct tcphdr *th = tcp_hdr(skb);
967
968         if (!th->syn)
969                 sk = cookie_v6_check(sk, skb);
970 #endif
971         return sk;
972 }
973
974 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
975 {
976         if (skb->protocol == htons(ETH_P_IP))
977                 return tcp_v4_conn_request(sk, skb);
978
979         if (!ipv6_unicast_destination(skb))
980                 goto drop;
981
982         return tcp_conn_request(&tcp6_request_sock_ops,
983                                 &tcp_request_sock_ipv6_ops, sk, skb);
984
985 drop:
986         tcp_listendrop(sk);
987         return 0; /* don't send reset */
988 }
989
990 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
991                                          struct request_sock *req,
992                                          struct dst_entry *dst,
993                                          struct request_sock *req_unhash,
994                                          bool *own_req)
995 {
996         struct inet_request_sock *ireq;
997         struct ipv6_pinfo *newnp;
998         const struct ipv6_pinfo *np = inet6_sk(sk);
999         struct ipv6_txoptions *opt;
1000         struct tcp6_sock *newtcp6sk;
1001         struct inet_sock *newinet;
1002         struct tcp_sock *newtp;
1003         struct sock *newsk;
1004 #ifdef CONFIG_TCP_MD5SIG
1005         struct tcp_md5sig_key *key;
1006 #endif
1007         struct flowi6 fl6;
1008
1009         if (skb->protocol == htons(ETH_P_IP)) {
1010                 /*
1011                  *      v6 mapped
1012                  */
1013
1014                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1015                                              req_unhash, own_req);
1016
1017                 if (!newsk)
1018                         return NULL;
1019
1020                 newtcp6sk = (struct tcp6_sock *)newsk;
1021                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1022
1023                 newinet = inet_sk(newsk);
1024                 newnp = inet6_sk(newsk);
1025                 newtp = tcp_sk(newsk);
1026
1027                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1028
1029                 newnp->saddr = newsk->sk_v6_rcv_saddr;
1030
1031                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1032                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1033 #ifdef CONFIG_TCP_MD5SIG
1034                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1035 #endif
1036
1037                 newnp->ipv6_ac_list = NULL;
1038                 newnp->ipv6_fl_list = NULL;
1039                 newnp->pktoptions  = NULL;
1040                 newnp->opt         = NULL;
1041                 newnp->mcast_oif   = tcp_v6_iif(skb);
1042                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1043                 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1044                 if (np->repflow)
1045                         newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1046
1047                 /*
1048                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1049                  * here, tcp_create_openreq_child now does this for us, see the comment in
1050                  * that function for the gory details. -acme
1051                  */
1052
1053                 /* It is tricky place. Until this moment IPv4 tcp
1054                    worked with IPv6 icsk.icsk_af_ops.
1055                    Sync it now.
1056                  */
1057                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1058
1059                 return newsk;
1060         }
1061
1062         ireq = inet_rsk(req);
1063
1064         if (sk_acceptq_is_full(sk))
1065                 goto out_overflow;
1066
1067         if (!dst) {
1068                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1069                 if (!dst)
1070                         goto out;
1071         }
1072
1073         newsk = tcp_create_openreq_child(sk, req, skb);
1074         if (!newsk)
1075                 goto out_nonewsk;
1076
1077         /*
1078          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1079          * count here, tcp_create_openreq_child now does this for us, see the
1080          * comment in that function for the gory details. -acme
1081          */
1082
1083         newsk->sk_gso_type = SKB_GSO_TCPV6;
1084         ip6_dst_store(newsk, dst, NULL, NULL);
1085         inet6_sk_rx_dst_set(newsk, skb);
1086
1087         newtcp6sk = (struct tcp6_sock *)newsk;
1088         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1089
1090         newtp = tcp_sk(newsk);
1091         newinet = inet_sk(newsk);
1092         newnp = inet6_sk(newsk);
1093
1094         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1095
1096         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1097         newnp->saddr = ireq->ir_v6_loc_addr;
1098         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1099         newsk->sk_bound_dev_if = ireq->ir_iif;
1100
1101         /* Now IPv6 options...
1102
1103            First: no IPv4 options.
1104          */
1105         newinet->inet_opt = NULL;
1106         newnp->ipv6_ac_list = NULL;
1107         newnp->ipv6_fl_list = NULL;
1108
1109         /* Clone RX bits */
1110         newnp->rxopt.all = np->rxopt.all;
1111
1112         newnp->pktoptions = NULL;
1113         newnp->opt        = NULL;
1114         newnp->mcast_oif  = tcp_v6_iif(skb);
1115         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1116         newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1117         if (np->repflow)
1118                 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1119
1120         /* Clone native IPv6 options from listening socket (if any)
1121
1122            Yes, keeping reference count would be much more clever,
1123            but we make one more one thing there: reattach optmem
1124            to newsk.
1125          */
1126         opt = ireq->ipv6_opt;
1127         if (!opt)
1128                 opt = rcu_dereference(np->opt);
1129         if (opt) {
1130                 opt = ipv6_dup_options(newsk, opt);
1131                 RCU_INIT_POINTER(newnp->opt, opt);
1132         }
1133         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1134         if (opt)
1135                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1136                                                     opt->opt_flen;
1137
1138         tcp_ca_openreq_child(newsk, dst);
1139
1140         tcp_sync_mss(newsk, dst_mtu(dst));
1141         newtp->advmss = dst_metric_advmss(dst);
1142         if (tcp_sk(sk)->rx_opt.user_mss &&
1143             tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1144                 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1145
1146         tcp_initialize_rcv_mss(newsk);
1147
1148         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1149         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1150
1151 #ifdef CONFIG_TCP_MD5SIG
1152         /* Copy over the MD5 key from the original socket */
1153         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1154         if (key) {
1155                 /* We're using one, so create a matching key
1156                  * on the newsk structure. If we fail to get
1157                  * memory, then we end up not copying the key
1158                  * across. Shucks.
1159                  */
1160                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1161                                AF_INET6, key->key, key->keylen,
1162                                sk_gfp_mask(sk, GFP_ATOMIC));
1163         }
1164 #endif
1165
1166         if (__inet_inherit_port(sk, newsk) < 0) {
1167                 inet_csk_prepare_forced_close(newsk);
1168                 tcp_done(newsk);
1169                 goto out;
1170         }
1171         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1172         if (*own_req) {
1173                 tcp_move_syn(newtp, req);
1174
1175                 /* Clone pktoptions received with SYN, if we own the req */
1176                 if (ireq->pktopts) {
1177                         newnp->pktoptions = skb_clone(ireq->pktopts,
1178                                                       sk_gfp_mask(sk, GFP_ATOMIC));
1179                         consume_skb(ireq->pktopts);
1180                         ireq->pktopts = NULL;
1181                         if (newnp->pktoptions)
1182                                 skb_set_owner_r(newnp->pktoptions, newsk);
1183                 }
1184         }
1185
1186         return newsk;
1187
1188 out_overflow:
1189         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1190 out_nonewsk:
1191         dst_release(dst);
1192 out:
1193         tcp_listendrop(sk);
1194         return NULL;
1195 }
1196
1197 /* The socket must have it's spinlock held when we get
1198  * here, unless it is a TCP_LISTEN socket.
1199  *
1200  * We have a potential double-lock case here, so even when
1201  * doing backlog processing we use the BH locking scheme.
1202  * This is because we cannot sleep with the original spinlock
1203  * held.
1204  */
1205 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1206 {
1207         struct ipv6_pinfo *np = inet6_sk(sk);
1208         struct tcp_sock *tp;
1209         struct sk_buff *opt_skb = NULL;
1210
1211         /* Imagine: socket is IPv6. IPv4 packet arrives,
1212            goes to IPv4 receive handler and backlogged.
1213            From backlog it always goes here. Kerboom...
1214            Fortunately, tcp_rcv_established and rcv_established
1215            handle them correctly, but it is not case with
1216            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1217          */
1218
1219         if (skb->protocol == htons(ETH_P_IP))
1220                 return tcp_v4_do_rcv(sk, skb);
1221
1222         if (sk_filter(sk, skb))
1223                 goto discard;
1224
1225         /*
1226          *      socket locking is here for SMP purposes as backlog rcv
1227          *      is currently called with bh processing disabled.
1228          */
1229
1230         /* Do Stevens' IPV6_PKTOPTIONS.
1231
1232            Yes, guys, it is the only place in our code, where we
1233            may make it not affecting IPv4.
1234            The rest of code is protocol independent,
1235            and I do not like idea to uglify IPv4.
1236
1237            Actually, all the idea behind IPV6_PKTOPTIONS
1238            looks not very well thought. For now we latch
1239            options, received in the last packet, enqueued
1240            by tcp. Feel free to propose better solution.
1241                                                --ANK (980728)
1242          */
1243         if (np->rxopt.all)
1244                 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1245
1246         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1247                 struct dst_entry *dst = sk->sk_rx_dst;
1248
1249                 sock_rps_save_rxhash(sk, skb);
1250                 sk_mark_napi_id(sk, skb);
1251                 if (dst) {
1252                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1253                             dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1254                                 dst_release(dst);
1255                                 sk->sk_rx_dst = NULL;
1256                         }
1257                 }
1258
1259                 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1260                 if (opt_skb)
1261                         goto ipv6_pktoptions;
1262                 return 0;
1263         }
1264
1265         if (tcp_checksum_complete(skb))
1266                 goto csum_err;
1267
1268         if (sk->sk_state == TCP_LISTEN) {
1269                 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1270
1271                 if (!nsk)
1272                         goto discard;
1273
1274                 if (nsk != sk) {
1275                         sock_rps_save_rxhash(nsk, skb);
1276                         sk_mark_napi_id(nsk, skb);
1277                         if (tcp_child_process(sk, nsk, skb))
1278                                 goto reset;
1279                         if (opt_skb)
1280                                 __kfree_skb(opt_skb);
1281                         return 0;
1282                 }
1283         } else
1284                 sock_rps_save_rxhash(sk, skb);
1285
1286         if (tcp_rcv_state_process(sk, skb))
1287                 goto reset;
1288         if (opt_skb)
1289                 goto ipv6_pktoptions;
1290         return 0;
1291
1292 reset:
1293         tcp_v6_send_reset(sk, skb);
1294 discard:
1295         if (opt_skb)
1296                 __kfree_skb(opt_skb);
1297         kfree_skb(skb);
1298         return 0;
1299 csum_err:
1300         TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1301         TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1302         goto discard;
1303
1304
1305 ipv6_pktoptions:
1306         /* Do you ask, what is it?
1307
1308            1. skb was enqueued by tcp.
1309            2. skb is added to tail of read queue, rather than out of order.
1310            3. socket is not in passive state.
1311            4. Finally, it really contains options, which user wants to receive.
1312          */
1313         tp = tcp_sk(sk);
1314         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1315             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1316                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1317                         np->mcast_oif = tcp_v6_iif(opt_skb);
1318                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1319                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1320                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1321                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1322                 if (np->repflow)
1323                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1324                 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1325                         skb_set_owner_r(opt_skb, sk);
1326                         opt_skb = xchg(&np->pktoptions, opt_skb);
1327                 } else {
1328                         __kfree_skb(opt_skb);
1329                         opt_skb = xchg(&np->pktoptions, NULL);
1330                 }
1331         }
1332
1333         kfree_skb(opt_skb);
1334         return 0;
1335 }
1336
1337 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1338                            const struct tcphdr *th)
1339 {
1340         /* This is tricky: we move IP6CB at its correct location into
1341          * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1342          * _decode_session6() uses IP6CB().
1343          * barrier() makes sure compiler won't play aliasing games.
1344          */
1345         memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1346                 sizeof(struct inet6_skb_parm));
1347         barrier();
1348
1349         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1350         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1351                                     skb->len - th->doff*4);
1352         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1353         TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1354         TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1355         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1356         TCP_SKB_CB(skb)->sacked = 0;
1357 }
1358
1359 static void tcp_v6_restore_cb(struct sk_buff *skb)
1360 {
1361         /* We need to move header back to the beginning if xfrm6_policy_check()
1362          * and tcp_v6_fill_cb() are going to be called again.
1363          */
1364         memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1365                 sizeof(struct inet6_skb_parm));
1366 }
1367
1368 static int tcp_v6_rcv(struct sk_buff *skb)
1369 {
1370         const struct tcphdr *th;
1371         const struct ipv6hdr *hdr;
1372         bool refcounted;
1373         struct sock *sk;
1374         int ret;
1375         struct net *net = dev_net(skb->dev);
1376
1377         if (skb->pkt_type != PACKET_HOST)
1378                 goto discard_it;
1379
1380         /*
1381          *      Count it even if it's bad.
1382          */
1383         __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1384
1385         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1386                 goto discard_it;
1387
1388         th = (const struct tcphdr *)skb->data;
1389
1390         if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1391                 goto bad_packet;
1392         if (!pskb_may_pull(skb, th->doff*4))
1393                 goto discard_it;
1394
1395         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1396                 goto csum_error;
1397
1398         th = (const struct tcphdr *)skb->data;
1399         hdr = ipv6_hdr(skb);
1400
1401 lookup:
1402         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1403                                 th->source, th->dest, inet6_iif(skb),
1404                                 &refcounted);
1405         if (!sk)
1406                 goto no_tcp_socket;
1407
1408 process:
1409         if (sk->sk_state == TCP_TIME_WAIT)
1410                 goto do_time_wait;
1411
1412         if (sk->sk_state == TCP_NEW_SYN_RECV) {
1413                 struct request_sock *req = inet_reqsk(sk);
1414                 struct sock *nsk;
1415
1416                 sk = req->rsk_listener;
1417                 tcp_v6_fill_cb(skb, hdr, th);
1418                 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1419                         sk_drops_add(sk, skb);
1420                         reqsk_put(req);
1421                         goto discard_it;
1422                 }
1423                 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1424                         inet_csk_reqsk_queue_drop_and_put(sk, req);
1425                         goto lookup;
1426                 }
1427                 sock_hold(sk);
1428                 refcounted = true;
1429                 nsk = tcp_check_req(sk, skb, req, false);
1430                 if (!nsk) {
1431                         reqsk_put(req);
1432                         goto discard_and_relse;
1433                 }
1434                 if (nsk == sk) {
1435                         reqsk_put(req);
1436                         tcp_v6_restore_cb(skb);
1437                 } else if (tcp_child_process(sk, nsk, skb)) {
1438                         tcp_v6_send_reset(nsk, skb);
1439                         goto discard_and_relse;
1440                 } else {
1441                         sock_put(sk);
1442                         return 0;
1443                 }
1444         }
1445         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1446                 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1447                 goto discard_and_relse;
1448         }
1449
1450         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1451                 goto discard_and_relse;
1452
1453         tcp_v6_fill_cb(skb, hdr, th);
1454
1455         if (tcp_v6_inbound_md5_hash(sk, skb))
1456                 goto discard_and_relse;
1457
1458         if (sk_filter(sk, skb))
1459                 goto discard_and_relse;
1460
1461         skb->dev = NULL;
1462
1463         if (sk->sk_state == TCP_LISTEN) {
1464                 ret = tcp_v6_do_rcv(sk, skb);
1465                 goto put_and_return;
1466         }
1467
1468         sk_incoming_cpu_update(sk);
1469
1470         bh_lock_sock_nested(sk);
1471         tcp_segs_in(tcp_sk(sk), skb);
1472         ret = 0;
1473         if (!sock_owned_by_user(sk)) {
1474                 if (!tcp_prequeue(sk, skb))
1475                         ret = tcp_v6_do_rcv(sk, skb);
1476         } else if (tcp_add_backlog(sk, skb)) {
1477                 goto discard_and_relse;
1478         }
1479         bh_unlock_sock(sk);
1480
1481 put_and_return:
1482         if (refcounted)
1483                 sock_put(sk);
1484         return ret ? -1 : 0;
1485
1486 no_tcp_socket:
1487         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1488                 goto discard_it;
1489
1490         tcp_v6_fill_cb(skb, hdr, th);
1491
1492         if (tcp_checksum_complete(skb)) {
1493 csum_error:
1494                 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1495 bad_packet:
1496                 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1497         } else {
1498                 tcp_v6_send_reset(NULL, skb);
1499         }
1500
1501 discard_it:
1502         kfree_skb(skb);
1503         return 0;
1504
1505 discard_and_relse:
1506         sk_drops_add(sk, skb);
1507         if (refcounted)
1508                 sock_put(sk);
1509         goto discard_it;
1510
1511 do_time_wait:
1512         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1513                 inet_twsk_put(inet_twsk(sk));
1514                 goto discard_it;
1515         }
1516
1517         tcp_v6_fill_cb(skb, hdr, th);
1518
1519         if (tcp_checksum_complete(skb)) {
1520                 inet_twsk_put(inet_twsk(sk));
1521                 goto csum_error;
1522         }
1523
1524         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1525         case TCP_TW_SYN:
1526         {
1527                 struct sock *sk2;
1528
1529                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1530                                             skb, __tcp_hdrlen(th),
1531                                             &ipv6_hdr(skb)->saddr, th->source,
1532                                             &ipv6_hdr(skb)->daddr,
1533                                             ntohs(th->dest), tcp_v6_iif(skb));
1534                 if (sk2) {
1535                         struct inet_timewait_sock *tw = inet_twsk(sk);
1536                         inet_twsk_deschedule_put(tw);
1537                         sk = sk2;
1538                         tcp_v6_restore_cb(skb);
1539                         refcounted = false;
1540                         goto process;
1541                 }
1542                 /* Fall through to ACK */
1543         }
1544         case TCP_TW_ACK:
1545                 tcp_v6_timewait_ack(sk, skb);
1546                 break;
1547         case TCP_TW_RST:
1548                 tcp_v6_restore_cb(skb);
1549                 tcp_v6_send_reset(sk, skb);
1550                 inet_twsk_deschedule_put(inet_twsk(sk));
1551                 goto discard_it;
1552         case TCP_TW_SUCCESS:
1553                 ;
1554         }
1555         goto discard_it;
1556 }
1557
1558 static void tcp_v6_early_demux(struct sk_buff *skb)
1559 {
1560         const struct ipv6hdr *hdr;
1561         const struct tcphdr *th;
1562         struct sock *sk;
1563
1564         if (skb->pkt_type != PACKET_HOST)
1565                 return;
1566
1567         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1568                 return;
1569
1570         hdr = ipv6_hdr(skb);
1571         th = tcp_hdr(skb);
1572
1573         if (th->doff < sizeof(struct tcphdr) / 4)
1574                 return;
1575
1576         /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1577         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1578                                         &hdr->saddr, th->source,
1579                                         &hdr->daddr, ntohs(th->dest),
1580                                         inet6_iif(skb));
1581         if (sk) {
1582                 skb->sk = sk;
1583                 skb->destructor = sock_edemux;
1584                 if (sk_fullsock(sk)) {
1585                         struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1586
1587                         if (dst)
1588                                 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1589                         if (dst &&
1590                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1591                                 skb_dst_set_noref(skb, dst);
1592                 }
1593         }
1594 }
1595
1596 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1597         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1598         .twsk_unique    = tcp_twsk_unique,
1599         .twsk_destructor = tcp_twsk_destructor,
1600 };
1601
1602 static const struct inet_connection_sock_af_ops ipv6_specific = {
1603         .queue_xmit        = inet6_csk_xmit,
1604         .send_check        = tcp_v6_send_check,
1605         .rebuild_header    = inet6_sk_rebuild_header,
1606         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1607         .conn_request      = tcp_v6_conn_request,
1608         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1609         .net_header_len    = sizeof(struct ipv6hdr),
1610         .net_frag_header_len = sizeof(struct frag_hdr),
1611         .setsockopt        = ipv6_setsockopt,
1612         .getsockopt        = ipv6_getsockopt,
1613         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1614         .sockaddr_len      = sizeof(struct sockaddr_in6),
1615         .bind_conflict     = inet6_csk_bind_conflict,
1616 #ifdef CONFIG_COMPAT
1617         .compat_setsockopt = compat_ipv6_setsockopt,
1618         .compat_getsockopt = compat_ipv6_getsockopt,
1619 #endif
1620         .mtu_reduced       = tcp_v6_mtu_reduced,
1621 };
1622
1623 #ifdef CONFIG_TCP_MD5SIG
1624 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1625         .md5_lookup     =       tcp_v6_md5_lookup,
1626         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1627         .md5_parse      =       tcp_v6_parse_md5_keys,
1628 };
1629 #endif
1630
1631 /*
1632  *      TCP over IPv4 via INET6 API
1633  */
1634 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1635         .queue_xmit        = ip_queue_xmit,
1636         .send_check        = tcp_v4_send_check,
1637         .rebuild_header    = inet_sk_rebuild_header,
1638         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1639         .conn_request      = tcp_v6_conn_request,
1640         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1641         .net_header_len    = sizeof(struct iphdr),
1642         .setsockopt        = ipv6_setsockopt,
1643         .getsockopt        = ipv6_getsockopt,
1644         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1645         .sockaddr_len      = sizeof(struct sockaddr_in6),
1646         .bind_conflict     = inet6_csk_bind_conflict,
1647 #ifdef CONFIG_COMPAT
1648         .compat_setsockopt = compat_ipv6_setsockopt,
1649         .compat_getsockopt = compat_ipv6_getsockopt,
1650 #endif
1651         .mtu_reduced       = tcp_v4_mtu_reduced,
1652 };
1653
1654 #ifdef CONFIG_TCP_MD5SIG
1655 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1656         .md5_lookup     =       tcp_v4_md5_lookup,
1657         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1658         .md5_parse      =       tcp_v6_parse_md5_keys,
1659 };
1660 #endif
1661
1662 /* NOTE: A lot of things set to zero explicitly by call to
1663  *       sk_alloc() so need not be done here.
1664  */
1665 static int tcp_v6_init_sock(struct sock *sk)
1666 {
1667         struct inet_connection_sock *icsk = inet_csk(sk);
1668
1669         tcp_init_sock(sk);
1670
1671         icsk->icsk_af_ops = &ipv6_specific;
1672
1673 #ifdef CONFIG_TCP_MD5SIG
1674         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1675 #endif
1676
1677         return 0;
1678 }
1679
1680 static void tcp_v6_destroy_sock(struct sock *sk)
1681 {
1682         tcp_v4_destroy_sock(sk);
1683         inet6_destroy_sock(sk);
1684 }
1685
1686 #ifdef CONFIG_PROC_FS
1687 /* Proc filesystem TCPv6 sock list dumping. */
1688 static void get_openreq6(struct seq_file *seq,
1689                          const struct request_sock *req, int i)
1690 {
1691         long ttd = req->rsk_timer.expires - jiffies;
1692         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1693         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1694
1695         if (ttd < 0)
1696                 ttd = 0;
1697
1698         seq_printf(seq,
1699                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1700                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1701                    i,
1702                    src->s6_addr32[0], src->s6_addr32[1],
1703                    src->s6_addr32[2], src->s6_addr32[3],
1704                    inet_rsk(req)->ir_num,
1705                    dest->s6_addr32[0], dest->s6_addr32[1],
1706                    dest->s6_addr32[2], dest->s6_addr32[3],
1707                    ntohs(inet_rsk(req)->ir_rmt_port),
1708                    TCP_SYN_RECV,
1709                    0, 0, /* could print option size, but that is af dependent. */
1710                    1,   /* timers active (only the expire timer) */
1711                    jiffies_to_clock_t(ttd),
1712                    req->num_timeout,
1713                    from_kuid_munged(seq_user_ns(seq),
1714                                     sock_i_uid(req->rsk_listener)),
1715                    0,  /* non standard timer */
1716                    0, /* open_requests have no inode */
1717                    0, req);
1718 }
1719
1720 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1721 {
1722         const struct in6_addr *dest, *src;
1723         __u16 destp, srcp;
1724         int timer_active;
1725         unsigned long timer_expires;
1726         const struct inet_sock *inet = inet_sk(sp);
1727         const struct tcp_sock *tp = tcp_sk(sp);
1728         const struct inet_connection_sock *icsk = inet_csk(sp);
1729         const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1730         int rx_queue;
1731         int state;
1732
1733         dest  = &sp->sk_v6_daddr;
1734         src   = &sp->sk_v6_rcv_saddr;
1735         destp = ntohs(inet->inet_dport);
1736         srcp  = ntohs(inet->inet_sport);
1737
1738         if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1739             icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
1740             icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1741                 timer_active    = 1;
1742                 timer_expires   = icsk->icsk_timeout;
1743         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1744                 timer_active    = 4;
1745                 timer_expires   = icsk->icsk_timeout;
1746         } else if (timer_pending(&sp->sk_timer)) {
1747                 timer_active    = 2;
1748                 timer_expires   = sp->sk_timer.expires;
1749         } else {
1750                 timer_active    = 0;
1751                 timer_expires = jiffies;
1752         }
1753
1754         state = sk_state_load(sp);
1755         if (state == TCP_LISTEN)
1756                 rx_queue = sp->sk_ack_backlog;
1757         else
1758                 /* Because we don't lock the socket,
1759                  * we might find a transient negative value.
1760                  */
1761                 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1762
1763         seq_printf(seq,
1764                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1765                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1766                    i,
1767                    src->s6_addr32[0], src->s6_addr32[1],
1768                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1769                    dest->s6_addr32[0], dest->s6_addr32[1],
1770                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1771                    state,
1772                    tp->write_seq - tp->snd_una,
1773                    rx_queue,
1774                    timer_active,
1775                    jiffies_delta_to_clock_t(timer_expires - jiffies),
1776                    icsk->icsk_retransmits,
1777                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1778                    icsk->icsk_probes_out,
1779                    sock_i_ino(sp),
1780                    atomic_read(&sp->sk_refcnt), sp,
1781                    jiffies_to_clock_t(icsk->icsk_rto),
1782                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1783                    (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1784                    tp->snd_cwnd,
1785                    state == TCP_LISTEN ?
1786                         fastopenq->max_qlen :
1787                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1788                    );
1789 }
1790
1791 static void get_timewait6_sock(struct seq_file *seq,
1792                                struct inet_timewait_sock *tw, int i)
1793 {
1794         long delta = tw->tw_timer.expires - jiffies;
1795         const struct in6_addr *dest, *src;
1796         __u16 destp, srcp;
1797
1798         dest = &tw->tw_v6_daddr;
1799         src  = &tw->tw_v6_rcv_saddr;
1800         destp = ntohs(tw->tw_dport);
1801         srcp  = ntohs(tw->tw_sport);
1802
1803         seq_printf(seq,
1804                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1805                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1806                    i,
1807                    src->s6_addr32[0], src->s6_addr32[1],
1808                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1809                    dest->s6_addr32[0], dest->s6_addr32[1],
1810                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1811                    tw->tw_substate, 0, 0,
1812                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1813                    atomic_read(&tw->tw_refcnt), tw);
1814 }
1815
1816 static int tcp6_seq_show(struct seq_file *seq, void *v)
1817 {
1818         struct tcp_iter_state *st;
1819         struct sock *sk = v;
1820
1821         if (v == SEQ_START_TOKEN) {
1822                 seq_puts(seq,
1823                          "  sl  "
1824                          "local_address                         "
1825                          "remote_address                        "
1826                          "st tx_queue rx_queue tr tm->when retrnsmt"
1827                          "   uid  timeout inode\n");
1828                 goto out;
1829         }
1830         st = seq->private;
1831
1832         if (sk->sk_state == TCP_TIME_WAIT)
1833                 get_timewait6_sock(seq, v, st->num);
1834         else if (sk->sk_state == TCP_NEW_SYN_RECV)
1835                 get_openreq6(seq, v, st->num);
1836         else
1837                 get_tcp6_sock(seq, v, st->num);
1838 out:
1839         return 0;
1840 }
1841
1842 static const struct file_operations tcp6_afinfo_seq_fops = {
1843         .owner   = THIS_MODULE,
1844         .open    = tcp_seq_open,
1845         .read    = seq_read,
1846         .llseek  = seq_lseek,
1847         .release = seq_release_net
1848 };
1849
1850 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1851         .name           = "tcp6",
1852         .family         = AF_INET6,
1853         .seq_fops       = &tcp6_afinfo_seq_fops,
1854         .seq_ops        = {
1855                 .show           = tcp6_seq_show,
1856         },
1857 };
1858
1859 int __net_init tcp6_proc_init(struct net *net)
1860 {
1861         return tcp_proc_register(net, &tcp6_seq_afinfo);
1862 }
1863
1864 void tcp6_proc_exit(struct net *net)
1865 {
1866         tcp_proc_unregister(net, &tcp6_seq_afinfo);
1867 }
1868 #endif
1869
1870 struct proto tcpv6_prot = {
1871         .name                   = "TCPv6",
1872         .owner                  = THIS_MODULE,
1873         .close                  = tcp_close,
1874         .connect                = tcp_v6_connect,
1875         .disconnect             = tcp_disconnect,
1876         .accept                 = inet_csk_accept,
1877         .ioctl                  = tcp_ioctl,
1878         .init                   = tcp_v6_init_sock,
1879         .destroy                = tcp_v6_destroy_sock,
1880         .shutdown               = tcp_shutdown,
1881         .setsockopt             = tcp_setsockopt,
1882         .getsockopt             = tcp_getsockopt,
1883         .recvmsg                = tcp_recvmsg,
1884         .sendmsg                = tcp_sendmsg,
1885         .sendpage               = tcp_sendpage,
1886         .backlog_rcv            = tcp_v6_do_rcv,
1887         .release_cb             = tcp_release_cb,
1888         .hash                   = inet6_hash,
1889         .unhash                 = inet_unhash,
1890         .get_port               = inet_csk_get_port,
1891         .enter_memory_pressure  = tcp_enter_memory_pressure,
1892         .stream_memory_free     = tcp_stream_memory_free,
1893         .sockets_allocated      = &tcp_sockets_allocated,
1894         .memory_allocated       = &tcp_memory_allocated,
1895         .memory_pressure        = &tcp_memory_pressure,
1896         .orphan_count           = &tcp_orphan_count,
1897         .sysctl_mem             = sysctl_tcp_mem,
1898         .sysctl_wmem            = sysctl_tcp_wmem,
1899         .sysctl_rmem            = sysctl_tcp_rmem,
1900         .max_header             = MAX_TCP_HEADER,
1901         .obj_size               = sizeof(struct tcp6_sock),
1902         .slab_flags             = SLAB_DESTROY_BY_RCU,
1903         .twsk_prot              = &tcp6_timewait_sock_ops,
1904         .rsk_prot               = &tcp6_request_sock_ops,
1905         .h.hashinfo             = &tcp_hashinfo,
1906         .no_autobind            = true,
1907 #ifdef CONFIG_COMPAT
1908         .compat_setsockopt      = compat_tcp_setsockopt,
1909         .compat_getsockopt      = compat_tcp_getsockopt,
1910 #endif
1911         .diag_destroy           = tcp_abort,
1912 };
1913
1914 static const struct inet6_protocol tcpv6_protocol = {
1915         .early_demux    =       tcp_v6_early_demux,
1916         .handler        =       tcp_v6_rcv,
1917         .err_handler    =       tcp_v6_err,
1918         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1919 };
1920
1921 static struct inet_protosw tcpv6_protosw = {
1922         .type           =       SOCK_STREAM,
1923         .protocol       =       IPPROTO_TCP,
1924         .prot           =       &tcpv6_prot,
1925         .ops            =       &inet6_stream_ops,
1926         .flags          =       INET_PROTOSW_PERMANENT |
1927                                 INET_PROTOSW_ICSK,
1928 };
1929
1930 static int __net_init tcpv6_net_init(struct net *net)
1931 {
1932         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1933                                     SOCK_RAW, IPPROTO_TCP, net);
1934 }
1935
1936 static void __net_exit tcpv6_net_exit(struct net *net)
1937 {
1938         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1939 }
1940
1941 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1942 {
1943         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1944 }
1945
1946 static struct pernet_operations tcpv6_net_ops = {
1947         .init       = tcpv6_net_init,
1948         .exit       = tcpv6_net_exit,
1949         .exit_batch = tcpv6_net_exit_batch,
1950 };
1951
1952 int __init tcpv6_init(void)
1953 {
1954         int ret;
1955
1956         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1957         if (ret)
1958                 goto out;
1959
1960         /* register inet6 protocol */
1961         ret = inet6_register_protosw(&tcpv6_protosw);
1962         if (ret)
1963                 goto out_tcpv6_protocol;
1964
1965         ret = register_pernet_subsys(&tcpv6_net_ops);
1966         if (ret)
1967                 goto out_tcpv6_protosw;
1968 out:
1969         return ret;
1970
1971 out_tcpv6_protosw:
1972         inet6_unregister_protosw(&tcpv6_protosw);
1973 out_tcpv6_protocol:
1974         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1975         goto out;
1976 }
1977
1978 void tcpv6_exit(void)
1979 {
1980         unregister_pernet_subsys(&tcpv6_net_ops);
1981         inet6_unregister_protosw(&tcpv6_protosw);
1982         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1983 }