53b6a4192b161eae3f44b226395e02a691ad120c
[linux-3.10.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
42 #include <linux/ipv6.h>
43 #include <linux/icmpv6.h>
44 #include <linux/random.h>
45
46 #include <net/tcp.h>
47 #include <net/ndisc.h>
48 #include <net/inet6_hashtables.h>
49 #include <net/inet6_connection_sock.h>
50 #include <net/ipv6.h>
51 #include <net/transp_v6.h>
52 #include <net/addrconf.h>
53 #include <net/ip6_route.h>
54 #include <net/ip6_checksum.h>
55 #include <net/inet_ecn.h>
56 #include <net/protocol.h>
57 #include <net/xfrm.h>
58 #include <net/snmp.h>
59 #include <net/dsfield.h>
60 #include <net/timewait_sock.h>
61 #include <net/netdma.h>
62 #include <net/inet_common.h>
63
64 #include <asm/uaccess.h>
65
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
68
69 #include <linux/crypto.h>
70 #include <linux/scatterlist.h>
71
72 static void     tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
73 static void     tcp_v6_reqsk_send_ack(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 struct inet_connection_sock_af_ops ipv6_mapped;
79 static struct inet_connection_sock_af_ops ipv6_specific;
80 #ifdef CONFIG_TCP_MD5SIG
81 static struct tcp_sock_af_ops tcp_sock_ipv6_specific;
82 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
83 #else
84 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
85                                                    struct in6_addr *addr)
86 {
87         return NULL;
88 }
89 #endif
90
91 static void tcp_v6_hash(struct sock *sk)
92 {
93         if (sk->sk_state != TCP_CLOSE) {
94                 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
95                         tcp_prot.hash(sk);
96                         return;
97                 }
98                 local_bh_disable();
99                 __inet6_hash(sk);
100                 local_bh_enable();
101         }
102 }
103
104 static __inline__ __sum16 tcp_v6_check(int len,
105                                    struct in6_addr *saddr,
106                                    struct in6_addr *daddr,
107                                    __wsum base)
108 {
109         return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
110 }
111
112 static __u32 tcp_v6_init_sequence(struct sk_buff *skb)
113 {
114         return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
115                                             ipv6_hdr(skb)->saddr.s6_addr32,
116                                             tcp_hdr(skb)->dest,
117                                             tcp_hdr(skb)->source);
118 }
119
120 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
121                           int addr_len)
122 {
123         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
124         struct inet_sock *inet = inet_sk(sk);
125         struct inet_connection_sock *icsk = inet_csk(sk);
126         struct ipv6_pinfo *np = inet6_sk(sk);
127         struct tcp_sock *tp = tcp_sk(sk);
128         struct in6_addr *saddr = NULL, *final_p = NULL, final;
129         struct flowi fl;
130         struct dst_entry *dst;
131         int addr_type;
132         int err;
133
134         if (addr_len < SIN6_LEN_RFC2133)
135                 return -EINVAL;
136
137         if (usin->sin6_family != AF_INET6)
138                 return(-EAFNOSUPPORT);
139
140         memset(&fl, 0, sizeof(fl));
141
142         if (np->sndflow) {
143                 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
144                 IP6_ECN_flow_init(fl.fl6_flowlabel);
145                 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
146                         struct ip6_flowlabel *flowlabel;
147                         flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
148                         if (flowlabel == NULL)
149                                 return -EINVAL;
150                         ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
151                         fl6_sock_release(flowlabel);
152                 }
153         }
154
155         /*
156          *      connect() to INADDR_ANY means loopback (BSD'ism).
157          */
158
159         if(ipv6_addr_any(&usin->sin6_addr))
160                 usin->sin6_addr.s6_addr[15] = 0x1;
161
162         addr_type = ipv6_addr_type(&usin->sin6_addr);
163
164         if(addr_type & IPV6_ADDR_MULTICAST)
165                 return -ENETUNREACH;
166
167         if (addr_type&IPV6_ADDR_LINKLOCAL) {
168                 if (addr_len >= sizeof(struct sockaddr_in6) &&
169                     usin->sin6_scope_id) {
170                         /* If interface is set while binding, indices
171                          * must coincide.
172                          */
173                         if (sk->sk_bound_dev_if &&
174                             sk->sk_bound_dev_if != usin->sin6_scope_id)
175                                 return -EINVAL;
176
177                         sk->sk_bound_dev_if = usin->sin6_scope_id;
178                 }
179
180                 /* Connect to link-local address requires an interface */
181                 if (!sk->sk_bound_dev_if)
182                         return -EINVAL;
183         }
184
185         if (tp->rx_opt.ts_recent_stamp &&
186             !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
187                 tp->rx_opt.ts_recent = 0;
188                 tp->rx_opt.ts_recent_stamp = 0;
189                 tp->write_seq = 0;
190         }
191
192         ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
193         np->flow_label = fl.fl6_flowlabel;
194
195         /*
196          *      TCP over IPv4
197          */
198
199         if (addr_type == IPV6_ADDR_MAPPED) {
200                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
201                 struct sockaddr_in sin;
202
203                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
204
205                 if (__ipv6_only_sock(sk))
206                         return -ENETUNREACH;
207
208                 sin.sin_family = AF_INET;
209                 sin.sin_port = usin->sin6_port;
210                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
211
212                 icsk->icsk_af_ops = &ipv6_mapped;
213                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
214 #ifdef CONFIG_TCP_MD5SIG
215                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
216 #endif
217
218                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
219
220                 if (err) {
221                         icsk->icsk_ext_hdr_len = exthdrlen;
222                         icsk->icsk_af_ops = &ipv6_specific;
223                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
224 #ifdef CONFIG_TCP_MD5SIG
225                         tp->af_specific = &tcp_sock_ipv6_specific;
226 #endif
227                         goto failure;
228                 } else {
229                         ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
230                                       inet->saddr);
231                         ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
232                                       inet->rcv_saddr);
233                 }
234
235                 return err;
236         }
237
238         if (!ipv6_addr_any(&np->rcv_saddr))
239                 saddr = &np->rcv_saddr;
240
241         fl.proto = IPPROTO_TCP;
242         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
243         ipv6_addr_copy(&fl.fl6_src,
244                        (saddr ? saddr : &np->saddr));
245         fl.oif = sk->sk_bound_dev_if;
246         fl.fl_ip_dport = usin->sin6_port;
247         fl.fl_ip_sport = inet->sport;
248
249         if (np->opt && np->opt->srcrt) {
250                 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
251                 ipv6_addr_copy(&final, &fl.fl6_dst);
252                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
253                 final_p = &final;
254         }
255
256         security_sk_classify_flow(sk, &fl);
257
258         err = ip6_dst_lookup(sk, &dst, &fl);
259         if (err)
260                 goto failure;
261         if (final_p)
262                 ipv6_addr_copy(&fl.fl6_dst, final_p);
263
264         err = __xfrm_lookup(sock_net(sk), &dst, &fl, sk, XFRM_LOOKUP_WAIT);
265         if (err < 0) {
266                 if (err == -EREMOTE)
267                         err = ip6_dst_blackhole(sk, &dst, &fl);
268                 if (err < 0)
269                         goto failure;
270         }
271
272         if (saddr == NULL) {
273                 saddr = &fl.fl6_src;
274                 ipv6_addr_copy(&np->rcv_saddr, saddr);
275         }
276
277         /* set the source address */
278         ipv6_addr_copy(&np->saddr, saddr);
279         inet->rcv_saddr = LOOPBACK4_IPV6;
280
281         sk->sk_gso_type = SKB_GSO_TCPV6;
282         __ip6_dst_store(sk, dst, NULL, NULL);
283
284         icsk->icsk_ext_hdr_len = 0;
285         if (np->opt)
286                 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
287                                           np->opt->opt_nflen);
288
289         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
290
291         inet->dport = usin->sin6_port;
292
293         tcp_set_state(sk, TCP_SYN_SENT);
294         err = inet6_hash_connect(&tcp_death_row, sk);
295         if (err)
296                 goto late_failure;
297
298         if (!tp->write_seq)
299                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
300                                                              np->daddr.s6_addr32,
301                                                              inet->sport,
302                                                              inet->dport);
303
304         err = tcp_connect(sk);
305         if (err)
306                 goto late_failure;
307
308         return 0;
309
310 late_failure:
311         tcp_set_state(sk, TCP_CLOSE);
312         __sk_dst_reset(sk);
313 failure:
314         inet->dport = 0;
315         sk->sk_route_caps = 0;
316         return err;
317 }
318
319 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
320                 int type, int code, int offset, __be32 info)
321 {
322         struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
323         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
324         struct ipv6_pinfo *np;
325         struct sock *sk;
326         int err;
327         struct tcp_sock *tp;
328         __u32 seq;
329         struct net *net = dev_net(skb->dev);
330
331         sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
332                         th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
333
334         if (sk == NULL) {
335                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
336                                    ICMP6_MIB_INERRORS);
337                 return;
338         }
339
340         if (sk->sk_state == TCP_TIME_WAIT) {
341                 inet_twsk_put(inet_twsk(sk));
342                 return;
343         }
344
345         bh_lock_sock(sk);
346         if (sock_owned_by_user(sk))
347                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
348
349         if (sk->sk_state == TCP_CLOSE)
350                 goto out;
351
352         tp = tcp_sk(sk);
353         seq = ntohl(th->seq);
354         if (sk->sk_state != TCP_LISTEN &&
355             !between(seq, tp->snd_una, tp->snd_nxt)) {
356                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
357                 goto out;
358         }
359
360         np = inet6_sk(sk);
361
362         if (type == ICMPV6_PKT_TOOBIG) {
363                 struct dst_entry *dst = NULL;
364
365                 if (sock_owned_by_user(sk))
366                         goto out;
367                 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
368                         goto out;
369
370                 /* icmp should have updated the destination cache entry */
371                 dst = __sk_dst_check(sk, np->dst_cookie);
372
373                 if (dst == NULL) {
374                         struct inet_sock *inet = inet_sk(sk);
375                         struct flowi fl;
376
377                         /* BUGGG_FUTURE: Again, it is not clear how
378                            to handle rthdr case. Ignore this complexity
379                            for now.
380                          */
381                         memset(&fl, 0, sizeof(fl));
382                         fl.proto = IPPROTO_TCP;
383                         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
384                         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
385                         fl.oif = sk->sk_bound_dev_if;
386                         fl.fl_ip_dport = inet->dport;
387                         fl.fl_ip_sport = inet->sport;
388                         security_skb_classify_flow(skb, &fl);
389
390                         if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
391                                 sk->sk_err_soft = -err;
392                                 goto out;
393                         }
394
395                         if ((err = xfrm_lookup(net, &dst, &fl, sk, 0)) < 0) {
396                                 sk->sk_err_soft = -err;
397                                 goto out;
398                         }
399
400                 } else
401                         dst_hold(dst);
402
403                 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
404                         tcp_sync_mss(sk, dst_mtu(dst));
405                         tcp_simple_retransmit(sk);
406                 } /* else let the usual retransmit timer handle it */
407                 dst_release(dst);
408                 goto out;
409         }
410
411         icmpv6_err_convert(type, code, &err);
412
413         /* Might be for an request_sock */
414         switch (sk->sk_state) {
415                 struct request_sock *req, **prev;
416         case TCP_LISTEN:
417                 if (sock_owned_by_user(sk))
418                         goto out;
419
420                 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
421                                            &hdr->saddr, inet6_iif(skb));
422                 if (!req)
423                         goto out;
424
425                 /* ICMPs are not backlogged, hence we cannot get
426                  * an established socket here.
427                  */
428                 WARN_ON(req->sk != NULL);
429
430                 if (seq != tcp_rsk(req)->snt_isn) {
431                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
432                         goto out;
433                 }
434
435                 inet_csk_reqsk_queue_drop(sk, req, prev);
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
457 out:
458         bh_unlock_sock(sk);
459         sock_put(sk);
460 }
461
462
463 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req)
464 {
465         struct inet6_request_sock *treq = inet6_rsk(req);
466         struct ipv6_pinfo *np = inet6_sk(sk);
467         struct sk_buff * skb;
468         struct ipv6_txoptions *opt = NULL;
469         struct in6_addr * final_p = NULL, final;
470         struct flowi fl;
471         struct dst_entry *dst;
472         int err = -1;
473
474         memset(&fl, 0, sizeof(fl));
475         fl.proto = IPPROTO_TCP;
476         ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
477         ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
478         fl.fl6_flowlabel = 0;
479         fl.oif = treq->iif;
480         fl.fl_ip_dport = inet_rsk(req)->rmt_port;
481         fl.fl_ip_sport = inet_rsk(req)->loc_port;
482         security_req_classify_flow(req, &fl);
483
484         opt = np->opt;
485         if (opt && opt->srcrt) {
486                 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
487                 ipv6_addr_copy(&final, &fl.fl6_dst);
488                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
489                 final_p = &final;
490         }
491
492         err = ip6_dst_lookup(sk, &dst, &fl);
493         if (err)
494                 goto done;
495         if (final_p)
496                 ipv6_addr_copy(&fl.fl6_dst, final_p);
497         if ((err = xfrm_lookup(sock_net(sk), &dst, &fl, sk, 0)) < 0)
498                 goto done;
499
500         skb = tcp_make_synack(sk, dst, req);
501         if (skb) {
502                 struct tcphdr *th = tcp_hdr(skb);
503
504                 th->check = tcp_v6_check(skb->len,
505                                          &treq->loc_addr, &treq->rmt_addr,
506                                          csum_partial(th, skb->len, skb->csum));
507
508                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
509                 err = ip6_xmit(sk, skb, &fl, opt, 0);
510                 err = net_xmit_eval(err);
511         }
512
513 done:
514         if (opt && opt != np->opt)
515                 sock_kfree_s(sk, opt, opt->tot_len);
516         dst_release(dst);
517         return err;
518 }
519
520 static inline void syn_flood_warning(struct sk_buff *skb)
521 {
522 #ifdef CONFIG_SYN_COOKIES
523         if (sysctl_tcp_syncookies)
524                 printk(KERN_INFO
525                        "TCPv6: Possible SYN flooding on port %d. "
526                        "Sending cookies.\n", ntohs(tcp_hdr(skb)->dest));
527         else
528 #endif
529                 printk(KERN_INFO
530                        "TCPv6: Possible SYN flooding on port %d. "
531                        "Dropping request.\n", ntohs(tcp_hdr(skb)->dest));
532 }
533
534 static void tcp_v6_reqsk_destructor(struct request_sock *req)
535 {
536         kfree_skb(inet6_rsk(req)->pktopts);
537 }
538
539 #ifdef CONFIG_TCP_MD5SIG
540 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
541                                                    struct in6_addr *addr)
542 {
543         struct tcp_sock *tp = tcp_sk(sk);
544         int i;
545
546         BUG_ON(tp == NULL);
547
548         if (!tp->md5sig_info || !tp->md5sig_info->entries6)
549                 return NULL;
550
551         for (i = 0; i < tp->md5sig_info->entries6; i++) {
552                 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, addr))
553                         return &tp->md5sig_info->keys6[i].base;
554         }
555         return NULL;
556 }
557
558 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
559                                                 struct sock *addr_sk)
560 {
561         return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
562 }
563
564 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
565                                                       struct request_sock *req)
566 {
567         return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
568 }
569
570 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
571                              char *newkey, u8 newkeylen)
572 {
573         /* Add key to the list */
574         struct tcp_md5sig_key *key;
575         struct tcp_sock *tp = tcp_sk(sk);
576         struct tcp6_md5sig_key *keys;
577
578         key = tcp_v6_md5_do_lookup(sk, peer);
579         if (key) {
580                 /* modify existing entry - just update that one */
581                 kfree(key->key);
582                 key->key = newkey;
583                 key->keylen = newkeylen;
584         } else {
585                 /* reallocate new list if current one is full. */
586                 if (!tp->md5sig_info) {
587                         tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
588                         if (!tp->md5sig_info) {
589                                 kfree(newkey);
590                                 return -ENOMEM;
591                         }
592                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
593                 }
594                 if (tcp_alloc_md5sig_pool() == NULL) {
595                         kfree(newkey);
596                         return -ENOMEM;
597                 }
598                 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
599                         keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
600                                        (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
601
602                         if (!keys) {
603                                 tcp_free_md5sig_pool();
604                                 kfree(newkey);
605                                 return -ENOMEM;
606                         }
607
608                         if (tp->md5sig_info->entries6)
609                                 memmove(keys, tp->md5sig_info->keys6,
610                                         (sizeof (tp->md5sig_info->keys6[0]) *
611                                          tp->md5sig_info->entries6));
612
613                         kfree(tp->md5sig_info->keys6);
614                         tp->md5sig_info->keys6 = keys;
615                         tp->md5sig_info->alloced6++;
616                 }
617
618                 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
619                                peer);
620                 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
621                 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
622
623                 tp->md5sig_info->entries6++;
624         }
625         return 0;
626 }
627
628 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
629                                u8 *newkey, __u8 newkeylen)
630 {
631         return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
632                                  newkey, newkeylen);
633 }
634
635 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
636 {
637         struct tcp_sock *tp = tcp_sk(sk);
638         int i;
639
640         for (i = 0; i < tp->md5sig_info->entries6; i++) {
641                 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, peer)) {
642                         /* Free the key */
643                         kfree(tp->md5sig_info->keys6[i].base.key);
644                         tp->md5sig_info->entries6--;
645
646                         if (tp->md5sig_info->entries6 == 0) {
647                                 kfree(tp->md5sig_info->keys6);
648                                 tp->md5sig_info->keys6 = NULL;
649                                 tp->md5sig_info->alloced6 = 0;
650                         } else {
651                                 /* shrink the database */
652                                 if (tp->md5sig_info->entries6 != i)
653                                         memmove(&tp->md5sig_info->keys6[i],
654                                                 &tp->md5sig_info->keys6[i+1],
655                                                 (tp->md5sig_info->entries6 - i)
656                                                 * sizeof (tp->md5sig_info->keys6[0]));
657                         }
658                         tcp_free_md5sig_pool();
659                         return 0;
660                 }
661         }
662         return -ENOENT;
663 }
664
665 static void tcp_v6_clear_md5_list (struct sock *sk)
666 {
667         struct tcp_sock *tp = tcp_sk(sk);
668         int i;
669
670         if (tp->md5sig_info->entries6) {
671                 for (i = 0; i < tp->md5sig_info->entries6; i++)
672                         kfree(tp->md5sig_info->keys6[i].base.key);
673                 tp->md5sig_info->entries6 = 0;
674                 tcp_free_md5sig_pool();
675         }
676
677         kfree(tp->md5sig_info->keys6);
678         tp->md5sig_info->keys6 = NULL;
679         tp->md5sig_info->alloced6 = 0;
680
681         if (tp->md5sig_info->entries4) {
682                 for (i = 0; i < tp->md5sig_info->entries4; i++)
683                         kfree(tp->md5sig_info->keys4[i].base.key);
684                 tp->md5sig_info->entries4 = 0;
685                 tcp_free_md5sig_pool();
686         }
687
688         kfree(tp->md5sig_info->keys4);
689         tp->md5sig_info->keys4 = NULL;
690         tp->md5sig_info->alloced4 = 0;
691 }
692
693 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
694                                   int optlen)
695 {
696         struct tcp_md5sig cmd;
697         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
698         u8 *newkey;
699
700         if (optlen < sizeof(cmd))
701                 return -EINVAL;
702
703         if (copy_from_user(&cmd, optval, sizeof(cmd)))
704                 return -EFAULT;
705
706         if (sin6->sin6_family != AF_INET6)
707                 return -EINVAL;
708
709         if (!cmd.tcpm_keylen) {
710                 if (!tcp_sk(sk)->md5sig_info)
711                         return -ENOENT;
712                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
713                         return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
714                 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
715         }
716
717         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
718                 return -EINVAL;
719
720         if (!tcp_sk(sk)->md5sig_info) {
721                 struct tcp_sock *tp = tcp_sk(sk);
722                 struct tcp_md5sig_info *p;
723
724                 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
725                 if (!p)
726                         return -ENOMEM;
727
728                 tp->md5sig_info = p;
729                 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
730         }
731
732         newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
733         if (!newkey)
734                 return -ENOMEM;
735         if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
736                 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
737                                          newkey, cmd.tcpm_keylen);
738         }
739         return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
740 }
741
742 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
743                                         struct in6_addr *daddr,
744                                         struct in6_addr *saddr, int nbytes)
745 {
746         struct tcp6_pseudohdr *bp;
747         struct scatterlist sg;
748
749         bp = &hp->md5_blk.ip6;
750         /* 1. TCP pseudo-header (RFC2460) */
751         ipv6_addr_copy(&bp->saddr, saddr);
752         ipv6_addr_copy(&bp->daddr, daddr);
753         bp->protocol = cpu_to_be32(IPPROTO_TCP);
754         bp->len = cpu_to_be32(nbytes);
755
756         sg_init_one(&sg, bp, sizeof(*bp));
757         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
758 }
759
760 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
761                                struct in6_addr *daddr, struct in6_addr *saddr,
762                                struct tcphdr *th)
763 {
764         struct tcp_md5sig_pool *hp;
765         struct hash_desc *desc;
766
767         hp = tcp_get_md5sig_pool();
768         if (!hp)
769                 goto clear_hash_noput;
770         desc = &hp->md5_desc;
771
772         if (crypto_hash_init(desc))
773                 goto clear_hash;
774         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
775                 goto clear_hash;
776         if (tcp_md5_hash_header(hp, th))
777                 goto clear_hash;
778         if (tcp_md5_hash_key(hp, key))
779                 goto clear_hash;
780         if (crypto_hash_final(desc, md5_hash))
781                 goto clear_hash;
782
783         tcp_put_md5sig_pool();
784         return 0;
785
786 clear_hash:
787         tcp_put_md5sig_pool();
788 clear_hash_noput:
789         memset(md5_hash, 0, 16);
790         return 1;
791 }
792
793 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
794                                struct sock *sk, struct request_sock *req,
795                                struct sk_buff *skb)
796 {
797         struct in6_addr *saddr, *daddr;
798         struct tcp_md5sig_pool *hp;
799         struct hash_desc *desc;
800         struct tcphdr *th = tcp_hdr(skb);
801
802         if (sk) {
803                 saddr = &inet6_sk(sk)->saddr;
804                 daddr = &inet6_sk(sk)->daddr;
805         } else if (req) {
806                 saddr = &inet6_rsk(req)->loc_addr;
807                 daddr = &inet6_rsk(req)->rmt_addr;
808         } else {
809                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
810                 saddr = &ip6h->saddr;
811                 daddr = &ip6h->daddr;
812         }
813
814         hp = tcp_get_md5sig_pool();
815         if (!hp)
816                 goto clear_hash_noput;
817         desc = &hp->md5_desc;
818
819         if (crypto_hash_init(desc))
820                 goto clear_hash;
821
822         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
823                 goto clear_hash;
824         if (tcp_md5_hash_header(hp, th))
825                 goto clear_hash;
826         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
827                 goto clear_hash;
828         if (tcp_md5_hash_key(hp, key))
829                 goto clear_hash;
830         if (crypto_hash_final(desc, md5_hash))
831                 goto clear_hash;
832
833         tcp_put_md5sig_pool();
834         return 0;
835
836 clear_hash:
837         tcp_put_md5sig_pool();
838 clear_hash_noput:
839         memset(md5_hash, 0, 16);
840         return 1;
841 }
842
843 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
844 {
845         __u8 *hash_location = NULL;
846         struct tcp_md5sig_key *hash_expected;
847         struct ipv6hdr *ip6h = ipv6_hdr(skb);
848         struct tcphdr *th = tcp_hdr(skb);
849         int genhash;
850         u8 newhash[16];
851
852         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
853         hash_location = tcp_parse_md5sig_option(th);
854
855         /* We've parsed the options - do we have a hash? */
856         if (!hash_expected && !hash_location)
857                 return 0;
858
859         if (hash_expected && !hash_location) {
860                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
861                 return 1;
862         }
863
864         if (!hash_expected && hash_location) {
865                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
866                 return 1;
867         }
868
869         /* check the signature */
870         genhash = tcp_v6_md5_hash_skb(newhash,
871                                       hash_expected,
872                                       NULL, NULL, skb);
873
874         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
875                 if (net_ratelimit()) {
876                         printk(KERN_INFO "MD5 Hash %s for (%pI6, %u)->(%pI6, %u)\n",
877                                genhash ? "failed" : "mismatch",
878                                &ip6h->saddr, ntohs(th->source),
879                                &ip6h->daddr, ntohs(th->dest));
880                 }
881                 return 1;
882         }
883         return 0;
884 }
885 #endif
886
887 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
888         .family         =       AF_INET6,
889         .obj_size       =       sizeof(struct tcp6_request_sock),
890         .rtx_syn_ack    =       tcp_v6_send_synack,
891         .send_ack       =       tcp_v6_reqsk_send_ack,
892         .destructor     =       tcp_v6_reqsk_destructor,
893         .send_reset     =       tcp_v6_send_reset
894 };
895
896 #ifdef CONFIG_TCP_MD5SIG
897 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
898         .md5_lookup     =       tcp_v6_reqsk_md5_lookup,
899 };
900 #endif
901
902 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
903         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
904         .twsk_unique    = tcp_twsk_unique,
905         .twsk_destructor= tcp_twsk_destructor,
906 };
907
908 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb)
909 {
910         struct ipv6_pinfo *np = inet6_sk(sk);
911         struct tcphdr *th = tcp_hdr(skb);
912
913         if (skb->ip_summed == CHECKSUM_PARTIAL) {
914                 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,  0);
915                 skb->csum_start = skb_transport_header(skb) - skb->head;
916                 skb->csum_offset = offsetof(struct tcphdr, check);
917         } else {
918                 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
919                                             csum_partial(th, th->doff<<2,
920                                                          skb->csum));
921         }
922 }
923
924 static int tcp_v6_gso_send_check(struct sk_buff *skb)
925 {
926         struct ipv6hdr *ipv6h;
927         struct tcphdr *th;
928
929         if (!pskb_may_pull(skb, sizeof(*th)))
930                 return -EINVAL;
931
932         ipv6h = ipv6_hdr(skb);
933         th = tcp_hdr(skb);
934
935         th->check = 0;
936         th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
937                                      IPPROTO_TCP, 0);
938         skb->csum_start = skb_transport_header(skb) - skb->head;
939         skb->csum_offset = offsetof(struct tcphdr, check);
940         skb->ip_summed = CHECKSUM_PARTIAL;
941         return 0;
942 }
943
944 static struct sk_buff **tcp6_gro_receive(struct sk_buff **head,
945                                          struct sk_buff *skb)
946 {
947         struct ipv6hdr *iph = skb_gro_network_header(skb);
948
949         switch (skb->ip_summed) {
950         case CHECKSUM_COMPLETE:
951                 if (!tcp_v6_check(skb_gro_len(skb), &iph->saddr, &iph->daddr,
952                                   skb->csum)) {
953                         skb->ip_summed = CHECKSUM_UNNECESSARY;
954                         break;
955                 }
956
957                 /* fall through */
958         case CHECKSUM_NONE:
959                 NAPI_GRO_CB(skb)->flush = 1;
960                 return NULL;
961         }
962
963         return tcp_gro_receive(head, skb);
964 }
965
966 static int tcp6_gro_complete(struct sk_buff *skb)
967 {
968         struct ipv6hdr *iph = ipv6_hdr(skb);
969         struct tcphdr *th = tcp_hdr(skb);
970
971         th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
972                                   &iph->saddr, &iph->daddr, 0);
973         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
974
975         return tcp_gro_complete(skb);
976 }
977
978 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
979                                  u32 ts, struct tcp_md5sig_key *key, int rst)
980 {
981         struct tcphdr *th = tcp_hdr(skb), *t1;
982         struct sk_buff *buff;
983         struct flowi fl;
984         struct net *net = dev_net(skb_dst(skb)->dev);
985         struct sock *ctl_sk = net->ipv6.tcp_sk;
986         unsigned int tot_len = sizeof(struct tcphdr);
987         struct dst_entry *dst;
988         __be32 *topt;
989
990         if (ts)
991                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
992 #ifdef CONFIG_TCP_MD5SIG
993         if (key)
994                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
995 #endif
996
997         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
998                          GFP_ATOMIC);
999         if (buff == NULL)
1000                 return;
1001
1002         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1003
1004         t1 = (struct tcphdr *) skb_push(buff, tot_len);
1005
1006         /* Swap the send and the receive. */
1007         memset(t1, 0, sizeof(*t1));
1008         t1->dest = th->source;
1009         t1->source = th->dest;
1010         t1->doff = tot_len / 4;
1011         t1->seq = htonl(seq);
1012         t1->ack_seq = htonl(ack);
1013         t1->ack = !rst || !th->ack;
1014         t1->rst = rst;
1015         t1->window = htons(win);
1016
1017         topt = (__be32 *)(t1 + 1);
1018
1019         if (ts) {
1020                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1021                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1022                 *topt++ = htonl(tcp_time_stamp);
1023                 *topt++ = htonl(ts);
1024         }
1025
1026 #ifdef CONFIG_TCP_MD5SIG
1027         if (key) {
1028                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1029                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1030                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
1031                                     &ipv6_hdr(skb)->saddr,
1032                                     &ipv6_hdr(skb)->daddr, t1);
1033         }
1034 #endif
1035
1036         buff->csum = csum_partial(t1, tot_len, 0);
1037
1038         memset(&fl, 0, sizeof(fl));
1039         ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1040         ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1041
1042         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1043                                     tot_len, IPPROTO_TCP,
1044                                     buff->csum);
1045
1046         fl.proto = IPPROTO_TCP;
1047         fl.oif = inet6_iif(skb);
1048         fl.fl_ip_dport = t1->dest;
1049         fl.fl_ip_sport = t1->source;
1050         security_skb_classify_flow(skb, &fl);
1051
1052         /* Pass a socket to ip6_dst_lookup either it is for RST
1053          * Underlying function will use this to retrieve the network
1054          * namespace
1055          */
1056         if (!ip6_dst_lookup(ctl_sk, &dst, &fl)) {
1057                 if (xfrm_lookup(net, &dst, &fl, NULL, 0) >= 0) {
1058                         skb_dst_set(buff, dst);
1059                         ip6_xmit(ctl_sk, buff, &fl, NULL, 0);
1060                         TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1061                         if (rst)
1062                                 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
1063                         return;
1064                 }
1065         }
1066
1067         kfree_skb(buff);
1068 }
1069
1070 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
1071 {
1072         struct tcphdr *th = tcp_hdr(skb);
1073         u32 seq = 0, ack_seq = 0;
1074         struct tcp_md5sig_key *key = NULL;
1075
1076         if (th->rst)
1077                 return;
1078
1079         if (!ipv6_unicast_destination(skb))
1080                 return;
1081
1082 #ifdef CONFIG_TCP_MD5SIG
1083         if (sk)
1084                 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1085 #endif
1086
1087         if (th->ack)
1088                 seq = ntohl(th->ack_seq);
1089         else
1090                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1091                           (th->doff << 2);
1092
1093         tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1);
1094 }
1095
1096 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
1097                             struct tcp_md5sig_key *key)
1098 {
1099         tcp_v6_send_response(skb, seq, ack, win, ts, key, 0);
1100 }
1101
1102 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1103 {
1104         struct inet_timewait_sock *tw = inet_twsk(sk);
1105         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1106
1107         tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1108                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1109                         tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw));
1110
1111         inet_twsk_put(tw);
1112 }
1113
1114 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1115                                   struct request_sock *req)
1116 {
1117         tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1118                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr));
1119 }
1120
1121
1122 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1123 {
1124         struct request_sock *req, **prev;
1125         const struct tcphdr *th = tcp_hdr(skb);
1126         struct sock *nsk;
1127
1128         /* Find possible connection requests. */
1129         req = inet6_csk_search_req(sk, &prev, th->source,
1130                                    &ipv6_hdr(skb)->saddr,
1131                                    &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1132         if (req)
1133                 return tcp_check_req(sk, skb, req, prev);
1134
1135         nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1136                         &ipv6_hdr(skb)->saddr, th->source,
1137                         &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1138
1139         if (nsk) {
1140                 if (nsk->sk_state != TCP_TIME_WAIT) {
1141                         bh_lock_sock(nsk);
1142                         return nsk;
1143                 }
1144                 inet_twsk_put(inet_twsk(nsk));
1145                 return NULL;
1146         }
1147
1148 #ifdef CONFIG_SYN_COOKIES
1149         if (!th->rst && !th->syn && th->ack)
1150                 sk = cookie_v6_check(sk, skb);
1151 #endif
1152         return sk;
1153 }
1154
1155 /* FIXME: this is substantially similar to the ipv4 code.
1156  * Can some kind of merge be done? -- erics
1157  */
1158 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1159 {
1160         struct inet6_request_sock *treq;
1161         struct ipv6_pinfo *np = inet6_sk(sk);
1162         struct tcp_options_received tmp_opt;
1163         struct tcp_sock *tp = tcp_sk(sk);
1164         struct request_sock *req = NULL;
1165         __u32 isn = TCP_SKB_CB(skb)->when;
1166 #ifdef CONFIG_SYN_COOKIES
1167         int want_cookie = 0;
1168 #else
1169 #define want_cookie 0
1170 #endif
1171
1172         if (skb->protocol == htons(ETH_P_IP))
1173                 return tcp_v4_conn_request(sk, skb);
1174
1175         if (!ipv6_unicast_destination(skb))
1176                 goto drop;
1177
1178         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1179                 if (net_ratelimit())
1180                         syn_flood_warning(skb);
1181 #ifdef CONFIG_SYN_COOKIES
1182                 if (sysctl_tcp_syncookies)
1183                         want_cookie = 1;
1184                 else
1185 #endif
1186                 goto drop;
1187         }
1188
1189         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1190                 goto drop;
1191
1192         req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1193         if (req == NULL)
1194                 goto drop;
1195
1196 #ifdef CONFIG_TCP_MD5SIG
1197         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1198 #endif
1199
1200         tcp_clear_options(&tmp_opt);
1201         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1202         tmp_opt.user_mss = tp->rx_opt.user_mss;
1203
1204         tcp_parse_options(skb, &tmp_opt, 0);
1205
1206         if (want_cookie && !tmp_opt.saw_tstamp)
1207                 tcp_clear_options(&tmp_opt);
1208
1209         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1210         tcp_openreq_init(req, &tmp_opt, skb);
1211
1212         treq = inet6_rsk(req);
1213         ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1214         ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1215         if (!want_cookie)
1216                 TCP_ECN_create_request(req, tcp_hdr(skb));
1217
1218         if (want_cookie) {
1219                 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1220                 req->cookie_ts = tmp_opt.tstamp_ok;
1221         } else if (!isn) {
1222                 if (ipv6_opt_accepted(sk, skb) ||
1223                     np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1224                     np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1225                         atomic_inc(&skb->users);
1226                         treq->pktopts = skb;
1227                 }
1228                 treq->iif = sk->sk_bound_dev_if;
1229
1230                 /* So that link locals have meaning */
1231                 if (!sk->sk_bound_dev_if &&
1232                     ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1233                         treq->iif = inet6_iif(skb);
1234
1235                 isn = tcp_v6_init_sequence(skb);
1236         }
1237
1238         tcp_rsk(req)->snt_isn = isn;
1239
1240         security_inet_conn_request(sk, skb, req);
1241
1242         if (tcp_v6_send_synack(sk, req))
1243                 goto drop;
1244
1245         if (!want_cookie) {
1246                 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1247                 return 0;
1248         }
1249
1250 drop:
1251         if (req)
1252                 reqsk_free(req);
1253
1254         return 0; /* don't send reset */
1255 }
1256
1257 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1258                                           struct request_sock *req,
1259                                           struct dst_entry *dst)
1260 {
1261         struct inet6_request_sock *treq;
1262         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1263         struct tcp6_sock *newtcp6sk;
1264         struct inet_sock *newinet;
1265         struct tcp_sock *newtp;
1266         struct sock *newsk;
1267         struct ipv6_txoptions *opt;
1268 #ifdef CONFIG_TCP_MD5SIG
1269         struct tcp_md5sig_key *key;
1270 #endif
1271
1272         if (skb->protocol == htons(ETH_P_IP)) {
1273                 /*
1274                  *      v6 mapped
1275                  */
1276
1277                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1278
1279                 if (newsk == NULL)
1280                         return NULL;
1281
1282                 newtcp6sk = (struct tcp6_sock *)newsk;
1283                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1284
1285                 newinet = inet_sk(newsk);
1286                 newnp = inet6_sk(newsk);
1287                 newtp = tcp_sk(newsk);
1288
1289                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1290
1291                 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1292                               newinet->daddr);
1293
1294                 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1295                               newinet->saddr);
1296
1297                 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1298
1299                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1300                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1301 #ifdef CONFIG_TCP_MD5SIG
1302                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1303 #endif
1304
1305                 newnp->pktoptions  = NULL;
1306                 newnp->opt         = NULL;
1307                 newnp->mcast_oif   = inet6_iif(skb);
1308                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1309
1310                 /*
1311                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1312                  * here, tcp_create_openreq_child now does this for us, see the comment in
1313                  * that function for the gory details. -acme
1314                  */
1315
1316                 /* It is tricky place. Until this moment IPv4 tcp
1317                    worked with IPv6 icsk.icsk_af_ops.
1318                    Sync it now.
1319                  */
1320                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1321
1322                 return newsk;
1323         }
1324
1325         treq = inet6_rsk(req);
1326         opt = np->opt;
1327
1328         if (sk_acceptq_is_full(sk))
1329                 goto out_overflow;
1330
1331         if (dst == NULL) {
1332                 struct in6_addr *final_p = NULL, final;
1333                 struct flowi fl;
1334
1335                 memset(&fl, 0, sizeof(fl));
1336                 fl.proto = IPPROTO_TCP;
1337                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1338                 if (opt && opt->srcrt) {
1339                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1340                         ipv6_addr_copy(&final, &fl.fl6_dst);
1341                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1342                         final_p = &final;
1343                 }
1344                 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1345                 fl.oif = sk->sk_bound_dev_if;
1346                 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1347                 fl.fl_ip_sport = inet_rsk(req)->loc_port;
1348                 security_req_classify_flow(req, &fl);
1349
1350                 if (ip6_dst_lookup(sk, &dst, &fl))
1351                         goto out;
1352
1353                 if (final_p)
1354                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1355
1356                 if ((xfrm_lookup(sock_net(sk), &dst, &fl, sk, 0)) < 0)
1357                         goto out;
1358         }
1359
1360         newsk = tcp_create_openreq_child(sk, req, skb);
1361         if (newsk == NULL)
1362                 goto out;
1363
1364         /*
1365          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1366          * count here, tcp_create_openreq_child now does this for us, see the
1367          * comment in that function for the gory details. -acme
1368          */
1369
1370         newsk->sk_gso_type = SKB_GSO_TCPV6;
1371         __ip6_dst_store(newsk, dst, NULL, NULL);
1372
1373         newtcp6sk = (struct tcp6_sock *)newsk;
1374         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1375
1376         newtp = tcp_sk(newsk);
1377         newinet = inet_sk(newsk);
1378         newnp = inet6_sk(newsk);
1379
1380         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1381
1382         ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1383         ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1384         ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1385         newsk->sk_bound_dev_if = treq->iif;
1386
1387         /* Now IPv6 options...
1388
1389            First: no IPv4 options.
1390          */
1391         newinet->opt = NULL;
1392         newnp->ipv6_fl_list = NULL;
1393
1394         /* Clone RX bits */
1395         newnp->rxopt.all = np->rxopt.all;
1396
1397         /* Clone pktoptions received with SYN */
1398         newnp->pktoptions = NULL;
1399         if (treq->pktopts != NULL) {
1400                 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1401                 kfree_skb(treq->pktopts);
1402                 treq->pktopts = NULL;
1403                 if (newnp->pktoptions)
1404                         skb_set_owner_r(newnp->pktoptions, newsk);
1405         }
1406         newnp->opt        = NULL;
1407         newnp->mcast_oif  = inet6_iif(skb);
1408         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1409
1410         /* Clone native IPv6 options from listening socket (if any)
1411
1412            Yes, keeping reference count would be much more clever,
1413            but we make one more one thing there: reattach optmem
1414            to newsk.
1415          */
1416         if (opt) {
1417                 newnp->opt = ipv6_dup_options(newsk, opt);
1418                 if (opt != np->opt)
1419                         sock_kfree_s(sk, opt, opt->tot_len);
1420         }
1421
1422         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1423         if (newnp->opt)
1424                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1425                                                      newnp->opt->opt_flen);
1426
1427         tcp_mtup_init(newsk);
1428         tcp_sync_mss(newsk, dst_mtu(dst));
1429         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1430         tcp_initialize_rcv_mss(newsk);
1431
1432         newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1433
1434 #ifdef CONFIG_TCP_MD5SIG
1435         /* Copy over the MD5 key from the original socket */
1436         if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1437                 /* We're using one, so create a matching key
1438                  * on the newsk structure. If we fail to get
1439                  * memory, then we end up not copying the key
1440                  * across. Shucks.
1441                  */
1442                 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1443                 if (newkey != NULL)
1444                         tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1445                                           newkey, key->keylen);
1446         }
1447 #endif
1448
1449         __inet6_hash(newsk);
1450         __inet_inherit_port(sk, newsk);
1451
1452         return newsk;
1453
1454 out_overflow:
1455         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1456 out:
1457         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1458         if (opt && opt != np->opt)
1459                 sock_kfree_s(sk, opt, opt->tot_len);
1460         dst_release(dst);
1461         return NULL;
1462 }
1463
1464 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1465 {
1466         if (skb->ip_summed == CHECKSUM_COMPLETE) {
1467                 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1468                                   &ipv6_hdr(skb)->daddr, skb->csum)) {
1469                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1470                         return 0;
1471                 }
1472         }
1473
1474         skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1475                                               &ipv6_hdr(skb)->saddr,
1476                                               &ipv6_hdr(skb)->daddr, 0));
1477
1478         if (skb->len <= 76) {
1479                 return __skb_checksum_complete(skb);
1480         }
1481         return 0;
1482 }
1483
1484 /* The socket must have it's spinlock held when we get
1485  * here.
1486  *
1487  * We have a potential double-lock case here, so even when
1488  * doing backlog processing we use the BH locking scheme.
1489  * This is because we cannot sleep with the original spinlock
1490  * held.
1491  */
1492 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1493 {
1494         struct ipv6_pinfo *np = inet6_sk(sk);
1495         struct tcp_sock *tp;
1496         struct sk_buff *opt_skb = NULL;
1497
1498         /* Imagine: socket is IPv6. IPv4 packet arrives,
1499            goes to IPv4 receive handler and backlogged.
1500            From backlog it always goes here. Kerboom...
1501            Fortunately, tcp_rcv_established and rcv_established
1502            handle them correctly, but it is not case with
1503            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1504          */
1505
1506         if (skb->protocol == htons(ETH_P_IP))
1507                 return tcp_v4_do_rcv(sk, skb);
1508
1509 #ifdef CONFIG_TCP_MD5SIG
1510         if (tcp_v6_inbound_md5_hash (sk, skb))
1511                 goto discard;
1512 #endif
1513
1514         if (sk_filter(sk, skb))
1515                 goto discard;
1516
1517         /*
1518          *      socket locking is here for SMP purposes as backlog rcv
1519          *      is currently called with bh processing disabled.
1520          */
1521
1522         /* Do Stevens' IPV6_PKTOPTIONS.
1523
1524            Yes, guys, it is the only place in our code, where we
1525            may make it not affecting IPv4.
1526            The rest of code is protocol independent,
1527            and I do not like idea to uglify IPv4.
1528
1529            Actually, all the idea behind IPV6_PKTOPTIONS
1530            looks not very well thought. For now we latch
1531            options, received in the last packet, enqueued
1532            by tcp. Feel free to propose better solution.
1533                                                --ANK (980728)
1534          */
1535         if (np->rxopt.all)
1536                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1537
1538         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1539                 TCP_CHECK_TIMER(sk);
1540                 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1541                         goto reset;
1542                 TCP_CHECK_TIMER(sk);
1543                 if (opt_skb)
1544                         goto ipv6_pktoptions;
1545                 return 0;
1546         }
1547
1548         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1549                 goto csum_err;
1550
1551         if (sk->sk_state == TCP_LISTEN) {
1552                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1553                 if (!nsk)
1554                         goto discard;
1555
1556                 /*
1557                  * Queue it on the new socket if the new socket is active,
1558                  * otherwise we just shortcircuit this and continue with
1559                  * the new socket..
1560                  */
1561                 if(nsk != sk) {
1562                         if (tcp_child_process(sk, nsk, skb))
1563                                 goto reset;
1564                         if (opt_skb)
1565                                 __kfree_skb(opt_skb);
1566                         return 0;
1567                 }
1568         }
1569
1570         TCP_CHECK_TIMER(sk);
1571         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1572                 goto reset;
1573         TCP_CHECK_TIMER(sk);
1574         if (opt_skb)
1575                 goto ipv6_pktoptions;
1576         return 0;
1577
1578 reset:
1579         tcp_v6_send_reset(sk, skb);
1580 discard:
1581         if (opt_skb)
1582                 __kfree_skb(opt_skb);
1583         kfree_skb(skb);
1584         return 0;
1585 csum_err:
1586         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1587         goto discard;
1588
1589
1590 ipv6_pktoptions:
1591         /* Do you ask, what is it?
1592
1593            1. skb was enqueued by tcp.
1594            2. skb is added to tail of read queue, rather than out of order.
1595            3. socket is not in passive state.
1596            4. Finally, it really contains options, which user wants to receive.
1597          */
1598         tp = tcp_sk(sk);
1599         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1600             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1601                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1602                         np->mcast_oif = inet6_iif(opt_skb);
1603                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1604                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1605                 if (ipv6_opt_accepted(sk, opt_skb)) {
1606                         skb_set_owner_r(opt_skb, sk);
1607                         opt_skb = xchg(&np->pktoptions, opt_skb);
1608                 } else {
1609                         __kfree_skb(opt_skb);
1610                         opt_skb = xchg(&np->pktoptions, NULL);
1611                 }
1612         }
1613
1614         kfree_skb(opt_skb);
1615         return 0;
1616 }
1617
1618 static int tcp_v6_rcv(struct sk_buff *skb)
1619 {
1620         struct tcphdr *th;
1621         struct sock *sk;
1622         int ret;
1623         struct net *net = dev_net(skb->dev);
1624
1625         if (skb->pkt_type != PACKET_HOST)
1626                 goto discard_it;
1627
1628         /*
1629          *      Count it even if it's bad.
1630          */
1631         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1632
1633         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1634                 goto discard_it;
1635
1636         th = tcp_hdr(skb);
1637
1638         if (th->doff < sizeof(struct tcphdr)/4)
1639                 goto bad_packet;
1640         if (!pskb_may_pull(skb, th->doff*4))
1641                 goto discard_it;
1642
1643         if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1644                 goto bad_packet;
1645
1646         th = tcp_hdr(skb);
1647         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1648         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1649                                     skb->len - th->doff*4);
1650         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1651         TCP_SKB_CB(skb)->when = 0;
1652         TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1653         TCP_SKB_CB(skb)->sacked = 0;
1654
1655         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1656         if (!sk)
1657                 goto no_tcp_socket;
1658
1659 process:
1660         if (sk->sk_state == TCP_TIME_WAIT)
1661                 goto do_time_wait;
1662
1663         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1664                 goto discard_and_relse;
1665
1666         if (sk_filter(sk, skb))
1667                 goto discard_and_relse;
1668
1669         skb->dev = NULL;
1670
1671         bh_lock_sock_nested(sk);
1672         ret = 0;
1673         if (!sock_owned_by_user(sk)) {
1674 #ifdef CONFIG_NET_DMA
1675                 struct tcp_sock *tp = tcp_sk(sk);
1676                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1677                         tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1678                 if (tp->ucopy.dma_chan)
1679                         ret = tcp_v6_do_rcv(sk, skb);
1680                 else
1681 #endif
1682                 {
1683                         if (!tcp_prequeue(sk, skb))
1684                                 ret = tcp_v6_do_rcv(sk, skb);
1685                 }
1686         } else
1687                 sk_add_backlog(sk, skb);
1688         bh_unlock_sock(sk);
1689
1690         sock_put(sk);
1691         return ret ? -1 : 0;
1692
1693 no_tcp_socket:
1694         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1695                 goto discard_it;
1696
1697         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1698 bad_packet:
1699                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1700         } else {
1701                 tcp_v6_send_reset(NULL, skb);
1702         }
1703
1704 discard_it:
1705
1706         /*
1707          *      Discard frame
1708          */
1709
1710         kfree_skb(skb);
1711         return 0;
1712
1713 discard_and_relse:
1714         sock_put(sk);
1715         goto discard_it;
1716
1717 do_time_wait:
1718         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1719                 inet_twsk_put(inet_twsk(sk));
1720                 goto discard_it;
1721         }
1722
1723         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1724                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1725                 inet_twsk_put(inet_twsk(sk));
1726                 goto discard_it;
1727         }
1728
1729         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1730         case TCP_TW_SYN:
1731         {
1732                 struct sock *sk2;
1733
1734                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1735                                             &ipv6_hdr(skb)->daddr,
1736                                             ntohs(th->dest), inet6_iif(skb));
1737                 if (sk2 != NULL) {
1738                         struct inet_timewait_sock *tw = inet_twsk(sk);
1739                         inet_twsk_deschedule(tw, &tcp_death_row);
1740                         inet_twsk_put(tw);
1741                         sk = sk2;
1742                         goto process;
1743                 }
1744                 /* Fall through to ACK */
1745         }
1746         case TCP_TW_ACK:
1747                 tcp_v6_timewait_ack(sk, skb);
1748                 break;
1749         case TCP_TW_RST:
1750                 goto no_tcp_socket;
1751         case TCP_TW_SUCCESS:;
1752         }
1753         goto discard_it;
1754 }
1755
1756 static int tcp_v6_remember_stamp(struct sock *sk)
1757 {
1758         /* Alas, not yet... */
1759         return 0;
1760 }
1761
1762 static struct inet_connection_sock_af_ops ipv6_specific = {
1763         .queue_xmit        = inet6_csk_xmit,
1764         .send_check        = tcp_v6_send_check,
1765         .rebuild_header    = inet6_sk_rebuild_header,
1766         .conn_request      = tcp_v6_conn_request,
1767         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1768         .remember_stamp    = tcp_v6_remember_stamp,
1769         .net_header_len    = sizeof(struct ipv6hdr),
1770         .setsockopt        = ipv6_setsockopt,
1771         .getsockopt        = ipv6_getsockopt,
1772         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1773         .sockaddr_len      = sizeof(struct sockaddr_in6),
1774         .bind_conflict     = inet6_csk_bind_conflict,
1775 #ifdef CONFIG_COMPAT
1776         .compat_setsockopt = compat_ipv6_setsockopt,
1777         .compat_getsockopt = compat_ipv6_getsockopt,
1778 #endif
1779 };
1780
1781 #ifdef CONFIG_TCP_MD5SIG
1782 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1783         .md5_lookup     =       tcp_v6_md5_lookup,
1784         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1785         .md5_add        =       tcp_v6_md5_add_func,
1786         .md5_parse      =       tcp_v6_parse_md5_keys,
1787 };
1788 #endif
1789
1790 /*
1791  *      TCP over IPv4 via INET6 API
1792  */
1793
1794 static struct inet_connection_sock_af_ops ipv6_mapped = {
1795         .queue_xmit        = ip_queue_xmit,
1796         .send_check        = tcp_v4_send_check,
1797         .rebuild_header    = inet_sk_rebuild_header,
1798         .conn_request      = tcp_v6_conn_request,
1799         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1800         .remember_stamp    = tcp_v4_remember_stamp,
1801         .net_header_len    = sizeof(struct iphdr),
1802         .setsockopt        = ipv6_setsockopt,
1803         .getsockopt        = ipv6_getsockopt,
1804         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1805         .sockaddr_len      = sizeof(struct sockaddr_in6),
1806         .bind_conflict     = inet6_csk_bind_conflict,
1807 #ifdef CONFIG_COMPAT
1808         .compat_setsockopt = compat_ipv6_setsockopt,
1809         .compat_getsockopt = compat_ipv6_getsockopt,
1810 #endif
1811 };
1812
1813 #ifdef CONFIG_TCP_MD5SIG
1814 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1815         .md5_lookup     =       tcp_v4_md5_lookup,
1816         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1817         .md5_add        =       tcp_v6_md5_add_func,
1818         .md5_parse      =       tcp_v6_parse_md5_keys,
1819 };
1820 #endif
1821
1822 /* NOTE: A lot of things set to zero explicitly by call to
1823  *       sk_alloc() so need not be done here.
1824  */
1825 static int tcp_v6_init_sock(struct sock *sk)
1826 {
1827         struct inet_connection_sock *icsk = inet_csk(sk);
1828         struct tcp_sock *tp = tcp_sk(sk);
1829
1830         skb_queue_head_init(&tp->out_of_order_queue);
1831         tcp_init_xmit_timers(sk);
1832         tcp_prequeue_init(tp);
1833
1834         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1835         tp->mdev = TCP_TIMEOUT_INIT;
1836
1837         /* So many TCP implementations out there (incorrectly) count the
1838          * initial SYN frame in their delayed-ACK and congestion control
1839          * algorithms that we must have the following bandaid to talk
1840          * efficiently to them.  -DaveM
1841          */
1842         tp->snd_cwnd = 2;
1843
1844         /* See draft-stevens-tcpca-spec-01 for discussion of the
1845          * initialization of these values.
1846          */
1847         tp->snd_ssthresh = 0x7fffffff;
1848         tp->snd_cwnd_clamp = ~0;
1849         tp->mss_cache = 536;
1850
1851         tp->reordering = sysctl_tcp_reordering;
1852
1853         sk->sk_state = TCP_CLOSE;
1854
1855         icsk->icsk_af_ops = &ipv6_specific;
1856         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1857         icsk->icsk_sync_mss = tcp_sync_mss;
1858         sk->sk_write_space = sk_stream_write_space;
1859         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1860
1861 #ifdef CONFIG_TCP_MD5SIG
1862         tp->af_specific = &tcp_sock_ipv6_specific;
1863 #endif
1864
1865         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1866         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1867
1868         local_bh_disable();
1869         percpu_counter_inc(&tcp_sockets_allocated);
1870         local_bh_enable();
1871
1872         return 0;
1873 }
1874
1875 static void tcp_v6_destroy_sock(struct sock *sk)
1876 {
1877 #ifdef CONFIG_TCP_MD5SIG
1878         /* Clean up the MD5 key list */
1879         if (tcp_sk(sk)->md5sig_info)
1880                 tcp_v6_clear_md5_list(sk);
1881 #endif
1882         tcp_v4_destroy_sock(sk);
1883         inet6_destroy_sock(sk);
1884 }
1885
1886 #ifdef CONFIG_PROC_FS
1887 /* Proc filesystem TCPv6 sock list dumping. */
1888 static void get_openreq6(struct seq_file *seq,
1889                          struct sock *sk, struct request_sock *req, int i, int uid)
1890 {
1891         int ttd = req->expires - jiffies;
1892         struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1893         struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1894
1895         if (ttd < 0)
1896                 ttd = 0;
1897
1898         seq_printf(seq,
1899                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1900                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1901                    i,
1902                    src->s6_addr32[0], src->s6_addr32[1],
1903                    src->s6_addr32[2], src->s6_addr32[3],
1904                    ntohs(inet_rsk(req)->loc_port),
1905                    dest->s6_addr32[0], dest->s6_addr32[1],
1906                    dest->s6_addr32[2], dest->s6_addr32[3],
1907                    ntohs(inet_rsk(req)->rmt_port),
1908                    TCP_SYN_RECV,
1909                    0,0, /* could print option size, but that is af dependent. */
1910                    1,   /* timers active (only the expire timer) */
1911                    jiffies_to_clock_t(ttd),
1912                    req->retrans,
1913                    uid,
1914                    0,  /* non standard timer */
1915                    0, /* open_requests have no inode */
1916                    0, req);
1917 }
1918
1919 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1920 {
1921         struct in6_addr *dest, *src;
1922         __u16 destp, srcp;
1923         int timer_active;
1924         unsigned long timer_expires;
1925         struct inet_sock *inet = inet_sk(sp);
1926         struct tcp_sock *tp = tcp_sk(sp);
1927         const struct inet_connection_sock *icsk = inet_csk(sp);
1928         struct ipv6_pinfo *np = inet6_sk(sp);
1929
1930         dest  = &np->daddr;
1931         src   = &np->rcv_saddr;
1932         destp = ntohs(inet->dport);
1933         srcp  = ntohs(inet->sport);
1934
1935         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1936                 timer_active    = 1;
1937                 timer_expires   = icsk->icsk_timeout;
1938         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1939                 timer_active    = 4;
1940                 timer_expires   = icsk->icsk_timeout;
1941         } else if (timer_pending(&sp->sk_timer)) {
1942                 timer_active    = 2;
1943                 timer_expires   = sp->sk_timer.expires;
1944         } else {
1945                 timer_active    = 0;
1946                 timer_expires = jiffies;
1947         }
1948
1949         seq_printf(seq,
1950                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1951                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %lu %lu %u %u %d\n",
1952                    i,
1953                    src->s6_addr32[0], src->s6_addr32[1],
1954                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1955                    dest->s6_addr32[0], dest->s6_addr32[1],
1956                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1957                    sp->sk_state,
1958                    tp->write_seq-tp->snd_una,
1959                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1960                    timer_active,
1961                    jiffies_to_clock_t(timer_expires - jiffies),
1962                    icsk->icsk_retransmits,
1963                    sock_i_uid(sp),
1964                    icsk->icsk_probes_out,
1965                    sock_i_ino(sp),
1966                    atomic_read(&sp->sk_refcnt), sp,
1967                    jiffies_to_clock_t(icsk->icsk_rto),
1968                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1969                    (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
1970                    tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
1971                    );
1972 }
1973
1974 static void get_timewait6_sock(struct seq_file *seq,
1975                                struct inet_timewait_sock *tw, int i)
1976 {
1977         struct in6_addr *dest, *src;
1978         __u16 destp, srcp;
1979         struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
1980         int ttd = tw->tw_ttd - jiffies;
1981
1982         if (ttd < 0)
1983                 ttd = 0;
1984
1985         dest = &tw6->tw_v6_daddr;
1986         src  = &tw6->tw_v6_rcv_saddr;
1987         destp = ntohs(tw->tw_dport);
1988         srcp  = ntohs(tw->tw_sport);
1989
1990         seq_printf(seq,
1991                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1992                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1993                    i,
1994                    src->s6_addr32[0], src->s6_addr32[1],
1995                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1996                    dest->s6_addr32[0], dest->s6_addr32[1],
1997                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1998                    tw->tw_substate, 0, 0,
1999                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2000                    atomic_read(&tw->tw_refcnt), tw);
2001 }
2002
2003 static int tcp6_seq_show(struct seq_file *seq, void *v)
2004 {
2005         struct tcp_iter_state *st;
2006
2007         if (v == SEQ_START_TOKEN) {
2008                 seq_puts(seq,
2009                          "  sl  "
2010                          "local_address                         "
2011                          "remote_address                        "
2012                          "st tx_queue rx_queue tr tm->when retrnsmt"
2013                          "   uid  timeout inode\n");
2014                 goto out;
2015         }
2016         st = seq->private;
2017
2018         switch (st->state) {
2019         case TCP_SEQ_STATE_LISTENING:
2020         case TCP_SEQ_STATE_ESTABLISHED:
2021                 get_tcp6_sock(seq, v, st->num);
2022                 break;
2023         case TCP_SEQ_STATE_OPENREQ:
2024                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2025                 break;
2026         case TCP_SEQ_STATE_TIME_WAIT:
2027                 get_timewait6_sock(seq, v, st->num);
2028                 break;
2029         }
2030 out:
2031         return 0;
2032 }
2033
2034 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2035         .name           = "tcp6",
2036         .family         = AF_INET6,
2037         .seq_fops       = {
2038                 .owner          = THIS_MODULE,
2039         },
2040         .seq_ops        = {
2041                 .show           = tcp6_seq_show,
2042         },
2043 };
2044
2045 int tcp6_proc_init(struct net *net)
2046 {
2047         return tcp_proc_register(net, &tcp6_seq_afinfo);
2048 }
2049
2050 void tcp6_proc_exit(struct net *net)
2051 {
2052         tcp_proc_unregister(net, &tcp6_seq_afinfo);
2053 }
2054 #endif
2055
2056 struct proto tcpv6_prot = {
2057         .name                   = "TCPv6",
2058         .owner                  = THIS_MODULE,
2059         .close                  = tcp_close,
2060         .connect                = tcp_v6_connect,
2061         .disconnect             = tcp_disconnect,
2062         .accept                 = inet_csk_accept,
2063         .ioctl                  = tcp_ioctl,
2064         .init                   = tcp_v6_init_sock,
2065         .destroy                = tcp_v6_destroy_sock,
2066         .shutdown               = tcp_shutdown,
2067         .setsockopt             = tcp_setsockopt,
2068         .getsockopt             = tcp_getsockopt,
2069         .recvmsg                = tcp_recvmsg,
2070         .backlog_rcv            = tcp_v6_do_rcv,
2071         .hash                   = tcp_v6_hash,
2072         .unhash                 = inet_unhash,
2073         .get_port               = inet_csk_get_port,
2074         .enter_memory_pressure  = tcp_enter_memory_pressure,
2075         .sockets_allocated      = &tcp_sockets_allocated,
2076         .memory_allocated       = &tcp_memory_allocated,
2077         .memory_pressure        = &tcp_memory_pressure,
2078         .orphan_count           = &tcp_orphan_count,
2079         .sysctl_mem             = sysctl_tcp_mem,
2080         .sysctl_wmem            = sysctl_tcp_wmem,
2081         .sysctl_rmem            = sysctl_tcp_rmem,
2082         .max_header             = MAX_TCP_HEADER,
2083         .obj_size               = sizeof(struct tcp6_sock),
2084         .slab_flags             = SLAB_DESTROY_BY_RCU,
2085         .twsk_prot              = &tcp6_timewait_sock_ops,
2086         .rsk_prot               = &tcp6_request_sock_ops,
2087         .h.hashinfo             = &tcp_hashinfo,
2088 #ifdef CONFIG_COMPAT
2089         .compat_setsockopt      = compat_tcp_setsockopt,
2090         .compat_getsockopt      = compat_tcp_getsockopt,
2091 #endif
2092 };
2093
2094 static struct inet6_protocol tcpv6_protocol = {
2095         .handler        =       tcp_v6_rcv,
2096         .err_handler    =       tcp_v6_err,
2097         .gso_send_check =       tcp_v6_gso_send_check,
2098         .gso_segment    =       tcp_tso_segment,
2099         .gro_receive    =       tcp6_gro_receive,
2100         .gro_complete   =       tcp6_gro_complete,
2101         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2102 };
2103
2104 static struct inet_protosw tcpv6_protosw = {
2105         .type           =       SOCK_STREAM,
2106         .protocol       =       IPPROTO_TCP,
2107         .prot           =       &tcpv6_prot,
2108         .ops            =       &inet6_stream_ops,
2109         .capability     =       -1,
2110         .no_check       =       0,
2111         .flags          =       INET_PROTOSW_PERMANENT |
2112                                 INET_PROTOSW_ICSK,
2113 };
2114
2115 static int tcpv6_net_init(struct net *net)
2116 {
2117         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2118                                     SOCK_RAW, IPPROTO_TCP, net);
2119 }
2120
2121 static void tcpv6_net_exit(struct net *net)
2122 {
2123         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2124         inet_twsk_purge(net, &tcp_hashinfo, &tcp_death_row, AF_INET6);
2125 }
2126
2127 static struct pernet_operations tcpv6_net_ops = {
2128         .init = tcpv6_net_init,
2129         .exit = tcpv6_net_exit,
2130 };
2131
2132 int __init tcpv6_init(void)
2133 {
2134         int ret;
2135
2136         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2137         if (ret)
2138                 goto out;
2139
2140         /* register inet6 protocol */
2141         ret = inet6_register_protosw(&tcpv6_protosw);
2142         if (ret)
2143                 goto out_tcpv6_protocol;
2144
2145         ret = register_pernet_subsys(&tcpv6_net_ops);
2146         if (ret)
2147                 goto out_tcpv6_protosw;
2148 out:
2149         return ret;
2150
2151 out_tcpv6_protocol:
2152         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2153 out_tcpv6_protosw:
2154         inet6_unregister_protosw(&tcpv6_protosw);
2155         goto out;
2156 }
2157
2158 void tcpv6_exit(void)
2159 {
2160         unregister_pernet_subsys(&tcpv6_net_ops);
2161         inet6_unregister_protosw(&tcpv6_protosw);
2162         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2163 }