]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - net/ipv6/tcp_ipv6.c
Merge tag 'ceph-for-4.10-rc1' of git://github.com/ceph/ceph-client
[mirror_ubuntu-zesty-kernel.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, u32 *tsoff)
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, tsoff);
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 fl6.flowi6_uid = sk->sk_uid;
237
238 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
239 final_p = fl6_update_dst(&fl6, opt, &final);
240
241 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
242
243 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
244 if (IS_ERR(dst)) {
245 err = PTR_ERR(dst);
246 goto failure;
247 }
248
249 if (!saddr) {
250 saddr = &fl6.saddr;
251 sk->sk_v6_rcv_saddr = *saddr;
252 }
253
254 /* set the source address */
255 np->saddr = *saddr;
256 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
257
258 sk->sk_gso_type = SKB_GSO_TCPV6;
259 ip6_dst_store(sk, dst, NULL, NULL);
260
261 if (tcp_death_row.sysctl_tw_recycle &&
262 !tp->rx_opt.ts_recent_stamp &&
263 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
264 tcp_fetch_timewait_stamp(sk, dst);
265
266 icsk->icsk_ext_hdr_len = 0;
267 if (opt)
268 icsk->icsk_ext_hdr_len = opt->opt_flen +
269 opt->opt_nflen;
270
271 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
272
273 inet->inet_dport = usin->sin6_port;
274
275 tcp_set_state(sk, TCP_SYN_SENT);
276 err = inet6_hash_connect(&tcp_death_row, sk);
277 if (err)
278 goto late_failure;
279
280 sk_set_txhash(sk);
281
282 if (!tp->write_seq && likely(!tp->repair))
283 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
284 sk->sk_v6_daddr.s6_addr32,
285 inet->inet_sport,
286 inet->inet_dport,
287 &tp->tsoffset);
288
289 err = tcp_connect(sk);
290 if (err)
291 goto late_failure;
292
293 return 0;
294
295 late_failure:
296 tcp_set_state(sk, TCP_CLOSE);
297 __sk_dst_reset(sk);
298 failure:
299 inet->inet_dport = 0;
300 sk->sk_route_caps = 0;
301 return err;
302 }
303
304 static void tcp_v6_mtu_reduced(struct sock *sk)
305 {
306 struct dst_entry *dst;
307
308 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
309 return;
310
311 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
312 if (!dst)
313 return;
314
315 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
316 tcp_sync_mss(sk, dst_mtu(dst));
317 tcp_simple_retransmit(sk);
318 }
319 }
320
321 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
322 u8 type, u8 code, int offset, __be32 info)
323 {
324 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
325 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
326 struct net *net = dev_net(skb->dev);
327 struct request_sock *fastopen;
328 struct ipv6_pinfo *np;
329 struct tcp_sock *tp;
330 __u32 seq, snd_una;
331 struct sock *sk;
332 bool fatal;
333 int err;
334
335 sk = __inet6_lookup_established(net, &tcp_hashinfo,
336 &hdr->daddr, th->dest,
337 &hdr->saddr, ntohs(th->source),
338 skb->dev->ifindex);
339
340 if (!sk) {
341 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
342 ICMP6_MIB_INERRORS);
343 return;
344 }
345
346 if (sk->sk_state == TCP_TIME_WAIT) {
347 inet_twsk_put(inet_twsk(sk));
348 return;
349 }
350 seq = ntohl(th->seq);
351 fatal = icmpv6_err_convert(type, code, &err);
352 if (sk->sk_state == TCP_NEW_SYN_RECV)
353 return tcp_req_err(sk, seq, fatal);
354
355 bh_lock_sock(sk);
356 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
357 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
358
359 if (sk->sk_state == TCP_CLOSE)
360 goto out;
361
362 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
363 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
364 goto out;
365 }
366
367 tp = tcp_sk(sk);
368 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
369 fastopen = tp->fastopen_rsk;
370 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
371 if (sk->sk_state != TCP_LISTEN &&
372 !between(seq, snd_una, tp->snd_nxt)) {
373 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
374 goto out;
375 }
376
377 np = inet6_sk(sk);
378
379 if (type == NDISC_REDIRECT) {
380 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
381
382 if (dst)
383 dst->ops->redirect(dst, sk, skb);
384 goto out;
385 }
386
387 if (type == ICMPV6_PKT_TOOBIG) {
388 /* We are not interested in TCP_LISTEN and open_requests
389 * (SYN-ACKs send out by Linux are always <576bytes so
390 * they should go through unfragmented).
391 */
392 if (sk->sk_state == TCP_LISTEN)
393 goto out;
394
395 if (!ip6_sk_accept_pmtu(sk))
396 goto out;
397
398 tp->mtu_info = ntohl(info);
399 if (!sock_owned_by_user(sk))
400 tcp_v6_mtu_reduced(sk);
401 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
402 &sk->sk_tsq_flags))
403 sock_hold(sk);
404 goto out;
405 }
406
407
408 /* Might be for an request_sock */
409 switch (sk->sk_state) {
410 case TCP_SYN_SENT:
411 case TCP_SYN_RECV:
412 /* Only in fast or simultaneous open. If a fast open socket is
413 * is already accepted it is treated as a connected one below.
414 */
415 if (fastopen && !fastopen->sk)
416 break;
417
418 if (!sock_owned_by_user(sk)) {
419 sk->sk_err = err;
420 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
421
422 tcp_done(sk);
423 } else
424 sk->sk_err_soft = err;
425 goto out;
426 }
427
428 if (!sock_owned_by_user(sk) && np->recverr) {
429 sk->sk_err = err;
430 sk->sk_error_report(sk);
431 } else
432 sk->sk_err_soft = err;
433
434 out:
435 bh_unlock_sock(sk);
436 sock_put(sk);
437 }
438
439
440 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
441 struct flowi *fl,
442 struct request_sock *req,
443 struct tcp_fastopen_cookie *foc,
444 enum tcp_synack_type synack_type)
445 {
446 struct inet_request_sock *ireq = inet_rsk(req);
447 struct ipv6_pinfo *np = inet6_sk(sk);
448 struct ipv6_txoptions *opt;
449 struct flowi6 *fl6 = &fl->u.ip6;
450 struct sk_buff *skb;
451 int err = -ENOMEM;
452
453 /* First, grab a route. */
454 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
455 IPPROTO_TCP)) == NULL)
456 goto done;
457
458 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
459
460 if (skb) {
461 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
462 &ireq->ir_v6_rmt_addr);
463
464 fl6->daddr = ireq->ir_v6_rmt_addr;
465 if (np->repflow && ireq->pktopts)
466 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
467
468 rcu_read_lock();
469 opt = ireq->ipv6_opt;
470 if (!opt)
471 opt = rcu_dereference(np->opt);
472 err = ip6_xmit(sk, skb, fl6, opt, np->tclass);
473 rcu_read_unlock();
474 err = net_xmit_eval(err);
475 }
476
477 done:
478 return err;
479 }
480
481
482 static void tcp_v6_reqsk_destructor(struct request_sock *req)
483 {
484 kfree(inet_rsk(req)->ipv6_opt);
485 kfree_skb(inet_rsk(req)->pktopts);
486 }
487
488 #ifdef CONFIG_TCP_MD5SIG
489 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
490 const struct in6_addr *addr)
491 {
492 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
493 }
494
495 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
496 const struct sock *addr_sk)
497 {
498 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
499 }
500
501 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
502 int optlen)
503 {
504 struct tcp_md5sig cmd;
505 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
506
507 if (optlen < sizeof(cmd))
508 return -EINVAL;
509
510 if (copy_from_user(&cmd, optval, sizeof(cmd)))
511 return -EFAULT;
512
513 if (sin6->sin6_family != AF_INET6)
514 return -EINVAL;
515
516 if (!cmd.tcpm_keylen) {
517 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
518 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
519 AF_INET);
520 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
521 AF_INET6);
522 }
523
524 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
525 return -EINVAL;
526
527 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
528 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
529 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
530
531 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
532 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
533 }
534
535 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
536 const struct in6_addr *daddr,
537 const struct in6_addr *saddr,
538 const struct tcphdr *th, int nbytes)
539 {
540 struct tcp6_pseudohdr *bp;
541 struct scatterlist sg;
542 struct tcphdr *_th;
543
544 bp = hp->scratch;
545 /* 1. TCP pseudo-header (RFC2460) */
546 bp->saddr = *saddr;
547 bp->daddr = *daddr;
548 bp->protocol = cpu_to_be32(IPPROTO_TCP);
549 bp->len = cpu_to_be32(nbytes);
550
551 _th = (struct tcphdr *)(bp + 1);
552 memcpy(_th, th, sizeof(*th));
553 _th->check = 0;
554
555 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
556 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
557 sizeof(*bp) + sizeof(*th));
558 return crypto_ahash_update(hp->md5_req);
559 }
560
561 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
562 const struct in6_addr *daddr, struct in6_addr *saddr,
563 const struct tcphdr *th)
564 {
565 struct tcp_md5sig_pool *hp;
566 struct ahash_request *req;
567
568 hp = tcp_get_md5sig_pool();
569 if (!hp)
570 goto clear_hash_noput;
571 req = hp->md5_req;
572
573 if (crypto_ahash_init(req))
574 goto clear_hash;
575 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
576 goto clear_hash;
577 if (tcp_md5_hash_key(hp, key))
578 goto clear_hash;
579 ahash_request_set_crypt(req, NULL, md5_hash, 0);
580 if (crypto_ahash_final(req))
581 goto clear_hash;
582
583 tcp_put_md5sig_pool();
584 return 0;
585
586 clear_hash:
587 tcp_put_md5sig_pool();
588 clear_hash_noput:
589 memset(md5_hash, 0, 16);
590 return 1;
591 }
592
593 static int tcp_v6_md5_hash_skb(char *md5_hash,
594 const struct tcp_md5sig_key *key,
595 const struct sock *sk,
596 const struct sk_buff *skb)
597 {
598 const struct in6_addr *saddr, *daddr;
599 struct tcp_md5sig_pool *hp;
600 struct ahash_request *req;
601 const struct tcphdr *th = tcp_hdr(skb);
602
603 if (sk) { /* valid for establish/request sockets */
604 saddr = &sk->sk_v6_rcv_saddr;
605 daddr = &sk->sk_v6_daddr;
606 } else {
607 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
608 saddr = &ip6h->saddr;
609 daddr = &ip6h->daddr;
610 }
611
612 hp = tcp_get_md5sig_pool();
613 if (!hp)
614 goto clear_hash_noput;
615 req = hp->md5_req;
616
617 if (crypto_ahash_init(req))
618 goto clear_hash;
619
620 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
621 goto clear_hash;
622 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
623 goto clear_hash;
624 if (tcp_md5_hash_key(hp, key))
625 goto clear_hash;
626 ahash_request_set_crypt(req, NULL, md5_hash, 0);
627 if (crypto_ahash_final(req))
628 goto clear_hash;
629
630 tcp_put_md5sig_pool();
631 return 0;
632
633 clear_hash:
634 tcp_put_md5sig_pool();
635 clear_hash_noput:
636 memset(md5_hash, 0, 16);
637 return 1;
638 }
639
640 #endif
641
642 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
643 const struct sk_buff *skb)
644 {
645 #ifdef CONFIG_TCP_MD5SIG
646 const __u8 *hash_location = NULL;
647 struct tcp_md5sig_key *hash_expected;
648 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
649 const struct tcphdr *th = tcp_hdr(skb);
650 int genhash;
651 u8 newhash[16];
652
653 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
654 hash_location = tcp_parse_md5sig_option(th);
655
656 /* We've parsed the options - do we have a hash? */
657 if (!hash_expected && !hash_location)
658 return false;
659
660 if (hash_expected && !hash_location) {
661 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
662 return true;
663 }
664
665 if (!hash_expected && hash_location) {
666 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
667 return true;
668 }
669
670 /* check the signature */
671 genhash = tcp_v6_md5_hash_skb(newhash,
672 hash_expected,
673 NULL, skb);
674
675 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
676 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
677 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
678 genhash ? "failed" : "mismatch",
679 &ip6h->saddr, ntohs(th->source),
680 &ip6h->daddr, ntohs(th->dest));
681 return true;
682 }
683 #endif
684 return false;
685 }
686
687 static void tcp_v6_init_req(struct request_sock *req,
688 const struct sock *sk_listener,
689 struct sk_buff *skb)
690 {
691 struct inet_request_sock *ireq = inet_rsk(req);
692 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
693
694 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
695 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
696
697 /* So that link locals have meaning */
698 if (!sk_listener->sk_bound_dev_if &&
699 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
700 ireq->ir_iif = tcp_v6_iif(skb);
701
702 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
703 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
704 np->rxopt.bits.rxinfo ||
705 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
706 np->rxopt.bits.rxohlim || np->repflow)) {
707 atomic_inc(&skb->users);
708 ireq->pktopts = skb;
709 }
710 }
711
712 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
713 struct flowi *fl,
714 const struct request_sock *req,
715 bool *strict)
716 {
717 if (strict)
718 *strict = true;
719 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
720 }
721
722 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
723 .family = AF_INET6,
724 .obj_size = sizeof(struct tcp6_request_sock),
725 .rtx_syn_ack = tcp_rtx_synack,
726 .send_ack = tcp_v6_reqsk_send_ack,
727 .destructor = tcp_v6_reqsk_destructor,
728 .send_reset = tcp_v6_send_reset,
729 .syn_ack_timeout = tcp_syn_ack_timeout,
730 };
731
732 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
733 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
734 sizeof(struct ipv6hdr),
735 #ifdef CONFIG_TCP_MD5SIG
736 .req_md5_lookup = tcp_v6_md5_lookup,
737 .calc_md5_hash = tcp_v6_md5_hash_skb,
738 #endif
739 .init_req = tcp_v6_init_req,
740 #ifdef CONFIG_SYN_COOKIES
741 .cookie_init_seq = cookie_v6_init_sequence,
742 #endif
743 .route_req = tcp_v6_route_req,
744 .init_seq = tcp_v6_init_sequence,
745 .send_synack = tcp_v6_send_synack,
746 };
747
748 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
749 u32 ack, u32 win, u32 tsval, u32 tsecr,
750 int oif, struct tcp_md5sig_key *key, int rst,
751 u8 tclass, __be32 label)
752 {
753 const struct tcphdr *th = tcp_hdr(skb);
754 struct tcphdr *t1;
755 struct sk_buff *buff;
756 struct flowi6 fl6;
757 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
758 struct sock *ctl_sk = net->ipv6.tcp_sk;
759 unsigned int tot_len = sizeof(struct tcphdr);
760 struct dst_entry *dst;
761 __be32 *topt;
762
763 if (tsecr)
764 tot_len += TCPOLEN_TSTAMP_ALIGNED;
765 #ifdef CONFIG_TCP_MD5SIG
766 if (key)
767 tot_len += TCPOLEN_MD5SIG_ALIGNED;
768 #endif
769
770 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
771 GFP_ATOMIC);
772 if (!buff)
773 return;
774
775 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
776
777 t1 = (struct tcphdr *) skb_push(buff, tot_len);
778 skb_reset_transport_header(buff);
779
780 /* Swap the send and the receive. */
781 memset(t1, 0, sizeof(*t1));
782 t1->dest = th->source;
783 t1->source = th->dest;
784 t1->doff = tot_len / 4;
785 t1->seq = htonl(seq);
786 t1->ack_seq = htonl(ack);
787 t1->ack = !rst || !th->ack;
788 t1->rst = rst;
789 t1->window = htons(win);
790
791 topt = (__be32 *)(t1 + 1);
792
793 if (tsecr) {
794 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
795 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
796 *topt++ = htonl(tsval);
797 *topt++ = htonl(tsecr);
798 }
799
800 #ifdef CONFIG_TCP_MD5SIG
801 if (key) {
802 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
803 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
804 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
805 &ipv6_hdr(skb)->saddr,
806 &ipv6_hdr(skb)->daddr, t1);
807 }
808 #endif
809
810 memset(&fl6, 0, sizeof(fl6));
811 fl6.daddr = ipv6_hdr(skb)->saddr;
812 fl6.saddr = ipv6_hdr(skb)->daddr;
813 fl6.flowlabel = label;
814
815 buff->ip_summed = CHECKSUM_PARTIAL;
816 buff->csum = 0;
817
818 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
819
820 fl6.flowi6_proto = IPPROTO_TCP;
821 if (rt6_need_strict(&fl6.daddr) && !oif)
822 fl6.flowi6_oif = tcp_v6_iif(skb);
823 else {
824 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
825 oif = skb->skb_iif;
826
827 fl6.flowi6_oif = oif;
828 }
829
830 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
831 fl6.fl6_dport = t1->dest;
832 fl6.fl6_sport = t1->source;
833 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
834 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
835
836 /* Pass a socket to ip6_dst_lookup either it is for RST
837 * Underlying function will use this to retrieve the network
838 * namespace
839 */
840 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
841 if (!IS_ERR(dst)) {
842 skb_dst_set(buff, dst);
843 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
844 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
845 if (rst)
846 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
847 return;
848 }
849
850 kfree_skb(buff);
851 }
852
853 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
854 {
855 const struct tcphdr *th = tcp_hdr(skb);
856 u32 seq = 0, ack_seq = 0;
857 struct tcp_md5sig_key *key = NULL;
858 #ifdef CONFIG_TCP_MD5SIG
859 const __u8 *hash_location = NULL;
860 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
861 unsigned char newhash[16];
862 int genhash;
863 struct sock *sk1 = NULL;
864 #endif
865 int oif;
866
867 if (th->rst)
868 return;
869
870 /* If sk not NULL, it means we did a successful lookup and incoming
871 * route had to be correct. prequeue might have dropped our dst.
872 */
873 if (!sk && !ipv6_unicast_destination(skb))
874 return;
875
876 #ifdef CONFIG_TCP_MD5SIG
877 rcu_read_lock();
878 hash_location = tcp_parse_md5sig_option(th);
879 if (sk && sk_fullsock(sk)) {
880 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
881 } else if (hash_location) {
882 /*
883 * active side is lost. Try to find listening socket through
884 * source port, and then find md5 key through listening socket.
885 * we are not loose security here:
886 * Incoming packet is checked with md5 hash with finding key,
887 * no RST generated if md5 hash doesn't match.
888 */
889 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
890 &tcp_hashinfo, NULL, 0,
891 &ipv6h->saddr,
892 th->source, &ipv6h->daddr,
893 ntohs(th->source), tcp_v6_iif(skb));
894 if (!sk1)
895 goto out;
896
897 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
898 if (!key)
899 goto out;
900
901 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
902 if (genhash || memcmp(hash_location, newhash, 16) != 0)
903 goto out;
904 }
905 #endif
906
907 if (th->ack)
908 seq = ntohl(th->ack_seq);
909 else
910 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
911 (th->doff << 2);
912
913 oif = sk ? sk->sk_bound_dev_if : 0;
914 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
915
916 #ifdef CONFIG_TCP_MD5SIG
917 out:
918 rcu_read_unlock();
919 #endif
920 }
921
922 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
923 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
924 struct tcp_md5sig_key *key, u8 tclass,
925 __be32 label)
926 {
927 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
928 tclass, label);
929 }
930
931 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
932 {
933 struct inet_timewait_sock *tw = inet_twsk(sk);
934 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
935
936 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
937 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
938 tcp_time_stamp + tcptw->tw_ts_offset,
939 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
940 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
941
942 inet_twsk_put(tw);
943 }
944
945 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
946 struct request_sock *req)
947 {
948 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
949 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
950 */
951 /* RFC 7323 2.3
952 * The window field (SEG.WND) of every outgoing segment, with the
953 * exception of <SYN> segments, MUST be right-shifted by
954 * Rcv.Wind.Shift bits:
955 */
956 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
957 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
958 tcp_rsk(req)->rcv_nxt,
959 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
960 tcp_time_stamp + tcp_rsk(req)->ts_off,
961 req->ts_recent, sk->sk_bound_dev_if,
962 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
963 0, 0);
964 }
965
966
967 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
968 {
969 #ifdef CONFIG_SYN_COOKIES
970 const struct tcphdr *th = tcp_hdr(skb);
971
972 if (!th->syn)
973 sk = cookie_v6_check(sk, skb);
974 #endif
975 return sk;
976 }
977
978 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
979 {
980 if (skb->protocol == htons(ETH_P_IP))
981 return tcp_v4_conn_request(sk, skb);
982
983 if (!ipv6_unicast_destination(skb))
984 goto drop;
985
986 return tcp_conn_request(&tcp6_request_sock_ops,
987 &tcp_request_sock_ipv6_ops, sk, skb);
988
989 drop:
990 tcp_listendrop(sk);
991 return 0; /* don't send reset */
992 }
993
994 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
995 struct request_sock *req,
996 struct dst_entry *dst,
997 struct request_sock *req_unhash,
998 bool *own_req)
999 {
1000 struct inet_request_sock *ireq;
1001 struct ipv6_pinfo *newnp;
1002 const struct ipv6_pinfo *np = inet6_sk(sk);
1003 struct ipv6_txoptions *opt;
1004 struct tcp6_sock *newtcp6sk;
1005 struct inet_sock *newinet;
1006 struct tcp_sock *newtp;
1007 struct sock *newsk;
1008 #ifdef CONFIG_TCP_MD5SIG
1009 struct tcp_md5sig_key *key;
1010 #endif
1011 struct flowi6 fl6;
1012
1013 if (skb->protocol == htons(ETH_P_IP)) {
1014 /*
1015 * v6 mapped
1016 */
1017
1018 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1019 req_unhash, own_req);
1020
1021 if (!newsk)
1022 return NULL;
1023
1024 newtcp6sk = (struct tcp6_sock *)newsk;
1025 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1026
1027 newinet = inet_sk(newsk);
1028 newnp = inet6_sk(newsk);
1029 newtp = tcp_sk(newsk);
1030
1031 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1032
1033 newnp->saddr = newsk->sk_v6_rcv_saddr;
1034
1035 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1036 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1037 #ifdef CONFIG_TCP_MD5SIG
1038 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1039 #endif
1040
1041 newnp->ipv6_ac_list = NULL;
1042 newnp->ipv6_fl_list = NULL;
1043 newnp->pktoptions = NULL;
1044 newnp->opt = NULL;
1045 newnp->mcast_oif = tcp_v6_iif(skb);
1046 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1047 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1048 if (np->repflow)
1049 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1050
1051 /*
1052 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1053 * here, tcp_create_openreq_child now does this for us, see the comment in
1054 * that function for the gory details. -acme
1055 */
1056
1057 /* It is tricky place. Until this moment IPv4 tcp
1058 worked with IPv6 icsk.icsk_af_ops.
1059 Sync it now.
1060 */
1061 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1062
1063 return newsk;
1064 }
1065
1066 ireq = inet_rsk(req);
1067
1068 if (sk_acceptq_is_full(sk))
1069 goto out_overflow;
1070
1071 if (!dst) {
1072 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1073 if (!dst)
1074 goto out;
1075 }
1076
1077 newsk = tcp_create_openreq_child(sk, req, skb);
1078 if (!newsk)
1079 goto out_nonewsk;
1080
1081 /*
1082 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1083 * count here, tcp_create_openreq_child now does this for us, see the
1084 * comment in that function for the gory details. -acme
1085 */
1086
1087 newsk->sk_gso_type = SKB_GSO_TCPV6;
1088 ip6_dst_store(newsk, dst, NULL, NULL);
1089 inet6_sk_rx_dst_set(newsk, skb);
1090
1091 newtcp6sk = (struct tcp6_sock *)newsk;
1092 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1093
1094 newtp = tcp_sk(newsk);
1095 newinet = inet_sk(newsk);
1096 newnp = inet6_sk(newsk);
1097
1098 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1099
1100 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1101 newnp->saddr = ireq->ir_v6_loc_addr;
1102 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1103 newsk->sk_bound_dev_if = ireq->ir_iif;
1104
1105 /* Now IPv6 options...
1106
1107 First: no IPv4 options.
1108 */
1109 newinet->inet_opt = NULL;
1110 newnp->ipv6_ac_list = NULL;
1111 newnp->ipv6_fl_list = NULL;
1112
1113 /* Clone RX bits */
1114 newnp->rxopt.all = np->rxopt.all;
1115
1116 newnp->pktoptions = NULL;
1117 newnp->opt = NULL;
1118 newnp->mcast_oif = tcp_v6_iif(skb);
1119 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1120 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1121 if (np->repflow)
1122 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1123
1124 /* Clone native IPv6 options from listening socket (if any)
1125
1126 Yes, keeping reference count would be much more clever,
1127 but we make one more one thing there: reattach optmem
1128 to newsk.
1129 */
1130 opt = ireq->ipv6_opt;
1131 if (!opt)
1132 opt = rcu_dereference(np->opt);
1133 if (opt) {
1134 opt = ipv6_dup_options(newsk, opt);
1135 RCU_INIT_POINTER(newnp->opt, opt);
1136 }
1137 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1138 if (opt)
1139 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1140 opt->opt_flen;
1141
1142 tcp_ca_openreq_child(newsk, dst);
1143
1144 tcp_sync_mss(newsk, dst_mtu(dst));
1145 newtp->advmss = dst_metric_advmss(dst);
1146 if (tcp_sk(sk)->rx_opt.user_mss &&
1147 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1148 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1149
1150 tcp_initialize_rcv_mss(newsk);
1151
1152 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1153 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1154
1155 #ifdef CONFIG_TCP_MD5SIG
1156 /* Copy over the MD5 key from the original socket */
1157 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1158 if (key) {
1159 /* We're using one, so create a matching key
1160 * on the newsk structure. If we fail to get
1161 * memory, then we end up not copying the key
1162 * across. Shucks.
1163 */
1164 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1165 AF_INET6, key->key, key->keylen,
1166 sk_gfp_mask(sk, GFP_ATOMIC));
1167 }
1168 #endif
1169
1170 if (__inet_inherit_port(sk, newsk) < 0) {
1171 inet_csk_prepare_forced_close(newsk);
1172 tcp_done(newsk);
1173 goto out;
1174 }
1175 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1176 if (*own_req) {
1177 tcp_move_syn(newtp, req);
1178
1179 /* Clone pktoptions received with SYN, if we own the req */
1180 if (ireq->pktopts) {
1181 newnp->pktoptions = skb_clone(ireq->pktopts,
1182 sk_gfp_mask(sk, GFP_ATOMIC));
1183 consume_skb(ireq->pktopts);
1184 ireq->pktopts = NULL;
1185 if (newnp->pktoptions)
1186 skb_set_owner_r(newnp->pktoptions, newsk);
1187 }
1188 }
1189
1190 return newsk;
1191
1192 out_overflow:
1193 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1194 out_nonewsk:
1195 dst_release(dst);
1196 out:
1197 tcp_listendrop(sk);
1198 return NULL;
1199 }
1200
1201 static void tcp_v6_restore_cb(struct sk_buff *skb)
1202 {
1203 /* We need to move header back to the beginning if xfrm6_policy_check()
1204 * and tcp_v6_fill_cb() are going to be called again.
1205 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1206 */
1207 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1208 sizeof(struct inet6_skb_parm));
1209 }
1210
1211 /* The socket must have it's spinlock held when we get
1212 * here, unless it is a TCP_LISTEN socket.
1213 *
1214 * We have a potential double-lock case here, so even when
1215 * doing backlog processing we use the BH locking scheme.
1216 * This is because we cannot sleep with the original spinlock
1217 * held.
1218 */
1219 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1220 {
1221 struct ipv6_pinfo *np = inet6_sk(sk);
1222 struct tcp_sock *tp;
1223 struct sk_buff *opt_skb = NULL;
1224
1225 /* Imagine: socket is IPv6. IPv4 packet arrives,
1226 goes to IPv4 receive handler and backlogged.
1227 From backlog it always goes here. Kerboom...
1228 Fortunately, tcp_rcv_established and rcv_established
1229 handle them correctly, but it is not case with
1230 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1231 */
1232
1233 if (skb->protocol == htons(ETH_P_IP))
1234 return tcp_v4_do_rcv(sk, skb);
1235
1236 if (tcp_filter(sk, skb))
1237 goto discard;
1238
1239 /*
1240 * socket locking is here for SMP purposes as backlog rcv
1241 * is currently called with bh processing disabled.
1242 */
1243
1244 /* Do Stevens' IPV6_PKTOPTIONS.
1245
1246 Yes, guys, it is the only place in our code, where we
1247 may make it not affecting IPv4.
1248 The rest of code is protocol independent,
1249 and I do not like idea to uglify IPv4.
1250
1251 Actually, all the idea behind IPV6_PKTOPTIONS
1252 looks not very well thought. For now we latch
1253 options, received in the last packet, enqueued
1254 by tcp. Feel free to propose better solution.
1255 --ANK (980728)
1256 */
1257 if (np->rxopt.all)
1258 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1259
1260 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1261 struct dst_entry *dst = sk->sk_rx_dst;
1262
1263 sock_rps_save_rxhash(sk, skb);
1264 sk_mark_napi_id(sk, skb);
1265 if (dst) {
1266 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1267 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1268 dst_release(dst);
1269 sk->sk_rx_dst = NULL;
1270 }
1271 }
1272
1273 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1274 if (opt_skb)
1275 goto ipv6_pktoptions;
1276 return 0;
1277 }
1278
1279 if (tcp_checksum_complete(skb))
1280 goto csum_err;
1281
1282 if (sk->sk_state == TCP_LISTEN) {
1283 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1284
1285 if (!nsk)
1286 goto discard;
1287
1288 if (nsk != sk) {
1289 sock_rps_save_rxhash(nsk, skb);
1290 sk_mark_napi_id(nsk, skb);
1291 if (tcp_child_process(sk, nsk, skb))
1292 goto reset;
1293 if (opt_skb)
1294 __kfree_skb(opt_skb);
1295 return 0;
1296 }
1297 } else
1298 sock_rps_save_rxhash(sk, skb);
1299
1300 if (tcp_rcv_state_process(sk, skb))
1301 goto reset;
1302 if (opt_skb)
1303 goto ipv6_pktoptions;
1304 return 0;
1305
1306 reset:
1307 tcp_v6_send_reset(sk, skb);
1308 discard:
1309 if (opt_skb)
1310 __kfree_skb(opt_skb);
1311 kfree_skb(skb);
1312 return 0;
1313 csum_err:
1314 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1315 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1316 goto discard;
1317
1318
1319 ipv6_pktoptions:
1320 /* Do you ask, what is it?
1321
1322 1. skb was enqueued by tcp.
1323 2. skb is added to tail of read queue, rather than out of order.
1324 3. socket is not in passive state.
1325 4. Finally, it really contains options, which user wants to receive.
1326 */
1327 tp = tcp_sk(sk);
1328 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1329 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1330 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1331 np->mcast_oif = tcp_v6_iif(opt_skb);
1332 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1333 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1334 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1335 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1336 if (np->repflow)
1337 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1338 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1339 skb_set_owner_r(opt_skb, sk);
1340 tcp_v6_restore_cb(opt_skb);
1341 opt_skb = xchg(&np->pktoptions, opt_skb);
1342 } else {
1343 __kfree_skb(opt_skb);
1344 opt_skb = xchg(&np->pktoptions, NULL);
1345 }
1346 }
1347
1348 kfree_skb(opt_skb);
1349 return 0;
1350 }
1351
1352 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1353 const struct tcphdr *th)
1354 {
1355 /* This is tricky: we move IP6CB at its correct location into
1356 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1357 * _decode_session6() uses IP6CB().
1358 * barrier() makes sure compiler won't play aliasing games.
1359 */
1360 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1361 sizeof(struct inet6_skb_parm));
1362 barrier();
1363
1364 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1365 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1366 skb->len - th->doff*4);
1367 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1368 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1369 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1370 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1371 TCP_SKB_CB(skb)->sacked = 0;
1372 }
1373
1374 static int tcp_v6_rcv(struct sk_buff *skb)
1375 {
1376 const struct tcphdr *th;
1377 const struct ipv6hdr *hdr;
1378 bool refcounted;
1379 struct sock *sk;
1380 int ret;
1381 struct net *net = dev_net(skb->dev);
1382
1383 if (skb->pkt_type != PACKET_HOST)
1384 goto discard_it;
1385
1386 /*
1387 * Count it even if it's bad.
1388 */
1389 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1390
1391 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1392 goto discard_it;
1393
1394 th = (const struct tcphdr *)skb->data;
1395
1396 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1397 goto bad_packet;
1398 if (!pskb_may_pull(skb, th->doff*4))
1399 goto discard_it;
1400
1401 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1402 goto csum_error;
1403
1404 th = (const struct tcphdr *)skb->data;
1405 hdr = ipv6_hdr(skb);
1406
1407 lookup:
1408 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1409 th->source, th->dest, inet6_iif(skb),
1410 &refcounted);
1411 if (!sk)
1412 goto no_tcp_socket;
1413
1414 process:
1415 if (sk->sk_state == TCP_TIME_WAIT)
1416 goto do_time_wait;
1417
1418 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1419 struct request_sock *req = inet_reqsk(sk);
1420 struct sock *nsk;
1421
1422 sk = req->rsk_listener;
1423 tcp_v6_fill_cb(skb, hdr, th);
1424 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1425 sk_drops_add(sk, skb);
1426 reqsk_put(req);
1427 goto discard_it;
1428 }
1429 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1430 inet_csk_reqsk_queue_drop_and_put(sk, req);
1431 goto lookup;
1432 }
1433 sock_hold(sk);
1434 refcounted = true;
1435 nsk = tcp_check_req(sk, skb, req, false);
1436 if (!nsk) {
1437 reqsk_put(req);
1438 goto discard_and_relse;
1439 }
1440 if (nsk == sk) {
1441 reqsk_put(req);
1442 tcp_v6_restore_cb(skb);
1443 } else if (tcp_child_process(sk, nsk, skb)) {
1444 tcp_v6_send_reset(nsk, skb);
1445 goto discard_and_relse;
1446 } else {
1447 sock_put(sk);
1448 return 0;
1449 }
1450 }
1451 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1452 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1453 goto discard_and_relse;
1454 }
1455
1456 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1457 goto discard_and_relse;
1458
1459 tcp_v6_fill_cb(skb, hdr, th);
1460
1461 if (tcp_v6_inbound_md5_hash(sk, skb))
1462 goto discard_and_relse;
1463
1464 if (tcp_filter(sk, skb))
1465 goto discard_and_relse;
1466 th = (const struct tcphdr *)skb->data;
1467 hdr = ipv6_hdr(skb);
1468
1469 skb->dev = NULL;
1470
1471 if (sk->sk_state == TCP_LISTEN) {
1472 ret = tcp_v6_do_rcv(sk, skb);
1473 goto put_and_return;
1474 }
1475
1476 sk_incoming_cpu_update(sk);
1477
1478 bh_lock_sock_nested(sk);
1479 tcp_segs_in(tcp_sk(sk), skb);
1480 ret = 0;
1481 if (!sock_owned_by_user(sk)) {
1482 if (!tcp_prequeue(sk, skb))
1483 ret = tcp_v6_do_rcv(sk, skb);
1484 } else if (tcp_add_backlog(sk, skb)) {
1485 goto discard_and_relse;
1486 }
1487 bh_unlock_sock(sk);
1488
1489 put_and_return:
1490 if (refcounted)
1491 sock_put(sk);
1492 return ret ? -1 : 0;
1493
1494 no_tcp_socket:
1495 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1496 goto discard_it;
1497
1498 tcp_v6_fill_cb(skb, hdr, th);
1499
1500 if (tcp_checksum_complete(skb)) {
1501 csum_error:
1502 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1503 bad_packet:
1504 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1505 } else {
1506 tcp_v6_send_reset(NULL, skb);
1507 }
1508
1509 discard_it:
1510 kfree_skb(skb);
1511 return 0;
1512
1513 discard_and_relse:
1514 sk_drops_add(sk, skb);
1515 if (refcounted)
1516 sock_put(sk);
1517 goto discard_it;
1518
1519 do_time_wait:
1520 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1521 inet_twsk_put(inet_twsk(sk));
1522 goto discard_it;
1523 }
1524
1525 tcp_v6_fill_cb(skb, hdr, th);
1526
1527 if (tcp_checksum_complete(skb)) {
1528 inet_twsk_put(inet_twsk(sk));
1529 goto csum_error;
1530 }
1531
1532 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1533 case TCP_TW_SYN:
1534 {
1535 struct sock *sk2;
1536
1537 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1538 skb, __tcp_hdrlen(th),
1539 &ipv6_hdr(skb)->saddr, th->source,
1540 &ipv6_hdr(skb)->daddr,
1541 ntohs(th->dest), tcp_v6_iif(skb));
1542 if (sk2) {
1543 struct inet_timewait_sock *tw = inet_twsk(sk);
1544 inet_twsk_deschedule_put(tw);
1545 sk = sk2;
1546 tcp_v6_restore_cb(skb);
1547 refcounted = false;
1548 goto process;
1549 }
1550 /* Fall through to ACK */
1551 }
1552 case TCP_TW_ACK:
1553 tcp_v6_timewait_ack(sk, skb);
1554 break;
1555 case TCP_TW_RST:
1556 tcp_v6_restore_cb(skb);
1557 tcp_v6_send_reset(sk, skb);
1558 inet_twsk_deschedule_put(inet_twsk(sk));
1559 goto discard_it;
1560 case TCP_TW_SUCCESS:
1561 ;
1562 }
1563 goto discard_it;
1564 }
1565
1566 static void tcp_v6_early_demux(struct sk_buff *skb)
1567 {
1568 const struct ipv6hdr *hdr;
1569 const struct tcphdr *th;
1570 struct sock *sk;
1571
1572 if (skb->pkt_type != PACKET_HOST)
1573 return;
1574
1575 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1576 return;
1577
1578 hdr = ipv6_hdr(skb);
1579 th = tcp_hdr(skb);
1580
1581 if (th->doff < sizeof(struct tcphdr) / 4)
1582 return;
1583
1584 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1585 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1586 &hdr->saddr, th->source,
1587 &hdr->daddr, ntohs(th->dest),
1588 inet6_iif(skb));
1589 if (sk) {
1590 skb->sk = sk;
1591 skb->destructor = sock_edemux;
1592 if (sk_fullsock(sk)) {
1593 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1594
1595 if (dst)
1596 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1597 if (dst &&
1598 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1599 skb_dst_set_noref(skb, dst);
1600 }
1601 }
1602 }
1603
1604 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1605 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1606 .twsk_unique = tcp_twsk_unique,
1607 .twsk_destructor = tcp_twsk_destructor,
1608 };
1609
1610 static const struct inet_connection_sock_af_ops ipv6_specific = {
1611 .queue_xmit = inet6_csk_xmit,
1612 .send_check = tcp_v6_send_check,
1613 .rebuild_header = inet6_sk_rebuild_header,
1614 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1615 .conn_request = tcp_v6_conn_request,
1616 .syn_recv_sock = tcp_v6_syn_recv_sock,
1617 .net_header_len = sizeof(struct ipv6hdr),
1618 .net_frag_header_len = sizeof(struct frag_hdr),
1619 .setsockopt = ipv6_setsockopt,
1620 .getsockopt = ipv6_getsockopt,
1621 .addr2sockaddr = inet6_csk_addr2sockaddr,
1622 .sockaddr_len = sizeof(struct sockaddr_in6),
1623 .bind_conflict = inet6_csk_bind_conflict,
1624 #ifdef CONFIG_COMPAT
1625 .compat_setsockopt = compat_ipv6_setsockopt,
1626 .compat_getsockopt = compat_ipv6_getsockopt,
1627 #endif
1628 .mtu_reduced = tcp_v6_mtu_reduced,
1629 };
1630
1631 #ifdef CONFIG_TCP_MD5SIG
1632 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1633 .md5_lookup = tcp_v6_md5_lookup,
1634 .calc_md5_hash = tcp_v6_md5_hash_skb,
1635 .md5_parse = tcp_v6_parse_md5_keys,
1636 };
1637 #endif
1638
1639 /*
1640 * TCP over IPv4 via INET6 API
1641 */
1642 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1643 .queue_xmit = ip_queue_xmit,
1644 .send_check = tcp_v4_send_check,
1645 .rebuild_header = inet_sk_rebuild_header,
1646 .sk_rx_dst_set = inet_sk_rx_dst_set,
1647 .conn_request = tcp_v6_conn_request,
1648 .syn_recv_sock = tcp_v6_syn_recv_sock,
1649 .net_header_len = sizeof(struct iphdr),
1650 .setsockopt = ipv6_setsockopt,
1651 .getsockopt = ipv6_getsockopt,
1652 .addr2sockaddr = inet6_csk_addr2sockaddr,
1653 .sockaddr_len = sizeof(struct sockaddr_in6),
1654 .bind_conflict = inet6_csk_bind_conflict,
1655 #ifdef CONFIG_COMPAT
1656 .compat_setsockopt = compat_ipv6_setsockopt,
1657 .compat_getsockopt = compat_ipv6_getsockopt,
1658 #endif
1659 .mtu_reduced = tcp_v4_mtu_reduced,
1660 };
1661
1662 #ifdef CONFIG_TCP_MD5SIG
1663 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1664 .md5_lookup = tcp_v4_md5_lookup,
1665 .calc_md5_hash = tcp_v4_md5_hash_skb,
1666 .md5_parse = tcp_v6_parse_md5_keys,
1667 };
1668 #endif
1669
1670 /* NOTE: A lot of things set to zero explicitly by call to
1671 * sk_alloc() so need not be done here.
1672 */
1673 static int tcp_v6_init_sock(struct sock *sk)
1674 {
1675 struct inet_connection_sock *icsk = inet_csk(sk);
1676
1677 tcp_init_sock(sk);
1678
1679 icsk->icsk_af_ops = &ipv6_specific;
1680
1681 #ifdef CONFIG_TCP_MD5SIG
1682 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1683 #endif
1684
1685 return 0;
1686 }
1687
1688 static void tcp_v6_destroy_sock(struct sock *sk)
1689 {
1690 tcp_v4_destroy_sock(sk);
1691 inet6_destroy_sock(sk);
1692 }
1693
1694 #ifdef CONFIG_PROC_FS
1695 /* Proc filesystem TCPv6 sock list dumping. */
1696 static void get_openreq6(struct seq_file *seq,
1697 const struct request_sock *req, int i)
1698 {
1699 long ttd = req->rsk_timer.expires - jiffies;
1700 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1701 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1702
1703 if (ttd < 0)
1704 ttd = 0;
1705
1706 seq_printf(seq,
1707 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1708 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1709 i,
1710 src->s6_addr32[0], src->s6_addr32[1],
1711 src->s6_addr32[2], src->s6_addr32[3],
1712 inet_rsk(req)->ir_num,
1713 dest->s6_addr32[0], dest->s6_addr32[1],
1714 dest->s6_addr32[2], dest->s6_addr32[3],
1715 ntohs(inet_rsk(req)->ir_rmt_port),
1716 TCP_SYN_RECV,
1717 0, 0, /* could print option size, but that is af dependent. */
1718 1, /* timers active (only the expire timer) */
1719 jiffies_to_clock_t(ttd),
1720 req->num_timeout,
1721 from_kuid_munged(seq_user_ns(seq),
1722 sock_i_uid(req->rsk_listener)),
1723 0, /* non standard timer */
1724 0, /* open_requests have no inode */
1725 0, req);
1726 }
1727
1728 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1729 {
1730 const struct in6_addr *dest, *src;
1731 __u16 destp, srcp;
1732 int timer_active;
1733 unsigned long timer_expires;
1734 const struct inet_sock *inet = inet_sk(sp);
1735 const struct tcp_sock *tp = tcp_sk(sp);
1736 const struct inet_connection_sock *icsk = inet_csk(sp);
1737 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1738 int rx_queue;
1739 int state;
1740
1741 dest = &sp->sk_v6_daddr;
1742 src = &sp->sk_v6_rcv_saddr;
1743 destp = ntohs(inet->inet_dport);
1744 srcp = ntohs(inet->inet_sport);
1745
1746 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1747 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
1748 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1749 timer_active = 1;
1750 timer_expires = icsk->icsk_timeout;
1751 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1752 timer_active = 4;
1753 timer_expires = icsk->icsk_timeout;
1754 } else if (timer_pending(&sp->sk_timer)) {
1755 timer_active = 2;
1756 timer_expires = sp->sk_timer.expires;
1757 } else {
1758 timer_active = 0;
1759 timer_expires = jiffies;
1760 }
1761
1762 state = sk_state_load(sp);
1763 if (state == TCP_LISTEN)
1764 rx_queue = sp->sk_ack_backlog;
1765 else
1766 /* Because we don't lock the socket,
1767 * we might find a transient negative value.
1768 */
1769 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1770
1771 seq_printf(seq,
1772 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1773 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1774 i,
1775 src->s6_addr32[0], src->s6_addr32[1],
1776 src->s6_addr32[2], src->s6_addr32[3], srcp,
1777 dest->s6_addr32[0], dest->s6_addr32[1],
1778 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1779 state,
1780 tp->write_seq - tp->snd_una,
1781 rx_queue,
1782 timer_active,
1783 jiffies_delta_to_clock_t(timer_expires - jiffies),
1784 icsk->icsk_retransmits,
1785 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1786 icsk->icsk_probes_out,
1787 sock_i_ino(sp),
1788 atomic_read(&sp->sk_refcnt), sp,
1789 jiffies_to_clock_t(icsk->icsk_rto),
1790 jiffies_to_clock_t(icsk->icsk_ack.ato),
1791 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1792 tp->snd_cwnd,
1793 state == TCP_LISTEN ?
1794 fastopenq->max_qlen :
1795 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1796 );
1797 }
1798
1799 static void get_timewait6_sock(struct seq_file *seq,
1800 struct inet_timewait_sock *tw, int i)
1801 {
1802 long delta = tw->tw_timer.expires - jiffies;
1803 const struct in6_addr *dest, *src;
1804 __u16 destp, srcp;
1805
1806 dest = &tw->tw_v6_daddr;
1807 src = &tw->tw_v6_rcv_saddr;
1808 destp = ntohs(tw->tw_dport);
1809 srcp = ntohs(tw->tw_sport);
1810
1811 seq_printf(seq,
1812 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1813 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1814 i,
1815 src->s6_addr32[0], src->s6_addr32[1],
1816 src->s6_addr32[2], src->s6_addr32[3], srcp,
1817 dest->s6_addr32[0], dest->s6_addr32[1],
1818 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1819 tw->tw_substate, 0, 0,
1820 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1821 atomic_read(&tw->tw_refcnt), tw);
1822 }
1823
1824 static int tcp6_seq_show(struct seq_file *seq, void *v)
1825 {
1826 struct tcp_iter_state *st;
1827 struct sock *sk = v;
1828
1829 if (v == SEQ_START_TOKEN) {
1830 seq_puts(seq,
1831 " sl "
1832 "local_address "
1833 "remote_address "
1834 "st tx_queue rx_queue tr tm->when retrnsmt"
1835 " uid timeout inode\n");
1836 goto out;
1837 }
1838 st = seq->private;
1839
1840 if (sk->sk_state == TCP_TIME_WAIT)
1841 get_timewait6_sock(seq, v, st->num);
1842 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1843 get_openreq6(seq, v, st->num);
1844 else
1845 get_tcp6_sock(seq, v, st->num);
1846 out:
1847 return 0;
1848 }
1849
1850 static const struct file_operations tcp6_afinfo_seq_fops = {
1851 .owner = THIS_MODULE,
1852 .open = tcp_seq_open,
1853 .read = seq_read,
1854 .llseek = seq_lseek,
1855 .release = seq_release_net
1856 };
1857
1858 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1859 .name = "tcp6",
1860 .family = AF_INET6,
1861 .seq_fops = &tcp6_afinfo_seq_fops,
1862 .seq_ops = {
1863 .show = tcp6_seq_show,
1864 },
1865 };
1866
1867 int __net_init tcp6_proc_init(struct net *net)
1868 {
1869 return tcp_proc_register(net, &tcp6_seq_afinfo);
1870 }
1871
1872 void tcp6_proc_exit(struct net *net)
1873 {
1874 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1875 }
1876 #endif
1877
1878 struct proto tcpv6_prot = {
1879 .name = "TCPv6",
1880 .owner = THIS_MODULE,
1881 .close = tcp_close,
1882 .connect = tcp_v6_connect,
1883 .disconnect = tcp_disconnect,
1884 .accept = inet_csk_accept,
1885 .ioctl = tcp_ioctl,
1886 .init = tcp_v6_init_sock,
1887 .destroy = tcp_v6_destroy_sock,
1888 .shutdown = tcp_shutdown,
1889 .setsockopt = tcp_setsockopt,
1890 .getsockopt = tcp_getsockopt,
1891 .recvmsg = tcp_recvmsg,
1892 .sendmsg = tcp_sendmsg,
1893 .sendpage = tcp_sendpage,
1894 .backlog_rcv = tcp_v6_do_rcv,
1895 .release_cb = tcp_release_cb,
1896 .hash = inet6_hash,
1897 .unhash = inet_unhash,
1898 .get_port = inet_csk_get_port,
1899 .enter_memory_pressure = tcp_enter_memory_pressure,
1900 .stream_memory_free = tcp_stream_memory_free,
1901 .sockets_allocated = &tcp_sockets_allocated,
1902 .memory_allocated = &tcp_memory_allocated,
1903 .memory_pressure = &tcp_memory_pressure,
1904 .orphan_count = &tcp_orphan_count,
1905 .sysctl_mem = sysctl_tcp_mem,
1906 .sysctl_wmem = sysctl_tcp_wmem,
1907 .sysctl_rmem = sysctl_tcp_rmem,
1908 .max_header = MAX_TCP_HEADER,
1909 .obj_size = sizeof(struct tcp6_sock),
1910 .slab_flags = SLAB_DESTROY_BY_RCU,
1911 .twsk_prot = &tcp6_timewait_sock_ops,
1912 .rsk_prot = &tcp6_request_sock_ops,
1913 .h.hashinfo = &tcp_hashinfo,
1914 .no_autobind = true,
1915 #ifdef CONFIG_COMPAT
1916 .compat_setsockopt = compat_tcp_setsockopt,
1917 .compat_getsockopt = compat_tcp_getsockopt,
1918 #endif
1919 .diag_destroy = tcp_abort,
1920 };
1921
1922 static const struct inet6_protocol tcpv6_protocol = {
1923 .early_demux = tcp_v6_early_demux,
1924 .handler = tcp_v6_rcv,
1925 .err_handler = tcp_v6_err,
1926 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1927 };
1928
1929 static struct inet_protosw tcpv6_protosw = {
1930 .type = SOCK_STREAM,
1931 .protocol = IPPROTO_TCP,
1932 .prot = &tcpv6_prot,
1933 .ops = &inet6_stream_ops,
1934 .flags = INET_PROTOSW_PERMANENT |
1935 INET_PROTOSW_ICSK,
1936 };
1937
1938 static int __net_init tcpv6_net_init(struct net *net)
1939 {
1940 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1941 SOCK_RAW, IPPROTO_TCP, net);
1942 }
1943
1944 static void __net_exit tcpv6_net_exit(struct net *net)
1945 {
1946 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1947 }
1948
1949 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1950 {
1951 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1952 }
1953
1954 static struct pernet_operations tcpv6_net_ops = {
1955 .init = tcpv6_net_init,
1956 .exit = tcpv6_net_exit,
1957 .exit_batch = tcpv6_net_exit_batch,
1958 };
1959
1960 int __init tcpv6_init(void)
1961 {
1962 int ret;
1963
1964 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1965 if (ret)
1966 goto out;
1967
1968 /* register inet6 protocol */
1969 ret = inet6_register_protosw(&tcpv6_protosw);
1970 if (ret)
1971 goto out_tcpv6_protocol;
1972
1973 ret = register_pernet_subsys(&tcpv6_net_ops);
1974 if (ret)
1975 goto out_tcpv6_protosw;
1976 out:
1977 return ret;
1978
1979 out_tcpv6_protosw:
1980 inet6_unregister_protosw(&tcpv6_protosw);
1981 out_tcpv6_protocol:
1982 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1983 goto out;
1984 }
1985
1986 void tcpv6_exit(void)
1987 {
1988 unregister_pernet_subsys(&tcpv6_net_ops);
1989 inet6_unregister_protosw(&tcpv6_protosw);
1990 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1991 }