xref: /openbmc/linux/net/ipv6/tcp_ipv6.c (revision 8dde5715)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	TCP over IPv6
4  *	Linux INET6 implementation
5  *
6  *	Authors:
7  *	Pedro Roque		<roque@di.fc.ul.pt>
8  *
9  *	Based on:
10  *	linux/net/ipv4/tcp.c
11  *	linux/net/ipv4/tcp_input.c
12  *	linux/net/ipv4/tcp_output.c
13  *
14  *	Fixes:
15  *	Hideaki YOSHIFUJI	:	sin6_scope_id support
16  *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
17  *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
18  *					a single port at the same time.
19  *	YOSHIFUJI Hideaki @USAGI:	convert /proc/net/tcp6 to seq_file.
20  */
21 
22 #include <linux/bottom_half.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/jiffies.h>
30 #include <linux/in.h>
31 #include <linux/in6.h>
32 #include <linux/netdevice.h>
33 #include <linux/init.h>
34 #include <linux/jhash.h>
35 #include <linux/ipsec.h>
36 #include <linux/times.h>
37 #include <linux/slab.h>
38 #include <linux/uaccess.h>
39 #include <linux/ipv6.h>
40 #include <linux/icmpv6.h>
41 #include <linux/random.h>
42 #include <linux/indirect_call_wrapper.h>
43 
44 #include <net/tcp.h>
45 #include <net/ndisc.h>
46 #include <net/inet6_hashtables.h>
47 #include <net/inet6_connection_sock.h>
48 #include <net/ipv6.h>
49 #include <net/transp_v6.h>
50 #include <net/addrconf.h>
51 #include <net/ip6_route.h>
52 #include <net/ip6_checksum.h>
53 #include <net/inet_ecn.h>
54 #include <net/protocol.h>
55 #include <net/xfrm.h>
56 #include <net/snmp.h>
57 #include <net/dsfield.h>
58 #include <net/timewait_sock.h>
59 #include <net/inet_common.h>
60 #include <net/secure_seq.h>
61 #include <net/busy_poll.h>
62 
63 #include <linux/proc_fs.h>
64 #include <linux/seq_file.h>
65 
66 #include <crypto/hash.h>
67 #include <linux/scatterlist.h>
68 
69 #include <trace/events/tcp.h>
70 
71 static void	tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
72 static void	tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
73 				      struct request_sock *req);
74 
75 static int	tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
76 
77 static const struct inet_connection_sock_af_ops ipv6_mapped;
78 static const struct inet_connection_sock_af_ops ipv6_specific;
79 #ifdef CONFIG_TCP_MD5SIG
80 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
81 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
82 #else
83 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
84 						   const struct in6_addr *addr)
85 {
86 	return NULL;
87 }
88 #endif
89 
90 /* Helper returning the inet6 address from a given tcp socket.
91  * It can be used in TCP stack instead of inet6_sk(sk).
92  * This avoids a dereference and allow compiler optimizations.
93  * It is a specialized version of inet6_sk_generic().
94  */
95 static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk)
96 {
97 	unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo);
98 
99 	return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
100 }
101 
102 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
103 {
104 	struct dst_entry *dst = skb_dst(skb);
105 
106 	if (dst && dst_hold_safe(dst)) {
107 		const struct rt6_info *rt = (const struct rt6_info *)dst;
108 
109 		sk->sk_rx_dst = dst;
110 		inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
111 		tcp_inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
112 	}
113 }
114 
115 static u32 tcp_v6_init_seq(const struct sk_buff *skb)
116 {
117 	return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
118 				ipv6_hdr(skb)->saddr.s6_addr32,
119 				tcp_hdr(skb)->dest,
120 				tcp_hdr(skb)->source);
121 }
122 
123 static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
124 {
125 	return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
126 				   ipv6_hdr(skb)->saddr.s6_addr32);
127 }
128 
129 static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
130 			      int addr_len)
131 {
132 	/* This check is replicated from tcp_v6_connect() and intended to
133 	 * prevent BPF program called below from accessing bytes that are out
134 	 * of the bound specified by user in addr_len.
135 	 */
136 	if (addr_len < SIN6_LEN_RFC2133)
137 		return -EINVAL;
138 
139 	sock_owned_by_me(sk);
140 
141 	return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr);
142 }
143 
144 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
145 			  int addr_len)
146 {
147 	struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
148 	struct inet_sock *inet = inet_sk(sk);
149 	struct inet_connection_sock *icsk = inet_csk(sk);
150 	struct ipv6_pinfo *np = tcp_inet6_sk(sk);
151 	struct tcp_sock *tp = tcp_sk(sk);
152 	struct in6_addr *saddr = NULL, *final_p, final;
153 	struct ipv6_txoptions *opt;
154 	struct flowi6 fl6;
155 	struct dst_entry *dst;
156 	int addr_type;
157 	int err;
158 	struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
159 
160 	if (addr_len < SIN6_LEN_RFC2133)
161 		return -EINVAL;
162 
163 	if (usin->sin6_family != AF_INET6)
164 		return -EAFNOSUPPORT;
165 
166 	memset(&fl6, 0, sizeof(fl6));
167 
168 	if (np->sndflow) {
169 		fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
170 		IP6_ECN_flow_init(fl6.flowlabel);
171 		if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
172 			struct ip6_flowlabel *flowlabel;
173 			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
174 			if (!flowlabel)
175 				return -EINVAL;
176 			fl6_sock_release(flowlabel);
177 		}
178 	}
179 
180 	/*
181 	 *	connect() to INADDR_ANY means loopback (BSD'ism).
182 	 */
183 
184 	if (ipv6_addr_any(&usin->sin6_addr)) {
185 		if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
186 			ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
187 					       &usin->sin6_addr);
188 		else
189 			usin->sin6_addr = in6addr_loopback;
190 	}
191 
192 	addr_type = ipv6_addr_type(&usin->sin6_addr);
193 
194 	if (addr_type & IPV6_ADDR_MULTICAST)
195 		return -ENETUNREACH;
196 
197 	if (addr_type&IPV6_ADDR_LINKLOCAL) {
198 		if (addr_len >= sizeof(struct sockaddr_in6) &&
199 		    usin->sin6_scope_id) {
200 			/* If interface is set while binding, indices
201 			 * must coincide.
202 			 */
203 			if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
204 				return -EINVAL;
205 
206 			sk->sk_bound_dev_if = usin->sin6_scope_id;
207 		}
208 
209 		/* Connect to link-local address requires an interface */
210 		if (!sk->sk_bound_dev_if)
211 			return -EINVAL;
212 	}
213 
214 	if (tp->rx_opt.ts_recent_stamp &&
215 	    !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
216 		tp->rx_opt.ts_recent = 0;
217 		tp->rx_opt.ts_recent_stamp = 0;
218 		tp->write_seq = 0;
219 	}
220 
221 	sk->sk_v6_daddr = usin->sin6_addr;
222 	np->flow_label = fl6.flowlabel;
223 
224 	/*
225 	 *	TCP over IPv4
226 	 */
227 
228 	if (addr_type & IPV6_ADDR_MAPPED) {
229 		u32 exthdrlen = icsk->icsk_ext_hdr_len;
230 		struct sockaddr_in sin;
231 
232 		if (__ipv6_only_sock(sk))
233 			return -ENETUNREACH;
234 
235 		sin.sin_family = AF_INET;
236 		sin.sin_port = usin->sin6_port;
237 		sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
238 
239 		icsk->icsk_af_ops = &ipv6_mapped;
240 		sk->sk_backlog_rcv = tcp_v4_do_rcv;
241 #ifdef CONFIG_TCP_MD5SIG
242 		tp->af_specific = &tcp_sock_ipv6_mapped_specific;
243 #endif
244 
245 		err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
246 
247 		if (err) {
248 			icsk->icsk_ext_hdr_len = exthdrlen;
249 			icsk->icsk_af_ops = &ipv6_specific;
250 			sk->sk_backlog_rcv = tcp_v6_do_rcv;
251 #ifdef CONFIG_TCP_MD5SIG
252 			tp->af_specific = &tcp_sock_ipv6_specific;
253 #endif
254 			goto failure;
255 		}
256 		np->saddr = sk->sk_v6_rcv_saddr;
257 
258 		return err;
259 	}
260 
261 	if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
262 		saddr = &sk->sk_v6_rcv_saddr;
263 
264 	fl6.flowi6_proto = IPPROTO_TCP;
265 	fl6.daddr = sk->sk_v6_daddr;
266 	fl6.saddr = saddr ? *saddr : np->saddr;
267 	fl6.flowi6_oif = sk->sk_bound_dev_if;
268 	fl6.flowi6_mark = sk->sk_mark;
269 	fl6.fl6_dport = usin->sin6_port;
270 	fl6.fl6_sport = inet->inet_sport;
271 	fl6.flowi6_uid = sk->sk_uid;
272 
273 	opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
274 	final_p = fl6_update_dst(&fl6, opt, &final);
275 
276 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
277 
278 	dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
279 	if (IS_ERR(dst)) {
280 		err = PTR_ERR(dst);
281 		goto failure;
282 	}
283 
284 	if (!saddr) {
285 		saddr = &fl6.saddr;
286 		sk->sk_v6_rcv_saddr = *saddr;
287 	}
288 
289 	/* set the source address */
290 	np->saddr = *saddr;
291 	inet->inet_rcv_saddr = LOOPBACK4_IPV6;
292 
293 	sk->sk_gso_type = SKB_GSO_TCPV6;
294 	ip6_dst_store(sk, dst, NULL, NULL);
295 
296 	icsk->icsk_ext_hdr_len = 0;
297 	if (opt)
298 		icsk->icsk_ext_hdr_len = opt->opt_flen +
299 					 opt->opt_nflen;
300 
301 	tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
302 
303 	inet->inet_dport = usin->sin6_port;
304 
305 	tcp_set_state(sk, TCP_SYN_SENT);
306 	err = inet6_hash_connect(tcp_death_row, sk);
307 	if (err)
308 		goto late_failure;
309 
310 	sk_set_txhash(sk);
311 
312 	if (likely(!tp->repair)) {
313 		if (!tp->write_seq)
314 			tp->write_seq = secure_tcpv6_seq(np->saddr.s6_addr32,
315 							 sk->sk_v6_daddr.s6_addr32,
316 							 inet->inet_sport,
317 							 inet->inet_dport);
318 		tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
319 						   np->saddr.s6_addr32,
320 						   sk->sk_v6_daddr.s6_addr32);
321 	}
322 
323 	if (tcp_fastopen_defer_connect(sk, &err))
324 		return err;
325 	if (err)
326 		goto late_failure;
327 
328 	err = tcp_connect(sk);
329 	if (err)
330 		goto late_failure;
331 
332 	return 0;
333 
334 late_failure:
335 	tcp_set_state(sk, TCP_CLOSE);
336 failure:
337 	inet->inet_dport = 0;
338 	sk->sk_route_caps = 0;
339 	return err;
340 }
341 
342 static void tcp_v6_mtu_reduced(struct sock *sk)
343 {
344 	struct dst_entry *dst;
345 
346 	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
347 		return;
348 
349 	dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
350 	if (!dst)
351 		return;
352 
353 	if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
354 		tcp_sync_mss(sk, dst_mtu(dst));
355 		tcp_simple_retransmit(sk);
356 	}
357 }
358 
359 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
360 		u8 type, u8 code, int offset, __be32 info)
361 {
362 	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
363 	const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
364 	struct net *net = dev_net(skb->dev);
365 	struct request_sock *fastopen;
366 	struct ipv6_pinfo *np;
367 	struct tcp_sock *tp;
368 	__u32 seq, snd_una;
369 	struct sock *sk;
370 	bool fatal;
371 	int err;
372 
373 	sk = __inet6_lookup_established(net, &tcp_hashinfo,
374 					&hdr->daddr, th->dest,
375 					&hdr->saddr, ntohs(th->source),
376 					skb->dev->ifindex, inet6_sdif(skb));
377 
378 	if (!sk) {
379 		__ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
380 				  ICMP6_MIB_INERRORS);
381 		return -ENOENT;
382 	}
383 
384 	if (sk->sk_state == TCP_TIME_WAIT) {
385 		inet_twsk_put(inet_twsk(sk));
386 		return 0;
387 	}
388 	seq = ntohl(th->seq);
389 	fatal = icmpv6_err_convert(type, code, &err);
390 	if (sk->sk_state == TCP_NEW_SYN_RECV) {
391 		tcp_req_err(sk, seq, fatal);
392 		return 0;
393 	}
394 
395 	bh_lock_sock(sk);
396 	if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
397 		__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
398 
399 	if (sk->sk_state == TCP_CLOSE)
400 		goto out;
401 
402 	if (ipv6_hdr(skb)->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
403 		__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
404 		goto out;
405 	}
406 
407 	tp = tcp_sk(sk);
408 	/* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
409 	fastopen = tp->fastopen_rsk;
410 	snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
411 	if (sk->sk_state != TCP_LISTEN &&
412 	    !between(seq, snd_una, tp->snd_nxt)) {
413 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
414 		goto out;
415 	}
416 
417 	np = tcp_inet6_sk(sk);
418 
419 	if (type == NDISC_REDIRECT) {
420 		if (!sock_owned_by_user(sk)) {
421 			struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
422 
423 			if (dst)
424 				dst->ops->redirect(dst, sk, skb);
425 		}
426 		goto out;
427 	}
428 
429 	if (type == ICMPV6_PKT_TOOBIG) {
430 		/* We are not interested in TCP_LISTEN and open_requests
431 		 * (SYN-ACKs send out by Linux are always <576bytes so
432 		 * they should go through unfragmented).
433 		 */
434 		if (sk->sk_state == TCP_LISTEN)
435 			goto out;
436 
437 		if (!ip6_sk_accept_pmtu(sk))
438 			goto out;
439 
440 		tp->mtu_info = ntohl(info);
441 		if (!sock_owned_by_user(sk))
442 			tcp_v6_mtu_reduced(sk);
443 		else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
444 					   &sk->sk_tsq_flags))
445 			sock_hold(sk);
446 		goto out;
447 	}
448 
449 
450 	/* Might be for an request_sock */
451 	switch (sk->sk_state) {
452 	case TCP_SYN_SENT:
453 	case TCP_SYN_RECV:
454 		/* Only in fast or simultaneous open. If a fast open socket is
455 		 * is already accepted it is treated as a connected one below.
456 		 */
457 		if (fastopen && !fastopen->sk)
458 			break;
459 
460 		if (!sock_owned_by_user(sk)) {
461 			sk->sk_err = err;
462 			sk->sk_error_report(sk);		/* Wake people up to see the error (see connect in sock.c) */
463 
464 			tcp_done(sk);
465 		} else
466 			sk->sk_err_soft = err;
467 		goto out;
468 	}
469 
470 	if (!sock_owned_by_user(sk) && np->recverr) {
471 		sk->sk_err = err;
472 		sk->sk_error_report(sk);
473 	} else
474 		sk->sk_err_soft = err;
475 
476 out:
477 	bh_unlock_sock(sk);
478 	sock_put(sk);
479 	return 0;
480 }
481 
482 
483 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
484 			      struct flowi *fl,
485 			      struct request_sock *req,
486 			      struct tcp_fastopen_cookie *foc,
487 			      enum tcp_synack_type synack_type)
488 {
489 	struct inet_request_sock *ireq = inet_rsk(req);
490 	struct ipv6_pinfo *np = tcp_inet6_sk(sk);
491 	struct ipv6_txoptions *opt;
492 	struct flowi6 *fl6 = &fl->u.ip6;
493 	struct sk_buff *skb;
494 	int err = -ENOMEM;
495 
496 	/* First, grab a route. */
497 	if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
498 					       IPPROTO_TCP)) == NULL)
499 		goto done;
500 
501 	skb = tcp_make_synack(sk, dst, req, foc, synack_type);
502 
503 	if (skb) {
504 		__tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
505 				    &ireq->ir_v6_rmt_addr);
506 
507 		fl6->daddr = ireq->ir_v6_rmt_addr;
508 		if (np->repflow && ireq->pktopts)
509 			fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
510 
511 		rcu_read_lock();
512 		opt = ireq->ipv6_opt;
513 		if (!opt)
514 			opt = rcu_dereference(np->opt);
515 		err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass);
516 		rcu_read_unlock();
517 		err = net_xmit_eval(err);
518 	}
519 
520 done:
521 	return err;
522 }
523 
524 
525 static void tcp_v6_reqsk_destructor(struct request_sock *req)
526 {
527 	kfree(inet_rsk(req)->ipv6_opt);
528 	kfree_skb(inet_rsk(req)->pktopts);
529 }
530 
531 #ifdef CONFIG_TCP_MD5SIG
532 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
533 						   const struct in6_addr *addr)
534 {
535 	return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
536 }
537 
538 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
539 						const struct sock *addr_sk)
540 {
541 	return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
542 }
543 
544 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
545 				 char __user *optval, int optlen)
546 {
547 	struct tcp_md5sig cmd;
548 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
549 	u8 prefixlen;
550 
551 	if (optlen < sizeof(cmd))
552 		return -EINVAL;
553 
554 	if (copy_from_user(&cmd, optval, sizeof(cmd)))
555 		return -EFAULT;
556 
557 	if (sin6->sin6_family != AF_INET6)
558 		return -EINVAL;
559 
560 	if (optname == TCP_MD5SIG_EXT &&
561 	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
562 		prefixlen = cmd.tcpm_prefixlen;
563 		if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
564 					prefixlen > 32))
565 			return -EINVAL;
566 	} else {
567 		prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
568 	}
569 
570 	if (!cmd.tcpm_keylen) {
571 		if (ipv6_addr_v4mapped(&sin6->sin6_addr))
572 			return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
573 					      AF_INET, prefixlen);
574 		return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
575 				      AF_INET6, prefixlen);
576 	}
577 
578 	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
579 		return -EINVAL;
580 
581 	if (ipv6_addr_v4mapped(&sin6->sin6_addr))
582 		return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
583 				      AF_INET, prefixlen, cmd.tcpm_key,
584 				      cmd.tcpm_keylen, GFP_KERNEL);
585 
586 	return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
587 			      AF_INET6, prefixlen, cmd.tcpm_key,
588 			      cmd.tcpm_keylen, GFP_KERNEL);
589 }
590 
591 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
592 				   const struct in6_addr *daddr,
593 				   const struct in6_addr *saddr,
594 				   const struct tcphdr *th, int nbytes)
595 {
596 	struct tcp6_pseudohdr *bp;
597 	struct scatterlist sg;
598 	struct tcphdr *_th;
599 
600 	bp = hp->scratch;
601 	/* 1. TCP pseudo-header (RFC2460) */
602 	bp->saddr = *saddr;
603 	bp->daddr = *daddr;
604 	bp->protocol = cpu_to_be32(IPPROTO_TCP);
605 	bp->len = cpu_to_be32(nbytes);
606 
607 	_th = (struct tcphdr *)(bp + 1);
608 	memcpy(_th, th, sizeof(*th));
609 	_th->check = 0;
610 
611 	sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
612 	ahash_request_set_crypt(hp->md5_req, &sg, NULL,
613 				sizeof(*bp) + sizeof(*th));
614 	return crypto_ahash_update(hp->md5_req);
615 }
616 
617 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
618 			       const struct in6_addr *daddr, struct in6_addr *saddr,
619 			       const struct tcphdr *th)
620 {
621 	struct tcp_md5sig_pool *hp;
622 	struct ahash_request *req;
623 
624 	hp = tcp_get_md5sig_pool();
625 	if (!hp)
626 		goto clear_hash_noput;
627 	req = hp->md5_req;
628 
629 	if (crypto_ahash_init(req))
630 		goto clear_hash;
631 	if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
632 		goto clear_hash;
633 	if (tcp_md5_hash_key(hp, key))
634 		goto clear_hash;
635 	ahash_request_set_crypt(req, NULL, md5_hash, 0);
636 	if (crypto_ahash_final(req))
637 		goto clear_hash;
638 
639 	tcp_put_md5sig_pool();
640 	return 0;
641 
642 clear_hash:
643 	tcp_put_md5sig_pool();
644 clear_hash_noput:
645 	memset(md5_hash, 0, 16);
646 	return 1;
647 }
648 
649 static int tcp_v6_md5_hash_skb(char *md5_hash,
650 			       const struct tcp_md5sig_key *key,
651 			       const struct sock *sk,
652 			       const struct sk_buff *skb)
653 {
654 	const struct in6_addr *saddr, *daddr;
655 	struct tcp_md5sig_pool *hp;
656 	struct ahash_request *req;
657 	const struct tcphdr *th = tcp_hdr(skb);
658 
659 	if (sk) { /* valid for establish/request sockets */
660 		saddr = &sk->sk_v6_rcv_saddr;
661 		daddr = &sk->sk_v6_daddr;
662 	} else {
663 		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
664 		saddr = &ip6h->saddr;
665 		daddr = &ip6h->daddr;
666 	}
667 
668 	hp = tcp_get_md5sig_pool();
669 	if (!hp)
670 		goto clear_hash_noput;
671 	req = hp->md5_req;
672 
673 	if (crypto_ahash_init(req))
674 		goto clear_hash;
675 
676 	if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
677 		goto clear_hash;
678 	if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
679 		goto clear_hash;
680 	if (tcp_md5_hash_key(hp, key))
681 		goto clear_hash;
682 	ahash_request_set_crypt(req, NULL, md5_hash, 0);
683 	if (crypto_ahash_final(req))
684 		goto clear_hash;
685 
686 	tcp_put_md5sig_pool();
687 	return 0;
688 
689 clear_hash:
690 	tcp_put_md5sig_pool();
691 clear_hash_noput:
692 	memset(md5_hash, 0, 16);
693 	return 1;
694 }
695 
696 #endif
697 
698 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
699 				    const struct sk_buff *skb)
700 {
701 #ifdef CONFIG_TCP_MD5SIG
702 	const __u8 *hash_location = NULL;
703 	struct tcp_md5sig_key *hash_expected;
704 	const struct ipv6hdr *ip6h = ipv6_hdr(skb);
705 	const struct tcphdr *th = tcp_hdr(skb);
706 	int genhash;
707 	u8 newhash[16];
708 
709 	hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
710 	hash_location = tcp_parse_md5sig_option(th);
711 
712 	/* We've parsed the options - do we have a hash? */
713 	if (!hash_expected && !hash_location)
714 		return false;
715 
716 	if (hash_expected && !hash_location) {
717 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
718 		return true;
719 	}
720 
721 	if (!hash_expected && hash_location) {
722 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
723 		return true;
724 	}
725 
726 	/* check the signature */
727 	genhash = tcp_v6_md5_hash_skb(newhash,
728 				      hash_expected,
729 				      NULL, skb);
730 
731 	if (genhash || memcmp(hash_location, newhash, 16) != 0) {
732 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
733 		net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
734 				     genhash ? "failed" : "mismatch",
735 				     &ip6h->saddr, ntohs(th->source),
736 				     &ip6h->daddr, ntohs(th->dest));
737 		return true;
738 	}
739 #endif
740 	return false;
741 }
742 
743 static void tcp_v6_init_req(struct request_sock *req,
744 			    const struct sock *sk_listener,
745 			    struct sk_buff *skb)
746 {
747 	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
748 	struct inet_request_sock *ireq = inet_rsk(req);
749 	const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
750 
751 	ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
752 	ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
753 
754 	/* So that link locals have meaning */
755 	if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
756 	    ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
757 		ireq->ir_iif = tcp_v6_iif(skb);
758 
759 	if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
760 	    (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
761 	     np->rxopt.bits.rxinfo ||
762 	     np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
763 	     np->rxopt.bits.rxohlim || np->repflow)) {
764 		refcount_inc(&skb->users);
765 		ireq->pktopts = skb;
766 	}
767 }
768 
769 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
770 					  struct flowi *fl,
771 					  const struct request_sock *req)
772 {
773 	return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
774 }
775 
776 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
777 	.family		=	AF_INET6,
778 	.obj_size	=	sizeof(struct tcp6_request_sock),
779 	.rtx_syn_ack	=	tcp_rtx_synack,
780 	.send_ack	=	tcp_v6_reqsk_send_ack,
781 	.destructor	=	tcp_v6_reqsk_destructor,
782 	.send_reset	=	tcp_v6_send_reset,
783 	.syn_ack_timeout =	tcp_syn_ack_timeout,
784 };
785 
786 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
787 	.mss_clamp	=	IPV6_MIN_MTU - sizeof(struct tcphdr) -
788 				sizeof(struct ipv6hdr),
789 #ifdef CONFIG_TCP_MD5SIG
790 	.req_md5_lookup	=	tcp_v6_md5_lookup,
791 	.calc_md5_hash	=	tcp_v6_md5_hash_skb,
792 #endif
793 	.init_req	=	tcp_v6_init_req,
794 #ifdef CONFIG_SYN_COOKIES
795 	.cookie_init_seq =	cookie_v6_init_sequence,
796 #endif
797 	.route_req	=	tcp_v6_route_req,
798 	.init_seq	=	tcp_v6_init_seq,
799 	.init_ts_off	=	tcp_v6_init_ts_off,
800 	.send_synack	=	tcp_v6_send_synack,
801 };
802 
803 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
804 				 u32 ack, u32 win, u32 tsval, u32 tsecr,
805 				 int oif, struct tcp_md5sig_key *key, int rst,
806 				 u8 tclass, __be32 label)
807 {
808 	const struct tcphdr *th = tcp_hdr(skb);
809 	struct tcphdr *t1;
810 	struct sk_buff *buff;
811 	struct flowi6 fl6;
812 	struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
813 	struct sock *ctl_sk = net->ipv6.tcp_sk;
814 	unsigned int tot_len = sizeof(struct tcphdr);
815 	struct dst_entry *dst;
816 	__be32 *topt;
817 	__u32 mark = 0;
818 
819 	if (tsecr)
820 		tot_len += TCPOLEN_TSTAMP_ALIGNED;
821 #ifdef CONFIG_TCP_MD5SIG
822 	if (key)
823 		tot_len += TCPOLEN_MD5SIG_ALIGNED;
824 #endif
825 
826 	buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
827 			 GFP_ATOMIC);
828 	if (!buff)
829 		return;
830 
831 	skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
832 
833 	t1 = skb_push(buff, tot_len);
834 	skb_reset_transport_header(buff);
835 
836 	/* Swap the send and the receive. */
837 	memset(t1, 0, sizeof(*t1));
838 	t1->dest = th->source;
839 	t1->source = th->dest;
840 	t1->doff = tot_len / 4;
841 	t1->seq = htonl(seq);
842 	t1->ack_seq = htonl(ack);
843 	t1->ack = !rst || !th->ack;
844 	t1->rst = rst;
845 	t1->window = htons(win);
846 
847 	topt = (__be32 *)(t1 + 1);
848 
849 	if (tsecr) {
850 		*topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
851 				(TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
852 		*topt++ = htonl(tsval);
853 		*topt++ = htonl(tsecr);
854 	}
855 
856 #ifdef CONFIG_TCP_MD5SIG
857 	if (key) {
858 		*topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
859 				(TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
860 		tcp_v6_md5_hash_hdr((__u8 *)topt, key,
861 				    &ipv6_hdr(skb)->saddr,
862 				    &ipv6_hdr(skb)->daddr, t1);
863 	}
864 #endif
865 
866 	memset(&fl6, 0, sizeof(fl6));
867 	fl6.daddr = ipv6_hdr(skb)->saddr;
868 	fl6.saddr = ipv6_hdr(skb)->daddr;
869 	fl6.flowlabel = label;
870 
871 	buff->ip_summed = CHECKSUM_PARTIAL;
872 	buff->csum = 0;
873 
874 	__tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
875 
876 	fl6.flowi6_proto = IPPROTO_TCP;
877 	if (rt6_need_strict(&fl6.daddr) && !oif)
878 		fl6.flowi6_oif = tcp_v6_iif(skb);
879 	else {
880 		if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
881 			oif = skb->skb_iif;
882 
883 		fl6.flowi6_oif = oif;
884 	}
885 
886 	if (sk)
887 		mark = (sk->sk_state == TCP_TIME_WAIT) ?
888 			inet_twsk(sk)->tw_mark : sk->sk_mark;
889 	fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
890 	fl6.fl6_dport = t1->dest;
891 	fl6.fl6_sport = t1->source;
892 	fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
893 	security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
894 
895 	/* Pass a socket to ip6_dst_lookup either it is for RST
896 	 * Underlying function will use this to retrieve the network
897 	 * namespace
898 	 */
899 	dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
900 	if (!IS_ERR(dst)) {
901 		skb_dst_set(buff, dst);
902 		ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, tclass);
903 		TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
904 		if (rst)
905 			TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
906 		return;
907 	}
908 
909 	kfree_skb(buff);
910 }
911 
912 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
913 {
914 	const struct tcphdr *th = tcp_hdr(skb);
915 	u32 seq = 0, ack_seq = 0;
916 	struct tcp_md5sig_key *key = NULL;
917 #ifdef CONFIG_TCP_MD5SIG
918 	const __u8 *hash_location = NULL;
919 	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
920 	unsigned char newhash[16];
921 	int genhash;
922 	struct sock *sk1 = NULL;
923 #endif
924 	int oif = 0;
925 
926 	if (th->rst)
927 		return;
928 
929 	/* If sk not NULL, it means we did a successful lookup and incoming
930 	 * route had to be correct. prequeue might have dropped our dst.
931 	 */
932 	if (!sk && !ipv6_unicast_destination(skb))
933 		return;
934 
935 #ifdef CONFIG_TCP_MD5SIG
936 	rcu_read_lock();
937 	hash_location = tcp_parse_md5sig_option(th);
938 	if (sk && sk_fullsock(sk)) {
939 		key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
940 	} else if (hash_location) {
941 		/*
942 		 * active side is lost. Try to find listening socket through
943 		 * source port, and then find md5 key through listening socket.
944 		 * we are not loose security here:
945 		 * Incoming packet is checked with md5 hash with finding key,
946 		 * no RST generated if md5 hash doesn't match.
947 		 */
948 		sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
949 					   &tcp_hashinfo, NULL, 0,
950 					   &ipv6h->saddr,
951 					   th->source, &ipv6h->daddr,
952 					   ntohs(th->source),
953 					   tcp_v6_iif_l3_slave(skb),
954 					   tcp_v6_sdif(skb));
955 		if (!sk1)
956 			goto out;
957 
958 		key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
959 		if (!key)
960 			goto out;
961 
962 		genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
963 		if (genhash || memcmp(hash_location, newhash, 16) != 0)
964 			goto out;
965 	}
966 #endif
967 
968 	if (th->ack)
969 		seq = ntohl(th->ack_seq);
970 	else
971 		ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
972 			  (th->doff << 2);
973 
974 	if (sk) {
975 		oif = sk->sk_bound_dev_if;
976 		if (sk_fullsock(sk))
977 			trace_tcp_send_reset(sk, skb);
978 	}
979 
980 	tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
981 
982 #ifdef CONFIG_TCP_MD5SIG
983 out:
984 	rcu_read_unlock();
985 #endif
986 }
987 
988 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
989 			    u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
990 			    struct tcp_md5sig_key *key, u8 tclass,
991 			    __be32 label)
992 {
993 	tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
994 			     tclass, label);
995 }
996 
997 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
998 {
999 	struct inet_timewait_sock *tw = inet_twsk(sk);
1000 	struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1001 
1002 	tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1003 			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1004 			tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1005 			tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1006 			tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
1007 
1008 	inet_twsk_put(tw);
1009 }
1010 
1011 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1012 				  struct request_sock *req)
1013 {
1014 	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1015 	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1016 	 */
1017 	/* RFC 7323 2.3
1018 	 * The window field (SEG.WND) of every outgoing segment, with the
1019 	 * exception of <SYN> segments, MUST be right-shifted by
1020 	 * Rcv.Wind.Shift bits:
1021 	 */
1022 	tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1023 			tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1024 			tcp_rsk(req)->rcv_nxt,
1025 			req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1026 			tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1027 			req->ts_recent, sk->sk_bound_dev_if,
1028 			tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr),
1029 			0, 0);
1030 }
1031 
1032 
1033 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1034 {
1035 #ifdef CONFIG_SYN_COOKIES
1036 	const struct tcphdr *th = tcp_hdr(skb);
1037 
1038 	if (!th->syn)
1039 		sk = cookie_v6_check(sk, skb);
1040 #endif
1041 	return sk;
1042 }
1043 
1044 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1045 {
1046 	if (skb->protocol == htons(ETH_P_IP))
1047 		return tcp_v4_conn_request(sk, skb);
1048 
1049 	if (!ipv6_unicast_destination(skb))
1050 		goto drop;
1051 
1052 	return tcp_conn_request(&tcp6_request_sock_ops,
1053 				&tcp_request_sock_ipv6_ops, sk, skb);
1054 
1055 drop:
1056 	tcp_listendrop(sk);
1057 	return 0; /* don't send reset */
1058 }
1059 
1060 static void tcp_v6_restore_cb(struct sk_buff *skb)
1061 {
1062 	/* We need to move header back to the beginning if xfrm6_policy_check()
1063 	 * and tcp_v6_fill_cb() are going to be called again.
1064 	 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1065 	 */
1066 	memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1067 		sizeof(struct inet6_skb_parm));
1068 }
1069 
1070 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1071 					 struct request_sock *req,
1072 					 struct dst_entry *dst,
1073 					 struct request_sock *req_unhash,
1074 					 bool *own_req)
1075 {
1076 	struct inet_request_sock *ireq;
1077 	struct ipv6_pinfo *newnp;
1078 	const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1079 	struct ipv6_txoptions *opt;
1080 	struct inet_sock *newinet;
1081 	struct tcp_sock *newtp;
1082 	struct sock *newsk;
1083 #ifdef CONFIG_TCP_MD5SIG
1084 	struct tcp_md5sig_key *key;
1085 #endif
1086 	struct flowi6 fl6;
1087 
1088 	if (skb->protocol == htons(ETH_P_IP)) {
1089 		/*
1090 		 *	v6 mapped
1091 		 */
1092 
1093 		newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1094 					     req_unhash, own_req);
1095 
1096 		if (!newsk)
1097 			return NULL;
1098 
1099 		inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1100 
1101 		newinet = inet_sk(newsk);
1102 		newnp = tcp_inet6_sk(newsk);
1103 		newtp = tcp_sk(newsk);
1104 
1105 		memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1106 
1107 		newnp->saddr = newsk->sk_v6_rcv_saddr;
1108 
1109 		inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1110 		newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1111 #ifdef CONFIG_TCP_MD5SIG
1112 		newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1113 #endif
1114 
1115 		newnp->ipv6_mc_list = NULL;
1116 		newnp->ipv6_ac_list = NULL;
1117 		newnp->ipv6_fl_list = NULL;
1118 		newnp->pktoptions  = NULL;
1119 		newnp->opt	   = NULL;
1120 		newnp->mcast_oif   = inet_iif(skb);
1121 		newnp->mcast_hops  = ip_hdr(skb)->ttl;
1122 		newnp->rcv_flowinfo = 0;
1123 		if (np->repflow)
1124 			newnp->flow_label = 0;
1125 
1126 		/*
1127 		 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1128 		 * here, tcp_create_openreq_child now does this for us, see the comment in
1129 		 * that function for the gory details. -acme
1130 		 */
1131 
1132 		/* It is tricky place. Until this moment IPv4 tcp
1133 		   worked with IPv6 icsk.icsk_af_ops.
1134 		   Sync it now.
1135 		 */
1136 		tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1137 
1138 		return newsk;
1139 	}
1140 
1141 	ireq = inet_rsk(req);
1142 
1143 	if (sk_acceptq_is_full(sk))
1144 		goto out_overflow;
1145 
1146 	if (!dst) {
1147 		dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1148 		if (!dst)
1149 			goto out;
1150 	}
1151 
1152 	newsk = tcp_create_openreq_child(sk, req, skb);
1153 	if (!newsk)
1154 		goto out_nonewsk;
1155 
1156 	/*
1157 	 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1158 	 * count here, tcp_create_openreq_child now does this for us, see the
1159 	 * comment in that function for the gory details. -acme
1160 	 */
1161 
1162 	newsk->sk_gso_type = SKB_GSO_TCPV6;
1163 	ip6_dst_store(newsk, dst, NULL, NULL);
1164 	inet6_sk_rx_dst_set(newsk, skb);
1165 
1166 	inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1167 
1168 	newtp = tcp_sk(newsk);
1169 	newinet = inet_sk(newsk);
1170 	newnp = tcp_inet6_sk(newsk);
1171 
1172 	memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1173 
1174 	newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1175 	newnp->saddr = ireq->ir_v6_loc_addr;
1176 	newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1177 	newsk->sk_bound_dev_if = ireq->ir_iif;
1178 
1179 	/* Now IPv6 options...
1180 
1181 	   First: no IPv4 options.
1182 	 */
1183 	newinet->inet_opt = NULL;
1184 	newnp->ipv6_mc_list = NULL;
1185 	newnp->ipv6_ac_list = NULL;
1186 	newnp->ipv6_fl_list = NULL;
1187 
1188 	/* Clone RX bits */
1189 	newnp->rxopt.all = np->rxopt.all;
1190 
1191 	newnp->pktoptions = NULL;
1192 	newnp->opt	  = NULL;
1193 	newnp->mcast_oif  = tcp_v6_iif(skb);
1194 	newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1195 	newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1196 	if (np->repflow)
1197 		newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1198 
1199 	/* Clone native IPv6 options from listening socket (if any)
1200 
1201 	   Yes, keeping reference count would be much more clever,
1202 	   but we make one more one thing there: reattach optmem
1203 	   to newsk.
1204 	 */
1205 	opt = ireq->ipv6_opt;
1206 	if (!opt)
1207 		opt = rcu_dereference(np->opt);
1208 	if (opt) {
1209 		opt = ipv6_dup_options(newsk, opt);
1210 		RCU_INIT_POINTER(newnp->opt, opt);
1211 	}
1212 	inet_csk(newsk)->icsk_ext_hdr_len = 0;
1213 	if (opt)
1214 		inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1215 						    opt->opt_flen;
1216 
1217 	tcp_ca_openreq_child(newsk, dst);
1218 
1219 	tcp_sync_mss(newsk, dst_mtu(dst));
1220 	newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1221 
1222 	tcp_initialize_rcv_mss(newsk);
1223 
1224 	newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1225 	newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1226 
1227 #ifdef CONFIG_TCP_MD5SIG
1228 	/* Copy over the MD5 key from the original socket */
1229 	key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1230 	if (key) {
1231 		/* We're using one, so create a matching key
1232 		 * on the newsk structure. If we fail to get
1233 		 * memory, then we end up not copying the key
1234 		 * across. Shucks.
1235 		 */
1236 		tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1237 			       AF_INET6, 128, key->key, key->keylen,
1238 			       sk_gfp_mask(sk, GFP_ATOMIC));
1239 	}
1240 #endif
1241 
1242 	if (__inet_inherit_port(sk, newsk) < 0) {
1243 		inet_csk_prepare_forced_close(newsk);
1244 		tcp_done(newsk);
1245 		goto out;
1246 	}
1247 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1248 	if (*own_req) {
1249 		tcp_move_syn(newtp, req);
1250 
1251 		/* Clone pktoptions received with SYN, if we own the req */
1252 		if (ireq->pktopts) {
1253 			newnp->pktoptions = skb_clone(ireq->pktopts,
1254 						      sk_gfp_mask(sk, GFP_ATOMIC));
1255 			consume_skb(ireq->pktopts);
1256 			ireq->pktopts = NULL;
1257 			if (newnp->pktoptions) {
1258 				tcp_v6_restore_cb(newnp->pktoptions);
1259 				skb_set_owner_r(newnp->pktoptions, newsk);
1260 			}
1261 		}
1262 	}
1263 
1264 	return newsk;
1265 
1266 out_overflow:
1267 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1268 out_nonewsk:
1269 	dst_release(dst);
1270 out:
1271 	tcp_listendrop(sk);
1272 	return NULL;
1273 }
1274 
1275 /* The socket must have it's spinlock held when we get
1276  * here, unless it is a TCP_LISTEN socket.
1277  *
1278  * We have a potential double-lock case here, so even when
1279  * doing backlog processing we use the BH locking scheme.
1280  * This is because we cannot sleep with the original spinlock
1281  * held.
1282  */
1283 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1284 {
1285 	struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1286 	struct sk_buff *opt_skb = NULL;
1287 	struct tcp_sock *tp;
1288 
1289 	/* Imagine: socket is IPv6. IPv4 packet arrives,
1290 	   goes to IPv4 receive handler and backlogged.
1291 	   From backlog it always goes here. Kerboom...
1292 	   Fortunately, tcp_rcv_established and rcv_established
1293 	   handle them correctly, but it is not case with
1294 	   tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1295 	 */
1296 
1297 	if (skb->protocol == htons(ETH_P_IP))
1298 		return tcp_v4_do_rcv(sk, skb);
1299 
1300 	/*
1301 	 *	socket locking is here for SMP purposes as backlog rcv
1302 	 *	is currently called with bh processing disabled.
1303 	 */
1304 
1305 	/* Do Stevens' IPV6_PKTOPTIONS.
1306 
1307 	   Yes, guys, it is the only place in our code, where we
1308 	   may make it not affecting IPv4.
1309 	   The rest of code is protocol independent,
1310 	   and I do not like idea to uglify IPv4.
1311 
1312 	   Actually, all the idea behind IPV6_PKTOPTIONS
1313 	   looks not very well thought. For now we latch
1314 	   options, received in the last packet, enqueued
1315 	   by tcp. Feel free to propose better solution.
1316 					       --ANK (980728)
1317 	 */
1318 	if (np->rxopt.all)
1319 		opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1320 
1321 	if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1322 		struct dst_entry *dst = sk->sk_rx_dst;
1323 
1324 		sock_rps_save_rxhash(sk, skb);
1325 		sk_mark_napi_id(sk, skb);
1326 		if (dst) {
1327 			if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1328 			    dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1329 				dst_release(dst);
1330 				sk->sk_rx_dst = NULL;
1331 			}
1332 		}
1333 
1334 		tcp_rcv_established(sk, skb);
1335 		if (opt_skb)
1336 			goto ipv6_pktoptions;
1337 		return 0;
1338 	}
1339 
1340 	if (tcp_checksum_complete(skb))
1341 		goto csum_err;
1342 
1343 	if (sk->sk_state == TCP_LISTEN) {
1344 		struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1345 
1346 		if (!nsk)
1347 			goto discard;
1348 
1349 		if (nsk != sk) {
1350 			if (tcp_child_process(sk, nsk, skb))
1351 				goto reset;
1352 			if (opt_skb)
1353 				__kfree_skb(opt_skb);
1354 			return 0;
1355 		}
1356 	} else
1357 		sock_rps_save_rxhash(sk, skb);
1358 
1359 	if (tcp_rcv_state_process(sk, skb))
1360 		goto reset;
1361 	if (opt_skb)
1362 		goto ipv6_pktoptions;
1363 	return 0;
1364 
1365 reset:
1366 	tcp_v6_send_reset(sk, skb);
1367 discard:
1368 	if (opt_skb)
1369 		__kfree_skb(opt_skb);
1370 	kfree_skb(skb);
1371 	return 0;
1372 csum_err:
1373 	TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1374 	TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1375 	goto discard;
1376 
1377 
1378 ipv6_pktoptions:
1379 	/* Do you ask, what is it?
1380 
1381 	   1. skb was enqueued by tcp.
1382 	   2. skb is added to tail of read queue, rather than out of order.
1383 	   3. socket is not in passive state.
1384 	   4. Finally, it really contains options, which user wants to receive.
1385 	 */
1386 	tp = tcp_sk(sk);
1387 	if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1388 	    !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1389 		if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1390 			np->mcast_oif = tcp_v6_iif(opt_skb);
1391 		if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1392 			np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1393 		if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1394 			np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1395 		if (np->repflow)
1396 			np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1397 		if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1398 			skb_set_owner_r(opt_skb, sk);
1399 			tcp_v6_restore_cb(opt_skb);
1400 			opt_skb = xchg(&np->pktoptions, opt_skb);
1401 		} else {
1402 			__kfree_skb(opt_skb);
1403 			opt_skb = xchg(&np->pktoptions, NULL);
1404 		}
1405 	}
1406 
1407 	kfree_skb(opt_skb);
1408 	return 0;
1409 }
1410 
1411 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1412 			   const struct tcphdr *th)
1413 {
1414 	/* This is tricky: we move IP6CB at its correct location into
1415 	 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1416 	 * _decode_session6() uses IP6CB().
1417 	 * barrier() makes sure compiler won't play aliasing games.
1418 	 */
1419 	memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1420 		sizeof(struct inet6_skb_parm));
1421 	barrier();
1422 
1423 	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1424 	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1425 				    skb->len - th->doff*4);
1426 	TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1427 	TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1428 	TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1429 	TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1430 	TCP_SKB_CB(skb)->sacked = 0;
1431 	TCP_SKB_CB(skb)->has_rxtstamp =
1432 			skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1433 }
1434 
1435 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1436 {
1437 	struct sk_buff *skb_to_free;
1438 	int sdif = inet6_sdif(skb);
1439 	const struct tcphdr *th;
1440 	const struct ipv6hdr *hdr;
1441 	bool refcounted;
1442 	struct sock *sk;
1443 	int ret;
1444 	struct net *net = dev_net(skb->dev);
1445 
1446 	if (skb->pkt_type != PACKET_HOST)
1447 		goto discard_it;
1448 
1449 	/*
1450 	 *	Count it even if it's bad.
1451 	 */
1452 	__TCP_INC_STATS(net, TCP_MIB_INSEGS);
1453 
1454 	if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1455 		goto discard_it;
1456 
1457 	th = (const struct tcphdr *)skb->data;
1458 
1459 	if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1460 		goto bad_packet;
1461 	if (!pskb_may_pull(skb, th->doff*4))
1462 		goto discard_it;
1463 
1464 	if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1465 		goto csum_error;
1466 
1467 	th = (const struct tcphdr *)skb->data;
1468 	hdr = ipv6_hdr(skb);
1469 
1470 lookup:
1471 	sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1472 				th->source, th->dest, inet6_iif(skb), sdif,
1473 				&refcounted);
1474 	if (!sk)
1475 		goto no_tcp_socket;
1476 
1477 process:
1478 	if (sk->sk_state == TCP_TIME_WAIT)
1479 		goto do_time_wait;
1480 
1481 	if (sk->sk_state == TCP_NEW_SYN_RECV) {
1482 		struct request_sock *req = inet_reqsk(sk);
1483 		bool req_stolen = false;
1484 		struct sock *nsk;
1485 
1486 		sk = req->rsk_listener;
1487 		if (tcp_v6_inbound_md5_hash(sk, skb)) {
1488 			sk_drops_add(sk, skb);
1489 			reqsk_put(req);
1490 			goto discard_it;
1491 		}
1492 		if (tcp_checksum_complete(skb)) {
1493 			reqsk_put(req);
1494 			goto csum_error;
1495 		}
1496 		if (unlikely(sk->sk_state != TCP_LISTEN)) {
1497 			inet_csk_reqsk_queue_drop_and_put(sk, req);
1498 			goto lookup;
1499 		}
1500 		sock_hold(sk);
1501 		refcounted = true;
1502 		nsk = NULL;
1503 		if (!tcp_filter(sk, skb)) {
1504 			th = (const struct tcphdr *)skb->data;
1505 			hdr = ipv6_hdr(skb);
1506 			tcp_v6_fill_cb(skb, hdr, th);
1507 			nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1508 		}
1509 		if (!nsk) {
1510 			reqsk_put(req);
1511 			if (req_stolen) {
1512 				/* Another cpu got exclusive access to req
1513 				 * and created a full blown socket.
1514 				 * Try to feed this packet to this socket
1515 				 * instead of discarding it.
1516 				 */
1517 				tcp_v6_restore_cb(skb);
1518 				sock_put(sk);
1519 				goto lookup;
1520 			}
1521 			goto discard_and_relse;
1522 		}
1523 		if (nsk == sk) {
1524 			reqsk_put(req);
1525 			tcp_v6_restore_cb(skb);
1526 		} else if (tcp_child_process(sk, nsk, skb)) {
1527 			tcp_v6_send_reset(nsk, skb);
1528 			goto discard_and_relse;
1529 		} else {
1530 			sock_put(sk);
1531 			return 0;
1532 		}
1533 	}
1534 	if (hdr->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
1535 		__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1536 		goto discard_and_relse;
1537 	}
1538 
1539 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1540 		goto discard_and_relse;
1541 
1542 	if (tcp_v6_inbound_md5_hash(sk, skb))
1543 		goto discard_and_relse;
1544 
1545 	if (tcp_filter(sk, skb))
1546 		goto discard_and_relse;
1547 	th = (const struct tcphdr *)skb->data;
1548 	hdr = ipv6_hdr(skb);
1549 	tcp_v6_fill_cb(skb, hdr, th);
1550 
1551 	skb->dev = NULL;
1552 
1553 	if (sk->sk_state == TCP_LISTEN) {
1554 		ret = tcp_v6_do_rcv(sk, skb);
1555 		goto put_and_return;
1556 	}
1557 
1558 	sk_incoming_cpu_update(sk);
1559 
1560 	bh_lock_sock_nested(sk);
1561 	tcp_segs_in(tcp_sk(sk), skb);
1562 	ret = 0;
1563 	if (!sock_owned_by_user(sk)) {
1564 		skb_to_free = sk->sk_rx_skb_cache;
1565 		sk->sk_rx_skb_cache = NULL;
1566 		ret = tcp_v6_do_rcv(sk, skb);
1567 	} else {
1568 		if (tcp_add_backlog(sk, skb))
1569 			goto discard_and_relse;
1570 		skb_to_free = NULL;
1571 	}
1572 	bh_unlock_sock(sk);
1573 	if (skb_to_free)
1574 		__kfree_skb(skb_to_free);
1575 put_and_return:
1576 	if (refcounted)
1577 		sock_put(sk);
1578 	return ret ? -1 : 0;
1579 
1580 no_tcp_socket:
1581 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1582 		goto discard_it;
1583 
1584 	tcp_v6_fill_cb(skb, hdr, th);
1585 
1586 	if (tcp_checksum_complete(skb)) {
1587 csum_error:
1588 		__TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1589 bad_packet:
1590 		__TCP_INC_STATS(net, TCP_MIB_INERRS);
1591 	} else {
1592 		tcp_v6_send_reset(NULL, skb);
1593 	}
1594 
1595 discard_it:
1596 	kfree_skb(skb);
1597 	return 0;
1598 
1599 discard_and_relse:
1600 	sk_drops_add(sk, skb);
1601 	if (refcounted)
1602 		sock_put(sk);
1603 	goto discard_it;
1604 
1605 do_time_wait:
1606 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1607 		inet_twsk_put(inet_twsk(sk));
1608 		goto discard_it;
1609 	}
1610 
1611 	tcp_v6_fill_cb(skb, hdr, th);
1612 
1613 	if (tcp_checksum_complete(skb)) {
1614 		inet_twsk_put(inet_twsk(sk));
1615 		goto csum_error;
1616 	}
1617 
1618 	switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1619 	case TCP_TW_SYN:
1620 	{
1621 		struct sock *sk2;
1622 
1623 		sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1624 					    skb, __tcp_hdrlen(th),
1625 					    &ipv6_hdr(skb)->saddr, th->source,
1626 					    &ipv6_hdr(skb)->daddr,
1627 					    ntohs(th->dest),
1628 					    tcp_v6_iif_l3_slave(skb),
1629 					    sdif);
1630 		if (sk2) {
1631 			struct inet_timewait_sock *tw = inet_twsk(sk);
1632 			inet_twsk_deschedule_put(tw);
1633 			sk = sk2;
1634 			tcp_v6_restore_cb(skb);
1635 			refcounted = false;
1636 			goto process;
1637 		}
1638 	}
1639 		/* to ACK */
1640 		/* fall through */
1641 	case TCP_TW_ACK:
1642 		tcp_v6_timewait_ack(sk, skb);
1643 		break;
1644 	case TCP_TW_RST:
1645 		tcp_v6_send_reset(sk, skb);
1646 		inet_twsk_deschedule_put(inet_twsk(sk));
1647 		goto discard_it;
1648 	case TCP_TW_SUCCESS:
1649 		;
1650 	}
1651 	goto discard_it;
1652 }
1653 
1654 INDIRECT_CALLABLE_SCOPE void tcp_v6_early_demux(struct sk_buff *skb)
1655 {
1656 	const struct ipv6hdr *hdr;
1657 	const struct tcphdr *th;
1658 	struct sock *sk;
1659 
1660 	if (skb->pkt_type != PACKET_HOST)
1661 		return;
1662 
1663 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1664 		return;
1665 
1666 	hdr = ipv6_hdr(skb);
1667 	th = tcp_hdr(skb);
1668 
1669 	if (th->doff < sizeof(struct tcphdr) / 4)
1670 		return;
1671 
1672 	/* Note : We use inet6_iif() here, not tcp_v6_iif() */
1673 	sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1674 					&hdr->saddr, th->source,
1675 					&hdr->daddr, ntohs(th->dest),
1676 					inet6_iif(skb), inet6_sdif(skb));
1677 	if (sk) {
1678 		skb->sk = sk;
1679 		skb->destructor = sock_edemux;
1680 		if (sk_fullsock(sk)) {
1681 			struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1682 
1683 			if (dst)
1684 				dst = dst_check(dst, tcp_inet6_sk(sk)->rx_dst_cookie);
1685 			if (dst &&
1686 			    inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1687 				skb_dst_set_noref(skb, dst);
1688 		}
1689 	}
1690 }
1691 
1692 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1693 	.twsk_obj_size	= sizeof(struct tcp6_timewait_sock),
1694 	.twsk_unique	= tcp_twsk_unique,
1695 	.twsk_destructor = tcp_twsk_destructor,
1696 };
1697 
1698 static const struct inet_connection_sock_af_ops ipv6_specific = {
1699 	.queue_xmit	   = inet6_csk_xmit,
1700 	.send_check	   = tcp_v6_send_check,
1701 	.rebuild_header	   = inet6_sk_rebuild_header,
1702 	.sk_rx_dst_set	   = inet6_sk_rx_dst_set,
1703 	.conn_request	   = tcp_v6_conn_request,
1704 	.syn_recv_sock	   = tcp_v6_syn_recv_sock,
1705 	.net_header_len	   = sizeof(struct ipv6hdr),
1706 	.net_frag_header_len = sizeof(struct frag_hdr),
1707 	.setsockopt	   = ipv6_setsockopt,
1708 	.getsockopt	   = ipv6_getsockopt,
1709 	.addr2sockaddr	   = inet6_csk_addr2sockaddr,
1710 	.sockaddr_len	   = sizeof(struct sockaddr_in6),
1711 #ifdef CONFIG_COMPAT
1712 	.compat_setsockopt = compat_ipv6_setsockopt,
1713 	.compat_getsockopt = compat_ipv6_getsockopt,
1714 #endif
1715 	.mtu_reduced	   = tcp_v6_mtu_reduced,
1716 };
1717 
1718 #ifdef CONFIG_TCP_MD5SIG
1719 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1720 	.md5_lookup	=	tcp_v6_md5_lookup,
1721 	.calc_md5_hash	=	tcp_v6_md5_hash_skb,
1722 	.md5_parse	=	tcp_v6_parse_md5_keys,
1723 };
1724 #endif
1725 
1726 /*
1727  *	TCP over IPv4 via INET6 API
1728  */
1729 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1730 	.queue_xmit	   = ip_queue_xmit,
1731 	.send_check	   = tcp_v4_send_check,
1732 	.rebuild_header	   = inet_sk_rebuild_header,
1733 	.sk_rx_dst_set	   = inet_sk_rx_dst_set,
1734 	.conn_request	   = tcp_v6_conn_request,
1735 	.syn_recv_sock	   = tcp_v6_syn_recv_sock,
1736 	.net_header_len	   = sizeof(struct iphdr),
1737 	.setsockopt	   = ipv6_setsockopt,
1738 	.getsockopt	   = ipv6_getsockopt,
1739 	.addr2sockaddr	   = inet6_csk_addr2sockaddr,
1740 	.sockaddr_len	   = sizeof(struct sockaddr_in6),
1741 #ifdef CONFIG_COMPAT
1742 	.compat_setsockopt = compat_ipv6_setsockopt,
1743 	.compat_getsockopt = compat_ipv6_getsockopt,
1744 #endif
1745 	.mtu_reduced	   = tcp_v4_mtu_reduced,
1746 };
1747 
1748 #ifdef CONFIG_TCP_MD5SIG
1749 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1750 	.md5_lookup	=	tcp_v4_md5_lookup,
1751 	.calc_md5_hash	=	tcp_v4_md5_hash_skb,
1752 	.md5_parse	=	tcp_v6_parse_md5_keys,
1753 };
1754 #endif
1755 
1756 /* NOTE: A lot of things set to zero explicitly by call to
1757  *       sk_alloc() so need not be done here.
1758  */
1759 static int tcp_v6_init_sock(struct sock *sk)
1760 {
1761 	struct inet_connection_sock *icsk = inet_csk(sk);
1762 
1763 	tcp_init_sock(sk);
1764 
1765 	icsk->icsk_af_ops = &ipv6_specific;
1766 
1767 #ifdef CONFIG_TCP_MD5SIG
1768 	tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1769 #endif
1770 
1771 	return 0;
1772 }
1773 
1774 static void tcp_v6_destroy_sock(struct sock *sk)
1775 {
1776 	tcp_v4_destroy_sock(sk);
1777 	inet6_destroy_sock(sk);
1778 }
1779 
1780 #ifdef CONFIG_PROC_FS
1781 /* Proc filesystem TCPv6 sock list dumping. */
1782 static void get_openreq6(struct seq_file *seq,
1783 			 const struct request_sock *req, int i)
1784 {
1785 	long ttd = req->rsk_timer.expires - jiffies;
1786 	const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1787 	const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1788 
1789 	if (ttd < 0)
1790 		ttd = 0;
1791 
1792 	seq_printf(seq,
1793 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1794 		   "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1795 		   i,
1796 		   src->s6_addr32[0], src->s6_addr32[1],
1797 		   src->s6_addr32[2], src->s6_addr32[3],
1798 		   inet_rsk(req)->ir_num,
1799 		   dest->s6_addr32[0], dest->s6_addr32[1],
1800 		   dest->s6_addr32[2], dest->s6_addr32[3],
1801 		   ntohs(inet_rsk(req)->ir_rmt_port),
1802 		   TCP_SYN_RECV,
1803 		   0, 0, /* could print option size, but that is af dependent. */
1804 		   1,   /* timers active (only the expire timer) */
1805 		   jiffies_to_clock_t(ttd),
1806 		   req->num_timeout,
1807 		   from_kuid_munged(seq_user_ns(seq),
1808 				    sock_i_uid(req->rsk_listener)),
1809 		   0,  /* non standard timer */
1810 		   0, /* open_requests have no inode */
1811 		   0, req);
1812 }
1813 
1814 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1815 {
1816 	const struct in6_addr *dest, *src;
1817 	__u16 destp, srcp;
1818 	int timer_active;
1819 	unsigned long timer_expires;
1820 	const struct inet_sock *inet = inet_sk(sp);
1821 	const struct tcp_sock *tp = tcp_sk(sp);
1822 	const struct inet_connection_sock *icsk = inet_csk(sp);
1823 	const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1824 	int rx_queue;
1825 	int state;
1826 
1827 	dest  = &sp->sk_v6_daddr;
1828 	src   = &sp->sk_v6_rcv_saddr;
1829 	destp = ntohs(inet->inet_dport);
1830 	srcp  = ntohs(inet->inet_sport);
1831 
1832 	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1833 	    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
1834 	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1835 		timer_active	= 1;
1836 		timer_expires	= icsk->icsk_timeout;
1837 	} else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1838 		timer_active	= 4;
1839 		timer_expires	= icsk->icsk_timeout;
1840 	} else if (timer_pending(&sp->sk_timer)) {
1841 		timer_active	= 2;
1842 		timer_expires	= sp->sk_timer.expires;
1843 	} else {
1844 		timer_active	= 0;
1845 		timer_expires = jiffies;
1846 	}
1847 
1848 	state = inet_sk_state_load(sp);
1849 	if (state == TCP_LISTEN)
1850 		rx_queue = sp->sk_ack_backlog;
1851 	else
1852 		/* Because we don't lock the socket,
1853 		 * we might find a transient negative value.
1854 		 */
1855 		rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1856 
1857 	seq_printf(seq,
1858 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1859 		   "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1860 		   i,
1861 		   src->s6_addr32[0], src->s6_addr32[1],
1862 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
1863 		   dest->s6_addr32[0], dest->s6_addr32[1],
1864 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
1865 		   state,
1866 		   tp->write_seq - tp->snd_una,
1867 		   rx_queue,
1868 		   timer_active,
1869 		   jiffies_delta_to_clock_t(timer_expires - jiffies),
1870 		   icsk->icsk_retransmits,
1871 		   from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1872 		   icsk->icsk_probes_out,
1873 		   sock_i_ino(sp),
1874 		   refcount_read(&sp->sk_refcnt), sp,
1875 		   jiffies_to_clock_t(icsk->icsk_rto),
1876 		   jiffies_to_clock_t(icsk->icsk_ack.ato),
1877 		   (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
1878 		   tp->snd_cwnd,
1879 		   state == TCP_LISTEN ?
1880 			fastopenq->max_qlen :
1881 			(tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1882 		   );
1883 }
1884 
1885 static void get_timewait6_sock(struct seq_file *seq,
1886 			       struct inet_timewait_sock *tw, int i)
1887 {
1888 	long delta = tw->tw_timer.expires - jiffies;
1889 	const struct in6_addr *dest, *src;
1890 	__u16 destp, srcp;
1891 
1892 	dest = &tw->tw_v6_daddr;
1893 	src  = &tw->tw_v6_rcv_saddr;
1894 	destp = ntohs(tw->tw_dport);
1895 	srcp  = ntohs(tw->tw_sport);
1896 
1897 	seq_printf(seq,
1898 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1899 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1900 		   i,
1901 		   src->s6_addr32[0], src->s6_addr32[1],
1902 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
1903 		   dest->s6_addr32[0], dest->s6_addr32[1],
1904 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
1905 		   tw->tw_substate, 0, 0,
1906 		   3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1907 		   refcount_read(&tw->tw_refcnt), tw);
1908 }
1909 
1910 static int tcp6_seq_show(struct seq_file *seq, void *v)
1911 {
1912 	struct tcp_iter_state *st;
1913 	struct sock *sk = v;
1914 
1915 	if (v == SEQ_START_TOKEN) {
1916 		seq_puts(seq,
1917 			 "  sl  "
1918 			 "local_address                         "
1919 			 "remote_address                        "
1920 			 "st tx_queue rx_queue tr tm->when retrnsmt"
1921 			 "   uid  timeout inode\n");
1922 		goto out;
1923 	}
1924 	st = seq->private;
1925 
1926 	if (sk->sk_state == TCP_TIME_WAIT)
1927 		get_timewait6_sock(seq, v, st->num);
1928 	else if (sk->sk_state == TCP_NEW_SYN_RECV)
1929 		get_openreq6(seq, v, st->num);
1930 	else
1931 		get_tcp6_sock(seq, v, st->num);
1932 out:
1933 	return 0;
1934 }
1935 
1936 static const struct seq_operations tcp6_seq_ops = {
1937 	.show		= tcp6_seq_show,
1938 	.start		= tcp_seq_start,
1939 	.next		= tcp_seq_next,
1940 	.stop		= tcp_seq_stop,
1941 };
1942 
1943 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1944 	.family		= AF_INET6,
1945 };
1946 
1947 int __net_init tcp6_proc_init(struct net *net)
1948 {
1949 	if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
1950 			sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
1951 		return -ENOMEM;
1952 	return 0;
1953 }
1954 
1955 void tcp6_proc_exit(struct net *net)
1956 {
1957 	remove_proc_entry("tcp6", net->proc_net);
1958 }
1959 #endif
1960 
1961 struct proto tcpv6_prot = {
1962 	.name			= "TCPv6",
1963 	.owner			= THIS_MODULE,
1964 	.close			= tcp_close,
1965 	.pre_connect		= tcp_v6_pre_connect,
1966 	.connect		= tcp_v6_connect,
1967 	.disconnect		= tcp_disconnect,
1968 	.accept			= inet_csk_accept,
1969 	.ioctl			= tcp_ioctl,
1970 	.init			= tcp_v6_init_sock,
1971 	.destroy		= tcp_v6_destroy_sock,
1972 	.shutdown		= tcp_shutdown,
1973 	.setsockopt		= tcp_setsockopt,
1974 	.getsockopt		= tcp_getsockopt,
1975 	.keepalive		= tcp_set_keepalive,
1976 	.recvmsg		= tcp_recvmsg,
1977 	.sendmsg		= tcp_sendmsg,
1978 	.sendpage		= tcp_sendpage,
1979 	.backlog_rcv		= tcp_v6_do_rcv,
1980 	.release_cb		= tcp_release_cb,
1981 	.hash			= inet6_hash,
1982 	.unhash			= inet_unhash,
1983 	.get_port		= inet_csk_get_port,
1984 	.enter_memory_pressure	= tcp_enter_memory_pressure,
1985 	.leave_memory_pressure	= tcp_leave_memory_pressure,
1986 	.stream_memory_free	= tcp_stream_memory_free,
1987 	.sockets_allocated	= &tcp_sockets_allocated,
1988 	.memory_allocated	= &tcp_memory_allocated,
1989 	.memory_pressure	= &tcp_memory_pressure,
1990 	.orphan_count		= &tcp_orphan_count,
1991 	.sysctl_mem		= sysctl_tcp_mem,
1992 	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
1993 	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
1994 	.max_header		= MAX_TCP_HEADER,
1995 	.obj_size		= sizeof(struct tcp6_sock),
1996 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
1997 	.twsk_prot		= &tcp6_timewait_sock_ops,
1998 	.rsk_prot		= &tcp6_request_sock_ops,
1999 	.h.hashinfo		= &tcp_hashinfo,
2000 	.no_autobind		= true,
2001 #ifdef CONFIG_COMPAT
2002 	.compat_setsockopt	= compat_tcp_setsockopt,
2003 	.compat_getsockopt	= compat_tcp_getsockopt,
2004 #endif
2005 	.diag_destroy		= tcp_abort,
2006 };
2007 
2008 /* thinking of making this const? Don't.
2009  * early_demux can change based on sysctl.
2010  */
2011 static struct inet6_protocol tcpv6_protocol = {
2012 	.early_demux	=	tcp_v6_early_demux,
2013 	.early_demux_handler =  tcp_v6_early_demux,
2014 	.handler	=	tcp_v6_rcv,
2015 	.err_handler	=	tcp_v6_err,
2016 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2017 };
2018 
2019 static struct inet_protosw tcpv6_protosw = {
2020 	.type		=	SOCK_STREAM,
2021 	.protocol	=	IPPROTO_TCP,
2022 	.prot		=	&tcpv6_prot,
2023 	.ops		=	&inet6_stream_ops,
2024 	.flags		=	INET_PROTOSW_PERMANENT |
2025 				INET_PROTOSW_ICSK,
2026 };
2027 
2028 static int __net_init tcpv6_net_init(struct net *net)
2029 {
2030 	return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2031 				    SOCK_RAW, IPPROTO_TCP, net);
2032 }
2033 
2034 static void __net_exit tcpv6_net_exit(struct net *net)
2035 {
2036 	inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2037 }
2038 
2039 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2040 {
2041 	inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2042 }
2043 
2044 static struct pernet_operations tcpv6_net_ops = {
2045 	.init	    = tcpv6_net_init,
2046 	.exit	    = tcpv6_net_exit,
2047 	.exit_batch = tcpv6_net_exit_batch,
2048 };
2049 
2050 int __init tcpv6_init(void)
2051 {
2052 	int ret;
2053 
2054 	ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2055 	if (ret)
2056 		goto out;
2057 
2058 	/* register inet6 protocol */
2059 	ret = inet6_register_protosw(&tcpv6_protosw);
2060 	if (ret)
2061 		goto out_tcpv6_protocol;
2062 
2063 	ret = register_pernet_subsys(&tcpv6_net_ops);
2064 	if (ret)
2065 		goto out_tcpv6_protosw;
2066 out:
2067 	return ret;
2068 
2069 out_tcpv6_protosw:
2070 	inet6_unregister_protosw(&tcpv6_protosw);
2071 out_tcpv6_protocol:
2072 	inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2073 	goto out;
2074 }
2075 
2076 void tcpv6_exit(void)
2077 {
2078 	unregister_pernet_subsys(&tcpv6_net_ops);
2079 	inet6_unregister_protosw(&tcpv6_protosw);
2080 	inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2081 }
2082