xref: /linux/net/ipv6/tcp_ipv6.c (revision 8be4d31cb8aaeea27bde4b7ddb26e28a89062ebf)
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/aligned_data.h>
45 #include <net/tcp.h>
46 #include <net/ndisc.h>
47 #include <net/inet6_hashtables.h>
48 #include <net/inet6_connection_sock.h>
49 #include <net/ipv6.h>
50 #include <net/transp_v6.h>
51 #include <net/addrconf.h>
52 #include <net/ip6_route.h>
53 #include <net/ip6_checksum.h>
54 #include <net/inet_ecn.h>
55 #include <net/protocol.h>
56 #include <net/xfrm.h>
57 #include <net/snmp.h>
58 #include <net/dsfield.h>
59 #include <net/timewait_sock.h>
60 #include <net/inet_common.h>
61 #include <net/secure_seq.h>
62 #include <net/hotdata.h>
63 #include <net/busy_poll.h>
64 #include <net/rstreason.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 #include <trace/events/tcp.h>
73 
74 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb,
75 			      enum sk_rst_reason reason);
76 static void	tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
77 				      struct request_sock *req);
78 
79 INDIRECT_CALLABLE_SCOPE int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
80 
81 static const struct inet_connection_sock_af_ops ipv6_mapped;
82 const struct inet_connection_sock_af_ops ipv6_specific;
83 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
84 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
85 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
86 #endif
87 
88 /* Helper returning the inet6 address from a given tcp socket.
89  * It can be used in TCP stack instead of inet6_sk(sk).
90  * This avoids a dereference and allow compiler optimizations.
91  * It is a specialized version of inet6_sk_generic().
92  */
93 #define tcp_inet6_sk(sk) (&container_of_const(tcp_sk(sk), \
94 					      struct tcp6_sock, tcp)->inet6)
95 
inet6_sk_rx_dst_set(struct sock * sk,const struct sk_buff * skb)96 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
97 {
98 	struct dst_entry *dst = skb_dst(skb);
99 
100 	if (dst && dst_hold_safe(dst)) {
101 		rcu_assign_pointer(sk->sk_rx_dst, dst);
102 		sk->sk_rx_dst_ifindex = skb->skb_iif;
103 		sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst));
104 	}
105 }
106 
tcp_v6_init_seq(const struct sk_buff * skb)107 static u32 tcp_v6_init_seq(const struct sk_buff *skb)
108 {
109 	return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
110 				ipv6_hdr(skb)->saddr.s6_addr32,
111 				tcp_hdr(skb)->dest,
112 				tcp_hdr(skb)->source);
113 }
114 
tcp_v6_init_ts_off(const struct net * net,const struct sk_buff * skb)115 static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
116 {
117 	return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
118 				   ipv6_hdr(skb)->saddr.s6_addr32);
119 }
120 
tcp_v6_pre_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)121 static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
122 			      int addr_len)
123 {
124 	/* This check is replicated from tcp_v6_connect() and intended to
125 	 * prevent BPF program called below from accessing bytes that are out
126 	 * of the bound specified by user in addr_len.
127 	 */
128 	if (addr_len < SIN6_LEN_RFC2133)
129 		return -EINVAL;
130 
131 	sock_owned_by_me(sk);
132 
133 	return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr, &addr_len);
134 }
135 
tcp_v6_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)136 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
137 			  int addr_len)
138 {
139 	struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
140 	struct inet_connection_sock *icsk = inet_csk(sk);
141 	struct in6_addr *saddr = NULL, *final_p, final;
142 	struct inet_timewait_death_row *tcp_death_row;
143 	struct ipv6_pinfo *np = tcp_inet6_sk(sk);
144 	struct inet_sock *inet = inet_sk(sk);
145 	struct tcp_sock *tp = tcp_sk(sk);
146 	struct net *net = sock_net(sk);
147 	struct ipv6_txoptions *opt;
148 	struct dst_entry *dst;
149 	struct flowi6 fl6;
150 	int addr_type;
151 	int err;
152 
153 	if (addr_len < SIN6_LEN_RFC2133)
154 		return -EINVAL;
155 
156 	if (usin->sin6_family != AF_INET6)
157 		return -EAFNOSUPPORT;
158 
159 	memset(&fl6, 0, sizeof(fl6));
160 
161 	if (inet6_test_bit(SNDFLOW, sk)) {
162 		fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
163 		IP6_ECN_flow_init(fl6.flowlabel);
164 		if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
165 			struct ip6_flowlabel *flowlabel;
166 			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
167 			if (IS_ERR(flowlabel))
168 				return -EINVAL;
169 			fl6_sock_release(flowlabel);
170 		}
171 	}
172 
173 	/*
174 	 *	connect() to INADDR_ANY means loopback (BSD'ism).
175 	 */
176 
177 	if (ipv6_addr_any(&usin->sin6_addr)) {
178 		if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
179 			ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
180 					       &usin->sin6_addr);
181 		else
182 			usin->sin6_addr = in6addr_loopback;
183 	}
184 
185 	addr_type = ipv6_addr_type(&usin->sin6_addr);
186 
187 	if (addr_type & IPV6_ADDR_MULTICAST)
188 		return -ENETUNREACH;
189 
190 	if (addr_type&IPV6_ADDR_LINKLOCAL) {
191 		if (addr_len >= sizeof(struct sockaddr_in6) &&
192 		    usin->sin6_scope_id) {
193 			/* If interface is set while binding, indices
194 			 * must coincide.
195 			 */
196 			if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
197 				return -EINVAL;
198 
199 			sk->sk_bound_dev_if = usin->sin6_scope_id;
200 		}
201 
202 		/* Connect to link-local address requires an interface */
203 		if (!sk->sk_bound_dev_if)
204 			return -EINVAL;
205 	}
206 
207 	if (tp->rx_opt.ts_recent_stamp &&
208 	    !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
209 		tp->rx_opt.ts_recent = 0;
210 		tp->rx_opt.ts_recent_stamp = 0;
211 		WRITE_ONCE(tp->write_seq, 0);
212 	}
213 
214 	sk->sk_v6_daddr = usin->sin6_addr;
215 	np->flow_label = fl6.flowlabel;
216 
217 	/*
218 	 *	TCP over IPv4
219 	 */
220 
221 	if (addr_type & IPV6_ADDR_MAPPED) {
222 		u32 exthdrlen = icsk->icsk_ext_hdr_len;
223 		struct sockaddr_in sin;
224 
225 		if (ipv6_only_sock(sk))
226 			return -ENETUNREACH;
227 
228 		sin.sin_family = AF_INET;
229 		sin.sin_port = usin->sin6_port;
230 		sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
231 
232 		/* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
233 		WRITE_ONCE(icsk->icsk_af_ops, &ipv6_mapped);
234 		if (sk_is_mptcp(sk))
235 			mptcpv6_handle_mapped(sk, true);
236 		sk->sk_backlog_rcv = tcp_v4_do_rcv;
237 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
238 		tp->af_specific = &tcp_sock_ipv6_mapped_specific;
239 #endif
240 
241 		err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
242 
243 		if (err) {
244 			icsk->icsk_ext_hdr_len = exthdrlen;
245 			/* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
246 			WRITE_ONCE(icsk->icsk_af_ops, &ipv6_specific);
247 			if (sk_is_mptcp(sk))
248 				mptcpv6_handle_mapped(sk, false);
249 			sk->sk_backlog_rcv = tcp_v6_do_rcv;
250 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
251 			tp->af_specific = &tcp_sock_ipv6_specific;
252 #endif
253 			goto failure;
254 		}
255 		np->saddr = sk->sk_v6_rcv_saddr;
256 
257 		return err;
258 	}
259 
260 	if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
261 		saddr = &sk->sk_v6_rcv_saddr;
262 
263 	fl6.flowi6_proto = IPPROTO_TCP;
264 	fl6.daddr = sk->sk_v6_daddr;
265 	fl6.saddr = saddr ? *saddr : np->saddr;
266 	fl6.flowlabel = ip6_make_flowinfo(np->tclass, np->flow_label);
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 	if (IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) && !fl6.fl6_sport)
272 		fl6.flowi6_flags = FLOWI_FLAG_ANY_SPORT;
273 	fl6.flowi6_uid = sk_uid(sk);
274 
275 	opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
276 	final_p = fl6_update_dst(&fl6, opt, &final);
277 
278 	security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
279 
280 	dst = ip6_dst_lookup_flow(net, sk, &fl6, final_p);
281 	if (IS_ERR(dst)) {
282 		err = PTR_ERR(dst);
283 		goto failure;
284 	}
285 
286 	tp->tcp_usec_ts = dst_tcp_usec_ts(dst);
287 	tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
288 
289 	if (!saddr) {
290 		saddr = &fl6.saddr;
291 
292 		err = inet_bhash2_update_saddr(sk, saddr, AF_INET6);
293 		if (err)
294 			goto failure;
295 	}
296 
297 	/* set the source address */
298 	np->saddr = *saddr;
299 	inet->inet_rcv_saddr = LOOPBACK4_IPV6;
300 
301 	sk->sk_gso_type = SKB_GSO_TCPV6;
302 	ip6_dst_store(sk, dst, NULL, NULL);
303 
304 	icsk->icsk_ext_hdr_len = 0;
305 	if (opt)
306 		icsk->icsk_ext_hdr_len = opt->opt_flen +
307 					 opt->opt_nflen;
308 
309 	tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
310 
311 	inet->inet_dport = usin->sin6_port;
312 
313 	tcp_set_state(sk, TCP_SYN_SENT);
314 	err = inet6_hash_connect(tcp_death_row, sk);
315 	if (err)
316 		goto late_failure;
317 
318 	sk_set_txhash(sk);
319 
320 	if (likely(!tp->repair)) {
321 		if (!tp->write_seq)
322 			WRITE_ONCE(tp->write_seq,
323 				   secure_tcpv6_seq(np->saddr.s6_addr32,
324 						    sk->sk_v6_daddr.s6_addr32,
325 						    inet->inet_sport,
326 						    inet->inet_dport));
327 		tp->tsoffset = secure_tcpv6_ts_off(net, np->saddr.s6_addr32,
328 						   sk->sk_v6_daddr.s6_addr32);
329 	}
330 
331 	if (tcp_fastopen_defer_connect(sk, &err))
332 		return err;
333 	if (err)
334 		goto late_failure;
335 
336 	err = tcp_connect(sk);
337 	if (err)
338 		goto late_failure;
339 
340 	return 0;
341 
342 late_failure:
343 	tcp_set_state(sk, TCP_CLOSE);
344 	inet_bhash2_reset_saddr(sk);
345 failure:
346 	inet->inet_dport = 0;
347 	sk->sk_route_caps = 0;
348 	return err;
349 }
350 
tcp_v6_mtu_reduced(struct sock * sk)351 static void tcp_v6_mtu_reduced(struct sock *sk)
352 {
353 	struct dst_entry *dst;
354 	u32 mtu;
355 
356 	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
357 		return;
358 
359 	mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
360 
361 	/* Drop requests trying to increase our current mss.
362 	 * Check done in __ip6_rt_update_pmtu() is too late.
363 	 */
364 	if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache)
365 		return;
366 
367 	dst = inet6_csk_update_pmtu(sk, mtu);
368 	if (!dst)
369 		return;
370 
371 	if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
372 		tcp_sync_mss(sk, dst_mtu(dst));
373 		tcp_simple_retransmit(sk);
374 	}
375 }
376 
tcp_v6_err(struct sk_buff * skb,struct inet6_skb_parm * opt,u8 type,u8 code,int offset,__be32 info)377 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
378 		u8 type, u8 code, int offset, __be32 info)
379 {
380 	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
381 	const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
382 	struct net *net = dev_net_rcu(skb->dev);
383 	struct request_sock *fastopen;
384 	struct ipv6_pinfo *np;
385 	struct tcp_sock *tp;
386 	__u32 seq, snd_una;
387 	struct sock *sk;
388 	bool fatal;
389 	int err;
390 
391 	sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
392 					&hdr->daddr, th->dest,
393 					&hdr->saddr, ntohs(th->source),
394 					skb->dev->ifindex, inet6_sdif(skb));
395 
396 	if (!sk) {
397 		__ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
398 				  ICMP6_MIB_INERRORS);
399 		return -ENOENT;
400 	}
401 
402 	if (sk->sk_state == TCP_TIME_WAIT) {
403 		/* To increase the counter of ignored icmps for TCP-AO */
404 		tcp_ao_ignore_icmp(sk, AF_INET6, type, code);
405 		inet_twsk_put(inet_twsk(sk));
406 		return 0;
407 	}
408 	seq = ntohl(th->seq);
409 	fatal = icmpv6_err_convert(type, code, &err);
410 	if (sk->sk_state == TCP_NEW_SYN_RECV) {
411 		tcp_req_err(sk, seq, fatal);
412 		return 0;
413 	}
414 
415 	if (tcp_ao_ignore_icmp(sk, AF_INET6, type, code)) {
416 		sock_put(sk);
417 		return 0;
418 	}
419 
420 	bh_lock_sock(sk);
421 	if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
422 		__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
423 
424 	if (sk->sk_state == TCP_CLOSE)
425 		goto out;
426 
427 	if (static_branch_unlikely(&ip6_min_hopcount)) {
428 		/* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
429 		if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
430 			__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
431 			goto out;
432 		}
433 	}
434 
435 	tp = tcp_sk(sk);
436 	/* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
437 	fastopen = rcu_dereference(tp->fastopen_rsk);
438 	snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
439 	if (sk->sk_state != TCP_LISTEN &&
440 	    !between(seq, snd_una, tp->snd_nxt)) {
441 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
442 		goto out;
443 	}
444 
445 	np = tcp_inet6_sk(sk);
446 
447 	if (type == NDISC_REDIRECT) {
448 		if (!sock_owned_by_user(sk)) {
449 			struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
450 
451 			if (dst)
452 				dst->ops->redirect(dst, sk, skb);
453 		}
454 		goto out;
455 	}
456 
457 	if (type == ICMPV6_PKT_TOOBIG) {
458 		u32 mtu = ntohl(info);
459 
460 		/* We are not interested in TCP_LISTEN and open_requests
461 		 * (SYN-ACKs send out by Linux are always <576bytes so
462 		 * they should go through unfragmented).
463 		 */
464 		if (sk->sk_state == TCP_LISTEN)
465 			goto out;
466 
467 		if (!ip6_sk_accept_pmtu(sk))
468 			goto out;
469 
470 		if (mtu < IPV6_MIN_MTU)
471 			goto out;
472 
473 		WRITE_ONCE(tp->mtu_info, mtu);
474 
475 		if (!sock_owned_by_user(sk))
476 			tcp_v6_mtu_reduced(sk);
477 		else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
478 					   &sk->sk_tsq_flags))
479 			sock_hold(sk);
480 		goto out;
481 	}
482 
483 
484 	/* Might be for an request_sock */
485 	switch (sk->sk_state) {
486 	case TCP_SYN_SENT:
487 	case TCP_SYN_RECV:
488 		/* Only in fast or simultaneous open. If a fast open socket is
489 		 * already accepted it is treated as a connected one below.
490 		 */
491 		if (fastopen && !fastopen->sk)
492 			break;
493 
494 		ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
495 
496 		if (!sock_owned_by_user(sk))
497 			tcp_done_with_error(sk, err);
498 		else
499 			WRITE_ONCE(sk->sk_err_soft, err);
500 		goto out;
501 	case TCP_LISTEN:
502 		break;
503 	default:
504 		/* check if this ICMP message allows revert of backoff.
505 		 * (see RFC 6069)
506 		 */
507 		if (!fastopen && type == ICMPV6_DEST_UNREACH &&
508 		    code == ICMPV6_NOROUTE)
509 			tcp_ld_RTO_revert(sk, seq);
510 	}
511 
512 	if (!sock_owned_by_user(sk) && inet6_test_bit(RECVERR6, sk)) {
513 		WRITE_ONCE(sk->sk_err, err);
514 		sk_error_report(sk);
515 	} else {
516 		WRITE_ONCE(sk->sk_err_soft, err);
517 	}
518 out:
519 	bh_unlock_sock(sk);
520 	sock_put(sk);
521 	return 0;
522 }
523 
524 
tcp_v6_send_synack(const struct sock * sk,struct dst_entry * dst,struct flowi * fl,struct request_sock * req,struct tcp_fastopen_cookie * foc,enum tcp_synack_type synack_type,struct sk_buff * syn_skb)525 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
526 			      struct flowi *fl,
527 			      struct request_sock *req,
528 			      struct tcp_fastopen_cookie *foc,
529 			      enum tcp_synack_type synack_type,
530 			      struct sk_buff *syn_skb)
531 {
532 	struct inet_request_sock *ireq = inet_rsk(req);
533 	const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
534 	struct ipv6_txoptions *opt;
535 	struct flowi6 *fl6 = &fl->u.ip6;
536 	struct sk_buff *skb;
537 	int err = -ENOMEM;
538 	u8 tclass;
539 
540 	/* First, grab a route. */
541 	if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
542 					       IPPROTO_TCP)) == NULL)
543 		goto done;
544 
545 	skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
546 
547 	if (skb) {
548 		__tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
549 				    &ireq->ir_v6_rmt_addr);
550 
551 		fl6->daddr = ireq->ir_v6_rmt_addr;
552 		if (inet6_test_bit(REPFLOW, sk) && ireq->pktopts)
553 			fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
554 
555 		tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ?
556 				(tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
557 				(np->tclass & INET_ECN_MASK) :
558 				np->tclass;
559 
560 		if (!INET_ECN_is_capable(tclass) &&
561 		    tcp_bpf_ca_needs_ecn((struct sock *)req))
562 			tclass |= INET_ECN_ECT_0;
563 
564 		rcu_read_lock();
565 		opt = ireq->ipv6_opt;
566 		if (!opt)
567 			opt = rcu_dereference(np->opt);
568 		err = ip6_xmit(sk, skb, fl6, skb->mark ? : READ_ONCE(sk->sk_mark),
569 			       opt, tclass, READ_ONCE(sk->sk_priority));
570 		rcu_read_unlock();
571 		err = net_xmit_eval(err);
572 	}
573 
574 done:
575 	return err;
576 }
577 
578 
tcp_v6_reqsk_destructor(struct request_sock * req)579 static void tcp_v6_reqsk_destructor(struct request_sock *req)
580 {
581 	kfree(inet_rsk(req)->ipv6_opt);
582 	consume_skb(inet_rsk(req)->pktopts);
583 }
584 
585 #ifdef CONFIG_TCP_MD5SIG
tcp_v6_md5_do_lookup(const struct sock * sk,const struct in6_addr * addr,int l3index)586 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
587 						   const struct in6_addr *addr,
588 						   int l3index)
589 {
590 	return tcp_md5_do_lookup(sk, l3index,
591 				 (union tcp_md5_addr *)addr, AF_INET6);
592 }
593 
tcp_v6_md5_lookup(const struct sock * sk,const struct sock * addr_sk)594 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
595 						const struct sock *addr_sk)
596 {
597 	int l3index;
598 
599 	l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
600 						 addr_sk->sk_bound_dev_if);
601 	return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
602 				    l3index);
603 }
604 
tcp_v6_parse_md5_keys(struct sock * sk,int optname,sockptr_t optval,int optlen)605 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
606 				 sockptr_t optval, int optlen)
607 {
608 	struct tcp_md5sig cmd;
609 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
610 	union tcp_ao_addr *addr;
611 	int l3index = 0;
612 	u8 prefixlen;
613 	bool l3flag;
614 	u8 flags;
615 
616 	if (optlen < sizeof(cmd))
617 		return -EINVAL;
618 
619 	if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
620 		return -EFAULT;
621 
622 	if (sin6->sin6_family != AF_INET6)
623 		return -EINVAL;
624 
625 	flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
626 	l3flag = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
627 
628 	if (optname == TCP_MD5SIG_EXT &&
629 	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
630 		prefixlen = cmd.tcpm_prefixlen;
631 		if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
632 					prefixlen > 32))
633 			return -EINVAL;
634 	} else {
635 		prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
636 	}
637 
638 	if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
639 	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
640 		struct net_device *dev;
641 
642 		rcu_read_lock();
643 		dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
644 		if (dev && netif_is_l3_master(dev))
645 			l3index = dev->ifindex;
646 		rcu_read_unlock();
647 
648 		/* ok to reference set/not set outside of rcu;
649 		 * right now device MUST be an L3 master
650 		 */
651 		if (!dev || !l3index)
652 			return -EINVAL;
653 	}
654 
655 	if (!cmd.tcpm_keylen) {
656 		if (ipv6_addr_v4mapped(&sin6->sin6_addr))
657 			return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
658 					      AF_INET, prefixlen,
659 					      l3index, flags);
660 		return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
661 				      AF_INET6, prefixlen, l3index, flags);
662 	}
663 
664 	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
665 		return -EINVAL;
666 
667 	if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
668 		addr = (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3];
669 
670 		/* Don't allow keys for peers that have a matching TCP-AO key.
671 		 * See the comment in tcp_ao_add_cmd()
672 		 */
673 		if (tcp_ao_required(sk, addr, AF_INET,
674 				    l3flag ? l3index : -1, false))
675 			return -EKEYREJECTED;
676 		return tcp_md5_do_add(sk, addr,
677 				      AF_INET, prefixlen, l3index, flags,
678 				      cmd.tcpm_key, cmd.tcpm_keylen);
679 	}
680 
681 	addr = (union tcp_md5_addr *)&sin6->sin6_addr;
682 
683 	/* Don't allow keys for peers that have a matching TCP-AO key.
684 	 * See the comment in tcp_ao_add_cmd()
685 	 */
686 	if (tcp_ao_required(sk, addr, AF_INET6, l3flag ? l3index : -1, false))
687 		return -EKEYREJECTED;
688 
689 	return tcp_md5_do_add(sk, addr, AF_INET6, prefixlen, l3index, flags,
690 			      cmd.tcpm_key, cmd.tcpm_keylen);
691 }
692 
tcp_v6_md5_hash_headers(struct tcp_sigpool * hp,const struct in6_addr * daddr,const struct in6_addr * saddr,const struct tcphdr * th,int nbytes)693 static int tcp_v6_md5_hash_headers(struct tcp_sigpool *hp,
694 				   const struct in6_addr *daddr,
695 				   const struct in6_addr *saddr,
696 				   const struct tcphdr *th, int nbytes)
697 {
698 	struct tcp6_pseudohdr *bp;
699 	struct scatterlist sg;
700 	struct tcphdr *_th;
701 
702 	bp = hp->scratch;
703 	/* 1. TCP pseudo-header (RFC2460) */
704 	bp->saddr = *saddr;
705 	bp->daddr = *daddr;
706 	bp->protocol = cpu_to_be32(IPPROTO_TCP);
707 	bp->len = cpu_to_be32(nbytes);
708 
709 	_th = (struct tcphdr *)(bp + 1);
710 	memcpy(_th, th, sizeof(*th));
711 	_th->check = 0;
712 
713 	sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
714 	ahash_request_set_crypt(hp->req, &sg, NULL,
715 				sizeof(*bp) + sizeof(*th));
716 	return crypto_ahash_update(hp->req);
717 }
718 
tcp_v6_md5_hash_hdr(char * md5_hash,const struct tcp_md5sig_key * key,const struct in6_addr * daddr,struct in6_addr * saddr,const struct tcphdr * th)719 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
720 			       const struct in6_addr *daddr, struct in6_addr *saddr,
721 			       const struct tcphdr *th)
722 {
723 	struct tcp_sigpool hp;
724 
725 	if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
726 		goto clear_hash_nostart;
727 
728 	if (crypto_ahash_init(hp.req))
729 		goto clear_hash;
730 	if (tcp_v6_md5_hash_headers(&hp, daddr, saddr, th, th->doff << 2))
731 		goto clear_hash;
732 	if (tcp_md5_hash_key(&hp, key))
733 		goto clear_hash;
734 	ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
735 	if (crypto_ahash_final(hp.req))
736 		goto clear_hash;
737 
738 	tcp_sigpool_end(&hp);
739 	return 0;
740 
741 clear_hash:
742 	tcp_sigpool_end(&hp);
743 clear_hash_nostart:
744 	memset(md5_hash, 0, 16);
745 	return 1;
746 }
747 
tcp_v6_md5_hash_skb(char * md5_hash,const struct tcp_md5sig_key * key,const struct sock * sk,const struct sk_buff * skb)748 static int tcp_v6_md5_hash_skb(char *md5_hash,
749 			       const struct tcp_md5sig_key *key,
750 			       const struct sock *sk,
751 			       const struct sk_buff *skb)
752 {
753 	const struct tcphdr *th = tcp_hdr(skb);
754 	const struct in6_addr *saddr, *daddr;
755 	struct tcp_sigpool hp;
756 
757 	if (sk) { /* valid for establish/request sockets */
758 		saddr = &sk->sk_v6_rcv_saddr;
759 		daddr = &sk->sk_v6_daddr;
760 	} else {
761 		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
762 		saddr = &ip6h->saddr;
763 		daddr = &ip6h->daddr;
764 	}
765 
766 	if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
767 		goto clear_hash_nostart;
768 
769 	if (crypto_ahash_init(hp.req))
770 		goto clear_hash;
771 
772 	if (tcp_v6_md5_hash_headers(&hp, daddr, saddr, th, skb->len))
773 		goto clear_hash;
774 	if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2))
775 		goto clear_hash;
776 	if (tcp_md5_hash_key(&hp, key))
777 		goto clear_hash;
778 	ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
779 	if (crypto_ahash_final(hp.req))
780 		goto clear_hash;
781 
782 	tcp_sigpool_end(&hp);
783 	return 0;
784 
785 clear_hash:
786 	tcp_sigpool_end(&hp);
787 clear_hash_nostart:
788 	memset(md5_hash, 0, 16);
789 	return 1;
790 }
791 #endif
792 
tcp_v6_init_req(struct request_sock * req,const struct sock * sk_listener,struct sk_buff * skb,u32 tw_isn)793 static void tcp_v6_init_req(struct request_sock *req,
794 			    const struct sock *sk_listener,
795 			    struct sk_buff *skb,
796 			    u32 tw_isn)
797 {
798 	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
799 	struct inet_request_sock *ireq = inet_rsk(req);
800 	const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
801 
802 	ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
803 	ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
804 	ireq->ir_rmt_addr = LOOPBACK4_IPV6;
805 	ireq->ir_loc_addr = LOOPBACK4_IPV6;
806 
807 	/* So that link locals have meaning */
808 	if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
809 	    ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
810 		ireq->ir_iif = tcp_v6_iif(skb);
811 
812 	if (!tw_isn &&
813 	    (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
814 	     np->rxopt.bits.rxinfo ||
815 	     np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
816 	     np->rxopt.bits.rxohlim || inet6_test_bit(REPFLOW, sk_listener))) {
817 		refcount_inc(&skb->users);
818 		ireq->pktopts = skb;
819 	}
820 }
821 
tcp_v6_route_req(const struct sock * sk,struct sk_buff * skb,struct flowi * fl,struct request_sock * req,u32 tw_isn)822 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
823 					  struct sk_buff *skb,
824 					  struct flowi *fl,
825 					  struct request_sock *req,
826 					  u32 tw_isn)
827 {
828 	tcp_v6_init_req(req, sk, skb, tw_isn);
829 
830 	if (security_inet_conn_request(sk, skb, req))
831 		return NULL;
832 
833 	return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
834 }
835 
836 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
837 	.family		=	AF_INET6,
838 	.obj_size	=	sizeof(struct tcp6_request_sock),
839 	.send_ack	=	tcp_v6_reqsk_send_ack,
840 	.destructor	=	tcp_v6_reqsk_destructor,
841 	.send_reset	=	tcp_v6_send_reset,
842 	.syn_ack_timeout =	tcp_syn_ack_timeout,
843 };
844 
845 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
846 	.mss_clamp	=	IPV6_MIN_MTU - sizeof(struct tcphdr) -
847 				sizeof(struct ipv6hdr),
848 #ifdef CONFIG_TCP_MD5SIG
849 	.req_md5_lookup	=	tcp_v6_md5_lookup,
850 	.calc_md5_hash	=	tcp_v6_md5_hash_skb,
851 #endif
852 #ifdef CONFIG_TCP_AO
853 	.ao_lookup	=	tcp_v6_ao_lookup_rsk,
854 	.ao_calc_key	=	tcp_v6_ao_calc_key_rsk,
855 	.ao_synack_hash =	tcp_v6_ao_synack_hash,
856 #endif
857 #ifdef CONFIG_SYN_COOKIES
858 	.cookie_init_seq =	cookie_v6_init_sequence,
859 #endif
860 	.route_req	=	tcp_v6_route_req,
861 	.init_seq	=	tcp_v6_init_seq,
862 	.init_ts_off	=	tcp_v6_init_ts_off,
863 	.send_synack	=	tcp_v6_send_synack,
864 };
865 
tcp_v6_send_response(const struct sock * sk,struct sk_buff * skb,u32 seq,u32 ack,u32 win,u32 tsval,u32 tsecr,int oif,int rst,u8 tclass,__be32 label,u32 priority,u32 txhash,struct tcp_key * key)866 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
867 				 u32 ack, u32 win, u32 tsval, u32 tsecr,
868 				 int oif, int rst, u8 tclass, __be32 label,
869 				 u32 priority, u32 txhash, struct tcp_key *key)
870 {
871 	struct net *net = sk ? sock_net(sk) : skb_dst_dev_net_rcu(skb);
872 	unsigned int tot_len = sizeof(struct tcphdr);
873 	struct sock *ctl_sk = net->ipv6.tcp_sk;
874 	const struct tcphdr *th = tcp_hdr(skb);
875 	__be32 mrst = 0, *topt;
876 	struct dst_entry *dst;
877 	struct sk_buff *buff;
878 	struct tcphdr *t1;
879 	struct flowi6 fl6;
880 	u32 mark = 0;
881 
882 	if (tsecr)
883 		tot_len += TCPOLEN_TSTAMP_ALIGNED;
884 	if (tcp_key_is_md5(key))
885 		tot_len += TCPOLEN_MD5SIG_ALIGNED;
886 	if (tcp_key_is_ao(key))
887 		tot_len += tcp_ao_len_aligned(key->ao_key);
888 
889 #ifdef CONFIG_MPTCP
890 	if (rst && !tcp_key_is_md5(key)) {
891 		mrst = mptcp_reset_option(skb);
892 
893 		if (mrst)
894 			tot_len += sizeof(__be32);
895 	}
896 #endif
897 
898 	buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
899 	if (!buff)
900 		return;
901 
902 	skb_reserve(buff, MAX_TCP_HEADER);
903 
904 	t1 = skb_push(buff, tot_len);
905 	skb_reset_transport_header(buff);
906 
907 	/* Swap the send and the receive. */
908 	memset(t1, 0, sizeof(*t1));
909 	t1->dest = th->source;
910 	t1->source = th->dest;
911 	t1->doff = tot_len / 4;
912 	t1->seq = htonl(seq);
913 	t1->ack_seq = htonl(ack);
914 	t1->ack = !rst || !th->ack;
915 	t1->rst = rst;
916 	t1->window = htons(win);
917 
918 	topt = (__be32 *)(t1 + 1);
919 
920 	if (tsecr) {
921 		*topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
922 				(TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
923 		*topt++ = htonl(tsval);
924 		*topt++ = htonl(tsecr);
925 	}
926 
927 	if (mrst)
928 		*topt++ = mrst;
929 
930 #ifdef CONFIG_TCP_MD5SIG
931 	if (tcp_key_is_md5(key)) {
932 		*topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
933 				(TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
934 		tcp_v6_md5_hash_hdr((__u8 *)topt, key->md5_key,
935 				    &ipv6_hdr(skb)->saddr,
936 				    &ipv6_hdr(skb)->daddr, t1);
937 	}
938 #endif
939 #ifdef CONFIG_TCP_AO
940 	if (tcp_key_is_ao(key)) {
941 		*topt++ = htonl((TCPOPT_AO << 24) |
942 				(tcp_ao_len(key->ao_key) << 16) |
943 				(key->ao_key->sndid << 8) |
944 				(key->rcv_next));
945 
946 		tcp_ao_hash_hdr(AF_INET6, (char *)topt, key->ao_key,
947 				key->traffic_key,
948 				(union tcp_ao_addr *)&ipv6_hdr(skb)->saddr,
949 				(union tcp_ao_addr *)&ipv6_hdr(skb)->daddr,
950 				t1, key->sne);
951 	}
952 #endif
953 
954 	memset(&fl6, 0, sizeof(fl6));
955 	fl6.daddr = ipv6_hdr(skb)->saddr;
956 	fl6.saddr = ipv6_hdr(skb)->daddr;
957 	fl6.flowlabel = label;
958 
959 	buff->ip_summed = CHECKSUM_PARTIAL;
960 
961 	__tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
962 
963 	fl6.flowi6_proto = IPPROTO_TCP;
964 	if (rt6_need_strict(&fl6.daddr) && !oif)
965 		fl6.flowi6_oif = tcp_v6_iif(skb);
966 	else {
967 		if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
968 			oif = skb->skb_iif;
969 
970 		fl6.flowi6_oif = oif;
971 	}
972 
973 	if (sk) {
974 		/* unconstify the socket only to attach it to buff with care. */
975 		skb_set_owner_edemux(buff, (struct sock *)sk);
976 
977 		if (sk->sk_state == TCP_TIME_WAIT)
978 			mark = inet_twsk(sk)->tw_mark;
979 		else
980 			mark = READ_ONCE(sk->sk_mark);
981 		skb_set_delivery_time(buff, tcp_transmit_time(sk), SKB_CLOCK_MONOTONIC);
982 	}
983 	if (txhash) {
984 		/* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */
985 		skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4);
986 	}
987 	fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
988 	fl6.fl6_dport = t1->dest;
989 	fl6.fl6_sport = t1->source;
990 	fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
991 	security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
992 
993 	/* Pass a socket to ip6_dst_lookup either it is for RST
994 	 * Underlying function will use this to retrieve the network
995 	 * namespace
996 	 */
997 	if (sk && sk->sk_state != TCP_TIME_WAIT)
998 		dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/
999 	else
1000 		dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL);
1001 	if (!IS_ERR(dst)) {
1002 		skb_dst_set(buff, dst);
1003 		ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
1004 			 tclass, priority);
1005 		TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
1006 		if (rst)
1007 			TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
1008 		return;
1009 	}
1010 
1011 	kfree_skb(buff);
1012 }
1013 
tcp_v6_send_reset(const struct sock * sk,struct sk_buff * skb,enum sk_rst_reason reason)1014 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb,
1015 			      enum sk_rst_reason reason)
1016 {
1017 	const struct tcphdr *th = tcp_hdr(skb);
1018 	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
1019 	const __u8 *md5_hash_location = NULL;
1020 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1021 	bool allocated_traffic_key = false;
1022 #endif
1023 	const struct tcp_ao_hdr *aoh;
1024 	struct tcp_key key = {};
1025 	u32 seq = 0, ack_seq = 0;
1026 	__be32 label = 0;
1027 	u32 priority = 0;
1028 	struct net *net;
1029 	u32 txhash = 0;
1030 	int oif = 0;
1031 #ifdef CONFIG_TCP_MD5SIG
1032 	unsigned char newhash[16];
1033 	int genhash;
1034 	struct sock *sk1 = NULL;
1035 #endif
1036 
1037 	if (th->rst)
1038 		return;
1039 
1040 	/* If sk not NULL, it means we did a successful lookup and incoming
1041 	 * route had to be correct. prequeue might have dropped our dst.
1042 	 */
1043 	if (!sk && !ipv6_unicast_destination(skb))
1044 		return;
1045 
1046 	net = sk ? sock_net(sk) : skb_dst_dev_net_rcu(skb);
1047 	/* Invalid TCP option size or twice included auth */
1048 	if (tcp_parse_auth_options(th, &md5_hash_location, &aoh))
1049 		return;
1050 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1051 	rcu_read_lock();
1052 #endif
1053 #ifdef CONFIG_TCP_MD5SIG
1054 	if (sk && sk_fullsock(sk)) {
1055 		int l3index;
1056 
1057 		/* sdif set, means packet ingressed via a device
1058 		 * in an L3 domain and inet_iif is set to it.
1059 		 */
1060 		l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1061 		key.md5_key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1062 		if (key.md5_key)
1063 			key.type = TCP_KEY_MD5;
1064 	} else if (md5_hash_location) {
1065 		int dif = tcp_v6_iif_l3_slave(skb);
1066 		int sdif = tcp_v6_sdif(skb);
1067 		int l3index;
1068 
1069 		/*
1070 		 * active side is lost. Try to find listening socket through
1071 		 * source port, and then find md5 key through listening socket.
1072 		 * we are not loose security here:
1073 		 * Incoming packet is checked with md5 hash with finding key,
1074 		 * no RST generated if md5 hash doesn't match.
1075 		 */
1076 		sk1 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1077 					    NULL, 0, &ipv6h->saddr, th->source,
1078 					    &ipv6h->daddr, ntohs(th->source),
1079 					    dif, sdif);
1080 		if (!sk1)
1081 			goto out;
1082 
1083 		/* sdif set, means packet ingressed via a device
1084 		 * in an L3 domain and dif is set to it.
1085 		 */
1086 		l3index = tcp_v6_sdif(skb) ? dif : 0;
1087 
1088 		key.md5_key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1089 		if (!key.md5_key)
1090 			goto out;
1091 		key.type = TCP_KEY_MD5;
1092 
1093 		genhash = tcp_v6_md5_hash_skb(newhash, key.md5_key, NULL, skb);
1094 		if (genhash || memcmp(md5_hash_location, newhash, 16) != 0)
1095 			goto out;
1096 	}
1097 #endif
1098 
1099 	if (th->ack)
1100 		seq = ntohl(th->ack_seq);
1101 	else
1102 		ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1103 			  (th->doff << 2);
1104 
1105 #ifdef CONFIG_TCP_AO
1106 	if (aoh) {
1107 		int l3index;
1108 
1109 		l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1110 		if (tcp_ao_prepare_reset(sk, skb, aoh, l3index, seq,
1111 					 &key.ao_key, &key.traffic_key,
1112 					 &allocated_traffic_key,
1113 					 &key.rcv_next, &key.sne))
1114 			goto out;
1115 		key.type = TCP_KEY_AO;
1116 	}
1117 #endif
1118 
1119 	if (sk) {
1120 		oif = sk->sk_bound_dev_if;
1121 		if (sk_fullsock(sk)) {
1122 			if (inet6_test_bit(REPFLOW, sk))
1123 				label = ip6_flowlabel(ipv6h);
1124 			priority = READ_ONCE(sk->sk_priority);
1125 			txhash = sk->sk_txhash;
1126 		}
1127 		if (sk->sk_state == TCP_TIME_WAIT) {
1128 			label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1129 			priority = inet_twsk(sk)->tw_priority;
1130 			txhash = inet_twsk(sk)->tw_txhash;
1131 		}
1132 	} else {
1133 		if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1134 			label = ip6_flowlabel(ipv6h);
1135 	}
1136 
1137 	trace_tcp_send_reset(sk, skb, reason);
1138 
1139 	tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, 1,
1140 			     ipv6_get_dsfield(ipv6h) & ~INET_ECN_MASK,
1141 			     label, priority, txhash,
1142 			     &key);
1143 
1144 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1145 out:
1146 	if (allocated_traffic_key)
1147 		kfree(key.traffic_key);
1148 	rcu_read_unlock();
1149 #endif
1150 }
1151 
tcp_v6_send_ack(const struct sock * sk,struct sk_buff * skb,u32 seq,u32 ack,u32 win,u32 tsval,u32 tsecr,int oif,struct tcp_key * key,u8 tclass,__be32 label,u32 priority,u32 txhash)1152 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1153 			    u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1154 			    struct tcp_key *key, u8 tclass,
1155 			    __be32 label, u32 priority, u32 txhash)
1156 {
1157 	tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, 0,
1158 			     tclass, label, priority, txhash, key);
1159 }
1160 
tcp_v6_timewait_ack(struct sock * sk,struct sk_buff * skb,enum tcp_tw_status tw_status)1161 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb,
1162 				enum tcp_tw_status tw_status)
1163 {
1164 	struct inet_timewait_sock *tw = inet_twsk(sk);
1165 	struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1166 	u8 tclass = tw->tw_tclass;
1167 	struct tcp_key key = {};
1168 
1169 	if (tw_status == TCP_TW_ACK_OOW)
1170 		tclass &= ~INET_ECN_MASK;
1171 #ifdef CONFIG_TCP_AO
1172 	struct tcp_ao_info *ao_info;
1173 
1174 	if (static_branch_unlikely(&tcp_ao_needed.key)) {
1175 
1176 		/* FIXME: the segment to-be-acked is not verified yet */
1177 		ao_info = rcu_dereference(tcptw->ao_info);
1178 		if (ao_info) {
1179 			const struct tcp_ao_hdr *aoh;
1180 
1181 			/* Invalid TCP option size or twice included auth */
1182 			if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1183 				goto out;
1184 			if (aoh)
1185 				key.ao_key = tcp_ao_established_key(sk, ao_info,
1186 								    aoh->rnext_keyid, -1);
1187 		}
1188 	}
1189 	if (key.ao_key) {
1190 		struct tcp_ao_key *rnext_key;
1191 
1192 		key.traffic_key = snd_other_key(key.ao_key);
1193 		/* rcv_next switches to our rcv_next */
1194 		rnext_key = READ_ONCE(ao_info->rnext_key);
1195 		key.rcv_next = rnext_key->rcvid;
1196 		key.sne = READ_ONCE(ao_info->snd_sne);
1197 		key.type = TCP_KEY_AO;
1198 #else
1199 	if (0) {
1200 #endif
1201 #ifdef CONFIG_TCP_MD5SIG
1202 	} else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1203 		key.md5_key = tcp_twsk_md5_key(tcptw);
1204 		if (key.md5_key)
1205 			key.type = TCP_KEY_MD5;
1206 #endif
1207 	}
1208 
1209 	tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt,
1210 			READ_ONCE(tcptw->tw_rcv_nxt),
1211 			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1212 			tcp_tw_tsval(tcptw),
1213 			READ_ONCE(tcptw->tw_ts_recent), tw->tw_bound_dev_if,
1214 			&key, tclass, cpu_to_be32(tw->tw_flowlabel),
1215 			tw->tw_priority, tw->tw_txhash);
1216 
1217 #ifdef CONFIG_TCP_AO
1218 out:
1219 #endif
1220 	inet_twsk_put(tw);
1221 }
1222 
1223 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1224 				  struct request_sock *req)
1225 {
1226 	struct tcp_key key = {};
1227 
1228 #ifdef CONFIG_TCP_AO
1229 	if (static_branch_unlikely(&tcp_ao_needed.key) &&
1230 	    tcp_rsk_used_ao(req)) {
1231 		const struct in6_addr *addr = &ipv6_hdr(skb)->saddr;
1232 		const struct tcp_ao_hdr *aoh;
1233 		int l3index;
1234 
1235 		l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1236 		/* Invalid TCP option size or twice included auth */
1237 		if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1238 			return;
1239 		if (!aoh)
1240 			return;
1241 		key.ao_key = tcp_ao_do_lookup(sk, l3index,
1242 					      (union tcp_ao_addr *)addr,
1243 					      AF_INET6, aoh->rnext_keyid, -1);
1244 		if (unlikely(!key.ao_key)) {
1245 			/* Send ACK with any matching MKT for the peer */
1246 			key.ao_key = tcp_ao_do_lookup(sk, l3index,
1247 						      (union tcp_ao_addr *)addr,
1248 						      AF_INET6, -1, -1);
1249 			/* Matching key disappeared (user removed the key?)
1250 			 * let the handshake timeout.
1251 			 */
1252 			if (!key.ao_key) {
1253 				net_info_ratelimited("TCP-AO key for (%pI6, %d)->(%pI6, %d) suddenly disappeared, won't ACK new connection\n",
1254 						     addr,
1255 						     ntohs(tcp_hdr(skb)->source),
1256 						     &ipv6_hdr(skb)->daddr,
1257 						     ntohs(tcp_hdr(skb)->dest));
1258 				return;
1259 			}
1260 		}
1261 		key.traffic_key = kmalloc(tcp_ao_digest_size(key.ao_key), GFP_ATOMIC);
1262 		if (!key.traffic_key)
1263 			return;
1264 
1265 		key.type = TCP_KEY_AO;
1266 		key.rcv_next = aoh->keyid;
1267 		tcp_v6_ao_calc_key_rsk(key.ao_key, key.traffic_key, req);
1268 #else
1269 	if (0) {
1270 #endif
1271 #ifdef CONFIG_TCP_MD5SIG
1272 	} else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1273 		int l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1274 
1275 		key.md5_key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr,
1276 						   l3index);
1277 		if (key.md5_key)
1278 			key.type = TCP_KEY_MD5;
1279 #endif
1280 	}
1281 
1282 	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1283 	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1284 	 */
1285 	tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1286 			tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1287 			tcp_rsk(req)->rcv_nxt,
1288 			tcp_synack_window(req) >> inet_rsk(req)->rcv_wscale,
1289 			tcp_rsk_tsval(tcp_rsk(req)),
1290 			req->ts_recent, sk->sk_bound_dev_if,
1291 			&key, ipv6_get_dsfield(ipv6_hdr(skb)) & ~INET_ECN_MASK,
1292 			0,
1293 			READ_ONCE(sk->sk_priority),
1294 			READ_ONCE(tcp_rsk(req)->txhash));
1295 	if (tcp_key_is_ao(&key))
1296 		kfree(key.traffic_key);
1297 }
1298 
1299 
1300 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1301 {
1302 #ifdef CONFIG_SYN_COOKIES
1303 	const struct tcphdr *th = tcp_hdr(skb);
1304 
1305 	if (!th->syn)
1306 		sk = cookie_v6_check(sk, skb);
1307 #endif
1308 	return sk;
1309 }
1310 
1311 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1312 			 struct tcphdr *th, u32 *cookie)
1313 {
1314 	u16 mss = 0;
1315 #ifdef CONFIG_SYN_COOKIES
1316 	mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1317 				    &tcp_request_sock_ipv6_ops, sk, th);
1318 	if (mss) {
1319 		*cookie = __cookie_v6_init_sequence(iph, th, &mss);
1320 		tcp_synq_overflow(sk);
1321 	}
1322 #endif
1323 	return mss;
1324 }
1325 
1326 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1327 {
1328 	if (skb->protocol == htons(ETH_P_IP))
1329 		return tcp_v4_conn_request(sk, skb);
1330 
1331 	if (!ipv6_unicast_destination(skb))
1332 		goto drop;
1333 
1334 	if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1335 		__IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1336 		return 0;
1337 	}
1338 
1339 	return tcp_conn_request(&tcp6_request_sock_ops,
1340 				&tcp_request_sock_ipv6_ops, sk, skb);
1341 
1342 drop:
1343 	tcp_listendrop(sk);
1344 	return 0; /* don't send reset */
1345 }
1346 
1347 static void tcp_v6_restore_cb(struct sk_buff *skb)
1348 {
1349 	/* We need to move header back to the beginning if xfrm6_policy_check()
1350 	 * and tcp_v6_fill_cb() are going to be called again.
1351 	 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1352 	 */
1353 	memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1354 		sizeof(struct inet6_skb_parm));
1355 }
1356 
1357 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1358 					 struct request_sock *req,
1359 					 struct dst_entry *dst,
1360 					 struct request_sock *req_unhash,
1361 					 bool *own_req)
1362 {
1363 	struct inet_request_sock *ireq;
1364 	struct ipv6_pinfo *newnp;
1365 	const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1366 	struct ipv6_txoptions *opt;
1367 	struct inet_sock *newinet;
1368 	bool found_dup_sk = false;
1369 	struct tcp_sock *newtp;
1370 	struct sock *newsk;
1371 #ifdef CONFIG_TCP_MD5SIG
1372 	struct tcp_md5sig_key *key;
1373 	int l3index;
1374 #endif
1375 	struct flowi6 fl6;
1376 
1377 	if (skb->protocol == htons(ETH_P_IP)) {
1378 		/*
1379 		 *	v6 mapped
1380 		 */
1381 
1382 		newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1383 					     req_unhash, own_req);
1384 
1385 		if (!newsk)
1386 			return NULL;
1387 
1388 		inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1389 
1390 		newnp = tcp_inet6_sk(newsk);
1391 		newtp = tcp_sk(newsk);
1392 
1393 		memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1394 
1395 		newnp->saddr = newsk->sk_v6_rcv_saddr;
1396 
1397 		inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1398 		if (sk_is_mptcp(newsk))
1399 			mptcpv6_handle_mapped(newsk, true);
1400 		newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1401 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1402 		newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1403 #endif
1404 
1405 		newnp->ipv6_mc_list = NULL;
1406 		newnp->ipv6_ac_list = NULL;
1407 		newnp->ipv6_fl_list = NULL;
1408 		newnp->pktoptions  = NULL;
1409 		newnp->opt	   = NULL;
1410 		newnp->mcast_oif   = inet_iif(skb);
1411 		newnp->mcast_hops  = ip_hdr(skb)->ttl;
1412 		newnp->rcv_flowinfo = 0;
1413 		if (inet6_test_bit(REPFLOW, sk))
1414 			newnp->flow_label = 0;
1415 
1416 		/*
1417 		 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1418 		 * here, tcp_create_openreq_child now does this for us, see the comment in
1419 		 * that function for the gory details. -acme
1420 		 */
1421 
1422 		/* It is tricky place. Until this moment IPv4 tcp
1423 		   worked with IPv6 icsk.icsk_af_ops.
1424 		   Sync it now.
1425 		 */
1426 		tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1427 
1428 		return newsk;
1429 	}
1430 
1431 	ireq = inet_rsk(req);
1432 
1433 	if (sk_acceptq_is_full(sk))
1434 		goto out_overflow;
1435 
1436 	if (!dst) {
1437 		dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1438 		if (!dst)
1439 			goto out;
1440 	}
1441 
1442 	newsk = tcp_create_openreq_child(sk, req, skb);
1443 	if (!newsk)
1444 		goto out_nonewsk;
1445 
1446 	/*
1447 	 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1448 	 * count here, tcp_create_openreq_child now does this for us, see the
1449 	 * comment in that function for the gory details. -acme
1450 	 */
1451 
1452 	newsk->sk_gso_type = SKB_GSO_TCPV6;
1453 	inet6_sk_rx_dst_set(newsk, skb);
1454 
1455 	inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1456 
1457 	newtp = tcp_sk(newsk);
1458 	newinet = inet_sk(newsk);
1459 	newnp = tcp_inet6_sk(newsk);
1460 
1461 	memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1462 
1463 	ip6_dst_store(newsk, dst, NULL, NULL);
1464 
1465 	newnp->saddr = ireq->ir_v6_loc_addr;
1466 
1467 	/* Now IPv6 options...
1468 
1469 	   First: no IPv4 options.
1470 	 */
1471 	newinet->inet_opt = NULL;
1472 	newnp->ipv6_mc_list = NULL;
1473 	newnp->ipv6_ac_list = NULL;
1474 	newnp->ipv6_fl_list = NULL;
1475 
1476 	/* Clone RX bits */
1477 	newnp->rxopt.all = np->rxopt.all;
1478 
1479 	newnp->pktoptions = NULL;
1480 	newnp->opt	  = NULL;
1481 	newnp->mcast_oif  = tcp_v6_iif(skb);
1482 	newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1483 	newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1484 	if (inet6_test_bit(REPFLOW, sk))
1485 		newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1486 
1487 	/* Set ToS of the new socket based upon the value of incoming SYN.
1488 	 * ECT bits are set later in tcp_init_transfer().
1489 	 */
1490 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1491 		newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1492 
1493 	/* Clone native IPv6 options from listening socket (if any)
1494 
1495 	   Yes, keeping reference count would be much more clever,
1496 	   but we make one more one thing there: reattach optmem
1497 	   to newsk.
1498 	 */
1499 	opt = ireq->ipv6_opt;
1500 	if (!opt)
1501 		opt = rcu_dereference(np->opt);
1502 	if (opt) {
1503 		opt = ipv6_dup_options(newsk, opt);
1504 		RCU_INIT_POINTER(newnp->opt, opt);
1505 	}
1506 	inet_csk(newsk)->icsk_ext_hdr_len = 0;
1507 	if (opt)
1508 		inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1509 						    opt->opt_flen;
1510 
1511 	tcp_ca_openreq_child(newsk, dst);
1512 
1513 	tcp_sync_mss(newsk, dst_mtu(dst));
1514 	newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1515 
1516 	tcp_initialize_rcv_mss(newsk);
1517 
1518 #ifdef CONFIG_TCP_MD5SIG
1519 	l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1520 
1521 	if (!tcp_rsk_used_ao(req)) {
1522 		/* Copy over the MD5 key from the original socket */
1523 		key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1524 		if (key) {
1525 			const union tcp_md5_addr *addr;
1526 
1527 			addr = (union tcp_md5_addr *)&newsk->sk_v6_daddr;
1528 			if (tcp_md5_key_copy(newsk, addr, AF_INET6, 128, l3index, key)) {
1529 				inet_csk_prepare_forced_close(newsk);
1530 				tcp_done(newsk);
1531 				goto out;
1532 			}
1533 		}
1534 	}
1535 #endif
1536 #ifdef CONFIG_TCP_AO
1537 	/* Copy over tcp_ao_info if any */
1538 	if (tcp_ao_copy_all_matching(sk, newsk, req, skb, AF_INET6))
1539 		goto out; /* OOM */
1540 #endif
1541 
1542 	if (__inet_inherit_port(sk, newsk) < 0) {
1543 		inet_csk_prepare_forced_close(newsk);
1544 		tcp_done(newsk);
1545 		goto out;
1546 	}
1547 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1548 				       &found_dup_sk);
1549 	if (*own_req) {
1550 		tcp_move_syn(newtp, req);
1551 
1552 		/* Clone pktoptions received with SYN, if we own the req */
1553 		if (ireq->pktopts) {
1554 			newnp->pktoptions = skb_clone_and_charge_r(ireq->pktopts, newsk);
1555 			consume_skb(ireq->pktopts);
1556 			ireq->pktopts = NULL;
1557 			if (newnp->pktoptions)
1558 				tcp_v6_restore_cb(newnp->pktoptions);
1559 		}
1560 	} else {
1561 		if (!req_unhash && found_dup_sk) {
1562 			/* This code path should only be executed in the
1563 			 * syncookie case only
1564 			 */
1565 			bh_unlock_sock(newsk);
1566 			sock_put(newsk);
1567 			newsk = NULL;
1568 		}
1569 	}
1570 
1571 	return newsk;
1572 
1573 out_overflow:
1574 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1575 out_nonewsk:
1576 	dst_release(dst);
1577 out:
1578 	tcp_listendrop(sk);
1579 	return NULL;
1580 }
1581 
1582 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1583 							   u32));
1584 /* The socket must have it's spinlock held when we get
1585  * here, unless it is a TCP_LISTEN socket.
1586  *
1587  * We have a potential double-lock case here, so even when
1588  * doing backlog processing we use the BH locking scheme.
1589  * This is because we cannot sleep with the original spinlock
1590  * held.
1591  */
1592 INDIRECT_CALLABLE_SCOPE
1593 int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1594 {
1595 	struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1596 	struct sk_buff *opt_skb = NULL;
1597 	enum skb_drop_reason reason;
1598 	struct tcp_sock *tp;
1599 
1600 	/* Imagine: socket is IPv6. IPv4 packet arrives,
1601 	   goes to IPv4 receive handler and backlogged.
1602 	   From backlog it always goes here. Kerboom...
1603 	   Fortunately, tcp_rcv_established and rcv_established
1604 	   handle them correctly, but it is not case with
1605 	   tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1606 	 */
1607 
1608 	if (skb->protocol == htons(ETH_P_IP))
1609 		return tcp_v4_do_rcv(sk, skb);
1610 
1611 	/*
1612 	 *	socket locking is here for SMP purposes as backlog rcv
1613 	 *	is currently called with bh processing disabled.
1614 	 */
1615 
1616 	/* Do Stevens' IPV6_PKTOPTIONS.
1617 
1618 	   Yes, guys, it is the only place in our code, where we
1619 	   may make it not affecting IPv4.
1620 	   The rest of code is protocol independent,
1621 	   and I do not like idea to uglify IPv4.
1622 
1623 	   Actually, all the idea behind IPV6_PKTOPTIONS
1624 	   looks not very well thought. For now we latch
1625 	   options, received in the last packet, enqueued
1626 	   by tcp. Feel free to propose better solution.
1627 					       --ANK (980728)
1628 	 */
1629 	if (np->rxopt.all && sk->sk_state != TCP_LISTEN)
1630 		opt_skb = skb_clone_and_charge_r(skb, sk);
1631 
1632 	if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1633 		struct dst_entry *dst;
1634 
1635 		dst = rcu_dereference_protected(sk->sk_rx_dst,
1636 						lockdep_sock_is_held(sk));
1637 
1638 		sock_rps_save_rxhash(sk, skb);
1639 		sk_mark_napi_id(sk, skb);
1640 		if (dst) {
1641 			if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1642 			    INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1643 					    dst, sk->sk_rx_dst_cookie) == NULL) {
1644 				RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1645 				dst_release(dst);
1646 			}
1647 		}
1648 
1649 		tcp_rcv_established(sk, skb);
1650 		if (opt_skb)
1651 			goto ipv6_pktoptions;
1652 		return 0;
1653 	}
1654 
1655 	if (tcp_checksum_complete(skb))
1656 		goto csum_err;
1657 
1658 	if (sk->sk_state == TCP_LISTEN) {
1659 		struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1660 
1661 		if (nsk != sk) {
1662 			if (nsk) {
1663 				reason = tcp_child_process(sk, nsk, skb);
1664 				if (reason)
1665 					goto reset;
1666 			}
1667 			return 0;
1668 		}
1669 	} else
1670 		sock_rps_save_rxhash(sk, skb);
1671 
1672 	reason = tcp_rcv_state_process(sk, skb);
1673 	if (reason)
1674 		goto reset;
1675 	if (opt_skb)
1676 		goto ipv6_pktoptions;
1677 	return 0;
1678 
1679 reset:
1680 	tcp_v6_send_reset(sk, skb, sk_rst_convert_drop_reason(reason));
1681 discard:
1682 	if (opt_skb)
1683 		__kfree_skb(opt_skb);
1684 	sk_skb_reason_drop(sk, skb, reason);
1685 	return 0;
1686 csum_err:
1687 	reason = SKB_DROP_REASON_TCP_CSUM;
1688 	trace_tcp_bad_csum(skb);
1689 	TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1690 	TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1691 	goto discard;
1692 
1693 
1694 ipv6_pktoptions:
1695 	/* Do you ask, what is it?
1696 
1697 	   1. skb was enqueued by tcp.
1698 	   2. skb is added to tail of read queue, rather than out of order.
1699 	   3. socket is not in passive state.
1700 	   4. Finally, it really contains options, which user wants to receive.
1701 	 */
1702 	tp = tcp_sk(sk);
1703 	if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1704 	    !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1705 		if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1706 			WRITE_ONCE(np->mcast_oif, tcp_v6_iif(opt_skb));
1707 		if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1708 			WRITE_ONCE(np->mcast_hops,
1709 				   ipv6_hdr(opt_skb)->hop_limit);
1710 		if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1711 			np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1712 		if (inet6_test_bit(REPFLOW, sk))
1713 			np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1714 		if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1715 			tcp_v6_restore_cb(opt_skb);
1716 			opt_skb = xchg(&np->pktoptions, opt_skb);
1717 		} else {
1718 			__kfree_skb(opt_skb);
1719 			opt_skb = xchg(&np->pktoptions, NULL);
1720 		}
1721 	}
1722 
1723 	consume_skb(opt_skb);
1724 	return 0;
1725 }
1726 
1727 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1728 			   const struct tcphdr *th)
1729 {
1730 	/* This is tricky: we move IP6CB at its correct location into
1731 	 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1732 	 * _decode_session6() uses IP6CB().
1733 	 * barrier() makes sure compiler won't play aliasing games.
1734 	 */
1735 	memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1736 		sizeof(struct inet6_skb_parm));
1737 	barrier();
1738 
1739 	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1740 	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1741 				    skb->len - th->doff*4);
1742 	TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1743 	TCP_SKB_CB(skb)->tcp_flags = tcp_flags_ntohs(th);
1744 	TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1745 	TCP_SKB_CB(skb)->sacked = 0;
1746 	TCP_SKB_CB(skb)->has_rxtstamp =
1747 			skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1748 }
1749 
1750 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1751 {
1752 	struct net *net = dev_net_rcu(skb->dev);
1753 	enum skb_drop_reason drop_reason;
1754 	enum tcp_tw_status tw_status;
1755 	int sdif = inet6_sdif(skb);
1756 	int dif = inet6_iif(skb);
1757 	const struct tcphdr *th;
1758 	const struct ipv6hdr *hdr;
1759 	struct sock *sk = NULL;
1760 	bool refcounted;
1761 	int ret;
1762 	u32 isn;
1763 
1764 	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1765 	if (skb->pkt_type != PACKET_HOST)
1766 		goto discard_it;
1767 
1768 	/*
1769 	 *	Count it even if it's bad.
1770 	 */
1771 	__TCP_INC_STATS(net, TCP_MIB_INSEGS);
1772 
1773 	if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1774 		goto discard_it;
1775 
1776 	th = (const struct tcphdr *)skb->data;
1777 
1778 	if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
1779 		drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1780 		goto bad_packet;
1781 	}
1782 	if (!pskb_may_pull(skb, th->doff*4))
1783 		goto discard_it;
1784 
1785 	if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1786 		goto csum_error;
1787 
1788 	th = (const struct tcphdr *)skb->data;
1789 	hdr = ipv6_hdr(skb);
1790 
1791 lookup:
1792 	sk = __inet6_lookup_skb(net->ipv4.tcp_death_row.hashinfo, skb, __tcp_hdrlen(th),
1793 				th->source, th->dest, inet6_iif(skb), sdif,
1794 				&refcounted);
1795 	if (!sk)
1796 		goto no_tcp_socket;
1797 
1798 	if (sk->sk_state == TCP_TIME_WAIT)
1799 		goto do_time_wait;
1800 
1801 	if (sk->sk_state == TCP_NEW_SYN_RECV) {
1802 		struct request_sock *req = inet_reqsk(sk);
1803 		bool req_stolen = false;
1804 		struct sock *nsk;
1805 
1806 		sk = req->rsk_listener;
1807 		if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1808 			drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1809 		else
1810 			drop_reason = tcp_inbound_hash(sk, req, skb,
1811 						       &hdr->saddr, &hdr->daddr,
1812 						       AF_INET6, dif, sdif);
1813 		if (drop_reason) {
1814 			sk_drops_add(sk, skb);
1815 			reqsk_put(req);
1816 			goto discard_it;
1817 		}
1818 		if (tcp_checksum_complete(skb)) {
1819 			reqsk_put(req);
1820 			goto csum_error;
1821 		}
1822 		if (unlikely(sk->sk_state != TCP_LISTEN)) {
1823 			nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
1824 			if (!nsk) {
1825 				inet_csk_reqsk_queue_drop_and_put(sk, req);
1826 				goto lookup;
1827 			}
1828 			sk = nsk;
1829 			/* reuseport_migrate_sock() has already held one sk_refcnt
1830 			 * before returning.
1831 			 */
1832 		} else {
1833 			sock_hold(sk);
1834 		}
1835 		refcounted = true;
1836 		nsk = NULL;
1837 		if (!tcp_filter(sk, skb, &drop_reason)) {
1838 			th = (const struct tcphdr *)skb->data;
1839 			hdr = ipv6_hdr(skb);
1840 			tcp_v6_fill_cb(skb, hdr, th);
1841 			nsk = tcp_check_req(sk, skb, req, false, &req_stolen,
1842 					    &drop_reason);
1843 		}
1844 		if (!nsk) {
1845 			reqsk_put(req);
1846 			if (req_stolen) {
1847 				/* Another cpu got exclusive access to req
1848 				 * and created a full blown socket.
1849 				 * Try to feed this packet to this socket
1850 				 * instead of discarding it.
1851 				 */
1852 				tcp_v6_restore_cb(skb);
1853 				sock_put(sk);
1854 				goto lookup;
1855 			}
1856 			goto discard_and_relse;
1857 		}
1858 		nf_reset_ct(skb);
1859 		if (nsk == sk) {
1860 			reqsk_put(req);
1861 			tcp_v6_restore_cb(skb);
1862 		} else {
1863 			drop_reason = tcp_child_process(sk, nsk, skb);
1864 			if (drop_reason) {
1865 				enum sk_rst_reason rst_reason;
1866 
1867 				rst_reason = sk_rst_convert_drop_reason(drop_reason);
1868 				tcp_v6_send_reset(nsk, skb, rst_reason);
1869 				goto discard_and_relse;
1870 			}
1871 			sock_put(sk);
1872 			return 0;
1873 		}
1874 	}
1875 
1876 process:
1877 	if (static_branch_unlikely(&ip6_min_hopcount)) {
1878 		/* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
1879 		if (unlikely(hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount))) {
1880 			__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1881 			drop_reason = SKB_DROP_REASON_TCP_MINTTL;
1882 			goto discard_and_relse;
1883 		}
1884 	}
1885 
1886 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
1887 		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1888 		goto discard_and_relse;
1889 	}
1890 
1891 	drop_reason = tcp_inbound_hash(sk, NULL, skb, &hdr->saddr, &hdr->daddr,
1892 				       AF_INET6, dif, sdif);
1893 	if (drop_reason)
1894 		goto discard_and_relse;
1895 
1896 	nf_reset_ct(skb);
1897 
1898 	if (tcp_filter(sk, skb, &drop_reason))
1899 		goto discard_and_relse;
1900 
1901 	th = (const struct tcphdr *)skb->data;
1902 	hdr = ipv6_hdr(skb);
1903 	tcp_v6_fill_cb(skb, hdr, th);
1904 
1905 	skb->dev = NULL;
1906 
1907 	if (sk->sk_state == TCP_LISTEN) {
1908 		ret = tcp_v6_do_rcv(sk, skb);
1909 		goto put_and_return;
1910 	}
1911 
1912 	sk_incoming_cpu_update(sk);
1913 
1914 	bh_lock_sock_nested(sk);
1915 	tcp_segs_in(tcp_sk(sk), skb);
1916 	ret = 0;
1917 	if (!sock_owned_by_user(sk)) {
1918 		ret = tcp_v6_do_rcv(sk, skb);
1919 	} else {
1920 		if (tcp_add_backlog(sk, skb, &drop_reason))
1921 			goto discard_and_relse;
1922 	}
1923 	bh_unlock_sock(sk);
1924 put_and_return:
1925 	if (refcounted)
1926 		sock_put(sk);
1927 	return ret ? -1 : 0;
1928 
1929 no_tcp_socket:
1930 	drop_reason = SKB_DROP_REASON_NO_SOCKET;
1931 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1932 		goto discard_it;
1933 
1934 	tcp_v6_fill_cb(skb, hdr, th);
1935 
1936 	if (tcp_checksum_complete(skb)) {
1937 csum_error:
1938 		drop_reason = SKB_DROP_REASON_TCP_CSUM;
1939 		trace_tcp_bad_csum(skb);
1940 		__TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1941 bad_packet:
1942 		__TCP_INC_STATS(net, TCP_MIB_INERRS);
1943 	} else {
1944 		tcp_v6_send_reset(NULL, skb, sk_rst_convert_drop_reason(drop_reason));
1945 	}
1946 
1947 discard_it:
1948 	SKB_DR_OR(drop_reason, NOT_SPECIFIED);
1949 	sk_skb_reason_drop(sk, skb, drop_reason);
1950 	return 0;
1951 
1952 discard_and_relse:
1953 	sk_drops_add(sk, skb);
1954 	if (refcounted)
1955 		sock_put(sk);
1956 	goto discard_it;
1957 
1958 do_time_wait:
1959 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1960 		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1961 		inet_twsk_put(inet_twsk(sk));
1962 		goto discard_it;
1963 	}
1964 
1965 	tcp_v6_fill_cb(skb, hdr, th);
1966 
1967 	if (tcp_checksum_complete(skb)) {
1968 		inet_twsk_put(inet_twsk(sk));
1969 		goto csum_error;
1970 	}
1971 
1972 	tw_status = tcp_timewait_state_process(inet_twsk(sk), skb, th, &isn,
1973 					       &drop_reason);
1974 	switch (tw_status) {
1975 	case TCP_TW_SYN:
1976 	{
1977 		struct sock *sk2;
1978 
1979 		sk2 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1980 					    skb, __tcp_hdrlen(th),
1981 					    &ipv6_hdr(skb)->saddr, th->source,
1982 					    &ipv6_hdr(skb)->daddr,
1983 					    ntohs(th->dest),
1984 					    tcp_v6_iif_l3_slave(skb),
1985 					    sdif);
1986 		if (sk2) {
1987 			struct inet_timewait_sock *tw = inet_twsk(sk);
1988 			inet_twsk_deschedule_put(tw);
1989 			sk = sk2;
1990 			tcp_v6_restore_cb(skb);
1991 			refcounted = false;
1992 			__this_cpu_write(tcp_tw_isn, isn);
1993 			goto process;
1994 		}
1995 	}
1996 		/* to ACK */
1997 		fallthrough;
1998 	case TCP_TW_ACK:
1999 	case TCP_TW_ACK_OOW:
2000 		tcp_v6_timewait_ack(sk, skb, tw_status);
2001 		break;
2002 	case TCP_TW_RST:
2003 		tcp_v6_send_reset(sk, skb, SK_RST_REASON_TCP_TIMEWAIT_SOCKET);
2004 		inet_twsk_deschedule_put(inet_twsk(sk));
2005 		goto discard_it;
2006 	case TCP_TW_SUCCESS:
2007 		;
2008 	}
2009 	goto discard_it;
2010 }
2011 
2012 void tcp_v6_early_demux(struct sk_buff *skb)
2013 {
2014 	struct net *net = dev_net_rcu(skb->dev);
2015 	const struct ipv6hdr *hdr;
2016 	const struct tcphdr *th;
2017 	struct sock *sk;
2018 
2019 	if (skb->pkt_type != PACKET_HOST)
2020 		return;
2021 
2022 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
2023 		return;
2024 
2025 	hdr = ipv6_hdr(skb);
2026 	th = tcp_hdr(skb);
2027 
2028 	if (th->doff < sizeof(struct tcphdr) / 4)
2029 		return;
2030 
2031 	/* Note : We use inet6_iif() here, not tcp_v6_iif() */
2032 	sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
2033 					&hdr->saddr, th->source,
2034 					&hdr->daddr, ntohs(th->dest),
2035 					inet6_iif(skb), inet6_sdif(skb));
2036 	if (sk) {
2037 		skb->sk = sk;
2038 		skb->destructor = sock_edemux;
2039 		if (sk_fullsock(sk)) {
2040 			struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
2041 
2042 			if (dst)
2043 				dst = dst_check(dst, sk->sk_rx_dst_cookie);
2044 			if (dst &&
2045 			    sk->sk_rx_dst_ifindex == skb->skb_iif)
2046 				skb_dst_set_noref(skb, dst);
2047 		}
2048 	}
2049 }
2050 
2051 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
2052 	.twsk_obj_size	= sizeof(struct tcp6_timewait_sock),
2053 	.twsk_destructor = tcp_twsk_destructor,
2054 };
2055 
2056 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
2057 {
2058 	__tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr);
2059 }
2060 
2061 const struct inet_connection_sock_af_ops ipv6_specific = {
2062 	.queue_xmit	   = inet6_csk_xmit,
2063 	.send_check	   = tcp_v6_send_check,
2064 	.rebuild_header	   = inet6_sk_rebuild_header,
2065 	.sk_rx_dst_set	   = inet6_sk_rx_dst_set,
2066 	.conn_request	   = tcp_v6_conn_request,
2067 	.syn_recv_sock	   = tcp_v6_syn_recv_sock,
2068 	.net_header_len	   = sizeof(struct ipv6hdr),
2069 	.setsockopt	   = ipv6_setsockopt,
2070 	.getsockopt	   = ipv6_getsockopt,
2071 	.mtu_reduced	   = tcp_v6_mtu_reduced,
2072 };
2073 
2074 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2075 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
2076 #ifdef CONFIG_TCP_MD5SIG
2077 	.md5_lookup	=	tcp_v6_md5_lookup,
2078 	.calc_md5_hash	=	tcp_v6_md5_hash_skb,
2079 	.md5_parse	=	tcp_v6_parse_md5_keys,
2080 #endif
2081 #ifdef CONFIG_TCP_AO
2082 	.ao_lookup	=	tcp_v6_ao_lookup,
2083 	.calc_ao_hash	=	tcp_v6_ao_hash_skb,
2084 	.ao_parse	=	tcp_v6_parse_ao,
2085 	.ao_calc_key_sk	=	tcp_v6_ao_calc_key_sk,
2086 #endif
2087 };
2088 #endif
2089 
2090 /*
2091  *	TCP over IPv4 via INET6 API
2092  */
2093 static const struct inet_connection_sock_af_ops ipv6_mapped = {
2094 	.queue_xmit	   = ip_queue_xmit,
2095 	.send_check	   = tcp_v4_send_check,
2096 	.rebuild_header	   = inet_sk_rebuild_header,
2097 	.sk_rx_dst_set	   = inet_sk_rx_dst_set,
2098 	.conn_request	   = tcp_v6_conn_request,
2099 	.syn_recv_sock	   = tcp_v6_syn_recv_sock,
2100 	.net_header_len	   = sizeof(struct iphdr),
2101 	.setsockopt	   = ipv6_setsockopt,
2102 	.getsockopt	   = ipv6_getsockopt,
2103 	.mtu_reduced	   = tcp_v4_mtu_reduced,
2104 };
2105 
2106 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2107 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
2108 #ifdef CONFIG_TCP_MD5SIG
2109 	.md5_lookup	=	tcp_v4_md5_lookup,
2110 	.calc_md5_hash	=	tcp_v4_md5_hash_skb,
2111 	.md5_parse	=	tcp_v6_parse_md5_keys,
2112 #endif
2113 #ifdef CONFIG_TCP_AO
2114 	.ao_lookup	=	tcp_v6_ao_lookup,
2115 	.calc_ao_hash	=	tcp_v4_ao_hash_skb,
2116 	.ao_parse	=	tcp_v6_parse_ao,
2117 	.ao_calc_key_sk	=	tcp_v4_ao_calc_key_sk,
2118 #endif
2119 };
2120 #endif
2121 
2122 /* NOTE: A lot of things set to zero explicitly by call to
2123  *       sk_alloc() so need not be done here.
2124  */
2125 static int tcp_v6_init_sock(struct sock *sk)
2126 {
2127 	struct inet_connection_sock *icsk = inet_csk(sk);
2128 
2129 	tcp_init_sock(sk);
2130 
2131 	icsk->icsk_af_ops = &ipv6_specific;
2132 
2133 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2134 	tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
2135 #endif
2136 
2137 	return 0;
2138 }
2139 
2140 #ifdef CONFIG_PROC_FS
2141 /* Proc filesystem TCPv6 sock list dumping. */
2142 static void get_openreq6(struct seq_file *seq,
2143 			 const struct request_sock *req, int i)
2144 {
2145 	long ttd = req->rsk_timer.expires - jiffies;
2146 	const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
2147 	const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
2148 
2149 	if (ttd < 0)
2150 		ttd = 0;
2151 
2152 	seq_printf(seq,
2153 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2154 		   "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
2155 		   i,
2156 		   src->s6_addr32[0], src->s6_addr32[1],
2157 		   src->s6_addr32[2], src->s6_addr32[3],
2158 		   inet_rsk(req)->ir_num,
2159 		   dest->s6_addr32[0], dest->s6_addr32[1],
2160 		   dest->s6_addr32[2], dest->s6_addr32[3],
2161 		   ntohs(inet_rsk(req)->ir_rmt_port),
2162 		   TCP_SYN_RECV,
2163 		   0, 0, /* could print option size, but that is af dependent. */
2164 		   1,   /* timers active (only the expire timer) */
2165 		   jiffies_to_clock_t(ttd),
2166 		   req->num_timeout,
2167 		   from_kuid_munged(seq_user_ns(seq),
2168 				    sk_uid(req->rsk_listener)),
2169 		   0,  /* non standard timer */
2170 		   0, /* open_requests have no inode */
2171 		   0, req);
2172 }
2173 
2174 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2175 {
2176 	const struct in6_addr *dest, *src;
2177 	__u16 destp, srcp;
2178 	int timer_active;
2179 	unsigned long timer_expires;
2180 	const struct inet_sock *inet = inet_sk(sp);
2181 	const struct tcp_sock *tp = tcp_sk(sp);
2182 	const struct inet_connection_sock *icsk = inet_csk(sp);
2183 	const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2184 	u8 icsk_pending;
2185 	int rx_queue;
2186 	int state;
2187 
2188 	dest  = &sp->sk_v6_daddr;
2189 	src   = &sp->sk_v6_rcv_saddr;
2190 	destp = ntohs(inet->inet_dport);
2191 	srcp  = ntohs(inet->inet_sport);
2192 
2193 	icsk_pending = smp_load_acquire(&icsk->icsk_pending);
2194 	if (icsk_pending == ICSK_TIME_RETRANS ||
2195 	    icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2196 	    icsk_pending == ICSK_TIME_LOSS_PROBE) {
2197 		timer_active	= 1;
2198 		timer_expires	= icsk_timeout(icsk);
2199 	} else if (icsk_pending == ICSK_TIME_PROBE0) {
2200 		timer_active	= 4;
2201 		timer_expires	= icsk_timeout(icsk);
2202 	} else if (timer_pending(&sp->sk_timer)) {
2203 		timer_active	= 2;
2204 		timer_expires	= sp->sk_timer.expires;
2205 	} else {
2206 		timer_active	= 0;
2207 		timer_expires = jiffies;
2208 	}
2209 
2210 	state = inet_sk_state_load(sp);
2211 	if (state == TCP_LISTEN)
2212 		rx_queue = READ_ONCE(sp->sk_ack_backlog);
2213 	else
2214 		/* Because we don't lock the socket,
2215 		 * we might find a transient negative value.
2216 		 */
2217 		rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2218 				      READ_ONCE(tp->copied_seq), 0);
2219 
2220 	seq_printf(seq,
2221 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2222 		   "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2223 		   i,
2224 		   src->s6_addr32[0], src->s6_addr32[1],
2225 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
2226 		   dest->s6_addr32[0], dest->s6_addr32[1],
2227 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
2228 		   state,
2229 		   READ_ONCE(tp->write_seq) - tp->snd_una,
2230 		   rx_queue,
2231 		   timer_active,
2232 		   jiffies_delta_to_clock_t(timer_expires - jiffies),
2233 		   icsk->icsk_retransmits,
2234 		   from_kuid_munged(seq_user_ns(seq), sk_uid(sp)),
2235 		   icsk->icsk_probes_out,
2236 		   sock_i_ino(sp),
2237 		   refcount_read(&sp->sk_refcnt), sp,
2238 		   jiffies_to_clock_t(icsk->icsk_rto),
2239 		   jiffies_to_clock_t(icsk->icsk_ack.ato),
2240 		   (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2241 		   tcp_snd_cwnd(tp),
2242 		   state == TCP_LISTEN ?
2243 			fastopenq->max_qlen :
2244 			(tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2245 		   );
2246 }
2247 
2248 static void get_timewait6_sock(struct seq_file *seq,
2249 			       struct inet_timewait_sock *tw, int i)
2250 {
2251 	long delta = tw->tw_timer.expires - jiffies;
2252 	const struct in6_addr *dest, *src;
2253 	__u16 destp, srcp;
2254 
2255 	dest = &tw->tw_v6_daddr;
2256 	src  = &tw->tw_v6_rcv_saddr;
2257 	destp = ntohs(tw->tw_dport);
2258 	srcp  = ntohs(tw->tw_sport);
2259 
2260 	seq_printf(seq,
2261 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2262 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2263 		   i,
2264 		   src->s6_addr32[0], src->s6_addr32[1],
2265 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
2266 		   dest->s6_addr32[0], dest->s6_addr32[1],
2267 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
2268 		   READ_ONCE(tw->tw_substate), 0, 0,
2269 		   3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2270 		   refcount_read(&tw->tw_refcnt), tw);
2271 }
2272 
2273 static int tcp6_seq_show(struct seq_file *seq, void *v)
2274 {
2275 	struct tcp_iter_state *st;
2276 	struct sock *sk = v;
2277 
2278 	if (v == SEQ_START_TOKEN) {
2279 		seq_puts(seq,
2280 			 "  sl  "
2281 			 "local_address                         "
2282 			 "remote_address                        "
2283 			 "st tx_queue rx_queue tr tm->when retrnsmt"
2284 			 "   uid  timeout inode\n");
2285 		goto out;
2286 	}
2287 	st = seq->private;
2288 
2289 	if (sk->sk_state == TCP_TIME_WAIT)
2290 		get_timewait6_sock(seq, v, st->num);
2291 	else if (sk->sk_state == TCP_NEW_SYN_RECV)
2292 		get_openreq6(seq, v, st->num);
2293 	else
2294 		get_tcp6_sock(seq, v, st->num);
2295 out:
2296 	return 0;
2297 }
2298 
2299 static const struct seq_operations tcp6_seq_ops = {
2300 	.show		= tcp6_seq_show,
2301 	.start		= tcp_seq_start,
2302 	.next		= tcp_seq_next,
2303 	.stop		= tcp_seq_stop,
2304 };
2305 
2306 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2307 	.family		= AF_INET6,
2308 };
2309 
2310 int __net_init tcp6_proc_init(struct net *net)
2311 {
2312 	if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2313 			sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2314 		return -ENOMEM;
2315 	return 0;
2316 }
2317 
2318 void tcp6_proc_exit(struct net *net)
2319 {
2320 	remove_proc_entry("tcp6", net->proc_net);
2321 }
2322 #endif
2323 
2324 struct proto tcpv6_prot = {
2325 	.name			= "TCPv6",
2326 	.owner			= THIS_MODULE,
2327 	.close			= tcp_close,
2328 	.pre_connect		= tcp_v6_pre_connect,
2329 	.connect		= tcp_v6_connect,
2330 	.disconnect		= tcp_disconnect,
2331 	.accept			= inet_csk_accept,
2332 	.ioctl			= tcp_ioctl,
2333 	.init			= tcp_v6_init_sock,
2334 	.destroy		= tcp_v4_destroy_sock,
2335 	.shutdown		= tcp_shutdown,
2336 	.setsockopt		= tcp_setsockopt,
2337 	.getsockopt		= tcp_getsockopt,
2338 	.bpf_bypass_getsockopt	= tcp_bpf_bypass_getsockopt,
2339 	.keepalive		= tcp_set_keepalive,
2340 	.recvmsg		= tcp_recvmsg,
2341 	.sendmsg		= tcp_sendmsg,
2342 	.splice_eof		= tcp_splice_eof,
2343 	.backlog_rcv		= tcp_v6_do_rcv,
2344 	.release_cb		= tcp_release_cb,
2345 	.hash			= inet6_hash,
2346 	.unhash			= inet_unhash,
2347 	.get_port		= inet_csk_get_port,
2348 	.put_port		= inet_put_port,
2349 #ifdef CONFIG_BPF_SYSCALL
2350 	.psock_update_sk_prot	= tcp_bpf_update_proto,
2351 #endif
2352 	.enter_memory_pressure	= tcp_enter_memory_pressure,
2353 	.leave_memory_pressure	= tcp_leave_memory_pressure,
2354 	.stream_memory_free	= tcp_stream_memory_free,
2355 	.sockets_allocated	= &tcp_sockets_allocated,
2356 
2357 	.memory_allocated	= &net_aligned_data.tcp_memory_allocated,
2358 	.per_cpu_fw_alloc	= &tcp_memory_per_cpu_fw_alloc,
2359 
2360 	.memory_pressure	= &tcp_memory_pressure,
2361 	.orphan_count		= &tcp_orphan_count,
2362 	.sysctl_mem		= sysctl_tcp_mem,
2363 	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
2364 	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
2365 	.max_header		= MAX_TCP_HEADER,
2366 	.obj_size		= sizeof(struct tcp6_sock),
2367 	.ipv6_pinfo_offset = offsetof(struct tcp6_sock, inet6),
2368 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
2369 	.twsk_prot		= &tcp6_timewait_sock_ops,
2370 	.rsk_prot		= &tcp6_request_sock_ops,
2371 	.h.hashinfo		= NULL,
2372 	.no_autobind		= true,
2373 	.diag_destroy		= tcp_abort,
2374 };
2375 EXPORT_SYMBOL_GPL(tcpv6_prot);
2376 
2377 
2378 static struct inet_protosw tcpv6_protosw = {
2379 	.type		=	SOCK_STREAM,
2380 	.protocol	=	IPPROTO_TCP,
2381 	.prot		=	&tcpv6_prot,
2382 	.ops		=	&inet6_stream_ops,
2383 	.flags		=	INET_PROTOSW_PERMANENT |
2384 				INET_PROTOSW_ICSK,
2385 };
2386 
2387 static int __net_init tcpv6_net_init(struct net *net)
2388 {
2389 	int res;
2390 
2391 	res = inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2392 				   SOCK_RAW, IPPROTO_TCP, net);
2393 	if (!res)
2394 		net->ipv6.tcp_sk->sk_clockid = CLOCK_MONOTONIC;
2395 
2396 	return res;
2397 }
2398 
2399 static void __net_exit tcpv6_net_exit(struct net *net)
2400 {
2401 	inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2402 }
2403 
2404 static struct pernet_operations tcpv6_net_ops = {
2405 	.init	    = tcpv6_net_init,
2406 	.exit	    = tcpv6_net_exit,
2407 };
2408 
2409 int __init tcpv6_init(void)
2410 {
2411 	int ret;
2412 
2413 	net_hotdata.tcpv6_protocol = (struct inet6_protocol) {
2414 		.handler     = tcp_v6_rcv,
2415 		.err_handler = tcp_v6_err,
2416 		.flags	     = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
2417 	};
2418 	ret = inet6_add_protocol(&net_hotdata.tcpv6_protocol, IPPROTO_TCP);
2419 	if (ret)
2420 		goto out;
2421 
2422 	/* register inet6 protocol */
2423 	ret = inet6_register_protosw(&tcpv6_protosw);
2424 	if (ret)
2425 		goto out_tcpv6_protocol;
2426 
2427 	ret = register_pernet_subsys(&tcpv6_net_ops);
2428 	if (ret)
2429 		goto out_tcpv6_protosw;
2430 
2431 	ret = mptcpv6_init();
2432 	if (ret)
2433 		goto out_tcpv6_pernet_subsys;
2434 
2435 out:
2436 	return ret;
2437 
2438 out_tcpv6_pernet_subsys:
2439 	unregister_pernet_subsys(&tcpv6_net_ops);
2440 out_tcpv6_protosw:
2441 	inet6_unregister_protosw(&tcpv6_protosw);
2442 out_tcpv6_protocol:
2443 	inet6_del_protocol(&net_hotdata.tcpv6_protocol, IPPROTO_TCP);
2444 	goto out;
2445 }
2446 
2447 void tcpv6_exit(void)
2448 {
2449 	unregister_pernet_subsys(&tcpv6_net_ops);
2450 	inet6_unregister_protosw(&tcpv6_protosw);
2451 	inet6_del_protocol(&net_hotdata.tcpv6_protocol, IPPROTO_TCP);
2452 }
2453