Loading...
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
71static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
72static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
73 struct request_sock *req);
74
75static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
76
77static const struct inet_connection_sock_af_ops ipv6_mapped;
78const struct inet_connection_sock_af_ops ipv6_specific;
79#ifdef CONFIG_TCP_MD5SIG
80static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
81static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
82#else
83static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
84 const struct in6_addr *addr,
85 int l3index)
86{
87 return NULL;
88}
89#endif
90
91/* Helper returning the inet6 address from a given tcp socket.
92 * It can be used in TCP stack instead of inet6_sk(sk).
93 * This avoids a dereference and allow compiler optimizations.
94 * It is a specialized version of inet6_sk_generic().
95 */
96static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk)
97{
98 unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo);
99
100 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
101}
102
103static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
104{
105 struct dst_entry *dst = skb_dst(skb);
106
107 if (dst && dst_hold_safe(dst)) {
108 const struct rt6_info *rt = (const struct rt6_info *)dst;
109
110 sk->sk_rx_dst = dst;
111 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
112 tcp_inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
113 }
114}
115
116static u32 tcp_v6_init_seq(const struct sk_buff *skb)
117{
118 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
119 ipv6_hdr(skb)->saddr.s6_addr32,
120 tcp_hdr(skb)->dest,
121 tcp_hdr(skb)->source);
122}
123
124static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
125{
126 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
127 ipv6_hdr(skb)->saddr.s6_addr32);
128}
129
130static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
131 int addr_len)
132{
133 /* This check is replicated from tcp_v6_connect() and intended to
134 * prevent BPF program called below from accessing bytes that are out
135 * of the bound specified by user in addr_len.
136 */
137 if (addr_len < SIN6_LEN_RFC2133)
138 return -EINVAL;
139
140 sock_owned_by_me(sk);
141
142 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr);
143}
144
145static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
146 int addr_len)
147{
148 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
149 struct inet_sock *inet = inet_sk(sk);
150 struct inet_connection_sock *icsk = inet_csk(sk);
151 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
152 struct tcp_sock *tp = tcp_sk(sk);
153 struct in6_addr *saddr = NULL, *final_p, final;
154 struct ipv6_txoptions *opt;
155 struct flowi6 fl6;
156 struct dst_entry *dst;
157 int addr_type;
158 int err;
159 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
160
161 if (addr_len < SIN6_LEN_RFC2133)
162 return -EINVAL;
163
164 if (usin->sin6_family != AF_INET6)
165 return -EAFNOSUPPORT;
166
167 memset(&fl6, 0, sizeof(fl6));
168
169 if (np->sndflow) {
170 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
171 IP6_ECN_flow_init(fl6.flowlabel);
172 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
173 struct ip6_flowlabel *flowlabel;
174 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
175 if (IS_ERR(flowlabel))
176 return -EINVAL;
177 fl6_sock_release(flowlabel);
178 }
179 }
180
181 /*
182 * connect() to INADDR_ANY means loopback (BSD'ism).
183 */
184
185 if (ipv6_addr_any(&usin->sin6_addr)) {
186 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
187 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
188 &usin->sin6_addr);
189 else
190 usin->sin6_addr = in6addr_loopback;
191 }
192
193 addr_type = ipv6_addr_type(&usin->sin6_addr);
194
195 if (addr_type & IPV6_ADDR_MULTICAST)
196 return -ENETUNREACH;
197
198 if (addr_type&IPV6_ADDR_LINKLOCAL) {
199 if (addr_len >= sizeof(struct sockaddr_in6) &&
200 usin->sin6_scope_id) {
201 /* If interface is set while binding, indices
202 * must coincide.
203 */
204 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
205 return -EINVAL;
206
207 sk->sk_bound_dev_if = usin->sin6_scope_id;
208 }
209
210 /* Connect to link-local address requires an interface */
211 if (!sk->sk_bound_dev_if)
212 return -EINVAL;
213 }
214
215 if (tp->rx_opt.ts_recent_stamp &&
216 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
217 tp->rx_opt.ts_recent = 0;
218 tp->rx_opt.ts_recent_stamp = 0;
219 WRITE_ONCE(tp->write_seq, 0);
220 }
221
222 sk->sk_v6_daddr = usin->sin6_addr;
223 np->flow_label = fl6.flowlabel;
224
225 /*
226 * TCP over IPv4
227 */
228
229 if (addr_type & IPV6_ADDR_MAPPED) {
230 u32 exthdrlen = icsk->icsk_ext_hdr_len;
231 struct sockaddr_in sin;
232
233 if (__ipv6_only_sock(sk))
234 return -ENETUNREACH;
235
236 sin.sin_family = AF_INET;
237 sin.sin_port = usin->sin6_port;
238 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
239
240 icsk->icsk_af_ops = &ipv6_mapped;
241 if (sk_is_mptcp(sk))
242 mptcpv6_handle_mapped(sk, true);
243 sk->sk_backlog_rcv = tcp_v4_do_rcv;
244#ifdef CONFIG_TCP_MD5SIG
245 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
246#endif
247
248 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
249
250 if (err) {
251 icsk->icsk_ext_hdr_len = exthdrlen;
252 icsk->icsk_af_ops = &ipv6_specific;
253 if (sk_is_mptcp(sk))
254 mptcpv6_handle_mapped(sk, false);
255 sk->sk_backlog_rcv = tcp_v6_do_rcv;
256#ifdef CONFIG_TCP_MD5SIG
257 tp->af_specific = &tcp_sock_ipv6_specific;
258#endif
259 goto failure;
260 }
261 np->saddr = sk->sk_v6_rcv_saddr;
262
263 return err;
264 }
265
266 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
267 saddr = &sk->sk_v6_rcv_saddr;
268
269 fl6.flowi6_proto = IPPROTO_TCP;
270 fl6.daddr = sk->sk_v6_daddr;
271 fl6.saddr = saddr ? *saddr : np->saddr;
272 fl6.flowi6_oif = sk->sk_bound_dev_if;
273 fl6.flowi6_mark = sk->sk_mark;
274 fl6.fl6_dport = usin->sin6_port;
275 fl6.fl6_sport = inet->inet_sport;
276 fl6.flowi6_uid = sk->sk_uid;
277
278 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
279 final_p = fl6_update_dst(&fl6, opt, &final);
280
281 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
282
283 dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
284 if (IS_ERR(dst)) {
285 err = PTR_ERR(dst);
286 goto failure;
287 }
288
289 if (!saddr) {
290 saddr = &fl6.saddr;
291 sk->sk_v6_rcv_saddr = *saddr;
292 }
293
294 /* set the source address */
295 np->saddr = *saddr;
296 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
297
298 sk->sk_gso_type = SKB_GSO_TCPV6;
299 ip6_dst_store(sk, dst, NULL, NULL);
300
301 icsk->icsk_ext_hdr_len = 0;
302 if (opt)
303 icsk->icsk_ext_hdr_len = opt->opt_flen +
304 opt->opt_nflen;
305
306 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
307
308 inet->inet_dport = usin->sin6_port;
309
310 tcp_set_state(sk, TCP_SYN_SENT);
311 err = inet6_hash_connect(tcp_death_row, sk);
312 if (err)
313 goto late_failure;
314
315 sk_set_txhash(sk);
316
317 if (likely(!tp->repair)) {
318 if (!tp->write_seq)
319 WRITE_ONCE(tp->write_seq,
320 secure_tcpv6_seq(np->saddr.s6_addr32,
321 sk->sk_v6_daddr.s6_addr32,
322 inet->inet_sport,
323 inet->inet_dport));
324 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
325 np->saddr.s6_addr32,
326 sk->sk_v6_daddr.s6_addr32);
327 }
328
329 if (tcp_fastopen_defer_connect(sk, &err))
330 return err;
331 if (err)
332 goto late_failure;
333
334 err = tcp_connect(sk);
335 if (err)
336 goto late_failure;
337
338 return 0;
339
340late_failure:
341 tcp_set_state(sk, TCP_CLOSE);
342failure:
343 inet->inet_dport = 0;
344 sk->sk_route_caps = 0;
345 return err;
346}
347
348static void tcp_v6_mtu_reduced(struct sock *sk)
349{
350 struct dst_entry *dst;
351
352 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
353 return;
354
355 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
356 if (!dst)
357 return;
358
359 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
360 tcp_sync_mss(sk, dst_mtu(dst));
361 tcp_simple_retransmit(sk);
362 }
363}
364
365static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
366 u8 type, u8 code, int offset, __be32 info)
367{
368 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
369 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
370 struct net *net = dev_net(skb->dev);
371 struct request_sock *fastopen;
372 struct ipv6_pinfo *np;
373 struct tcp_sock *tp;
374 __u32 seq, snd_una;
375 struct sock *sk;
376 bool fatal;
377 int err;
378
379 sk = __inet6_lookup_established(net, &tcp_hashinfo,
380 &hdr->daddr, th->dest,
381 &hdr->saddr, ntohs(th->source),
382 skb->dev->ifindex, inet6_sdif(skb));
383
384 if (!sk) {
385 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
386 ICMP6_MIB_INERRORS);
387 return -ENOENT;
388 }
389
390 if (sk->sk_state == TCP_TIME_WAIT) {
391 inet_twsk_put(inet_twsk(sk));
392 return 0;
393 }
394 seq = ntohl(th->seq);
395 fatal = icmpv6_err_convert(type, code, &err);
396 if (sk->sk_state == TCP_NEW_SYN_RECV) {
397 tcp_req_err(sk, seq, fatal);
398 return 0;
399 }
400
401 bh_lock_sock(sk);
402 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
403 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
404
405 if (sk->sk_state == TCP_CLOSE)
406 goto out;
407
408 if (ipv6_hdr(skb)->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
409 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
410 goto out;
411 }
412
413 tp = tcp_sk(sk);
414 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
415 fastopen = rcu_dereference(tp->fastopen_rsk);
416 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
417 if (sk->sk_state != TCP_LISTEN &&
418 !between(seq, snd_una, tp->snd_nxt)) {
419 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
420 goto out;
421 }
422
423 np = tcp_inet6_sk(sk);
424
425 if (type == NDISC_REDIRECT) {
426 if (!sock_owned_by_user(sk)) {
427 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
428
429 if (dst)
430 dst->ops->redirect(dst, sk, skb);
431 }
432 goto out;
433 }
434
435 if (type == ICMPV6_PKT_TOOBIG) {
436 /* We are not interested in TCP_LISTEN and open_requests
437 * (SYN-ACKs send out by Linux are always <576bytes so
438 * they should go through unfragmented).
439 */
440 if (sk->sk_state == TCP_LISTEN)
441 goto out;
442
443 if (!ip6_sk_accept_pmtu(sk))
444 goto out;
445
446 tp->mtu_info = ntohl(info);
447 if (!sock_owned_by_user(sk))
448 tcp_v6_mtu_reduced(sk);
449 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
450 &sk->sk_tsq_flags))
451 sock_hold(sk);
452 goto out;
453 }
454
455
456 /* Might be for an request_sock */
457 switch (sk->sk_state) {
458 case TCP_SYN_SENT:
459 case TCP_SYN_RECV:
460 /* Only in fast or simultaneous open. If a fast open socket is
461 * is already accepted it is treated as a connected one below.
462 */
463 if (fastopen && !fastopen->sk)
464 break;
465
466 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
467
468 if (!sock_owned_by_user(sk)) {
469 sk->sk_err = err;
470 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
471
472 tcp_done(sk);
473 } else
474 sk->sk_err_soft = err;
475 goto out;
476 case TCP_LISTEN:
477 break;
478 default:
479 /* check if this ICMP message allows revert of backoff.
480 * (see RFC 6069)
481 */
482 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
483 code == ICMPV6_NOROUTE)
484 tcp_ld_RTO_revert(sk, seq);
485 }
486
487 if (!sock_owned_by_user(sk) && np->recverr) {
488 sk->sk_err = err;
489 sk->sk_error_report(sk);
490 } else
491 sk->sk_err_soft = err;
492
493out:
494 bh_unlock_sock(sk);
495 sock_put(sk);
496 return 0;
497}
498
499
500static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
501 struct flowi *fl,
502 struct request_sock *req,
503 struct tcp_fastopen_cookie *foc,
504 enum tcp_synack_type synack_type)
505{
506 struct inet_request_sock *ireq = inet_rsk(req);
507 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
508 struct ipv6_txoptions *opt;
509 struct flowi6 *fl6 = &fl->u.ip6;
510 struct sk_buff *skb;
511 int err = -ENOMEM;
512
513 /* First, grab a route. */
514 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
515 IPPROTO_TCP)) == NULL)
516 goto done;
517
518 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
519
520 if (skb) {
521 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
522 &ireq->ir_v6_rmt_addr);
523
524 fl6->daddr = ireq->ir_v6_rmt_addr;
525 if (np->repflow && ireq->pktopts)
526 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
527
528 rcu_read_lock();
529 opt = ireq->ipv6_opt;
530 if (!opt)
531 opt = rcu_dereference(np->opt);
532 err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass,
533 sk->sk_priority);
534 rcu_read_unlock();
535 err = net_xmit_eval(err);
536 }
537
538done:
539 return err;
540}
541
542
543static void tcp_v6_reqsk_destructor(struct request_sock *req)
544{
545 kfree(inet_rsk(req)->ipv6_opt);
546 kfree_skb(inet_rsk(req)->pktopts);
547}
548
549#ifdef CONFIG_TCP_MD5SIG
550static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
551 const struct in6_addr *addr,
552 int l3index)
553{
554 return tcp_md5_do_lookup(sk, l3index,
555 (union tcp_md5_addr *)addr, AF_INET6);
556}
557
558static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
559 const struct sock *addr_sk)
560{
561 int l3index;
562
563 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
564 addr_sk->sk_bound_dev_if);
565 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
566 l3index);
567}
568
569static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
570 sockptr_t optval, int optlen)
571{
572 struct tcp_md5sig cmd;
573 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
574 int l3index = 0;
575 u8 prefixlen;
576
577 if (optlen < sizeof(cmd))
578 return -EINVAL;
579
580 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
581 return -EFAULT;
582
583 if (sin6->sin6_family != AF_INET6)
584 return -EINVAL;
585
586 if (optname == TCP_MD5SIG_EXT &&
587 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
588 prefixlen = cmd.tcpm_prefixlen;
589 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
590 prefixlen > 32))
591 return -EINVAL;
592 } else {
593 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
594 }
595
596 if (optname == TCP_MD5SIG_EXT &&
597 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
598 struct net_device *dev;
599
600 rcu_read_lock();
601 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
602 if (dev && netif_is_l3_master(dev))
603 l3index = dev->ifindex;
604 rcu_read_unlock();
605
606 /* ok to reference set/not set outside of rcu;
607 * right now device MUST be an L3 master
608 */
609 if (!dev || !l3index)
610 return -EINVAL;
611 }
612
613 if (!cmd.tcpm_keylen) {
614 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
615 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
616 AF_INET, prefixlen,
617 l3index);
618 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
619 AF_INET6, prefixlen, l3index);
620 }
621
622 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
623 return -EINVAL;
624
625 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
626 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
627 AF_INET, prefixlen, l3index,
628 cmd.tcpm_key, cmd.tcpm_keylen,
629 GFP_KERNEL);
630
631 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
632 AF_INET6, prefixlen, l3index,
633 cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
634}
635
636static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
637 const struct in6_addr *daddr,
638 const struct in6_addr *saddr,
639 const struct tcphdr *th, int nbytes)
640{
641 struct tcp6_pseudohdr *bp;
642 struct scatterlist sg;
643 struct tcphdr *_th;
644
645 bp = hp->scratch;
646 /* 1. TCP pseudo-header (RFC2460) */
647 bp->saddr = *saddr;
648 bp->daddr = *daddr;
649 bp->protocol = cpu_to_be32(IPPROTO_TCP);
650 bp->len = cpu_to_be32(nbytes);
651
652 _th = (struct tcphdr *)(bp + 1);
653 memcpy(_th, th, sizeof(*th));
654 _th->check = 0;
655
656 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
657 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
658 sizeof(*bp) + sizeof(*th));
659 return crypto_ahash_update(hp->md5_req);
660}
661
662static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
663 const struct in6_addr *daddr, struct in6_addr *saddr,
664 const struct tcphdr *th)
665{
666 struct tcp_md5sig_pool *hp;
667 struct ahash_request *req;
668
669 hp = tcp_get_md5sig_pool();
670 if (!hp)
671 goto clear_hash_noput;
672 req = hp->md5_req;
673
674 if (crypto_ahash_init(req))
675 goto clear_hash;
676 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
677 goto clear_hash;
678 if (tcp_md5_hash_key(hp, key))
679 goto clear_hash;
680 ahash_request_set_crypt(req, NULL, md5_hash, 0);
681 if (crypto_ahash_final(req))
682 goto clear_hash;
683
684 tcp_put_md5sig_pool();
685 return 0;
686
687clear_hash:
688 tcp_put_md5sig_pool();
689clear_hash_noput:
690 memset(md5_hash, 0, 16);
691 return 1;
692}
693
694static int tcp_v6_md5_hash_skb(char *md5_hash,
695 const struct tcp_md5sig_key *key,
696 const struct sock *sk,
697 const struct sk_buff *skb)
698{
699 const struct in6_addr *saddr, *daddr;
700 struct tcp_md5sig_pool *hp;
701 struct ahash_request *req;
702 const struct tcphdr *th = tcp_hdr(skb);
703
704 if (sk) { /* valid for establish/request sockets */
705 saddr = &sk->sk_v6_rcv_saddr;
706 daddr = &sk->sk_v6_daddr;
707 } else {
708 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
709 saddr = &ip6h->saddr;
710 daddr = &ip6h->daddr;
711 }
712
713 hp = tcp_get_md5sig_pool();
714 if (!hp)
715 goto clear_hash_noput;
716 req = hp->md5_req;
717
718 if (crypto_ahash_init(req))
719 goto clear_hash;
720
721 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
722 goto clear_hash;
723 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
724 goto clear_hash;
725 if (tcp_md5_hash_key(hp, key))
726 goto clear_hash;
727 ahash_request_set_crypt(req, NULL, md5_hash, 0);
728 if (crypto_ahash_final(req))
729 goto clear_hash;
730
731 tcp_put_md5sig_pool();
732 return 0;
733
734clear_hash:
735 tcp_put_md5sig_pool();
736clear_hash_noput:
737 memset(md5_hash, 0, 16);
738 return 1;
739}
740
741#endif
742
743static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
744 const struct sk_buff *skb,
745 int dif, int sdif)
746{
747#ifdef CONFIG_TCP_MD5SIG
748 const __u8 *hash_location = NULL;
749 struct tcp_md5sig_key *hash_expected;
750 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
751 const struct tcphdr *th = tcp_hdr(skb);
752 int genhash, l3index;
753 u8 newhash[16];
754
755 /* sdif set, means packet ingressed via a device
756 * in an L3 domain and dif is set to the l3mdev
757 */
758 l3index = sdif ? dif : 0;
759
760 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr, l3index);
761 hash_location = tcp_parse_md5sig_option(th);
762
763 /* We've parsed the options - do we have a hash? */
764 if (!hash_expected && !hash_location)
765 return false;
766
767 if (hash_expected && !hash_location) {
768 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
769 return true;
770 }
771
772 if (!hash_expected && hash_location) {
773 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
774 return true;
775 }
776
777 /* check the signature */
778 genhash = tcp_v6_md5_hash_skb(newhash,
779 hash_expected,
780 NULL, skb);
781
782 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
783 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
784 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u L3 index %d\n",
785 genhash ? "failed" : "mismatch",
786 &ip6h->saddr, ntohs(th->source),
787 &ip6h->daddr, ntohs(th->dest), l3index);
788 return true;
789 }
790#endif
791 return false;
792}
793
794static void tcp_v6_init_req(struct request_sock *req,
795 const struct sock *sk_listener,
796 struct sk_buff *skb)
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
805 /* So that link locals have meaning */
806 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
807 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
808 ireq->ir_iif = tcp_v6_iif(skb);
809
810 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
811 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
812 np->rxopt.bits.rxinfo ||
813 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
814 np->rxopt.bits.rxohlim || np->repflow)) {
815 refcount_inc(&skb->users);
816 ireq->pktopts = skb;
817 }
818}
819
820static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
821 struct flowi *fl,
822 const struct request_sock *req)
823{
824 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
825}
826
827struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
828 .family = AF_INET6,
829 .obj_size = sizeof(struct tcp6_request_sock),
830 .rtx_syn_ack = tcp_rtx_synack,
831 .send_ack = tcp_v6_reqsk_send_ack,
832 .destructor = tcp_v6_reqsk_destructor,
833 .send_reset = tcp_v6_send_reset,
834 .syn_ack_timeout = tcp_syn_ack_timeout,
835};
836
837const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
838 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
839 sizeof(struct ipv6hdr),
840#ifdef CONFIG_TCP_MD5SIG
841 .req_md5_lookup = tcp_v6_md5_lookup,
842 .calc_md5_hash = tcp_v6_md5_hash_skb,
843#endif
844 .init_req = tcp_v6_init_req,
845#ifdef CONFIG_SYN_COOKIES
846 .cookie_init_seq = cookie_v6_init_sequence,
847#endif
848 .route_req = tcp_v6_route_req,
849 .init_seq = tcp_v6_init_seq,
850 .init_ts_off = tcp_v6_init_ts_off,
851 .send_synack = tcp_v6_send_synack,
852};
853
854static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
855 u32 ack, u32 win, u32 tsval, u32 tsecr,
856 int oif, struct tcp_md5sig_key *key, int rst,
857 u8 tclass, __be32 label, u32 priority)
858{
859 const struct tcphdr *th = tcp_hdr(skb);
860 struct tcphdr *t1;
861 struct sk_buff *buff;
862 struct flowi6 fl6;
863 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
864 struct sock *ctl_sk = net->ipv6.tcp_sk;
865 unsigned int tot_len = sizeof(struct tcphdr);
866 struct dst_entry *dst;
867 __be32 *topt;
868 __u32 mark = 0;
869
870 if (tsecr)
871 tot_len += TCPOLEN_TSTAMP_ALIGNED;
872#ifdef CONFIG_TCP_MD5SIG
873 if (key)
874 tot_len += TCPOLEN_MD5SIG_ALIGNED;
875#endif
876
877 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
878 GFP_ATOMIC);
879 if (!buff)
880 return;
881
882 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
883
884 t1 = skb_push(buff, tot_len);
885 skb_reset_transport_header(buff);
886
887 /* Swap the send and the receive. */
888 memset(t1, 0, sizeof(*t1));
889 t1->dest = th->source;
890 t1->source = th->dest;
891 t1->doff = tot_len / 4;
892 t1->seq = htonl(seq);
893 t1->ack_seq = htonl(ack);
894 t1->ack = !rst || !th->ack;
895 t1->rst = rst;
896 t1->window = htons(win);
897
898 topt = (__be32 *)(t1 + 1);
899
900 if (tsecr) {
901 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
902 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
903 *topt++ = htonl(tsval);
904 *topt++ = htonl(tsecr);
905 }
906
907#ifdef CONFIG_TCP_MD5SIG
908 if (key) {
909 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
910 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
911 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
912 &ipv6_hdr(skb)->saddr,
913 &ipv6_hdr(skb)->daddr, t1);
914 }
915#endif
916
917 memset(&fl6, 0, sizeof(fl6));
918 fl6.daddr = ipv6_hdr(skb)->saddr;
919 fl6.saddr = ipv6_hdr(skb)->daddr;
920 fl6.flowlabel = label;
921
922 buff->ip_summed = CHECKSUM_PARTIAL;
923 buff->csum = 0;
924
925 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
926
927 fl6.flowi6_proto = IPPROTO_TCP;
928 if (rt6_need_strict(&fl6.daddr) && !oif)
929 fl6.flowi6_oif = tcp_v6_iif(skb);
930 else {
931 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
932 oif = skb->skb_iif;
933
934 fl6.flowi6_oif = oif;
935 }
936
937 if (sk) {
938 if (sk->sk_state == TCP_TIME_WAIT) {
939 mark = inet_twsk(sk)->tw_mark;
940 /* autoflowlabel relies on buff->hash */
941 skb_set_hash(buff, inet_twsk(sk)->tw_txhash,
942 PKT_HASH_TYPE_L4);
943 } else {
944 mark = sk->sk_mark;
945 }
946 buff->tstamp = tcp_transmit_time(sk);
947 }
948 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
949 fl6.fl6_dport = t1->dest;
950 fl6.fl6_sport = t1->source;
951 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
952 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
953
954 /* Pass a socket to ip6_dst_lookup either it is for RST
955 * Underlying function will use this to retrieve the network
956 * namespace
957 */
958 dst = ip6_dst_lookup_flow(sock_net(ctl_sk), ctl_sk, &fl6, NULL);
959 if (!IS_ERR(dst)) {
960 skb_dst_set(buff, dst);
961 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, tclass,
962 priority);
963 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
964 if (rst)
965 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
966 return;
967 }
968
969 kfree_skb(buff);
970}
971
972static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
973{
974 const struct tcphdr *th = tcp_hdr(skb);
975 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
976 u32 seq = 0, ack_seq = 0;
977 struct tcp_md5sig_key *key = NULL;
978#ifdef CONFIG_TCP_MD5SIG
979 const __u8 *hash_location = NULL;
980 unsigned char newhash[16];
981 int genhash;
982 struct sock *sk1 = NULL;
983#endif
984 __be32 label = 0;
985 u32 priority = 0;
986 struct net *net;
987 int oif = 0;
988
989 if (th->rst)
990 return;
991
992 /* If sk not NULL, it means we did a successful lookup and incoming
993 * route had to be correct. prequeue might have dropped our dst.
994 */
995 if (!sk && !ipv6_unicast_destination(skb))
996 return;
997
998 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
999#ifdef CONFIG_TCP_MD5SIG
1000 rcu_read_lock();
1001 hash_location = tcp_parse_md5sig_option(th);
1002 if (sk && sk_fullsock(sk)) {
1003 int l3index;
1004
1005 /* sdif set, means packet ingressed via a device
1006 * in an L3 domain and inet_iif is set to it.
1007 */
1008 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1009 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1010 } else if (hash_location) {
1011 int dif = tcp_v6_iif_l3_slave(skb);
1012 int sdif = tcp_v6_sdif(skb);
1013 int l3index;
1014
1015 /*
1016 * active side is lost. Try to find listening socket through
1017 * source port, and then find md5 key through listening socket.
1018 * we are not loose security here:
1019 * Incoming packet is checked with md5 hash with finding key,
1020 * no RST generated if md5 hash doesn't match.
1021 */
1022 sk1 = inet6_lookup_listener(net,
1023 &tcp_hashinfo, NULL, 0,
1024 &ipv6h->saddr,
1025 th->source, &ipv6h->daddr,
1026 ntohs(th->source), dif, sdif);
1027 if (!sk1)
1028 goto out;
1029
1030 /* sdif set, means packet ingressed via a device
1031 * in an L3 domain and dif is set to it.
1032 */
1033 l3index = tcp_v6_sdif(skb) ? dif : 0;
1034
1035 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1036 if (!key)
1037 goto out;
1038
1039 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
1040 if (genhash || memcmp(hash_location, newhash, 16) != 0)
1041 goto out;
1042 }
1043#endif
1044
1045 if (th->ack)
1046 seq = ntohl(th->ack_seq);
1047 else
1048 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1049 (th->doff << 2);
1050
1051 if (sk) {
1052 oif = sk->sk_bound_dev_if;
1053 if (sk_fullsock(sk)) {
1054 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1055
1056 trace_tcp_send_reset(sk, skb);
1057 if (np->repflow)
1058 label = ip6_flowlabel(ipv6h);
1059 priority = sk->sk_priority;
1060 }
1061 if (sk->sk_state == TCP_TIME_WAIT) {
1062 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1063 priority = inet_twsk(sk)->tw_priority;
1064 }
1065 } else {
1066 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1067 label = ip6_flowlabel(ipv6h);
1068 }
1069
1070 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0,
1071 label, priority);
1072
1073#ifdef CONFIG_TCP_MD5SIG
1074out:
1075 rcu_read_unlock();
1076#endif
1077}
1078
1079static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1080 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1081 struct tcp_md5sig_key *key, u8 tclass,
1082 __be32 label, u32 priority)
1083{
1084 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1085 tclass, label, priority);
1086}
1087
1088static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1089{
1090 struct inet_timewait_sock *tw = inet_twsk(sk);
1091 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1092
1093 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1094 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1095 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1096 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1097 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority);
1098
1099 inet_twsk_put(tw);
1100}
1101
1102static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1103 struct request_sock *req)
1104{
1105 int l3index;
1106
1107 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1108
1109 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1110 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1111 */
1112 /* RFC 7323 2.3
1113 * The window field (SEG.WND) of every outgoing segment, with the
1114 * exception of <SYN> segments, MUST be right-shifted by
1115 * Rcv.Wind.Shift bits:
1116 */
1117 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1118 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1119 tcp_rsk(req)->rcv_nxt,
1120 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1121 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1122 req->ts_recent, sk->sk_bound_dev_if,
1123 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
1124 0, 0, sk->sk_priority);
1125}
1126
1127
1128static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1129{
1130#ifdef CONFIG_SYN_COOKIES
1131 const struct tcphdr *th = tcp_hdr(skb);
1132
1133 if (!th->syn)
1134 sk = cookie_v6_check(sk, skb);
1135#endif
1136 return sk;
1137}
1138
1139u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1140 struct tcphdr *th, u32 *cookie)
1141{
1142 u16 mss = 0;
1143#ifdef CONFIG_SYN_COOKIES
1144 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1145 &tcp_request_sock_ipv6_ops, sk, th);
1146 if (mss) {
1147 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1148 tcp_synq_overflow(sk);
1149 }
1150#endif
1151 return mss;
1152}
1153
1154static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1155{
1156 if (skb->protocol == htons(ETH_P_IP))
1157 return tcp_v4_conn_request(sk, skb);
1158
1159 if (!ipv6_unicast_destination(skb))
1160 goto drop;
1161
1162 return tcp_conn_request(&tcp6_request_sock_ops,
1163 &tcp_request_sock_ipv6_ops, sk, skb);
1164
1165drop:
1166 tcp_listendrop(sk);
1167 return 0; /* don't send reset */
1168}
1169
1170static void tcp_v6_restore_cb(struct sk_buff *skb)
1171{
1172 /* We need to move header back to the beginning if xfrm6_policy_check()
1173 * and tcp_v6_fill_cb() are going to be called again.
1174 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1175 */
1176 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1177 sizeof(struct inet6_skb_parm));
1178}
1179
1180static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1181 struct request_sock *req,
1182 struct dst_entry *dst,
1183 struct request_sock *req_unhash,
1184 bool *own_req)
1185{
1186 struct inet_request_sock *ireq;
1187 struct ipv6_pinfo *newnp;
1188 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1189 struct ipv6_txoptions *opt;
1190 struct inet_sock *newinet;
1191 struct tcp_sock *newtp;
1192 struct sock *newsk;
1193#ifdef CONFIG_TCP_MD5SIG
1194 struct tcp_md5sig_key *key;
1195 int l3index;
1196#endif
1197 struct flowi6 fl6;
1198
1199 if (skb->protocol == htons(ETH_P_IP)) {
1200 /*
1201 * v6 mapped
1202 */
1203
1204 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1205 req_unhash, own_req);
1206
1207 if (!newsk)
1208 return NULL;
1209
1210 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1211
1212 newinet = inet_sk(newsk);
1213 newnp = tcp_inet6_sk(newsk);
1214 newtp = tcp_sk(newsk);
1215
1216 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1217
1218 newnp->saddr = newsk->sk_v6_rcv_saddr;
1219
1220 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1221 if (sk_is_mptcp(newsk))
1222 mptcpv6_handle_mapped(newsk, true);
1223 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1224#ifdef CONFIG_TCP_MD5SIG
1225 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1226#endif
1227
1228 newnp->ipv6_mc_list = NULL;
1229 newnp->ipv6_ac_list = NULL;
1230 newnp->ipv6_fl_list = NULL;
1231 newnp->pktoptions = NULL;
1232 newnp->opt = NULL;
1233 newnp->mcast_oif = inet_iif(skb);
1234 newnp->mcast_hops = ip_hdr(skb)->ttl;
1235 newnp->rcv_flowinfo = 0;
1236 if (np->repflow)
1237 newnp->flow_label = 0;
1238
1239 /*
1240 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1241 * here, tcp_create_openreq_child now does this for us, see the comment in
1242 * that function for the gory details. -acme
1243 */
1244
1245 /* It is tricky place. Until this moment IPv4 tcp
1246 worked with IPv6 icsk.icsk_af_ops.
1247 Sync it now.
1248 */
1249 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1250
1251 return newsk;
1252 }
1253
1254 ireq = inet_rsk(req);
1255
1256 if (sk_acceptq_is_full(sk))
1257 goto out_overflow;
1258
1259 if (!dst) {
1260 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1261 if (!dst)
1262 goto out;
1263 }
1264
1265 newsk = tcp_create_openreq_child(sk, req, skb);
1266 if (!newsk)
1267 goto out_nonewsk;
1268
1269 /*
1270 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1271 * count here, tcp_create_openreq_child now does this for us, see the
1272 * comment in that function for the gory details. -acme
1273 */
1274
1275 newsk->sk_gso_type = SKB_GSO_TCPV6;
1276 ip6_dst_store(newsk, dst, NULL, NULL);
1277 inet6_sk_rx_dst_set(newsk, skb);
1278
1279 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1280
1281 newtp = tcp_sk(newsk);
1282 newinet = inet_sk(newsk);
1283 newnp = tcp_inet6_sk(newsk);
1284
1285 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1286
1287 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1288 newnp->saddr = ireq->ir_v6_loc_addr;
1289 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1290 newsk->sk_bound_dev_if = ireq->ir_iif;
1291
1292 /* Now IPv6 options...
1293
1294 First: no IPv4 options.
1295 */
1296 newinet->inet_opt = NULL;
1297 newnp->ipv6_mc_list = NULL;
1298 newnp->ipv6_ac_list = NULL;
1299 newnp->ipv6_fl_list = NULL;
1300
1301 /* Clone RX bits */
1302 newnp->rxopt.all = np->rxopt.all;
1303
1304 newnp->pktoptions = NULL;
1305 newnp->opt = NULL;
1306 newnp->mcast_oif = tcp_v6_iif(skb);
1307 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1308 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1309 if (np->repflow)
1310 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1311
1312 /* Clone native IPv6 options from listening socket (if any)
1313
1314 Yes, keeping reference count would be much more clever,
1315 but we make one more one thing there: reattach optmem
1316 to newsk.
1317 */
1318 opt = ireq->ipv6_opt;
1319 if (!opt)
1320 opt = rcu_dereference(np->opt);
1321 if (opt) {
1322 opt = ipv6_dup_options(newsk, opt);
1323 RCU_INIT_POINTER(newnp->opt, opt);
1324 }
1325 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1326 if (opt)
1327 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1328 opt->opt_flen;
1329
1330 tcp_ca_openreq_child(newsk, dst);
1331
1332 tcp_sync_mss(newsk, dst_mtu(dst));
1333 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1334
1335 tcp_initialize_rcv_mss(newsk);
1336
1337 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1338 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1339
1340#ifdef CONFIG_TCP_MD5SIG
1341 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1342
1343 /* Copy over the MD5 key from the original socket */
1344 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1345 if (key) {
1346 /* We're using one, so create a matching key
1347 * on the newsk structure. If we fail to get
1348 * memory, then we end up not copying the key
1349 * across. Shucks.
1350 */
1351 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1352 AF_INET6, 128, l3index, key->key, key->keylen,
1353 sk_gfp_mask(sk, GFP_ATOMIC));
1354 }
1355#endif
1356
1357 if (__inet_inherit_port(sk, newsk) < 0) {
1358 inet_csk_prepare_forced_close(newsk);
1359 tcp_done(newsk);
1360 goto out;
1361 }
1362 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1363 if (*own_req) {
1364 tcp_move_syn(newtp, req);
1365
1366 /* Clone pktoptions received with SYN, if we own the req */
1367 if (ireq->pktopts) {
1368 newnp->pktoptions = skb_clone(ireq->pktopts,
1369 sk_gfp_mask(sk, GFP_ATOMIC));
1370 consume_skb(ireq->pktopts);
1371 ireq->pktopts = NULL;
1372 if (newnp->pktoptions) {
1373 tcp_v6_restore_cb(newnp->pktoptions);
1374 skb_set_owner_r(newnp->pktoptions, newsk);
1375 }
1376 }
1377 }
1378
1379 return newsk;
1380
1381out_overflow:
1382 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1383out_nonewsk:
1384 dst_release(dst);
1385out:
1386 tcp_listendrop(sk);
1387 return NULL;
1388}
1389
1390/* The socket must have it's spinlock held when we get
1391 * here, unless it is a TCP_LISTEN socket.
1392 *
1393 * We have a potential double-lock case here, so even when
1394 * doing backlog processing we use the BH locking scheme.
1395 * This is because we cannot sleep with the original spinlock
1396 * held.
1397 */
1398static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1399{
1400 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1401 struct sk_buff *opt_skb = NULL;
1402 struct tcp_sock *tp;
1403
1404 /* Imagine: socket is IPv6. IPv4 packet arrives,
1405 goes to IPv4 receive handler and backlogged.
1406 From backlog it always goes here. Kerboom...
1407 Fortunately, tcp_rcv_established and rcv_established
1408 handle them correctly, but it is not case with
1409 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1410 */
1411
1412 if (skb->protocol == htons(ETH_P_IP))
1413 return tcp_v4_do_rcv(sk, skb);
1414
1415 /*
1416 * socket locking is here for SMP purposes as backlog rcv
1417 * is currently called with bh processing disabled.
1418 */
1419
1420 /* Do Stevens' IPV6_PKTOPTIONS.
1421
1422 Yes, guys, it is the only place in our code, where we
1423 may make it not affecting IPv4.
1424 The rest of code is protocol independent,
1425 and I do not like idea to uglify IPv4.
1426
1427 Actually, all the idea behind IPV6_PKTOPTIONS
1428 looks not very well thought. For now we latch
1429 options, received in the last packet, enqueued
1430 by tcp. Feel free to propose better solution.
1431 --ANK (980728)
1432 */
1433 if (np->rxopt.all)
1434 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1435
1436 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1437 struct dst_entry *dst = sk->sk_rx_dst;
1438
1439 sock_rps_save_rxhash(sk, skb);
1440 sk_mark_napi_id(sk, skb);
1441 if (dst) {
1442 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1443 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1444 dst_release(dst);
1445 sk->sk_rx_dst = NULL;
1446 }
1447 }
1448
1449 tcp_rcv_established(sk, skb);
1450 if (opt_skb)
1451 goto ipv6_pktoptions;
1452 return 0;
1453 }
1454
1455 if (tcp_checksum_complete(skb))
1456 goto csum_err;
1457
1458 if (sk->sk_state == TCP_LISTEN) {
1459 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1460
1461 if (!nsk)
1462 goto discard;
1463
1464 if (nsk != sk) {
1465 if (tcp_child_process(sk, nsk, skb))
1466 goto reset;
1467 if (opt_skb)
1468 __kfree_skb(opt_skb);
1469 return 0;
1470 }
1471 } else
1472 sock_rps_save_rxhash(sk, skb);
1473
1474 if (tcp_rcv_state_process(sk, skb))
1475 goto reset;
1476 if (opt_skb)
1477 goto ipv6_pktoptions;
1478 return 0;
1479
1480reset:
1481 tcp_v6_send_reset(sk, skb);
1482discard:
1483 if (opt_skb)
1484 __kfree_skb(opt_skb);
1485 kfree_skb(skb);
1486 return 0;
1487csum_err:
1488 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1489 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1490 goto discard;
1491
1492
1493ipv6_pktoptions:
1494 /* Do you ask, what is it?
1495
1496 1. skb was enqueued by tcp.
1497 2. skb is added to tail of read queue, rather than out of order.
1498 3. socket is not in passive state.
1499 4. Finally, it really contains options, which user wants to receive.
1500 */
1501 tp = tcp_sk(sk);
1502 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1503 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1504 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1505 np->mcast_oif = tcp_v6_iif(opt_skb);
1506 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1507 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1508 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1509 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1510 if (np->repflow)
1511 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1512 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1513 skb_set_owner_r(opt_skb, sk);
1514 tcp_v6_restore_cb(opt_skb);
1515 opt_skb = xchg(&np->pktoptions, opt_skb);
1516 } else {
1517 __kfree_skb(opt_skb);
1518 opt_skb = xchg(&np->pktoptions, NULL);
1519 }
1520 }
1521
1522 kfree_skb(opt_skb);
1523 return 0;
1524}
1525
1526static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1527 const struct tcphdr *th)
1528{
1529 /* This is tricky: we move IP6CB at its correct location into
1530 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1531 * _decode_session6() uses IP6CB().
1532 * barrier() makes sure compiler won't play aliasing games.
1533 */
1534 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1535 sizeof(struct inet6_skb_parm));
1536 barrier();
1537
1538 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1539 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1540 skb->len - th->doff*4);
1541 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1542 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1543 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1544 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1545 TCP_SKB_CB(skb)->sacked = 0;
1546 TCP_SKB_CB(skb)->has_rxtstamp =
1547 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1548}
1549
1550INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1551{
1552 struct sk_buff *skb_to_free;
1553 int sdif = inet6_sdif(skb);
1554 int dif = inet6_iif(skb);
1555 const struct tcphdr *th;
1556 const struct ipv6hdr *hdr;
1557 bool refcounted;
1558 struct sock *sk;
1559 int ret;
1560 struct net *net = dev_net(skb->dev);
1561
1562 if (skb->pkt_type != PACKET_HOST)
1563 goto discard_it;
1564
1565 /*
1566 * Count it even if it's bad.
1567 */
1568 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1569
1570 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1571 goto discard_it;
1572
1573 th = (const struct tcphdr *)skb->data;
1574
1575 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1576 goto bad_packet;
1577 if (!pskb_may_pull(skb, th->doff*4))
1578 goto discard_it;
1579
1580 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1581 goto csum_error;
1582
1583 th = (const struct tcphdr *)skb->data;
1584 hdr = ipv6_hdr(skb);
1585
1586lookup:
1587 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1588 th->source, th->dest, inet6_iif(skb), sdif,
1589 &refcounted);
1590 if (!sk)
1591 goto no_tcp_socket;
1592
1593process:
1594 if (sk->sk_state == TCP_TIME_WAIT)
1595 goto do_time_wait;
1596
1597 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1598 struct request_sock *req = inet_reqsk(sk);
1599 bool req_stolen = false;
1600 struct sock *nsk;
1601
1602 sk = req->rsk_listener;
1603 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif)) {
1604 sk_drops_add(sk, skb);
1605 reqsk_put(req);
1606 goto discard_it;
1607 }
1608 if (tcp_checksum_complete(skb)) {
1609 reqsk_put(req);
1610 goto csum_error;
1611 }
1612 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1613 inet_csk_reqsk_queue_drop_and_put(sk, req);
1614 goto lookup;
1615 }
1616 sock_hold(sk);
1617 refcounted = true;
1618 nsk = NULL;
1619 if (!tcp_filter(sk, skb)) {
1620 th = (const struct tcphdr *)skb->data;
1621 hdr = ipv6_hdr(skb);
1622 tcp_v6_fill_cb(skb, hdr, th);
1623 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1624 }
1625 if (!nsk) {
1626 reqsk_put(req);
1627 if (req_stolen) {
1628 /* Another cpu got exclusive access to req
1629 * and created a full blown socket.
1630 * Try to feed this packet to this socket
1631 * instead of discarding it.
1632 */
1633 tcp_v6_restore_cb(skb);
1634 sock_put(sk);
1635 goto lookup;
1636 }
1637 goto discard_and_relse;
1638 }
1639 if (nsk == sk) {
1640 reqsk_put(req);
1641 tcp_v6_restore_cb(skb);
1642 } else if (tcp_child_process(sk, nsk, skb)) {
1643 tcp_v6_send_reset(nsk, skb);
1644 goto discard_and_relse;
1645 } else {
1646 sock_put(sk);
1647 return 0;
1648 }
1649 }
1650 if (hdr->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
1651 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1652 goto discard_and_relse;
1653 }
1654
1655 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1656 goto discard_and_relse;
1657
1658 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif))
1659 goto discard_and_relse;
1660
1661 if (tcp_filter(sk, skb))
1662 goto discard_and_relse;
1663 th = (const struct tcphdr *)skb->data;
1664 hdr = ipv6_hdr(skb);
1665 tcp_v6_fill_cb(skb, hdr, th);
1666
1667 skb->dev = NULL;
1668
1669 if (sk->sk_state == TCP_LISTEN) {
1670 ret = tcp_v6_do_rcv(sk, skb);
1671 goto put_and_return;
1672 }
1673
1674 sk_incoming_cpu_update(sk);
1675
1676 bh_lock_sock_nested(sk);
1677 tcp_segs_in(tcp_sk(sk), skb);
1678 ret = 0;
1679 if (!sock_owned_by_user(sk)) {
1680 skb_to_free = sk->sk_rx_skb_cache;
1681 sk->sk_rx_skb_cache = NULL;
1682 ret = tcp_v6_do_rcv(sk, skb);
1683 } else {
1684 if (tcp_add_backlog(sk, skb))
1685 goto discard_and_relse;
1686 skb_to_free = NULL;
1687 }
1688 bh_unlock_sock(sk);
1689 if (skb_to_free)
1690 __kfree_skb(skb_to_free);
1691put_and_return:
1692 if (refcounted)
1693 sock_put(sk);
1694 return ret ? -1 : 0;
1695
1696no_tcp_socket:
1697 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1698 goto discard_it;
1699
1700 tcp_v6_fill_cb(skb, hdr, th);
1701
1702 if (tcp_checksum_complete(skb)) {
1703csum_error:
1704 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1705bad_packet:
1706 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1707 } else {
1708 tcp_v6_send_reset(NULL, skb);
1709 }
1710
1711discard_it:
1712 kfree_skb(skb);
1713 return 0;
1714
1715discard_and_relse:
1716 sk_drops_add(sk, skb);
1717 if (refcounted)
1718 sock_put(sk);
1719 goto discard_it;
1720
1721do_time_wait:
1722 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1723 inet_twsk_put(inet_twsk(sk));
1724 goto discard_it;
1725 }
1726
1727 tcp_v6_fill_cb(skb, hdr, th);
1728
1729 if (tcp_checksum_complete(skb)) {
1730 inet_twsk_put(inet_twsk(sk));
1731 goto csum_error;
1732 }
1733
1734 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1735 case TCP_TW_SYN:
1736 {
1737 struct sock *sk2;
1738
1739 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1740 skb, __tcp_hdrlen(th),
1741 &ipv6_hdr(skb)->saddr, th->source,
1742 &ipv6_hdr(skb)->daddr,
1743 ntohs(th->dest),
1744 tcp_v6_iif_l3_slave(skb),
1745 sdif);
1746 if (sk2) {
1747 struct inet_timewait_sock *tw = inet_twsk(sk);
1748 inet_twsk_deschedule_put(tw);
1749 sk = sk2;
1750 tcp_v6_restore_cb(skb);
1751 refcounted = false;
1752 goto process;
1753 }
1754 }
1755 /* to ACK */
1756 fallthrough;
1757 case TCP_TW_ACK:
1758 tcp_v6_timewait_ack(sk, skb);
1759 break;
1760 case TCP_TW_RST:
1761 tcp_v6_send_reset(sk, skb);
1762 inet_twsk_deschedule_put(inet_twsk(sk));
1763 goto discard_it;
1764 case TCP_TW_SUCCESS:
1765 ;
1766 }
1767 goto discard_it;
1768}
1769
1770INDIRECT_CALLABLE_SCOPE void tcp_v6_early_demux(struct sk_buff *skb)
1771{
1772 const struct ipv6hdr *hdr;
1773 const struct tcphdr *th;
1774 struct sock *sk;
1775
1776 if (skb->pkt_type != PACKET_HOST)
1777 return;
1778
1779 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1780 return;
1781
1782 hdr = ipv6_hdr(skb);
1783 th = tcp_hdr(skb);
1784
1785 if (th->doff < sizeof(struct tcphdr) / 4)
1786 return;
1787
1788 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1789 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1790 &hdr->saddr, th->source,
1791 &hdr->daddr, ntohs(th->dest),
1792 inet6_iif(skb), inet6_sdif(skb));
1793 if (sk) {
1794 skb->sk = sk;
1795 skb->destructor = sock_edemux;
1796 if (sk_fullsock(sk)) {
1797 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1798
1799 if (dst)
1800 dst = dst_check(dst, tcp_inet6_sk(sk)->rx_dst_cookie);
1801 if (dst &&
1802 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1803 skb_dst_set_noref(skb, dst);
1804 }
1805 }
1806}
1807
1808static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1809 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1810 .twsk_unique = tcp_twsk_unique,
1811 .twsk_destructor = tcp_twsk_destructor,
1812};
1813
1814INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
1815{
1816 struct ipv6_pinfo *np = inet6_sk(sk);
1817
1818 __tcp_v6_send_check(skb, &np->saddr, &sk->sk_v6_daddr);
1819}
1820
1821const struct inet_connection_sock_af_ops ipv6_specific = {
1822 .queue_xmit = inet6_csk_xmit,
1823 .send_check = tcp_v6_send_check,
1824 .rebuild_header = inet6_sk_rebuild_header,
1825 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1826 .conn_request = tcp_v6_conn_request,
1827 .syn_recv_sock = tcp_v6_syn_recv_sock,
1828 .net_header_len = sizeof(struct ipv6hdr),
1829 .net_frag_header_len = sizeof(struct frag_hdr),
1830 .setsockopt = ipv6_setsockopt,
1831 .getsockopt = ipv6_getsockopt,
1832 .addr2sockaddr = inet6_csk_addr2sockaddr,
1833 .sockaddr_len = sizeof(struct sockaddr_in6),
1834 .mtu_reduced = tcp_v6_mtu_reduced,
1835};
1836
1837#ifdef CONFIG_TCP_MD5SIG
1838static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1839 .md5_lookup = tcp_v6_md5_lookup,
1840 .calc_md5_hash = tcp_v6_md5_hash_skb,
1841 .md5_parse = tcp_v6_parse_md5_keys,
1842};
1843#endif
1844
1845/*
1846 * TCP over IPv4 via INET6 API
1847 */
1848static const struct inet_connection_sock_af_ops ipv6_mapped = {
1849 .queue_xmit = ip_queue_xmit,
1850 .send_check = tcp_v4_send_check,
1851 .rebuild_header = inet_sk_rebuild_header,
1852 .sk_rx_dst_set = inet_sk_rx_dst_set,
1853 .conn_request = tcp_v6_conn_request,
1854 .syn_recv_sock = tcp_v6_syn_recv_sock,
1855 .net_header_len = sizeof(struct iphdr),
1856 .setsockopt = ipv6_setsockopt,
1857 .getsockopt = ipv6_getsockopt,
1858 .addr2sockaddr = inet6_csk_addr2sockaddr,
1859 .sockaddr_len = sizeof(struct sockaddr_in6),
1860 .mtu_reduced = tcp_v4_mtu_reduced,
1861};
1862
1863#ifdef CONFIG_TCP_MD5SIG
1864static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1865 .md5_lookup = tcp_v4_md5_lookup,
1866 .calc_md5_hash = tcp_v4_md5_hash_skb,
1867 .md5_parse = tcp_v6_parse_md5_keys,
1868};
1869#endif
1870
1871/* NOTE: A lot of things set to zero explicitly by call to
1872 * sk_alloc() so need not be done here.
1873 */
1874static int tcp_v6_init_sock(struct sock *sk)
1875{
1876 struct inet_connection_sock *icsk = inet_csk(sk);
1877
1878 tcp_init_sock(sk);
1879
1880 icsk->icsk_af_ops = &ipv6_specific;
1881
1882#ifdef CONFIG_TCP_MD5SIG
1883 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1884#endif
1885
1886 return 0;
1887}
1888
1889static void tcp_v6_destroy_sock(struct sock *sk)
1890{
1891 tcp_v4_destroy_sock(sk);
1892 inet6_destroy_sock(sk);
1893}
1894
1895#ifdef CONFIG_PROC_FS
1896/* Proc filesystem TCPv6 sock list dumping. */
1897static void get_openreq6(struct seq_file *seq,
1898 const struct request_sock *req, int i)
1899{
1900 long ttd = req->rsk_timer.expires - jiffies;
1901 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1902 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1903
1904 if (ttd < 0)
1905 ttd = 0;
1906
1907 seq_printf(seq,
1908 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1909 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1910 i,
1911 src->s6_addr32[0], src->s6_addr32[1],
1912 src->s6_addr32[2], src->s6_addr32[3],
1913 inet_rsk(req)->ir_num,
1914 dest->s6_addr32[0], dest->s6_addr32[1],
1915 dest->s6_addr32[2], dest->s6_addr32[3],
1916 ntohs(inet_rsk(req)->ir_rmt_port),
1917 TCP_SYN_RECV,
1918 0, 0, /* could print option size, but that is af dependent. */
1919 1, /* timers active (only the expire timer) */
1920 jiffies_to_clock_t(ttd),
1921 req->num_timeout,
1922 from_kuid_munged(seq_user_ns(seq),
1923 sock_i_uid(req->rsk_listener)),
1924 0, /* non standard timer */
1925 0, /* open_requests have no inode */
1926 0, req);
1927}
1928
1929static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1930{
1931 const struct in6_addr *dest, *src;
1932 __u16 destp, srcp;
1933 int timer_active;
1934 unsigned long timer_expires;
1935 const struct inet_sock *inet = inet_sk(sp);
1936 const struct tcp_sock *tp = tcp_sk(sp);
1937 const struct inet_connection_sock *icsk = inet_csk(sp);
1938 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1939 int rx_queue;
1940 int state;
1941
1942 dest = &sp->sk_v6_daddr;
1943 src = &sp->sk_v6_rcv_saddr;
1944 destp = ntohs(inet->inet_dport);
1945 srcp = ntohs(inet->inet_sport);
1946
1947 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1948 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
1949 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1950 timer_active = 1;
1951 timer_expires = icsk->icsk_timeout;
1952 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1953 timer_active = 4;
1954 timer_expires = icsk->icsk_timeout;
1955 } else if (timer_pending(&sp->sk_timer)) {
1956 timer_active = 2;
1957 timer_expires = sp->sk_timer.expires;
1958 } else {
1959 timer_active = 0;
1960 timer_expires = jiffies;
1961 }
1962
1963 state = inet_sk_state_load(sp);
1964 if (state == TCP_LISTEN)
1965 rx_queue = READ_ONCE(sp->sk_ack_backlog);
1966 else
1967 /* Because we don't lock the socket,
1968 * we might find a transient negative value.
1969 */
1970 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
1971 READ_ONCE(tp->copied_seq), 0);
1972
1973 seq_printf(seq,
1974 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1975 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1976 i,
1977 src->s6_addr32[0], src->s6_addr32[1],
1978 src->s6_addr32[2], src->s6_addr32[3], srcp,
1979 dest->s6_addr32[0], dest->s6_addr32[1],
1980 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1981 state,
1982 READ_ONCE(tp->write_seq) - tp->snd_una,
1983 rx_queue,
1984 timer_active,
1985 jiffies_delta_to_clock_t(timer_expires - jiffies),
1986 icsk->icsk_retransmits,
1987 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1988 icsk->icsk_probes_out,
1989 sock_i_ino(sp),
1990 refcount_read(&sp->sk_refcnt), sp,
1991 jiffies_to_clock_t(icsk->icsk_rto),
1992 jiffies_to_clock_t(icsk->icsk_ack.ato),
1993 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
1994 tp->snd_cwnd,
1995 state == TCP_LISTEN ?
1996 fastopenq->max_qlen :
1997 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1998 );
1999}
2000
2001static void get_timewait6_sock(struct seq_file *seq,
2002 struct inet_timewait_sock *tw, int i)
2003{
2004 long delta = tw->tw_timer.expires - jiffies;
2005 const struct in6_addr *dest, *src;
2006 __u16 destp, srcp;
2007
2008 dest = &tw->tw_v6_daddr;
2009 src = &tw->tw_v6_rcv_saddr;
2010 destp = ntohs(tw->tw_dport);
2011 srcp = ntohs(tw->tw_sport);
2012
2013 seq_printf(seq,
2014 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2015 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2016 i,
2017 src->s6_addr32[0], src->s6_addr32[1],
2018 src->s6_addr32[2], src->s6_addr32[3], srcp,
2019 dest->s6_addr32[0], dest->s6_addr32[1],
2020 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2021 tw->tw_substate, 0, 0,
2022 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2023 refcount_read(&tw->tw_refcnt), tw);
2024}
2025
2026static int tcp6_seq_show(struct seq_file *seq, void *v)
2027{
2028 struct tcp_iter_state *st;
2029 struct sock *sk = v;
2030
2031 if (v == SEQ_START_TOKEN) {
2032 seq_puts(seq,
2033 " sl "
2034 "local_address "
2035 "remote_address "
2036 "st tx_queue rx_queue tr tm->when retrnsmt"
2037 " uid timeout inode\n");
2038 goto out;
2039 }
2040 st = seq->private;
2041
2042 if (sk->sk_state == TCP_TIME_WAIT)
2043 get_timewait6_sock(seq, v, st->num);
2044 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2045 get_openreq6(seq, v, st->num);
2046 else
2047 get_tcp6_sock(seq, v, st->num);
2048out:
2049 return 0;
2050}
2051
2052static const struct seq_operations tcp6_seq_ops = {
2053 .show = tcp6_seq_show,
2054 .start = tcp_seq_start,
2055 .next = tcp_seq_next,
2056 .stop = tcp_seq_stop,
2057};
2058
2059static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2060 .family = AF_INET6,
2061};
2062
2063int __net_init tcp6_proc_init(struct net *net)
2064{
2065 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2066 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2067 return -ENOMEM;
2068 return 0;
2069}
2070
2071void tcp6_proc_exit(struct net *net)
2072{
2073 remove_proc_entry("tcp6", net->proc_net);
2074}
2075#endif
2076
2077struct proto tcpv6_prot = {
2078 .name = "TCPv6",
2079 .owner = THIS_MODULE,
2080 .close = tcp_close,
2081 .pre_connect = tcp_v6_pre_connect,
2082 .connect = tcp_v6_connect,
2083 .disconnect = tcp_disconnect,
2084 .accept = inet_csk_accept,
2085 .ioctl = tcp_ioctl,
2086 .init = tcp_v6_init_sock,
2087 .destroy = tcp_v6_destroy_sock,
2088 .shutdown = tcp_shutdown,
2089 .setsockopt = tcp_setsockopt,
2090 .getsockopt = tcp_getsockopt,
2091 .keepalive = tcp_set_keepalive,
2092 .recvmsg = tcp_recvmsg,
2093 .sendmsg = tcp_sendmsg,
2094 .sendpage = tcp_sendpage,
2095 .backlog_rcv = tcp_v6_do_rcv,
2096 .release_cb = tcp_release_cb,
2097 .hash = inet6_hash,
2098 .unhash = inet_unhash,
2099 .get_port = inet_csk_get_port,
2100 .enter_memory_pressure = tcp_enter_memory_pressure,
2101 .leave_memory_pressure = tcp_leave_memory_pressure,
2102 .stream_memory_free = tcp_stream_memory_free,
2103 .sockets_allocated = &tcp_sockets_allocated,
2104 .memory_allocated = &tcp_memory_allocated,
2105 .memory_pressure = &tcp_memory_pressure,
2106 .orphan_count = &tcp_orphan_count,
2107 .sysctl_mem = sysctl_tcp_mem,
2108 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2109 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2110 .max_header = MAX_TCP_HEADER,
2111 .obj_size = sizeof(struct tcp6_sock),
2112 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2113 .twsk_prot = &tcp6_timewait_sock_ops,
2114 .rsk_prot = &tcp6_request_sock_ops,
2115 .h.hashinfo = &tcp_hashinfo,
2116 .no_autobind = true,
2117 .diag_destroy = tcp_abort,
2118};
2119EXPORT_SYMBOL_GPL(tcpv6_prot);
2120
2121/* thinking of making this const? Don't.
2122 * early_demux can change based on sysctl.
2123 */
2124static struct inet6_protocol tcpv6_protocol = {
2125 .early_demux = tcp_v6_early_demux,
2126 .early_demux_handler = tcp_v6_early_demux,
2127 .handler = tcp_v6_rcv,
2128 .err_handler = tcp_v6_err,
2129 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2130};
2131
2132static struct inet_protosw tcpv6_protosw = {
2133 .type = SOCK_STREAM,
2134 .protocol = IPPROTO_TCP,
2135 .prot = &tcpv6_prot,
2136 .ops = &inet6_stream_ops,
2137 .flags = INET_PROTOSW_PERMANENT |
2138 INET_PROTOSW_ICSK,
2139};
2140
2141static int __net_init tcpv6_net_init(struct net *net)
2142{
2143 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2144 SOCK_RAW, IPPROTO_TCP, net);
2145}
2146
2147static void __net_exit tcpv6_net_exit(struct net *net)
2148{
2149 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2150}
2151
2152static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2153{
2154 inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2155}
2156
2157static struct pernet_operations tcpv6_net_ops = {
2158 .init = tcpv6_net_init,
2159 .exit = tcpv6_net_exit,
2160 .exit_batch = tcpv6_net_exit_batch,
2161};
2162
2163int __init tcpv6_init(void)
2164{
2165 int ret;
2166
2167 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2168 if (ret)
2169 goto out;
2170
2171 /* register inet6 protocol */
2172 ret = inet6_register_protosw(&tcpv6_protosw);
2173 if (ret)
2174 goto out_tcpv6_protocol;
2175
2176 ret = register_pernet_subsys(&tcpv6_net_ops);
2177 if (ret)
2178 goto out_tcpv6_protosw;
2179
2180 ret = mptcpv6_init();
2181 if (ret)
2182 goto out_tcpv6_pernet_subsys;
2183
2184out:
2185 return ret;
2186
2187out_tcpv6_pernet_subsys:
2188 unregister_pernet_subsys(&tcpv6_net_ops);
2189out_tcpv6_protosw:
2190 inet6_unregister_protosw(&tcpv6_protosw);
2191out_tcpv6_protocol:
2192 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2193 goto out;
2194}
2195
2196void tcpv6_exit(void)
2197{
2198 unregister_pernet_subsys(&tcpv6_net_ops);
2199 inet6_unregister_protosw(&tcpv6_protosw);
2200 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2201}
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/hotdata.h>
62#include <net/busy_poll.h>
63#include <net/rstreason.h>
64
65#include <linux/proc_fs.h>
66#include <linux/seq_file.h>
67
68#include <crypto/hash.h>
69#include <linux/scatterlist.h>
70
71#include <trace/events/tcp.h>
72
73static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb,
74 enum sk_rst_reason reason);
75static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
76 struct request_sock *req);
77
78INDIRECT_CALLABLE_SCOPE int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
79
80static const struct inet_connection_sock_af_ops ipv6_mapped;
81const struct inet_connection_sock_af_ops ipv6_specific;
82#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
83static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
84static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
85#endif
86
87/* Helper returning the inet6 address from a given tcp socket.
88 * It can be used in TCP stack instead of inet6_sk(sk).
89 * This avoids a dereference and allow compiler optimizations.
90 * It is a specialized version of inet6_sk_generic().
91 */
92#define tcp_inet6_sk(sk) (&container_of_const(tcp_sk(sk), \
93 struct tcp6_sock, tcp)->inet6)
94
95static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
96{
97 struct dst_entry *dst = skb_dst(skb);
98
99 if (dst && dst_hold_safe(dst)) {
100 rcu_assign_pointer(sk->sk_rx_dst, dst);
101 sk->sk_rx_dst_ifindex = skb->skb_iif;
102 sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst));
103 }
104}
105
106static u32 tcp_v6_init_seq(const struct sk_buff *skb)
107{
108 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
109 ipv6_hdr(skb)->saddr.s6_addr32,
110 tcp_hdr(skb)->dest,
111 tcp_hdr(skb)->source);
112}
113
114static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
115{
116 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
117 ipv6_hdr(skb)->saddr.s6_addr32);
118}
119
120static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
121 int addr_len)
122{
123 /* This check is replicated from tcp_v6_connect() and intended to
124 * prevent BPF program called below from accessing bytes that are out
125 * of the bound specified by user in addr_len.
126 */
127 if (addr_len < SIN6_LEN_RFC2133)
128 return -EINVAL;
129
130 sock_owned_by_me(sk);
131
132 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr, &addr_len);
133}
134
135static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
136 int addr_len)
137{
138 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
139 struct inet_connection_sock *icsk = inet_csk(sk);
140 struct in6_addr *saddr = NULL, *final_p, final;
141 struct inet_timewait_death_row *tcp_death_row;
142 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
143 struct inet_sock *inet = inet_sk(sk);
144 struct tcp_sock *tp = tcp_sk(sk);
145 struct net *net = sock_net(sk);
146 struct ipv6_txoptions *opt;
147 struct dst_entry *dst;
148 struct flowi6 fl6;
149 int addr_type;
150 int err;
151
152 if (addr_len < SIN6_LEN_RFC2133)
153 return -EINVAL;
154
155 if (usin->sin6_family != AF_INET6)
156 return -EAFNOSUPPORT;
157
158 memset(&fl6, 0, sizeof(fl6));
159
160 if (inet6_test_bit(SNDFLOW, sk)) {
161 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
162 IP6_ECN_flow_init(fl6.flowlabel);
163 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
164 struct ip6_flowlabel *flowlabel;
165 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
166 if (IS_ERR(flowlabel))
167 return -EINVAL;
168 fl6_sock_release(flowlabel);
169 }
170 }
171
172 /*
173 * connect() to INADDR_ANY means loopback (BSD'ism).
174 */
175
176 if (ipv6_addr_any(&usin->sin6_addr)) {
177 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
178 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
179 &usin->sin6_addr);
180 else
181 usin->sin6_addr = in6addr_loopback;
182 }
183
184 addr_type = ipv6_addr_type(&usin->sin6_addr);
185
186 if (addr_type & IPV6_ADDR_MULTICAST)
187 return -ENETUNREACH;
188
189 if (addr_type&IPV6_ADDR_LINKLOCAL) {
190 if (addr_len >= sizeof(struct sockaddr_in6) &&
191 usin->sin6_scope_id) {
192 /* If interface is set while binding, indices
193 * must coincide.
194 */
195 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
196 return -EINVAL;
197
198 sk->sk_bound_dev_if = usin->sin6_scope_id;
199 }
200
201 /* Connect to link-local address requires an interface */
202 if (!sk->sk_bound_dev_if)
203 return -EINVAL;
204 }
205
206 if (tp->rx_opt.ts_recent_stamp &&
207 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
208 tp->rx_opt.ts_recent = 0;
209 tp->rx_opt.ts_recent_stamp = 0;
210 WRITE_ONCE(tp->write_seq, 0);
211 }
212
213 sk->sk_v6_daddr = usin->sin6_addr;
214 np->flow_label = fl6.flowlabel;
215
216 /*
217 * TCP over IPv4
218 */
219
220 if (addr_type & IPV6_ADDR_MAPPED) {
221 u32 exthdrlen = icsk->icsk_ext_hdr_len;
222 struct sockaddr_in sin;
223
224 if (ipv6_only_sock(sk))
225 return -ENETUNREACH;
226
227 sin.sin_family = AF_INET;
228 sin.sin_port = usin->sin6_port;
229 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
230
231 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
232 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_mapped);
233 if (sk_is_mptcp(sk))
234 mptcpv6_handle_mapped(sk, true);
235 sk->sk_backlog_rcv = tcp_v4_do_rcv;
236#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
237 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
238#endif
239
240 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
241
242 if (err) {
243 icsk->icsk_ext_hdr_len = exthdrlen;
244 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
245 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_specific);
246 if (sk_is_mptcp(sk))
247 mptcpv6_handle_mapped(sk, false);
248 sk->sk_backlog_rcv = tcp_v6_do_rcv;
249#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
250 tp->af_specific = &tcp_sock_ipv6_specific;
251#endif
252 goto failure;
253 }
254 np->saddr = sk->sk_v6_rcv_saddr;
255
256 return err;
257 }
258
259 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
260 saddr = &sk->sk_v6_rcv_saddr;
261
262 fl6.flowi6_proto = IPPROTO_TCP;
263 fl6.daddr = sk->sk_v6_daddr;
264 fl6.saddr = saddr ? *saddr : np->saddr;
265 fl6.flowlabel = ip6_make_flowinfo(np->tclass, np->flow_label);
266 fl6.flowi6_oif = sk->sk_bound_dev_if;
267 fl6.flowi6_mark = sk->sk_mark;
268 fl6.fl6_dport = usin->sin6_port;
269 fl6.fl6_sport = inet->inet_sport;
270 fl6.flowi6_uid = sk->sk_uid;
271
272 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
273 final_p = fl6_update_dst(&fl6, opt, &final);
274
275 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
276
277 dst = ip6_dst_lookup_flow(net, sk, &fl6, final_p);
278 if (IS_ERR(dst)) {
279 err = PTR_ERR(dst);
280 goto failure;
281 }
282
283 tp->tcp_usec_ts = dst_tcp_usec_ts(dst);
284 tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
285
286 if (!saddr) {
287 saddr = &fl6.saddr;
288
289 err = inet_bhash2_update_saddr(sk, saddr, AF_INET6);
290 if (err)
291 goto failure;
292 }
293
294 /* set the source address */
295 np->saddr = *saddr;
296 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
297
298 sk->sk_gso_type = SKB_GSO_TCPV6;
299 ip6_dst_store(sk, dst, NULL, NULL);
300
301 icsk->icsk_ext_hdr_len = 0;
302 if (opt)
303 icsk->icsk_ext_hdr_len = opt->opt_flen +
304 opt->opt_nflen;
305
306 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
307
308 inet->inet_dport = usin->sin6_port;
309
310 tcp_set_state(sk, TCP_SYN_SENT);
311 err = inet6_hash_connect(tcp_death_row, sk);
312 if (err)
313 goto late_failure;
314
315 sk_set_txhash(sk);
316
317 if (likely(!tp->repair)) {
318 if (!tp->write_seq)
319 WRITE_ONCE(tp->write_seq,
320 secure_tcpv6_seq(np->saddr.s6_addr32,
321 sk->sk_v6_daddr.s6_addr32,
322 inet->inet_sport,
323 inet->inet_dport));
324 tp->tsoffset = secure_tcpv6_ts_off(net, np->saddr.s6_addr32,
325 sk->sk_v6_daddr.s6_addr32);
326 }
327
328 if (tcp_fastopen_defer_connect(sk, &err))
329 return err;
330 if (err)
331 goto late_failure;
332
333 err = tcp_connect(sk);
334 if (err)
335 goto late_failure;
336
337 return 0;
338
339late_failure:
340 tcp_set_state(sk, TCP_CLOSE);
341 inet_bhash2_reset_saddr(sk);
342failure:
343 inet->inet_dport = 0;
344 sk->sk_route_caps = 0;
345 return err;
346}
347
348static void tcp_v6_mtu_reduced(struct sock *sk)
349{
350 struct dst_entry *dst;
351 u32 mtu;
352
353 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
354 return;
355
356 mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
357
358 /* Drop requests trying to increase our current mss.
359 * Check done in __ip6_rt_update_pmtu() is too late.
360 */
361 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache)
362 return;
363
364 dst = inet6_csk_update_pmtu(sk, mtu);
365 if (!dst)
366 return;
367
368 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
369 tcp_sync_mss(sk, dst_mtu(dst));
370 tcp_simple_retransmit(sk);
371 }
372}
373
374static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
375 u8 type, u8 code, int offset, __be32 info)
376{
377 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
378 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
379 struct net *net = dev_net(skb->dev);
380 struct request_sock *fastopen;
381 struct ipv6_pinfo *np;
382 struct tcp_sock *tp;
383 __u32 seq, snd_una;
384 struct sock *sk;
385 bool fatal;
386 int err;
387
388 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
389 &hdr->daddr, th->dest,
390 &hdr->saddr, ntohs(th->source),
391 skb->dev->ifindex, inet6_sdif(skb));
392
393 if (!sk) {
394 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
395 ICMP6_MIB_INERRORS);
396 return -ENOENT;
397 }
398
399 if (sk->sk_state == TCP_TIME_WAIT) {
400 /* To increase the counter of ignored icmps for TCP-AO */
401 tcp_ao_ignore_icmp(sk, AF_INET6, type, code);
402 inet_twsk_put(inet_twsk(sk));
403 return 0;
404 }
405 seq = ntohl(th->seq);
406 fatal = icmpv6_err_convert(type, code, &err);
407 if (sk->sk_state == TCP_NEW_SYN_RECV) {
408 tcp_req_err(sk, seq, fatal);
409 return 0;
410 }
411
412 if (tcp_ao_ignore_icmp(sk, AF_INET6, type, code)) {
413 sock_put(sk);
414 return 0;
415 }
416
417 bh_lock_sock(sk);
418 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
419 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
420
421 if (sk->sk_state == TCP_CLOSE)
422 goto out;
423
424 if (static_branch_unlikely(&ip6_min_hopcount)) {
425 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
426 if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
427 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
428 goto out;
429 }
430 }
431
432 tp = tcp_sk(sk);
433 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
434 fastopen = rcu_dereference(tp->fastopen_rsk);
435 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
436 if (sk->sk_state != TCP_LISTEN &&
437 !between(seq, snd_una, tp->snd_nxt)) {
438 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
439 goto out;
440 }
441
442 np = tcp_inet6_sk(sk);
443
444 if (type == NDISC_REDIRECT) {
445 if (!sock_owned_by_user(sk)) {
446 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
447
448 if (dst)
449 dst->ops->redirect(dst, sk, skb);
450 }
451 goto out;
452 }
453
454 if (type == ICMPV6_PKT_TOOBIG) {
455 u32 mtu = ntohl(info);
456
457 /* We are not interested in TCP_LISTEN and open_requests
458 * (SYN-ACKs send out by Linux are always <576bytes so
459 * they should go through unfragmented).
460 */
461 if (sk->sk_state == TCP_LISTEN)
462 goto out;
463
464 if (!ip6_sk_accept_pmtu(sk))
465 goto out;
466
467 if (mtu < IPV6_MIN_MTU)
468 goto out;
469
470 WRITE_ONCE(tp->mtu_info, mtu);
471
472 if (!sock_owned_by_user(sk))
473 tcp_v6_mtu_reduced(sk);
474 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
475 &sk->sk_tsq_flags))
476 sock_hold(sk);
477 goto out;
478 }
479
480
481 /* Might be for an request_sock */
482 switch (sk->sk_state) {
483 case TCP_SYN_SENT:
484 case TCP_SYN_RECV:
485 /* Only in fast or simultaneous open. If a fast open socket is
486 * already accepted it is treated as a connected one below.
487 */
488 if (fastopen && !fastopen->sk)
489 break;
490
491 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
492
493 if (!sock_owned_by_user(sk))
494 tcp_done_with_error(sk, err);
495 else
496 WRITE_ONCE(sk->sk_err_soft, err);
497 goto out;
498 case TCP_LISTEN:
499 break;
500 default:
501 /* check if this ICMP message allows revert of backoff.
502 * (see RFC 6069)
503 */
504 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
505 code == ICMPV6_NOROUTE)
506 tcp_ld_RTO_revert(sk, seq);
507 }
508
509 if (!sock_owned_by_user(sk) && inet6_test_bit(RECVERR6, sk)) {
510 WRITE_ONCE(sk->sk_err, err);
511 sk_error_report(sk);
512 } else {
513 WRITE_ONCE(sk->sk_err_soft, err);
514 }
515out:
516 bh_unlock_sock(sk);
517 sock_put(sk);
518 return 0;
519}
520
521
522static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
523 struct flowi *fl,
524 struct request_sock *req,
525 struct tcp_fastopen_cookie *foc,
526 enum tcp_synack_type synack_type,
527 struct sk_buff *syn_skb)
528{
529 struct inet_request_sock *ireq = inet_rsk(req);
530 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
531 struct ipv6_txoptions *opt;
532 struct flowi6 *fl6 = &fl->u.ip6;
533 struct sk_buff *skb;
534 int err = -ENOMEM;
535 u8 tclass;
536
537 /* First, grab a route. */
538 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
539 IPPROTO_TCP)) == NULL)
540 goto done;
541
542 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
543
544 if (skb) {
545 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
546 &ireq->ir_v6_rmt_addr);
547
548 fl6->daddr = ireq->ir_v6_rmt_addr;
549 if (inet6_test_bit(REPFLOW, sk) && ireq->pktopts)
550 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
551
552 tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ?
553 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
554 (np->tclass & INET_ECN_MASK) :
555 np->tclass;
556
557 if (!INET_ECN_is_capable(tclass) &&
558 tcp_bpf_ca_needs_ecn((struct sock *)req))
559 tclass |= INET_ECN_ECT_0;
560
561 rcu_read_lock();
562 opt = ireq->ipv6_opt;
563 if (!opt)
564 opt = rcu_dereference(np->opt);
565 err = ip6_xmit(sk, skb, fl6, skb->mark ? : READ_ONCE(sk->sk_mark),
566 opt, tclass, READ_ONCE(sk->sk_priority));
567 rcu_read_unlock();
568 err = net_xmit_eval(err);
569 }
570
571done:
572 return err;
573}
574
575
576static void tcp_v6_reqsk_destructor(struct request_sock *req)
577{
578 kfree(inet_rsk(req)->ipv6_opt);
579 consume_skb(inet_rsk(req)->pktopts);
580}
581
582#ifdef CONFIG_TCP_MD5SIG
583static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
584 const struct in6_addr *addr,
585 int l3index)
586{
587 return tcp_md5_do_lookup(sk, l3index,
588 (union tcp_md5_addr *)addr, AF_INET6);
589}
590
591static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
592 const struct sock *addr_sk)
593{
594 int l3index;
595
596 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
597 addr_sk->sk_bound_dev_if);
598 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
599 l3index);
600}
601
602static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
603 sockptr_t optval, int optlen)
604{
605 struct tcp_md5sig cmd;
606 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
607 union tcp_ao_addr *addr;
608 int l3index = 0;
609 u8 prefixlen;
610 bool l3flag;
611 u8 flags;
612
613 if (optlen < sizeof(cmd))
614 return -EINVAL;
615
616 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
617 return -EFAULT;
618
619 if (sin6->sin6_family != AF_INET6)
620 return -EINVAL;
621
622 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
623 l3flag = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
624
625 if (optname == TCP_MD5SIG_EXT &&
626 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
627 prefixlen = cmd.tcpm_prefixlen;
628 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
629 prefixlen > 32))
630 return -EINVAL;
631 } else {
632 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
633 }
634
635 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
636 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
637 struct net_device *dev;
638
639 rcu_read_lock();
640 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
641 if (dev && netif_is_l3_master(dev))
642 l3index = dev->ifindex;
643 rcu_read_unlock();
644
645 /* ok to reference set/not set outside of rcu;
646 * right now device MUST be an L3 master
647 */
648 if (!dev || !l3index)
649 return -EINVAL;
650 }
651
652 if (!cmd.tcpm_keylen) {
653 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
654 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
655 AF_INET, prefixlen,
656 l3index, flags);
657 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
658 AF_INET6, prefixlen, l3index, flags);
659 }
660
661 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
662 return -EINVAL;
663
664 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
665 addr = (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3];
666
667 /* Don't allow keys for peers that have a matching TCP-AO key.
668 * See the comment in tcp_ao_add_cmd()
669 */
670 if (tcp_ao_required(sk, addr, AF_INET,
671 l3flag ? l3index : -1, false))
672 return -EKEYREJECTED;
673 return tcp_md5_do_add(sk, addr,
674 AF_INET, prefixlen, l3index, flags,
675 cmd.tcpm_key, cmd.tcpm_keylen);
676 }
677
678 addr = (union tcp_md5_addr *)&sin6->sin6_addr;
679
680 /* Don't allow keys for peers that have a matching TCP-AO key.
681 * See the comment in tcp_ao_add_cmd()
682 */
683 if (tcp_ao_required(sk, addr, AF_INET6, l3flag ? l3index : -1, false))
684 return -EKEYREJECTED;
685
686 return tcp_md5_do_add(sk, addr, AF_INET6, prefixlen, l3index, flags,
687 cmd.tcpm_key, cmd.tcpm_keylen);
688}
689
690static int tcp_v6_md5_hash_headers(struct tcp_sigpool *hp,
691 const struct in6_addr *daddr,
692 const struct in6_addr *saddr,
693 const struct tcphdr *th, int nbytes)
694{
695 struct tcp6_pseudohdr *bp;
696 struct scatterlist sg;
697 struct tcphdr *_th;
698
699 bp = hp->scratch;
700 /* 1. TCP pseudo-header (RFC2460) */
701 bp->saddr = *saddr;
702 bp->daddr = *daddr;
703 bp->protocol = cpu_to_be32(IPPROTO_TCP);
704 bp->len = cpu_to_be32(nbytes);
705
706 _th = (struct tcphdr *)(bp + 1);
707 memcpy(_th, th, sizeof(*th));
708 _th->check = 0;
709
710 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
711 ahash_request_set_crypt(hp->req, &sg, NULL,
712 sizeof(*bp) + sizeof(*th));
713 return crypto_ahash_update(hp->req);
714}
715
716static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
717 const struct in6_addr *daddr, struct in6_addr *saddr,
718 const struct tcphdr *th)
719{
720 struct tcp_sigpool hp;
721
722 if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
723 goto clear_hash_nostart;
724
725 if (crypto_ahash_init(hp.req))
726 goto clear_hash;
727 if (tcp_v6_md5_hash_headers(&hp, daddr, saddr, th, th->doff << 2))
728 goto clear_hash;
729 if (tcp_md5_hash_key(&hp, key))
730 goto clear_hash;
731 ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
732 if (crypto_ahash_final(hp.req))
733 goto clear_hash;
734
735 tcp_sigpool_end(&hp);
736 return 0;
737
738clear_hash:
739 tcp_sigpool_end(&hp);
740clear_hash_nostart:
741 memset(md5_hash, 0, 16);
742 return 1;
743}
744
745static int tcp_v6_md5_hash_skb(char *md5_hash,
746 const struct tcp_md5sig_key *key,
747 const struct sock *sk,
748 const struct sk_buff *skb)
749{
750 const struct tcphdr *th = tcp_hdr(skb);
751 const struct in6_addr *saddr, *daddr;
752 struct tcp_sigpool hp;
753
754 if (sk) { /* valid for establish/request sockets */
755 saddr = &sk->sk_v6_rcv_saddr;
756 daddr = &sk->sk_v6_daddr;
757 } else {
758 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
759 saddr = &ip6h->saddr;
760 daddr = &ip6h->daddr;
761 }
762
763 if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
764 goto clear_hash_nostart;
765
766 if (crypto_ahash_init(hp.req))
767 goto clear_hash;
768
769 if (tcp_v6_md5_hash_headers(&hp, daddr, saddr, th, skb->len))
770 goto clear_hash;
771 if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2))
772 goto clear_hash;
773 if (tcp_md5_hash_key(&hp, key))
774 goto clear_hash;
775 ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
776 if (crypto_ahash_final(hp.req))
777 goto clear_hash;
778
779 tcp_sigpool_end(&hp);
780 return 0;
781
782clear_hash:
783 tcp_sigpool_end(&hp);
784clear_hash_nostart:
785 memset(md5_hash, 0, 16);
786 return 1;
787}
788#endif
789
790static void tcp_v6_init_req(struct request_sock *req,
791 const struct sock *sk_listener,
792 struct sk_buff *skb,
793 u32 tw_isn)
794{
795 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
796 struct inet_request_sock *ireq = inet_rsk(req);
797 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
798
799 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
800 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
801
802 /* So that link locals have meaning */
803 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
804 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
805 ireq->ir_iif = tcp_v6_iif(skb);
806
807 if (!tw_isn &&
808 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
809 np->rxopt.bits.rxinfo ||
810 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
811 np->rxopt.bits.rxohlim || inet6_test_bit(REPFLOW, sk_listener))) {
812 refcount_inc(&skb->users);
813 ireq->pktopts = skb;
814 }
815}
816
817static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
818 struct sk_buff *skb,
819 struct flowi *fl,
820 struct request_sock *req,
821 u32 tw_isn)
822{
823 tcp_v6_init_req(req, sk, skb, tw_isn);
824
825 if (security_inet_conn_request(sk, skb, req))
826 return NULL;
827
828 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
829}
830
831struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
832 .family = AF_INET6,
833 .obj_size = sizeof(struct tcp6_request_sock),
834 .rtx_syn_ack = tcp_rtx_synack,
835 .send_ack = tcp_v6_reqsk_send_ack,
836 .destructor = tcp_v6_reqsk_destructor,
837 .send_reset = tcp_v6_send_reset,
838 .syn_ack_timeout = tcp_syn_ack_timeout,
839};
840
841const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
842 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
843 sizeof(struct ipv6hdr),
844#ifdef CONFIG_TCP_MD5SIG
845 .req_md5_lookup = tcp_v6_md5_lookup,
846 .calc_md5_hash = tcp_v6_md5_hash_skb,
847#endif
848#ifdef CONFIG_TCP_AO
849 .ao_lookup = tcp_v6_ao_lookup_rsk,
850 .ao_calc_key = tcp_v6_ao_calc_key_rsk,
851 .ao_synack_hash = tcp_v6_ao_synack_hash,
852#endif
853#ifdef CONFIG_SYN_COOKIES
854 .cookie_init_seq = cookie_v6_init_sequence,
855#endif
856 .route_req = tcp_v6_route_req,
857 .init_seq = tcp_v6_init_seq,
858 .init_ts_off = tcp_v6_init_ts_off,
859 .send_synack = tcp_v6_send_synack,
860};
861
862static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
863 u32 ack, u32 win, u32 tsval, u32 tsecr,
864 int oif, int rst, u8 tclass, __be32 label,
865 u32 priority, u32 txhash, struct tcp_key *key)
866{
867 const struct tcphdr *th = tcp_hdr(skb);
868 struct tcphdr *t1;
869 struct sk_buff *buff;
870 struct flowi6 fl6;
871 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
872 struct sock *ctl_sk = net->ipv6.tcp_sk;
873 unsigned int tot_len = sizeof(struct tcphdr);
874 __be32 mrst = 0, *topt;
875 struct dst_entry *dst;
876 __u32 mark = 0;
877
878 if (tsecr)
879 tot_len += TCPOLEN_TSTAMP_ALIGNED;
880 if (tcp_key_is_md5(key))
881 tot_len += TCPOLEN_MD5SIG_ALIGNED;
882 if (tcp_key_is_ao(key))
883 tot_len += tcp_ao_len_aligned(key->ao_key);
884
885#ifdef CONFIG_MPTCP
886 if (rst && !tcp_key_is_md5(key)) {
887 mrst = mptcp_reset_option(skb);
888
889 if (mrst)
890 tot_len += sizeof(__be32);
891 }
892#endif
893
894 buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
895 if (!buff)
896 return;
897
898 skb_reserve(buff, MAX_TCP_HEADER);
899
900 t1 = skb_push(buff, tot_len);
901 skb_reset_transport_header(buff);
902
903 /* Swap the send and the receive. */
904 memset(t1, 0, sizeof(*t1));
905 t1->dest = th->source;
906 t1->source = th->dest;
907 t1->doff = tot_len / 4;
908 t1->seq = htonl(seq);
909 t1->ack_seq = htonl(ack);
910 t1->ack = !rst || !th->ack;
911 t1->rst = rst;
912 t1->window = htons(win);
913
914 topt = (__be32 *)(t1 + 1);
915
916 if (tsecr) {
917 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
918 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
919 *topt++ = htonl(tsval);
920 *topt++ = htonl(tsecr);
921 }
922
923 if (mrst)
924 *topt++ = mrst;
925
926#ifdef CONFIG_TCP_MD5SIG
927 if (tcp_key_is_md5(key)) {
928 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
929 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
930 tcp_v6_md5_hash_hdr((__u8 *)topt, key->md5_key,
931 &ipv6_hdr(skb)->saddr,
932 &ipv6_hdr(skb)->daddr, t1);
933 }
934#endif
935#ifdef CONFIG_TCP_AO
936 if (tcp_key_is_ao(key)) {
937 *topt++ = htonl((TCPOPT_AO << 24) |
938 (tcp_ao_len(key->ao_key) << 16) |
939 (key->ao_key->sndid << 8) |
940 (key->rcv_next));
941
942 tcp_ao_hash_hdr(AF_INET6, (char *)topt, key->ao_key,
943 key->traffic_key,
944 (union tcp_ao_addr *)&ipv6_hdr(skb)->saddr,
945 (union tcp_ao_addr *)&ipv6_hdr(skb)->daddr,
946 t1, key->sne);
947 }
948#endif
949
950 memset(&fl6, 0, sizeof(fl6));
951 fl6.daddr = ipv6_hdr(skb)->saddr;
952 fl6.saddr = ipv6_hdr(skb)->daddr;
953 fl6.flowlabel = label;
954
955 buff->ip_summed = CHECKSUM_PARTIAL;
956
957 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
958
959 fl6.flowi6_proto = IPPROTO_TCP;
960 if (rt6_need_strict(&fl6.daddr) && !oif)
961 fl6.flowi6_oif = tcp_v6_iif(skb);
962 else {
963 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
964 oif = skb->skb_iif;
965
966 fl6.flowi6_oif = oif;
967 }
968
969 if (sk) {
970 /* unconstify the socket only to attach it to buff with care. */
971 skb_set_owner_edemux(buff, (struct sock *)sk);
972
973 if (sk->sk_state == TCP_TIME_WAIT)
974 mark = inet_twsk(sk)->tw_mark;
975 else
976 mark = READ_ONCE(sk->sk_mark);
977 skb_set_delivery_time(buff, tcp_transmit_time(sk), SKB_CLOCK_MONOTONIC);
978 }
979 if (txhash) {
980 /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */
981 skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4);
982 }
983 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
984 fl6.fl6_dport = t1->dest;
985 fl6.fl6_sport = t1->source;
986 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
987 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
988
989 /* Pass a socket to ip6_dst_lookup either it is for RST
990 * Underlying function will use this to retrieve the network
991 * namespace
992 */
993 if (sk && sk->sk_state != TCP_TIME_WAIT)
994 dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/
995 else
996 dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL);
997 if (!IS_ERR(dst)) {
998 skb_dst_set(buff, dst);
999 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
1000 tclass & ~INET_ECN_MASK, priority);
1001 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
1002 if (rst)
1003 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
1004 return;
1005 }
1006
1007 kfree_skb(buff);
1008}
1009
1010static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb,
1011 enum sk_rst_reason reason)
1012{
1013 const struct tcphdr *th = tcp_hdr(skb);
1014 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
1015 const __u8 *md5_hash_location = NULL;
1016#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1017 bool allocated_traffic_key = false;
1018#endif
1019 const struct tcp_ao_hdr *aoh;
1020 struct tcp_key key = {};
1021 u32 seq = 0, ack_seq = 0;
1022 __be32 label = 0;
1023 u32 priority = 0;
1024 struct net *net;
1025 u32 txhash = 0;
1026 int oif = 0;
1027#ifdef CONFIG_TCP_MD5SIG
1028 unsigned char newhash[16];
1029 int genhash;
1030 struct sock *sk1 = NULL;
1031#endif
1032
1033 if (th->rst)
1034 return;
1035
1036 /* If sk not NULL, it means we did a successful lookup and incoming
1037 * route had to be correct. prequeue might have dropped our dst.
1038 */
1039 if (!sk && !ipv6_unicast_destination(skb))
1040 return;
1041
1042 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
1043 /* Invalid TCP option size or twice included auth */
1044 if (tcp_parse_auth_options(th, &md5_hash_location, &aoh))
1045 return;
1046#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1047 rcu_read_lock();
1048#endif
1049#ifdef CONFIG_TCP_MD5SIG
1050 if (sk && sk_fullsock(sk)) {
1051 int l3index;
1052
1053 /* sdif set, means packet ingressed via a device
1054 * in an L3 domain and inet_iif is set to it.
1055 */
1056 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1057 key.md5_key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1058 if (key.md5_key)
1059 key.type = TCP_KEY_MD5;
1060 } else if (md5_hash_location) {
1061 int dif = tcp_v6_iif_l3_slave(skb);
1062 int sdif = tcp_v6_sdif(skb);
1063 int l3index;
1064
1065 /*
1066 * active side is lost. Try to find listening socket through
1067 * source port, and then find md5 key through listening socket.
1068 * we are not loose security here:
1069 * Incoming packet is checked with md5 hash with finding key,
1070 * no RST generated if md5 hash doesn't match.
1071 */
1072 sk1 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1073 NULL, 0, &ipv6h->saddr, th->source,
1074 &ipv6h->daddr, ntohs(th->source),
1075 dif, sdif);
1076 if (!sk1)
1077 goto out;
1078
1079 /* sdif set, means packet ingressed via a device
1080 * in an L3 domain and dif is set to it.
1081 */
1082 l3index = tcp_v6_sdif(skb) ? dif : 0;
1083
1084 key.md5_key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1085 if (!key.md5_key)
1086 goto out;
1087 key.type = TCP_KEY_MD5;
1088
1089 genhash = tcp_v6_md5_hash_skb(newhash, key.md5_key, NULL, skb);
1090 if (genhash || memcmp(md5_hash_location, newhash, 16) != 0)
1091 goto out;
1092 }
1093#endif
1094
1095 if (th->ack)
1096 seq = ntohl(th->ack_seq);
1097 else
1098 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1099 (th->doff << 2);
1100
1101#ifdef CONFIG_TCP_AO
1102 if (aoh) {
1103 int l3index;
1104
1105 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1106 if (tcp_ao_prepare_reset(sk, skb, aoh, l3index, seq,
1107 &key.ao_key, &key.traffic_key,
1108 &allocated_traffic_key,
1109 &key.rcv_next, &key.sne))
1110 goto out;
1111 key.type = TCP_KEY_AO;
1112 }
1113#endif
1114
1115 if (sk) {
1116 oif = sk->sk_bound_dev_if;
1117 if (sk_fullsock(sk)) {
1118 if (inet6_test_bit(REPFLOW, sk))
1119 label = ip6_flowlabel(ipv6h);
1120 priority = READ_ONCE(sk->sk_priority);
1121 txhash = sk->sk_txhash;
1122 }
1123 if (sk->sk_state == TCP_TIME_WAIT) {
1124 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1125 priority = inet_twsk(sk)->tw_priority;
1126 txhash = inet_twsk(sk)->tw_txhash;
1127 }
1128 } else {
1129 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1130 label = ip6_flowlabel(ipv6h);
1131 }
1132
1133 trace_tcp_send_reset(sk, skb, reason);
1134
1135 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, 1,
1136 ipv6_get_dsfield(ipv6h), label, priority, txhash,
1137 &key);
1138
1139#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1140out:
1141 if (allocated_traffic_key)
1142 kfree(key.traffic_key);
1143 rcu_read_unlock();
1144#endif
1145}
1146
1147static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1148 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1149 struct tcp_key *key, u8 tclass,
1150 __be32 label, u32 priority, u32 txhash)
1151{
1152 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, 0,
1153 tclass, label, priority, txhash, key);
1154}
1155
1156static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1157{
1158 struct inet_timewait_sock *tw = inet_twsk(sk);
1159 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1160 struct tcp_key key = {};
1161#ifdef CONFIG_TCP_AO
1162 struct tcp_ao_info *ao_info;
1163
1164 if (static_branch_unlikely(&tcp_ao_needed.key)) {
1165
1166 /* FIXME: the segment to-be-acked is not verified yet */
1167 ao_info = rcu_dereference(tcptw->ao_info);
1168 if (ao_info) {
1169 const struct tcp_ao_hdr *aoh;
1170
1171 /* Invalid TCP option size or twice included auth */
1172 if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1173 goto out;
1174 if (aoh)
1175 key.ao_key = tcp_ao_established_key(sk, ao_info,
1176 aoh->rnext_keyid, -1);
1177 }
1178 }
1179 if (key.ao_key) {
1180 struct tcp_ao_key *rnext_key;
1181
1182 key.traffic_key = snd_other_key(key.ao_key);
1183 /* rcv_next switches to our rcv_next */
1184 rnext_key = READ_ONCE(ao_info->rnext_key);
1185 key.rcv_next = rnext_key->rcvid;
1186 key.sne = READ_ONCE(ao_info->snd_sne);
1187 key.type = TCP_KEY_AO;
1188#else
1189 if (0) {
1190#endif
1191#ifdef CONFIG_TCP_MD5SIG
1192 } else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1193 key.md5_key = tcp_twsk_md5_key(tcptw);
1194 if (key.md5_key)
1195 key.type = TCP_KEY_MD5;
1196#endif
1197 }
1198
1199 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt,
1200 READ_ONCE(tcptw->tw_rcv_nxt),
1201 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1202 tcp_tw_tsval(tcptw),
1203 READ_ONCE(tcptw->tw_ts_recent), tw->tw_bound_dev_if,
1204 &key, tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel),
1205 tw->tw_priority, tw->tw_txhash);
1206
1207#ifdef CONFIG_TCP_AO
1208out:
1209#endif
1210 inet_twsk_put(tw);
1211}
1212
1213static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1214 struct request_sock *req)
1215{
1216 struct tcp_key key = {};
1217
1218#ifdef CONFIG_TCP_AO
1219 if (static_branch_unlikely(&tcp_ao_needed.key) &&
1220 tcp_rsk_used_ao(req)) {
1221 const struct in6_addr *addr = &ipv6_hdr(skb)->saddr;
1222 const struct tcp_ao_hdr *aoh;
1223 int l3index;
1224
1225 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1226 /* Invalid TCP option size or twice included auth */
1227 if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1228 return;
1229 if (!aoh)
1230 return;
1231 key.ao_key = tcp_ao_do_lookup(sk, l3index,
1232 (union tcp_ao_addr *)addr,
1233 AF_INET6, aoh->rnext_keyid, -1);
1234 if (unlikely(!key.ao_key)) {
1235 /* Send ACK with any matching MKT for the peer */
1236 key.ao_key = tcp_ao_do_lookup(sk, l3index,
1237 (union tcp_ao_addr *)addr,
1238 AF_INET6, -1, -1);
1239 /* Matching key disappeared (user removed the key?)
1240 * let the handshake timeout.
1241 */
1242 if (!key.ao_key) {
1243 net_info_ratelimited("TCP-AO key for (%pI6, %d)->(%pI6, %d) suddenly disappeared, won't ACK new connection\n",
1244 addr,
1245 ntohs(tcp_hdr(skb)->source),
1246 &ipv6_hdr(skb)->daddr,
1247 ntohs(tcp_hdr(skb)->dest));
1248 return;
1249 }
1250 }
1251 key.traffic_key = kmalloc(tcp_ao_digest_size(key.ao_key), GFP_ATOMIC);
1252 if (!key.traffic_key)
1253 return;
1254
1255 key.type = TCP_KEY_AO;
1256 key.rcv_next = aoh->keyid;
1257 tcp_v6_ao_calc_key_rsk(key.ao_key, key.traffic_key, req);
1258#else
1259 if (0) {
1260#endif
1261#ifdef CONFIG_TCP_MD5SIG
1262 } else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1263 int l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1264
1265 key.md5_key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr,
1266 l3index);
1267 if (key.md5_key)
1268 key.type = TCP_KEY_MD5;
1269#endif
1270 }
1271
1272 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1273 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1274 */
1275 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1276 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1277 tcp_rsk(req)->rcv_nxt,
1278 tcp_synack_window(req) >> inet_rsk(req)->rcv_wscale,
1279 tcp_rsk_tsval(tcp_rsk(req)),
1280 READ_ONCE(req->ts_recent), sk->sk_bound_dev_if,
1281 &key, ipv6_get_dsfield(ipv6_hdr(skb)), 0,
1282 READ_ONCE(sk->sk_priority),
1283 READ_ONCE(tcp_rsk(req)->txhash));
1284 if (tcp_key_is_ao(&key))
1285 kfree(key.traffic_key);
1286}
1287
1288
1289static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1290{
1291#ifdef CONFIG_SYN_COOKIES
1292 const struct tcphdr *th = tcp_hdr(skb);
1293
1294 if (!th->syn)
1295 sk = cookie_v6_check(sk, skb);
1296#endif
1297 return sk;
1298}
1299
1300u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1301 struct tcphdr *th, u32 *cookie)
1302{
1303 u16 mss = 0;
1304#ifdef CONFIG_SYN_COOKIES
1305 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1306 &tcp_request_sock_ipv6_ops, sk, th);
1307 if (mss) {
1308 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1309 tcp_synq_overflow(sk);
1310 }
1311#endif
1312 return mss;
1313}
1314
1315static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1316{
1317 if (skb->protocol == htons(ETH_P_IP))
1318 return tcp_v4_conn_request(sk, skb);
1319
1320 if (!ipv6_unicast_destination(skb))
1321 goto drop;
1322
1323 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1324 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1325 return 0;
1326 }
1327
1328 return tcp_conn_request(&tcp6_request_sock_ops,
1329 &tcp_request_sock_ipv6_ops, sk, skb);
1330
1331drop:
1332 tcp_listendrop(sk);
1333 return 0; /* don't send reset */
1334}
1335
1336static void tcp_v6_restore_cb(struct sk_buff *skb)
1337{
1338 /* We need to move header back to the beginning if xfrm6_policy_check()
1339 * and tcp_v6_fill_cb() are going to be called again.
1340 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1341 */
1342 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1343 sizeof(struct inet6_skb_parm));
1344}
1345
1346static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1347 struct request_sock *req,
1348 struct dst_entry *dst,
1349 struct request_sock *req_unhash,
1350 bool *own_req)
1351{
1352 struct inet_request_sock *ireq;
1353 struct ipv6_pinfo *newnp;
1354 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1355 struct ipv6_txoptions *opt;
1356 struct inet_sock *newinet;
1357 bool found_dup_sk = false;
1358 struct tcp_sock *newtp;
1359 struct sock *newsk;
1360#ifdef CONFIG_TCP_MD5SIG
1361 struct tcp_md5sig_key *key;
1362 int l3index;
1363#endif
1364 struct flowi6 fl6;
1365
1366 if (skb->protocol == htons(ETH_P_IP)) {
1367 /*
1368 * v6 mapped
1369 */
1370
1371 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1372 req_unhash, own_req);
1373
1374 if (!newsk)
1375 return NULL;
1376
1377 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1378
1379 newnp = tcp_inet6_sk(newsk);
1380 newtp = tcp_sk(newsk);
1381
1382 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1383
1384 newnp->saddr = newsk->sk_v6_rcv_saddr;
1385
1386 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1387 if (sk_is_mptcp(newsk))
1388 mptcpv6_handle_mapped(newsk, true);
1389 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1390#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
1391 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1392#endif
1393
1394 newnp->ipv6_mc_list = NULL;
1395 newnp->ipv6_ac_list = NULL;
1396 newnp->ipv6_fl_list = NULL;
1397 newnp->pktoptions = NULL;
1398 newnp->opt = NULL;
1399 newnp->mcast_oif = inet_iif(skb);
1400 newnp->mcast_hops = ip_hdr(skb)->ttl;
1401 newnp->rcv_flowinfo = 0;
1402 if (inet6_test_bit(REPFLOW, sk))
1403 newnp->flow_label = 0;
1404
1405 /*
1406 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1407 * here, tcp_create_openreq_child now does this for us, see the comment in
1408 * that function for the gory details. -acme
1409 */
1410
1411 /* It is tricky place. Until this moment IPv4 tcp
1412 worked with IPv6 icsk.icsk_af_ops.
1413 Sync it now.
1414 */
1415 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1416
1417 return newsk;
1418 }
1419
1420 ireq = inet_rsk(req);
1421
1422 if (sk_acceptq_is_full(sk))
1423 goto out_overflow;
1424
1425 if (!dst) {
1426 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1427 if (!dst)
1428 goto out;
1429 }
1430
1431 newsk = tcp_create_openreq_child(sk, req, skb);
1432 if (!newsk)
1433 goto out_nonewsk;
1434
1435 /*
1436 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1437 * count here, tcp_create_openreq_child now does this for us, see the
1438 * comment in that function for the gory details. -acme
1439 */
1440
1441 newsk->sk_gso_type = SKB_GSO_TCPV6;
1442 inet6_sk_rx_dst_set(newsk, skb);
1443
1444 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1445
1446 newtp = tcp_sk(newsk);
1447 newinet = inet_sk(newsk);
1448 newnp = tcp_inet6_sk(newsk);
1449
1450 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1451
1452 ip6_dst_store(newsk, dst, NULL, NULL);
1453
1454 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1455 newnp->saddr = ireq->ir_v6_loc_addr;
1456 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1457 newsk->sk_bound_dev_if = ireq->ir_iif;
1458
1459 /* Now IPv6 options...
1460
1461 First: no IPv4 options.
1462 */
1463 newinet->inet_opt = NULL;
1464 newnp->ipv6_mc_list = NULL;
1465 newnp->ipv6_ac_list = NULL;
1466 newnp->ipv6_fl_list = NULL;
1467
1468 /* Clone RX bits */
1469 newnp->rxopt.all = np->rxopt.all;
1470
1471 newnp->pktoptions = NULL;
1472 newnp->opt = NULL;
1473 newnp->mcast_oif = tcp_v6_iif(skb);
1474 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1475 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1476 if (inet6_test_bit(REPFLOW, sk))
1477 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1478
1479 /* Set ToS of the new socket based upon the value of incoming SYN.
1480 * ECT bits are set later in tcp_init_transfer().
1481 */
1482 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1483 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1484
1485 /* Clone native IPv6 options from listening socket (if any)
1486
1487 Yes, keeping reference count would be much more clever,
1488 but we make one more one thing there: reattach optmem
1489 to newsk.
1490 */
1491 opt = ireq->ipv6_opt;
1492 if (!opt)
1493 opt = rcu_dereference(np->opt);
1494 if (opt) {
1495 opt = ipv6_dup_options(newsk, opt);
1496 RCU_INIT_POINTER(newnp->opt, opt);
1497 }
1498 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1499 if (opt)
1500 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1501 opt->opt_flen;
1502
1503 tcp_ca_openreq_child(newsk, dst);
1504
1505 tcp_sync_mss(newsk, dst_mtu(dst));
1506 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1507
1508 tcp_initialize_rcv_mss(newsk);
1509
1510 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1511 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1512
1513#ifdef CONFIG_TCP_MD5SIG
1514 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1515
1516 if (!tcp_rsk_used_ao(req)) {
1517 /* Copy over the MD5 key from the original socket */
1518 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1519 if (key) {
1520 const union tcp_md5_addr *addr;
1521
1522 addr = (union tcp_md5_addr *)&newsk->sk_v6_daddr;
1523 if (tcp_md5_key_copy(newsk, addr, AF_INET6, 128, l3index, key)) {
1524 inet_csk_prepare_forced_close(newsk);
1525 tcp_done(newsk);
1526 goto out;
1527 }
1528 }
1529 }
1530#endif
1531#ifdef CONFIG_TCP_AO
1532 /* Copy over tcp_ao_info if any */
1533 if (tcp_ao_copy_all_matching(sk, newsk, req, skb, AF_INET6))
1534 goto out; /* OOM */
1535#endif
1536
1537 if (__inet_inherit_port(sk, newsk) < 0) {
1538 inet_csk_prepare_forced_close(newsk);
1539 tcp_done(newsk);
1540 goto out;
1541 }
1542 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1543 &found_dup_sk);
1544 if (*own_req) {
1545 tcp_move_syn(newtp, req);
1546
1547 /* Clone pktoptions received with SYN, if we own the req */
1548 if (ireq->pktopts) {
1549 newnp->pktoptions = skb_clone_and_charge_r(ireq->pktopts, newsk);
1550 consume_skb(ireq->pktopts);
1551 ireq->pktopts = NULL;
1552 if (newnp->pktoptions)
1553 tcp_v6_restore_cb(newnp->pktoptions);
1554 }
1555 } else {
1556 if (!req_unhash && found_dup_sk) {
1557 /* This code path should only be executed in the
1558 * syncookie case only
1559 */
1560 bh_unlock_sock(newsk);
1561 sock_put(newsk);
1562 newsk = NULL;
1563 }
1564 }
1565
1566 return newsk;
1567
1568out_overflow:
1569 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1570out_nonewsk:
1571 dst_release(dst);
1572out:
1573 tcp_listendrop(sk);
1574 return NULL;
1575}
1576
1577INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1578 u32));
1579/* The socket must have it's spinlock held when we get
1580 * here, unless it is a TCP_LISTEN socket.
1581 *
1582 * We have a potential double-lock case here, so even when
1583 * doing backlog processing we use the BH locking scheme.
1584 * This is because we cannot sleep with the original spinlock
1585 * held.
1586 */
1587INDIRECT_CALLABLE_SCOPE
1588int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1589{
1590 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1591 struct sk_buff *opt_skb = NULL;
1592 enum skb_drop_reason reason;
1593 struct tcp_sock *tp;
1594
1595 /* Imagine: socket is IPv6. IPv4 packet arrives,
1596 goes to IPv4 receive handler and backlogged.
1597 From backlog it always goes here. Kerboom...
1598 Fortunately, tcp_rcv_established and rcv_established
1599 handle them correctly, but it is not case with
1600 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1601 */
1602
1603 if (skb->protocol == htons(ETH_P_IP))
1604 return tcp_v4_do_rcv(sk, skb);
1605
1606 /*
1607 * socket locking is here for SMP purposes as backlog rcv
1608 * is currently called with bh processing disabled.
1609 */
1610
1611 /* Do Stevens' IPV6_PKTOPTIONS.
1612
1613 Yes, guys, it is the only place in our code, where we
1614 may make it not affecting IPv4.
1615 The rest of code is protocol independent,
1616 and I do not like idea to uglify IPv4.
1617
1618 Actually, all the idea behind IPV6_PKTOPTIONS
1619 looks not very well thought. For now we latch
1620 options, received in the last packet, enqueued
1621 by tcp. Feel free to propose better solution.
1622 --ANK (980728)
1623 */
1624 if (np->rxopt.all && sk->sk_state != TCP_LISTEN)
1625 opt_skb = skb_clone_and_charge_r(skb, sk);
1626
1627 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1628 struct dst_entry *dst;
1629
1630 dst = rcu_dereference_protected(sk->sk_rx_dst,
1631 lockdep_sock_is_held(sk));
1632
1633 sock_rps_save_rxhash(sk, skb);
1634 sk_mark_napi_id(sk, skb);
1635 if (dst) {
1636 if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1637 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1638 dst, sk->sk_rx_dst_cookie) == NULL) {
1639 RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1640 dst_release(dst);
1641 }
1642 }
1643
1644 tcp_rcv_established(sk, skb);
1645 if (opt_skb)
1646 goto ipv6_pktoptions;
1647 return 0;
1648 }
1649
1650 if (tcp_checksum_complete(skb))
1651 goto csum_err;
1652
1653 if (sk->sk_state == TCP_LISTEN) {
1654 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1655
1656 if (nsk != sk) {
1657 if (nsk) {
1658 reason = tcp_child_process(sk, nsk, skb);
1659 if (reason)
1660 goto reset;
1661 }
1662 return 0;
1663 }
1664 } else
1665 sock_rps_save_rxhash(sk, skb);
1666
1667 reason = tcp_rcv_state_process(sk, skb);
1668 if (reason)
1669 goto reset;
1670 if (opt_skb)
1671 goto ipv6_pktoptions;
1672 return 0;
1673
1674reset:
1675 tcp_v6_send_reset(sk, skb, sk_rst_convert_drop_reason(reason));
1676discard:
1677 if (opt_skb)
1678 __kfree_skb(opt_skb);
1679 sk_skb_reason_drop(sk, skb, reason);
1680 return 0;
1681csum_err:
1682 reason = SKB_DROP_REASON_TCP_CSUM;
1683 trace_tcp_bad_csum(skb);
1684 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1685 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1686 goto discard;
1687
1688
1689ipv6_pktoptions:
1690 /* Do you ask, what is it?
1691
1692 1. skb was enqueued by tcp.
1693 2. skb is added to tail of read queue, rather than out of order.
1694 3. socket is not in passive state.
1695 4. Finally, it really contains options, which user wants to receive.
1696 */
1697 tp = tcp_sk(sk);
1698 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1699 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1700 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1701 WRITE_ONCE(np->mcast_oif, tcp_v6_iif(opt_skb));
1702 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1703 WRITE_ONCE(np->mcast_hops,
1704 ipv6_hdr(opt_skb)->hop_limit);
1705 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1706 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1707 if (inet6_test_bit(REPFLOW, sk))
1708 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1709 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1710 tcp_v6_restore_cb(opt_skb);
1711 opt_skb = xchg(&np->pktoptions, opt_skb);
1712 } else {
1713 __kfree_skb(opt_skb);
1714 opt_skb = xchg(&np->pktoptions, NULL);
1715 }
1716 }
1717
1718 consume_skb(opt_skb);
1719 return 0;
1720}
1721
1722static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1723 const struct tcphdr *th)
1724{
1725 /* This is tricky: we move IP6CB at its correct location into
1726 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1727 * _decode_session6() uses IP6CB().
1728 * barrier() makes sure compiler won't play aliasing games.
1729 */
1730 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1731 sizeof(struct inet6_skb_parm));
1732 barrier();
1733
1734 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1735 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1736 skb->len - th->doff*4);
1737 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1738 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1739 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1740 TCP_SKB_CB(skb)->sacked = 0;
1741 TCP_SKB_CB(skb)->has_rxtstamp =
1742 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1743}
1744
1745INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1746{
1747 enum skb_drop_reason drop_reason;
1748 int sdif = inet6_sdif(skb);
1749 int dif = inet6_iif(skb);
1750 const struct tcphdr *th;
1751 const struct ipv6hdr *hdr;
1752 struct sock *sk = NULL;
1753 bool refcounted;
1754 int ret;
1755 u32 isn;
1756 struct net *net = dev_net(skb->dev);
1757
1758 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1759 if (skb->pkt_type != PACKET_HOST)
1760 goto discard_it;
1761
1762 /*
1763 * Count it even if it's bad.
1764 */
1765 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1766
1767 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1768 goto discard_it;
1769
1770 th = (const struct tcphdr *)skb->data;
1771
1772 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
1773 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1774 goto bad_packet;
1775 }
1776 if (!pskb_may_pull(skb, th->doff*4))
1777 goto discard_it;
1778
1779 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1780 goto csum_error;
1781
1782 th = (const struct tcphdr *)skb->data;
1783 hdr = ipv6_hdr(skb);
1784
1785lookup:
1786 sk = __inet6_lookup_skb(net->ipv4.tcp_death_row.hashinfo, skb, __tcp_hdrlen(th),
1787 th->source, th->dest, inet6_iif(skb), sdif,
1788 &refcounted);
1789 if (!sk)
1790 goto no_tcp_socket;
1791
1792 if (sk->sk_state == TCP_TIME_WAIT)
1793 goto do_time_wait;
1794
1795 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1796 struct request_sock *req = inet_reqsk(sk);
1797 bool req_stolen = false;
1798 struct sock *nsk;
1799
1800 sk = req->rsk_listener;
1801 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1802 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1803 else
1804 drop_reason = tcp_inbound_hash(sk, req, skb,
1805 &hdr->saddr, &hdr->daddr,
1806 AF_INET6, dif, sdif);
1807 if (drop_reason) {
1808 sk_drops_add(sk, skb);
1809 reqsk_put(req);
1810 goto discard_it;
1811 }
1812 if (tcp_checksum_complete(skb)) {
1813 reqsk_put(req);
1814 goto csum_error;
1815 }
1816 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1817 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
1818 if (!nsk) {
1819 inet_csk_reqsk_queue_drop_and_put(sk, req);
1820 goto lookup;
1821 }
1822 sk = nsk;
1823 /* reuseport_migrate_sock() has already held one sk_refcnt
1824 * before returning.
1825 */
1826 } else {
1827 sock_hold(sk);
1828 }
1829 refcounted = true;
1830 nsk = NULL;
1831 if (!tcp_filter(sk, skb)) {
1832 th = (const struct tcphdr *)skb->data;
1833 hdr = ipv6_hdr(skb);
1834 tcp_v6_fill_cb(skb, hdr, th);
1835 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1836 } else {
1837 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1838 }
1839 if (!nsk) {
1840 reqsk_put(req);
1841 if (req_stolen) {
1842 /* Another cpu got exclusive access to req
1843 * and created a full blown socket.
1844 * Try to feed this packet to this socket
1845 * instead of discarding it.
1846 */
1847 tcp_v6_restore_cb(skb);
1848 sock_put(sk);
1849 goto lookup;
1850 }
1851 goto discard_and_relse;
1852 }
1853 nf_reset_ct(skb);
1854 if (nsk == sk) {
1855 reqsk_put(req);
1856 tcp_v6_restore_cb(skb);
1857 } else {
1858 drop_reason = tcp_child_process(sk, nsk, skb);
1859 if (drop_reason) {
1860 enum sk_rst_reason rst_reason;
1861
1862 rst_reason = sk_rst_convert_drop_reason(drop_reason);
1863 tcp_v6_send_reset(nsk, skb, rst_reason);
1864 goto discard_and_relse;
1865 }
1866 sock_put(sk);
1867 return 0;
1868 }
1869 }
1870
1871process:
1872 if (static_branch_unlikely(&ip6_min_hopcount)) {
1873 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
1874 if (unlikely(hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount))) {
1875 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1876 drop_reason = SKB_DROP_REASON_TCP_MINTTL;
1877 goto discard_and_relse;
1878 }
1879 }
1880
1881 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
1882 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1883 goto discard_and_relse;
1884 }
1885
1886 drop_reason = tcp_inbound_hash(sk, NULL, skb, &hdr->saddr, &hdr->daddr,
1887 AF_INET6, dif, sdif);
1888 if (drop_reason)
1889 goto discard_and_relse;
1890
1891 nf_reset_ct(skb);
1892
1893 if (tcp_filter(sk, skb)) {
1894 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1895 goto discard_and_relse;
1896 }
1897 th = (const struct tcphdr *)skb->data;
1898 hdr = ipv6_hdr(skb);
1899 tcp_v6_fill_cb(skb, hdr, th);
1900
1901 skb->dev = NULL;
1902
1903 if (sk->sk_state == TCP_LISTEN) {
1904 ret = tcp_v6_do_rcv(sk, skb);
1905 goto put_and_return;
1906 }
1907
1908 sk_incoming_cpu_update(sk);
1909
1910 bh_lock_sock_nested(sk);
1911 tcp_segs_in(tcp_sk(sk), skb);
1912 ret = 0;
1913 if (!sock_owned_by_user(sk)) {
1914 ret = tcp_v6_do_rcv(sk, skb);
1915 } else {
1916 if (tcp_add_backlog(sk, skb, &drop_reason))
1917 goto discard_and_relse;
1918 }
1919 bh_unlock_sock(sk);
1920put_and_return:
1921 if (refcounted)
1922 sock_put(sk);
1923 return ret ? -1 : 0;
1924
1925no_tcp_socket:
1926 drop_reason = SKB_DROP_REASON_NO_SOCKET;
1927 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1928 goto discard_it;
1929
1930 tcp_v6_fill_cb(skb, hdr, th);
1931
1932 if (tcp_checksum_complete(skb)) {
1933csum_error:
1934 drop_reason = SKB_DROP_REASON_TCP_CSUM;
1935 trace_tcp_bad_csum(skb);
1936 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1937bad_packet:
1938 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1939 } else {
1940 tcp_v6_send_reset(NULL, skb, sk_rst_convert_drop_reason(drop_reason));
1941 }
1942
1943discard_it:
1944 SKB_DR_OR(drop_reason, NOT_SPECIFIED);
1945 sk_skb_reason_drop(sk, skb, drop_reason);
1946 return 0;
1947
1948discard_and_relse:
1949 sk_drops_add(sk, skb);
1950 if (refcounted)
1951 sock_put(sk);
1952 goto discard_it;
1953
1954do_time_wait:
1955 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1956 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1957 inet_twsk_put(inet_twsk(sk));
1958 goto discard_it;
1959 }
1960
1961 tcp_v6_fill_cb(skb, hdr, th);
1962
1963 if (tcp_checksum_complete(skb)) {
1964 inet_twsk_put(inet_twsk(sk));
1965 goto csum_error;
1966 }
1967
1968 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th, &isn)) {
1969 case TCP_TW_SYN:
1970 {
1971 struct sock *sk2;
1972
1973 sk2 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1974 skb, __tcp_hdrlen(th),
1975 &ipv6_hdr(skb)->saddr, th->source,
1976 &ipv6_hdr(skb)->daddr,
1977 ntohs(th->dest),
1978 tcp_v6_iif_l3_slave(skb),
1979 sdif);
1980 if (sk2) {
1981 struct inet_timewait_sock *tw = inet_twsk(sk);
1982 inet_twsk_deschedule_put(tw);
1983 sk = sk2;
1984 tcp_v6_restore_cb(skb);
1985 refcounted = false;
1986 __this_cpu_write(tcp_tw_isn, isn);
1987 goto process;
1988 }
1989 }
1990 /* to ACK */
1991 fallthrough;
1992 case TCP_TW_ACK:
1993 tcp_v6_timewait_ack(sk, skb);
1994 break;
1995 case TCP_TW_RST:
1996 tcp_v6_send_reset(sk, skb, SK_RST_REASON_TCP_TIMEWAIT_SOCKET);
1997 inet_twsk_deschedule_put(inet_twsk(sk));
1998 goto discard_it;
1999 case TCP_TW_SUCCESS:
2000 ;
2001 }
2002 goto discard_it;
2003}
2004
2005void tcp_v6_early_demux(struct sk_buff *skb)
2006{
2007 struct net *net = dev_net(skb->dev);
2008 const struct ipv6hdr *hdr;
2009 const struct tcphdr *th;
2010 struct sock *sk;
2011
2012 if (skb->pkt_type != PACKET_HOST)
2013 return;
2014
2015 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
2016 return;
2017
2018 hdr = ipv6_hdr(skb);
2019 th = tcp_hdr(skb);
2020
2021 if (th->doff < sizeof(struct tcphdr) / 4)
2022 return;
2023
2024 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
2025 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
2026 &hdr->saddr, th->source,
2027 &hdr->daddr, ntohs(th->dest),
2028 inet6_iif(skb), inet6_sdif(skb));
2029 if (sk) {
2030 skb->sk = sk;
2031 skb->destructor = sock_edemux;
2032 if (sk_fullsock(sk)) {
2033 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
2034
2035 if (dst)
2036 dst = dst_check(dst, sk->sk_rx_dst_cookie);
2037 if (dst &&
2038 sk->sk_rx_dst_ifindex == skb->skb_iif)
2039 skb_dst_set_noref(skb, dst);
2040 }
2041 }
2042}
2043
2044static struct timewait_sock_ops tcp6_timewait_sock_ops = {
2045 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
2046 .twsk_destructor = tcp_twsk_destructor,
2047};
2048
2049INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
2050{
2051 __tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr);
2052}
2053
2054const struct inet_connection_sock_af_ops ipv6_specific = {
2055 .queue_xmit = inet6_csk_xmit,
2056 .send_check = tcp_v6_send_check,
2057 .rebuild_header = inet6_sk_rebuild_header,
2058 .sk_rx_dst_set = inet6_sk_rx_dst_set,
2059 .conn_request = tcp_v6_conn_request,
2060 .syn_recv_sock = tcp_v6_syn_recv_sock,
2061 .net_header_len = sizeof(struct ipv6hdr),
2062 .setsockopt = ipv6_setsockopt,
2063 .getsockopt = ipv6_getsockopt,
2064 .addr2sockaddr = inet6_csk_addr2sockaddr,
2065 .sockaddr_len = sizeof(struct sockaddr_in6),
2066 .mtu_reduced = tcp_v6_mtu_reduced,
2067};
2068
2069#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2070static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
2071#ifdef CONFIG_TCP_MD5SIG
2072 .md5_lookup = tcp_v6_md5_lookup,
2073 .calc_md5_hash = tcp_v6_md5_hash_skb,
2074 .md5_parse = tcp_v6_parse_md5_keys,
2075#endif
2076#ifdef CONFIG_TCP_AO
2077 .ao_lookup = tcp_v6_ao_lookup,
2078 .calc_ao_hash = tcp_v6_ao_hash_skb,
2079 .ao_parse = tcp_v6_parse_ao,
2080 .ao_calc_key_sk = tcp_v6_ao_calc_key_sk,
2081#endif
2082};
2083#endif
2084
2085/*
2086 * TCP over IPv4 via INET6 API
2087 */
2088static const struct inet_connection_sock_af_ops ipv6_mapped = {
2089 .queue_xmit = ip_queue_xmit,
2090 .send_check = tcp_v4_send_check,
2091 .rebuild_header = inet_sk_rebuild_header,
2092 .sk_rx_dst_set = inet_sk_rx_dst_set,
2093 .conn_request = tcp_v6_conn_request,
2094 .syn_recv_sock = tcp_v6_syn_recv_sock,
2095 .net_header_len = sizeof(struct iphdr),
2096 .setsockopt = ipv6_setsockopt,
2097 .getsockopt = ipv6_getsockopt,
2098 .addr2sockaddr = inet6_csk_addr2sockaddr,
2099 .sockaddr_len = sizeof(struct sockaddr_in6),
2100 .mtu_reduced = tcp_v4_mtu_reduced,
2101};
2102
2103#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2104static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
2105#ifdef CONFIG_TCP_MD5SIG
2106 .md5_lookup = tcp_v4_md5_lookup,
2107 .calc_md5_hash = tcp_v4_md5_hash_skb,
2108 .md5_parse = tcp_v6_parse_md5_keys,
2109#endif
2110#ifdef CONFIG_TCP_AO
2111 .ao_lookup = tcp_v6_ao_lookup,
2112 .calc_ao_hash = tcp_v4_ao_hash_skb,
2113 .ao_parse = tcp_v6_parse_ao,
2114 .ao_calc_key_sk = tcp_v4_ao_calc_key_sk,
2115#endif
2116};
2117#endif
2118
2119/* NOTE: A lot of things set to zero explicitly by call to
2120 * sk_alloc() so need not be done here.
2121 */
2122static int tcp_v6_init_sock(struct sock *sk)
2123{
2124 struct inet_connection_sock *icsk = inet_csk(sk);
2125
2126 tcp_init_sock(sk);
2127
2128 icsk->icsk_af_ops = &ipv6_specific;
2129
2130#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2131 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
2132#endif
2133
2134 return 0;
2135}
2136
2137#ifdef CONFIG_PROC_FS
2138/* Proc filesystem TCPv6 sock list dumping. */
2139static void get_openreq6(struct seq_file *seq,
2140 const struct request_sock *req, int i)
2141{
2142 long ttd = req->rsk_timer.expires - jiffies;
2143 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
2144 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
2145
2146 if (ttd < 0)
2147 ttd = 0;
2148
2149 seq_printf(seq,
2150 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2151 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
2152 i,
2153 src->s6_addr32[0], src->s6_addr32[1],
2154 src->s6_addr32[2], src->s6_addr32[3],
2155 inet_rsk(req)->ir_num,
2156 dest->s6_addr32[0], dest->s6_addr32[1],
2157 dest->s6_addr32[2], dest->s6_addr32[3],
2158 ntohs(inet_rsk(req)->ir_rmt_port),
2159 TCP_SYN_RECV,
2160 0, 0, /* could print option size, but that is af dependent. */
2161 1, /* timers active (only the expire timer) */
2162 jiffies_to_clock_t(ttd),
2163 req->num_timeout,
2164 from_kuid_munged(seq_user_ns(seq),
2165 sock_i_uid(req->rsk_listener)),
2166 0, /* non standard timer */
2167 0, /* open_requests have no inode */
2168 0, req);
2169}
2170
2171static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2172{
2173 const struct in6_addr *dest, *src;
2174 __u16 destp, srcp;
2175 int timer_active;
2176 unsigned long timer_expires;
2177 const struct inet_sock *inet = inet_sk(sp);
2178 const struct tcp_sock *tp = tcp_sk(sp);
2179 const struct inet_connection_sock *icsk = inet_csk(sp);
2180 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2181 u8 icsk_pending;
2182 int rx_queue;
2183 int state;
2184
2185 dest = &sp->sk_v6_daddr;
2186 src = &sp->sk_v6_rcv_saddr;
2187 destp = ntohs(inet->inet_dport);
2188 srcp = ntohs(inet->inet_sport);
2189
2190 icsk_pending = smp_load_acquire(&icsk->icsk_pending);
2191 if (icsk_pending == ICSK_TIME_RETRANS ||
2192 icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2193 icsk_pending == ICSK_TIME_LOSS_PROBE) {
2194 timer_active = 1;
2195 timer_expires = icsk->icsk_timeout;
2196 } else if (icsk_pending == ICSK_TIME_PROBE0) {
2197 timer_active = 4;
2198 timer_expires = icsk->icsk_timeout;
2199 } else if (timer_pending(&sp->sk_timer)) {
2200 timer_active = 2;
2201 timer_expires = sp->sk_timer.expires;
2202 } else {
2203 timer_active = 0;
2204 timer_expires = jiffies;
2205 }
2206
2207 state = inet_sk_state_load(sp);
2208 if (state == TCP_LISTEN)
2209 rx_queue = READ_ONCE(sp->sk_ack_backlog);
2210 else
2211 /* Because we don't lock the socket,
2212 * we might find a transient negative value.
2213 */
2214 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2215 READ_ONCE(tp->copied_seq), 0);
2216
2217 seq_printf(seq,
2218 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2219 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2220 i,
2221 src->s6_addr32[0], src->s6_addr32[1],
2222 src->s6_addr32[2], src->s6_addr32[3], srcp,
2223 dest->s6_addr32[0], dest->s6_addr32[1],
2224 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2225 state,
2226 READ_ONCE(tp->write_seq) - tp->snd_una,
2227 rx_queue,
2228 timer_active,
2229 jiffies_delta_to_clock_t(timer_expires - jiffies),
2230 icsk->icsk_retransmits,
2231 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
2232 icsk->icsk_probes_out,
2233 sock_i_ino(sp),
2234 refcount_read(&sp->sk_refcnt), sp,
2235 jiffies_to_clock_t(icsk->icsk_rto),
2236 jiffies_to_clock_t(icsk->icsk_ack.ato),
2237 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2238 tcp_snd_cwnd(tp),
2239 state == TCP_LISTEN ?
2240 fastopenq->max_qlen :
2241 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2242 );
2243}
2244
2245static void get_timewait6_sock(struct seq_file *seq,
2246 struct inet_timewait_sock *tw, int i)
2247{
2248 long delta = tw->tw_timer.expires - jiffies;
2249 const struct in6_addr *dest, *src;
2250 __u16 destp, srcp;
2251
2252 dest = &tw->tw_v6_daddr;
2253 src = &tw->tw_v6_rcv_saddr;
2254 destp = ntohs(tw->tw_dport);
2255 srcp = ntohs(tw->tw_sport);
2256
2257 seq_printf(seq,
2258 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2259 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2260 i,
2261 src->s6_addr32[0], src->s6_addr32[1],
2262 src->s6_addr32[2], src->s6_addr32[3], srcp,
2263 dest->s6_addr32[0], dest->s6_addr32[1],
2264 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2265 READ_ONCE(tw->tw_substate), 0, 0,
2266 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2267 refcount_read(&tw->tw_refcnt), tw);
2268}
2269
2270static int tcp6_seq_show(struct seq_file *seq, void *v)
2271{
2272 struct tcp_iter_state *st;
2273 struct sock *sk = v;
2274
2275 if (v == SEQ_START_TOKEN) {
2276 seq_puts(seq,
2277 " sl "
2278 "local_address "
2279 "remote_address "
2280 "st tx_queue rx_queue tr tm->when retrnsmt"
2281 " uid timeout inode\n");
2282 goto out;
2283 }
2284 st = seq->private;
2285
2286 if (sk->sk_state == TCP_TIME_WAIT)
2287 get_timewait6_sock(seq, v, st->num);
2288 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2289 get_openreq6(seq, v, st->num);
2290 else
2291 get_tcp6_sock(seq, v, st->num);
2292out:
2293 return 0;
2294}
2295
2296static const struct seq_operations tcp6_seq_ops = {
2297 .show = tcp6_seq_show,
2298 .start = tcp_seq_start,
2299 .next = tcp_seq_next,
2300 .stop = tcp_seq_stop,
2301};
2302
2303static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2304 .family = AF_INET6,
2305};
2306
2307int __net_init tcp6_proc_init(struct net *net)
2308{
2309 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2310 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2311 return -ENOMEM;
2312 return 0;
2313}
2314
2315void tcp6_proc_exit(struct net *net)
2316{
2317 remove_proc_entry("tcp6", net->proc_net);
2318}
2319#endif
2320
2321struct proto tcpv6_prot = {
2322 .name = "TCPv6",
2323 .owner = THIS_MODULE,
2324 .close = tcp_close,
2325 .pre_connect = tcp_v6_pre_connect,
2326 .connect = tcp_v6_connect,
2327 .disconnect = tcp_disconnect,
2328 .accept = inet_csk_accept,
2329 .ioctl = tcp_ioctl,
2330 .init = tcp_v6_init_sock,
2331 .destroy = tcp_v4_destroy_sock,
2332 .shutdown = tcp_shutdown,
2333 .setsockopt = tcp_setsockopt,
2334 .getsockopt = tcp_getsockopt,
2335 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
2336 .keepalive = tcp_set_keepalive,
2337 .recvmsg = tcp_recvmsg,
2338 .sendmsg = tcp_sendmsg,
2339 .splice_eof = tcp_splice_eof,
2340 .backlog_rcv = tcp_v6_do_rcv,
2341 .release_cb = tcp_release_cb,
2342 .hash = inet6_hash,
2343 .unhash = inet_unhash,
2344 .get_port = inet_csk_get_port,
2345 .put_port = inet_put_port,
2346#ifdef CONFIG_BPF_SYSCALL
2347 .psock_update_sk_prot = tcp_bpf_update_proto,
2348#endif
2349 .enter_memory_pressure = tcp_enter_memory_pressure,
2350 .leave_memory_pressure = tcp_leave_memory_pressure,
2351 .stream_memory_free = tcp_stream_memory_free,
2352 .sockets_allocated = &tcp_sockets_allocated,
2353
2354 .memory_allocated = &tcp_memory_allocated,
2355 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
2356
2357 .memory_pressure = &tcp_memory_pressure,
2358 .orphan_count = &tcp_orphan_count,
2359 .sysctl_mem = sysctl_tcp_mem,
2360 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2361 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2362 .max_header = MAX_TCP_HEADER,
2363 .obj_size = sizeof(struct tcp6_sock),
2364 .ipv6_pinfo_offset = offsetof(struct tcp6_sock, inet6),
2365 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2366 .twsk_prot = &tcp6_timewait_sock_ops,
2367 .rsk_prot = &tcp6_request_sock_ops,
2368 .h.hashinfo = NULL,
2369 .no_autobind = true,
2370 .diag_destroy = tcp_abort,
2371};
2372EXPORT_SYMBOL_GPL(tcpv6_prot);
2373
2374
2375static struct inet_protosw tcpv6_protosw = {
2376 .type = SOCK_STREAM,
2377 .protocol = IPPROTO_TCP,
2378 .prot = &tcpv6_prot,
2379 .ops = &inet6_stream_ops,
2380 .flags = INET_PROTOSW_PERMANENT |
2381 INET_PROTOSW_ICSK,
2382};
2383
2384static int __net_init tcpv6_net_init(struct net *net)
2385{
2386 int res;
2387
2388 res = inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2389 SOCK_RAW, IPPROTO_TCP, net);
2390 if (!res)
2391 net->ipv6.tcp_sk->sk_clockid = CLOCK_MONOTONIC;
2392
2393 return res;
2394}
2395
2396static void __net_exit tcpv6_net_exit(struct net *net)
2397{
2398 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2399}
2400
2401static struct pernet_operations tcpv6_net_ops = {
2402 .init = tcpv6_net_init,
2403 .exit = tcpv6_net_exit,
2404};
2405
2406int __init tcpv6_init(void)
2407{
2408 int ret;
2409
2410 net_hotdata.tcpv6_protocol = (struct inet6_protocol) {
2411 .handler = tcp_v6_rcv,
2412 .err_handler = tcp_v6_err,
2413 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
2414 };
2415 ret = inet6_add_protocol(&net_hotdata.tcpv6_protocol, IPPROTO_TCP);
2416 if (ret)
2417 goto out;
2418
2419 /* register inet6 protocol */
2420 ret = inet6_register_protosw(&tcpv6_protosw);
2421 if (ret)
2422 goto out_tcpv6_protocol;
2423
2424 ret = register_pernet_subsys(&tcpv6_net_ops);
2425 if (ret)
2426 goto out_tcpv6_protosw;
2427
2428 ret = mptcpv6_init();
2429 if (ret)
2430 goto out_tcpv6_pernet_subsys;
2431
2432out:
2433 return ret;
2434
2435out_tcpv6_pernet_subsys:
2436 unregister_pernet_subsys(&tcpv6_net_ops);
2437out_tcpv6_protosw:
2438 inet6_unregister_protosw(&tcpv6_protosw);
2439out_tcpv6_protocol:
2440 inet6_del_protocol(&net_hotdata.tcpv6_protocol, IPPROTO_TCP);
2441 goto out;
2442}
2443
2444void tcpv6_exit(void)
2445{
2446 unregister_pernet_subsys(&tcpv6_net_ops);
2447 inet6_unregister_protosw(&tcpv6_protosw);
2448 inet6_del_protocol(&net_hotdata.tcpv6_protocol, IPPROTO_TCP);
2449}