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/*
2 * TCP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on:
9 * linux/net/ipv4/tcp.c
10 * linux/net/ipv4/tcp_input.c
11 * linux/net/ipv4/tcp_output.c
12 *
13 * Fixes:
14 * Hideaki YOSHIFUJI : sin6_scope_id support
15 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
16 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
17 * a single port at the same time.
18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
19 *
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
24 */
25
26#include <linux/bottom_half.h>
27#include <linux/module.h>
28#include <linux/errno.h>
29#include <linux/types.h>
30#include <linux/socket.h>
31#include <linux/sockios.h>
32#include <linux/net.h>
33#include <linux/jiffies.h>
34#include <linux/in.h>
35#include <linux/in6.h>
36#include <linux/netdevice.h>
37#include <linux/init.h>
38#include <linux/jhash.h>
39#include <linux/ipsec.h>
40#include <linux/times.h>
41#include <linux/slab.h>
42#include <linux/uaccess.h>
43#include <linux/ipv6.h>
44#include <linux/icmpv6.h>
45#include <linux/random.h>
46
47#include <net/tcp.h>
48#include <net/ndisc.h>
49#include <net/inet6_hashtables.h>
50#include <net/inet6_connection_sock.h>
51#include <net/ipv6.h>
52#include <net/transp_v6.h>
53#include <net/addrconf.h>
54#include <net/ip6_route.h>
55#include <net/ip6_checksum.h>
56#include <net/inet_ecn.h>
57#include <net/protocol.h>
58#include <net/xfrm.h>
59#include <net/snmp.h>
60#include <net/dsfield.h>
61#include <net/timewait_sock.h>
62#include <net/inet_common.h>
63#include <net/secure_seq.h>
64#include <net/busy_poll.h>
65
66#include <linux/proc_fs.h>
67#include <linux/seq_file.h>
68
69#include <crypto/hash.h>
70#include <linux/scatterlist.h>
71
72#include <trace/events/tcp.h>
73
74static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
75static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
76 struct request_sock *req);
77
78static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
79
80static const struct inet_connection_sock_af_ops ipv6_mapped;
81static const struct inet_connection_sock_af_ops ipv6_specific;
82#ifdef CONFIG_TCP_MD5SIG
83static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
84static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
85#else
86static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
87 const struct in6_addr *addr)
88{
89 return NULL;
90}
91#endif
92
93static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
94{
95 struct dst_entry *dst = skb_dst(skb);
96
97 if (dst && dst_hold_safe(dst)) {
98 const struct rt6_info *rt = (const struct rt6_info *)dst;
99
100 sk->sk_rx_dst = dst;
101 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
102 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
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);
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_sock *inet = inet_sk(sk);
140 struct inet_connection_sock *icsk = inet_csk(sk);
141 struct ipv6_pinfo *np = inet6_sk(sk);
142 struct tcp_sock *tp = tcp_sk(sk);
143 struct in6_addr *saddr = NULL, *final_p, final;
144 struct ipv6_txoptions *opt;
145 struct flowi6 fl6;
146 struct dst_entry *dst;
147 int addr_type;
148 int err;
149 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
150
151 if (addr_len < SIN6_LEN_RFC2133)
152 return -EINVAL;
153
154 if (usin->sin6_family != AF_INET6)
155 return -EAFNOSUPPORT;
156
157 memset(&fl6, 0, sizeof(fl6));
158
159 if (np->sndflow) {
160 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
161 IP6_ECN_flow_init(fl6.flowlabel);
162 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
163 struct ip6_flowlabel *flowlabel;
164 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
165 if (!flowlabel)
166 return -EINVAL;
167 fl6_sock_release(flowlabel);
168 }
169 }
170
171 /*
172 * connect() to INADDR_ANY means loopback (BSD'ism).
173 */
174
175 if (ipv6_addr_any(&usin->sin6_addr)) {
176 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
177 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
178 &usin->sin6_addr);
179 else
180 usin->sin6_addr = in6addr_loopback;
181 }
182
183 addr_type = ipv6_addr_type(&usin->sin6_addr);
184
185 if (addr_type & IPV6_ADDR_MULTICAST)
186 return -ENETUNREACH;
187
188 if (addr_type&IPV6_ADDR_LINKLOCAL) {
189 if (addr_len >= sizeof(struct sockaddr_in6) &&
190 usin->sin6_scope_id) {
191 /* If interface is set while binding, indices
192 * must coincide.
193 */
194 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
195 return -EINVAL;
196
197 sk->sk_bound_dev_if = usin->sin6_scope_id;
198 }
199
200 /* Connect to link-local address requires an interface */
201 if (!sk->sk_bound_dev_if)
202 return -EINVAL;
203 }
204
205 if (tp->rx_opt.ts_recent_stamp &&
206 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
207 tp->rx_opt.ts_recent = 0;
208 tp->rx_opt.ts_recent_stamp = 0;
209 tp->write_seq = 0;
210 }
211
212 sk->sk_v6_daddr = usin->sin6_addr;
213 np->flow_label = fl6.flowlabel;
214
215 /*
216 * TCP over IPv4
217 */
218
219 if (addr_type & IPV6_ADDR_MAPPED) {
220 u32 exthdrlen = icsk->icsk_ext_hdr_len;
221 struct sockaddr_in sin;
222
223 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
224
225 if (__ipv6_only_sock(sk))
226 return -ENETUNREACH;
227
228 sin.sin_family = AF_INET;
229 sin.sin_port = usin->sin6_port;
230 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
231
232 icsk->icsk_af_ops = &ipv6_mapped;
233 sk->sk_backlog_rcv = tcp_v4_do_rcv;
234#ifdef CONFIG_TCP_MD5SIG
235 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
236#endif
237
238 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
239
240 if (err) {
241 icsk->icsk_ext_hdr_len = exthdrlen;
242 icsk->icsk_af_ops = &ipv6_specific;
243 sk->sk_backlog_rcv = tcp_v6_do_rcv;
244#ifdef CONFIG_TCP_MD5SIG
245 tp->af_specific = &tcp_sock_ipv6_specific;
246#endif
247 goto failure;
248 }
249 np->saddr = sk->sk_v6_rcv_saddr;
250
251 return err;
252 }
253
254 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
255 saddr = &sk->sk_v6_rcv_saddr;
256
257 fl6.flowi6_proto = IPPROTO_TCP;
258 fl6.daddr = sk->sk_v6_daddr;
259 fl6.saddr = saddr ? *saddr : np->saddr;
260 fl6.flowi6_oif = sk->sk_bound_dev_if;
261 fl6.flowi6_mark = sk->sk_mark;
262 fl6.fl6_dport = usin->sin6_port;
263 fl6.fl6_sport = inet->inet_sport;
264 fl6.flowi6_uid = sk->sk_uid;
265
266 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
267 final_p = fl6_update_dst(&fl6, opt, &final);
268
269 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
270
271 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
272 if (IS_ERR(dst)) {
273 err = PTR_ERR(dst);
274 goto failure;
275 }
276
277 if (!saddr) {
278 saddr = &fl6.saddr;
279 sk->sk_v6_rcv_saddr = *saddr;
280 }
281
282 /* set the source address */
283 np->saddr = *saddr;
284 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
285
286 sk->sk_gso_type = SKB_GSO_TCPV6;
287 ip6_dst_store(sk, dst, NULL, NULL);
288
289 icsk->icsk_ext_hdr_len = 0;
290 if (opt)
291 icsk->icsk_ext_hdr_len = opt->opt_flen +
292 opt->opt_nflen;
293
294 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
295
296 inet->inet_dport = usin->sin6_port;
297
298 tcp_set_state(sk, TCP_SYN_SENT);
299 err = inet6_hash_connect(tcp_death_row, sk);
300 if (err)
301 goto late_failure;
302
303 sk_set_txhash(sk);
304
305 if (likely(!tp->repair)) {
306 if (!tp->write_seq)
307 tp->write_seq = secure_tcpv6_seq(np->saddr.s6_addr32,
308 sk->sk_v6_daddr.s6_addr32,
309 inet->inet_sport,
310 inet->inet_dport);
311 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
312 np->saddr.s6_addr32,
313 sk->sk_v6_daddr.s6_addr32);
314 }
315
316 if (tcp_fastopen_defer_connect(sk, &err))
317 return err;
318 if (err)
319 goto late_failure;
320
321 err = tcp_connect(sk);
322 if (err)
323 goto late_failure;
324
325 return 0;
326
327late_failure:
328 tcp_set_state(sk, TCP_CLOSE);
329failure:
330 inet->inet_dport = 0;
331 sk->sk_route_caps = 0;
332 return err;
333}
334
335static void tcp_v6_mtu_reduced(struct sock *sk)
336{
337 struct dst_entry *dst;
338
339 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
340 return;
341
342 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
343 if (!dst)
344 return;
345
346 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
347 tcp_sync_mss(sk, dst_mtu(dst));
348 tcp_simple_retransmit(sk);
349 }
350}
351
352static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
353 u8 type, u8 code, int offset, __be32 info)
354{
355 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
356 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
357 struct net *net = dev_net(skb->dev);
358 struct request_sock *fastopen;
359 struct ipv6_pinfo *np;
360 struct tcp_sock *tp;
361 __u32 seq, snd_una;
362 struct sock *sk;
363 bool fatal;
364 int err;
365
366 sk = __inet6_lookup_established(net, &tcp_hashinfo,
367 &hdr->daddr, th->dest,
368 &hdr->saddr, ntohs(th->source),
369 skb->dev->ifindex, inet6_sdif(skb));
370
371 if (!sk) {
372 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
373 ICMP6_MIB_INERRORS);
374 return;
375 }
376
377 if (sk->sk_state == TCP_TIME_WAIT) {
378 inet_twsk_put(inet_twsk(sk));
379 return;
380 }
381 seq = ntohl(th->seq);
382 fatal = icmpv6_err_convert(type, code, &err);
383 if (sk->sk_state == TCP_NEW_SYN_RECV)
384 return tcp_req_err(sk, seq, fatal);
385
386 bh_lock_sock(sk);
387 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
388 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
389
390 if (sk->sk_state == TCP_CLOSE)
391 goto out;
392
393 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
394 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
395 goto out;
396 }
397
398 tp = tcp_sk(sk);
399 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
400 fastopen = tp->fastopen_rsk;
401 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
402 if (sk->sk_state != TCP_LISTEN &&
403 !between(seq, snd_una, tp->snd_nxt)) {
404 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
405 goto out;
406 }
407
408 np = inet6_sk(sk);
409
410 if (type == NDISC_REDIRECT) {
411 if (!sock_owned_by_user(sk)) {
412 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
413
414 if (dst)
415 dst->ops->redirect(dst, sk, skb);
416 }
417 goto out;
418 }
419
420 if (type == ICMPV6_PKT_TOOBIG) {
421 /* We are not interested in TCP_LISTEN and open_requests
422 * (SYN-ACKs send out by Linux are always <576bytes so
423 * they should go through unfragmented).
424 */
425 if (sk->sk_state == TCP_LISTEN)
426 goto out;
427
428 if (!ip6_sk_accept_pmtu(sk))
429 goto out;
430
431 tp->mtu_info = ntohl(info);
432 if (!sock_owned_by_user(sk))
433 tcp_v6_mtu_reduced(sk);
434 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
435 &sk->sk_tsq_flags))
436 sock_hold(sk);
437 goto out;
438 }
439
440
441 /* Might be for an request_sock */
442 switch (sk->sk_state) {
443 case TCP_SYN_SENT:
444 case TCP_SYN_RECV:
445 /* Only in fast or simultaneous open. If a fast open socket is
446 * is already accepted it is treated as a connected one below.
447 */
448 if (fastopen && !fastopen->sk)
449 break;
450
451 if (!sock_owned_by_user(sk)) {
452 sk->sk_err = err;
453 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
454
455 tcp_done(sk);
456 } else
457 sk->sk_err_soft = err;
458 goto out;
459 }
460
461 if (!sock_owned_by_user(sk) && np->recverr) {
462 sk->sk_err = err;
463 sk->sk_error_report(sk);
464 } else
465 sk->sk_err_soft = err;
466
467out:
468 bh_unlock_sock(sk);
469 sock_put(sk);
470}
471
472
473static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
474 struct flowi *fl,
475 struct request_sock *req,
476 struct tcp_fastopen_cookie *foc,
477 enum tcp_synack_type synack_type)
478{
479 struct inet_request_sock *ireq = inet_rsk(req);
480 struct ipv6_pinfo *np = inet6_sk(sk);
481 struct ipv6_txoptions *opt;
482 struct flowi6 *fl6 = &fl->u.ip6;
483 struct sk_buff *skb;
484 int err = -ENOMEM;
485
486 /* First, grab a route. */
487 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
488 IPPROTO_TCP)) == NULL)
489 goto done;
490
491 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
492
493 if (skb) {
494 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
495 &ireq->ir_v6_rmt_addr);
496
497 fl6->daddr = ireq->ir_v6_rmt_addr;
498 if (np->repflow && ireq->pktopts)
499 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
500
501 rcu_read_lock();
502 opt = ireq->ipv6_opt;
503 if (!opt)
504 opt = rcu_dereference(np->opt);
505 err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass);
506 rcu_read_unlock();
507 err = net_xmit_eval(err);
508 }
509
510done:
511 return err;
512}
513
514
515static void tcp_v6_reqsk_destructor(struct request_sock *req)
516{
517 kfree(inet_rsk(req)->ipv6_opt);
518 kfree_skb(inet_rsk(req)->pktopts);
519}
520
521#ifdef CONFIG_TCP_MD5SIG
522static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
523 const struct in6_addr *addr)
524{
525 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
526}
527
528static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
529 const struct sock *addr_sk)
530{
531 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
532}
533
534static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
535 char __user *optval, int optlen)
536{
537 struct tcp_md5sig cmd;
538 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
539 u8 prefixlen;
540
541 if (optlen < sizeof(cmd))
542 return -EINVAL;
543
544 if (copy_from_user(&cmd, optval, sizeof(cmd)))
545 return -EFAULT;
546
547 if (sin6->sin6_family != AF_INET6)
548 return -EINVAL;
549
550 if (optname == TCP_MD5SIG_EXT &&
551 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
552 prefixlen = cmd.tcpm_prefixlen;
553 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
554 prefixlen > 32))
555 return -EINVAL;
556 } else {
557 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
558 }
559
560 if (!cmd.tcpm_keylen) {
561 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
562 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
563 AF_INET, prefixlen);
564 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
565 AF_INET6, prefixlen);
566 }
567
568 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
569 return -EINVAL;
570
571 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
572 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
573 AF_INET, prefixlen, cmd.tcpm_key,
574 cmd.tcpm_keylen, GFP_KERNEL);
575
576 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
577 AF_INET6, prefixlen, cmd.tcpm_key,
578 cmd.tcpm_keylen, GFP_KERNEL);
579}
580
581static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
582 const struct in6_addr *daddr,
583 const struct in6_addr *saddr,
584 const struct tcphdr *th, int nbytes)
585{
586 struct tcp6_pseudohdr *bp;
587 struct scatterlist sg;
588 struct tcphdr *_th;
589
590 bp = hp->scratch;
591 /* 1. TCP pseudo-header (RFC2460) */
592 bp->saddr = *saddr;
593 bp->daddr = *daddr;
594 bp->protocol = cpu_to_be32(IPPROTO_TCP);
595 bp->len = cpu_to_be32(nbytes);
596
597 _th = (struct tcphdr *)(bp + 1);
598 memcpy(_th, th, sizeof(*th));
599 _th->check = 0;
600
601 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
602 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
603 sizeof(*bp) + sizeof(*th));
604 return crypto_ahash_update(hp->md5_req);
605}
606
607static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
608 const struct in6_addr *daddr, struct in6_addr *saddr,
609 const struct tcphdr *th)
610{
611 struct tcp_md5sig_pool *hp;
612 struct ahash_request *req;
613
614 hp = tcp_get_md5sig_pool();
615 if (!hp)
616 goto clear_hash_noput;
617 req = hp->md5_req;
618
619 if (crypto_ahash_init(req))
620 goto clear_hash;
621 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
622 goto clear_hash;
623 if (tcp_md5_hash_key(hp, key))
624 goto clear_hash;
625 ahash_request_set_crypt(req, NULL, md5_hash, 0);
626 if (crypto_ahash_final(req))
627 goto clear_hash;
628
629 tcp_put_md5sig_pool();
630 return 0;
631
632clear_hash:
633 tcp_put_md5sig_pool();
634clear_hash_noput:
635 memset(md5_hash, 0, 16);
636 return 1;
637}
638
639static int tcp_v6_md5_hash_skb(char *md5_hash,
640 const struct tcp_md5sig_key *key,
641 const struct sock *sk,
642 const struct sk_buff *skb)
643{
644 const struct in6_addr *saddr, *daddr;
645 struct tcp_md5sig_pool *hp;
646 struct ahash_request *req;
647 const struct tcphdr *th = tcp_hdr(skb);
648
649 if (sk) { /* valid for establish/request sockets */
650 saddr = &sk->sk_v6_rcv_saddr;
651 daddr = &sk->sk_v6_daddr;
652 } else {
653 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
654 saddr = &ip6h->saddr;
655 daddr = &ip6h->daddr;
656 }
657
658 hp = tcp_get_md5sig_pool();
659 if (!hp)
660 goto clear_hash_noput;
661 req = hp->md5_req;
662
663 if (crypto_ahash_init(req))
664 goto clear_hash;
665
666 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
667 goto clear_hash;
668 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
669 goto clear_hash;
670 if (tcp_md5_hash_key(hp, key))
671 goto clear_hash;
672 ahash_request_set_crypt(req, NULL, md5_hash, 0);
673 if (crypto_ahash_final(req))
674 goto clear_hash;
675
676 tcp_put_md5sig_pool();
677 return 0;
678
679clear_hash:
680 tcp_put_md5sig_pool();
681clear_hash_noput:
682 memset(md5_hash, 0, 16);
683 return 1;
684}
685
686#endif
687
688static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
689 const struct sk_buff *skb)
690{
691#ifdef CONFIG_TCP_MD5SIG
692 const __u8 *hash_location = NULL;
693 struct tcp_md5sig_key *hash_expected;
694 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
695 const struct tcphdr *th = tcp_hdr(skb);
696 int genhash;
697 u8 newhash[16];
698
699 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
700 hash_location = tcp_parse_md5sig_option(th);
701
702 /* We've parsed the options - do we have a hash? */
703 if (!hash_expected && !hash_location)
704 return false;
705
706 if (hash_expected && !hash_location) {
707 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
708 return true;
709 }
710
711 if (!hash_expected && hash_location) {
712 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
713 return true;
714 }
715
716 /* check the signature */
717 genhash = tcp_v6_md5_hash_skb(newhash,
718 hash_expected,
719 NULL, skb);
720
721 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
722 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
723 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
724 genhash ? "failed" : "mismatch",
725 &ip6h->saddr, ntohs(th->source),
726 &ip6h->daddr, ntohs(th->dest));
727 return true;
728 }
729#endif
730 return false;
731}
732
733static void tcp_v6_init_req(struct request_sock *req,
734 const struct sock *sk_listener,
735 struct sk_buff *skb)
736{
737 struct inet_request_sock *ireq = inet_rsk(req);
738 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
739
740 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
741 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
742
743 /* So that link locals have meaning */
744 if (!sk_listener->sk_bound_dev_if &&
745 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
746 ireq->ir_iif = tcp_v6_iif(skb);
747
748 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
749 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
750 np->rxopt.bits.rxinfo ||
751 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
752 np->rxopt.bits.rxohlim || np->repflow)) {
753 refcount_inc(&skb->users);
754 ireq->pktopts = skb;
755 }
756}
757
758static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
759 struct flowi *fl,
760 const struct request_sock *req)
761{
762 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
763}
764
765struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
766 .family = AF_INET6,
767 .obj_size = sizeof(struct tcp6_request_sock),
768 .rtx_syn_ack = tcp_rtx_synack,
769 .send_ack = tcp_v6_reqsk_send_ack,
770 .destructor = tcp_v6_reqsk_destructor,
771 .send_reset = tcp_v6_send_reset,
772 .syn_ack_timeout = tcp_syn_ack_timeout,
773};
774
775static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
776 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
777 sizeof(struct ipv6hdr),
778#ifdef CONFIG_TCP_MD5SIG
779 .req_md5_lookup = tcp_v6_md5_lookup,
780 .calc_md5_hash = tcp_v6_md5_hash_skb,
781#endif
782 .init_req = tcp_v6_init_req,
783#ifdef CONFIG_SYN_COOKIES
784 .cookie_init_seq = cookie_v6_init_sequence,
785#endif
786 .route_req = tcp_v6_route_req,
787 .init_seq = tcp_v6_init_seq,
788 .init_ts_off = tcp_v6_init_ts_off,
789 .send_synack = tcp_v6_send_synack,
790};
791
792static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
793 u32 ack, u32 win, u32 tsval, u32 tsecr,
794 int oif, struct tcp_md5sig_key *key, int rst,
795 u8 tclass, __be32 label)
796{
797 const struct tcphdr *th = tcp_hdr(skb);
798 struct tcphdr *t1;
799 struct sk_buff *buff;
800 struct flowi6 fl6;
801 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
802 struct sock *ctl_sk = net->ipv6.tcp_sk;
803 unsigned int tot_len = sizeof(struct tcphdr);
804 struct dst_entry *dst;
805 __be32 *topt;
806
807 if (tsecr)
808 tot_len += TCPOLEN_TSTAMP_ALIGNED;
809#ifdef CONFIG_TCP_MD5SIG
810 if (key)
811 tot_len += TCPOLEN_MD5SIG_ALIGNED;
812#endif
813
814 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
815 GFP_ATOMIC);
816 if (!buff)
817 return;
818
819 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
820
821 t1 = skb_push(buff, tot_len);
822 skb_reset_transport_header(buff);
823
824 /* Swap the send and the receive. */
825 memset(t1, 0, sizeof(*t1));
826 t1->dest = th->source;
827 t1->source = th->dest;
828 t1->doff = tot_len / 4;
829 t1->seq = htonl(seq);
830 t1->ack_seq = htonl(ack);
831 t1->ack = !rst || !th->ack;
832 t1->rst = rst;
833 t1->window = htons(win);
834
835 topt = (__be32 *)(t1 + 1);
836
837 if (tsecr) {
838 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
839 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
840 *topt++ = htonl(tsval);
841 *topt++ = htonl(tsecr);
842 }
843
844#ifdef CONFIG_TCP_MD5SIG
845 if (key) {
846 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
847 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
848 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
849 &ipv6_hdr(skb)->saddr,
850 &ipv6_hdr(skb)->daddr, t1);
851 }
852#endif
853
854 memset(&fl6, 0, sizeof(fl6));
855 fl6.daddr = ipv6_hdr(skb)->saddr;
856 fl6.saddr = ipv6_hdr(skb)->daddr;
857 fl6.flowlabel = label;
858
859 buff->ip_summed = CHECKSUM_PARTIAL;
860 buff->csum = 0;
861
862 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
863
864 fl6.flowi6_proto = IPPROTO_TCP;
865 if (rt6_need_strict(&fl6.daddr) && !oif)
866 fl6.flowi6_oif = tcp_v6_iif(skb);
867 else {
868 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
869 oif = skb->skb_iif;
870
871 fl6.flowi6_oif = oif;
872 }
873
874 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
875 fl6.fl6_dport = t1->dest;
876 fl6.fl6_sport = t1->source;
877 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
878 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
879
880 /* Pass a socket to ip6_dst_lookup either it is for RST
881 * Underlying function will use this to retrieve the network
882 * namespace
883 */
884 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
885 if (!IS_ERR(dst)) {
886 skb_dst_set(buff, dst);
887 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, tclass);
888 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
889 if (rst)
890 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
891 return;
892 }
893
894 kfree_skb(buff);
895}
896
897static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
898{
899 const struct tcphdr *th = tcp_hdr(skb);
900 u32 seq = 0, ack_seq = 0;
901 struct tcp_md5sig_key *key = NULL;
902#ifdef CONFIG_TCP_MD5SIG
903 const __u8 *hash_location = NULL;
904 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
905 unsigned char newhash[16];
906 int genhash;
907 struct sock *sk1 = NULL;
908#endif
909 int oif = 0;
910
911 if (th->rst)
912 return;
913
914 /* If sk not NULL, it means we did a successful lookup and incoming
915 * route had to be correct. prequeue might have dropped our dst.
916 */
917 if (!sk && !ipv6_unicast_destination(skb))
918 return;
919
920#ifdef CONFIG_TCP_MD5SIG
921 rcu_read_lock();
922 hash_location = tcp_parse_md5sig_option(th);
923 if (sk && sk_fullsock(sk)) {
924 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
925 } else if (hash_location) {
926 /*
927 * active side is lost. Try to find listening socket through
928 * source port, and then find md5 key through listening socket.
929 * we are not loose security here:
930 * Incoming packet is checked with md5 hash with finding key,
931 * no RST generated if md5 hash doesn't match.
932 */
933 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
934 &tcp_hashinfo, NULL, 0,
935 &ipv6h->saddr,
936 th->source, &ipv6h->daddr,
937 ntohs(th->source), tcp_v6_iif(skb),
938 tcp_v6_sdif(skb));
939 if (!sk1)
940 goto out;
941
942 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
943 if (!key)
944 goto out;
945
946 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
947 if (genhash || memcmp(hash_location, newhash, 16) != 0)
948 goto out;
949 }
950#endif
951
952 if (th->ack)
953 seq = ntohl(th->ack_seq);
954 else
955 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
956 (th->doff << 2);
957
958 if (sk) {
959 oif = sk->sk_bound_dev_if;
960 if (sk_fullsock(sk))
961 trace_tcp_send_reset(sk, skb);
962 }
963
964 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
965
966#ifdef CONFIG_TCP_MD5SIG
967out:
968 rcu_read_unlock();
969#endif
970}
971
972static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
973 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
974 struct tcp_md5sig_key *key, u8 tclass,
975 __be32 label)
976{
977 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
978 tclass, label);
979}
980
981static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
982{
983 struct inet_timewait_sock *tw = inet_twsk(sk);
984 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
985
986 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
987 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
988 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
989 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
990 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
991
992 inet_twsk_put(tw);
993}
994
995static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
996 struct request_sock *req)
997{
998 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
999 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1000 */
1001 /* RFC 7323 2.3
1002 * The window field (SEG.WND) of every outgoing segment, with the
1003 * exception of <SYN> segments, MUST be right-shifted by
1004 * Rcv.Wind.Shift bits:
1005 */
1006 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1007 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1008 tcp_rsk(req)->rcv_nxt,
1009 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1010 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1011 req->ts_recent, sk->sk_bound_dev_if,
1012 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr),
1013 0, 0);
1014}
1015
1016
1017static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1018{
1019#ifdef CONFIG_SYN_COOKIES
1020 const struct tcphdr *th = tcp_hdr(skb);
1021
1022 if (!th->syn)
1023 sk = cookie_v6_check(sk, skb);
1024#endif
1025 return sk;
1026}
1027
1028static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1029{
1030 if (skb->protocol == htons(ETH_P_IP))
1031 return tcp_v4_conn_request(sk, skb);
1032
1033 if (!ipv6_unicast_destination(skb))
1034 goto drop;
1035
1036 return tcp_conn_request(&tcp6_request_sock_ops,
1037 &tcp_request_sock_ipv6_ops, sk, skb);
1038
1039drop:
1040 tcp_listendrop(sk);
1041 return 0; /* don't send reset */
1042}
1043
1044static void tcp_v6_restore_cb(struct sk_buff *skb)
1045{
1046 /* We need to move header back to the beginning if xfrm6_policy_check()
1047 * and tcp_v6_fill_cb() are going to be called again.
1048 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1049 */
1050 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1051 sizeof(struct inet6_skb_parm));
1052}
1053
1054static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1055 struct request_sock *req,
1056 struct dst_entry *dst,
1057 struct request_sock *req_unhash,
1058 bool *own_req)
1059{
1060 struct inet_request_sock *ireq;
1061 struct ipv6_pinfo *newnp;
1062 const struct ipv6_pinfo *np = inet6_sk(sk);
1063 struct ipv6_txoptions *opt;
1064 struct tcp6_sock *newtcp6sk;
1065 struct inet_sock *newinet;
1066 struct tcp_sock *newtp;
1067 struct sock *newsk;
1068#ifdef CONFIG_TCP_MD5SIG
1069 struct tcp_md5sig_key *key;
1070#endif
1071 struct flowi6 fl6;
1072
1073 if (skb->protocol == htons(ETH_P_IP)) {
1074 /*
1075 * v6 mapped
1076 */
1077
1078 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1079 req_unhash, own_req);
1080
1081 if (!newsk)
1082 return NULL;
1083
1084 newtcp6sk = (struct tcp6_sock *)newsk;
1085 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1086
1087 newinet = inet_sk(newsk);
1088 newnp = inet6_sk(newsk);
1089 newtp = tcp_sk(newsk);
1090
1091 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1092
1093 newnp->saddr = newsk->sk_v6_rcv_saddr;
1094
1095 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1096 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1097#ifdef CONFIG_TCP_MD5SIG
1098 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1099#endif
1100
1101 newnp->ipv6_mc_list = NULL;
1102 newnp->ipv6_ac_list = NULL;
1103 newnp->ipv6_fl_list = NULL;
1104 newnp->pktoptions = NULL;
1105 newnp->opt = NULL;
1106 newnp->mcast_oif = tcp_v6_iif(skb);
1107 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1108 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1109 if (np->repflow)
1110 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1111
1112 /*
1113 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1114 * here, tcp_create_openreq_child now does this for us, see the comment in
1115 * that function for the gory details. -acme
1116 */
1117
1118 /* It is tricky place. Until this moment IPv4 tcp
1119 worked with IPv6 icsk.icsk_af_ops.
1120 Sync it now.
1121 */
1122 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1123
1124 return newsk;
1125 }
1126
1127 ireq = inet_rsk(req);
1128
1129 if (sk_acceptq_is_full(sk))
1130 goto out_overflow;
1131
1132 if (!dst) {
1133 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1134 if (!dst)
1135 goto out;
1136 }
1137
1138 newsk = tcp_create_openreq_child(sk, req, skb);
1139 if (!newsk)
1140 goto out_nonewsk;
1141
1142 /*
1143 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1144 * count here, tcp_create_openreq_child now does this for us, see the
1145 * comment in that function for the gory details. -acme
1146 */
1147
1148 newsk->sk_gso_type = SKB_GSO_TCPV6;
1149 ip6_dst_store(newsk, dst, NULL, NULL);
1150 inet6_sk_rx_dst_set(newsk, skb);
1151
1152 newtcp6sk = (struct tcp6_sock *)newsk;
1153 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1154
1155 newtp = tcp_sk(newsk);
1156 newinet = inet_sk(newsk);
1157 newnp = inet6_sk(newsk);
1158
1159 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1160
1161 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1162 newnp->saddr = ireq->ir_v6_loc_addr;
1163 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1164 newsk->sk_bound_dev_if = ireq->ir_iif;
1165
1166 /* Now IPv6 options...
1167
1168 First: no IPv4 options.
1169 */
1170 newinet->inet_opt = NULL;
1171 newnp->ipv6_mc_list = NULL;
1172 newnp->ipv6_ac_list = NULL;
1173 newnp->ipv6_fl_list = NULL;
1174
1175 /* Clone RX bits */
1176 newnp->rxopt.all = np->rxopt.all;
1177
1178 newnp->pktoptions = NULL;
1179 newnp->opt = NULL;
1180 newnp->mcast_oif = tcp_v6_iif(skb);
1181 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1182 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1183 if (np->repflow)
1184 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1185
1186 /* Clone native IPv6 options from listening socket (if any)
1187
1188 Yes, keeping reference count would be much more clever,
1189 but we make one more one thing there: reattach optmem
1190 to newsk.
1191 */
1192 opt = ireq->ipv6_opt;
1193 if (!opt)
1194 opt = rcu_dereference(np->opt);
1195 if (opt) {
1196 opt = ipv6_dup_options(newsk, opt);
1197 RCU_INIT_POINTER(newnp->opt, opt);
1198 }
1199 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1200 if (opt)
1201 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1202 opt->opt_flen;
1203
1204 tcp_ca_openreq_child(newsk, dst);
1205
1206 tcp_sync_mss(newsk, dst_mtu(dst));
1207 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1208
1209 tcp_initialize_rcv_mss(newsk);
1210
1211 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1212 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1213
1214#ifdef CONFIG_TCP_MD5SIG
1215 /* Copy over the MD5 key from the original socket */
1216 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1217 if (key) {
1218 /* We're using one, so create a matching key
1219 * on the newsk structure. If we fail to get
1220 * memory, then we end up not copying the key
1221 * across. Shucks.
1222 */
1223 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1224 AF_INET6, 128, key->key, key->keylen,
1225 sk_gfp_mask(sk, GFP_ATOMIC));
1226 }
1227#endif
1228
1229 if (__inet_inherit_port(sk, newsk) < 0) {
1230 inet_csk_prepare_forced_close(newsk);
1231 tcp_done(newsk);
1232 goto out;
1233 }
1234 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1235 if (*own_req) {
1236 tcp_move_syn(newtp, req);
1237
1238 /* Clone pktoptions received with SYN, if we own the req */
1239 if (ireq->pktopts) {
1240 newnp->pktoptions = skb_clone(ireq->pktopts,
1241 sk_gfp_mask(sk, GFP_ATOMIC));
1242 consume_skb(ireq->pktopts);
1243 ireq->pktopts = NULL;
1244 if (newnp->pktoptions) {
1245 tcp_v6_restore_cb(newnp->pktoptions);
1246 skb_set_owner_r(newnp->pktoptions, newsk);
1247 }
1248 }
1249 }
1250
1251 return newsk;
1252
1253out_overflow:
1254 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1255out_nonewsk:
1256 dst_release(dst);
1257out:
1258 tcp_listendrop(sk);
1259 return NULL;
1260}
1261
1262/* The socket must have it's spinlock held when we get
1263 * here, unless it is a TCP_LISTEN socket.
1264 *
1265 * We have a potential double-lock case here, so even when
1266 * doing backlog processing we use the BH locking scheme.
1267 * This is because we cannot sleep with the original spinlock
1268 * held.
1269 */
1270static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1271{
1272 struct ipv6_pinfo *np = inet6_sk(sk);
1273 struct tcp_sock *tp;
1274 struct sk_buff *opt_skb = NULL;
1275
1276 /* Imagine: socket is IPv6. IPv4 packet arrives,
1277 goes to IPv4 receive handler and backlogged.
1278 From backlog it always goes here. Kerboom...
1279 Fortunately, tcp_rcv_established and rcv_established
1280 handle them correctly, but it is not case with
1281 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1282 */
1283
1284 if (skb->protocol == htons(ETH_P_IP))
1285 return tcp_v4_do_rcv(sk, skb);
1286
1287 /*
1288 * socket locking is here for SMP purposes as backlog rcv
1289 * is currently called with bh processing disabled.
1290 */
1291
1292 /* Do Stevens' IPV6_PKTOPTIONS.
1293
1294 Yes, guys, it is the only place in our code, where we
1295 may make it not affecting IPv4.
1296 The rest of code is protocol independent,
1297 and I do not like idea to uglify IPv4.
1298
1299 Actually, all the idea behind IPV6_PKTOPTIONS
1300 looks not very well thought. For now we latch
1301 options, received in the last packet, enqueued
1302 by tcp. Feel free to propose better solution.
1303 --ANK (980728)
1304 */
1305 if (np->rxopt.all)
1306 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1307
1308 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1309 struct dst_entry *dst = sk->sk_rx_dst;
1310
1311 sock_rps_save_rxhash(sk, skb);
1312 sk_mark_napi_id(sk, skb);
1313 if (dst) {
1314 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1315 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1316 dst_release(dst);
1317 sk->sk_rx_dst = NULL;
1318 }
1319 }
1320
1321 tcp_rcv_established(sk, skb, tcp_hdr(skb));
1322 if (opt_skb)
1323 goto ipv6_pktoptions;
1324 return 0;
1325 }
1326
1327 if (tcp_checksum_complete(skb))
1328 goto csum_err;
1329
1330 if (sk->sk_state == TCP_LISTEN) {
1331 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1332
1333 if (!nsk)
1334 goto discard;
1335
1336 if (nsk != sk) {
1337 if (tcp_child_process(sk, nsk, skb))
1338 goto reset;
1339 if (opt_skb)
1340 __kfree_skb(opt_skb);
1341 return 0;
1342 }
1343 } else
1344 sock_rps_save_rxhash(sk, skb);
1345
1346 if (tcp_rcv_state_process(sk, skb))
1347 goto reset;
1348 if (opt_skb)
1349 goto ipv6_pktoptions;
1350 return 0;
1351
1352reset:
1353 tcp_v6_send_reset(sk, skb);
1354discard:
1355 if (opt_skb)
1356 __kfree_skb(opt_skb);
1357 kfree_skb(skb);
1358 return 0;
1359csum_err:
1360 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1361 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1362 goto discard;
1363
1364
1365ipv6_pktoptions:
1366 /* Do you ask, what is it?
1367
1368 1. skb was enqueued by tcp.
1369 2. skb is added to tail of read queue, rather than out of order.
1370 3. socket is not in passive state.
1371 4. Finally, it really contains options, which user wants to receive.
1372 */
1373 tp = tcp_sk(sk);
1374 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1375 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1376 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1377 np->mcast_oif = tcp_v6_iif(opt_skb);
1378 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1379 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1380 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1381 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1382 if (np->repflow)
1383 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1384 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1385 skb_set_owner_r(opt_skb, sk);
1386 tcp_v6_restore_cb(opt_skb);
1387 opt_skb = xchg(&np->pktoptions, opt_skb);
1388 } else {
1389 __kfree_skb(opt_skb);
1390 opt_skb = xchg(&np->pktoptions, NULL);
1391 }
1392 }
1393
1394 kfree_skb(opt_skb);
1395 return 0;
1396}
1397
1398static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1399 const struct tcphdr *th)
1400{
1401 /* This is tricky: we move IP6CB at its correct location into
1402 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1403 * _decode_session6() uses IP6CB().
1404 * barrier() makes sure compiler won't play aliasing games.
1405 */
1406 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1407 sizeof(struct inet6_skb_parm));
1408 barrier();
1409
1410 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1411 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1412 skb->len - th->doff*4);
1413 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1414 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1415 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1416 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1417 TCP_SKB_CB(skb)->sacked = 0;
1418 TCP_SKB_CB(skb)->has_rxtstamp =
1419 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1420}
1421
1422static int tcp_v6_rcv(struct sk_buff *skb)
1423{
1424 int sdif = inet6_sdif(skb);
1425 const struct tcphdr *th;
1426 const struct ipv6hdr *hdr;
1427 bool refcounted;
1428 struct sock *sk;
1429 int ret;
1430 struct net *net = dev_net(skb->dev);
1431
1432 if (skb->pkt_type != PACKET_HOST)
1433 goto discard_it;
1434
1435 /*
1436 * Count it even if it's bad.
1437 */
1438 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1439
1440 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1441 goto discard_it;
1442
1443 th = (const struct tcphdr *)skb->data;
1444
1445 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1446 goto bad_packet;
1447 if (!pskb_may_pull(skb, th->doff*4))
1448 goto discard_it;
1449
1450 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1451 goto csum_error;
1452
1453 th = (const struct tcphdr *)skb->data;
1454 hdr = ipv6_hdr(skb);
1455
1456lookup:
1457 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1458 th->source, th->dest, inet6_iif(skb), sdif,
1459 &refcounted);
1460 if (!sk)
1461 goto no_tcp_socket;
1462
1463process:
1464 if (sk->sk_state == TCP_TIME_WAIT)
1465 goto do_time_wait;
1466
1467 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1468 struct request_sock *req = inet_reqsk(sk);
1469 bool req_stolen = false;
1470 struct sock *nsk;
1471
1472 sk = req->rsk_listener;
1473 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1474 sk_drops_add(sk, skb);
1475 reqsk_put(req);
1476 goto discard_it;
1477 }
1478 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1479 inet_csk_reqsk_queue_drop_and_put(sk, req);
1480 goto lookup;
1481 }
1482 sock_hold(sk);
1483 refcounted = true;
1484 nsk = NULL;
1485 if (!tcp_filter(sk, skb)) {
1486 th = (const struct tcphdr *)skb->data;
1487 hdr = ipv6_hdr(skb);
1488 tcp_v6_fill_cb(skb, hdr, th);
1489 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1490 }
1491 if (!nsk) {
1492 reqsk_put(req);
1493 if (req_stolen) {
1494 /* Another cpu got exclusive access to req
1495 * and created a full blown socket.
1496 * Try to feed this packet to this socket
1497 * instead of discarding it.
1498 */
1499 tcp_v6_restore_cb(skb);
1500 sock_put(sk);
1501 goto lookup;
1502 }
1503 goto discard_and_relse;
1504 }
1505 if (nsk == sk) {
1506 reqsk_put(req);
1507 tcp_v6_restore_cb(skb);
1508 } else if (tcp_child_process(sk, nsk, skb)) {
1509 tcp_v6_send_reset(nsk, skb);
1510 goto discard_and_relse;
1511 } else {
1512 sock_put(sk);
1513 return 0;
1514 }
1515 }
1516 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1517 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1518 goto discard_and_relse;
1519 }
1520
1521 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1522 goto discard_and_relse;
1523
1524 if (tcp_v6_inbound_md5_hash(sk, skb))
1525 goto discard_and_relse;
1526
1527 if (tcp_filter(sk, skb))
1528 goto discard_and_relse;
1529 th = (const struct tcphdr *)skb->data;
1530 hdr = ipv6_hdr(skb);
1531 tcp_v6_fill_cb(skb, hdr, th);
1532
1533 skb->dev = NULL;
1534
1535 if (sk->sk_state == TCP_LISTEN) {
1536 ret = tcp_v6_do_rcv(sk, skb);
1537 goto put_and_return;
1538 }
1539
1540 sk_incoming_cpu_update(sk);
1541
1542 bh_lock_sock_nested(sk);
1543 tcp_segs_in(tcp_sk(sk), skb);
1544 ret = 0;
1545 if (!sock_owned_by_user(sk)) {
1546 ret = tcp_v6_do_rcv(sk, skb);
1547 } else if (tcp_add_backlog(sk, skb)) {
1548 goto discard_and_relse;
1549 }
1550 bh_unlock_sock(sk);
1551
1552put_and_return:
1553 if (refcounted)
1554 sock_put(sk);
1555 return ret ? -1 : 0;
1556
1557no_tcp_socket:
1558 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1559 goto discard_it;
1560
1561 tcp_v6_fill_cb(skb, hdr, th);
1562
1563 if (tcp_checksum_complete(skb)) {
1564csum_error:
1565 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1566bad_packet:
1567 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1568 } else {
1569 tcp_v6_send_reset(NULL, skb);
1570 }
1571
1572discard_it:
1573 kfree_skb(skb);
1574 return 0;
1575
1576discard_and_relse:
1577 sk_drops_add(sk, skb);
1578 if (refcounted)
1579 sock_put(sk);
1580 goto discard_it;
1581
1582do_time_wait:
1583 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1584 inet_twsk_put(inet_twsk(sk));
1585 goto discard_it;
1586 }
1587
1588 tcp_v6_fill_cb(skb, hdr, th);
1589
1590 if (tcp_checksum_complete(skb)) {
1591 inet_twsk_put(inet_twsk(sk));
1592 goto csum_error;
1593 }
1594
1595 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1596 case TCP_TW_SYN:
1597 {
1598 struct sock *sk2;
1599
1600 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1601 skb, __tcp_hdrlen(th),
1602 &ipv6_hdr(skb)->saddr, th->source,
1603 &ipv6_hdr(skb)->daddr,
1604 ntohs(th->dest), tcp_v6_iif(skb),
1605 sdif);
1606 if (sk2) {
1607 struct inet_timewait_sock *tw = inet_twsk(sk);
1608 inet_twsk_deschedule_put(tw);
1609 sk = sk2;
1610 tcp_v6_restore_cb(skb);
1611 refcounted = false;
1612 goto process;
1613 }
1614 }
1615 /* to ACK */
1616 /* fall through */
1617 case TCP_TW_ACK:
1618 tcp_v6_timewait_ack(sk, skb);
1619 break;
1620 case TCP_TW_RST:
1621 tcp_v6_send_reset(sk, skb);
1622 inet_twsk_deschedule_put(inet_twsk(sk));
1623 goto discard_it;
1624 case TCP_TW_SUCCESS:
1625 ;
1626 }
1627 goto discard_it;
1628}
1629
1630static void tcp_v6_early_demux(struct sk_buff *skb)
1631{
1632 const struct ipv6hdr *hdr;
1633 const struct tcphdr *th;
1634 struct sock *sk;
1635
1636 if (skb->pkt_type != PACKET_HOST)
1637 return;
1638
1639 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1640 return;
1641
1642 hdr = ipv6_hdr(skb);
1643 th = tcp_hdr(skb);
1644
1645 if (th->doff < sizeof(struct tcphdr) / 4)
1646 return;
1647
1648 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1649 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1650 &hdr->saddr, th->source,
1651 &hdr->daddr, ntohs(th->dest),
1652 inet6_iif(skb), inet6_sdif(skb));
1653 if (sk) {
1654 skb->sk = sk;
1655 skb->destructor = sock_edemux;
1656 if (sk_fullsock(sk)) {
1657 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1658
1659 if (dst)
1660 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1661 if (dst &&
1662 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1663 skb_dst_set_noref(skb, dst);
1664 }
1665 }
1666}
1667
1668static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1669 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1670 .twsk_unique = tcp_twsk_unique,
1671 .twsk_destructor = tcp_twsk_destructor,
1672};
1673
1674static const struct inet_connection_sock_af_ops ipv6_specific = {
1675 .queue_xmit = inet6_csk_xmit,
1676 .send_check = tcp_v6_send_check,
1677 .rebuild_header = inet6_sk_rebuild_header,
1678 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1679 .conn_request = tcp_v6_conn_request,
1680 .syn_recv_sock = tcp_v6_syn_recv_sock,
1681 .net_header_len = sizeof(struct ipv6hdr),
1682 .net_frag_header_len = sizeof(struct frag_hdr),
1683 .setsockopt = ipv6_setsockopt,
1684 .getsockopt = ipv6_getsockopt,
1685 .addr2sockaddr = inet6_csk_addr2sockaddr,
1686 .sockaddr_len = sizeof(struct sockaddr_in6),
1687#ifdef CONFIG_COMPAT
1688 .compat_setsockopt = compat_ipv6_setsockopt,
1689 .compat_getsockopt = compat_ipv6_getsockopt,
1690#endif
1691 .mtu_reduced = tcp_v6_mtu_reduced,
1692};
1693
1694#ifdef CONFIG_TCP_MD5SIG
1695static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1696 .md5_lookup = tcp_v6_md5_lookup,
1697 .calc_md5_hash = tcp_v6_md5_hash_skb,
1698 .md5_parse = tcp_v6_parse_md5_keys,
1699};
1700#endif
1701
1702/*
1703 * TCP over IPv4 via INET6 API
1704 */
1705static const struct inet_connection_sock_af_ops ipv6_mapped = {
1706 .queue_xmit = ip_queue_xmit,
1707 .send_check = tcp_v4_send_check,
1708 .rebuild_header = inet_sk_rebuild_header,
1709 .sk_rx_dst_set = inet_sk_rx_dst_set,
1710 .conn_request = tcp_v6_conn_request,
1711 .syn_recv_sock = tcp_v6_syn_recv_sock,
1712 .net_header_len = sizeof(struct iphdr),
1713 .setsockopt = ipv6_setsockopt,
1714 .getsockopt = ipv6_getsockopt,
1715 .addr2sockaddr = inet6_csk_addr2sockaddr,
1716 .sockaddr_len = sizeof(struct sockaddr_in6),
1717#ifdef CONFIG_COMPAT
1718 .compat_setsockopt = compat_ipv6_setsockopt,
1719 .compat_getsockopt = compat_ipv6_getsockopt,
1720#endif
1721 .mtu_reduced = tcp_v4_mtu_reduced,
1722};
1723
1724#ifdef CONFIG_TCP_MD5SIG
1725static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1726 .md5_lookup = tcp_v4_md5_lookup,
1727 .calc_md5_hash = tcp_v4_md5_hash_skb,
1728 .md5_parse = tcp_v6_parse_md5_keys,
1729};
1730#endif
1731
1732/* NOTE: A lot of things set to zero explicitly by call to
1733 * sk_alloc() so need not be done here.
1734 */
1735static int tcp_v6_init_sock(struct sock *sk)
1736{
1737 struct inet_connection_sock *icsk = inet_csk(sk);
1738
1739 tcp_init_sock(sk);
1740
1741 icsk->icsk_af_ops = &ipv6_specific;
1742
1743#ifdef CONFIG_TCP_MD5SIG
1744 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1745#endif
1746
1747 return 0;
1748}
1749
1750static void tcp_v6_destroy_sock(struct sock *sk)
1751{
1752 tcp_v4_destroy_sock(sk);
1753 inet6_destroy_sock(sk);
1754}
1755
1756#ifdef CONFIG_PROC_FS
1757/* Proc filesystem TCPv6 sock list dumping. */
1758static void get_openreq6(struct seq_file *seq,
1759 const struct request_sock *req, int i)
1760{
1761 long ttd = req->rsk_timer.expires - jiffies;
1762 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1763 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1764
1765 if (ttd < 0)
1766 ttd = 0;
1767
1768 seq_printf(seq,
1769 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1770 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1771 i,
1772 src->s6_addr32[0], src->s6_addr32[1],
1773 src->s6_addr32[2], src->s6_addr32[3],
1774 inet_rsk(req)->ir_num,
1775 dest->s6_addr32[0], dest->s6_addr32[1],
1776 dest->s6_addr32[2], dest->s6_addr32[3],
1777 ntohs(inet_rsk(req)->ir_rmt_port),
1778 TCP_SYN_RECV,
1779 0, 0, /* could print option size, but that is af dependent. */
1780 1, /* timers active (only the expire timer) */
1781 jiffies_to_clock_t(ttd),
1782 req->num_timeout,
1783 from_kuid_munged(seq_user_ns(seq),
1784 sock_i_uid(req->rsk_listener)),
1785 0, /* non standard timer */
1786 0, /* open_requests have no inode */
1787 0, req);
1788}
1789
1790static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1791{
1792 const struct in6_addr *dest, *src;
1793 __u16 destp, srcp;
1794 int timer_active;
1795 unsigned long timer_expires;
1796 const struct inet_sock *inet = inet_sk(sp);
1797 const struct tcp_sock *tp = tcp_sk(sp);
1798 const struct inet_connection_sock *icsk = inet_csk(sp);
1799 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1800 int rx_queue;
1801 int state;
1802
1803 dest = &sp->sk_v6_daddr;
1804 src = &sp->sk_v6_rcv_saddr;
1805 destp = ntohs(inet->inet_dport);
1806 srcp = ntohs(inet->inet_sport);
1807
1808 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1809 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
1810 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1811 timer_active = 1;
1812 timer_expires = icsk->icsk_timeout;
1813 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1814 timer_active = 4;
1815 timer_expires = icsk->icsk_timeout;
1816 } else if (timer_pending(&sp->sk_timer)) {
1817 timer_active = 2;
1818 timer_expires = sp->sk_timer.expires;
1819 } else {
1820 timer_active = 0;
1821 timer_expires = jiffies;
1822 }
1823
1824 state = inet_sk_state_load(sp);
1825 if (state == TCP_LISTEN)
1826 rx_queue = sp->sk_ack_backlog;
1827 else
1828 /* Because we don't lock the socket,
1829 * we might find a transient negative value.
1830 */
1831 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1832
1833 seq_printf(seq,
1834 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1835 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1836 i,
1837 src->s6_addr32[0], src->s6_addr32[1],
1838 src->s6_addr32[2], src->s6_addr32[3], srcp,
1839 dest->s6_addr32[0], dest->s6_addr32[1],
1840 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1841 state,
1842 tp->write_seq - tp->snd_una,
1843 rx_queue,
1844 timer_active,
1845 jiffies_delta_to_clock_t(timer_expires - jiffies),
1846 icsk->icsk_retransmits,
1847 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1848 icsk->icsk_probes_out,
1849 sock_i_ino(sp),
1850 refcount_read(&sp->sk_refcnt), sp,
1851 jiffies_to_clock_t(icsk->icsk_rto),
1852 jiffies_to_clock_t(icsk->icsk_ack.ato),
1853 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1854 tp->snd_cwnd,
1855 state == TCP_LISTEN ?
1856 fastopenq->max_qlen :
1857 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1858 );
1859}
1860
1861static void get_timewait6_sock(struct seq_file *seq,
1862 struct inet_timewait_sock *tw, int i)
1863{
1864 long delta = tw->tw_timer.expires - jiffies;
1865 const struct in6_addr *dest, *src;
1866 __u16 destp, srcp;
1867
1868 dest = &tw->tw_v6_daddr;
1869 src = &tw->tw_v6_rcv_saddr;
1870 destp = ntohs(tw->tw_dport);
1871 srcp = ntohs(tw->tw_sport);
1872
1873 seq_printf(seq,
1874 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1875 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1876 i,
1877 src->s6_addr32[0], src->s6_addr32[1],
1878 src->s6_addr32[2], src->s6_addr32[3], srcp,
1879 dest->s6_addr32[0], dest->s6_addr32[1],
1880 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1881 tw->tw_substate, 0, 0,
1882 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1883 refcount_read(&tw->tw_refcnt), tw);
1884}
1885
1886static int tcp6_seq_show(struct seq_file *seq, void *v)
1887{
1888 struct tcp_iter_state *st;
1889 struct sock *sk = v;
1890
1891 if (v == SEQ_START_TOKEN) {
1892 seq_puts(seq,
1893 " sl "
1894 "local_address "
1895 "remote_address "
1896 "st tx_queue rx_queue tr tm->when retrnsmt"
1897 " uid timeout inode\n");
1898 goto out;
1899 }
1900 st = seq->private;
1901
1902 if (sk->sk_state == TCP_TIME_WAIT)
1903 get_timewait6_sock(seq, v, st->num);
1904 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1905 get_openreq6(seq, v, st->num);
1906 else
1907 get_tcp6_sock(seq, v, st->num);
1908out:
1909 return 0;
1910}
1911
1912static const struct file_operations tcp6_afinfo_seq_fops = {
1913 .open = tcp_seq_open,
1914 .read = seq_read,
1915 .llseek = seq_lseek,
1916 .release = seq_release_net
1917};
1918
1919static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1920 .name = "tcp6",
1921 .family = AF_INET6,
1922 .seq_fops = &tcp6_afinfo_seq_fops,
1923 .seq_ops = {
1924 .show = tcp6_seq_show,
1925 },
1926};
1927
1928int __net_init tcp6_proc_init(struct net *net)
1929{
1930 return tcp_proc_register(net, &tcp6_seq_afinfo);
1931}
1932
1933void tcp6_proc_exit(struct net *net)
1934{
1935 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1936}
1937#endif
1938
1939struct proto tcpv6_prot = {
1940 .name = "TCPv6",
1941 .owner = THIS_MODULE,
1942 .close = tcp_close,
1943 .pre_connect = tcp_v6_pre_connect,
1944 .connect = tcp_v6_connect,
1945 .disconnect = tcp_disconnect,
1946 .accept = inet_csk_accept,
1947 .ioctl = tcp_ioctl,
1948 .init = tcp_v6_init_sock,
1949 .destroy = tcp_v6_destroy_sock,
1950 .shutdown = tcp_shutdown,
1951 .setsockopt = tcp_setsockopt,
1952 .getsockopt = tcp_getsockopt,
1953 .keepalive = tcp_set_keepalive,
1954 .recvmsg = tcp_recvmsg,
1955 .sendmsg = tcp_sendmsg,
1956 .sendpage = tcp_sendpage,
1957 .backlog_rcv = tcp_v6_do_rcv,
1958 .release_cb = tcp_release_cb,
1959 .hash = inet6_hash,
1960 .unhash = inet_unhash,
1961 .get_port = inet_csk_get_port,
1962 .enter_memory_pressure = tcp_enter_memory_pressure,
1963 .leave_memory_pressure = tcp_leave_memory_pressure,
1964 .stream_memory_free = tcp_stream_memory_free,
1965 .sockets_allocated = &tcp_sockets_allocated,
1966 .memory_allocated = &tcp_memory_allocated,
1967 .memory_pressure = &tcp_memory_pressure,
1968 .orphan_count = &tcp_orphan_count,
1969 .sysctl_mem = sysctl_tcp_mem,
1970 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
1971 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
1972 .max_header = MAX_TCP_HEADER,
1973 .obj_size = sizeof(struct tcp6_sock),
1974 .slab_flags = SLAB_TYPESAFE_BY_RCU,
1975 .twsk_prot = &tcp6_timewait_sock_ops,
1976 .rsk_prot = &tcp6_request_sock_ops,
1977 .h.hashinfo = &tcp_hashinfo,
1978 .no_autobind = true,
1979#ifdef CONFIG_COMPAT
1980 .compat_setsockopt = compat_tcp_setsockopt,
1981 .compat_getsockopt = compat_tcp_getsockopt,
1982#endif
1983 .diag_destroy = tcp_abort,
1984};
1985
1986/* thinking of making this const? Don't.
1987 * early_demux can change based on sysctl.
1988 */
1989static struct inet6_protocol tcpv6_protocol = {
1990 .early_demux = tcp_v6_early_demux,
1991 .early_demux_handler = tcp_v6_early_demux,
1992 .handler = tcp_v6_rcv,
1993 .err_handler = tcp_v6_err,
1994 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1995};
1996
1997static struct inet_protosw tcpv6_protosw = {
1998 .type = SOCK_STREAM,
1999 .protocol = IPPROTO_TCP,
2000 .prot = &tcpv6_prot,
2001 .ops = &inet6_stream_ops,
2002 .flags = INET_PROTOSW_PERMANENT |
2003 INET_PROTOSW_ICSK,
2004};
2005
2006static int __net_init tcpv6_net_init(struct net *net)
2007{
2008 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2009 SOCK_RAW, IPPROTO_TCP, net);
2010}
2011
2012static void __net_exit tcpv6_net_exit(struct net *net)
2013{
2014 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2015}
2016
2017static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2018{
2019 inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2020}
2021
2022static struct pernet_operations tcpv6_net_ops = {
2023 .init = tcpv6_net_init,
2024 .exit = tcpv6_net_exit,
2025 .exit_batch = tcpv6_net_exit_batch,
2026};
2027
2028int __init tcpv6_init(void)
2029{
2030 int ret;
2031
2032 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2033 if (ret)
2034 goto out;
2035
2036 /* register inet6 protocol */
2037 ret = inet6_register_protosw(&tcpv6_protosw);
2038 if (ret)
2039 goto out_tcpv6_protocol;
2040
2041 ret = register_pernet_subsys(&tcpv6_net_ops);
2042 if (ret)
2043 goto out_tcpv6_protosw;
2044out:
2045 return ret;
2046
2047out_tcpv6_protosw:
2048 inet6_unregister_protosw(&tcpv6_protosw);
2049out_tcpv6_protocol:
2050 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2051 goto out;
2052}
2053
2054void tcpv6_exit(void)
2055{
2056 unregister_pernet_subsys(&tcpv6_net_ops);
2057 inet6_unregister_protosw(&tcpv6_protosw);
2058 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2059}