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/netdma.h>
63#include <net/inet_common.h>
64#include <net/secure_seq.h>
65#include <net/tcp_memcontrol.h>
66#include <net/busy_poll.h>
67
68#include <linux/proc_fs.h>
69#include <linux/seq_file.h>
70
71#include <linux/crypto.h>
72#include <linux/scatterlist.h>
73
74static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
75static void tcp_v6_reqsk_send_ack(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(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 const struct rt6_info *rt = (const struct rt6_info *)dst;
97
98 dst_hold(dst);
99 sk->sk_rx_dst = dst;
100 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
101 if (rt->rt6i_node)
102 inet6_sk(sk)->rx_dst_cookie = rt->rt6i_node->fn_sernum;
103}
104
105static void tcp_v6_hash(struct sock *sk)
106{
107 if (sk->sk_state != TCP_CLOSE) {
108 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
109 tcp_prot.hash(sk);
110 return;
111 }
112 local_bh_disable();
113 __inet6_hash(sk, NULL);
114 local_bh_enable();
115 }
116}
117
118static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
119{
120 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
121 ipv6_hdr(skb)->saddr.s6_addr32,
122 tcp_hdr(skb)->dest,
123 tcp_hdr(skb)->source);
124}
125
126static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
127 int addr_len)
128{
129 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
130 struct inet_sock *inet = inet_sk(sk);
131 struct inet_connection_sock *icsk = inet_csk(sk);
132 struct ipv6_pinfo *np = inet6_sk(sk);
133 struct tcp_sock *tp = tcp_sk(sk);
134 struct in6_addr *saddr = NULL, *final_p, final;
135 struct rt6_info *rt;
136 struct flowi6 fl6;
137 struct dst_entry *dst;
138 int addr_type;
139 int err;
140
141 if (addr_len < SIN6_LEN_RFC2133)
142 return -EINVAL;
143
144 if (usin->sin6_family != AF_INET6)
145 return -EAFNOSUPPORT;
146
147 memset(&fl6, 0, sizeof(fl6));
148
149 if (np->sndflow) {
150 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
151 IP6_ECN_flow_init(fl6.flowlabel);
152 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
153 struct ip6_flowlabel *flowlabel;
154 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
155 if (flowlabel == NULL)
156 return -EINVAL;
157 fl6_sock_release(flowlabel);
158 }
159 }
160
161 /*
162 * connect() to INADDR_ANY means loopback (BSD'ism).
163 */
164
165 if (ipv6_addr_any(&usin->sin6_addr))
166 usin->sin6_addr.s6_addr[15] = 0x1;
167
168 addr_type = ipv6_addr_type(&usin->sin6_addr);
169
170 if (addr_type & IPV6_ADDR_MULTICAST)
171 return -ENETUNREACH;
172
173 if (addr_type&IPV6_ADDR_LINKLOCAL) {
174 if (addr_len >= sizeof(struct sockaddr_in6) &&
175 usin->sin6_scope_id) {
176 /* If interface is set while binding, indices
177 * must coincide.
178 */
179 if (sk->sk_bound_dev_if &&
180 sk->sk_bound_dev_if != usin->sin6_scope_id)
181 return -EINVAL;
182
183 sk->sk_bound_dev_if = usin->sin6_scope_id;
184 }
185
186 /* Connect to link-local address requires an interface */
187 if (!sk->sk_bound_dev_if)
188 return -EINVAL;
189 }
190
191 if (tp->rx_opt.ts_recent_stamp &&
192 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
193 tp->rx_opt.ts_recent = 0;
194 tp->rx_opt.ts_recent_stamp = 0;
195 tp->write_seq = 0;
196 }
197
198 sk->sk_v6_daddr = usin->sin6_addr;
199 np->flow_label = fl6.flowlabel;
200
201 /*
202 * TCP over IPv4
203 */
204
205 if (addr_type == IPV6_ADDR_MAPPED) {
206 u32 exthdrlen = icsk->icsk_ext_hdr_len;
207 struct sockaddr_in sin;
208
209 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
210
211 if (__ipv6_only_sock(sk))
212 return -ENETUNREACH;
213
214 sin.sin_family = AF_INET;
215 sin.sin_port = usin->sin6_port;
216 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
217
218 icsk->icsk_af_ops = &ipv6_mapped;
219 sk->sk_backlog_rcv = tcp_v4_do_rcv;
220#ifdef CONFIG_TCP_MD5SIG
221 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
222#endif
223
224 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
225
226 if (err) {
227 icsk->icsk_ext_hdr_len = exthdrlen;
228 icsk->icsk_af_ops = &ipv6_specific;
229 sk->sk_backlog_rcv = tcp_v6_do_rcv;
230#ifdef CONFIG_TCP_MD5SIG
231 tp->af_specific = &tcp_sock_ipv6_specific;
232#endif
233 goto failure;
234 } else {
235 ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
236 ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
237 &sk->sk_v6_rcv_saddr);
238 }
239
240 return err;
241 }
242
243 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
244 saddr = &sk->sk_v6_rcv_saddr;
245
246 fl6.flowi6_proto = IPPROTO_TCP;
247 fl6.daddr = sk->sk_v6_daddr;
248 fl6.saddr = saddr ? *saddr : np->saddr;
249 fl6.flowi6_oif = sk->sk_bound_dev_if;
250 fl6.flowi6_mark = sk->sk_mark;
251 fl6.fl6_dport = usin->sin6_port;
252 fl6.fl6_sport = inet->inet_sport;
253
254 final_p = fl6_update_dst(&fl6, np->opt, &final);
255
256 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
257
258 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
259 if (IS_ERR(dst)) {
260 err = PTR_ERR(dst);
261 goto failure;
262 }
263
264 if (saddr == NULL) {
265 saddr = &fl6.saddr;
266 sk->sk_v6_rcv_saddr = *saddr;
267 }
268
269 /* set the source address */
270 np->saddr = *saddr;
271 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
272
273 sk->sk_gso_type = SKB_GSO_TCPV6;
274 __ip6_dst_store(sk, dst, NULL, NULL);
275
276 rt = (struct rt6_info *) dst;
277 if (tcp_death_row.sysctl_tw_recycle &&
278 !tp->rx_opt.ts_recent_stamp &&
279 ipv6_addr_equal(&rt->rt6i_dst.addr, &sk->sk_v6_daddr))
280 tcp_fetch_timewait_stamp(sk, dst);
281
282 icsk->icsk_ext_hdr_len = 0;
283 if (np->opt)
284 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
285 np->opt->opt_nflen);
286
287 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
288
289 inet->inet_dport = usin->sin6_port;
290
291 tcp_set_state(sk, TCP_SYN_SENT);
292 err = inet6_hash_connect(&tcp_death_row, sk);
293 if (err)
294 goto late_failure;
295
296 if (!tp->write_seq && likely(!tp->repair))
297 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
298 sk->sk_v6_daddr.s6_addr32,
299 inet->inet_sport,
300 inet->inet_dport);
301
302 err = tcp_connect(sk);
303 if (err)
304 goto late_failure;
305
306 return 0;
307
308late_failure:
309 tcp_set_state(sk, TCP_CLOSE);
310 __sk_dst_reset(sk);
311failure:
312 inet->inet_dport = 0;
313 sk->sk_route_caps = 0;
314 return err;
315}
316
317static void tcp_v6_mtu_reduced(struct sock *sk)
318{
319 struct dst_entry *dst;
320
321 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
322 return;
323
324 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
325 if (!dst)
326 return;
327
328 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
329 tcp_sync_mss(sk, dst_mtu(dst));
330 tcp_simple_retransmit(sk);
331 }
332}
333
334static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
335 u8 type, u8 code, int offset, __be32 info)
336{
337 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
338 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
339 struct ipv6_pinfo *np;
340 struct sock *sk;
341 int err;
342 struct tcp_sock *tp;
343 __u32 seq;
344 struct net *net = dev_net(skb->dev);
345
346 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
347 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
348
349 if (sk == NULL) {
350 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
351 ICMP6_MIB_INERRORS);
352 return;
353 }
354
355 if (sk->sk_state == TCP_TIME_WAIT) {
356 inet_twsk_put(inet_twsk(sk));
357 return;
358 }
359
360 bh_lock_sock(sk);
361 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
362 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
363
364 if (sk->sk_state == TCP_CLOSE)
365 goto out;
366
367 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
368 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
369 goto out;
370 }
371
372 tp = tcp_sk(sk);
373 seq = ntohl(th->seq);
374 if (sk->sk_state != TCP_LISTEN &&
375 !between(seq, tp->snd_una, tp->snd_nxt)) {
376 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
377 goto out;
378 }
379
380 np = inet6_sk(sk);
381
382 if (type == NDISC_REDIRECT) {
383 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
384
385 if (dst)
386 dst->ops->redirect(dst, sk, skb);
387 goto out;
388 }
389
390 if (type == ICMPV6_PKT_TOOBIG) {
391 /* We are not interested in TCP_LISTEN and open_requests
392 * (SYN-ACKs send out by Linux are always <576bytes so
393 * they should go through unfragmented).
394 */
395 if (sk->sk_state == TCP_LISTEN)
396 goto out;
397
398 if (!ip6_sk_accept_pmtu(sk))
399 goto out;
400
401 tp->mtu_info = ntohl(info);
402 if (!sock_owned_by_user(sk))
403 tcp_v6_mtu_reduced(sk);
404 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
405 &tp->tsq_flags))
406 sock_hold(sk);
407 goto out;
408 }
409
410 icmpv6_err_convert(type, code, &err);
411
412 /* Might be for an request_sock */
413 switch (sk->sk_state) {
414 struct request_sock *req, **prev;
415 case TCP_LISTEN:
416 if (sock_owned_by_user(sk))
417 goto out;
418
419 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
420 &hdr->saddr, inet6_iif(skb));
421 if (!req)
422 goto out;
423
424 /* ICMPs are not backlogged, hence we cannot get
425 * an established socket here.
426 */
427 WARN_ON(req->sk != NULL);
428
429 if (seq != tcp_rsk(req)->snt_isn) {
430 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
431 goto out;
432 }
433
434 inet_csk_reqsk_queue_drop(sk, req, prev);
435 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
436 goto out;
437
438 case TCP_SYN_SENT:
439 case TCP_SYN_RECV: /* Cannot happen.
440 It can, it SYNs are crossed. --ANK */
441 if (!sock_owned_by_user(sk)) {
442 sk->sk_err = err;
443 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
444
445 tcp_done(sk);
446 } else
447 sk->sk_err_soft = err;
448 goto out;
449 }
450
451 if (!sock_owned_by_user(sk) && np->recverr) {
452 sk->sk_err = err;
453 sk->sk_error_report(sk);
454 } else
455 sk->sk_err_soft = err;
456
457out:
458 bh_unlock_sock(sk);
459 sock_put(sk);
460}
461
462
463static int tcp_v6_send_synack(struct sock *sk, struct dst_entry *dst,
464 struct flowi6 *fl6,
465 struct request_sock *req,
466 u16 queue_mapping)
467{
468 struct inet_request_sock *ireq = inet_rsk(req);
469 struct ipv6_pinfo *np = inet6_sk(sk);
470 struct sk_buff *skb;
471 int err = -ENOMEM;
472
473 /* First, grab a route. */
474 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
475 goto done;
476
477 skb = tcp_make_synack(sk, dst, req, NULL);
478
479 if (skb) {
480 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
481 &ireq->ir_v6_rmt_addr);
482
483 fl6->daddr = ireq->ir_v6_rmt_addr;
484 if (np->repflow && (ireq->pktopts != NULL))
485 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
486
487 skb_set_queue_mapping(skb, queue_mapping);
488 err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass);
489 err = net_xmit_eval(err);
490 }
491
492done:
493 return err;
494}
495
496static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req)
497{
498 struct flowi6 fl6;
499 int res;
500
501 res = tcp_v6_send_synack(sk, NULL, &fl6, req, 0);
502 if (!res) {
503 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
504 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
505 }
506 return res;
507}
508
509static void tcp_v6_reqsk_destructor(struct request_sock *req)
510{
511 kfree_skb(inet_rsk(req)->pktopts);
512}
513
514#ifdef CONFIG_TCP_MD5SIG
515static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
516 const struct in6_addr *addr)
517{
518 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
519}
520
521static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
522 struct sock *addr_sk)
523{
524 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
525}
526
527static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
528 struct request_sock *req)
529{
530 return tcp_v6_md5_do_lookup(sk, &inet_rsk(req)->ir_v6_rmt_addr);
531}
532
533static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
534 int optlen)
535{
536 struct tcp_md5sig cmd;
537 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
538
539 if (optlen < sizeof(cmd))
540 return -EINVAL;
541
542 if (copy_from_user(&cmd, optval, sizeof(cmd)))
543 return -EFAULT;
544
545 if (sin6->sin6_family != AF_INET6)
546 return -EINVAL;
547
548 if (!cmd.tcpm_keylen) {
549 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
550 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
551 AF_INET);
552 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
553 AF_INET6);
554 }
555
556 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
557 return -EINVAL;
558
559 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
560 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
561 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
562
563 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
564 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
565}
566
567static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
568 const struct in6_addr *daddr,
569 const struct in6_addr *saddr, int nbytes)
570{
571 struct tcp6_pseudohdr *bp;
572 struct scatterlist sg;
573
574 bp = &hp->md5_blk.ip6;
575 /* 1. TCP pseudo-header (RFC2460) */
576 bp->saddr = *saddr;
577 bp->daddr = *daddr;
578 bp->protocol = cpu_to_be32(IPPROTO_TCP);
579 bp->len = cpu_to_be32(nbytes);
580
581 sg_init_one(&sg, bp, sizeof(*bp));
582 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
583}
584
585static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
586 const struct in6_addr *daddr, struct in6_addr *saddr,
587 const struct tcphdr *th)
588{
589 struct tcp_md5sig_pool *hp;
590 struct hash_desc *desc;
591
592 hp = tcp_get_md5sig_pool();
593 if (!hp)
594 goto clear_hash_noput;
595 desc = &hp->md5_desc;
596
597 if (crypto_hash_init(desc))
598 goto clear_hash;
599 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
600 goto clear_hash;
601 if (tcp_md5_hash_header(hp, th))
602 goto clear_hash;
603 if (tcp_md5_hash_key(hp, key))
604 goto clear_hash;
605 if (crypto_hash_final(desc, md5_hash))
606 goto clear_hash;
607
608 tcp_put_md5sig_pool();
609 return 0;
610
611clear_hash:
612 tcp_put_md5sig_pool();
613clear_hash_noput:
614 memset(md5_hash, 0, 16);
615 return 1;
616}
617
618static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
619 const struct sock *sk,
620 const struct request_sock *req,
621 const struct sk_buff *skb)
622{
623 const struct in6_addr *saddr, *daddr;
624 struct tcp_md5sig_pool *hp;
625 struct hash_desc *desc;
626 const struct tcphdr *th = tcp_hdr(skb);
627
628 if (sk) {
629 saddr = &inet6_sk(sk)->saddr;
630 daddr = &sk->sk_v6_daddr;
631 } else if (req) {
632 saddr = &inet_rsk(req)->ir_v6_loc_addr;
633 daddr = &inet_rsk(req)->ir_v6_rmt_addr;
634 } else {
635 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
636 saddr = &ip6h->saddr;
637 daddr = &ip6h->daddr;
638 }
639
640 hp = tcp_get_md5sig_pool();
641 if (!hp)
642 goto clear_hash_noput;
643 desc = &hp->md5_desc;
644
645 if (crypto_hash_init(desc))
646 goto clear_hash;
647
648 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
649 goto clear_hash;
650 if (tcp_md5_hash_header(hp, th))
651 goto clear_hash;
652 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
653 goto clear_hash;
654 if (tcp_md5_hash_key(hp, key))
655 goto clear_hash;
656 if (crypto_hash_final(desc, md5_hash))
657 goto clear_hash;
658
659 tcp_put_md5sig_pool();
660 return 0;
661
662clear_hash:
663 tcp_put_md5sig_pool();
664clear_hash_noput:
665 memset(md5_hash, 0, 16);
666 return 1;
667}
668
669static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
670{
671 const __u8 *hash_location = NULL;
672 struct tcp_md5sig_key *hash_expected;
673 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
674 const struct tcphdr *th = tcp_hdr(skb);
675 int genhash;
676 u8 newhash[16];
677
678 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
679 hash_location = tcp_parse_md5sig_option(th);
680
681 /* We've parsed the options - do we have a hash? */
682 if (!hash_expected && !hash_location)
683 return 0;
684
685 if (hash_expected && !hash_location) {
686 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
687 return 1;
688 }
689
690 if (!hash_expected && hash_location) {
691 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
692 return 1;
693 }
694
695 /* check the signature */
696 genhash = tcp_v6_md5_hash_skb(newhash,
697 hash_expected,
698 NULL, NULL, skb);
699
700 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
701 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
702 genhash ? "failed" : "mismatch",
703 &ip6h->saddr, ntohs(th->source),
704 &ip6h->daddr, ntohs(th->dest));
705 return 1;
706 }
707 return 0;
708}
709#endif
710
711struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
712 .family = AF_INET6,
713 .obj_size = sizeof(struct tcp6_request_sock),
714 .rtx_syn_ack = tcp_v6_rtx_synack,
715 .send_ack = tcp_v6_reqsk_send_ack,
716 .destructor = tcp_v6_reqsk_destructor,
717 .send_reset = tcp_v6_send_reset,
718 .syn_ack_timeout = tcp_syn_ack_timeout,
719};
720
721#ifdef CONFIG_TCP_MD5SIG
722static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
723 .md5_lookup = tcp_v6_reqsk_md5_lookup,
724 .calc_md5_hash = tcp_v6_md5_hash_skb,
725};
726#endif
727
728static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
729 u32 tsval, u32 tsecr, int oif,
730 struct tcp_md5sig_key *key, int rst, u8 tclass,
731 u32 label)
732{
733 const struct tcphdr *th = tcp_hdr(skb);
734 struct tcphdr *t1;
735 struct sk_buff *buff;
736 struct flowi6 fl6;
737 struct net *net = dev_net(skb_dst(skb)->dev);
738 struct sock *ctl_sk = net->ipv6.tcp_sk;
739 unsigned int tot_len = sizeof(struct tcphdr);
740 struct dst_entry *dst;
741 __be32 *topt;
742
743 if (tsecr)
744 tot_len += TCPOLEN_TSTAMP_ALIGNED;
745#ifdef CONFIG_TCP_MD5SIG
746 if (key)
747 tot_len += TCPOLEN_MD5SIG_ALIGNED;
748#endif
749
750 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
751 GFP_ATOMIC);
752 if (buff == NULL)
753 return;
754
755 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
756
757 t1 = (struct tcphdr *) skb_push(buff, tot_len);
758 skb_reset_transport_header(buff);
759
760 /* Swap the send and the receive. */
761 memset(t1, 0, sizeof(*t1));
762 t1->dest = th->source;
763 t1->source = th->dest;
764 t1->doff = tot_len / 4;
765 t1->seq = htonl(seq);
766 t1->ack_seq = htonl(ack);
767 t1->ack = !rst || !th->ack;
768 t1->rst = rst;
769 t1->window = htons(win);
770
771 topt = (__be32 *)(t1 + 1);
772
773 if (tsecr) {
774 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
775 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
776 *topt++ = htonl(tsval);
777 *topt++ = htonl(tsecr);
778 }
779
780#ifdef CONFIG_TCP_MD5SIG
781 if (key) {
782 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
783 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
784 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
785 &ipv6_hdr(skb)->saddr,
786 &ipv6_hdr(skb)->daddr, t1);
787 }
788#endif
789
790 memset(&fl6, 0, sizeof(fl6));
791 fl6.daddr = ipv6_hdr(skb)->saddr;
792 fl6.saddr = ipv6_hdr(skb)->daddr;
793 fl6.flowlabel = label;
794
795 buff->ip_summed = CHECKSUM_PARTIAL;
796 buff->csum = 0;
797
798 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
799
800 fl6.flowi6_proto = IPPROTO_TCP;
801 if (rt6_need_strict(&fl6.daddr) && !oif)
802 fl6.flowi6_oif = inet6_iif(skb);
803 else
804 fl6.flowi6_oif = oif;
805 fl6.fl6_dport = t1->dest;
806 fl6.fl6_sport = t1->source;
807 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
808
809 /* Pass a socket to ip6_dst_lookup either it is for RST
810 * Underlying function will use this to retrieve the network
811 * namespace
812 */
813 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
814 if (!IS_ERR(dst)) {
815 skb_dst_set(buff, dst);
816 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
817 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
818 if (rst)
819 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
820 return;
821 }
822
823 kfree_skb(buff);
824}
825
826static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
827{
828 const struct tcphdr *th = tcp_hdr(skb);
829 u32 seq = 0, ack_seq = 0;
830 struct tcp_md5sig_key *key = NULL;
831#ifdef CONFIG_TCP_MD5SIG
832 const __u8 *hash_location = NULL;
833 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
834 unsigned char newhash[16];
835 int genhash;
836 struct sock *sk1 = NULL;
837#endif
838 int oif;
839
840 if (th->rst)
841 return;
842
843 if (!ipv6_unicast_destination(skb))
844 return;
845
846#ifdef CONFIG_TCP_MD5SIG
847 hash_location = tcp_parse_md5sig_option(th);
848 if (!sk && hash_location) {
849 /*
850 * active side is lost. Try to find listening socket through
851 * source port, and then find md5 key through listening socket.
852 * we are not loose security here:
853 * Incoming packet is checked with md5 hash with finding key,
854 * no RST generated if md5 hash doesn't match.
855 */
856 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
857 &tcp_hashinfo, &ipv6h->saddr,
858 th->source, &ipv6h->daddr,
859 ntohs(th->source), inet6_iif(skb));
860 if (!sk1)
861 return;
862
863 rcu_read_lock();
864 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
865 if (!key)
866 goto release_sk1;
867
868 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
869 if (genhash || memcmp(hash_location, newhash, 16) != 0)
870 goto release_sk1;
871 } else {
872 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
873 }
874#endif
875
876 if (th->ack)
877 seq = ntohl(th->ack_seq);
878 else
879 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
880 (th->doff << 2);
881
882 oif = sk ? sk->sk_bound_dev_if : 0;
883 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
884
885#ifdef CONFIG_TCP_MD5SIG
886release_sk1:
887 if (sk1) {
888 rcu_read_unlock();
889 sock_put(sk1);
890 }
891#endif
892}
893
894static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
895 u32 win, u32 tsval, u32 tsecr, int oif,
896 struct tcp_md5sig_key *key, u8 tclass,
897 u32 label)
898{
899 tcp_v6_send_response(skb, seq, ack, win, tsval, tsecr, oif, key, 0, tclass,
900 label);
901}
902
903static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
904{
905 struct inet_timewait_sock *tw = inet_twsk(sk);
906 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
907
908 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
909 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
910 tcp_time_stamp + tcptw->tw_ts_offset,
911 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
912 tw->tw_tclass, (tw->tw_flowlabel << 12));
913
914 inet_twsk_put(tw);
915}
916
917static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
918 struct request_sock *req)
919{
920 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1,
921 req->rcv_wnd, tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
922 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
923 0, 0);
924}
925
926
927static struct sock *tcp_v6_hnd_req(struct sock *sk, struct sk_buff *skb)
928{
929 struct request_sock *req, **prev;
930 const struct tcphdr *th = tcp_hdr(skb);
931 struct sock *nsk;
932
933 /* Find possible connection requests. */
934 req = inet6_csk_search_req(sk, &prev, th->source,
935 &ipv6_hdr(skb)->saddr,
936 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
937 if (req)
938 return tcp_check_req(sk, skb, req, prev, false);
939
940 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
941 &ipv6_hdr(skb)->saddr, th->source,
942 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
943
944 if (nsk) {
945 if (nsk->sk_state != TCP_TIME_WAIT) {
946 bh_lock_sock(nsk);
947 return nsk;
948 }
949 inet_twsk_put(inet_twsk(nsk));
950 return NULL;
951 }
952
953#ifdef CONFIG_SYN_COOKIES
954 if (!th->syn)
955 sk = cookie_v6_check(sk, skb);
956#endif
957 return sk;
958}
959
960/* FIXME: this is substantially similar to the ipv4 code.
961 * Can some kind of merge be done? -- erics
962 */
963static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
964{
965 struct tcp_options_received tmp_opt;
966 struct request_sock *req;
967 struct inet_request_sock *ireq;
968 struct ipv6_pinfo *np = inet6_sk(sk);
969 struct tcp_sock *tp = tcp_sk(sk);
970 __u32 isn = TCP_SKB_CB(skb)->when;
971 struct dst_entry *dst = NULL;
972 struct flowi6 fl6;
973 bool want_cookie = false;
974
975 if (skb->protocol == htons(ETH_P_IP))
976 return tcp_v4_conn_request(sk, skb);
977
978 if (!ipv6_unicast_destination(skb))
979 goto drop;
980
981 if ((sysctl_tcp_syncookies == 2 ||
982 inet_csk_reqsk_queue_is_full(sk)) && !isn) {
983 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
984 if (!want_cookie)
985 goto drop;
986 }
987
988 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
989 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
990 goto drop;
991 }
992
993 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
994 if (req == NULL)
995 goto drop;
996
997#ifdef CONFIG_TCP_MD5SIG
998 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
999#endif
1000
1001 tcp_clear_options(&tmp_opt);
1002 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1003 tmp_opt.user_mss = tp->rx_opt.user_mss;
1004 tcp_parse_options(skb, &tmp_opt, 0, NULL);
1005
1006 if (want_cookie && !tmp_opt.saw_tstamp)
1007 tcp_clear_options(&tmp_opt);
1008
1009 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1010 tcp_openreq_init(req, &tmp_opt, skb);
1011
1012 ireq = inet_rsk(req);
1013 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
1014 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
1015 if (!want_cookie || tmp_opt.tstamp_ok)
1016 TCP_ECN_create_request(req, skb, sock_net(sk));
1017
1018 ireq->ir_iif = sk->sk_bound_dev_if;
1019
1020 /* So that link locals have meaning */
1021 if (!sk->sk_bound_dev_if &&
1022 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
1023 ireq->ir_iif = inet6_iif(skb);
1024
1025 if (!isn) {
1026 if (ipv6_opt_accepted(sk, skb) ||
1027 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1028 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim ||
1029 np->repflow) {
1030 atomic_inc(&skb->users);
1031 ireq->pktopts = skb;
1032 }
1033
1034 if (want_cookie) {
1035 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1036 req->cookie_ts = tmp_opt.tstamp_ok;
1037 goto have_isn;
1038 }
1039
1040 /* VJ's idea. We save last timestamp seen
1041 * from the destination in peer table, when entering
1042 * state TIME-WAIT, and check against it before
1043 * accepting new connection request.
1044 *
1045 * If "isn" is not zero, this request hit alive
1046 * timewait bucket, so that all the necessary checks
1047 * are made in the function processing timewait state.
1048 */
1049 if (tmp_opt.saw_tstamp &&
1050 tcp_death_row.sysctl_tw_recycle &&
1051 (dst = inet6_csk_route_req(sk, &fl6, req)) != NULL) {
1052 if (!tcp_peer_is_proven(req, dst, true)) {
1053 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1054 goto drop_and_release;
1055 }
1056 }
1057 /* Kill the following clause, if you dislike this way. */
1058 else if (!sysctl_tcp_syncookies &&
1059 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1060 (sysctl_max_syn_backlog >> 2)) &&
1061 !tcp_peer_is_proven(req, dst, false)) {
1062 /* Without syncookies last quarter of
1063 * backlog is filled with destinations,
1064 * proven to be alive.
1065 * It means that we continue to communicate
1066 * to destinations, already remembered
1067 * to the moment of synflood.
1068 */
1069 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1070 &ireq->ir_v6_rmt_addr, ntohs(tcp_hdr(skb)->source));
1071 goto drop_and_release;
1072 }
1073
1074 isn = tcp_v6_init_sequence(skb);
1075 }
1076have_isn:
1077 tcp_rsk(req)->snt_isn = isn;
1078
1079 if (security_inet_conn_request(sk, skb, req))
1080 goto drop_and_release;
1081
1082 if (tcp_v6_send_synack(sk, dst, &fl6, req,
1083 skb_get_queue_mapping(skb)) ||
1084 want_cookie)
1085 goto drop_and_free;
1086
1087 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1088 tcp_rsk(req)->listener = NULL;
1089 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1090 return 0;
1091
1092drop_and_release:
1093 dst_release(dst);
1094drop_and_free:
1095 reqsk_free(req);
1096drop:
1097 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1098 return 0; /* don't send reset */
1099}
1100
1101static struct sock *tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1102 struct request_sock *req,
1103 struct dst_entry *dst)
1104{
1105 struct inet_request_sock *ireq;
1106 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1107 struct tcp6_sock *newtcp6sk;
1108 struct inet_sock *newinet;
1109 struct tcp_sock *newtp;
1110 struct sock *newsk;
1111#ifdef CONFIG_TCP_MD5SIG
1112 struct tcp_md5sig_key *key;
1113#endif
1114 struct flowi6 fl6;
1115
1116 if (skb->protocol == htons(ETH_P_IP)) {
1117 /*
1118 * v6 mapped
1119 */
1120
1121 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1122
1123 if (newsk == NULL)
1124 return NULL;
1125
1126 newtcp6sk = (struct tcp6_sock *)newsk;
1127 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1128
1129 newinet = inet_sk(newsk);
1130 newnp = inet6_sk(newsk);
1131 newtp = tcp_sk(newsk);
1132
1133 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1134
1135 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newsk->sk_v6_daddr);
1136
1137 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1138
1139 newsk->sk_v6_rcv_saddr = newnp->saddr;
1140
1141 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1142 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1143#ifdef CONFIG_TCP_MD5SIG
1144 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1145#endif
1146
1147 newnp->ipv6_ac_list = NULL;
1148 newnp->ipv6_fl_list = NULL;
1149 newnp->pktoptions = NULL;
1150 newnp->opt = NULL;
1151 newnp->mcast_oif = inet6_iif(skb);
1152 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1153 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1154 if (np->repflow)
1155 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1156
1157 /*
1158 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1159 * here, tcp_create_openreq_child now does this for us, see the comment in
1160 * that function for the gory details. -acme
1161 */
1162
1163 /* It is tricky place. Until this moment IPv4 tcp
1164 worked with IPv6 icsk.icsk_af_ops.
1165 Sync it now.
1166 */
1167 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1168
1169 return newsk;
1170 }
1171
1172 ireq = inet_rsk(req);
1173
1174 if (sk_acceptq_is_full(sk))
1175 goto out_overflow;
1176
1177 if (!dst) {
1178 dst = inet6_csk_route_req(sk, &fl6, req);
1179 if (!dst)
1180 goto out;
1181 }
1182
1183 newsk = tcp_create_openreq_child(sk, req, skb);
1184 if (newsk == NULL)
1185 goto out_nonewsk;
1186
1187 /*
1188 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1189 * count here, tcp_create_openreq_child now does this for us, see the
1190 * comment in that function for the gory details. -acme
1191 */
1192
1193 newsk->sk_gso_type = SKB_GSO_TCPV6;
1194 __ip6_dst_store(newsk, dst, NULL, NULL);
1195 inet6_sk_rx_dst_set(newsk, skb);
1196
1197 newtcp6sk = (struct tcp6_sock *)newsk;
1198 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1199
1200 newtp = tcp_sk(newsk);
1201 newinet = inet_sk(newsk);
1202 newnp = inet6_sk(newsk);
1203
1204 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1205
1206 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1207 newnp->saddr = ireq->ir_v6_loc_addr;
1208 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1209 newsk->sk_bound_dev_if = ireq->ir_iif;
1210
1211 /* Now IPv6 options...
1212
1213 First: no IPv4 options.
1214 */
1215 newinet->inet_opt = NULL;
1216 newnp->ipv6_ac_list = NULL;
1217 newnp->ipv6_fl_list = NULL;
1218
1219 /* Clone RX bits */
1220 newnp->rxopt.all = np->rxopt.all;
1221
1222 /* Clone pktoptions received with SYN */
1223 newnp->pktoptions = NULL;
1224 if (ireq->pktopts != NULL) {
1225 newnp->pktoptions = skb_clone(ireq->pktopts,
1226 sk_gfp_atomic(sk, GFP_ATOMIC));
1227 consume_skb(ireq->pktopts);
1228 ireq->pktopts = NULL;
1229 if (newnp->pktoptions)
1230 skb_set_owner_r(newnp->pktoptions, newsk);
1231 }
1232 newnp->opt = NULL;
1233 newnp->mcast_oif = inet6_iif(skb);
1234 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1235 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1236 if (np->repflow)
1237 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1238
1239 /* Clone native IPv6 options from listening socket (if any)
1240
1241 Yes, keeping reference count would be much more clever,
1242 but we make one more one thing there: reattach optmem
1243 to newsk.
1244 */
1245 if (np->opt)
1246 newnp->opt = ipv6_dup_options(newsk, np->opt);
1247
1248 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1249 if (newnp->opt)
1250 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1251 newnp->opt->opt_flen);
1252
1253 tcp_sync_mss(newsk, dst_mtu(dst));
1254 newtp->advmss = dst_metric_advmss(dst);
1255 if (tcp_sk(sk)->rx_opt.user_mss &&
1256 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1257 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1258
1259 tcp_initialize_rcv_mss(newsk);
1260
1261 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1262 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1263
1264#ifdef CONFIG_TCP_MD5SIG
1265 /* Copy over the MD5 key from the original socket */
1266 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1267 if (key != NULL) {
1268 /* We're using one, so create a matching key
1269 * on the newsk structure. If we fail to get
1270 * memory, then we end up not copying the key
1271 * across. Shucks.
1272 */
1273 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1274 AF_INET6, key->key, key->keylen,
1275 sk_gfp_atomic(sk, GFP_ATOMIC));
1276 }
1277#endif
1278
1279 if (__inet_inherit_port(sk, newsk) < 0) {
1280 inet_csk_prepare_forced_close(newsk);
1281 tcp_done(newsk);
1282 goto out;
1283 }
1284 __inet6_hash(newsk, NULL);
1285
1286 return newsk;
1287
1288out_overflow:
1289 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1290out_nonewsk:
1291 dst_release(dst);
1292out:
1293 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1294 return NULL;
1295}
1296
1297static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1298{
1299 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1300 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1301 &ipv6_hdr(skb)->daddr, skb->csum)) {
1302 skb->ip_summed = CHECKSUM_UNNECESSARY;
1303 return 0;
1304 }
1305 }
1306
1307 skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1308 &ipv6_hdr(skb)->saddr,
1309 &ipv6_hdr(skb)->daddr, 0));
1310
1311 if (skb->len <= 76)
1312 return __skb_checksum_complete(skb);
1313 return 0;
1314}
1315
1316/* The socket must have it's spinlock held when we get
1317 * here.
1318 *
1319 * We have a potential double-lock case here, so even when
1320 * doing backlog processing we use the BH locking scheme.
1321 * This is because we cannot sleep with the original spinlock
1322 * held.
1323 */
1324static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1325{
1326 struct ipv6_pinfo *np = inet6_sk(sk);
1327 struct tcp_sock *tp;
1328 struct sk_buff *opt_skb = NULL;
1329
1330 /* Imagine: socket is IPv6. IPv4 packet arrives,
1331 goes to IPv4 receive handler and backlogged.
1332 From backlog it always goes here. Kerboom...
1333 Fortunately, tcp_rcv_established and rcv_established
1334 handle them correctly, but it is not case with
1335 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1336 */
1337
1338 if (skb->protocol == htons(ETH_P_IP))
1339 return tcp_v4_do_rcv(sk, skb);
1340
1341#ifdef CONFIG_TCP_MD5SIG
1342 if (tcp_v6_inbound_md5_hash(sk, skb))
1343 goto discard;
1344#endif
1345
1346 if (sk_filter(sk, skb))
1347 goto discard;
1348
1349 /*
1350 * socket locking is here for SMP purposes as backlog rcv
1351 * is currently called with bh processing disabled.
1352 */
1353
1354 /* Do Stevens' IPV6_PKTOPTIONS.
1355
1356 Yes, guys, it is the only place in our code, where we
1357 may make it not affecting IPv4.
1358 The rest of code is protocol independent,
1359 and I do not like idea to uglify IPv4.
1360
1361 Actually, all the idea behind IPV6_PKTOPTIONS
1362 looks not very well thought. For now we latch
1363 options, received in the last packet, enqueued
1364 by tcp. Feel free to propose better solution.
1365 --ANK (980728)
1366 */
1367 if (np->rxopt.all)
1368 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1369
1370 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1371 struct dst_entry *dst = sk->sk_rx_dst;
1372
1373 sock_rps_save_rxhash(sk, skb);
1374 if (dst) {
1375 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1376 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1377 dst_release(dst);
1378 sk->sk_rx_dst = NULL;
1379 }
1380 }
1381
1382 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1383 if (opt_skb)
1384 goto ipv6_pktoptions;
1385 return 0;
1386 }
1387
1388 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1389 goto csum_err;
1390
1391 if (sk->sk_state == TCP_LISTEN) {
1392 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1393 if (!nsk)
1394 goto discard;
1395
1396 /*
1397 * Queue it on the new socket if the new socket is active,
1398 * otherwise we just shortcircuit this and continue with
1399 * the new socket..
1400 */
1401 if (nsk != sk) {
1402 sock_rps_save_rxhash(nsk, skb);
1403 if (tcp_child_process(sk, nsk, skb))
1404 goto reset;
1405 if (opt_skb)
1406 __kfree_skb(opt_skb);
1407 return 0;
1408 }
1409 } else
1410 sock_rps_save_rxhash(sk, skb);
1411
1412 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1413 goto reset;
1414 if (opt_skb)
1415 goto ipv6_pktoptions;
1416 return 0;
1417
1418reset:
1419 tcp_v6_send_reset(sk, skb);
1420discard:
1421 if (opt_skb)
1422 __kfree_skb(opt_skb);
1423 kfree_skb(skb);
1424 return 0;
1425csum_err:
1426 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1427 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1428 goto discard;
1429
1430
1431ipv6_pktoptions:
1432 /* Do you ask, what is it?
1433
1434 1. skb was enqueued by tcp.
1435 2. skb is added to tail of read queue, rather than out of order.
1436 3. socket is not in passive state.
1437 4. Finally, it really contains options, which user wants to receive.
1438 */
1439 tp = tcp_sk(sk);
1440 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1441 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1442 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1443 np->mcast_oif = inet6_iif(opt_skb);
1444 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1445 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1446 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1447 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1448 if (np->repflow)
1449 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1450 if (ipv6_opt_accepted(sk, opt_skb)) {
1451 skb_set_owner_r(opt_skb, sk);
1452 opt_skb = xchg(&np->pktoptions, opt_skb);
1453 } else {
1454 __kfree_skb(opt_skb);
1455 opt_skb = xchg(&np->pktoptions, NULL);
1456 }
1457 }
1458
1459 kfree_skb(opt_skb);
1460 return 0;
1461}
1462
1463static int tcp_v6_rcv(struct sk_buff *skb)
1464{
1465 const struct tcphdr *th;
1466 const struct ipv6hdr *hdr;
1467 struct sock *sk;
1468 int ret;
1469 struct net *net = dev_net(skb->dev);
1470
1471 if (skb->pkt_type != PACKET_HOST)
1472 goto discard_it;
1473
1474 /*
1475 * Count it even if it's bad.
1476 */
1477 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1478
1479 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1480 goto discard_it;
1481
1482 th = tcp_hdr(skb);
1483
1484 if (th->doff < sizeof(struct tcphdr)/4)
1485 goto bad_packet;
1486 if (!pskb_may_pull(skb, th->doff*4))
1487 goto discard_it;
1488
1489 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1490 goto csum_error;
1491
1492 th = tcp_hdr(skb);
1493 hdr = ipv6_hdr(skb);
1494 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1495 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1496 skb->len - th->doff*4);
1497 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1498 TCP_SKB_CB(skb)->when = 0;
1499 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1500 TCP_SKB_CB(skb)->sacked = 0;
1501
1502 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1503 if (!sk)
1504 goto no_tcp_socket;
1505
1506process:
1507 if (sk->sk_state == TCP_TIME_WAIT)
1508 goto do_time_wait;
1509
1510 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1511 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1512 goto discard_and_relse;
1513 }
1514
1515 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1516 goto discard_and_relse;
1517
1518 if (sk_filter(sk, skb))
1519 goto discard_and_relse;
1520
1521 sk_mark_napi_id(sk, skb);
1522 skb->dev = NULL;
1523
1524 bh_lock_sock_nested(sk);
1525 ret = 0;
1526 if (!sock_owned_by_user(sk)) {
1527#ifdef CONFIG_NET_DMA
1528 struct tcp_sock *tp = tcp_sk(sk);
1529 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1530 tp->ucopy.dma_chan = net_dma_find_channel();
1531 if (tp->ucopy.dma_chan)
1532 ret = tcp_v6_do_rcv(sk, skb);
1533 else
1534#endif
1535 {
1536 if (!tcp_prequeue(sk, skb))
1537 ret = tcp_v6_do_rcv(sk, skb);
1538 }
1539 } else if (unlikely(sk_add_backlog(sk, skb,
1540 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1541 bh_unlock_sock(sk);
1542 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1543 goto discard_and_relse;
1544 }
1545 bh_unlock_sock(sk);
1546
1547 sock_put(sk);
1548 return ret ? -1 : 0;
1549
1550no_tcp_socket:
1551 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1552 goto discard_it;
1553
1554 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1555csum_error:
1556 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1557bad_packet:
1558 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1559 } else {
1560 tcp_v6_send_reset(NULL, skb);
1561 }
1562
1563discard_it:
1564 kfree_skb(skb);
1565 return 0;
1566
1567discard_and_relse:
1568 sock_put(sk);
1569 goto discard_it;
1570
1571do_time_wait:
1572 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1573 inet_twsk_put(inet_twsk(sk));
1574 goto discard_it;
1575 }
1576
1577 if (skb->len < (th->doff<<2)) {
1578 inet_twsk_put(inet_twsk(sk));
1579 goto bad_packet;
1580 }
1581 if (tcp_checksum_complete(skb)) {
1582 inet_twsk_put(inet_twsk(sk));
1583 goto csum_error;
1584 }
1585
1586 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1587 case TCP_TW_SYN:
1588 {
1589 struct sock *sk2;
1590
1591 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1592 &ipv6_hdr(skb)->saddr, th->source,
1593 &ipv6_hdr(skb)->daddr,
1594 ntohs(th->dest), inet6_iif(skb));
1595 if (sk2 != NULL) {
1596 struct inet_timewait_sock *tw = inet_twsk(sk);
1597 inet_twsk_deschedule(tw, &tcp_death_row);
1598 inet_twsk_put(tw);
1599 sk = sk2;
1600 goto process;
1601 }
1602 /* Fall through to ACK */
1603 }
1604 case TCP_TW_ACK:
1605 tcp_v6_timewait_ack(sk, skb);
1606 break;
1607 case TCP_TW_RST:
1608 goto no_tcp_socket;
1609 case TCP_TW_SUCCESS:
1610 ;
1611 }
1612 goto discard_it;
1613}
1614
1615static void tcp_v6_early_demux(struct sk_buff *skb)
1616{
1617 const struct ipv6hdr *hdr;
1618 const struct tcphdr *th;
1619 struct sock *sk;
1620
1621 if (skb->pkt_type != PACKET_HOST)
1622 return;
1623
1624 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1625 return;
1626
1627 hdr = ipv6_hdr(skb);
1628 th = tcp_hdr(skb);
1629
1630 if (th->doff < sizeof(struct tcphdr) / 4)
1631 return;
1632
1633 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1634 &hdr->saddr, th->source,
1635 &hdr->daddr, ntohs(th->dest),
1636 inet6_iif(skb));
1637 if (sk) {
1638 skb->sk = sk;
1639 skb->destructor = sock_edemux;
1640 if (sk->sk_state != TCP_TIME_WAIT) {
1641 struct dst_entry *dst = sk->sk_rx_dst;
1642
1643 if (dst)
1644 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1645 if (dst &&
1646 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1647 skb_dst_set_noref(skb, dst);
1648 }
1649 }
1650}
1651
1652static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1653 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1654 .twsk_unique = tcp_twsk_unique,
1655 .twsk_destructor = tcp_twsk_destructor,
1656};
1657
1658static const struct inet_connection_sock_af_ops ipv6_specific = {
1659 .queue_xmit = inet6_csk_xmit,
1660 .send_check = tcp_v6_send_check,
1661 .rebuild_header = inet6_sk_rebuild_header,
1662 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1663 .conn_request = tcp_v6_conn_request,
1664 .syn_recv_sock = tcp_v6_syn_recv_sock,
1665 .net_header_len = sizeof(struct ipv6hdr),
1666 .net_frag_header_len = sizeof(struct frag_hdr),
1667 .setsockopt = ipv6_setsockopt,
1668 .getsockopt = ipv6_getsockopt,
1669 .addr2sockaddr = inet6_csk_addr2sockaddr,
1670 .sockaddr_len = sizeof(struct sockaddr_in6),
1671 .bind_conflict = inet6_csk_bind_conflict,
1672#ifdef CONFIG_COMPAT
1673 .compat_setsockopt = compat_ipv6_setsockopt,
1674 .compat_getsockopt = compat_ipv6_getsockopt,
1675#endif
1676};
1677
1678#ifdef CONFIG_TCP_MD5SIG
1679static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1680 .md5_lookup = tcp_v6_md5_lookup,
1681 .calc_md5_hash = tcp_v6_md5_hash_skb,
1682 .md5_parse = tcp_v6_parse_md5_keys,
1683};
1684#endif
1685
1686/*
1687 * TCP over IPv4 via INET6 API
1688 */
1689static const struct inet_connection_sock_af_ops ipv6_mapped = {
1690 .queue_xmit = ip_queue_xmit,
1691 .send_check = tcp_v4_send_check,
1692 .rebuild_header = inet_sk_rebuild_header,
1693 .sk_rx_dst_set = inet_sk_rx_dst_set,
1694 .conn_request = tcp_v6_conn_request,
1695 .syn_recv_sock = tcp_v6_syn_recv_sock,
1696 .net_header_len = sizeof(struct iphdr),
1697 .setsockopt = ipv6_setsockopt,
1698 .getsockopt = ipv6_getsockopt,
1699 .addr2sockaddr = inet6_csk_addr2sockaddr,
1700 .sockaddr_len = sizeof(struct sockaddr_in6),
1701 .bind_conflict = inet6_csk_bind_conflict,
1702#ifdef CONFIG_COMPAT
1703 .compat_setsockopt = compat_ipv6_setsockopt,
1704 .compat_getsockopt = compat_ipv6_getsockopt,
1705#endif
1706};
1707
1708#ifdef CONFIG_TCP_MD5SIG
1709static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1710 .md5_lookup = tcp_v4_md5_lookup,
1711 .calc_md5_hash = tcp_v4_md5_hash_skb,
1712 .md5_parse = tcp_v6_parse_md5_keys,
1713};
1714#endif
1715
1716/* NOTE: A lot of things set to zero explicitly by call to
1717 * sk_alloc() so need not be done here.
1718 */
1719static int tcp_v6_init_sock(struct sock *sk)
1720{
1721 struct inet_connection_sock *icsk = inet_csk(sk);
1722
1723 tcp_init_sock(sk);
1724
1725 icsk->icsk_af_ops = &ipv6_specific;
1726
1727#ifdef CONFIG_TCP_MD5SIG
1728 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1729#endif
1730
1731 return 0;
1732}
1733
1734static void tcp_v6_destroy_sock(struct sock *sk)
1735{
1736 tcp_v4_destroy_sock(sk);
1737 inet6_destroy_sock(sk);
1738}
1739
1740#ifdef CONFIG_PROC_FS
1741/* Proc filesystem TCPv6 sock list dumping. */
1742static void get_openreq6(struct seq_file *seq,
1743 const struct sock *sk, struct request_sock *req, int i, kuid_t uid)
1744{
1745 int ttd = req->expires - jiffies;
1746 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1747 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1748
1749 if (ttd < 0)
1750 ttd = 0;
1751
1752 seq_printf(seq,
1753 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1754 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1755 i,
1756 src->s6_addr32[0], src->s6_addr32[1],
1757 src->s6_addr32[2], src->s6_addr32[3],
1758 inet_rsk(req)->ir_num,
1759 dest->s6_addr32[0], dest->s6_addr32[1],
1760 dest->s6_addr32[2], dest->s6_addr32[3],
1761 ntohs(inet_rsk(req)->ir_rmt_port),
1762 TCP_SYN_RECV,
1763 0, 0, /* could print option size, but that is af dependent. */
1764 1, /* timers active (only the expire timer) */
1765 jiffies_to_clock_t(ttd),
1766 req->num_timeout,
1767 from_kuid_munged(seq_user_ns(seq), uid),
1768 0, /* non standard timer */
1769 0, /* open_requests have no inode */
1770 0, req);
1771}
1772
1773static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1774{
1775 const struct in6_addr *dest, *src;
1776 __u16 destp, srcp;
1777 int timer_active;
1778 unsigned long timer_expires;
1779 const struct inet_sock *inet = inet_sk(sp);
1780 const struct tcp_sock *tp = tcp_sk(sp);
1781 const struct inet_connection_sock *icsk = inet_csk(sp);
1782
1783 dest = &sp->sk_v6_daddr;
1784 src = &sp->sk_v6_rcv_saddr;
1785 destp = ntohs(inet->inet_dport);
1786 srcp = ntohs(inet->inet_sport);
1787
1788 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1789 timer_active = 1;
1790 timer_expires = icsk->icsk_timeout;
1791 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1792 timer_active = 4;
1793 timer_expires = icsk->icsk_timeout;
1794 } else if (timer_pending(&sp->sk_timer)) {
1795 timer_active = 2;
1796 timer_expires = sp->sk_timer.expires;
1797 } else {
1798 timer_active = 0;
1799 timer_expires = jiffies;
1800 }
1801
1802 seq_printf(seq,
1803 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1804 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1805 i,
1806 src->s6_addr32[0], src->s6_addr32[1],
1807 src->s6_addr32[2], src->s6_addr32[3], srcp,
1808 dest->s6_addr32[0], dest->s6_addr32[1],
1809 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1810 sp->sk_state,
1811 tp->write_seq-tp->snd_una,
1812 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1813 timer_active,
1814 jiffies_delta_to_clock_t(timer_expires - jiffies),
1815 icsk->icsk_retransmits,
1816 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1817 icsk->icsk_probes_out,
1818 sock_i_ino(sp),
1819 atomic_read(&sp->sk_refcnt), sp,
1820 jiffies_to_clock_t(icsk->icsk_rto),
1821 jiffies_to_clock_t(icsk->icsk_ack.ato),
1822 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1823 tp->snd_cwnd,
1824 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
1825 );
1826}
1827
1828static void get_timewait6_sock(struct seq_file *seq,
1829 struct inet_timewait_sock *tw, int i)
1830{
1831 const struct in6_addr *dest, *src;
1832 __u16 destp, srcp;
1833 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
1834
1835 dest = &tw->tw_v6_daddr;
1836 src = &tw->tw_v6_rcv_saddr;
1837 destp = ntohs(tw->tw_dport);
1838 srcp = ntohs(tw->tw_sport);
1839
1840 seq_printf(seq,
1841 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1842 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1843 i,
1844 src->s6_addr32[0], src->s6_addr32[1],
1845 src->s6_addr32[2], src->s6_addr32[3], srcp,
1846 dest->s6_addr32[0], dest->s6_addr32[1],
1847 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1848 tw->tw_substate, 0, 0,
1849 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1850 atomic_read(&tw->tw_refcnt), tw);
1851}
1852
1853static int tcp6_seq_show(struct seq_file *seq, void *v)
1854{
1855 struct tcp_iter_state *st;
1856 struct sock *sk = v;
1857
1858 if (v == SEQ_START_TOKEN) {
1859 seq_puts(seq,
1860 " sl "
1861 "local_address "
1862 "remote_address "
1863 "st tx_queue rx_queue tr tm->when retrnsmt"
1864 " uid timeout inode\n");
1865 goto out;
1866 }
1867 st = seq->private;
1868
1869 switch (st->state) {
1870 case TCP_SEQ_STATE_LISTENING:
1871 case TCP_SEQ_STATE_ESTABLISHED:
1872 if (sk->sk_state == TCP_TIME_WAIT)
1873 get_timewait6_sock(seq, v, st->num);
1874 else
1875 get_tcp6_sock(seq, v, st->num);
1876 break;
1877 case TCP_SEQ_STATE_OPENREQ:
1878 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1879 break;
1880 }
1881out:
1882 return 0;
1883}
1884
1885static const struct file_operations tcp6_afinfo_seq_fops = {
1886 .owner = THIS_MODULE,
1887 .open = tcp_seq_open,
1888 .read = seq_read,
1889 .llseek = seq_lseek,
1890 .release = seq_release_net
1891};
1892
1893static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1894 .name = "tcp6",
1895 .family = AF_INET6,
1896 .seq_fops = &tcp6_afinfo_seq_fops,
1897 .seq_ops = {
1898 .show = tcp6_seq_show,
1899 },
1900};
1901
1902int __net_init tcp6_proc_init(struct net *net)
1903{
1904 return tcp_proc_register(net, &tcp6_seq_afinfo);
1905}
1906
1907void tcp6_proc_exit(struct net *net)
1908{
1909 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1910}
1911#endif
1912
1913static void tcp_v6_clear_sk(struct sock *sk, int size)
1914{
1915 struct inet_sock *inet = inet_sk(sk);
1916
1917 /* we do not want to clear pinet6 field, because of RCU lookups */
1918 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1919
1920 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1921 memset(&inet->pinet6 + 1, 0, size);
1922}
1923
1924struct proto tcpv6_prot = {
1925 .name = "TCPv6",
1926 .owner = THIS_MODULE,
1927 .close = tcp_close,
1928 .connect = tcp_v6_connect,
1929 .disconnect = tcp_disconnect,
1930 .accept = inet_csk_accept,
1931 .ioctl = tcp_ioctl,
1932 .init = tcp_v6_init_sock,
1933 .destroy = tcp_v6_destroy_sock,
1934 .shutdown = tcp_shutdown,
1935 .setsockopt = tcp_setsockopt,
1936 .getsockopt = tcp_getsockopt,
1937 .recvmsg = tcp_recvmsg,
1938 .sendmsg = tcp_sendmsg,
1939 .sendpage = tcp_sendpage,
1940 .backlog_rcv = tcp_v6_do_rcv,
1941 .release_cb = tcp_release_cb,
1942 .mtu_reduced = tcp_v6_mtu_reduced,
1943 .hash = tcp_v6_hash,
1944 .unhash = inet_unhash,
1945 .get_port = inet_csk_get_port,
1946 .enter_memory_pressure = tcp_enter_memory_pressure,
1947 .stream_memory_free = tcp_stream_memory_free,
1948 .sockets_allocated = &tcp_sockets_allocated,
1949 .memory_allocated = &tcp_memory_allocated,
1950 .memory_pressure = &tcp_memory_pressure,
1951 .orphan_count = &tcp_orphan_count,
1952 .sysctl_mem = sysctl_tcp_mem,
1953 .sysctl_wmem = sysctl_tcp_wmem,
1954 .sysctl_rmem = sysctl_tcp_rmem,
1955 .max_header = MAX_TCP_HEADER,
1956 .obj_size = sizeof(struct tcp6_sock),
1957 .slab_flags = SLAB_DESTROY_BY_RCU,
1958 .twsk_prot = &tcp6_timewait_sock_ops,
1959 .rsk_prot = &tcp6_request_sock_ops,
1960 .h.hashinfo = &tcp_hashinfo,
1961 .no_autobind = true,
1962#ifdef CONFIG_COMPAT
1963 .compat_setsockopt = compat_tcp_setsockopt,
1964 .compat_getsockopt = compat_tcp_getsockopt,
1965#endif
1966#ifdef CONFIG_MEMCG_KMEM
1967 .proto_cgroup = tcp_proto_cgroup,
1968#endif
1969 .clear_sk = tcp_v6_clear_sk,
1970};
1971
1972static const struct inet6_protocol tcpv6_protocol = {
1973 .early_demux = tcp_v6_early_demux,
1974 .handler = tcp_v6_rcv,
1975 .err_handler = tcp_v6_err,
1976 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1977};
1978
1979static struct inet_protosw tcpv6_protosw = {
1980 .type = SOCK_STREAM,
1981 .protocol = IPPROTO_TCP,
1982 .prot = &tcpv6_prot,
1983 .ops = &inet6_stream_ops,
1984 .no_check = 0,
1985 .flags = INET_PROTOSW_PERMANENT |
1986 INET_PROTOSW_ICSK,
1987};
1988
1989static int __net_init tcpv6_net_init(struct net *net)
1990{
1991 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1992 SOCK_RAW, IPPROTO_TCP, net);
1993}
1994
1995static void __net_exit tcpv6_net_exit(struct net *net)
1996{
1997 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1998}
1999
2000static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2001{
2002 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2003}
2004
2005static struct pernet_operations tcpv6_net_ops = {
2006 .init = tcpv6_net_init,
2007 .exit = tcpv6_net_exit,
2008 .exit_batch = tcpv6_net_exit_batch,
2009};
2010
2011int __init tcpv6_init(void)
2012{
2013 int ret;
2014
2015 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2016 if (ret)
2017 goto out;
2018
2019 /* register inet6 protocol */
2020 ret = inet6_register_protosw(&tcpv6_protosw);
2021 if (ret)
2022 goto out_tcpv6_protocol;
2023
2024 ret = register_pernet_subsys(&tcpv6_net_ops);
2025 if (ret)
2026 goto out_tcpv6_protosw;
2027out:
2028 return ret;
2029
2030out_tcpv6_protosw:
2031 inet6_unregister_protosw(&tcpv6_protosw);
2032out_tcpv6_protocol:
2033 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2034 goto out;
2035}
2036
2037void tcpv6_exit(void)
2038{
2039 unregister_pernet_subsys(&tcpv6_net_ops);
2040 inet6_unregister_protosw(&tcpv6_protosw);
2041 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2042}