Loading...
1/*
2 * TCP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on:
9 * linux/net/ipv4/tcp.c
10 * linux/net/ipv4/tcp_input.c
11 * linux/net/ipv4/tcp_output.c
12 *
13 * Fixes:
14 * Hideaki YOSHIFUJI : sin6_scope_id support
15 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
16 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
17 * a single port at the same time.
18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
19 *
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
24 */
25
26#include <linux/bottom_half.h>
27#include <linux/module.h>
28#include <linux/errno.h>
29#include <linux/types.h>
30#include <linux/socket.h>
31#include <linux/sockios.h>
32#include <linux/net.h>
33#include <linux/jiffies.h>
34#include <linux/in.h>
35#include <linux/in6.h>
36#include <linux/netdevice.h>
37#include <linux/init.h>
38#include <linux/jhash.h>
39#include <linux/ipsec.h>
40#include <linux/times.h>
41#include <linux/slab.h>
42#include <linux/uaccess.h>
43#include <linux/ipv6.h>
44#include <linux/icmpv6.h>
45#include <linux/random.h>
46
47#include <net/tcp.h>
48#include <net/ndisc.h>
49#include <net/inet6_hashtables.h>
50#include <net/inet6_connection_sock.h>
51#include <net/ipv6.h>
52#include <net/transp_v6.h>
53#include <net/addrconf.h>
54#include <net/ip6_route.h>
55#include <net/ip6_checksum.h>
56#include <net/inet_ecn.h>
57#include <net/protocol.h>
58#include <net/xfrm.h>
59#include <net/snmp.h>
60#include <net/dsfield.h>
61#include <net/timewait_sock.h>
62#include <net/inet_common.h>
63#include <net/secure_seq.h>
64#include <net/busy_poll.h>
65
66#include <linux/proc_fs.h>
67#include <linux/seq_file.h>
68
69#include <crypto/hash.h>
70#include <linux/scatterlist.h>
71
72static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
73static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
74 struct request_sock *req);
75
76static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
77
78static const struct inet_connection_sock_af_ops ipv6_mapped;
79static const struct inet_connection_sock_af_ops ipv6_specific;
80#ifdef CONFIG_TCP_MD5SIG
81static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
82static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
83#else
84static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
85 const struct in6_addr *addr)
86{
87 return NULL;
88}
89#endif
90
91static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
92{
93 struct dst_entry *dst = skb_dst(skb);
94
95 if (dst && dst_hold_safe(dst)) {
96 const struct rt6_info *rt = (const struct rt6_info *)dst;
97
98 sk->sk_rx_dst = dst;
99 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
100 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
101 }
102}
103
104static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
105{
106 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
107 ipv6_hdr(skb)->saddr.s6_addr32,
108 tcp_hdr(skb)->dest,
109 tcp_hdr(skb)->source);
110}
111
112static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
113 int addr_len)
114{
115 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
116 struct inet_sock *inet = inet_sk(sk);
117 struct inet_connection_sock *icsk = inet_csk(sk);
118 struct ipv6_pinfo *np = inet6_sk(sk);
119 struct tcp_sock *tp = tcp_sk(sk);
120 struct in6_addr *saddr = NULL, *final_p, final;
121 struct ipv6_txoptions *opt;
122 struct flowi6 fl6;
123 struct dst_entry *dst;
124 int addr_type;
125 int err;
126
127 if (addr_len < SIN6_LEN_RFC2133)
128 return -EINVAL;
129
130 if (usin->sin6_family != AF_INET6)
131 return -EAFNOSUPPORT;
132
133 memset(&fl6, 0, sizeof(fl6));
134
135 if (np->sndflow) {
136 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
137 IP6_ECN_flow_init(fl6.flowlabel);
138 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
139 struct ip6_flowlabel *flowlabel;
140 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
141 if (!flowlabel)
142 return -EINVAL;
143 fl6_sock_release(flowlabel);
144 }
145 }
146
147 /*
148 * connect() to INADDR_ANY means loopback (BSD'ism).
149 */
150
151 if (ipv6_addr_any(&usin->sin6_addr))
152 usin->sin6_addr.s6_addr[15] = 0x1;
153
154 addr_type = ipv6_addr_type(&usin->sin6_addr);
155
156 if (addr_type & IPV6_ADDR_MULTICAST)
157 return -ENETUNREACH;
158
159 if (addr_type&IPV6_ADDR_LINKLOCAL) {
160 if (addr_len >= sizeof(struct sockaddr_in6) &&
161 usin->sin6_scope_id) {
162 /* If interface is set while binding, indices
163 * must coincide.
164 */
165 if (sk->sk_bound_dev_if &&
166 sk->sk_bound_dev_if != usin->sin6_scope_id)
167 return -EINVAL;
168
169 sk->sk_bound_dev_if = usin->sin6_scope_id;
170 }
171
172 /* Connect to link-local address requires an interface */
173 if (!sk->sk_bound_dev_if)
174 return -EINVAL;
175 }
176
177 if (tp->rx_opt.ts_recent_stamp &&
178 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
179 tp->rx_opt.ts_recent = 0;
180 tp->rx_opt.ts_recent_stamp = 0;
181 tp->write_seq = 0;
182 }
183
184 sk->sk_v6_daddr = usin->sin6_addr;
185 np->flow_label = fl6.flowlabel;
186
187 /*
188 * TCP over IPv4
189 */
190
191 if (addr_type == IPV6_ADDR_MAPPED) {
192 u32 exthdrlen = icsk->icsk_ext_hdr_len;
193 struct sockaddr_in sin;
194
195 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
196
197 if (__ipv6_only_sock(sk))
198 return -ENETUNREACH;
199
200 sin.sin_family = AF_INET;
201 sin.sin_port = usin->sin6_port;
202 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
203
204 icsk->icsk_af_ops = &ipv6_mapped;
205 sk->sk_backlog_rcv = tcp_v4_do_rcv;
206#ifdef CONFIG_TCP_MD5SIG
207 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
208#endif
209
210 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
211
212 if (err) {
213 icsk->icsk_ext_hdr_len = exthdrlen;
214 icsk->icsk_af_ops = &ipv6_specific;
215 sk->sk_backlog_rcv = tcp_v6_do_rcv;
216#ifdef CONFIG_TCP_MD5SIG
217 tp->af_specific = &tcp_sock_ipv6_specific;
218#endif
219 goto failure;
220 }
221 np->saddr = sk->sk_v6_rcv_saddr;
222
223 return err;
224 }
225
226 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
227 saddr = &sk->sk_v6_rcv_saddr;
228
229 fl6.flowi6_proto = IPPROTO_TCP;
230 fl6.daddr = sk->sk_v6_daddr;
231 fl6.saddr = saddr ? *saddr : np->saddr;
232 fl6.flowi6_oif = sk->sk_bound_dev_if;
233 fl6.flowi6_mark = sk->sk_mark;
234 fl6.fl6_dport = usin->sin6_port;
235 fl6.fl6_sport = inet->inet_sport;
236
237 opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
238 final_p = fl6_update_dst(&fl6, opt, &final);
239
240 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
241
242 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
243 if (IS_ERR(dst)) {
244 err = PTR_ERR(dst);
245 goto failure;
246 }
247
248 if (!saddr) {
249 saddr = &fl6.saddr;
250 sk->sk_v6_rcv_saddr = *saddr;
251 }
252
253 /* set the source address */
254 np->saddr = *saddr;
255 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
256
257 sk->sk_gso_type = SKB_GSO_TCPV6;
258 ip6_dst_store(sk, dst, NULL, NULL);
259
260 if (tcp_death_row.sysctl_tw_recycle &&
261 !tp->rx_opt.ts_recent_stamp &&
262 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
263 tcp_fetch_timewait_stamp(sk, dst);
264
265 icsk->icsk_ext_hdr_len = 0;
266 if (opt)
267 icsk->icsk_ext_hdr_len = opt->opt_flen +
268 opt->opt_nflen;
269
270 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
271
272 inet->inet_dport = usin->sin6_port;
273
274 tcp_set_state(sk, TCP_SYN_SENT);
275 err = inet6_hash_connect(&tcp_death_row, sk);
276 if (err)
277 goto late_failure;
278
279 sk_set_txhash(sk);
280
281 if (!tp->write_seq && likely(!tp->repair))
282 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
283 sk->sk_v6_daddr.s6_addr32,
284 inet->inet_sport,
285 inet->inet_dport);
286
287 err = tcp_connect(sk);
288 if (err)
289 goto late_failure;
290
291 return 0;
292
293late_failure:
294 tcp_set_state(sk, TCP_CLOSE);
295 __sk_dst_reset(sk);
296failure:
297 inet->inet_dport = 0;
298 sk->sk_route_caps = 0;
299 return err;
300}
301
302static void tcp_v6_mtu_reduced(struct sock *sk)
303{
304 struct dst_entry *dst;
305
306 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
307 return;
308
309 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
310 if (!dst)
311 return;
312
313 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
314 tcp_sync_mss(sk, dst_mtu(dst));
315 tcp_simple_retransmit(sk);
316 }
317}
318
319static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
320 u8 type, u8 code, int offset, __be32 info)
321{
322 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
323 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
324 struct net *net = dev_net(skb->dev);
325 struct request_sock *fastopen;
326 struct ipv6_pinfo *np;
327 struct tcp_sock *tp;
328 __u32 seq, snd_una;
329 struct sock *sk;
330 bool fatal;
331 int err;
332
333 sk = __inet6_lookup_established(net, &tcp_hashinfo,
334 &hdr->daddr, th->dest,
335 &hdr->saddr, ntohs(th->source),
336 skb->dev->ifindex);
337
338 if (!sk) {
339 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
340 ICMP6_MIB_INERRORS);
341 return;
342 }
343
344 if (sk->sk_state == TCP_TIME_WAIT) {
345 inet_twsk_put(inet_twsk(sk));
346 return;
347 }
348 seq = ntohl(th->seq);
349 fatal = icmpv6_err_convert(type, code, &err);
350 if (sk->sk_state == TCP_NEW_SYN_RECV)
351 return tcp_req_err(sk, seq, fatal);
352
353 bh_lock_sock(sk);
354 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
355 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
356
357 if (sk->sk_state == TCP_CLOSE)
358 goto out;
359
360 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
361 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
362 goto out;
363 }
364
365 tp = tcp_sk(sk);
366 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
367 fastopen = tp->fastopen_rsk;
368 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
369 if (sk->sk_state != TCP_LISTEN &&
370 !between(seq, snd_una, tp->snd_nxt)) {
371 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
372 goto out;
373 }
374
375 np = inet6_sk(sk);
376
377 if (type == NDISC_REDIRECT) {
378 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
379
380 if (dst)
381 dst->ops->redirect(dst, sk, skb);
382 goto out;
383 }
384
385 if (type == ICMPV6_PKT_TOOBIG) {
386 /* We are not interested in TCP_LISTEN and open_requests
387 * (SYN-ACKs send out by Linux are always <576bytes so
388 * they should go through unfragmented).
389 */
390 if (sk->sk_state == TCP_LISTEN)
391 goto out;
392
393 if (!ip6_sk_accept_pmtu(sk))
394 goto out;
395
396 tp->mtu_info = ntohl(info);
397 if (!sock_owned_by_user(sk))
398 tcp_v6_mtu_reduced(sk);
399 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
400 &tp->tsq_flags))
401 sock_hold(sk);
402 goto out;
403 }
404
405
406 /* Might be for an request_sock */
407 switch (sk->sk_state) {
408 case TCP_SYN_SENT:
409 case TCP_SYN_RECV:
410 /* Only in fast or simultaneous open. If a fast open socket is
411 * is already accepted it is treated as a connected one below.
412 */
413 if (fastopen && !fastopen->sk)
414 break;
415
416 if (!sock_owned_by_user(sk)) {
417 sk->sk_err = err;
418 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
419
420 tcp_done(sk);
421 } else
422 sk->sk_err_soft = err;
423 goto out;
424 }
425
426 if (!sock_owned_by_user(sk) && np->recverr) {
427 sk->sk_err = err;
428 sk->sk_error_report(sk);
429 } else
430 sk->sk_err_soft = err;
431
432out:
433 bh_unlock_sock(sk);
434 sock_put(sk);
435}
436
437
438static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
439 struct flowi *fl,
440 struct request_sock *req,
441 struct tcp_fastopen_cookie *foc,
442 bool attach_req)
443{
444 struct inet_request_sock *ireq = inet_rsk(req);
445 struct ipv6_pinfo *np = inet6_sk(sk);
446 struct flowi6 *fl6 = &fl->u.ip6;
447 struct sk_buff *skb;
448 int err = -ENOMEM;
449
450 /* First, grab a route. */
451 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
452 IPPROTO_TCP)) == NULL)
453 goto done;
454
455 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
456
457 if (skb) {
458 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
459 &ireq->ir_v6_rmt_addr);
460
461 fl6->daddr = ireq->ir_v6_rmt_addr;
462 if (np->repflow && ireq->pktopts)
463 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
464
465 rcu_read_lock();
466 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
467 np->tclass);
468 rcu_read_unlock();
469 err = net_xmit_eval(err);
470 }
471
472done:
473 return err;
474}
475
476
477static void tcp_v6_reqsk_destructor(struct request_sock *req)
478{
479 kfree_skb(inet_rsk(req)->pktopts);
480}
481
482#ifdef CONFIG_TCP_MD5SIG
483static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
484 const struct in6_addr *addr)
485{
486 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
487}
488
489static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
490 const struct sock *addr_sk)
491{
492 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
493}
494
495static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
496 int optlen)
497{
498 struct tcp_md5sig cmd;
499 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
500
501 if (optlen < sizeof(cmd))
502 return -EINVAL;
503
504 if (copy_from_user(&cmd, optval, sizeof(cmd)))
505 return -EFAULT;
506
507 if (sin6->sin6_family != AF_INET6)
508 return -EINVAL;
509
510 if (!cmd.tcpm_keylen) {
511 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
512 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
513 AF_INET);
514 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
515 AF_INET6);
516 }
517
518 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
519 return -EINVAL;
520
521 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
522 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
523 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
524
525 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
526 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
527}
528
529static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
530 const struct in6_addr *daddr,
531 const struct in6_addr *saddr, int nbytes)
532{
533 struct tcp6_pseudohdr *bp;
534 struct scatterlist sg;
535
536 bp = &hp->md5_blk.ip6;
537 /* 1. TCP pseudo-header (RFC2460) */
538 bp->saddr = *saddr;
539 bp->daddr = *daddr;
540 bp->protocol = cpu_to_be32(IPPROTO_TCP);
541 bp->len = cpu_to_be32(nbytes);
542
543 sg_init_one(&sg, bp, sizeof(*bp));
544 ahash_request_set_crypt(hp->md5_req, &sg, NULL, sizeof(*bp));
545 return crypto_ahash_update(hp->md5_req);
546}
547
548static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
549 const struct in6_addr *daddr, struct in6_addr *saddr,
550 const struct tcphdr *th)
551{
552 struct tcp_md5sig_pool *hp;
553 struct ahash_request *req;
554
555 hp = tcp_get_md5sig_pool();
556 if (!hp)
557 goto clear_hash_noput;
558 req = hp->md5_req;
559
560 if (crypto_ahash_init(req))
561 goto clear_hash;
562 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
563 goto clear_hash;
564 if (tcp_md5_hash_header(hp, th))
565 goto clear_hash;
566 if (tcp_md5_hash_key(hp, key))
567 goto clear_hash;
568 ahash_request_set_crypt(req, NULL, md5_hash, 0);
569 if (crypto_ahash_final(req))
570 goto clear_hash;
571
572 tcp_put_md5sig_pool();
573 return 0;
574
575clear_hash:
576 tcp_put_md5sig_pool();
577clear_hash_noput:
578 memset(md5_hash, 0, 16);
579 return 1;
580}
581
582static int tcp_v6_md5_hash_skb(char *md5_hash,
583 const struct tcp_md5sig_key *key,
584 const struct sock *sk,
585 const struct sk_buff *skb)
586{
587 const struct in6_addr *saddr, *daddr;
588 struct tcp_md5sig_pool *hp;
589 struct ahash_request *req;
590 const struct tcphdr *th = tcp_hdr(skb);
591
592 if (sk) { /* valid for establish/request sockets */
593 saddr = &sk->sk_v6_rcv_saddr;
594 daddr = &sk->sk_v6_daddr;
595 } else {
596 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
597 saddr = &ip6h->saddr;
598 daddr = &ip6h->daddr;
599 }
600
601 hp = tcp_get_md5sig_pool();
602 if (!hp)
603 goto clear_hash_noput;
604 req = hp->md5_req;
605
606 if (crypto_ahash_init(req))
607 goto clear_hash;
608
609 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
610 goto clear_hash;
611 if (tcp_md5_hash_header(hp, th))
612 goto clear_hash;
613 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
614 goto clear_hash;
615 if (tcp_md5_hash_key(hp, key))
616 goto clear_hash;
617 ahash_request_set_crypt(req, NULL, md5_hash, 0);
618 if (crypto_ahash_final(req))
619 goto clear_hash;
620
621 tcp_put_md5sig_pool();
622 return 0;
623
624clear_hash:
625 tcp_put_md5sig_pool();
626clear_hash_noput:
627 memset(md5_hash, 0, 16);
628 return 1;
629}
630
631#endif
632
633static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
634 const struct sk_buff *skb)
635{
636#ifdef CONFIG_TCP_MD5SIG
637 const __u8 *hash_location = NULL;
638 struct tcp_md5sig_key *hash_expected;
639 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
640 const struct tcphdr *th = tcp_hdr(skb);
641 int genhash;
642 u8 newhash[16];
643
644 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
645 hash_location = tcp_parse_md5sig_option(th);
646
647 /* We've parsed the options - do we have a hash? */
648 if (!hash_expected && !hash_location)
649 return false;
650
651 if (hash_expected && !hash_location) {
652 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
653 return true;
654 }
655
656 if (!hash_expected && hash_location) {
657 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
658 return true;
659 }
660
661 /* check the signature */
662 genhash = tcp_v6_md5_hash_skb(newhash,
663 hash_expected,
664 NULL, skb);
665
666 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
667 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
668 genhash ? "failed" : "mismatch",
669 &ip6h->saddr, ntohs(th->source),
670 &ip6h->daddr, ntohs(th->dest));
671 return true;
672 }
673#endif
674 return false;
675}
676
677static void tcp_v6_init_req(struct request_sock *req,
678 const struct sock *sk_listener,
679 struct sk_buff *skb)
680{
681 struct inet_request_sock *ireq = inet_rsk(req);
682 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
683
684 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
685 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
686
687 /* So that link locals have meaning */
688 if (!sk_listener->sk_bound_dev_if &&
689 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
690 ireq->ir_iif = tcp_v6_iif(skb);
691
692 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
693 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
694 np->rxopt.bits.rxinfo ||
695 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
696 np->rxopt.bits.rxohlim || np->repflow)) {
697 atomic_inc(&skb->users);
698 ireq->pktopts = skb;
699 }
700}
701
702static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
703 struct flowi *fl,
704 const struct request_sock *req,
705 bool *strict)
706{
707 if (strict)
708 *strict = true;
709 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
710}
711
712struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
713 .family = AF_INET6,
714 .obj_size = sizeof(struct tcp6_request_sock),
715 .rtx_syn_ack = tcp_rtx_synack,
716 .send_ack = tcp_v6_reqsk_send_ack,
717 .destructor = tcp_v6_reqsk_destructor,
718 .send_reset = tcp_v6_send_reset,
719 .syn_ack_timeout = tcp_syn_ack_timeout,
720};
721
722static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
723 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
724 sizeof(struct ipv6hdr),
725#ifdef CONFIG_TCP_MD5SIG
726 .req_md5_lookup = tcp_v6_md5_lookup,
727 .calc_md5_hash = tcp_v6_md5_hash_skb,
728#endif
729 .init_req = tcp_v6_init_req,
730#ifdef CONFIG_SYN_COOKIES
731 .cookie_init_seq = cookie_v6_init_sequence,
732#endif
733 .route_req = tcp_v6_route_req,
734 .init_seq = tcp_v6_init_sequence,
735 .send_synack = tcp_v6_send_synack,
736};
737
738static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
739 u32 ack, u32 win, u32 tsval, u32 tsecr,
740 int oif, struct tcp_md5sig_key *key, int rst,
741 u8 tclass, u32 label)
742{
743 const struct tcphdr *th = tcp_hdr(skb);
744 struct tcphdr *t1;
745 struct sk_buff *buff;
746 struct flowi6 fl6;
747 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
748 struct sock *ctl_sk = net->ipv6.tcp_sk;
749 unsigned int tot_len = sizeof(struct tcphdr);
750 struct dst_entry *dst;
751 __be32 *topt;
752
753 if (tsecr)
754 tot_len += TCPOLEN_TSTAMP_ALIGNED;
755#ifdef CONFIG_TCP_MD5SIG
756 if (key)
757 tot_len += TCPOLEN_MD5SIG_ALIGNED;
758#endif
759
760 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
761 GFP_ATOMIC);
762 if (!buff)
763 return;
764
765 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
766
767 t1 = (struct tcphdr *) skb_push(buff, tot_len);
768 skb_reset_transport_header(buff);
769
770 /* Swap the send and the receive. */
771 memset(t1, 0, sizeof(*t1));
772 t1->dest = th->source;
773 t1->source = th->dest;
774 t1->doff = tot_len / 4;
775 t1->seq = htonl(seq);
776 t1->ack_seq = htonl(ack);
777 t1->ack = !rst || !th->ack;
778 t1->rst = rst;
779 t1->window = htons(win);
780
781 topt = (__be32 *)(t1 + 1);
782
783 if (tsecr) {
784 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
785 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
786 *topt++ = htonl(tsval);
787 *topt++ = htonl(tsecr);
788 }
789
790#ifdef CONFIG_TCP_MD5SIG
791 if (key) {
792 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
793 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
794 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
795 &ipv6_hdr(skb)->saddr,
796 &ipv6_hdr(skb)->daddr, t1);
797 }
798#endif
799
800 memset(&fl6, 0, sizeof(fl6));
801 fl6.daddr = ipv6_hdr(skb)->saddr;
802 fl6.saddr = ipv6_hdr(skb)->daddr;
803 fl6.flowlabel = label;
804
805 buff->ip_summed = CHECKSUM_PARTIAL;
806 buff->csum = 0;
807
808 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
809
810 fl6.flowi6_proto = IPPROTO_TCP;
811 if (rt6_need_strict(&fl6.daddr) && !oif)
812 fl6.flowi6_oif = tcp_v6_iif(skb);
813 else {
814 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
815 oif = skb->skb_iif;
816
817 fl6.flowi6_oif = oif;
818 }
819
820 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
821 fl6.fl6_dport = t1->dest;
822 fl6.fl6_sport = t1->source;
823 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
824
825 /* Pass a socket to ip6_dst_lookup either it is for RST
826 * Underlying function will use this to retrieve the network
827 * namespace
828 */
829 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
830 if (!IS_ERR(dst)) {
831 skb_dst_set(buff, dst);
832 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
833 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
834 if (rst)
835 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
836 return;
837 }
838
839 kfree_skb(buff);
840}
841
842static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
843{
844 const struct tcphdr *th = tcp_hdr(skb);
845 u32 seq = 0, ack_seq = 0;
846 struct tcp_md5sig_key *key = NULL;
847#ifdef CONFIG_TCP_MD5SIG
848 const __u8 *hash_location = NULL;
849 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
850 unsigned char newhash[16];
851 int genhash;
852 struct sock *sk1 = NULL;
853#endif
854 int oif;
855
856 if (th->rst)
857 return;
858
859 /* If sk not NULL, it means we did a successful lookup and incoming
860 * route had to be correct. prequeue might have dropped our dst.
861 */
862 if (!sk && !ipv6_unicast_destination(skb))
863 return;
864
865#ifdef CONFIG_TCP_MD5SIG
866 hash_location = tcp_parse_md5sig_option(th);
867 if (sk && sk_fullsock(sk)) {
868 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
869 } else if (hash_location) {
870 /*
871 * active side is lost. Try to find listening socket through
872 * source port, and then find md5 key through listening socket.
873 * we are not loose security here:
874 * Incoming packet is checked with md5 hash with finding key,
875 * no RST generated if md5 hash doesn't match.
876 */
877 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
878 &tcp_hashinfo, NULL, 0,
879 &ipv6h->saddr,
880 th->source, &ipv6h->daddr,
881 ntohs(th->source), tcp_v6_iif(skb));
882 if (!sk1)
883 return;
884
885 rcu_read_lock();
886 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
887 if (!key)
888 goto release_sk1;
889
890 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
891 if (genhash || memcmp(hash_location, newhash, 16) != 0)
892 goto release_sk1;
893 }
894#endif
895
896 if (th->ack)
897 seq = ntohl(th->ack_seq);
898 else
899 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
900 (th->doff << 2);
901
902 oif = sk ? sk->sk_bound_dev_if : 0;
903 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
904
905#ifdef CONFIG_TCP_MD5SIG
906release_sk1:
907 if (sk1) {
908 rcu_read_unlock();
909 sock_put(sk1);
910 }
911#endif
912}
913
914static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
915 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
916 struct tcp_md5sig_key *key, u8 tclass,
917 u32 label)
918{
919 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
920 tclass, label);
921}
922
923static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
924{
925 struct inet_timewait_sock *tw = inet_twsk(sk);
926 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
927
928 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
929 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
930 tcp_time_stamp + tcptw->tw_ts_offset,
931 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
932 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
933
934 inet_twsk_put(tw);
935}
936
937static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
938 struct request_sock *req)
939{
940 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
941 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
942 */
943 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
944 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
945 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
946 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
947 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
948 0, 0);
949}
950
951
952static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
953{
954#ifdef CONFIG_SYN_COOKIES
955 const struct tcphdr *th = tcp_hdr(skb);
956
957 if (!th->syn)
958 sk = cookie_v6_check(sk, skb);
959#endif
960 return sk;
961}
962
963static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
964{
965 if (skb->protocol == htons(ETH_P_IP))
966 return tcp_v4_conn_request(sk, skb);
967
968 if (!ipv6_unicast_destination(skb))
969 goto drop;
970
971 return tcp_conn_request(&tcp6_request_sock_ops,
972 &tcp_request_sock_ipv6_ops, sk, skb);
973
974drop:
975 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
976 return 0; /* don't send reset */
977}
978
979static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
980 struct request_sock *req,
981 struct dst_entry *dst,
982 struct request_sock *req_unhash,
983 bool *own_req)
984{
985 struct inet_request_sock *ireq;
986 struct ipv6_pinfo *newnp;
987 const struct ipv6_pinfo *np = inet6_sk(sk);
988 struct ipv6_txoptions *opt;
989 struct tcp6_sock *newtcp6sk;
990 struct inet_sock *newinet;
991 struct tcp_sock *newtp;
992 struct sock *newsk;
993#ifdef CONFIG_TCP_MD5SIG
994 struct tcp_md5sig_key *key;
995#endif
996 struct flowi6 fl6;
997
998 if (skb->protocol == htons(ETH_P_IP)) {
999 /*
1000 * v6 mapped
1001 */
1002
1003 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1004 req_unhash, own_req);
1005
1006 if (!newsk)
1007 return NULL;
1008
1009 newtcp6sk = (struct tcp6_sock *)newsk;
1010 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1011
1012 newinet = inet_sk(newsk);
1013 newnp = inet6_sk(newsk);
1014 newtp = tcp_sk(newsk);
1015
1016 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1017
1018 newnp->saddr = newsk->sk_v6_rcv_saddr;
1019
1020 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1021 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1022#ifdef CONFIG_TCP_MD5SIG
1023 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1024#endif
1025
1026 newnp->ipv6_ac_list = NULL;
1027 newnp->ipv6_fl_list = NULL;
1028 newnp->pktoptions = NULL;
1029 newnp->opt = NULL;
1030 newnp->mcast_oif = tcp_v6_iif(skb);
1031 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1032 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1033 if (np->repflow)
1034 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1035
1036 /*
1037 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1038 * here, tcp_create_openreq_child now does this for us, see the comment in
1039 * that function for the gory details. -acme
1040 */
1041
1042 /* It is tricky place. Until this moment IPv4 tcp
1043 worked with IPv6 icsk.icsk_af_ops.
1044 Sync it now.
1045 */
1046 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1047
1048 return newsk;
1049 }
1050
1051 ireq = inet_rsk(req);
1052
1053 if (sk_acceptq_is_full(sk))
1054 goto out_overflow;
1055
1056 if (!dst) {
1057 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1058 if (!dst)
1059 goto out;
1060 }
1061
1062 newsk = tcp_create_openreq_child(sk, req, skb);
1063 if (!newsk)
1064 goto out_nonewsk;
1065
1066 /*
1067 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1068 * count here, tcp_create_openreq_child now does this for us, see the
1069 * comment in that function for the gory details. -acme
1070 */
1071
1072 newsk->sk_gso_type = SKB_GSO_TCPV6;
1073 ip6_dst_store(newsk, dst, NULL, NULL);
1074 inet6_sk_rx_dst_set(newsk, skb);
1075
1076 newtcp6sk = (struct tcp6_sock *)newsk;
1077 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1078
1079 newtp = tcp_sk(newsk);
1080 newinet = inet_sk(newsk);
1081 newnp = inet6_sk(newsk);
1082
1083 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1084
1085 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1086 newnp->saddr = ireq->ir_v6_loc_addr;
1087 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1088 newsk->sk_bound_dev_if = ireq->ir_iif;
1089
1090 /* Now IPv6 options...
1091
1092 First: no IPv4 options.
1093 */
1094 newinet->inet_opt = NULL;
1095 newnp->ipv6_ac_list = NULL;
1096 newnp->ipv6_fl_list = NULL;
1097
1098 /* Clone RX bits */
1099 newnp->rxopt.all = np->rxopt.all;
1100
1101 newnp->pktoptions = NULL;
1102 newnp->opt = NULL;
1103 newnp->mcast_oif = tcp_v6_iif(skb);
1104 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1105 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1106 if (np->repflow)
1107 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1108
1109 /* Clone native IPv6 options from listening socket (if any)
1110
1111 Yes, keeping reference count would be much more clever,
1112 but we make one more one thing there: reattach optmem
1113 to newsk.
1114 */
1115 opt = rcu_dereference(np->opt);
1116 if (opt) {
1117 opt = ipv6_dup_options(newsk, opt);
1118 RCU_INIT_POINTER(newnp->opt, opt);
1119 }
1120 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1121 if (opt)
1122 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1123 opt->opt_flen;
1124
1125 tcp_ca_openreq_child(newsk, dst);
1126
1127 tcp_sync_mss(newsk, dst_mtu(dst));
1128 newtp->advmss = dst_metric_advmss(dst);
1129 if (tcp_sk(sk)->rx_opt.user_mss &&
1130 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1131 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1132
1133 tcp_initialize_rcv_mss(newsk);
1134
1135 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1136 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1137
1138#ifdef CONFIG_TCP_MD5SIG
1139 /* Copy over the MD5 key from the original socket */
1140 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1141 if (key) {
1142 /* We're using one, so create a matching key
1143 * on the newsk structure. If we fail to get
1144 * memory, then we end up not copying the key
1145 * across. Shucks.
1146 */
1147 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1148 AF_INET6, key->key, key->keylen,
1149 sk_gfp_mask(sk, GFP_ATOMIC));
1150 }
1151#endif
1152
1153 if (__inet_inherit_port(sk, newsk) < 0) {
1154 inet_csk_prepare_forced_close(newsk);
1155 tcp_done(newsk);
1156 goto out;
1157 }
1158 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1159 if (*own_req) {
1160 tcp_move_syn(newtp, req);
1161
1162 /* Clone pktoptions received with SYN, if we own the req */
1163 if (ireq->pktopts) {
1164 newnp->pktoptions = skb_clone(ireq->pktopts,
1165 sk_gfp_mask(sk, GFP_ATOMIC));
1166 consume_skb(ireq->pktopts);
1167 ireq->pktopts = NULL;
1168 if (newnp->pktoptions)
1169 skb_set_owner_r(newnp->pktoptions, newsk);
1170 }
1171 }
1172
1173 return newsk;
1174
1175out_overflow:
1176 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1177out_nonewsk:
1178 dst_release(dst);
1179out:
1180 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1181 return NULL;
1182}
1183
1184/* The socket must have it's spinlock held when we get
1185 * here, unless it is a TCP_LISTEN socket.
1186 *
1187 * We have a potential double-lock case here, so even when
1188 * doing backlog processing we use the BH locking scheme.
1189 * This is because we cannot sleep with the original spinlock
1190 * held.
1191 */
1192static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1193{
1194 struct ipv6_pinfo *np = inet6_sk(sk);
1195 struct tcp_sock *tp;
1196 struct sk_buff *opt_skb = NULL;
1197
1198 /* Imagine: socket is IPv6. IPv4 packet arrives,
1199 goes to IPv4 receive handler and backlogged.
1200 From backlog it always goes here. Kerboom...
1201 Fortunately, tcp_rcv_established and rcv_established
1202 handle them correctly, but it is not case with
1203 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1204 */
1205
1206 if (skb->protocol == htons(ETH_P_IP))
1207 return tcp_v4_do_rcv(sk, skb);
1208
1209 if (sk_filter(sk, skb))
1210 goto discard;
1211
1212 /*
1213 * socket locking is here for SMP purposes as backlog rcv
1214 * is currently called with bh processing disabled.
1215 */
1216
1217 /* Do Stevens' IPV6_PKTOPTIONS.
1218
1219 Yes, guys, it is the only place in our code, where we
1220 may make it not affecting IPv4.
1221 The rest of code is protocol independent,
1222 and I do not like idea to uglify IPv4.
1223
1224 Actually, all the idea behind IPV6_PKTOPTIONS
1225 looks not very well thought. For now we latch
1226 options, received in the last packet, enqueued
1227 by tcp. Feel free to propose better solution.
1228 --ANK (980728)
1229 */
1230 if (np->rxopt.all)
1231 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1232
1233 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1234 struct dst_entry *dst = sk->sk_rx_dst;
1235
1236 sock_rps_save_rxhash(sk, skb);
1237 sk_mark_napi_id(sk, skb);
1238 if (dst) {
1239 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1240 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1241 dst_release(dst);
1242 sk->sk_rx_dst = NULL;
1243 }
1244 }
1245
1246 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1247 if (opt_skb)
1248 goto ipv6_pktoptions;
1249 return 0;
1250 }
1251
1252 if (tcp_checksum_complete(skb))
1253 goto csum_err;
1254
1255 if (sk->sk_state == TCP_LISTEN) {
1256 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1257
1258 if (!nsk)
1259 goto discard;
1260
1261 if (nsk != sk) {
1262 sock_rps_save_rxhash(nsk, skb);
1263 sk_mark_napi_id(nsk, skb);
1264 if (tcp_child_process(sk, nsk, skb))
1265 goto reset;
1266 if (opt_skb)
1267 __kfree_skb(opt_skb);
1268 return 0;
1269 }
1270 } else
1271 sock_rps_save_rxhash(sk, skb);
1272
1273 if (tcp_rcv_state_process(sk, skb))
1274 goto reset;
1275 if (opt_skb)
1276 goto ipv6_pktoptions;
1277 return 0;
1278
1279reset:
1280 tcp_v6_send_reset(sk, skb);
1281discard:
1282 if (opt_skb)
1283 __kfree_skb(opt_skb);
1284 kfree_skb(skb);
1285 return 0;
1286csum_err:
1287 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1288 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1289 goto discard;
1290
1291
1292ipv6_pktoptions:
1293 /* Do you ask, what is it?
1294
1295 1. skb was enqueued by tcp.
1296 2. skb is added to tail of read queue, rather than out of order.
1297 3. socket is not in passive state.
1298 4. Finally, it really contains options, which user wants to receive.
1299 */
1300 tp = tcp_sk(sk);
1301 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1302 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1303 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1304 np->mcast_oif = tcp_v6_iif(opt_skb);
1305 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1306 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1307 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1308 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1309 if (np->repflow)
1310 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1311 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1312 skb_set_owner_r(opt_skb, sk);
1313 opt_skb = xchg(&np->pktoptions, opt_skb);
1314 } else {
1315 __kfree_skb(opt_skb);
1316 opt_skb = xchg(&np->pktoptions, NULL);
1317 }
1318 }
1319
1320 kfree_skb(opt_skb);
1321 return 0;
1322}
1323
1324static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1325 const struct tcphdr *th)
1326{
1327 /* This is tricky: we move IP6CB at its correct location into
1328 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1329 * _decode_session6() uses IP6CB().
1330 * barrier() makes sure compiler won't play aliasing games.
1331 */
1332 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1333 sizeof(struct inet6_skb_parm));
1334 barrier();
1335
1336 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1337 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1338 skb->len - th->doff*4);
1339 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1340 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1341 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1342 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1343 TCP_SKB_CB(skb)->sacked = 0;
1344}
1345
1346static void tcp_v6_restore_cb(struct sk_buff *skb)
1347{
1348 /* We need to move header back to the beginning if xfrm6_policy_check()
1349 * and tcp_v6_fill_cb() are going to be called again.
1350 */
1351 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1352 sizeof(struct inet6_skb_parm));
1353}
1354
1355static int tcp_v6_rcv(struct sk_buff *skb)
1356{
1357 const struct tcphdr *th;
1358 const struct ipv6hdr *hdr;
1359 struct sock *sk;
1360 int ret;
1361 struct net *net = dev_net(skb->dev);
1362
1363 if (skb->pkt_type != PACKET_HOST)
1364 goto discard_it;
1365
1366 /*
1367 * Count it even if it's bad.
1368 */
1369 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1370
1371 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1372 goto discard_it;
1373
1374 th = tcp_hdr(skb);
1375
1376 if (th->doff < sizeof(struct tcphdr)/4)
1377 goto bad_packet;
1378 if (!pskb_may_pull(skb, th->doff*4))
1379 goto discard_it;
1380
1381 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1382 goto csum_error;
1383
1384 th = tcp_hdr(skb);
1385 hdr = ipv6_hdr(skb);
1386
1387lookup:
1388 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1389 th->source, th->dest, inet6_iif(skb));
1390 if (!sk)
1391 goto no_tcp_socket;
1392
1393process:
1394 if (sk->sk_state == TCP_TIME_WAIT)
1395 goto do_time_wait;
1396
1397 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1398 struct request_sock *req = inet_reqsk(sk);
1399 struct sock *nsk;
1400
1401 sk = req->rsk_listener;
1402 tcp_v6_fill_cb(skb, hdr, th);
1403 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1404 reqsk_put(req);
1405 goto discard_it;
1406 }
1407 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1408 inet_csk_reqsk_queue_drop_and_put(sk, req);
1409 goto lookup;
1410 }
1411 sock_hold(sk);
1412 nsk = tcp_check_req(sk, skb, req, false);
1413 if (!nsk) {
1414 reqsk_put(req);
1415 goto discard_and_relse;
1416 }
1417 if (nsk == sk) {
1418 reqsk_put(req);
1419 tcp_v6_restore_cb(skb);
1420 } else if (tcp_child_process(sk, nsk, skb)) {
1421 tcp_v6_send_reset(nsk, skb);
1422 goto discard_and_relse;
1423 } else {
1424 sock_put(sk);
1425 return 0;
1426 }
1427 }
1428 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1429 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1430 goto discard_and_relse;
1431 }
1432
1433 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1434 goto discard_and_relse;
1435
1436 tcp_v6_fill_cb(skb, hdr, th);
1437
1438 if (tcp_v6_inbound_md5_hash(sk, skb))
1439 goto discard_and_relse;
1440
1441 if (sk_filter(sk, skb))
1442 goto discard_and_relse;
1443
1444 skb->dev = NULL;
1445
1446 if (sk->sk_state == TCP_LISTEN) {
1447 ret = tcp_v6_do_rcv(sk, skb);
1448 goto put_and_return;
1449 }
1450
1451 sk_incoming_cpu_update(sk);
1452
1453 bh_lock_sock_nested(sk);
1454 tcp_segs_in(tcp_sk(sk), skb);
1455 ret = 0;
1456 if (!sock_owned_by_user(sk)) {
1457 if (!tcp_prequeue(sk, skb))
1458 ret = tcp_v6_do_rcv(sk, skb);
1459 } else if (unlikely(sk_add_backlog(sk, skb,
1460 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1461 bh_unlock_sock(sk);
1462 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1463 goto discard_and_relse;
1464 }
1465 bh_unlock_sock(sk);
1466
1467put_and_return:
1468 sock_put(sk);
1469 return ret ? -1 : 0;
1470
1471no_tcp_socket:
1472 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1473 goto discard_it;
1474
1475 tcp_v6_fill_cb(skb, hdr, th);
1476
1477 if (tcp_checksum_complete(skb)) {
1478csum_error:
1479 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1480bad_packet:
1481 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1482 } else {
1483 tcp_v6_send_reset(NULL, skb);
1484 }
1485
1486discard_it:
1487 kfree_skb(skb);
1488 return 0;
1489
1490discard_and_relse:
1491 sock_put(sk);
1492 goto discard_it;
1493
1494do_time_wait:
1495 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1496 inet_twsk_put(inet_twsk(sk));
1497 goto discard_it;
1498 }
1499
1500 tcp_v6_fill_cb(skb, hdr, th);
1501
1502 if (tcp_checksum_complete(skb)) {
1503 inet_twsk_put(inet_twsk(sk));
1504 goto csum_error;
1505 }
1506
1507 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1508 case TCP_TW_SYN:
1509 {
1510 struct sock *sk2;
1511
1512 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1513 skb, __tcp_hdrlen(th),
1514 &ipv6_hdr(skb)->saddr, th->source,
1515 &ipv6_hdr(skb)->daddr,
1516 ntohs(th->dest), tcp_v6_iif(skb));
1517 if (sk2) {
1518 struct inet_timewait_sock *tw = inet_twsk(sk);
1519 inet_twsk_deschedule_put(tw);
1520 sk = sk2;
1521 tcp_v6_restore_cb(skb);
1522 goto process;
1523 }
1524 /* Fall through to ACK */
1525 }
1526 case TCP_TW_ACK:
1527 tcp_v6_timewait_ack(sk, skb);
1528 break;
1529 case TCP_TW_RST:
1530 tcp_v6_restore_cb(skb);
1531 tcp_v6_send_reset(sk, skb);
1532 inet_twsk_deschedule_put(inet_twsk(sk));
1533 goto discard_it;
1534 case TCP_TW_SUCCESS:
1535 ;
1536 }
1537 goto discard_it;
1538}
1539
1540static void tcp_v6_early_demux(struct sk_buff *skb)
1541{
1542 const struct ipv6hdr *hdr;
1543 const struct tcphdr *th;
1544 struct sock *sk;
1545
1546 if (skb->pkt_type != PACKET_HOST)
1547 return;
1548
1549 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1550 return;
1551
1552 hdr = ipv6_hdr(skb);
1553 th = tcp_hdr(skb);
1554
1555 if (th->doff < sizeof(struct tcphdr) / 4)
1556 return;
1557
1558 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1559 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1560 &hdr->saddr, th->source,
1561 &hdr->daddr, ntohs(th->dest),
1562 inet6_iif(skb));
1563 if (sk) {
1564 skb->sk = sk;
1565 skb->destructor = sock_edemux;
1566 if (sk_fullsock(sk)) {
1567 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1568
1569 if (dst)
1570 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1571 if (dst &&
1572 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1573 skb_dst_set_noref(skb, dst);
1574 }
1575 }
1576}
1577
1578static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1579 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1580 .twsk_unique = tcp_twsk_unique,
1581 .twsk_destructor = tcp_twsk_destructor,
1582};
1583
1584static const struct inet_connection_sock_af_ops ipv6_specific = {
1585 .queue_xmit = inet6_csk_xmit,
1586 .send_check = tcp_v6_send_check,
1587 .rebuild_header = inet6_sk_rebuild_header,
1588 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1589 .conn_request = tcp_v6_conn_request,
1590 .syn_recv_sock = tcp_v6_syn_recv_sock,
1591 .net_header_len = sizeof(struct ipv6hdr),
1592 .net_frag_header_len = sizeof(struct frag_hdr),
1593 .setsockopt = ipv6_setsockopt,
1594 .getsockopt = ipv6_getsockopt,
1595 .addr2sockaddr = inet6_csk_addr2sockaddr,
1596 .sockaddr_len = sizeof(struct sockaddr_in6),
1597 .bind_conflict = inet6_csk_bind_conflict,
1598#ifdef CONFIG_COMPAT
1599 .compat_setsockopt = compat_ipv6_setsockopt,
1600 .compat_getsockopt = compat_ipv6_getsockopt,
1601#endif
1602 .mtu_reduced = tcp_v6_mtu_reduced,
1603};
1604
1605#ifdef CONFIG_TCP_MD5SIG
1606static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1607 .md5_lookup = tcp_v6_md5_lookup,
1608 .calc_md5_hash = tcp_v6_md5_hash_skb,
1609 .md5_parse = tcp_v6_parse_md5_keys,
1610};
1611#endif
1612
1613/*
1614 * TCP over IPv4 via INET6 API
1615 */
1616static const struct inet_connection_sock_af_ops ipv6_mapped = {
1617 .queue_xmit = ip_queue_xmit,
1618 .send_check = tcp_v4_send_check,
1619 .rebuild_header = inet_sk_rebuild_header,
1620 .sk_rx_dst_set = inet_sk_rx_dst_set,
1621 .conn_request = tcp_v6_conn_request,
1622 .syn_recv_sock = tcp_v6_syn_recv_sock,
1623 .net_header_len = sizeof(struct iphdr),
1624 .setsockopt = ipv6_setsockopt,
1625 .getsockopt = ipv6_getsockopt,
1626 .addr2sockaddr = inet6_csk_addr2sockaddr,
1627 .sockaddr_len = sizeof(struct sockaddr_in6),
1628 .bind_conflict = inet6_csk_bind_conflict,
1629#ifdef CONFIG_COMPAT
1630 .compat_setsockopt = compat_ipv6_setsockopt,
1631 .compat_getsockopt = compat_ipv6_getsockopt,
1632#endif
1633 .mtu_reduced = tcp_v4_mtu_reduced,
1634};
1635
1636#ifdef CONFIG_TCP_MD5SIG
1637static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1638 .md5_lookup = tcp_v4_md5_lookup,
1639 .calc_md5_hash = tcp_v4_md5_hash_skb,
1640 .md5_parse = tcp_v6_parse_md5_keys,
1641};
1642#endif
1643
1644/* NOTE: A lot of things set to zero explicitly by call to
1645 * sk_alloc() so need not be done here.
1646 */
1647static int tcp_v6_init_sock(struct sock *sk)
1648{
1649 struct inet_connection_sock *icsk = inet_csk(sk);
1650
1651 tcp_init_sock(sk);
1652
1653 icsk->icsk_af_ops = &ipv6_specific;
1654
1655#ifdef CONFIG_TCP_MD5SIG
1656 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1657#endif
1658
1659 return 0;
1660}
1661
1662static void tcp_v6_destroy_sock(struct sock *sk)
1663{
1664 tcp_v4_destroy_sock(sk);
1665 inet6_destroy_sock(sk);
1666}
1667
1668#ifdef CONFIG_PROC_FS
1669/* Proc filesystem TCPv6 sock list dumping. */
1670static void get_openreq6(struct seq_file *seq,
1671 const struct request_sock *req, int i)
1672{
1673 long ttd = req->rsk_timer.expires - jiffies;
1674 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1675 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1676
1677 if (ttd < 0)
1678 ttd = 0;
1679
1680 seq_printf(seq,
1681 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1682 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1683 i,
1684 src->s6_addr32[0], src->s6_addr32[1],
1685 src->s6_addr32[2], src->s6_addr32[3],
1686 inet_rsk(req)->ir_num,
1687 dest->s6_addr32[0], dest->s6_addr32[1],
1688 dest->s6_addr32[2], dest->s6_addr32[3],
1689 ntohs(inet_rsk(req)->ir_rmt_port),
1690 TCP_SYN_RECV,
1691 0, 0, /* could print option size, but that is af dependent. */
1692 1, /* timers active (only the expire timer) */
1693 jiffies_to_clock_t(ttd),
1694 req->num_timeout,
1695 from_kuid_munged(seq_user_ns(seq),
1696 sock_i_uid(req->rsk_listener)),
1697 0, /* non standard timer */
1698 0, /* open_requests have no inode */
1699 0, req);
1700}
1701
1702static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1703{
1704 const struct in6_addr *dest, *src;
1705 __u16 destp, srcp;
1706 int timer_active;
1707 unsigned long timer_expires;
1708 const struct inet_sock *inet = inet_sk(sp);
1709 const struct tcp_sock *tp = tcp_sk(sp);
1710 const struct inet_connection_sock *icsk = inet_csk(sp);
1711 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1712 int rx_queue;
1713 int state;
1714
1715 dest = &sp->sk_v6_daddr;
1716 src = &sp->sk_v6_rcv_saddr;
1717 destp = ntohs(inet->inet_dport);
1718 srcp = ntohs(inet->inet_sport);
1719
1720 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1721 timer_active = 1;
1722 timer_expires = icsk->icsk_timeout;
1723 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1724 timer_active = 4;
1725 timer_expires = icsk->icsk_timeout;
1726 } else if (timer_pending(&sp->sk_timer)) {
1727 timer_active = 2;
1728 timer_expires = sp->sk_timer.expires;
1729 } else {
1730 timer_active = 0;
1731 timer_expires = jiffies;
1732 }
1733
1734 state = sk_state_load(sp);
1735 if (state == TCP_LISTEN)
1736 rx_queue = sp->sk_ack_backlog;
1737 else
1738 /* Because we don't lock the socket,
1739 * we might find a transient negative value.
1740 */
1741 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1742
1743 seq_printf(seq,
1744 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1745 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1746 i,
1747 src->s6_addr32[0], src->s6_addr32[1],
1748 src->s6_addr32[2], src->s6_addr32[3], srcp,
1749 dest->s6_addr32[0], dest->s6_addr32[1],
1750 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1751 state,
1752 tp->write_seq - tp->snd_una,
1753 rx_queue,
1754 timer_active,
1755 jiffies_delta_to_clock_t(timer_expires - jiffies),
1756 icsk->icsk_retransmits,
1757 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1758 icsk->icsk_probes_out,
1759 sock_i_ino(sp),
1760 atomic_read(&sp->sk_refcnt), sp,
1761 jiffies_to_clock_t(icsk->icsk_rto),
1762 jiffies_to_clock_t(icsk->icsk_ack.ato),
1763 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1764 tp->snd_cwnd,
1765 state == TCP_LISTEN ?
1766 fastopenq->max_qlen :
1767 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1768 );
1769}
1770
1771static void get_timewait6_sock(struct seq_file *seq,
1772 struct inet_timewait_sock *tw, int i)
1773{
1774 long delta = tw->tw_timer.expires - jiffies;
1775 const struct in6_addr *dest, *src;
1776 __u16 destp, srcp;
1777
1778 dest = &tw->tw_v6_daddr;
1779 src = &tw->tw_v6_rcv_saddr;
1780 destp = ntohs(tw->tw_dport);
1781 srcp = ntohs(tw->tw_sport);
1782
1783 seq_printf(seq,
1784 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1785 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1786 i,
1787 src->s6_addr32[0], src->s6_addr32[1],
1788 src->s6_addr32[2], src->s6_addr32[3], srcp,
1789 dest->s6_addr32[0], dest->s6_addr32[1],
1790 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1791 tw->tw_substate, 0, 0,
1792 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1793 atomic_read(&tw->tw_refcnt), tw);
1794}
1795
1796static int tcp6_seq_show(struct seq_file *seq, void *v)
1797{
1798 struct tcp_iter_state *st;
1799 struct sock *sk = v;
1800
1801 if (v == SEQ_START_TOKEN) {
1802 seq_puts(seq,
1803 " sl "
1804 "local_address "
1805 "remote_address "
1806 "st tx_queue rx_queue tr tm->when retrnsmt"
1807 " uid timeout inode\n");
1808 goto out;
1809 }
1810 st = seq->private;
1811
1812 if (sk->sk_state == TCP_TIME_WAIT)
1813 get_timewait6_sock(seq, v, st->num);
1814 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1815 get_openreq6(seq, v, st->num);
1816 else
1817 get_tcp6_sock(seq, v, st->num);
1818out:
1819 return 0;
1820}
1821
1822static const struct file_operations tcp6_afinfo_seq_fops = {
1823 .owner = THIS_MODULE,
1824 .open = tcp_seq_open,
1825 .read = seq_read,
1826 .llseek = seq_lseek,
1827 .release = seq_release_net
1828};
1829
1830static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1831 .name = "tcp6",
1832 .family = AF_INET6,
1833 .seq_fops = &tcp6_afinfo_seq_fops,
1834 .seq_ops = {
1835 .show = tcp6_seq_show,
1836 },
1837};
1838
1839int __net_init tcp6_proc_init(struct net *net)
1840{
1841 return tcp_proc_register(net, &tcp6_seq_afinfo);
1842}
1843
1844void tcp6_proc_exit(struct net *net)
1845{
1846 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1847}
1848#endif
1849
1850static void tcp_v6_clear_sk(struct sock *sk, int size)
1851{
1852 struct inet_sock *inet = inet_sk(sk);
1853
1854 /* we do not want to clear pinet6 field, because of RCU lookups */
1855 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1856
1857 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1858 memset(&inet->pinet6 + 1, 0, size);
1859}
1860
1861struct proto tcpv6_prot = {
1862 .name = "TCPv6",
1863 .owner = THIS_MODULE,
1864 .close = tcp_close,
1865 .connect = tcp_v6_connect,
1866 .disconnect = tcp_disconnect,
1867 .accept = inet_csk_accept,
1868 .ioctl = tcp_ioctl,
1869 .init = tcp_v6_init_sock,
1870 .destroy = tcp_v6_destroy_sock,
1871 .shutdown = tcp_shutdown,
1872 .setsockopt = tcp_setsockopt,
1873 .getsockopt = tcp_getsockopt,
1874 .recvmsg = tcp_recvmsg,
1875 .sendmsg = tcp_sendmsg,
1876 .sendpage = tcp_sendpage,
1877 .backlog_rcv = tcp_v6_do_rcv,
1878 .release_cb = tcp_release_cb,
1879 .hash = inet6_hash,
1880 .unhash = inet_unhash,
1881 .get_port = inet_csk_get_port,
1882 .enter_memory_pressure = tcp_enter_memory_pressure,
1883 .stream_memory_free = tcp_stream_memory_free,
1884 .sockets_allocated = &tcp_sockets_allocated,
1885 .memory_allocated = &tcp_memory_allocated,
1886 .memory_pressure = &tcp_memory_pressure,
1887 .orphan_count = &tcp_orphan_count,
1888 .sysctl_mem = sysctl_tcp_mem,
1889 .sysctl_wmem = sysctl_tcp_wmem,
1890 .sysctl_rmem = sysctl_tcp_rmem,
1891 .max_header = MAX_TCP_HEADER,
1892 .obj_size = sizeof(struct tcp6_sock),
1893 .slab_flags = SLAB_DESTROY_BY_RCU,
1894 .twsk_prot = &tcp6_timewait_sock_ops,
1895 .rsk_prot = &tcp6_request_sock_ops,
1896 .h.hashinfo = &tcp_hashinfo,
1897 .no_autobind = true,
1898#ifdef CONFIG_COMPAT
1899 .compat_setsockopt = compat_tcp_setsockopt,
1900 .compat_getsockopt = compat_tcp_getsockopt,
1901#endif
1902 .clear_sk = tcp_v6_clear_sk,
1903 .diag_destroy = tcp_abort,
1904};
1905
1906static const struct inet6_protocol tcpv6_protocol = {
1907 .early_demux = tcp_v6_early_demux,
1908 .handler = tcp_v6_rcv,
1909 .err_handler = tcp_v6_err,
1910 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1911};
1912
1913static struct inet_protosw tcpv6_protosw = {
1914 .type = SOCK_STREAM,
1915 .protocol = IPPROTO_TCP,
1916 .prot = &tcpv6_prot,
1917 .ops = &inet6_stream_ops,
1918 .flags = INET_PROTOSW_PERMANENT |
1919 INET_PROTOSW_ICSK,
1920};
1921
1922static int __net_init tcpv6_net_init(struct net *net)
1923{
1924 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1925 SOCK_RAW, IPPROTO_TCP, net);
1926}
1927
1928static void __net_exit tcpv6_net_exit(struct net *net)
1929{
1930 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1931}
1932
1933static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1934{
1935 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1936}
1937
1938static struct pernet_operations tcpv6_net_ops = {
1939 .init = tcpv6_net_init,
1940 .exit = tcpv6_net_exit,
1941 .exit_batch = tcpv6_net_exit_batch,
1942};
1943
1944int __init tcpv6_init(void)
1945{
1946 int ret;
1947
1948 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1949 if (ret)
1950 goto out;
1951
1952 /* register inet6 protocol */
1953 ret = inet6_register_protosw(&tcpv6_protosw);
1954 if (ret)
1955 goto out_tcpv6_protocol;
1956
1957 ret = register_pernet_subsys(&tcpv6_net_ops);
1958 if (ret)
1959 goto out_tcpv6_protosw;
1960out:
1961 return ret;
1962
1963out_tcpv6_protosw:
1964 inet6_unregister_protosw(&tcpv6_protosw);
1965out_tcpv6_protocol:
1966 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1967 goto out;
1968}
1969
1970void tcpv6_exit(void)
1971{
1972 unregister_pernet_subsys(&tcpv6_net_ops);
1973 inet6_unregister_protosw(&tcpv6_protosw);
1974 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1975}
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
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
66#include <asm/uaccess.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);
79static void __tcp_v6_send_check(struct sk_buff *skb,
80 const struct in6_addr *saddr,
81 const struct in6_addr *daddr);
82
83static const struct inet_connection_sock_af_ops ipv6_mapped;
84static const struct inet_connection_sock_af_ops ipv6_specific;
85#ifdef CONFIG_TCP_MD5SIG
86static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
87static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
88#else
89static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
90 const struct in6_addr *addr)
91{
92 return NULL;
93}
94#endif
95
96static void tcp_v6_hash(struct sock *sk)
97{
98 if (sk->sk_state != TCP_CLOSE) {
99 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
100 tcp_prot.hash(sk);
101 return;
102 }
103 local_bh_disable();
104 __inet6_hash(sk, NULL);
105 local_bh_enable();
106 }
107}
108
109static __inline__ __sum16 tcp_v6_check(int len,
110 const struct in6_addr *saddr,
111 const struct in6_addr *daddr,
112 __wsum base)
113{
114 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
115}
116
117static __u32 tcp_v6_init_sequence(struct sk_buff *skb)
118{
119 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
120 ipv6_hdr(skb)->saddr.s6_addr32,
121 tcp_hdr(skb)->dest,
122 tcp_hdr(skb)->source);
123}
124
125static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
126 int addr_len)
127{
128 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
129 struct inet_sock *inet = inet_sk(sk);
130 struct inet_connection_sock *icsk = inet_csk(sk);
131 struct ipv6_pinfo *np = inet6_sk(sk);
132 struct tcp_sock *tp = tcp_sk(sk);
133 struct in6_addr *saddr = NULL, *final_p, final;
134 struct rt6_info *rt;
135 struct flowi6 fl6;
136 struct dst_entry *dst;
137 int addr_type;
138 int err;
139
140 if (addr_len < SIN6_LEN_RFC2133)
141 return -EINVAL;
142
143 if (usin->sin6_family != AF_INET6)
144 return -EAFNOSUPPORT;
145
146 memset(&fl6, 0, sizeof(fl6));
147
148 if (np->sndflow) {
149 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
150 IP6_ECN_flow_init(fl6.flowlabel);
151 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
152 struct ip6_flowlabel *flowlabel;
153 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
154 if (flowlabel == NULL)
155 return -EINVAL;
156 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
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(&np->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 ipv6_addr_copy(&np->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 &np->rcv_saddr);
238 }
239
240 return err;
241 }
242
243 if (!ipv6_addr_any(&np->rcv_saddr))
244 saddr = &np->rcv_saddr;
245
246 fl6.flowi6_proto = IPPROTO_TCP;
247 ipv6_addr_copy(&fl6.daddr, &np->daddr);
248 ipv6_addr_copy(&fl6.saddr,
249 (saddr ? saddr : &np->saddr));
250 fl6.flowi6_oif = sk->sk_bound_dev_if;
251 fl6.flowi6_mark = sk->sk_mark;
252 fl6.fl6_dport = usin->sin6_port;
253 fl6.fl6_sport = inet->inet_sport;
254
255 final_p = fl6_update_dst(&fl6, np->opt, &final);
256
257 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
258
259 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, true);
260 if (IS_ERR(dst)) {
261 err = PTR_ERR(dst);
262 goto failure;
263 }
264
265 if (saddr == NULL) {
266 saddr = &fl6.saddr;
267 ipv6_addr_copy(&np->rcv_saddr, saddr);
268 }
269
270 /* set the source address */
271 ipv6_addr_copy(&np->saddr, saddr);
272 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
273
274 sk->sk_gso_type = SKB_GSO_TCPV6;
275 __ip6_dst_store(sk, dst, NULL, NULL);
276
277 rt = (struct rt6_info *) dst;
278 if (tcp_death_row.sysctl_tw_recycle &&
279 !tp->rx_opt.ts_recent_stamp &&
280 ipv6_addr_equal(&rt->rt6i_dst.addr, &np->daddr)) {
281 struct inet_peer *peer = rt6_get_peer(rt);
282 /*
283 * VJ's idea. We save last timestamp seen from
284 * the destination in peer table, when entering state
285 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
286 * when trying new connection.
287 */
288 if (peer) {
289 inet_peer_refcheck(peer);
290 if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
291 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
292 tp->rx_opt.ts_recent = peer->tcp_ts;
293 }
294 }
295 }
296
297 icsk->icsk_ext_hdr_len = 0;
298 if (np->opt)
299 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
300 np->opt->opt_nflen);
301
302 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
303
304 inet->inet_dport = usin->sin6_port;
305
306 tcp_set_state(sk, TCP_SYN_SENT);
307 err = inet6_hash_connect(&tcp_death_row, sk);
308 if (err)
309 goto late_failure;
310
311 if (!tp->write_seq)
312 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
313 np->daddr.s6_addr32,
314 inet->inet_sport,
315 inet->inet_dport);
316
317 err = tcp_connect(sk);
318 if (err)
319 goto late_failure;
320
321 return 0;
322
323late_failure:
324 tcp_set_state(sk, TCP_CLOSE);
325 __sk_dst_reset(sk);
326failure:
327 inet->inet_dport = 0;
328 sk->sk_route_caps = 0;
329 return err;
330}
331
332static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
333 u8 type, u8 code, int offset, __be32 info)
334{
335 const struct ipv6hdr *hdr = (const struct ipv6hdr*)skb->data;
336 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
337 struct ipv6_pinfo *np;
338 struct sock *sk;
339 int err;
340 struct tcp_sock *tp;
341 __u32 seq;
342 struct net *net = dev_net(skb->dev);
343
344 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
345 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
346
347 if (sk == NULL) {
348 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
349 ICMP6_MIB_INERRORS);
350 return;
351 }
352
353 if (sk->sk_state == TCP_TIME_WAIT) {
354 inet_twsk_put(inet_twsk(sk));
355 return;
356 }
357
358 bh_lock_sock(sk);
359 if (sock_owned_by_user(sk))
360 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
361
362 if (sk->sk_state == TCP_CLOSE)
363 goto out;
364
365 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
366 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
367 goto out;
368 }
369
370 tp = tcp_sk(sk);
371 seq = ntohl(th->seq);
372 if (sk->sk_state != TCP_LISTEN &&
373 !between(seq, tp->snd_una, tp->snd_nxt)) {
374 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
375 goto out;
376 }
377
378 np = inet6_sk(sk);
379
380 if (type == ICMPV6_PKT_TOOBIG) {
381 struct dst_entry *dst;
382
383 if (sock_owned_by_user(sk))
384 goto out;
385 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
386 goto out;
387
388 /* icmp should have updated the destination cache entry */
389 dst = __sk_dst_check(sk, np->dst_cookie);
390
391 if (dst == NULL) {
392 struct inet_sock *inet = inet_sk(sk);
393 struct flowi6 fl6;
394
395 /* BUGGG_FUTURE: Again, it is not clear how
396 to handle rthdr case. Ignore this complexity
397 for now.
398 */
399 memset(&fl6, 0, sizeof(fl6));
400 fl6.flowi6_proto = IPPROTO_TCP;
401 ipv6_addr_copy(&fl6.daddr, &np->daddr);
402 ipv6_addr_copy(&fl6.saddr, &np->saddr);
403 fl6.flowi6_oif = sk->sk_bound_dev_if;
404 fl6.flowi6_mark = sk->sk_mark;
405 fl6.fl6_dport = inet->inet_dport;
406 fl6.fl6_sport = inet->inet_sport;
407 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
408
409 dst = ip6_dst_lookup_flow(sk, &fl6, NULL, false);
410 if (IS_ERR(dst)) {
411 sk->sk_err_soft = -PTR_ERR(dst);
412 goto out;
413 }
414
415 } else
416 dst_hold(dst);
417
418 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
419 tcp_sync_mss(sk, dst_mtu(dst));
420 tcp_simple_retransmit(sk);
421 } /* else let the usual retransmit timer handle it */
422 dst_release(dst);
423 goto out;
424 }
425
426 icmpv6_err_convert(type, code, &err);
427
428 /* Might be for an request_sock */
429 switch (sk->sk_state) {
430 struct request_sock *req, **prev;
431 case TCP_LISTEN:
432 if (sock_owned_by_user(sk))
433 goto out;
434
435 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
436 &hdr->saddr, inet6_iif(skb));
437 if (!req)
438 goto out;
439
440 /* ICMPs are not backlogged, hence we cannot get
441 * an established socket here.
442 */
443 WARN_ON(req->sk != NULL);
444
445 if (seq != tcp_rsk(req)->snt_isn) {
446 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
447 goto out;
448 }
449
450 inet_csk_reqsk_queue_drop(sk, req, prev);
451 goto out;
452
453 case TCP_SYN_SENT:
454 case TCP_SYN_RECV: /* Cannot happen.
455 It can, it SYNs are crossed. --ANK */
456 if (!sock_owned_by_user(sk)) {
457 sk->sk_err = err;
458 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
459
460 tcp_done(sk);
461 } else
462 sk->sk_err_soft = err;
463 goto out;
464 }
465
466 if (!sock_owned_by_user(sk) && np->recverr) {
467 sk->sk_err = err;
468 sk->sk_error_report(sk);
469 } else
470 sk->sk_err_soft = err;
471
472out:
473 bh_unlock_sock(sk);
474 sock_put(sk);
475}
476
477
478static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
479 struct request_values *rvp)
480{
481 struct inet6_request_sock *treq = inet6_rsk(req);
482 struct ipv6_pinfo *np = inet6_sk(sk);
483 struct sk_buff * skb;
484 struct ipv6_txoptions *opt = NULL;
485 struct in6_addr * final_p, final;
486 struct flowi6 fl6;
487 struct dst_entry *dst;
488 int err;
489
490 memset(&fl6, 0, sizeof(fl6));
491 fl6.flowi6_proto = IPPROTO_TCP;
492 ipv6_addr_copy(&fl6.daddr, &treq->rmt_addr);
493 ipv6_addr_copy(&fl6.saddr, &treq->loc_addr);
494 fl6.flowlabel = 0;
495 fl6.flowi6_oif = treq->iif;
496 fl6.flowi6_mark = sk->sk_mark;
497 fl6.fl6_dport = inet_rsk(req)->rmt_port;
498 fl6.fl6_sport = inet_rsk(req)->loc_port;
499 security_req_classify_flow(req, flowi6_to_flowi(&fl6));
500
501 opt = np->opt;
502 final_p = fl6_update_dst(&fl6, opt, &final);
503
504 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false);
505 if (IS_ERR(dst)) {
506 err = PTR_ERR(dst);
507 dst = NULL;
508 goto done;
509 }
510 skb = tcp_make_synack(sk, dst, req, rvp);
511 err = -ENOMEM;
512 if (skb) {
513 __tcp_v6_send_check(skb, &treq->loc_addr, &treq->rmt_addr);
514
515 ipv6_addr_copy(&fl6.daddr, &treq->rmt_addr);
516 err = ip6_xmit(sk, skb, &fl6, opt);
517 err = net_xmit_eval(err);
518 }
519
520done:
521 if (opt && opt != np->opt)
522 sock_kfree_s(sk, opt, opt->tot_len);
523 dst_release(dst);
524 return err;
525}
526
527static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req,
528 struct request_values *rvp)
529{
530 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
531 return tcp_v6_send_synack(sk, req, rvp);
532}
533
534static void tcp_v6_reqsk_destructor(struct request_sock *req)
535{
536 kfree_skb(inet6_rsk(req)->pktopts);
537}
538
539#ifdef CONFIG_TCP_MD5SIG
540static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
541 const struct in6_addr *addr)
542{
543 struct tcp_sock *tp = tcp_sk(sk);
544 int i;
545
546 BUG_ON(tp == NULL);
547
548 if (!tp->md5sig_info || !tp->md5sig_info->entries6)
549 return NULL;
550
551 for (i = 0; i < tp->md5sig_info->entries6; i++) {
552 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, addr))
553 return &tp->md5sig_info->keys6[i].base;
554 }
555 return NULL;
556}
557
558static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
559 struct sock *addr_sk)
560{
561 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
562}
563
564static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
565 struct request_sock *req)
566{
567 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
568}
569
570static int tcp_v6_md5_do_add(struct sock *sk, const struct in6_addr *peer,
571 char *newkey, u8 newkeylen)
572{
573 /* Add key to the list */
574 struct tcp_md5sig_key *key;
575 struct tcp_sock *tp = tcp_sk(sk);
576 struct tcp6_md5sig_key *keys;
577
578 key = tcp_v6_md5_do_lookup(sk, peer);
579 if (key) {
580 /* modify existing entry - just update that one */
581 kfree(key->key);
582 key->key = newkey;
583 key->keylen = newkeylen;
584 } else {
585 /* reallocate new list if current one is full. */
586 if (!tp->md5sig_info) {
587 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
588 if (!tp->md5sig_info) {
589 kfree(newkey);
590 return -ENOMEM;
591 }
592 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
593 }
594 if (tp->md5sig_info->entries6 == 0 &&
595 tcp_alloc_md5sig_pool(sk) == NULL) {
596 kfree(newkey);
597 return -ENOMEM;
598 }
599 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
600 keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
601 (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
602
603 if (!keys) {
604 kfree(newkey);
605 if (tp->md5sig_info->entries6 == 0)
606 tcp_free_md5sig_pool();
607 return -ENOMEM;
608 }
609
610 if (tp->md5sig_info->entries6)
611 memmove(keys, tp->md5sig_info->keys6,
612 (sizeof (tp->md5sig_info->keys6[0]) *
613 tp->md5sig_info->entries6));
614
615 kfree(tp->md5sig_info->keys6);
616 tp->md5sig_info->keys6 = keys;
617 tp->md5sig_info->alloced6++;
618 }
619
620 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
621 peer);
622 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
623 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
624
625 tp->md5sig_info->entries6++;
626 }
627 return 0;
628}
629
630static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
631 u8 *newkey, __u8 newkeylen)
632{
633 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
634 newkey, newkeylen);
635}
636
637static int tcp_v6_md5_do_del(struct sock *sk, const struct in6_addr *peer)
638{
639 struct tcp_sock *tp = tcp_sk(sk);
640 int i;
641
642 for (i = 0; i < tp->md5sig_info->entries6; i++) {
643 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, peer)) {
644 /* Free the key */
645 kfree(tp->md5sig_info->keys6[i].base.key);
646 tp->md5sig_info->entries6--;
647
648 if (tp->md5sig_info->entries6 == 0) {
649 kfree(tp->md5sig_info->keys6);
650 tp->md5sig_info->keys6 = NULL;
651 tp->md5sig_info->alloced6 = 0;
652 tcp_free_md5sig_pool();
653 } else {
654 /* shrink the database */
655 if (tp->md5sig_info->entries6 != i)
656 memmove(&tp->md5sig_info->keys6[i],
657 &tp->md5sig_info->keys6[i+1],
658 (tp->md5sig_info->entries6 - i)
659 * sizeof (tp->md5sig_info->keys6[0]));
660 }
661 return 0;
662 }
663 }
664 return -ENOENT;
665}
666
667static void tcp_v6_clear_md5_list (struct sock *sk)
668{
669 struct tcp_sock *tp = tcp_sk(sk);
670 int i;
671
672 if (tp->md5sig_info->entries6) {
673 for (i = 0; i < tp->md5sig_info->entries6; i++)
674 kfree(tp->md5sig_info->keys6[i].base.key);
675 tp->md5sig_info->entries6 = 0;
676 tcp_free_md5sig_pool();
677 }
678
679 kfree(tp->md5sig_info->keys6);
680 tp->md5sig_info->keys6 = NULL;
681 tp->md5sig_info->alloced6 = 0;
682
683 if (tp->md5sig_info->entries4) {
684 for (i = 0; i < tp->md5sig_info->entries4; i++)
685 kfree(tp->md5sig_info->keys4[i].base.key);
686 tp->md5sig_info->entries4 = 0;
687 tcp_free_md5sig_pool();
688 }
689
690 kfree(tp->md5sig_info->keys4);
691 tp->md5sig_info->keys4 = NULL;
692 tp->md5sig_info->alloced4 = 0;
693}
694
695static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
696 int optlen)
697{
698 struct tcp_md5sig cmd;
699 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
700 u8 *newkey;
701
702 if (optlen < sizeof(cmd))
703 return -EINVAL;
704
705 if (copy_from_user(&cmd, optval, sizeof(cmd)))
706 return -EFAULT;
707
708 if (sin6->sin6_family != AF_INET6)
709 return -EINVAL;
710
711 if (!cmd.tcpm_keylen) {
712 if (!tcp_sk(sk)->md5sig_info)
713 return -ENOENT;
714 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
715 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
716 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
717 }
718
719 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
720 return -EINVAL;
721
722 if (!tcp_sk(sk)->md5sig_info) {
723 struct tcp_sock *tp = tcp_sk(sk);
724 struct tcp_md5sig_info *p;
725
726 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
727 if (!p)
728 return -ENOMEM;
729
730 tp->md5sig_info = p;
731 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
732 }
733
734 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
735 if (!newkey)
736 return -ENOMEM;
737 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
738 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
739 newkey, cmd.tcpm_keylen);
740 }
741 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
742}
743
744static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
745 const struct in6_addr *daddr,
746 const struct in6_addr *saddr, int nbytes)
747{
748 struct tcp6_pseudohdr *bp;
749 struct scatterlist sg;
750
751 bp = &hp->md5_blk.ip6;
752 /* 1. TCP pseudo-header (RFC2460) */
753 ipv6_addr_copy(&bp->saddr, saddr);
754 ipv6_addr_copy(&bp->daddr, daddr);
755 bp->protocol = cpu_to_be32(IPPROTO_TCP);
756 bp->len = cpu_to_be32(nbytes);
757
758 sg_init_one(&sg, bp, sizeof(*bp));
759 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
760}
761
762static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
763 const struct in6_addr *daddr, struct in6_addr *saddr,
764 struct tcphdr *th)
765{
766 struct tcp_md5sig_pool *hp;
767 struct hash_desc *desc;
768
769 hp = tcp_get_md5sig_pool();
770 if (!hp)
771 goto clear_hash_noput;
772 desc = &hp->md5_desc;
773
774 if (crypto_hash_init(desc))
775 goto clear_hash;
776 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
777 goto clear_hash;
778 if (tcp_md5_hash_header(hp, th))
779 goto clear_hash;
780 if (tcp_md5_hash_key(hp, key))
781 goto clear_hash;
782 if (crypto_hash_final(desc, md5_hash))
783 goto clear_hash;
784
785 tcp_put_md5sig_pool();
786 return 0;
787
788clear_hash:
789 tcp_put_md5sig_pool();
790clear_hash_noput:
791 memset(md5_hash, 0, 16);
792 return 1;
793}
794
795static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
796 struct sock *sk, struct request_sock *req,
797 struct sk_buff *skb)
798{
799 const struct in6_addr *saddr, *daddr;
800 struct tcp_md5sig_pool *hp;
801 struct hash_desc *desc;
802 struct tcphdr *th = tcp_hdr(skb);
803
804 if (sk) {
805 saddr = &inet6_sk(sk)->saddr;
806 daddr = &inet6_sk(sk)->daddr;
807 } else if (req) {
808 saddr = &inet6_rsk(req)->loc_addr;
809 daddr = &inet6_rsk(req)->rmt_addr;
810 } else {
811 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
812 saddr = &ip6h->saddr;
813 daddr = &ip6h->daddr;
814 }
815
816 hp = tcp_get_md5sig_pool();
817 if (!hp)
818 goto clear_hash_noput;
819 desc = &hp->md5_desc;
820
821 if (crypto_hash_init(desc))
822 goto clear_hash;
823
824 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
825 goto clear_hash;
826 if (tcp_md5_hash_header(hp, th))
827 goto clear_hash;
828 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
829 goto clear_hash;
830 if (tcp_md5_hash_key(hp, key))
831 goto clear_hash;
832 if (crypto_hash_final(desc, md5_hash))
833 goto clear_hash;
834
835 tcp_put_md5sig_pool();
836 return 0;
837
838clear_hash:
839 tcp_put_md5sig_pool();
840clear_hash_noput:
841 memset(md5_hash, 0, 16);
842 return 1;
843}
844
845static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
846{
847 __u8 *hash_location = NULL;
848 struct tcp_md5sig_key *hash_expected;
849 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
850 struct tcphdr *th = tcp_hdr(skb);
851 int genhash;
852 u8 newhash[16];
853
854 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
855 hash_location = tcp_parse_md5sig_option(th);
856
857 /* We've parsed the options - do we have a hash? */
858 if (!hash_expected && !hash_location)
859 return 0;
860
861 if (hash_expected && !hash_location) {
862 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
863 return 1;
864 }
865
866 if (!hash_expected && hash_location) {
867 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
868 return 1;
869 }
870
871 /* check the signature */
872 genhash = tcp_v6_md5_hash_skb(newhash,
873 hash_expected,
874 NULL, NULL, skb);
875
876 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
877 if (net_ratelimit()) {
878 printk(KERN_INFO "MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
879 genhash ? "failed" : "mismatch",
880 &ip6h->saddr, ntohs(th->source),
881 &ip6h->daddr, ntohs(th->dest));
882 }
883 return 1;
884 }
885 return 0;
886}
887#endif
888
889struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
890 .family = AF_INET6,
891 .obj_size = sizeof(struct tcp6_request_sock),
892 .rtx_syn_ack = tcp_v6_rtx_synack,
893 .send_ack = tcp_v6_reqsk_send_ack,
894 .destructor = tcp_v6_reqsk_destructor,
895 .send_reset = tcp_v6_send_reset,
896 .syn_ack_timeout = tcp_syn_ack_timeout,
897};
898
899#ifdef CONFIG_TCP_MD5SIG
900static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
901 .md5_lookup = tcp_v6_reqsk_md5_lookup,
902 .calc_md5_hash = tcp_v6_md5_hash_skb,
903};
904#endif
905
906static void __tcp_v6_send_check(struct sk_buff *skb,
907 const struct in6_addr *saddr, const struct in6_addr *daddr)
908{
909 struct tcphdr *th = tcp_hdr(skb);
910
911 if (skb->ip_summed == CHECKSUM_PARTIAL) {
912 th->check = ~tcp_v6_check(skb->len, saddr, daddr, 0);
913 skb->csum_start = skb_transport_header(skb) - skb->head;
914 skb->csum_offset = offsetof(struct tcphdr, check);
915 } else {
916 th->check = tcp_v6_check(skb->len, saddr, daddr,
917 csum_partial(th, th->doff << 2,
918 skb->csum));
919 }
920}
921
922static void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
923{
924 struct ipv6_pinfo *np = inet6_sk(sk);
925
926 __tcp_v6_send_check(skb, &np->saddr, &np->daddr);
927}
928
929static int tcp_v6_gso_send_check(struct sk_buff *skb)
930{
931 const struct ipv6hdr *ipv6h;
932 struct tcphdr *th;
933
934 if (!pskb_may_pull(skb, sizeof(*th)))
935 return -EINVAL;
936
937 ipv6h = ipv6_hdr(skb);
938 th = tcp_hdr(skb);
939
940 th->check = 0;
941 skb->ip_summed = CHECKSUM_PARTIAL;
942 __tcp_v6_send_check(skb, &ipv6h->saddr, &ipv6h->daddr);
943 return 0;
944}
945
946static struct sk_buff **tcp6_gro_receive(struct sk_buff **head,
947 struct sk_buff *skb)
948{
949 const struct ipv6hdr *iph = skb_gro_network_header(skb);
950
951 switch (skb->ip_summed) {
952 case CHECKSUM_COMPLETE:
953 if (!tcp_v6_check(skb_gro_len(skb), &iph->saddr, &iph->daddr,
954 skb->csum)) {
955 skb->ip_summed = CHECKSUM_UNNECESSARY;
956 break;
957 }
958
959 /* fall through */
960 case CHECKSUM_NONE:
961 NAPI_GRO_CB(skb)->flush = 1;
962 return NULL;
963 }
964
965 return tcp_gro_receive(head, skb);
966}
967
968static int tcp6_gro_complete(struct sk_buff *skb)
969{
970 const struct ipv6hdr *iph = ipv6_hdr(skb);
971 struct tcphdr *th = tcp_hdr(skb);
972
973 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
974 &iph->saddr, &iph->daddr, 0);
975 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
976
977 return tcp_gro_complete(skb);
978}
979
980static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
981 u32 ts, struct tcp_md5sig_key *key, int rst)
982{
983 struct tcphdr *th = tcp_hdr(skb), *t1;
984 struct sk_buff *buff;
985 struct flowi6 fl6;
986 struct net *net = dev_net(skb_dst(skb)->dev);
987 struct sock *ctl_sk = net->ipv6.tcp_sk;
988 unsigned int tot_len = sizeof(struct tcphdr);
989 struct dst_entry *dst;
990 __be32 *topt;
991
992 if (ts)
993 tot_len += TCPOLEN_TSTAMP_ALIGNED;
994#ifdef CONFIG_TCP_MD5SIG
995 if (key)
996 tot_len += TCPOLEN_MD5SIG_ALIGNED;
997#endif
998
999 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1000 GFP_ATOMIC);
1001 if (buff == NULL)
1002 return;
1003
1004 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1005
1006 t1 = (struct tcphdr *) skb_push(buff, tot_len);
1007 skb_reset_transport_header(buff);
1008
1009 /* Swap the send and the receive. */
1010 memset(t1, 0, sizeof(*t1));
1011 t1->dest = th->source;
1012 t1->source = th->dest;
1013 t1->doff = tot_len / 4;
1014 t1->seq = htonl(seq);
1015 t1->ack_seq = htonl(ack);
1016 t1->ack = !rst || !th->ack;
1017 t1->rst = rst;
1018 t1->window = htons(win);
1019
1020 topt = (__be32 *)(t1 + 1);
1021
1022 if (ts) {
1023 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1024 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1025 *topt++ = htonl(tcp_time_stamp);
1026 *topt++ = htonl(ts);
1027 }
1028
1029#ifdef CONFIG_TCP_MD5SIG
1030 if (key) {
1031 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1032 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1033 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
1034 &ipv6_hdr(skb)->saddr,
1035 &ipv6_hdr(skb)->daddr, t1);
1036 }
1037#endif
1038
1039 memset(&fl6, 0, sizeof(fl6));
1040 ipv6_addr_copy(&fl6.daddr, &ipv6_hdr(skb)->saddr);
1041 ipv6_addr_copy(&fl6.saddr, &ipv6_hdr(skb)->daddr);
1042
1043 buff->ip_summed = CHECKSUM_PARTIAL;
1044 buff->csum = 0;
1045
1046 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
1047
1048 fl6.flowi6_proto = IPPROTO_TCP;
1049 fl6.flowi6_oif = inet6_iif(skb);
1050 fl6.fl6_dport = t1->dest;
1051 fl6.fl6_sport = t1->source;
1052 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
1053
1054 /* Pass a socket to ip6_dst_lookup either it is for RST
1055 * Underlying function will use this to retrieve the network
1056 * namespace
1057 */
1058 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL, false);
1059 if (!IS_ERR(dst)) {
1060 skb_dst_set(buff, dst);
1061 ip6_xmit(ctl_sk, buff, &fl6, NULL);
1062 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1063 if (rst)
1064 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
1065 return;
1066 }
1067
1068 kfree_skb(buff);
1069}
1070
1071static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
1072{
1073 struct tcphdr *th = tcp_hdr(skb);
1074 u32 seq = 0, ack_seq = 0;
1075 struct tcp_md5sig_key *key = NULL;
1076
1077 if (th->rst)
1078 return;
1079
1080 if (!ipv6_unicast_destination(skb))
1081 return;
1082
1083#ifdef CONFIG_TCP_MD5SIG
1084 if (sk)
1085 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1086#endif
1087
1088 if (th->ack)
1089 seq = ntohl(th->ack_seq);
1090 else
1091 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1092 (th->doff << 2);
1093
1094 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1);
1095}
1096
1097static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
1098 struct tcp_md5sig_key *key)
1099{
1100 tcp_v6_send_response(skb, seq, ack, win, ts, key, 0);
1101}
1102
1103static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1104{
1105 struct inet_timewait_sock *tw = inet_twsk(sk);
1106 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1107
1108 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1109 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1110 tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw));
1111
1112 inet_twsk_put(tw);
1113}
1114
1115static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1116 struct request_sock *req)
1117{
1118 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1119 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr));
1120}
1121
1122
1123static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1124{
1125 struct request_sock *req, **prev;
1126 const struct tcphdr *th = tcp_hdr(skb);
1127 struct sock *nsk;
1128
1129 /* Find possible connection requests. */
1130 req = inet6_csk_search_req(sk, &prev, th->source,
1131 &ipv6_hdr(skb)->saddr,
1132 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1133 if (req)
1134 return tcp_check_req(sk, skb, req, prev);
1135
1136 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1137 &ipv6_hdr(skb)->saddr, th->source,
1138 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1139
1140 if (nsk) {
1141 if (nsk->sk_state != TCP_TIME_WAIT) {
1142 bh_lock_sock(nsk);
1143 return nsk;
1144 }
1145 inet_twsk_put(inet_twsk(nsk));
1146 return NULL;
1147 }
1148
1149#ifdef CONFIG_SYN_COOKIES
1150 if (!th->syn)
1151 sk = cookie_v6_check(sk, skb);
1152#endif
1153 return sk;
1154}
1155
1156/* FIXME: this is substantially similar to the ipv4 code.
1157 * Can some kind of merge be done? -- erics
1158 */
1159static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1160{
1161 struct tcp_extend_values tmp_ext;
1162 struct tcp_options_received tmp_opt;
1163 u8 *hash_location;
1164 struct request_sock *req;
1165 struct inet6_request_sock *treq;
1166 struct ipv6_pinfo *np = inet6_sk(sk);
1167 struct tcp_sock *tp = tcp_sk(sk);
1168 __u32 isn = TCP_SKB_CB(skb)->when;
1169 struct dst_entry *dst = NULL;
1170 int want_cookie = 0;
1171
1172 if (skb->protocol == htons(ETH_P_IP))
1173 return tcp_v4_conn_request(sk, skb);
1174
1175 if (!ipv6_unicast_destination(skb))
1176 goto drop;
1177
1178 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1179 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
1180 if (!want_cookie)
1181 goto drop;
1182 }
1183
1184 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1185 goto drop;
1186
1187 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1188 if (req == NULL)
1189 goto drop;
1190
1191#ifdef CONFIG_TCP_MD5SIG
1192 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1193#endif
1194
1195 tcp_clear_options(&tmp_opt);
1196 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1197 tmp_opt.user_mss = tp->rx_opt.user_mss;
1198 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1199
1200 if (tmp_opt.cookie_plus > 0 &&
1201 tmp_opt.saw_tstamp &&
1202 !tp->rx_opt.cookie_out_never &&
1203 (sysctl_tcp_cookie_size > 0 ||
1204 (tp->cookie_values != NULL &&
1205 tp->cookie_values->cookie_desired > 0))) {
1206 u8 *c;
1207 u32 *d;
1208 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1209 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1210
1211 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1212 goto drop_and_free;
1213
1214 /* Secret recipe starts with IP addresses */
1215 d = (__force u32 *)&ipv6_hdr(skb)->daddr.s6_addr32[0];
1216 *mess++ ^= *d++;
1217 *mess++ ^= *d++;
1218 *mess++ ^= *d++;
1219 *mess++ ^= *d++;
1220 d = (__force u32 *)&ipv6_hdr(skb)->saddr.s6_addr32[0];
1221 *mess++ ^= *d++;
1222 *mess++ ^= *d++;
1223 *mess++ ^= *d++;
1224 *mess++ ^= *d++;
1225
1226 /* plus variable length Initiator Cookie */
1227 c = (u8 *)mess;
1228 while (l-- > 0)
1229 *c++ ^= *hash_location++;
1230
1231 want_cookie = 0; /* not our kind of cookie */
1232 tmp_ext.cookie_out_never = 0; /* false */
1233 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1234 } else if (!tp->rx_opt.cookie_in_always) {
1235 /* redundant indications, but ensure initialization. */
1236 tmp_ext.cookie_out_never = 1; /* true */
1237 tmp_ext.cookie_plus = 0;
1238 } else {
1239 goto drop_and_free;
1240 }
1241 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1242
1243 if (want_cookie && !tmp_opt.saw_tstamp)
1244 tcp_clear_options(&tmp_opt);
1245
1246 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1247 tcp_openreq_init(req, &tmp_opt, skb);
1248
1249 treq = inet6_rsk(req);
1250 ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1251 ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1252 if (!want_cookie || tmp_opt.tstamp_ok)
1253 TCP_ECN_create_request(req, tcp_hdr(skb));
1254
1255 if (!isn) {
1256 struct inet_peer *peer = NULL;
1257
1258 if (ipv6_opt_accepted(sk, skb) ||
1259 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1260 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1261 atomic_inc(&skb->users);
1262 treq->pktopts = skb;
1263 }
1264 treq->iif = sk->sk_bound_dev_if;
1265
1266 /* So that link locals have meaning */
1267 if (!sk->sk_bound_dev_if &&
1268 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1269 treq->iif = inet6_iif(skb);
1270
1271 if (want_cookie) {
1272 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1273 req->cookie_ts = tmp_opt.tstamp_ok;
1274 goto have_isn;
1275 }
1276
1277 /* VJ's idea. We save last timestamp seen
1278 * from the destination in peer table, when entering
1279 * state TIME-WAIT, and check against it before
1280 * accepting new connection request.
1281 *
1282 * If "isn" is not zero, this request hit alive
1283 * timewait bucket, so that all the necessary checks
1284 * are made in the function processing timewait state.
1285 */
1286 if (tmp_opt.saw_tstamp &&
1287 tcp_death_row.sysctl_tw_recycle &&
1288 (dst = inet6_csk_route_req(sk, req)) != NULL &&
1289 (peer = rt6_get_peer((struct rt6_info *)dst)) != NULL &&
1290 ipv6_addr_equal((struct in6_addr *)peer->daddr.addr.a6,
1291 &treq->rmt_addr)) {
1292 inet_peer_refcheck(peer);
1293 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1294 (s32)(peer->tcp_ts - req->ts_recent) >
1295 TCP_PAWS_WINDOW) {
1296 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1297 goto drop_and_release;
1298 }
1299 }
1300 /* Kill the following clause, if you dislike this way. */
1301 else if (!sysctl_tcp_syncookies &&
1302 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1303 (sysctl_max_syn_backlog >> 2)) &&
1304 (!peer || !peer->tcp_ts_stamp) &&
1305 (!dst || !dst_metric(dst, RTAX_RTT))) {
1306 /* Without syncookies last quarter of
1307 * backlog is filled with destinations,
1308 * proven to be alive.
1309 * It means that we continue to communicate
1310 * to destinations, already remembered
1311 * to the moment of synflood.
1312 */
1313 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1314 &treq->rmt_addr, ntohs(tcp_hdr(skb)->source));
1315 goto drop_and_release;
1316 }
1317
1318 isn = tcp_v6_init_sequence(skb);
1319 }
1320have_isn:
1321 tcp_rsk(req)->snt_isn = isn;
1322 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1323
1324 security_inet_conn_request(sk, skb, req);
1325
1326 if (tcp_v6_send_synack(sk, req,
1327 (struct request_values *)&tmp_ext) ||
1328 want_cookie)
1329 goto drop_and_free;
1330
1331 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1332 return 0;
1333
1334drop_and_release:
1335 dst_release(dst);
1336drop_and_free:
1337 reqsk_free(req);
1338drop:
1339 return 0; /* don't send reset */
1340}
1341
1342static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1343 struct request_sock *req,
1344 struct dst_entry *dst)
1345{
1346 struct inet6_request_sock *treq;
1347 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1348 struct tcp6_sock *newtcp6sk;
1349 struct inet_sock *newinet;
1350 struct tcp_sock *newtp;
1351 struct sock *newsk;
1352 struct ipv6_txoptions *opt;
1353#ifdef CONFIG_TCP_MD5SIG
1354 struct tcp_md5sig_key *key;
1355#endif
1356
1357 if (skb->protocol == htons(ETH_P_IP)) {
1358 /*
1359 * v6 mapped
1360 */
1361
1362 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1363
1364 if (newsk == NULL)
1365 return NULL;
1366
1367 newtcp6sk = (struct tcp6_sock *)newsk;
1368 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1369
1370 newinet = inet_sk(newsk);
1371 newnp = inet6_sk(newsk);
1372 newtp = tcp_sk(newsk);
1373
1374 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1375
1376 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newnp->daddr);
1377
1378 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1379
1380 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1381
1382 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1383 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1384#ifdef CONFIG_TCP_MD5SIG
1385 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1386#endif
1387
1388 newnp->ipv6_ac_list = NULL;
1389 newnp->ipv6_fl_list = NULL;
1390 newnp->pktoptions = NULL;
1391 newnp->opt = NULL;
1392 newnp->mcast_oif = inet6_iif(skb);
1393 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1394
1395 /*
1396 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1397 * here, tcp_create_openreq_child now does this for us, see the comment in
1398 * that function for the gory details. -acme
1399 */
1400
1401 /* It is tricky place. Until this moment IPv4 tcp
1402 worked with IPv6 icsk.icsk_af_ops.
1403 Sync it now.
1404 */
1405 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1406
1407 return newsk;
1408 }
1409
1410 treq = inet6_rsk(req);
1411 opt = np->opt;
1412
1413 if (sk_acceptq_is_full(sk))
1414 goto out_overflow;
1415
1416 if (!dst) {
1417 dst = inet6_csk_route_req(sk, req);
1418 if (!dst)
1419 goto out;
1420 }
1421
1422 newsk = tcp_create_openreq_child(sk, req, skb);
1423 if (newsk == NULL)
1424 goto out_nonewsk;
1425
1426 /*
1427 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1428 * count here, tcp_create_openreq_child now does this for us, see the
1429 * comment in that function for the gory details. -acme
1430 */
1431
1432 newsk->sk_gso_type = SKB_GSO_TCPV6;
1433 __ip6_dst_store(newsk, dst, NULL, NULL);
1434
1435 newtcp6sk = (struct tcp6_sock *)newsk;
1436 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1437
1438 newtp = tcp_sk(newsk);
1439 newinet = inet_sk(newsk);
1440 newnp = inet6_sk(newsk);
1441
1442 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1443
1444 ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1445 ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1446 ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1447 newsk->sk_bound_dev_if = treq->iif;
1448
1449 /* Now IPv6 options...
1450
1451 First: no IPv4 options.
1452 */
1453 newinet->inet_opt = NULL;
1454 newnp->ipv6_ac_list = NULL;
1455 newnp->ipv6_fl_list = NULL;
1456
1457 /* Clone RX bits */
1458 newnp->rxopt.all = np->rxopt.all;
1459
1460 /* Clone pktoptions received with SYN */
1461 newnp->pktoptions = NULL;
1462 if (treq->pktopts != NULL) {
1463 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1464 kfree_skb(treq->pktopts);
1465 treq->pktopts = NULL;
1466 if (newnp->pktoptions)
1467 skb_set_owner_r(newnp->pktoptions, newsk);
1468 }
1469 newnp->opt = NULL;
1470 newnp->mcast_oif = inet6_iif(skb);
1471 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1472
1473 /* Clone native IPv6 options from listening socket (if any)
1474
1475 Yes, keeping reference count would be much more clever,
1476 but we make one more one thing there: reattach optmem
1477 to newsk.
1478 */
1479 if (opt) {
1480 newnp->opt = ipv6_dup_options(newsk, opt);
1481 if (opt != np->opt)
1482 sock_kfree_s(sk, opt, opt->tot_len);
1483 }
1484
1485 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1486 if (newnp->opt)
1487 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1488 newnp->opt->opt_flen);
1489
1490 tcp_mtup_init(newsk);
1491 tcp_sync_mss(newsk, dst_mtu(dst));
1492 newtp->advmss = dst_metric_advmss(dst);
1493 tcp_initialize_rcv_mss(newsk);
1494 if (tcp_rsk(req)->snt_synack)
1495 tcp_valid_rtt_meas(newsk,
1496 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1497 newtp->total_retrans = req->retrans;
1498
1499 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1500 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1501
1502#ifdef CONFIG_TCP_MD5SIG
1503 /* Copy over the MD5 key from the original socket */
1504 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1505 /* We're using one, so create a matching key
1506 * on the newsk structure. If we fail to get
1507 * memory, then we end up not copying the key
1508 * across. Shucks.
1509 */
1510 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1511 if (newkey != NULL)
1512 tcp_v6_md5_do_add(newsk, &newnp->daddr,
1513 newkey, key->keylen);
1514 }
1515#endif
1516
1517 if (__inet_inherit_port(sk, newsk) < 0) {
1518 sock_put(newsk);
1519 goto out;
1520 }
1521 __inet6_hash(newsk, NULL);
1522
1523 return newsk;
1524
1525out_overflow:
1526 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1527out_nonewsk:
1528 if (opt && opt != np->opt)
1529 sock_kfree_s(sk, opt, opt->tot_len);
1530 dst_release(dst);
1531out:
1532 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1533 return NULL;
1534}
1535
1536static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1537{
1538 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1539 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1540 &ipv6_hdr(skb)->daddr, skb->csum)) {
1541 skb->ip_summed = CHECKSUM_UNNECESSARY;
1542 return 0;
1543 }
1544 }
1545
1546 skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1547 &ipv6_hdr(skb)->saddr,
1548 &ipv6_hdr(skb)->daddr, 0));
1549
1550 if (skb->len <= 76) {
1551 return __skb_checksum_complete(skb);
1552 }
1553 return 0;
1554}
1555
1556/* The socket must have it's spinlock held when we get
1557 * here.
1558 *
1559 * We have a potential double-lock case here, so even when
1560 * doing backlog processing we use the BH locking scheme.
1561 * This is because we cannot sleep with the original spinlock
1562 * held.
1563 */
1564static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1565{
1566 struct ipv6_pinfo *np = inet6_sk(sk);
1567 struct tcp_sock *tp;
1568 struct sk_buff *opt_skb = NULL;
1569
1570 /* Imagine: socket is IPv6. IPv4 packet arrives,
1571 goes to IPv4 receive handler and backlogged.
1572 From backlog it always goes here. Kerboom...
1573 Fortunately, tcp_rcv_established and rcv_established
1574 handle them correctly, but it is not case with
1575 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1576 */
1577
1578 if (skb->protocol == htons(ETH_P_IP))
1579 return tcp_v4_do_rcv(sk, skb);
1580
1581#ifdef CONFIG_TCP_MD5SIG
1582 if (tcp_v6_inbound_md5_hash (sk, skb))
1583 goto discard;
1584#endif
1585
1586 if (sk_filter(sk, skb))
1587 goto discard;
1588
1589 /*
1590 * socket locking is here for SMP purposes as backlog rcv
1591 * is currently called with bh processing disabled.
1592 */
1593
1594 /* Do Stevens' IPV6_PKTOPTIONS.
1595
1596 Yes, guys, it is the only place in our code, where we
1597 may make it not affecting IPv4.
1598 The rest of code is protocol independent,
1599 and I do not like idea to uglify IPv4.
1600
1601 Actually, all the idea behind IPV6_PKTOPTIONS
1602 looks not very well thought. For now we latch
1603 options, received in the last packet, enqueued
1604 by tcp. Feel free to propose better solution.
1605 --ANK (980728)
1606 */
1607 if (np->rxopt.all)
1608 opt_skb = skb_clone(skb, GFP_ATOMIC);
1609
1610 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1611 sock_rps_save_rxhash(sk, skb->rxhash);
1612 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1613 goto reset;
1614 if (opt_skb)
1615 goto ipv6_pktoptions;
1616 return 0;
1617 }
1618
1619 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1620 goto csum_err;
1621
1622 if (sk->sk_state == TCP_LISTEN) {
1623 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1624 if (!nsk)
1625 goto discard;
1626
1627 /*
1628 * Queue it on the new socket if the new socket is active,
1629 * otherwise we just shortcircuit this and continue with
1630 * the new socket..
1631 */
1632 if(nsk != sk) {
1633 sock_rps_save_rxhash(nsk, skb->rxhash);
1634 if (tcp_child_process(sk, nsk, skb))
1635 goto reset;
1636 if (opt_skb)
1637 __kfree_skb(opt_skb);
1638 return 0;
1639 }
1640 } else
1641 sock_rps_save_rxhash(sk, skb->rxhash);
1642
1643 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1644 goto reset;
1645 if (opt_skb)
1646 goto ipv6_pktoptions;
1647 return 0;
1648
1649reset:
1650 tcp_v6_send_reset(sk, skb);
1651discard:
1652 if (opt_skb)
1653 __kfree_skb(opt_skb);
1654 kfree_skb(skb);
1655 return 0;
1656csum_err:
1657 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1658 goto discard;
1659
1660
1661ipv6_pktoptions:
1662 /* Do you ask, what is it?
1663
1664 1. skb was enqueued by tcp.
1665 2. skb is added to tail of read queue, rather than out of order.
1666 3. socket is not in passive state.
1667 4. Finally, it really contains options, which user wants to receive.
1668 */
1669 tp = tcp_sk(sk);
1670 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1671 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1672 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1673 np->mcast_oif = inet6_iif(opt_skb);
1674 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1675 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1676 if (ipv6_opt_accepted(sk, opt_skb)) {
1677 skb_set_owner_r(opt_skb, sk);
1678 opt_skb = xchg(&np->pktoptions, opt_skb);
1679 } else {
1680 __kfree_skb(opt_skb);
1681 opt_skb = xchg(&np->pktoptions, NULL);
1682 }
1683 }
1684
1685 kfree_skb(opt_skb);
1686 return 0;
1687}
1688
1689static int tcp_v6_rcv(struct sk_buff *skb)
1690{
1691 struct tcphdr *th;
1692 const struct ipv6hdr *hdr;
1693 struct sock *sk;
1694 int ret;
1695 struct net *net = dev_net(skb->dev);
1696
1697 if (skb->pkt_type != PACKET_HOST)
1698 goto discard_it;
1699
1700 /*
1701 * Count it even if it's bad.
1702 */
1703 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1704
1705 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1706 goto discard_it;
1707
1708 th = tcp_hdr(skb);
1709
1710 if (th->doff < sizeof(struct tcphdr)/4)
1711 goto bad_packet;
1712 if (!pskb_may_pull(skb, th->doff*4))
1713 goto discard_it;
1714
1715 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1716 goto bad_packet;
1717
1718 th = tcp_hdr(skb);
1719 hdr = ipv6_hdr(skb);
1720 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1721 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1722 skb->len - th->doff*4);
1723 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1724 TCP_SKB_CB(skb)->when = 0;
1725 TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(hdr);
1726 TCP_SKB_CB(skb)->sacked = 0;
1727
1728 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1729 if (!sk)
1730 goto no_tcp_socket;
1731
1732process:
1733 if (sk->sk_state == TCP_TIME_WAIT)
1734 goto do_time_wait;
1735
1736 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1737 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1738 goto discard_and_relse;
1739 }
1740
1741 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1742 goto discard_and_relse;
1743
1744 if (sk_filter(sk, skb))
1745 goto discard_and_relse;
1746
1747 skb->dev = NULL;
1748
1749 bh_lock_sock_nested(sk);
1750 ret = 0;
1751 if (!sock_owned_by_user(sk)) {
1752#ifdef CONFIG_NET_DMA
1753 struct tcp_sock *tp = tcp_sk(sk);
1754 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1755 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1756 if (tp->ucopy.dma_chan)
1757 ret = tcp_v6_do_rcv(sk, skb);
1758 else
1759#endif
1760 {
1761 if (!tcp_prequeue(sk, skb))
1762 ret = tcp_v6_do_rcv(sk, skb);
1763 }
1764 } else if (unlikely(sk_add_backlog(sk, skb))) {
1765 bh_unlock_sock(sk);
1766 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1767 goto discard_and_relse;
1768 }
1769 bh_unlock_sock(sk);
1770
1771 sock_put(sk);
1772 return ret ? -1 : 0;
1773
1774no_tcp_socket:
1775 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1776 goto discard_it;
1777
1778 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1779bad_packet:
1780 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1781 } else {
1782 tcp_v6_send_reset(NULL, skb);
1783 }
1784
1785discard_it:
1786
1787 /*
1788 * Discard frame
1789 */
1790
1791 kfree_skb(skb);
1792 return 0;
1793
1794discard_and_relse:
1795 sock_put(sk);
1796 goto discard_it;
1797
1798do_time_wait:
1799 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1800 inet_twsk_put(inet_twsk(sk));
1801 goto discard_it;
1802 }
1803
1804 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1805 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1806 inet_twsk_put(inet_twsk(sk));
1807 goto discard_it;
1808 }
1809
1810 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1811 case TCP_TW_SYN:
1812 {
1813 struct sock *sk2;
1814
1815 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1816 &ipv6_hdr(skb)->daddr,
1817 ntohs(th->dest), inet6_iif(skb));
1818 if (sk2 != NULL) {
1819 struct inet_timewait_sock *tw = inet_twsk(sk);
1820 inet_twsk_deschedule(tw, &tcp_death_row);
1821 inet_twsk_put(tw);
1822 sk = sk2;
1823 goto process;
1824 }
1825 /* Fall through to ACK */
1826 }
1827 case TCP_TW_ACK:
1828 tcp_v6_timewait_ack(sk, skb);
1829 break;
1830 case TCP_TW_RST:
1831 goto no_tcp_socket;
1832 case TCP_TW_SUCCESS:;
1833 }
1834 goto discard_it;
1835}
1836
1837static struct inet_peer *tcp_v6_get_peer(struct sock *sk, bool *release_it)
1838{
1839 struct rt6_info *rt = (struct rt6_info *) __sk_dst_get(sk);
1840 struct ipv6_pinfo *np = inet6_sk(sk);
1841 struct inet_peer *peer;
1842
1843 if (!rt ||
1844 !ipv6_addr_equal(&np->daddr, &rt->rt6i_dst.addr)) {
1845 peer = inet_getpeer_v6(&np->daddr, 1);
1846 *release_it = true;
1847 } else {
1848 if (!rt->rt6i_peer)
1849 rt6_bind_peer(rt, 1);
1850 peer = rt->rt6i_peer;
1851 *release_it = false;
1852 }
1853
1854 return peer;
1855}
1856
1857static void *tcp_v6_tw_get_peer(struct sock *sk)
1858{
1859 struct inet6_timewait_sock *tw6 = inet6_twsk(sk);
1860 struct inet_timewait_sock *tw = inet_twsk(sk);
1861
1862 if (tw->tw_family == AF_INET)
1863 return tcp_v4_tw_get_peer(sk);
1864
1865 return inet_getpeer_v6(&tw6->tw_v6_daddr, 1);
1866}
1867
1868static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1869 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1870 .twsk_unique = tcp_twsk_unique,
1871 .twsk_destructor= tcp_twsk_destructor,
1872 .twsk_getpeer = tcp_v6_tw_get_peer,
1873};
1874
1875static const struct inet_connection_sock_af_ops ipv6_specific = {
1876 .queue_xmit = inet6_csk_xmit,
1877 .send_check = tcp_v6_send_check,
1878 .rebuild_header = inet6_sk_rebuild_header,
1879 .conn_request = tcp_v6_conn_request,
1880 .syn_recv_sock = tcp_v6_syn_recv_sock,
1881 .get_peer = tcp_v6_get_peer,
1882 .net_header_len = sizeof(struct ipv6hdr),
1883 .setsockopt = ipv6_setsockopt,
1884 .getsockopt = ipv6_getsockopt,
1885 .addr2sockaddr = inet6_csk_addr2sockaddr,
1886 .sockaddr_len = sizeof(struct sockaddr_in6),
1887 .bind_conflict = inet6_csk_bind_conflict,
1888#ifdef CONFIG_COMPAT
1889 .compat_setsockopt = compat_ipv6_setsockopt,
1890 .compat_getsockopt = compat_ipv6_getsockopt,
1891#endif
1892};
1893
1894#ifdef CONFIG_TCP_MD5SIG
1895static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1896 .md5_lookup = tcp_v6_md5_lookup,
1897 .calc_md5_hash = tcp_v6_md5_hash_skb,
1898 .md5_add = tcp_v6_md5_add_func,
1899 .md5_parse = tcp_v6_parse_md5_keys,
1900};
1901#endif
1902
1903/*
1904 * TCP over IPv4 via INET6 API
1905 */
1906
1907static const struct inet_connection_sock_af_ops ipv6_mapped = {
1908 .queue_xmit = ip_queue_xmit,
1909 .send_check = tcp_v4_send_check,
1910 .rebuild_header = inet_sk_rebuild_header,
1911 .conn_request = tcp_v6_conn_request,
1912 .syn_recv_sock = tcp_v6_syn_recv_sock,
1913 .get_peer = tcp_v4_get_peer,
1914 .net_header_len = sizeof(struct iphdr),
1915 .setsockopt = ipv6_setsockopt,
1916 .getsockopt = ipv6_getsockopt,
1917 .addr2sockaddr = inet6_csk_addr2sockaddr,
1918 .sockaddr_len = sizeof(struct sockaddr_in6),
1919 .bind_conflict = inet6_csk_bind_conflict,
1920#ifdef CONFIG_COMPAT
1921 .compat_setsockopt = compat_ipv6_setsockopt,
1922 .compat_getsockopt = compat_ipv6_getsockopt,
1923#endif
1924};
1925
1926#ifdef CONFIG_TCP_MD5SIG
1927static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1928 .md5_lookup = tcp_v4_md5_lookup,
1929 .calc_md5_hash = tcp_v4_md5_hash_skb,
1930 .md5_add = tcp_v6_md5_add_func,
1931 .md5_parse = tcp_v6_parse_md5_keys,
1932};
1933#endif
1934
1935/* NOTE: A lot of things set to zero explicitly by call to
1936 * sk_alloc() so need not be done here.
1937 */
1938static int tcp_v6_init_sock(struct sock *sk)
1939{
1940 struct inet_connection_sock *icsk = inet_csk(sk);
1941 struct tcp_sock *tp = tcp_sk(sk);
1942
1943 skb_queue_head_init(&tp->out_of_order_queue);
1944 tcp_init_xmit_timers(sk);
1945 tcp_prequeue_init(tp);
1946
1947 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1948 tp->mdev = TCP_TIMEOUT_INIT;
1949
1950 /* So many TCP implementations out there (incorrectly) count the
1951 * initial SYN frame in their delayed-ACK and congestion control
1952 * algorithms that we must have the following bandaid to talk
1953 * efficiently to them. -DaveM
1954 */
1955 tp->snd_cwnd = 2;
1956
1957 /* See draft-stevens-tcpca-spec-01 for discussion of the
1958 * initialization of these values.
1959 */
1960 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1961 tp->snd_cwnd_clamp = ~0;
1962 tp->mss_cache = TCP_MSS_DEFAULT;
1963
1964 tp->reordering = sysctl_tcp_reordering;
1965
1966 sk->sk_state = TCP_CLOSE;
1967
1968 icsk->icsk_af_ops = &ipv6_specific;
1969 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1970 icsk->icsk_sync_mss = tcp_sync_mss;
1971 sk->sk_write_space = sk_stream_write_space;
1972 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1973
1974#ifdef CONFIG_TCP_MD5SIG
1975 tp->af_specific = &tcp_sock_ipv6_specific;
1976#endif
1977
1978 /* TCP Cookie Transactions */
1979 if (sysctl_tcp_cookie_size > 0) {
1980 /* Default, cookies without s_data_payload. */
1981 tp->cookie_values =
1982 kzalloc(sizeof(*tp->cookie_values),
1983 sk->sk_allocation);
1984 if (tp->cookie_values != NULL)
1985 kref_init(&tp->cookie_values->kref);
1986 }
1987 /* Presumed zeroed, in order of appearance:
1988 * cookie_in_always, cookie_out_never,
1989 * s_data_constant, s_data_in, s_data_out
1990 */
1991 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1992 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1993
1994 local_bh_disable();
1995 percpu_counter_inc(&tcp_sockets_allocated);
1996 local_bh_enable();
1997
1998 return 0;
1999}
2000
2001static void tcp_v6_destroy_sock(struct sock *sk)
2002{
2003#ifdef CONFIG_TCP_MD5SIG
2004 /* Clean up the MD5 key list */
2005 if (tcp_sk(sk)->md5sig_info)
2006 tcp_v6_clear_md5_list(sk);
2007#endif
2008 tcp_v4_destroy_sock(sk);
2009 inet6_destroy_sock(sk);
2010}
2011
2012#ifdef CONFIG_PROC_FS
2013/* Proc filesystem TCPv6 sock list dumping. */
2014static void get_openreq6(struct seq_file *seq,
2015 struct sock *sk, struct request_sock *req, int i, int uid)
2016{
2017 int ttd = req->expires - jiffies;
2018 const struct in6_addr *src = &inet6_rsk(req)->loc_addr;
2019 const struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
2020
2021 if (ttd < 0)
2022 ttd = 0;
2023
2024 seq_printf(seq,
2025 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2026 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2027 i,
2028 src->s6_addr32[0], src->s6_addr32[1],
2029 src->s6_addr32[2], src->s6_addr32[3],
2030 ntohs(inet_rsk(req)->loc_port),
2031 dest->s6_addr32[0], dest->s6_addr32[1],
2032 dest->s6_addr32[2], dest->s6_addr32[3],
2033 ntohs(inet_rsk(req)->rmt_port),
2034 TCP_SYN_RECV,
2035 0,0, /* could print option size, but that is af dependent. */
2036 1, /* timers active (only the expire timer) */
2037 jiffies_to_clock_t(ttd),
2038 req->retrans,
2039 uid,
2040 0, /* non standard timer */
2041 0, /* open_requests have no inode */
2042 0, req);
2043}
2044
2045static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2046{
2047 const struct in6_addr *dest, *src;
2048 __u16 destp, srcp;
2049 int timer_active;
2050 unsigned long timer_expires;
2051 struct inet_sock *inet = inet_sk(sp);
2052 struct tcp_sock *tp = tcp_sk(sp);
2053 const struct inet_connection_sock *icsk = inet_csk(sp);
2054 struct ipv6_pinfo *np = inet6_sk(sp);
2055
2056 dest = &np->daddr;
2057 src = &np->rcv_saddr;
2058 destp = ntohs(inet->inet_dport);
2059 srcp = ntohs(inet->inet_sport);
2060
2061 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2062 timer_active = 1;
2063 timer_expires = icsk->icsk_timeout;
2064 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2065 timer_active = 4;
2066 timer_expires = icsk->icsk_timeout;
2067 } else if (timer_pending(&sp->sk_timer)) {
2068 timer_active = 2;
2069 timer_expires = sp->sk_timer.expires;
2070 } else {
2071 timer_active = 0;
2072 timer_expires = jiffies;
2073 }
2074
2075 seq_printf(seq,
2076 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2077 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %lu %lu %u %u %d\n",
2078 i,
2079 src->s6_addr32[0], src->s6_addr32[1],
2080 src->s6_addr32[2], src->s6_addr32[3], srcp,
2081 dest->s6_addr32[0], dest->s6_addr32[1],
2082 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2083 sp->sk_state,
2084 tp->write_seq-tp->snd_una,
2085 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2086 timer_active,
2087 jiffies_to_clock_t(timer_expires - jiffies),
2088 icsk->icsk_retransmits,
2089 sock_i_uid(sp),
2090 icsk->icsk_probes_out,
2091 sock_i_ino(sp),
2092 atomic_read(&sp->sk_refcnt), sp,
2093 jiffies_to_clock_t(icsk->icsk_rto),
2094 jiffies_to_clock_t(icsk->icsk_ack.ato),
2095 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2096 tp->snd_cwnd,
2097 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
2098 );
2099}
2100
2101static void get_timewait6_sock(struct seq_file *seq,
2102 struct inet_timewait_sock *tw, int i)
2103{
2104 const struct in6_addr *dest, *src;
2105 __u16 destp, srcp;
2106 struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2107 int ttd = tw->tw_ttd - jiffies;
2108
2109 if (ttd < 0)
2110 ttd = 0;
2111
2112 dest = &tw6->tw_v6_daddr;
2113 src = &tw6->tw_v6_rcv_saddr;
2114 destp = ntohs(tw->tw_dport);
2115 srcp = ntohs(tw->tw_sport);
2116
2117 seq_printf(seq,
2118 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2119 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2120 i,
2121 src->s6_addr32[0], src->s6_addr32[1],
2122 src->s6_addr32[2], src->s6_addr32[3], srcp,
2123 dest->s6_addr32[0], dest->s6_addr32[1],
2124 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2125 tw->tw_substate, 0, 0,
2126 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2127 atomic_read(&tw->tw_refcnt), tw);
2128}
2129
2130static int tcp6_seq_show(struct seq_file *seq, void *v)
2131{
2132 struct tcp_iter_state *st;
2133
2134 if (v == SEQ_START_TOKEN) {
2135 seq_puts(seq,
2136 " sl "
2137 "local_address "
2138 "remote_address "
2139 "st tx_queue rx_queue tr tm->when retrnsmt"
2140 " uid timeout inode\n");
2141 goto out;
2142 }
2143 st = seq->private;
2144
2145 switch (st->state) {
2146 case TCP_SEQ_STATE_LISTENING:
2147 case TCP_SEQ_STATE_ESTABLISHED:
2148 get_tcp6_sock(seq, v, st->num);
2149 break;
2150 case TCP_SEQ_STATE_OPENREQ:
2151 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2152 break;
2153 case TCP_SEQ_STATE_TIME_WAIT:
2154 get_timewait6_sock(seq, v, st->num);
2155 break;
2156 }
2157out:
2158 return 0;
2159}
2160
2161static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2162 .name = "tcp6",
2163 .family = AF_INET6,
2164 .seq_fops = {
2165 .owner = THIS_MODULE,
2166 },
2167 .seq_ops = {
2168 .show = tcp6_seq_show,
2169 },
2170};
2171
2172int __net_init tcp6_proc_init(struct net *net)
2173{
2174 return tcp_proc_register(net, &tcp6_seq_afinfo);
2175}
2176
2177void tcp6_proc_exit(struct net *net)
2178{
2179 tcp_proc_unregister(net, &tcp6_seq_afinfo);
2180}
2181#endif
2182
2183struct proto tcpv6_prot = {
2184 .name = "TCPv6",
2185 .owner = THIS_MODULE,
2186 .close = tcp_close,
2187 .connect = tcp_v6_connect,
2188 .disconnect = tcp_disconnect,
2189 .accept = inet_csk_accept,
2190 .ioctl = tcp_ioctl,
2191 .init = tcp_v6_init_sock,
2192 .destroy = tcp_v6_destroy_sock,
2193 .shutdown = tcp_shutdown,
2194 .setsockopt = tcp_setsockopt,
2195 .getsockopt = tcp_getsockopt,
2196 .recvmsg = tcp_recvmsg,
2197 .sendmsg = tcp_sendmsg,
2198 .sendpage = tcp_sendpage,
2199 .backlog_rcv = tcp_v6_do_rcv,
2200 .hash = tcp_v6_hash,
2201 .unhash = inet_unhash,
2202 .get_port = inet_csk_get_port,
2203 .enter_memory_pressure = tcp_enter_memory_pressure,
2204 .sockets_allocated = &tcp_sockets_allocated,
2205 .memory_allocated = &tcp_memory_allocated,
2206 .memory_pressure = &tcp_memory_pressure,
2207 .orphan_count = &tcp_orphan_count,
2208 .sysctl_mem = sysctl_tcp_mem,
2209 .sysctl_wmem = sysctl_tcp_wmem,
2210 .sysctl_rmem = sysctl_tcp_rmem,
2211 .max_header = MAX_TCP_HEADER,
2212 .obj_size = sizeof(struct tcp6_sock),
2213 .slab_flags = SLAB_DESTROY_BY_RCU,
2214 .twsk_prot = &tcp6_timewait_sock_ops,
2215 .rsk_prot = &tcp6_request_sock_ops,
2216 .h.hashinfo = &tcp_hashinfo,
2217 .no_autobind = true,
2218#ifdef CONFIG_COMPAT
2219 .compat_setsockopt = compat_tcp_setsockopt,
2220 .compat_getsockopt = compat_tcp_getsockopt,
2221#endif
2222};
2223
2224static const struct inet6_protocol tcpv6_protocol = {
2225 .handler = tcp_v6_rcv,
2226 .err_handler = tcp_v6_err,
2227 .gso_send_check = tcp_v6_gso_send_check,
2228 .gso_segment = tcp_tso_segment,
2229 .gro_receive = tcp6_gro_receive,
2230 .gro_complete = tcp6_gro_complete,
2231 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2232};
2233
2234static struct inet_protosw tcpv6_protosw = {
2235 .type = SOCK_STREAM,
2236 .protocol = IPPROTO_TCP,
2237 .prot = &tcpv6_prot,
2238 .ops = &inet6_stream_ops,
2239 .no_check = 0,
2240 .flags = INET_PROTOSW_PERMANENT |
2241 INET_PROTOSW_ICSK,
2242};
2243
2244static int __net_init tcpv6_net_init(struct net *net)
2245{
2246 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2247 SOCK_RAW, IPPROTO_TCP, net);
2248}
2249
2250static void __net_exit tcpv6_net_exit(struct net *net)
2251{
2252 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2253}
2254
2255static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2256{
2257 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2258}
2259
2260static struct pernet_operations tcpv6_net_ops = {
2261 .init = tcpv6_net_init,
2262 .exit = tcpv6_net_exit,
2263 .exit_batch = tcpv6_net_exit_batch,
2264};
2265
2266int __init tcpv6_init(void)
2267{
2268 int ret;
2269
2270 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2271 if (ret)
2272 goto out;
2273
2274 /* register inet6 protocol */
2275 ret = inet6_register_protosw(&tcpv6_protosw);
2276 if (ret)
2277 goto out_tcpv6_protocol;
2278
2279 ret = register_pernet_subsys(&tcpv6_net_ops);
2280 if (ret)
2281 goto out_tcpv6_protosw;
2282out:
2283 return ret;
2284
2285out_tcpv6_protocol:
2286 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2287out_tcpv6_protosw:
2288 inet6_unregister_protosw(&tcpv6_protosw);
2289 goto out;
2290}
2291
2292void tcpv6_exit(void)
2293{
2294 unregister_pernet_subsys(&tcpv6_net_ops);
2295 inet6_unregister_protosw(&tcpv6_protosw);
2296 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2297}