]> git.ipfire.org Git - people/ms/linux.git/blob - net/ipv6/tcp_ipv6.c
projects / people / ms / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[people/ms/linux.git] / net / ipv6 / tcp_ipv6.c
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/tcp_memcontrol.h>
65 #include <net/busy_poll.h>
66
67 #include <linux/proc_fs.h>
68 #include <linux/seq_file.h>
69
70 #include <linux/crypto.h>
71 #include <linux/scatterlist.h>
72
73 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
74 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
75 struct request_sock *req);
76
77 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
78
79 static const struct inet_connection_sock_af_ops ipv6_mapped;
80 static const struct inet_connection_sock_af_ops ipv6_specific;
81 #ifdef CONFIG_TCP_MD5SIG
82 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
83 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
84 #else
85 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
86 const struct in6_addr *addr)
87 {
88 return NULL;
89 }
90 #endif
91
92 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
93 {
94 struct dst_entry *dst = skb_dst(skb);
95
96 if (dst) {
97 const struct rt6_info *rt = (const struct rt6_info *)dst;
98
99 dst_hold(dst);
100 sk->sk_rx_dst = dst;
101 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
102 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
103 }
104 }
105
106 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
107 {
108 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
109 ipv6_hdr(skb)->saddr.s6_addr32,
110 tcp_hdr(skb)->dest,
111 tcp_hdr(skb)->source);
112 }
113
114 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
115 int addr_len)
116 {
117 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
118 struct inet_sock *inet = inet_sk(sk);
119 struct inet_connection_sock *icsk = inet_csk(sk);
120 struct ipv6_pinfo *np = inet6_sk(sk);
121 struct tcp_sock *tp = tcp_sk(sk);
122 struct in6_addr *saddr = NULL, *final_p, final;
123 struct ipv6_txoptions *opt;
124 struct flowi6 fl6;
125 struct dst_entry *dst;
126 int addr_type;
127 int err;
128
129 if (addr_len < SIN6_LEN_RFC2133)
130 return -EINVAL;
131
132 if (usin->sin6_family != AF_INET6)
133 return -EAFNOSUPPORT;
134
135 memset(&fl6, 0, sizeof(fl6));
136
137 if (np->sndflow) {
138 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
139 IP6_ECN_flow_init(fl6.flowlabel);
140 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
141 struct ip6_flowlabel *flowlabel;
142 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
143 if (!flowlabel)
144 return -EINVAL;
145 fl6_sock_release(flowlabel);
146 }
147 }
148
149 /*
150 * connect() to INADDR_ANY means loopback (BSD'ism).
151 */
152
153 if (ipv6_addr_any(&usin->sin6_addr))
154 usin->sin6_addr.s6_addr[15] = 0x1;
155
156 addr_type = ipv6_addr_type(&usin->sin6_addr);
157
158 if (addr_type & IPV6_ADDR_MULTICAST)
159 return -ENETUNREACH;
160
161 if (addr_type&IPV6_ADDR_LINKLOCAL) {
162 if (addr_len >= sizeof(struct sockaddr_in6) &&
163 usin->sin6_scope_id) {
164 /* If interface is set while binding, indices
165 * must coincide.
166 */
167 if (sk->sk_bound_dev_if &&
168 sk->sk_bound_dev_if != usin->sin6_scope_id)
169 return -EINVAL;
170
171 sk->sk_bound_dev_if = usin->sin6_scope_id;
172 }
173
174 /* Connect to link-local address requires an interface */
175 if (!sk->sk_bound_dev_if)
176 return -EINVAL;
177 }
178
179 if (tp->rx_opt.ts_recent_stamp &&
180 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
181 tp->rx_opt.ts_recent = 0;
182 tp->rx_opt.ts_recent_stamp = 0;
183 tp->write_seq = 0;
184 }
185
186 sk->sk_v6_daddr = usin->sin6_addr;
187 np->flow_label = fl6.flowlabel;
188
189 /*
190 * TCP over IPv4
191 */
192
193 if (addr_type == IPV6_ADDR_MAPPED) {
194 u32 exthdrlen = icsk->icsk_ext_hdr_len;
195 struct sockaddr_in sin;
196
197 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
198
199 if (__ipv6_only_sock(sk))
200 return -ENETUNREACH;
201
202 sin.sin_family = AF_INET;
203 sin.sin_port = usin->sin6_port;
204 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
205
206 icsk->icsk_af_ops = &ipv6_mapped;
207 sk->sk_backlog_rcv = tcp_v4_do_rcv;
208 #ifdef CONFIG_TCP_MD5SIG
209 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
210 #endif
211
212 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
213
214 if (err) {
215 icsk->icsk_ext_hdr_len = exthdrlen;
216 icsk->icsk_af_ops = &ipv6_specific;
217 sk->sk_backlog_rcv = tcp_v6_do_rcv;
218 #ifdef CONFIG_TCP_MD5SIG
219 tp->af_specific = &tcp_sock_ipv6_specific;
220 #endif
221 goto failure;
222 }
223 np->saddr = sk->sk_v6_rcv_saddr;
224
225 return err;
226 }
227
228 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
229 saddr = &sk->sk_v6_rcv_saddr;
230
231 fl6.flowi6_proto = IPPROTO_TCP;
232 fl6.daddr = sk->sk_v6_daddr;
233 fl6.saddr = saddr ? *saddr : np->saddr;
234 fl6.flowi6_oif = sk->sk_bound_dev_if;
235 fl6.flowi6_mark = sk->sk_mark;
236 fl6.fl6_dport = usin->sin6_port;
237 fl6.fl6_sport = inet->inet_sport;
238
239 opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
240 final_p = fl6_update_dst(&fl6, opt, &final);
241
242 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
243
244 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
245 if (IS_ERR(dst)) {
246 err = PTR_ERR(dst);
247 goto failure;
248 }
249
250 if (!saddr) {
251 saddr = &fl6.saddr;
252 sk->sk_v6_rcv_saddr = *saddr;
253 }
254
255 /* set the source address */
256 np->saddr = *saddr;
257 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
258
259 sk->sk_gso_type = SKB_GSO_TCPV6;
260 ip6_dst_store(sk, dst, NULL, NULL);
261
262 if (tcp_death_row.sysctl_tw_recycle &&
263 !tp->rx_opt.ts_recent_stamp &&
264 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
265 tcp_fetch_timewait_stamp(sk, dst);
266
267 icsk->icsk_ext_hdr_len = 0;
268 if (opt)
269 icsk->icsk_ext_hdr_len = opt->opt_flen +
270 opt->opt_nflen;
271
272 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
273
274 inet->inet_dport = usin->sin6_port;
275
276 tcp_set_state(sk, TCP_SYN_SENT);
277 err = inet6_hash_connect(&tcp_death_row, sk);
278 if (err)
279 goto late_failure;
280
281 sk_set_txhash(sk);
282
283 if (!tp->write_seq && likely(!tp->repair))
284 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
285 sk->sk_v6_daddr.s6_addr32,
286 inet->inet_sport,
287 inet->inet_dport);
288
289 err = tcp_connect(sk);
290 if (err)
291 goto late_failure;
292
293 return 0;
294
295 late_failure:
296 tcp_set_state(sk, TCP_CLOSE);
297 __sk_dst_reset(sk);
298 failure:
299 inet->inet_dport = 0;
300 sk->sk_route_caps = 0;
301 return err;
302 }
303
304 static void tcp_v6_mtu_reduced(struct sock *sk)
305 {
306 struct dst_entry *dst;
307
308 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
309 return;
310
311 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
312 if (!dst)
313 return;
314
315 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
316 tcp_sync_mss(sk, dst_mtu(dst));
317 tcp_simple_retransmit(sk);
318 }
319 }
320
321 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
322 u8 type, u8 code, int offset, __be32 info)
323 {
324 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
325 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
326 struct net *net = dev_net(skb->dev);
327 struct request_sock *fastopen;
328 struct ipv6_pinfo *np;
329 struct tcp_sock *tp;
330 __u32 seq, snd_una;
331 struct sock *sk;
332 int err;
333
334 sk = __inet6_lookup_established(net, &tcp_hashinfo,
335 &hdr->daddr, th->dest,
336 &hdr->saddr, ntohs(th->source),
337 skb->dev->ifindex);
338
339 if (!sk) {
340 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
341 ICMP6_MIB_INERRORS);
342 return;
343 }
344
345 if (sk->sk_state == TCP_TIME_WAIT) {
346 inet_twsk_put(inet_twsk(sk));
347 return;
348 }
349 seq = ntohl(th->seq);
350 if (sk->sk_state == TCP_NEW_SYN_RECV)
351 return tcp_req_err(sk, seq);
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 icmpv6_err_convert(type, code, &err);
406
407 /* Might be for an request_sock */
408 switch (sk->sk_state) {
409 case TCP_SYN_SENT:
410 case TCP_SYN_RECV:
411 /* Only in fast or simultaneous open. If a fast open socket is
412 * is already accepted it is treated as a connected one below.
413 */
414 if (fastopen && !fastopen->sk)
415 break;
416
417 if (!sock_owned_by_user(sk)) {
418 sk->sk_err = err;
419 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
420
421 tcp_done(sk);
422 } else
423 sk->sk_err_soft = err;
424 goto out;
425 }
426
427 if (!sock_owned_by_user(sk) && np->recverr) {
428 sk->sk_err = err;
429 sk->sk_error_report(sk);
430 } else
431 sk->sk_err_soft = err;
432
433 out:
434 bh_unlock_sock(sk);
435 sock_put(sk);
436 }
437
438
439 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
440 struct flowi *fl,
441 struct request_sock *req,
442 struct tcp_fastopen_cookie *foc,
443 bool attach_req)
444 {
445 struct inet_request_sock *ireq = inet_rsk(req);
446 struct ipv6_pinfo *np = inet6_sk(sk);
447 struct flowi6 *fl6 = &fl->u.ip6;
448 struct sk_buff *skb;
449 int err = -ENOMEM;
450
451 /* First, grab a route. */
452 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
453 IPPROTO_TCP)) == NULL)
454 goto done;
455
456 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
457
458 if (skb) {
459 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
460 &ireq->ir_v6_rmt_addr);
461
462 fl6->daddr = ireq->ir_v6_rmt_addr;
463 if (np->repflow && ireq->pktopts)
464 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
465
466 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
467 np->tclass);
468 err = net_xmit_eval(err);
469 }
470
471 done:
472 return err;
473 }
474
475
476 static void tcp_v6_reqsk_destructor(struct request_sock *req)
477 {
478 kfree_skb(inet_rsk(req)->pktopts);
479 }
480
481 #ifdef CONFIG_TCP_MD5SIG
482 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
483 const struct in6_addr *addr)
484 {
485 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
486 }
487
488 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
489 const struct sock *addr_sk)
490 {
491 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
492 }
493
494 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
495 int optlen)
496 {
497 struct tcp_md5sig cmd;
498 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
499
500 if (optlen < sizeof(cmd))
501 return -EINVAL;
502
503 if (copy_from_user(&cmd, optval, sizeof(cmd)))
504 return -EFAULT;
505
506 if (sin6->sin6_family != AF_INET6)
507 return -EINVAL;
508
509 if (!cmd.tcpm_keylen) {
510 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
511 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
512 AF_INET);
513 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
514 AF_INET6);
515 }
516
517 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
518 return -EINVAL;
519
520 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
521 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
522 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
523
524 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
525 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
526 }
527
528 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
529 const struct in6_addr *daddr,
530 const struct in6_addr *saddr, int nbytes)
531 {
532 struct tcp6_pseudohdr *bp;
533 struct scatterlist sg;
534
535 bp = &hp->md5_blk.ip6;
536 /* 1. TCP pseudo-header (RFC2460) */
537 bp->saddr = *saddr;
538 bp->daddr = *daddr;
539 bp->protocol = cpu_to_be32(IPPROTO_TCP);
540 bp->len = cpu_to_be32(nbytes);
541
542 sg_init_one(&sg, bp, sizeof(*bp));
543 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
544 }
545
546 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
547 const struct in6_addr *daddr, struct in6_addr *saddr,
548 const struct tcphdr *th)
549 {
550 struct tcp_md5sig_pool *hp;
551 struct hash_desc *desc;
552
553 hp = tcp_get_md5sig_pool();
554 if (!hp)
555 goto clear_hash_noput;
556 desc = &hp->md5_desc;
557
558 if (crypto_hash_init(desc))
559 goto clear_hash;
560 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
561 goto clear_hash;
562 if (tcp_md5_hash_header(hp, th))
563 goto clear_hash;
564 if (tcp_md5_hash_key(hp, key))
565 goto clear_hash;
566 if (crypto_hash_final(desc, md5_hash))
567 goto clear_hash;
568
569 tcp_put_md5sig_pool();
570 return 0;
571
572 clear_hash:
573 tcp_put_md5sig_pool();
574 clear_hash_noput:
575 memset(md5_hash, 0, 16);
576 return 1;
577 }
578
579 static int tcp_v6_md5_hash_skb(char *md5_hash,
580 const struct tcp_md5sig_key *key,
581 const struct sock *sk,
582 const struct sk_buff *skb)
583 {
584 const struct in6_addr *saddr, *daddr;
585 struct tcp_md5sig_pool *hp;
586 struct hash_desc *desc;
587 const struct tcphdr *th = tcp_hdr(skb);
588
589 if (sk) { /* valid for establish/request sockets */
590 saddr = &sk->sk_v6_rcv_saddr;
591 daddr = &sk->sk_v6_daddr;
592 } else {
593 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
594 saddr = &ip6h->saddr;
595 daddr = &ip6h->daddr;
596 }
597
598 hp = tcp_get_md5sig_pool();
599 if (!hp)
600 goto clear_hash_noput;
601 desc = &hp->md5_desc;
602
603 if (crypto_hash_init(desc))
604 goto clear_hash;
605
606 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
607 goto clear_hash;
608 if (tcp_md5_hash_header(hp, th))
609 goto clear_hash;
610 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
611 goto clear_hash;
612 if (tcp_md5_hash_key(hp, key))
613 goto clear_hash;
614 if (crypto_hash_final(desc, md5_hash))
615 goto clear_hash;
616
617 tcp_put_md5sig_pool();
618 return 0;
619
620 clear_hash:
621 tcp_put_md5sig_pool();
622 clear_hash_noput:
623 memset(md5_hash, 0, 16);
624 return 1;
625 }
626
627 #endif
628
629 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
630 const struct sk_buff *skb)
631 {
632 #ifdef CONFIG_TCP_MD5SIG
633 const __u8 *hash_location = NULL;
634 struct tcp_md5sig_key *hash_expected;
635 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
636 const struct tcphdr *th = tcp_hdr(skb);
637 int genhash;
638 u8 newhash[16];
639
640 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
641 hash_location = tcp_parse_md5sig_option(th);
642
643 /* We've parsed the options - do we have a hash? */
644 if (!hash_expected && !hash_location)
645 return false;
646
647 if (hash_expected && !hash_location) {
648 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
649 return true;
650 }
651
652 if (!hash_expected && hash_location) {
653 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
654 return true;
655 }
656
657 /* check the signature */
658 genhash = tcp_v6_md5_hash_skb(newhash,
659 hash_expected,
660 NULL, skb);
661
662 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
663 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
664 genhash ? "failed" : "mismatch",
665 &ip6h->saddr, ntohs(th->source),
666 &ip6h->daddr, ntohs(th->dest));
667 return true;
668 }
669 #endif
670 return false;
671 }
672
673 static void tcp_v6_init_req(struct request_sock *req,
674 const struct sock *sk_listener,
675 struct sk_buff *skb)
676 {
677 struct inet_request_sock *ireq = inet_rsk(req);
678 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
679
680 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
681 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
682
683 /* So that link locals have meaning */
684 if (!sk_listener->sk_bound_dev_if &&
685 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
686 ireq->ir_iif = tcp_v6_iif(skb);
687
688 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
689 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
690 np->rxopt.bits.rxinfo ||
691 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
692 np->rxopt.bits.rxohlim || np->repflow)) {
693 atomic_inc(&skb->users);
694 ireq->pktopts = skb;
695 }
696 }
697
698 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
699 struct flowi *fl,
700 const struct request_sock *req,
701 bool *strict)
702 {
703 if (strict)
704 *strict = true;
705 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
706 }
707
708 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
709 .family = AF_INET6,
710 .obj_size = sizeof(struct tcp6_request_sock),
711 .rtx_syn_ack = tcp_rtx_synack,
712 .send_ack = tcp_v6_reqsk_send_ack,
713 .destructor = tcp_v6_reqsk_destructor,
714 .send_reset = tcp_v6_send_reset,
715 .syn_ack_timeout = tcp_syn_ack_timeout,
716 };
717
718 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
719 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
720 sizeof(struct ipv6hdr),
721 #ifdef CONFIG_TCP_MD5SIG
722 .req_md5_lookup = tcp_v6_md5_lookup,
723 .calc_md5_hash = tcp_v6_md5_hash_skb,
724 #endif
725 .init_req = tcp_v6_init_req,
726 #ifdef CONFIG_SYN_COOKIES
727 .cookie_init_seq = cookie_v6_init_sequence,
728 #endif
729 .route_req = tcp_v6_route_req,
730 .init_seq = tcp_v6_init_sequence,
731 .send_synack = tcp_v6_send_synack,
732 };
733
734 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
735 u32 ack, u32 win, u32 tsval, u32 tsecr,
736 int oif, struct tcp_md5sig_key *key, int rst,
737 u8 tclass, u32 label)
738 {
739 const struct tcphdr *th = tcp_hdr(skb);
740 struct tcphdr *t1;
741 struct sk_buff *buff;
742 struct flowi6 fl6;
743 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
744 struct sock *ctl_sk = net->ipv6.tcp_sk;
745 unsigned int tot_len = sizeof(struct tcphdr);
746 struct dst_entry *dst;
747 __be32 *topt;
748
749 if (tsecr)
750 tot_len += TCPOLEN_TSTAMP_ALIGNED;
751 #ifdef CONFIG_TCP_MD5SIG
752 if (key)
753 tot_len += TCPOLEN_MD5SIG_ALIGNED;
754 #endif
755
756 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
757 GFP_ATOMIC);
758 if (!buff)
759 return;
760
761 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
762
763 t1 = (struct tcphdr *) skb_push(buff, tot_len);
764 skb_reset_transport_header(buff);
765
766 /* Swap the send and the receive. */
767 memset(t1, 0, sizeof(*t1));
768 t1->dest = th->source;
769 t1->source = th->dest;
770 t1->doff = tot_len / 4;
771 t1->seq = htonl(seq);
772 t1->ack_seq = htonl(ack);
773 t1->ack = !rst || !th->ack;
774 t1->rst = rst;
775 t1->window = htons(win);
776
777 topt = (__be32 *)(t1 + 1);
778
779 if (tsecr) {
780 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
781 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
782 *topt++ = htonl(tsval);
783 *topt++ = htonl(tsecr);
784 }
785
786 #ifdef CONFIG_TCP_MD5SIG
787 if (key) {
788 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
789 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
790 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
791 &ipv6_hdr(skb)->saddr,
792 &ipv6_hdr(skb)->daddr, t1);
793 }
794 #endif
795
796 memset(&fl6, 0, sizeof(fl6));
797 fl6.daddr = ipv6_hdr(skb)->saddr;
798 fl6.saddr = ipv6_hdr(skb)->daddr;
799 fl6.flowlabel = label;
800
801 buff->ip_summed = CHECKSUM_PARTIAL;
802 buff->csum = 0;
803
804 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
805
806 fl6.flowi6_proto = IPPROTO_TCP;
807 if (rt6_need_strict(&fl6.daddr) && !oif)
808 fl6.flowi6_oif = tcp_v6_iif(skb);
809 else
810 fl6.flowi6_oif = oif;
811 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
812 fl6.fl6_dport = t1->dest;
813 fl6.fl6_sport = t1->source;
814 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
815
816 /* Pass a socket to ip6_dst_lookup either it is for RST
817 * Underlying function will use this to retrieve the network
818 * namespace
819 */
820 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
821 if (!IS_ERR(dst)) {
822 skb_dst_set(buff, dst);
823 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
824 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
825 if (rst)
826 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
827 return;
828 }
829
830 kfree_skb(buff);
831 }
832
833 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
834 {
835 const struct tcphdr *th = tcp_hdr(skb);
836 u32 seq = 0, ack_seq = 0;
837 struct tcp_md5sig_key *key = NULL;
838 #ifdef CONFIG_TCP_MD5SIG
839 const __u8 *hash_location = NULL;
840 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
841 unsigned char newhash[16];
842 int genhash;
843 struct sock *sk1 = NULL;
844 #endif
845 int oif;
846
847 if (th->rst)
848 return;
849
850 /* If sk not NULL, it means we did a successful lookup and incoming
851 * route had to be correct. prequeue might have dropped our dst.
852 */
853 if (!sk && !ipv6_unicast_destination(skb))
854 return;
855
856 #ifdef CONFIG_TCP_MD5SIG
857 hash_location = tcp_parse_md5sig_option(th);
858 if (!sk && hash_location) {
859 /*
860 * active side is lost. Try to find listening socket through
861 * source port, and then find md5 key through listening socket.
862 * we are not loose security here:
863 * Incoming packet is checked with md5 hash with finding key,
864 * no RST generated if md5 hash doesn't match.
865 */
866 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
867 &tcp_hashinfo, &ipv6h->saddr,
868 th->source, &ipv6h->daddr,
869 ntohs(th->source), tcp_v6_iif(skb));
870 if (!sk1)
871 return;
872
873 rcu_read_lock();
874 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
875 if (!key)
876 goto release_sk1;
877
878 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
879 if (genhash || memcmp(hash_location, newhash, 16) != 0)
880 goto release_sk1;
881 } else {
882 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
883 }
884 #endif
885
886 if (th->ack)
887 seq = ntohl(th->ack_seq);
888 else
889 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
890 (th->doff << 2);
891
892 oif = sk ? sk->sk_bound_dev_if : 0;
893 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
894
895 #ifdef CONFIG_TCP_MD5SIG
896 release_sk1:
897 if (sk1) {
898 rcu_read_unlock();
899 sock_put(sk1);
900 }
901 #endif
902 }
903
904 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
905 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
906 struct tcp_md5sig_key *key, u8 tclass,
907 u32 label)
908 {
909 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
910 tclass, label);
911 }
912
913 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
914 {
915 struct inet_timewait_sock *tw = inet_twsk(sk);
916 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
917
918 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
919 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
920 tcp_time_stamp + tcptw->tw_ts_offset,
921 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
922 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
923
924 inet_twsk_put(tw);
925 }
926
927 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
928 struct request_sock *req)
929 {
930 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
931 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
932 */
933 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
934 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
935 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
936 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
937 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
938 0, 0);
939 }
940
941
942 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
943 {
944 #ifdef CONFIG_SYN_COOKIES
945 const struct tcphdr *th = tcp_hdr(skb);
946
947 if (!th->syn)
948 sk = cookie_v6_check(sk, skb);
949 #endif
950 return sk;
951 }
952
953 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
954 {
955 if (skb->protocol == htons(ETH_P_IP))
956 return tcp_v4_conn_request(sk, skb);
957
958 if (!ipv6_unicast_destination(skb))
959 goto drop;
960
961 return tcp_conn_request(&tcp6_request_sock_ops,
962 &tcp_request_sock_ipv6_ops, sk, skb);
963
964 drop:
965 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
966 return 0; /* don't send reset */
967 }
968
969 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
970 struct request_sock *req,
971 struct dst_entry *dst,
972 struct request_sock *req_unhash,
973 bool *own_req)
974 {
975 struct inet_request_sock *ireq;
976 struct ipv6_pinfo *newnp;
977 const struct ipv6_pinfo *np = inet6_sk(sk);
978 struct ipv6_txoptions *opt;
979 struct tcp6_sock *newtcp6sk;
980 struct inet_sock *newinet;
981 struct tcp_sock *newtp;
982 struct sock *newsk;
983 #ifdef CONFIG_TCP_MD5SIG
984 struct tcp_md5sig_key *key;
985 #endif
986 struct flowi6 fl6;
987
988 if (skb->protocol == htons(ETH_P_IP)) {
989 /*
990 * v6 mapped
991 */
992
993 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
994 req_unhash, own_req);
995
996 if (!newsk)
997 return NULL;
998
999 newtcp6sk = (struct tcp6_sock *)newsk;
1000 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1001
1002 newinet = inet_sk(newsk);
1003 newnp = inet6_sk(newsk);
1004 newtp = tcp_sk(newsk);
1005
1006 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1007
1008 newnp->saddr = newsk->sk_v6_rcv_saddr;
1009
1010 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1011 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1012 #ifdef CONFIG_TCP_MD5SIG
1013 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1014 #endif
1015
1016 newnp->ipv6_ac_list = NULL;
1017 newnp->ipv6_fl_list = NULL;
1018 newnp->pktoptions = NULL;
1019 newnp->opt = NULL;
1020 newnp->mcast_oif = tcp_v6_iif(skb);
1021 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1022 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1023 if (np->repflow)
1024 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1025
1026 /*
1027 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1028 * here, tcp_create_openreq_child now does this for us, see the comment in
1029 * that function for the gory details. -acme
1030 */
1031
1032 /* It is tricky place. Until this moment IPv4 tcp
1033 worked with IPv6 icsk.icsk_af_ops.
1034 Sync it now.
1035 */
1036 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1037
1038 return newsk;
1039 }
1040
1041 ireq = inet_rsk(req);
1042
1043 if (sk_acceptq_is_full(sk))
1044 goto out_overflow;
1045
1046 if (!dst) {
1047 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1048 if (!dst)
1049 goto out;
1050 }
1051
1052 newsk = tcp_create_openreq_child(sk, req, skb);
1053 if (!newsk)
1054 goto out_nonewsk;
1055
1056 /*
1057 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1058 * count here, tcp_create_openreq_child now does this for us, see the
1059 * comment in that function for the gory details. -acme
1060 */
1061
1062 newsk->sk_gso_type = SKB_GSO_TCPV6;
1063 ip6_dst_store(newsk, dst, NULL, NULL);
1064 inet6_sk_rx_dst_set(newsk, skb);
1065
1066 newtcp6sk = (struct tcp6_sock *)newsk;
1067 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1068
1069 newtp = tcp_sk(newsk);
1070 newinet = inet_sk(newsk);
1071 newnp = inet6_sk(newsk);
1072
1073 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1074
1075 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1076 newnp->saddr = ireq->ir_v6_loc_addr;
1077 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1078 newsk->sk_bound_dev_if = ireq->ir_iif;
1079
1080 /* Now IPv6 options...
1081
1082 First: no IPv4 options.
1083 */
1084 newinet->inet_opt = NULL;
1085 newnp->ipv6_ac_list = NULL;
1086 newnp->ipv6_fl_list = NULL;
1087
1088 /* Clone RX bits */
1089 newnp->rxopt.all = np->rxopt.all;
1090
1091 newnp->pktoptions = NULL;
1092 newnp->opt = NULL;
1093 newnp->mcast_oif = tcp_v6_iif(skb);
1094 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1095 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1096 if (np->repflow)
1097 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1098
1099 /* Clone native IPv6 options from listening socket (if any)
1100
1101 Yes, keeping reference count would be much more clever,
1102 but we make one more one thing there: reattach optmem
1103 to newsk.
1104 */
1105 opt = rcu_dereference(np->opt);
1106 if (opt) {
1107 opt = ipv6_dup_options(newsk, opt);
1108 RCU_INIT_POINTER(newnp->opt, opt);
1109 }
1110 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1111 if (opt)
1112 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1113 opt->opt_flen;
1114
1115 tcp_ca_openreq_child(newsk, dst);
1116
1117 tcp_sync_mss(newsk, dst_mtu(dst));
1118 newtp->advmss = dst_metric_advmss(dst);
1119 if (tcp_sk(sk)->rx_opt.user_mss &&
1120 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1121 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1122
1123 tcp_initialize_rcv_mss(newsk);
1124
1125 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1126 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1127
1128 #ifdef CONFIG_TCP_MD5SIG
1129 /* Copy over the MD5 key from the original socket */
1130 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1131 if (key) {
1132 /* We're using one, so create a matching key
1133 * on the newsk structure. If we fail to get
1134 * memory, then we end up not copying the key
1135 * across. Shucks.
1136 */
1137 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1138 AF_INET6, key->key, key->keylen,
1139 sk_gfp_atomic(sk, GFP_ATOMIC));
1140 }
1141 #endif
1142
1143 if (__inet_inherit_port(sk, newsk) < 0) {
1144 inet_csk_prepare_forced_close(newsk);
1145 tcp_done(newsk);
1146 goto out;
1147 }
1148 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1149 if (*own_req) {
1150 tcp_move_syn(newtp, req);
1151
1152 /* Clone pktoptions received with SYN, if we own the req */
1153 if (ireq->pktopts) {
1154 newnp->pktoptions = skb_clone(ireq->pktopts,
1155 sk_gfp_atomic(sk, GFP_ATOMIC));
1156 consume_skb(ireq->pktopts);
1157 ireq->pktopts = NULL;
1158 if (newnp->pktoptions)
1159 skb_set_owner_r(newnp->pktoptions, newsk);
1160 }
1161 }
1162
1163 return newsk;
1164
1165 out_overflow:
1166 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1167 out_nonewsk:
1168 dst_release(dst);
1169 out:
1170 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1171 return NULL;
1172 }
1173
1174 /* The socket must have it's spinlock held when we get
1175 * here, unless it is a TCP_LISTEN socket.
1176 *
1177 * We have a potential double-lock case here, so even when
1178 * doing backlog processing we use the BH locking scheme.
1179 * This is because we cannot sleep with the original spinlock
1180 * held.
1181 */
1182 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1183 {
1184 struct ipv6_pinfo *np = inet6_sk(sk);
1185 struct tcp_sock *tp;
1186 struct sk_buff *opt_skb = NULL;
1187
1188 /* Imagine: socket is IPv6. IPv4 packet arrives,
1189 goes to IPv4 receive handler and backlogged.
1190 From backlog it always goes here. Kerboom...
1191 Fortunately, tcp_rcv_established and rcv_established
1192 handle them correctly, but it is not case with
1193 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1194 */
1195
1196 if (skb->protocol == htons(ETH_P_IP))
1197 return tcp_v4_do_rcv(sk, skb);
1198
1199 if (sk_filter(sk, skb))
1200 goto discard;
1201
1202 /*
1203 * socket locking is here for SMP purposes as backlog rcv
1204 * is currently called with bh processing disabled.
1205 */
1206
1207 /* Do Stevens' IPV6_PKTOPTIONS.
1208
1209 Yes, guys, it is the only place in our code, where we
1210 may make it not affecting IPv4.
1211 The rest of code is protocol independent,
1212 and I do not like idea to uglify IPv4.
1213
1214 Actually, all the idea behind IPV6_PKTOPTIONS
1215 looks not very well thought. For now we latch
1216 options, received in the last packet, enqueued
1217 by tcp. Feel free to propose better solution.
1218 --ANK (980728)
1219 */
1220 if (np->rxopt.all)
1221 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1222
1223 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1224 struct dst_entry *dst = sk->sk_rx_dst;
1225
1226 sock_rps_save_rxhash(sk, skb);
1227 sk_mark_napi_id(sk, skb);
1228 if (dst) {
1229 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1230 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1231 dst_release(dst);
1232 sk->sk_rx_dst = NULL;
1233 }
1234 }
1235
1236 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1237 if (opt_skb)
1238 goto ipv6_pktoptions;
1239 return 0;
1240 }
1241
1242 if (tcp_checksum_complete(skb))
1243 goto csum_err;
1244
1245 if (sk->sk_state == TCP_LISTEN) {
1246 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1247
1248 if (!nsk)
1249 goto discard;
1250
1251 if (nsk != sk) {
1252 sock_rps_save_rxhash(nsk, skb);
1253 sk_mark_napi_id(nsk, skb);
1254 if (tcp_child_process(sk, nsk, skb))
1255 goto reset;
1256 if (opt_skb)
1257 __kfree_skb(opt_skb);
1258 return 0;
1259 }
1260 } else
1261 sock_rps_save_rxhash(sk, skb);
1262
1263 if (tcp_rcv_state_process(sk, skb))
1264 goto reset;
1265 if (opt_skb)
1266 goto ipv6_pktoptions;
1267 return 0;
1268
1269 reset:
1270 tcp_v6_send_reset(sk, skb);
1271 discard:
1272 if (opt_skb)
1273 __kfree_skb(opt_skb);
1274 kfree_skb(skb);
1275 return 0;
1276 csum_err:
1277 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1278 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1279 goto discard;
1280
1281
1282 ipv6_pktoptions:
1283 /* Do you ask, what is it?
1284
1285 1. skb was enqueued by tcp.
1286 2. skb is added to tail of read queue, rather than out of order.
1287 3. socket is not in passive state.
1288 4. Finally, it really contains options, which user wants to receive.
1289 */
1290 tp = tcp_sk(sk);
1291 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1292 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1293 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1294 np->mcast_oif = tcp_v6_iif(opt_skb);
1295 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1296 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1297 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1298 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1299 if (np->repflow)
1300 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1301 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1302 skb_set_owner_r(opt_skb, sk);
1303 opt_skb = xchg(&np->pktoptions, opt_skb);
1304 } else {
1305 __kfree_skb(opt_skb);
1306 opt_skb = xchg(&np->pktoptions, NULL);
1307 }
1308 }
1309
1310 kfree_skb(opt_skb);
1311 return 0;
1312 }
1313
1314 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1315 const struct tcphdr *th)
1316 {
1317 /* This is tricky: we move IP6CB at its correct location into
1318 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1319 * _decode_session6() uses IP6CB().
1320 * barrier() makes sure compiler won't play aliasing games.
1321 */
1322 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1323 sizeof(struct inet6_skb_parm));
1324 barrier();
1325
1326 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1327 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1328 skb->len - th->doff*4);
1329 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1330 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1331 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1332 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1333 TCP_SKB_CB(skb)->sacked = 0;
1334 }
1335
1336 static void tcp_v6_restore_cb(struct sk_buff *skb)
1337 {
1338 /* We need to move header back to the beginning if xfrm6_policy_check()
1339 * and tcp_v6_fill_cb() are going to be called again.
1340 */
1341 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1342 sizeof(struct inet6_skb_parm));
1343 }
1344
1345 static int tcp_v6_rcv(struct sk_buff *skb)
1346 {
1347 const struct tcphdr *th;
1348 const struct ipv6hdr *hdr;
1349 struct sock *sk;
1350 int ret;
1351 struct net *net = dev_net(skb->dev);
1352
1353 if (skb->pkt_type != PACKET_HOST)
1354 goto discard_it;
1355
1356 /*
1357 * Count it even if it's bad.
1358 */
1359 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1360
1361 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1362 goto discard_it;
1363
1364 th = tcp_hdr(skb);
1365
1366 if (th->doff < sizeof(struct tcphdr)/4)
1367 goto bad_packet;
1368 if (!pskb_may_pull(skb, th->doff*4))
1369 goto discard_it;
1370
1371 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1372 goto csum_error;
1373
1374 th = tcp_hdr(skb);
1375 hdr = ipv6_hdr(skb);
1376
1377 lookup:
1378 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest,
1379 inet6_iif(skb));
1380 if (!sk)
1381 goto no_tcp_socket;
1382
1383 process:
1384 if (sk->sk_state == TCP_TIME_WAIT)
1385 goto do_time_wait;
1386
1387 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1388 struct request_sock *req = inet_reqsk(sk);
1389 struct sock *nsk = NULL;
1390
1391 sk = req->rsk_listener;
1392 tcp_v6_fill_cb(skb, hdr, th);
1393 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1394 reqsk_put(req);
1395 goto discard_it;
1396 }
1397 if (likely(sk->sk_state == TCP_LISTEN)) {
1398 nsk = tcp_check_req(sk, skb, req, false);
1399 } else {
1400 inet_csk_reqsk_queue_drop_and_put(sk, req);
1401 goto lookup;
1402 }
1403 if (!nsk) {
1404 reqsk_put(req);
1405 goto discard_it;
1406 }
1407 if (nsk == sk) {
1408 sock_hold(sk);
1409 reqsk_put(req);
1410 tcp_v6_restore_cb(skb);
1411 } else if (tcp_child_process(sk, nsk, skb)) {
1412 tcp_v6_send_reset(nsk, skb);
1413 goto discard_it;
1414 } else {
1415 return 0;
1416 }
1417 }
1418 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1419 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1420 goto discard_and_relse;
1421 }
1422
1423 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1424 goto discard_and_relse;
1425
1426 tcp_v6_fill_cb(skb, hdr, th);
1427
1428 if (tcp_v6_inbound_md5_hash(sk, skb))
1429 goto discard_and_relse;
1430
1431 if (sk_filter(sk, skb))
1432 goto discard_and_relse;
1433
1434 skb->dev = NULL;
1435
1436 if (sk->sk_state == TCP_LISTEN) {
1437 ret = tcp_v6_do_rcv(sk, skb);
1438 goto put_and_return;
1439 }
1440
1441 sk_incoming_cpu_update(sk);
1442
1443 bh_lock_sock_nested(sk);
1444 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1445 ret = 0;
1446 if (!sock_owned_by_user(sk)) {
1447 if (!tcp_prequeue(sk, skb))
1448 ret = tcp_v6_do_rcv(sk, skb);
1449 } else if (unlikely(sk_add_backlog(sk, skb,
1450 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1451 bh_unlock_sock(sk);
1452 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1453 goto discard_and_relse;
1454 }
1455 bh_unlock_sock(sk);
1456
1457 put_and_return:
1458 sock_put(sk);
1459 return ret ? -1 : 0;
1460
1461 no_tcp_socket:
1462 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1463 goto discard_it;
1464
1465 tcp_v6_fill_cb(skb, hdr, th);
1466
1467 if (tcp_checksum_complete(skb)) {
1468 csum_error:
1469 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1470 bad_packet:
1471 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1472 } else {
1473 tcp_v6_send_reset(NULL, skb);
1474 }
1475
1476 discard_it:
1477 kfree_skb(skb);
1478 return 0;
1479
1480 discard_and_relse:
1481 sock_put(sk);
1482 goto discard_it;
1483
1484 do_time_wait:
1485 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1486 inet_twsk_put(inet_twsk(sk));
1487 goto discard_it;
1488 }
1489
1490 tcp_v6_fill_cb(skb, hdr, th);
1491
1492 if (tcp_checksum_complete(skb)) {
1493 inet_twsk_put(inet_twsk(sk));
1494 goto csum_error;
1495 }
1496
1497 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1498 case TCP_TW_SYN:
1499 {
1500 struct sock *sk2;
1501
1502 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1503 &ipv6_hdr(skb)->saddr, th->source,
1504 &ipv6_hdr(skb)->daddr,
1505 ntohs(th->dest), tcp_v6_iif(skb));
1506 if (sk2) {
1507 struct inet_timewait_sock *tw = inet_twsk(sk);
1508 inet_twsk_deschedule_put(tw);
1509 sk = sk2;
1510 tcp_v6_restore_cb(skb);
1511 goto process;
1512 }
1513 /* Fall through to ACK */
1514 }
1515 case TCP_TW_ACK:
1516 tcp_v6_timewait_ack(sk, skb);
1517 break;
1518 case TCP_TW_RST:
1519 tcp_v6_restore_cb(skb);
1520 goto no_tcp_socket;
1521 case TCP_TW_SUCCESS:
1522 ;
1523 }
1524 goto discard_it;
1525 }
1526
1527 static void tcp_v6_early_demux(struct sk_buff *skb)
1528 {
1529 const struct ipv6hdr *hdr;
1530 const struct tcphdr *th;
1531 struct sock *sk;
1532
1533 if (skb->pkt_type != PACKET_HOST)
1534 return;
1535
1536 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1537 return;
1538
1539 hdr = ipv6_hdr(skb);
1540 th = tcp_hdr(skb);
1541
1542 if (th->doff < sizeof(struct tcphdr) / 4)
1543 return;
1544
1545 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1546 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1547 &hdr->saddr, th->source,
1548 &hdr->daddr, ntohs(th->dest),
1549 inet6_iif(skb));
1550 if (sk) {
1551 skb->sk = sk;
1552 skb->destructor = sock_edemux;
1553 if (sk_fullsock(sk)) {
1554 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1555
1556 if (dst)
1557 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1558 if (dst &&
1559 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1560 skb_dst_set_noref(skb, dst);
1561 }
1562 }
1563 }
1564
1565 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1566 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1567 .twsk_unique = tcp_twsk_unique,
1568 .twsk_destructor = tcp_twsk_destructor,
1569 };
1570
1571 static const struct inet_connection_sock_af_ops ipv6_specific = {
1572 .queue_xmit = inet6_csk_xmit,
1573 .send_check = tcp_v6_send_check,
1574 .rebuild_header = inet6_sk_rebuild_header,
1575 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1576 .conn_request = tcp_v6_conn_request,
1577 .syn_recv_sock = tcp_v6_syn_recv_sock,
1578 .net_header_len = sizeof(struct ipv6hdr),
1579 .net_frag_header_len = sizeof(struct frag_hdr),
1580 .setsockopt = ipv6_setsockopt,
1581 .getsockopt = ipv6_getsockopt,
1582 .addr2sockaddr = inet6_csk_addr2sockaddr,
1583 .sockaddr_len = sizeof(struct sockaddr_in6),
1584 .bind_conflict = inet6_csk_bind_conflict,
1585 #ifdef CONFIG_COMPAT
1586 .compat_setsockopt = compat_ipv6_setsockopt,
1587 .compat_getsockopt = compat_ipv6_getsockopt,
1588 #endif
1589 .mtu_reduced = tcp_v6_mtu_reduced,
1590 };
1591
1592 #ifdef CONFIG_TCP_MD5SIG
1593 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1594 .md5_lookup = tcp_v6_md5_lookup,
1595 .calc_md5_hash = tcp_v6_md5_hash_skb,
1596 .md5_parse = tcp_v6_parse_md5_keys,
1597 };
1598 #endif
1599
1600 /*
1601 * TCP over IPv4 via INET6 API
1602 */
1603 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1604 .queue_xmit = ip_queue_xmit,
1605 .send_check = tcp_v4_send_check,
1606 .rebuild_header = inet_sk_rebuild_header,
1607 .sk_rx_dst_set = inet_sk_rx_dst_set,
1608 .conn_request = tcp_v6_conn_request,
1609 .syn_recv_sock = tcp_v6_syn_recv_sock,
1610 .net_header_len = sizeof(struct iphdr),
1611 .setsockopt = ipv6_setsockopt,
1612 .getsockopt = ipv6_getsockopt,
1613 .addr2sockaddr = inet6_csk_addr2sockaddr,
1614 .sockaddr_len = sizeof(struct sockaddr_in6),
1615 .bind_conflict = inet6_csk_bind_conflict,
1616 #ifdef CONFIG_COMPAT
1617 .compat_setsockopt = compat_ipv6_setsockopt,
1618 .compat_getsockopt = compat_ipv6_getsockopt,
1619 #endif
1620 .mtu_reduced = tcp_v4_mtu_reduced,
1621 };
1622
1623 #ifdef CONFIG_TCP_MD5SIG
1624 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1625 .md5_lookup = tcp_v4_md5_lookup,
1626 .calc_md5_hash = tcp_v4_md5_hash_skb,
1627 .md5_parse = tcp_v6_parse_md5_keys,
1628 };
1629 #endif
1630
1631 /* NOTE: A lot of things set to zero explicitly by call to
1632 * sk_alloc() so need not be done here.
1633 */
1634 static int tcp_v6_init_sock(struct sock *sk)
1635 {
1636 struct inet_connection_sock *icsk = inet_csk(sk);
1637
1638 tcp_init_sock(sk);
1639
1640 icsk->icsk_af_ops = &ipv6_specific;
1641
1642 #ifdef CONFIG_TCP_MD5SIG
1643 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1644 #endif
1645
1646 return 0;
1647 }
1648
1649 static void tcp_v6_destroy_sock(struct sock *sk)
1650 {
1651 tcp_v4_destroy_sock(sk);
1652 inet6_destroy_sock(sk);
1653 }
1654
1655 #ifdef CONFIG_PROC_FS
1656 /* Proc filesystem TCPv6 sock list dumping. */
1657 static void get_openreq6(struct seq_file *seq,
1658 const struct request_sock *req, int i)
1659 {
1660 long ttd = req->rsk_timer.expires - jiffies;
1661 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1662 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1663
1664 if (ttd < 0)
1665 ttd = 0;
1666
1667 seq_printf(seq,
1668 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1669 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1670 i,
1671 src->s6_addr32[0], src->s6_addr32[1],
1672 src->s6_addr32[2], src->s6_addr32[3],
1673 inet_rsk(req)->ir_num,
1674 dest->s6_addr32[0], dest->s6_addr32[1],
1675 dest->s6_addr32[2], dest->s6_addr32[3],
1676 ntohs(inet_rsk(req)->ir_rmt_port),
1677 TCP_SYN_RECV,
1678 0, 0, /* could print option size, but that is af dependent. */
1679 1, /* timers active (only the expire timer) */
1680 jiffies_to_clock_t(ttd),
1681 req->num_timeout,
1682 from_kuid_munged(seq_user_ns(seq),
1683 sock_i_uid(req->rsk_listener)),
1684 0, /* non standard timer */
1685 0, /* open_requests have no inode */
1686 0, req);
1687 }
1688
1689 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1690 {
1691 const struct in6_addr *dest, *src;
1692 __u16 destp, srcp;
1693 int timer_active;
1694 unsigned long timer_expires;
1695 const struct inet_sock *inet = inet_sk(sp);
1696 const struct tcp_sock *tp = tcp_sk(sp);
1697 const struct inet_connection_sock *icsk = inet_csk(sp);
1698 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1699 int rx_queue;
1700 int state;
1701
1702 dest = &sp->sk_v6_daddr;
1703 src = &sp->sk_v6_rcv_saddr;
1704 destp = ntohs(inet->inet_dport);
1705 srcp = ntohs(inet->inet_sport);
1706
1707 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1708 timer_active = 1;
1709 timer_expires = icsk->icsk_timeout;
1710 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1711 timer_active = 4;
1712 timer_expires = icsk->icsk_timeout;
1713 } else if (timer_pending(&sp->sk_timer)) {
1714 timer_active = 2;
1715 timer_expires = sp->sk_timer.expires;
1716 } else {
1717 timer_active = 0;
1718 timer_expires = jiffies;
1719 }
1720
1721 state = sk_state_load(sp);
1722 if (state == TCP_LISTEN)
1723 rx_queue = sp->sk_ack_backlog;
1724 else
1725 /* Because we don't lock the socket,
1726 * we might find a transient negative value.
1727 */
1728 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1729
1730 seq_printf(seq,
1731 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1732 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1733 i,
1734 src->s6_addr32[0], src->s6_addr32[1],
1735 src->s6_addr32[2], src->s6_addr32[3], srcp,
1736 dest->s6_addr32[0], dest->s6_addr32[1],
1737 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1738 state,
1739 tp->write_seq - tp->snd_una,
1740 rx_queue,
1741 timer_active,
1742 jiffies_delta_to_clock_t(timer_expires - jiffies),
1743 icsk->icsk_retransmits,
1744 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1745 icsk->icsk_probes_out,
1746 sock_i_ino(sp),
1747 atomic_read(&sp->sk_refcnt), sp,
1748 jiffies_to_clock_t(icsk->icsk_rto),
1749 jiffies_to_clock_t(icsk->icsk_ack.ato),
1750 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1751 tp->snd_cwnd,
1752 state == TCP_LISTEN ?
1753 fastopenq->max_qlen :
1754 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1755 );
1756 }
1757
1758 static void get_timewait6_sock(struct seq_file *seq,
1759 struct inet_timewait_sock *tw, int i)
1760 {
1761 long delta = tw->tw_timer.expires - jiffies;
1762 const struct in6_addr *dest, *src;
1763 __u16 destp, srcp;
1764
1765 dest = &tw->tw_v6_daddr;
1766 src = &tw->tw_v6_rcv_saddr;
1767 destp = ntohs(tw->tw_dport);
1768 srcp = ntohs(tw->tw_sport);
1769
1770 seq_printf(seq,
1771 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1772 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1773 i,
1774 src->s6_addr32[0], src->s6_addr32[1],
1775 src->s6_addr32[2], src->s6_addr32[3], srcp,
1776 dest->s6_addr32[0], dest->s6_addr32[1],
1777 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1778 tw->tw_substate, 0, 0,
1779 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1780 atomic_read(&tw->tw_refcnt), tw);
1781 }
1782
1783 static int tcp6_seq_show(struct seq_file *seq, void *v)
1784 {
1785 struct tcp_iter_state *st;
1786 struct sock *sk = v;
1787
1788 if (v == SEQ_START_TOKEN) {
1789 seq_puts(seq,
1790 " sl "
1791 "local_address "
1792 "remote_address "
1793 "st tx_queue rx_queue tr tm->when retrnsmt"
1794 " uid timeout inode\n");
1795 goto out;
1796 }
1797 st = seq->private;
1798
1799 if (sk->sk_state == TCP_TIME_WAIT)
1800 get_timewait6_sock(seq, v, st->num);
1801 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1802 get_openreq6(seq, v, st->num);
1803 else
1804 get_tcp6_sock(seq, v, st->num);
1805 out:
1806 return 0;
1807 }
1808
1809 static const struct file_operations tcp6_afinfo_seq_fops = {
1810 .owner = THIS_MODULE,
1811 .open = tcp_seq_open,
1812 .read = seq_read,
1813 .llseek = seq_lseek,
1814 .release = seq_release_net
1815 };
1816
1817 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1818 .name = "tcp6",
1819 .family = AF_INET6,
1820 .seq_fops = &tcp6_afinfo_seq_fops,
1821 .seq_ops = {
1822 .show = tcp6_seq_show,
1823 },
1824 };
1825
1826 int __net_init tcp6_proc_init(struct net *net)
1827 {
1828 return tcp_proc_register(net, &tcp6_seq_afinfo);
1829 }
1830
1831 void tcp6_proc_exit(struct net *net)
1832 {
1833 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1834 }
1835 #endif
1836
1837 static void tcp_v6_clear_sk(struct sock *sk, int size)
1838 {
1839 struct inet_sock *inet = inet_sk(sk);
1840
1841 /* we do not want to clear pinet6 field, because of RCU lookups */
1842 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1843
1844 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1845 memset(&inet->pinet6 + 1, 0, size);
1846 }
1847
1848 struct proto tcpv6_prot = {
1849 .name = "TCPv6",
1850 .owner = THIS_MODULE,
1851 .close = tcp_close,
1852 .connect = tcp_v6_connect,
1853 .disconnect = tcp_disconnect,
1854 .accept = inet_csk_accept,
1855 .ioctl = tcp_ioctl,
1856 .init = tcp_v6_init_sock,
1857 .destroy = tcp_v6_destroy_sock,
1858 .shutdown = tcp_shutdown,
1859 .setsockopt = tcp_setsockopt,
1860 .getsockopt = tcp_getsockopt,
1861 .recvmsg = tcp_recvmsg,
1862 .sendmsg = tcp_sendmsg,
1863 .sendpage = tcp_sendpage,
1864 .backlog_rcv = tcp_v6_do_rcv,
1865 .release_cb = tcp_release_cb,
1866 .hash = inet_hash,
1867 .unhash = inet_unhash,
1868 .get_port = inet_csk_get_port,
1869 .enter_memory_pressure = tcp_enter_memory_pressure,
1870 .stream_memory_free = tcp_stream_memory_free,
1871 .sockets_allocated = &tcp_sockets_allocated,
1872 .memory_allocated = &tcp_memory_allocated,
1873 .memory_pressure = &tcp_memory_pressure,
1874 .orphan_count = &tcp_orphan_count,
1875 .sysctl_mem = sysctl_tcp_mem,
1876 .sysctl_wmem = sysctl_tcp_wmem,
1877 .sysctl_rmem = sysctl_tcp_rmem,
1878 .max_header = MAX_TCP_HEADER,
1879 .obj_size = sizeof(struct tcp6_sock),
1880 .slab_flags = SLAB_DESTROY_BY_RCU,
1881 .twsk_prot = &tcp6_timewait_sock_ops,
1882 .rsk_prot = &tcp6_request_sock_ops,
1883 .h.hashinfo = &tcp_hashinfo,
1884 .no_autobind = true,
1885 #ifdef CONFIG_COMPAT
1886 .compat_setsockopt = compat_tcp_setsockopt,
1887 .compat_getsockopt = compat_tcp_getsockopt,
1888 #endif
1889 #ifdef CONFIG_MEMCG_KMEM
1890 .proto_cgroup = tcp_proto_cgroup,
1891 #endif
1892 .clear_sk = tcp_v6_clear_sk,
1893 };
1894
1895 static const struct inet6_protocol tcpv6_protocol = {
1896 .early_demux = tcp_v6_early_demux,
1897 .handler = tcp_v6_rcv,
1898 .err_handler = tcp_v6_err,
1899 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1900 };
1901
1902 static struct inet_protosw tcpv6_protosw = {
1903 .type = SOCK_STREAM,
1904 .protocol = IPPROTO_TCP,
1905 .prot = &tcpv6_prot,
1906 .ops = &inet6_stream_ops,
1907 .flags = INET_PROTOSW_PERMANENT |
1908 INET_PROTOSW_ICSK,
1909 };
1910
1911 static int __net_init tcpv6_net_init(struct net *net)
1912 {
1913 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1914 SOCK_RAW, IPPROTO_TCP, net);
1915 }
1916
1917 static void __net_exit tcpv6_net_exit(struct net *net)
1918 {
1919 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1920 }
1921
1922 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1923 {
1924 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1925 }
1926
1927 static struct pernet_operations tcpv6_net_ops = {
1928 .init = tcpv6_net_init,
1929 .exit = tcpv6_net_exit,
1930 .exit_batch = tcpv6_net_exit_batch,
1931 };
1932
1933 int __init tcpv6_init(void)
1934 {
1935 int ret;
1936
1937 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1938 if (ret)
1939 goto out;
1940
1941 /* register inet6 protocol */
1942 ret = inet6_register_protosw(&tcpv6_protosw);
1943 if (ret)
1944 goto out_tcpv6_protocol;
1945
1946 ret = register_pernet_subsys(&tcpv6_net_ops);
1947 if (ret)
1948 goto out_tcpv6_protosw;
1949 out:
1950 return ret;
1951
1952 out_tcpv6_protosw:
1953 inet6_unregister_protosw(&tcpv6_protosw);
1954 out_tcpv6_protocol:
1955 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1956 goto out;
1957 }
1958
1959 void tcpv6_exit(void)
1960 {
1961 unregister_pernet_subsys(&tcpv6_net_ops);
1962 inet6_unregister_protosw(&tcpv6_protosw);
1963 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1964 }
Michael's Linux tree.