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#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, u32 *tsoff)
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, tsoff);
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 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
153 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
154 &usin->sin6_addr);
155 else
156 usin->sin6_addr = in6addr_loopback;
157 }
158
159 addr_type = ipv6_addr_type(&usin->sin6_addr);
160
161 if (addr_type & IPV6_ADDR_MULTICAST)
162 return -ENETUNREACH;
163
164 if (addr_type&IPV6_ADDR_LINKLOCAL) {
165 if (addr_len >= sizeof(struct sockaddr_in6) &&
166 usin->sin6_scope_id) {
167 /* If interface is set while binding, indices
168 * must coincide.
169 */
170 if (sk->sk_bound_dev_if &&
171 sk->sk_bound_dev_if != usin->sin6_scope_id)
172 return -EINVAL;
173
174 sk->sk_bound_dev_if = usin->sin6_scope_id;
175 }
176
177 /* Connect to link-local address requires an interface */
178 if (!sk->sk_bound_dev_if)
179 return -EINVAL;
180 }
181
182 if (tp->rx_opt.ts_recent_stamp &&
183 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
184 tp->rx_opt.ts_recent = 0;
185 tp->rx_opt.ts_recent_stamp = 0;
186 tp->write_seq = 0;
187 }
188
189 sk->sk_v6_daddr = usin->sin6_addr;
190 np->flow_label = fl6.flowlabel;
191
192 /*
193 * TCP over IPv4
194 */
195
196 if (addr_type & IPV6_ADDR_MAPPED) {
197 u32 exthdrlen = icsk->icsk_ext_hdr_len;
198 struct sockaddr_in sin;
199
200 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
201
202 if (__ipv6_only_sock(sk))
203 return -ENETUNREACH;
204
205 sin.sin_family = AF_INET;
206 sin.sin_port = usin->sin6_port;
207 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
208
209 icsk->icsk_af_ops = &ipv6_mapped;
210 sk->sk_backlog_rcv = tcp_v4_do_rcv;
211#ifdef CONFIG_TCP_MD5SIG
212 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
213#endif
214
215 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
216
217 if (err) {
218 icsk->icsk_ext_hdr_len = exthdrlen;
219 icsk->icsk_af_ops = &ipv6_specific;
220 sk->sk_backlog_rcv = tcp_v6_do_rcv;
221#ifdef CONFIG_TCP_MD5SIG
222 tp->af_specific = &tcp_sock_ipv6_specific;
223#endif
224 goto failure;
225 }
226 np->saddr = sk->sk_v6_rcv_saddr;
227
228 return err;
229 }
230
231 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
232 saddr = &sk->sk_v6_rcv_saddr;
233
234 fl6.flowi6_proto = IPPROTO_TCP;
235 fl6.daddr = sk->sk_v6_daddr;
236 fl6.saddr = saddr ? *saddr : np->saddr;
237 fl6.flowi6_oif = sk->sk_bound_dev_if;
238 fl6.flowi6_mark = sk->sk_mark;
239 fl6.fl6_dport = usin->sin6_port;
240 fl6.fl6_sport = inet->inet_sport;
241 fl6.flowi6_uid = sk->sk_uid;
242
243 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
244 final_p = fl6_update_dst(&fl6, opt, &final);
245
246 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
247
248 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
249 if (IS_ERR(dst)) {
250 err = PTR_ERR(dst);
251 goto failure;
252 }
253
254 if (!saddr) {
255 saddr = &fl6.saddr;
256 sk->sk_v6_rcv_saddr = *saddr;
257 }
258
259 /* set the source address */
260 np->saddr = *saddr;
261 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
262
263 sk->sk_gso_type = SKB_GSO_TCPV6;
264 ip6_dst_store(sk, dst, NULL, NULL);
265
266 if (tcp_death_row.sysctl_tw_recycle &&
267 !tp->rx_opt.ts_recent_stamp &&
268 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
269 tcp_fetch_timewait_stamp(sk, dst);
270
271 icsk->icsk_ext_hdr_len = 0;
272 if (opt)
273 icsk->icsk_ext_hdr_len = opt->opt_flen +
274 opt->opt_nflen;
275
276 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
277
278 inet->inet_dport = usin->sin6_port;
279
280 tcp_set_state(sk, TCP_SYN_SENT);
281 err = inet6_hash_connect(&tcp_death_row, sk);
282 if (err)
283 goto late_failure;
284
285 sk_set_txhash(sk);
286
287 if (!tp->write_seq && likely(!tp->repair))
288 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
289 sk->sk_v6_daddr.s6_addr32,
290 inet->inet_sport,
291 inet->inet_dport,
292 &tp->tsoffset);
293
294 err = tcp_connect(sk);
295 if (err)
296 goto late_failure;
297
298 return 0;
299
300late_failure:
301 tcp_set_state(sk, TCP_CLOSE);
302 __sk_dst_reset(sk);
303failure:
304 inet->inet_dport = 0;
305 sk->sk_route_caps = 0;
306 return err;
307}
308
309static void tcp_v6_mtu_reduced(struct sock *sk)
310{
311 struct dst_entry *dst;
312
313 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
314 return;
315
316 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
317 if (!dst)
318 return;
319
320 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
321 tcp_sync_mss(sk, dst_mtu(dst));
322 tcp_simple_retransmit(sk);
323 }
324}
325
326static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
327 u8 type, u8 code, int offset, __be32 info)
328{
329 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
330 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
331 struct net *net = dev_net(skb->dev);
332 struct request_sock *fastopen;
333 struct ipv6_pinfo *np;
334 struct tcp_sock *tp;
335 __u32 seq, snd_una;
336 struct sock *sk;
337 bool fatal;
338 int err;
339
340 sk = __inet6_lookup_established(net, &tcp_hashinfo,
341 &hdr->daddr, th->dest,
342 &hdr->saddr, ntohs(th->source),
343 skb->dev->ifindex);
344
345 if (!sk) {
346 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
347 ICMP6_MIB_INERRORS);
348 return;
349 }
350
351 if (sk->sk_state == TCP_TIME_WAIT) {
352 inet_twsk_put(inet_twsk(sk));
353 return;
354 }
355 seq = ntohl(th->seq);
356 fatal = icmpv6_err_convert(type, code, &err);
357 if (sk->sk_state == TCP_NEW_SYN_RECV)
358 return tcp_req_err(sk, seq, fatal);
359
360 bh_lock_sock(sk);
361 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
362 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
363
364 if (sk->sk_state == TCP_CLOSE)
365 goto out;
366
367 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
368 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
369 goto out;
370 }
371
372 tp = tcp_sk(sk);
373 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
374 fastopen = tp->fastopen_rsk;
375 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
376 if (sk->sk_state != TCP_LISTEN &&
377 !between(seq, snd_una, tp->snd_nxt)) {
378 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
379 goto out;
380 }
381
382 np = inet6_sk(sk);
383
384 if (type == NDISC_REDIRECT) {
385 if (!sock_owned_by_user(sk)) {
386 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
387
388 if (dst)
389 dst->ops->redirect(dst, sk, skb);
390 }
391 goto out;
392 }
393
394 if (type == ICMPV6_PKT_TOOBIG) {
395 /* We are not interested in TCP_LISTEN and open_requests
396 * (SYN-ACKs send out by Linux are always <576bytes so
397 * they should go through unfragmented).
398 */
399 if (sk->sk_state == TCP_LISTEN)
400 goto out;
401
402 if (!ip6_sk_accept_pmtu(sk))
403 goto out;
404
405 tp->mtu_info = ntohl(info);
406 if (!sock_owned_by_user(sk))
407 tcp_v6_mtu_reduced(sk);
408 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
409 &sk->sk_tsq_flags))
410 sock_hold(sk);
411 goto out;
412 }
413
414
415 /* Might be for an request_sock */
416 switch (sk->sk_state) {
417 case TCP_SYN_SENT:
418 case TCP_SYN_RECV:
419 /* Only in fast or simultaneous open. If a fast open socket is
420 * is already accepted it is treated as a connected one below.
421 */
422 if (fastopen && !fastopen->sk)
423 break;
424
425 if (!sock_owned_by_user(sk)) {
426 sk->sk_err = err;
427 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
428
429 tcp_done(sk);
430 } else
431 sk->sk_err_soft = err;
432 goto out;
433 }
434
435 if (!sock_owned_by_user(sk) && np->recverr) {
436 sk->sk_err = err;
437 sk->sk_error_report(sk);
438 } else
439 sk->sk_err_soft = err;
440
441out:
442 bh_unlock_sock(sk);
443 sock_put(sk);
444}
445
446
447static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
448 struct flowi *fl,
449 struct request_sock *req,
450 struct tcp_fastopen_cookie *foc,
451 enum tcp_synack_type synack_type)
452{
453 struct inet_request_sock *ireq = inet_rsk(req);
454 struct ipv6_pinfo *np = inet6_sk(sk);
455 struct ipv6_txoptions *opt;
456 struct flowi6 *fl6 = &fl->u.ip6;
457 struct sk_buff *skb;
458 int err = -ENOMEM;
459
460 /* First, grab a route. */
461 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
462 IPPROTO_TCP)) == NULL)
463 goto done;
464
465 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
466
467 if (skb) {
468 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
469 &ireq->ir_v6_rmt_addr);
470
471 fl6->daddr = ireq->ir_v6_rmt_addr;
472 if (np->repflow && ireq->pktopts)
473 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
474
475 rcu_read_lock();
476 opt = ireq->ipv6_opt;
477 if (!opt)
478 opt = rcu_dereference(np->opt);
479 err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass);
480 rcu_read_unlock();
481 err = net_xmit_eval(err);
482 }
483
484done:
485 return err;
486}
487
488
489static void tcp_v6_reqsk_destructor(struct request_sock *req)
490{
491 kfree(inet_rsk(req)->ipv6_opt);
492 kfree_skb(inet_rsk(req)->pktopts);
493}
494
495#ifdef CONFIG_TCP_MD5SIG
496static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
497 const struct in6_addr *addr)
498{
499 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
500}
501
502static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
503 const struct sock *addr_sk)
504{
505 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
506}
507
508static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
509 int optlen)
510{
511 struct tcp_md5sig cmd;
512 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
513
514 if (optlen < sizeof(cmd))
515 return -EINVAL;
516
517 if (copy_from_user(&cmd, optval, sizeof(cmd)))
518 return -EFAULT;
519
520 if (sin6->sin6_family != AF_INET6)
521 return -EINVAL;
522
523 if (!cmd.tcpm_keylen) {
524 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
525 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
526 AF_INET);
527 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
528 AF_INET6);
529 }
530
531 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
532 return -EINVAL;
533
534 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
535 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
536 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
537
538 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
539 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
540}
541
542static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
543 const struct in6_addr *daddr,
544 const struct in6_addr *saddr,
545 const struct tcphdr *th, int nbytes)
546{
547 struct tcp6_pseudohdr *bp;
548 struct scatterlist sg;
549 struct tcphdr *_th;
550
551 bp = hp->scratch;
552 /* 1. TCP pseudo-header (RFC2460) */
553 bp->saddr = *saddr;
554 bp->daddr = *daddr;
555 bp->protocol = cpu_to_be32(IPPROTO_TCP);
556 bp->len = cpu_to_be32(nbytes);
557
558 _th = (struct tcphdr *)(bp + 1);
559 memcpy(_th, th, sizeof(*th));
560 _th->check = 0;
561
562 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
563 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
564 sizeof(*bp) + sizeof(*th));
565 return crypto_ahash_update(hp->md5_req);
566}
567
568static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
569 const struct in6_addr *daddr, struct in6_addr *saddr,
570 const struct tcphdr *th)
571{
572 struct tcp_md5sig_pool *hp;
573 struct ahash_request *req;
574
575 hp = tcp_get_md5sig_pool();
576 if (!hp)
577 goto clear_hash_noput;
578 req = hp->md5_req;
579
580 if (crypto_ahash_init(req))
581 goto clear_hash;
582 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
583 goto clear_hash;
584 if (tcp_md5_hash_key(hp, key))
585 goto clear_hash;
586 ahash_request_set_crypt(req, NULL, md5_hash, 0);
587 if (crypto_ahash_final(req))
588 goto clear_hash;
589
590 tcp_put_md5sig_pool();
591 return 0;
592
593clear_hash:
594 tcp_put_md5sig_pool();
595clear_hash_noput:
596 memset(md5_hash, 0, 16);
597 return 1;
598}
599
600static int tcp_v6_md5_hash_skb(char *md5_hash,
601 const struct tcp_md5sig_key *key,
602 const struct sock *sk,
603 const struct sk_buff *skb)
604{
605 const struct in6_addr *saddr, *daddr;
606 struct tcp_md5sig_pool *hp;
607 struct ahash_request *req;
608 const struct tcphdr *th = tcp_hdr(skb);
609
610 if (sk) { /* valid for establish/request sockets */
611 saddr = &sk->sk_v6_rcv_saddr;
612 daddr = &sk->sk_v6_daddr;
613 } else {
614 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
615 saddr = &ip6h->saddr;
616 daddr = &ip6h->daddr;
617 }
618
619 hp = tcp_get_md5sig_pool();
620 if (!hp)
621 goto clear_hash_noput;
622 req = hp->md5_req;
623
624 if (crypto_ahash_init(req))
625 goto clear_hash;
626
627 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
628 goto clear_hash;
629 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
630 goto clear_hash;
631 if (tcp_md5_hash_key(hp, key))
632 goto clear_hash;
633 ahash_request_set_crypt(req, NULL, md5_hash, 0);
634 if (crypto_ahash_final(req))
635 goto clear_hash;
636
637 tcp_put_md5sig_pool();
638 return 0;
639
640clear_hash:
641 tcp_put_md5sig_pool();
642clear_hash_noput:
643 memset(md5_hash, 0, 16);
644 return 1;
645}
646
647#endif
648
649static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
650 const struct sk_buff *skb)
651{
652#ifdef CONFIG_TCP_MD5SIG
653 const __u8 *hash_location = NULL;
654 struct tcp_md5sig_key *hash_expected;
655 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
656 const struct tcphdr *th = tcp_hdr(skb);
657 int genhash;
658 u8 newhash[16];
659
660 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
661 hash_location = tcp_parse_md5sig_option(th);
662
663 /* We've parsed the options - do we have a hash? */
664 if (!hash_expected && !hash_location)
665 return false;
666
667 if (hash_expected && !hash_location) {
668 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
669 return true;
670 }
671
672 if (!hash_expected && hash_location) {
673 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
674 return true;
675 }
676
677 /* check the signature */
678 genhash = tcp_v6_md5_hash_skb(newhash,
679 hash_expected,
680 NULL, skb);
681
682 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
683 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
684 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
685 genhash ? "failed" : "mismatch",
686 &ip6h->saddr, ntohs(th->source),
687 &ip6h->daddr, ntohs(th->dest));
688 return true;
689 }
690#endif
691 return false;
692}
693
694static void tcp_v6_init_req(struct request_sock *req,
695 const struct sock *sk_listener,
696 struct sk_buff *skb)
697{
698 struct inet_request_sock *ireq = inet_rsk(req);
699 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
700
701 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
702 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
703
704 /* So that link locals have meaning */
705 if (!sk_listener->sk_bound_dev_if &&
706 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
707 ireq->ir_iif = tcp_v6_iif(skb);
708
709 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
710 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
711 np->rxopt.bits.rxinfo ||
712 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
713 np->rxopt.bits.rxohlim || np->repflow)) {
714 atomic_inc(&skb->users);
715 ireq->pktopts = skb;
716 }
717}
718
719static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
720 struct flowi *fl,
721 const struct request_sock *req,
722 bool *strict)
723{
724 if (strict)
725 *strict = true;
726 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
727}
728
729struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
730 .family = AF_INET6,
731 .obj_size = sizeof(struct tcp6_request_sock),
732 .rtx_syn_ack = tcp_rtx_synack,
733 .send_ack = tcp_v6_reqsk_send_ack,
734 .destructor = tcp_v6_reqsk_destructor,
735 .send_reset = tcp_v6_send_reset,
736 .syn_ack_timeout = tcp_syn_ack_timeout,
737};
738
739static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
740 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
741 sizeof(struct ipv6hdr),
742#ifdef CONFIG_TCP_MD5SIG
743 .req_md5_lookup = tcp_v6_md5_lookup,
744 .calc_md5_hash = tcp_v6_md5_hash_skb,
745#endif
746 .init_req = tcp_v6_init_req,
747#ifdef CONFIG_SYN_COOKIES
748 .cookie_init_seq = cookie_v6_init_sequence,
749#endif
750 .route_req = tcp_v6_route_req,
751 .init_seq = tcp_v6_init_sequence,
752 .send_synack = tcp_v6_send_synack,
753};
754
755static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
756 u32 ack, u32 win, u32 tsval, u32 tsecr,
757 int oif, struct tcp_md5sig_key *key, int rst,
758 u8 tclass, __be32 label)
759{
760 const struct tcphdr *th = tcp_hdr(skb);
761 struct tcphdr *t1;
762 struct sk_buff *buff;
763 struct flowi6 fl6;
764 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
765 struct sock *ctl_sk = net->ipv6.tcp_sk;
766 unsigned int tot_len = sizeof(struct tcphdr);
767 struct dst_entry *dst;
768 __be32 *topt;
769
770 if (tsecr)
771 tot_len += TCPOLEN_TSTAMP_ALIGNED;
772#ifdef CONFIG_TCP_MD5SIG
773 if (key)
774 tot_len += TCPOLEN_MD5SIG_ALIGNED;
775#endif
776
777 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
778 GFP_ATOMIC);
779 if (!buff)
780 return;
781
782 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
783
784 t1 = (struct tcphdr *) skb_push(buff, tot_len);
785 skb_reset_transport_header(buff);
786
787 /* Swap the send and the receive. */
788 memset(t1, 0, sizeof(*t1));
789 t1->dest = th->source;
790 t1->source = th->dest;
791 t1->doff = tot_len / 4;
792 t1->seq = htonl(seq);
793 t1->ack_seq = htonl(ack);
794 t1->ack = !rst || !th->ack;
795 t1->rst = rst;
796 t1->window = htons(win);
797
798 topt = (__be32 *)(t1 + 1);
799
800 if (tsecr) {
801 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
802 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
803 *topt++ = htonl(tsval);
804 *topt++ = htonl(tsecr);
805 }
806
807#ifdef CONFIG_TCP_MD5SIG
808 if (key) {
809 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
810 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
811 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
812 &ipv6_hdr(skb)->saddr,
813 &ipv6_hdr(skb)->daddr, t1);
814 }
815#endif
816
817 memset(&fl6, 0, sizeof(fl6));
818 fl6.daddr = ipv6_hdr(skb)->saddr;
819 fl6.saddr = ipv6_hdr(skb)->daddr;
820 fl6.flowlabel = label;
821
822 buff->ip_summed = CHECKSUM_PARTIAL;
823 buff->csum = 0;
824
825 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
826
827 fl6.flowi6_proto = IPPROTO_TCP;
828 if (rt6_need_strict(&fl6.daddr) && !oif)
829 fl6.flowi6_oif = tcp_v6_iif(skb);
830 else {
831 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
832 oif = skb->skb_iif;
833
834 fl6.flowi6_oif = oif;
835 }
836
837 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
838 fl6.fl6_dport = t1->dest;
839 fl6.fl6_sport = t1->source;
840 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
841 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
842
843 /* Pass a socket to ip6_dst_lookup either it is for RST
844 * Underlying function will use this to retrieve the network
845 * namespace
846 */
847 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
848 if (!IS_ERR(dst)) {
849 skb_dst_set(buff, dst);
850 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, tclass);
851 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
852 if (rst)
853 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
854 return;
855 }
856
857 kfree_skb(buff);
858}
859
860static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
861{
862 const struct tcphdr *th = tcp_hdr(skb);
863 u32 seq = 0, ack_seq = 0;
864 struct tcp_md5sig_key *key = NULL;
865#ifdef CONFIG_TCP_MD5SIG
866 const __u8 *hash_location = NULL;
867 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
868 unsigned char newhash[16];
869 int genhash;
870 struct sock *sk1 = NULL;
871#endif
872 int oif;
873
874 if (th->rst)
875 return;
876
877 /* If sk not NULL, it means we did a successful lookup and incoming
878 * route had to be correct. prequeue might have dropped our dst.
879 */
880 if (!sk && !ipv6_unicast_destination(skb))
881 return;
882
883#ifdef CONFIG_TCP_MD5SIG
884 rcu_read_lock();
885 hash_location = tcp_parse_md5sig_option(th);
886 if (sk && sk_fullsock(sk)) {
887 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
888 } else if (hash_location) {
889 /*
890 * active side is lost. Try to find listening socket through
891 * source port, and then find md5 key through listening socket.
892 * we are not loose security here:
893 * Incoming packet is checked with md5 hash with finding key,
894 * no RST generated if md5 hash doesn't match.
895 */
896 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
897 &tcp_hashinfo, NULL, 0,
898 &ipv6h->saddr,
899 th->source, &ipv6h->daddr,
900 ntohs(th->source), tcp_v6_iif(skb));
901 if (!sk1)
902 goto out;
903
904 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
905 if (!key)
906 goto out;
907
908 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
909 if (genhash || memcmp(hash_location, newhash, 16) != 0)
910 goto out;
911 }
912#endif
913
914 if (th->ack)
915 seq = ntohl(th->ack_seq);
916 else
917 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
918 (th->doff << 2);
919
920 oif = sk ? sk->sk_bound_dev_if : 0;
921 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
922
923#ifdef CONFIG_TCP_MD5SIG
924out:
925 rcu_read_unlock();
926#endif
927}
928
929static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
930 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
931 struct tcp_md5sig_key *key, u8 tclass,
932 __be32 label)
933{
934 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
935 tclass, label);
936}
937
938static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
939{
940 struct inet_timewait_sock *tw = inet_twsk(sk);
941 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
942
943 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
944 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
945 tcp_time_stamp + tcptw->tw_ts_offset,
946 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
947 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
948
949 inet_twsk_put(tw);
950}
951
952static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
953 struct request_sock *req)
954{
955 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
956 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
957 */
958 /* RFC 7323 2.3
959 * The window field (SEG.WND) of every outgoing segment, with the
960 * exception of <SYN> segments, MUST be right-shifted by
961 * Rcv.Wind.Shift bits:
962 */
963 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
964 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
965 tcp_rsk(req)->rcv_nxt,
966 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
967 tcp_time_stamp + tcp_rsk(req)->ts_off,
968 req->ts_recent, sk->sk_bound_dev_if,
969 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
970 0, 0);
971}
972
973
974static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
975{
976#ifdef CONFIG_SYN_COOKIES
977 const struct tcphdr *th = tcp_hdr(skb);
978
979 if (!th->syn)
980 sk = cookie_v6_check(sk, skb);
981#endif
982 return sk;
983}
984
985static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
986{
987 if (skb->protocol == htons(ETH_P_IP))
988 return tcp_v4_conn_request(sk, skb);
989
990 if (!ipv6_unicast_destination(skb))
991 goto drop;
992
993 return tcp_conn_request(&tcp6_request_sock_ops,
994 &tcp_request_sock_ipv6_ops, sk, skb);
995
996drop:
997 tcp_listendrop(sk);
998 return 0; /* don't send reset */
999}
1000
1001static void tcp_v6_restore_cb(struct sk_buff *skb)
1002{
1003 /* We need to move header back to the beginning if xfrm6_policy_check()
1004 * and tcp_v6_fill_cb() are going to be called again.
1005 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1006 */
1007 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1008 sizeof(struct inet6_skb_parm));
1009}
1010
1011static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1012 struct request_sock *req,
1013 struct dst_entry *dst,
1014 struct request_sock *req_unhash,
1015 bool *own_req)
1016{
1017 struct inet_request_sock *ireq;
1018 struct ipv6_pinfo *newnp;
1019 const struct ipv6_pinfo *np = inet6_sk(sk);
1020 struct ipv6_txoptions *opt;
1021 struct tcp6_sock *newtcp6sk;
1022 struct inet_sock *newinet;
1023 struct tcp_sock *newtp;
1024 struct sock *newsk;
1025#ifdef CONFIG_TCP_MD5SIG
1026 struct tcp_md5sig_key *key;
1027#endif
1028 struct flowi6 fl6;
1029
1030 if (skb->protocol == htons(ETH_P_IP)) {
1031 /*
1032 * v6 mapped
1033 */
1034
1035 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1036 req_unhash, own_req);
1037
1038 if (!newsk)
1039 return NULL;
1040
1041 newtcp6sk = (struct tcp6_sock *)newsk;
1042 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1043
1044 newinet = inet_sk(newsk);
1045 newnp = inet6_sk(newsk);
1046 newtp = tcp_sk(newsk);
1047
1048 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1049
1050 newnp->saddr = newsk->sk_v6_rcv_saddr;
1051
1052 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1053 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1054#ifdef CONFIG_TCP_MD5SIG
1055 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1056#endif
1057
1058 newnp->ipv6_ac_list = NULL;
1059 newnp->ipv6_fl_list = NULL;
1060 newnp->pktoptions = NULL;
1061 newnp->opt = NULL;
1062 newnp->mcast_oif = tcp_v6_iif(skb);
1063 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1064 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1065 if (np->repflow)
1066 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1067
1068 /*
1069 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1070 * here, tcp_create_openreq_child now does this for us, see the comment in
1071 * that function for the gory details. -acme
1072 */
1073
1074 /* It is tricky place. Until this moment IPv4 tcp
1075 worked with IPv6 icsk.icsk_af_ops.
1076 Sync it now.
1077 */
1078 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1079
1080 return newsk;
1081 }
1082
1083 ireq = inet_rsk(req);
1084
1085 if (sk_acceptq_is_full(sk))
1086 goto out_overflow;
1087
1088 if (!dst) {
1089 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1090 if (!dst)
1091 goto out;
1092 }
1093
1094 newsk = tcp_create_openreq_child(sk, req, skb);
1095 if (!newsk)
1096 goto out_nonewsk;
1097
1098 /*
1099 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1100 * count here, tcp_create_openreq_child now does this for us, see the
1101 * comment in that function for the gory details. -acme
1102 */
1103
1104 newsk->sk_gso_type = SKB_GSO_TCPV6;
1105 ip6_dst_store(newsk, dst, NULL, NULL);
1106 inet6_sk_rx_dst_set(newsk, skb);
1107
1108 newtcp6sk = (struct tcp6_sock *)newsk;
1109 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1110
1111 newtp = tcp_sk(newsk);
1112 newinet = inet_sk(newsk);
1113 newnp = inet6_sk(newsk);
1114
1115 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1116
1117 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1118 newnp->saddr = ireq->ir_v6_loc_addr;
1119 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1120 newsk->sk_bound_dev_if = ireq->ir_iif;
1121
1122 /* Now IPv6 options...
1123
1124 First: no IPv4 options.
1125 */
1126 newinet->inet_opt = NULL;
1127 newnp->ipv6_ac_list = NULL;
1128 newnp->ipv6_fl_list = NULL;
1129
1130 /* Clone RX bits */
1131 newnp->rxopt.all = np->rxopt.all;
1132
1133 newnp->pktoptions = NULL;
1134 newnp->opt = NULL;
1135 newnp->mcast_oif = tcp_v6_iif(skb);
1136 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1137 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1138 if (np->repflow)
1139 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1140
1141 /* Clone native IPv6 options from listening socket (if any)
1142
1143 Yes, keeping reference count would be much more clever,
1144 but we make one more one thing there: reattach optmem
1145 to newsk.
1146 */
1147 opt = ireq->ipv6_opt;
1148 if (!opt)
1149 opt = rcu_dereference(np->opt);
1150 if (opt) {
1151 opt = ipv6_dup_options(newsk, opt);
1152 RCU_INIT_POINTER(newnp->opt, opt);
1153 }
1154 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1155 if (opt)
1156 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1157 opt->opt_flen;
1158
1159 tcp_ca_openreq_child(newsk, dst);
1160
1161 tcp_sync_mss(newsk, dst_mtu(dst));
1162 newtp->advmss = dst_metric_advmss(dst);
1163 if (tcp_sk(sk)->rx_opt.user_mss &&
1164 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1165 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1166
1167 tcp_initialize_rcv_mss(newsk);
1168
1169 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1170 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1171
1172#ifdef CONFIG_TCP_MD5SIG
1173 /* Copy over the MD5 key from the original socket */
1174 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1175 if (key) {
1176 /* We're using one, so create a matching key
1177 * on the newsk structure. If we fail to get
1178 * memory, then we end up not copying the key
1179 * across. Shucks.
1180 */
1181 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1182 AF_INET6, key->key, key->keylen,
1183 sk_gfp_mask(sk, GFP_ATOMIC));
1184 }
1185#endif
1186
1187 if (__inet_inherit_port(sk, newsk) < 0) {
1188 inet_csk_prepare_forced_close(newsk);
1189 tcp_done(newsk);
1190 goto out;
1191 }
1192 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1193 if (*own_req) {
1194 tcp_move_syn(newtp, req);
1195
1196 /* Clone pktoptions received with SYN, if we own the req */
1197 if (ireq->pktopts) {
1198 newnp->pktoptions = skb_clone(ireq->pktopts,
1199 sk_gfp_mask(sk, GFP_ATOMIC));
1200 consume_skb(ireq->pktopts);
1201 ireq->pktopts = NULL;
1202 if (newnp->pktoptions) {
1203 tcp_v6_restore_cb(newnp->pktoptions);
1204 skb_set_owner_r(newnp->pktoptions, newsk);
1205 }
1206 }
1207 }
1208
1209 return newsk;
1210
1211out_overflow:
1212 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1213out_nonewsk:
1214 dst_release(dst);
1215out:
1216 tcp_listendrop(sk);
1217 return NULL;
1218}
1219
1220/* The socket must have it's spinlock held when we get
1221 * here, unless it is a TCP_LISTEN socket.
1222 *
1223 * We have a potential double-lock case here, so even when
1224 * doing backlog processing we use the BH locking scheme.
1225 * This is because we cannot sleep with the original spinlock
1226 * held.
1227 */
1228static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1229{
1230 struct ipv6_pinfo *np = inet6_sk(sk);
1231 struct tcp_sock *tp;
1232 struct sk_buff *opt_skb = NULL;
1233
1234 /* Imagine: socket is IPv6. IPv4 packet arrives,
1235 goes to IPv4 receive handler and backlogged.
1236 From backlog it always goes here. Kerboom...
1237 Fortunately, tcp_rcv_established and rcv_established
1238 handle them correctly, but it is not case with
1239 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1240 */
1241
1242 if (skb->protocol == htons(ETH_P_IP))
1243 return tcp_v4_do_rcv(sk, skb);
1244
1245 if (tcp_filter(sk, skb))
1246 goto discard;
1247
1248 /*
1249 * socket locking is here for SMP purposes as backlog rcv
1250 * is currently called with bh processing disabled.
1251 */
1252
1253 /* Do Stevens' IPV6_PKTOPTIONS.
1254
1255 Yes, guys, it is the only place in our code, where we
1256 may make it not affecting IPv4.
1257 The rest of code is protocol independent,
1258 and I do not like idea to uglify IPv4.
1259
1260 Actually, all the idea behind IPV6_PKTOPTIONS
1261 looks not very well thought. For now we latch
1262 options, received in the last packet, enqueued
1263 by tcp. Feel free to propose better solution.
1264 --ANK (980728)
1265 */
1266 if (np->rxopt.all)
1267 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1268
1269 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1270 struct dst_entry *dst = sk->sk_rx_dst;
1271
1272 sock_rps_save_rxhash(sk, skb);
1273 sk_mark_napi_id(sk, skb);
1274 if (dst) {
1275 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1276 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1277 dst_release(dst);
1278 sk->sk_rx_dst = NULL;
1279 }
1280 }
1281
1282 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1283 if (opt_skb)
1284 goto ipv6_pktoptions;
1285 return 0;
1286 }
1287
1288 if (tcp_checksum_complete(skb))
1289 goto csum_err;
1290
1291 if (sk->sk_state == TCP_LISTEN) {
1292 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1293
1294 if (!nsk)
1295 goto discard;
1296
1297 if (nsk != sk) {
1298 sock_rps_save_rxhash(nsk, skb);
1299 sk_mark_napi_id(nsk, skb);
1300 if (tcp_child_process(sk, nsk, skb))
1301 goto reset;
1302 if (opt_skb)
1303 __kfree_skb(opt_skb);
1304 return 0;
1305 }
1306 } else
1307 sock_rps_save_rxhash(sk, skb);
1308
1309 if (tcp_rcv_state_process(sk, skb))
1310 goto reset;
1311 if (opt_skb)
1312 goto ipv6_pktoptions;
1313 return 0;
1314
1315reset:
1316 tcp_v6_send_reset(sk, skb);
1317discard:
1318 if (opt_skb)
1319 __kfree_skb(opt_skb);
1320 kfree_skb(skb);
1321 return 0;
1322csum_err:
1323 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1324 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1325 goto discard;
1326
1327
1328ipv6_pktoptions:
1329 /* Do you ask, what is it?
1330
1331 1. skb was enqueued by tcp.
1332 2. skb is added to tail of read queue, rather than out of order.
1333 3. socket is not in passive state.
1334 4. Finally, it really contains options, which user wants to receive.
1335 */
1336 tp = tcp_sk(sk);
1337 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1338 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1339 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1340 np->mcast_oif = tcp_v6_iif(opt_skb);
1341 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1342 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1343 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1344 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1345 if (np->repflow)
1346 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1347 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1348 skb_set_owner_r(opt_skb, sk);
1349 tcp_v6_restore_cb(opt_skb);
1350 opt_skb = xchg(&np->pktoptions, opt_skb);
1351 } else {
1352 __kfree_skb(opt_skb);
1353 opt_skb = xchg(&np->pktoptions, NULL);
1354 }
1355 }
1356
1357 kfree_skb(opt_skb);
1358 return 0;
1359}
1360
1361static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1362 const struct tcphdr *th)
1363{
1364 /* This is tricky: we move IP6CB at its correct location into
1365 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1366 * _decode_session6() uses IP6CB().
1367 * barrier() makes sure compiler won't play aliasing games.
1368 */
1369 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1370 sizeof(struct inet6_skb_parm));
1371 barrier();
1372
1373 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1374 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1375 skb->len - th->doff*4);
1376 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1377 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1378 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1379 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1380 TCP_SKB_CB(skb)->sacked = 0;
1381}
1382
1383static int tcp_v6_rcv(struct sk_buff *skb)
1384{
1385 const struct tcphdr *th;
1386 const struct ipv6hdr *hdr;
1387 bool refcounted;
1388 struct sock *sk;
1389 int ret;
1390 struct net *net = dev_net(skb->dev);
1391
1392 if (skb->pkt_type != PACKET_HOST)
1393 goto discard_it;
1394
1395 /*
1396 * Count it even if it's bad.
1397 */
1398 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1399
1400 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1401 goto discard_it;
1402
1403 th = (const struct tcphdr *)skb->data;
1404
1405 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1406 goto bad_packet;
1407 if (!pskb_may_pull(skb, th->doff*4))
1408 goto discard_it;
1409
1410 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1411 goto csum_error;
1412
1413 th = (const struct tcphdr *)skb->data;
1414 hdr = ipv6_hdr(skb);
1415
1416lookup:
1417 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1418 th->source, th->dest, inet6_iif(skb),
1419 &refcounted);
1420 if (!sk)
1421 goto no_tcp_socket;
1422
1423process:
1424 if (sk->sk_state == TCP_TIME_WAIT)
1425 goto do_time_wait;
1426
1427 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1428 struct request_sock *req = inet_reqsk(sk);
1429 struct sock *nsk;
1430
1431 sk = req->rsk_listener;
1432 tcp_v6_fill_cb(skb, hdr, th);
1433 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1434 sk_drops_add(sk, skb);
1435 reqsk_put(req);
1436 goto discard_it;
1437 }
1438 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1439 inet_csk_reqsk_queue_drop_and_put(sk, req);
1440 goto lookup;
1441 }
1442 sock_hold(sk);
1443 refcounted = true;
1444 nsk = tcp_check_req(sk, skb, req, false);
1445 if (!nsk) {
1446 reqsk_put(req);
1447 goto discard_and_relse;
1448 }
1449 if (nsk == sk) {
1450 reqsk_put(req);
1451 tcp_v6_restore_cb(skb);
1452 } else if (tcp_child_process(sk, nsk, skb)) {
1453 tcp_v6_send_reset(nsk, skb);
1454 goto discard_and_relse;
1455 } else {
1456 sock_put(sk);
1457 return 0;
1458 }
1459 }
1460 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1461 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1462 goto discard_and_relse;
1463 }
1464
1465 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1466 goto discard_and_relse;
1467
1468 tcp_v6_fill_cb(skb, hdr, th);
1469
1470 if (tcp_v6_inbound_md5_hash(sk, skb))
1471 goto discard_and_relse;
1472
1473 if (tcp_filter(sk, skb))
1474 goto discard_and_relse;
1475 th = (const struct tcphdr *)skb->data;
1476 hdr = ipv6_hdr(skb);
1477
1478 skb->dev = NULL;
1479
1480 if (sk->sk_state == TCP_LISTEN) {
1481 ret = tcp_v6_do_rcv(sk, skb);
1482 goto put_and_return;
1483 }
1484
1485 sk_incoming_cpu_update(sk);
1486
1487 bh_lock_sock_nested(sk);
1488 tcp_segs_in(tcp_sk(sk), skb);
1489 ret = 0;
1490 if (!sock_owned_by_user(sk)) {
1491 if (!tcp_prequeue(sk, skb))
1492 ret = tcp_v6_do_rcv(sk, skb);
1493 } else if (tcp_add_backlog(sk, skb)) {
1494 goto discard_and_relse;
1495 }
1496 bh_unlock_sock(sk);
1497
1498put_and_return:
1499 if (refcounted)
1500 sock_put(sk);
1501 return ret ? -1 : 0;
1502
1503no_tcp_socket:
1504 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1505 goto discard_it;
1506
1507 tcp_v6_fill_cb(skb, hdr, th);
1508
1509 if (tcp_checksum_complete(skb)) {
1510csum_error:
1511 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1512bad_packet:
1513 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1514 } else {
1515 tcp_v6_send_reset(NULL, skb);
1516 }
1517
1518discard_it:
1519 kfree_skb(skb);
1520 return 0;
1521
1522discard_and_relse:
1523 sk_drops_add(sk, skb);
1524 if (refcounted)
1525 sock_put(sk);
1526 goto discard_it;
1527
1528do_time_wait:
1529 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1530 inet_twsk_put(inet_twsk(sk));
1531 goto discard_it;
1532 }
1533
1534 tcp_v6_fill_cb(skb, hdr, th);
1535
1536 if (tcp_checksum_complete(skb)) {
1537 inet_twsk_put(inet_twsk(sk));
1538 goto csum_error;
1539 }
1540
1541 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1542 case TCP_TW_SYN:
1543 {
1544 struct sock *sk2;
1545
1546 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1547 skb, __tcp_hdrlen(th),
1548 &ipv6_hdr(skb)->saddr, th->source,
1549 &ipv6_hdr(skb)->daddr,
1550 ntohs(th->dest), tcp_v6_iif(skb));
1551 if (sk2) {
1552 struct inet_timewait_sock *tw = inet_twsk(sk);
1553 inet_twsk_deschedule_put(tw);
1554 sk = sk2;
1555 tcp_v6_restore_cb(skb);
1556 refcounted = false;
1557 goto process;
1558 }
1559 /* Fall through to ACK */
1560 }
1561 case TCP_TW_ACK:
1562 tcp_v6_timewait_ack(sk, skb);
1563 break;
1564 case TCP_TW_RST:
1565 tcp_v6_restore_cb(skb);
1566 tcp_v6_send_reset(sk, skb);
1567 inet_twsk_deschedule_put(inet_twsk(sk));
1568 goto discard_it;
1569 case TCP_TW_SUCCESS:
1570 ;
1571 }
1572 goto discard_it;
1573}
1574
1575static void tcp_v6_early_demux(struct sk_buff *skb)
1576{
1577 const struct ipv6hdr *hdr;
1578 const struct tcphdr *th;
1579 struct sock *sk;
1580
1581 if (skb->pkt_type != PACKET_HOST)
1582 return;
1583
1584 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1585 return;
1586
1587 hdr = ipv6_hdr(skb);
1588 th = tcp_hdr(skb);
1589
1590 if (th->doff < sizeof(struct tcphdr) / 4)
1591 return;
1592
1593 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1594 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1595 &hdr->saddr, th->source,
1596 &hdr->daddr, ntohs(th->dest),
1597 inet6_iif(skb));
1598 if (sk) {
1599 skb->sk = sk;
1600 skb->destructor = sock_edemux;
1601 if (sk_fullsock(sk)) {
1602 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1603
1604 if (dst)
1605 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1606 if (dst &&
1607 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1608 skb_dst_set_noref(skb, dst);
1609 }
1610 }
1611}
1612
1613static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1614 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1615 .twsk_unique = tcp_twsk_unique,
1616 .twsk_destructor = tcp_twsk_destructor,
1617};
1618
1619static const struct inet_connection_sock_af_ops ipv6_specific = {
1620 .queue_xmit = inet6_csk_xmit,
1621 .send_check = tcp_v6_send_check,
1622 .rebuild_header = inet6_sk_rebuild_header,
1623 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1624 .conn_request = tcp_v6_conn_request,
1625 .syn_recv_sock = tcp_v6_syn_recv_sock,
1626 .net_header_len = sizeof(struct ipv6hdr),
1627 .net_frag_header_len = sizeof(struct frag_hdr),
1628 .setsockopt = ipv6_setsockopt,
1629 .getsockopt = ipv6_getsockopt,
1630 .addr2sockaddr = inet6_csk_addr2sockaddr,
1631 .sockaddr_len = sizeof(struct sockaddr_in6),
1632 .bind_conflict = inet6_csk_bind_conflict,
1633#ifdef CONFIG_COMPAT
1634 .compat_setsockopt = compat_ipv6_setsockopt,
1635 .compat_getsockopt = compat_ipv6_getsockopt,
1636#endif
1637 .mtu_reduced = tcp_v6_mtu_reduced,
1638};
1639
1640#ifdef CONFIG_TCP_MD5SIG
1641static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1642 .md5_lookup = tcp_v6_md5_lookup,
1643 .calc_md5_hash = tcp_v6_md5_hash_skb,
1644 .md5_parse = tcp_v6_parse_md5_keys,
1645};
1646#endif
1647
1648/*
1649 * TCP over IPv4 via INET6 API
1650 */
1651static const struct inet_connection_sock_af_ops ipv6_mapped = {
1652 .queue_xmit = ip_queue_xmit,
1653 .send_check = tcp_v4_send_check,
1654 .rebuild_header = inet_sk_rebuild_header,
1655 .sk_rx_dst_set = inet_sk_rx_dst_set,
1656 .conn_request = tcp_v6_conn_request,
1657 .syn_recv_sock = tcp_v6_syn_recv_sock,
1658 .net_header_len = sizeof(struct iphdr),
1659 .setsockopt = ipv6_setsockopt,
1660 .getsockopt = ipv6_getsockopt,
1661 .addr2sockaddr = inet6_csk_addr2sockaddr,
1662 .sockaddr_len = sizeof(struct sockaddr_in6),
1663 .bind_conflict = inet6_csk_bind_conflict,
1664#ifdef CONFIG_COMPAT
1665 .compat_setsockopt = compat_ipv6_setsockopt,
1666 .compat_getsockopt = compat_ipv6_getsockopt,
1667#endif
1668 .mtu_reduced = tcp_v4_mtu_reduced,
1669};
1670
1671#ifdef CONFIG_TCP_MD5SIG
1672static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1673 .md5_lookup = tcp_v4_md5_lookup,
1674 .calc_md5_hash = tcp_v4_md5_hash_skb,
1675 .md5_parse = tcp_v6_parse_md5_keys,
1676};
1677#endif
1678
1679/* NOTE: A lot of things set to zero explicitly by call to
1680 * sk_alloc() so need not be done here.
1681 */
1682static int tcp_v6_init_sock(struct sock *sk)
1683{
1684 struct inet_connection_sock *icsk = inet_csk(sk);
1685
1686 tcp_init_sock(sk);
1687
1688 icsk->icsk_af_ops = &ipv6_specific;
1689
1690#ifdef CONFIG_TCP_MD5SIG
1691 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1692#endif
1693
1694 return 0;
1695}
1696
1697static void tcp_v6_destroy_sock(struct sock *sk)
1698{
1699 tcp_v4_destroy_sock(sk);
1700 inet6_destroy_sock(sk);
1701}
1702
1703#ifdef CONFIG_PROC_FS
1704/* Proc filesystem TCPv6 sock list dumping. */
1705static void get_openreq6(struct seq_file *seq,
1706 const struct request_sock *req, int i)
1707{
1708 long ttd = req->rsk_timer.expires - jiffies;
1709 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1710 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1711
1712 if (ttd < 0)
1713 ttd = 0;
1714
1715 seq_printf(seq,
1716 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1717 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1718 i,
1719 src->s6_addr32[0], src->s6_addr32[1],
1720 src->s6_addr32[2], src->s6_addr32[3],
1721 inet_rsk(req)->ir_num,
1722 dest->s6_addr32[0], dest->s6_addr32[1],
1723 dest->s6_addr32[2], dest->s6_addr32[3],
1724 ntohs(inet_rsk(req)->ir_rmt_port),
1725 TCP_SYN_RECV,
1726 0, 0, /* could print option size, but that is af dependent. */
1727 1, /* timers active (only the expire timer) */
1728 jiffies_to_clock_t(ttd),
1729 req->num_timeout,
1730 from_kuid_munged(seq_user_ns(seq),
1731 sock_i_uid(req->rsk_listener)),
1732 0, /* non standard timer */
1733 0, /* open_requests have no inode */
1734 0, req);
1735}
1736
1737static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1738{
1739 const struct in6_addr *dest, *src;
1740 __u16 destp, srcp;
1741 int timer_active;
1742 unsigned long timer_expires;
1743 const struct inet_sock *inet = inet_sk(sp);
1744 const struct tcp_sock *tp = tcp_sk(sp);
1745 const struct inet_connection_sock *icsk = inet_csk(sp);
1746 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1747 int rx_queue;
1748 int state;
1749
1750 dest = &sp->sk_v6_daddr;
1751 src = &sp->sk_v6_rcv_saddr;
1752 destp = ntohs(inet->inet_dport);
1753 srcp = ntohs(inet->inet_sport);
1754
1755 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1756 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
1757 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1758 timer_active = 1;
1759 timer_expires = icsk->icsk_timeout;
1760 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1761 timer_active = 4;
1762 timer_expires = icsk->icsk_timeout;
1763 } else if (timer_pending(&sp->sk_timer)) {
1764 timer_active = 2;
1765 timer_expires = sp->sk_timer.expires;
1766 } else {
1767 timer_active = 0;
1768 timer_expires = jiffies;
1769 }
1770
1771 state = sk_state_load(sp);
1772 if (state == TCP_LISTEN)
1773 rx_queue = sp->sk_ack_backlog;
1774 else
1775 /* Because we don't lock the socket,
1776 * we might find a transient negative value.
1777 */
1778 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1779
1780 seq_printf(seq,
1781 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1782 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1783 i,
1784 src->s6_addr32[0], src->s6_addr32[1],
1785 src->s6_addr32[2], src->s6_addr32[3], srcp,
1786 dest->s6_addr32[0], dest->s6_addr32[1],
1787 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1788 state,
1789 tp->write_seq - tp->snd_una,
1790 rx_queue,
1791 timer_active,
1792 jiffies_delta_to_clock_t(timer_expires - jiffies),
1793 icsk->icsk_retransmits,
1794 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1795 icsk->icsk_probes_out,
1796 sock_i_ino(sp),
1797 atomic_read(&sp->sk_refcnt), sp,
1798 jiffies_to_clock_t(icsk->icsk_rto),
1799 jiffies_to_clock_t(icsk->icsk_ack.ato),
1800 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1801 tp->snd_cwnd,
1802 state == TCP_LISTEN ?
1803 fastopenq->max_qlen :
1804 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1805 );
1806}
1807
1808static void get_timewait6_sock(struct seq_file *seq,
1809 struct inet_timewait_sock *tw, int i)
1810{
1811 long delta = tw->tw_timer.expires - jiffies;
1812 const struct in6_addr *dest, *src;
1813 __u16 destp, srcp;
1814
1815 dest = &tw->tw_v6_daddr;
1816 src = &tw->tw_v6_rcv_saddr;
1817 destp = ntohs(tw->tw_dport);
1818 srcp = ntohs(tw->tw_sport);
1819
1820 seq_printf(seq,
1821 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1822 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1823 i,
1824 src->s6_addr32[0], src->s6_addr32[1],
1825 src->s6_addr32[2], src->s6_addr32[3], srcp,
1826 dest->s6_addr32[0], dest->s6_addr32[1],
1827 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1828 tw->tw_substate, 0, 0,
1829 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1830 atomic_read(&tw->tw_refcnt), tw);
1831}
1832
1833static int tcp6_seq_show(struct seq_file *seq, void *v)
1834{
1835 struct tcp_iter_state *st;
1836 struct sock *sk = v;
1837
1838 if (v == SEQ_START_TOKEN) {
1839 seq_puts(seq,
1840 " sl "
1841 "local_address "
1842 "remote_address "
1843 "st tx_queue rx_queue tr tm->when retrnsmt"
1844 " uid timeout inode\n");
1845 goto out;
1846 }
1847 st = seq->private;
1848
1849 if (sk->sk_state == TCP_TIME_WAIT)
1850 get_timewait6_sock(seq, v, st->num);
1851 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1852 get_openreq6(seq, v, st->num);
1853 else
1854 get_tcp6_sock(seq, v, st->num);
1855out:
1856 return 0;
1857}
1858
1859static const struct file_operations tcp6_afinfo_seq_fops = {
1860 .owner = THIS_MODULE,
1861 .open = tcp_seq_open,
1862 .read = seq_read,
1863 .llseek = seq_lseek,
1864 .release = seq_release_net
1865};
1866
1867static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1868 .name = "tcp6",
1869 .family = AF_INET6,
1870 .seq_fops = &tcp6_afinfo_seq_fops,
1871 .seq_ops = {
1872 .show = tcp6_seq_show,
1873 },
1874};
1875
1876int __net_init tcp6_proc_init(struct net *net)
1877{
1878 return tcp_proc_register(net, &tcp6_seq_afinfo);
1879}
1880
1881void tcp6_proc_exit(struct net *net)
1882{
1883 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1884}
1885#endif
1886
1887struct proto tcpv6_prot = {
1888 .name = "TCPv6",
1889 .owner = THIS_MODULE,
1890 .close = tcp_close,
1891 .connect = tcp_v6_connect,
1892 .disconnect = tcp_disconnect,
1893 .accept = inet_csk_accept,
1894 .ioctl = tcp_ioctl,
1895 .init = tcp_v6_init_sock,
1896 .destroy = tcp_v6_destroy_sock,
1897 .shutdown = tcp_shutdown,
1898 .setsockopt = tcp_setsockopt,
1899 .getsockopt = tcp_getsockopt,
1900 .recvmsg = tcp_recvmsg,
1901 .sendmsg = tcp_sendmsg,
1902 .sendpage = tcp_sendpage,
1903 .backlog_rcv = tcp_v6_do_rcv,
1904 .release_cb = tcp_release_cb,
1905 .hash = inet6_hash,
1906 .unhash = inet_unhash,
1907 .get_port = inet_csk_get_port,
1908 .enter_memory_pressure = tcp_enter_memory_pressure,
1909 .stream_memory_free = tcp_stream_memory_free,
1910 .sockets_allocated = &tcp_sockets_allocated,
1911 .memory_allocated = &tcp_memory_allocated,
1912 .memory_pressure = &tcp_memory_pressure,
1913 .orphan_count = &tcp_orphan_count,
1914 .sysctl_mem = sysctl_tcp_mem,
1915 .sysctl_wmem = sysctl_tcp_wmem,
1916 .sysctl_rmem = sysctl_tcp_rmem,
1917 .max_header = MAX_TCP_HEADER,
1918 .obj_size = sizeof(struct tcp6_sock),
1919 .slab_flags = SLAB_DESTROY_BY_RCU,
1920 .twsk_prot = &tcp6_timewait_sock_ops,
1921 .rsk_prot = &tcp6_request_sock_ops,
1922 .h.hashinfo = &tcp_hashinfo,
1923 .no_autobind = true,
1924#ifdef CONFIG_COMPAT
1925 .compat_setsockopt = compat_tcp_setsockopt,
1926 .compat_getsockopt = compat_tcp_getsockopt,
1927#endif
1928 .diag_destroy = tcp_abort,
1929};
1930
1931static const struct inet6_protocol tcpv6_protocol = {
1932 .early_demux = tcp_v6_early_demux,
1933 .handler = tcp_v6_rcv,
1934 .err_handler = tcp_v6_err,
1935 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1936};
1937
1938static struct inet_protosw tcpv6_protosw = {
1939 .type = SOCK_STREAM,
1940 .protocol = IPPROTO_TCP,
1941 .prot = &tcpv6_prot,
1942 .ops = &inet6_stream_ops,
1943 .flags = INET_PROTOSW_PERMANENT |
1944 INET_PROTOSW_ICSK,
1945};
1946
1947static int __net_init tcpv6_net_init(struct net *net)
1948{
1949 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1950 SOCK_RAW, IPPROTO_TCP, net);
1951}
1952
1953static void __net_exit tcpv6_net_exit(struct net *net)
1954{
1955 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1956}
1957
1958static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1959{
1960 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1961}
1962
1963static struct pernet_operations tcpv6_net_ops = {
1964 .init = tcpv6_net_init,
1965 .exit = tcpv6_net_exit,
1966 .exit_batch = tcpv6_net_exit_batch,
1967};
1968
1969int __init tcpv6_init(void)
1970{
1971 int ret;
1972
1973 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1974 if (ret)
1975 goto out;
1976
1977 /* register inet6 protocol */
1978 ret = inet6_register_protosw(&tcpv6_protosw);
1979 if (ret)
1980 goto out_tcpv6_protocol;
1981
1982 ret = register_pernet_subsys(&tcpv6_net_ops);
1983 if (ret)
1984 goto out_tcpv6_protosw;
1985out:
1986 return ret;
1987
1988out_tcpv6_protosw:
1989 inet6_unregister_protosw(&tcpv6_protosw);
1990out_tcpv6_protocol:
1991 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1992 goto out;
1993}
1994
1995void tcpv6_exit(void)
1996{
1997 unregister_pernet_subsys(&tcpv6_net_ops);
1998 inet6_unregister_protosw(&tcpv6_protosw);
1999 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2000}