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/netdma.h>
63#include <net/inet_common.h>
64#include <net/secure_seq.h>
65#include <net/tcp_memcontrol.h>
66#include <net/busy_poll.h>
67
68#include <linux/proc_fs.h>
69#include <linux/seq_file.h>
70
71#include <linux/crypto.h>
72#include <linux/scatterlist.h>
73
74static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
75static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
76 struct request_sock *req);
77
78static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
79
80static const struct inet_connection_sock_af_ops ipv6_mapped;
81static const struct inet_connection_sock_af_ops ipv6_specific;
82#ifdef CONFIG_TCP_MD5SIG
83static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
84static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
85#else
86static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
87 const struct in6_addr *addr)
88{
89 return NULL;
90}
91#endif
92
93static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
94{
95 struct dst_entry *dst = skb_dst(skb);
96 const struct rt6_info *rt = (const struct rt6_info *)dst;
97
98 dst_hold(dst);
99 sk->sk_rx_dst = dst;
100 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
101 if (rt->rt6i_node)
102 inet6_sk(sk)->rx_dst_cookie = rt->rt6i_node->fn_sernum;
103}
104
105static void tcp_v6_hash(struct sock *sk)
106{
107 if (sk->sk_state != TCP_CLOSE) {
108 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
109 tcp_prot.hash(sk);
110 return;
111 }
112 local_bh_disable();
113 __inet6_hash(sk, NULL);
114 local_bh_enable();
115 }
116}
117
118static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
119{
120 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
121 ipv6_hdr(skb)->saddr.s6_addr32,
122 tcp_hdr(skb)->dest,
123 tcp_hdr(skb)->source);
124}
125
126static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
127 int addr_len)
128{
129 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
130 struct inet_sock *inet = inet_sk(sk);
131 struct inet_connection_sock *icsk = inet_csk(sk);
132 struct ipv6_pinfo *np = inet6_sk(sk);
133 struct tcp_sock *tp = tcp_sk(sk);
134 struct in6_addr *saddr = NULL, *final_p, final;
135 struct rt6_info *rt;
136 struct flowi6 fl6;
137 struct dst_entry *dst;
138 int addr_type;
139 int err;
140
141 if (addr_len < SIN6_LEN_RFC2133)
142 return -EINVAL;
143
144 if (usin->sin6_family != AF_INET6)
145 return -EAFNOSUPPORT;
146
147 memset(&fl6, 0, sizeof(fl6));
148
149 if (np->sndflow) {
150 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
151 IP6_ECN_flow_init(fl6.flowlabel);
152 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
153 struct ip6_flowlabel *flowlabel;
154 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
155 if (flowlabel == NULL)
156 return -EINVAL;
157 fl6_sock_release(flowlabel);
158 }
159 }
160
161 /*
162 * connect() to INADDR_ANY means loopback (BSD'ism).
163 */
164
165 if (ipv6_addr_any(&usin->sin6_addr))
166 usin->sin6_addr.s6_addr[15] = 0x1;
167
168 addr_type = ipv6_addr_type(&usin->sin6_addr);
169
170 if (addr_type & IPV6_ADDR_MULTICAST)
171 return -ENETUNREACH;
172
173 if (addr_type&IPV6_ADDR_LINKLOCAL) {
174 if (addr_len >= sizeof(struct sockaddr_in6) &&
175 usin->sin6_scope_id) {
176 /* If interface is set while binding, indices
177 * must coincide.
178 */
179 if (sk->sk_bound_dev_if &&
180 sk->sk_bound_dev_if != usin->sin6_scope_id)
181 return -EINVAL;
182
183 sk->sk_bound_dev_if = usin->sin6_scope_id;
184 }
185
186 /* Connect to link-local address requires an interface */
187 if (!sk->sk_bound_dev_if)
188 return -EINVAL;
189 }
190
191 if (tp->rx_opt.ts_recent_stamp &&
192 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
193 tp->rx_opt.ts_recent = 0;
194 tp->rx_opt.ts_recent_stamp = 0;
195 tp->write_seq = 0;
196 }
197
198 sk->sk_v6_daddr = usin->sin6_addr;
199 np->flow_label = fl6.flowlabel;
200
201 /*
202 * TCP over IPv4
203 */
204
205 if (addr_type == IPV6_ADDR_MAPPED) {
206 u32 exthdrlen = icsk->icsk_ext_hdr_len;
207 struct sockaddr_in sin;
208
209 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
210
211 if (__ipv6_only_sock(sk))
212 return -ENETUNREACH;
213
214 sin.sin_family = AF_INET;
215 sin.sin_port = usin->sin6_port;
216 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
217
218 icsk->icsk_af_ops = &ipv6_mapped;
219 sk->sk_backlog_rcv = tcp_v4_do_rcv;
220#ifdef CONFIG_TCP_MD5SIG
221 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
222#endif
223
224 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
225
226 if (err) {
227 icsk->icsk_ext_hdr_len = exthdrlen;
228 icsk->icsk_af_ops = &ipv6_specific;
229 sk->sk_backlog_rcv = tcp_v6_do_rcv;
230#ifdef CONFIG_TCP_MD5SIG
231 tp->af_specific = &tcp_sock_ipv6_specific;
232#endif
233 goto failure;
234 } else {
235 ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
236 ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
237 &sk->sk_v6_rcv_saddr);
238 }
239
240 return err;
241 }
242
243 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
244 saddr = &sk->sk_v6_rcv_saddr;
245
246 fl6.flowi6_proto = IPPROTO_TCP;
247 fl6.daddr = sk->sk_v6_daddr;
248 fl6.saddr = saddr ? *saddr : np->saddr;
249 fl6.flowi6_oif = sk->sk_bound_dev_if;
250 fl6.flowi6_mark = sk->sk_mark;
251 fl6.fl6_dport = usin->sin6_port;
252 fl6.fl6_sport = inet->inet_sport;
253
254 final_p = fl6_update_dst(&fl6, np->opt, &final);
255
256 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
257
258 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
259 if (IS_ERR(dst)) {
260 err = PTR_ERR(dst);
261 goto failure;
262 }
263
264 if (saddr == NULL) {
265 saddr = &fl6.saddr;
266 sk->sk_v6_rcv_saddr = *saddr;
267 }
268
269 /* set the source address */
270 np->saddr = *saddr;
271 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
272
273 sk->sk_gso_type = SKB_GSO_TCPV6;
274 __ip6_dst_store(sk, dst, NULL, NULL);
275
276 rt = (struct rt6_info *) dst;
277 if (tcp_death_row.sysctl_tw_recycle &&
278 !tp->rx_opt.ts_recent_stamp &&
279 ipv6_addr_equal(&rt->rt6i_dst.addr, &sk->sk_v6_daddr))
280 tcp_fetch_timewait_stamp(sk, dst);
281
282 icsk->icsk_ext_hdr_len = 0;
283 if (np->opt)
284 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
285 np->opt->opt_nflen);
286
287 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
288
289 inet->inet_dport = usin->sin6_port;
290
291 tcp_set_state(sk, TCP_SYN_SENT);
292 err = inet6_hash_connect(&tcp_death_row, sk);
293 if (err)
294 goto late_failure;
295
296 if (!tp->write_seq && likely(!tp->repair))
297 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
298 sk->sk_v6_daddr.s6_addr32,
299 inet->inet_sport,
300 inet->inet_dport);
301
302 err = tcp_connect(sk);
303 if (err)
304 goto late_failure;
305
306 return 0;
307
308late_failure:
309 tcp_set_state(sk, TCP_CLOSE);
310 __sk_dst_reset(sk);
311failure:
312 inet->inet_dport = 0;
313 sk->sk_route_caps = 0;
314 return err;
315}
316
317static void tcp_v6_mtu_reduced(struct sock *sk)
318{
319 struct dst_entry *dst;
320
321 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
322 return;
323
324 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
325 if (!dst)
326 return;
327
328 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
329 tcp_sync_mss(sk, dst_mtu(dst));
330 tcp_simple_retransmit(sk);
331 }
332}
333
334static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
335 u8 type, u8 code, int offset, __be32 info)
336{
337 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
338 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
339 struct ipv6_pinfo *np;
340 struct sock *sk;
341 int err;
342 struct tcp_sock *tp;
343 __u32 seq;
344 struct net *net = dev_net(skb->dev);
345
346 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
347 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
348
349 if (sk == NULL) {
350 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
351 ICMP6_MIB_INERRORS);
352 return;
353 }
354
355 if (sk->sk_state == TCP_TIME_WAIT) {
356 inet_twsk_put(inet_twsk(sk));
357 return;
358 }
359
360 bh_lock_sock(sk);
361 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
362 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
363
364 if (sk->sk_state == TCP_CLOSE)
365 goto out;
366
367 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
368 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
369 goto out;
370 }
371
372 tp = tcp_sk(sk);
373 seq = ntohl(th->seq);
374 if (sk->sk_state != TCP_LISTEN &&
375 !between(seq, tp->snd_una, tp->snd_nxt)) {
376 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
377 goto out;
378 }
379
380 np = inet6_sk(sk);
381
382 if (type == NDISC_REDIRECT) {
383 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
384
385 if (dst)
386 dst->ops->redirect(dst, sk, skb);
387 goto out;
388 }
389
390 if (type == ICMPV6_PKT_TOOBIG) {
391 /* We are not interested in TCP_LISTEN and open_requests
392 * (SYN-ACKs send out by Linux are always <576bytes so
393 * they should go through unfragmented).
394 */
395 if (sk->sk_state == TCP_LISTEN)
396 goto out;
397
398 if (!ip6_sk_accept_pmtu(sk))
399 goto out;
400
401 tp->mtu_info = ntohl(info);
402 if (!sock_owned_by_user(sk))
403 tcp_v6_mtu_reduced(sk);
404 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
405 &tp->tsq_flags))
406 sock_hold(sk);
407 goto out;
408 }
409
410 icmpv6_err_convert(type, code, &err);
411
412 /* Might be for an request_sock */
413 switch (sk->sk_state) {
414 struct request_sock *req, **prev;
415 case TCP_LISTEN:
416 if (sock_owned_by_user(sk))
417 goto out;
418
419 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
420 &hdr->saddr, inet6_iif(skb));
421 if (!req)
422 goto out;
423
424 /* ICMPs are not backlogged, hence we cannot get
425 * an established socket here.
426 */
427 WARN_ON(req->sk != NULL);
428
429 if (seq != tcp_rsk(req)->snt_isn) {
430 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
431 goto out;
432 }
433
434 inet_csk_reqsk_queue_drop(sk, req, prev);
435 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
436 goto out;
437
438 case TCP_SYN_SENT:
439 case TCP_SYN_RECV: /* Cannot happen.
440 It can, it SYNs are crossed. --ANK */
441 if (!sock_owned_by_user(sk)) {
442 sk->sk_err = err;
443 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
444
445 tcp_done(sk);
446 } else
447 sk->sk_err_soft = err;
448 goto out;
449 }
450
451 if (!sock_owned_by_user(sk) && np->recverr) {
452 sk->sk_err = err;
453 sk->sk_error_report(sk);
454 } else
455 sk->sk_err_soft = err;
456
457out:
458 bh_unlock_sock(sk);
459 sock_put(sk);
460}
461
462
463static int tcp_v6_send_synack(struct sock *sk, struct dst_entry *dst,
464 struct flowi6 *fl6,
465 struct request_sock *req,
466 u16 queue_mapping)
467{
468 struct inet_request_sock *ireq = inet_rsk(req);
469 struct ipv6_pinfo *np = inet6_sk(sk);
470 struct sk_buff *skb;
471 int err = -ENOMEM;
472
473 /* First, grab a route. */
474 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
475 goto done;
476
477 skb = tcp_make_synack(sk, dst, req, NULL);
478
479 if (skb) {
480 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
481 &ireq->ir_v6_rmt_addr);
482
483 fl6->daddr = ireq->ir_v6_rmt_addr;
484 if (np->repflow && (ireq->pktopts != NULL))
485 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
486
487 skb_set_queue_mapping(skb, queue_mapping);
488 err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass);
489 err = net_xmit_eval(err);
490 }
491
492done:
493 return err;
494}
495
496static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req)
497{
498 struct flowi6 fl6;
499 int res;
500
501 res = tcp_v6_send_synack(sk, NULL, &fl6, req, 0);
502 if (!res) {
503 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
504 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
505 }
506 return res;
507}
508
509static void tcp_v6_reqsk_destructor(struct request_sock *req)
510{
511 kfree_skb(inet_rsk(req)->pktopts);
512}
513
514#ifdef CONFIG_TCP_MD5SIG
515static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
516 const struct in6_addr *addr)
517{
518 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
519}
520
521static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
522 struct sock *addr_sk)
523{
524 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
525}
526
527static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
528 struct request_sock *req)
529{
530 return tcp_v6_md5_do_lookup(sk, &inet_rsk(req)->ir_v6_rmt_addr);
531}
532
533static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
534 int optlen)
535{
536 struct tcp_md5sig cmd;
537 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
538
539 if (optlen < sizeof(cmd))
540 return -EINVAL;
541
542 if (copy_from_user(&cmd, optval, sizeof(cmd)))
543 return -EFAULT;
544
545 if (sin6->sin6_family != AF_INET6)
546 return -EINVAL;
547
548 if (!cmd.tcpm_keylen) {
549 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
550 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
551 AF_INET);
552 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
553 AF_INET6);
554 }
555
556 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
557 return -EINVAL;
558
559 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
560 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
561 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
562
563 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
564 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
565}
566
567static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
568 const struct in6_addr *daddr,
569 const struct in6_addr *saddr, int nbytes)
570{
571 struct tcp6_pseudohdr *bp;
572 struct scatterlist sg;
573
574 bp = &hp->md5_blk.ip6;
575 /* 1. TCP pseudo-header (RFC2460) */
576 bp->saddr = *saddr;
577 bp->daddr = *daddr;
578 bp->protocol = cpu_to_be32(IPPROTO_TCP);
579 bp->len = cpu_to_be32(nbytes);
580
581 sg_init_one(&sg, bp, sizeof(*bp));
582 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
583}
584
585static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
586 const struct in6_addr *daddr, struct in6_addr *saddr,
587 const struct tcphdr *th)
588{
589 struct tcp_md5sig_pool *hp;
590 struct hash_desc *desc;
591
592 hp = tcp_get_md5sig_pool();
593 if (!hp)
594 goto clear_hash_noput;
595 desc = &hp->md5_desc;
596
597 if (crypto_hash_init(desc))
598 goto clear_hash;
599 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
600 goto clear_hash;
601 if (tcp_md5_hash_header(hp, th))
602 goto clear_hash;
603 if (tcp_md5_hash_key(hp, key))
604 goto clear_hash;
605 if (crypto_hash_final(desc, md5_hash))
606 goto clear_hash;
607
608 tcp_put_md5sig_pool();
609 return 0;
610
611clear_hash:
612 tcp_put_md5sig_pool();
613clear_hash_noput:
614 memset(md5_hash, 0, 16);
615 return 1;
616}
617
618static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
619 const struct sock *sk,
620 const struct request_sock *req,
621 const struct sk_buff *skb)
622{
623 const struct in6_addr *saddr, *daddr;
624 struct tcp_md5sig_pool *hp;
625 struct hash_desc *desc;
626 const struct tcphdr *th = tcp_hdr(skb);
627
628 if (sk) {
629 saddr = &inet6_sk(sk)->saddr;
630 daddr = &sk->sk_v6_daddr;
631 } else if (req) {
632 saddr = &inet_rsk(req)->ir_v6_loc_addr;
633 daddr = &inet_rsk(req)->ir_v6_rmt_addr;
634 } else {
635 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
636 saddr = &ip6h->saddr;
637 daddr = &ip6h->daddr;
638 }
639
640 hp = tcp_get_md5sig_pool();
641 if (!hp)
642 goto clear_hash_noput;
643 desc = &hp->md5_desc;
644
645 if (crypto_hash_init(desc))
646 goto clear_hash;
647
648 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
649 goto clear_hash;
650 if (tcp_md5_hash_header(hp, th))
651 goto clear_hash;
652 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
653 goto clear_hash;
654 if (tcp_md5_hash_key(hp, key))
655 goto clear_hash;
656 if (crypto_hash_final(desc, md5_hash))
657 goto clear_hash;
658
659 tcp_put_md5sig_pool();
660 return 0;
661
662clear_hash:
663 tcp_put_md5sig_pool();
664clear_hash_noput:
665 memset(md5_hash, 0, 16);
666 return 1;
667}
668
669static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
670{
671 const __u8 *hash_location = NULL;
672 struct tcp_md5sig_key *hash_expected;
673 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
674 const struct tcphdr *th = tcp_hdr(skb);
675 int genhash;
676 u8 newhash[16];
677
678 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
679 hash_location = tcp_parse_md5sig_option(th);
680
681 /* We've parsed the options - do we have a hash? */
682 if (!hash_expected && !hash_location)
683 return 0;
684
685 if (hash_expected && !hash_location) {
686 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
687 return 1;
688 }
689
690 if (!hash_expected && hash_location) {
691 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
692 return 1;
693 }
694
695 /* check the signature */
696 genhash = tcp_v6_md5_hash_skb(newhash,
697 hash_expected,
698 NULL, NULL, skb);
699
700 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
701 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
702 genhash ? "failed" : "mismatch",
703 &ip6h->saddr, ntohs(th->source),
704 &ip6h->daddr, ntohs(th->dest));
705 return 1;
706 }
707 return 0;
708}
709#endif
710
711struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
712 .family = AF_INET6,
713 .obj_size = sizeof(struct tcp6_request_sock),
714 .rtx_syn_ack = tcp_v6_rtx_synack,
715 .send_ack = tcp_v6_reqsk_send_ack,
716 .destructor = tcp_v6_reqsk_destructor,
717 .send_reset = tcp_v6_send_reset,
718 .syn_ack_timeout = tcp_syn_ack_timeout,
719};
720
721#ifdef CONFIG_TCP_MD5SIG
722static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
723 .md5_lookup = tcp_v6_reqsk_md5_lookup,
724 .calc_md5_hash = tcp_v6_md5_hash_skb,
725};
726#endif
727
728static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
729 u32 tsval, u32 tsecr, int oif,
730 struct tcp_md5sig_key *key, int rst, u8 tclass,
731 u32 label)
732{
733 const struct tcphdr *th = tcp_hdr(skb);
734 struct tcphdr *t1;
735 struct sk_buff *buff;
736 struct flowi6 fl6;
737 struct net *net = dev_net(skb_dst(skb)->dev);
738 struct sock *ctl_sk = net->ipv6.tcp_sk;
739 unsigned int tot_len = sizeof(struct tcphdr);
740 struct dst_entry *dst;
741 __be32 *topt;
742
743 if (tsecr)
744 tot_len += TCPOLEN_TSTAMP_ALIGNED;
745#ifdef CONFIG_TCP_MD5SIG
746 if (key)
747 tot_len += TCPOLEN_MD5SIG_ALIGNED;
748#endif
749
750 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
751 GFP_ATOMIC);
752 if (buff == NULL)
753 return;
754
755 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
756
757 t1 = (struct tcphdr *) skb_push(buff, tot_len);
758 skb_reset_transport_header(buff);
759
760 /* Swap the send and the receive. */
761 memset(t1, 0, sizeof(*t1));
762 t1->dest = th->source;
763 t1->source = th->dest;
764 t1->doff = tot_len / 4;
765 t1->seq = htonl(seq);
766 t1->ack_seq = htonl(ack);
767 t1->ack = !rst || !th->ack;
768 t1->rst = rst;
769 t1->window = htons(win);
770
771 topt = (__be32 *)(t1 + 1);
772
773 if (tsecr) {
774 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
775 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
776 *topt++ = htonl(tsval);
777 *topt++ = htonl(tsecr);
778 }
779
780#ifdef CONFIG_TCP_MD5SIG
781 if (key) {
782 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
783 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
784 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
785 &ipv6_hdr(skb)->saddr,
786 &ipv6_hdr(skb)->daddr, t1);
787 }
788#endif
789
790 memset(&fl6, 0, sizeof(fl6));
791 fl6.daddr = ipv6_hdr(skb)->saddr;
792 fl6.saddr = ipv6_hdr(skb)->daddr;
793 fl6.flowlabel = label;
794
795 buff->ip_summed = CHECKSUM_PARTIAL;
796 buff->csum = 0;
797
798 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
799
800 fl6.flowi6_proto = IPPROTO_TCP;
801 if (rt6_need_strict(&fl6.daddr) && !oif)
802 fl6.flowi6_oif = inet6_iif(skb);
803 else
804 fl6.flowi6_oif = oif;
805 fl6.fl6_dport = t1->dest;
806 fl6.fl6_sport = t1->source;
807 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
808
809 /* Pass a socket to ip6_dst_lookup either it is for RST
810 * Underlying function will use this to retrieve the network
811 * namespace
812 */
813 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
814 if (!IS_ERR(dst)) {
815 skb_dst_set(buff, dst);
816 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
817 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
818 if (rst)
819 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
820 return;
821 }
822
823 kfree_skb(buff);
824}
825
826static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
827{
828 const struct tcphdr *th = tcp_hdr(skb);
829 u32 seq = 0, ack_seq = 0;
830 struct tcp_md5sig_key *key = NULL;
831#ifdef CONFIG_TCP_MD5SIG
832 const __u8 *hash_location = NULL;
833 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
834 unsigned char newhash[16];
835 int genhash;
836 struct sock *sk1 = NULL;
837#endif
838 int oif;
839
840 if (th->rst)
841 return;
842
843 if (!ipv6_unicast_destination(skb))
844 return;
845
846#ifdef CONFIG_TCP_MD5SIG
847 hash_location = tcp_parse_md5sig_option(th);
848 if (!sk && hash_location) {
849 /*
850 * active side is lost. Try to find listening socket through
851 * source port, and then find md5 key through listening socket.
852 * we are not loose security here:
853 * Incoming packet is checked with md5 hash with finding key,
854 * no RST generated if md5 hash doesn't match.
855 */
856 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
857 &tcp_hashinfo, &ipv6h->saddr,
858 th->source, &ipv6h->daddr,
859 ntohs(th->source), inet6_iif(skb));
860 if (!sk1)
861 return;
862
863 rcu_read_lock();
864 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
865 if (!key)
866 goto release_sk1;
867
868 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
869 if (genhash || memcmp(hash_location, newhash, 16) != 0)
870 goto release_sk1;
871 } else {
872 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
873 }
874#endif
875
876 if (th->ack)
877 seq = ntohl(th->ack_seq);
878 else
879 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
880 (th->doff << 2);
881
882 oif = sk ? sk->sk_bound_dev_if : 0;
883 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
884
885#ifdef CONFIG_TCP_MD5SIG
886release_sk1:
887 if (sk1) {
888 rcu_read_unlock();
889 sock_put(sk1);
890 }
891#endif
892}
893
894static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
895 u32 win, u32 tsval, u32 tsecr, int oif,
896 struct tcp_md5sig_key *key, u8 tclass,
897 u32 label)
898{
899 tcp_v6_send_response(skb, seq, ack, win, tsval, tsecr, oif, key, 0, tclass,
900 label);
901}
902
903static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
904{
905 struct inet_timewait_sock *tw = inet_twsk(sk);
906 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
907
908 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
909 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
910 tcp_time_stamp + tcptw->tw_ts_offset,
911 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
912 tw->tw_tclass, (tw->tw_flowlabel << 12));
913
914 inet_twsk_put(tw);
915}
916
917static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
918 struct request_sock *req)
919{
920 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1,
921 req->rcv_wnd, tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
922 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
923 0, 0);
924}
925
926
927static struct sock *tcp_v6_hnd_req(struct sock *sk, struct sk_buff *skb)
928{
929 struct request_sock *req, **prev;
930 const struct tcphdr *th = tcp_hdr(skb);
931 struct sock *nsk;
932
933 /* Find possible connection requests. */
934 req = inet6_csk_search_req(sk, &prev, th->source,
935 &ipv6_hdr(skb)->saddr,
936 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
937 if (req)
938 return tcp_check_req(sk, skb, req, prev, false);
939
940 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
941 &ipv6_hdr(skb)->saddr, th->source,
942 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
943
944 if (nsk) {
945 if (nsk->sk_state != TCP_TIME_WAIT) {
946 bh_lock_sock(nsk);
947 return nsk;
948 }
949 inet_twsk_put(inet_twsk(nsk));
950 return NULL;
951 }
952
953#ifdef CONFIG_SYN_COOKIES
954 if (!th->syn)
955 sk = cookie_v6_check(sk, skb);
956#endif
957 return sk;
958}
959
960/* FIXME: this is substantially similar to the ipv4 code.
961 * Can some kind of merge be done? -- erics
962 */
963static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
964{
965 struct tcp_options_received tmp_opt;
966 struct request_sock *req;
967 struct inet_request_sock *ireq;
968 struct ipv6_pinfo *np = inet6_sk(sk);
969 struct tcp_sock *tp = tcp_sk(sk);
970 __u32 isn = TCP_SKB_CB(skb)->when;
971 struct dst_entry *dst = NULL;
972 struct flowi6 fl6;
973 bool want_cookie = false;
974
975 if (skb->protocol == htons(ETH_P_IP))
976 return tcp_v4_conn_request(sk, skb);
977
978 if (!ipv6_unicast_destination(skb))
979 goto drop;
980
981 if ((sysctl_tcp_syncookies == 2 ||
982 inet_csk_reqsk_queue_is_full(sk)) && !isn) {
983 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
984 if (!want_cookie)
985 goto drop;
986 }
987
988 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
989 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
990 goto drop;
991 }
992
993 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
994 if (req == NULL)
995 goto drop;
996
997#ifdef CONFIG_TCP_MD5SIG
998 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
999#endif
1000
1001 tcp_clear_options(&tmp_opt);
1002 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1003 tmp_opt.user_mss = tp->rx_opt.user_mss;
1004 tcp_parse_options(skb, &tmp_opt, 0, NULL);
1005
1006 if (want_cookie && !tmp_opt.saw_tstamp)
1007 tcp_clear_options(&tmp_opt);
1008
1009 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1010 tcp_openreq_init(req, &tmp_opt, skb);
1011
1012 ireq = inet_rsk(req);
1013 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
1014 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
1015 if (!want_cookie || tmp_opt.tstamp_ok)
1016 TCP_ECN_create_request(req, skb, sock_net(sk));
1017
1018 ireq->ir_iif = sk->sk_bound_dev_if;
1019
1020 /* So that link locals have meaning */
1021 if (!sk->sk_bound_dev_if &&
1022 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
1023 ireq->ir_iif = inet6_iif(skb);
1024
1025 if (!isn) {
1026 if (ipv6_opt_accepted(sk, skb) ||
1027 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1028 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim ||
1029 np->repflow) {
1030 atomic_inc(&skb->users);
1031 ireq->pktopts = skb;
1032 }
1033
1034 if (want_cookie) {
1035 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1036 req->cookie_ts = tmp_opt.tstamp_ok;
1037 goto have_isn;
1038 }
1039
1040 /* VJ's idea. We save last timestamp seen
1041 * from the destination in peer table, when entering
1042 * state TIME-WAIT, and check against it before
1043 * accepting new connection request.
1044 *
1045 * If "isn" is not zero, this request hit alive
1046 * timewait bucket, so that all the necessary checks
1047 * are made in the function processing timewait state.
1048 */
1049 if (tmp_opt.saw_tstamp &&
1050 tcp_death_row.sysctl_tw_recycle &&
1051 (dst = inet6_csk_route_req(sk, &fl6, req)) != NULL) {
1052 if (!tcp_peer_is_proven(req, dst, true)) {
1053 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1054 goto drop_and_release;
1055 }
1056 }
1057 /* Kill the following clause, if you dislike this way. */
1058 else if (!sysctl_tcp_syncookies &&
1059 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1060 (sysctl_max_syn_backlog >> 2)) &&
1061 !tcp_peer_is_proven(req, dst, false)) {
1062 /* Without syncookies last quarter of
1063 * backlog is filled with destinations,
1064 * proven to be alive.
1065 * It means that we continue to communicate
1066 * to destinations, already remembered
1067 * to the moment of synflood.
1068 */
1069 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1070 &ireq->ir_v6_rmt_addr, ntohs(tcp_hdr(skb)->source));
1071 goto drop_and_release;
1072 }
1073
1074 isn = tcp_v6_init_sequence(skb);
1075 }
1076have_isn:
1077 tcp_rsk(req)->snt_isn = isn;
1078
1079 if (security_inet_conn_request(sk, skb, req))
1080 goto drop_and_release;
1081
1082 if (tcp_v6_send_synack(sk, dst, &fl6, req,
1083 skb_get_queue_mapping(skb)) ||
1084 want_cookie)
1085 goto drop_and_free;
1086
1087 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1088 tcp_rsk(req)->listener = NULL;
1089 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1090 return 0;
1091
1092drop_and_release:
1093 dst_release(dst);
1094drop_and_free:
1095 reqsk_free(req);
1096drop:
1097 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1098 return 0; /* don't send reset */
1099}
1100
1101static struct sock *tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1102 struct request_sock *req,
1103 struct dst_entry *dst)
1104{
1105 struct inet_request_sock *ireq;
1106 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1107 struct tcp6_sock *newtcp6sk;
1108 struct inet_sock *newinet;
1109 struct tcp_sock *newtp;
1110 struct sock *newsk;
1111#ifdef CONFIG_TCP_MD5SIG
1112 struct tcp_md5sig_key *key;
1113#endif
1114 struct flowi6 fl6;
1115
1116 if (skb->protocol == htons(ETH_P_IP)) {
1117 /*
1118 * v6 mapped
1119 */
1120
1121 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1122
1123 if (newsk == NULL)
1124 return NULL;
1125
1126 newtcp6sk = (struct tcp6_sock *)newsk;
1127 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1128
1129 newinet = inet_sk(newsk);
1130 newnp = inet6_sk(newsk);
1131 newtp = tcp_sk(newsk);
1132
1133 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1134
1135 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newsk->sk_v6_daddr);
1136
1137 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1138
1139 newsk->sk_v6_rcv_saddr = newnp->saddr;
1140
1141 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1142 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1143#ifdef CONFIG_TCP_MD5SIG
1144 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1145#endif
1146
1147 newnp->ipv6_ac_list = NULL;
1148 newnp->ipv6_fl_list = NULL;
1149 newnp->pktoptions = NULL;
1150 newnp->opt = NULL;
1151 newnp->mcast_oif = inet6_iif(skb);
1152 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1153 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1154 if (np->repflow)
1155 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1156
1157 /*
1158 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1159 * here, tcp_create_openreq_child now does this for us, see the comment in
1160 * that function for the gory details. -acme
1161 */
1162
1163 /* It is tricky place. Until this moment IPv4 tcp
1164 worked with IPv6 icsk.icsk_af_ops.
1165 Sync it now.
1166 */
1167 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1168
1169 return newsk;
1170 }
1171
1172 ireq = inet_rsk(req);
1173
1174 if (sk_acceptq_is_full(sk))
1175 goto out_overflow;
1176
1177 if (!dst) {
1178 dst = inet6_csk_route_req(sk, &fl6, req);
1179 if (!dst)
1180 goto out;
1181 }
1182
1183 newsk = tcp_create_openreq_child(sk, req, skb);
1184 if (newsk == NULL)
1185 goto out_nonewsk;
1186
1187 /*
1188 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1189 * count here, tcp_create_openreq_child now does this for us, see the
1190 * comment in that function for the gory details. -acme
1191 */
1192
1193 newsk->sk_gso_type = SKB_GSO_TCPV6;
1194 __ip6_dst_store(newsk, dst, NULL, NULL);
1195 inet6_sk_rx_dst_set(newsk, skb);
1196
1197 newtcp6sk = (struct tcp6_sock *)newsk;
1198 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1199
1200 newtp = tcp_sk(newsk);
1201 newinet = inet_sk(newsk);
1202 newnp = inet6_sk(newsk);
1203
1204 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1205
1206 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1207 newnp->saddr = ireq->ir_v6_loc_addr;
1208 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1209 newsk->sk_bound_dev_if = ireq->ir_iif;
1210
1211 /* Now IPv6 options...
1212
1213 First: no IPv4 options.
1214 */
1215 newinet->inet_opt = NULL;
1216 newnp->ipv6_ac_list = NULL;
1217 newnp->ipv6_fl_list = NULL;
1218
1219 /* Clone RX bits */
1220 newnp->rxopt.all = np->rxopt.all;
1221
1222 /* Clone pktoptions received with SYN */
1223 newnp->pktoptions = NULL;
1224 if (ireq->pktopts != NULL) {
1225 newnp->pktoptions = skb_clone(ireq->pktopts,
1226 sk_gfp_atomic(sk, GFP_ATOMIC));
1227 consume_skb(ireq->pktopts);
1228 ireq->pktopts = NULL;
1229 if (newnp->pktoptions)
1230 skb_set_owner_r(newnp->pktoptions, newsk);
1231 }
1232 newnp->opt = NULL;
1233 newnp->mcast_oif = inet6_iif(skb);
1234 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1235 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1236 if (np->repflow)
1237 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1238
1239 /* Clone native IPv6 options from listening socket (if any)
1240
1241 Yes, keeping reference count would be much more clever,
1242 but we make one more one thing there: reattach optmem
1243 to newsk.
1244 */
1245 if (np->opt)
1246 newnp->opt = ipv6_dup_options(newsk, np->opt);
1247
1248 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1249 if (newnp->opt)
1250 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1251 newnp->opt->opt_flen);
1252
1253 tcp_sync_mss(newsk, dst_mtu(dst));
1254 newtp->advmss = dst_metric_advmss(dst);
1255 if (tcp_sk(sk)->rx_opt.user_mss &&
1256 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1257 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1258
1259 tcp_initialize_rcv_mss(newsk);
1260
1261 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1262 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1263
1264#ifdef CONFIG_TCP_MD5SIG
1265 /* Copy over the MD5 key from the original socket */
1266 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1267 if (key != NULL) {
1268 /* We're using one, so create a matching key
1269 * on the newsk structure. If we fail to get
1270 * memory, then we end up not copying the key
1271 * across. Shucks.
1272 */
1273 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1274 AF_INET6, key->key, key->keylen,
1275 sk_gfp_atomic(sk, GFP_ATOMIC));
1276 }
1277#endif
1278
1279 if (__inet_inherit_port(sk, newsk) < 0) {
1280 inet_csk_prepare_forced_close(newsk);
1281 tcp_done(newsk);
1282 goto out;
1283 }
1284 __inet6_hash(newsk, NULL);
1285
1286 return newsk;
1287
1288out_overflow:
1289 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1290out_nonewsk:
1291 dst_release(dst);
1292out:
1293 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1294 return NULL;
1295}
1296
1297static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1298{
1299 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1300 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1301 &ipv6_hdr(skb)->daddr, skb->csum)) {
1302 skb->ip_summed = CHECKSUM_UNNECESSARY;
1303 return 0;
1304 }
1305 }
1306
1307 skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1308 &ipv6_hdr(skb)->saddr,
1309 &ipv6_hdr(skb)->daddr, 0));
1310
1311 if (skb->len <= 76)
1312 return __skb_checksum_complete(skb);
1313 return 0;
1314}
1315
1316/* The socket must have it's spinlock held when we get
1317 * here.
1318 *
1319 * We have a potential double-lock case here, so even when
1320 * doing backlog processing we use the BH locking scheme.
1321 * This is because we cannot sleep with the original spinlock
1322 * held.
1323 */
1324static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1325{
1326 struct ipv6_pinfo *np = inet6_sk(sk);
1327 struct tcp_sock *tp;
1328 struct sk_buff *opt_skb = NULL;
1329
1330 /* Imagine: socket is IPv6. IPv4 packet arrives,
1331 goes to IPv4 receive handler and backlogged.
1332 From backlog it always goes here. Kerboom...
1333 Fortunately, tcp_rcv_established and rcv_established
1334 handle them correctly, but it is not case with
1335 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1336 */
1337
1338 if (skb->protocol == htons(ETH_P_IP))
1339 return tcp_v4_do_rcv(sk, skb);
1340
1341#ifdef CONFIG_TCP_MD5SIG
1342 if (tcp_v6_inbound_md5_hash(sk, skb))
1343 goto discard;
1344#endif
1345
1346 if (sk_filter(sk, skb))
1347 goto discard;
1348
1349 /*
1350 * socket locking is here for SMP purposes as backlog rcv
1351 * is currently called with bh processing disabled.
1352 */
1353
1354 /* Do Stevens' IPV6_PKTOPTIONS.
1355
1356 Yes, guys, it is the only place in our code, where we
1357 may make it not affecting IPv4.
1358 The rest of code is protocol independent,
1359 and I do not like idea to uglify IPv4.
1360
1361 Actually, all the idea behind IPV6_PKTOPTIONS
1362 looks not very well thought. For now we latch
1363 options, received in the last packet, enqueued
1364 by tcp. Feel free to propose better solution.
1365 --ANK (980728)
1366 */
1367 if (np->rxopt.all)
1368 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1369
1370 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1371 struct dst_entry *dst = sk->sk_rx_dst;
1372
1373 sock_rps_save_rxhash(sk, skb);
1374 if (dst) {
1375 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1376 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1377 dst_release(dst);
1378 sk->sk_rx_dst = NULL;
1379 }
1380 }
1381
1382 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1383 if (opt_skb)
1384 goto ipv6_pktoptions;
1385 return 0;
1386 }
1387
1388 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1389 goto csum_err;
1390
1391 if (sk->sk_state == TCP_LISTEN) {
1392 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1393 if (!nsk)
1394 goto discard;
1395
1396 /*
1397 * Queue it on the new socket if the new socket is active,
1398 * otherwise we just shortcircuit this and continue with
1399 * the new socket..
1400 */
1401 if (nsk != sk) {
1402 sock_rps_save_rxhash(nsk, skb);
1403 if (tcp_child_process(sk, nsk, skb))
1404 goto reset;
1405 if (opt_skb)
1406 __kfree_skb(opt_skb);
1407 return 0;
1408 }
1409 } else
1410 sock_rps_save_rxhash(sk, skb);
1411
1412 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1413 goto reset;
1414 if (opt_skb)
1415 goto ipv6_pktoptions;
1416 return 0;
1417
1418reset:
1419 tcp_v6_send_reset(sk, skb);
1420discard:
1421 if (opt_skb)
1422 __kfree_skb(opt_skb);
1423 kfree_skb(skb);
1424 return 0;
1425csum_err:
1426 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1427 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1428 goto discard;
1429
1430
1431ipv6_pktoptions:
1432 /* Do you ask, what is it?
1433
1434 1. skb was enqueued by tcp.
1435 2. skb is added to tail of read queue, rather than out of order.
1436 3. socket is not in passive state.
1437 4. Finally, it really contains options, which user wants to receive.
1438 */
1439 tp = tcp_sk(sk);
1440 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1441 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1442 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1443 np->mcast_oif = inet6_iif(opt_skb);
1444 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1445 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1446 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1447 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1448 if (np->repflow)
1449 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1450 if (ipv6_opt_accepted(sk, opt_skb)) {
1451 skb_set_owner_r(opt_skb, sk);
1452 opt_skb = xchg(&np->pktoptions, opt_skb);
1453 } else {
1454 __kfree_skb(opt_skb);
1455 opt_skb = xchg(&np->pktoptions, NULL);
1456 }
1457 }
1458
1459 kfree_skb(opt_skb);
1460 return 0;
1461}
1462
1463static int tcp_v6_rcv(struct sk_buff *skb)
1464{
1465 const struct tcphdr *th;
1466 const struct ipv6hdr *hdr;
1467 struct sock *sk;
1468 int ret;
1469 struct net *net = dev_net(skb->dev);
1470
1471 if (skb->pkt_type != PACKET_HOST)
1472 goto discard_it;
1473
1474 /*
1475 * Count it even if it's bad.
1476 */
1477 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1478
1479 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1480 goto discard_it;
1481
1482 th = tcp_hdr(skb);
1483
1484 if (th->doff < sizeof(struct tcphdr)/4)
1485 goto bad_packet;
1486 if (!pskb_may_pull(skb, th->doff*4))
1487 goto discard_it;
1488
1489 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1490 goto csum_error;
1491
1492 th = tcp_hdr(skb);
1493 hdr = ipv6_hdr(skb);
1494 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1495 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1496 skb->len - th->doff*4);
1497 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1498 TCP_SKB_CB(skb)->when = 0;
1499 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1500 TCP_SKB_CB(skb)->sacked = 0;
1501
1502 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1503 if (!sk)
1504 goto no_tcp_socket;
1505
1506process:
1507 if (sk->sk_state == TCP_TIME_WAIT)
1508 goto do_time_wait;
1509
1510 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1511 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1512 goto discard_and_relse;
1513 }
1514
1515 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1516 goto discard_and_relse;
1517
1518 if (sk_filter(sk, skb))
1519 goto discard_and_relse;
1520
1521 sk_mark_napi_id(sk, skb);
1522 skb->dev = NULL;
1523
1524 bh_lock_sock_nested(sk);
1525 ret = 0;
1526 if (!sock_owned_by_user(sk)) {
1527#ifdef CONFIG_NET_DMA
1528 struct tcp_sock *tp = tcp_sk(sk);
1529 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1530 tp->ucopy.dma_chan = net_dma_find_channel();
1531 if (tp->ucopy.dma_chan)
1532 ret = tcp_v6_do_rcv(sk, skb);
1533 else
1534#endif
1535 {
1536 if (!tcp_prequeue(sk, skb))
1537 ret = tcp_v6_do_rcv(sk, skb);
1538 }
1539 } else if (unlikely(sk_add_backlog(sk, skb,
1540 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1541 bh_unlock_sock(sk);
1542 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1543 goto discard_and_relse;
1544 }
1545 bh_unlock_sock(sk);
1546
1547 sock_put(sk);
1548 return ret ? -1 : 0;
1549
1550no_tcp_socket:
1551 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1552 goto discard_it;
1553
1554 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1555csum_error:
1556 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1557bad_packet:
1558 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1559 } else {
1560 tcp_v6_send_reset(NULL, skb);
1561 }
1562
1563discard_it:
1564 kfree_skb(skb);
1565 return 0;
1566
1567discard_and_relse:
1568 sock_put(sk);
1569 goto discard_it;
1570
1571do_time_wait:
1572 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1573 inet_twsk_put(inet_twsk(sk));
1574 goto discard_it;
1575 }
1576
1577 if (skb->len < (th->doff<<2)) {
1578 inet_twsk_put(inet_twsk(sk));
1579 goto bad_packet;
1580 }
1581 if (tcp_checksum_complete(skb)) {
1582 inet_twsk_put(inet_twsk(sk));
1583 goto csum_error;
1584 }
1585
1586 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1587 case TCP_TW_SYN:
1588 {
1589 struct sock *sk2;
1590
1591 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1592 &ipv6_hdr(skb)->saddr, th->source,
1593 &ipv6_hdr(skb)->daddr,
1594 ntohs(th->dest), inet6_iif(skb));
1595 if (sk2 != NULL) {
1596 struct inet_timewait_sock *tw = inet_twsk(sk);
1597 inet_twsk_deschedule(tw, &tcp_death_row);
1598 inet_twsk_put(tw);
1599 sk = sk2;
1600 goto process;
1601 }
1602 /* Fall through to ACK */
1603 }
1604 case TCP_TW_ACK:
1605 tcp_v6_timewait_ack(sk, skb);
1606 break;
1607 case TCP_TW_RST:
1608 goto no_tcp_socket;
1609 case TCP_TW_SUCCESS:
1610 ;
1611 }
1612 goto discard_it;
1613}
1614
1615static void tcp_v6_early_demux(struct sk_buff *skb)
1616{
1617 const struct ipv6hdr *hdr;
1618 const struct tcphdr *th;
1619 struct sock *sk;
1620
1621 if (skb->pkt_type != PACKET_HOST)
1622 return;
1623
1624 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1625 return;
1626
1627 hdr = ipv6_hdr(skb);
1628 th = tcp_hdr(skb);
1629
1630 if (th->doff < sizeof(struct tcphdr) / 4)
1631 return;
1632
1633 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1634 &hdr->saddr, th->source,
1635 &hdr->daddr, ntohs(th->dest),
1636 inet6_iif(skb));
1637 if (sk) {
1638 skb->sk = sk;
1639 skb->destructor = sock_edemux;
1640 if (sk->sk_state != TCP_TIME_WAIT) {
1641 struct dst_entry *dst = sk->sk_rx_dst;
1642
1643 if (dst)
1644 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1645 if (dst &&
1646 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1647 skb_dst_set_noref(skb, dst);
1648 }
1649 }
1650}
1651
1652static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1653 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1654 .twsk_unique = tcp_twsk_unique,
1655 .twsk_destructor = tcp_twsk_destructor,
1656};
1657
1658static const struct inet_connection_sock_af_ops ipv6_specific = {
1659 .queue_xmit = inet6_csk_xmit,
1660 .send_check = tcp_v6_send_check,
1661 .rebuild_header = inet6_sk_rebuild_header,
1662 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1663 .conn_request = tcp_v6_conn_request,
1664 .syn_recv_sock = tcp_v6_syn_recv_sock,
1665 .net_header_len = sizeof(struct ipv6hdr),
1666 .net_frag_header_len = sizeof(struct frag_hdr),
1667 .setsockopt = ipv6_setsockopt,
1668 .getsockopt = ipv6_getsockopt,
1669 .addr2sockaddr = inet6_csk_addr2sockaddr,
1670 .sockaddr_len = sizeof(struct sockaddr_in6),
1671 .bind_conflict = inet6_csk_bind_conflict,
1672#ifdef CONFIG_COMPAT
1673 .compat_setsockopt = compat_ipv6_setsockopt,
1674 .compat_getsockopt = compat_ipv6_getsockopt,
1675#endif
1676};
1677
1678#ifdef CONFIG_TCP_MD5SIG
1679static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1680 .md5_lookup = tcp_v6_md5_lookup,
1681 .calc_md5_hash = tcp_v6_md5_hash_skb,
1682 .md5_parse = tcp_v6_parse_md5_keys,
1683};
1684#endif
1685
1686/*
1687 * TCP over IPv4 via INET6 API
1688 */
1689static const struct inet_connection_sock_af_ops ipv6_mapped = {
1690 .queue_xmit = ip_queue_xmit,
1691 .send_check = tcp_v4_send_check,
1692 .rebuild_header = inet_sk_rebuild_header,
1693 .sk_rx_dst_set = inet_sk_rx_dst_set,
1694 .conn_request = tcp_v6_conn_request,
1695 .syn_recv_sock = tcp_v6_syn_recv_sock,
1696 .net_header_len = sizeof(struct iphdr),
1697 .setsockopt = ipv6_setsockopt,
1698 .getsockopt = ipv6_getsockopt,
1699 .addr2sockaddr = inet6_csk_addr2sockaddr,
1700 .sockaddr_len = sizeof(struct sockaddr_in6),
1701 .bind_conflict = inet6_csk_bind_conflict,
1702#ifdef CONFIG_COMPAT
1703 .compat_setsockopt = compat_ipv6_setsockopt,
1704 .compat_getsockopt = compat_ipv6_getsockopt,
1705#endif
1706};
1707
1708#ifdef CONFIG_TCP_MD5SIG
1709static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1710 .md5_lookup = tcp_v4_md5_lookup,
1711 .calc_md5_hash = tcp_v4_md5_hash_skb,
1712 .md5_parse = tcp_v6_parse_md5_keys,
1713};
1714#endif
1715
1716/* NOTE: A lot of things set to zero explicitly by call to
1717 * sk_alloc() so need not be done here.
1718 */
1719static int tcp_v6_init_sock(struct sock *sk)
1720{
1721 struct inet_connection_sock *icsk = inet_csk(sk);
1722
1723 tcp_init_sock(sk);
1724
1725 icsk->icsk_af_ops = &ipv6_specific;
1726
1727#ifdef CONFIG_TCP_MD5SIG
1728 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1729#endif
1730
1731 return 0;
1732}
1733
1734static void tcp_v6_destroy_sock(struct sock *sk)
1735{
1736 tcp_v4_destroy_sock(sk);
1737 inet6_destroy_sock(sk);
1738}
1739
1740#ifdef CONFIG_PROC_FS
1741/* Proc filesystem TCPv6 sock list dumping. */
1742static void get_openreq6(struct seq_file *seq,
1743 const struct sock *sk, struct request_sock *req, int i, kuid_t uid)
1744{
1745 int ttd = req->expires - jiffies;
1746 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1747 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1748
1749 if (ttd < 0)
1750 ttd = 0;
1751
1752 seq_printf(seq,
1753 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1754 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1755 i,
1756 src->s6_addr32[0], src->s6_addr32[1],
1757 src->s6_addr32[2], src->s6_addr32[3],
1758 inet_rsk(req)->ir_num,
1759 dest->s6_addr32[0], dest->s6_addr32[1],
1760 dest->s6_addr32[2], dest->s6_addr32[3],
1761 ntohs(inet_rsk(req)->ir_rmt_port),
1762 TCP_SYN_RECV,
1763 0, 0, /* could print option size, but that is af dependent. */
1764 1, /* timers active (only the expire timer) */
1765 jiffies_to_clock_t(ttd),
1766 req->num_timeout,
1767 from_kuid_munged(seq_user_ns(seq), uid),
1768 0, /* non standard timer */
1769 0, /* open_requests have no inode */
1770 0, req);
1771}
1772
1773static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1774{
1775 const struct in6_addr *dest, *src;
1776 __u16 destp, srcp;
1777 int timer_active;
1778 unsigned long timer_expires;
1779 const struct inet_sock *inet = inet_sk(sp);
1780 const struct tcp_sock *tp = tcp_sk(sp);
1781 const struct inet_connection_sock *icsk = inet_csk(sp);
1782
1783 dest = &sp->sk_v6_daddr;
1784 src = &sp->sk_v6_rcv_saddr;
1785 destp = ntohs(inet->inet_dport);
1786 srcp = ntohs(inet->inet_sport);
1787
1788 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1789 timer_active = 1;
1790 timer_expires = icsk->icsk_timeout;
1791 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1792 timer_active = 4;
1793 timer_expires = icsk->icsk_timeout;
1794 } else if (timer_pending(&sp->sk_timer)) {
1795 timer_active = 2;
1796 timer_expires = sp->sk_timer.expires;
1797 } else {
1798 timer_active = 0;
1799 timer_expires = jiffies;
1800 }
1801
1802 seq_printf(seq,
1803 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1804 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1805 i,
1806 src->s6_addr32[0], src->s6_addr32[1],
1807 src->s6_addr32[2], src->s6_addr32[3], srcp,
1808 dest->s6_addr32[0], dest->s6_addr32[1],
1809 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1810 sp->sk_state,
1811 tp->write_seq-tp->snd_una,
1812 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1813 timer_active,
1814 jiffies_delta_to_clock_t(timer_expires - jiffies),
1815 icsk->icsk_retransmits,
1816 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1817 icsk->icsk_probes_out,
1818 sock_i_ino(sp),
1819 atomic_read(&sp->sk_refcnt), sp,
1820 jiffies_to_clock_t(icsk->icsk_rto),
1821 jiffies_to_clock_t(icsk->icsk_ack.ato),
1822 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1823 tp->snd_cwnd,
1824 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
1825 );
1826}
1827
1828static void get_timewait6_sock(struct seq_file *seq,
1829 struct inet_timewait_sock *tw, int i)
1830{
1831 const struct in6_addr *dest, *src;
1832 __u16 destp, srcp;
1833 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
1834
1835 dest = &tw->tw_v6_daddr;
1836 src = &tw->tw_v6_rcv_saddr;
1837 destp = ntohs(tw->tw_dport);
1838 srcp = ntohs(tw->tw_sport);
1839
1840 seq_printf(seq,
1841 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1842 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1843 i,
1844 src->s6_addr32[0], src->s6_addr32[1],
1845 src->s6_addr32[2], src->s6_addr32[3], srcp,
1846 dest->s6_addr32[0], dest->s6_addr32[1],
1847 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1848 tw->tw_substate, 0, 0,
1849 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1850 atomic_read(&tw->tw_refcnt), tw);
1851}
1852
1853static int tcp6_seq_show(struct seq_file *seq, void *v)
1854{
1855 struct tcp_iter_state *st;
1856 struct sock *sk = v;
1857
1858 if (v == SEQ_START_TOKEN) {
1859 seq_puts(seq,
1860 " sl "
1861 "local_address "
1862 "remote_address "
1863 "st tx_queue rx_queue tr tm->when retrnsmt"
1864 " uid timeout inode\n");
1865 goto out;
1866 }
1867 st = seq->private;
1868
1869 switch (st->state) {
1870 case TCP_SEQ_STATE_LISTENING:
1871 case TCP_SEQ_STATE_ESTABLISHED:
1872 if (sk->sk_state == TCP_TIME_WAIT)
1873 get_timewait6_sock(seq, v, st->num);
1874 else
1875 get_tcp6_sock(seq, v, st->num);
1876 break;
1877 case TCP_SEQ_STATE_OPENREQ:
1878 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1879 break;
1880 }
1881out:
1882 return 0;
1883}
1884
1885static const struct file_operations tcp6_afinfo_seq_fops = {
1886 .owner = THIS_MODULE,
1887 .open = tcp_seq_open,
1888 .read = seq_read,
1889 .llseek = seq_lseek,
1890 .release = seq_release_net
1891};
1892
1893static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1894 .name = "tcp6",
1895 .family = AF_INET6,
1896 .seq_fops = &tcp6_afinfo_seq_fops,
1897 .seq_ops = {
1898 .show = tcp6_seq_show,
1899 },
1900};
1901
1902int __net_init tcp6_proc_init(struct net *net)
1903{
1904 return tcp_proc_register(net, &tcp6_seq_afinfo);
1905}
1906
1907void tcp6_proc_exit(struct net *net)
1908{
1909 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1910}
1911#endif
1912
1913static void tcp_v6_clear_sk(struct sock *sk, int size)
1914{
1915 struct inet_sock *inet = inet_sk(sk);
1916
1917 /* we do not want to clear pinet6 field, because of RCU lookups */
1918 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1919
1920 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1921 memset(&inet->pinet6 + 1, 0, size);
1922}
1923
1924struct proto tcpv6_prot = {
1925 .name = "TCPv6",
1926 .owner = THIS_MODULE,
1927 .close = tcp_close,
1928 .connect = tcp_v6_connect,
1929 .disconnect = tcp_disconnect,
1930 .accept = inet_csk_accept,
1931 .ioctl = tcp_ioctl,
1932 .init = tcp_v6_init_sock,
1933 .destroy = tcp_v6_destroy_sock,
1934 .shutdown = tcp_shutdown,
1935 .setsockopt = tcp_setsockopt,
1936 .getsockopt = tcp_getsockopt,
1937 .recvmsg = tcp_recvmsg,
1938 .sendmsg = tcp_sendmsg,
1939 .sendpage = tcp_sendpage,
1940 .backlog_rcv = tcp_v6_do_rcv,
1941 .release_cb = tcp_release_cb,
1942 .mtu_reduced = tcp_v6_mtu_reduced,
1943 .hash = tcp_v6_hash,
1944 .unhash = inet_unhash,
1945 .get_port = inet_csk_get_port,
1946 .enter_memory_pressure = tcp_enter_memory_pressure,
1947 .stream_memory_free = tcp_stream_memory_free,
1948 .sockets_allocated = &tcp_sockets_allocated,
1949 .memory_allocated = &tcp_memory_allocated,
1950 .memory_pressure = &tcp_memory_pressure,
1951 .orphan_count = &tcp_orphan_count,
1952 .sysctl_mem = sysctl_tcp_mem,
1953 .sysctl_wmem = sysctl_tcp_wmem,
1954 .sysctl_rmem = sysctl_tcp_rmem,
1955 .max_header = MAX_TCP_HEADER,
1956 .obj_size = sizeof(struct tcp6_sock),
1957 .slab_flags = SLAB_DESTROY_BY_RCU,
1958 .twsk_prot = &tcp6_timewait_sock_ops,
1959 .rsk_prot = &tcp6_request_sock_ops,
1960 .h.hashinfo = &tcp_hashinfo,
1961 .no_autobind = true,
1962#ifdef CONFIG_COMPAT
1963 .compat_setsockopt = compat_tcp_setsockopt,
1964 .compat_getsockopt = compat_tcp_getsockopt,
1965#endif
1966#ifdef CONFIG_MEMCG_KMEM
1967 .proto_cgroup = tcp_proto_cgroup,
1968#endif
1969 .clear_sk = tcp_v6_clear_sk,
1970};
1971
1972static const struct inet6_protocol tcpv6_protocol = {
1973 .early_demux = tcp_v6_early_demux,
1974 .handler = tcp_v6_rcv,
1975 .err_handler = tcp_v6_err,
1976 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1977};
1978
1979static struct inet_protosw tcpv6_protosw = {
1980 .type = SOCK_STREAM,
1981 .protocol = IPPROTO_TCP,
1982 .prot = &tcpv6_prot,
1983 .ops = &inet6_stream_ops,
1984 .no_check = 0,
1985 .flags = INET_PROTOSW_PERMANENT |
1986 INET_PROTOSW_ICSK,
1987};
1988
1989static int __net_init tcpv6_net_init(struct net *net)
1990{
1991 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1992 SOCK_RAW, IPPROTO_TCP, net);
1993}
1994
1995static void __net_exit tcpv6_net_exit(struct net *net)
1996{
1997 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1998}
1999
2000static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2001{
2002 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2003}
2004
2005static struct pernet_operations tcpv6_net_ops = {
2006 .init = tcpv6_net_init,
2007 .exit = tcpv6_net_exit,
2008 .exit_batch = tcpv6_net_exit_batch,
2009};
2010
2011int __init tcpv6_init(void)
2012{
2013 int ret;
2014
2015 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2016 if (ret)
2017 goto out;
2018
2019 /* register inet6 protocol */
2020 ret = inet6_register_protosw(&tcpv6_protosw);
2021 if (ret)
2022 goto out_tcpv6_protocol;
2023
2024 ret = register_pernet_subsys(&tcpv6_net_ops);
2025 if (ret)
2026 goto out_tcpv6_protosw;
2027out:
2028 return ret;
2029
2030out_tcpv6_protosw:
2031 inet6_unregister_protosw(&tcpv6_protosw);
2032out_tcpv6_protocol:
2033 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2034 goto out;
2035}
2036
2037void tcpv6_exit(void)
2038{
2039 unregister_pernet_subsys(&tcpv6_net_ops);
2040 inet6_unregister_protosw(&tcpv6_protosw);
2041 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2042}
1/*
2 * TCP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on:
9 * linux/net/ipv4/tcp.c
10 * linux/net/ipv4/tcp_input.c
11 * linux/net/ipv4/tcp_output.c
12 *
13 * Fixes:
14 * Hideaki YOSHIFUJI : sin6_scope_id support
15 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
16 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
17 * a single port at the same time.
18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
19 *
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
24 */
25
26#include <linux/bottom_half.h>
27#include <linux/module.h>
28#include <linux/errno.h>
29#include <linux/types.h>
30#include <linux/socket.h>
31#include <linux/sockios.h>
32#include <linux/net.h>
33#include <linux/jiffies.h>
34#include <linux/in.h>
35#include <linux/in6.h>
36#include <linux/netdevice.h>
37#include <linux/init.h>
38#include <linux/jhash.h>
39#include <linux/ipsec.h>
40#include <linux/times.h>
41#include <linux/slab.h>
42#include <linux/uaccess.h>
43#include <linux/ipv6.h>
44#include <linux/icmpv6.h>
45#include <linux/random.h>
46
47#include <net/tcp.h>
48#include <net/ndisc.h>
49#include <net/inet6_hashtables.h>
50#include <net/inet6_connection_sock.h>
51#include <net/ipv6.h>
52#include <net/transp_v6.h>
53#include <net/addrconf.h>
54#include <net/ip6_route.h>
55#include <net/ip6_checksum.h>
56#include <net/inet_ecn.h>
57#include <net/protocol.h>
58#include <net/xfrm.h>
59#include <net/snmp.h>
60#include <net/dsfield.h>
61#include <net/timewait_sock.h>
62#include <net/inet_common.h>
63#include <net/secure_seq.h>
64#include <net/busy_poll.h>
65
66#include <linux/proc_fs.h>
67#include <linux/seq_file.h>
68
69#include <crypto/hash.h>
70#include <linux/scatterlist.h>
71
72#include <trace/events/tcp.h>
73
74static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
75static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
76 struct request_sock *req);
77
78static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
79
80static const struct inet_connection_sock_af_ops ipv6_mapped;
81static const struct inet_connection_sock_af_ops ipv6_specific;
82#ifdef CONFIG_TCP_MD5SIG
83static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
84static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
85#else
86static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
87 const struct in6_addr *addr)
88{
89 return NULL;
90}
91#endif
92
93static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
94{
95 struct dst_entry *dst = skb_dst(skb);
96
97 if (dst && dst_hold_safe(dst)) {
98 const struct rt6_info *rt = (const struct rt6_info *)dst;
99
100 sk->sk_rx_dst = dst;
101 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
102 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
103 }
104}
105
106static u32 tcp_v6_init_seq(const struct sk_buff *skb)
107{
108 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
109 ipv6_hdr(skb)->saddr.s6_addr32,
110 tcp_hdr(skb)->dest,
111 tcp_hdr(skb)->source);
112}
113
114static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
115{
116 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
117 ipv6_hdr(skb)->saddr.s6_addr32);
118}
119
120static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
121 int addr_len)
122{
123 /* This check is replicated from tcp_v6_connect() and intended to
124 * prevent BPF program called below from accessing bytes that are out
125 * of the bound specified by user in addr_len.
126 */
127 if (addr_len < SIN6_LEN_RFC2133)
128 return -EINVAL;
129
130 sock_owned_by_me(sk);
131
132 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr);
133}
134
135static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
136 int addr_len)
137{
138 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
139 struct inet_sock *inet = inet_sk(sk);
140 struct inet_connection_sock *icsk = inet_csk(sk);
141 struct ipv6_pinfo *np = inet6_sk(sk);
142 struct tcp_sock *tp = tcp_sk(sk);
143 struct in6_addr *saddr = NULL, *final_p, final;
144 struct ipv6_txoptions *opt;
145 struct flowi6 fl6;
146 struct dst_entry *dst;
147 int addr_type;
148 int err;
149 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
150
151 if (addr_len < SIN6_LEN_RFC2133)
152 return -EINVAL;
153
154 if (usin->sin6_family != AF_INET6)
155 return -EAFNOSUPPORT;
156
157 memset(&fl6, 0, sizeof(fl6));
158
159 if (np->sndflow) {
160 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
161 IP6_ECN_flow_init(fl6.flowlabel);
162 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
163 struct ip6_flowlabel *flowlabel;
164 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
165 if (!flowlabel)
166 return -EINVAL;
167 fl6_sock_release(flowlabel);
168 }
169 }
170
171 /*
172 * connect() to INADDR_ANY means loopback (BSD'ism).
173 */
174
175 if (ipv6_addr_any(&usin->sin6_addr)) {
176 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
177 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
178 &usin->sin6_addr);
179 else
180 usin->sin6_addr = in6addr_loopback;
181 }
182
183 addr_type = ipv6_addr_type(&usin->sin6_addr);
184
185 if (addr_type & IPV6_ADDR_MULTICAST)
186 return -ENETUNREACH;
187
188 if (addr_type&IPV6_ADDR_LINKLOCAL) {
189 if (addr_len >= sizeof(struct sockaddr_in6) &&
190 usin->sin6_scope_id) {
191 /* If interface is set while binding, indices
192 * must coincide.
193 */
194 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
195 return -EINVAL;
196
197 sk->sk_bound_dev_if = usin->sin6_scope_id;
198 }
199
200 /* Connect to link-local address requires an interface */
201 if (!sk->sk_bound_dev_if)
202 return -EINVAL;
203 }
204
205 if (tp->rx_opt.ts_recent_stamp &&
206 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
207 tp->rx_opt.ts_recent = 0;
208 tp->rx_opt.ts_recent_stamp = 0;
209 tp->write_seq = 0;
210 }
211
212 sk->sk_v6_daddr = usin->sin6_addr;
213 np->flow_label = fl6.flowlabel;
214
215 /*
216 * TCP over IPv4
217 */
218
219 if (addr_type & IPV6_ADDR_MAPPED) {
220 u32 exthdrlen = icsk->icsk_ext_hdr_len;
221 struct sockaddr_in sin;
222
223 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
224
225 if (__ipv6_only_sock(sk))
226 return -ENETUNREACH;
227
228 sin.sin_family = AF_INET;
229 sin.sin_port = usin->sin6_port;
230 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
231
232 icsk->icsk_af_ops = &ipv6_mapped;
233 sk->sk_backlog_rcv = tcp_v4_do_rcv;
234#ifdef CONFIG_TCP_MD5SIG
235 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
236#endif
237
238 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
239
240 if (err) {
241 icsk->icsk_ext_hdr_len = exthdrlen;
242 icsk->icsk_af_ops = &ipv6_specific;
243 sk->sk_backlog_rcv = tcp_v6_do_rcv;
244#ifdef CONFIG_TCP_MD5SIG
245 tp->af_specific = &tcp_sock_ipv6_specific;
246#endif
247 goto failure;
248 }
249 np->saddr = sk->sk_v6_rcv_saddr;
250
251 return err;
252 }
253
254 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
255 saddr = &sk->sk_v6_rcv_saddr;
256
257 fl6.flowi6_proto = IPPROTO_TCP;
258 fl6.daddr = sk->sk_v6_daddr;
259 fl6.saddr = saddr ? *saddr : np->saddr;
260 fl6.flowi6_oif = sk->sk_bound_dev_if;
261 fl6.flowi6_mark = sk->sk_mark;
262 fl6.fl6_dport = usin->sin6_port;
263 fl6.fl6_sport = inet->inet_sport;
264 fl6.flowi6_uid = sk->sk_uid;
265
266 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
267 final_p = fl6_update_dst(&fl6, opt, &final);
268
269 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
270
271 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
272 if (IS_ERR(dst)) {
273 err = PTR_ERR(dst);
274 goto failure;
275 }
276
277 if (!saddr) {
278 saddr = &fl6.saddr;
279 sk->sk_v6_rcv_saddr = *saddr;
280 }
281
282 /* set the source address */
283 np->saddr = *saddr;
284 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
285
286 sk->sk_gso_type = SKB_GSO_TCPV6;
287 ip6_dst_store(sk, dst, NULL, NULL);
288
289 icsk->icsk_ext_hdr_len = 0;
290 if (opt)
291 icsk->icsk_ext_hdr_len = opt->opt_flen +
292 opt->opt_nflen;
293
294 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
295
296 inet->inet_dport = usin->sin6_port;
297
298 tcp_set_state(sk, TCP_SYN_SENT);
299 err = inet6_hash_connect(tcp_death_row, sk);
300 if (err)
301 goto late_failure;
302
303 sk_set_txhash(sk);
304
305 if (likely(!tp->repair)) {
306 if (!tp->write_seq)
307 tp->write_seq = secure_tcpv6_seq(np->saddr.s6_addr32,
308 sk->sk_v6_daddr.s6_addr32,
309 inet->inet_sport,
310 inet->inet_dport);
311 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
312 np->saddr.s6_addr32,
313 sk->sk_v6_daddr.s6_addr32);
314 }
315
316 if (tcp_fastopen_defer_connect(sk, &err))
317 return err;
318 if (err)
319 goto late_failure;
320
321 err = tcp_connect(sk);
322 if (err)
323 goto late_failure;
324
325 return 0;
326
327late_failure:
328 tcp_set_state(sk, TCP_CLOSE);
329failure:
330 inet->inet_dport = 0;
331 sk->sk_route_caps = 0;
332 return err;
333}
334
335static void tcp_v6_mtu_reduced(struct sock *sk)
336{
337 struct dst_entry *dst;
338
339 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
340 return;
341
342 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
343 if (!dst)
344 return;
345
346 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
347 tcp_sync_mss(sk, dst_mtu(dst));
348 tcp_simple_retransmit(sk);
349 }
350}
351
352static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
353 u8 type, u8 code, int offset, __be32 info)
354{
355 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
356 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
357 struct net *net = dev_net(skb->dev);
358 struct request_sock *fastopen;
359 struct ipv6_pinfo *np;
360 struct tcp_sock *tp;
361 __u32 seq, snd_una;
362 struct sock *sk;
363 bool fatal;
364 int err;
365
366 sk = __inet6_lookup_established(net, &tcp_hashinfo,
367 &hdr->daddr, th->dest,
368 &hdr->saddr, ntohs(th->source),
369 skb->dev->ifindex, inet6_sdif(skb));
370
371 if (!sk) {
372 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
373 ICMP6_MIB_INERRORS);
374 return;
375 }
376
377 if (sk->sk_state == TCP_TIME_WAIT) {
378 inet_twsk_put(inet_twsk(sk));
379 return;
380 }
381 seq = ntohl(th->seq);
382 fatal = icmpv6_err_convert(type, code, &err);
383 if (sk->sk_state == TCP_NEW_SYN_RECV)
384 return tcp_req_err(sk, seq, fatal);
385
386 bh_lock_sock(sk);
387 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
388 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
389
390 if (sk->sk_state == TCP_CLOSE)
391 goto out;
392
393 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
394 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
395 goto out;
396 }
397
398 tp = tcp_sk(sk);
399 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
400 fastopen = tp->fastopen_rsk;
401 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
402 if (sk->sk_state != TCP_LISTEN &&
403 !between(seq, snd_una, tp->snd_nxt)) {
404 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
405 goto out;
406 }
407
408 np = inet6_sk(sk);
409
410 if (type == NDISC_REDIRECT) {
411 if (!sock_owned_by_user(sk)) {
412 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
413
414 if (dst)
415 dst->ops->redirect(dst, sk, skb);
416 }
417 goto out;
418 }
419
420 if (type == ICMPV6_PKT_TOOBIG) {
421 /* We are not interested in TCP_LISTEN and open_requests
422 * (SYN-ACKs send out by Linux are always <576bytes so
423 * they should go through unfragmented).
424 */
425 if (sk->sk_state == TCP_LISTEN)
426 goto out;
427
428 if (!ip6_sk_accept_pmtu(sk))
429 goto out;
430
431 tp->mtu_info = ntohl(info);
432 if (!sock_owned_by_user(sk))
433 tcp_v6_mtu_reduced(sk);
434 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
435 &sk->sk_tsq_flags))
436 sock_hold(sk);
437 goto out;
438 }
439
440
441 /* Might be for an request_sock */
442 switch (sk->sk_state) {
443 case TCP_SYN_SENT:
444 case TCP_SYN_RECV:
445 /* Only in fast or simultaneous open. If a fast open socket is
446 * is already accepted it is treated as a connected one below.
447 */
448 if (fastopen && !fastopen->sk)
449 break;
450
451 if (!sock_owned_by_user(sk)) {
452 sk->sk_err = err;
453 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
454
455 tcp_done(sk);
456 } else
457 sk->sk_err_soft = err;
458 goto out;
459 }
460
461 if (!sock_owned_by_user(sk) && np->recverr) {
462 sk->sk_err = err;
463 sk->sk_error_report(sk);
464 } else
465 sk->sk_err_soft = err;
466
467out:
468 bh_unlock_sock(sk);
469 sock_put(sk);
470}
471
472
473static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
474 struct flowi *fl,
475 struct request_sock *req,
476 struct tcp_fastopen_cookie *foc,
477 enum tcp_synack_type synack_type)
478{
479 struct inet_request_sock *ireq = inet_rsk(req);
480 struct ipv6_pinfo *np = inet6_sk(sk);
481 struct ipv6_txoptions *opt;
482 struct flowi6 *fl6 = &fl->u.ip6;
483 struct sk_buff *skb;
484 int err = -ENOMEM;
485
486 /* First, grab a route. */
487 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
488 IPPROTO_TCP)) == NULL)
489 goto done;
490
491 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
492
493 if (skb) {
494 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
495 &ireq->ir_v6_rmt_addr);
496
497 fl6->daddr = ireq->ir_v6_rmt_addr;
498 if (np->repflow && ireq->pktopts)
499 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
500
501 rcu_read_lock();
502 opt = ireq->ipv6_opt;
503 if (!opt)
504 opt = rcu_dereference(np->opt);
505 err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass);
506 rcu_read_unlock();
507 err = net_xmit_eval(err);
508 }
509
510done:
511 return err;
512}
513
514
515static void tcp_v6_reqsk_destructor(struct request_sock *req)
516{
517 kfree(inet_rsk(req)->ipv6_opt);
518 kfree_skb(inet_rsk(req)->pktopts);
519}
520
521#ifdef CONFIG_TCP_MD5SIG
522static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
523 const struct in6_addr *addr)
524{
525 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
526}
527
528static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
529 const struct sock *addr_sk)
530{
531 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
532}
533
534static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
535 char __user *optval, int optlen)
536{
537 struct tcp_md5sig cmd;
538 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
539 u8 prefixlen;
540
541 if (optlen < sizeof(cmd))
542 return -EINVAL;
543
544 if (copy_from_user(&cmd, optval, sizeof(cmd)))
545 return -EFAULT;
546
547 if (sin6->sin6_family != AF_INET6)
548 return -EINVAL;
549
550 if (optname == TCP_MD5SIG_EXT &&
551 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
552 prefixlen = cmd.tcpm_prefixlen;
553 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
554 prefixlen > 32))
555 return -EINVAL;
556 } else {
557 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
558 }
559
560 if (!cmd.tcpm_keylen) {
561 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
562 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
563 AF_INET, prefixlen);
564 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
565 AF_INET6, prefixlen);
566 }
567
568 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
569 return -EINVAL;
570
571 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
572 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
573 AF_INET, prefixlen, cmd.tcpm_key,
574 cmd.tcpm_keylen, GFP_KERNEL);
575
576 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
577 AF_INET6, prefixlen, cmd.tcpm_key,
578 cmd.tcpm_keylen, GFP_KERNEL);
579}
580
581static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
582 const struct in6_addr *daddr,
583 const struct in6_addr *saddr,
584 const struct tcphdr *th, int nbytes)
585{
586 struct tcp6_pseudohdr *bp;
587 struct scatterlist sg;
588 struct tcphdr *_th;
589
590 bp = hp->scratch;
591 /* 1. TCP pseudo-header (RFC2460) */
592 bp->saddr = *saddr;
593 bp->daddr = *daddr;
594 bp->protocol = cpu_to_be32(IPPROTO_TCP);
595 bp->len = cpu_to_be32(nbytes);
596
597 _th = (struct tcphdr *)(bp + 1);
598 memcpy(_th, th, sizeof(*th));
599 _th->check = 0;
600
601 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
602 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
603 sizeof(*bp) + sizeof(*th));
604 return crypto_ahash_update(hp->md5_req);
605}
606
607static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
608 const struct in6_addr *daddr, struct in6_addr *saddr,
609 const struct tcphdr *th)
610{
611 struct tcp_md5sig_pool *hp;
612 struct ahash_request *req;
613
614 hp = tcp_get_md5sig_pool();
615 if (!hp)
616 goto clear_hash_noput;
617 req = hp->md5_req;
618
619 if (crypto_ahash_init(req))
620 goto clear_hash;
621 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
622 goto clear_hash;
623 if (tcp_md5_hash_key(hp, key))
624 goto clear_hash;
625 ahash_request_set_crypt(req, NULL, md5_hash, 0);
626 if (crypto_ahash_final(req))
627 goto clear_hash;
628
629 tcp_put_md5sig_pool();
630 return 0;
631
632clear_hash:
633 tcp_put_md5sig_pool();
634clear_hash_noput:
635 memset(md5_hash, 0, 16);
636 return 1;
637}
638
639static int tcp_v6_md5_hash_skb(char *md5_hash,
640 const struct tcp_md5sig_key *key,
641 const struct sock *sk,
642 const struct sk_buff *skb)
643{
644 const struct in6_addr *saddr, *daddr;
645 struct tcp_md5sig_pool *hp;
646 struct ahash_request *req;
647 const struct tcphdr *th = tcp_hdr(skb);
648
649 if (sk) { /* valid for establish/request sockets */
650 saddr = &sk->sk_v6_rcv_saddr;
651 daddr = &sk->sk_v6_daddr;
652 } else {
653 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
654 saddr = &ip6h->saddr;
655 daddr = &ip6h->daddr;
656 }
657
658 hp = tcp_get_md5sig_pool();
659 if (!hp)
660 goto clear_hash_noput;
661 req = hp->md5_req;
662
663 if (crypto_ahash_init(req))
664 goto clear_hash;
665
666 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
667 goto clear_hash;
668 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
669 goto clear_hash;
670 if (tcp_md5_hash_key(hp, key))
671 goto clear_hash;
672 ahash_request_set_crypt(req, NULL, md5_hash, 0);
673 if (crypto_ahash_final(req))
674 goto clear_hash;
675
676 tcp_put_md5sig_pool();
677 return 0;
678
679clear_hash:
680 tcp_put_md5sig_pool();
681clear_hash_noput:
682 memset(md5_hash, 0, 16);
683 return 1;
684}
685
686#endif
687
688static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
689 const struct sk_buff *skb)
690{
691#ifdef CONFIG_TCP_MD5SIG
692 const __u8 *hash_location = NULL;
693 struct tcp_md5sig_key *hash_expected;
694 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
695 const struct tcphdr *th = tcp_hdr(skb);
696 int genhash;
697 u8 newhash[16];
698
699 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
700 hash_location = tcp_parse_md5sig_option(th);
701
702 /* We've parsed the options - do we have a hash? */
703 if (!hash_expected && !hash_location)
704 return false;
705
706 if (hash_expected && !hash_location) {
707 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
708 return true;
709 }
710
711 if (!hash_expected && hash_location) {
712 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
713 return true;
714 }
715
716 /* check the signature */
717 genhash = tcp_v6_md5_hash_skb(newhash,
718 hash_expected,
719 NULL, skb);
720
721 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
722 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
723 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
724 genhash ? "failed" : "mismatch",
725 &ip6h->saddr, ntohs(th->source),
726 &ip6h->daddr, ntohs(th->dest));
727 return true;
728 }
729#endif
730 return false;
731}
732
733static void tcp_v6_init_req(struct request_sock *req,
734 const struct sock *sk_listener,
735 struct sk_buff *skb)
736{
737 struct inet_request_sock *ireq = inet_rsk(req);
738 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
739
740 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
741 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
742
743 /* So that link locals have meaning */
744 if (!sk_listener->sk_bound_dev_if &&
745 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
746 ireq->ir_iif = tcp_v6_iif(skb);
747
748 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
749 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
750 np->rxopt.bits.rxinfo ||
751 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
752 np->rxopt.bits.rxohlim || np->repflow)) {
753 refcount_inc(&skb->users);
754 ireq->pktopts = skb;
755 }
756}
757
758static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
759 struct flowi *fl,
760 const struct request_sock *req)
761{
762 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
763}
764
765struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
766 .family = AF_INET6,
767 .obj_size = sizeof(struct tcp6_request_sock),
768 .rtx_syn_ack = tcp_rtx_synack,
769 .send_ack = tcp_v6_reqsk_send_ack,
770 .destructor = tcp_v6_reqsk_destructor,
771 .send_reset = tcp_v6_send_reset,
772 .syn_ack_timeout = tcp_syn_ack_timeout,
773};
774
775static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
776 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
777 sizeof(struct ipv6hdr),
778#ifdef CONFIG_TCP_MD5SIG
779 .req_md5_lookup = tcp_v6_md5_lookup,
780 .calc_md5_hash = tcp_v6_md5_hash_skb,
781#endif
782 .init_req = tcp_v6_init_req,
783#ifdef CONFIG_SYN_COOKIES
784 .cookie_init_seq = cookie_v6_init_sequence,
785#endif
786 .route_req = tcp_v6_route_req,
787 .init_seq = tcp_v6_init_seq,
788 .init_ts_off = tcp_v6_init_ts_off,
789 .send_synack = tcp_v6_send_synack,
790};
791
792static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
793 u32 ack, u32 win, u32 tsval, u32 tsecr,
794 int oif, struct tcp_md5sig_key *key, int rst,
795 u8 tclass, __be32 label)
796{
797 const struct tcphdr *th = tcp_hdr(skb);
798 struct tcphdr *t1;
799 struct sk_buff *buff;
800 struct flowi6 fl6;
801 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
802 struct sock *ctl_sk = net->ipv6.tcp_sk;
803 unsigned int tot_len = sizeof(struct tcphdr);
804 struct dst_entry *dst;
805 __be32 *topt;
806
807 if (tsecr)
808 tot_len += TCPOLEN_TSTAMP_ALIGNED;
809#ifdef CONFIG_TCP_MD5SIG
810 if (key)
811 tot_len += TCPOLEN_MD5SIG_ALIGNED;
812#endif
813
814 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
815 GFP_ATOMIC);
816 if (!buff)
817 return;
818
819 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
820
821 t1 = skb_push(buff, tot_len);
822 skb_reset_transport_header(buff);
823
824 /* Swap the send and the receive. */
825 memset(t1, 0, sizeof(*t1));
826 t1->dest = th->source;
827 t1->source = th->dest;
828 t1->doff = tot_len / 4;
829 t1->seq = htonl(seq);
830 t1->ack_seq = htonl(ack);
831 t1->ack = !rst || !th->ack;
832 t1->rst = rst;
833 t1->window = htons(win);
834
835 topt = (__be32 *)(t1 + 1);
836
837 if (tsecr) {
838 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
839 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
840 *topt++ = htonl(tsval);
841 *topt++ = htonl(tsecr);
842 }
843
844#ifdef CONFIG_TCP_MD5SIG
845 if (key) {
846 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
847 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
848 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
849 &ipv6_hdr(skb)->saddr,
850 &ipv6_hdr(skb)->daddr, t1);
851 }
852#endif
853
854 memset(&fl6, 0, sizeof(fl6));
855 fl6.daddr = ipv6_hdr(skb)->saddr;
856 fl6.saddr = ipv6_hdr(skb)->daddr;
857 fl6.flowlabel = label;
858
859 buff->ip_summed = CHECKSUM_PARTIAL;
860 buff->csum = 0;
861
862 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
863
864 fl6.flowi6_proto = IPPROTO_TCP;
865 if (rt6_need_strict(&fl6.daddr) && !oif)
866 fl6.flowi6_oif = tcp_v6_iif(skb);
867 else {
868 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
869 oif = skb->skb_iif;
870
871 fl6.flowi6_oif = oif;
872 }
873
874 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
875 fl6.fl6_dport = t1->dest;
876 fl6.fl6_sport = t1->source;
877 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
878 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
879
880 /* Pass a socket to ip6_dst_lookup either it is for RST
881 * Underlying function will use this to retrieve the network
882 * namespace
883 */
884 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
885 if (!IS_ERR(dst)) {
886 skb_dst_set(buff, dst);
887 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, tclass);
888 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
889 if (rst)
890 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
891 return;
892 }
893
894 kfree_skb(buff);
895}
896
897static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
898{
899 const struct tcphdr *th = tcp_hdr(skb);
900 u32 seq = 0, ack_seq = 0;
901 struct tcp_md5sig_key *key = NULL;
902#ifdef CONFIG_TCP_MD5SIG
903 const __u8 *hash_location = NULL;
904 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
905 unsigned char newhash[16];
906 int genhash;
907 struct sock *sk1 = NULL;
908#endif
909 int oif = 0;
910
911 if (th->rst)
912 return;
913
914 /* If sk not NULL, it means we did a successful lookup and incoming
915 * route had to be correct. prequeue might have dropped our dst.
916 */
917 if (!sk && !ipv6_unicast_destination(skb))
918 return;
919
920#ifdef CONFIG_TCP_MD5SIG
921 rcu_read_lock();
922 hash_location = tcp_parse_md5sig_option(th);
923 if (sk && sk_fullsock(sk)) {
924 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
925 } else if (hash_location) {
926 /*
927 * active side is lost. Try to find listening socket through
928 * source port, and then find md5 key through listening socket.
929 * we are not loose security here:
930 * Incoming packet is checked with md5 hash with finding key,
931 * no RST generated if md5 hash doesn't match.
932 */
933 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
934 &tcp_hashinfo, NULL, 0,
935 &ipv6h->saddr,
936 th->source, &ipv6h->daddr,
937 ntohs(th->source), tcp_v6_iif(skb),
938 tcp_v6_sdif(skb));
939 if (!sk1)
940 goto out;
941
942 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
943 if (!key)
944 goto out;
945
946 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
947 if (genhash || memcmp(hash_location, newhash, 16) != 0)
948 goto out;
949 }
950#endif
951
952 if (th->ack)
953 seq = ntohl(th->ack_seq);
954 else
955 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
956 (th->doff << 2);
957
958 if (sk) {
959 oif = sk->sk_bound_dev_if;
960 if (sk_fullsock(sk))
961 trace_tcp_send_reset(sk, skb);
962 }
963
964 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
965
966#ifdef CONFIG_TCP_MD5SIG
967out:
968 rcu_read_unlock();
969#endif
970}
971
972static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
973 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
974 struct tcp_md5sig_key *key, u8 tclass,
975 __be32 label)
976{
977 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
978 tclass, label);
979}
980
981static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
982{
983 struct inet_timewait_sock *tw = inet_twsk(sk);
984 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
985
986 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
987 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
988 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
989 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
990 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
991
992 inet_twsk_put(tw);
993}
994
995static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
996 struct request_sock *req)
997{
998 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
999 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1000 */
1001 /* RFC 7323 2.3
1002 * The window field (SEG.WND) of every outgoing segment, with the
1003 * exception of <SYN> segments, MUST be right-shifted by
1004 * Rcv.Wind.Shift bits:
1005 */
1006 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1007 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1008 tcp_rsk(req)->rcv_nxt,
1009 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1010 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1011 req->ts_recent, sk->sk_bound_dev_if,
1012 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr),
1013 0, 0);
1014}
1015
1016
1017static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1018{
1019#ifdef CONFIG_SYN_COOKIES
1020 const struct tcphdr *th = tcp_hdr(skb);
1021
1022 if (!th->syn)
1023 sk = cookie_v6_check(sk, skb);
1024#endif
1025 return sk;
1026}
1027
1028static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1029{
1030 if (skb->protocol == htons(ETH_P_IP))
1031 return tcp_v4_conn_request(sk, skb);
1032
1033 if (!ipv6_unicast_destination(skb))
1034 goto drop;
1035
1036 return tcp_conn_request(&tcp6_request_sock_ops,
1037 &tcp_request_sock_ipv6_ops, sk, skb);
1038
1039drop:
1040 tcp_listendrop(sk);
1041 return 0; /* don't send reset */
1042}
1043
1044static void tcp_v6_restore_cb(struct sk_buff *skb)
1045{
1046 /* We need to move header back to the beginning if xfrm6_policy_check()
1047 * and tcp_v6_fill_cb() are going to be called again.
1048 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1049 */
1050 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1051 sizeof(struct inet6_skb_parm));
1052}
1053
1054static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1055 struct request_sock *req,
1056 struct dst_entry *dst,
1057 struct request_sock *req_unhash,
1058 bool *own_req)
1059{
1060 struct inet_request_sock *ireq;
1061 struct ipv6_pinfo *newnp;
1062 const struct ipv6_pinfo *np = inet6_sk(sk);
1063 struct ipv6_txoptions *opt;
1064 struct tcp6_sock *newtcp6sk;
1065 struct inet_sock *newinet;
1066 struct tcp_sock *newtp;
1067 struct sock *newsk;
1068#ifdef CONFIG_TCP_MD5SIG
1069 struct tcp_md5sig_key *key;
1070#endif
1071 struct flowi6 fl6;
1072
1073 if (skb->protocol == htons(ETH_P_IP)) {
1074 /*
1075 * v6 mapped
1076 */
1077
1078 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1079 req_unhash, own_req);
1080
1081 if (!newsk)
1082 return NULL;
1083
1084 newtcp6sk = (struct tcp6_sock *)newsk;
1085 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1086
1087 newinet = inet_sk(newsk);
1088 newnp = inet6_sk(newsk);
1089 newtp = tcp_sk(newsk);
1090
1091 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1092
1093 newnp->saddr = newsk->sk_v6_rcv_saddr;
1094
1095 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1096 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1097#ifdef CONFIG_TCP_MD5SIG
1098 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1099#endif
1100
1101 newnp->ipv6_mc_list = NULL;
1102 newnp->ipv6_ac_list = NULL;
1103 newnp->ipv6_fl_list = NULL;
1104 newnp->pktoptions = NULL;
1105 newnp->opt = NULL;
1106 newnp->mcast_oif = tcp_v6_iif(skb);
1107 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1108 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1109 if (np->repflow)
1110 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1111
1112 /*
1113 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1114 * here, tcp_create_openreq_child now does this for us, see the comment in
1115 * that function for the gory details. -acme
1116 */
1117
1118 /* It is tricky place. Until this moment IPv4 tcp
1119 worked with IPv6 icsk.icsk_af_ops.
1120 Sync it now.
1121 */
1122 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1123
1124 return newsk;
1125 }
1126
1127 ireq = inet_rsk(req);
1128
1129 if (sk_acceptq_is_full(sk))
1130 goto out_overflow;
1131
1132 if (!dst) {
1133 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1134 if (!dst)
1135 goto out;
1136 }
1137
1138 newsk = tcp_create_openreq_child(sk, req, skb);
1139 if (!newsk)
1140 goto out_nonewsk;
1141
1142 /*
1143 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1144 * count here, tcp_create_openreq_child now does this for us, see the
1145 * comment in that function for the gory details. -acme
1146 */
1147
1148 newsk->sk_gso_type = SKB_GSO_TCPV6;
1149 ip6_dst_store(newsk, dst, NULL, NULL);
1150 inet6_sk_rx_dst_set(newsk, skb);
1151
1152 newtcp6sk = (struct tcp6_sock *)newsk;
1153 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1154
1155 newtp = tcp_sk(newsk);
1156 newinet = inet_sk(newsk);
1157 newnp = inet6_sk(newsk);
1158
1159 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1160
1161 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1162 newnp->saddr = ireq->ir_v6_loc_addr;
1163 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1164 newsk->sk_bound_dev_if = ireq->ir_iif;
1165
1166 /* Now IPv6 options...
1167
1168 First: no IPv4 options.
1169 */
1170 newinet->inet_opt = NULL;
1171 newnp->ipv6_mc_list = NULL;
1172 newnp->ipv6_ac_list = NULL;
1173 newnp->ipv6_fl_list = NULL;
1174
1175 /* Clone RX bits */
1176 newnp->rxopt.all = np->rxopt.all;
1177
1178 newnp->pktoptions = NULL;
1179 newnp->opt = NULL;
1180 newnp->mcast_oif = tcp_v6_iif(skb);
1181 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1182 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1183 if (np->repflow)
1184 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1185
1186 /* Clone native IPv6 options from listening socket (if any)
1187
1188 Yes, keeping reference count would be much more clever,
1189 but we make one more one thing there: reattach optmem
1190 to newsk.
1191 */
1192 opt = ireq->ipv6_opt;
1193 if (!opt)
1194 opt = rcu_dereference(np->opt);
1195 if (opt) {
1196 opt = ipv6_dup_options(newsk, opt);
1197 RCU_INIT_POINTER(newnp->opt, opt);
1198 }
1199 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1200 if (opt)
1201 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1202 opt->opt_flen;
1203
1204 tcp_ca_openreq_child(newsk, dst);
1205
1206 tcp_sync_mss(newsk, dst_mtu(dst));
1207 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1208
1209 tcp_initialize_rcv_mss(newsk);
1210
1211 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1212 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1213
1214#ifdef CONFIG_TCP_MD5SIG
1215 /* Copy over the MD5 key from the original socket */
1216 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1217 if (key) {
1218 /* We're using one, so create a matching key
1219 * on the newsk structure. If we fail to get
1220 * memory, then we end up not copying the key
1221 * across. Shucks.
1222 */
1223 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1224 AF_INET6, 128, key->key, key->keylen,
1225 sk_gfp_mask(sk, GFP_ATOMIC));
1226 }
1227#endif
1228
1229 if (__inet_inherit_port(sk, newsk) < 0) {
1230 inet_csk_prepare_forced_close(newsk);
1231 tcp_done(newsk);
1232 goto out;
1233 }
1234 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1235 if (*own_req) {
1236 tcp_move_syn(newtp, req);
1237
1238 /* Clone pktoptions received with SYN, if we own the req */
1239 if (ireq->pktopts) {
1240 newnp->pktoptions = skb_clone(ireq->pktopts,
1241 sk_gfp_mask(sk, GFP_ATOMIC));
1242 consume_skb(ireq->pktopts);
1243 ireq->pktopts = NULL;
1244 if (newnp->pktoptions) {
1245 tcp_v6_restore_cb(newnp->pktoptions);
1246 skb_set_owner_r(newnp->pktoptions, newsk);
1247 }
1248 }
1249 }
1250
1251 return newsk;
1252
1253out_overflow:
1254 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1255out_nonewsk:
1256 dst_release(dst);
1257out:
1258 tcp_listendrop(sk);
1259 return NULL;
1260}
1261
1262/* The socket must have it's spinlock held when we get
1263 * here, unless it is a TCP_LISTEN socket.
1264 *
1265 * We have a potential double-lock case here, so even when
1266 * doing backlog processing we use the BH locking scheme.
1267 * This is because we cannot sleep with the original spinlock
1268 * held.
1269 */
1270static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1271{
1272 struct ipv6_pinfo *np = inet6_sk(sk);
1273 struct tcp_sock *tp;
1274 struct sk_buff *opt_skb = NULL;
1275
1276 /* Imagine: socket is IPv6. IPv4 packet arrives,
1277 goes to IPv4 receive handler and backlogged.
1278 From backlog it always goes here. Kerboom...
1279 Fortunately, tcp_rcv_established and rcv_established
1280 handle them correctly, but it is not case with
1281 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1282 */
1283
1284 if (skb->protocol == htons(ETH_P_IP))
1285 return tcp_v4_do_rcv(sk, skb);
1286
1287 /*
1288 * socket locking is here for SMP purposes as backlog rcv
1289 * is currently called with bh processing disabled.
1290 */
1291
1292 /* Do Stevens' IPV6_PKTOPTIONS.
1293
1294 Yes, guys, it is the only place in our code, where we
1295 may make it not affecting IPv4.
1296 The rest of code is protocol independent,
1297 and I do not like idea to uglify IPv4.
1298
1299 Actually, all the idea behind IPV6_PKTOPTIONS
1300 looks not very well thought. For now we latch
1301 options, received in the last packet, enqueued
1302 by tcp. Feel free to propose better solution.
1303 --ANK (980728)
1304 */
1305 if (np->rxopt.all)
1306 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1307
1308 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1309 struct dst_entry *dst = sk->sk_rx_dst;
1310
1311 sock_rps_save_rxhash(sk, skb);
1312 sk_mark_napi_id(sk, skb);
1313 if (dst) {
1314 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1315 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1316 dst_release(dst);
1317 sk->sk_rx_dst = NULL;
1318 }
1319 }
1320
1321 tcp_rcv_established(sk, skb, tcp_hdr(skb));
1322 if (opt_skb)
1323 goto ipv6_pktoptions;
1324 return 0;
1325 }
1326
1327 if (tcp_checksum_complete(skb))
1328 goto csum_err;
1329
1330 if (sk->sk_state == TCP_LISTEN) {
1331 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1332
1333 if (!nsk)
1334 goto discard;
1335
1336 if (nsk != sk) {
1337 if (tcp_child_process(sk, nsk, skb))
1338 goto reset;
1339 if (opt_skb)
1340 __kfree_skb(opt_skb);
1341 return 0;
1342 }
1343 } else
1344 sock_rps_save_rxhash(sk, skb);
1345
1346 if (tcp_rcv_state_process(sk, skb))
1347 goto reset;
1348 if (opt_skb)
1349 goto ipv6_pktoptions;
1350 return 0;
1351
1352reset:
1353 tcp_v6_send_reset(sk, skb);
1354discard:
1355 if (opt_skb)
1356 __kfree_skb(opt_skb);
1357 kfree_skb(skb);
1358 return 0;
1359csum_err:
1360 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1361 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1362 goto discard;
1363
1364
1365ipv6_pktoptions:
1366 /* Do you ask, what is it?
1367
1368 1. skb was enqueued by tcp.
1369 2. skb is added to tail of read queue, rather than out of order.
1370 3. socket is not in passive state.
1371 4. Finally, it really contains options, which user wants to receive.
1372 */
1373 tp = tcp_sk(sk);
1374 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1375 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1376 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1377 np->mcast_oif = tcp_v6_iif(opt_skb);
1378 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1379 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1380 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1381 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1382 if (np->repflow)
1383 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1384 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1385 skb_set_owner_r(opt_skb, sk);
1386 tcp_v6_restore_cb(opt_skb);
1387 opt_skb = xchg(&np->pktoptions, opt_skb);
1388 } else {
1389 __kfree_skb(opt_skb);
1390 opt_skb = xchg(&np->pktoptions, NULL);
1391 }
1392 }
1393
1394 kfree_skb(opt_skb);
1395 return 0;
1396}
1397
1398static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1399 const struct tcphdr *th)
1400{
1401 /* This is tricky: we move IP6CB at its correct location into
1402 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1403 * _decode_session6() uses IP6CB().
1404 * barrier() makes sure compiler won't play aliasing games.
1405 */
1406 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1407 sizeof(struct inet6_skb_parm));
1408 barrier();
1409
1410 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1411 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1412 skb->len - th->doff*4);
1413 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1414 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1415 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1416 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1417 TCP_SKB_CB(skb)->sacked = 0;
1418 TCP_SKB_CB(skb)->has_rxtstamp =
1419 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1420}
1421
1422static int tcp_v6_rcv(struct sk_buff *skb)
1423{
1424 int sdif = inet6_sdif(skb);
1425 const struct tcphdr *th;
1426 const struct ipv6hdr *hdr;
1427 bool refcounted;
1428 struct sock *sk;
1429 int ret;
1430 struct net *net = dev_net(skb->dev);
1431
1432 if (skb->pkt_type != PACKET_HOST)
1433 goto discard_it;
1434
1435 /*
1436 * Count it even if it's bad.
1437 */
1438 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1439
1440 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1441 goto discard_it;
1442
1443 th = (const struct tcphdr *)skb->data;
1444
1445 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1446 goto bad_packet;
1447 if (!pskb_may_pull(skb, th->doff*4))
1448 goto discard_it;
1449
1450 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1451 goto csum_error;
1452
1453 th = (const struct tcphdr *)skb->data;
1454 hdr = ipv6_hdr(skb);
1455
1456lookup:
1457 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1458 th->source, th->dest, inet6_iif(skb), sdif,
1459 &refcounted);
1460 if (!sk)
1461 goto no_tcp_socket;
1462
1463process:
1464 if (sk->sk_state == TCP_TIME_WAIT)
1465 goto do_time_wait;
1466
1467 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1468 struct request_sock *req = inet_reqsk(sk);
1469 bool req_stolen = false;
1470 struct sock *nsk;
1471
1472 sk = req->rsk_listener;
1473 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1474 sk_drops_add(sk, skb);
1475 reqsk_put(req);
1476 goto discard_it;
1477 }
1478 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1479 inet_csk_reqsk_queue_drop_and_put(sk, req);
1480 goto lookup;
1481 }
1482 sock_hold(sk);
1483 refcounted = true;
1484 nsk = NULL;
1485 if (!tcp_filter(sk, skb)) {
1486 th = (const struct tcphdr *)skb->data;
1487 hdr = ipv6_hdr(skb);
1488 tcp_v6_fill_cb(skb, hdr, th);
1489 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1490 }
1491 if (!nsk) {
1492 reqsk_put(req);
1493 if (req_stolen) {
1494 /* Another cpu got exclusive access to req
1495 * and created a full blown socket.
1496 * Try to feed this packet to this socket
1497 * instead of discarding it.
1498 */
1499 tcp_v6_restore_cb(skb);
1500 sock_put(sk);
1501 goto lookup;
1502 }
1503 goto discard_and_relse;
1504 }
1505 if (nsk == sk) {
1506 reqsk_put(req);
1507 tcp_v6_restore_cb(skb);
1508 } else if (tcp_child_process(sk, nsk, skb)) {
1509 tcp_v6_send_reset(nsk, skb);
1510 goto discard_and_relse;
1511 } else {
1512 sock_put(sk);
1513 return 0;
1514 }
1515 }
1516 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1517 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1518 goto discard_and_relse;
1519 }
1520
1521 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1522 goto discard_and_relse;
1523
1524 if (tcp_v6_inbound_md5_hash(sk, skb))
1525 goto discard_and_relse;
1526
1527 if (tcp_filter(sk, skb))
1528 goto discard_and_relse;
1529 th = (const struct tcphdr *)skb->data;
1530 hdr = ipv6_hdr(skb);
1531 tcp_v6_fill_cb(skb, hdr, th);
1532
1533 skb->dev = NULL;
1534
1535 if (sk->sk_state == TCP_LISTEN) {
1536 ret = tcp_v6_do_rcv(sk, skb);
1537 goto put_and_return;
1538 }
1539
1540 sk_incoming_cpu_update(sk);
1541
1542 bh_lock_sock_nested(sk);
1543 tcp_segs_in(tcp_sk(sk), skb);
1544 ret = 0;
1545 if (!sock_owned_by_user(sk)) {
1546 ret = tcp_v6_do_rcv(sk, skb);
1547 } else if (tcp_add_backlog(sk, skb)) {
1548 goto discard_and_relse;
1549 }
1550 bh_unlock_sock(sk);
1551
1552put_and_return:
1553 if (refcounted)
1554 sock_put(sk);
1555 return ret ? -1 : 0;
1556
1557no_tcp_socket:
1558 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1559 goto discard_it;
1560
1561 tcp_v6_fill_cb(skb, hdr, th);
1562
1563 if (tcp_checksum_complete(skb)) {
1564csum_error:
1565 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1566bad_packet:
1567 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1568 } else {
1569 tcp_v6_send_reset(NULL, skb);
1570 }
1571
1572discard_it:
1573 kfree_skb(skb);
1574 return 0;
1575
1576discard_and_relse:
1577 sk_drops_add(sk, skb);
1578 if (refcounted)
1579 sock_put(sk);
1580 goto discard_it;
1581
1582do_time_wait:
1583 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1584 inet_twsk_put(inet_twsk(sk));
1585 goto discard_it;
1586 }
1587
1588 tcp_v6_fill_cb(skb, hdr, th);
1589
1590 if (tcp_checksum_complete(skb)) {
1591 inet_twsk_put(inet_twsk(sk));
1592 goto csum_error;
1593 }
1594
1595 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1596 case TCP_TW_SYN:
1597 {
1598 struct sock *sk2;
1599
1600 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1601 skb, __tcp_hdrlen(th),
1602 &ipv6_hdr(skb)->saddr, th->source,
1603 &ipv6_hdr(skb)->daddr,
1604 ntohs(th->dest), tcp_v6_iif(skb),
1605 sdif);
1606 if (sk2) {
1607 struct inet_timewait_sock *tw = inet_twsk(sk);
1608 inet_twsk_deschedule_put(tw);
1609 sk = sk2;
1610 tcp_v6_restore_cb(skb);
1611 refcounted = false;
1612 goto process;
1613 }
1614 }
1615 /* to ACK */
1616 /* fall through */
1617 case TCP_TW_ACK:
1618 tcp_v6_timewait_ack(sk, skb);
1619 break;
1620 case TCP_TW_RST:
1621 tcp_v6_send_reset(sk, skb);
1622 inet_twsk_deschedule_put(inet_twsk(sk));
1623 goto discard_it;
1624 case TCP_TW_SUCCESS:
1625 ;
1626 }
1627 goto discard_it;
1628}
1629
1630static void tcp_v6_early_demux(struct sk_buff *skb)
1631{
1632 const struct ipv6hdr *hdr;
1633 const struct tcphdr *th;
1634 struct sock *sk;
1635
1636 if (skb->pkt_type != PACKET_HOST)
1637 return;
1638
1639 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1640 return;
1641
1642 hdr = ipv6_hdr(skb);
1643 th = tcp_hdr(skb);
1644
1645 if (th->doff < sizeof(struct tcphdr) / 4)
1646 return;
1647
1648 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1649 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1650 &hdr->saddr, th->source,
1651 &hdr->daddr, ntohs(th->dest),
1652 inet6_iif(skb), inet6_sdif(skb));
1653 if (sk) {
1654 skb->sk = sk;
1655 skb->destructor = sock_edemux;
1656 if (sk_fullsock(sk)) {
1657 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1658
1659 if (dst)
1660 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1661 if (dst &&
1662 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1663 skb_dst_set_noref(skb, dst);
1664 }
1665 }
1666}
1667
1668static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1669 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1670 .twsk_unique = tcp_twsk_unique,
1671 .twsk_destructor = tcp_twsk_destructor,
1672};
1673
1674static const struct inet_connection_sock_af_ops ipv6_specific = {
1675 .queue_xmit = inet6_csk_xmit,
1676 .send_check = tcp_v6_send_check,
1677 .rebuild_header = inet6_sk_rebuild_header,
1678 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1679 .conn_request = tcp_v6_conn_request,
1680 .syn_recv_sock = tcp_v6_syn_recv_sock,
1681 .net_header_len = sizeof(struct ipv6hdr),
1682 .net_frag_header_len = sizeof(struct frag_hdr),
1683 .setsockopt = ipv6_setsockopt,
1684 .getsockopt = ipv6_getsockopt,
1685 .addr2sockaddr = inet6_csk_addr2sockaddr,
1686 .sockaddr_len = sizeof(struct sockaddr_in6),
1687#ifdef CONFIG_COMPAT
1688 .compat_setsockopt = compat_ipv6_setsockopt,
1689 .compat_getsockopt = compat_ipv6_getsockopt,
1690#endif
1691 .mtu_reduced = tcp_v6_mtu_reduced,
1692};
1693
1694#ifdef CONFIG_TCP_MD5SIG
1695static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1696 .md5_lookup = tcp_v6_md5_lookup,
1697 .calc_md5_hash = tcp_v6_md5_hash_skb,
1698 .md5_parse = tcp_v6_parse_md5_keys,
1699};
1700#endif
1701
1702/*
1703 * TCP over IPv4 via INET6 API
1704 */
1705static const struct inet_connection_sock_af_ops ipv6_mapped = {
1706 .queue_xmit = ip_queue_xmit,
1707 .send_check = tcp_v4_send_check,
1708 .rebuild_header = inet_sk_rebuild_header,
1709 .sk_rx_dst_set = inet_sk_rx_dst_set,
1710 .conn_request = tcp_v6_conn_request,
1711 .syn_recv_sock = tcp_v6_syn_recv_sock,
1712 .net_header_len = sizeof(struct iphdr),
1713 .setsockopt = ipv6_setsockopt,
1714 .getsockopt = ipv6_getsockopt,
1715 .addr2sockaddr = inet6_csk_addr2sockaddr,
1716 .sockaddr_len = sizeof(struct sockaddr_in6),
1717#ifdef CONFIG_COMPAT
1718 .compat_setsockopt = compat_ipv6_setsockopt,
1719 .compat_getsockopt = compat_ipv6_getsockopt,
1720#endif
1721 .mtu_reduced = tcp_v4_mtu_reduced,
1722};
1723
1724#ifdef CONFIG_TCP_MD5SIG
1725static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1726 .md5_lookup = tcp_v4_md5_lookup,
1727 .calc_md5_hash = tcp_v4_md5_hash_skb,
1728 .md5_parse = tcp_v6_parse_md5_keys,
1729};
1730#endif
1731
1732/* NOTE: A lot of things set to zero explicitly by call to
1733 * sk_alloc() so need not be done here.
1734 */
1735static int tcp_v6_init_sock(struct sock *sk)
1736{
1737 struct inet_connection_sock *icsk = inet_csk(sk);
1738
1739 tcp_init_sock(sk);
1740
1741 icsk->icsk_af_ops = &ipv6_specific;
1742
1743#ifdef CONFIG_TCP_MD5SIG
1744 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1745#endif
1746
1747 return 0;
1748}
1749
1750static void tcp_v6_destroy_sock(struct sock *sk)
1751{
1752 tcp_v4_destroy_sock(sk);
1753 inet6_destroy_sock(sk);
1754}
1755
1756#ifdef CONFIG_PROC_FS
1757/* Proc filesystem TCPv6 sock list dumping. */
1758static void get_openreq6(struct seq_file *seq,
1759 const struct request_sock *req, int i)
1760{
1761 long ttd = req->rsk_timer.expires - jiffies;
1762 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1763 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1764
1765 if (ttd < 0)
1766 ttd = 0;
1767
1768 seq_printf(seq,
1769 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1770 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1771 i,
1772 src->s6_addr32[0], src->s6_addr32[1],
1773 src->s6_addr32[2], src->s6_addr32[3],
1774 inet_rsk(req)->ir_num,
1775 dest->s6_addr32[0], dest->s6_addr32[1],
1776 dest->s6_addr32[2], dest->s6_addr32[3],
1777 ntohs(inet_rsk(req)->ir_rmt_port),
1778 TCP_SYN_RECV,
1779 0, 0, /* could print option size, but that is af dependent. */
1780 1, /* timers active (only the expire timer) */
1781 jiffies_to_clock_t(ttd),
1782 req->num_timeout,
1783 from_kuid_munged(seq_user_ns(seq),
1784 sock_i_uid(req->rsk_listener)),
1785 0, /* non standard timer */
1786 0, /* open_requests have no inode */
1787 0, req);
1788}
1789
1790static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1791{
1792 const struct in6_addr *dest, *src;
1793 __u16 destp, srcp;
1794 int timer_active;
1795 unsigned long timer_expires;
1796 const struct inet_sock *inet = inet_sk(sp);
1797 const struct tcp_sock *tp = tcp_sk(sp);
1798 const struct inet_connection_sock *icsk = inet_csk(sp);
1799 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1800 int rx_queue;
1801 int state;
1802
1803 dest = &sp->sk_v6_daddr;
1804 src = &sp->sk_v6_rcv_saddr;
1805 destp = ntohs(inet->inet_dport);
1806 srcp = ntohs(inet->inet_sport);
1807
1808 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1809 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
1810 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1811 timer_active = 1;
1812 timer_expires = icsk->icsk_timeout;
1813 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1814 timer_active = 4;
1815 timer_expires = icsk->icsk_timeout;
1816 } else if (timer_pending(&sp->sk_timer)) {
1817 timer_active = 2;
1818 timer_expires = sp->sk_timer.expires;
1819 } else {
1820 timer_active = 0;
1821 timer_expires = jiffies;
1822 }
1823
1824 state = inet_sk_state_load(sp);
1825 if (state == TCP_LISTEN)
1826 rx_queue = sp->sk_ack_backlog;
1827 else
1828 /* Because we don't lock the socket,
1829 * we might find a transient negative value.
1830 */
1831 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1832
1833 seq_printf(seq,
1834 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1835 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1836 i,
1837 src->s6_addr32[0], src->s6_addr32[1],
1838 src->s6_addr32[2], src->s6_addr32[3], srcp,
1839 dest->s6_addr32[0], dest->s6_addr32[1],
1840 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1841 state,
1842 tp->write_seq - tp->snd_una,
1843 rx_queue,
1844 timer_active,
1845 jiffies_delta_to_clock_t(timer_expires - jiffies),
1846 icsk->icsk_retransmits,
1847 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1848 icsk->icsk_probes_out,
1849 sock_i_ino(sp),
1850 refcount_read(&sp->sk_refcnt), sp,
1851 jiffies_to_clock_t(icsk->icsk_rto),
1852 jiffies_to_clock_t(icsk->icsk_ack.ato),
1853 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1854 tp->snd_cwnd,
1855 state == TCP_LISTEN ?
1856 fastopenq->max_qlen :
1857 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1858 );
1859}
1860
1861static void get_timewait6_sock(struct seq_file *seq,
1862 struct inet_timewait_sock *tw, int i)
1863{
1864 long delta = tw->tw_timer.expires - jiffies;
1865 const struct in6_addr *dest, *src;
1866 __u16 destp, srcp;
1867
1868 dest = &tw->tw_v6_daddr;
1869 src = &tw->tw_v6_rcv_saddr;
1870 destp = ntohs(tw->tw_dport);
1871 srcp = ntohs(tw->tw_sport);
1872
1873 seq_printf(seq,
1874 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1875 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1876 i,
1877 src->s6_addr32[0], src->s6_addr32[1],
1878 src->s6_addr32[2], src->s6_addr32[3], srcp,
1879 dest->s6_addr32[0], dest->s6_addr32[1],
1880 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1881 tw->tw_substate, 0, 0,
1882 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1883 refcount_read(&tw->tw_refcnt), tw);
1884}
1885
1886static int tcp6_seq_show(struct seq_file *seq, void *v)
1887{
1888 struct tcp_iter_state *st;
1889 struct sock *sk = v;
1890
1891 if (v == SEQ_START_TOKEN) {
1892 seq_puts(seq,
1893 " sl "
1894 "local_address "
1895 "remote_address "
1896 "st tx_queue rx_queue tr tm->when retrnsmt"
1897 " uid timeout inode\n");
1898 goto out;
1899 }
1900 st = seq->private;
1901
1902 if (sk->sk_state == TCP_TIME_WAIT)
1903 get_timewait6_sock(seq, v, st->num);
1904 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1905 get_openreq6(seq, v, st->num);
1906 else
1907 get_tcp6_sock(seq, v, st->num);
1908out:
1909 return 0;
1910}
1911
1912static const struct file_operations tcp6_afinfo_seq_fops = {
1913 .open = tcp_seq_open,
1914 .read = seq_read,
1915 .llseek = seq_lseek,
1916 .release = seq_release_net
1917};
1918
1919static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1920 .name = "tcp6",
1921 .family = AF_INET6,
1922 .seq_fops = &tcp6_afinfo_seq_fops,
1923 .seq_ops = {
1924 .show = tcp6_seq_show,
1925 },
1926};
1927
1928int __net_init tcp6_proc_init(struct net *net)
1929{
1930 return tcp_proc_register(net, &tcp6_seq_afinfo);
1931}
1932
1933void tcp6_proc_exit(struct net *net)
1934{
1935 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1936}
1937#endif
1938
1939struct proto tcpv6_prot = {
1940 .name = "TCPv6",
1941 .owner = THIS_MODULE,
1942 .close = tcp_close,
1943 .pre_connect = tcp_v6_pre_connect,
1944 .connect = tcp_v6_connect,
1945 .disconnect = tcp_disconnect,
1946 .accept = inet_csk_accept,
1947 .ioctl = tcp_ioctl,
1948 .init = tcp_v6_init_sock,
1949 .destroy = tcp_v6_destroy_sock,
1950 .shutdown = tcp_shutdown,
1951 .setsockopt = tcp_setsockopt,
1952 .getsockopt = tcp_getsockopt,
1953 .keepalive = tcp_set_keepalive,
1954 .recvmsg = tcp_recvmsg,
1955 .sendmsg = tcp_sendmsg,
1956 .sendpage = tcp_sendpage,
1957 .backlog_rcv = tcp_v6_do_rcv,
1958 .release_cb = tcp_release_cb,
1959 .hash = inet6_hash,
1960 .unhash = inet_unhash,
1961 .get_port = inet_csk_get_port,
1962 .enter_memory_pressure = tcp_enter_memory_pressure,
1963 .leave_memory_pressure = tcp_leave_memory_pressure,
1964 .stream_memory_free = tcp_stream_memory_free,
1965 .sockets_allocated = &tcp_sockets_allocated,
1966 .memory_allocated = &tcp_memory_allocated,
1967 .memory_pressure = &tcp_memory_pressure,
1968 .orphan_count = &tcp_orphan_count,
1969 .sysctl_mem = sysctl_tcp_mem,
1970 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
1971 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
1972 .max_header = MAX_TCP_HEADER,
1973 .obj_size = sizeof(struct tcp6_sock),
1974 .slab_flags = SLAB_TYPESAFE_BY_RCU,
1975 .twsk_prot = &tcp6_timewait_sock_ops,
1976 .rsk_prot = &tcp6_request_sock_ops,
1977 .h.hashinfo = &tcp_hashinfo,
1978 .no_autobind = true,
1979#ifdef CONFIG_COMPAT
1980 .compat_setsockopt = compat_tcp_setsockopt,
1981 .compat_getsockopt = compat_tcp_getsockopt,
1982#endif
1983 .diag_destroy = tcp_abort,
1984};
1985
1986/* thinking of making this const? Don't.
1987 * early_demux can change based on sysctl.
1988 */
1989static struct inet6_protocol tcpv6_protocol = {
1990 .early_demux = tcp_v6_early_demux,
1991 .early_demux_handler = tcp_v6_early_demux,
1992 .handler = tcp_v6_rcv,
1993 .err_handler = tcp_v6_err,
1994 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1995};
1996
1997static struct inet_protosw tcpv6_protosw = {
1998 .type = SOCK_STREAM,
1999 .protocol = IPPROTO_TCP,
2000 .prot = &tcpv6_prot,
2001 .ops = &inet6_stream_ops,
2002 .flags = INET_PROTOSW_PERMANENT |
2003 INET_PROTOSW_ICSK,
2004};
2005
2006static int __net_init tcpv6_net_init(struct net *net)
2007{
2008 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2009 SOCK_RAW, IPPROTO_TCP, net);
2010}
2011
2012static void __net_exit tcpv6_net_exit(struct net *net)
2013{
2014 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2015}
2016
2017static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2018{
2019 inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2020}
2021
2022static struct pernet_operations tcpv6_net_ops = {
2023 .init = tcpv6_net_init,
2024 .exit = tcpv6_net_exit,
2025 .exit_batch = tcpv6_net_exit_batch,
2026};
2027
2028int __init tcpv6_init(void)
2029{
2030 int ret;
2031
2032 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2033 if (ret)
2034 goto out;
2035
2036 /* register inet6 protocol */
2037 ret = inet6_register_protosw(&tcpv6_protosw);
2038 if (ret)
2039 goto out_tcpv6_protocol;
2040
2041 ret = register_pernet_subsys(&tcpv6_net_ops);
2042 if (ret)
2043 goto out_tcpv6_protosw;
2044out:
2045 return ret;
2046
2047out_tcpv6_protosw:
2048 inet6_unregister_protosw(&tcpv6_protosw);
2049out_tcpv6_protocol:
2050 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2051 goto out;
2052}
2053
2054void tcpv6_exit(void)
2055{
2056 unregister_pernet_subsys(&tcpv6_net_ops);
2057 inet6_unregister_protosw(&tcpv6_protosw);
2058 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2059}