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// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * TCP over IPv6
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
8 *
9 * Based on:
10 * linux/net/ipv4/tcp.c
11 * linux/net/ipv4/tcp_input.c
12 * linux/net/ipv4/tcp_output.c
13 *
14 * Fixes:
15 * Hideaki YOSHIFUJI : sin6_scope_id support
16 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
17 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
18 * a single port at the same time.
19 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
20 */
21
22#include <linux/bottom_half.h>
23#include <linux/module.h>
24#include <linux/errno.h>
25#include <linux/types.h>
26#include <linux/socket.h>
27#include <linux/sockios.h>
28#include <linux/net.h>
29#include <linux/jiffies.h>
30#include <linux/in.h>
31#include <linux/in6.h>
32#include <linux/netdevice.h>
33#include <linux/init.h>
34#include <linux/jhash.h>
35#include <linux/ipsec.h>
36#include <linux/times.h>
37#include <linux/slab.h>
38#include <linux/uaccess.h>
39#include <linux/ipv6.h>
40#include <linux/icmpv6.h>
41#include <linux/random.h>
42#include <linux/indirect_call_wrapper.h>
43
44#include <net/tcp.h>
45#include <net/ndisc.h>
46#include <net/inet6_hashtables.h>
47#include <net/inet6_connection_sock.h>
48#include <net/ipv6.h>
49#include <net/transp_v6.h>
50#include <net/addrconf.h>
51#include <net/ip6_route.h>
52#include <net/ip6_checksum.h>
53#include <net/inet_ecn.h>
54#include <net/protocol.h>
55#include <net/xfrm.h>
56#include <net/snmp.h>
57#include <net/dsfield.h>
58#include <net/timewait_sock.h>
59#include <net/inet_common.h>
60#include <net/secure_seq.h>
61#include <net/busy_poll.h>
62
63#include <linux/proc_fs.h>
64#include <linux/seq_file.h>
65
66#include <crypto/hash.h>
67#include <linux/scatterlist.h>
68
69#include <trace/events/tcp.h>
70
71static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
72static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
73 struct request_sock *req);
74
75static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
76
77static const struct inet_connection_sock_af_ops ipv6_mapped;
78static const struct inet_connection_sock_af_ops ipv6_specific;
79#ifdef CONFIG_TCP_MD5SIG
80static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
81static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
82#else
83static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
84 const struct in6_addr *addr)
85{
86 return NULL;
87}
88#endif
89
90/* Helper returning the inet6 address from a given tcp socket.
91 * It can be used in TCP stack instead of inet6_sk(sk).
92 * This avoids a dereference and allow compiler optimizations.
93 * It is a specialized version of inet6_sk_generic().
94 */
95static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk)
96{
97 unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo);
98
99 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
100}
101
102static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
103{
104 struct dst_entry *dst = skb_dst(skb);
105
106 if (dst && dst_hold_safe(dst)) {
107 const struct rt6_info *rt = (const struct rt6_info *)dst;
108
109 sk->sk_rx_dst = dst;
110 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
111 tcp_inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
112 }
113}
114
115static u32 tcp_v6_init_seq(const struct sk_buff *skb)
116{
117 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
118 ipv6_hdr(skb)->saddr.s6_addr32,
119 tcp_hdr(skb)->dest,
120 tcp_hdr(skb)->source);
121}
122
123static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
124{
125 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
126 ipv6_hdr(skb)->saddr.s6_addr32);
127}
128
129static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
130 int addr_len)
131{
132 /* This check is replicated from tcp_v6_connect() and intended to
133 * prevent BPF program called below from accessing bytes that are out
134 * of the bound specified by user in addr_len.
135 */
136 if (addr_len < SIN6_LEN_RFC2133)
137 return -EINVAL;
138
139 sock_owned_by_me(sk);
140
141 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr);
142}
143
144static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
145 int addr_len)
146{
147 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
148 struct inet_sock *inet = inet_sk(sk);
149 struct inet_connection_sock *icsk = inet_csk(sk);
150 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
151 struct tcp_sock *tp = tcp_sk(sk);
152 struct in6_addr *saddr = NULL, *final_p, final;
153 struct ipv6_txoptions *opt;
154 struct flowi6 fl6;
155 struct dst_entry *dst;
156 int addr_type;
157 int err;
158 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
159
160 if (addr_len < SIN6_LEN_RFC2133)
161 return -EINVAL;
162
163 if (usin->sin6_family != AF_INET6)
164 return -EAFNOSUPPORT;
165
166 memset(&fl6, 0, sizeof(fl6));
167
168 if (np->sndflow) {
169 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
170 IP6_ECN_flow_init(fl6.flowlabel);
171 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
172 struct ip6_flowlabel *flowlabel;
173 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
174 if (IS_ERR(flowlabel))
175 return -EINVAL;
176 fl6_sock_release(flowlabel);
177 }
178 }
179
180 /*
181 * connect() to INADDR_ANY means loopback (BSD'ism).
182 */
183
184 if (ipv6_addr_any(&usin->sin6_addr)) {
185 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
186 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
187 &usin->sin6_addr);
188 else
189 usin->sin6_addr = in6addr_loopback;
190 }
191
192 addr_type = ipv6_addr_type(&usin->sin6_addr);
193
194 if (addr_type & IPV6_ADDR_MULTICAST)
195 return -ENETUNREACH;
196
197 if (addr_type&IPV6_ADDR_LINKLOCAL) {
198 if (addr_len >= sizeof(struct sockaddr_in6) &&
199 usin->sin6_scope_id) {
200 /* If interface is set while binding, indices
201 * must coincide.
202 */
203 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
204 return -EINVAL;
205
206 sk->sk_bound_dev_if = usin->sin6_scope_id;
207 }
208
209 /* Connect to link-local address requires an interface */
210 if (!sk->sk_bound_dev_if)
211 return -EINVAL;
212 }
213
214 if (tp->rx_opt.ts_recent_stamp &&
215 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
216 tp->rx_opt.ts_recent = 0;
217 tp->rx_opt.ts_recent_stamp = 0;
218 WRITE_ONCE(tp->write_seq, 0);
219 }
220
221 sk->sk_v6_daddr = usin->sin6_addr;
222 np->flow_label = fl6.flowlabel;
223
224 /*
225 * TCP over IPv4
226 */
227
228 if (addr_type & IPV6_ADDR_MAPPED) {
229 u32 exthdrlen = icsk->icsk_ext_hdr_len;
230 struct sockaddr_in sin;
231
232 if (__ipv6_only_sock(sk))
233 return -ENETUNREACH;
234
235 sin.sin_family = AF_INET;
236 sin.sin_port = usin->sin6_port;
237 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
238
239 icsk->icsk_af_ops = &ipv6_mapped;
240 sk->sk_backlog_rcv = tcp_v4_do_rcv;
241#ifdef CONFIG_TCP_MD5SIG
242 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
243#endif
244
245 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
246
247 if (err) {
248 icsk->icsk_ext_hdr_len = exthdrlen;
249 icsk->icsk_af_ops = &ipv6_specific;
250 sk->sk_backlog_rcv = tcp_v6_do_rcv;
251#ifdef CONFIG_TCP_MD5SIG
252 tp->af_specific = &tcp_sock_ipv6_specific;
253#endif
254 goto failure;
255 }
256 np->saddr = sk->sk_v6_rcv_saddr;
257
258 return err;
259 }
260
261 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
262 saddr = &sk->sk_v6_rcv_saddr;
263
264 fl6.flowi6_proto = IPPROTO_TCP;
265 fl6.daddr = sk->sk_v6_daddr;
266 fl6.saddr = saddr ? *saddr : np->saddr;
267 fl6.flowi6_oif = sk->sk_bound_dev_if;
268 fl6.flowi6_mark = sk->sk_mark;
269 fl6.fl6_dport = usin->sin6_port;
270 fl6.fl6_sport = inet->inet_sport;
271 fl6.flowi6_uid = sk->sk_uid;
272
273 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
274 final_p = fl6_update_dst(&fl6, opt, &final);
275
276 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
277
278 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
279 if (IS_ERR(dst)) {
280 err = PTR_ERR(dst);
281 goto failure;
282 }
283
284 if (!saddr) {
285 saddr = &fl6.saddr;
286 sk->sk_v6_rcv_saddr = *saddr;
287 }
288
289 /* set the source address */
290 np->saddr = *saddr;
291 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
292
293 sk->sk_gso_type = SKB_GSO_TCPV6;
294 ip6_dst_store(sk, dst, NULL, NULL);
295
296 icsk->icsk_ext_hdr_len = 0;
297 if (opt)
298 icsk->icsk_ext_hdr_len = opt->opt_flen +
299 opt->opt_nflen;
300
301 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
302
303 inet->inet_dport = usin->sin6_port;
304
305 tcp_set_state(sk, TCP_SYN_SENT);
306 err = inet6_hash_connect(tcp_death_row, sk);
307 if (err)
308 goto late_failure;
309
310 sk_set_txhash(sk);
311
312 if (likely(!tp->repair)) {
313 if (!tp->write_seq)
314 WRITE_ONCE(tp->write_seq,
315 secure_tcpv6_seq(np->saddr.s6_addr32,
316 sk->sk_v6_daddr.s6_addr32,
317 inet->inet_sport,
318 inet->inet_dport));
319 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
320 np->saddr.s6_addr32,
321 sk->sk_v6_daddr.s6_addr32);
322 }
323
324 if (tcp_fastopen_defer_connect(sk, &err))
325 return err;
326 if (err)
327 goto late_failure;
328
329 err = tcp_connect(sk);
330 if (err)
331 goto late_failure;
332
333 return 0;
334
335late_failure:
336 tcp_set_state(sk, TCP_CLOSE);
337failure:
338 inet->inet_dport = 0;
339 sk->sk_route_caps = 0;
340 return err;
341}
342
343static void tcp_v6_mtu_reduced(struct sock *sk)
344{
345 struct dst_entry *dst;
346
347 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
348 return;
349
350 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
351 if (!dst)
352 return;
353
354 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
355 tcp_sync_mss(sk, dst_mtu(dst));
356 tcp_simple_retransmit(sk);
357 }
358}
359
360static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
361 u8 type, u8 code, int offset, __be32 info)
362{
363 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
364 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
365 struct net *net = dev_net(skb->dev);
366 struct request_sock *fastopen;
367 struct ipv6_pinfo *np;
368 struct tcp_sock *tp;
369 __u32 seq, snd_una;
370 struct sock *sk;
371 bool fatal;
372 int err;
373
374 sk = __inet6_lookup_established(net, &tcp_hashinfo,
375 &hdr->daddr, th->dest,
376 &hdr->saddr, ntohs(th->source),
377 skb->dev->ifindex, inet6_sdif(skb));
378
379 if (!sk) {
380 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
381 ICMP6_MIB_INERRORS);
382 return -ENOENT;
383 }
384
385 if (sk->sk_state == TCP_TIME_WAIT) {
386 inet_twsk_put(inet_twsk(sk));
387 return 0;
388 }
389 seq = ntohl(th->seq);
390 fatal = icmpv6_err_convert(type, code, &err);
391 if (sk->sk_state == TCP_NEW_SYN_RECV) {
392 tcp_req_err(sk, seq, fatal);
393 return 0;
394 }
395
396 bh_lock_sock(sk);
397 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
398 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
399
400 if (sk->sk_state == TCP_CLOSE)
401 goto out;
402
403 if (ipv6_hdr(skb)->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
404 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
405 goto out;
406 }
407
408 tp = tcp_sk(sk);
409 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
410 fastopen = rcu_dereference(tp->fastopen_rsk);
411 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
412 if (sk->sk_state != TCP_LISTEN &&
413 !between(seq, snd_una, tp->snd_nxt)) {
414 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
415 goto out;
416 }
417
418 np = tcp_inet6_sk(sk);
419
420 if (type == NDISC_REDIRECT) {
421 if (!sock_owned_by_user(sk)) {
422 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
423
424 if (dst)
425 dst->ops->redirect(dst, sk, skb);
426 }
427 goto out;
428 }
429
430 if (type == ICMPV6_PKT_TOOBIG) {
431 /* We are not interested in TCP_LISTEN and open_requests
432 * (SYN-ACKs send out by Linux are always <576bytes so
433 * they should go through unfragmented).
434 */
435 if (sk->sk_state == TCP_LISTEN)
436 goto out;
437
438 if (!ip6_sk_accept_pmtu(sk))
439 goto out;
440
441 tp->mtu_info = ntohl(info);
442 if (!sock_owned_by_user(sk))
443 tcp_v6_mtu_reduced(sk);
444 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
445 &sk->sk_tsq_flags))
446 sock_hold(sk);
447 goto out;
448 }
449
450
451 /* Might be for an request_sock */
452 switch (sk->sk_state) {
453 case TCP_SYN_SENT:
454 case TCP_SYN_RECV:
455 /* Only in fast or simultaneous open. If a fast open socket is
456 * is already accepted it is treated as a connected one below.
457 */
458 if (fastopen && !fastopen->sk)
459 break;
460
461 if (!sock_owned_by_user(sk)) {
462 sk->sk_err = err;
463 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
464
465 tcp_done(sk);
466 } else
467 sk->sk_err_soft = err;
468 goto out;
469 }
470
471 if (!sock_owned_by_user(sk) && np->recverr) {
472 sk->sk_err = err;
473 sk->sk_error_report(sk);
474 } else
475 sk->sk_err_soft = err;
476
477out:
478 bh_unlock_sock(sk);
479 sock_put(sk);
480 return 0;
481}
482
483
484static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
485 struct flowi *fl,
486 struct request_sock *req,
487 struct tcp_fastopen_cookie *foc,
488 enum tcp_synack_type synack_type)
489{
490 struct inet_request_sock *ireq = inet_rsk(req);
491 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
492 struct ipv6_txoptions *opt;
493 struct flowi6 *fl6 = &fl->u.ip6;
494 struct sk_buff *skb;
495 int err = -ENOMEM;
496
497 /* First, grab a route. */
498 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
499 IPPROTO_TCP)) == NULL)
500 goto done;
501
502 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
503
504 if (skb) {
505 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
506 &ireq->ir_v6_rmt_addr);
507
508 fl6->daddr = ireq->ir_v6_rmt_addr;
509 if (np->repflow && ireq->pktopts)
510 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
511
512 rcu_read_lock();
513 opt = ireq->ipv6_opt;
514 if (!opt)
515 opt = rcu_dereference(np->opt);
516 err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass,
517 sk->sk_priority);
518 rcu_read_unlock();
519 err = net_xmit_eval(err);
520 }
521
522done:
523 return err;
524}
525
526
527static void tcp_v6_reqsk_destructor(struct request_sock *req)
528{
529 kfree(inet_rsk(req)->ipv6_opt);
530 kfree_skb(inet_rsk(req)->pktopts);
531}
532
533#ifdef CONFIG_TCP_MD5SIG
534static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
535 const struct in6_addr *addr)
536{
537 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
538}
539
540static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
541 const struct sock *addr_sk)
542{
543 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
544}
545
546static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
547 char __user *optval, int optlen)
548{
549 struct tcp_md5sig cmd;
550 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
551 u8 prefixlen;
552
553 if (optlen < sizeof(cmd))
554 return -EINVAL;
555
556 if (copy_from_user(&cmd, optval, sizeof(cmd)))
557 return -EFAULT;
558
559 if (sin6->sin6_family != AF_INET6)
560 return -EINVAL;
561
562 if (optname == TCP_MD5SIG_EXT &&
563 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
564 prefixlen = cmd.tcpm_prefixlen;
565 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
566 prefixlen > 32))
567 return -EINVAL;
568 } else {
569 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
570 }
571
572 if (!cmd.tcpm_keylen) {
573 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
574 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
575 AF_INET, prefixlen);
576 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
577 AF_INET6, prefixlen);
578 }
579
580 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
581 return -EINVAL;
582
583 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
584 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
585 AF_INET, prefixlen, cmd.tcpm_key,
586 cmd.tcpm_keylen, GFP_KERNEL);
587
588 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
589 AF_INET6, prefixlen, cmd.tcpm_key,
590 cmd.tcpm_keylen, GFP_KERNEL);
591}
592
593static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
594 const struct in6_addr *daddr,
595 const struct in6_addr *saddr,
596 const struct tcphdr *th, int nbytes)
597{
598 struct tcp6_pseudohdr *bp;
599 struct scatterlist sg;
600 struct tcphdr *_th;
601
602 bp = hp->scratch;
603 /* 1. TCP pseudo-header (RFC2460) */
604 bp->saddr = *saddr;
605 bp->daddr = *daddr;
606 bp->protocol = cpu_to_be32(IPPROTO_TCP);
607 bp->len = cpu_to_be32(nbytes);
608
609 _th = (struct tcphdr *)(bp + 1);
610 memcpy(_th, th, sizeof(*th));
611 _th->check = 0;
612
613 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
614 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
615 sizeof(*bp) + sizeof(*th));
616 return crypto_ahash_update(hp->md5_req);
617}
618
619static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
620 const struct in6_addr *daddr, struct in6_addr *saddr,
621 const struct tcphdr *th)
622{
623 struct tcp_md5sig_pool *hp;
624 struct ahash_request *req;
625
626 hp = tcp_get_md5sig_pool();
627 if (!hp)
628 goto clear_hash_noput;
629 req = hp->md5_req;
630
631 if (crypto_ahash_init(req))
632 goto clear_hash;
633 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
634 goto clear_hash;
635 if (tcp_md5_hash_key(hp, key))
636 goto clear_hash;
637 ahash_request_set_crypt(req, NULL, md5_hash, 0);
638 if (crypto_ahash_final(req))
639 goto clear_hash;
640
641 tcp_put_md5sig_pool();
642 return 0;
643
644clear_hash:
645 tcp_put_md5sig_pool();
646clear_hash_noput:
647 memset(md5_hash, 0, 16);
648 return 1;
649}
650
651static int tcp_v6_md5_hash_skb(char *md5_hash,
652 const struct tcp_md5sig_key *key,
653 const struct sock *sk,
654 const struct sk_buff *skb)
655{
656 const struct in6_addr *saddr, *daddr;
657 struct tcp_md5sig_pool *hp;
658 struct ahash_request *req;
659 const struct tcphdr *th = tcp_hdr(skb);
660
661 if (sk) { /* valid for establish/request sockets */
662 saddr = &sk->sk_v6_rcv_saddr;
663 daddr = &sk->sk_v6_daddr;
664 } else {
665 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
666 saddr = &ip6h->saddr;
667 daddr = &ip6h->daddr;
668 }
669
670 hp = tcp_get_md5sig_pool();
671 if (!hp)
672 goto clear_hash_noput;
673 req = hp->md5_req;
674
675 if (crypto_ahash_init(req))
676 goto clear_hash;
677
678 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
679 goto clear_hash;
680 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
681 goto clear_hash;
682 if (tcp_md5_hash_key(hp, key))
683 goto clear_hash;
684 ahash_request_set_crypt(req, NULL, md5_hash, 0);
685 if (crypto_ahash_final(req))
686 goto clear_hash;
687
688 tcp_put_md5sig_pool();
689 return 0;
690
691clear_hash:
692 tcp_put_md5sig_pool();
693clear_hash_noput:
694 memset(md5_hash, 0, 16);
695 return 1;
696}
697
698#endif
699
700static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
701 const struct sk_buff *skb)
702{
703#ifdef CONFIG_TCP_MD5SIG
704 const __u8 *hash_location = NULL;
705 struct tcp_md5sig_key *hash_expected;
706 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
707 const struct tcphdr *th = tcp_hdr(skb);
708 int genhash;
709 u8 newhash[16];
710
711 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
712 hash_location = tcp_parse_md5sig_option(th);
713
714 /* We've parsed the options - do we have a hash? */
715 if (!hash_expected && !hash_location)
716 return false;
717
718 if (hash_expected && !hash_location) {
719 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
720 return true;
721 }
722
723 if (!hash_expected && hash_location) {
724 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
725 return true;
726 }
727
728 /* check the signature */
729 genhash = tcp_v6_md5_hash_skb(newhash,
730 hash_expected,
731 NULL, skb);
732
733 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
734 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
735 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
736 genhash ? "failed" : "mismatch",
737 &ip6h->saddr, ntohs(th->source),
738 &ip6h->daddr, ntohs(th->dest));
739 return true;
740 }
741#endif
742 return false;
743}
744
745static void tcp_v6_init_req(struct request_sock *req,
746 const struct sock *sk_listener,
747 struct sk_buff *skb)
748{
749 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
750 struct inet_request_sock *ireq = inet_rsk(req);
751 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
752
753 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
754 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
755
756 /* So that link locals have meaning */
757 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
758 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
759 ireq->ir_iif = tcp_v6_iif(skb);
760
761 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
762 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
763 np->rxopt.bits.rxinfo ||
764 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
765 np->rxopt.bits.rxohlim || np->repflow)) {
766 refcount_inc(&skb->users);
767 ireq->pktopts = skb;
768 }
769}
770
771static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
772 struct flowi *fl,
773 const struct request_sock *req)
774{
775 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
776}
777
778struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
779 .family = AF_INET6,
780 .obj_size = sizeof(struct tcp6_request_sock),
781 .rtx_syn_ack = tcp_rtx_synack,
782 .send_ack = tcp_v6_reqsk_send_ack,
783 .destructor = tcp_v6_reqsk_destructor,
784 .send_reset = tcp_v6_send_reset,
785 .syn_ack_timeout = tcp_syn_ack_timeout,
786};
787
788static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
789 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
790 sizeof(struct ipv6hdr),
791#ifdef CONFIG_TCP_MD5SIG
792 .req_md5_lookup = tcp_v6_md5_lookup,
793 .calc_md5_hash = tcp_v6_md5_hash_skb,
794#endif
795 .init_req = tcp_v6_init_req,
796#ifdef CONFIG_SYN_COOKIES
797 .cookie_init_seq = cookie_v6_init_sequence,
798#endif
799 .route_req = tcp_v6_route_req,
800 .init_seq = tcp_v6_init_seq,
801 .init_ts_off = tcp_v6_init_ts_off,
802 .send_synack = tcp_v6_send_synack,
803};
804
805static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
806 u32 ack, u32 win, u32 tsval, u32 tsecr,
807 int oif, struct tcp_md5sig_key *key, int rst,
808 u8 tclass, __be32 label, u32 priority)
809{
810 const struct tcphdr *th = tcp_hdr(skb);
811 struct tcphdr *t1;
812 struct sk_buff *buff;
813 struct flowi6 fl6;
814 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
815 struct sock *ctl_sk = net->ipv6.tcp_sk;
816 unsigned int tot_len = sizeof(struct tcphdr);
817 struct dst_entry *dst;
818 __be32 *topt;
819 __u32 mark = 0;
820
821 if (tsecr)
822 tot_len += TCPOLEN_TSTAMP_ALIGNED;
823#ifdef CONFIG_TCP_MD5SIG
824 if (key)
825 tot_len += TCPOLEN_MD5SIG_ALIGNED;
826#endif
827
828 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
829 GFP_ATOMIC);
830 if (!buff)
831 return;
832
833 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
834
835 t1 = skb_push(buff, tot_len);
836 skb_reset_transport_header(buff);
837
838 /* Swap the send and the receive. */
839 memset(t1, 0, sizeof(*t1));
840 t1->dest = th->source;
841 t1->source = th->dest;
842 t1->doff = tot_len / 4;
843 t1->seq = htonl(seq);
844 t1->ack_seq = htonl(ack);
845 t1->ack = !rst || !th->ack;
846 t1->rst = rst;
847 t1->window = htons(win);
848
849 topt = (__be32 *)(t1 + 1);
850
851 if (tsecr) {
852 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
853 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
854 *topt++ = htonl(tsval);
855 *topt++ = htonl(tsecr);
856 }
857
858#ifdef CONFIG_TCP_MD5SIG
859 if (key) {
860 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
861 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
862 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
863 &ipv6_hdr(skb)->saddr,
864 &ipv6_hdr(skb)->daddr, t1);
865 }
866#endif
867
868 memset(&fl6, 0, sizeof(fl6));
869 fl6.daddr = ipv6_hdr(skb)->saddr;
870 fl6.saddr = ipv6_hdr(skb)->daddr;
871 fl6.flowlabel = label;
872
873 buff->ip_summed = CHECKSUM_PARTIAL;
874 buff->csum = 0;
875
876 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
877
878 fl6.flowi6_proto = IPPROTO_TCP;
879 if (rt6_need_strict(&fl6.daddr) && !oif)
880 fl6.flowi6_oif = tcp_v6_iif(skb);
881 else {
882 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
883 oif = skb->skb_iif;
884
885 fl6.flowi6_oif = oif;
886 }
887
888 if (sk) {
889 if (sk->sk_state == TCP_TIME_WAIT) {
890 mark = inet_twsk(sk)->tw_mark;
891 /* autoflowlabel relies on buff->hash */
892 skb_set_hash(buff, inet_twsk(sk)->tw_txhash,
893 PKT_HASH_TYPE_L4);
894 } else {
895 mark = sk->sk_mark;
896 }
897 buff->tstamp = tcp_transmit_time(sk);
898 }
899 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
900 fl6.fl6_dport = t1->dest;
901 fl6.fl6_sport = t1->source;
902 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
903 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
904
905 /* Pass a socket to ip6_dst_lookup either it is for RST
906 * Underlying function will use this to retrieve the network
907 * namespace
908 */
909 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
910 if (!IS_ERR(dst)) {
911 skb_dst_set(buff, dst);
912 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, tclass,
913 priority);
914 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
915 if (rst)
916 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
917 return;
918 }
919
920 kfree_skb(buff);
921}
922
923static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
924{
925 const struct tcphdr *th = tcp_hdr(skb);
926 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
927 u32 seq = 0, ack_seq = 0;
928 struct tcp_md5sig_key *key = NULL;
929#ifdef CONFIG_TCP_MD5SIG
930 const __u8 *hash_location = NULL;
931 unsigned char newhash[16];
932 int genhash;
933 struct sock *sk1 = NULL;
934#endif
935 __be32 label = 0;
936 u32 priority = 0;
937 struct net *net;
938 int oif = 0;
939
940 if (th->rst)
941 return;
942
943 /* If sk not NULL, it means we did a successful lookup and incoming
944 * route had to be correct. prequeue might have dropped our dst.
945 */
946 if (!sk && !ipv6_unicast_destination(skb))
947 return;
948
949 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
950#ifdef CONFIG_TCP_MD5SIG
951 rcu_read_lock();
952 hash_location = tcp_parse_md5sig_option(th);
953 if (sk && sk_fullsock(sk)) {
954 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
955 } else if (hash_location) {
956 /*
957 * active side is lost. Try to find listening socket through
958 * source port, and then find md5 key through listening socket.
959 * we are not loose security here:
960 * Incoming packet is checked with md5 hash with finding key,
961 * no RST generated if md5 hash doesn't match.
962 */
963 sk1 = inet6_lookup_listener(net,
964 &tcp_hashinfo, NULL, 0,
965 &ipv6h->saddr,
966 th->source, &ipv6h->daddr,
967 ntohs(th->source),
968 tcp_v6_iif_l3_slave(skb),
969 tcp_v6_sdif(skb));
970 if (!sk1)
971 goto out;
972
973 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
974 if (!key)
975 goto out;
976
977 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
978 if (genhash || memcmp(hash_location, newhash, 16) != 0)
979 goto out;
980 }
981#endif
982
983 if (th->ack)
984 seq = ntohl(th->ack_seq);
985 else
986 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
987 (th->doff << 2);
988
989 if (sk) {
990 oif = sk->sk_bound_dev_if;
991 if (sk_fullsock(sk)) {
992 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
993
994 trace_tcp_send_reset(sk, skb);
995 if (np->repflow)
996 label = ip6_flowlabel(ipv6h);
997 priority = sk->sk_priority;
998 }
999 if (sk->sk_state == TCP_TIME_WAIT) {
1000 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1001 priority = inet_twsk(sk)->tw_priority;
1002 }
1003 } else {
1004 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1005 label = ip6_flowlabel(ipv6h);
1006 }
1007
1008 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0,
1009 label, priority);
1010
1011#ifdef CONFIG_TCP_MD5SIG
1012out:
1013 rcu_read_unlock();
1014#endif
1015}
1016
1017static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1018 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1019 struct tcp_md5sig_key *key, u8 tclass,
1020 __be32 label, u32 priority)
1021{
1022 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1023 tclass, label, priority);
1024}
1025
1026static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1027{
1028 struct inet_timewait_sock *tw = inet_twsk(sk);
1029 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1030
1031 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1032 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1033 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1034 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1035 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority);
1036
1037 inet_twsk_put(tw);
1038}
1039
1040static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1041 struct request_sock *req)
1042{
1043 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1044 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1045 */
1046 /* RFC 7323 2.3
1047 * The window field (SEG.WND) of every outgoing segment, with the
1048 * exception of <SYN> segments, MUST be right-shifted by
1049 * Rcv.Wind.Shift bits:
1050 */
1051 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1052 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1053 tcp_rsk(req)->rcv_nxt,
1054 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1055 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1056 req->ts_recent, sk->sk_bound_dev_if,
1057 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr),
1058 0, 0, sk->sk_priority);
1059}
1060
1061
1062static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1063{
1064#ifdef CONFIG_SYN_COOKIES
1065 const struct tcphdr *th = tcp_hdr(skb);
1066
1067 if (!th->syn)
1068 sk = cookie_v6_check(sk, skb);
1069#endif
1070 return sk;
1071}
1072
1073u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1074 struct tcphdr *th, u32 *cookie)
1075{
1076 u16 mss = 0;
1077#ifdef CONFIG_SYN_COOKIES
1078 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1079 &tcp_request_sock_ipv6_ops, sk, th);
1080 if (mss) {
1081 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1082 tcp_synq_overflow(sk);
1083 }
1084#endif
1085 return mss;
1086}
1087
1088static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1089{
1090 if (skb->protocol == htons(ETH_P_IP))
1091 return tcp_v4_conn_request(sk, skb);
1092
1093 if (!ipv6_unicast_destination(skb))
1094 goto drop;
1095
1096 return tcp_conn_request(&tcp6_request_sock_ops,
1097 &tcp_request_sock_ipv6_ops, sk, skb);
1098
1099drop:
1100 tcp_listendrop(sk);
1101 return 0; /* don't send reset */
1102}
1103
1104static void tcp_v6_restore_cb(struct sk_buff *skb)
1105{
1106 /* We need to move header back to the beginning if xfrm6_policy_check()
1107 * and tcp_v6_fill_cb() are going to be called again.
1108 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1109 */
1110 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1111 sizeof(struct inet6_skb_parm));
1112}
1113
1114static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1115 struct request_sock *req,
1116 struct dst_entry *dst,
1117 struct request_sock *req_unhash,
1118 bool *own_req)
1119{
1120 struct inet_request_sock *ireq;
1121 struct ipv6_pinfo *newnp;
1122 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1123 struct ipv6_txoptions *opt;
1124 struct inet_sock *newinet;
1125 struct tcp_sock *newtp;
1126 struct sock *newsk;
1127#ifdef CONFIG_TCP_MD5SIG
1128 struct tcp_md5sig_key *key;
1129#endif
1130 struct flowi6 fl6;
1131
1132 if (skb->protocol == htons(ETH_P_IP)) {
1133 /*
1134 * v6 mapped
1135 */
1136
1137 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1138 req_unhash, own_req);
1139
1140 if (!newsk)
1141 return NULL;
1142
1143 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1144
1145 newinet = inet_sk(newsk);
1146 newnp = tcp_inet6_sk(newsk);
1147 newtp = tcp_sk(newsk);
1148
1149 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1150
1151 newnp->saddr = newsk->sk_v6_rcv_saddr;
1152
1153 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1154 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1155#ifdef CONFIG_TCP_MD5SIG
1156 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1157#endif
1158
1159 newnp->ipv6_mc_list = NULL;
1160 newnp->ipv6_ac_list = NULL;
1161 newnp->ipv6_fl_list = NULL;
1162 newnp->pktoptions = NULL;
1163 newnp->opt = NULL;
1164 newnp->mcast_oif = inet_iif(skb);
1165 newnp->mcast_hops = ip_hdr(skb)->ttl;
1166 newnp->rcv_flowinfo = 0;
1167 if (np->repflow)
1168 newnp->flow_label = 0;
1169
1170 /*
1171 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1172 * here, tcp_create_openreq_child now does this for us, see the comment in
1173 * that function for the gory details. -acme
1174 */
1175
1176 /* It is tricky place. Until this moment IPv4 tcp
1177 worked with IPv6 icsk.icsk_af_ops.
1178 Sync it now.
1179 */
1180 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1181
1182 return newsk;
1183 }
1184
1185 ireq = inet_rsk(req);
1186
1187 if (sk_acceptq_is_full(sk))
1188 goto out_overflow;
1189
1190 if (!dst) {
1191 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1192 if (!dst)
1193 goto out;
1194 }
1195
1196 newsk = tcp_create_openreq_child(sk, req, skb);
1197 if (!newsk)
1198 goto out_nonewsk;
1199
1200 /*
1201 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1202 * count here, tcp_create_openreq_child now does this for us, see the
1203 * comment in that function for the gory details. -acme
1204 */
1205
1206 newsk->sk_gso_type = SKB_GSO_TCPV6;
1207 ip6_dst_store(newsk, dst, NULL, NULL);
1208 inet6_sk_rx_dst_set(newsk, skb);
1209
1210 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1211
1212 newtp = tcp_sk(newsk);
1213 newinet = inet_sk(newsk);
1214 newnp = tcp_inet6_sk(newsk);
1215
1216 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1217
1218 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1219 newnp->saddr = ireq->ir_v6_loc_addr;
1220 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1221 newsk->sk_bound_dev_if = ireq->ir_iif;
1222
1223 /* Now IPv6 options...
1224
1225 First: no IPv4 options.
1226 */
1227 newinet->inet_opt = NULL;
1228 newnp->ipv6_mc_list = NULL;
1229 newnp->ipv6_ac_list = NULL;
1230 newnp->ipv6_fl_list = NULL;
1231
1232 /* Clone RX bits */
1233 newnp->rxopt.all = np->rxopt.all;
1234
1235 newnp->pktoptions = NULL;
1236 newnp->opt = NULL;
1237 newnp->mcast_oif = tcp_v6_iif(skb);
1238 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1239 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1240 if (np->repflow)
1241 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1242
1243 /* Clone native IPv6 options from listening socket (if any)
1244
1245 Yes, keeping reference count would be much more clever,
1246 but we make one more one thing there: reattach optmem
1247 to newsk.
1248 */
1249 opt = ireq->ipv6_opt;
1250 if (!opt)
1251 opt = rcu_dereference(np->opt);
1252 if (opt) {
1253 opt = ipv6_dup_options(newsk, opt);
1254 RCU_INIT_POINTER(newnp->opt, opt);
1255 }
1256 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1257 if (opt)
1258 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1259 opt->opt_flen;
1260
1261 tcp_ca_openreq_child(newsk, dst);
1262
1263 tcp_sync_mss(newsk, dst_mtu(dst));
1264 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1265
1266 tcp_initialize_rcv_mss(newsk);
1267
1268 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1269 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1270
1271#ifdef CONFIG_TCP_MD5SIG
1272 /* Copy over the MD5 key from the original socket */
1273 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1274 if (key) {
1275 /* We're using one, so create a matching key
1276 * on the newsk structure. If we fail to get
1277 * memory, then we end up not copying the key
1278 * across. Shucks.
1279 */
1280 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1281 AF_INET6, 128, key->key, key->keylen,
1282 sk_gfp_mask(sk, GFP_ATOMIC));
1283 }
1284#endif
1285
1286 if (__inet_inherit_port(sk, newsk) < 0) {
1287 inet_csk_prepare_forced_close(newsk);
1288 tcp_done(newsk);
1289 goto out;
1290 }
1291 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1292 if (*own_req) {
1293 tcp_move_syn(newtp, req);
1294
1295 /* Clone pktoptions received with SYN, if we own the req */
1296 if (ireq->pktopts) {
1297 newnp->pktoptions = skb_clone(ireq->pktopts,
1298 sk_gfp_mask(sk, GFP_ATOMIC));
1299 consume_skb(ireq->pktopts);
1300 ireq->pktopts = NULL;
1301 if (newnp->pktoptions) {
1302 tcp_v6_restore_cb(newnp->pktoptions);
1303 skb_set_owner_r(newnp->pktoptions, newsk);
1304 }
1305 }
1306 }
1307
1308 return newsk;
1309
1310out_overflow:
1311 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1312out_nonewsk:
1313 dst_release(dst);
1314out:
1315 tcp_listendrop(sk);
1316 return NULL;
1317}
1318
1319/* The socket must have it's spinlock held when we get
1320 * here, unless it is a TCP_LISTEN socket.
1321 *
1322 * We have a potential double-lock case here, so even when
1323 * doing backlog processing we use the BH locking scheme.
1324 * This is because we cannot sleep with the original spinlock
1325 * held.
1326 */
1327static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1328{
1329 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1330 struct sk_buff *opt_skb = NULL;
1331 struct tcp_sock *tp;
1332
1333 /* Imagine: socket is IPv6. IPv4 packet arrives,
1334 goes to IPv4 receive handler and backlogged.
1335 From backlog it always goes here. Kerboom...
1336 Fortunately, tcp_rcv_established and rcv_established
1337 handle them correctly, but it is not case with
1338 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1339 */
1340
1341 if (skb->protocol == htons(ETH_P_IP))
1342 return tcp_v4_do_rcv(sk, skb);
1343
1344 /*
1345 * socket locking is here for SMP purposes as backlog rcv
1346 * is currently called with bh processing disabled.
1347 */
1348
1349 /* Do Stevens' IPV6_PKTOPTIONS.
1350
1351 Yes, guys, it is the only place in our code, where we
1352 may make it not affecting IPv4.
1353 The rest of code is protocol independent,
1354 and I do not like idea to uglify IPv4.
1355
1356 Actually, all the idea behind IPV6_PKTOPTIONS
1357 looks not very well thought. For now we latch
1358 options, received in the last packet, enqueued
1359 by tcp. Feel free to propose better solution.
1360 --ANK (980728)
1361 */
1362 if (np->rxopt.all)
1363 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1364
1365 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1366 struct dst_entry *dst = sk->sk_rx_dst;
1367
1368 sock_rps_save_rxhash(sk, skb);
1369 sk_mark_napi_id(sk, skb);
1370 if (dst) {
1371 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1372 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1373 dst_release(dst);
1374 sk->sk_rx_dst = NULL;
1375 }
1376 }
1377
1378 tcp_rcv_established(sk, skb);
1379 if (opt_skb)
1380 goto ipv6_pktoptions;
1381 return 0;
1382 }
1383
1384 if (tcp_checksum_complete(skb))
1385 goto csum_err;
1386
1387 if (sk->sk_state == TCP_LISTEN) {
1388 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1389
1390 if (!nsk)
1391 goto discard;
1392
1393 if (nsk != sk) {
1394 if (tcp_child_process(sk, nsk, skb))
1395 goto reset;
1396 if (opt_skb)
1397 __kfree_skb(opt_skb);
1398 return 0;
1399 }
1400 } else
1401 sock_rps_save_rxhash(sk, skb);
1402
1403 if (tcp_rcv_state_process(sk, skb))
1404 goto reset;
1405 if (opt_skb)
1406 goto ipv6_pktoptions;
1407 return 0;
1408
1409reset:
1410 tcp_v6_send_reset(sk, skb);
1411discard:
1412 if (opt_skb)
1413 __kfree_skb(opt_skb);
1414 kfree_skb(skb);
1415 return 0;
1416csum_err:
1417 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1418 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1419 goto discard;
1420
1421
1422ipv6_pktoptions:
1423 /* Do you ask, what is it?
1424
1425 1. skb was enqueued by tcp.
1426 2. skb is added to tail of read queue, rather than out of order.
1427 3. socket is not in passive state.
1428 4. Finally, it really contains options, which user wants to receive.
1429 */
1430 tp = tcp_sk(sk);
1431 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1432 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1433 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1434 np->mcast_oif = tcp_v6_iif(opt_skb);
1435 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1436 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1437 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1438 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1439 if (np->repflow)
1440 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1441 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1442 skb_set_owner_r(opt_skb, sk);
1443 tcp_v6_restore_cb(opt_skb);
1444 opt_skb = xchg(&np->pktoptions, opt_skb);
1445 } else {
1446 __kfree_skb(opt_skb);
1447 opt_skb = xchg(&np->pktoptions, NULL);
1448 }
1449 }
1450
1451 kfree_skb(opt_skb);
1452 return 0;
1453}
1454
1455static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1456 const struct tcphdr *th)
1457{
1458 /* This is tricky: we move IP6CB at its correct location into
1459 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1460 * _decode_session6() uses IP6CB().
1461 * barrier() makes sure compiler won't play aliasing games.
1462 */
1463 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1464 sizeof(struct inet6_skb_parm));
1465 barrier();
1466
1467 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1468 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1469 skb->len - th->doff*4);
1470 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1471 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1472 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1473 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1474 TCP_SKB_CB(skb)->sacked = 0;
1475 TCP_SKB_CB(skb)->has_rxtstamp =
1476 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1477}
1478
1479INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1480{
1481 struct sk_buff *skb_to_free;
1482 int sdif = inet6_sdif(skb);
1483 const struct tcphdr *th;
1484 const struct ipv6hdr *hdr;
1485 bool refcounted;
1486 struct sock *sk;
1487 int ret;
1488 struct net *net = dev_net(skb->dev);
1489
1490 if (skb->pkt_type != PACKET_HOST)
1491 goto discard_it;
1492
1493 /*
1494 * Count it even if it's bad.
1495 */
1496 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1497
1498 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1499 goto discard_it;
1500
1501 th = (const struct tcphdr *)skb->data;
1502
1503 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1504 goto bad_packet;
1505 if (!pskb_may_pull(skb, th->doff*4))
1506 goto discard_it;
1507
1508 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1509 goto csum_error;
1510
1511 th = (const struct tcphdr *)skb->data;
1512 hdr = ipv6_hdr(skb);
1513
1514lookup:
1515 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1516 th->source, th->dest, inet6_iif(skb), sdif,
1517 &refcounted);
1518 if (!sk)
1519 goto no_tcp_socket;
1520
1521process:
1522 if (sk->sk_state == TCP_TIME_WAIT)
1523 goto do_time_wait;
1524
1525 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1526 struct request_sock *req = inet_reqsk(sk);
1527 bool req_stolen = false;
1528 struct sock *nsk;
1529
1530 sk = req->rsk_listener;
1531 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1532 sk_drops_add(sk, skb);
1533 reqsk_put(req);
1534 goto discard_it;
1535 }
1536 if (tcp_checksum_complete(skb)) {
1537 reqsk_put(req);
1538 goto csum_error;
1539 }
1540 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1541 inet_csk_reqsk_queue_drop_and_put(sk, req);
1542 goto lookup;
1543 }
1544 sock_hold(sk);
1545 refcounted = true;
1546 nsk = NULL;
1547 if (!tcp_filter(sk, skb)) {
1548 th = (const struct tcphdr *)skb->data;
1549 hdr = ipv6_hdr(skb);
1550 tcp_v6_fill_cb(skb, hdr, th);
1551 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1552 }
1553 if (!nsk) {
1554 reqsk_put(req);
1555 if (req_stolen) {
1556 /* Another cpu got exclusive access to req
1557 * and created a full blown socket.
1558 * Try to feed this packet to this socket
1559 * instead of discarding it.
1560 */
1561 tcp_v6_restore_cb(skb);
1562 sock_put(sk);
1563 goto lookup;
1564 }
1565 goto discard_and_relse;
1566 }
1567 if (nsk == sk) {
1568 reqsk_put(req);
1569 tcp_v6_restore_cb(skb);
1570 } else if (tcp_child_process(sk, nsk, skb)) {
1571 tcp_v6_send_reset(nsk, skb);
1572 goto discard_and_relse;
1573 } else {
1574 sock_put(sk);
1575 return 0;
1576 }
1577 }
1578 if (hdr->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
1579 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1580 goto discard_and_relse;
1581 }
1582
1583 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1584 goto discard_and_relse;
1585
1586 if (tcp_v6_inbound_md5_hash(sk, skb))
1587 goto discard_and_relse;
1588
1589 if (tcp_filter(sk, skb))
1590 goto discard_and_relse;
1591 th = (const struct tcphdr *)skb->data;
1592 hdr = ipv6_hdr(skb);
1593 tcp_v6_fill_cb(skb, hdr, th);
1594
1595 skb->dev = NULL;
1596
1597 if (sk->sk_state == TCP_LISTEN) {
1598 ret = tcp_v6_do_rcv(sk, skb);
1599 goto put_and_return;
1600 }
1601
1602 sk_incoming_cpu_update(sk);
1603
1604 bh_lock_sock_nested(sk);
1605 tcp_segs_in(tcp_sk(sk), skb);
1606 ret = 0;
1607 if (!sock_owned_by_user(sk)) {
1608 skb_to_free = sk->sk_rx_skb_cache;
1609 sk->sk_rx_skb_cache = NULL;
1610 ret = tcp_v6_do_rcv(sk, skb);
1611 } else {
1612 if (tcp_add_backlog(sk, skb))
1613 goto discard_and_relse;
1614 skb_to_free = NULL;
1615 }
1616 bh_unlock_sock(sk);
1617 if (skb_to_free)
1618 __kfree_skb(skb_to_free);
1619put_and_return:
1620 if (refcounted)
1621 sock_put(sk);
1622 return ret ? -1 : 0;
1623
1624no_tcp_socket:
1625 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1626 goto discard_it;
1627
1628 tcp_v6_fill_cb(skb, hdr, th);
1629
1630 if (tcp_checksum_complete(skb)) {
1631csum_error:
1632 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1633bad_packet:
1634 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1635 } else {
1636 tcp_v6_send_reset(NULL, skb);
1637 }
1638
1639discard_it:
1640 kfree_skb(skb);
1641 return 0;
1642
1643discard_and_relse:
1644 sk_drops_add(sk, skb);
1645 if (refcounted)
1646 sock_put(sk);
1647 goto discard_it;
1648
1649do_time_wait:
1650 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1651 inet_twsk_put(inet_twsk(sk));
1652 goto discard_it;
1653 }
1654
1655 tcp_v6_fill_cb(skb, hdr, th);
1656
1657 if (tcp_checksum_complete(skb)) {
1658 inet_twsk_put(inet_twsk(sk));
1659 goto csum_error;
1660 }
1661
1662 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1663 case TCP_TW_SYN:
1664 {
1665 struct sock *sk2;
1666
1667 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1668 skb, __tcp_hdrlen(th),
1669 &ipv6_hdr(skb)->saddr, th->source,
1670 &ipv6_hdr(skb)->daddr,
1671 ntohs(th->dest),
1672 tcp_v6_iif_l3_slave(skb),
1673 sdif);
1674 if (sk2) {
1675 struct inet_timewait_sock *tw = inet_twsk(sk);
1676 inet_twsk_deschedule_put(tw);
1677 sk = sk2;
1678 tcp_v6_restore_cb(skb);
1679 refcounted = false;
1680 goto process;
1681 }
1682 }
1683 /* to ACK */
1684 /* fall through */
1685 case TCP_TW_ACK:
1686 tcp_v6_timewait_ack(sk, skb);
1687 break;
1688 case TCP_TW_RST:
1689 tcp_v6_send_reset(sk, skb);
1690 inet_twsk_deschedule_put(inet_twsk(sk));
1691 goto discard_it;
1692 case TCP_TW_SUCCESS:
1693 ;
1694 }
1695 goto discard_it;
1696}
1697
1698INDIRECT_CALLABLE_SCOPE void tcp_v6_early_demux(struct sk_buff *skb)
1699{
1700 const struct ipv6hdr *hdr;
1701 const struct tcphdr *th;
1702 struct sock *sk;
1703
1704 if (skb->pkt_type != PACKET_HOST)
1705 return;
1706
1707 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1708 return;
1709
1710 hdr = ipv6_hdr(skb);
1711 th = tcp_hdr(skb);
1712
1713 if (th->doff < sizeof(struct tcphdr) / 4)
1714 return;
1715
1716 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1717 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1718 &hdr->saddr, th->source,
1719 &hdr->daddr, ntohs(th->dest),
1720 inet6_iif(skb), inet6_sdif(skb));
1721 if (sk) {
1722 skb->sk = sk;
1723 skb->destructor = sock_edemux;
1724 if (sk_fullsock(sk)) {
1725 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1726
1727 if (dst)
1728 dst = dst_check(dst, tcp_inet6_sk(sk)->rx_dst_cookie);
1729 if (dst &&
1730 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1731 skb_dst_set_noref(skb, dst);
1732 }
1733 }
1734}
1735
1736static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1737 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1738 .twsk_unique = tcp_twsk_unique,
1739 .twsk_destructor = tcp_twsk_destructor,
1740};
1741
1742static const struct inet_connection_sock_af_ops ipv6_specific = {
1743 .queue_xmit = inet6_csk_xmit,
1744 .send_check = tcp_v6_send_check,
1745 .rebuild_header = inet6_sk_rebuild_header,
1746 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1747 .conn_request = tcp_v6_conn_request,
1748 .syn_recv_sock = tcp_v6_syn_recv_sock,
1749 .net_header_len = sizeof(struct ipv6hdr),
1750 .net_frag_header_len = sizeof(struct frag_hdr),
1751 .setsockopt = ipv6_setsockopt,
1752 .getsockopt = ipv6_getsockopt,
1753 .addr2sockaddr = inet6_csk_addr2sockaddr,
1754 .sockaddr_len = sizeof(struct sockaddr_in6),
1755#ifdef CONFIG_COMPAT
1756 .compat_setsockopt = compat_ipv6_setsockopt,
1757 .compat_getsockopt = compat_ipv6_getsockopt,
1758#endif
1759 .mtu_reduced = tcp_v6_mtu_reduced,
1760};
1761
1762#ifdef CONFIG_TCP_MD5SIG
1763static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1764 .md5_lookup = tcp_v6_md5_lookup,
1765 .calc_md5_hash = tcp_v6_md5_hash_skb,
1766 .md5_parse = tcp_v6_parse_md5_keys,
1767};
1768#endif
1769
1770/*
1771 * TCP over IPv4 via INET6 API
1772 */
1773static const struct inet_connection_sock_af_ops ipv6_mapped = {
1774 .queue_xmit = ip_queue_xmit,
1775 .send_check = tcp_v4_send_check,
1776 .rebuild_header = inet_sk_rebuild_header,
1777 .sk_rx_dst_set = inet_sk_rx_dst_set,
1778 .conn_request = tcp_v6_conn_request,
1779 .syn_recv_sock = tcp_v6_syn_recv_sock,
1780 .net_header_len = sizeof(struct iphdr),
1781 .setsockopt = ipv6_setsockopt,
1782 .getsockopt = ipv6_getsockopt,
1783 .addr2sockaddr = inet6_csk_addr2sockaddr,
1784 .sockaddr_len = sizeof(struct sockaddr_in6),
1785#ifdef CONFIG_COMPAT
1786 .compat_setsockopt = compat_ipv6_setsockopt,
1787 .compat_getsockopt = compat_ipv6_getsockopt,
1788#endif
1789 .mtu_reduced = tcp_v4_mtu_reduced,
1790};
1791
1792#ifdef CONFIG_TCP_MD5SIG
1793static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1794 .md5_lookup = tcp_v4_md5_lookup,
1795 .calc_md5_hash = tcp_v4_md5_hash_skb,
1796 .md5_parse = tcp_v6_parse_md5_keys,
1797};
1798#endif
1799
1800/* NOTE: A lot of things set to zero explicitly by call to
1801 * sk_alloc() so need not be done here.
1802 */
1803static int tcp_v6_init_sock(struct sock *sk)
1804{
1805 struct inet_connection_sock *icsk = inet_csk(sk);
1806
1807 tcp_init_sock(sk);
1808
1809 icsk->icsk_af_ops = &ipv6_specific;
1810
1811#ifdef CONFIG_TCP_MD5SIG
1812 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1813#endif
1814
1815 return 0;
1816}
1817
1818static void tcp_v6_destroy_sock(struct sock *sk)
1819{
1820 tcp_v4_destroy_sock(sk);
1821 inet6_destroy_sock(sk);
1822}
1823
1824#ifdef CONFIG_PROC_FS
1825/* Proc filesystem TCPv6 sock list dumping. */
1826static void get_openreq6(struct seq_file *seq,
1827 const struct request_sock *req, int i)
1828{
1829 long ttd = req->rsk_timer.expires - jiffies;
1830 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1831 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1832
1833 if (ttd < 0)
1834 ttd = 0;
1835
1836 seq_printf(seq,
1837 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1838 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1839 i,
1840 src->s6_addr32[0], src->s6_addr32[1],
1841 src->s6_addr32[2], src->s6_addr32[3],
1842 inet_rsk(req)->ir_num,
1843 dest->s6_addr32[0], dest->s6_addr32[1],
1844 dest->s6_addr32[2], dest->s6_addr32[3],
1845 ntohs(inet_rsk(req)->ir_rmt_port),
1846 TCP_SYN_RECV,
1847 0, 0, /* could print option size, but that is af dependent. */
1848 1, /* timers active (only the expire timer) */
1849 jiffies_to_clock_t(ttd),
1850 req->num_timeout,
1851 from_kuid_munged(seq_user_ns(seq),
1852 sock_i_uid(req->rsk_listener)),
1853 0, /* non standard timer */
1854 0, /* open_requests have no inode */
1855 0, req);
1856}
1857
1858static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1859{
1860 const struct in6_addr *dest, *src;
1861 __u16 destp, srcp;
1862 int timer_active;
1863 unsigned long timer_expires;
1864 const struct inet_sock *inet = inet_sk(sp);
1865 const struct tcp_sock *tp = tcp_sk(sp);
1866 const struct inet_connection_sock *icsk = inet_csk(sp);
1867 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1868 int rx_queue;
1869 int state;
1870
1871 dest = &sp->sk_v6_daddr;
1872 src = &sp->sk_v6_rcv_saddr;
1873 destp = ntohs(inet->inet_dport);
1874 srcp = ntohs(inet->inet_sport);
1875
1876 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1877 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
1878 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1879 timer_active = 1;
1880 timer_expires = icsk->icsk_timeout;
1881 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1882 timer_active = 4;
1883 timer_expires = icsk->icsk_timeout;
1884 } else if (timer_pending(&sp->sk_timer)) {
1885 timer_active = 2;
1886 timer_expires = sp->sk_timer.expires;
1887 } else {
1888 timer_active = 0;
1889 timer_expires = jiffies;
1890 }
1891
1892 state = inet_sk_state_load(sp);
1893 if (state == TCP_LISTEN)
1894 rx_queue = sp->sk_ack_backlog;
1895 else
1896 /* Because we don't lock the socket,
1897 * we might find a transient negative value.
1898 */
1899 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
1900 READ_ONCE(tp->copied_seq), 0);
1901
1902 seq_printf(seq,
1903 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1904 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1905 i,
1906 src->s6_addr32[0], src->s6_addr32[1],
1907 src->s6_addr32[2], src->s6_addr32[3], srcp,
1908 dest->s6_addr32[0], dest->s6_addr32[1],
1909 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1910 state,
1911 READ_ONCE(tp->write_seq) - tp->snd_una,
1912 rx_queue,
1913 timer_active,
1914 jiffies_delta_to_clock_t(timer_expires - jiffies),
1915 icsk->icsk_retransmits,
1916 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1917 icsk->icsk_probes_out,
1918 sock_i_ino(sp),
1919 refcount_read(&sp->sk_refcnt), sp,
1920 jiffies_to_clock_t(icsk->icsk_rto),
1921 jiffies_to_clock_t(icsk->icsk_ack.ato),
1922 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
1923 tp->snd_cwnd,
1924 state == TCP_LISTEN ?
1925 fastopenq->max_qlen :
1926 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1927 );
1928}
1929
1930static void get_timewait6_sock(struct seq_file *seq,
1931 struct inet_timewait_sock *tw, int i)
1932{
1933 long delta = tw->tw_timer.expires - jiffies;
1934 const struct in6_addr *dest, *src;
1935 __u16 destp, srcp;
1936
1937 dest = &tw->tw_v6_daddr;
1938 src = &tw->tw_v6_rcv_saddr;
1939 destp = ntohs(tw->tw_dport);
1940 srcp = ntohs(tw->tw_sport);
1941
1942 seq_printf(seq,
1943 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1944 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1945 i,
1946 src->s6_addr32[0], src->s6_addr32[1],
1947 src->s6_addr32[2], src->s6_addr32[3], srcp,
1948 dest->s6_addr32[0], dest->s6_addr32[1],
1949 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1950 tw->tw_substate, 0, 0,
1951 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1952 refcount_read(&tw->tw_refcnt), tw);
1953}
1954
1955static int tcp6_seq_show(struct seq_file *seq, void *v)
1956{
1957 struct tcp_iter_state *st;
1958 struct sock *sk = v;
1959
1960 if (v == SEQ_START_TOKEN) {
1961 seq_puts(seq,
1962 " sl "
1963 "local_address "
1964 "remote_address "
1965 "st tx_queue rx_queue tr tm->when retrnsmt"
1966 " uid timeout inode\n");
1967 goto out;
1968 }
1969 st = seq->private;
1970
1971 if (sk->sk_state == TCP_TIME_WAIT)
1972 get_timewait6_sock(seq, v, st->num);
1973 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1974 get_openreq6(seq, v, st->num);
1975 else
1976 get_tcp6_sock(seq, v, st->num);
1977out:
1978 return 0;
1979}
1980
1981static const struct seq_operations tcp6_seq_ops = {
1982 .show = tcp6_seq_show,
1983 .start = tcp_seq_start,
1984 .next = tcp_seq_next,
1985 .stop = tcp_seq_stop,
1986};
1987
1988static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1989 .family = AF_INET6,
1990};
1991
1992int __net_init tcp6_proc_init(struct net *net)
1993{
1994 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
1995 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
1996 return -ENOMEM;
1997 return 0;
1998}
1999
2000void tcp6_proc_exit(struct net *net)
2001{
2002 remove_proc_entry("tcp6", net->proc_net);
2003}
2004#endif
2005
2006struct proto tcpv6_prot = {
2007 .name = "TCPv6",
2008 .owner = THIS_MODULE,
2009 .close = tcp_close,
2010 .pre_connect = tcp_v6_pre_connect,
2011 .connect = tcp_v6_connect,
2012 .disconnect = tcp_disconnect,
2013 .accept = inet_csk_accept,
2014 .ioctl = tcp_ioctl,
2015 .init = tcp_v6_init_sock,
2016 .destroy = tcp_v6_destroy_sock,
2017 .shutdown = tcp_shutdown,
2018 .setsockopt = tcp_setsockopt,
2019 .getsockopt = tcp_getsockopt,
2020 .keepalive = tcp_set_keepalive,
2021 .recvmsg = tcp_recvmsg,
2022 .sendmsg = tcp_sendmsg,
2023 .sendpage = tcp_sendpage,
2024 .backlog_rcv = tcp_v6_do_rcv,
2025 .release_cb = tcp_release_cb,
2026 .hash = inet6_hash,
2027 .unhash = inet_unhash,
2028 .get_port = inet_csk_get_port,
2029 .enter_memory_pressure = tcp_enter_memory_pressure,
2030 .leave_memory_pressure = tcp_leave_memory_pressure,
2031 .stream_memory_free = tcp_stream_memory_free,
2032 .sockets_allocated = &tcp_sockets_allocated,
2033 .memory_allocated = &tcp_memory_allocated,
2034 .memory_pressure = &tcp_memory_pressure,
2035 .orphan_count = &tcp_orphan_count,
2036 .sysctl_mem = sysctl_tcp_mem,
2037 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2038 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2039 .max_header = MAX_TCP_HEADER,
2040 .obj_size = sizeof(struct tcp6_sock),
2041 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2042 .twsk_prot = &tcp6_timewait_sock_ops,
2043 .rsk_prot = &tcp6_request_sock_ops,
2044 .h.hashinfo = &tcp_hashinfo,
2045 .no_autobind = true,
2046#ifdef CONFIG_COMPAT
2047 .compat_setsockopt = compat_tcp_setsockopt,
2048 .compat_getsockopt = compat_tcp_getsockopt,
2049#endif
2050 .diag_destroy = tcp_abort,
2051};
2052
2053/* thinking of making this const? Don't.
2054 * early_demux can change based on sysctl.
2055 */
2056static struct inet6_protocol tcpv6_protocol = {
2057 .early_demux = tcp_v6_early_demux,
2058 .early_demux_handler = tcp_v6_early_demux,
2059 .handler = tcp_v6_rcv,
2060 .err_handler = tcp_v6_err,
2061 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2062};
2063
2064static struct inet_protosw tcpv6_protosw = {
2065 .type = SOCK_STREAM,
2066 .protocol = IPPROTO_TCP,
2067 .prot = &tcpv6_prot,
2068 .ops = &inet6_stream_ops,
2069 .flags = INET_PROTOSW_PERMANENT |
2070 INET_PROTOSW_ICSK,
2071};
2072
2073static int __net_init tcpv6_net_init(struct net *net)
2074{
2075 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2076 SOCK_RAW, IPPROTO_TCP, net);
2077}
2078
2079static void __net_exit tcpv6_net_exit(struct net *net)
2080{
2081 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2082}
2083
2084static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2085{
2086 inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2087}
2088
2089static struct pernet_operations tcpv6_net_ops = {
2090 .init = tcpv6_net_init,
2091 .exit = tcpv6_net_exit,
2092 .exit_batch = tcpv6_net_exit_batch,
2093};
2094
2095int __init tcpv6_init(void)
2096{
2097 int ret;
2098
2099 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2100 if (ret)
2101 goto out;
2102
2103 /* register inet6 protocol */
2104 ret = inet6_register_protosw(&tcpv6_protosw);
2105 if (ret)
2106 goto out_tcpv6_protocol;
2107
2108 ret = register_pernet_subsys(&tcpv6_net_ops);
2109 if (ret)
2110 goto out_tcpv6_protosw;
2111out:
2112 return ret;
2113
2114out_tcpv6_protosw:
2115 inet6_unregister_protosw(&tcpv6_protosw);
2116out_tcpv6_protocol:
2117 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2118 goto out;
2119}
2120
2121void tcpv6_exit(void)
2122{
2123 unregister_pernet_subsys(&tcpv6_net_ops);
2124 inet6_unregister_protosw(&tcpv6_protosw);
2125 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2126}