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