3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
10 * linux/net/ipv4/tcp_input.c
11 * linux/net/ipv4/tcp_output.c
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.
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.
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>
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>
48 #include <net/ndisc.h>
49 #include <net/inet6_hashtables.h>
50 #include <net/inet6_connection_sock.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>
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>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
69 #include <crypto/hash.h>
70 #include <linux/scatterlist.h>
72 static void tcp_v6_send_reset(const struct sock
*sk
, struct sk_buff
*skb
);
73 static void tcp_v6_reqsk_send_ack(const struct sock
*sk
, struct sk_buff
*skb
,
74 struct request_sock
*req
);
76 static int tcp_v6_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
78 static const struct inet_connection_sock_af_ops ipv6_mapped
;
79 static const struct inet_connection_sock_af_ops ipv6_specific
;
80 #ifdef CONFIG_TCP_MD5SIG
81 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific
;
82 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific
;
84 static struct tcp_md5sig_key
*tcp_v6_md5_do_lookup(const struct sock
*sk
,
85 const struct in6_addr
*addr
)
91 static void inet6_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
93 struct dst_entry
*dst
= skb_dst(skb
);
95 if (dst
&& dst_hold_safe(dst
)) {
96 const struct rt6_info
*rt
= (const struct rt6_info
*)dst
;
99 inet_sk(sk
)->rx_dst_ifindex
= skb
->skb_iif
;
100 inet6_sk(sk
)->rx_dst_cookie
= rt6_get_cookie(rt
);
104 static u32
tcp_v6_init_seq(const struct sk_buff
*skb
)
106 return secure_tcpv6_seq(ipv6_hdr(skb
)->daddr
.s6_addr32
,
107 ipv6_hdr(skb
)->saddr
.s6_addr32
,
109 tcp_hdr(skb
)->source
);
112 static u32
tcp_v6_init_ts_off(const struct net
*net
, const struct sk_buff
*skb
)
114 return secure_tcpv6_ts_off(net
, ipv6_hdr(skb
)->daddr
.s6_addr32
,
115 ipv6_hdr(skb
)->saddr
.s6_addr32
);
118 static int tcp_v6_connect(struct sock
*sk
, struct sockaddr
*uaddr
,
121 struct sockaddr_in6
*usin
= (struct sockaddr_in6
*) uaddr
;
122 struct inet_sock
*inet
= inet_sk(sk
);
123 struct inet_connection_sock
*icsk
= inet_csk(sk
);
124 struct ipv6_pinfo
*np
= inet6_sk(sk
);
125 struct tcp_sock
*tp
= tcp_sk(sk
);
126 struct in6_addr
*saddr
= NULL
, *final_p
, final
;
127 struct ipv6_txoptions
*opt
;
129 struct dst_entry
*dst
;
132 struct inet_timewait_death_row
*tcp_death_row
= &sock_net(sk
)->ipv4
.tcp_death_row
;
134 if (addr_len
< SIN6_LEN_RFC2133
)
137 if (usin
->sin6_family
!= AF_INET6
)
138 return -EAFNOSUPPORT
;
140 memset(&fl6
, 0, sizeof(fl6
));
143 fl6
.flowlabel
= usin
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
144 IP6_ECN_flow_init(fl6
.flowlabel
);
145 if (fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) {
146 struct ip6_flowlabel
*flowlabel
;
147 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
150 fl6_sock_release(flowlabel
);
155 * connect() to INADDR_ANY means loopback (BSD'ism).
158 if (ipv6_addr_any(&usin
->sin6_addr
)) {
159 if (ipv6_addr_v4mapped(&sk
->sk_v6_rcv_saddr
))
160 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK
),
163 usin
->sin6_addr
= in6addr_loopback
;
166 addr_type
= ipv6_addr_type(&usin
->sin6_addr
);
168 if (addr_type
& IPV6_ADDR_MULTICAST
)
171 if (addr_type
&IPV6_ADDR_LINKLOCAL
) {
172 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
173 usin
->sin6_scope_id
) {
174 /* If interface is set while binding, indices
177 if (sk
->sk_bound_dev_if
&&
178 sk
->sk_bound_dev_if
!= usin
->sin6_scope_id
)
181 sk
->sk_bound_dev_if
= usin
->sin6_scope_id
;
184 /* Connect to link-local address requires an interface */
185 if (!sk
->sk_bound_dev_if
)
189 if (tp
->rx_opt
.ts_recent_stamp
&&
190 !ipv6_addr_equal(&sk
->sk_v6_daddr
, &usin
->sin6_addr
)) {
191 tp
->rx_opt
.ts_recent
= 0;
192 tp
->rx_opt
.ts_recent_stamp
= 0;
196 sk
->sk_v6_daddr
= usin
->sin6_addr
;
197 np
->flow_label
= fl6
.flowlabel
;
203 if (addr_type
& IPV6_ADDR_MAPPED
) {
204 u32 exthdrlen
= icsk
->icsk_ext_hdr_len
;
205 struct sockaddr_in sin
;
207 SOCK_DEBUG(sk
, "connect: ipv4 mapped\n");
209 if (__ipv6_only_sock(sk
))
212 sin
.sin_family
= AF_INET
;
213 sin
.sin_port
= usin
->sin6_port
;
214 sin
.sin_addr
.s_addr
= usin
->sin6_addr
.s6_addr32
[3];
216 icsk
->icsk_af_ops
= &ipv6_mapped
;
217 sk
->sk_backlog_rcv
= tcp_v4_do_rcv
;
218 #ifdef CONFIG_TCP_MD5SIG
219 tp
->af_specific
= &tcp_sock_ipv6_mapped_specific
;
222 err
= tcp_v4_connect(sk
, (struct sockaddr
*)&sin
, sizeof(sin
));
225 icsk
->icsk_ext_hdr_len
= exthdrlen
;
226 icsk
->icsk_af_ops
= &ipv6_specific
;
227 sk
->sk_backlog_rcv
= tcp_v6_do_rcv
;
228 #ifdef CONFIG_TCP_MD5SIG
229 tp
->af_specific
= &tcp_sock_ipv6_specific
;
233 np
->saddr
= sk
->sk_v6_rcv_saddr
;
238 if (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
))
239 saddr
= &sk
->sk_v6_rcv_saddr
;
241 fl6
.flowi6_proto
= IPPROTO_TCP
;
242 fl6
.daddr
= sk
->sk_v6_daddr
;
243 fl6
.saddr
= saddr
? *saddr
: np
->saddr
;
244 fl6
.flowi6_oif
= sk
->sk_bound_dev_if
;
245 fl6
.flowi6_mark
= sk
->sk_mark
;
246 fl6
.fl6_dport
= usin
->sin6_port
;
247 fl6
.fl6_sport
= inet
->inet_sport
;
248 fl6
.flowi6_uid
= sk
->sk_uid
;
250 opt
= rcu_dereference_protected(np
->opt
, lockdep_sock_is_held(sk
));
251 final_p
= fl6_update_dst(&fl6
, opt
, &final
);
253 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
255 dst
= ip6_dst_lookup_flow(sk
, &fl6
, final_p
);
263 sk
->sk_v6_rcv_saddr
= *saddr
;
266 /* set the source address */
268 inet
->inet_rcv_saddr
= LOOPBACK4_IPV6
;
270 sk
->sk_gso_type
= SKB_GSO_TCPV6
;
271 ip6_dst_store(sk
, dst
, NULL
, NULL
);
273 icsk
->icsk_ext_hdr_len
= 0;
275 icsk
->icsk_ext_hdr_len
= opt
->opt_flen
+
278 tp
->rx_opt
.mss_clamp
= IPV6_MIN_MTU
- sizeof(struct tcphdr
) - sizeof(struct ipv6hdr
);
280 inet
->inet_dport
= usin
->sin6_port
;
282 tcp_set_state(sk
, TCP_SYN_SENT
);
283 err
= inet6_hash_connect(tcp_death_row
, sk
);
289 if (likely(!tp
->repair
)) {
291 tp
->write_seq
= secure_tcpv6_seq(np
->saddr
.s6_addr32
,
292 sk
->sk_v6_daddr
.s6_addr32
,
295 tp
->tsoffset
= secure_tcpv6_ts_off(sock_net(sk
),
297 sk
->sk_v6_daddr
.s6_addr32
);
300 if (tcp_fastopen_defer_connect(sk
, &err
))
305 err
= tcp_connect(sk
);
312 tcp_set_state(sk
, TCP_CLOSE
);
314 inet
->inet_dport
= 0;
315 sk
->sk_route_caps
= 0;
319 static void tcp_v6_mtu_reduced(struct sock
*sk
)
321 struct dst_entry
*dst
;
323 if ((1 << sk
->sk_state
) & (TCPF_LISTEN
| TCPF_CLOSE
))
326 dst
= inet6_csk_update_pmtu(sk
, tcp_sk(sk
)->mtu_info
);
330 if (inet_csk(sk
)->icsk_pmtu_cookie
> dst_mtu(dst
)) {
331 tcp_sync_mss(sk
, dst_mtu(dst
));
332 tcp_simple_retransmit(sk
);
336 static void tcp_v6_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
337 u8 type
, u8 code
, int offset
, __be32 info
)
339 const struct ipv6hdr
*hdr
= (const struct ipv6hdr
*)skb
->data
;
340 const struct tcphdr
*th
= (struct tcphdr
*)(skb
->data
+offset
);
341 struct net
*net
= dev_net(skb
->dev
);
342 struct request_sock
*fastopen
;
343 struct ipv6_pinfo
*np
;
350 sk
= __inet6_lookup_established(net
, &tcp_hashinfo
,
351 &hdr
->daddr
, th
->dest
,
352 &hdr
->saddr
, ntohs(th
->source
),
353 skb
->dev
->ifindex
, inet6_sdif(skb
));
356 __ICMP6_INC_STATS(net
, __in6_dev_get(skb
->dev
),
361 if (sk
->sk_state
== TCP_TIME_WAIT
) {
362 inet_twsk_put(inet_twsk(sk
));
365 seq
= ntohl(th
->seq
);
366 fatal
= icmpv6_err_convert(type
, code
, &err
);
367 if (sk
->sk_state
== TCP_NEW_SYN_RECV
)
368 return tcp_req_err(sk
, seq
, fatal
);
371 if (sock_owned_by_user(sk
) && type
!= ICMPV6_PKT_TOOBIG
)
372 __NET_INC_STATS(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
374 if (sk
->sk_state
== TCP_CLOSE
)
377 if (ipv6_hdr(skb
)->hop_limit
< inet6_sk(sk
)->min_hopcount
) {
378 __NET_INC_STATS(net
, LINUX_MIB_TCPMINTTLDROP
);
383 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
384 fastopen
= tp
->fastopen_rsk
;
385 snd_una
= fastopen
? tcp_rsk(fastopen
)->snt_isn
: tp
->snd_una
;
386 if (sk
->sk_state
!= TCP_LISTEN
&&
387 !between(seq
, snd_una
, tp
->snd_nxt
)) {
388 __NET_INC_STATS(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
394 if (type
== NDISC_REDIRECT
) {
395 if (!sock_owned_by_user(sk
)) {
396 struct dst_entry
*dst
= __sk_dst_check(sk
, np
->dst_cookie
);
399 dst
->ops
->redirect(dst
, sk
, skb
);
404 if (type
== ICMPV6_PKT_TOOBIG
) {
405 /* We are not interested in TCP_LISTEN and open_requests
406 * (SYN-ACKs send out by Linux are always <576bytes so
407 * they should go through unfragmented).
409 if (sk
->sk_state
== TCP_LISTEN
)
412 if (!ip6_sk_accept_pmtu(sk
))
415 tp
->mtu_info
= ntohl(info
);
416 if (!sock_owned_by_user(sk
))
417 tcp_v6_mtu_reduced(sk
);
418 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED
,
425 /* Might be for an request_sock */
426 switch (sk
->sk_state
) {
429 /* Only in fast or simultaneous open. If a fast open socket is
430 * is already accepted it is treated as a connected one below.
432 if (fastopen
&& !fastopen
->sk
)
435 if (!sock_owned_by_user(sk
)) {
437 sk
->sk_error_report(sk
); /* Wake people up to see the error (see connect in sock.c) */
441 sk
->sk_err_soft
= err
;
445 if (!sock_owned_by_user(sk
) && np
->recverr
) {
447 sk
->sk_error_report(sk
);
449 sk
->sk_err_soft
= err
;
457 static int tcp_v6_send_synack(const struct sock
*sk
, struct dst_entry
*dst
,
459 struct request_sock
*req
,
460 struct tcp_fastopen_cookie
*foc
,
461 enum tcp_synack_type synack_type
)
463 struct inet_request_sock
*ireq
= inet_rsk(req
);
464 struct ipv6_pinfo
*np
= inet6_sk(sk
);
465 struct ipv6_txoptions
*opt
;
466 struct flowi6
*fl6
= &fl
->u
.ip6
;
470 /* First, grab a route. */
471 if (!dst
&& (dst
= inet6_csk_route_req(sk
, fl6
, req
,
472 IPPROTO_TCP
)) == NULL
)
475 skb
= tcp_make_synack(sk
, dst
, req
, foc
, synack_type
);
478 __tcp_v6_send_check(skb
, &ireq
->ir_v6_loc_addr
,
479 &ireq
->ir_v6_rmt_addr
);
481 fl6
->daddr
= ireq
->ir_v6_rmt_addr
;
482 if (np
->repflow
&& ireq
->pktopts
)
483 fl6
->flowlabel
= ip6_flowlabel(ipv6_hdr(ireq
->pktopts
));
486 opt
= ireq
->ipv6_opt
;
488 opt
= rcu_dereference(np
->opt
);
489 err
= ip6_xmit(sk
, skb
, fl6
, sk
->sk_mark
, opt
, np
->tclass
);
491 err
= net_xmit_eval(err
);
499 static void tcp_v6_reqsk_destructor(struct request_sock
*req
)
501 kfree(inet_rsk(req
)->ipv6_opt
);
502 kfree_skb(inet_rsk(req
)->pktopts
);
505 #ifdef CONFIG_TCP_MD5SIG
506 static struct tcp_md5sig_key
*tcp_v6_md5_do_lookup(const struct sock
*sk
,
507 const struct in6_addr
*addr
)
509 return tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)addr
, AF_INET6
);
512 static struct tcp_md5sig_key
*tcp_v6_md5_lookup(const struct sock
*sk
,
513 const struct sock
*addr_sk
)
515 return tcp_v6_md5_do_lookup(sk
, &addr_sk
->sk_v6_daddr
);
518 static int tcp_v6_parse_md5_keys(struct sock
*sk
, int optname
,
519 char __user
*optval
, int optlen
)
521 struct tcp_md5sig cmd
;
522 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*)&cmd
.tcpm_addr
;
525 if (optlen
< sizeof(cmd
))
528 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
531 if (sin6
->sin6_family
!= AF_INET6
)
534 if (optname
== TCP_MD5SIG_EXT
&&
535 cmd
.tcpm_flags
& TCP_MD5SIG_FLAG_PREFIX
) {
536 prefixlen
= cmd
.tcpm_prefixlen
;
537 if (prefixlen
> 128 || (ipv6_addr_v4mapped(&sin6
->sin6_addr
) &&
541 prefixlen
= ipv6_addr_v4mapped(&sin6
->sin6_addr
) ? 32 : 128;
544 if (!cmd
.tcpm_keylen
) {
545 if (ipv6_addr_v4mapped(&sin6
->sin6_addr
))
546 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
.s6_addr32
[3],
548 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
,
549 AF_INET6
, prefixlen
);
552 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
555 if (ipv6_addr_v4mapped(&sin6
->sin6_addr
))
556 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
.s6_addr32
[3],
557 AF_INET
, prefixlen
, cmd
.tcpm_key
,
558 cmd
.tcpm_keylen
, GFP_KERNEL
);
560 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
,
561 AF_INET6
, prefixlen
, cmd
.tcpm_key
,
562 cmd
.tcpm_keylen
, GFP_KERNEL
);
565 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool
*hp
,
566 const struct in6_addr
*daddr
,
567 const struct in6_addr
*saddr
,
568 const struct tcphdr
*th
, int nbytes
)
570 struct tcp6_pseudohdr
*bp
;
571 struct scatterlist sg
;
575 /* 1. TCP pseudo-header (RFC2460) */
578 bp
->protocol
= cpu_to_be32(IPPROTO_TCP
);
579 bp
->len
= cpu_to_be32(nbytes
);
581 _th
= (struct tcphdr
*)(bp
+ 1);
582 memcpy(_th
, th
, sizeof(*th
));
585 sg_init_one(&sg
, bp
, sizeof(*bp
) + sizeof(*th
));
586 ahash_request_set_crypt(hp
->md5_req
, &sg
, NULL
,
587 sizeof(*bp
) + sizeof(*th
));
588 return crypto_ahash_update(hp
->md5_req
);
591 static int tcp_v6_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
592 const struct in6_addr
*daddr
, struct in6_addr
*saddr
,
593 const struct tcphdr
*th
)
595 struct tcp_md5sig_pool
*hp
;
596 struct ahash_request
*req
;
598 hp
= tcp_get_md5sig_pool();
600 goto clear_hash_noput
;
603 if (crypto_ahash_init(req
))
605 if (tcp_v6_md5_hash_headers(hp
, daddr
, saddr
, th
, th
->doff
<< 2))
607 if (tcp_md5_hash_key(hp
, key
))
609 ahash_request_set_crypt(req
, NULL
, md5_hash
, 0);
610 if (crypto_ahash_final(req
))
613 tcp_put_md5sig_pool();
617 tcp_put_md5sig_pool();
619 memset(md5_hash
, 0, 16);
623 static int tcp_v6_md5_hash_skb(char *md5_hash
,
624 const struct tcp_md5sig_key
*key
,
625 const struct sock
*sk
,
626 const struct sk_buff
*skb
)
628 const struct in6_addr
*saddr
, *daddr
;
629 struct tcp_md5sig_pool
*hp
;
630 struct ahash_request
*req
;
631 const struct tcphdr
*th
= tcp_hdr(skb
);
633 if (sk
) { /* valid for establish/request sockets */
634 saddr
= &sk
->sk_v6_rcv_saddr
;
635 daddr
= &sk
->sk_v6_daddr
;
637 const struct ipv6hdr
*ip6h
= ipv6_hdr(skb
);
638 saddr
= &ip6h
->saddr
;
639 daddr
= &ip6h
->daddr
;
642 hp
= tcp_get_md5sig_pool();
644 goto clear_hash_noput
;
647 if (crypto_ahash_init(req
))
650 if (tcp_v6_md5_hash_headers(hp
, daddr
, saddr
, th
, skb
->len
))
652 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
654 if (tcp_md5_hash_key(hp
, key
))
656 ahash_request_set_crypt(req
, NULL
, md5_hash
, 0);
657 if (crypto_ahash_final(req
))
660 tcp_put_md5sig_pool();
664 tcp_put_md5sig_pool();
666 memset(md5_hash
, 0, 16);
672 static bool tcp_v6_inbound_md5_hash(const struct sock
*sk
,
673 const struct sk_buff
*skb
)
675 #ifdef CONFIG_TCP_MD5SIG
676 const __u8
*hash_location
= NULL
;
677 struct tcp_md5sig_key
*hash_expected
;
678 const struct ipv6hdr
*ip6h
= ipv6_hdr(skb
);
679 const struct tcphdr
*th
= tcp_hdr(skb
);
683 hash_expected
= tcp_v6_md5_do_lookup(sk
, &ip6h
->saddr
);
684 hash_location
= tcp_parse_md5sig_option(th
);
686 /* We've parsed the options - do we have a hash? */
687 if (!hash_expected
&& !hash_location
)
690 if (hash_expected
&& !hash_location
) {
691 NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
695 if (!hash_expected
&& hash_location
) {
696 NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
700 /* check the signature */
701 genhash
= tcp_v6_md5_hash_skb(newhash
,
705 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
706 NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPMD5FAILURE
);
707 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
708 genhash
? "failed" : "mismatch",
709 &ip6h
->saddr
, ntohs(th
->source
),
710 &ip6h
->daddr
, ntohs(th
->dest
));
717 static void tcp_v6_init_req(struct request_sock
*req
,
718 const struct sock
*sk_listener
,
721 struct inet_request_sock
*ireq
= inet_rsk(req
);
722 const struct ipv6_pinfo
*np
= inet6_sk(sk_listener
);
724 ireq
->ir_v6_rmt_addr
= ipv6_hdr(skb
)->saddr
;
725 ireq
->ir_v6_loc_addr
= ipv6_hdr(skb
)->daddr
;
727 /* So that link locals have meaning */
728 if (!sk_listener
->sk_bound_dev_if
&&
729 ipv6_addr_type(&ireq
->ir_v6_rmt_addr
) & IPV6_ADDR_LINKLOCAL
)
730 ireq
->ir_iif
= tcp_v6_iif(skb
);
732 if (!TCP_SKB_CB(skb
)->tcp_tw_isn
&&
733 (ipv6_opt_accepted(sk_listener
, skb
, &TCP_SKB_CB(skb
)->header
.h6
) ||
734 np
->rxopt
.bits
.rxinfo
||
735 np
->rxopt
.bits
.rxoinfo
|| np
->rxopt
.bits
.rxhlim
||
736 np
->rxopt
.bits
.rxohlim
|| np
->repflow
)) {
737 refcount_inc(&skb
->users
);
742 static struct dst_entry
*tcp_v6_route_req(const struct sock
*sk
,
744 const struct request_sock
*req
)
746 return inet6_csk_route_req(sk
, &fl
->u
.ip6
, req
, IPPROTO_TCP
);
749 struct request_sock_ops tcp6_request_sock_ops __read_mostly
= {
751 .obj_size
= sizeof(struct tcp6_request_sock
),
752 .rtx_syn_ack
= tcp_rtx_synack
,
753 .send_ack
= tcp_v6_reqsk_send_ack
,
754 .destructor
= tcp_v6_reqsk_destructor
,
755 .send_reset
= tcp_v6_send_reset
,
756 .syn_ack_timeout
= tcp_syn_ack_timeout
,
759 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops
= {
760 .mss_clamp
= IPV6_MIN_MTU
- sizeof(struct tcphdr
) -
761 sizeof(struct ipv6hdr
),
762 #ifdef CONFIG_TCP_MD5SIG
763 .req_md5_lookup
= tcp_v6_md5_lookup
,
764 .calc_md5_hash
= tcp_v6_md5_hash_skb
,
766 .init_req
= tcp_v6_init_req
,
767 #ifdef CONFIG_SYN_COOKIES
768 .cookie_init_seq
= cookie_v6_init_sequence
,
770 .route_req
= tcp_v6_route_req
,
771 .init_seq
= tcp_v6_init_seq
,
772 .init_ts_off
= tcp_v6_init_ts_off
,
773 .send_synack
= tcp_v6_send_synack
,
776 static void tcp_v6_send_response(const struct sock
*sk
, struct sk_buff
*skb
, u32 seq
,
777 u32 ack
, u32 win
, u32 tsval
, u32 tsecr
,
778 int oif
, struct tcp_md5sig_key
*key
, int rst
,
779 u8 tclass
, __be32 label
)
781 const struct tcphdr
*th
= tcp_hdr(skb
);
783 struct sk_buff
*buff
;
785 struct net
*net
= sk
? sock_net(sk
) : dev_net(skb_dst(skb
)->dev
);
786 struct sock
*ctl_sk
= net
->ipv6
.tcp_sk
;
787 unsigned int tot_len
= sizeof(struct tcphdr
);
788 struct dst_entry
*dst
;
792 tot_len
+= TCPOLEN_TSTAMP_ALIGNED
;
793 #ifdef CONFIG_TCP_MD5SIG
795 tot_len
+= TCPOLEN_MD5SIG_ALIGNED
;
798 buff
= alloc_skb(MAX_HEADER
+ sizeof(struct ipv6hdr
) + tot_len
,
803 skb_reserve(buff
, MAX_HEADER
+ sizeof(struct ipv6hdr
) + tot_len
);
805 t1
= skb_push(buff
, tot_len
);
806 skb_reset_transport_header(buff
);
808 /* Swap the send and the receive. */
809 memset(t1
, 0, sizeof(*t1
));
810 t1
->dest
= th
->source
;
811 t1
->source
= th
->dest
;
812 t1
->doff
= tot_len
/ 4;
813 t1
->seq
= htonl(seq
);
814 t1
->ack_seq
= htonl(ack
);
815 t1
->ack
= !rst
|| !th
->ack
;
817 t1
->window
= htons(win
);
819 topt
= (__be32
*)(t1
+ 1);
822 *topt
++ = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
823 (TCPOPT_TIMESTAMP
<< 8) | TCPOLEN_TIMESTAMP
);
824 *topt
++ = htonl(tsval
);
825 *topt
++ = htonl(tsecr
);
828 #ifdef CONFIG_TCP_MD5SIG
830 *topt
++ = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
831 (TCPOPT_MD5SIG
<< 8) | TCPOLEN_MD5SIG
);
832 tcp_v6_md5_hash_hdr((__u8
*)topt
, key
,
833 &ipv6_hdr(skb
)->saddr
,
834 &ipv6_hdr(skb
)->daddr
, t1
);
838 memset(&fl6
, 0, sizeof(fl6
));
839 fl6
.daddr
= ipv6_hdr(skb
)->saddr
;
840 fl6
.saddr
= ipv6_hdr(skb
)->daddr
;
841 fl6
.flowlabel
= label
;
843 buff
->ip_summed
= CHECKSUM_PARTIAL
;
846 __tcp_v6_send_check(buff
, &fl6
.saddr
, &fl6
.daddr
);
848 fl6
.flowi6_proto
= IPPROTO_TCP
;
849 if (rt6_need_strict(&fl6
.daddr
) && !oif
)
850 fl6
.flowi6_oif
= tcp_v6_iif(skb
);
852 if (!oif
&& netif_index_is_l3_master(net
, skb
->skb_iif
))
855 fl6
.flowi6_oif
= oif
;
858 fl6
.flowi6_mark
= IP6_REPLY_MARK(net
, skb
->mark
);
859 fl6
.fl6_dport
= t1
->dest
;
860 fl6
.fl6_sport
= t1
->source
;
861 fl6
.flowi6_uid
= sock_net_uid(net
, sk
&& sk_fullsock(sk
) ? sk
: NULL
);
862 security_skb_classify_flow(skb
, flowi6_to_flowi(&fl6
));
864 /* Pass a socket to ip6_dst_lookup either it is for RST
865 * Underlying function will use this to retrieve the network
868 dst
= ip6_dst_lookup_flow(ctl_sk
, &fl6
, NULL
);
870 skb_dst_set(buff
, dst
);
871 ip6_xmit(ctl_sk
, buff
, &fl6
, fl6
.flowi6_mark
, NULL
, tclass
);
872 TCP_INC_STATS(net
, TCP_MIB_OUTSEGS
);
874 TCP_INC_STATS(net
, TCP_MIB_OUTRSTS
);
881 static void tcp_v6_send_reset(const struct sock
*sk
, struct sk_buff
*skb
)
883 const struct tcphdr
*th
= tcp_hdr(skb
);
884 u32 seq
= 0, ack_seq
= 0;
885 struct tcp_md5sig_key
*key
= NULL
;
886 #ifdef CONFIG_TCP_MD5SIG
887 const __u8
*hash_location
= NULL
;
888 struct ipv6hdr
*ipv6h
= ipv6_hdr(skb
);
889 unsigned char newhash
[16];
891 struct sock
*sk1
= NULL
;
898 /* If sk not NULL, it means we did a successful lookup and incoming
899 * route had to be correct. prequeue might have dropped our dst.
901 if (!sk
&& !ipv6_unicast_destination(skb
))
904 #ifdef CONFIG_TCP_MD5SIG
906 hash_location
= tcp_parse_md5sig_option(th
);
907 if (sk
&& sk_fullsock(sk
)) {
908 key
= tcp_v6_md5_do_lookup(sk
, &ipv6h
->saddr
);
909 } else if (hash_location
) {
911 * active side is lost. Try to find listening socket through
912 * source port, and then find md5 key through listening socket.
913 * we are not loose security here:
914 * Incoming packet is checked with md5 hash with finding key,
915 * no RST generated if md5 hash doesn't match.
917 sk1
= inet6_lookup_listener(dev_net(skb_dst(skb
)->dev
),
918 &tcp_hashinfo
, NULL
, 0,
920 th
->source
, &ipv6h
->daddr
,
922 tcp_v6_iif_l3_slave(skb
),
927 key
= tcp_v6_md5_do_lookup(sk1
, &ipv6h
->saddr
);
931 genhash
= tcp_v6_md5_hash_skb(newhash
, key
, NULL
, skb
);
932 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
938 seq
= ntohl(th
->ack_seq
);
940 ack_seq
= ntohl(th
->seq
) + th
->syn
+ th
->fin
+ skb
->len
-
943 oif
= sk
? sk
->sk_bound_dev_if
: 0;
944 tcp_v6_send_response(sk
, skb
, seq
, ack_seq
, 0, 0, 0, oif
, key
, 1, 0, 0);
946 #ifdef CONFIG_TCP_MD5SIG
952 static void tcp_v6_send_ack(const struct sock
*sk
, struct sk_buff
*skb
, u32 seq
,
953 u32 ack
, u32 win
, u32 tsval
, u32 tsecr
, int oif
,
954 struct tcp_md5sig_key
*key
, u8 tclass
,
957 tcp_v6_send_response(sk
, skb
, seq
, ack
, win
, tsval
, tsecr
, oif
, key
, 0,
961 static void tcp_v6_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
963 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
964 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
966 tcp_v6_send_ack(sk
, skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
967 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
968 tcp_time_stamp_raw() + tcptw
->tw_ts_offset
,
969 tcptw
->tw_ts_recent
, tw
->tw_bound_dev_if
, tcp_twsk_md5_key(tcptw
),
970 tw
->tw_tclass
, cpu_to_be32(tw
->tw_flowlabel
));
975 static void tcp_v6_reqsk_send_ack(const struct sock
*sk
, struct sk_buff
*skb
,
976 struct request_sock
*req
)
978 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
979 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
982 * The window field (SEG.WND) of every outgoing segment, with the
983 * exception of <SYN> segments, MUST be right-shifted by
984 * Rcv.Wind.Shift bits:
986 tcp_v6_send_ack(sk
, skb
, (sk
->sk_state
== TCP_LISTEN
) ?
987 tcp_rsk(req
)->snt_isn
+ 1 : tcp_sk(sk
)->snd_nxt
,
988 tcp_rsk(req
)->rcv_nxt
,
989 req
->rsk_rcv_wnd
>> inet_rsk(req
)->rcv_wscale
,
990 tcp_time_stamp_raw() + tcp_rsk(req
)->ts_off
,
991 req
->ts_recent
, sk
->sk_bound_dev_if
,
992 tcp_v6_md5_do_lookup(sk
, &ipv6_hdr(skb
)->saddr
),
997 static struct sock
*tcp_v6_cookie_check(struct sock
*sk
, struct sk_buff
*skb
)
999 #ifdef CONFIG_SYN_COOKIES
1000 const struct tcphdr
*th
= tcp_hdr(skb
);
1003 sk
= cookie_v6_check(sk
, skb
);
1008 static int tcp_v6_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1010 if (skb
->protocol
== htons(ETH_P_IP
))
1011 return tcp_v4_conn_request(sk
, skb
);
1013 if (!ipv6_unicast_destination(skb
))
1016 return tcp_conn_request(&tcp6_request_sock_ops
,
1017 &tcp_request_sock_ipv6_ops
, sk
, skb
);
1021 return 0; /* don't send reset */
1024 static void tcp_v6_restore_cb(struct sk_buff
*skb
)
1026 /* We need to move header back to the beginning if xfrm6_policy_check()
1027 * and tcp_v6_fill_cb() are going to be called again.
1028 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1030 memmove(IP6CB(skb
), &TCP_SKB_CB(skb
)->header
.h6
,
1031 sizeof(struct inet6_skb_parm
));
1034 static struct sock
*tcp_v6_syn_recv_sock(const struct sock
*sk
, struct sk_buff
*skb
,
1035 struct request_sock
*req
,
1036 struct dst_entry
*dst
,
1037 struct request_sock
*req_unhash
,
1040 struct inet_request_sock
*ireq
;
1041 struct ipv6_pinfo
*newnp
;
1042 const struct ipv6_pinfo
*np
= inet6_sk(sk
);
1043 struct ipv6_txoptions
*opt
;
1044 struct tcp6_sock
*newtcp6sk
;
1045 struct inet_sock
*newinet
;
1046 struct tcp_sock
*newtp
;
1048 #ifdef CONFIG_TCP_MD5SIG
1049 struct tcp_md5sig_key
*key
;
1053 if (skb
->protocol
== htons(ETH_P_IP
)) {
1058 newsk
= tcp_v4_syn_recv_sock(sk
, skb
, req
, dst
,
1059 req_unhash
, own_req
);
1064 newtcp6sk
= (struct tcp6_sock
*)newsk
;
1065 inet_sk(newsk
)->pinet6
= &newtcp6sk
->inet6
;
1067 newinet
= inet_sk(newsk
);
1068 newnp
= inet6_sk(newsk
);
1069 newtp
= tcp_sk(newsk
);
1071 memcpy(newnp
, np
, sizeof(struct ipv6_pinfo
));
1073 newnp
->saddr
= newsk
->sk_v6_rcv_saddr
;
1075 inet_csk(newsk
)->icsk_af_ops
= &ipv6_mapped
;
1076 newsk
->sk_backlog_rcv
= tcp_v4_do_rcv
;
1077 #ifdef CONFIG_TCP_MD5SIG
1078 newtp
->af_specific
= &tcp_sock_ipv6_mapped_specific
;
1081 newnp
->ipv6_mc_list
= NULL
;
1082 newnp
->ipv6_ac_list
= NULL
;
1083 newnp
->ipv6_fl_list
= NULL
;
1084 newnp
->pktoptions
= NULL
;
1086 newnp
->mcast_oif
= inet_iif(skb
);
1087 newnp
->mcast_hops
= ip_hdr(skb
)->ttl
;
1088 newnp
->rcv_flowinfo
= 0;
1090 newnp
->flow_label
= 0;
1093 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1094 * here, tcp_create_openreq_child now does this for us, see the comment in
1095 * that function for the gory details. -acme
1098 /* It is tricky place. Until this moment IPv4 tcp
1099 worked with IPv6 icsk.icsk_af_ops.
1102 tcp_sync_mss(newsk
, inet_csk(newsk
)->icsk_pmtu_cookie
);
1107 ireq
= inet_rsk(req
);
1109 if (sk_acceptq_is_full(sk
))
1113 dst
= inet6_csk_route_req(sk
, &fl6
, req
, IPPROTO_TCP
);
1118 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1123 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1124 * count here, tcp_create_openreq_child now does this for us, see the
1125 * comment in that function for the gory details. -acme
1128 newsk
->sk_gso_type
= SKB_GSO_TCPV6
;
1129 ip6_dst_store(newsk
, dst
, NULL
, NULL
);
1130 inet6_sk_rx_dst_set(newsk
, skb
);
1132 newtcp6sk
= (struct tcp6_sock
*)newsk
;
1133 inet_sk(newsk
)->pinet6
= &newtcp6sk
->inet6
;
1135 newtp
= tcp_sk(newsk
);
1136 newinet
= inet_sk(newsk
);
1137 newnp
= inet6_sk(newsk
);
1139 memcpy(newnp
, np
, sizeof(struct ipv6_pinfo
));
1141 newsk
->sk_v6_daddr
= ireq
->ir_v6_rmt_addr
;
1142 newnp
->saddr
= ireq
->ir_v6_loc_addr
;
1143 newsk
->sk_v6_rcv_saddr
= ireq
->ir_v6_loc_addr
;
1144 newsk
->sk_bound_dev_if
= ireq
->ir_iif
;
1146 /* Now IPv6 options...
1148 First: no IPv4 options.
1150 newinet
->inet_opt
= NULL
;
1151 newnp
->ipv6_mc_list
= NULL
;
1152 newnp
->ipv6_ac_list
= NULL
;
1153 newnp
->ipv6_fl_list
= NULL
;
1156 newnp
->rxopt
.all
= np
->rxopt
.all
;
1158 newnp
->pktoptions
= NULL
;
1160 newnp
->mcast_oif
= tcp_v6_iif(skb
);
1161 newnp
->mcast_hops
= ipv6_hdr(skb
)->hop_limit
;
1162 newnp
->rcv_flowinfo
= ip6_flowinfo(ipv6_hdr(skb
));
1164 newnp
->flow_label
= ip6_flowlabel(ipv6_hdr(skb
));
1166 /* Clone native IPv6 options from listening socket (if any)
1168 Yes, keeping reference count would be much more clever,
1169 but we make one more one thing there: reattach optmem
1172 opt
= ireq
->ipv6_opt
;
1174 opt
= rcu_dereference(np
->opt
);
1176 opt
= ipv6_dup_options(newsk
, opt
);
1177 RCU_INIT_POINTER(newnp
->opt
, opt
);
1179 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1181 inet_csk(newsk
)->icsk_ext_hdr_len
= opt
->opt_nflen
+
1184 tcp_ca_openreq_child(newsk
, dst
);
1186 tcp_sync_mss(newsk
, dst_mtu(dst
));
1187 newtp
->advmss
= tcp_mss_clamp(tcp_sk(sk
), dst_metric_advmss(dst
));
1189 tcp_initialize_rcv_mss(newsk
);
1191 newinet
->inet_daddr
= newinet
->inet_saddr
= LOOPBACK4_IPV6
;
1192 newinet
->inet_rcv_saddr
= LOOPBACK4_IPV6
;
1194 #ifdef CONFIG_TCP_MD5SIG
1195 /* Copy over the MD5 key from the original socket */
1196 key
= tcp_v6_md5_do_lookup(sk
, &newsk
->sk_v6_daddr
);
1198 /* We're using one, so create a matching key
1199 * on the newsk structure. If we fail to get
1200 * memory, then we end up not copying the key
1203 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newsk
->sk_v6_daddr
,
1204 AF_INET6
, 128, key
->key
, key
->keylen
,
1205 sk_gfp_mask(sk
, GFP_ATOMIC
));
1209 if (__inet_inherit_port(sk
, newsk
) < 0) {
1210 inet_csk_prepare_forced_close(newsk
);
1214 *own_req
= inet_ehash_nolisten(newsk
, req_to_sk(req_unhash
));
1216 tcp_move_syn(newtp
, req
);
1218 /* Clone pktoptions received with SYN, if we own the req */
1219 if (ireq
->pktopts
) {
1220 newnp
->pktoptions
= skb_clone(ireq
->pktopts
,
1221 sk_gfp_mask(sk
, GFP_ATOMIC
));
1222 consume_skb(ireq
->pktopts
);
1223 ireq
->pktopts
= NULL
;
1224 if (newnp
->pktoptions
) {
1225 tcp_v6_restore_cb(newnp
->pktoptions
);
1226 skb_set_owner_r(newnp
->pktoptions
, newsk
);
1234 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1242 /* The socket must have it's spinlock held when we get
1243 * here, unless it is a TCP_LISTEN socket.
1245 * We have a potential double-lock case here, so even when
1246 * doing backlog processing we use the BH locking scheme.
1247 * This is because we cannot sleep with the original spinlock
1250 static int tcp_v6_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1252 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1253 struct tcp_sock
*tp
;
1254 struct sk_buff
*opt_skb
= NULL
;
1256 /* Imagine: socket is IPv6. IPv4 packet arrives,
1257 goes to IPv4 receive handler and backlogged.
1258 From backlog it always goes here. Kerboom...
1259 Fortunately, tcp_rcv_established and rcv_established
1260 handle them correctly, but it is not case with
1261 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1264 if (skb
->protocol
== htons(ETH_P_IP
))
1265 return tcp_v4_do_rcv(sk
, skb
);
1268 * socket locking is here for SMP purposes as backlog rcv
1269 * is currently called with bh processing disabled.
1272 /* Do Stevens' IPV6_PKTOPTIONS.
1274 Yes, guys, it is the only place in our code, where we
1275 may make it not affecting IPv4.
1276 The rest of code is protocol independent,
1277 and I do not like idea to uglify IPv4.
1279 Actually, all the idea behind IPV6_PKTOPTIONS
1280 looks not very well thought. For now we latch
1281 options, received in the last packet, enqueued
1282 by tcp. Feel free to propose better solution.
1286 opt_skb
= skb_clone(skb
, sk_gfp_mask(sk
, GFP_ATOMIC
));
1288 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1289 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1291 sock_rps_save_rxhash(sk
, skb
);
1292 sk_mark_napi_id(sk
, skb
);
1294 if (inet_sk(sk
)->rx_dst_ifindex
!= skb
->skb_iif
||
1295 dst
->ops
->check(dst
, np
->rx_dst_cookie
) == NULL
) {
1297 sk
->sk_rx_dst
= NULL
;
1301 tcp_rcv_established(sk
, skb
, tcp_hdr(skb
));
1303 goto ipv6_pktoptions
;
1307 if (tcp_checksum_complete(skb
))
1310 if (sk
->sk_state
== TCP_LISTEN
) {
1311 struct sock
*nsk
= tcp_v6_cookie_check(sk
, skb
);
1317 if (tcp_child_process(sk
, nsk
, skb
))
1320 __kfree_skb(opt_skb
);
1324 sock_rps_save_rxhash(sk
, skb
);
1326 if (tcp_rcv_state_process(sk
, skb
))
1329 goto ipv6_pktoptions
;
1333 tcp_v6_send_reset(sk
, skb
);
1336 __kfree_skb(opt_skb
);
1340 TCP_INC_STATS(sock_net(sk
), TCP_MIB_CSUMERRORS
);
1341 TCP_INC_STATS(sock_net(sk
), TCP_MIB_INERRS
);
1346 /* Do you ask, what is it?
1348 1. skb was enqueued by tcp.
1349 2. skb is added to tail of read queue, rather than out of order.
1350 3. socket is not in passive state.
1351 4. Finally, it really contains options, which user wants to receive.
1354 if (TCP_SKB_CB(opt_skb
)->end_seq
== tp
->rcv_nxt
&&
1355 !((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
))) {
1356 if (np
->rxopt
.bits
.rxinfo
|| np
->rxopt
.bits
.rxoinfo
)
1357 np
->mcast_oif
= tcp_v6_iif(opt_skb
);
1358 if (np
->rxopt
.bits
.rxhlim
|| np
->rxopt
.bits
.rxohlim
)
1359 np
->mcast_hops
= ipv6_hdr(opt_skb
)->hop_limit
;
1360 if (np
->rxopt
.bits
.rxflow
|| np
->rxopt
.bits
.rxtclass
)
1361 np
->rcv_flowinfo
= ip6_flowinfo(ipv6_hdr(opt_skb
));
1363 np
->flow_label
= ip6_flowlabel(ipv6_hdr(opt_skb
));
1364 if (ipv6_opt_accepted(sk
, opt_skb
, &TCP_SKB_CB(opt_skb
)->header
.h6
)) {
1365 skb_set_owner_r(opt_skb
, sk
);
1366 tcp_v6_restore_cb(opt_skb
);
1367 opt_skb
= xchg(&np
->pktoptions
, opt_skb
);
1369 __kfree_skb(opt_skb
);
1370 opt_skb
= xchg(&np
->pktoptions
, NULL
);
1378 static void tcp_v6_fill_cb(struct sk_buff
*skb
, const struct ipv6hdr
*hdr
,
1379 const struct tcphdr
*th
)
1381 /* This is tricky: we move IP6CB at its correct location into
1382 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1383 * _decode_session6() uses IP6CB().
1384 * barrier() makes sure compiler won't play aliasing games.
1386 memmove(&TCP_SKB_CB(skb
)->header
.h6
, IP6CB(skb
),
1387 sizeof(struct inet6_skb_parm
));
1390 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1391 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1392 skb
->len
- th
->doff
*4);
1393 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1394 TCP_SKB_CB(skb
)->tcp_flags
= tcp_flag_byte(th
);
1395 TCP_SKB_CB(skb
)->tcp_tw_isn
= 0;
1396 TCP_SKB_CB(skb
)->ip_dsfield
= ipv6_get_dsfield(hdr
);
1397 TCP_SKB_CB(skb
)->sacked
= 0;
1398 TCP_SKB_CB(skb
)->has_rxtstamp
=
1399 skb
->tstamp
|| skb_hwtstamps(skb
)->hwtstamp
;
1402 static int tcp_v6_rcv(struct sk_buff
*skb
)
1404 int sdif
= inet6_sdif(skb
);
1405 const struct tcphdr
*th
;
1406 const struct ipv6hdr
*hdr
;
1410 struct net
*net
= dev_net(skb
->dev
);
1412 if (skb
->pkt_type
!= PACKET_HOST
)
1416 * Count it even if it's bad.
1418 __TCP_INC_STATS(net
, TCP_MIB_INSEGS
);
1420 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1423 th
= (const struct tcphdr
*)skb
->data
;
1425 if (unlikely(th
->doff
< sizeof(struct tcphdr
)/4))
1427 if (!pskb_may_pull(skb
, th
->doff
*4))
1430 if (skb_checksum_init(skb
, IPPROTO_TCP
, ip6_compute_pseudo
))
1433 th
= (const struct tcphdr
*)skb
->data
;
1434 hdr
= ipv6_hdr(skb
);
1437 sk
= __inet6_lookup_skb(&tcp_hashinfo
, skb
, __tcp_hdrlen(th
),
1438 th
->source
, th
->dest
, inet6_iif(skb
), sdif
,
1444 if (sk
->sk_state
== TCP_TIME_WAIT
)
1447 if (sk
->sk_state
== TCP_NEW_SYN_RECV
) {
1448 struct request_sock
*req
= inet_reqsk(sk
);
1451 sk
= req
->rsk_listener
;
1452 if (tcp_v6_inbound_md5_hash(sk
, skb
)) {
1453 sk_drops_add(sk
, skb
);
1457 if (tcp_checksum_complete(skb
)) {
1461 if (unlikely(sk
->sk_state
!= TCP_LISTEN
)) {
1462 inet_csk_reqsk_queue_drop_and_put(sk
, req
);
1468 if (!tcp_filter(sk
, skb
)) {
1469 th
= (const struct tcphdr
*)skb
->data
;
1470 hdr
= ipv6_hdr(skb
);
1471 tcp_v6_fill_cb(skb
, hdr
, th
);
1472 nsk
= tcp_check_req(sk
, skb
, req
, false);
1476 goto discard_and_relse
;
1480 tcp_v6_restore_cb(skb
);
1481 } else if (tcp_child_process(sk
, nsk
, skb
)) {
1482 tcp_v6_send_reset(nsk
, skb
);
1483 goto discard_and_relse
;
1489 if (hdr
->hop_limit
< inet6_sk(sk
)->min_hopcount
) {
1490 __NET_INC_STATS(net
, LINUX_MIB_TCPMINTTLDROP
);
1491 goto discard_and_relse
;
1494 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1495 goto discard_and_relse
;
1497 if (tcp_v6_inbound_md5_hash(sk
, skb
))
1498 goto discard_and_relse
;
1500 if (tcp_filter(sk
, skb
))
1501 goto discard_and_relse
;
1502 th
= (const struct tcphdr
*)skb
->data
;
1503 hdr
= ipv6_hdr(skb
);
1504 tcp_v6_fill_cb(skb
, hdr
, th
);
1508 if (sk
->sk_state
== TCP_LISTEN
) {
1509 ret
= tcp_v6_do_rcv(sk
, skb
);
1510 goto put_and_return
;
1513 sk_incoming_cpu_update(sk
);
1515 bh_lock_sock_nested(sk
);
1516 tcp_segs_in(tcp_sk(sk
), skb
);
1518 if (!sock_owned_by_user(sk
)) {
1519 ret
= tcp_v6_do_rcv(sk
, skb
);
1520 } else if (tcp_add_backlog(sk
, skb
)) {
1521 goto discard_and_relse
;
1528 return ret
? -1 : 0;
1531 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1534 tcp_v6_fill_cb(skb
, hdr
, th
);
1536 if (tcp_checksum_complete(skb
)) {
1538 __TCP_INC_STATS(net
, TCP_MIB_CSUMERRORS
);
1540 __TCP_INC_STATS(net
, TCP_MIB_INERRS
);
1542 tcp_v6_send_reset(NULL
, skb
);
1550 sk_drops_add(sk
, skb
);
1556 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1557 inet_twsk_put(inet_twsk(sk
));
1561 tcp_v6_fill_cb(skb
, hdr
, th
);
1563 if (tcp_checksum_complete(skb
)) {
1564 inet_twsk_put(inet_twsk(sk
));
1568 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1573 sk2
= inet6_lookup_listener(dev_net(skb
->dev
), &tcp_hashinfo
,
1574 skb
, __tcp_hdrlen(th
),
1575 &ipv6_hdr(skb
)->saddr
, th
->source
,
1576 &ipv6_hdr(skb
)->daddr
,
1578 tcp_v6_iif_l3_slave(skb
),
1581 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
1582 inet_twsk_deschedule_put(tw
);
1584 tcp_v6_restore_cb(skb
);
1588 /* Fall through to ACK */
1591 tcp_v6_timewait_ack(sk
, skb
);
1594 tcp_v6_send_reset(sk
, skb
);
1595 inet_twsk_deschedule_put(inet_twsk(sk
));
1597 case TCP_TW_SUCCESS
:
1603 static void tcp_v6_early_demux(struct sk_buff
*skb
)
1605 const struct ipv6hdr
*hdr
;
1606 const struct tcphdr
*th
;
1609 if (skb
->pkt_type
!= PACKET_HOST
)
1612 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct tcphdr
)))
1615 hdr
= ipv6_hdr(skb
);
1618 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1621 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1622 sk
= __inet6_lookup_established(dev_net(skb
->dev
), &tcp_hashinfo
,
1623 &hdr
->saddr
, th
->source
,
1624 &hdr
->daddr
, ntohs(th
->dest
),
1625 inet6_iif(skb
), inet6_sdif(skb
));
1628 skb
->destructor
= sock_edemux
;
1629 if (sk_fullsock(sk
)) {
1630 struct dst_entry
*dst
= READ_ONCE(sk
->sk_rx_dst
);
1633 dst
= dst_check(dst
, inet6_sk(sk
)->rx_dst_cookie
);
1635 inet_sk(sk
)->rx_dst_ifindex
== skb
->skb_iif
)
1636 skb_dst_set_noref(skb
, dst
);
1641 static struct timewait_sock_ops tcp6_timewait_sock_ops
= {
1642 .twsk_obj_size
= sizeof(struct tcp6_timewait_sock
),
1643 .twsk_unique
= tcp_twsk_unique
,
1644 .twsk_destructor
= tcp_twsk_destructor
,
1647 static const struct inet_connection_sock_af_ops ipv6_specific
= {
1648 .queue_xmit
= inet6_csk_xmit
,
1649 .send_check
= tcp_v6_send_check
,
1650 .rebuild_header
= inet6_sk_rebuild_header
,
1651 .sk_rx_dst_set
= inet6_sk_rx_dst_set
,
1652 .conn_request
= tcp_v6_conn_request
,
1653 .syn_recv_sock
= tcp_v6_syn_recv_sock
,
1654 .net_header_len
= sizeof(struct ipv6hdr
),
1655 .net_frag_header_len
= sizeof(struct frag_hdr
),
1656 .setsockopt
= ipv6_setsockopt
,
1657 .getsockopt
= ipv6_getsockopt
,
1658 .addr2sockaddr
= inet6_csk_addr2sockaddr
,
1659 .sockaddr_len
= sizeof(struct sockaddr_in6
),
1660 #ifdef CONFIG_COMPAT
1661 .compat_setsockopt
= compat_ipv6_setsockopt
,
1662 .compat_getsockopt
= compat_ipv6_getsockopt
,
1664 .mtu_reduced
= tcp_v6_mtu_reduced
,
1667 #ifdef CONFIG_TCP_MD5SIG
1668 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific
= {
1669 .md5_lookup
= tcp_v6_md5_lookup
,
1670 .calc_md5_hash
= tcp_v6_md5_hash_skb
,
1671 .md5_parse
= tcp_v6_parse_md5_keys
,
1676 * TCP over IPv4 via INET6 API
1678 static const struct inet_connection_sock_af_ops ipv6_mapped
= {
1679 .queue_xmit
= ip_queue_xmit
,
1680 .send_check
= tcp_v4_send_check
,
1681 .rebuild_header
= inet_sk_rebuild_header
,
1682 .sk_rx_dst_set
= inet_sk_rx_dst_set
,
1683 .conn_request
= tcp_v6_conn_request
,
1684 .syn_recv_sock
= tcp_v6_syn_recv_sock
,
1685 .net_header_len
= sizeof(struct iphdr
),
1686 .setsockopt
= ipv6_setsockopt
,
1687 .getsockopt
= ipv6_getsockopt
,
1688 .addr2sockaddr
= inet6_csk_addr2sockaddr
,
1689 .sockaddr_len
= sizeof(struct sockaddr_in6
),
1690 #ifdef CONFIG_COMPAT
1691 .compat_setsockopt
= compat_ipv6_setsockopt
,
1692 .compat_getsockopt
= compat_ipv6_getsockopt
,
1694 .mtu_reduced
= tcp_v4_mtu_reduced
,
1697 #ifdef CONFIG_TCP_MD5SIG
1698 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific
= {
1699 .md5_lookup
= tcp_v4_md5_lookup
,
1700 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1701 .md5_parse
= tcp_v6_parse_md5_keys
,
1705 /* NOTE: A lot of things set to zero explicitly by call to
1706 * sk_alloc() so need not be done here.
1708 static int tcp_v6_init_sock(struct sock
*sk
)
1710 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1714 icsk
->icsk_af_ops
= &ipv6_specific
;
1716 #ifdef CONFIG_TCP_MD5SIG
1717 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv6_specific
;
1723 static void tcp_v6_destroy_sock(struct sock
*sk
)
1725 tcp_v4_destroy_sock(sk
);
1726 inet6_destroy_sock(sk
);
1729 #ifdef CONFIG_PROC_FS
1730 /* Proc filesystem TCPv6 sock list dumping. */
1731 static void get_openreq6(struct seq_file
*seq
,
1732 const struct request_sock
*req
, int i
)
1734 long ttd
= req
->rsk_timer
.expires
- jiffies
;
1735 const struct in6_addr
*src
= &inet_rsk(req
)->ir_v6_loc_addr
;
1736 const struct in6_addr
*dest
= &inet_rsk(req
)->ir_v6_rmt_addr
;
1742 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1743 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1745 src
->s6_addr32
[0], src
->s6_addr32
[1],
1746 src
->s6_addr32
[2], src
->s6_addr32
[3],
1747 inet_rsk(req
)->ir_num
,
1748 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1749 dest
->s6_addr32
[2], dest
->s6_addr32
[3],
1750 ntohs(inet_rsk(req
)->ir_rmt_port
),
1752 0, 0, /* could print option size, but that is af dependent. */
1753 1, /* timers active (only the expire timer) */
1754 jiffies_to_clock_t(ttd
),
1756 from_kuid_munged(seq_user_ns(seq
),
1757 sock_i_uid(req
->rsk_listener
)),
1758 0, /* non standard timer */
1759 0, /* open_requests have no inode */
1763 static void get_tcp6_sock(struct seq_file
*seq
, struct sock
*sp
, int i
)
1765 const struct in6_addr
*dest
, *src
;
1768 unsigned long timer_expires
;
1769 const struct inet_sock
*inet
= inet_sk(sp
);
1770 const struct tcp_sock
*tp
= tcp_sk(sp
);
1771 const struct inet_connection_sock
*icsk
= inet_csk(sp
);
1772 const struct fastopen_queue
*fastopenq
= &icsk
->icsk_accept_queue
.fastopenq
;
1776 dest
= &sp
->sk_v6_daddr
;
1777 src
= &sp
->sk_v6_rcv_saddr
;
1778 destp
= ntohs(inet
->inet_dport
);
1779 srcp
= ntohs(inet
->inet_sport
);
1781 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
||
1782 icsk
->icsk_pending
== ICSK_TIME_REO_TIMEOUT
||
1783 icsk
->icsk_pending
== ICSK_TIME_LOSS_PROBE
) {
1785 timer_expires
= icsk
->icsk_timeout
;
1786 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
1788 timer_expires
= icsk
->icsk_timeout
;
1789 } else if (timer_pending(&sp
->sk_timer
)) {
1791 timer_expires
= sp
->sk_timer
.expires
;
1794 timer_expires
= jiffies
;
1797 state
= sk_state_load(sp
);
1798 if (state
== TCP_LISTEN
)
1799 rx_queue
= sp
->sk_ack_backlog
;
1801 /* Because we don't lock the socket,
1802 * we might find a transient negative value.
1804 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
1807 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1808 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1810 src
->s6_addr32
[0], src
->s6_addr32
[1],
1811 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1812 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1813 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1815 tp
->write_seq
- tp
->snd_una
,
1818 jiffies_delta_to_clock_t(timer_expires
- jiffies
),
1819 icsk
->icsk_retransmits
,
1820 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(sp
)),
1821 icsk
->icsk_probes_out
,
1823 refcount_read(&sp
->sk_refcnt
), sp
,
1824 jiffies_to_clock_t(icsk
->icsk_rto
),
1825 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
1826 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
1828 state
== TCP_LISTEN
?
1829 fastopenq
->max_qlen
:
1830 (tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
)
1834 static void get_timewait6_sock(struct seq_file
*seq
,
1835 struct inet_timewait_sock
*tw
, int i
)
1837 long delta
= tw
->tw_timer
.expires
- jiffies
;
1838 const struct in6_addr
*dest
, *src
;
1841 dest
= &tw
->tw_v6_daddr
;
1842 src
= &tw
->tw_v6_rcv_saddr
;
1843 destp
= ntohs(tw
->tw_dport
);
1844 srcp
= ntohs(tw
->tw_sport
);
1847 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1848 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1850 src
->s6_addr32
[0], src
->s6_addr32
[1],
1851 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1852 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1853 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1854 tw
->tw_substate
, 0, 0,
1855 3, jiffies_delta_to_clock_t(delta
), 0, 0, 0, 0,
1856 refcount_read(&tw
->tw_refcnt
), tw
);
1859 static int tcp6_seq_show(struct seq_file
*seq
, void *v
)
1861 struct tcp_iter_state
*st
;
1862 struct sock
*sk
= v
;
1864 if (v
== SEQ_START_TOKEN
) {
1869 "st tx_queue rx_queue tr tm->when retrnsmt"
1870 " uid timeout inode\n");
1875 if (sk
->sk_state
== TCP_TIME_WAIT
)
1876 get_timewait6_sock(seq
, v
, st
->num
);
1877 else if (sk
->sk_state
== TCP_NEW_SYN_RECV
)
1878 get_openreq6(seq
, v
, st
->num
);
1880 get_tcp6_sock(seq
, v
, st
->num
);
1885 static const struct file_operations tcp6_afinfo_seq_fops
= {
1886 .owner
= THIS_MODULE
,
1887 .open
= tcp_seq_open
,
1889 .llseek
= seq_lseek
,
1890 .release
= seq_release_net
1893 static struct tcp_seq_afinfo tcp6_seq_afinfo
= {
1896 .seq_fops
= &tcp6_afinfo_seq_fops
,
1898 .show
= tcp6_seq_show
,
1902 int __net_init
tcp6_proc_init(struct net
*net
)
1904 return tcp_proc_register(net
, &tcp6_seq_afinfo
);
1907 void tcp6_proc_exit(struct net
*net
)
1909 tcp_proc_unregister(net
, &tcp6_seq_afinfo
);
1913 struct proto tcpv6_prot
= {
1915 .owner
= THIS_MODULE
,
1917 .connect
= tcp_v6_connect
,
1918 .disconnect
= tcp_disconnect
,
1919 .accept
= inet_csk_accept
,
1921 .init
= tcp_v6_init_sock
,
1922 .destroy
= tcp_v6_destroy_sock
,
1923 .shutdown
= tcp_shutdown
,
1924 .setsockopt
= tcp_setsockopt
,
1925 .getsockopt
= tcp_getsockopt
,
1926 .keepalive
= tcp_set_keepalive
,
1927 .recvmsg
= tcp_recvmsg
,
1928 .sendmsg
= tcp_sendmsg
,
1929 .sendpage
= tcp_sendpage
,
1930 .backlog_rcv
= tcp_v6_do_rcv
,
1931 .release_cb
= tcp_release_cb
,
1933 .unhash
= inet_unhash
,
1934 .get_port
= inet_csk_get_port
,
1935 .enter_memory_pressure
= tcp_enter_memory_pressure
,
1936 .leave_memory_pressure
= tcp_leave_memory_pressure
,
1937 .stream_memory_free
= tcp_stream_memory_free
,
1938 .sockets_allocated
= &tcp_sockets_allocated
,
1939 .memory_allocated
= &tcp_memory_allocated
,
1940 .memory_pressure
= &tcp_memory_pressure
,
1941 .orphan_count
= &tcp_orphan_count
,
1942 .sysctl_mem
= sysctl_tcp_mem
,
1943 .sysctl_wmem
= sysctl_tcp_wmem
,
1944 .sysctl_rmem
= sysctl_tcp_rmem
,
1945 .max_header
= MAX_TCP_HEADER
,
1946 .obj_size
= sizeof(struct tcp6_sock
),
1947 .slab_flags
= SLAB_TYPESAFE_BY_RCU
,
1948 .twsk_prot
= &tcp6_timewait_sock_ops
,
1949 .rsk_prot
= &tcp6_request_sock_ops
,
1950 .h
.hashinfo
= &tcp_hashinfo
,
1951 .no_autobind
= true,
1952 #ifdef CONFIG_COMPAT
1953 .compat_setsockopt
= compat_tcp_setsockopt
,
1954 .compat_getsockopt
= compat_tcp_getsockopt
,
1956 .diag_destroy
= tcp_abort
,
1959 /* thinking of making this const? Don't.
1960 * early_demux can change based on sysctl.
1962 static struct inet6_protocol tcpv6_protocol
= {
1963 .early_demux
= tcp_v6_early_demux
,
1964 .early_demux_handler
= tcp_v6_early_demux
,
1965 .handler
= tcp_v6_rcv
,
1966 .err_handler
= tcp_v6_err
,
1967 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1970 static struct inet_protosw tcpv6_protosw
= {
1971 .type
= SOCK_STREAM
,
1972 .protocol
= IPPROTO_TCP
,
1973 .prot
= &tcpv6_prot
,
1974 .ops
= &inet6_stream_ops
,
1975 .flags
= INET_PROTOSW_PERMANENT
|
1979 static int __net_init
tcpv6_net_init(struct net
*net
)
1981 return inet_ctl_sock_create(&net
->ipv6
.tcp_sk
, PF_INET6
,
1982 SOCK_RAW
, IPPROTO_TCP
, net
);
1985 static void __net_exit
tcpv6_net_exit(struct net
*net
)
1987 inet_ctl_sock_destroy(net
->ipv6
.tcp_sk
);
1990 static void __net_exit
tcpv6_net_exit_batch(struct list_head
*net_exit_list
)
1992 inet_twsk_purge(&tcp_hashinfo
, AF_INET6
);
1995 static struct pernet_operations tcpv6_net_ops
= {
1996 .init
= tcpv6_net_init
,
1997 .exit
= tcpv6_net_exit
,
1998 .exit_batch
= tcpv6_net_exit_batch
,
2001 int __init
tcpv6_init(void)
2005 ret
= inet6_add_protocol(&tcpv6_protocol
, IPPROTO_TCP
);
2009 /* register inet6 protocol */
2010 ret
= inet6_register_protosw(&tcpv6_protosw
);
2012 goto out_tcpv6_protocol
;
2014 ret
= register_pernet_subsys(&tcpv6_net_ops
);
2016 goto out_tcpv6_protosw
;
2021 inet6_unregister_protosw(&tcpv6_protosw
);
2023 inet6_del_protocol(&tcpv6_protocol
, IPPROTO_TCP
);
2027 void tcpv6_exit(void)
2029 unregister_pernet_subsys(&tcpv6_net_ops
);
2030 inet6_unregister_protosw(&tcpv6_protosw
);
2031 inet6_del_protocol(&tcpv6_protocol
, IPPROTO_TCP
);