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_sequence(const struct sk_buff
*skb
, u32
*tsoff
)
106 return secure_tcpv6_sequence_number(ipv6_hdr(skb
)->daddr
.s6_addr32
,
107 ipv6_hdr(skb
)->saddr
.s6_addr32
,
109 tcp_hdr(skb
)->source
, tsoff
);
112 static int tcp_v6_connect(struct sock
*sk
, struct sockaddr
*uaddr
,
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
;
123 struct dst_entry
*dst
;
127 if (addr_len
< SIN6_LEN_RFC2133
)
130 if (usin
->sin6_family
!= AF_INET6
)
131 return -EAFNOSUPPORT
;
133 memset(&fl6
, 0, sizeof(fl6
));
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
);
143 fl6_sock_release(flowlabel
);
148 * connect() to INADDR_ANY means loopback (BSD'ism).
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
),
156 usin
->sin6_addr
= in6addr_loopback
;
159 addr_type
= ipv6_addr_type(&usin
->sin6_addr
);
161 if (addr_type
& IPV6_ADDR_MULTICAST
)
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
170 if (sk
->sk_bound_dev_if
&&
171 sk
->sk_bound_dev_if
!= usin
->sin6_scope_id
)
174 sk
->sk_bound_dev_if
= usin
->sin6_scope_id
;
177 /* Connect to link-local address requires an interface */
178 if (!sk
->sk_bound_dev_if
)
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;
189 sk
->sk_v6_daddr
= usin
->sin6_addr
;
190 np
->flow_label
= fl6
.flowlabel
;
196 if (addr_type
& IPV6_ADDR_MAPPED
) {
197 u32 exthdrlen
= icsk
->icsk_ext_hdr_len
;
198 struct sockaddr_in sin
;
200 SOCK_DEBUG(sk
, "connect: ipv4 mapped\n");
202 if (__ipv6_only_sock(sk
))
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];
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
;
215 err
= tcp_v4_connect(sk
, (struct sockaddr
*)&sin
, sizeof(sin
));
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
;
226 np
->saddr
= sk
->sk_v6_rcv_saddr
;
231 if (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
))
232 saddr
= &sk
->sk_v6_rcv_saddr
;
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
;
243 opt
= rcu_dereference_protected(np
->opt
, lockdep_sock_is_held(sk
));
244 final_p
= fl6_update_dst(&fl6
, opt
, &final
);
246 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
248 dst
= ip6_dst_lookup_flow(sk
, &fl6
, final_p
);
256 sk
->sk_v6_rcv_saddr
= *saddr
;
259 /* set the source address */
261 inet
->inet_rcv_saddr
= LOOPBACK4_IPV6
;
263 sk
->sk_gso_type
= SKB_GSO_TCPV6
;
264 ip6_dst_store(sk
, dst
, NULL
, NULL
);
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
);
271 icsk
->icsk_ext_hdr_len
= 0;
273 icsk
->icsk_ext_hdr_len
= opt
->opt_flen
+
276 tp
->rx_opt
.mss_clamp
= IPV6_MIN_MTU
- sizeof(struct tcphdr
) - sizeof(struct ipv6hdr
);
278 inet
->inet_dport
= usin
->sin6_port
;
280 tcp_set_state(sk
, TCP_SYN_SENT
);
281 err
= inet6_hash_connect(&tcp_death_row
, sk
);
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
,
294 err
= tcp_connect(sk
);
301 tcp_set_state(sk
, TCP_CLOSE
);
304 inet
->inet_dport
= 0;
305 sk
->sk_route_caps
= 0;
309 static void tcp_v6_mtu_reduced(struct sock
*sk
)
311 struct dst_entry
*dst
;
313 if ((1 << sk
->sk_state
) & (TCPF_LISTEN
| TCPF_CLOSE
))
316 dst
= inet6_csk_update_pmtu(sk
, tcp_sk(sk
)->mtu_info
);
320 if (inet_csk(sk
)->icsk_pmtu_cookie
> dst_mtu(dst
)) {
321 tcp_sync_mss(sk
, dst_mtu(dst
));
322 tcp_simple_retransmit(sk
);
326 static void tcp_v6_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
327 u8 type
, u8 code
, int offset
, __be32 info
)
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
;
340 sk
= __inet6_lookup_established(net
, &tcp_hashinfo
,
341 &hdr
->daddr
, th
->dest
,
342 &hdr
->saddr
, ntohs(th
->source
),
346 __ICMP6_INC_STATS(net
, __in6_dev_get(skb
->dev
),
351 if (sk
->sk_state
== TCP_TIME_WAIT
) {
352 inet_twsk_put(inet_twsk(sk
));
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
);
361 if (sock_owned_by_user(sk
) && type
!= ICMPV6_PKT_TOOBIG
)
362 __NET_INC_STATS(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
364 if (sk
->sk_state
== TCP_CLOSE
)
367 if (ipv6_hdr(skb
)->hop_limit
< inet6_sk(sk
)->min_hopcount
) {
368 __NET_INC_STATS(net
, LINUX_MIB_TCPMINTTLDROP
);
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
);
384 if (type
== NDISC_REDIRECT
) {
385 if (!sock_owned_by_user(sk
)) {
386 struct dst_entry
*dst
= __sk_dst_check(sk
, np
->dst_cookie
);
389 dst
->ops
->redirect(dst
, sk
, skb
);
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).
399 if (sk
->sk_state
== TCP_LISTEN
)
402 if (!ip6_sk_accept_pmtu(sk
))
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
,
415 /* Might be for an request_sock */
416 switch (sk
->sk_state
) {
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.
422 if (fastopen
&& !fastopen
->sk
)
425 if (!sock_owned_by_user(sk
)) {
427 sk
->sk_error_report(sk
); /* Wake people up to see the error (see connect in sock.c) */
431 sk
->sk_err_soft
= err
;
435 if (!sock_owned_by_user(sk
) && np
->recverr
) {
437 sk
->sk_error_report(sk
);
439 sk
->sk_err_soft
= err
;
447 static int tcp_v6_send_synack(const struct sock
*sk
, struct dst_entry
*dst
,
449 struct request_sock
*req
,
450 struct tcp_fastopen_cookie
*foc
,
451 enum tcp_synack_type synack_type
)
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
;
460 /* First, grab a route. */
461 if (!dst
&& (dst
= inet6_csk_route_req(sk
, fl6
, req
,
462 IPPROTO_TCP
)) == NULL
)
465 skb
= tcp_make_synack(sk
, dst
, req
, foc
, synack_type
);
468 __tcp_v6_send_check(skb
, &ireq
->ir_v6_loc_addr
,
469 &ireq
->ir_v6_rmt_addr
);
471 fl6
->daddr
= ireq
->ir_v6_rmt_addr
;
472 if (np
->repflow
&& ireq
->pktopts
)
473 fl6
->flowlabel
= ip6_flowlabel(ipv6_hdr(ireq
->pktopts
));
476 opt
= ireq
->ipv6_opt
;
478 opt
= rcu_dereference(np
->opt
);
479 err
= ip6_xmit(sk
, skb
, fl6
, sk
->sk_mark
, opt
, np
->tclass
);
481 err
= net_xmit_eval(err
);
489 static void tcp_v6_reqsk_destructor(struct request_sock
*req
)
491 kfree(inet_rsk(req
)->ipv6_opt
);
492 kfree_skb(inet_rsk(req
)->pktopts
);
495 #ifdef CONFIG_TCP_MD5SIG
496 static struct tcp_md5sig_key
*tcp_v6_md5_do_lookup(const struct sock
*sk
,
497 const struct in6_addr
*addr
)
499 return tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)addr
, AF_INET6
);
502 static struct tcp_md5sig_key
*tcp_v6_md5_lookup(const struct sock
*sk
,
503 const struct sock
*addr_sk
)
505 return tcp_v6_md5_do_lookup(sk
, &addr_sk
->sk_v6_daddr
);
508 static int tcp_v6_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
511 struct tcp_md5sig cmd
;
512 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*)&cmd
.tcpm_addr
;
514 if (optlen
< sizeof(cmd
))
517 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
520 if (sin6
->sin6_family
!= AF_INET6
)
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],
527 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
,
531 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
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
);
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
);
542 static 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
)
547 struct tcp6_pseudohdr
*bp
;
548 struct scatterlist sg
;
552 /* 1. TCP pseudo-header (RFC2460) */
555 bp
->protocol
= cpu_to_be32(IPPROTO_TCP
);
556 bp
->len
= cpu_to_be32(nbytes
);
558 _th
= (struct tcphdr
*)(bp
+ 1);
559 memcpy(_th
, th
, sizeof(*th
));
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
);
568 static 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
)
572 struct tcp_md5sig_pool
*hp
;
573 struct ahash_request
*req
;
575 hp
= tcp_get_md5sig_pool();
577 goto clear_hash_noput
;
580 if (crypto_ahash_init(req
))
582 if (tcp_v6_md5_hash_headers(hp
, daddr
, saddr
, th
, th
->doff
<< 2))
584 if (tcp_md5_hash_key(hp
, key
))
586 ahash_request_set_crypt(req
, NULL
, md5_hash
, 0);
587 if (crypto_ahash_final(req
))
590 tcp_put_md5sig_pool();
594 tcp_put_md5sig_pool();
596 memset(md5_hash
, 0, 16);
600 static 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
)
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
);
610 if (sk
) { /* valid for establish/request sockets */
611 saddr
= &sk
->sk_v6_rcv_saddr
;
612 daddr
= &sk
->sk_v6_daddr
;
614 const struct ipv6hdr
*ip6h
= ipv6_hdr(skb
);
615 saddr
= &ip6h
->saddr
;
616 daddr
= &ip6h
->daddr
;
619 hp
= tcp_get_md5sig_pool();
621 goto clear_hash_noput
;
624 if (crypto_ahash_init(req
))
627 if (tcp_v6_md5_hash_headers(hp
, daddr
, saddr
, th
, skb
->len
))
629 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
631 if (tcp_md5_hash_key(hp
, key
))
633 ahash_request_set_crypt(req
, NULL
, md5_hash
, 0);
634 if (crypto_ahash_final(req
))
637 tcp_put_md5sig_pool();
641 tcp_put_md5sig_pool();
643 memset(md5_hash
, 0, 16);
649 static bool tcp_v6_inbound_md5_hash(const struct sock
*sk
,
650 const struct sk_buff
*skb
)
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
);
660 hash_expected
= tcp_v6_md5_do_lookup(sk
, &ip6h
->saddr
);
661 hash_location
= tcp_parse_md5sig_option(th
);
663 /* We've parsed the options - do we have a hash? */
664 if (!hash_expected
&& !hash_location
)
667 if (hash_expected
&& !hash_location
) {
668 NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
672 if (!hash_expected
&& hash_location
) {
673 NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
677 /* check the signature */
678 genhash
= tcp_v6_md5_hash_skb(newhash
,
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
));
694 static void tcp_v6_init_req(struct request_sock
*req
,
695 const struct sock
*sk_listener
,
698 struct inet_request_sock
*ireq
= inet_rsk(req
);
699 const struct ipv6_pinfo
*np
= inet6_sk(sk_listener
);
701 ireq
->ir_v6_rmt_addr
= ipv6_hdr(skb
)->saddr
;
702 ireq
->ir_v6_loc_addr
= ipv6_hdr(skb
)->daddr
;
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
);
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
);
719 static struct dst_entry
*tcp_v6_route_req(const struct sock
*sk
,
721 const struct request_sock
*req
,
726 return inet6_csk_route_req(sk
, &fl
->u
.ip6
, req
, IPPROTO_TCP
);
729 struct request_sock_ops tcp6_request_sock_ops __read_mostly
= {
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
,
739 static 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
,
746 .init_req
= tcp_v6_init_req
,
747 #ifdef CONFIG_SYN_COOKIES
748 .cookie_init_seq
= cookie_v6_init_sequence
,
750 .route_req
= tcp_v6_route_req
,
751 .init_seq
= tcp_v6_init_sequence
,
752 .send_synack
= tcp_v6_send_synack
,
755 static 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
)
760 const struct tcphdr
*th
= tcp_hdr(skb
);
762 struct sk_buff
*buff
;
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
;
771 tot_len
+= TCPOLEN_TSTAMP_ALIGNED
;
772 #ifdef CONFIG_TCP_MD5SIG
774 tot_len
+= TCPOLEN_MD5SIG_ALIGNED
;
777 buff
= alloc_skb(MAX_HEADER
+ sizeof(struct ipv6hdr
) + tot_len
,
782 skb_reserve(buff
, MAX_HEADER
+ sizeof(struct ipv6hdr
) + tot_len
);
784 t1
= (struct tcphdr
*) skb_push(buff
, tot_len
);
785 skb_reset_transport_header(buff
);
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
;
796 t1
->window
= htons(win
);
798 topt
= (__be32
*)(t1
+ 1);
801 *topt
++ = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
802 (TCPOPT_TIMESTAMP
<< 8) | TCPOLEN_TIMESTAMP
);
803 *topt
++ = htonl(tsval
);
804 *topt
++ = htonl(tsecr
);
807 #ifdef CONFIG_TCP_MD5SIG
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
);
817 memset(&fl6
, 0, sizeof(fl6
));
818 fl6
.daddr
= ipv6_hdr(skb
)->saddr
;
819 fl6
.saddr
= ipv6_hdr(skb
)->daddr
;
820 fl6
.flowlabel
= label
;
822 buff
->ip_summed
= CHECKSUM_PARTIAL
;
825 __tcp_v6_send_check(buff
, &fl6
.saddr
, &fl6
.daddr
);
827 fl6
.flowi6_proto
= IPPROTO_TCP
;
828 if (rt6_need_strict(&fl6
.daddr
) && !oif
)
829 fl6
.flowi6_oif
= tcp_v6_iif(skb
);
831 if (!oif
&& netif_index_is_l3_master(net
, skb
->skb_iif
))
834 fl6
.flowi6_oif
= oif
;
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
));
843 /* Pass a socket to ip6_dst_lookup either it is for RST
844 * Underlying function will use this to retrieve the network
847 dst
= ip6_dst_lookup_flow(ctl_sk
, &fl6
, NULL
);
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
);
853 TCP_INC_STATS(net
, TCP_MIB_OUTRSTS
);
860 static void tcp_v6_send_reset(const struct sock
*sk
, struct sk_buff
*skb
)
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];
870 struct sock
*sk1
= NULL
;
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.
880 if (!sk
&& !ipv6_unicast_destination(skb
))
883 #ifdef CONFIG_TCP_MD5SIG
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
) {
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.
896 sk1
= inet6_lookup_listener(dev_net(skb_dst(skb
)->dev
),
897 &tcp_hashinfo
, NULL
, 0,
899 th
->source
, &ipv6h
->daddr
,
900 ntohs(th
->source
), tcp_v6_iif(skb
));
904 key
= tcp_v6_md5_do_lookup(sk1
, &ipv6h
->saddr
);
908 genhash
= tcp_v6_md5_hash_skb(newhash
, key
, NULL
, skb
);
909 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
915 seq
= ntohl(th
->ack_seq
);
917 ack_seq
= ntohl(th
->seq
) + th
->syn
+ th
->fin
+ skb
->len
-
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);
923 #ifdef CONFIG_TCP_MD5SIG
929 static 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
,
934 tcp_v6_send_response(sk
, skb
, seq
, ack
, win
, tsval
, tsecr
, oif
, key
, 0,
938 static void tcp_v6_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
940 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
941 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
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
));
952 static void tcp_v6_reqsk_send_ack(const struct sock
*sk
, struct sk_buff
*skb
,
953 struct request_sock
*req
)
955 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
956 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
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:
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
),
974 static struct sock
*tcp_v6_cookie_check(struct sock
*sk
, struct sk_buff
*skb
)
976 #ifdef CONFIG_SYN_COOKIES
977 const struct tcphdr
*th
= tcp_hdr(skb
);
980 sk
= cookie_v6_check(sk
, skb
);
985 static int tcp_v6_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
987 if (skb
->protocol
== htons(ETH_P_IP
))
988 return tcp_v4_conn_request(sk
, skb
);
990 if (!ipv6_unicast_destination(skb
))
993 return tcp_conn_request(&tcp6_request_sock_ops
,
994 &tcp_request_sock_ipv6_ops
, sk
, skb
);
998 return 0; /* don't send reset */
1001 static void tcp_v6_restore_cb(struct sk_buff
*skb
)
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.
1007 memmove(IP6CB(skb
), &TCP_SKB_CB(skb
)->header
.h6
,
1008 sizeof(struct inet6_skb_parm
));
1011 static 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
,
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
;
1025 #ifdef CONFIG_TCP_MD5SIG
1026 struct tcp_md5sig_key
*key
;
1030 if (skb
->protocol
== htons(ETH_P_IP
)) {
1035 newsk
= tcp_v4_syn_recv_sock(sk
, skb
, req
, dst
,
1036 req_unhash
, own_req
);
1041 newtcp6sk
= (struct tcp6_sock
*)newsk
;
1042 inet_sk(newsk
)->pinet6
= &newtcp6sk
->inet6
;
1044 newinet
= inet_sk(newsk
);
1045 newnp
= inet6_sk(newsk
);
1046 newtp
= tcp_sk(newsk
);
1048 memcpy(newnp
, np
, sizeof(struct ipv6_pinfo
));
1050 newnp
->saddr
= newsk
->sk_v6_rcv_saddr
;
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
;
1058 newnp
->ipv6_mc_list
= NULL
;
1059 newnp
->ipv6_ac_list
= NULL
;
1060 newnp
->ipv6_fl_list
= NULL
;
1061 newnp
->pktoptions
= NULL
;
1063 newnp
->mcast_oif
= tcp_v6_iif(skb
);
1064 newnp
->mcast_hops
= ipv6_hdr(skb
)->hop_limit
;
1065 newnp
->rcv_flowinfo
= ip6_flowinfo(ipv6_hdr(skb
));
1067 newnp
->flow_label
= ip6_flowlabel(ipv6_hdr(skb
));
1070 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1071 * here, tcp_create_openreq_child now does this for us, see the comment in
1072 * that function for the gory details. -acme
1075 /* It is tricky place. Until this moment IPv4 tcp
1076 worked with IPv6 icsk.icsk_af_ops.
1079 tcp_sync_mss(newsk
, inet_csk(newsk
)->icsk_pmtu_cookie
);
1084 ireq
= inet_rsk(req
);
1086 if (sk_acceptq_is_full(sk
))
1090 dst
= inet6_csk_route_req(sk
, &fl6
, req
, IPPROTO_TCP
);
1095 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1100 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1101 * count here, tcp_create_openreq_child now does this for us, see the
1102 * comment in that function for the gory details. -acme
1105 newsk
->sk_gso_type
= SKB_GSO_TCPV6
;
1106 ip6_dst_store(newsk
, dst
, NULL
, NULL
);
1107 inet6_sk_rx_dst_set(newsk
, skb
);
1109 newtcp6sk
= (struct tcp6_sock
*)newsk
;
1110 inet_sk(newsk
)->pinet6
= &newtcp6sk
->inet6
;
1112 newtp
= tcp_sk(newsk
);
1113 newinet
= inet_sk(newsk
);
1114 newnp
= inet6_sk(newsk
);
1116 memcpy(newnp
, np
, sizeof(struct ipv6_pinfo
));
1118 newsk
->sk_v6_daddr
= ireq
->ir_v6_rmt_addr
;
1119 newnp
->saddr
= ireq
->ir_v6_loc_addr
;
1120 newsk
->sk_v6_rcv_saddr
= ireq
->ir_v6_loc_addr
;
1121 newsk
->sk_bound_dev_if
= ireq
->ir_iif
;
1123 /* Now IPv6 options...
1125 First: no IPv4 options.
1127 newinet
->inet_opt
= NULL
;
1128 newnp
->ipv6_mc_list
= NULL
;
1129 newnp
->ipv6_ac_list
= NULL
;
1130 newnp
->ipv6_fl_list
= NULL
;
1133 newnp
->rxopt
.all
= np
->rxopt
.all
;
1135 newnp
->pktoptions
= NULL
;
1137 newnp
->mcast_oif
= tcp_v6_iif(skb
);
1138 newnp
->mcast_hops
= ipv6_hdr(skb
)->hop_limit
;
1139 newnp
->rcv_flowinfo
= ip6_flowinfo(ipv6_hdr(skb
));
1141 newnp
->flow_label
= ip6_flowlabel(ipv6_hdr(skb
));
1143 /* Clone native IPv6 options from listening socket (if any)
1145 Yes, keeping reference count would be much more clever,
1146 but we make one more one thing there: reattach optmem
1149 opt
= ireq
->ipv6_opt
;
1151 opt
= rcu_dereference(np
->opt
);
1153 opt
= ipv6_dup_options(newsk
, opt
);
1154 RCU_INIT_POINTER(newnp
->opt
, opt
);
1156 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1158 inet_csk(newsk
)->icsk_ext_hdr_len
= opt
->opt_nflen
+
1161 tcp_ca_openreq_child(newsk
, dst
);
1163 tcp_sync_mss(newsk
, dst_mtu(dst
));
1164 newtp
->advmss
= dst_metric_advmss(dst
);
1165 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1166 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1167 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1169 tcp_initialize_rcv_mss(newsk
);
1171 newinet
->inet_daddr
= newinet
->inet_saddr
= LOOPBACK4_IPV6
;
1172 newinet
->inet_rcv_saddr
= LOOPBACK4_IPV6
;
1174 #ifdef CONFIG_TCP_MD5SIG
1175 /* Copy over the MD5 key from the original socket */
1176 key
= tcp_v6_md5_do_lookup(sk
, &newsk
->sk_v6_daddr
);
1178 /* We're using one, so create a matching key
1179 * on the newsk structure. If we fail to get
1180 * memory, then we end up not copying the key
1183 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newsk
->sk_v6_daddr
,
1184 AF_INET6
, key
->key
, key
->keylen
,
1185 sk_gfp_mask(sk
, GFP_ATOMIC
));
1189 if (__inet_inherit_port(sk
, newsk
) < 0) {
1190 inet_csk_prepare_forced_close(newsk
);
1194 *own_req
= inet_ehash_nolisten(newsk
, req_to_sk(req_unhash
));
1196 tcp_move_syn(newtp
, req
);
1198 /* Clone pktoptions received with SYN, if we own the req */
1199 if (ireq
->pktopts
) {
1200 newnp
->pktoptions
= skb_clone(ireq
->pktopts
,
1201 sk_gfp_mask(sk
, GFP_ATOMIC
));
1202 consume_skb(ireq
->pktopts
);
1203 ireq
->pktopts
= NULL
;
1204 if (newnp
->pktoptions
) {
1205 tcp_v6_restore_cb(newnp
->pktoptions
);
1206 skb_set_owner_r(newnp
->pktoptions
, newsk
);
1214 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1222 /* The socket must have it's spinlock held when we get
1223 * here, unless it is a TCP_LISTEN socket.
1225 * We have a potential double-lock case here, so even when
1226 * doing backlog processing we use the BH locking scheme.
1227 * This is because we cannot sleep with the original spinlock
1230 static int tcp_v6_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1232 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1233 struct tcp_sock
*tp
;
1234 struct sk_buff
*opt_skb
= NULL
;
1236 /* Imagine: socket is IPv6. IPv4 packet arrives,
1237 goes to IPv4 receive handler and backlogged.
1238 From backlog it always goes here. Kerboom...
1239 Fortunately, tcp_rcv_established and rcv_established
1240 handle them correctly, but it is not case with
1241 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1244 if (skb
->protocol
== htons(ETH_P_IP
))
1245 return tcp_v4_do_rcv(sk
, skb
);
1247 if (tcp_filter(sk
, skb
))
1251 * socket locking is here for SMP purposes as backlog rcv
1252 * is currently called with bh processing disabled.
1255 /* Do Stevens' IPV6_PKTOPTIONS.
1257 Yes, guys, it is the only place in our code, where we
1258 may make it not affecting IPv4.
1259 The rest of code is protocol independent,
1260 and I do not like idea to uglify IPv4.
1262 Actually, all the idea behind IPV6_PKTOPTIONS
1263 looks not very well thought. For now we latch
1264 options, received in the last packet, enqueued
1265 by tcp. Feel free to propose better solution.
1269 opt_skb
= skb_clone(skb
, sk_gfp_mask(sk
, GFP_ATOMIC
));
1271 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1272 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1274 sock_rps_save_rxhash(sk
, skb
);
1275 sk_mark_napi_id(sk
, skb
);
1277 if (inet_sk(sk
)->rx_dst_ifindex
!= skb
->skb_iif
||
1278 dst
->ops
->check(dst
, np
->rx_dst_cookie
) == NULL
) {
1280 sk
->sk_rx_dst
= NULL
;
1284 tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
);
1286 goto ipv6_pktoptions
;
1290 if (tcp_checksum_complete(skb
))
1293 if (sk
->sk_state
== TCP_LISTEN
) {
1294 struct sock
*nsk
= tcp_v6_cookie_check(sk
, skb
);
1300 sock_rps_save_rxhash(nsk
, skb
);
1301 sk_mark_napi_id(nsk
, skb
);
1302 if (tcp_child_process(sk
, nsk
, skb
))
1305 __kfree_skb(opt_skb
);
1309 sock_rps_save_rxhash(sk
, skb
);
1311 if (tcp_rcv_state_process(sk
, skb
))
1314 goto ipv6_pktoptions
;
1318 tcp_v6_send_reset(sk
, skb
);
1321 __kfree_skb(opt_skb
);
1325 TCP_INC_STATS(sock_net(sk
), TCP_MIB_CSUMERRORS
);
1326 TCP_INC_STATS(sock_net(sk
), TCP_MIB_INERRS
);
1331 /* Do you ask, what is it?
1333 1. skb was enqueued by tcp.
1334 2. skb is added to tail of read queue, rather than out of order.
1335 3. socket is not in passive state.
1336 4. Finally, it really contains options, which user wants to receive.
1339 if (TCP_SKB_CB(opt_skb
)->end_seq
== tp
->rcv_nxt
&&
1340 !((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
))) {
1341 if (np
->rxopt
.bits
.rxinfo
|| np
->rxopt
.bits
.rxoinfo
)
1342 np
->mcast_oif
= tcp_v6_iif(opt_skb
);
1343 if (np
->rxopt
.bits
.rxhlim
|| np
->rxopt
.bits
.rxohlim
)
1344 np
->mcast_hops
= ipv6_hdr(opt_skb
)->hop_limit
;
1345 if (np
->rxopt
.bits
.rxflow
|| np
->rxopt
.bits
.rxtclass
)
1346 np
->rcv_flowinfo
= ip6_flowinfo(ipv6_hdr(opt_skb
));
1348 np
->flow_label
= ip6_flowlabel(ipv6_hdr(opt_skb
));
1349 if (ipv6_opt_accepted(sk
, opt_skb
, &TCP_SKB_CB(opt_skb
)->header
.h6
)) {
1350 skb_set_owner_r(opt_skb
, sk
);
1351 tcp_v6_restore_cb(opt_skb
);
1352 opt_skb
= xchg(&np
->pktoptions
, opt_skb
);
1354 __kfree_skb(opt_skb
);
1355 opt_skb
= xchg(&np
->pktoptions
, NULL
);
1363 static void tcp_v6_fill_cb(struct sk_buff
*skb
, const struct ipv6hdr
*hdr
,
1364 const struct tcphdr
*th
)
1366 /* This is tricky: we move IP6CB at its correct location into
1367 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1368 * _decode_session6() uses IP6CB().
1369 * barrier() makes sure compiler won't play aliasing games.
1371 memmove(&TCP_SKB_CB(skb
)->header
.h6
, IP6CB(skb
),
1372 sizeof(struct inet6_skb_parm
));
1375 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1376 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1377 skb
->len
- th
->doff
*4);
1378 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1379 TCP_SKB_CB(skb
)->tcp_flags
= tcp_flag_byte(th
);
1380 TCP_SKB_CB(skb
)->tcp_tw_isn
= 0;
1381 TCP_SKB_CB(skb
)->ip_dsfield
= ipv6_get_dsfield(hdr
);
1382 TCP_SKB_CB(skb
)->sacked
= 0;
1385 static int tcp_v6_rcv(struct sk_buff
*skb
)
1387 const struct tcphdr
*th
;
1388 const struct ipv6hdr
*hdr
;
1392 struct net
*net
= dev_net(skb
->dev
);
1394 if (skb
->pkt_type
!= PACKET_HOST
)
1398 * Count it even if it's bad.
1400 __TCP_INC_STATS(net
, TCP_MIB_INSEGS
);
1402 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1405 th
= (const struct tcphdr
*)skb
->data
;
1407 if (unlikely(th
->doff
< sizeof(struct tcphdr
)/4))
1409 if (!pskb_may_pull(skb
, th
->doff
*4))
1412 if (skb_checksum_init(skb
, IPPROTO_TCP
, ip6_compute_pseudo
))
1415 th
= (const struct tcphdr
*)skb
->data
;
1416 hdr
= ipv6_hdr(skb
);
1419 sk
= __inet6_lookup_skb(&tcp_hashinfo
, skb
, __tcp_hdrlen(th
),
1420 th
->source
, th
->dest
, inet6_iif(skb
),
1426 if (sk
->sk_state
== TCP_TIME_WAIT
)
1429 if (sk
->sk_state
== TCP_NEW_SYN_RECV
) {
1430 struct request_sock
*req
= inet_reqsk(sk
);
1433 sk
= req
->rsk_listener
;
1434 tcp_v6_fill_cb(skb
, hdr
, th
);
1435 if (tcp_v6_inbound_md5_hash(sk
, skb
)) {
1436 sk_drops_add(sk
, skb
);
1440 if (unlikely(sk
->sk_state
!= TCP_LISTEN
)) {
1441 inet_csk_reqsk_queue_drop_and_put(sk
, req
);
1446 nsk
= tcp_check_req(sk
, skb
, req
, false);
1449 goto discard_and_relse
;
1453 tcp_v6_restore_cb(skb
);
1454 } else if (tcp_child_process(sk
, nsk
, skb
)) {
1455 tcp_v6_send_reset(nsk
, skb
);
1456 goto discard_and_relse
;
1462 if (hdr
->hop_limit
< inet6_sk(sk
)->min_hopcount
) {
1463 __NET_INC_STATS(net
, LINUX_MIB_TCPMINTTLDROP
);
1464 goto discard_and_relse
;
1467 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1468 goto discard_and_relse
;
1470 tcp_v6_fill_cb(skb
, hdr
, th
);
1472 if (tcp_v6_inbound_md5_hash(sk
, skb
))
1473 goto discard_and_relse
;
1475 if (tcp_filter(sk
, skb
))
1476 goto discard_and_relse
;
1477 th
= (const struct tcphdr
*)skb
->data
;
1478 hdr
= ipv6_hdr(skb
);
1482 if (sk
->sk_state
== TCP_LISTEN
) {
1483 ret
= tcp_v6_do_rcv(sk
, skb
);
1484 goto put_and_return
;
1487 sk_incoming_cpu_update(sk
);
1489 bh_lock_sock_nested(sk
);
1490 tcp_segs_in(tcp_sk(sk
), skb
);
1492 if (!sock_owned_by_user(sk
)) {
1493 if (!tcp_prequeue(sk
, skb
))
1494 ret
= tcp_v6_do_rcv(sk
, skb
);
1495 } else if (tcp_add_backlog(sk
, skb
)) {
1496 goto discard_and_relse
;
1503 return ret
? -1 : 0;
1506 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1509 tcp_v6_fill_cb(skb
, hdr
, th
);
1511 if (tcp_checksum_complete(skb
)) {
1513 __TCP_INC_STATS(net
, TCP_MIB_CSUMERRORS
);
1515 __TCP_INC_STATS(net
, TCP_MIB_INERRS
);
1517 tcp_v6_send_reset(NULL
, skb
);
1525 sk_drops_add(sk
, skb
);
1531 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1532 inet_twsk_put(inet_twsk(sk
));
1536 tcp_v6_fill_cb(skb
, hdr
, th
);
1538 if (tcp_checksum_complete(skb
)) {
1539 inet_twsk_put(inet_twsk(sk
));
1543 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1548 sk2
= inet6_lookup_listener(dev_net(skb
->dev
), &tcp_hashinfo
,
1549 skb
, __tcp_hdrlen(th
),
1550 &ipv6_hdr(skb
)->saddr
, th
->source
,
1551 &ipv6_hdr(skb
)->daddr
,
1552 ntohs(th
->dest
), tcp_v6_iif(skb
));
1554 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
1555 inet_twsk_deschedule_put(tw
);
1557 tcp_v6_restore_cb(skb
);
1561 /* Fall through to ACK */
1564 tcp_v6_timewait_ack(sk
, skb
);
1567 tcp_v6_restore_cb(skb
);
1568 tcp_v6_send_reset(sk
, skb
);
1569 inet_twsk_deschedule_put(inet_twsk(sk
));
1571 case TCP_TW_SUCCESS
:
1577 static void tcp_v6_early_demux(struct sk_buff
*skb
)
1579 const struct ipv6hdr
*hdr
;
1580 const struct tcphdr
*th
;
1583 if (skb
->pkt_type
!= PACKET_HOST
)
1586 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct tcphdr
)))
1589 hdr
= ipv6_hdr(skb
);
1592 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1595 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1596 sk
= __inet6_lookup_established(dev_net(skb
->dev
), &tcp_hashinfo
,
1597 &hdr
->saddr
, th
->source
,
1598 &hdr
->daddr
, ntohs(th
->dest
),
1602 skb
->destructor
= sock_edemux
;
1603 if (sk_fullsock(sk
)) {
1604 struct dst_entry
*dst
= READ_ONCE(sk
->sk_rx_dst
);
1607 dst
= dst_check(dst
, inet6_sk(sk
)->rx_dst_cookie
);
1609 inet_sk(sk
)->rx_dst_ifindex
== skb
->skb_iif
)
1610 skb_dst_set_noref(skb
, dst
);
1615 static struct timewait_sock_ops tcp6_timewait_sock_ops
= {
1616 .twsk_obj_size
= sizeof(struct tcp6_timewait_sock
),
1617 .twsk_unique
= tcp_twsk_unique
,
1618 .twsk_destructor
= tcp_twsk_destructor
,
1621 static const struct inet_connection_sock_af_ops ipv6_specific
= {
1622 .queue_xmit
= inet6_csk_xmit
,
1623 .send_check
= tcp_v6_send_check
,
1624 .rebuild_header
= inet6_sk_rebuild_header
,
1625 .sk_rx_dst_set
= inet6_sk_rx_dst_set
,
1626 .conn_request
= tcp_v6_conn_request
,
1627 .syn_recv_sock
= tcp_v6_syn_recv_sock
,
1628 .net_header_len
= sizeof(struct ipv6hdr
),
1629 .net_frag_header_len
= sizeof(struct frag_hdr
),
1630 .setsockopt
= ipv6_setsockopt
,
1631 .getsockopt
= ipv6_getsockopt
,
1632 .addr2sockaddr
= inet6_csk_addr2sockaddr
,
1633 .sockaddr_len
= sizeof(struct sockaddr_in6
),
1634 .bind_conflict
= inet6_csk_bind_conflict
,
1635 #ifdef CONFIG_COMPAT
1636 .compat_setsockopt
= compat_ipv6_setsockopt
,
1637 .compat_getsockopt
= compat_ipv6_getsockopt
,
1639 .mtu_reduced
= tcp_v6_mtu_reduced
,
1642 #ifdef CONFIG_TCP_MD5SIG
1643 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific
= {
1644 .md5_lookup
= tcp_v6_md5_lookup
,
1645 .calc_md5_hash
= tcp_v6_md5_hash_skb
,
1646 .md5_parse
= tcp_v6_parse_md5_keys
,
1651 * TCP over IPv4 via INET6 API
1653 static const struct inet_connection_sock_af_ops ipv6_mapped
= {
1654 .queue_xmit
= ip_queue_xmit
,
1655 .send_check
= tcp_v4_send_check
,
1656 .rebuild_header
= inet_sk_rebuild_header
,
1657 .sk_rx_dst_set
= inet_sk_rx_dst_set
,
1658 .conn_request
= tcp_v6_conn_request
,
1659 .syn_recv_sock
= tcp_v6_syn_recv_sock
,
1660 .net_header_len
= sizeof(struct iphdr
),
1661 .setsockopt
= ipv6_setsockopt
,
1662 .getsockopt
= ipv6_getsockopt
,
1663 .addr2sockaddr
= inet6_csk_addr2sockaddr
,
1664 .sockaddr_len
= sizeof(struct sockaddr_in6
),
1665 .bind_conflict
= inet6_csk_bind_conflict
,
1666 #ifdef CONFIG_COMPAT
1667 .compat_setsockopt
= compat_ipv6_setsockopt
,
1668 .compat_getsockopt
= compat_ipv6_getsockopt
,
1670 .mtu_reduced
= tcp_v4_mtu_reduced
,
1673 #ifdef CONFIG_TCP_MD5SIG
1674 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific
= {
1675 .md5_lookup
= tcp_v4_md5_lookup
,
1676 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1677 .md5_parse
= tcp_v6_parse_md5_keys
,
1681 /* NOTE: A lot of things set to zero explicitly by call to
1682 * sk_alloc() so need not be done here.
1684 static int tcp_v6_init_sock(struct sock
*sk
)
1686 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1690 icsk
->icsk_af_ops
= &ipv6_specific
;
1692 #ifdef CONFIG_TCP_MD5SIG
1693 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv6_specific
;
1699 static void tcp_v6_destroy_sock(struct sock
*sk
)
1701 tcp_v4_destroy_sock(sk
);
1702 inet6_destroy_sock(sk
);
1705 #ifdef CONFIG_PROC_FS
1706 /* Proc filesystem TCPv6 sock list dumping. */
1707 static void get_openreq6(struct seq_file
*seq
,
1708 const struct request_sock
*req
, int i
)
1710 long ttd
= req
->rsk_timer
.expires
- jiffies
;
1711 const struct in6_addr
*src
= &inet_rsk(req
)->ir_v6_loc_addr
;
1712 const struct in6_addr
*dest
= &inet_rsk(req
)->ir_v6_rmt_addr
;
1718 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1719 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1721 src
->s6_addr32
[0], src
->s6_addr32
[1],
1722 src
->s6_addr32
[2], src
->s6_addr32
[3],
1723 inet_rsk(req
)->ir_num
,
1724 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1725 dest
->s6_addr32
[2], dest
->s6_addr32
[3],
1726 ntohs(inet_rsk(req
)->ir_rmt_port
),
1728 0, 0, /* could print option size, but that is af dependent. */
1729 1, /* timers active (only the expire timer) */
1730 jiffies_to_clock_t(ttd
),
1732 from_kuid_munged(seq_user_ns(seq
),
1733 sock_i_uid(req
->rsk_listener
)),
1734 0, /* non standard timer */
1735 0, /* open_requests have no inode */
1739 static void get_tcp6_sock(struct seq_file
*seq
, struct sock
*sp
, int i
)
1741 const struct in6_addr
*dest
, *src
;
1744 unsigned long timer_expires
;
1745 const struct inet_sock
*inet
= inet_sk(sp
);
1746 const struct tcp_sock
*tp
= tcp_sk(sp
);
1747 const struct inet_connection_sock
*icsk
= inet_csk(sp
);
1748 const struct fastopen_queue
*fastopenq
= &icsk
->icsk_accept_queue
.fastopenq
;
1752 dest
= &sp
->sk_v6_daddr
;
1753 src
= &sp
->sk_v6_rcv_saddr
;
1754 destp
= ntohs(inet
->inet_dport
);
1755 srcp
= ntohs(inet
->inet_sport
);
1757 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
||
1758 icsk
->icsk_pending
== ICSK_TIME_EARLY_RETRANS
||
1759 icsk
->icsk_pending
== ICSK_TIME_LOSS_PROBE
) {
1761 timer_expires
= icsk
->icsk_timeout
;
1762 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
1764 timer_expires
= icsk
->icsk_timeout
;
1765 } else if (timer_pending(&sp
->sk_timer
)) {
1767 timer_expires
= sp
->sk_timer
.expires
;
1770 timer_expires
= jiffies
;
1773 state
= sk_state_load(sp
);
1774 if (state
== TCP_LISTEN
)
1775 rx_queue
= sp
->sk_ack_backlog
;
1777 /* Because we don't lock the socket,
1778 * we might find a transient negative value.
1780 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
1783 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1784 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1786 src
->s6_addr32
[0], src
->s6_addr32
[1],
1787 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1788 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1789 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1791 tp
->write_seq
- tp
->snd_una
,
1794 jiffies_delta_to_clock_t(timer_expires
- jiffies
),
1795 icsk
->icsk_retransmits
,
1796 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(sp
)),
1797 icsk
->icsk_probes_out
,
1799 atomic_read(&sp
->sk_refcnt
), sp
,
1800 jiffies_to_clock_t(icsk
->icsk_rto
),
1801 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
1802 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
1804 state
== TCP_LISTEN
?
1805 fastopenq
->max_qlen
:
1806 (tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
)
1810 static void get_timewait6_sock(struct seq_file
*seq
,
1811 struct inet_timewait_sock
*tw
, int i
)
1813 long delta
= tw
->tw_timer
.expires
- jiffies
;
1814 const struct in6_addr
*dest
, *src
;
1817 dest
= &tw
->tw_v6_daddr
;
1818 src
= &tw
->tw_v6_rcv_saddr
;
1819 destp
= ntohs(tw
->tw_dport
);
1820 srcp
= ntohs(tw
->tw_sport
);
1823 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1824 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1826 src
->s6_addr32
[0], src
->s6_addr32
[1],
1827 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1828 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1829 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1830 tw
->tw_substate
, 0, 0,
1831 3, jiffies_delta_to_clock_t(delta
), 0, 0, 0, 0,
1832 atomic_read(&tw
->tw_refcnt
), tw
);
1835 static int tcp6_seq_show(struct seq_file
*seq
, void *v
)
1837 struct tcp_iter_state
*st
;
1838 struct sock
*sk
= v
;
1840 if (v
== SEQ_START_TOKEN
) {
1845 "st tx_queue rx_queue tr tm->when retrnsmt"
1846 " uid timeout inode\n");
1851 if (sk
->sk_state
== TCP_TIME_WAIT
)
1852 get_timewait6_sock(seq
, v
, st
->num
);
1853 else if (sk
->sk_state
== TCP_NEW_SYN_RECV
)
1854 get_openreq6(seq
, v
, st
->num
);
1856 get_tcp6_sock(seq
, v
, st
->num
);
1861 static const struct file_operations tcp6_afinfo_seq_fops
= {
1862 .owner
= THIS_MODULE
,
1863 .open
= tcp_seq_open
,
1865 .llseek
= seq_lseek
,
1866 .release
= seq_release_net
1869 static struct tcp_seq_afinfo tcp6_seq_afinfo
= {
1872 .seq_fops
= &tcp6_afinfo_seq_fops
,
1874 .show
= tcp6_seq_show
,
1878 int __net_init
tcp6_proc_init(struct net
*net
)
1880 return tcp_proc_register(net
, &tcp6_seq_afinfo
);
1883 void tcp6_proc_exit(struct net
*net
)
1885 tcp_proc_unregister(net
, &tcp6_seq_afinfo
);
1889 struct proto tcpv6_prot
= {
1891 .owner
= THIS_MODULE
,
1893 .connect
= tcp_v6_connect
,
1894 .disconnect
= tcp_disconnect
,
1895 .accept
= inet_csk_accept
,
1897 .init
= tcp_v6_init_sock
,
1898 .destroy
= tcp_v6_destroy_sock
,
1899 .shutdown
= tcp_shutdown
,
1900 .setsockopt
= tcp_setsockopt
,
1901 .getsockopt
= tcp_getsockopt
,
1902 .recvmsg
= tcp_recvmsg
,
1903 .sendmsg
= tcp_sendmsg
,
1904 .sendpage
= tcp_sendpage
,
1905 .backlog_rcv
= tcp_v6_do_rcv
,
1906 .release_cb
= tcp_release_cb
,
1908 .unhash
= inet_unhash
,
1909 .get_port
= inet_csk_get_port
,
1910 .enter_memory_pressure
= tcp_enter_memory_pressure
,
1911 .stream_memory_free
= tcp_stream_memory_free
,
1912 .sockets_allocated
= &tcp_sockets_allocated
,
1913 .memory_allocated
= &tcp_memory_allocated
,
1914 .memory_pressure
= &tcp_memory_pressure
,
1915 .orphan_count
= &tcp_orphan_count
,
1916 .sysctl_mem
= sysctl_tcp_mem
,
1917 .sysctl_wmem
= sysctl_tcp_wmem
,
1918 .sysctl_rmem
= sysctl_tcp_rmem
,
1919 .max_header
= MAX_TCP_HEADER
,
1920 .obj_size
= sizeof(struct tcp6_sock
),
1921 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1922 .twsk_prot
= &tcp6_timewait_sock_ops
,
1923 .rsk_prot
= &tcp6_request_sock_ops
,
1924 .h
.hashinfo
= &tcp_hashinfo
,
1925 .no_autobind
= true,
1926 #ifdef CONFIG_COMPAT
1927 .compat_setsockopt
= compat_tcp_setsockopt
,
1928 .compat_getsockopt
= compat_tcp_getsockopt
,
1930 .diag_destroy
= tcp_abort
,
1933 static const struct inet6_protocol tcpv6_protocol
= {
1934 .early_demux
= tcp_v6_early_demux
,
1935 .handler
= tcp_v6_rcv
,
1936 .err_handler
= tcp_v6_err
,
1937 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1940 static struct inet_protosw tcpv6_protosw
= {
1941 .type
= SOCK_STREAM
,
1942 .protocol
= IPPROTO_TCP
,
1943 .prot
= &tcpv6_prot
,
1944 .ops
= &inet6_stream_ops
,
1945 .flags
= INET_PROTOSW_PERMANENT
|
1949 static int __net_init
tcpv6_net_init(struct net
*net
)
1951 return inet_ctl_sock_create(&net
->ipv6
.tcp_sk
, PF_INET6
,
1952 SOCK_RAW
, IPPROTO_TCP
, net
);
1955 static void __net_exit
tcpv6_net_exit(struct net
*net
)
1957 inet_ctl_sock_destroy(net
->ipv6
.tcp_sk
);
1960 static void __net_exit
tcpv6_net_exit_batch(struct list_head
*net_exit_list
)
1962 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET6
);
1965 static struct pernet_operations tcpv6_net_ops
= {
1966 .init
= tcpv6_net_init
,
1967 .exit
= tcpv6_net_exit
,
1968 .exit_batch
= tcpv6_net_exit_batch
,
1971 int __init
tcpv6_init(void)
1975 ret
= inet6_add_protocol(&tcpv6_protocol
, IPPROTO_TCP
);
1979 /* register inet6 protocol */
1980 ret
= inet6_register_protosw(&tcpv6_protosw
);
1982 goto out_tcpv6_protocol
;
1984 ret
= register_pernet_subsys(&tcpv6_net_ops
);
1986 goto out_tcpv6_protosw
;
1991 inet6_unregister_protosw(&tcpv6_protosw
);
1993 inet6_del_protocol(&tcpv6_protocol
, IPPROTO_TCP
);
1997 void tcpv6_exit(void)
1999 unregister_pernet_subsys(&tcpv6_net_ops
);
2000 inet6_unregister_protosw(&tcpv6_protosw
);
2001 inet6_del_protocol(&tcpv6_protocol
, IPPROTO_TCP
);