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 #include <trace/events/tcp.h>
74 static void tcp_v6_send_reset(const struct sock
*sk
, struct sk_buff
*skb
);
75 static void tcp_v6_reqsk_send_ack(const struct sock
*sk
, struct sk_buff
*skb
,
76 struct request_sock
*req
);
78 static int tcp_v6_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
80 static const struct inet_connection_sock_af_ops ipv6_mapped
;
81 static const struct inet_connection_sock_af_ops ipv6_specific
;
82 #ifdef CONFIG_TCP_MD5SIG
83 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific
;
84 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific
;
86 static struct tcp_md5sig_key
*tcp_v6_md5_do_lookup(const struct sock
*sk
,
87 const struct in6_addr
*addr
)
93 static void inet6_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
95 struct dst_entry
*dst
= skb_dst(skb
);
97 if (dst
&& dst_hold_safe(dst
)) {
98 const struct rt6_info
*rt
= (const struct rt6_info
*)dst
;
101 inet_sk(sk
)->rx_dst_ifindex
= skb
->skb_iif
;
102 inet6_sk(sk
)->rx_dst_cookie
= rt6_get_cookie(rt
);
106 static u32
tcp_v6_init_seq(const struct sk_buff
*skb
)
108 return secure_tcpv6_seq(ipv6_hdr(skb
)->daddr
.s6_addr32
,
109 ipv6_hdr(skb
)->saddr
.s6_addr32
,
111 tcp_hdr(skb
)->source
);
114 static u32
tcp_v6_init_ts_off(const struct net
*net
, const struct sk_buff
*skb
)
116 return secure_tcpv6_ts_off(net
, ipv6_hdr(skb
)->daddr
.s6_addr32
,
117 ipv6_hdr(skb
)->saddr
.s6_addr32
);
120 static int tcp_v6_connect(struct sock
*sk
, struct sockaddr
*uaddr
,
123 struct sockaddr_in6
*usin
= (struct sockaddr_in6
*) uaddr
;
124 struct inet_sock
*inet
= inet_sk(sk
);
125 struct inet_connection_sock
*icsk
= inet_csk(sk
);
126 struct ipv6_pinfo
*np
= inet6_sk(sk
);
127 struct tcp_sock
*tp
= tcp_sk(sk
);
128 struct in6_addr
*saddr
= NULL
, *final_p
, final
;
129 struct ipv6_txoptions
*opt
;
131 struct dst_entry
*dst
;
134 struct inet_timewait_death_row
*tcp_death_row
= &sock_net(sk
)->ipv4
.tcp_death_row
;
136 if (addr_len
< SIN6_LEN_RFC2133
)
139 if (usin
->sin6_family
!= AF_INET6
)
140 return -EAFNOSUPPORT
;
142 memset(&fl6
, 0, sizeof(fl6
));
145 fl6
.flowlabel
= usin
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
146 IP6_ECN_flow_init(fl6
.flowlabel
);
147 if (fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) {
148 struct ip6_flowlabel
*flowlabel
;
149 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
152 fl6_sock_release(flowlabel
);
157 * connect() to INADDR_ANY means loopback (BSD'ism).
160 if (ipv6_addr_any(&usin
->sin6_addr
)) {
161 if (ipv6_addr_v4mapped(&sk
->sk_v6_rcv_saddr
))
162 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK
),
165 usin
->sin6_addr
= in6addr_loopback
;
168 addr_type
= ipv6_addr_type(&usin
->sin6_addr
);
170 if (addr_type
& IPV6_ADDR_MULTICAST
)
173 if (addr_type
&IPV6_ADDR_LINKLOCAL
) {
174 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
175 usin
->sin6_scope_id
) {
176 /* If interface is set while binding, indices
179 if (sk
->sk_bound_dev_if
&&
180 sk
->sk_bound_dev_if
!= usin
->sin6_scope_id
)
183 sk
->sk_bound_dev_if
= usin
->sin6_scope_id
;
186 /* Connect to link-local address requires an interface */
187 if (!sk
->sk_bound_dev_if
)
191 if (tp
->rx_opt
.ts_recent_stamp
&&
192 !ipv6_addr_equal(&sk
->sk_v6_daddr
, &usin
->sin6_addr
)) {
193 tp
->rx_opt
.ts_recent
= 0;
194 tp
->rx_opt
.ts_recent_stamp
= 0;
198 sk
->sk_v6_daddr
= usin
->sin6_addr
;
199 np
->flow_label
= fl6
.flowlabel
;
205 if (addr_type
& IPV6_ADDR_MAPPED
) {
206 u32 exthdrlen
= icsk
->icsk_ext_hdr_len
;
207 struct sockaddr_in sin
;
209 SOCK_DEBUG(sk
, "connect: ipv4 mapped\n");
211 if (__ipv6_only_sock(sk
))
214 sin
.sin_family
= AF_INET
;
215 sin
.sin_port
= usin
->sin6_port
;
216 sin
.sin_addr
.s_addr
= usin
->sin6_addr
.s6_addr32
[3];
218 icsk
->icsk_af_ops
= &ipv6_mapped
;
219 sk
->sk_backlog_rcv
= tcp_v4_do_rcv
;
220 #ifdef CONFIG_TCP_MD5SIG
221 tp
->af_specific
= &tcp_sock_ipv6_mapped_specific
;
224 err
= tcp_v4_connect(sk
, (struct sockaddr
*)&sin
, sizeof(sin
));
227 icsk
->icsk_ext_hdr_len
= exthdrlen
;
228 icsk
->icsk_af_ops
= &ipv6_specific
;
229 sk
->sk_backlog_rcv
= tcp_v6_do_rcv
;
230 #ifdef CONFIG_TCP_MD5SIG
231 tp
->af_specific
= &tcp_sock_ipv6_specific
;
235 np
->saddr
= sk
->sk_v6_rcv_saddr
;
240 if (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
))
241 saddr
= &sk
->sk_v6_rcv_saddr
;
243 fl6
.flowi6_proto
= IPPROTO_TCP
;
244 fl6
.daddr
= sk
->sk_v6_daddr
;
245 fl6
.saddr
= saddr
? *saddr
: np
->saddr
;
246 fl6
.flowi6_oif
= sk
->sk_bound_dev_if
;
247 fl6
.flowi6_mark
= sk
->sk_mark
;
248 fl6
.fl6_dport
= usin
->sin6_port
;
249 fl6
.fl6_sport
= inet
->inet_sport
;
250 fl6
.flowi6_uid
= sk
->sk_uid
;
252 opt
= rcu_dereference_protected(np
->opt
, lockdep_sock_is_held(sk
));
253 final_p
= fl6_update_dst(&fl6
, opt
, &final
);
255 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
257 dst
= ip6_dst_lookup_flow(sk
, &fl6
, final_p
);
265 sk
->sk_v6_rcv_saddr
= *saddr
;
268 /* set the source address */
270 inet
->inet_rcv_saddr
= LOOPBACK4_IPV6
;
272 sk
->sk_gso_type
= SKB_GSO_TCPV6
;
273 ip6_dst_store(sk
, dst
, NULL
, NULL
);
275 icsk
->icsk_ext_hdr_len
= 0;
277 icsk
->icsk_ext_hdr_len
= opt
->opt_flen
+
280 tp
->rx_opt
.mss_clamp
= IPV6_MIN_MTU
- sizeof(struct tcphdr
) - sizeof(struct ipv6hdr
);
282 inet
->inet_dport
= usin
->sin6_port
;
284 tcp_set_state(sk
, TCP_SYN_SENT
);
285 err
= inet6_hash_connect(tcp_death_row
, sk
);
291 if (likely(!tp
->repair
)) {
293 tp
->write_seq
= secure_tcpv6_seq(np
->saddr
.s6_addr32
,
294 sk
->sk_v6_daddr
.s6_addr32
,
297 tp
->tsoffset
= secure_tcpv6_ts_off(sock_net(sk
),
299 sk
->sk_v6_daddr
.s6_addr32
);
302 if (tcp_fastopen_defer_connect(sk
, &err
))
307 err
= tcp_connect(sk
);
314 tcp_set_state(sk
, TCP_CLOSE
);
316 inet
->inet_dport
= 0;
317 sk
->sk_route_caps
= 0;
321 static void tcp_v6_mtu_reduced(struct sock
*sk
)
323 struct dst_entry
*dst
;
325 if ((1 << sk
->sk_state
) & (TCPF_LISTEN
| TCPF_CLOSE
))
328 dst
= inet6_csk_update_pmtu(sk
, tcp_sk(sk
)->mtu_info
);
332 if (inet_csk(sk
)->icsk_pmtu_cookie
> dst_mtu(dst
)) {
333 tcp_sync_mss(sk
, dst_mtu(dst
));
334 tcp_simple_retransmit(sk
);
338 static void tcp_v6_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
339 u8 type
, u8 code
, int offset
, __be32 info
)
341 const struct ipv6hdr
*hdr
= (const struct ipv6hdr
*)skb
->data
;
342 const struct tcphdr
*th
= (struct tcphdr
*)(skb
->data
+offset
);
343 struct net
*net
= dev_net(skb
->dev
);
344 struct request_sock
*fastopen
;
345 struct ipv6_pinfo
*np
;
352 sk
= __inet6_lookup_established(net
, &tcp_hashinfo
,
353 &hdr
->daddr
, th
->dest
,
354 &hdr
->saddr
, ntohs(th
->source
),
355 skb
->dev
->ifindex
, inet6_sdif(skb
));
358 __ICMP6_INC_STATS(net
, __in6_dev_get(skb
->dev
),
363 if (sk
->sk_state
== TCP_TIME_WAIT
) {
364 inet_twsk_put(inet_twsk(sk
));
367 seq
= ntohl(th
->seq
);
368 fatal
= icmpv6_err_convert(type
, code
, &err
);
369 if (sk
->sk_state
== TCP_NEW_SYN_RECV
)
370 return tcp_req_err(sk
, seq
, fatal
);
373 if (sock_owned_by_user(sk
) && type
!= ICMPV6_PKT_TOOBIG
)
374 __NET_INC_STATS(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
376 if (sk
->sk_state
== TCP_CLOSE
)
379 if (ipv6_hdr(skb
)->hop_limit
< inet6_sk(sk
)->min_hopcount
) {
380 __NET_INC_STATS(net
, LINUX_MIB_TCPMINTTLDROP
);
385 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
386 fastopen
= tp
->fastopen_rsk
;
387 snd_una
= fastopen
? tcp_rsk(fastopen
)->snt_isn
: tp
->snd_una
;
388 if (sk
->sk_state
!= TCP_LISTEN
&&
389 !between(seq
, snd_una
, tp
->snd_nxt
)) {
390 __NET_INC_STATS(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
396 if (type
== NDISC_REDIRECT
) {
397 if (!sock_owned_by_user(sk
)) {
398 struct dst_entry
*dst
= __sk_dst_check(sk
, np
->dst_cookie
);
401 dst
->ops
->redirect(dst
, sk
, skb
);
406 if (type
== ICMPV6_PKT_TOOBIG
) {
407 /* We are not interested in TCP_LISTEN and open_requests
408 * (SYN-ACKs send out by Linux are always <576bytes so
409 * they should go through unfragmented).
411 if (sk
->sk_state
== TCP_LISTEN
)
414 if (!ip6_sk_accept_pmtu(sk
))
417 tp
->mtu_info
= ntohl(info
);
418 if (!sock_owned_by_user(sk
))
419 tcp_v6_mtu_reduced(sk
);
420 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED
,
427 /* Might be for an request_sock */
428 switch (sk
->sk_state
) {
431 /* Only in fast or simultaneous open. If a fast open socket is
432 * is already accepted it is treated as a connected one below.
434 if (fastopen
&& !fastopen
->sk
)
437 if (!sock_owned_by_user(sk
)) {
439 sk
->sk_error_report(sk
); /* Wake people up to see the error (see connect in sock.c) */
443 sk
->sk_err_soft
= err
;
447 if (!sock_owned_by_user(sk
) && np
->recverr
) {
449 sk
->sk_error_report(sk
);
451 sk
->sk_err_soft
= err
;
459 static int tcp_v6_send_synack(const struct sock
*sk
, struct dst_entry
*dst
,
461 struct request_sock
*req
,
462 struct tcp_fastopen_cookie
*foc
,
463 enum tcp_synack_type synack_type
)
465 struct inet_request_sock
*ireq
= inet_rsk(req
);
466 struct ipv6_pinfo
*np
= inet6_sk(sk
);
467 struct ipv6_txoptions
*opt
;
468 struct flowi6
*fl6
= &fl
->u
.ip6
;
472 /* First, grab a route. */
473 if (!dst
&& (dst
= inet6_csk_route_req(sk
, fl6
, req
,
474 IPPROTO_TCP
)) == NULL
)
477 skb
= tcp_make_synack(sk
, dst
, req
, foc
, synack_type
);
480 __tcp_v6_send_check(skb
, &ireq
->ir_v6_loc_addr
,
481 &ireq
->ir_v6_rmt_addr
);
483 fl6
->daddr
= ireq
->ir_v6_rmt_addr
;
484 if (np
->repflow
&& ireq
->pktopts
)
485 fl6
->flowlabel
= ip6_flowlabel(ipv6_hdr(ireq
->pktopts
));
488 opt
= ireq
->ipv6_opt
;
490 opt
= rcu_dereference(np
->opt
);
491 err
= ip6_xmit(sk
, skb
, fl6
, sk
->sk_mark
, opt
, np
->tclass
);
493 err
= net_xmit_eval(err
);
501 static void tcp_v6_reqsk_destructor(struct request_sock
*req
)
503 kfree(inet_rsk(req
)->ipv6_opt
);
504 kfree_skb(inet_rsk(req
)->pktopts
);
507 #ifdef CONFIG_TCP_MD5SIG
508 static struct tcp_md5sig_key
*tcp_v6_md5_do_lookup(const struct sock
*sk
,
509 const struct in6_addr
*addr
)
511 return tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)addr
, AF_INET6
);
514 static struct tcp_md5sig_key
*tcp_v6_md5_lookup(const struct sock
*sk
,
515 const struct sock
*addr_sk
)
517 return tcp_v6_md5_do_lookup(sk
, &addr_sk
->sk_v6_daddr
);
520 static int tcp_v6_parse_md5_keys(struct sock
*sk
, int optname
,
521 char __user
*optval
, int optlen
)
523 struct tcp_md5sig cmd
;
524 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*)&cmd
.tcpm_addr
;
527 if (optlen
< sizeof(cmd
))
530 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
533 if (sin6
->sin6_family
!= AF_INET6
)
536 if (optname
== TCP_MD5SIG_EXT
&&
537 cmd
.tcpm_flags
& TCP_MD5SIG_FLAG_PREFIX
) {
538 prefixlen
= cmd
.tcpm_prefixlen
;
539 if (prefixlen
> 128 || (ipv6_addr_v4mapped(&sin6
->sin6_addr
) &&
543 prefixlen
= ipv6_addr_v4mapped(&sin6
->sin6_addr
) ? 32 : 128;
546 if (!cmd
.tcpm_keylen
) {
547 if (ipv6_addr_v4mapped(&sin6
->sin6_addr
))
548 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
.s6_addr32
[3],
550 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
,
551 AF_INET6
, prefixlen
);
554 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
557 if (ipv6_addr_v4mapped(&sin6
->sin6_addr
))
558 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
.s6_addr32
[3],
559 AF_INET
, prefixlen
, cmd
.tcpm_key
,
560 cmd
.tcpm_keylen
, GFP_KERNEL
);
562 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
,
563 AF_INET6
, prefixlen
, cmd
.tcpm_key
,
564 cmd
.tcpm_keylen
, GFP_KERNEL
);
567 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool
*hp
,
568 const struct in6_addr
*daddr
,
569 const struct in6_addr
*saddr
,
570 const struct tcphdr
*th
, int nbytes
)
572 struct tcp6_pseudohdr
*bp
;
573 struct scatterlist sg
;
577 /* 1. TCP pseudo-header (RFC2460) */
580 bp
->protocol
= cpu_to_be32(IPPROTO_TCP
);
581 bp
->len
= cpu_to_be32(nbytes
);
583 _th
= (struct tcphdr
*)(bp
+ 1);
584 memcpy(_th
, th
, sizeof(*th
));
587 sg_init_one(&sg
, bp
, sizeof(*bp
) + sizeof(*th
));
588 ahash_request_set_crypt(hp
->md5_req
, &sg
, NULL
,
589 sizeof(*bp
) + sizeof(*th
));
590 return crypto_ahash_update(hp
->md5_req
);
593 static int tcp_v6_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
594 const struct in6_addr
*daddr
, struct in6_addr
*saddr
,
595 const struct tcphdr
*th
)
597 struct tcp_md5sig_pool
*hp
;
598 struct ahash_request
*req
;
600 hp
= tcp_get_md5sig_pool();
602 goto clear_hash_noput
;
605 if (crypto_ahash_init(req
))
607 if (tcp_v6_md5_hash_headers(hp
, daddr
, saddr
, th
, th
->doff
<< 2))
609 if (tcp_md5_hash_key(hp
, key
))
611 ahash_request_set_crypt(req
, NULL
, md5_hash
, 0);
612 if (crypto_ahash_final(req
))
615 tcp_put_md5sig_pool();
619 tcp_put_md5sig_pool();
621 memset(md5_hash
, 0, 16);
625 static int tcp_v6_md5_hash_skb(char *md5_hash
,
626 const struct tcp_md5sig_key
*key
,
627 const struct sock
*sk
,
628 const struct sk_buff
*skb
)
630 const struct in6_addr
*saddr
, *daddr
;
631 struct tcp_md5sig_pool
*hp
;
632 struct ahash_request
*req
;
633 const struct tcphdr
*th
= tcp_hdr(skb
);
635 if (sk
) { /* valid for establish/request sockets */
636 saddr
= &sk
->sk_v6_rcv_saddr
;
637 daddr
= &sk
->sk_v6_daddr
;
639 const struct ipv6hdr
*ip6h
= ipv6_hdr(skb
);
640 saddr
= &ip6h
->saddr
;
641 daddr
= &ip6h
->daddr
;
644 hp
= tcp_get_md5sig_pool();
646 goto clear_hash_noput
;
649 if (crypto_ahash_init(req
))
652 if (tcp_v6_md5_hash_headers(hp
, daddr
, saddr
, th
, skb
->len
))
654 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
656 if (tcp_md5_hash_key(hp
, key
))
658 ahash_request_set_crypt(req
, NULL
, md5_hash
, 0);
659 if (crypto_ahash_final(req
))
662 tcp_put_md5sig_pool();
666 tcp_put_md5sig_pool();
668 memset(md5_hash
, 0, 16);
674 static bool tcp_v6_inbound_md5_hash(const struct sock
*sk
,
675 const struct sk_buff
*skb
)
677 #ifdef CONFIG_TCP_MD5SIG
678 const __u8
*hash_location
= NULL
;
679 struct tcp_md5sig_key
*hash_expected
;
680 const struct ipv6hdr
*ip6h
= ipv6_hdr(skb
);
681 const struct tcphdr
*th
= tcp_hdr(skb
);
685 hash_expected
= tcp_v6_md5_do_lookup(sk
, &ip6h
->saddr
);
686 hash_location
= tcp_parse_md5sig_option(th
);
688 /* We've parsed the options - do we have a hash? */
689 if (!hash_expected
&& !hash_location
)
692 if (hash_expected
&& !hash_location
) {
693 NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
697 if (!hash_expected
&& hash_location
) {
698 NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
702 /* check the signature */
703 genhash
= tcp_v6_md5_hash_skb(newhash
,
707 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
708 NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPMD5FAILURE
);
709 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
710 genhash
? "failed" : "mismatch",
711 &ip6h
->saddr
, ntohs(th
->source
),
712 &ip6h
->daddr
, ntohs(th
->dest
));
719 static void tcp_v6_init_req(struct request_sock
*req
,
720 const struct sock
*sk_listener
,
723 struct inet_request_sock
*ireq
= inet_rsk(req
);
724 const struct ipv6_pinfo
*np
= inet6_sk(sk_listener
);
726 ireq
->ir_v6_rmt_addr
= ipv6_hdr(skb
)->saddr
;
727 ireq
->ir_v6_loc_addr
= ipv6_hdr(skb
)->daddr
;
729 /* So that link locals have meaning */
730 if (!sk_listener
->sk_bound_dev_if
&&
731 ipv6_addr_type(&ireq
->ir_v6_rmt_addr
) & IPV6_ADDR_LINKLOCAL
)
732 ireq
->ir_iif
= tcp_v6_iif(skb
);
734 if (!TCP_SKB_CB(skb
)->tcp_tw_isn
&&
735 (ipv6_opt_accepted(sk_listener
, skb
, &TCP_SKB_CB(skb
)->header
.h6
) ||
736 np
->rxopt
.bits
.rxinfo
||
737 np
->rxopt
.bits
.rxoinfo
|| np
->rxopt
.bits
.rxhlim
||
738 np
->rxopt
.bits
.rxohlim
|| np
->repflow
)) {
739 refcount_inc(&skb
->users
);
744 static struct dst_entry
*tcp_v6_route_req(const struct sock
*sk
,
746 const struct request_sock
*req
)
748 return inet6_csk_route_req(sk
, &fl
->u
.ip6
, req
, IPPROTO_TCP
);
751 struct request_sock_ops tcp6_request_sock_ops __read_mostly
= {
753 .obj_size
= sizeof(struct tcp6_request_sock
),
754 .rtx_syn_ack
= tcp_rtx_synack
,
755 .send_ack
= tcp_v6_reqsk_send_ack
,
756 .destructor
= tcp_v6_reqsk_destructor
,
757 .send_reset
= tcp_v6_send_reset
,
758 .syn_ack_timeout
= tcp_syn_ack_timeout
,
761 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops
= {
762 .mss_clamp
= IPV6_MIN_MTU
- sizeof(struct tcphdr
) -
763 sizeof(struct ipv6hdr
),
764 #ifdef CONFIG_TCP_MD5SIG
765 .req_md5_lookup
= tcp_v6_md5_lookup
,
766 .calc_md5_hash
= tcp_v6_md5_hash_skb
,
768 .init_req
= tcp_v6_init_req
,
769 #ifdef CONFIG_SYN_COOKIES
770 .cookie_init_seq
= cookie_v6_init_sequence
,
772 .route_req
= tcp_v6_route_req
,
773 .init_seq
= tcp_v6_init_seq
,
774 .init_ts_off
= tcp_v6_init_ts_off
,
775 .send_synack
= tcp_v6_send_synack
,
778 static void tcp_v6_send_response(const struct sock
*sk
, struct sk_buff
*skb
, u32 seq
,
779 u32 ack
, u32 win
, u32 tsval
, u32 tsecr
,
780 int oif
, struct tcp_md5sig_key
*key
, int rst
,
781 u8 tclass
, __be32 label
)
783 const struct tcphdr
*th
= tcp_hdr(skb
);
785 struct sk_buff
*buff
;
787 struct net
*net
= sk
? sock_net(sk
) : dev_net(skb_dst(skb
)->dev
);
788 struct sock
*ctl_sk
= net
->ipv6
.tcp_sk
;
789 unsigned int tot_len
= sizeof(struct tcphdr
);
790 struct dst_entry
*dst
;
794 tot_len
+= TCPOLEN_TSTAMP_ALIGNED
;
795 #ifdef CONFIG_TCP_MD5SIG
797 tot_len
+= TCPOLEN_MD5SIG_ALIGNED
;
800 buff
= alloc_skb(MAX_HEADER
+ sizeof(struct ipv6hdr
) + tot_len
,
805 skb_reserve(buff
, MAX_HEADER
+ sizeof(struct ipv6hdr
) + tot_len
);
807 t1
= skb_push(buff
, tot_len
);
808 skb_reset_transport_header(buff
);
810 /* Swap the send and the receive. */
811 memset(t1
, 0, sizeof(*t1
));
812 t1
->dest
= th
->source
;
813 t1
->source
= th
->dest
;
814 t1
->doff
= tot_len
/ 4;
815 t1
->seq
= htonl(seq
);
816 t1
->ack_seq
= htonl(ack
);
817 t1
->ack
= !rst
|| !th
->ack
;
819 t1
->window
= htons(win
);
821 topt
= (__be32
*)(t1
+ 1);
824 *topt
++ = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
825 (TCPOPT_TIMESTAMP
<< 8) | TCPOLEN_TIMESTAMP
);
826 *topt
++ = htonl(tsval
);
827 *topt
++ = htonl(tsecr
);
830 #ifdef CONFIG_TCP_MD5SIG
832 *topt
++ = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
833 (TCPOPT_MD5SIG
<< 8) | TCPOLEN_MD5SIG
);
834 tcp_v6_md5_hash_hdr((__u8
*)topt
, key
,
835 &ipv6_hdr(skb
)->saddr
,
836 &ipv6_hdr(skb
)->daddr
, t1
);
840 memset(&fl6
, 0, sizeof(fl6
));
841 fl6
.daddr
= ipv6_hdr(skb
)->saddr
;
842 fl6
.saddr
= ipv6_hdr(skb
)->daddr
;
843 fl6
.flowlabel
= label
;
845 buff
->ip_summed
= CHECKSUM_PARTIAL
;
848 __tcp_v6_send_check(buff
, &fl6
.saddr
, &fl6
.daddr
);
850 fl6
.flowi6_proto
= IPPROTO_TCP
;
851 if (rt6_need_strict(&fl6
.daddr
) && !oif
)
852 fl6
.flowi6_oif
= tcp_v6_iif(skb
);
854 if (!oif
&& netif_index_is_l3_master(net
, skb
->skb_iif
))
857 fl6
.flowi6_oif
= oif
;
860 fl6
.flowi6_mark
= IP6_REPLY_MARK(net
, skb
->mark
);
861 fl6
.fl6_dport
= t1
->dest
;
862 fl6
.fl6_sport
= t1
->source
;
863 fl6
.flowi6_uid
= sock_net_uid(net
, sk
&& sk_fullsock(sk
) ? sk
: NULL
);
864 security_skb_classify_flow(skb
, flowi6_to_flowi(&fl6
));
866 /* Pass a socket to ip6_dst_lookup either it is for RST
867 * Underlying function will use this to retrieve the network
870 dst
= ip6_dst_lookup_flow(ctl_sk
, &fl6
, NULL
);
872 skb_dst_set(buff
, dst
);
873 ip6_xmit(ctl_sk
, buff
, &fl6
, fl6
.flowi6_mark
, NULL
, tclass
);
874 TCP_INC_STATS(net
, TCP_MIB_OUTSEGS
);
876 TCP_INC_STATS(net
, TCP_MIB_OUTRSTS
);
883 static void tcp_v6_send_reset(const struct sock
*sk
, struct sk_buff
*skb
)
885 const struct tcphdr
*th
= tcp_hdr(skb
);
886 u32 seq
= 0, ack_seq
= 0;
887 struct tcp_md5sig_key
*key
= NULL
;
888 #ifdef CONFIG_TCP_MD5SIG
889 const __u8
*hash_location
= NULL
;
890 struct ipv6hdr
*ipv6h
= ipv6_hdr(skb
);
891 unsigned char newhash
[16];
893 struct sock
*sk1
= NULL
;
900 /* If sk not NULL, it means we did a successful lookup and incoming
901 * route had to be correct. prequeue might have dropped our dst.
903 if (!sk
&& !ipv6_unicast_destination(skb
))
906 #ifdef CONFIG_TCP_MD5SIG
908 hash_location
= tcp_parse_md5sig_option(th
);
909 if (sk
&& sk_fullsock(sk
)) {
910 key
= tcp_v6_md5_do_lookup(sk
, &ipv6h
->saddr
);
911 } else if (hash_location
) {
913 * active side is lost. Try to find listening socket through
914 * source port, and then find md5 key through listening socket.
915 * we are not loose security here:
916 * Incoming packet is checked with md5 hash with finding key,
917 * no RST generated if md5 hash doesn't match.
919 sk1
= inet6_lookup_listener(dev_net(skb_dst(skb
)->dev
),
920 &tcp_hashinfo
, NULL
, 0,
922 th
->source
, &ipv6h
->daddr
,
924 tcp_v6_iif_l3_slave(skb
),
929 key
= tcp_v6_md5_do_lookup(sk1
, &ipv6h
->saddr
);
933 genhash
= tcp_v6_md5_hash_skb(newhash
, key
, NULL
, skb
);
934 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
940 seq
= ntohl(th
->ack_seq
);
942 ack_seq
= ntohl(th
->seq
) + th
->syn
+ th
->fin
+ skb
->len
-
946 oif
= sk
->sk_bound_dev_if
;
948 trace_tcp_send_reset(sk
, skb
);
951 tcp_v6_send_response(sk
, skb
, seq
, ack_seq
, 0, 0, 0, oif
, key
, 1, 0, 0);
953 #ifdef CONFIG_TCP_MD5SIG
959 static void tcp_v6_send_ack(const struct sock
*sk
, struct sk_buff
*skb
, u32 seq
,
960 u32 ack
, u32 win
, u32 tsval
, u32 tsecr
, int oif
,
961 struct tcp_md5sig_key
*key
, u8 tclass
,
964 tcp_v6_send_response(sk
, skb
, seq
, ack
, win
, tsval
, tsecr
, oif
, key
, 0,
968 static void tcp_v6_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
970 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
971 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
973 tcp_v6_send_ack(sk
, skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
974 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
975 tcp_time_stamp_raw() + tcptw
->tw_ts_offset
,
976 tcptw
->tw_ts_recent
, tw
->tw_bound_dev_if
, tcp_twsk_md5_key(tcptw
),
977 tw
->tw_tclass
, cpu_to_be32(tw
->tw_flowlabel
));
982 static void tcp_v6_reqsk_send_ack(const struct sock
*sk
, struct sk_buff
*skb
,
983 struct request_sock
*req
)
985 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
986 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
989 * The window field (SEG.WND) of every outgoing segment, with the
990 * exception of <SYN> segments, MUST be right-shifted by
991 * Rcv.Wind.Shift bits:
993 tcp_v6_send_ack(sk
, skb
, (sk
->sk_state
== TCP_LISTEN
) ?
994 tcp_rsk(req
)->snt_isn
+ 1 : tcp_sk(sk
)->snd_nxt
,
995 tcp_rsk(req
)->rcv_nxt
,
996 req
->rsk_rcv_wnd
>> inet_rsk(req
)->rcv_wscale
,
997 tcp_time_stamp_raw() + tcp_rsk(req
)->ts_off
,
998 req
->ts_recent
, sk
->sk_bound_dev_if
,
999 tcp_v6_md5_do_lookup(sk
, &ipv6_hdr(skb
)->saddr
),
1004 static struct sock
*tcp_v6_cookie_check(struct sock
*sk
, struct sk_buff
*skb
)
1006 #ifdef CONFIG_SYN_COOKIES
1007 const struct tcphdr
*th
= tcp_hdr(skb
);
1010 sk
= cookie_v6_check(sk
, skb
);
1015 static int tcp_v6_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1017 if (skb
->protocol
== htons(ETH_P_IP
))
1018 return tcp_v4_conn_request(sk
, skb
);
1020 if (!ipv6_unicast_destination(skb
))
1023 return tcp_conn_request(&tcp6_request_sock_ops
,
1024 &tcp_request_sock_ipv6_ops
, sk
, skb
);
1028 return 0; /* don't send reset */
1031 static void tcp_v6_restore_cb(struct sk_buff
*skb
)
1033 /* We need to move header back to the beginning if xfrm6_policy_check()
1034 * and tcp_v6_fill_cb() are going to be called again.
1035 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1037 memmove(IP6CB(skb
), &TCP_SKB_CB(skb
)->header
.h6
,
1038 sizeof(struct inet6_skb_parm
));
1041 static struct sock
*tcp_v6_syn_recv_sock(const struct sock
*sk
, struct sk_buff
*skb
,
1042 struct request_sock
*req
,
1043 struct dst_entry
*dst
,
1044 struct request_sock
*req_unhash
,
1047 struct inet_request_sock
*ireq
;
1048 struct ipv6_pinfo
*newnp
;
1049 const struct ipv6_pinfo
*np
= inet6_sk(sk
);
1050 struct ipv6_txoptions
*opt
;
1051 struct tcp6_sock
*newtcp6sk
;
1052 struct inet_sock
*newinet
;
1053 struct tcp_sock
*newtp
;
1055 #ifdef CONFIG_TCP_MD5SIG
1056 struct tcp_md5sig_key
*key
;
1060 if (skb
->protocol
== htons(ETH_P_IP
)) {
1065 newsk
= tcp_v4_syn_recv_sock(sk
, skb
, req
, dst
,
1066 req_unhash
, own_req
);
1071 newtcp6sk
= (struct tcp6_sock
*)newsk
;
1072 inet_sk(newsk
)->pinet6
= &newtcp6sk
->inet6
;
1074 newinet
= inet_sk(newsk
);
1075 newnp
= inet6_sk(newsk
);
1076 newtp
= tcp_sk(newsk
);
1078 memcpy(newnp
, np
, sizeof(struct ipv6_pinfo
));
1080 newnp
->saddr
= newsk
->sk_v6_rcv_saddr
;
1082 inet_csk(newsk
)->icsk_af_ops
= &ipv6_mapped
;
1083 newsk
->sk_backlog_rcv
= tcp_v4_do_rcv
;
1084 #ifdef CONFIG_TCP_MD5SIG
1085 newtp
->af_specific
= &tcp_sock_ipv6_mapped_specific
;
1088 newnp
->ipv6_mc_list
= NULL
;
1089 newnp
->ipv6_ac_list
= NULL
;
1090 newnp
->ipv6_fl_list
= NULL
;
1091 newnp
->pktoptions
= NULL
;
1093 newnp
->mcast_oif
= tcp_v6_iif(skb
);
1094 newnp
->mcast_hops
= ipv6_hdr(skb
)->hop_limit
;
1095 newnp
->rcv_flowinfo
= ip6_flowinfo(ipv6_hdr(skb
));
1097 newnp
->flow_label
= ip6_flowlabel(ipv6_hdr(skb
));
1100 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1101 * here, tcp_create_openreq_child now does this for us, see the comment in
1102 * that function for the gory details. -acme
1105 /* It is tricky place. Until this moment IPv4 tcp
1106 worked with IPv6 icsk.icsk_af_ops.
1109 tcp_sync_mss(newsk
, inet_csk(newsk
)->icsk_pmtu_cookie
);
1114 ireq
= inet_rsk(req
);
1116 if (sk_acceptq_is_full(sk
))
1120 dst
= inet6_csk_route_req(sk
, &fl6
, req
, IPPROTO_TCP
);
1125 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1130 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1131 * count here, tcp_create_openreq_child now does this for us, see the
1132 * comment in that function for the gory details. -acme
1135 newsk
->sk_gso_type
= SKB_GSO_TCPV6
;
1136 ip6_dst_store(newsk
, dst
, NULL
, NULL
);
1137 inet6_sk_rx_dst_set(newsk
, skb
);
1139 newtcp6sk
= (struct tcp6_sock
*)newsk
;
1140 inet_sk(newsk
)->pinet6
= &newtcp6sk
->inet6
;
1142 newtp
= tcp_sk(newsk
);
1143 newinet
= inet_sk(newsk
);
1144 newnp
= inet6_sk(newsk
);
1146 memcpy(newnp
, np
, sizeof(struct ipv6_pinfo
));
1148 newsk
->sk_v6_daddr
= ireq
->ir_v6_rmt_addr
;
1149 newnp
->saddr
= ireq
->ir_v6_loc_addr
;
1150 newsk
->sk_v6_rcv_saddr
= ireq
->ir_v6_loc_addr
;
1151 newsk
->sk_bound_dev_if
= ireq
->ir_iif
;
1153 /* Now IPv6 options...
1155 First: no IPv4 options.
1157 newinet
->inet_opt
= NULL
;
1158 newnp
->ipv6_mc_list
= NULL
;
1159 newnp
->ipv6_ac_list
= NULL
;
1160 newnp
->ipv6_fl_list
= NULL
;
1163 newnp
->rxopt
.all
= np
->rxopt
.all
;
1165 newnp
->pktoptions
= NULL
;
1167 newnp
->mcast_oif
= tcp_v6_iif(skb
);
1168 newnp
->mcast_hops
= ipv6_hdr(skb
)->hop_limit
;
1169 newnp
->rcv_flowinfo
= ip6_flowinfo(ipv6_hdr(skb
));
1171 newnp
->flow_label
= ip6_flowlabel(ipv6_hdr(skb
));
1173 /* Clone native IPv6 options from listening socket (if any)
1175 Yes, keeping reference count would be much more clever,
1176 but we make one more one thing there: reattach optmem
1179 opt
= ireq
->ipv6_opt
;
1181 opt
= rcu_dereference(np
->opt
);
1183 opt
= ipv6_dup_options(newsk
, opt
);
1184 RCU_INIT_POINTER(newnp
->opt
, opt
);
1186 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1188 inet_csk(newsk
)->icsk_ext_hdr_len
= opt
->opt_nflen
+
1191 tcp_ca_openreq_child(newsk
, dst
);
1193 tcp_sync_mss(newsk
, dst_mtu(dst
));
1194 newtp
->advmss
= tcp_mss_clamp(tcp_sk(sk
), dst_metric_advmss(dst
));
1196 tcp_initialize_rcv_mss(newsk
);
1198 newinet
->inet_daddr
= newinet
->inet_saddr
= LOOPBACK4_IPV6
;
1199 newinet
->inet_rcv_saddr
= LOOPBACK4_IPV6
;
1201 #ifdef CONFIG_TCP_MD5SIG
1202 /* Copy over the MD5 key from the original socket */
1203 key
= tcp_v6_md5_do_lookup(sk
, &newsk
->sk_v6_daddr
);
1205 /* We're using one, so create a matching key
1206 * on the newsk structure. If we fail to get
1207 * memory, then we end up not copying the key
1210 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newsk
->sk_v6_daddr
,
1211 AF_INET6
, 128, key
->key
, key
->keylen
,
1212 sk_gfp_mask(sk
, GFP_ATOMIC
));
1216 if (__inet_inherit_port(sk
, newsk
) < 0) {
1217 inet_csk_prepare_forced_close(newsk
);
1221 *own_req
= inet_ehash_nolisten(newsk
, req_to_sk(req_unhash
));
1223 tcp_move_syn(newtp
, req
);
1225 /* Clone pktoptions received with SYN, if we own the req */
1226 if (ireq
->pktopts
) {
1227 newnp
->pktoptions
= skb_clone(ireq
->pktopts
,
1228 sk_gfp_mask(sk
, GFP_ATOMIC
));
1229 consume_skb(ireq
->pktopts
);
1230 ireq
->pktopts
= NULL
;
1231 if (newnp
->pktoptions
) {
1232 tcp_v6_restore_cb(newnp
->pktoptions
);
1233 skb_set_owner_r(newnp
->pktoptions
, newsk
);
1241 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1249 /* The socket must have it's spinlock held when we get
1250 * here, unless it is a TCP_LISTEN socket.
1252 * We have a potential double-lock case here, so even when
1253 * doing backlog processing we use the BH locking scheme.
1254 * This is because we cannot sleep with the original spinlock
1257 static int tcp_v6_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1259 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1260 struct tcp_sock
*tp
;
1261 struct sk_buff
*opt_skb
= NULL
;
1263 /* Imagine: socket is IPv6. IPv4 packet arrives,
1264 goes to IPv4 receive handler and backlogged.
1265 From backlog it always goes here. Kerboom...
1266 Fortunately, tcp_rcv_established and rcv_established
1267 handle them correctly, but it is not case with
1268 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1271 if (skb
->protocol
== htons(ETH_P_IP
))
1272 return tcp_v4_do_rcv(sk
, skb
);
1275 * socket locking is here for SMP purposes as backlog rcv
1276 * is currently called with bh processing disabled.
1279 /* Do Stevens' IPV6_PKTOPTIONS.
1281 Yes, guys, it is the only place in our code, where we
1282 may make it not affecting IPv4.
1283 The rest of code is protocol independent,
1284 and I do not like idea to uglify IPv4.
1286 Actually, all the idea behind IPV6_PKTOPTIONS
1287 looks not very well thought. For now we latch
1288 options, received in the last packet, enqueued
1289 by tcp. Feel free to propose better solution.
1293 opt_skb
= skb_clone(skb
, sk_gfp_mask(sk
, GFP_ATOMIC
));
1295 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1296 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1298 sock_rps_save_rxhash(sk
, skb
);
1299 sk_mark_napi_id(sk
, skb
);
1301 if (inet_sk(sk
)->rx_dst_ifindex
!= skb
->skb_iif
||
1302 dst
->ops
->check(dst
, np
->rx_dst_cookie
) == NULL
) {
1304 sk
->sk_rx_dst
= NULL
;
1308 tcp_rcv_established(sk
, skb
, tcp_hdr(skb
));
1310 goto ipv6_pktoptions
;
1314 if (tcp_checksum_complete(skb
))
1317 if (sk
->sk_state
== TCP_LISTEN
) {
1318 struct sock
*nsk
= tcp_v6_cookie_check(sk
, skb
);
1324 if (tcp_child_process(sk
, nsk
, skb
))
1327 __kfree_skb(opt_skb
);
1331 sock_rps_save_rxhash(sk
, skb
);
1333 if (tcp_rcv_state_process(sk
, skb
))
1336 goto ipv6_pktoptions
;
1340 tcp_v6_send_reset(sk
, skb
);
1343 __kfree_skb(opt_skb
);
1347 TCP_INC_STATS(sock_net(sk
), TCP_MIB_CSUMERRORS
);
1348 TCP_INC_STATS(sock_net(sk
), TCP_MIB_INERRS
);
1353 /* Do you ask, what is it?
1355 1. skb was enqueued by tcp.
1356 2. skb is added to tail of read queue, rather than out of order.
1357 3. socket is not in passive state.
1358 4. Finally, it really contains options, which user wants to receive.
1361 if (TCP_SKB_CB(opt_skb
)->end_seq
== tp
->rcv_nxt
&&
1362 !((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
))) {
1363 if (np
->rxopt
.bits
.rxinfo
|| np
->rxopt
.bits
.rxoinfo
)
1364 np
->mcast_oif
= tcp_v6_iif(opt_skb
);
1365 if (np
->rxopt
.bits
.rxhlim
|| np
->rxopt
.bits
.rxohlim
)
1366 np
->mcast_hops
= ipv6_hdr(opt_skb
)->hop_limit
;
1367 if (np
->rxopt
.bits
.rxflow
|| np
->rxopt
.bits
.rxtclass
)
1368 np
->rcv_flowinfo
= ip6_flowinfo(ipv6_hdr(opt_skb
));
1370 np
->flow_label
= ip6_flowlabel(ipv6_hdr(opt_skb
));
1371 if (ipv6_opt_accepted(sk
, opt_skb
, &TCP_SKB_CB(opt_skb
)->header
.h6
)) {
1372 skb_set_owner_r(opt_skb
, sk
);
1373 tcp_v6_restore_cb(opt_skb
);
1374 opt_skb
= xchg(&np
->pktoptions
, opt_skb
);
1376 __kfree_skb(opt_skb
);
1377 opt_skb
= xchg(&np
->pktoptions
, NULL
);
1385 static void tcp_v6_fill_cb(struct sk_buff
*skb
, const struct ipv6hdr
*hdr
,
1386 const struct tcphdr
*th
)
1388 /* This is tricky: we move IP6CB at its correct location into
1389 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1390 * _decode_session6() uses IP6CB().
1391 * barrier() makes sure compiler won't play aliasing games.
1393 memmove(&TCP_SKB_CB(skb
)->header
.h6
, IP6CB(skb
),
1394 sizeof(struct inet6_skb_parm
));
1397 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1398 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1399 skb
->len
- th
->doff
*4);
1400 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1401 TCP_SKB_CB(skb
)->tcp_flags
= tcp_flag_byte(th
);
1402 TCP_SKB_CB(skb
)->tcp_tw_isn
= 0;
1403 TCP_SKB_CB(skb
)->ip_dsfield
= ipv6_get_dsfield(hdr
);
1404 TCP_SKB_CB(skb
)->sacked
= 0;
1405 TCP_SKB_CB(skb
)->has_rxtstamp
=
1406 skb
->tstamp
|| skb_hwtstamps(skb
)->hwtstamp
;
1409 static int tcp_v6_rcv(struct sk_buff
*skb
)
1411 int sdif
= inet6_sdif(skb
);
1412 const struct tcphdr
*th
;
1413 const struct ipv6hdr
*hdr
;
1417 struct net
*net
= dev_net(skb
->dev
);
1419 if (skb
->pkt_type
!= PACKET_HOST
)
1423 * Count it even if it's bad.
1425 __TCP_INC_STATS(net
, TCP_MIB_INSEGS
);
1427 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1430 th
= (const struct tcphdr
*)skb
->data
;
1432 if (unlikely(th
->doff
< sizeof(struct tcphdr
)/4))
1434 if (!pskb_may_pull(skb
, th
->doff
*4))
1437 if (skb_checksum_init(skb
, IPPROTO_TCP
, ip6_compute_pseudo
))
1440 th
= (const struct tcphdr
*)skb
->data
;
1441 hdr
= ipv6_hdr(skb
);
1444 sk
= __inet6_lookup_skb(&tcp_hashinfo
, skb
, __tcp_hdrlen(th
),
1445 th
->source
, th
->dest
, inet6_iif(skb
), sdif
,
1451 if (sk
->sk_state
== TCP_TIME_WAIT
)
1454 if (sk
->sk_state
== TCP_NEW_SYN_RECV
) {
1455 struct request_sock
*req
= inet_reqsk(sk
);
1458 sk
= req
->rsk_listener
;
1459 if (tcp_v6_inbound_md5_hash(sk
, skb
)) {
1460 sk_drops_add(sk
, skb
);
1464 if (tcp_checksum_complete(skb
)) {
1468 if (unlikely(sk
->sk_state
!= TCP_LISTEN
)) {
1469 inet_csk_reqsk_queue_drop_and_put(sk
, req
);
1475 if (!tcp_filter(sk
, skb
)) {
1476 th
= (const struct tcphdr
*)skb
->data
;
1477 hdr
= ipv6_hdr(skb
);
1478 tcp_v6_fill_cb(skb
, hdr
, th
);
1479 nsk
= tcp_check_req(sk
, skb
, req
, false);
1483 goto discard_and_relse
;
1487 tcp_v6_restore_cb(skb
);
1488 } else if (tcp_child_process(sk
, nsk
, skb
)) {
1489 tcp_v6_send_reset(nsk
, skb
);
1490 goto discard_and_relse
;
1496 if (hdr
->hop_limit
< inet6_sk(sk
)->min_hopcount
) {
1497 __NET_INC_STATS(net
, LINUX_MIB_TCPMINTTLDROP
);
1498 goto discard_and_relse
;
1501 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1502 goto discard_and_relse
;
1504 if (tcp_v6_inbound_md5_hash(sk
, skb
))
1505 goto discard_and_relse
;
1507 if (tcp_filter(sk
, skb
))
1508 goto discard_and_relse
;
1509 th
= (const struct tcphdr
*)skb
->data
;
1510 hdr
= ipv6_hdr(skb
);
1511 tcp_v6_fill_cb(skb
, hdr
, th
);
1515 if (sk
->sk_state
== TCP_LISTEN
) {
1516 ret
= tcp_v6_do_rcv(sk
, skb
);
1517 goto put_and_return
;
1520 sk_incoming_cpu_update(sk
);
1522 bh_lock_sock_nested(sk
);
1523 tcp_segs_in(tcp_sk(sk
), skb
);
1525 if (!sock_owned_by_user(sk
)) {
1526 ret
= tcp_v6_do_rcv(sk
, skb
);
1527 } else if (tcp_add_backlog(sk
, skb
)) {
1528 goto discard_and_relse
;
1535 return ret
? -1 : 0;
1538 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1541 tcp_v6_fill_cb(skb
, hdr
, th
);
1543 if (tcp_checksum_complete(skb
)) {
1545 __TCP_INC_STATS(net
, TCP_MIB_CSUMERRORS
);
1547 __TCP_INC_STATS(net
, TCP_MIB_INERRS
);
1549 tcp_v6_send_reset(NULL
, skb
);
1557 sk_drops_add(sk
, skb
);
1563 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1564 inet_twsk_put(inet_twsk(sk
));
1568 tcp_v6_fill_cb(skb
, hdr
, th
);
1570 if (tcp_checksum_complete(skb
)) {
1571 inet_twsk_put(inet_twsk(sk
));
1575 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1580 sk2
= inet6_lookup_listener(dev_net(skb
->dev
), &tcp_hashinfo
,
1581 skb
, __tcp_hdrlen(th
),
1582 &ipv6_hdr(skb
)->saddr
, th
->source
,
1583 &ipv6_hdr(skb
)->daddr
,
1585 tcp_v6_iif_l3_slave(skb
),
1588 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
1589 inet_twsk_deschedule_put(tw
);
1591 tcp_v6_restore_cb(skb
);
1599 tcp_v6_timewait_ack(sk
, skb
);
1602 tcp_v6_send_reset(sk
, skb
);
1603 inet_twsk_deschedule_put(inet_twsk(sk
));
1605 case TCP_TW_SUCCESS
:
1611 static void tcp_v6_early_demux(struct sk_buff
*skb
)
1613 const struct ipv6hdr
*hdr
;
1614 const struct tcphdr
*th
;
1617 if (skb
->pkt_type
!= PACKET_HOST
)
1620 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct tcphdr
)))
1623 hdr
= ipv6_hdr(skb
);
1626 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1629 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1630 sk
= __inet6_lookup_established(dev_net(skb
->dev
), &tcp_hashinfo
,
1631 &hdr
->saddr
, th
->source
,
1632 &hdr
->daddr
, ntohs(th
->dest
),
1633 inet6_iif(skb
), inet6_sdif(skb
));
1636 skb
->destructor
= sock_edemux
;
1637 if (sk_fullsock(sk
)) {
1638 struct dst_entry
*dst
= READ_ONCE(sk
->sk_rx_dst
);
1641 dst
= dst_check(dst
, inet6_sk(sk
)->rx_dst_cookie
);
1643 inet_sk(sk
)->rx_dst_ifindex
== skb
->skb_iif
)
1644 skb_dst_set_noref(skb
, dst
);
1649 static struct timewait_sock_ops tcp6_timewait_sock_ops
= {
1650 .twsk_obj_size
= sizeof(struct tcp6_timewait_sock
),
1651 .twsk_unique
= tcp_twsk_unique
,
1652 .twsk_destructor
= tcp_twsk_destructor
,
1655 static const struct inet_connection_sock_af_ops ipv6_specific
= {
1656 .queue_xmit
= inet6_csk_xmit
,
1657 .send_check
= tcp_v6_send_check
,
1658 .rebuild_header
= inet6_sk_rebuild_header
,
1659 .sk_rx_dst_set
= inet6_sk_rx_dst_set
,
1660 .conn_request
= tcp_v6_conn_request
,
1661 .syn_recv_sock
= tcp_v6_syn_recv_sock
,
1662 .net_header_len
= sizeof(struct ipv6hdr
),
1663 .net_frag_header_len
= sizeof(struct frag_hdr
),
1664 .setsockopt
= ipv6_setsockopt
,
1665 .getsockopt
= ipv6_getsockopt
,
1666 .addr2sockaddr
= inet6_csk_addr2sockaddr
,
1667 .sockaddr_len
= sizeof(struct sockaddr_in6
),
1668 #ifdef CONFIG_COMPAT
1669 .compat_setsockopt
= compat_ipv6_setsockopt
,
1670 .compat_getsockopt
= compat_ipv6_getsockopt
,
1672 .mtu_reduced
= tcp_v6_mtu_reduced
,
1675 #ifdef CONFIG_TCP_MD5SIG
1676 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific
= {
1677 .md5_lookup
= tcp_v6_md5_lookup
,
1678 .calc_md5_hash
= tcp_v6_md5_hash_skb
,
1679 .md5_parse
= tcp_v6_parse_md5_keys
,
1684 * TCP over IPv4 via INET6 API
1686 static const struct inet_connection_sock_af_ops ipv6_mapped
= {
1687 .queue_xmit
= ip_queue_xmit
,
1688 .send_check
= tcp_v4_send_check
,
1689 .rebuild_header
= inet_sk_rebuild_header
,
1690 .sk_rx_dst_set
= inet_sk_rx_dst_set
,
1691 .conn_request
= tcp_v6_conn_request
,
1692 .syn_recv_sock
= tcp_v6_syn_recv_sock
,
1693 .net_header_len
= sizeof(struct iphdr
),
1694 .setsockopt
= ipv6_setsockopt
,
1695 .getsockopt
= ipv6_getsockopt
,
1696 .addr2sockaddr
= inet6_csk_addr2sockaddr
,
1697 .sockaddr_len
= sizeof(struct sockaddr_in6
),
1698 #ifdef CONFIG_COMPAT
1699 .compat_setsockopt
= compat_ipv6_setsockopt
,
1700 .compat_getsockopt
= compat_ipv6_getsockopt
,
1702 .mtu_reduced
= tcp_v4_mtu_reduced
,
1705 #ifdef CONFIG_TCP_MD5SIG
1706 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific
= {
1707 .md5_lookup
= tcp_v4_md5_lookup
,
1708 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1709 .md5_parse
= tcp_v6_parse_md5_keys
,
1713 /* NOTE: A lot of things set to zero explicitly by call to
1714 * sk_alloc() so need not be done here.
1716 static int tcp_v6_init_sock(struct sock
*sk
)
1718 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1722 icsk
->icsk_af_ops
= &ipv6_specific
;
1724 #ifdef CONFIG_TCP_MD5SIG
1725 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv6_specific
;
1731 static void tcp_v6_destroy_sock(struct sock
*sk
)
1733 tcp_v4_destroy_sock(sk
);
1734 inet6_destroy_sock(sk
);
1737 #ifdef CONFIG_PROC_FS
1738 /* Proc filesystem TCPv6 sock list dumping. */
1739 static void get_openreq6(struct seq_file
*seq
,
1740 const struct request_sock
*req
, int i
)
1742 long ttd
= req
->rsk_timer
.expires
- jiffies
;
1743 const struct in6_addr
*src
= &inet_rsk(req
)->ir_v6_loc_addr
;
1744 const struct in6_addr
*dest
= &inet_rsk(req
)->ir_v6_rmt_addr
;
1750 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1751 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1753 src
->s6_addr32
[0], src
->s6_addr32
[1],
1754 src
->s6_addr32
[2], src
->s6_addr32
[3],
1755 inet_rsk(req
)->ir_num
,
1756 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1757 dest
->s6_addr32
[2], dest
->s6_addr32
[3],
1758 ntohs(inet_rsk(req
)->ir_rmt_port
),
1760 0, 0, /* could print option size, but that is af dependent. */
1761 1, /* timers active (only the expire timer) */
1762 jiffies_to_clock_t(ttd
),
1764 from_kuid_munged(seq_user_ns(seq
),
1765 sock_i_uid(req
->rsk_listener
)),
1766 0, /* non standard timer */
1767 0, /* open_requests have no inode */
1771 static void get_tcp6_sock(struct seq_file
*seq
, struct sock
*sp
, int i
)
1773 const struct in6_addr
*dest
, *src
;
1776 unsigned long timer_expires
;
1777 const struct inet_sock
*inet
= inet_sk(sp
);
1778 const struct tcp_sock
*tp
= tcp_sk(sp
);
1779 const struct inet_connection_sock
*icsk
= inet_csk(sp
);
1780 const struct fastopen_queue
*fastopenq
= &icsk
->icsk_accept_queue
.fastopenq
;
1784 dest
= &sp
->sk_v6_daddr
;
1785 src
= &sp
->sk_v6_rcv_saddr
;
1786 destp
= ntohs(inet
->inet_dport
);
1787 srcp
= ntohs(inet
->inet_sport
);
1789 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
||
1790 icsk
->icsk_pending
== ICSK_TIME_REO_TIMEOUT
||
1791 icsk
->icsk_pending
== ICSK_TIME_LOSS_PROBE
) {
1793 timer_expires
= icsk
->icsk_timeout
;
1794 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
1796 timer_expires
= icsk
->icsk_timeout
;
1797 } else if (timer_pending(&sp
->sk_timer
)) {
1799 timer_expires
= sp
->sk_timer
.expires
;
1802 timer_expires
= jiffies
;
1805 state
= sk_state_load(sp
);
1806 if (state
== TCP_LISTEN
)
1807 rx_queue
= sp
->sk_ack_backlog
;
1809 /* Because we don't lock the socket,
1810 * we might find a transient negative value.
1812 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
1815 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1816 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1818 src
->s6_addr32
[0], src
->s6_addr32
[1],
1819 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1820 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1821 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1823 tp
->write_seq
- tp
->snd_una
,
1826 jiffies_delta_to_clock_t(timer_expires
- jiffies
),
1827 icsk
->icsk_retransmits
,
1828 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(sp
)),
1829 icsk
->icsk_probes_out
,
1831 refcount_read(&sp
->sk_refcnt
), sp
,
1832 jiffies_to_clock_t(icsk
->icsk_rto
),
1833 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
1834 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
1836 state
== TCP_LISTEN
?
1837 fastopenq
->max_qlen
:
1838 (tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
)
1842 static void get_timewait6_sock(struct seq_file
*seq
,
1843 struct inet_timewait_sock
*tw
, int i
)
1845 long delta
= tw
->tw_timer
.expires
- jiffies
;
1846 const struct in6_addr
*dest
, *src
;
1849 dest
= &tw
->tw_v6_daddr
;
1850 src
= &tw
->tw_v6_rcv_saddr
;
1851 destp
= ntohs(tw
->tw_dport
);
1852 srcp
= ntohs(tw
->tw_sport
);
1855 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1856 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1858 src
->s6_addr32
[0], src
->s6_addr32
[1],
1859 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1860 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1861 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1862 tw
->tw_substate
, 0, 0,
1863 3, jiffies_delta_to_clock_t(delta
), 0, 0, 0, 0,
1864 refcount_read(&tw
->tw_refcnt
), tw
);
1867 static int tcp6_seq_show(struct seq_file
*seq
, void *v
)
1869 struct tcp_iter_state
*st
;
1870 struct sock
*sk
= v
;
1872 if (v
== SEQ_START_TOKEN
) {
1877 "st tx_queue rx_queue tr tm->when retrnsmt"
1878 " uid timeout inode\n");
1883 if (sk
->sk_state
== TCP_TIME_WAIT
)
1884 get_timewait6_sock(seq
, v
, st
->num
);
1885 else if (sk
->sk_state
== TCP_NEW_SYN_RECV
)
1886 get_openreq6(seq
, v
, st
->num
);
1888 get_tcp6_sock(seq
, v
, st
->num
);
1893 static const struct file_operations tcp6_afinfo_seq_fops
= {
1894 .owner
= THIS_MODULE
,
1895 .open
= tcp_seq_open
,
1897 .llseek
= seq_lseek
,
1898 .release
= seq_release_net
1901 static struct tcp_seq_afinfo tcp6_seq_afinfo
= {
1904 .seq_fops
= &tcp6_afinfo_seq_fops
,
1906 .show
= tcp6_seq_show
,
1910 int __net_init
tcp6_proc_init(struct net
*net
)
1912 return tcp_proc_register(net
, &tcp6_seq_afinfo
);
1915 void tcp6_proc_exit(struct net
*net
)
1917 tcp_proc_unregister(net
, &tcp6_seq_afinfo
);
1921 struct proto tcpv6_prot
= {
1923 .owner
= THIS_MODULE
,
1925 .connect
= tcp_v6_connect
,
1926 .disconnect
= tcp_disconnect
,
1927 .accept
= inet_csk_accept
,
1929 .init
= tcp_v6_init_sock
,
1930 .destroy
= tcp_v6_destroy_sock
,
1931 .shutdown
= tcp_shutdown
,
1932 .setsockopt
= tcp_setsockopt
,
1933 .getsockopt
= tcp_getsockopt
,
1934 .keepalive
= tcp_set_keepalive
,
1935 .recvmsg
= tcp_recvmsg
,
1936 .sendmsg
= tcp_sendmsg
,
1937 .sendpage
= tcp_sendpage
,
1938 .backlog_rcv
= tcp_v6_do_rcv
,
1939 .release_cb
= tcp_release_cb
,
1941 .unhash
= inet_unhash
,
1942 .get_port
= inet_csk_get_port
,
1943 .enter_memory_pressure
= tcp_enter_memory_pressure
,
1944 .leave_memory_pressure
= tcp_leave_memory_pressure
,
1945 .stream_memory_free
= tcp_stream_memory_free
,
1946 .sockets_allocated
= &tcp_sockets_allocated
,
1947 .memory_allocated
= &tcp_memory_allocated
,
1948 .memory_pressure
= &tcp_memory_pressure
,
1949 .orphan_count
= &tcp_orphan_count
,
1950 .sysctl_mem
= sysctl_tcp_mem
,
1951 .sysctl_wmem_offset
= offsetof(struct net
, ipv4
.sysctl_tcp_wmem
),
1952 .sysctl_rmem_offset
= offsetof(struct net
, ipv4
.sysctl_tcp_rmem
),
1953 .max_header
= MAX_TCP_HEADER
,
1954 .obj_size
= sizeof(struct tcp6_sock
),
1955 .slab_flags
= SLAB_TYPESAFE_BY_RCU
,
1956 .twsk_prot
= &tcp6_timewait_sock_ops
,
1957 .rsk_prot
= &tcp6_request_sock_ops
,
1958 .h
.hashinfo
= &tcp_hashinfo
,
1959 .no_autobind
= true,
1960 #ifdef CONFIG_COMPAT
1961 .compat_setsockopt
= compat_tcp_setsockopt
,
1962 .compat_getsockopt
= compat_tcp_getsockopt
,
1964 .diag_destroy
= tcp_abort
,
1967 /* thinking of making this const? Don't.
1968 * early_demux can change based on sysctl.
1970 static struct inet6_protocol tcpv6_protocol
= {
1971 .early_demux
= tcp_v6_early_demux
,
1972 .early_demux_handler
= tcp_v6_early_demux
,
1973 .handler
= tcp_v6_rcv
,
1974 .err_handler
= tcp_v6_err
,
1975 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1978 static struct inet_protosw tcpv6_protosw
= {
1979 .type
= SOCK_STREAM
,
1980 .protocol
= IPPROTO_TCP
,
1981 .prot
= &tcpv6_prot
,
1982 .ops
= &inet6_stream_ops
,
1983 .flags
= INET_PROTOSW_PERMANENT
|
1987 static int __net_init
tcpv6_net_init(struct net
*net
)
1989 return inet_ctl_sock_create(&net
->ipv6
.tcp_sk
, PF_INET6
,
1990 SOCK_RAW
, IPPROTO_TCP
, net
);
1993 static void __net_exit
tcpv6_net_exit(struct net
*net
)
1995 inet_ctl_sock_destroy(net
->ipv6
.tcp_sk
);
1998 static void __net_exit
tcpv6_net_exit_batch(struct list_head
*net_exit_list
)
2000 inet_twsk_purge(&tcp_hashinfo
, AF_INET6
);
2003 static struct pernet_operations tcpv6_net_ops
= {
2004 .init
= tcpv6_net_init
,
2005 .exit
= tcpv6_net_exit
,
2006 .exit_batch
= tcpv6_net_exit_batch
,
2009 int __init
tcpv6_init(void)
2013 ret
= inet6_add_protocol(&tcpv6_protocol
, IPPROTO_TCP
);
2017 /* register inet6 protocol */
2018 ret
= inet6_register_protosw(&tcpv6_protosw
);
2020 goto out_tcpv6_protocol
;
2022 ret
= register_pernet_subsys(&tcpv6_net_ops
);
2024 goto out_tcpv6_protosw
;
2029 inet6_unregister_protosw(&tcpv6_protosw
);
2031 inet6_del_protocol(&tcpv6_protocol
, IPPROTO_TCP
);
2035 void tcpv6_exit(void)
2037 unregister_pernet_subsys(&tcpv6_net_ops
);
2038 inet6_unregister_protosw(&tcpv6_protosw
);
2039 inet6_del_protocol(&tcpv6_protocol
, IPPROTO_TCP
);