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>
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/netdma.h>
63 #include <net/inet_common.h>
64 #include <net/secure_seq.h>
65 #include <net/tcp_memcontrol.h>
66 #include <net/busy_poll.h>
68 #include <asm/uaccess.h>
70 #include <linux/proc_fs.h>
71 #include <linux/seq_file.h>
73 #include <linux/crypto.h>
74 #include <linux/scatterlist.h>
76 static void tcp_v6_send_reset(struct sock
*sk
, struct sk_buff
*skb
);
77 static void tcp_v6_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
78 struct request_sock
*req
);
80 static int tcp_v6_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
82 static const struct inet_connection_sock_af_ops ipv6_mapped
;
83 static const struct inet_connection_sock_af_ops ipv6_specific
;
84 #ifdef CONFIG_TCP_MD5SIG
85 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific
;
86 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific
;
88 static struct tcp_md5sig_key
*tcp_v6_md5_do_lookup(struct sock
*sk
,
89 const struct in6_addr
*addr
)
95 static void inet6_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
97 struct dst_entry
*dst
= skb_dst(skb
);
98 const struct rt6_info
*rt
= (const struct rt6_info
*)dst
;
102 inet_sk(sk
)->rx_dst_ifindex
= skb
->skb_iif
;
104 inet6_sk(sk
)->rx_dst_cookie
= rt
->rt6i_node
->fn_sernum
;
107 static void tcp_v6_hash(struct sock
*sk
)
109 if (sk
->sk_state
!= TCP_CLOSE
) {
110 if (inet_csk(sk
)->icsk_af_ops
== &ipv6_mapped
) {
115 __inet6_hash(sk
, NULL
);
120 static __u32
tcp_v6_init_sequence(const struct sk_buff
*skb
)
122 return secure_tcpv6_sequence_number(ipv6_hdr(skb
)->daddr
.s6_addr32
,
123 ipv6_hdr(skb
)->saddr
.s6_addr32
,
125 tcp_hdr(skb
)->source
);
128 static int tcp_v6_connect(struct sock
*sk
, struct sockaddr
*uaddr
,
131 struct sockaddr_in6
*usin
= (struct sockaddr_in6
*) uaddr
;
132 struct inet_sock
*inet
= inet_sk(sk
);
133 struct inet_connection_sock
*icsk
= inet_csk(sk
);
134 struct ipv6_pinfo
*np
= inet6_sk(sk
);
135 struct tcp_sock
*tp
= tcp_sk(sk
);
136 struct in6_addr
*saddr
= NULL
, *final_p
, final
;
139 struct dst_entry
*dst
;
143 if (addr_len
< SIN6_LEN_RFC2133
)
146 if (usin
->sin6_family
!= AF_INET6
)
147 return -EAFNOSUPPORT
;
149 memset(&fl6
, 0, sizeof(fl6
));
152 fl6
.flowlabel
= usin
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
153 IP6_ECN_flow_init(fl6
.flowlabel
);
154 if (fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) {
155 struct ip6_flowlabel
*flowlabel
;
156 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
157 if (flowlabel
== NULL
)
159 usin
->sin6_addr
= flowlabel
->dst
;
160 fl6_sock_release(flowlabel
);
165 * connect() to INADDR_ANY means loopback (BSD'ism).
168 if(ipv6_addr_any(&usin
->sin6_addr
))
169 usin
->sin6_addr
.s6_addr
[15] = 0x1;
171 addr_type
= ipv6_addr_type(&usin
->sin6_addr
);
173 if(addr_type
& IPV6_ADDR_MULTICAST
)
176 if (addr_type
&IPV6_ADDR_LINKLOCAL
) {
177 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
178 usin
->sin6_scope_id
) {
179 /* If interface is set while binding, indices
182 if (sk
->sk_bound_dev_if
&&
183 sk
->sk_bound_dev_if
!= usin
->sin6_scope_id
)
186 sk
->sk_bound_dev_if
= usin
->sin6_scope_id
;
189 /* Connect to link-local address requires an interface */
190 if (!sk
->sk_bound_dev_if
)
194 if (tp
->rx_opt
.ts_recent_stamp
&&
195 !ipv6_addr_equal(&np
->daddr
, &usin
->sin6_addr
)) {
196 tp
->rx_opt
.ts_recent
= 0;
197 tp
->rx_opt
.ts_recent_stamp
= 0;
201 np
->daddr
= usin
->sin6_addr
;
202 np
->flow_label
= fl6
.flowlabel
;
208 if (addr_type
== IPV6_ADDR_MAPPED
) {
209 u32 exthdrlen
= icsk
->icsk_ext_hdr_len
;
210 struct sockaddr_in sin
;
212 SOCK_DEBUG(sk
, "connect: ipv4 mapped\n");
214 if (__ipv6_only_sock(sk
))
217 sin
.sin_family
= AF_INET
;
218 sin
.sin_port
= usin
->sin6_port
;
219 sin
.sin_addr
.s_addr
= usin
->sin6_addr
.s6_addr32
[3];
221 icsk
->icsk_af_ops
= &ipv6_mapped
;
222 sk
->sk_backlog_rcv
= tcp_v4_do_rcv
;
223 #ifdef CONFIG_TCP_MD5SIG
224 tp
->af_specific
= &tcp_sock_ipv6_mapped_specific
;
227 err
= tcp_v4_connect(sk
, (struct sockaddr
*)&sin
, sizeof(sin
));
230 icsk
->icsk_ext_hdr_len
= exthdrlen
;
231 icsk
->icsk_af_ops
= &ipv6_specific
;
232 sk
->sk_backlog_rcv
= tcp_v6_do_rcv
;
233 #ifdef CONFIG_TCP_MD5SIG
234 tp
->af_specific
= &tcp_sock_ipv6_specific
;
238 ipv6_addr_set_v4mapped(inet
->inet_saddr
, &np
->saddr
);
239 ipv6_addr_set_v4mapped(inet
->inet_rcv_saddr
,
246 if (!ipv6_addr_any(&np
->rcv_saddr
))
247 saddr
= &np
->rcv_saddr
;
249 fl6
.flowi6_proto
= IPPROTO_TCP
;
250 fl6
.daddr
= np
->daddr
;
251 fl6
.saddr
= saddr
? *saddr
: np
->saddr
;
252 fl6
.flowi6_oif
= sk
->sk_bound_dev_if
;
253 fl6
.flowi6_mark
= sk
->sk_mark
;
254 fl6
.fl6_dport
= usin
->sin6_port
;
255 fl6
.fl6_sport
= inet
->inet_sport
;
257 final_p
= fl6_update_dst(&fl6
, np
->opt
, &final
);
259 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
261 dst
= ip6_dst_lookup_flow(sk
, &fl6
, final_p
, true);
269 np
->rcv_saddr
= *saddr
;
272 /* set the source address */
274 inet
->inet_rcv_saddr
= LOOPBACK4_IPV6
;
276 sk
->sk_gso_type
= SKB_GSO_TCPV6
;
277 __ip6_dst_store(sk
, dst
, NULL
, NULL
);
279 rt
= (struct rt6_info
*) dst
;
280 if (tcp_death_row
.sysctl_tw_recycle
&&
281 !tp
->rx_opt
.ts_recent_stamp
&&
282 ipv6_addr_equal(&rt
->rt6i_dst
.addr
, &np
->daddr
))
283 tcp_fetch_timewait_stamp(sk
, dst
);
285 icsk
->icsk_ext_hdr_len
= 0;
287 icsk
->icsk_ext_hdr_len
= (np
->opt
->opt_flen
+
290 tp
->rx_opt
.mss_clamp
= IPV6_MIN_MTU
- sizeof(struct tcphdr
) - sizeof(struct ipv6hdr
);
292 inet
->inet_dport
= usin
->sin6_port
;
294 tcp_set_state(sk
, TCP_SYN_SENT
);
295 err
= inet6_hash_connect(&tcp_death_row
, sk
);
299 if (!tp
->write_seq
&& likely(!tp
->repair
))
300 tp
->write_seq
= secure_tcpv6_sequence_number(np
->saddr
.s6_addr32
,
305 err
= tcp_connect(sk
);
312 tcp_set_state(sk
, TCP_CLOSE
);
315 inet
->inet_dport
= 0;
316 sk
->sk_route_caps
= 0;
320 static void tcp_v6_mtu_reduced(struct sock
*sk
)
322 struct dst_entry
*dst
;
324 if ((1 << sk
->sk_state
) & (TCPF_LISTEN
| TCPF_CLOSE
))
327 dst
= inet6_csk_update_pmtu(sk
, tcp_sk(sk
)->mtu_info
);
331 if (inet_csk(sk
)->icsk_pmtu_cookie
> dst_mtu(dst
)) {
332 tcp_sync_mss(sk
, dst_mtu(dst
));
333 tcp_simple_retransmit(sk
);
337 static void tcp_v6_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
338 u8 type
, u8 code
, int offset
, __be32 info
)
340 const struct ipv6hdr
*hdr
= (const struct ipv6hdr
*)skb
->data
;
341 const struct tcphdr
*th
= (struct tcphdr
*)(skb
->data
+offset
);
342 struct ipv6_pinfo
*np
;
347 struct net
*net
= dev_net(skb
->dev
);
349 sk
= inet6_lookup(net
, &tcp_hashinfo
, &hdr
->daddr
,
350 th
->dest
, &hdr
->saddr
, th
->source
, skb
->dev
->ifindex
);
353 ICMP6_INC_STATS_BH(net
, __in6_dev_get(skb
->dev
),
358 if (sk
->sk_state
== TCP_TIME_WAIT
) {
359 inet_twsk_put(inet_twsk(sk
));
364 if (sock_owned_by_user(sk
) && type
!= ICMPV6_PKT_TOOBIG
)
365 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
367 if (sk
->sk_state
== TCP_CLOSE
)
370 if (ipv6_hdr(skb
)->hop_limit
< inet6_sk(sk
)->min_hopcount
) {
371 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
376 seq
= ntohl(th
->seq
);
377 if (sk
->sk_state
!= TCP_LISTEN
&&
378 !between(seq
, tp
->snd_una
, tp
->snd_nxt
)) {
379 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
385 if (type
== NDISC_REDIRECT
) {
386 struct dst_entry
*dst
= __sk_dst_check(sk
, np
->dst_cookie
);
389 dst
->ops
->redirect(dst
, sk
, skb
);
393 if (type
== ICMPV6_PKT_TOOBIG
) {
394 /* We are not interested in TCP_LISTEN and open_requests
395 * (SYN-ACKs send out by Linux are always <576bytes so
396 * they should go through unfragmented).
398 if (sk
->sk_state
== TCP_LISTEN
)
401 tp
->mtu_info
= ntohl(info
);
402 if (!sock_owned_by_user(sk
))
403 tcp_v6_mtu_reduced(sk
);
404 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED
,
410 icmpv6_err_convert(type
, code
, &err
);
412 /* Might be for an request_sock */
413 switch (sk
->sk_state
) {
414 struct request_sock
*req
, **prev
;
416 if (sock_owned_by_user(sk
))
419 req
= inet6_csk_search_req(sk
, &prev
, th
->dest
, &hdr
->daddr
,
420 &hdr
->saddr
, inet6_iif(skb
));
424 /* ICMPs are not backlogged, hence we cannot get
425 * an established socket here.
427 WARN_ON(req
->sk
!= NULL
);
429 if (seq
!= tcp_rsk(req
)->snt_isn
) {
430 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
434 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
435 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
439 case TCP_SYN_RECV
: /* Cannot happen.
440 It can, it SYNs are crossed. --ANK */
441 if (!sock_owned_by_user(sk
)) {
443 sk
->sk_error_report(sk
); /* Wake people up to see the error (see connect in sock.c) */
447 sk
->sk_err_soft
= err
;
451 if (!sock_owned_by_user(sk
) && np
->recverr
) {
453 sk
->sk_error_report(sk
);
455 sk
->sk_err_soft
= err
;
463 static int tcp_v6_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
465 struct request_sock
*req
,
468 struct inet6_request_sock
*treq
= inet6_rsk(req
);
469 struct ipv6_pinfo
*np
= inet6_sk(sk
);
470 struct sk_buff
* skb
;
473 /* First, grab a route. */
474 if (!dst
&& (dst
= inet6_csk_route_req(sk
, fl6
, req
)) == NULL
)
477 skb
= tcp_make_synack(sk
, dst
, req
, NULL
);
480 __tcp_v6_send_check(skb
, &treq
->loc_addr
, &treq
->rmt_addr
);
482 fl6
->daddr
= treq
->rmt_addr
;
483 skb_set_queue_mapping(skb
, queue_mapping
);
484 err
= ip6_xmit(sk
, skb
, fl6
, np
->opt
, np
->tclass
);
485 err
= net_xmit_eval(err
);
492 static int tcp_v6_rtx_synack(struct sock
*sk
, struct request_sock
*req
)
497 res
= tcp_v6_send_synack(sk
, NULL
, &fl6
, req
, 0);
499 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_RETRANSSEGS
);
503 static void tcp_v6_reqsk_destructor(struct request_sock
*req
)
505 kfree_skb(inet6_rsk(req
)->pktopts
);
508 #ifdef CONFIG_TCP_MD5SIG
509 static struct tcp_md5sig_key
*tcp_v6_md5_do_lookup(struct sock
*sk
,
510 const struct in6_addr
*addr
)
512 return tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)addr
, AF_INET6
);
515 static struct tcp_md5sig_key
*tcp_v6_md5_lookup(struct sock
*sk
,
516 struct sock
*addr_sk
)
518 return tcp_v6_md5_do_lookup(sk
, &inet6_sk(addr_sk
)->daddr
);
521 static struct tcp_md5sig_key
*tcp_v6_reqsk_md5_lookup(struct sock
*sk
,
522 struct request_sock
*req
)
524 return tcp_v6_md5_do_lookup(sk
, &inet6_rsk(req
)->rmt_addr
);
527 static int tcp_v6_parse_md5_keys (struct sock
*sk
, char __user
*optval
,
530 struct tcp_md5sig cmd
;
531 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*)&cmd
.tcpm_addr
;
533 if (optlen
< sizeof(cmd
))
536 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
539 if (sin6
->sin6_family
!= AF_INET6
)
542 if (!cmd
.tcpm_keylen
) {
543 if (ipv6_addr_v4mapped(&sin6
->sin6_addr
))
544 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
.s6_addr32
[3],
546 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
,
550 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
553 if (ipv6_addr_v4mapped(&sin6
->sin6_addr
))
554 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
.s6_addr32
[3],
555 AF_INET
, cmd
.tcpm_key
, cmd
.tcpm_keylen
, GFP_KERNEL
);
557 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin6
->sin6_addr
,
558 AF_INET6
, cmd
.tcpm_key
, cmd
.tcpm_keylen
, GFP_KERNEL
);
561 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
562 const struct in6_addr
*daddr
,
563 const struct in6_addr
*saddr
, int nbytes
)
565 struct tcp6_pseudohdr
*bp
;
566 struct scatterlist sg
;
568 bp
= &hp
->md5_blk
.ip6
;
569 /* 1. TCP pseudo-header (RFC2460) */
572 bp
->protocol
= cpu_to_be32(IPPROTO_TCP
);
573 bp
->len
= cpu_to_be32(nbytes
);
575 sg_init_one(&sg
, bp
, sizeof(*bp
));
576 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
579 static int tcp_v6_md5_hash_hdr(char *md5_hash
, struct tcp_md5sig_key
*key
,
580 const struct in6_addr
*daddr
, struct in6_addr
*saddr
,
581 const struct tcphdr
*th
)
583 struct tcp_md5sig_pool
*hp
;
584 struct hash_desc
*desc
;
586 hp
= tcp_get_md5sig_pool();
588 goto clear_hash_noput
;
589 desc
= &hp
->md5_desc
;
591 if (crypto_hash_init(desc
))
593 if (tcp_v6_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
595 if (tcp_md5_hash_header(hp
, th
))
597 if (tcp_md5_hash_key(hp
, key
))
599 if (crypto_hash_final(desc
, md5_hash
))
602 tcp_put_md5sig_pool();
606 tcp_put_md5sig_pool();
608 memset(md5_hash
, 0, 16);
612 static int tcp_v6_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
613 const struct sock
*sk
,
614 const struct request_sock
*req
,
615 const struct sk_buff
*skb
)
617 const struct in6_addr
*saddr
, *daddr
;
618 struct tcp_md5sig_pool
*hp
;
619 struct hash_desc
*desc
;
620 const struct tcphdr
*th
= tcp_hdr(skb
);
623 saddr
= &inet6_sk(sk
)->saddr
;
624 daddr
= &inet6_sk(sk
)->daddr
;
626 saddr
= &inet6_rsk(req
)->loc_addr
;
627 daddr
= &inet6_rsk(req
)->rmt_addr
;
629 const struct ipv6hdr
*ip6h
= ipv6_hdr(skb
);
630 saddr
= &ip6h
->saddr
;
631 daddr
= &ip6h
->daddr
;
634 hp
= tcp_get_md5sig_pool();
636 goto clear_hash_noput
;
637 desc
= &hp
->md5_desc
;
639 if (crypto_hash_init(desc
))
642 if (tcp_v6_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
644 if (tcp_md5_hash_header(hp
, th
))
646 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
648 if (tcp_md5_hash_key(hp
, key
))
650 if (crypto_hash_final(desc
, md5_hash
))
653 tcp_put_md5sig_pool();
657 tcp_put_md5sig_pool();
659 memset(md5_hash
, 0, 16);
663 static int tcp_v6_inbound_md5_hash(struct sock
*sk
, const struct sk_buff
*skb
)
665 const __u8
*hash_location
= NULL
;
666 struct tcp_md5sig_key
*hash_expected
;
667 const struct ipv6hdr
*ip6h
= ipv6_hdr(skb
);
668 const struct tcphdr
*th
= tcp_hdr(skb
);
672 hash_expected
= tcp_v6_md5_do_lookup(sk
, &ip6h
->saddr
);
673 hash_location
= tcp_parse_md5sig_option(th
);
675 /* We've parsed the options - do we have a hash? */
676 if (!hash_expected
&& !hash_location
)
679 if (hash_expected
&& !hash_location
) {
680 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
684 if (!hash_expected
&& hash_location
) {
685 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
689 /* check the signature */
690 genhash
= tcp_v6_md5_hash_skb(newhash
,
694 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
695 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
696 genhash
? "failed" : "mismatch",
697 &ip6h
->saddr
, ntohs(th
->source
),
698 &ip6h
->daddr
, ntohs(th
->dest
));
705 struct request_sock_ops tcp6_request_sock_ops __read_mostly
= {
707 .obj_size
= sizeof(struct tcp6_request_sock
),
708 .rtx_syn_ack
= tcp_v6_rtx_synack
,
709 .send_ack
= tcp_v6_reqsk_send_ack
,
710 .destructor
= tcp_v6_reqsk_destructor
,
711 .send_reset
= tcp_v6_send_reset
,
712 .syn_ack_timeout
= tcp_syn_ack_timeout
,
715 #ifdef CONFIG_TCP_MD5SIG
716 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops
= {
717 .md5_lookup
= tcp_v6_reqsk_md5_lookup
,
718 .calc_md5_hash
= tcp_v6_md5_hash_skb
,
722 static void tcp_v6_send_response(struct sk_buff
*skb
, u32 seq
, u32 ack
, u32 win
,
723 u32 tsval
, u32 tsecr
,
724 struct tcp_md5sig_key
*key
, int rst
, u8 tclass
)
726 const struct tcphdr
*th
= tcp_hdr(skb
);
728 struct sk_buff
*buff
;
730 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
731 struct sock
*ctl_sk
= net
->ipv6
.tcp_sk
;
732 unsigned int tot_len
= sizeof(struct tcphdr
);
733 struct dst_entry
*dst
;
737 tot_len
+= TCPOLEN_TSTAMP_ALIGNED
;
738 #ifdef CONFIG_TCP_MD5SIG
740 tot_len
+= TCPOLEN_MD5SIG_ALIGNED
;
743 buff
= alloc_skb(MAX_HEADER
+ sizeof(struct ipv6hdr
) + tot_len
,
748 skb_reserve(buff
, MAX_HEADER
+ sizeof(struct ipv6hdr
) + tot_len
);
750 t1
= (struct tcphdr
*) skb_push(buff
, tot_len
);
751 skb_reset_transport_header(buff
);
753 /* Swap the send and the receive. */
754 memset(t1
, 0, sizeof(*t1
));
755 t1
->dest
= th
->source
;
756 t1
->source
= th
->dest
;
757 t1
->doff
= tot_len
/ 4;
758 t1
->seq
= htonl(seq
);
759 t1
->ack_seq
= htonl(ack
);
760 t1
->ack
= !rst
|| !th
->ack
;
762 t1
->window
= htons(win
);
764 topt
= (__be32
*)(t1
+ 1);
767 *topt
++ = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
768 (TCPOPT_TIMESTAMP
<< 8) | TCPOLEN_TIMESTAMP
);
769 *topt
++ = htonl(tsval
);
770 *topt
++ = htonl(tsecr
);
773 #ifdef CONFIG_TCP_MD5SIG
775 *topt
++ = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
776 (TCPOPT_MD5SIG
<< 8) | TCPOLEN_MD5SIG
);
777 tcp_v6_md5_hash_hdr((__u8
*)topt
, key
,
778 &ipv6_hdr(skb
)->saddr
,
779 &ipv6_hdr(skb
)->daddr
, t1
);
783 memset(&fl6
, 0, sizeof(fl6
));
784 fl6
.daddr
= ipv6_hdr(skb
)->saddr
;
785 fl6
.saddr
= ipv6_hdr(skb
)->daddr
;
787 buff
->ip_summed
= CHECKSUM_PARTIAL
;
790 __tcp_v6_send_check(buff
, &fl6
.saddr
, &fl6
.daddr
);
792 fl6
.flowi6_proto
= IPPROTO_TCP
;
793 if (ipv6_addr_type(&fl6
.daddr
) & IPV6_ADDR_LINKLOCAL
)
794 fl6
.flowi6_oif
= inet6_iif(skb
);
795 fl6
.fl6_dport
= t1
->dest
;
796 fl6
.fl6_sport
= t1
->source
;
797 security_skb_classify_flow(skb
, flowi6_to_flowi(&fl6
));
799 /* Pass a socket to ip6_dst_lookup either it is for RST
800 * Underlying function will use this to retrieve the network
803 dst
= ip6_dst_lookup_flow(ctl_sk
, &fl6
, NULL
, false);
805 skb_dst_set(buff
, dst
);
806 ip6_xmit(ctl_sk
, buff
, &fl6
, NULL
, tclass
);
807 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
809 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
816 static void tcp_v6_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
818 const struct tcphdr
*th
= tcp_hdr(skb
);
819 u32 seq
= 0, ack_seq
= 0;
820 struct tcp_md5sig_key
*key
= NULL
;
821 #ifdef CONFIG_TCP_MD5SIG
822 const __u8
*hash_location
= NULL
;
823 struct ipv6hdr
*ipv6h
= ipv6_hdr(skb
);
824 unsigned char newhash
[16];
826 struct sock
*sk1
= NULL
;
832 if (!ipv6_unicast_destination(skb
))
835 #ifdef CONFIG_TCP_MD5SIG
836 hash_location
= tcp_parse_md5sig_option(th
);
837 if (!sk
&& hash_location
) {
839 * active side is lost. Try to find listening socket through
840 * source port, and then find md5 key through listening socket.
841 * we are not loose security here:
842 * Incoming packet is checked with md5 hash with finding key,
843 * no RST generated if md5 hash doesn't match.
845 sk1
= inet6_lookup_listener(dev_net(skb_dst(skb
)->dev
),
846 &tcp_hashinfo
, &ipv6h
->saddr
,
847 th
->source
, &ipv6h
->daddr
,
848 ntohs(th
->source
), inet6_iif(skb
));
853 key
= tcp_v6_md5_do_lookup(sk1
, &ipv6h
->saddr
);
857 genhash
= tcp_v6_md5_hash_skb(newhash
, key
, NULL
, NULL
, skb
);
858 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
861 key
= sk
? tcp_v6_md5_do_lookup(sk
, &ipv6h
->saddr
) : NULL
;
866 seq
= ntohl(th
->ack_seq
);
868 ack_seq
= ntohl(th
->seq
) + th
->syn
+ th
->fin
+ skb
->len
-
871 tcp_v6_send_response(skb
, seq
, ack_seq
, 0, 0, 0, key
, 1, 0);
873 #ifdef CONFIG_TCP_MD5SIG
882 static void tcp_v6_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
883 u32 win
, u32 tsval
, u32 tsecr
,
884 struct tcp_md5sig_key
*key
, u8 tclass
)
886 tcp_v6_send_response(skb
, seq
, ack
, win
, tsval
, tsecr
, key
, 0, tclass
);
889 static void tcp_v6_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
891 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
892 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
894 tcp_v6_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
895 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
896 tcp_time_stamp
+ tcptw
->tw_ts_offset
,
897 tcptw
->tw_ts_recent
, tcp_twsk_md5_key(tcptw
),
903 static void tcp_v6_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
904 struct request_sock
*req
)
906 tcp_v6_send_ack(skb
, tcp_rsk(req
)->snt_isn
+ 1, tcp_rsk(req
)->rcv_isn
+ 1,
907 req
->rcv_wnd
, tcp_time_stamp
, req
->ts_recent
,
908 tcp_v6_md5_do_lookup(sk
, &ipv6_hdr(skb
)->daddr
), 0);
912 static struct sock
*tcp_v6_hnd_req(struct sock
*sk
,struct sk_buff
*skb
)
914 struct request_sock
*req
, **prev
;
915 const struct tcphdr
*th
= tcp_hdr(skb
);
918 /* Find possible connection requests. */
919 req
= inet6_csk_search_req(sk
, &prev
, th
->source
,
920 &ipv6_hdr(skb
)->saddr
,
921 &ipv6_hdr(skb
)->daddr
, inet6_iif(skb
));
923 return tcp_check_req(sk
, skb
, req
, prev
, false);
925 nsk
= __inet6_lookup_established(sock_net(sk
), &tcp_hashinfo
,
926 &ipv6_hdr(skb
)->saddr
, th
->source
,
927 &ipv6_hdr(skb
)->daddr
, ntohs(th
->dest
), inet6_iif(skb
));
930 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
934 inet_twsk_put(inet_twsk(nsk
));
938 #ifdef CONFIG_SYN_COOKIES
940 sk
= cookie_v6_check(sk
, skb
);
945 /* FIXME: this is substantially similar to the ipv4 code.
946 * Can some kind of merge be done? -- erics
948 static int tcp_v6_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
950 struct tcp_options_received tmp_opt
;
951 struct request_sock
*req
;
952 struct inet6_request_sock
*treq
;
953 struct ipv6_pinfo
*np
= inet6_sk(sk
);
954 struct tcp_sock
*tp
= tcp_sk(sk
);
955 __u32 isn
= TCP_SKB_CB(skb
)->when
;
956 struct dst_entry
*dst
= NULL
;
958 bool want_cookie
= false;
960 if (skb
->protocol
== htons(ETH_P_IP
))
961 return tcp_v4_conn_request(sk
, skb
);
963 if (!ipv6_unicast_destination(skb
))
966 if (inet_csk_reqsk_queue_is_full(sk
) && !isn
) {
967 want_cookie
= tcp_syn_flood_action(sk
, skb
, "TCPv6");
972 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1) {
973 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
977 req
= inet6_reqsk_alloc(&tcp6_request_sock_ops
);
981 #ifdef CONFIG_TCP_MD5SIG
982 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv6_ops
;
985 tcp_clear_options(&tmp_opt
);
986 tmp_opt
.mss_clamp
= IPV6_MIN_MTU
- sizeof(struct tcphdr
) - sizeof(struct ipv6hdr
);
987 tmp_opt
.user_mss
= tp
->rx_opt
.user_mss
;
988 tcp_parse_options(skb
, &tmp_opt
, 0, NULL
);
990 if (want_cookie
&& !tmp_opt
.saw_tstamp
)
991 tcp_clear_options(&tmp_opt
);
993 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
994 tcp_openreq_init(req
, &tmp_opt
, skb
);
996 treq
= inet6_rsk(req
);
997 treq
->rmt_addr
= ipv6_hdr(skb
)->saddr
;
998 treq
->loc_addr
= ipv6_hdr(skb
)->daddr
;
999 if (!want_cookie
|| tmp_opt
.tstamp_ok
)
1000 TCP_ECN_create_request(req
, skb
, sock_net(sk
));
1002 treq
->iif
= sk
->sk_bound_dev_if
;
1004 /* So that link locals have meaning */
1005 if (!sk
->sk_bound_dev_if
&&
1006 ipv6_addr_type(&treq
->rmt_addr
) & IPV6_ADDR_LINKLOCAL
)
1007 treq
->iif
= inet6_iif(skb
);
1010 if (ipv6_opt_accepted(sk
, skb
) ||
1011 np
->rxopt
.bits
.rxinfo
|| np
->rxopt
.bits
.rxoinfo
||
1012 np
->rxopt
.bits
.rxhlim
|| np
->rxopt
.bits
.rxohlim
) {
1013 atomic_inc(&skb
->users
);
1014 treq
->pktopts
= skb
;
1018 isn
= cookie_v6_init_sequence(sk
, skb
, &req
->mss
);
1019 req
->cookie_ts
= tmp_opt
.tstamp_ok
;
1023 /* VJ's idea. We save last timestamp seen
1024 * from the destination in peer table, when entering
1025 * state TIME-WAIT, and check against it before
1026 * accepting new connection request.
1028 * If "isn" is not zero, this request hit alive
1029 * timewait bucket, so that all the necessary checks
1030 * are made in the function processing timewait state.
1032 if (tmp_opt
.saw_tstamp
&&
1033 tcp_death_row
.sysctl_tw_recycle
&&
1034 (dst
= inet6_csk_route_req(sk
, &fl6
, req
)) != NULL
) {
1035 if (!tcp_peer_is_proven(req
, dst
, true)) {
1036 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_PAWSPASSIVEREJECTED
);
1037 goto drop_and_release
;
1040 /* Kill the following clause, if you dislike this way. */
1041 else if (!sysctl_tcp_syncookies
&&
1042 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1043 (sysctl_max_syn_backlog
>> 2)) &&
1044 !tcp_peer_is_proven(req
, dst
, false)) {
1045 /* Without syncookies last quarter of
1046 * backlog is filled with destinations,
1047 * proven to be alive.
1048 * It means that we continue to communicate
1049 * to destinations, already remembered
1050 * to the moment of synflood.
1052 LIMIT_NETDEBUG(KERN_DEBUG
"TCP: drop open request from %pI6/%u\n",
1053 &treq
->rmt_addr
, ntohs(tcp_hdr(skb
)->source
));
1054 goto drop_and_release
;
1057 isn
= tcp_v6_init_sequence(skb
);
1060 tcp_rsk(req
)->snt_isn
= isn
;
1062 if (security_inet_conn_request(sk
, skb
, req
))
1063 goto drop_and_release
;
1065 if (tcp_v6_send_synack(sk
, dst
, &fl6
, req
,
1066 skb_get_queue_mapping(skb
)) ||
1070 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1071 tcp_rsk(req
)->listener
= NULL
;
1072 inet6_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1080 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1081 return 0; /* don't send reset */
1084 static struct sock
* tcp_v6_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1085 struct request_sock
*req
,
1086 struct dst_entry
*dst
)
1088 struct inet6_request_sock
*treq
;
1089 struct ipv6_pinfo
*newnp
, *np
= inet6_sk(sk
);
1090 struct tcp6_sock
*newtcp6sk
;
1091 struct inet_sock
*newinet
;
1092 struct tcp_sock
*newtp
;
1094 #ifdef CONFIG_TCP_MD5SIG
1095 struct tcp_md5sig_key
*key
;
1099 if (skb
->protocol
== htons(ETH_P_IP
)) {
1104 newsk
= tcp_v4_syn_recv_sock(sk
, skb
, req
, dst
);
1109 newtcp6sk
= (struct tcp6_sock
*)newsk
;
1110 inet_sk(newsk
)->pinet6
= &newtcp6sk
->inet6
;
1112 newinet
= inet_sk(newsk
);
1113 newnp
= inet6_sk(newsk
);
1114 newtp
= tcp_sk(newsk
);
1116 memcpy(newnp
, np
, sizeof(struct ipv6_pinfo
));
1118 ipv6_addr_set_v4mapped(newinet
->inet_daddr
, &newnp
->daddr
);
1120 ipv6_addr_set_v4mapped(newinet
->inet_saddr
, &newnp
->saddr
);
1122 newnp
->rcv_saddr
= newnp
->saddr
;
1124 inet_csk(newsk
)->icsk_af_ops
= &ipv6_mapped
;
1125 newsk
->sk_backlog_rcv
= tcp_v4_do_rcv
;
1126 #ifdef CONFIG_TCP_MD5SIG
1127 newtp
->af_specific
= &tcp_sock_ipv6_mapped_specific
;
1130 newnp
->ipv6_ac_list
= NULL
;
1131 newnp
->ipv6_fl_list
= NULL
;
1132 newnp
->pktoptions
= NULL
;
1134 newnp
->mcast_oif
= inet6_iif(skb
);
1135 newnp
->mcast_hops
= ipv6_hdr(skb
)->hop_limit
;
1136 newnp
->rcv_tclass
= ipv6_get_dsfield(ipv6_hdr(skb
));
1139 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1140 * here, tcp_create_openreq_child now does this for us, see the comment in
1141 * that function for the gory details. -acme
1144 /* It is tricky place. Until this moment IPv4 tcp
1145 worked with IPv6 icsk.icsk_af_ops.
1148 tcp_sync_mss(newsk
, inet_csk(newsk
)->icsk_pmtu_cookie
);
1153 treq
= inet6_rsk(req
);
1155 if (sk_acceptq_is_full(sk
))
1159 dst
= inet6_csk_route_req(sk
, &fl6
, req
);
1164 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1169 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1170 * count here, tcp_create_openreq_child now does this for us, see the
1171 * comment in that function for the gory details. -acme
1174 newsk
->sk_gso_type
= SKB_GSO_TCPV6
;
1175 __ip6_dst_store(newsk
, dst
, NULL
, NULL
);
1176 inet6_sk_rx_dst_set(newsk
, skb
);
1178 newtcp6sk
= (struct tcp6_sock
*)newsk
;
1179 inet_sk(newsk
)->pinet6
= &newtcp6sk
->inet6
;
1181 newtp
= tcp_sk(newsk
);
1182 newinet
= inet_sk(newsk
);
1183 newnp
= inet6_sk(newsk
);
1185 memcpy(newnp
, np
, sizeof(struct ipv6_pinfo
));
1187 newnp
->daddr
= treq
->rmt_addr
;
1188 newnp
->saddr
= treq
->loc_addr
;
1189 newnp
->rcv_saddr
= treq
->loc_addr
;
1190 newsk
->sk_bound_dev_if
= treq
->iif
;
1192 /* Now IPv6 options...
1194 First: no IPv4 options.
1196 newinet
->inet_opt
= NULL
;
1197 newnp
->ipv6_ac_list
= NULL
;
1198 newnp
->ipv6_fl_list
= NULL
;
1201 newnp
->rxopt
.all
= np
->rxopt
.all
;
1203 /* Clone pktoptions received with SYN */
1204 newnp
->pktoptions
= NULL
;
1205 if (treq
->pktopts
!= NULL
) {
1206 newnp
->pktoptions
= skb_clone(treq
->pktopts
,
1207 sk_gfp_atomic(sk
, GFP_ATOMIC
));
1208 consume_skb(treq
->pktopts
);
1209 treq
->pktopts
= NULL
;
1210 if (newnp
->pktoptions
)
1211 skb_set_owner_r(newnp
->pktoptions
, newsk
);
1214 newnp
->mcast_oif
= inet6_iif(skb
);
1215 newnp
->mcast_hops
= ipv6_hdr(skb
)->hop_limit
;
1216 newnp
->rcv_tclass
= ipv6_get_dsfield(ipv6_hdr(skb
));
1218 /* Clone native IPv6 options from listening socket (if any)
1220 Yes, keeping reference count would be much more clever,
1221 but we make one more one thing there: reattach optmem
1225 newnp
->opt
= ipv6_dup_options(newsk
, np
->opt
);
1227 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1229 inet_csk(newsk
)->icsk_ext_hdr_len
= (newnp
->opt
->opt_nflen
+
1230 newnp
->opt
->opt_flen
);
1232 tcp_mtup_init(newsk
);
1233 tcp_sync_mss(newsk
, dst_mtu(dst
));
1234 newtp
->advmss
= dst_metric_advmss(dst
);
1235 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1236 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1237 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1239 tcp_initialize_rcv_mss(newsk
);
1241 newinet
->inet_daddr
= newinet
->inet_saddr
= LOOPBACK4_IPV6
;
1242 newinet
->inet_rcv_saddr
= LOOPBACK4_IPV6
;
1244 #ifdef CONFIG_TCP_MD5SIG
1245 /* Copy over the MD5 key from the original socket */
1246 if ((key
= tcp_v6_md5_do_lookup(sk
, &newnp
->daddr
)) != NULL
) {
1247 /* We're using one, so create a matching key
1248 * on the newsk structure. If we fail to get
1249 * memory, then we end up not copying the key
1252 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newnp
->daddr
,
1253 AF_INET6
, key
->key
, key
->keylen
,
1254 sk_gfp_atomic(sk
, GFP_ATOMIC
));
1258 if (__inet_inherit_port(sk
, newsk
) < 0) {
1259 inet_csk_prepare_forced_close(newsk
);
1263 __inet6_hash(newsk
, NULL
);
1268 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1272 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1276 static __sum16
tcp_v6_checksum_init(struct sk_buff
*skb
)
1278 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1279 if (!tcp_v6_check(skb
->len
, &ipv6_hdr(skb
)->saddr
,
1280 &ipv6_hdr(skb
)->daddr
, skb
->csum
)) {
1281 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1286 skb
->csum
= ~csum_unfold(tcp_v6_check(skb
->len
,
1287 &ipv6_hdr(skb
)->saddr
,
1288 &ipv6_hdr(skb
)->daddr
, 0));
1290 if (skb
->len
<= 76) {
1291 return __skb_checksum_complete(skb
);
1296 /* The socket must have it's spinlock held when we get
1299 * We have a potential double-lock case here, so even when
1300 * doing backlog processing we use the BH locking scheme.
1301 * This is because we cannot sleep with the original spinlock
1304 static int tcp_v6_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1306 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1307 struct tcp_sock
*tp
;
1308 struct sk_buff
*opt_skb
= NULL
;
1310 /* Imagine: socket is IPv6. IPv4 packet arrives,
1311 goes to IPv4 receive handler and backlogged.
1312 From backlog it always goes here. Kerboom...
1313 Fortunately, tcp_rcv_established and rcv_established
1314 handle them correctly, but it is not case with
1315 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1318 if (skb
->protocol
== htons(ETH_P_IP
))
1319 return tcp_v4_do_rcv(sk
, skb
);
1321 #ifdef CONFIG_TCP_MD5SIG
1322 if (tcp_v6_inbound_md5_hash (sk
, skb
))
1326 if (sk_filter(sk
, skb
))
1330 * socket locking is here for SMP purposes as backlog rcv
1331 * is currently called with bh processing disabled.
1334 /* Do Stevens' IPV6_PKTOPTIONS.
1336 Yes, guys, it is the only place in our code, where we
1337 may make it not affecting IPv4.
1338 The rest of code is protocol independent,
1339 and I do not like idea to uglify IPv4.
1341 Actually, all the idea behind IPV6_PKTOPTIONS
1342 looks not very well thought. For now we latch
1343 options, received in the last packet, enqueued
1344 by tcp. Feel free to propose better solution.
1348 opt_skb
= skb_clone(skb
, sk_gfp_atomic(sk
, GFP_ATOMIC
));
1350 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1351 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1353 sock_rps_save_rxhash(sk
, skb
);
1355 if (inet_sk(sk
)->rx_dst_ifindex
!= skb
->skb_iif
||
1356 dst
->ops
->check(dst
, np
->rx_dst_cookie
) == NULL
) {
1358 sk
->sk_rx_dst
= NULL
;
1362 if (tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
))
1365 goto ipv6_pktoptions
;
1369 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1372 if (sk
->sk_state
== TCP_LISTEN
) {
1373 struct sock
*nsk
= tcp_v6_hnd_req(sk
, skb
);
1378 * Queue it on the new socket if the new socket is active,
1379 * otherwise we just shortcircuit this and continue with
1383 sock_rps_save_rxhash(nsk
, skb
);
1384 if (tcp_child_process(sk
, nsk
, skb
))
1387 __kfree_skb(opt_skb
);
1391 sock_rps_save_rxhash(sk
, skb
);
1393 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
))
1396 goto ipv6_pktoptions
;
1400 tcp_v6_send_reset(sk
, skb
);
1403 __kfree_skb(opt_skb
);
1407 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_CSUMERRORS
);
1408 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1413 /* Do you ask, what is it?
1415 1. skb was enqueued by tcp.
1416 2. skb is added to tail of read queue, rather than out of order.
1417 3. socket is not in passive state.
1418 4. Finally, it really contains options, which user wants to receive.
1421 if (TCP_SKB_CB(opt_skb
)->end_seq
== tp
->rcv_nxt
&&
1422 !((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
))) {
1423 if (np
->rxopt
.bits
.rxinfo
|| np
->rxopt
.bits
.rxoinfo
)
1424 np
->mcast_oif
= inet6_iif(opt_skb
);
1425 if (np
->rxopt
.bits
.rxhlim
|| np
->rxopt
.bits
.rxohlim
)
1426 np
->mcast_hops
= ipv6_hdr(opt_skb
)->hop_limit
;
1427 if (np
->rxopt
.bits
.rxtclass
)
1428 np
->rcv_tclass
= ipv6_get_dsfield(ipv6_hdr(skb
));
1429 if (ipv6_opt_accepted(sk
, opt_skb
)) {
1430 skb_set_owner_r(opt_skb
, sk
);
1431 opt_skb
= xchg(&np
->pktoptions
, opt_skb
);
1433 __kfree_skb(opt_skb
);
1434 opt_skb
= xchg(&np
->pktoptions
, NULL
);
1442 static int tcp_v6_rcv(struct sk_buff
*skb
)
1444 const struct tcphdr
*th
;
1445 const struct ipv6hdr
*hdr
;
1448 struct net
*net
= dev_net(skb
->dev
);
1450 if (skb
->pkt_type
!= PACKET_HOST
)
1454 * Count it even if it's bad.
1456 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1458 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1463 if (th
->doff
< sizeof(struct tcphdr
)/4)
1465 if (!pskb_may_pull(skb
, th
->doff
*4))
1468 if (!skb_csum_unnecessary(skb
) && tcp_v6_checksum_init(skb
))
1472 hdr
= ipv6_hdr(skb
);
1473 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1474 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1475 skb
->len
- th
->doff
*4);
1476 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1477 TCP_SKB_CB(skb
)->when
= 0;
1478 TCP_SKB_CB(skb
)->ip_dsfield
= ipv6_get_dsfield(hdr
);
1479 TCP_SKB_CB(skb
)->sacked
= 0;
1481 sk
= __inet6_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1486 if (sk
->sk_state
== TCP_TIME_WAIT
)
1489 if (hdr
->hop_limit
< inet6_sk(sk
)->min_hopcount
) {
1490 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
1491 goto discard_and_relse
;
1494 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1495 goto discard_and_relse
;
1497 if (sk_filter(sk
, skb
))
1498 goto discard_and_relse
;
1500 sk_mark_napi_id(sk
, skb
);
1503 bh_lock_sock_nested(sk
);
1505 if (!sock_owned_by_user(sk
)) {
1506 #ifdef CONFIG_NET_DMA
1507 struct tcp_sock
*tp
= tcp_sk(sk
);
1508 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1509 tp
->ucopy
.dma_chan
= net_dma_find_channel();
1510 if (tp
->ucopy
.dma_chan
)
1511 ret
= tcp_v6_do_rcv(sk
, skb
);
1515 if (!tcp_prequeue(sk
, skb
))
1516 ret
= tcp_v6_do_rcv(sk
, skb
);
1518 } else if (unlikely(sk_add_backlog(sk
, skb
,
1519 sk
->sk_rcvbuf
+ sk
->sk_sndbuf
))) {
1521 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
1522 goto discard_and_relse
;
1527 return ret
? -1 : 0;
1530 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1533 if (skb
->len
< (th
->doff
<<2) || tcp_checksum_complete(skb
)) {
1535 TCP_INC_STATS_BH(net
, TCP_MIB_CSUMERRORS
);
1537 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1539 tcp_v6_send_reset(NULL
, skb
);
1551 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1552 inet_twsk_put(inet_twsk(sk
));
1556 if (skb
->len
< (th
->doff
<<2)) {
1557 inet_twsk_put(inet_twsk(sk
));
1560 if (tcp_checksum_complete(skb
)) {
1561 inet_twsk_put(inet_twsk(sk
));
1565 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1570 sk2
= inet6_lookup_listener(dev_net(skb
->dev
), &tcp_hashinfo
,
1571 &ipv6_hdr(skb
)->saddr
, th
->source
,
1572 &ipv6_hdr(skb
)->daddr
,
1573 ntohs(th
->dest
), inet6_iif(skb
));
1575 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
1576 inet_twsk_deschedule(tw
, &tcp_death_row
);
1581 /* Fall through to ACK */
1584 tcp_v6_timewait_ack(sk
, skb
);
1588 case TCP_TW_SUCCESS
:;
1593 static void tcp_v6_early_demux(struct sk_buff
*skb
)
1595 const struct ipv6hdr
*hdr
;
1596 const struct tcphdr
*th
;
1599 if (skb
->pkt_type
!= PACKET_HOST
)
1602 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct tcphdr
)))
1605 hdr
= ipv6_hdr(skb
);
1608 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1611 sk
= __inet6_lookup_established(dev_net(skb
->dev
), &tcp_hashinfo
,
1612 &hdr
->saddr
, th
->source
,
1613 &hdr
->daddr
, ntohs(th
->dest
),
1617 skb
->destructor
= sock_edemux
;
1618 if (sk
->sk_state
!= TCP_TIME_WAIT
) {
1619 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1622 dst
= dst_check(dst
, inet6_sk(sk
)->rx_dst_cookie
);
1624 inet_sk(sk
)->rx_dst_ifindex
== skb
->skb_iif
)
1625 skb_dst_set_noref(skb
, dst
);
1630 static struct timewait_sock_ops tcp6_timewait_sock_ops
= {
1631 .twsk_obj_size
= sizeof(struct tcp6_timewait_sock
),
1632 .twsk_unique
= tcp_twsk_unique
,
1633 .twsk_destructor
= tcp_twsk_destructor
,
1636 static const struct inet_connection_sock_af_ops ipv6_specific
= {
1637 .queue_xmit
= inet6_csk_xmit
,
1638 .send_check
= tcp_v6_send_check
,
1639 .rebuild_header
= inet6_sk_rebuild_header
,
1640 .sk_rx_dst_set
= inet6_sk_rx_dst_set
,
1641 .conn_request
= tcp_v6_conn_request
,
1642 .syn_recv_sock
= tcp_v6_syn_recv_sock
,
1643 .net_header_len
= sizeof(struct ipv6hdr
),
1644 .net_frag_header_len
= sizeof(struct frag_hdr
),
1645 .setsockopt
= ipv6_setsockopt
,
1646 .getsockopt
= ipv6_getsockopt
,
1647 .addr2sockaddr
= inet6_csk_addr2sockaddr
,
1648 .sockaddr_len
= sizeof(struct sockaddr_in6
),
1649 .bind_conflict
= inet6_csk_bind_conflict
,
1650 #ifdef CONFIG_COMPAT
1651 .compat_setsockopt
= compat_ipv6_setsockopt
,
1652 .compat_getsockopt
= compat_ipv6_getsockopt
,
1656 #ifdef CONFIG_TCP_MD5SIG
1657 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific
= {
1658 .md5_lookup
= tcp_v6_md5_lookup
,
1659 .calc_md5_hash
= tcp_v6_md5_hash_skb
,
1660 .md5_parse
= tcp_v6_parse_md5_keys
,
1665 * TCP over IPv4 via INET6 API
1668 static const struct inet_connection_sock_af_ops ipv6_mapped
= {
1669 .queue_xmit
= ip_queue_xmit
,
1670 .send_check
= tcp_v4_send_check
,
1671 .rebuild_header
= inet_sk_rebuild_header
,
1672 .sk_rx_dst_set
= inet_sk_rx_dst_set
,
1673 .conn_request
= tcp_v6_conn_request
,
1674 .syn_recv_sock
= tcp_v6_syn_recv_sock
,
1675 .net_header_len
= sizeof(struct iphdr
),
1676 .setsockopt
= ipv6_setsockopt
,
1677 .getsockopt
= ipv6_getsockopt
,
1678 .addr2sockaddr
= inet6_csk_addr2sockaddr
,
1679 .sockaddr_len
= sizeof(struct sockaddr_in6
),
1680 .bind_conflict
= inet6_csk_bind_conflict
,
1681 #ifdef CONFIG_COMPAT
1682 .compat_setsockopt
= compat_ipv6_setsockopt
,
1683 .compat_getsockopt
= compat_ipv6_getsockopt
,
1687 #ifdef CONFIG_TCP_MD5SIG
1688 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific
= {
1689 .md5_lookup
= tcp_v4_md5_lookup
,
1690 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1691 .md5_parse
= tcp_v6_parse_md5_keys
,
1695 /* NOTE: A lot of things set to zero explicitly by call to
1696 * sk_alloc() so need not be done here.
1698 static int tcp_v6_init_sock(struct sock
*sk
)
1700 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1704 icsk
->icsk_af_ops
= &ipv6_specific
;
1706 #ifdef CONFIG_TCP_MD5SIG
1707 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv6_specific
;
1713 static void tcp_v6_destroy_sock(struct sock
*sk
)
1715 tcp_v4_destroy_sock(sk
);
1716 inet6_destroy_sock(sk
);
1719 #ifdef CONFIG_PROC_FS
1720 /* Proc filesystem TCPv6 sock list dumping. */
1721 static void get_openreq6(struct seq_file
*seq
,
1722 const struct sock
*sk
, struct request_sock
*req
, int i
, kuid_t uid
)
1724 int ttd
= req
->expires
- jiffies
;
1725 const struct in6_addr
*src
= &inet6_rsk(req
)->loc_addr
;
1726 const struct in6_addr
*dest
= &inet6_rsk(req
)->rmt_addr
;
1732 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1733 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1735 src
->s6_addr32
[0], src
->s6_addr32
[1],
1736 src
->s6_addr32
[2], src
->s6_addr32
[3],
1737 ntohs(inet_rsk(req
)->loc_port
),
1738 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1739 dest
->s6_addr32
[2], dest
->s6_addr32
[3],
1740 ntohs(inet_rsk(req
)->rmt_port
),
1742 0,0, /* could print option size, but that is af dependent. */
1743 1, /* timers active (only the expire timer) */
1744 jiffies_to_clock_t(ttd
),
1746 from_kuid_munged(seq_user_ns(seq
), uid
),
1747 0, /* non standard timer */
1748 0, /* open_requests have no inode */
1752 static void get_tcp6_sock(struct seq_file
*seq
, struct sock
*sp
, int i
)
1754 const struct in6_addr
*dest
, *src
;
1757 unsigned long timer_expires
;
1758 const struct inet_sock
*inet
= inet_sk(sp
);
1759 const struct tcp_sock
*tp
= tcp_sk(sp
);
1760 const struct inet_connection_sock
*icsk
= inet_csk(sp
);
1761 const struct ipv6_pinfo
*np
= inet6_sk(sp
);
1764 src
= &np
->rcv_saddr
;
1765 destp
= ntohs(inet
->inet_dport
);
1766 srcp
= ntohs(inet
->inet_sport
);
1768 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
) {
1770 timer_expires
= icsk
->icsk_timeout
;
1771 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
1773 timer_expires
= icsk
->icsk_timeout
;
1774 } else if (timer_pending(&sp
->sk_timer
)) {
1776 timer_expires
= sp
->sk_timer
.expires
;
1779 timer_expires
= jiffies
;
1783 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1784 "%02X %08X:%08X %02X:%08lX %08X %5d %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
,
1792 (sp
->sk_state
== TCP_LISTEN
) ? sp
->sk_ack_backlog
: (tp
->rcv_nxt
- tp
->copied_seq
),
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 tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
1808 static void get_timewait6_sock(struct seq_file
*seq
,
1809 struct inet_timewait_sock
*tw
, int i
)
1811 const struct in6_addr
*dest
, *src
;
1813 const struct inet6_timewait_sock
*tw6
= inet6_twsk((struct sock
*)tw
);
1814 long delta
= tw
->tw_ttd
- jiffies
;
1816 dest
= &tw6
->tw_v6_daddr
;
1817 src
= &tw6
->tw_v6_rcv_saddr
;
1818 destp
= ntohs(tw
->tw_dport
);
1819 srcp
= ntohs(tw
->tw_sport
);
1822 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1823 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1825 src
->s6_addr32
[0], src
->s6_addr32
[1],
1826 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1827 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1828 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1829 tw
->tw_substate
, 0, 0,
1830 3, jiffies_delta_to_clock_t(delta
), 0, 0, 0, 0,
1831 atomic_read(&tw
->tw_refcnt
), tw
);
1834 static int tcp6_seq_show(struct seq_file
*seq
, void *v
)
1836 struct tcp_iter_state
*st
;
1838 if (v
== SEQ_START_TOKEN
) {
1843 "st tx_queue rx_queue tr tm->when retrnsmt"
1844 " uid timeout inode\n");
1849 switch (st
->state
) {
1850 case TCP_SEQ_STATE_LISTENING
:
1851 case TCP_SEQ_STATE_ESTABLISHED
:
1852 get_tcp6_sock(seq
, v
, st
->num
);
1854 case TCP_SEQ_STATE_OPENREQ
:
1855 get_openreq6(seq
, st
->syn_wait_sk
, v
, st
->num
, st
->uid
);
1857 case TCP_SEQ_STATE_TIME_WAIT
:
1858 get_timewait6_sock(seq
, v
, st
->num
);
1865 static const struct file_operations tcp6_afinfo_seq_fops
= {
1866 .owner
= THIS_MODULE
,
1867 .open
= tcp_seq_open
,
1869 .llseek
= seq_lseek
,
1870 .release
= seq_release_net
1873 static struct tcp_seq_afinfo tcp6_seq_afinfo
= {
1876 .seq_fops
= &tcp6_afinfo_seq_fops
,
1878 .show
= tcp6_seq_show
,
1882 int __net_init
tcp6_proc_init(struct net
*net
)
1884 return tcp_proc_register(net
, &tcp6_seq_afinfo
);
1887 void tcp6_proc_exit(struct net
*net
)
1889 tcp_proc_unregister(net
, &tcp6_seq_afinfo
);
1893 static void tcp_v6_clear_sk(struct sock
*sk
, int size
)
1895 struct inet_sock
*inet
= inet_sk(sk
);
1897 /* we do not want to clear pinet6 field, because of RCU lookups */
1898 sk_prot_clear_nulls(sk
, offsetof(struct inet_sock
, pinet6
));
1900 size
-= offsetof(struct inet_sock
, pinet6
) + sizeof(inet
->pinet6
);
1901 memset(&inet
->pinet6
+ 1, 0, size
);
1904 struct proto tcpv6_prot
= {
1906 .owner
= THIS_MODULE
,
1908 .connect
= tcp_v6_connect
,
1909 .disconnect
= tcp_disconnect
,
1910 .accept
= inet_csk_accept
,
1912 .init
= tcp_v6_init_sock
,
1913 .destroy
= tcp_v6_destroy_sock
,
1914 .shutdown
= tcp_shutdown
,
1915 .setsockopt
= tcp_setsockopt
,
1916 .getsockopt
= tcp_getsockopt
,
1917 .recvmsg
= tcp_recvmsg
,
1918 .sendmsg
= tcp_sendmsg
,
1919 .sendpage
= tcp_sendpage
,
1920 .backlog_rcv
= tcp_v6_do_rcv
,
1921 .release_cb
= tcp_release_cb
,
1922 .mtu_reduced
= tcp_v6_mtu_reduced
,
1923 .hash
= tcp_v6_hash
,
1924 .unhash
= inet_unhash
,
1925 .get_port
= inet_csk_get_port
,
1926 .enter_memory_pressure
= tcp_enter_memory_pressure
,
1927 .stream_memory_free
= tcp_stream_memory_free
,
1928 .sockets_allocated
= &tcp_sockets_allocated
,
1929 .memory_allocated
= &tcp_memory_allocated
,
1930 .memory_pressure
= &tcp_memory_pressure
,
1931 .orphan_count
= &tcp_orphan_count
,
1932 .sysctl_wmem
= sysctl_tcp_wmem
,
1933 .sysctl_rmem
= sysctl_tcp_rmem
,
1934 .max_header
= MAX_TCP_HEADER
,
1935 .obj_size
= sizeof(struct tcp6_sock
),
1936 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1937 .twsk_prot
= &tcp6_timewait_sock_ops
,
1938 .rsk_prot
= &tcp6_request_sock_ops
,
1939 .h
.hashinfo
= &tcp_hashinfo
,
1940 .no_autobind
= true,
1941 #ifdef CONFIG_COMPAT
1942 .compat_setsockopt
= compat_tcp_setsockopt
,
1943 .compat_getsockopt
= compat_tcp_getsockopt
,
1945 #ifdef CONFIG_MEMCG_KMEM
1946 .proto_cgroup
= tcp_proto_cgroup
,
1948 .clear_sk
= tcp_v6_clear_sk
,
1951 static const struct inet6_protocol tcpv6_protocol
= {
1952 .early_demux
= tcp_v6_early_demux
,
1953 .handler
= tcp_v6_rcv
,
1954 .err_handler
= tcp_v6_err
,
1955 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1958 static struct inet_protosw tcpv6_protosw
= {
1959 .type
= SOCK_STREAM
,
1960 .protocol
= IPPROTO_TCP
,
1961 .prot
= &tcpv6_prot
,
1962 .ops
= &inet6_stream_ops
,
1964 .flags
= INET_PROTOSW_PERMANENT
|
1968 static int __net_init
tcpv6_net_init(struct net
*net
)
1970 return inet_ctl_sock_create(&net
->ipv6
.tcp_sk
, PF_INET6
,
1971 SOCK_RAW
, IPPROTO_TCP
, net
);
1974 static void __net_exit
tcpv6_net_exit(struct net
*net
)
1976 inet_ctl_sock_destroy(net
->ipv6
.tcp_sk
);
1979 static void __net_exit
tcpv6_net_exit_batch(struct list_head
*net_exit_list
)
1981 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET6
);
1984 static struct pernet_operations tcpv6_net_ops
= {
1985 .init
= tcpv6_net_init
,
1986 .exit
= tcpv6_net_exit
,
1987 .exit_batch
= tcpv6_net_exit_batch
,
1990 int __init
tcpv6_init(void)
1994 ret
= inet6_add_protocol(&tcpv6_protocol
, IPPROTO_TCP
);
1998 /* register inet6 protocol */
1999 ret
= inet6_register_protosw(&tcpv6_protosw
);
2001 goto out_tcpv6_protocol
;
2003 ret
= register_pernet_subsys(&tcpv6_net_ops
);
2005 goto out_tcpv6_protosw
;
2010 inet6_unregister_protosw(&tcpv6_protosw
);
2012 inet6_del_protocol(&tcpv6_protocol
, IPPROTO_TCP
);
2016 void tcpv6_exit(void)
2018 unregister_pernet_subsys(&tcpv6_net_ops
);
2019 inet6_unregister_protosw(&tcpv6_protosw
);
2020 inet6_del_protocol(&tcpv6_protocol
, IPPROTO_TCP
);