3 * Linux INET6 implementation
5 * Based on net/dccp6/ipv6.c
7 * Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License
11 * as published by the Free Software Foundation; either version
12 * 2 of the License, or (at your option) any later version.
15 #include <linux/config.h>
16 #include <linux/module.h>
17 #include <linux/random.h>
18 #include <linux/xfrm.h>
20 #include <net/addrconf.h>
21 #include <net/inet_common.h>
22 #include <net/inet_hashtables.h>
23 #include <net/inet6_connection_sock.h>
24 #include <net/inet6_hashtables.h>
25 #include <net/ip6_route.h>
27 #include <net/protocol.h>
28 #include <net/transp_v6.h>
34 static void dccp_v6_ctl_send_reset(struct sk_buff
*skb
);
35 static void dccp_v6_reqsk_send_ack(struct sk_buff
*skb
,
36 struct request_sock
*req
);
37 static void dccp_v6_send_check(struct sock
*sk
, int len
, struct sk_buff
*skb
);
39 static int dccp_v6_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
41 static struct inet_connection_sock_af_ops dccp_ipv6_mapped
;
42 static struct inet_connection_sock_af_ops dccp_ipv6_af_ops
;
44 static int dccp_v6_get_port(struct sock
*sk
, unsigned short snum
)
46 return inet_csk_get_port(&dccp_hashinfo
, sk
, snum
,
47 inet6_csk_bind_conflict
);
50 static void dccp_v6_hash(struct sock
*sk
)
52 if (sk
->sk_state
!= DCCP_CLOSED
) {
53 if (inet_csk(sk
)->icsk_af_ops
== &dccp_ipv6_mapped
) {
58 __inet6_hash(&dccp_hashinfo
, sk
);
63 static inline u16
dccp_v6_check(struct dccp_hdr
*dh
, int len
,
64 struct in6_addr
*saddr
,
65 struct in6_addr
*daddr
,
68 return csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_DCCP
, base
);
71 static __u32
dccp_v6_init_sequence(struct sock
*sk
, struct sk_buff
*skb
)
73 const struct dccp_hdr
*dh
= dccp_hdr(skb
);
75 if (skb
->protocol
== htons(ETH_P_IPV6
))
76 return secure_tcpv6_sequence_number(skb
->nh
.ipv6h
->daddr
.s6_addr32
,
77 skb
->nh
.ipv6h
->saddr
.s6_addr32
,
81 return secure_dccp_sequence_number(skb
->nh
.iph
->daddr
,
87 static int __dccp_v6_check_established(struct sock
*sk
, const __u16 lport
,
88 struct inet_timewait_sock
**twp
)
90 struct inet_sock
*inet
= inet_sk(sk
);
91 const struct ipv6_pinfo
*np
= inet6_sk(sk
);
92 const struct in6_addr
*daddr
= &np
->rcv_saddr
;
93 const struct in6_addr
*saddr
= &np
->daddr
;
94 const int dif
= sk
->sk_bound_dev_if
;
95 const u32 ports
= INET_COMBINED_PORTS(inet
->dport
, lport
);
96 const unsigned int hash
= inet6_ehashfn(daddr
, inet
->num
,
98 struct inet_ehash_bucket
*head
= inet_ehash_bucket(&dccp_hashinfo
, hash
);
100 const struct hlist_node
*node
;
101 struct inet_timewait_sock
*tw
;
103 prefetch(head
->chain
.first
);
104 write_lock(&head
->lock
);
106 /* Check TIME-WAIT sockets first. */
107 sk_for_each(sk2
, node
, &(head
+ dccp_hashinfo
.ehash_size
)->chain
) {
108 const struct inet6_timewait_sock
*tw6
= inet6_twsk(sk2
);
112 if(*((__u32
*)&(tw
->tw_dport
)) == ports
&&
113 sk2
->sk_family
== PF_INET6
&&
114 ipv6_addr_equal(&tw6
->tw_v6_daddr
, saddr
) &&
115 ipv6_addr_equal(&tw6
->tw_v6_rcv_saddr
, daddr
) &&
116 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
)
121 /* And established part... */
122 sk_for_each(sk2
, node
, &head
->chain
) {
123 if (INET6_MATCH(sk2
, hash
, saddr
, daddr
, ports
, dif
))
127 BUG_TRAP(sk_unhashed(sk
));
128 __sk_add_node(sk
, &head
->chain
);
130 sock_prot_inc_use(sk
->sk_prot
);
131 write_unlock(&head
->lock
);
135 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED
);
137 /* Silly. Should hash-dance instead... */
138 inet_twsk_deschedule(tw
, &dccp_death_row
);
139 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED
);
146 write_unlock(&head
->lock
);
147 return -EADDRNOTAVAIL
;
150 static inline u32
dccp_v6_port_offset(const struct sock
*sk
)
152 const struct inet_sock
*inet
= inet_sk(sk
);
153 const struct ipv6_pinfo
*np
= inet6_sk(sk
);
155 return secure_tcpv6_port_ephemeral(np
->rcv_saddr
.s6_addr32
,
160 static int dccp_v6_hash_connect(struct sock
*sk
)
162 const unsigned short snum
= inet_sk(sk
)->num
;
163 struct inet_bind_hashbucket
*head
;
164 struct inet_bind_bucket
*tb
;
168 int low
= sysctl_local_port_range
[0];
169 int high
= sysctl_local_port_range
[1];
170 int range
= high
- low
;
174 u32 offset
= hint
+ dccp_v6_port_offset(sk
);
175 struct hlist_node
*node
;
176 struct inet_timewait_sock
*tw
= NULL
;
179 for (i
= 1; i
<= range
; i
++) {
180 port
= low
+ (i
+ offset
) % range
;
181 head
= &dccp_hashinfo
.bhash
[inet_bhashfn(port
,
182 dccp_hashinfo
.bhash_size
)];
183 spin_lock(&head
->lock
);
185 /* Does not bother with rcv_saddr checks,
186 * because the established check is already
189 inet_bind_bucket_for_each(tb
, node
, &head
->chain
) {
190 if (tb
->port
== port
) {
191 BUG_TRAP(!hlist_empty(&tb
->owners
));
192 if (tb
->fastreuse
>= 0)
194 if (!__dccp_v6_check_established(sk
,
202 tb
= inet_bind_bucket_create(dccp_hashinfo
.bind_bucket_cachep
,
205 spin_unlock(&head
->lock
);
212 spin_unlock(&head
->lock
);
216 return -EADDRNOTAVAIL
;
220 /* Head lock still held and bh's disabled */
221 inet_bind_hash(sk
, tb
, port
);
222 if (sk_unhashed(sk
)) {
223 inet_sk(sk
)->sport
= htons(port
);
224 __inet6_hash(&dccp_hashinfo
, sk
);
226 spin_unlock(&head
->lock
);
229 inet_twsk_deschedule(tw
, &dccp_death_row
);
237 head
= &dccp_hashinfo
.bhash
[inet_bhashfn(snum
,
238 dccp_hashinfo
.bhash_size
)];
239 tb
= inet_csk(sk
)->icsk_bind_hash
;
240 spin_lock_bh(&head
->lock
);
242 if (sk_head(&tb
->owners
) == sk
&& !sk
->sk_bind_node
.next
) {
243 __inet6_hash(&dccp_hashinfo
, sk
);
244 spin_unlock_bh(&head
->lock
);
247 spin_unlock(&head
->lock
);
248 /* No definite answer... Walk to established hash table */
249 ret
= __dccp_v6_check_established(sk
, snum
, NULL
);
256 static int dccp_v6_connect(struct sock
*sk
, struct sockaddr
*uaddr
,
259 struct sockaddr_in6
*usin
= (struct sockaddr_in6
*) uaddr
;
260 struct inet_sock
*inet
= inet_sk(sk
);
261 struct ipv6_pinfo
*np
= inet6_sk(sk
);
262 struct dccp_sock
*dp
= dccp_sk(sk
);
263 struct in6_addr
*saddr
= NULL
, *final_p
= NULL
, final
;
265 struct dst_entry
*dst
;
269 dp
->dccps_role
= DCCP_ROLE_CLIENT
;
271 if (addr_len
< SIN6_LEN_RFC2133
)
274 if (usin
->sin6_family
!= AF_INET6
)
275 return -EAFNOSUPPORT
;
277 memset(&fl
, 0, sizeof(fl
));
280 fl
.fl6_flowlabel
= usin
->sin6_flowinfo
& IPV6_FLOWINFO_MASK
;
281 IP6_ECN_flow_init(fl
.fl6_flowlabel
);
282 if (fl
.fl6_flowlabel
& IPV6_FLOWLABEL_MASK
) {
283 struct ip6_flowlabel
*flowlabel
;
284 flowlabel
= fl6_sock_lookup(sk
, fl
.fl6_flowlabel
);
285 if (flowlabel
== NULL
)
287 ipv6_addr_copy(&usin
->sin6_addr
, &flowlabel
->dst
);
288 fl6_sock_release(flowlabel
);
293 * connect() to INADDR_ANY means loopback (BSD'ism).
296 if (ipv6_addr_any(&usin
->sin6_addr
))
297 usin
->sin6_addr
.s6_addr
[15] = 0x1;
299 addr_type
= ipv6_addr_type(&usin
->sin6_addr
);
301 if(addr_type
& IPV6_ADDR_MULTICAST
)
304 if (addr_type
& IPV6_ADDR_LINKLOCAL
) {
305 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
306 usin
->sin6_scope_id
) {
307 /* If interface is set while binding, indices
310 if (sk
->sk_bound_dev_if
&&
311 sk
->sk_bound_dev_if
!= usin
->sin6_scope_id
)
314 sk
->sk_bound_dev_if
= usin
->sin6_scope_id
;
317 /* Connect to link-local address requires an interface */
318 if (!sk
->sk_bound_dev_if
)
322 ipv6_addr_copy(&np
->daddr
, &usin
->sin6_addr
);
323 np
->flow_label
= fl
.fl6_flowlabel
;
329 if (addr_type
== IPV6_ADDR_MAPPED
) {
330 u32 exthdrlen
= dp
->dccps_ext_header_len
;
331 struct sockaddr_in sin
;
333 SOCK_DEBUG(sk
, "connect: ipv4 mapped\n");
335 if (__ipv6_only_sock(sk
))
338 sin
.sin_family
= AF_INET
;
339 sin
.sin_port
= usin
->sin6_port
;
340 sin
.sin_addr
.s_addr
= usin
->sin6_addr
.s6_addr32
[3];
342 inet_csk(sk
)->icsk_af_ops
= &dccp_ipv6_mapped
;
343 sk
->sk_backlog_rcv
= dccp_v4_do_rcv
;
345 err
= dccp_v4_connect(sk
, (struct sockaddr
*)&sin
, sizeof(sin
));
348 dp
->dccps_ext_header_len
= exthdrlen
;
349 inet_csk(sk
)->icsk_af_ops
= &dccp_ipv6_af_ops
;
350 sk
->sk_backlog_rcv
= dccp_v6_do_rcv
;
353 ipv6_addr_set(&np
->saddr
, 0, 0, htonl(0x0000FFFF),
355 ipv6_addr_set(&np
->rcv_saddr
, 0, 0, htonl(0x0000FFFF),
362 if (!ipv6_addr_any(&np
->rcv_saddr
))
363 saddr
= &np
->rcv_saddr
;
365 fl
.proto
= IPPROTO_DCCP
;
366 ipv6_addr_copy(&fl
.fl6_dst
, &np
->daddr
);
367 ipv6_addr_copy(&fl
.fl6_src
, saddr
? saddr
: &np
->saddr
);
368 fl
.oif
= sk
->sk_bound_dev_if
;
369 fl
.fl_ip_dport
= usin
->sin6_port
;
370 fl
.fl_ip_sport
= inet
->sport
;
372 if (np
->opt
&& np
->opt
->srcrt
) {
373 struct rt0_hdr
*rt0
= (struct rt0_hdr
*)np
->opt
->srcrt
;
374 ipv6_addr_copy(&final
, &fl
.fl6_dst
);
375 ipv6_addr_copy(&fl
.fl6_dst
, rt0
->addr
);
379 err
= ip6_dst_lookup(sk
, &dst
, &fl
);
383 ipv6_addr_copy(&fl
.fl6_dst
, final_p
);
385 if ((err
= xfrm_lookup(&dst
, &fl
, sk
, 0)) < 0)
390 ipv6_addr_copy(&np
->rcv_saddr
, saddr
);
393 /* set the source address */
394 ipv6_addr_copy(&np
->saddr
, saddr
);
395 inet
->rcv_saddr
= LOOPBACK4_IPV6
;
397 ip6_dst_store(sk
, dst
, NULL
);
399 dp
->dccps_ext_header_len
= 0;
401 dp
->dccps_ext_header_len
= np
->opt
->opt_flen
+ np
->opt
->opt_nflen
;
403 inet
->dport
= usin
->sin6_port
;
405 dccp_set_state(sk
, DCCP_REQUESTING
);
406 err
= dccp_v6_hash_connect(sk
);
411 dp
->dccps_gar
= secure_dccp_v6_sequence_number(np
->saddr
.s6_addr32
,
416 err
= dccp_connect(sk
);
423 dccp_set_state(sk
, DCCP_CLOSED
);
427 sk
->sk_route_caps
= 0;
431 static void dccp_v6_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
432 int type
, int code
, int offset
, __u32 info
)
434 struct ipv6hdr
*hdr
= (struct ipv6hdr
*)skb
->data
;
435 const struct dccp_hdr
*dh
= (struct dccp_hdr
*)(skb
->data
+ offset
);
436 struct ipv6_pinfo
*np
;
441 sk
= inet6_lookup(&dccp_hashinfo
, &hdr
->daddr
, dh
->dccph_dport
,
442 &hdr
->saddr
, dh
->dccph_sport
, skb
->dev
->ifindex
);
445 ICMP6_INC_STATS_BH(__in6_dev_get(skb
->dev
), ICMP6_MIB_INERRORS
);
449 if (sk
->sk_state
== DCCP_TIME_WAIT
) {
450 inet_twsk_put((struct inet_timewait_sock
*)sk
);
455 if (sock_owned_by_user(sk
))
456 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS
);
458 if (sk
->sk_state
== DCCP_CLOSED
)
463 if (type
== ICMPV6_PKT_TOOBIG
) {
464 struct dccp_sock
*dp
= dccp_sk(sk
);
465 struct dst_entry
*dst
= NULL
;
467 if (sock_owned_by_user(sk
))
469 if ((1 << sk
->sk_state
) & (DCCPF_LISTEN
| DCCPF_CLOSED
))
472 /* icmp should have updated the destination cache entry */
473 dst
= __sk_dst_check(sk
, np
->dst_cookie
);
476 struct inet_sock
*inet
= inet_sk(sk
);
479 /* BUGGG_FUTURE: Again, it is not clear how
480 to handle rthdr case. Ignore this complexity
483 memset(&fl
, 0, sizeof(fl
));
484 fl
.proto
= IPPROTO_DCCP
;
485 ipv6_addr_copy(&fl
.fl6_dst
, &np
->daddr
);
486 ipv6_addr_copy(&fl
.fl6_src
, &np
->saddr
);
487 fl
.oif
= sk
->sk_bound_dev_if
;
488 fl
.fl_ip_dport
= inet
->dport
;
489 fl
.fl_ip_sport
= inet
->sport
;
491 if ((err
= ip6_dst_lookup(sk
, &dst
, &fl
))) {
492 sk
->sk_err_soft
= -err
;
496 if ((err
= xfrm_lookup(&dst
, &fl
, sk
, 0)) < 0) {
497 sk
->sk_err_soft
= -err
;
504 if (dp
->dccps_pmtu_cookie
> dst_mtu(dst
)) {
505 dccp_sync_mss(sk
, dst_mtu(dst
));
506 } /* else let the usual retransmit timer handle it */
511 icmpv6_err_convert(type
, code
, &err
);
513 seq
= DCCP_SKB_CB(skb
)->dccpd_seq
;
514 /* Might be for an request_sock */
515 switch (sk
->sk_state
) {
516 struct request_sock
*req
, **prev
;
518 if (sock_owned_by_user(sk
))
521 req
= inet6_csk_search_req(sk
, &prev
, dh
->dccph_dport
,
522 &hdr
->daddr
, &hdr
->saddr
,
527 /* ICMPs are not backlogged, hence we cannot get
528 * an established socket here.
530 BUG_TRAP(req
->sk
== NULL
);
532 if (seq
!= dccp_rsk(req
)->dreq_iss
) {
533 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS
);
537 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
540 case DCCP_REQUESTING
:
541 case DCCP_RESPOND
: /* Cannot happen.
542 It can, it SYNs are crossed. --ANK */
543 if (!sock_owned_by_user(sk
)) {
544 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS
);
547 * Wake people up to see the error
548 * (see connect in sock.c)
550 sk
->sk_error_report(sk
);
554 sk
->sk_err_soft
= err
;
558 if (!sock_owned_by_user(sk
) && np
->recverr
) {
560 sk
->sk_error_report(sk
);
562 sk
->sk_err_soft
= err
;
570 static int dccp_v6_send_response(struct sock
*sk
, struct request_sock
*req
,
571 struct dst_entry
*dst
)
573 struct inet6_request_sock
*ireq6
= inet6_rsk(req
);
574 struct ipv6_pinfo
*np
= inet6_sk(sk
);
576 struct ipv6_txoptions
*opt
= NULL
;
577 struct in6_addr
*final_p
= NULL
, final
;
581 memset(&fl
, 0, sizeof(fl
));
582 fl
.proto
= IPPROTO_DCCP
;
583 ipv6_addr_copy(&fl
.fl6_dst
, &ireq6
->rmt_addr
);
584 ipv6_addr_copy(&fl
.fl6_src
, &ireq6
->loc_addr
);
585 fl
.fl6_flowlabel
= 0;
587 fl
.fl_ip_dport
= inet_rsk(req
)->rmt_port
;
588 fl
.fl_ip_sport
= inet_sk(sk
)->sport
;
593 np
->rxopt
.bits
.osrcrt
== 2 &&
595 struct sk_buff
*pktopts
= ireq6
->pktopts
;
596 struct inet6_skb_parm
*rxopt
= IP6CB(pktopts
);
598 opt
= ipv6_invert_rthdr(sk
,
599 (struct ipv6_rt_hdr
*)(pktopts
->nh
.raw
+
603 if (opt
&& opt
->srcrt
) {
604 struct rt0_hdr
*rt0
= (struct rt0_hdr
*)opt
->srcrt
;
605 ipv6_addr_copy(&final
, &fl
.fl6_dst
);
606 ipv6_addr_copy(&fl
.fl6_dst
, rt0
->addr
);
610 err
= ip6_dst_lookup(sk
, &dst
, &fl
);
614 ipv6_addr_copy(&fl
.fl6_dst
, final_p
);
615 if ((err
= xfrm_lookup(&dst
, &fl
, sk
, 0)) < 0)
619 skb
= dccp_make_response(sk
, dst
, req
);
621 struct dccp_hdr
*dh
= dccp_hdr(skb
);
622 dh
->dccph_checksum
= dccp_v6_check(dh
, skb
->len
,
625 csum_partial((char *)dh
,
628 ipv6_addr_copy(&fl
.fl6_dst
, &ireq6
->rmt_addr
);
629 err
= ip6_xmit(sk
, skb
, &fl
, opt
, 0);
630 if (err
== NET_XMIT_CN
)
635 if (opt
&& opt
!= np
->opt
)
636 sock_kfree_s(sk
, opt
, opt
->tot_len
);
640 static void dccp_v6_reqsk_destructor(struct request_sock
*req
)
642 if (inet6_rsk(req
)->pktopts
!= NULL
)
643 kfree_skb(inet6_rsk(req
)->pktopts
);
646 static struct request_sock_ops dccp6_request_sock_ops
= {
648 .obj_size
= sizeof(struct dccp6_request_sock
),
649 .rtx_syn_ack
= dccp_v6_send_response
,
650 .send_ack
= dccp_v6_reqsk_send_ack
,
651 .destructor
= dccp_v6_reqsk_destructor
,
652 .send_reset
= dccp_v6_ctl_send_reset
,
655 static void dccp_v6_send_check(struct sock
*sk
, int len
, struct sk_buff
*skb
)
657 struct ipv6_pinfo
*np
= inet6_sk(sk
);
658 struct dccp_hdr
*dh
= dccp_hdr(skb
);
660 dh
->dccph_checksum
= csum_ipv6_magic(&np
->saddr
, &np
->daddr
,
662 csum_partial((char *)dh
,
667 static void dccp_v6_ctl_send_reset(struct sk_buff
*rxskb
)
669 struct dccp_hdr
*rxdh
= dccp_hdr(rxskb
), *dh
;
670 const int dccp_hdr_reset_len
= sizeof(struct dccp_hdr
) +
671 sizeof(struct dccp_hdr_ext
) +
672 sizeof(struct dccp_hdr_reset
);
677 if (rxdh
->dccph_type
== DCCP_PKT_RESET
)
680 if (!ipv6_unicast_destination(rxskb
))
684 * We need to grab some memory, and put together an RST,
685 * and then put it into the queue to be sent.
688 skb
= alloc_skb(MAX_HEADER
+ sizeof(struct ipv6hdr
) +
689 dccp_hdr_reset_len
, GFP_ATOMIC
);
693 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct ipv6hdr
) +
696 skb
->h
.raw
= skb_push(skb
, dccp_hdr_reset_len
);
698 memset(dh
, 0, dccp_hdr_reset_len
);
700 /* Swap the send and the receive. */
701 dh
->dccph_type
= DCCP_PKT_RESET
;
702 dh
->dccph_sport
= rxdh
->dccph_dport
;
703 dh
->dccph_dport
= rxdh
->dccph_sport
;
704 dh
->dccph_doff
= dccp_hdr_reset_len
/ 4;
706 dccp_hdr_reset(skb
)->dccph_reset_code
=
707 DCCP_SKB_CB(rxskb
)->dccpd_reset_code
;
709 /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
711 if (DCCP_SKB_CB(rxskb
)->dccpd_ack_seq
!= DCCP_PKT_WITHOUT_ACK_SEQ
)
712 dccp_set_seqno(&seqno
, DCCP_SKB_CB(rxskb
)->dccpd_ack_seq
+ 1);
714 dccp_hdr_set_seq(dh
, seqno
);
715 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb
),
716 DCCP_SKB_CB(rxskb
)->dccpd_seq
);
718 memset(&fl
, 0, sizeof(fl
));
719 ipv6_addr_copy(&fl
.fl6_dst
, &rxskb
->nh
.ipv6h
->saddr
);
720 ipv6_addr_copy(&fl
.fl6_src
, &rxskb
->nh
.ipv6h
->daddr
);
721 dh
->dccph_checksum
= csum_ipv6_magic(&fl
.fl6_src
, &fl
.fl6_dst
,
722 sizeof(*dh
), IPPROTO_DCCP
,
724 fl
.proto
= IPPROTO_DCCP
;
725 fl
.oif
= inet6_iif(rxskb
);
726 fl
.fl_ip_dport
= dh
->dccph_dport
;
727 fl
.fl_ip_sport
= dh
->dccph_sport
;
729 /* sk = NULL, but it is safe for now. RST socket required. */
730 if (!ip6_dst_lookup(NULL
, &skb
->dst
, &fl
)) {
731 if (xfrm_lookup(&skb
->dst
, &fl
, NULL
, 0) >= 0) {
732 ip6_xmit(NULL
, skb
, &fl
, NULL
, 0);
733 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS
);
734 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS
);
742 static void dccp_v6_ctl_send_ack(struct sk_buff
*rxskb
)
745 struct dccp_hdr
*rxdh
= dccp_hdr(rxskb
), *dh
;
746 const int dccp_hdr_ack_len
= sizeof(struct dccp_hdr
) +
747 sizeof(struct dccp_hdr_ext
) +
748 sizeof(struct dccp_hdr_ack_bits
);
751 skb
= alloc_skb(MAX_HEADER
+ sizeof(struct ipv6hdr
) +
752 dccp_hdr_ack_len
, GFP_ATOMIC
);
756 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct ipv6hdr
) +
759 skb
->h
.raw
= skb_push(skb
, dccp_hdr_ack_len
);
761 memset(dh
, 0, dccp_hdr_ack_len
);
763 /* Build DCCP header and checksum it. */
764 dh
->dccph_type
= DCCP_PKT_ACK
;
765 dh
->dccph_sport
= rxdh
->dccph_dport
;
766 dh
->dccph_dport
= rxdh
->dccph_sport
;
767 dh
->dccph_doff
= dccp_hdr_ack_len
/ 4;
770 dccp_hdr_set_seq(dh
, DCCP_SKB_CB(rxskb
)->dccpd_ack_seq
);
771 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb
),
772 DCCP_SKB_CB(rxskb
)->dccpd_seq
);
774 memset(&fl
, 0, sizeof(fl
));
775 ipv6_addr_copy(&fl
.fl6_dst
, &rxskb
->nh
.ipv6h
->saddr
);
776 ipv6_addr_copy(&fl
.fl6_src
, &rxskb
->nh
.ipv6h
->daddr
);
778 /* FIXME: calculate checksum, IPv4 also should... */
780 fl
.proto
= IPPROTO_DCCP
;
781 fl
.oif
= inet6_iif(rxskb
);
782 fl
.fl_ip_dport
= dh
->dccph_dport
;
783 fl
.fl_ip_sport
= dh
->dccph_sport
;
785 if (!ip6_dst_lookup(NULL
, &skb
->dst
, &fl
)) {
786 if (xfrm_lookup(&skb
->dst
, &fl
, NULL
, 0) >= 0) {
787 ip6_xmit(NULL
, skb
, &fl
, NULL
, 0);
788 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS
);
796 static void dccp_v6_reqsk_send_ack(struct sk_buff
*skb
,
797 struct request_sock
*req
)
799 dccp_v6_ctl_send_ack(skb
);
802 static struct sock
*dccp_v6_hnd_req(struct sock
*sk
,struct sk_buff
*skb
)
804 const struct dccp_hdr
*dh
= dccp_hdr(skb
);
805 const struct ipv6hdr
*iph
= skb
->nh
.ipv6h
;
807 struct request_sock
**prev
;
808 /* Find possible connection requests. */
809 struct request_sock
*req
= inet6_csk_search_req(sk
, &prev
,
815 return dccp_check_req(sk
, skb
, req
, prev
);
817 nsk
= __inet6_lookup_established(&dccp_hashinfo
,
818 &iph
->saddr
, dh
->dccph_sport
,
819 &iph
->daddr
, ntohs(dh
->dccph_dport
),
823 if (nsk
->sk_state
!= DCCP_TIME_WAIT
) {
827 inet_twsk_put((struct inet_timewait_sock
*)nsk
);
834 static int dccp_v6_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
836 struct inet_request_sock
*ireq
;
838 struct request_sock
*req
;
839 struct dccp_request_sock
*dreq
;
840 struct inet6_request_sock
*ireq6
;
841 struct ipv6_pinfo
*np
= inet6_sk(sk
);
842 const __u32 service
= dccp_hdr_request(skb
)->dccph_req_service
;
843 struct dccp_skb_cb
*dcb
= DCCP_SKB_CB(skb
);
844 __u8 reset_code
= DCCP_RESET_CODE_TOO_BUSY
;
846 if (skb
->protocol
== htons(ETH_P_IP
))
847 return dccp_v4_conn_request(sk
, skb
);
849 if (!ipv6_unicast_destination(skb
))
852 if (dccp_bad_service_code(sk
, service
)) {
853 reset_code
= DCCP_RESET_CODE_BAD_SERVICE_CODE
;
857 * There are no SYN attacks on IPv6, yet...
859 if (inet_csk_reqsk_queue_is_full(sk
))
862 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1)
865 req
= inet6_reqsk_alloc(sk
->sk_prot
->rsk_prot
);
869 /* FIXME: process options */
871 dccp_openreq_init(req
, &dp
, skb
);
873 ireq6
= inet6_rsk(req
);
874 ireq
= inet_rsk(req
);
875 ipv6_addr_copy(&ireq6
->rmt_addr
, &skb
->nh
.ipv6h
->saddr
);
876 ipv6_addr_copy(&ireq6
->loc_addr
, &skb
->nh
.ipv6h
->daddr
);
877 req
->rcv_wnd
= 100; /* Fake, option parsing will get the
879 ireq6
->pktopts
= NULL
;
881 if (ipv6_opt_accepted(sk
, skb
) ||
882 np
->rxopt
.bits
.rxinfo
|| np
->rxopt
.bits
.rxoinfo
||
883 np
->rxopt
.bits
.rxhlim
|| np
->rxopt
.bits
.rxohlim
) {
884 atomic_inc(&skb
->users
);
885 ireq6
->pktopts
= skb
;
887 ireq6
->iif
= sk
->sk_bound_dev_if
;
889 /* So that link locals have meaning */
890 if (!sk
->sk_bound_dev_if
&&
891 ipv6_addr_type(&ireq6
->rmt_addr
) & IPV6_ADDR_LINKLOCAL
)
892 ireq6
->iif
= inet6_iif(skb
);
895 * Step 3: Process LISTEN state
897 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
899 * In fact we defer setting S.GSR, S.SWL, S.SWH to
900 * dccp_create_openreq_child.
902 dreq
= dccp_rsk(req
);
903 dreq
->dreq_isr
= dcb
->dccpd_seq
;
904 dreq
->dreq_iss
= dccp_v6_init_sequence(sk
, skb
);
905 dreq
->dreq_service
= service
;
907 if (dccp_v6_send_response(sk
, req
, NULL
))
910 inet6_csk_reqsk_queue_hash_add(sk
, req
, DCCP_TIMEOUT_INIT
);
916 DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS
);
917 dcb
->dccpd_reset_code
= reset_code
;
921 static struct sock
*dccp_v6_request_recv_sock(struct sock
*sk
,
923 struct request_sock
*req
,
924 struct dst_entry
*dst
)
926 struct inet6_request_sock
*ireq6
= inet6_rsk(req
);
927 struct ipv6_pinfo
*newnp
, *np
= inet6_sk(sk
);
928 struct inet_sock
*newinet
;
929 struct dccp_sock
*newdp
;
930 struct dccp6_sock
*newdp6
;
932 struct ipv6_txoptions
*opt
;
934 if (skb
->protocol
== htons(ETH_P_IP
)) {
939 newsk
= dccp_v4_request_recv_sock(sk
, skb
, req
, dst
);
943 newdp6
= (struct dccp6_sock
*)newsk
;
944 newdp
= dccp_sk(newsk
);
945 newinet
= inet_sk(newsk
);
946 newinet
->pinet6
= &newdp6
->inet6
;
947 newnp
= inet6_sk(newsk
);
949 memcpy(newnp
, np
, sizeof(struct ipv6_pinfo
));
951 ipv6_addr_set(&newnp
->daddr
, 0, 0, htonl(0x0000FFFF),
954 ipv6_addr_set(&newnp
->saddr
, 0, 0, htonl(0x0000FFFF),
957 ipv6_addr_copy(&newnp
->rcv_saddr
, &newnp
->saddr
);
959 inet_csk(newsk
)->icsk_af_ops
= &dccp_ipv6_mapped
;
960 newsk
->sk_backlog_rcv
= dccp_v4_do_rcv
;
961 newnp
->pktoptions
= NULL
;
963 newnp
->mcast_oif
= inet6_iif(skb
);
964 newnp
->mcast_hops
= skb
->nh
.ipv6h
->hop_limit
;
967 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
968 * here, dccp_create_openreq_child now does this for us, see the comment in
969 * that function for the gory details. -acme
972 /* It is tricky place. Until this moment IPv4 tcp
973 worked with IPv6 icsk.icsk_af_ops.
976 dccp_sync_mss(newsk
, newdp
->dccps_pmtu_cookie
);
983 if (sk_acceptq_is_full(sk
))
986 if (np
->rxopt
.bits
.osrcrt
== 2 &&
987 opt
== NULL
&& ireq6
->pktopts
) {
988 struct inet6_skb_parm
*rxopt
= IP6CB(ireq6
->pktopts
);
990 opt
= ipv6_invert_rthdr(sk
,
991 (struct ipv6_rt_hdr
*)(ireq6
->pktopts
->nh
.raw
+
996 struct in6_addr
*final_p
= NULL
, final
;
999 memset(&fl
, 0, sizeof(fl
));
1000 fl
.proto
= IPPROTO_DCCP
;
1001 ipv6_addr_copy(&fl
.fl6_dst
, &ireq6
->rmt_addr
);
1002 if (opt
&& opt
->srcrt
) {
1003 struct rt0_hdr
*rt0
= (struct rt0_hdr
*) opt
->srcrt
;
1004 ipv6_addr_copy(&final
, &fl
.fl6_dst
);
1005 ipv6_addr_copy(&fl
.fl6_dst
, rt0
->addr
);
1008 ipv6_addr_copy(&fl
.fl6_src
, &ireq6
->loc_addr
);
1009 fl
.oif
= sk
->sk_bound_dev_if
;
1010 fl
.fl_ip_dport
= inet_rsk(req
)->rmt_port
;
1011 fl
.fl_ip_sport
= inet_sk(sk
)->sport
;
1013 if (ip6_dst_lookup(sk
, &dst
, &fl
))
1017 ipv6_addr_copy(&fl
.fl6_dst
, final_p
);
1019 if ((xfrm_lookup(&dst
, &fl
, sk
, 0)) < 0)
1023 newsk
= dccp_create_openreq_child(sk
, req
, skb
);
1028 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1029 * count here, dccp_create_openreq_child now does this for us, see the
1030 * comment in that function for the gory details. -acme
1033 ip6_dst_store(newsk
, dst
, NULL
);
1034 newsk
->sk_route_caps
= dst
->dev
->features
&
1035 ~(NETIF_F_IP_CSUM
| NETIF_F_TSO
);
1037 newdp6
= (struct dccp6_sock
*)newsk
;
1038 newinet
= inet_sk(newsk
);
1039 newinet
->pinet6
= &newdp6
->inet6
;
1040 newdp
= dccp_sk(newsk
);
1041 newnp
= inet6_sk(newsk
);
1043 memcpy(newnp
, np
, sizeof(struct ipv6_pinfo
));
1045 ipv6_addr_copy(&newnp
->daddr
, &ireq6
->rmt_addr
);
1046 ipv6_addr_copy(&newnp
->saddr
, &ireq6
->loc_addr
);
1047 ipv6_addr_copy(&newnp
->rcv_saddr
, &ireq6
->loc_addr
);
1048 newsk
->sk_bound_dev_if
= ireq6
->iif
;
1050 /* Now IPv6 options...
1052 First: no IPv4 options.
1054 newinet
->opt
= NULL
;
1057 newnp
->rxopt
.all
= np
->rxopt
.all
;
1059 /* Clone pktoptions received with SYN */
1060 newnp
->pktoptions
= NULL
;
1061 if (ireq6
->pktopts
!= NULL
) {
1062 newnp
->pktoptions
= skb_clone(ireq6
->pktopts
, GFP_ATOMIC
);
1063 kfree_skb(ireq6
->pktopts
);
1064 ireq6
->pktopts
= NULL
;
1065 if (newnp
->pktoptions
)
1066 skb_set_owner_r(newnp
->pktoptions
, newsk
);
1069 newnp
->mcast_oif
= inet6_iif(skb
);
1070 newnp
->mcast_hops
= skb
->nh
.ipv6h
->hop_limit
;
1072 /* Clone native IPv6 options from listening socket (if any)
1074 Yes, keeping reference count would be much more clever,
1075 but we make one more one thing there: reattach optmem
1079 newnp
->opt
= ipv6_dup_options(newsk
, opt
);
1081 sock_kfree_s(sk
, opt
, opt
->tot_len
);
1084 newdp
->dccps_ext_header_len
= 0;
1086 newdp
->dccps_ext_header_len
= newnp
->opt
->opt_nflen
+
1087 newnp
->opt
->opt_flen
;
1089 dccp_sync_mss(newsk
, dst_mtu(dst
));
1091 newinet
->daddr
= newinet
->saddr
= newinet
->rcv_saddr
= LOOPBACK4_IPV6
;
1093 __inet6_hash(&dccp_hashinfo
, newsk
);
1094 inet_inherit_port(&dccp_hashinfo
, sk
, newsk
);
1099 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS
);
1101 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS
);
1102 if (opt
&& opt
!= np
->opt
)
1103 sock_kfree_s(sk
, opt
, opt
->tot_len
);
1108 /* The socket must have it's spinlock held when we get
1111 * We have a potential double-lock case here, so even when
1112 * doing backlog processing we use the BH locking scheme.
1113 * This is because we cannot sleep with the original spinlock
1116 static int dccp_v6_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1118 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1119 struct sk_buff
*opt_skb
= NULL
;
1121 /* Imagine: socket is IPv6. IPv4 packet arrives,
1122 goes to IPv4 receive handler and backlogged.
1123 From backlog it always goes here. Kerboom...
1124 Fortunately, dccp_rcv_established and rcv_established
1125 handle them correctly, but it is not case with
1126 dccp_v6_hnd_req and dccp_v6_ctl_send_reset(). --ANK
1129 if (skb
->protocol
== htons(ETH_P_IP
))
1130 return dccp_v4_do_rcv(sk
, skb
);
1132 if (sk_filter(sk
, skb
, 0))
1136 * socket locking is here for SMP purposes as backlog rcv
1137 * is currently called with bh processing disabled.
1140 /* Do Stevens' IPV6_PKTOPTIONS.
1142 Yes, guys, it is the only place in our code, where we
1143 may make it not affecting IPv4.
1144 The rest of code is protocol independent,
1145 and I do not like idea to uglify IPv4.
1147 Actually, all the idea behind IPV6_PKTOPTIONS
1148 looks not very well thought. For now we latch
1149 options, received in the last packet, enqueued
1150 by tcp. Feel free to propose better solution.
1154 opt_skb
= skb_clone(skb
, GFP_ATOMIC
);
1156 if (sk
->sk_state
== DCCP_OPEN
) { /* Fast path */
1157 if (dccp_rcv_established(sk
, skb
, dccp_hdr(skb
), skb
->len
))
1162 if (sk
->sk_state
== DCCP_LISTEN
) {
1163 struct sock
*nsk
= dccp_v6_hnd_req(sk
, skb
);
1168 * Queue it on the new socket if the new socket is active,
1169 * otherwise we just shortcircuit this and continue with
1173 if (dccp_child_process(sk
, nsk
, skb
))
1176 __kfree_skb(opt_skb
);
1181 if (dccp_rcv_state_process(sk
, skb
, dccp_hdr(skb
), skb
->len
))
1186 dccp_v6_ctl_send_reset(skb
);
1189 __kfree_skb(opt_skb
);
1194 static int dccp_v6_rcv(struct sk_buff
**pskb
, unsigned int *nhoffp
)
1196 const struct dccp_hdr
*dh
;
1197 struct sk_buff
*skb
= *pskb
;
1201 /* Step 1: Check header basics: */
1203 if (dccp_invalid_packet(skb
))
1208 DCCP_SKB_CB(skb
)->dccpd_seq
= dccp_hdr_seq(skb
);
1209 DCCP_SKB_CB(skb
)->dccpd_type
= dh
->dccph_type
;
1211 if (dccp_packet_without_ack(skb
))
1212 DCCP_SKB_CB(skb
)->dccpd_ack_seq
= DCCP_PKT_WITHOUT_ACK_SEQ
;
1214 DCCP_SKB_CB(skb
)->dccpd_ack_seq
= dccp_hdr_ack_seq(skb
);
1217 * Look up flow ID in table and get corresponding socket */
1218 sk
= __inet6_lookup(&dccp_hashinfo
, &skb
->nh
.ipv6h
->saddr
,
1220 &skb
->nh
.ipv6h
->daddr
, ntohs(dh
->dccph_dport
),
1225 * Generate Reset(No Connection) unless P.type == Reset
1226 * Drop packet and return
1229 goto no_dccp_socket
;
1233 * ... or S.state == TIMEWAIT,
1234 * Generate Reset(No Connection) unless P.type == Reset
1235 * Drop packet and return
1238 if (sk
->sk_state
== DCCP_TIME_WAIT
)
1241 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1242 goto discard_and_relse
;
1244 if (sk_filter(sk
, skb
, 0))
1245 goto discard_and_relse
;
1251 if (!sock_owned_by_user(sk
))
1252 rc
= dccp_v6_do_rcv(sk
, skb
);
1254 sk_add_backlog(sk
, skb
);
1261 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1265 * Generate Reset(No Connection) unless P.type == Reset
1266 * Drop packet and return
1268 if (dh
->dccph_type
!= DCCP_PKT_RESET
) {
1269 DCCP_SKB_CB(skb
)->dccpd_reset_code
=
1270 DCCP_RESET_CODE_NO_CONNECTION
;
1271 dccp_v6_ctl_send_reset(skb
);
1287 inet_twsk_put((struct inet_timewait_sock
*)sk
);
1288 goto no_dccp_socket
;
1291 static struct inet_connection_sock_af_ops dccp_ipv6_af_ops
= {
1292 .queue_xmit
= inet6_csk_xmit
,
1293 .send_check
= dccp_v6_send_check
,
1294 .rebuild_header
= inet6_sk_rebuild_header
,
1295 .conn_request
= dccp_v6_conn_request
,
1296 .syn_recv_sock
= dccp_v6_request_recv_sock
,
1297 .net_header_len
= sizeof(struct ipv6hdr
),
1298 .setsockopt
= ipv6_setsockopt
,
1299 .getsockopt
= ipv6_getsockopt
,
1300 .addr2sockaddr
= inet6_csk_addr2sockaddr
,
1301 .sockaddr_len
= sizeof(struct sockaddr_in6
)
1305 * DCCP over IPv4 via INET6 API
1307 static struct inet_connection_sock_af_ops dccp_ipv6_mapped
= {
1308 .queue_xmit
= ip_queue_xmit
,
1309 .send_check
= dccp_v4_send_check
,
1310 .rebuild_header
= inet_sk_rebuild_header
,
1311 .conn_request
= dccp_v6_conn_request
,
1312 .syn_recv_sock
= dccp_v6_request_recv_sock
,
1313 .net_header_len
= sizeof(struct iphdr
),
1314 .setsockopt
= ipv6_setsockopt
,
1315 .getsockopt
= ipv6_getsockopt
,
1316 .addr2sockaddr
= inet6_csk_addr2sockaddr
,
1317 .sockaddr_len
= sizeof(struct sockaddr_in6
)
1320 /* NOTE: A lot of things set to zero explicitly by call to
1321 * sk_alloc() so need not be done here.
1323 static int dccp_v6_init_sock(struct sock
*sk
)
1325 int err
= dccp_v4_init_sock(sk
);
1328 inet_csk(sk
)->icsk_af_ops
= &dccp_ipv6_af_ops
;
1333 static int dccp_v6_destroy_sock(struct sock
*sk
)
1335 dccp_v4_destroy_sock(sk
);
1336 return inet6_destroy_sock(sk
);
1339 static struct proto dccp_v6_prot
= {
1341 .owner
= THIS_MODULE
,
1342 .close
= dccp_close
,
1343 .connect
= dccp_v6_connect
,
1344 .disconnect
= dccp_disconnect
,
1345 .ioctl
= dccp_ioctl
,
1346 .init
= dccp_v6_init_sock
,
1347 .setsockopt
= dccp_setsockopt
,
1348 .getsockopt
= dccp_getsockopt
,
1349 .sendmsg
= dccp_sendmsg
,
1350 .recvmsg
= dccp_recvmsg
,
1351 .backlog_rcv
= dccp_v6_do_rcv
,
1352 .hash
= dccp_v6_hash
,
1353 .unhash
= dccp_unhash
,
1354 .accept
= inet_csk_accept
,
1355 .get_port
= dccp_v6_get_port
,
1356 .shutdown
= dccp_shutdown
,
1357 .destroy
= dccp_v6_destroy_sock
,
1358 .orphan_count
= &dccp_orphan_count
,
1359 .max_header
= MAX_DCCP_HEADER
,
1360 .obj_size
= sizeof(struct dccp6_sock
),
1361 .rsk_prot
= &dccp6_request_sock_ops
,
1362 .twsk_obj_size
= sizeof(struct dccp6_timewait_sock
),
1365 static struct inet6_protocol dccp_v6_protocol
= {
1366 .handler
= dccp_v6_rcv
,
1367 .err_handler
= dccp_v6_err
,
1368 .flags
= INET6_PROTO_NOPOLICY
| INET6_PROTO_FINAL
,
1371 static struct proto_ops inet6_dccp_ops
= {
1373 .owner
= THIS_MODULE
,
1374 .release
= inet6_release
,
1376 .connect
= inet_stream_connect
,
1377 .socketpair
= sock_no_socketpair
,
1378 .accept
= inet_accept
,
1379 .getname
= inet6_getname
,
1381 .ioctl
= inet6_ioctl
,
1382 .listen
= inet_dccp_listen
,
1383 .shutdown
= inet_shutdown
,
1384 .setsockopt
= sock_common_setsockopt
,
1385 .getsockopt
= sock_common_getsockopt
,
1386 .sendmsg
= inet_sendmsg
,
1387 .recvmsg
= sock_common_recvmsg
,
1388 .mmap
= sock_no_mmap
,
1389 .sendpage
= sock_no_sendpage
,
1392 static struct inet_protosw dccp_v6_protosw
= {
1394 .protocol
= IPPROTO_DCCP
,
1395 .prot
= &dccp_v6_prot
,
1396 .ops
= &inet6_dccp_ops
,
1400 static int __init
dccp_v6_init(void)
1402 int err
= proto_register(&dccp_v6_prot
, 1);
1407 err
= inet6_add_protocol(&dccp_v6_protocol
, IPPROTO_DCCP
);
1409 goto out_unregister_proto
;
1411 inet6_register_protosw(&dccp_v6_protosw
);
1414 out_unregister_proto
:
1415 proto_unregister(&dccp_v6_prot
);
1419 static void __exit
dccp_v6_exit(void)
1421 inet6_del_protocol(&dccp_v6_protocol
, IPPROTO_DCCP
);
1422 inet6_unregister_protosw(&dccp_v6_protosw
);
1423 proto_unregister(&dccp_v6_prot
);
1426 module_init(dccp_v6_init
);
1427 module_exit(dccp_v6_exit
);
1430 * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1431 * values directly, Also cover the case where the protocol is not specified,
1432 * i.e. net-pf-PF_INET6-proto-0-type-SOCK_DCCP
1434 MODULE_ALIAS("net-pf-" __stringify(PF_INET6
) "-proto-33-type-6");
1435 MODULE_ALIAS("net-pf-" __stringify(PF_INET6
) "-proto-0-type-6");
1436 MODULE_LICENSE("GPL");
1437 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1438 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol");