tcp/dccp: install syn_recv requests into ehash table
[linux.git] / net / dccp / ipv6.c
1 /*
2  *      DCCP over IPv6
3  *      Linux INET6 implementation
4  *
5  *      Based on net/dccp6/ipv6.c
6  *
7  *      Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
8  *
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.
13  */
14
15 #include <linux/module.h>
16 #include <linux/random.h>
17 #include <linux/slab.h>
18 #include <linux/xfrm.h>
19
20 #include <net/addrconf.h>
21 #include <net/inet_common.h>
22 #include <net/inet_hashtables.h>
23 #include <net/inet_sock.h>
24 #include <net/inet6_connection_sock.h>
25 #include <net/inet6_hashtables.h>
26 #include <net/ip6_route.h>
27 #include <net/ipv6.h>
28 #include <net/protocol.h>
29 #include <net/transp_v6.h>
30 #include <net/ip6_checksum.h>
31 #include <net/xfrm.h>
32 #include <net/secure_seq.h>
33
34 #include "dccp.h"
35 #include "ipv6.h"
36 #include "feat.h"
37
38 /* The per-net dccp.v6_ctl_sk is used for sending RSTs and ACKs */
39
40 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped;
41 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops;
42
43 /* add pseudo-header to DCCP checksum stored in skb->csum */
44 static inline __sum16 dccp_v6_csum_finish(struct sk_buff *skb,
45                                       const struct in6_addr *saddr,
46                                       const struct in6_addr *daddr)
47 {
48         return csum_ipv6_magic(saddr, daddr, skb->len, IPPROTO_DCCP, skb->csum);
49 }
50
51 static inline void dccp_v6_send_check(struct sock *sk, struct sk_buff *skb)
52 {
53         struct ipv6_pinfo *np = inet6_sk(sk);
54         struct dccp_hdr *dh = dccp_hdr(skb);
55
56         dccp_csum_outgoing(skb);
57         dh->dccph_checksum = dccp_v6_csum_finish(skb, &np->saddr, &sk->sk_v6_daddr);
58 }
59
60 static inline __u64 dccp_v6_init_sequence(struct sk_buff *skb)
61 {
62         return secure_dccpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
63                                              ipv6_hdr(skb)->saddr.s6_addr32,
64                                              dccp_hdr(skb)->dccph_dport,
65                                              dccp_hdr(skb)->dccph_sport     );
66
67 }
68
69 static void dccp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
70                         u8 type, u8 code, int offset, __be32 info)
71 {
72         const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
73         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
74         struct dccp_sock *dp;
75         struct ipv6_pinfo *np;
76         struct sock *sk;
77         int err;
78         __u64 seq;
79         struct net *net = dev_net(skb->dev);
80
81         if (skb->len < offset + sizeof(*dh) ||
82             skb->len < offset + __dccp_basic_hdr_len(dh)) {
83                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
84                                    ICMP6_MIB_INERRORS);
85                 return;
86         }
87
88         sk = __inet6_lookup_established(net, &dccp_hashinfo,
89                                         &hdr->daddr, dh->dccph_dport,
90                                         &hdr->saddr, ntohs(dh->dccph_sport),
91                                         inet6_iif(skb));
92
93         if (!sk) {
94                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
95                                    ICMP6_MIB_INERRORS);
96                 return;
97         }
98
99         if (sk->sk_state == DCCP_TIME_WAIT) {
100                 inet_twsk_put(inet_twsk(sk));
101                 return;
102         }
103         seq = dccp_hdr_seq(dh);
104         if (sk->sk_state == DCCP_NEW_SYN_RECV)
105                 return dccp_req_err(sk, seq);
106
107         bh_lock_sock(sk);
108         if (sock_owned_by_user(sk))
109                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
110
111         if (sk->sk_state == DCCP_CLOSED)
112                 goto out;
113
114         dp = dccp_sk(sk);
115         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
116             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
117                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
118                 goto out;
119         }
120
121         np = inet6_sk(sk);
122
123         if (type == NDISC_REDIRECT) {
124                 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
125
126                 if (dst)
127                         dst->ops->redirect(dst, sk, skb);
128                 goto out;
129         }
130
131         if (type == ICMPV6_PKT_TOOBIG) {
132                 struct dst_entry *dst = NULL;
133
134                 if (!ip6_sk_accept_pmtu(sk))
135                         goto out;
136
137                 if (sock_owned_by_user(sk))
138                         goto out;
139                 if ((1 << sk->sk_state) & (DCCPF_LISTEN | DCCPF_CLOSED))
140                         goto out;
141
142                 dst = inet6_csk_update_pmtu(sk, ntohl(info));
143                 if (!dst)
144                         goto out;
145
146                 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst))
147                         dccp_sync_mss(sk, dst_mtu(dst));
148                 goto out;
149         }
150
151         icmpv6_err_convert(type, code, &err);
152
153         /* Might be for an request_sock */
154         switch (sk->sk_state) {
155         case DCCP_REQUESTING:
156         case DCCP_RESPOND:  /* Cannot happen.
157                                It can, it SYNs are crossed. --ANK */
158                 if (!sock_owned_by_user(sk)) {
159                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
160                         sk->sk_err = err;
161                         /*
162                          * Wake people up to see the error
163                          * (see connect in sock.c)
164                          */
165                         sk->sk_error_report(sk);
166                         dccp_done(sk);
167                 } else
168                         sk->sk_err_soft = err;
169                 goto out;
170         }
171
172         if (!sock_owned_by_user(sk) && np->recverr) {
173                 sk->sk_err = err;
174                 sk->sk_error_report(sk);
175         } else
176                 sk->sk_err_soft = err;
177
178 out:
179         bh_unlock_sock(sk);
180         sock_put(sk);
181 }
182
183
184 static int dccp_v6_send_response(const struct sock *sk, struct request_sock *req)
185 {
186         struct inet_request_sock *ireq = inet_rsk(req);
187         struct ipv6_pinfo *np = inet6_sk(sk);
188         struct sk_buff *skb;
189         struct in6_addr *final_p, final;
190         struct flowi6 fl6;
191         int err = -1;
192         struct dst_entry *dst;
193
194         memset(&fl6, 0, sizeof(fl6));
195         fl6.flowi6_proto = IPPROTO_DCCP;
196         fl6.daddr = ireq->ir_v6_rmt_addr;
197         fl6.saddr = ireq->ir_v6_loc_addr;
198         fl6.flowlabel = 0;
199         fl6.flowi6_oif = ireq->ir_iif;
200         fl6.fl6_dport = ireq->ir_rmt_port;
201         fl6.fl6_sport = htons(ireq->ir_num);
202         security_req_classify_flow(req, flowi6_to_flowi(&fl6));
203
204
205         final_p = fl6_update_dst(&fl6, np->opt, &final);
206
207         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
208         if (IS_ERR(dst)) {
209                 err = PTR_ERR(dst);
210                 dst = NULL;
211                 goto done;
212         }
213
214         skb = dccp_make_response(sk, dst, req);
215         if (skb != NULL) {
216                 struct dccp_hdr *dh = dccp_hdr(skb);
217
218                 dh->dccph_checksum = dccp_v6_csum_finish(skb,
219                                                          &ireq->ir_v6_loc_addr,
220                                                          &ireq->ir_v6_rmt_addr);
221                 fl6.daddr = ireq->ir_v6_rmt_addr;
222                 err = ip6_xmit(sk, skb, &fl6, np->opt, np->tclass);
223                 err = net_xmit_eval(err);
224         }
225
226 done:
227         dst_release(dst);
228         return err;
229 }
230
231 static void dccp_v6_reqsk_destructor(struct request_sock *req)
232 {
233         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
234         kfree_skb(inet_rsk(req)->pktopts);
235 }
236
237 static void dccp_v6_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
238 {
239         const struct ipv6hdr *rxip6h;
240         struct sk_buff *skb;
241         struct flowi6 fl6;
242         struct net *net = dev_net(skb_dst(rxskb)->dev);
243         struct sock *ctl_sk = net->dccp.v6_ctl_sk;
244         struct dst_entry *dst;
245
246         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
247                 return;
248
249         if (!ipv6_unicast_destination(rxskb))
250                 return;
251
252         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
253         if (skb == NULL)
254                 return;
255
256         rxip6h = ipv6_hdr(rxskb);
257         dccp_hdr(skb)->dccph_checksum = dccp_v6_csum_finish(skb, &rxip6h->saddr,
258                                                             &rxip6h->daddr);
259
260         memset(&fl6, 0, sizeof(fl6));
261         fl6.daddr = rxip6h->saddr;
262         fl6.saddr = rxip6h->daddr;
263
264         fl6.flowi6_proto = IPPROTO_DCCP;
265         fl6.flowi6_oif = inet6_iif(rxskb);
266         fl6.fl6_dport = dccp_hdr(skb)->dccph_dport;
267         fl6.fl6_sport = dccp_hdr(skb)->dccph_sport;
268         security_skb_classify_flow(rxskb, flowi6_to_flowi(&fl6));
269
270         /* sk = NULL, but it is safe for now. RST socket required. */
271         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
272         if (!IS_ERR(dst)) {
273                 skb_dst_set(skb, dst);
274                 ip6_xmit(ctl_sk, skb, &fl6, NULL, 0);
275                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
276                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
277                 return;
278         }
279
280         kfree_skb(skb);
281 }
282
283 static struct request_sock_ops dccp6_request_sock_ops = {
284         .family         = AF_INET6,
285         .obj_size       = sizeof(struct dccp6_request_sock),
286         .rtx_syn_ack    = dccp_v6_send_response,
287         .send_ack       = dccp_reqsk_send_ack,
288         .destructor     = dccp_v6_reqsk_destructor,
289         .send_reset     = dccp_v6_ctl_send_reset,
290         .syn_ack_timeout = dccp_syn_ack_timeout,
291 };
292
293 static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
294 {
295         struct request_sock *req;
296         struct dccp_request_sock *dreq;
297         struct inet_request_sock *ireq;
298         struct ipv6_pinfo *np = inet6_sk(sk);
299         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
300         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
301
302         if (skb->protocol == htons(ETH_P_IP))
303                 return dccp_v4_conn_request(sk, skb);
304
305         if (!ipv6_unicast_destination(skb))
306                 return 0;       /* discard, don't send a reset here */
307
308         if (dccp_bad_service_code(sk, service)) {
309                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
310                 goto drop;
311         }
312         /*
313          * There are no SYN attacks on IPv6, yet...
314          */
315         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
316         if (inet_csk_reqsk_queue_is_full(sk))
317                 goto drop;
318
319         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
320                 goto drop;
321
322         req = inet_reqsk_alloc(&dccp6_request_sock_ops, sk);
323         if (req == NULL)
324                 goto drop;
325
326         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
327                 goto drop_and_free;
328
329         dreq = dccp_rsk(req);
330         if (dccp_parse_options(sk, dreq, skb))
331                 goto drop_and_free;
332
333         if (security_inet_conn_request(sk, skb, req))
334                 goto drop_and_free;
335
336         ireq = inet_rsk(req);
337         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
338         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
339         ireq->ireq_family = AF_INET6;
340
341         if (ipv6_opt_accepted(sk, skb, IP6CB(skb)) ||
342             np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
343             np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
344                 atomic_inc(&skb->users);
345                 ireq->pktopts = skb;
346         }
347         ireq->ir_iif = sk->sk_bound_dev_if;
348
349         /* So that link locals have meaning */
350         if (!sk->sk_bound_dev_if &&
351             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
352                 ireq->ir_iif = inet6_iif(skb);
353
354         /*
355          * Step 3: Process LISTEN state
356          *
357          *   Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
358          *
359          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
360          */
361         dreq->dreq_isr     = dcb->dccpd_seq;
362         dreq->dreq_gsr     = dreq->dreq_isr;
363         dreq->dreq_iss     = dccp_v6_init_sequence(skb);
364         dreq->dreq_gss     = dreq->dreq_iss;
365         dreq->dreq_service = service;
366
367         if (dccp_v6_send_response(sk, req))
368                 goto drop_and_free;
369
370         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
371         return 0;
372
373 drop_and_free:
374         reqsk_free(req);
375 drop:
376         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
377         return -1;
378 }
379
380 static struct sock *dccp_v6_request_recv_sock(const struct sock *sk,
381                                               struct sk_buff *skb,
382                                               struct request_sock *req,
383                                               struct dst_entry *dst)
384 {
385         struct inet_request_sock *ireq = inet_rsk(req);
386         struct ipv6_pinfo *newnp;
387         const struct ipv6_pinfo *np = inet6_sk(sk);
388         struct inet_sock *newinet;
389         struct dccp6_sock *newdp6;
390         struct sock *newsk;
391
392         if (skb->protocol == htons(ETH_P_IP)) {
393                 /*
394                  *      v6 mapped
395                  */
396                 newsk = dccp_v4_request_recv_sock(sk, skb, req, dst);
397                 if (newsk == NULL)
398                         return NULL;
399
400                 newdp6 = (struct dccp6_sock *)newsk;
401                 newinet = inet_sk(newsk);
402                 newinet->pinet6 = &newdp6->inet6;
403                 newnp = inet6_sk(newsk);
404
405                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
406
407                 newnp->saddr = newsk->sk_v6_rcv_saddr;
408
409                 inet_csk(newsk)->icsk_af_ops = &dccp_ipv6_mapped;
410                 newsk->sk_backlog_rcv = dccp_v4_do_rcv;
411                 newnp->pktoptions  = NULL;
412                 newnp->opt         = NULL;
413                 newnp->mcast_oif   = inet6_iif(skb);
414                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
415
416                 /*
417                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
418                  * here, dccp_create_openreq_child now does this for us, see the comment in
419                  * that function for the gory details. -acme
420                  */
421
422                 /* It is tricky place. Until this moment IPv4 tcp
423                    worked with IPv6 icsk.icsk_af_ops.
424                    Sync it now.
425                  */
426                 dccp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
427
428                 return newsk;
429         }
430
431
432         if (sk_acceptq_is_full(sk))
433                 goto out_overflow;
434
435         if (!dst) {
436                 struct flowi6 fl6;
437
438                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_DCCP);
439                 if (!dst)
440                         goto out;
441         }
442
443         newsk = dccp_create_openreq_child(sk, req, skb);
444         if (newsk == NULL)
445                 goto out_nonewsk;
446
447         /*
448          * No need to charge this sock to the relevant IPv6 refcnt debug socks
449          * count here, dccp_create_openreq_child now does this for us, see the
450          * comment in that function for the gory details. -acme
451          */
452
453         __ip6_dst_store(newsk, dst, NULL, NULL);
454         newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM |
455                                                       NETIF_F_TSO);
456         newdp6 = (struct dccp6_sock *)newsk;
457         newinet = inet_sk(newsk);
458         newinet->pinet6 = &newdp6->inet6;
459         newnp = inet6_sk(newsk);
460
461         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
462
463         newsk->sk_v6_daddr      = ireq->ir_v6_rmt_addr;
464         newnp->saddr            = ireq->ir_v6_loc_addr;
465         newsk->sk_v6_rcv_saddr  = ireq->ir_v6_loc_addr;
466         newsk->sk_bound_dev_if  = ireq->ir_iif;
467
468         /* Now IPv6 options...
469
470            First: no IPv4 options.
471          */
472         newinet->inet_opt = NULL;
473
474         /* Clone RX bits */
475         newnp->rxopt.all = np->rxopt.all;
476
477         /* Clone pktoptions received with SYN */
478         newnp->pktoptions = NULL;
479         if (ireq->pktopts != NULL) {
480                 newnp->pktoptions = skb_clone(ireq->pktopts, GFP_ATOMIC);
481                 consume_skb(ireq->pktopts);
482                 ireq->pktopts = NULL;
483                 if (newnp->pktoptions)
484                         skb_set_owner_r(newnp->pktoptions, newsk);
485         }
486         newnp->opt        = NULL;
487         newnp->mcast_oif  = inet6_iif(skb);
488         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
489
490         /*
491          * Clone native IPv6 options from listening socket (if any)
492          *
493          * Yes, keeping reference count would be much more clever, but we make
494          * one more one thing there: reattach optmem to newsk.
495          */
496         if (np->opt != NULL)
497                 newnp->opt = ipv6_dup_options(newsk, np->opt);
498
499         inet_csk(newsk)->icsk_ext_hdr_len = 0;
500         if (newnp->opt != NULL)
501                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
502                                                      newnp->opt->opt_flen);
503
504         dccp_sync_mss(newsk, dst_mtu(dst));
505
506         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
507         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
508
509         if (__inet_inherit_port(sk, newsk) < 0) {
510                 inet_csk_prepare_forced_close(newsk);
511                 dccp_done(newsk);
512                 goto out;
513         }
514         __inet_hash(newsk, NULL);
515
516         return newsk;
517
518 out_overflow:
519         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
520 out_nonewsk:
521         dst_release(dst);
522 out:
523         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
524         return NULL;
525 }
526
527 /* The socket must have it's spinlock held when we get
528  * here.
529  *
530  * We have a potential double-lock case here, so even when
531  * doing backlog processing we use the BH locking scheme.
532  * This is because we cannot sleep with the original spinlock
533  * held.
534  */
535 static int dccp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
536 {
537         struct ipv6_pinfo *np = inet6_sk(sk);
538         struct sk_buff *opt_skb = NULL;
539
540         /* Imagine: socket is IPv6. IPv4 packet arrives,
541            goes to IPv4 receive handler and backlogged.
542            From backlog it always goes here. Kerboom...
543            Fortunately, dccp_rcv_established and rcv_established
544            handle them correctly, but it is not case with
545            dccp_v6_hnd_req and dccp_v6_ctl_send_reset().   --ANK
546          */
547
548         if (skb->protocol == htons(ETH_P_IP))
549                 return dccp_v4_do_rcv(sk, skb);
550
551         if (sk_filter(sk, skb))
552                 goto discard;
553
554         /*
555          * socket locking is here for SMP purposes as backlog rcv is currently
556          * called with bh processing disabled.
557          */
558
559         /* Do Stevens' IPV6_PKTOPTIONS.
560
561            Yes, guys, it is the only place in our code, where we
562            may make it not affecting IPv4.
563            The rest of code is protocol independent,
564            and I do not like idea to uglify IPv4.
565
566            Actually, all the idea behind IPV6_PKTOPTIONS
567            looks not very well thought. For now we latch
568            options, received in the last packet, enqueued
569            by tcp. Feel free to propose better solution.
570                                                --ANK (980728)
571          */
572         if (np->rxopt.all)
573         /*
574          * FIXME: Add handling of IPV6_PKTOPTIONS skb. See the comments below
575          *        (wrt ipv6_pktopions) and net/ipv6/tcp_ipv6.c for an example.
576          */
577                 opt_skb = skb_clone(skb, GFP_ATOMIC);
578
579         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
580                 if (dccp_rcv_established(sk, skb, dccp_hdr(skb), skb->len))
581                         goto reset;
582                 if (opt_skb) {
583                         /* XXX This is where we would goto ipv6_pktoptions. */
584                         __kfree_skb(opt_skb);
585                 }
586                 return 0;
587         }
588
589         /*
590          *  Step 3: Process LISTEN state
591          *     If S.state == LISTEN,
592          *       If P.type == Request or P contains a valid Init Cookie option,
593          *            (* Must scan the packet's options to check for Init
594          *               Cookies.  Only Init Cookies are processed here,
595          *               however; other options are processed in Step 8.  This
596          *               scan need only be performed if the endpoint uses Init
597          *               Cookies *)
598          *            (* Generate a new socket and switch to that socket *)
599          *            Set S := new socket for this port pair
600          *            S.state = RESPOND
601          *            Choose S.ISS (initial seqno) or set from Init Cookies
602          *            Initialize S.GAR := S.ISS
603          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
604          *            Continue with S.state == RESPOND
605          *            (* A Response packet will be generated in Step 11 *)
606          *       Otherwise,
607          *            Generate Reset(No Connection) unless P.type == Reset
608          *            Drop packet and return
609          *
610          * NOTE: the check for the packet types is done in
611          *       dccp_rcv_state_process
612          */
613
614         if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len))
615                 goto reset;
616         if (opt_skb) {
617                 /* XXX This is where we would goto ipv6_pktoptions. */
618                 __kfree_skb(opt_skb);
619         }
620         return 0;
621
622 reset:
623         dccp_v6_ctl_send_reset(sk, skb);
624 discard:
625         if (opt_skb != NULL)
626                 __kfree_skb(opt_skb);
627         kfree_skb(skb);
628         return 0;
629 }
630
631 static int dccp_v6_rcv(struct sk_buff *skb)
632 {
633         const struct dccp_hdr *dh;
634         struct sock *sk;
635         int min_cov;
636
637         /* Step 1: Check header basics */
638
639         if (dccp_invalid_packet(skb))
640                 goto discard_it;
641
642         /* Step 1: If header checksum is incorrect, drop packet and return. */
643         if (dccp_v6_csum_finish(skb, &ipv6_hdr(skb)->saddr,
644                                      &ipv6_hdr(skb)->daddr)) {
645                 DCCP_WARN("dropped packet with invalid checksum\n");
646                 goto discard_it;
647         }
648
649         dh = dccp_hdr(skb);
650
651         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
652         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
653
654         if (dccp_packet_without_ack(skb))
655                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
656         else
657                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
658
659         /* Step 2:
660          *      Look up flow ID in table and get corresponding socket */
661         sk = __inet6_lookup_skb(&dccp_hashinfo, skb,
662                                 dh->dccph_sport, dh->dccph_dport,
663                                 inet6_iif(skb));
664         /*
665          * Step 2:
666          *      If no socket ...
667          */
668         if (sk == NULL) {
669                 dccp_pr_debug("failed to look up flow ID in table and "
670                               "get corresponding socket\n");
671                 goto no_dccp_socket;
672         }
673
674         /*
675          * Step 2:
676          *      ... or S.state == TIMEWAIT,
677          *              Generate Reset(No Connection) unless P.type == Reset
678          *              Drop packet and return
679          */
680         if (sk->sk_state == DCCP_TIME_WAIT) {
681                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
682                 inet_twsk_put(inet_twsk(sk));
683                 goto no_dccp_socket;
684         }
685
686         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
687                 struct request_sock *req = inet_reqsk(sk);
688                 struct sock *nsk = NULL;
689
690                 sk = req->rsk_listener;
691                 if (sk->sk_state == DCCP_LISTEN)
692                         nsk = dccp_check_req(sk, skb, req);
693                 if (!nsk) {
694                         reqsk_put(req);
695                         goto discard_it;
696                 }
697                 if (nsk == sk) {
698                         sock_hold(sk);
699                         reqsk_put(req);
700                 } else if (dccp_child_process(sk, nsk, skb)) {
701                         dccp_v6_ctl_send_reset(sk, skb);
702                         goto discard_it;
703                 } else {
704                         return 0;
705                 }
706         }
707         /*
708          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
709          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
710          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
711          */
712         min_cov = dccp_sk(sk)->dccps_pcrlen;
713         if (dh->dccph_cscov  &&  (min_cov == 0 || dh->dccph_cscov < min_cov))  {
714                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
715                               dh->dccph_cscov, min_cov);
716                 /* FIXME: send Data Dropped option (see also dccp_v4_rcv) */
717                 goto discard_and_relse;
718         }
719
720         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
721                 goto discard_and_relse;
722
723         return sk_receive_skb(sk, skb, 1) ? -1 : 0;
724
725 no_dccp_socket:
726         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
727                 goto discard_it;
728         /*
729          * Step 2:
730          *      If no socket ...
731          *              Generate Reset(No Connection) unless P.type == Reset
732          *              Drop packet and return
733          */
734         if (dh->dccph_type != DCCP_PKT_RESET) {
735                 DCCP_SKB_CB(skb)->dccpd_reset_code =
736                                         DCCP_RESET_CODE_NO_CONNECTION;
737                 dccp_v6_ctl_send_reset(sk, skb);
738         }
739
740 discard_it:
741         kfree_skb(skb);
742         return 0;
743
744 discard_and_relse:
745         sock_put(sk);
746         goto discard_it;
747 }
748
749 static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
750                            int addr_len)
751 {
752         struct sockaddr_in6 *usin = (struct sockaddr_in6 *)uaddr;
753         struct inet_connection_sock *icsk = inet_csk(sk);
754         struct inet_sock *inet = inet_sk(sk);
755         struct ipv6_pinfo *np = inet6_sk(sk);
756         struct dccp_sock *dp = dccp_sk(sk);
757         struct in6_addr *saddr = NULL, *final_p, final;
758         struct flowi6 fl6;
759         struct dst_entry *dst;
760         int addr_type;
761         int err;
762
763         dp->dccps_role = DCCP_ROLE_CLIENT;
764
765         if (addr_len < SIN6_LEN_RFC2133)
766                 return -EINVAL;
767
768         if (usin->sin6_family != AF_INET6)
769                 return -EAFNOSUPPORT;
770
771         memset(&fl6, 0, sizeof(fl6));
772
773         if (np->sndflow) {
774                 fl6.flowlabel = usin->sin6_flowinfo & IPV6_FLOWINFO_MASK;
775                 IP6_ECN_flow_init(fl6.flowlabel);
776                 if (fl6.flowlabel & IPV6_FLOWLABEL_MASK) {
777                         struct ip6_flowlabel *flowlabel;
778                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
779                         if (flowlabel == NULL)
780                                 return -EINVAL;
781                         fl6_sock_release(flowlabel);
782                 }
783         }
784         /*
785          * connect() to INADDR_ANY means loopback (BSD'ism).
786          */
787         if (ipv6_addr_any(&usin->sin6_addr))
788                 usin->sin6_addr.s6_addr[15] = 1;
789
790         addr_type = ipv6_addr_type(&usin->sin6_addr);
791
792         if (addr_type & IPV6_ADDR_MULTICAST)
793                 return -ENETUNREACH;
794
795         if (addr_type & IPV6_ADDR_LINKLOCAL) {
796                 if (addr_len >= sizeof(struct sockaddr_in6) &&
797                     usin->sin6_scope_id) {
798                         /* If interface is set while binding, indices
799                          * must coincide.
800                          */
801                         if (sk->sk_bound_dev_if &&
802                             sk->sk_bound_dev_if != usin->sin6_scope_id)
803                                 return -EINVAL;
804
805                         sk->sk_bound_dev_if = usin->sin6_scope_id;
806                 }
807
808                 /* Connect to link-local address requires an interface */
809                 if (!sk->sk_bound_dev_if)
810                         return -EINVAL;
811         }
812
813         sk->sk_v6_daddr = usin->sin6_addr;
814         np->flow_label = fl6.flowlabel;
815
816         /*
817          * DCCP over IPv4
818          */
819         if (addr_type == IPV6_ADDR_MAPPED) {
820                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
821                 struct sockaddr_in sin;
822
823                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
824
825                 if (__ipv6_only_sock(sk))
826                         return -ENETUNREACH;
827
828                 sin.sin_family = AF_INET;
829                 sin.sin_port = usin->sin6_port;
830                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
831
832                 icsk->icsk_af_ops = &dccp_ipv6_mapped;
833                 sk->sk_backlog_rcv = dccp_v4_do_rcv;
834
835                 err = dccp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
836                 if (err) {
837                         icsk->icsk_ext_hdr_len = exthdrlen;
838                         icsk->icsk_af_ops = &dccp_ipv6_af_ops;
839                         sk->sk_backlog_rcv = dccp_v6_do_rcv;
840                         goto failure;
841                 }
842                 np->saddr = sk->sk_v6_rcv_saddr;
843                 return err;
844         }
845
846         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
847                 saddr = &sk->sk_v6_rcv_saddr;
848
849         fl6.flowi6_proto = IPPROTO_DCCP;
850         fl6.daddr = sk->sk_v6_daddr;
851         fl6.saddr = saddr ? *saddr : np->saddr;
852         fl6.flowi6_oif = sk->sk_bound_dev_if;
853         fl6.fl6_dport = usin->sin6_port;
854         fl6.fl6_sport = inet->inet_sport;
855         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
856
857         final_p = fl6_update_dst(&fl6, np->opt, &final);
858
859         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
860         if (IS_ERR(dst)) {
861                 err = PTR_ERR(dst);
862                 goto failure;
863         }
864
865         if (saddr == NULL) {
866                 saddr = &fl6.saddr;
867                 sk->sk_v6_rcv_saddr = *saddr;
868         }
869
870         /* set the source address */
871         np->saddr = *saddr;
872         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
873
874         __ip6_dst_store(sk, dst, NULL, NULL);
875
876         icsk->icsk_ext_hdr_len = 0;
877         if (np->opt != NULL)
878                 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
879                                           np->opt->opt_nflen);
880
881         inet->inet_dport = usin->sin6_port;
882
883         dccp_set_state(sk, DCCP_REQUESTING);
884         err = inet6_hash_connect(&dccp_death_row, sk);
885         if (err)
886                 goto late_failure;
887
888         dp->dccps_iss = secure_dccpv6_sequence_number(np->saddr.s6_addr32,
889                                                       sk->sk_v6_daddr.s6_addr32,
890                                                       inet->inet_sport,
891                                                       inet->inet_dport);
892         err = dccp_connect(sk);
893         if (err)
894                 goto late_failure;
895
896         return 0;
897
898 late_failure:
899         dccp_set_state(sk, DCCP_CLOSED);
900         __sk_dst_reset(sk);
901 failure:
902         inet->inet_dport = 0;
903         sk->sk_route_caps = 0;
904         return err;
905 }
906
907 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops = {
908         .queue_xmit        = inet6_csk_xmit,
909         .send_check        = dccp_v6_send_check,
910         .rebuild_header    = inet6_sk_rebuild_header,
911         .conn_request      = dccp_v6_conn_request,
912         .syn_recv_sock     = dccp_v6_request_recv_sock,
913         .net_header_len    = sizeof(struct ipv6hdr),
914         .setsockopt        = ipv6_setsockopt,
915         .getsockopt        = ipv6_getsockopt,
916         .addr2sockaddr     = inet6_csk_addr2sockaddr,
917         .sockaddr_len      = sizeof(struct sockaddr_in6),
918         .bind_conflict     = inet6_csk_bind_conflict,
919 #ifdef CONFIG_COMPAT
920         .compat_setsockopt = compat_ipv6_setsockopt,
921         .compat_getsockopt = compat_ipv6_getsockopt,
922 #endif
923 };
924
925 /*
926  *      DCCP over IPv4 via INET6 API
927  */
928 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped = {
929         .queue_xmit        = ip_queue_xmit,
930         .send_check        = dccp_v4_send_check,
931         .rebuild_header    = inet_sk_rebuild_header,
932         .conn_request      = dccp_v6_conn_request,
933         .syn_recv_sock     = dccp_v6_request_recv_sock,
934         .net_header_len    = sizeof(struct iphdr),
935         .setsockopt        = ipv6_setsockopt,
936         .getsockopt        = ipv6_getsockopt,
937         .addr2sockaddr     = inet6_csk_addr2sockaddr,
938         .sockaddr_len      = sizeof(struct sockaddr_in6),
939 #ifdef CONFIG_COMPAT
940         .compat_setsockopt = compat_ipv6_setsockopt,
941         .compat_getsockopt = compat_ipv6_getsockopt,
942 #endif
943 };
944
945 /* NOTE: A lot of things set to zero explicitly by call to
946  *       sk_alloc() so need not be done here.
947  */
948 static int dccp_v6_init_sock(struct sock *sk)
949 {
950         static __u8 dccp_v6_ctl_sock_initialized;
951         int err = dccp_init_sock(sk, dccp_v6_ctl_sock_initialized);
952
953         if (err == 0) {
954                 if (unlikely(!dccp_v6_ctl_sock_initialized))
955                         dccp_v6_ctl_sock_initialized = 1;
956                 inet_csk(sk)->icsk_af_ops = &dccp_ipv6_af_ops;
957         }
958
959         return err;
960 }
961
962 static void dccp_v6_destroy_sock(struct sock *sk)
963 {
964         dccp_destroy_sock(sk);
965         inet6_destroy_sock(sk);
966 }
967
968 static struct timewait_sock_ops dccp6_timewait_sock_ops = {
969         .twsk_obj_size  = sizeof(struct dccp6_timewait_sock),
970 };
971
972 static struct proto dccp_v6_prot = {
973         .name              = "DCCPv6",
974         .owner             = THIS_MODULE,
975         .close             = dccp_close,
976         .connect           = dccp_v6_connect,
977         .disconnect        = dccp_disconnect,
978         .ioctl             = dccp_ioctl,
979         .init              = dccp_v6_init_sock,
980         .setsockopt        = dccp_setsockopt,
981         .getsockopt        = dccp_getsockopt,
982         .sendmsg           = dccp_sendmsg,
983         .recvmsg           = dccp_recvmsg,
984         .backlog_rcv       = dccp_v6_do_rcv,
985         .hash              = inet_hash,
986         .unhash            = inet_unhash,
987         .accept            = inet_csk_accept,
988         .get_port          = inet_csk_get_port,
989         .shutdown          = dccp_shutdown,
990         .destroy           = dccp_v6_destroy_sock,
991         .orphan_count      = &dccp_orphan_count,
992         .max_header        = MAX_DCCP_HEADER,
993         .obj_size          = sizeof(struct dccp6_sock),
994         .slab_flags        = SLAB_DESTROY_BY_RCU,
995         .rsk_prot          = &dccp6_request_sock_ops,
996         .twsk_prot         = &dccp6_timewait_sock_ops,
997         .h.hashinfo        = &dccp_hashinfo,
998 #ifdef CONFIG_COMPAT
999         .compat_setsockopt = compat_dccp_setsockopt,
1000         .compat_getsockopt = compat_dccp_getsockopt,
1001 #endif
1002 };
1003
1004 static const struct inet6_protocol dccp_v6_protocol = {
1005         .handler        = dccp_v6_rcv,
1006         .err_handler    = dccp_v6_err,
1007         .flags          = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1008 };
1009
1010 static const struct proto_ops inet6_dccp_ops = {
1011         .family            = PF_INET6,
1012         .owner             = THIS_MODULE,
1013         .release           = inet6_release,
1014         .bind              = inet6_bind,
1015         .connect           = inet_stream_connect,
1016         .socketpair        = sock_no_socketpair,
1017         .accept            = inet_accept,
1018         .getname           = inet6_getname,
1019         .poll              = dccp_poll,
1020         .ioctl             = inet6_ioctl,
1021         .listen            = inet_dccp_listen,
1022         .shutdown          = inet_shutdown,
1023         .setsockopt        = sock_common_setsockopt,
1024         .getsockopt        = sock_common_getsockopt,
1025         .sendmsg           = inet_sendmsg,
1026         .recvmsg           = sock_common_recvmsg,
1027         .mmap              = sock_no_mmap,
1028         .sendpage          = sock_no_sendpage,
1029 #ifdef CONFIG_COMPAT
1030         .compat_setsockopt = compat_sock_common_setsockopt,
1031         .compat_getsockopt = compat_sock_common_getsockopt,
1032 #endif
1033 };
1034
1035 static struct inet_protosw dccp_v6_protosw = {
1036         .type           = SOCK_DCCP,
1037         .protocol       = IPPROTO_DCCP,
1038         .prot           = &dccp_v6_prot,
1039         .ops            = &inet6_dccp_ops,
1040         .flags          = INET_PROTOSW_ICSK,
1041 };
1042
1043 static int __net_init dccp_v6_init_net(struct net *net)
1044 {
1045         if (dccp_hashinfo.bhash == NULL)
1046                 return -ESOCKTNOSUPPORT;
1047
1048         return inet_ctl_sock_create(&net->dccp.v6_ctl_sk, PF_INET6,
1049                                     SOCK_DCCP, IPPROTO_DCCP, net);
1050 }
1051
1052 static void __net_exit dccp_v6_exit_net(struct net *net)
1053 {
1054         inet_ctl_sock_destroy(net->dccp.v6_ctl_sk);
1055 }
1056
1057 static struct pernet_operations dccp_v6_ops = {
1058         .init   = dccp_v6_init_net,
1059         .exit   = dccp_v6_exit_net,
1060 };
1061
1062 static int __init dccp_v6_init(void)
1063 {
1064         int err = proto_register(&dccp_v6_prot, 1);
1065
1066         if (err != 0)
1067                 goto out;
1068
1069         err = inet6_add_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1070         if (err != 0)
1071                 goto out_unregister_proto;
1072
1073         inet6_register_protosw(&dccp_v6_protosw);
1074
1075         err = register_pernet_subsys(&dccp_v6_ops);
1076         if (err != 0)
1077                 goto out_destroy_ctl_sock;
1078 out:
1079         return err;
1080
1081 out_destroy_ctl_sock:
1082         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1083         inet6_unregister_protosw(&dccp_v6_protosw);
1084 out_unregister_proto:
1085         proto_unregister(&dccp_v6_prot);
1086         goto out;
1087 }
1088
1089 static void __exit dccp_v6_exit(void)
1090 {
1091         unregister_pernet_subsys(&dccp_v6_ops);
1092         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1093         inet6_unregister_protosw(&dccp_v6_protosw);
1094         proto_unregister(&dccp_v6_prot);
1095 }
1096
1097 module_init(dccp_v6_init);
1098 module_exit(dccp_v6_exit);
1099
1100 /*
1101  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1102  * values directly, Also cover the case where the protocol is not specified,
1103  * i.e. net-pf-PF_INET6-proto-0-type-SOCK_DCCP
1104  */
1105 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 33, 6);
1106 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 0, 6);
1107 MODULE_LICENSE("GPL");
1108 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1109 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol");