dccp: limit sk_filter trim to payload
[pandora-kernel.git] / net / dccp / ipv4.c
1 /*
2  *  net/dccp/ipv4.c
3  *
4  *  An implementation of the DCCP protocol
5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
6  *
7  *      This program is free software; you can redistribute it and/or
8  *      modify it under the terms of the GNU General Public License
9  *      as published by the Free Software Foundation; either version
10  *      2 of the License, or (at your option) any later version.
11  */
12
13 #include <linux/dccp.h>
14 #include <linux/icmp.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/random.h>
19
20 #include <net/icmp.h>
21 #include <net/inet_common.h>
22 #include <net/inet_hashtables.h>
23 #include <net/inet_sock.h>
24 #include <net/protocol.h>
25 #include <net/sock.h>
26 #include <net/timewait_sock.h>
27 #include <net/tcp_states.h>
28 #include <net/xfrm.h>
29 #include <net/secure_seq.h>
30
31 #include "ackvec.h"
32 #include "ccid.h"
33 #include "dccp.h"
34 #include "feat.h"
35
36 /*
37  * The per-net dccp.v4_ctl_sk socket is used for responding to
38  * the Out-of-the-blue (OOTB) packets. A control sock will be created
39  * for this socket at the initialization time.
40  */
41
42 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
43 {
44         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
45         struct inet_sock *inet = inet_sk(sk);
46         struct dccp_sock *dp = dccp_sk(sk);
47         __be16 orig_sport, orig_dport;
48         __be32 daddr, nexthop;
49         struct flowi4 *fl4;
50         struct rtable *rt;
51         int err;
52         struct ip_options_rcu *inet_opt;
53
54         dp->dccps_role = DCCP_ROLE_CLIENT;
55
56         if (addr_len < sizeof(struct sockaddr_in))
57                 return -EINVAL;
58
59         if (usin->sin_family != AF_INET)
60                 return -EAFNOSUPPORT;
61
62         nexthop = daddr = usin->sin_addr.s_addr;
63
64         inet_opt = rcu_dereference_protected(inet->inet_opt,
65                                              sock_owned_by_user(sk));
66         if (inet_opt != NULL && inet_opt->opt.srr) {
67                 if (daddr == 0)
68                         return -EINVAL;
69                 nexthop = inet_opt->opt.faddr;
70         }
71
72         orig_sport = inet->inet_sport;
73         orig_dport = usin->sin_port;
74         fl4 = &inet->cork.fl.u.ip4;
75         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
76                               RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
77                               IPPROTO_DCCP,
78                               orig_sport, orig_dport, sk, true);
79         if (IS_ERR(rt))
80                 return PTR_ERR(rt);
81
82         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
83                 ip_rt_put(rt);
84                 return -ENETUNREACH;
85         }
86
87         if (inet_opt == NULL || !inet_opt->opt.srr)
88                 daddr = fl4->daddr;
89
90         if (inet->inet_saddr == 0)
91                 inet->inet_saddr = fl4->saddr;
92         inet->inet_rcv_saddr = inet->inet_saddr;
93
94         inet->inet_dport = usin->sin_port;
95         inet->inet_daddr = daddr;
96
97         inet_csk(sk)->icsk_ext_hdr_len = 0;
98         if (inet_opt)
99                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
100         /*
101          * Socket identity is still unknown (sport may be zero).
102          * However we set state to DCCP_REQUESTING and not releasing socket
103          * lock select source port, enter ourselves into the hash tables and
104          * complete initialization after this.
105          */
106         dccp_set_state(sk, DCCP_REQUESTING);
107         err = inet_hash_connect(&dccp_death_row, sk);
108         if (err != 0)
109                 goto failure;
110
111         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
112                                inet->inet_sport, inet->inet_dport, sk);
113         if (IS_ERR(rt)) {
114                 err = PTR_ERR(rt);
115                 rt = NULL;
116                 goto failure;
117         }
118         /* OK, now commit destination to socket.  */
119         sk_setup_caps(sk, &rt->dst);
120
121         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
122                                                     inet->inet_daddr,
123                                                     inet->inet_sport,
124                                                     inet->inet_dport);
125         inet->inet_id = dp->dccps_iss ^ jiffies;
126
127         err = dccp_connect(sk);
128         rt = NULL;
129         if (err != 0)
130                 goto failure;
131 out:
132         return err;
133 failure:
134         /*
135          * This unhashes the socket and releases the local port, if necessary.
136          */
137         dccp_set_state(sk, DCCP_CLOSED);
138         ip_rt_put(rt);
139         sk->sk_route_caps = 0;
140         inet->inet_dport = 0;
141         goto out;
142 }
143
144 EXPORT_SYMBOL_GPL(dccp_v4_connect);
145
146 /*
147  * This routine does path mtu discovery as defined in RFC1191.
148  */
149 static inline void dccp_do_pmtu_discovery(struct sock *sk,
150                                           const struct iphdr *iph,
151                                           u32 mtu)
152 {
153         struct dst_entry *dst;
154         const struct inet_sock *inet = inet_sk(sk);
155         const struct dccp_sock *dp = dccp_sk(sk);
156
157         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
158          * send out by Linux are always < 576bytes so they should go through
159          * unfragmented).
160          */
161         if (sk->sk_state == DCCP_LISTEN)
162                 return;
163
164         /* We don't check in the destentry if pmtu discovery is forbidden
165          * on this route. We just assume that no packet_to_big packets
166          * are send back when pmtu discovery is not active.
167          * There is a small race when the user changes this flag in the
168          * route, but I think that's acceptable.
169          */
170         if ((dst = __sk_dst_check(sk, 0)) == NULL)
171                 return;
172
173         dst->ops->update_pmtu(dst, mtu);
174
175         /* Something is about to be wrong... Remember soft error
176          * for the case, if this connection will not able to recover.
177          */
178         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
179                 sk->sk_err_soft = EMSGSIZE;
180
181         mtu = dst_mtu(dst);
182
183         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
184             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
185                 dccp_sync_mss(sk, mtu);
186
187                 /*
188                  * From RFC 4340, sec. 14.1:
189                  *
190                  *      DCCP-Sync packets are the best choice for upward
191                  *      probing, since DCCP-Sync probes do not risk application
192                  *      data loss.
193                  */
194                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
195         } /* else let the usual retransmit timer handle it */
196 }
197
198 /*
199  * This routine is called by the ICMP module when it gets some sort of error
200  * condition. If err < 0 then the socket should be closed and the error
201  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
202  * After adjustment header points to the first 8 bytes of the tcp header. We
203  * need to find the appropriate port.
204  *
205  * The locking strategy used here is very "optimistic". When someone else
206  * accesses the socket the ICMP is just dropped and for some paths there is no
207  * check at all. A more general error queue to queue errors for later handling
208  * is probably better.
209  */
210 static void dccp_v4_err(struct sk_buff *skb, u32 info)
211 {
212         const struct iphdr *iph = (struct iphdr *)skb->data;
213         const u8 offset = iph->ihl << 2;
214         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
215         struct dccp_sock *dp;
216         struct inet_sock *inet;
217         const int type = icmp_hdr(skb)->type;
218         const int code = icmp_hdr(skb)->code;
219         struct sock *sk;
220         __u64 seq;
221         int err;
222         struct net *net = dev_net(skb->dev);
223
224         if (skb->len < offset + sizeof(*dh) ||
225             skb->len < offset + __dccp_basic_hdr_len(dh)) {
226                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
227                 return;
228         }
229
230         sk = inet_lookup(net, &dccp_hashinfo,
231                         iph->daddr, dh->dccph_dport,
232                         iph->saddr, dh->dccph_sport, inet_iif(skb));
233         if (sk == NULL) {
234                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
235                 return;
236         }
237
238         if (sk->sk_state == DCCP_TIME_WAIT) {
239                 inet_twsk_put(inet_twsk(sk));
240                 return;
241         }
242
243         bh_lock_sock(sk);
244         /* If too many ICMPs get dropped on busy
245          * servers this needs to be solved differently.
246          */
247         if (sock_owned_by_user(sk))
248                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
249
250         if (sk->sk_state == DCCP_CLOSED)
251                 goto out;
252
253         dp = dccp_sk(sk);
254         seq = dccp_hdr_seq(dh);
255         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
256             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
257                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
258                 goto out;
259         }
260
261         switch (type) {
262         case ICMP_SOURCE_QUENCH:
263                 /* Just silently ignore these. */
264                 goto out;
265         case ICMP_PARAMETERPROB:
266                 err = EPROTO;
267                 break;
268         case ICMP_DEST_UNREACH:
269                 if (code > NR_ICMP_UNREACH)
270                         goto out;
271
272                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
273                         if (!sock_owned_by_user(sk))
274                                 dccp_do_pmtu_discovery(sk, iph, info);
275                         goto out;
276                 }
277
278                 err = icmp_err_convert[code].errno;
279                 break;
280         case ICMP_TIME_EXCEEDED:
281                 err = EHOSTUNREACH;
282                 break;
283         default:
284                 goto out;
285         }
286
287         switch (sk->sk_state) {
288                 struct request_sock *req , **prev;
289         case DCCP_LISTEN:
290                 if (sock_owned_by_user(sk))
291                         goto out;
292                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
293                                           iph->daddr, iph->saddr);
294                 if (!req)
295                         goto out;
296
297                 /*
298                  * ICMPs are not backlogged, hence we cannot get an established
299                  * socket here.
300                  */
301                 WARN_ON(req->sk);
302
303                 if (seq != dccp_rsk(req)->dreq_iss) {
304                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
305                         goto out;
306                 }
307                 /*
308                  * Still in RESPOND, just remove it silently.
309                  * There is no good way to pass the error to the newly
310                  * created socket, and POSIX does not want network
311                  * errors returned from accept().
312                  */
313                 inet_csk_reqsk_queue_drop(sk, req, prev);
314                 goto out;
315
316         case DCCP_REQUESTING:
317         case DCCP_RESPOND:
318                 if (!sock_owned_by_user(sk)) {
319                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
320                         sk->sk_err = err;
321
322                         sk->sk_error_report(sk);
323
324                         dccp_done(sk);
325                 } else
326                         sk->sk_err_soft = err;
327                 goto out;
328         }
329
330         /* If we've already connected we will keep trying
331          * until we time out, or the user gives up.
332          *
333          * rfc1122 4.2.3.9 allows to consider as hard errors
334          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
335          * but it is obsoleted by pmtu discovery).
336          *
337          * Note, that in modern internet, where routing is unreliable
338          * and in each dark corner broken firewalls sit, sending random
339          * errors ordered by their masters even this two messages finally lose
340          * their original sense (even Linux sends invalid PORT_UNREACHs)
341          *
342          * Now we are in compliance with RFCs.
343          *                                                      --ANK (980905)
344          */
345
346         inet = inet_sk(sk);
347         if (!sock_owned_by_user(sk) && inet->recverr) {
348                 sk->sk_err = err;
349                 sk->sk_error_report(sk);
350         } else /* Only an error on timeout */
351                 sk->sk_err_soft = err;
352 out:
353         bh_unlock_sock(sk);
354         sock_put(sk);
355 }
356
357 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
358                                       __be32 src, __be32 dst)
359 {
360         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
361 }
362
363 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
364 {
365         const struct inet_sock *inet = inet_sk(sk);
366         struct dccp_hdr *dh = dccp_hdr(skb);
367
368         dccp_csum_outgoing(skb);
369         dh->dccph_checksum = dccp_v4_csum_finish(skb,
370                                                  inet->inet_saddr,
371                                                  inet->inet_daddr);
372 }
373
374 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
375
376 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
377 {
378         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
379                                            ip_hdr(skb)->saddr,
380                                            dccp_hdr(skb)->dccph_dport,
381                                            dccp_hdr(skb)->dccph_sport);
382 }
383
384 /*
385  * The three way handshake has completed - we got a valid ACK or DATAACK -
386  * now create the new socket.
387  *
388  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
389  */
390 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
391                                        struct request_sock *req,
392                                        struct dst_entry *dst)
393 {
394         struct inet_request_sock *ireq;
395         struct inet_sock *newinet;
396         struct sock *newsk;
397
398         if (sk_acceptq_is_full(sk))
399                 goto exit_overflow;
400
401         newsk = dccp_create_openreq_child(sk, req, skb);
402         if (newsk == NULL)
403                 goto exit_nonewsk;
404
405         newinet            = inet_sk(newsk);
406         ireq               = inet_rsk(req);
407         newinet->inet_daddr     = ireq->rmt_addr;
408         newinet->inet_rcv_saddr = ireq->loc_addr;
409         newinet->inet_saddr     = ireq->loc_addr;
410         newinet->inet_opt       = ireq->opt;
411         ireq->opt          = NULL;
412         newinet->mc_index  = inet_iif(skb);
413         newinet->mc_ttl    = ip_hdr(skb)->ttl;
414         newinet->inet_id   = jiffies;
415
416         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
417                 goto put_and_exit;
418
419         sk_setup_caps(newsk, dst);
420
421         dccp_sync_mss(newsk, dst_mtu(dst));
422
423         if (__inet_inherit_port(sk, newsk) < 0)
424                 goto put_and_exit;
425         __inet_hash_nolisten(newsk, NULL);
426
427         return newsk;
428
429 exit_overflow:
430         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
431 exit_nonewsk:
432         dst_release(dst);
433 exit:
434         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
435         return NULL;
436 put_and_exit:
437         inet_csk_prepare_forced_close(newsk);
438         dccp_done(newsk);
439         goto exit;
440 }
441
442 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
443
444 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
445 {
446         const struct dccp_hdr *dh = dccp_hdr(skb);
447         const struct iphdr *iph = ip_hdr(skb);
448         struct sock *nsk;
449         struct request_sock **prev;
450         /* Find possible connection requests. */
451         struct request_sock *req = inet_csk_search_req(sk, &prev,
452                                                        dh->dccph_sport,
453                                                        iph->saddr, iph->daddr);
454         if (req != NULL)
455                 return dccp_check_req(sk, skb, req, prev);
456
457         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
458                                       iph->saddr, dh->dccph_sport,
459                                       iph->daddr, dh->dccph_dport,
460                                       inet_iif(skb));
461         if (nsk != NULL) {
462                 if (nsk->sk_state != DCCP_TIME_WAIT) {
463                         bh_lock_sock(nsk);
464                         return nsk;
465                 }
466                 inet_twsk_put(inet_twsk(nsk));
467                 return NULL;
468         }
469
470         return sk;
471 }
472
473 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
474                                            struct sk_buff *skb)
475 {
476         struct rtable *rt;
477         struct flowi4 fl4 = {
478                 .flowi4_oif = skb_rtable(skb)->rt_iif,
479                 .daddr = ip_hdr(skb)->saddr,
480                 .saddr = ip_hdr(skb)->daddr,
481                 .flowi4_tos = RT_CONN_FLAGS(sk),
482                 .flowi4_proto = sk->sk_protocol,
483                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
484                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
485         };
486
487         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
488         rt = ip_route_output_flow(net, &fl4, sk);
489         if (IS_ERR(rt)) {
490                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
491                 return NULL;
492         }
493
494         return &rt->dst;
495 }
496
497 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
498                                  struct request_values *rv_unused)
499 {
500         int err = -1;
501         struct sk_buff *skb;
502         struct dst_entry *dst;
503         struct flowi4 fl4;
504
505         dst = inet_csk_route_req(sk, &fl4, req);
506         if (dst == NULL)
507                 goto out;
508
509         skb = dccp_make_response(sk, dst, req);
510         if (skb != NULL) {
511                 const struct inet_request_sock *ireq = inet_rsk(req);
512                 struct dccp_hdr *dh = dccp_hdr(skb);
513
514                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
515                                                               ireq->rmt_addr);
516                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
517                                             ireq->rmt_addr,
518                                             ireq->opt);
519                 err = net_xmit_eval(err);
520         }
521
522 out:
523         dst_release(dst);
524         return err;
525 }
526
527 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
528 {
529         int err;
530         const struct iphdr *rxiph;
531         struct sk_buff *skb;
532         struct dst_entry *dst;
533         struct net *net = dev_net(skb_dst(rxskb)->dev);
534         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
535
536         /* Never send a reset in response to a reset. */
537         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
538                 return;
539
540         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
541                 return;
542
543         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
544         if (dst == NULL)
545                 return;
546
547         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
548         if (skb == NULL)
549                 goto out;
550
551         rxiph = ip_hdr(rxskb);
552         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
553                                                                  rxiph->daddr);
554         skb_dst_set(skb, dst_clone(dst));
555
556         bh_lock_sock(ctl_sk);
557         err = ip_build_and_send_pkt(skb, ctl_sk,
558                                     rxiph->daddr, rxiph->saddr, NULL);
559         bh_unlock_sock(ctl_sk);
560
561         if (net_xmit_eval(err) == 0) {
562                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
563                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
564         }
565 out:
566          dst_release(dst);
567 }
568
569 static void dccp_v4_reqsk_destructor(struct request_sock *req)
570 {
571         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
572         kfree(inet_rsk(req)->opt);
573 }
574
575 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
576         .family         = PF_INET,
577         .obj_size       = sizeof(struct dccp_request_sock),
578         .rtx_syn_ack    = dccp_v4_send_response,
579         .send_ack       = dccp_reqsk_send_ack,
580         .destructor     = dccp_v4_reqsk_destructor,
581         .send_reset     = dccp_v4_ctl_send_reset,
582 };
583
584 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
585 {
586         struct inet_request_sock *ireq;
587         struct request_sock *req;
588         struct dccp_request_sock *dreq;
589         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
590         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
591
592         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
593         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
594                 return 0;       /* discard, don't send a reset here */
595
596         if (dccp_bad_service_code(sk, service)) {
597                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
598                 goto drop;
599         }
600         /*
601          * TW buckets are converted to open requests without
602          * limitations, they conserve resources and peer is
603          * evidently real one.
604          */
605         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
606         if (inet_csk_reqsk_queue_is_full(sk))
607                 goto drop;
608
609         /*
610          * Accept backlog is full. If we have already queued enough
611          * of warm entries in syn queue, drop request. It is better than
612          * clogging syn queue with openreqs with exponentially increasing
613          * timeout.
614          */
615         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
616                 goto drop;
617
618         req = inet_reqsk_alloc(&dccp_request_sock_ops);
619         if (req == NULL)
620                 goto drop;
621
622         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
623                 goto drop_and_free;
624
625         dreq = dccp_rsk(req);
626         if (dccp_parse_options(sk, dreq, skb))
627                 goto drop_and_free;
628
629         if (security_inet_conn_request(sk, skb, req))
630                 goto drop_and_free;
631
632         ireq = inet_rsk(req);
633         ireq->loc_addr = ip_hdr(skb)->daddr;
634         ireq->rmt_addr = ip_hdr(skb)->saddr;
635
636         /*
637          * Step 3: Process LISTEN state
638          *
639          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
640          *
641          * In fact we defer setting S.GSR, S.SWL, S.SWH to
642          * dccp_create_openreq_child.
643          */
644         dreq->dreq_isr     = dcb->dccpd_seq;
645         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
646         dreq->dreq_service = service;
647
648         if (dccp_v4_send_response(sk, req, NULL))
649                 goto drop_and_free;
650
651         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
652         return 0;
653
654 drop_and_free:
655         reqsk_free(req);
656 drop:
657         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
658         return -1;
659 }
660
661 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
662
663 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
664 {
665         struct dccp_hdr *dh = dccp_hdr(skb);
666
667         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
668                 if (dccp_rcv_established(sk, skb, dh, skb->len))
669                         goto reset;
670                 return 0;
671         }
672
673         /*
674          *  Step 3: Process LISTEN state
675          *       If P.type == Request or P contains a valid Init Cookie option,
676          *            (* Must scan the packet's options to check for Init
677          *               Cookies.  Only Init Cookies are processed here,
678          *               however; other options are processed in Step 8.  This
679          *               scan need only be performed if the endpoint uses Init
680          *               Cookies *)
681          *            (* Generate a new socket and switch to that socket *)
682          *            Set S := new socket for this port pair
683          *            S.state = RESPOND
684          *            Choose S.ISS (initial seqno) or set from Init Cookies
685          *            Initialize S.GAR := S.ISS
686          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
687          *            Continue with S.state == RESPOND
688          *            (* A Response packet will be generated in Step 11 *)
689          *       Otherwise,
690          *            Generate Reset(No Connection) unless P.type == Reset
691          *            Drop packet and return
692          *
693          * NOTE: the check for the packet types is done in
694          *       dccp_rcv_state_process
695          */
696         if (sk->sk_state == DCCP_LISTEN) {
697                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
698
699                 if (nsk == NULL)
700                         goto discard;
701
702                 if (nsk != sk) {
703                         if (dccp_child_process(sk, nsk, skb))
704                                 goto reset;
705                         return 0;
706                 }
707         }
708
709         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
710                 goto reset;
711         return 0;
712
713 reset:
714         dccp_v4_ctl_send_reset(sk, skb);
715 discard:
716         kfree_skb(skb);
717         return 0;
718 }
719
720 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
721
722 /**
723  *      dccp_invalid_packet  -  check for malformed packets
724  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
725  *      Packets that fail these checks are ignored and do not receive Resets.
726  */
727 int dccp_invalid_packet(struct sk_buff *skb)
728 {
729         const struct dccp_hdr *dh;
730         unsigned int cscov;
731
732         if (skb->pkt_type != PACKET_HOST)
733                 return 1;
734
735         /* If the packet is shorter than 12 bytes, drop packet and return */
736         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
737                 DCCP_WARN("pskb_may_pull failed\n");
738                 return 1;
739         }
740
741         dh = dccp_hdr(skb);
742
743         /* If P.type is not understood, drop packet and return */
744         if (dh->dccph_type >= DCCP_PKT_INVALID) {
745                 DCCP_WARN("invalid packet type\n");
746                 return 1;
747         }
748
749         /*
750          * If P.Data Offset is too small for packet type, drop packet and return
751          */
752         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
753                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
754                 return 1;
755         }
756         /*
757          * If P.Data Offset is too too large for packet, drop packet and return
758          */
759         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
760                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
761                 return 1;
762         }
763
764         /*
765          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
766          * has short sequence numbers), drop packet and return
767          */
768         if ((dh->dccph_type < DCCP_PKT_DATA    ||
769             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
770                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
771                           dccp_packet_name(dh->dccph_type));
772                 return 1;
773         }
774
775         /*
776          * If P.CsCov is too large for the packet size, drop packet and return.
777          * This must come _before_ checksumming (not as RFC 4340 suggests).
778          */
779         cscov = dccp_csum_coverage(skb);
780         if (cscov > skb->len) {
781                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
782                           dh->dccph_cscov, skb->len);
783                 return 1;
784         }
785
786         /* If header checksum is incorrect, drop packet and return.
787          * (This step is completed in the AF-dependent functions.) */
788         skb->csum = skb_checksum(skb, 0, cscov, 0);
789
790         return 0;
791 }
792
793 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
794
795 /* this is called when real data arrives */
796 static int dccp_v4_rcv(struct sk_buff *skb)
797 {
798         const struct dccp_hdr *dh;
799         const struct iphdr *iph;
800         struct sock *sk;
801         int min_cov;
802
803         /* Step 1: Check header basics */
804
805         if (dccp_invalid_packet(skb))
806                 goto discard_it;
807
808         iph = ip_hdr(skb);
809         /* Step 1: If header checksum is incorrect, drop packet and return */
810         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
811                 DCCP_WARN("dropped packet with invalid checksum\n");
812                 goto discard_it;
813         }
814
815         dh = dccp_hdr(skb);
816
817         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
818         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
819
820         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
821                       dccp_packet_name(dh->dccph_type),
822                       &iph->saddr, ntohs(dh->dccph_sport),
823                       &iph->daddr, ntohs(dh->dccph_dport),
824                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
825
826         if (dccp_packet_without_ack(skb)) {
827                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
828                 dccp_pr_debug_cat("\n");
829         } else {
830                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
831                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
832                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
833         }
834
835         /* Step 2:
836          *      Look up flow ID in table and get corresponding socket */
837         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
838                                dh->dccph_sport, dh->dccph_dport);
839         /*
840          * Step 2:
841          *      If no socket ...
842          */
843         if (sk == NULL) {
844                 dccp_pr_debug("failed to look up flow ID in table and "
845                               "get corresponding socket\n");
846                 goto no_dccp_socket;
847         }
848
849         /*
850          * Step 2:
851          *      ... or S.state == TIMEWAIT,
852          *              Generate Reset(No Connection) unless P.type == Reset
853          *              Drop packet and return
854          */
855         if (sk->sk_state == DCCP_TIME_WAIT) {
856                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
857                 inet_twsk_put(inet_twsk(sk));
858                 goto no_dccp_socket;
859         }
860
861         /*
862          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
863          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
864          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
865          */
866         min_cov = dccp_sk(sk)->dccps_pcrlen;
867         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
868                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
869                               dh->dccph_cscov, min_cov);
870                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
871                  *         options (Section 11.7) with Drop Code 0, Protocol
872                  *         Constraints."                                     */
873                 goto discard_and_relse;
874         }
875
876         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
877                 goto discard_and_relse;
878         nf_reset(skb);
879
880         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4);
881
882 no_dccp_socket:
883         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
884                 goto discard_it;
885         /*
886          * Step 2:
887          *      If no socket ...
888          *              Generate Reset(No Connection) unless P.type == Reset
889          *              Drop packet and return
890          */
891         if (dh->dccph_type != DCCP_PKT_RESET) {
892                 DCCP_SKB_CB(skb)->dccpd_reset_code =
893                                         DCCP_RESET_CODE_NO_CONNECTION;
894                 dccp_v4_ctl_send_reset(sk, skb);
895         }
896
897 discard_it:
898         kfree_skb(skb);
899         return 0;
900
901 discard_and_relse:
902         sock_put(sk);
903         goto discard_it;
904 }
905
906 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
907         .queue_xmit        = ip_queue_xmit,
908         .send_check        = dccp_v4_send_check,
909         .rebuild_header    = inet_sk_rebuild_header,
910         .conn_request      = dccp_v4_conn_request,
911         .syn_recv_sock     = dccp_v4_request_recv_sock,
912         .net_header_len    = sizeof(struct iphdr),
913         .setsockopt        = ip_setsockopt,
914         .getsockopt        = ip_getsockopt,
915         .addr2sockaddr     = inet_csk_addr2sockaddr,
916         .sockaddr_len      = sizeof(struct sockaddr_in),
917         .bind_conflict     = inet_csk_bind_conflict,
918 #ifdef CONFIG_COMPAT
919         .compat_setsockopt = compat_ip_setsockopt,
920         .compat_getsockopt = compat_ip_getsockopt,
921 #endif
922 };
923
924 static int dccp_v4_init_sock(struct sock *sk)
925 {
926         static __u8 dccp_v4_ctl_sock_initialized;
927         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
928
929         if (err == 0) {
930                 if (unlikely(!dccp_v4_ctl_sock_initialized))
931                         dccp_v4_ctl_sock_initialized = 1;
932                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
933         }
934
935         return err;
936 }
937
938 static struct timewait_sock_ops dccp_timewait_sock_ops = {
939         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
940 };
941
942 static struct proto dccp_v4_prot = {
943         .name                   = "DCCP",
944         .owner                  = THIS_MODULE,
945         .close                  = dccp_close,
946         .connect                = dccp_v4_connect,
947         .disconnect             = dccp_disconnect,
948         .ioctl                  = dccp_ioctl,
949         .init                   = dccp_v4_init_sock,
950         .setsockopt             = dccp_setsockopt,
951         .getsockopt             = dccp_getsockopt,
952         .sendmsg                = dccp_sendmsg,
953         .recvmsg                = dccp_recvmsg,
954         .backlog_rcv            = dccp_v4_do_rcv,
955         .hash                   = inet_hash,
956         .unhash                 = inet_unhash,
957         .accept                 = inet_csk_accept,
958         .get_port               = inet_csk_get_port,
959         .shutdown               = dccp_shutdown,
960         .destroy                = dccp_destroy_sock,
961         .orphan_count           = &dccp_orphan_count,
962         .max_header             = MAX_DCCP_HEADER,
963         .obj_size               = sizeof(struct dccp_sock),
964         .slab_flags             = SLAB_DESTROY_BY_RCU,
965         .rsk_prot               = &dccp_request_sock_ops,
966         .twsk_prot              = &dccp_timewait_sock_ops,
967         .h.hashinfo             = &dccp_hashinfo,
968 #ifdef CONFIG_COMPAT
969         .compat_setsockopt      = compat_dccp_setsockopt,
970         .compat_getsockopt      = compat_dccp_getsockopt,
971 #endif
972 };
973
974 static const struct net_protocol dccp_v4_protocol = {
975         .handler        = dccp_v4_rcv,
976         .err_handler    = dccp_v4_err,
977         .no_policy      = 1,
978         .netns_ok       = 1,
979 };
980
981 static const struct proto_ops inet_dccp_ops = {
982         .family            = PF_INET,
983         .owner             = THIS_MODULE,
984         .release           = inet_release,
985         .bind              = inet_bind,
986         .connect           = inet_stream_connect,
987         .socketpair        = sock_no_socketpair,
988         .accept            = inet_accept,
989         .getname           = inet_getname,
990         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
991         .poll              = dccp_poll,
992         .ioctl             = inet_ioctl,
993         /* FIXME: work on inet_listen to rename it to sock_common_listen */
994         .listen            = inet_dccp_listen,
995         .shutdown          = inet_shutdown,
996         .setsockopt        = sock_common_setsockopt,
997         .getsockopt        = sock_common_getsockopt,
998         .sendmsg           = inet_sendmsg,
999         .recvmsg           = sock_common_recvmsg,
1000         .mmap              = sock_no_mmap,
1001         .sendpage          = sock_no_sendpage,
1002 #ifdef CONFIG_COMPAT
1003         .compat_setsockopt = compat_sock_common_setsockopt,
1004         .compat_getsockopt = compat_sock_common_getsockopt,
1005 #endif
1006 };
1007
1008 static struct inet_protosw dccp_v4_protosw = {
1009         .type           = SOCK_DCCP,
1010         .protocol       = IPPROTO_DCCP,
1011         .prot           = &dccp_v4_prot,
1012         .ops            = &inet_dccp_ops,
1013         .no_check       = 0,
1014         .flags          = INET_PROTOSW_ICSK,
1015 };
1016
1017 static int __net_init dccp_v4_init_net(struct net *net)
1018 {
1019         if (dccp_hashinfo.bhash == NULL)
1020                 return -ESOCKTNOSUPPORT;
1021
1022         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1023                                     SOCK_DCCP, IPPROTO_DCCP, net);
1024 }
1025
1026 static void __net_exit dccp_v4_exit_net(struct net *net)
1027 {
1028         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1029 }
1030
1031 static struct pernet_operations dccp_v4_ops = {
1032         .init   = dccp_v4_init_net,
1033         .exit   = dccp_v4_exit_net,
1034 };
1035
1036 static int __init dccp_v4_init(void)
1037 {
1038         int err = proto_register(&dccp_v4_prot, 1);
1039
1040         if (err != 0)
1041                 goto out;
1042
1043         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1044         if (err != 0)
1045                 goto out_proto_unregister;
1046
1047         inet_register_protosw(&dccp_v4_protosw);
1048
1049         err = register_pernet_subsys(&dccp_v4_ops);
1050         if (err)
1051                 goto out_destroy_ctl_sock;
1052 out:
1053         return err;
1054 out_destroy_ctl_sock:
1055         inet_unregister_protosw(&dccp_v4_protosw);
1056         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1057 out_proto_unregister:
1058         proto_unregister(&dccp_v4_prot);
1059         goto out;
1060 }
1061
1062 static void __exit dccp_v4_exit(void)
1063 {
1064         unregister_pernet_subsys(&dccp_v4_ops);
1065         inet_unregister_protosw(&dccp_v4_protosw);
1066         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1067         proto_unregister(&dccp_v4_prot);
1068 }
1069
1070 module_init(dccp_v4_init);
1071 module_exit(dccp_v4_exit);
1072
1073 /*
1074  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1075  * values directly, Also cover the case where the protocol is not specified,
1076  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1077  */
1078 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1079 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1080 MODULE_LICENSE("GPL");
1081 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1082 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");