Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[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
30 #include "ackvec.h"
31 #include "ccid.h"
32 #include "dccp.h"
33 #include "feat.h"
34
35 /*
36  * The per-net dccp.v4_ctl_sk socket is used for responding to
37  * the Out-of-the-blue (OOTB) packets. A control sock will be created
38  * for this socket at the initialization time.
39  */
40
41 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
42 {
43         struct inet_sock *inet = inet_sk(sk);
44         struct dccp_sock *dp = dccp_sk(sk);
45         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
46         struct rtable *rt;
47         __be32 daddr, nexthop;
48         int tmp;
49         int err;
50
51         dp->dccps_role = DCCP_ROLE_CLIENT;
52
53         if (addr_len < sizeof(struct sockaddr_in))
54                 return -EINVAL;
55
56         if (usin->sin_family != AF_INET)
57                 return -EAFNOSUPPORT;
58
59         nexthop = daddr = usin->sin_addr.s_addr;
60         if (inet->opt != NULL && inet->opt->srr) {
61                 if (daddr == 0)
62                         return -EINVAL;
63                 nexthop = inet->opt->faddr;
64         }
65
66         tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr,
67                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
68                                IPPROTO_DCCP,
69                                inet->inet_sport, usin->sin_port, sk, 1);
70         if (tmp < 0)
71                 return tmp;
72
73         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
74                 ip_rt_put(rt);
75                 return -ENETUNREACH;
76         }
77
78         if (inet->opt == NULL || !inet->opt->srr)
79                 daddr = rt->rt_dst;
80
81         if (inet->inet_saddr == 0)
82                 inet->inet_saddr = rt->rt_src;
83         inet->inet_rcv_saddr = inet->inet_saddr;
84
85         inet->inet_dport = usin->sin_port;
86         inet->inet_daddr = daddr;
87
88         inet_csk(sk)->icsk_ext_hdr_len = 0;
89         if (inet->opt != NULL)
90                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
91         /*
92          * Socket identity is still unknown (sport may be zero).
93          * However we set state to DCCP_REQUESTING and not releasing socket
94          * lock select source port, enter ourselves into the hash tables and
95          * complete initialization after this.
96          */
97         dccp_set_state(sk, DCCP_REQUESTING);
98         err = inet_hash_connect(&dccp_death_row, sk);
99         if (err != 0)
100                 goto failure;
101
102         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->inet_sport,
103                                 inet->inet_dport, sk);
104         if (err != 0)
105                 goto failure;
106
107         /* OK, now commit destination to socket.  */
108         sk_setup_caps(sk, &rt->dst);
109
110         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
111                                                     inet->inet_daddr,
112                                                     inet->inet_sport,
113                                                     inet->inet_dport);
114         inet->inet_id = dp->dccps_iss ^ jiffies;
115
116         err = dccp_connect(sk);
117         rt = NULL;
118         if (err != 0)
119                 goto failure;
120 out:
121         return err;
122 failure:
123         /*
124          * This unhashes the socket and releases the local port, if necessary.
125          */
126         dccp_set_state(sk, DCCP_CLOSED);
127         ip_rt_put(rt);
128         sk->sk_route_caps = 0;
129         inet->inet_dport = 0;
130         goto out;
131 }
132
133 EXPORT_SYMBOL_GPL(dccp_v4_connect);
134
135 /*
136  * This routine does path mtu discovery as defined in RFC1191.
137  */
138 static inline void dccp_do_pmtu_discovery(struct sock *sk,
139                                           const struct iphdr *iph,
140                                           u32 mtu)
141 {
142         struct dst_entry *dst;
143         const struct inet_sock *inet = inet_sk(sk);
144         const struct dccp_sock *dp = dccp_sk(sk);
145
146         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
147          * send out by Linux are always < 576bytes so they should go through
148          * unfragmented).
149          */
150         if (sk->sk_state == DCCP_LISTEN)
151                 return;
152
153         /* We don't check in the destentry if pmtu discovery is forbidden
154          * on this route. We just assume that no packet_to_big packets
155          * are send back when pmtu discovery is not active.
156          * There is a small race when the user changes this flag in the
157          * route, but I think that's acceptable.
158          */
159         if ((dst = __sk_dst_check(sk, 0)) == NULL)
160                 return;
161
162         dst->ops->update_pmtu(dst, mtu);
163
164         /* Something is about to be wrong... Remember soft error
165          * for the case, if this connection will not able to recover.
166          */
167         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
168                 sk->sk_err_soft = EMSGSIZE;
169
170         mtu = dst_mtu(dst);
171
172         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
173             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
174                 dccp_sync_mss(sk, mtu);
175
176                 /*
177                  * From RFC 4340, sec. 14.1:
178                  *
179                  *      DCCP-Sync packets are the best choice for upward
180                  *      probing, since DCCP-Sync probes do not risk application
181                  *      data loss.
182                  */
183                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
184         } /* else let the usual retransmit timer handle it */
185 }
186
187 /*
188  * This routine is called by the ICMP module when it gets some sort of error
189  * condition. If err < 0 then the socket should be closed and the error
190  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
191  * After adjustment header points to the first 8 bytes of the tcp header. We
192  * need to find the appropriate port.
193  *
194  * The locking strategy used here is very "optimistic". When someone else
195  * accesses the socket the ICMP is just dropped and for some paths there is no
196  * check at all. A more general error queue to queue errors for later handling
197  * is probably better.
198  */
199 static void dccp_v4_err(struct sk_buff *skb, u32 info)
200 {
201         const struct iphdr *iph = (struct iphdr *)skb->data;
202         const u8 offset = iph->ihl << 2;
203         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
204         struct dccp_sock *dp;
205         struct inet_sock *inet;
206         const int type = icmp_hdr(skb)->type;
207         const int code = icmp_hdr(skb)->code;
208         struct sock *sk;
209         __u64 seq;
210         int err;
211         struct net *net = dev_net(skb->dev);
212
213         if (skb->len < offset + sizeof(*dh) ||
214             skb->len < offset + __dccp_basic_hdr_len(dh)) {
215                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
216                 return;
217         }
218
219         sk = inet_lookup(net, &dccp_hashinfo,
220                         iph->daddr, dh->dccph_dport,
221                         iph->saddr, dh->dccph_sport, inet_iif(skb));
222         if (sk == NULL) {
223                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
224                 return;
225         }
226
227         if (sk->sk_state == DCCP_TIME_WAIT) {
228                 inet_twsk_put(inet_twsk(sk));
229                 return;
230         }
231
232         bh_lock_sock(sk);
233         /* If too many ICMPs get dropped on busy
234          * servers this needs to be solved differently.
235          */
236         if (sock_owned_by_user(sk))
237                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
238
239         if (sk->sk_state == DCCP_CLOSED)
240                 goto out;
241
242         dp = dccp_sk(sk);
243         seq = dccp_hdr_seq(dh);
244         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
245             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
246                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
247                 goto out;
248         }
249
250         switch (type) {
251         case ICMP_SOURCE_QUENCH:
252                 /* Just silently ignore these. */
253                 goto out;
254         case ICMP_PARAMETERPROB:
255                 err = EPROTO;
256                 break;
257         case ICMP_DEST_UNREACH:
258                 if (code > NR_ICMP_UNREACH)
259                         goto out;
260
261                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
262                         if (!sock_owned_by_user(sk))
263                                 dccp_do_pmtu_discovery(sk, iph, info);
264                         goto out;
265                 }
266
267                 err = icmp_err_convert[code].errno;
268                 break;
269         case ICMP_TIME_EXCEEDED:
270                 err = EHOSTUNREACH;
271                 break;
272         default:
273                 goto out;
274         }
275
276         switch (sk->sk_state) {
277                 struct request_sock *req , **prev;
278         case DCCP_LISTEN:
279                 if (sock_owned_by_user(sk))
280                         goto out;
281                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
282                                           iph->daddr, iph->saddr);
283                 if (!req)
284                         goto out;
285
286                 /*
287                  * ICMPs are not backlogged, hence we cannot get an established
288                  * socket here.
289                  */
290                 WARN_ON(req->sk);
291
292                 if (seq != dccp_rsk(req)->dreq_iss) {
293                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
294                         goto out;
295                 }
296                 /*
297                  * Still in RESPOND, just remove it silently.
298                  * There is no good way to pass the error to the newly
299                  * created socket, and POSIX does not want network
300                  * errors returned from accept().
301                  */
302                 inet_csk_reqsk_queue_drop(sk, req, prev);
303                 goto out;
304
305         case DCCP_REQUESTING:
306         case DCCP_RESPOND:
307                 if (!sock_owned_by_user(sk)) {
308                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
309                         sk->sk_err = err;
310
311                         sk->sk_error_report(sk);
312
313                         dccp_done(sk);
314                 } else
315                         sk->sk_err_soft = err;
316                 goto out;
317         }
318
319         /* If we've already connected we will keep trying
320          * until we time out, or the user gives up.
321          *
322          * rfc1122 4.2.3.9 allows to consider as hard errors
323          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
324          * but it is obsoleted by pmtu discovery).
325          *
326          * Note, that in modern internet, where routing is unreliable
327          * and in each dark corner broken firewalls sit, sending random
328          * errors ordered by their masters even this two messages finally lose
329          * their original sense (even Linux sends invalid PORT_UNREACHs)
330          *
331          * Now we are in compliance with RFCs.
332          *                                                      --ANK (980905)
333          */
334
335         inet = inet_sk(sk);
336         if (!sock_owned_by_user(sk) && inet->recverr) {
337                 sk->sk_err = err;
338                 sk->sk_error_report(sk);
339         } else /* Only an error on timeout */
340                 sk->sk_err_soft = err;
341 out:
342         bh_unlock_sock(sk);
343         sock_put(sk);
344 }
345
346 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
347                                       __be32 src, __be32 dst)
348 {
349         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
350 }
351
352 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
353 {
354         const struct inet_sock *inet = inet_sk(sk);
355         struct dccp_hdr *dh = dccp_hdr(skb);
356
357         dccp_csum_outgoing(skb);
358         dh->dccph_checksum = dccp_v4_csum_finish(skb,
359                                                  inet->inet_saddr,
360                                                  inet->inet_daddr);
361 }
362
363 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
364
365 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
366 {
367         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
368                                            ip_hdr(skb)->saddr,
369                                            dccp_hdr(skb)->dccph_dport,
370                                            dccp_hdr(skb)->dccph_sport);
371 }
372
373 /*
374  * The three way handshake has completed - we got a valid ACK or DATAACK -
375  * now create the new socket.
376  *
377  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
378  */
379 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
380                                        struct request_sock *req,
381                                        struct dst_entry *dst)
382 {
383         struct inet_request_sock *ireq;
384         struct inet_sock *newinet;
385         struct sock *newsk;
386
387         if (sk_acceptq_is_full(sk))
388                 goto exit_overflow;
389
390         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
391                 goto exit;
392
393         newsk = dccp_create_openreq_child(sk, req, skb);
394         if (newsk == NULL)
395                 goto exit_nonewsk;
396
397         sk_setup_caps(newsk, dst);
398
399         newinet            = inet_sk(newsk);
400         ireq               = inet_rsk(req);
401         newinet->inet_daddr     = ireq->rmt_addr;
402         newinet->inet_rcv_saddr = ireq->loc_addr;
403         newinet->inet_saddr     = ireq->loc_addr;
404         newinet->opt       = ireq->opt;
405         ireq->opt          = NULL;
406         newinet->mc_index  = inet_iif(skb);
407         newinet->mc_ttl    = ip_hdr(skb)->ttl;
408         newinet->inet_id   = jiffies;
409
410         dccp_sync_mss(newsk, dst_mtu(dst));
411
412         if (__inet_inherit_port(sk, newsk) < 0) {
413                 sock_put(newsk);
414                 goto exit;
415         }
416         __inet_hash_nolisten(newsk, NULL);
417
418         return newsk;
419
420 exit_overflow:
421         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
422 exit_nonewsk:
423         dst_release(dst);
424 exit:
425         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
426         return NULL;
427 }
428
429 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
430
431 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
432 {
433         const struct dccp_hdr *dh = dccp_hdr(skb);
434         const struct iphdr *iph = ip_hdr(skb);
435         struct sock *nsk;
436         struct request_sock **prev;
437         /* Find possible connection requests. */
438         struct request_sock *req = inet_csk_search_req(sk, &prev,
439                                                        dh->dccph_sport,
440                                                        iph->saddr, iph->daddr);
441         if (req != NULL)
442                 return dccp_check_req(sk, skb, req, prev);
443
444         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
445                                       iph->saddr, dh->dccph_sport,
446                                       iph->daddr, dh->dccph_dport,
447                                       inet_iif(skb));
448         if (nsk != NULL) {
449                 if (nsk->sk_state != DCCP_TIME_WAIT) {
450                         bh_lock_sock(nsk);
451                         return nsk;
452                 }
453                 inet_twsk_put(inet_twsk(nsk));
454                 return NULL;
455         }
456
457         return sk;
458 }
459
460 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
461                                            struct sk_buff *skb)
462 {
463         struct rtable *rt;
464         struct flowi fl = { .oif = skb_rtable(skb)->rt_iif,
465                             .fl4_dst = ip_hdr(skb)->saddr,
466                             .fl4_src = ip_hdr(skb)->daddr,
467                             .fl4_tos = RT_CONN_FLAGS(sk),
468                             .proto = sk->sk_protocol,
469                             .fl_ip_sport = dccp_hdr(skb)->dccph_dport,
470                             .fl_ip_dport = dccp_hdr(skb)->dccph_sport
471                           };
472
473         security_skb_classify_flow(skb, &fl);
474         if (ip_route_output_flow(net, &rt, &fl, sk, 0)) {
475                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
476                 return NULL;
477         }
478
479         return &rt->dst;
480 }
481
482 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
483                                  struct request_values *rv_unused)
484 {
485         int err = -1;
486         struct sk_buff *skb;
487         struct dst_entry *dst;
488
489         dst = inet_csk_route_req(sk, req);
490         if (dst == NULL)
491                 goto out;
492
493         skb = dccp_make_response(sk, dst, req);
494         if (skb != NULL) {
495                 const struct inet_request_sock *ireq = inet_rsk(req);
496                 struct dccp_hdr *dh = dccp_hdr(skb);
497
498                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
499                                                               ireq->rmt_addr);
500                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
501                                             ireq->rmt_addr,
502                                             ireq->opt);
503                 err = net_xmit_eval(err);
504         }
505
506 out:
507         dst_release(dst);
508         return err;
509 }
510
511 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
512 {
513         int err;
514         const struct iphdr *rxiph;
515         struct sk_buff *skb;
516         struct dst_entry *dst;
517         struct net *net = dev_net(skb_dst(rxskb)->dev);
518         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
519
520         /* Never send a reset in response to a reset. */
521         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
522                 return;
523
524         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
525                 return;
526
527         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
528         if (dst == NULL)
529                 return;
530
531         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
532         if (skb == NULL)
533                 goto out;
534
535         rxiph = ip_hdr(rxskb);
536         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
537                                                                  rxiph->daddr);
538         skb_dst_set(skb, dst_clone(dst));
539
540         bh_lock_sock(ctl_sk);
541         err = ip_build_and_send_pkt(skb, ctl_sk,
542                                     rxiph->daddr, rxiph->saddr, NULL);
543         bh_unlock_sock(ctl_sk);
544
545         if (net_xmit_eval(err) == 0) {
546                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
547                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
548         }
549 out:
550          dst_release(dst);
551 }
552
553 static void dccp_v4_reqsk_destructor(struct request_sock *req)
554 {
555         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
556         kfree(inet_rsk(req)->opt);
557 }
558
559 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
560         .family         = PF_INET,
561         .obj_size       = sizeof(struct dccp_request_sock),
562         .rtx_syn_ack    = dccp_v4_send_response,
563         .send_ack       = dccp_reqsk_send_ack,
564         .destructor     = dccp_v4_reqsk_destructor,
565         .send_reset     = dccp_v4_ctl_send_reset,
566 };
567
568 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
569 {
570         struct inet_request_sock *ireq;
571         struct request_sock *req;
572         struct dccp_request_sock *dreq;
573         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
574         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
575
576         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
577         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
578                 return 0;       /* discard, don't send a reset here */
579
580         if (dccp_bad_service_code(sk, service)) {
581                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
582                 goto drop;
583         }
584         /*
585          * TW buckets are converted to open requests without
586          * limitations, they conserve resources and peer is
587          * evidently real one.
588          */
589         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
590         if (inet_csk_reqsk_queue_is_full(sk))
591                 goto drop;
592
593         /*
594          * Accept backlog is full. If we have already queued enough
595          * of warm entries in syn queue, drop request. It is better than
596          * clogging syn queue with openreqs with exponentially increasing
597          * timeout.
598          */
599         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
600                 goto drop;
601
602         req = inet_reqsk_alloc(&dccp_request_sock_ops);
603         if (req == NULL)
604                 goto drop;
605
606         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
607                 goto drop_and_free;
608
609         dreq = dccp_rsk(req);
610         if (dccp_parse_options(sk, dreq, skb))
611                 goto drop_and_free;
612
613         if (security_inet_conn_request(sk, skb, req))
614                 goto drop_and_free;
615
616         ireq = inet_rsk(req);
617         ireq->loc_addr = ip_hdr(skb)->daddr;
618         ireq->rmt_addr = ip_hdr(skb)->saddr;
619
620         /*
621          * Step 3: Process LISTEN state
622          *
623          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
624          *
625          * In fact we defer setting S.GSR, S.SWL, S.SWH to
626          * dccp_create_openreq_child.
627          */
628         dreq->dreq_isr     = dcb->dccpd_seq;
629         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
630         dreq->dreq_service = service;
631
632         if (dccp_v4_send_response(sk, req, NULL))
633                 goto drop_and_free;
634
635         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
636         return 0;
637
638 drop_and_free:
639         reqsk_free(req);
640 drop:
641         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
642         return -1;
643 }
644
645 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
646
647 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
648 {
649         struct dccp_hdr *dh = dccp_hdr(skb);
650
651         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
652                 if (dccp_rcv_established(sk, skb, dh, skb->len))
653                         goto reset;
654                 return 0;
655         }
656
657         /*
658          *  Step 3: Process LISTEN state
659          *       If P.type == Request or P contains a valid Init Cookie option,
660          *            (* Must scan the packet's options to check for Init
661          *               Cookies.  Only Init Cookies are processed here,
662          *               however; other options are processed in Step 8.  This
663          *               scan need only be performed if the endpoint uses Init
664          *               Cookies *)
665          *            (* Generate a new socket and switch to that socket *)
666          *            Set S := new socket for this port pair
667          *            S.state = RESPOND
668          *            Choose S.ISS (initial seqno) or set from Init Cookies
669          *            Initialize S.GAR := S.ISS
670          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
671          *            Continue with S.state == RESPOND
672          *            (* A Response packet will be generated in Step 11 *)
673          *       Otherwise,
674          *            Generate Reset(No Connection) unless P.type == Reset
675          *            Drop packet and return
676          *
677          * NOTE: the check for the packet types is done in
678          *       dccp_rcv_state_process
679          */
680         if (sk->sk_state == DCCP_LISTEN) {
681                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
682
683                 if (nsk == NULL)
684                         goto discard;
685
686                 if (nsk != sk) {
687                         if (dccp_child_process(sk, nsk, skb))
688                                 goto reset;
689                         return 0;
690                 }
691         }
692
693         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
694                 goto reset;
695         return 0;
696
697 reset:
698         dccp_v4_ctl_send_reset(sk, skb);
699 discard:
700         kfree_skb(skb);
701         return 0;
702 }
703
704 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
705
706 /**
707  *      dccp_invalid_packet  -  check for malformed packets
708  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
709  *      Packets that fail these checks are ignored and do not receive Resets.
710  */
711 int dccp_invalid_packet(struct sk_buff *skb)
712 {
713         const struct dccp_hdr *dh;
714         unsigned int cscov;
715
716         if (skb->pkt_type != PACKET_HOST)
717                 return 1;
718
719         /* If the packet is shorter than 12 bytes, drop packet and return */
720         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
721                 DCCP_WARN("pskb_may_pull failed\n");
722                 return 1;
723         }
724
725         dh = dccp_hdr(skb);
726
727         /* If P.type is not understood, drop packet and return */
728         if (dh->dccph_type >= DCCP_PKT_INVALID) {
729                 DCCP_WARN("invalid packet type\n");
730                 return 1;
731         }
732
733         /*
734          * If P.Data Offset is too small for packet type, drop packet and return
735          */
736         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
737                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
738                 return 1;
739         }
740         /*
741          * If P.Data Offset is too too large for packet, drop packet and return
742          */
743         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
744                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
745                 return 1;
746         }
747
748         /*
749          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
750          * has short sequence numbers), drop packet and return
751          */
752         if ((dh->dccph_type < DCCP_PKT_DATA    ||
753             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
754                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
755                           dccp_packet_name(dh->dccph_type));
756                 return 1;
757         }
758
759         /*
760          * If P.CsCov is too large for the packet size, drop packet and return.
761          * This must come _before_ checksumming (not as RFC 4340 suggests).
762          */
763         cscov = dccp_csum_coverage(skb);
764         if (cscov > skb->len) {
765                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
766                           dh->dccph_cscov, skb->len);
767                 return 1;
768         }
769
770         /* If header checksum is incorrect, drop packet and return.
771          * (This step is completed in the AF-dependent functions.) */
772         skb->csum = skb_checksum(skb, 0, cscov, 0);
773
774         return 0;
775 }
776
777 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
778
779 /* this is called when real data arrives */
780 static int dccp_v4_rcv(struct sk_buff *skb)
781 {
782         const struct dccp_hdr *dh;
783         const struct iphdr *iph;
784         struct sock *sk;
785         int min_cov;
786
787         /* Step 1: Check header basics */
788
789         if (dccp_invalid_packet(skb))
790                 goto discard_it;
791
792         iph = ip_hdr(skb);
793         /* Step 1: If header checksum is incorrect, drop packet and return */
794         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
795                 DCCP_WARN("dropped packet with invalid checksum\n");
796                 goto discard_it;
797         }
798
799         dh = dccp_hdr(skb);
800
801         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
802         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
803
804         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
805                       dccp_packet_name(dh->dccph_type),
806                       &iph->saddr, ntohs(dh->dccph_sport),
807                       &iph->daddr, ntohs(dh->dccph_dport),
808                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
809
810         if (dccp_packet_without_ack(skb)) {
811                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
812                 dccp_pr_debug_cat("\n");
813         } else {
814                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
815                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
816                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
817         }
818
819         /* Step 2:
820          *      Look up flow ID in table and get corresponding socket */
821         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
822                                dh->dccph_sport, dh->dccph_dport);
823         /*
824          * Step 2:
825          *      If no socket ...
826          */
827         if (sk == NULL) {
828                 dccp_pr_debug("failed to look up flow ID in table and "
829                               "get corresponding socket\n");
830                 goto no_dccp_socket;
831         }
832
833         /*
834          * Step 2:
835          *      ... or S.state == TIMEWAIT,
836          *              Generate Reset(No Connection) unless P.type == Reset
837          *              Drop packet and return
838          */
839         if (sk->sk_state == DCCP_TIME_WAIT) {
840                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
841                 inet_twsk_put(inet_twsk(sk));
842                 goto no_dccp_socket;
843         }
844
845         /*
846          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
847          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
848          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
849          */
850         min_cov = dccp_sk(sk)->dccps_pcrlen;
851         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
852                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
853                               dh->dccph_cscov, min_cov);
854                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
855                  *         options (Section 11.7) with Drop Code 0, Protocol
856                  *         Constraints."                                     */
857                 goto discard_and_relse;
858         }
859
860         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
861                 goto discard_and_relse;
862         nf_reset(skb);
863
864         return sk_receive_skb(sk, skb, 1);
865
866 no_dccp_socket:
867         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
868                 goto discard_it;
869         /*
870          * Step 2:
871          *      If no socket ...
872          *              Generate Reset(No Connection) unless P.type == Reset
873          *              Drop packet and return
874          */
875         if (dh->dccph_type != DCCP_PKT_RESET) {
876                 DCCP_SKB_CB(skb)->dccpd_reset_code =
877                                         DCCP_RESET_CODE_NO_CONNECTION;
878                 dccp_v4_ctl_send_reset(sk, skb);
879         }
880
881 discard_it:
882         kfree_skb(skb);
883         return 0;
884
885 discard_and_relse:
886         sock_put(sk);
887         goto discard_it;
888 }
889
890 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
891         .queue_xmit        = ip_queue_xmit,
892         .send_check        = dccp_v4_send_check,
893         .rebuild_header    = inet_sk_rebuild_header,
894         .conn_request      = dccp_v4_conn_request,
895         .syn_recv_sock     = dccp_v4_request_recv_sock,
896         .net_header_len    = sizeof(struct iphdr),
897         .setsockopt        = ip_setsockopt,
898         .getsockopt        = ip_getsockopt,
899         .addr2sockaddr     = inet_csk_addr2sockaddr,
900         .sockaddr_len      = sizeof(struct sockaddr_in),
901         .bind_conflict     = inet_csk_bind_conflict,
902 #ifdef CONFIG_COMPAT
903         .compat_setsockopt = compat_ip_setsockopt,
904         .compat_getsockopt = compat_ip_getsockopt,
905 #endif
906 };
907
908 static int dccp_v4_init_sock(struct sock *sk)
909 {
910         static __u8 dccp_v4_ctl_sock_initialized;
911         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
912
913         if (err == 0) {
914                 if (unlikely(!dccp_v4_ctl_sock_initialized))
915                         dccp_v4_ctl_sock_initialized = 1;
916                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
917         }
918
919         return err;
920 }
921
922 static struct timewait_sock_ops dccp_timewait_sock_ops = {
923         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
924 };
925
926 static struct proto dccp_v4_prot = {
927         .name                   = "DCCP",
928         .owner                  = THIS_MODULE,
929         .close                  = dccp_close,
930         .connect                = dccp_v4_connect,
931         .disconnect             = dccp_disconnect,
932         .ioctl                  = dccp_ioctl,
933         .init                   = dccp_v4_init_sock,
934         .setsockopt             = dccp_setsockopt,
935         .getsockopt             = dccp_getsockopt,
936         .sendmsg                = dccp_sendmsg,
937         .recvmsg                = dccp_recvmsg,
938         .backlog_rcv            = dccp_v4_do_rcv,
939         .hash                   = inet_hash,
940         .unhash                 = inet_unhash,
941         .accept                 = inet_csk_accept,
942         .get_port               = inet_csk_get_port,
943         .shutdown               = dccp_shutdown,
944         .destroy                = dccp_destroy_sock,
945         .orphan_count           = &dccp_orphan_count,
946         .max_header             = MAX_DCCP_HEADER,
947         .obj_size               = sizeof(struct dccp_sock),
948         .slab_flags             = SLAB_DESTROY_BY_RCU,
949         .rsk_prot               = &dccp_request_sock_ops,
950         .twsk_prot              = &dccp_timewait_sock_ops,
951         .h.hashinfo             = &dccp_hashinfo,
952 #ifdef CONFIG_COMPAT
953         .compat_setsockopt      = compat_dccp_setsockopt,
954         .compat_getsockopt      = compat_dccp_getsockopt,
955 #endif
956 };
957
958 static const struct net_protocol dccp_v4_protocol = {
959         .handler        = dccp_v4_rcv,
960         .err_handler    = dccp_v4_err,
961         .no_policy      = 1,
962         .netns_ok       = 1,
963 };
964
965 static const struct proto_ops inet_dccp_ops = {
966         .family            = PF_INET,
967         .owner             = THIS_MODULE,
968         .release           = inet_release,
969         .bind              = inet_bind,
970         .connect           = inet_stream_connect,
971         .socketpair        = sock_no_socketpair,
972         .accept            = inet_accept,
973         .getname           = inet_getname,
974         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
975         .poll              = dccp_poll,
976         .ioctl             = inet_ioctl,
977         /* FIXME: work on inet_listen to rename it to sock_common_listen */
978         .listen            = inet_dccp_listen,
979         .shutdown          = inet_shutdown,
980         .setsockopt        = sock_common_setsockopt,
981         .getsockopt        = sock_common_getsockopt,
982         .sendmsg           = inet_sendmsg,
983         .recvmsg           = sock_common_recvmsg,
984         .mmap              = sock_no_mmap,
985         .sendpage          = sock_no_sendpage,
986 #ifdef CONFIG_COMPAT
987         .compat_setsockopt = compat_sock_common_setsockopt,
988         .compat_getsockopt = compat_sock_common_getsockopt,
989 #endif
990 };
991
992 static struct inet_protosw dccp_v4_protosw = {
993         .type           = SOCK_DCCP,
994         .protocol       = IPPROTO_DCCP,
995         .prot           = &dccp_v4_prot,
996         .ops            = &inet_dccp_ops,
997         .no_check       = 0,
998         .flags          = INET_PROTOSW_ICSK,
999 };
1000
1001 static int __net_init dccp_v4_init_net(struct net *net)
1002 {
1003         if (dccp_hashinfo.bhash == NULL)
1004                 return -ESOCKTNOSUPPORT;
1005
1006         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1007                                     SOCK_DCCP, IPPROTO_DCCP, net);
1008 }
1009
1010 static void __net_exit dccp_v4_exit_net(struct net *net)
1011 {
1012         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1013 }
1014
1015 static struct pernet_operations dccp_v4_ops = {
1016         .init   = dccp_v4_init_net,
1017         .exit   = dccp_v4_exit_net,
1018 };
1019
1020 static int __init dccp_v4_init(void)
1021 {
1022         int err = proto_register(&dccp_v4_prot, 1);
1023
1024         if (err != 0)
1025                 goto out;
1026
1027         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1028         if (err != 0)
1029                 goto out_proto_unregister;
1030
1031         inet_register_protosw(&dccp_v4_protosw);
1032
1033         err = register_pernet_subsys(&dccp_v4_ops);
1034         if (err)
1035                 goto out_destroy_ctl_sock;
1036 out:
1037         return err;
1038 out_destroy_ctl_sock:
1039         inet_unregister_protosw(&dccp_v4_protosw);
1040         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1041 out_proto_unregister:
1042         proto_unregister(&dccp_v4_prot);
1043         goto out;
1044 }
1045
1046 static void __exit dccp_v4_exit(void)
1047 {
1048         unregister_pernet_subsys(&dccp_v4_ops);
1049         inet_unregister_protosw(&dccp_v4_protosw);
1050         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1051         proto_unregister(&dccp_v4_prot);
1052 }
1053
1054 module_init(dccp_v4_init);
1055 module_exit(dccp_v4_exit);
1056
1057 /*
1058  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1059  * values directly, Also cover the case where the protocol is not specified,
1060  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1061  */
1062 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1063 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1064 MODULE_LICENSE("GPL");
1065 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1066 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");