Merge branch 'x86/uaccess' into core/percpu
[linux-2.6] / 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/module.h>
16 #include <linux/skbuff.h>
17 #include <linux/random.h>
18
19 #include <net/icmp.h>
20 #include <net/inet_common.h>
21 #include <net/inet_hashtables.h>
22 #include <net/inet_sock.h>
23 #include <net/protocol.h>
24 #include <net/sock.h>
25 #include <net/timewait_sock.h>
26 #include <net/tcp_states.h>
27 #include <net/xfrm.h>
28
29 #include "ackvec.h"
30 #include "ccid.h"
31 #include "dccp.h"
32 #include "feat.h"
33
34 /*
35  * The per-net dccp.v4_ctl_sk socket is used for responding to
36  * the Out-of-the-blue (OOTB) packets. A control sock will be created
37  * for this socket at the initialization time.
38  */
39
40 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
41 {
42         struct inet_sock *inet = inet_sk(sk);
43         struct dccp_sock *dp = dccp_sk(sk);
44         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
45         struct rtable *rt;
46         __be32 daddr, nexthop;
47         int tmp;
48         int err;
49
50         dp->dccps_role = DCCP_ROLE_CLIENT;
51
52         if (addr_len < sizeof(struct sockaddr_in))
53                 return -EINVAL;
54
55         if (usin->sin_family != AF_INET)
56                 return -EAFNOSUPPORT;
57
58         nexthop = daddr = usin->sin_addr.s_addr;
59         if (inet->opt != NULL && inet->opt->srr) {
60                 if (daddr == 0)
61                         return -EINVAL;
62                 nexthop = inet->opt->faddr;
63         }
64
65         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
66                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
67                                IPPROTO_DCCP,
68                                inet->sport, usin->sin_port, sk, 1);
69         if (tmp < 0)
70                 return tmp;
71
72         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
73                 ip_rt_put(rt);
74                 return -ENETUNREACH;
75         }
76
77         if (inet->opt == NULL || !inet->opt->srr)
78                 daddr = rt->rt_dst;
79
80         if (inet->saddr == 0)
81                 inet->saddr = rt->rt_src;
82         inet->rcv_saddr = inet->saddr;
83
84         inet->dport = usin->sin_port;
85         inet->daddr = daddr;
86
87         inet_csk(sk)->icsk_ext_hdr_len = 0;
88         if (inet->opt != NULL)
89                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
90         /*
91          * Socket identity is still unknown (sport may be zero).
92          * However we set state to DCCP_REQUESTING and not releasing socket
93          * lock select source port, enter ourselves into the hash tables and
94          * complete initialization after this.
95          */
96         dccp_set_state(sk, DCCP_REQUESTING);
97         err = inet_hash_connect(&dccp_death_row, sk);
98         if (err != 0)
99                 goto failure;
100
101         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
102                                 sk);
103         if (err != 0)
104                 goto failure;
105
106         /* OK, now commit destination to socket.  */
107         sk_setup_caps(sk, &rt->u.dst);
108
109         dp->dccps_iss = secure_dccp_sequence_number(inet->saddr, inet->daddr,
110                                                     inet->sport, inet->dport);
111         inet->id = dp->dccps_iss ^ jiffies;
112
113         err = dccp_connect(sk);
114         rt = NULL;
115         if (err != 0)
116                 goto failure;
117 out:
118         return err;
119 failure:
120         /*
121          * This unhashes the socket and releases the local port, if necessary.
122          */
123         dccp_set_state(sk, DCCP_CLOSED);
124         ip_rt_put(rt);
125         sk->sk_route_caps = 0;
126         inet->dport = 0;
127         goto out;
128 }
129
130 EXPORT_SYMBOL_GPL(dccp_v4_connect);
131
132 /*
133  * This routine does path mtu discovery as defined in RFC1191.
134  */
135 static inline void dccp_do_pmtu_discovery(struct sock *sk,
136                                           const struct iphdr *iph,
137                                           u32 mtu)
138 {
139         struct dst_entry *dst;
140         const struct inet_sock *inet = inet_sk(sk);
141         const struct dccp_sock *dp = dccp_sk(sk);
142
143         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
144          * send out by Linux are always < 576bytes so they should go through
145          * unfragmented).
146          */
147         if (sk->sk_state == DCCP_LISTEN)
148                 return;
149
150         /* We don't check in the destentry if pmtu discovery is forbidden
151          * on this route. We just assume that no packet_to_big packets
152          * are send back when pmtu discovery is not active.
153          * There is a small race when the user changes this flag in the
154          * route, but I think that's acceptable.
155          */
156         if ((dst = __sk_dst_check(sk, 0)) == NULL)
157                 return;
158
159         dst->ops->update_pmtu(dst, mtu);
160
161         /* Something is about to be wrong... Remember soft error
162          * for the case, if this connection will not able to recover.
163          */
164         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
165                 sk->sk_err_soft = EMSGSIZE;
166
167         mtu = dst_mtu(dst);
168
169         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
170             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
171                 dccp_sync_mss(sk, mtu);
172
173                 /*
174                  * From RFC 4340, sec. 14.1:
175                  *
176                  *      DCCP-Sync packets are the best choice for upward
177                  *      probing, since DCCP-Sync probes do not risk application
178                  *      data loss.
179                  */
180                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
181         } /* else let the usual retransmit timer handle it */
182 }
183
184 /*
185  * This routine is called by the ICMP module when it gets some sort of error
186  * condition. If err < 0 then the socket should be closed and the error
187  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
188  * After adjustment header points to the first 8 bytes of the tcp header. We
189  * need to find the appropriate port.
190  *
191  * The locking strategy used here is very "optimistic". When someone else
192  * accesses the socket the ICMP is just dropped and for some paths there is no
193  * check at all. A more general error queue to queue errors for later handling
194  * is probably better.
195  */
196 static void dccp_v4_err(struct sk_buff *skb, u32 info)
197 {
198         const struct iphdr *iph = (struct iphdr *)skb->data;
199         const u8 offset = iph->ihl << 2;
200         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
201         struct dccp_sock *dp;
202         struct inet_sock *inet;
203         const int type = icmp_hdr(skb)->type;
204         const int code = icmp_hdr(skb)->code;
205         struct sock *sk;
206         __u64 seq;
207         int err;
208         struct net *net = dev_net(skb->dev);
209
210         if (skb->len < offset + sizeof(*dh) ||
211             skb->len < offset + __dccp_basic_hdr_len(dh)) {
212                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
213                 return;
214         }
215
216         sk = inet_lookup(net, &dccp_hashinfo,
217                         iph->daddr, dh->dccph_dport,
218                         iph->saddr, dh->dccph_sport, inet_iif(skb));
219         if (sk == NULL) {
220                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
221                 return;
222         }
223
224         if (sk->sk_state == DCCP_TIME_WAIT) {
225                 inet_twsk_put(inet_twsk(sk));
226                 return;
227         }
228
229         bh_lock_sock(sk);
230         /* If too many ICMPs get dropped on busy
231          * servers this needs to be solved differently.
232          */
233         if (sock_owned_by_user(sk))
234                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
235
236         if (sk->sk_state == DCCP_CLOSED)
237                 goto out;
238
239         dp = dccp_sk(sk);
240         seq = dccp_hdr_seq(dh);
241         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
242             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
243                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
244                 goto out;
245         }
246
247         switch (type) {
248         case ICMP_SOURCE_QUENCH:
249                 /* Just silently ignore these. */
250                 goto out;
251         case ICMP_PARAMETERPROB:
252                 err = EPROTO;
253                 break;
254         case ICMP_DEST_UNREACH:
255                 if (code > NR_ICMP_UNREACH)
256                         goto out;
257
258                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
259                         if (!sock_owned_by_user(sk))
260                                 dccp_do_pmtu_discovery(sk, iph, info);
261                         goto out;
262                 }
263
264                 err = icmp_err_convert[code].errno;
265                 break;
266         case ICMP_TIME_EXCEEDED:
267                 err = EHOSTUNREACH;
268                 break;
269         default:
270                 goto out;
271         }
272
273         switch (sk->sk_state) {
274                 struct request_sock *req , **prev;
275         case DCCP_LISTEN:
276                 if (sock_owned_by_user(sk))
277                         goto out;
278                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
279                                           iph->daddr, iph->saddr);
280                 if (!req)
281                         goto out;
282
283                 /*
284                  * ICMPs are not backlogged, hence we cannot get an established
285                  * socket here.
286                  */
287                 WARN_ON(req->sk);
288
289                 if (seq != dccp_rsk(req)->dreq_iss) {
290                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
291                         goto out;
292                 }
293                 /*
294                  * Still in RESPOND, just remove it silently.
295                  * There is no good way to pass the error to the newly
296                  * created socket, and POSIX does not want network
297                  * errors returned from accept().
298                  */
299                 inet_csk_reqsk_queue_drop(sk, req, prev);
300                 goto out;
301
302         case DCCP_REQUESTING:
303         case DCCP_RESPOND:
304                 if (!sock_owned_by_user(sk)) {
305                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
306                         sk->sk_err = err;
307
308                         sk->sk_error_report(sk);
309
310                         dccp_done(sk);
311                 } else
312                         sk->sk_err_soft = err;
313                 goto out;
314         }
315
316         /* If we've already connected we will keep trying
317          * until we time out, or the user gives up.
318          *
319          * rfc1122 4.2.3.9 allows to consider as hard errors
320          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
321          * but it is obsoleted by pmtu discovery).
322          *
323          * Note, that in modern internet, where routing is unreliable
324          * and in each dark corner broken firewalls sit, sending random
325          * errors ordered by their masters even this two messages finally lose
326          * their original sense (even Linux sends invalid PORT_UNREACHs)
327          *
328          * Now we are in compliance with RFCs.
329          *                                                      --ANK (980905)
330          */
331
332         inet = inet_sk(sk);
333         if (!sock_owned_by_user(sk) && inet->recverr) {
334                 sk->sk_err = err;
335                 sk->sk_error_report(sk);
336         } else /* Only an error on timeout */
337                 sk->sk_err_soft = err;
338 out:
339         bh_unlock_sock(sk);
340         sock_put(sk);
341 }
342
343 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
344                                       __be32 src, __be32 dst)
345 {
346         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
347 }
348
349 void dccp_v4_send_check(struct sock *sk, int unused, struct sk_buff *skb)
350 {
351         const struct inet_sock *inet = inet_sk(sk);
352         struct dccp_hdr *dh = dccp_hdr(skb);
353
354         dccp_csum_outgoing(skb);
355         dh->dccph_checksum = dccp_v4_csum_finish(skb, inet->saddr, inet->daddr);
356 }
357
358 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
359
360 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
361 {
362         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
363                                            ip_hdr(skb)->saddr,
364                                            dccp_hdr(skb)->dccph_dport,
365                                            dccp_hdr(skb)->dccph_sport);
366 }
367
368 /*
369  * The three way handshake has completed - we got a valid ACK or DATAACK -
370  * now create the new socket.
371  *
372  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
373  */
374 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
375                                        struct request_sock *req,
376                                        struct dst_entry *dst)
377 {
378         struct inet_request_sock *ireq;
379         struct inet_sock *newinet;
380         struct sock *newsk;
381
382         if (sk_acceptq_is_full(sk))
383                 goto exit_overflow;
384
385         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
386                 goto exit;
387
388         newsk = dccp_create_openreq_child(sk, req, skb);
389         if (newsk == NULL)
390                 goto exit;
391
392         sk_setup_caps(newsk, dst);
393
394         newinet            = inet_sk(newsk);
395         ireq               = inet_rsk(req);
396         newinet->daddr     = ireq->rmt_addr;
397         newinet->rcv_saddr = ireq->loc_addr;
398         newinet->saddr     = ireq->loc_addr;
399         newinet->opt       = ireq->opt;
400         ireq->opt          = NULL;
401         newinet->mc_index  = inet_iif(skb);
402         newinet->mc_ttl    = ip_hdr(skb)->ttl;
403         newinet->id        = jiffies;
404
405         dccp_sync_mss(newsk, dst_mtu(dst));
406
407         __inet_hash_nolisten(newsk);
408         __inet_inherit_port(sk, newsk);
409
410         return newsk;
411
412 exit_overflow:
413         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
414 exit:
415         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
416         dst_release(dst);
417         return NULL;
418 }
419
420 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
421
422 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
423 {
424         const struct dccp_hdr *dh = dccp_hdr(skb);
425         const struct iphdr *iph = ip_hdr(skb);
426         struct sock *nsk;
427         struct request_sock **prev;
428         /* Find possible connection requests. */
429         struct request_sock *req = inet_csk_search_req(sk, &prev,
430                                                        dh->dccph_sport,
431                                                        iph->saddr, iph->daddr);
432         if (req != NULL)
433                 return dccp_check_req(sk, skb, req, prev);
434
435         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
436                                       iph->saddr, dh->dccph_sport,
437                                       iph->daddr, dh->dccph_dport,
438                                       inet_iif(skb));
439         if (nsk != NULL) {
440                 if (nsk->sk_state != DCCP_TIME_WAIT) {
441                         bh_lock_sock(nsk);
442                         return nsk;
443                 }
444                 inet_twsk_put(inet_twsk(nsk));
445                 return NULL;
446         }
447
448         return sk;
449 }
450
451 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
452                                            struct sk_buff *skb)
453 {
454         struct rtable *rt;
455         struct flowi fl = { .oif = skb->rtable->rt_iif,
456                             .nl_u = { .ip4_u =
457                                       { .daddr = ip_hdr(skb)->saddr,
458                                         .saddr = ip_hdr(skb)->daddr,
459                                         .tos = RT_CONN_FLAGS(sk) } },
460                             .proto = sk->sk_protocol,
461                             .uli_u = { .ports =
462                                        { .sport = dccp_hdr(skb)->dccph_dport,
463                                          .dport = dccp_hdr(skb)->dccph_sport }
464                                      }
465                           };
466
467         security_skb_classify_flow(skb, &fl);
468         if (ip_route_output_flow(net, &rt, &fl, sk, 0)) {
469                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
470                 return NULL;
471         }
472
473         return &rt->u.dst;
474 }
475
476 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req)
477 {
478         int err = -1;
479         struct sk_buff *skb;
480         struct dst_entry *dst;
481
482         dst = inet_csk_route_req(sk, req);
483         if (dst == NULL)
484                 goto out;
485
486         skb = dccp_make_response(sk, dst, req);
487         if (skb != NULL) {
488                 const struct inet_request_sock *ireq = inet_rsk(req);
489                 struct dccp_hdr *dh = dccp_hdr(skb);
490
491                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
492                                                               ireq->rmt_addr);
493                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
494                                             ireq->rmt_addr,
495                                             ireq->opt);
496                 err = net_xmit_eval(err);
497         }
498
499 out:
500         dst_release(dst);
501         return err;
502 }
503
504 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
505 {
506         int err;
507         const struct iphdr *rxiph;
508         struct sk_buff *skb;
509         struct dst_entry *dst;
510         struct net *net = dev_net(rxskb->dst->dev);
511         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
512
513         /* Never send a reset in response to a reset. */
514         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
515                 return;
516
517         if (rxskb->rtable->rt_type != RTN_LOCAL)
518                 return;
519
520         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
521         if (dst == NULL)
522                 return;
523
524         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
525         if (skb == NULL)
526                 goto out;
527
528         rxiph = ip_hdr(rxskb);
529         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
530                                                                  rxiph->daddr);
531         skb->dst = dst_clone(dst);
532
533         bh_lock_sock(ctl_sk);
534         err = ip_build_and_send_pkt(skb, ctl_sk,
535                                     rxiph->daddr, rxiph->saddr, NULL);
536         bh_unlock_sock(ctl_sk);
537
538         if (net_xmit_eval(err) == 0) {
539                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
540                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
541         }
542 out:
543          dst_release(dst);
544 }
545
546 static void dccp_v4_reqsk_destructor(struct request_sock *req)
547 {
548         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
549         kfree(inet_rsk(req)->opt);
550 }
551
552 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
553         .family         = PF_INET,
554         .obj_size       = sizeof(struct dccp_request_sock),
555         .rtx_syn_ack    = dccp_v4_send_response,
556         .send_ack       = dccp_reqsk_send_ack,
557         .destructor     = dccp_v4_reqsk_destructor,
558         .send_reset     = dccp_v4_ctl_send_reset,
559 };
560
561 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
562 {
563         struct inet_request_sock *ireq;
564         struct request_sock *req;
565         struct dccp_request_sock *dreq;
566         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
567         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
568
569         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
570         if (skb->rtable->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
571                 return 0;       /* discard, don't send a reset here */
572
573         if (dccp_bad_service_code(sk, service)) {
574                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
575                 goto drop;
576         }
577         /*
578          * TW buckets are converted to open requests without
579          * limitations, they conserve resources and peer is
580          * evidently real one.
581          */
582         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
583         if (inet_csk_reqsk_queue_is_full(sk))
584                 goto drop;
585
586         /*
587          * Accept backlog is full. If we have already queued enough
588          * of warm entries in syn queue, drop request. It is better than
589          * clogging syn queue with openreqs with exponentially increasing
590          * timeout.
591          */
592         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
593                 goto drop;
594
595         req = inet_reqsk_alloc(&dccp_request_sock_ops);
596         if (req == NULL)
597                 goto drop;
598
599         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
600                 goto drop_and_free;
601
602         dreq = dccp_rsk(req);
603         if (dccp_parse_options(sk, dreq, skb))
604                 goto drop_and_free;
605
606         if (security_inet_conn_request(sk, skb, req))
607                 goto drop_and_free;
608
609         ireq = inet_rsk(req);
610         ireq->loc_addr = ip_hdr(skb)->daddr;
611         ireq->rmt_addr = ip_hdr(skb)->saddr;
612
613         /*
614          * Step 3: Process LISTEN state
615          *
616          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
617          *
618          * In fact we defer setting S.GSR, S.SWL, S.SWH to
619          * dccp_create_openreq_child.
620          */
621         dreq->dreq_isr     = dcb->dccpd_seq;
622         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
623         dreq->dreq_service = service;
624
625         if (dccp_v4_send_response(sk, req))
626                 goto drop_and_free;
627
628         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
629         return 0;
630
631 drop_and_free:
632         reqsk_free(req);
633 drop:
634         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
635         return -1;
636 }
637
638 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
639
640 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
641 {
642         struct dccp_hdr *dh = dccp_hdr(skb);
643
644         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
645                 if (dccp_rcv_established(sk, skb, dh, skb->len))
646                         goto reset;
647                 return 0;
648         }
649
650         /*
651          *  Step 3: Process LISTEN state
652          *       If P.type == Request or P contains a valid Init Cookie option,
653          *            (* Must scan the packet's options to check for Init
654          *               Cookies.  Only Init Cookies are processed here,
655          *               however; other options are processed in Step 8.  This
656          *               scan need only be performed if the endpoint uses Init
657          *               Cookies *)
658          *            (* Generate a new socket and switch to that socket *)
659          *            Set S := new socket for this port pair
660          *            S.state = RESPOND
661          *            Choose S.ISS (initial seqno) or set from Init Cookies
662          *            Initialize S.GAR := S.ISS
663          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
664          *            Continue with S.state == RESPOND
665          *            (* A Response packet will be generated in Step 11 *)
666          *       Otherwise,
667          *            Generate Reset(No Connection) unless P.type == Reset
668          *            Drop packet and return
669          *
670          * NOTE: the check for the packet types is done in
671          *       dccp_rcv_state_process
672          */
673         if (sk->sk_state == DCCP_LISTEN) {
674                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
675
676                 if (nsk == NULL)
677                         goto discard;
678
679                 if (nsk != sk) {
680                         if (dccp_child_process(sk, nsk, skb))
681                                 goto reset;
682                         return 0;
683                 }
684         }
685
686         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
687                 goto reset;
688         return 0;
689
690 reset:
691         dccp_v4_ctl_send_reset(sk, skb);
692 discard:
693         kfree_skb(skb);
694         return 0;
695 }
696
697 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
698
699 /**
700  *      dccp_invalid_packet  -  check for malformed packets
701  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
702  *      Packets that fail these checks are ignored and do not receive Resets.
703  */
704 int dccp_invalid_packet(struct sk_buff *skb)
705 {
706         const struct dccp_hdr *dh;
707         unsigned int cscov;
708
709         if (skb->pkt_type != PACKET_HOST)
710                 return 1;
711
712         /* If the packet is shorter than 12 bytes, drop packet and return */
713         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
714                 DCCP_WARN("pskb_may_pull failed\n");
715                 return 1;
716         }
717
718         dh = dccp_hdr(skb);
719
720         /* If P.type is not understood, drop packet and return */
721         if (dh->dccph_type >= DCCP_PKT_INVALID) {
722                 DCCP_WARN("invalid packet type\n");
723                 return 1;
724         }
725
726         /*
727          * If P.Data Offset is too small for packet type, drop packet and return
728          */
729         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
730                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
731                 return 1;
732         }
733         /*
734          * If P.Data Offset is too too large for packet, drop packet and return
735          */
736         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
737                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
738                 return 1;
739         }
740
741         /*
742          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
743          * has short sequence numbers), drop packet and return
744          */
745         if ((dh->dccph_type < DCCP_PKT_DATA    ||
746             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
747                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
748                           dccp_packet_name(dh->dccph_type));
749                 return 1;
750         }
751
752         /*
753          * If P.CsCov is too large for the packet size, drop packet and return.
754          * This must come _before_ checksumming (not as RFC 4340 suggests).
755          */
756         cscov = dccp_csum_coverage(skb);
757         if (cscov > skb->len) {
758                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
759                           dh->dccph_cscov, skb->len);
760                 return 1;
761         }
762
763         /* If header checksum is incorrect, drop packet and return.
764          * (This step is completed in the AF-dependent functions.) */
765         skb->csum = skb_checksum(skb, 0, cscov, 0);
766
767         return 0;
768 }
769
770 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
771
772 /* this is called when real data arrives */
773 static int dccp_v4_rcv(struct sk_buff *skb)
774 {
775         const struct dccp_hdr *dh;
776         const struct iphdr *iph;
777         struct sock *sk;
778         int min_cov;
779
780         /* Step 1: Check header basics */
781
782         if (dccp_invalid_packet(skb))
783                 goto discard_it;
784
785         iph = ip_hdr(skb);
786         /* Step 1: If header checksum is incorrect, drop packet and return */
787         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
788                 DCCP_WARN("dropped packet with invalid checksum\n");
789                 goto discard_it;
790         }
791
792         dh = dccp_hdr(skb);
793
794         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
795         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
796
797         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
798                       dccp_packet_name(dh->dccph_type),
799                       &iph->saddr, ntohs(dh->dccph_sport),
800                       &iph->daddr, ntohs(dh->dccph_dport),
801                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
802
803         if (dccp_packet_without_ack(skb)) {
804                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
805                 dccp_pr_debug_cat("\n");
806         } else {
807                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
808                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
809                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
810         }
811
812         /* Step 2:
813          *      Look up flow ID in table and get corresponding socket */
814         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
815                                dh->dccph_sport, dh->dccph_dport);
816         /*
817          * Step 2:
818          *      If no socket ...
819          */
820         if (sk == NULL) {
821                 dccp_pr_debug("failed to look up flow ID in table and "
822                               "get corresponding socket\n");
823                 goto no_dccp_socket;
824         }
825
826         /*
827          * Step 2:
828          *      ... or S.state == TIMEWAIT,
829          *              Generate Reset(No Connection) unless P.type == Reset
830          *              Drop packet and return
831          */
832         if (sk->sk_state == DCCP_TIME_WAIT) {
833                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
834                 inet_twsk_put(inet_twsk(sk));
835                 goto no_dccp_socket;
836         }
837
838         /*
839          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
840          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
841          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
842          */
843         min_cov = dccp_sk(sk)->dccps_pcrlen;
844         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
845                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
846                               dh->dccph_cscov, min_cov);
847                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
848                  *         options (Section 11.7) with Drop Code 0, Protocol
849                  *         Constraints."                                     */
850                 goto discard_and_relse;
851         }
852
853         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
854                 goto discard_and_relse;
855         nf_reset(skb);
856
857         return sk_receive_skb(sk, skb, 1);
858
859 no_dccp_socket:
860         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
861                 goto discard_it;
862         /*
863          * Step 2:
864          *      If no socket ...
865          *              Generate Reset(No Connection) unless P.type == Reset
866          *              Drop packet and return
867          */
868         if (dh->dccph_type != DCCP_PKT_RESET) {
869                 DCCP_SKB_CB(skb)->dccpd_reset_code =
870                                         DCCP_RESET_CODE_NO_CONNECTION;
871                 dccp_v4_ctl_send_reset(sk, skb);
872         }
873
874 discard_it:
875         kfree_skb(skb);
876         return 0;
877
878 discard_and_relse:
879         sock_put(sk);
880         goto discard_it;
881 }
882
883 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
884         .queue_xmit        = ip_queue_xmit,
885         .send_check        = dccp_v4_send_check,
886         .rebuild_header    = inet_sk_rebuild_header,
887         .conn_request      = dccp_v4_conn_request,
888         .syn_recv_sock     = dccp_v4_request_recv_sock,
889         .net_header_len    = sizeof(struct iphdr),
890         .setsockopt        = ip_setsockopt,
891         .getsockopt        = ip_getsockopt,
892         .addr2sockaddr     = inet_csk_addr2sockaddr,
893         .sockaddr_len      = sizeof(struct sockaddr_in),
894         .bind_conflict     = inet_csk_bind_conflict,
895 #ifdef CONFIG_COMPAT
896         .compat_setsockopt = compat_ip_setsockopt,
897         .compat_getsockopt = compat_ip_getsockopt,
898 #endif
899 };
900
901 static int dccp_v4_init_sock(struct sock *sk)
902 {
903         static __u8 dccp_v4_ctl_sock_initialized;
904         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
905
906         if (err == 0) {
907                 if (unlikely(!dccp_v4_ctl_sock_initialized))
908                         dccp_v4_ctl_sock_initialized = 1;
909                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
910         }
911
912         return err;
913 }
914
915 static struct timewait_sock_ops dccp_timewait_sock_ops = {
916         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
917 };
918
919 static struct proto dccp_v4_prot = {
920         .name                   = "DCCP",
921         .owner                  = THIS_MODULE,
922         .close                  = dccp_close,
923         .connect                = dccp_v4_connect,
924         .disconnect             = dccp_disconnect,
925         .ioctl                  = dccp_ioctl,
926         .init                   = dccp_v4_init_sock,
927         .setsockopt             = dccp_setsockopt,
928         .getsockopt             = dccp_getsockopt,
929         .sendmsg                = dccp_sendmsg,
930         .recvmsg                = dccp_recvmsg,
931         .backlog_rcv            = dccp_v4_do_rcv,
932         .hash                   = inet_hash,
933         .unhash                 = inet_unhash,
934         .accept                 = inet_csk_accept,
935         .get_port               = inet_csk_get_port,
936         .shutdown               = dccp_shutdown,
937         .destroy                = dccp_destroy_sock,
938         .orphan_count           = &dccp_orphan_count,
939         .max_header             = MAX_DCCP_HEADER,
940         .obj_size               = sizeof(struct dccp_sock),
941         .slab_flags             = SLAB_DESTROY_BY_RCU,
942         .rsk_prot               = &dccp_request_sock_ops,
943         .twsk_prot              = &dccp_timewait_sock_ops,
944         .h.hashinfo             = &dccp_hashinfo,
945 #ifdef CONFIG_COMPAT
946         .compat_setsockopt      = compat_dccp_setsockopt,
947         .compat_getsockopt      = compat_dccp_getsockopt,
948 #endif
949 };
950
951 static struct net_protocol dccp_v4_protocol = {
952         .handler        = dccp_v4_rcv,
953         .err_handler    = dccp_v4_err,
954         .no_policy      = 1,
955         .netns_ok       = 1,
956 };
957
958 static const struct proto_ops inet_dccp_ops = {
959         .family            = PF_INET,
960         .owner             = THIS_MODULE,
961         .release           = inet_release,
962         .bind              = inet_bind,
963         .connect           = inet_stream_connect,
964         .socketpair        = sock_no_socketpair,
965         .accept            = inet_accept,
966         .getname           = inet_getname,
967         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
968         .poll              = dccp_poll,
969         .ioctl             = inet_ioctl,
970         /* FIXME: work on inet_listen to rename it to sock_common_listen */
971         .listen            = inet_dccp_listen,
972         .shutdown          = inet_shutdown,
973         .setsockopt        = sock_common_setsockopt,
974         .getsockopt        = sock_common_getsockopt,
975         .sendmsg           = inet_sendmsg,
976         .recvmsg           = sock_common_recvmsg,
977         .mmap              = sock_no_mmap,
978         .sendpage          = sock_no_sendpage,
979 #ifdef CONFIG_COMPAT
980         .compat_setsockopt = compat_sock_common_setsockopt,
981         .compat_getsockopt = compat_sock_common_getsockopt,
982 #endif
983 };
984
985 static struct inet_protosw dccp_v4_protosw = {
986         .type           = SOCK_DCCP,
987         .protocol       = IPPROTO_DCCP,
988         .prot           = &dccp_v4_prot,
989         .ops            = &inet_dccp_ops,
990         .capability     = -1,
991         .no_check       = 0,
992         .flags          = INET_PROTOSW_ICSK,
993 };
994
995 static int dccp_v4_init_net(struct net *net)
996 {
997         int err;
998
999         err = inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1000                                    SOCK_DCCP, IPPROTO_DCCP, net);
1001         return err;
1002 }
1003
1004 static void dccp_v4_exit_net(struct net *net)
1005 {
1006         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1007 }
1008
1009 static struct pernet_operations dccp_v4_ops = {
1010         .init   = dccp_v4_init_net,
1011         .exit   = dccp_v4_exit_net,
1012 };
1013
1014 static int __init dccp_v4_init(void)
1015 {
1016         int err = proto_register(&dccp_v4_prot, 1);
1017
1018         if (err != 0)
1019                 goto out;
1020
1021         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1022         if (err != 0)
1023                 goto out_proto_unregister;
1024
1025         inet_register_protosw(&dccp_v4_protosw);
1026
1027         err = register_pernet_subsys(&dccp_v4_ops);
1028         if (err)
1029                 goto out_destroy_ctl_sock;
1030 out:
1031         return err;
1032 out_destroy_ctl_sock:
1033         inet_unregister_protosw(&dccp_v4_protosw);
1034         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1035 out_proto_unregister:
1036         proto_unregister(&dccp_v4_prot);
1037         goto out;
1038 }
1039
1040 static void __exit dccp_v4_exit(void)
1041 {
1042         unregister_pernet_subsys(&dccp_v4_ops);
1043         inet_unregister_protosw(&dccp_v4_protosw);
1044         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1045         proto_unregister(&dccp_v4_prot);
1046 }
1047
1048 module_init(dccp_v4_init);
1049 module_exit(dccp_v4_exit);
1050
1051 /*
1052  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1053  * values directly, Also cover the case where the protocol is not specified,
1054  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1055  */
1056 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1057 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1058 MODULE_LICENSE("GPL");
1059 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1060 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");