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