[DCCP]: Support for partial checksums (RFC 4340, sec. 9.2)
[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  * This is the global socket data structure 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 static struct socket *dccp_v4_ctl_socket;
40
41 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
42 {
43         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
44                                  inet_csk_bind_conflict);
45 }
46
47 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
48 {
49         struct inet_sock *inet = inet_sk(sk);
50         struct dccp_sock *dp = dccp_sk(sk);
51         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
52         struct rtable *rt;
53         __be32 daddr, nexthop;
54         int tmp;
55         int err;
56
57         dp->dccps_role = DCCP_ROLE_CLIENT;
58
59         if (addr_len < sizeof(struct sockaddr_in))
60                 return -EINVAL;
61
62         if (usin->sin_family != AF_INET)
63                 return -EAFNOSUPPORT;
64
65         nexthop = daddr = usin->sin_addr.s_addr;
66         if (inet->opt != NULL && inet->opt->srr) {
67                 if (daddr == 0)
68                         return -EINVAL;
69                 nexthop = inet->opt->faddr;
70         }
71
72         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
73                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
74                                IPPROTO_DCCP,
75                                inet->sport, usin->sin_port, sk);
76         if (tmp < 0)
77                 return tmp;
78
79         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
80                 ip_rt_put(rt);
81                 return -ENETUNREACH;
82         }
83
84         if (inet->opt == NULL || !inet->opt->srr)
85                 daddr = rt->rt_dst;
86
87         if (inet->saddr == 0)
88                 inet->saddr = rt->rt_src;
89         inet->rcv_saddr = inet->saddr;
90
91         inet->dport = usin->sin_port;
92         inet->daddr = daddr;
93
94         inet_csk(sk)->icsk_ext_hdr_len = 0;
95         if (inet->opt != NULL)
96                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
97         /*
98          * Socket identity is still unknown (sport may be zero).
99          * However we set state to DCCP_REQUESTING and not releasing socket
100          * lock select source port, enter ourselves into the hash tables and
101          * complete initialization after this.
102          */
103         dccp_set_state(sk, DCCP_REQUESTING);
104         err = inet_hash_connect(&dccp_death_row, sk);
105         if (err != 0)
106                 goto failure;
107
108         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
109                                 sk);
110         if (err != 0)
111                 goto failure;
112
113         /* OK, now commit destination to socket.  */
114         sk_setup_caps(sk, &rt->u.dst);
115
116         dp->dccps_gar =
117                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
118                                                             inet->daddr,
119                                                             inet->sport,
120                                                             usin->sin_port);
121         dccp_update_gss(sk, dp->dccps_iss);
122
123         inet->id = dp->dccps_iss ^ jiffies;
124
125         err = dccp_connect(sk);
126         rt = NULL;
127         if (err != 0)
128                 goto failure;
129 out:
130         return err;
131 failure:
132         /*
133          * This unhashes the socket and releases the local port, if necessary.
134          */
135         dccp_set_state(sk, DCCP_CLOSED);
136         ip_rt_put(rt);
137         sk->sk_route_caps = 0;
138         inet->dport = 0;
139         goto out;
140 }
141
142 EXPORT_SYMBOL_GPL(dccp_v4_connect);
143
144 /*
145  * This routine does path mtu discovery as defined in RFC1191.
146  */
147 static inline void dccp_do_pmtu_discovery(struct sock *sk,
148                                           const struct iphdr *iph,
149                                           u32 mtu)
150 {
151         struct dst_entry *dst;
152         const struct inet_sock *inet = inet_sk(sk);
153         const struct dccp_sock *dp = dccp_sk(sk);
154
155         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
156          * send out by Linux are always < 576bytes so they should go through
157          * unfragmented).
158          */
159         if (sk->sk_state == DCCP_LISTEN)
160                 return;
161
162         /* We don't check in the destentry if pmtu discovery is forbidden
163          * on this route. We just assume that no packet_to_big packets
164          * are send back when pmtu discovery is not active.
165          * There is a small race when the user changes this flag in the
166          * route, but I think that's acceptable.
167          */
168         if ((dst = __sk_dst_check(sk, 0)) == NULL)
169                 return;
170
171         dst->ops->update_pmtu(dst, mtu);
172
173         /* Something is about to be wrong... Remember soft error
174          * for the case, if this connection will not able to recover.
175          */
176         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
177                 sk->sk_err_soft = EMSGSIZE;
178
179         mtu = dst_mtu(dst);
180
181         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
182             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
183                 dccp_sync_mss(sk, mtu);
184
185                 /*
186                  * From RFC 4340, sec. 14.1:
187                  *
188                  *      DCCP-Sync packets are the best choice for upward
189                  *      probing, since DCCP-Sync probes do not risk application
190                  *      data loss.
191                  */
192                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
193         } /* else let the usual retransmit timer handle it */
194 }
195
196 /*
197  * This routine is called by the ICMP module when it gets some sort of error
198  * condition. If err < 0 then the socket should be closed and the error
199  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
200  * After adjustment header points to the first 8 bytes of the tcp header. We
201  * need to find the appropriate port.
202  *
203  * The locking strategy used here is very "optimistic". When someone else
204  * accesses the socket the ICMP is just dropped and for some paths there is no
205  * check at all. A more general error queue to queue errors for later handling
206  * is probably better.
207  */
208 static void dccp_v4_err(struct sk_buff *skb, u32 info)
209 {
210         const struct iphdr *iph = (struct iphdr *)skb->data;
211         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
212                                                         (iph->ihl << 2));
213         struct dccp_sock *dp;
214         struct inet_sock *inet;
215         const int type = skb->h.icmph->type;
216         const int code = skb->h.icmph->code;
217         struct sock *sk;
218         __u64 seq;
219         int err;
220
221         if (skb->len < (iph->ihl << 2) + 8) {
222                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
223                 return;
224         }
225
226         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
227                          iph->saddr, dh->dccph_sport, inet_iif(skb));
228         if (sk == NULL) {
229                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
230                 return;
231         }
232
233         if (sk->sk_state == DCCP_TIME_WAIT) {
234                 inet_twsk_put(inet_twsk(sk));
235                 return;
236         }
237
238         bh_lock_sock(sk);
239         /* If too many ICMPs get dropped on busy
240          * servers this needs to be solved differently.
241          */
242         if (sock_owned_by_user(sk))
243                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
244
245         if (sk->sk_state == DCCP_CLOSED)
246                 goto out;
247
248         dp = dccp_sk(sk);
249         seq = dccp_hdr_seq(skb);
250         if (sk->sk_state != DCCP_LISTEN &&
251             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
252                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
253                 goto out;
254         }
255
256         switch (type) {
257         case ICMP_SOURCE_QUENCH:
258                 /* Just silently ignore these. */
259                 goto out;
260         case ICMP_PARAMETERPROB:
261                 err = EPROTO;
262                 break;
263         case ICMP_DEST_UNREACH:
264                 if (code > NR_ICMP_UNREACH)
265                         goto out;
266
267                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
268                         if (!sock_owned_by_user(sk))
269                                 dccp_do_pmtu_discovery(sk, iph, info);
270                         goto out;
271                 }
272
273                 err = icmp_err_convert[code].errno;
274                 break;
275         case ICMP_TIME_EXCEEDED:
276                 err = EHOSTUNREACH;
277                 break;
278         default:
279                 goto out;
280         }
281
282         switch (sk->sk_state) {
283                 struct request_sock *req , **prev;
284         case DCCP_LISTEN:
285                 if (sock_owned_by_user(sk))
286                         goto out;
287                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
288                                           iph->daddr, iph->saddr);
289                 if (!req)
290                         goto out;
291
292                 /*
293                  * ICMPs are not backlogged, hence we cannot get an established
294                  * socket here.
295                  */
296                 BUG_TRAP(!req->sk);
297
298                 if (seq != dccp_rsk(req)->dreq_iss) {
299                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
300                         goto out;
301                 }
302                 /*
303                  * Still in RESPOND, just remove it silently.
304                  * There is no good way to pass the error to the newly
305                  * created socket, and POSIX does not want network
306                  * errors returned from accept().
307                  */
308                 inet_csk_reqsk_queue_drop(sk, req, prev);
309                 goto out;
310
311         case DCCP_REQUESTING:
312         case DCCP_RESPOND:
313                 if (!sock_owned_by_user(sk)) {
314                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
315                         sk->sk_err = err;
316
317                         sk->sk_error_report(sk);
318
319                         dccp_done(sk);
320                 } else
321                         sk->sk_err_soft = err;
322                 goto out;
323         }
324
325         /* If we've already connected we will keep trying
326          * until we time out, or the user gives up.
327          *
328          * rfc1122 4.2.3.9 allows to consider as hard errors
329          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
330          * but it is obsoleted by pmtu discovery).
331          *
332          * Note, that in modern internet, where routing is unreliable
333          * and in each dark corner broken firewalls sit, sending random
334          * errors ordered by their masters even this two messages finally lose
335          * their original sense (even Linux sends invalid PORT_UNREACHs)
336          *
337          * Now we are in compliance with RFCs.
338          *                                                      --ANK (980905)
339          */
340
341         inet = inet_sk(sk);
342         if (!sock_owned_by_user(sk) && inet->recverr) {
343                 sk->sk_err = err;
344                 sk->sk_error_report(sk);
345         } else /* Only an error on timeout */
346                 sk->sk_err_soft = err;
347 out:
348         bh_unlock_sock(sk);
349         sock_put(sk);
350 }
351
352 static inline u16 dccp_v4_csum_finish(struct sk_buff *skb,
353                                       __be32 src, __be32 dst)
354 {
355         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
356 }
357
358 void dccp_v4_send_check(struct sock *sk, int unused, struct sk_buff *skb)
359 {
360         const struct inet_sock *inet = inet_sk(sk);
361         struct dccp_hdr *dh = dccp_hdr(skb);
362
363         dccp_csum_outgoing(skb);
364         dh->dccph_checksum = dccp_v4_csum_finish(skb, inet->saddr, inet->daddr);
365 }
366
367 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
368
369 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
370                                         const struct sk_buff *skb)
371 {
372         return secure_dccp_sequence_number(skb->nh.iph->daddr,
373                                            skb->nh.iph->saddr,
374                                            dccp_hdr(skb)->dccph_dport,
375                                            dccp_hdr(skb)->dccph_sport);
376 }
377
378 /*
379  * The three way handshake has completed - we got a valid ACK or DATAACK -
380  * now create the new socket.
381  *
382  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
383  */
384 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
385                                        struct request_sock *req,
386                                        struct dst_entry *dst)
387 {
388         struct inet_request_sock *ireq;
389         struct inet_sock *newinet;
390         struct dccp_sock *newdp;
391         struct sock *newsk;
392
393         if (sk_acceptq_is_full(sk))
394                 goto exit_overflow;
395
396         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
397                 goto exit;
398
399         newsk = dccp_create_openreq_child(sk, req, skb);
400         if (newsk == NULL)
401                 goto exit;
402
403         sk_setup_caps(newsk, dst);
404
405         newdp              = dccp_sk(newsk);
406         newinet            = inet_sk(newsk);
407         ireq               = inet_rsk(req);
408         newinet->daddr     = ireq->rmt_addr;
409         newinet->rcv_saddr = ireq->loc_addr;
410         newinet->saddr     = ireq->loc_addr;
411         newinet->opt       = ireq->opt;
412         ireq->opt          = NULL;
413         newinet->mc_index  = inet_iif(skb);
414         newinet->mc_ttl    = skb->nh.iph->ttl;
415         newinet->id        = jiffies;
416
417         dccp_sync_mss(newsk, dst_mtu(dst));
418
419         __inet_hash(&dccp_hashinfo, newsk, 0);
420         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
421
422         return newsk;
423
424 exit_overflow:
425         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
426 exit:
427         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
428         dst_release(dst);
429         return NULL;
430 }
431
432 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
433
434 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
435 {
436         const struct dccp_hdr *dh = dccp_hdr(skb);
437         const struct iphdr *iph = skb->nh.iph;
438         struct sock *nsk;
439         struct request_sock **prev;
440         /* Find possible connection requests. */
441         struct request_sock *req = inet_csk_search_req(sk, &prev,
442                                                        dh->dccph_sport,
443                                                        iph->saddr, iph->daddr);
444         if (req != NULL)
445                 return dccp_check_req(sk, skb, req, prev);
446
447         nsk = inet_lookup_established(&dccp_hashinfo,
448                                       iph->saddr, dh->dccph_sport,
449                                       iph->daddr, dh->dccph_dport,
450                                       inet_iif(skb));
451         if (nsk != NULL) {
452                 if (nsk->sk_state != DCCP_TIME_WAIT) {
453                         bh_lock_sock(nsk);
454                         return nsk;
455                 }
456                 inet_twsk_put(inet_twsk(nsk));
457                 return NULL;
458         }
459
460         return sk;
461 }
462
463 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
464                                            struct sk_buff *skb)
465 {
466         struct rtable *rt;
467         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
468                             .nl_u = { .ip4_u =
469                                       { .daddr = skb->nh.iph->saddr,
470                                         .saddr = skb->nh.iph->daddr,
471                                         .tos = RT_CONN_FLAGS(sk) } },
472                             .proto = sk->sk_protocol,
473                             .uli_u = { .ports =
474                                        { .sport = dccp_hdr(skb)->dccph_dport,
475                                          .dport = dccp_hdr(skb)->dccph_sport }
476                                      }
477                           };
478
479         security_skb_classify_flow(skb, &fl);
480         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
481                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
482                 return NULL;
483         }
484
485         return &rt->u.dst;
486 }
487
488 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
489                                  struct dst_entry *dst)
490 {
491         int err = -1;
492         struct sk_buff *skb;
493
494         /* First, grab a route. */
495         
496         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
497                 goto out;
498
499         skb = dccp_make_response(sk, dst, req);
500         if (skb != NULL) {
501                 const struct inet_request_sock *ireq = inet_rsk(req);
502                 struct dccp_hdr *dh = dccp_hdr(skb);
503
504                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
505                                                               ireq->rmt_addr);
506                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
507                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
508                                             ireq->rmt_addr,
509                                             ireq->opt);
510                 if (err == NET_XMIT_CN)
511                         err = 0;
512         }
513
514 out:
515         dst_release(dst);
516         return err;
517 }
518
519 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
520 {
521         int err;
522         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
523         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
524                                        sizeof(struct dccp_hdr_ext) +
525                                        sizeof(struct dccp_hdr_reset);
526         struct sk_buff *skb;
527         struct dst_entry *dst;
528         u64 seqno;
529
530         /* Never send a reset in response to a reset. */
531         if (rxdh->dccph_type == DCCP_PKT_RESET)
532                 return;
533
534         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
535                 return;
536
537         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
538         if (dst == NULL)
539                 return;
540
541         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header,
542                         GFP_ATOMIC);
543         if (skb == NULL)
544                 goto out;
545
546         /* Reserve space for headers. */
547         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
548         skb->dst = dst_clone(dst);
549
550         dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
551
552         /* Build DCCP header and checksum it. */
553         dh->dccph_type     = DCCP_PKT_RESET;
554         dh->dccph_sport    = rxdh->dccph_dport;
555         dh->dccph_dport    = rxdh->dccph_sport;
556         dh->dccph_doff     = dccp_hdr_reset_len / 4;
557         dh->dccph_x        = 1;
558         dccp_hdr_reset(skb)->dccph_reset_code =
559                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
560
561         /* See "8.3.1. Abnormal Termination" in RFC 4340 */
562         seqno = 0;
563         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
564                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
565
566         dccp_hdr_set_seq(dh, seqno);
567         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
568                          DCCP_SKB_CB(rxskb)->dccpd_seq);
569
570         dccp_csum_outgoing(skb);
571         dh->dccph_checksum = dccp_v4_csum_finish(skb, rxskb->nh.iph->saddr,
572                                                       rxskb->nh.iph->daddr);
573
574         bh_lock_sock(dccp_v4_ctl_socket->sk);
575         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
576                                     rxskb->nh.iph->daddr,
577                                     rxskb->nh.iph->saddr, NULL);
578         bh_unlock_sock(dccp_v4_ctl_socket->sk);
579
580         if (err == NET_XMIT_CN || err == 0) {
581                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
582                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
583         }
584 out:
585          dst_release(dst);
586 }
587
588 static void dccp_v4_reqsk_destructor(struct request_sock *req)
589 {
590         kfree(inet_rsk(req)->opt);
591 }
592
593 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
594         .family         = PF_INET,
595         .obj_size       = sizeof(struct dccp_request_sock),
596         .rtx_syn_ack    = dccp_v4_send_response,
597         .send_ack       = dccp_reqsk_send_ack,
598         .destructor     = dccp_v4_reqsk_destructor,
599         .send_reset     = dccp_v4_ctl_send_reset,
600 };
601
602 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
603 {
604         struct inet_request_sock *ireq;
605         struct request_sock *req;
606         struct dccp_request_sock *dreq;
607         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
608         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
609         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
610
611         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
612         if (((struct rtable *)skb->dst)->rt_flags &
613             (RTCF_BROADCAST | RTCF_MULTICAST)) {
614                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
615                 goto drop;
616         }
617
618         if (dccp_bad_service_code(sk, service)) {
619                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
620                 goto drop;
621         }
622         /*
623          * TW buckets are converted to open requests without
624          * limitations, they conserve resources and peer is
625          * evidently real one.
626          */
627         if (inet_csk_reqsk_queue_is_full(sk))
628                 goto drop;
629
630         /*
631          * Accept backlog is full. If we have already queued enough
632          * of warm entries in syn queue, drop request. It is better than
633          * clogging syn queue with openreqs with exponentially increasing
634          * timeout.
635          */
636         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
637                 goto drop;
638
639         req = reqsk_alloc(&dccp_request_sock_ops);
640         if (req == NULL)
641                 goto drop;
642
643         if (dccp_parse_options(sk, skb))
644                 goto drop_and_free;
645
646         dccp_reqsk_init(req, skb);
647
648         if (security_inet_conn_request(sk, skb, req))
649                 goto drop_and_free;
650
651         ireq = inet_rsk(req);
652         ireq->loc_addr = skb->nh.iph->daddr;
653         ireq->rmt_addr = skb->nh.iph->saddr;
654         req->rcv_wnd    = dccp_feat_default_sequence_window;
655         ireq->opt       = NULL;
656
657         /* 
658          * Step 3: Process LISTEN state
659          *
660          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
661          *
662          * In fact we defer setting S.GSR, S.SWL, S.SWH to
663          * dccp_create_openreq_child.
664          */
665         dreq = dccp_rsk(req);
666         dreq->dreq_isr     = dcb->dccpd_seq;
667         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
668         dreq->dreq_service = service;
669
670         if (dccp_v4_send_response(sk, req, NULL))
671                 goto drop_and_free;
672
673         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
674         return 0;
675
676 drop_and_free:
677         reqsk_free(req);
678 drop:
679         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
680         dcb->dccpd_reset_code = reset_code;
681         return -1;
682 }
683
684 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
685
686 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
687 {
688         struct dccp_hdr *dh = dccp_hdr(skb);
689
690         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
691                 if (dccp_rcv_established(sk, skb, dh, skb->len))
692                         goto reset;
693                 return 0;
694         }
695
696         /*
697          *  Step 3: Process LISTEN state
698          *       If P.type == Request or P contains a valid Init Cookie option,
699          *            (* Must scan the packet's options to check for Init
700          *               Cookies.  Only Init Cookies are processed here,
701          *               however; other options are processed in Step 8.  This
702          *               scan need only be performed if the endpoint uses Init
703          *               Cookies *)
704          *            (* Generate a new socket and switch to that socket *)
705          *            Set S := new socket for this port pair
706          *            S.state = RESPOND
707          *            Choose S.ISS (initial seqno) or set from Init Cookies
708          *            Initialize S.GAR := S.ISS
709          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
710          *            Continue with S.state == RESPOND
711          *            (* A Response packet will be generated in Step 11 *)
712          *       Otherwise,
713          *            Generate Reset(No Connection) unless P.type == Reset
714          *            Drop packet and return
715          *
716          * NOTE: the check for the packet types is done in
717          *       dccp_rcv_state_process
718          */
719         if (sk->sk_state == DCCP_LISTEN) {
720                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
721
722                 if (nsk == NULL)
723                         goto discard;
724
725                 if (nsk != sk) {
726                         if (dccp_child_process(sk, nsk, skb))
727                                 goto reset;
728                         return 0;
729                 }
730         }
731
732         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
733                 goto reset;
734         return 0;
735
736 reset:
737         dccp_v4_ctl_send_reset(skb);
738 discard:
739         kfree_skb(skb);
740         return 0;
741 }
742
743 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
744
745 int dccp_invalid_packet(struct sk_buff *skb)
746 {
747         const struct dccp_hdr *dh;
748         unsigned int cscov;
749
750         if (skb->pkt_type != PACKET_HOST)
751                 return 1;
752
753         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
754                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
755                 return 1;
756         }
757
758         dh = dccp_hdr(skb);
759
760         /* If the packet type is not understood, drop packet and return */
761         if (dh->dccph_type >= DCCP_PKT_INVALID) {
762                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
763                 return 1;
764         }
765
766         /*
767          * If P.Data Offset is too small for packet type, or too large for
768          * packet, drop packet and return
769          */
770         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
771                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
772                                             "too small 1\n",
773                                dh->dccph_doff);
774                 return 1;
775         }
776
777         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
778                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
779                                             "too small 2\n",
780                                dh->dccph_doff);
781                 return 1;
782         }
783
784         dh = dccp_hdr(skb);
785
786         /*
787          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
788          * has short sequence numbers), drop packet and return
789          */
790         if (dh->dccph_x == 0 &&
791             dh->dccph_type != DCCP_PKT_DATA &&
792             dh->dccph_type != DCCP_PKT_ACK &&
793             dh->dccph_type != DCCP_PKT_DATAACK) {
794                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
795                                             "nor DataAck and P.X == 0\n",
796                                dccp_packet_name(dh->dccph_type));
797                 return 1;
798         }
799
800         /*
801          * If P.CsCov is too large for the packet size, drop packet and return.
802          * This must come _before_ checksumming (not as RFC 4340 suggests).
803          */
804         cscov = dccp_csum_coverage(skb);
805         if (cscov > skb->len) {
806                 LIMIT_NETDEBUG(KERN_WARNING
807                                "DCCP: P.CsCov %u exceeds packet length %d\n",
808                                dh->dccph_cscov, skb->len);
809                 return 1;
810         }
811
812         /* If header checksum is incorrect, drop packet and return.
813          * (This step is completed in the AF-dependent functions.) */
814         skb->csum = skb_checksum(skb, 0, cscov, 0);
815
816         return 0;
817 }
818
819 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
820
821 /* this is called when real data arrives */
822 static int dccp_v4_rcv(struct sk_buff *skb)
823 {
824         const struct dccp_hdr *dh;
825         struct sock *sk;
826         int min_cov;
827
828         /* Step 1: Check header basics */
829
830         if (dccp_invalid_packet(skb))
831                 goto discard_it;
832
833         /* Step 1: If header checksum is incorrect, drop packet and return */
834         if (dccp_v4_csum_finish(skb, skb->nh.iph->saddr, skb->nh.iph->daddr)) {
835                 LIMIT_NETDEBUG(KERN_WARNING
836                                "%s: dropped packet with invalid checksum\n",
837                                __FUNCTION__);
838                 goto discard_it;
839         }
840
841         dh = dccp_hdr(skb);
842
843         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
844         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
845
846         dccp_pr_debug("%8.8s "
847                       "src=%u.%u.%u.%u@%-5d "
848                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
849                       dccp_packet_name(dh->dccph_type),
850                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
851                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
852                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
853
854         if (dccp_packet_without_ack(skb)) {
855                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
856                 dccp_pr_debug_cat("\n");
857         } else {
858                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
859                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
860                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
861         }
862
863         /* Step 2:
864          *      Look up flow ID in table and get corresponding socket */
865         sk = __inet_lookup(&dccp_hashinfo,
866                            skb->nh.iph->saddr, dh->dccph_sport,
867                            skb->nh.iph->daddr, dh->dccph_dport,
868                            inet_iif(skb));
869
870         /* 
871          * Step 2:
872          *      If no socket ...
873          */
874         if (sk == NULL) {
875                 dccp_pr_debug("failed to look up flow ID in table and "
876                               "get corresponding socket\n");
877                 goto no_dccp_socket;
878         }
879
880         /* 
881          * Step 2:
882          *      ... or S.state == TIMEWAIT,
883          *              Generate Reset(No Connection) unless P.type == Reset
884          *              Drop packet and return
885          */
886         if (sk->sk_state == DCCP_TIME_WAIT) {
887                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
888                 inet_twsk_put(inet_twsk(sk));
889                 goto no_dccp_socket;
890         }
891
892         /*
893          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
894          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
895          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
896          */
897         min_cov = dccp_sk(sk)->dccps_pcrlen;
898         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
899                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
900                               dh->dccph_cscov, min_cov);
901                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
902                  *         options (Section 11.7) with Drop Code 0, Protocol
903                  *         Constraints."                                     */
904                 goto discard_and_relse;
905         }
906
907         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
908                 goto discard_and_relse;
909         nf_reset(skb);
910
911         return sk_receive_skb(sk, skb);
912
913 no_dccp_socket:
914         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
915                 goto discard_it;
916         /*
917          * Step 2:
918          *      If no socket ...
919          *              Generate Reset(No Connection) unless P.type == Reset
920          *              Drop packet and return
921          */
922         if (dh->dccph_type != DCCP_PKT_RESET) {
923                 DCCP_SKB_CB(skb)->dccpd_reset_code =
924                                         DCCP_RESET_CODE_NO_CONNECTION;
925                 dccp_v4_ctl_send_reset(skb);
926         }
927
928 discard_it:
929         kfree_skb(skb);
930         return 0;
931
932 discard_and_relse:
933         sock_put(sk);
934         goto discard_it;
935 }
936
937 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
938         .queue_xmit        = ip_queue_xmit,
939         .send_check        = dccp_v4_send_check,
940         .rebuild_header    = inet_sk_rebuild_header,
941         .conn_request      = dccp_v4_conn_request,
942         .syn_recv_sock     = dccp_v4_request_recv_sock,
943         .net_header_len    = sizeof(struct iphdr),
944         .setsockopt        = ip_setsockopt,
945         .getsockopt        = ip_getsockopt,
946         .addr2sockaddr     = inet_csk_addr2sockaddr,
947         .sockaddr_len      = sizeof(struct sockaddr_in),
948 #ifdef CONFIG_COMPAT
949         .compat_setsockopt = compat_ip_setsockopt,
950         .compat_getsockopt = compat_ip_getsockopt,
951 #endif
952 };
953
954 static int dccp_v4_init_sock(struct sock *sk)
955 {
956         static __u8 dccp_v4_ctl_sock_initialized;
957         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
958
959         if (err == 0) {
960                 if (unlikely(!dccp_v4_ctl_sock_initialized))
961                         dccp_v4_ctl_sock_initialized = 1;
962                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
963         }
964
965         return err;
966 }
967
968 static struct timewait_sock_ops dccp_timewait_sock_ops = {
969         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
970 };
971
972 static struct proto dccp_v4_prot = {
973         .name                   = "DCCP",
974         .owner                  = THIS_MODULE,
975         .close                  = dccp_close,
976         .connect                = dccp_v4_connect,
977         .disconnect             = dccp_disconnect,
978         .ioctl                  = dccp_ioctl,
979         .init                   = dccp_v4_init_sock,
980         .setsockopt             = dccp_setsockopt,
981         .getsockopt             = dccp_getsockopt,
982         .sendmsg                = dccp_sendmsg,
983         .recvmsg                = dccp_recvmsg,
984         .backlog_rcv            = dccp_v4_do_rcv,
985         .hash                   = dccp_hash,
986         .unhash                 = dccp_unhash,
987         .accept                 = inet_csk_accept,
988         .get_port               = dccp_v4_get_port,
989         .shutdown               = dccp_shutdown,
990         .destroy                = dccp_destroy_sock,
991         .orphan_count           = &dccp_orphan_count,
992         .max_header             = MAX_DCCP_HEADER,
993         .obj_size               = sizeof(struct dccp_sock),
994         .rsk_prot               = &dccp_request_sock_ops,
995         .twsk_prot              = &dccp_timewait_sock_ops,
996 #ifdef CONFIG_COMPAT
997         .compat_setsockopt      = compat_dccp_setsockopt,
998         .compat_getsockopt      = compat_dccp_getsockopt,
999 #endif
1000 };
1001
1002 static struct net_protocol dccp_v4_protocol = {
1003         .handler        = dccp_v4_rcv,
1004         .err_handler    = dccp_v4_err,
1005         .no_policy      = 1,
1006 };
1007
1008 static const struct proto_ops inet_dccp_ops = {
1009         .family            = PF_INET,
1010         .owner             = THIS_MODULE,
1011         .release           = inet_release,
1012         .bind              = inet_bind,
1013         .connect           = inet_stream_connect,
1014         .socketpair        = sock_no_socketpair,
1015         .accept            = inet_accept,
1016         .getname           = inet_getname,
1017         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1018         .poll              = dccp_poll,
1019         .ioctl             = inet_ioctl,
1020         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1021         .listen            = inet_dccp_listen,
1022         .shutdown          = inet_shutdown,
1023         .setsockopt        = sock_common_setsockopt,
1024         .getsockopt        = sock_common_getsockopt,
1025         .sendmsg           = inet_sendmsg,
1026         .recvmsg           = sock_common_recvmsg,
1027         .mmap              = sock_no_mmap,
1028         .sendpage          = sock_no_sendpage,
1029 #ifdef CONFIG_COMPAT
1030         .compat_setsockopt = compat_sock_common_setsockopt,
1031         .compat_getsockopt = compat_sock_common_getsockopt,
1032 #endif
1033 };
1034
1035 static struct inet_protosw dccp_v4_protosw = {
1036         .type           = SOCK_DCCP,
1037         .protocol       = IPPROTO_DCCP,
1038         .prot           = &dccp_v4_prot,
1039         .ops            = &inet_dccp_ops,
1040         .capability     = -1,
1041         .no_check       = 0,
1042         .flags          = INET_PROTOSW_ICSK,
1043 };
1044
1045 static int __init dccp_v4_init(void)
1046 {
1047         int err = proto_register(&dccp_v4_prot, 1);
1048
1049         if (err != 0)
1050                 goto out;
1051
1052         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1053         if (err != 0)
1054                 goto out_proto_unregister;
1055
1056         inet_register_protosw(&dccp_v4_protosw);
1057
1058         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1059                                        SOCK_DCCP, IPPROTO_DCCP);
1060         if (err)
1061                 goto out_unregister_protosw;
1062 out:
1063         return err;
1064 out_unregister_protosw:
1065         inet_unregister_protosw(&dccp_v4_protosw);
1066         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1067 out_proto_unregister:
1068         proto_unregister(&dccp_v4_prot);
1069         goto out;
1070 }
1071
1072 static void __exit dccp_v4_exit(void)
1073 {
1074         inet_unregister_protosw(&dccp_v4_protosw);
1075         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1076         proto_unregister(&dccp_v4_prot);
1077 }
1078
1079 module_init(dccp_v4_init);
1080 module_exit(dccp_v4_exit);
1081
1082 /*
1083  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1084  * values directly, Also cover the case where the protocol is not specified,
1085  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1086  */
1087 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1088 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1089 MODULE_LICENSE("GPL");
1090 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1091 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");