Merge branches 'release', 'cpuidle-2.6.25' and 'idle' into release
[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  * 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 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
42 {
43         struct inet_sock *inet = inet_sk(sk);
44         struct dccp_sock *dp = dccp_sk(sk);
45         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
46         struct rtable *rt;
47         __be32 daddr, nexthop;
48         int tmp;
49         int err;
50
51         dp->dccps_role = DCCP_ROLE_CLIENT;
52
53         if (addr_len < sizeof(struct sockaddr_in))
54                 return -EINVAL;
55
56         if (usin->sin_family != AF_INET)
57                 return -EAFNOSUPPORT;
58
59         nexthop = daddr = usin->sin_addr.s_addr;
60         if (inet->opt != NULL && inet->opt->srr) {
61                 if (daddr == 0)
62                         return -EINVAL;
63                 nexthop = inet->opt->faddr;
64         }
65
66         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
67                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
68                                IPPROTO_DCCP,
69                                inet->sport, usin->sin_port, sk, 1);
70         if (tmp < 0)
71                 return tmp;
72
73         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
74                 ip_rt_put(rt);
75                 return -ENETUNREACH;
76         }
77
78         if (inet->opt == NULL || !inet->opt->srr)
79                 daddr = rt->rt_dst;
80
81         if (inet->saddr == 0)
82                 inet->saddr = rt->rt_src;
83         inet->rcv_saddr = inet->saddr;
84
85         inet->dport = usin->sin_port;
86         inet->daddr = daddr;
87
88         inet_csk(sk)->icsk_ext_hdr_len = 0;
89         if (inet->opt != NULL)
90                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
91         /*
92          * Socket identity is still unknown (sport may be zero).
93          * However we set state to DCCP_REQUESTING and not releasing socket
94          * lock select source port, enter ourselves into the hash tables and
95          * complete initialization after this.
96          */
97         dccp_set_state(sk, DCCP_REQUESTING);
98         err = inet_hash_connect(&dccp_death_row, sk);
99         if (err != 0)
100                 goto failure;
101
102         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
103                                 sk);
104         if (err != 0)
105                 goto failure;
106
107         /* OK, now commit destination to socket.  */
108         sk_setup_caps(sk, &rt->u.dst);
109
110         dp->dccps_iss = secure_dccp_sequence_number(inet->saddr, inet->daddr,
111                                                     inet->sport, inet->dport);
112         inet->id = dp->dccps_iss ^ jiffies;
113
114         err = dccp_connect(sk);
115         rt = NULL;
116         if (err != 0)
117                 goto failure;
118 out:
119         return err;
120 failure:
121         /*
122          * This unhashes the socket and releases the local port, if necessary.
123          */
124         dccp_set_state(sk, DCCP_CLOSED);
125         ip_rt_put(rt);
126         sk->sk_route_caps = 0;
127         inet->dport = 0;
128         goto out;
129 }
130
131 EXPORT_SYMBOL_GPL(dccp_v4_connect);
132
133 /*
134  * This routine does path mtu discovery as defined in RFC1191.
135  */
136 static inline void dccp_do_pmtu_discovery(struct sock *sk,
137                                           const struct iphdr *iph,
138                                           u32 mtu)
139 {
140         struct dst_entry *dst;
141         const struct inet_sock *inet = inet_sk(sk);
142         const struct dccp_sock *dp = dccp_sk(sk);
143
144         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
145          * send out by Linux are always < 576bytes so they should go through
146          * unfragmented).
147          */
148         if (sk->sk_state == DCCP_LISTEN)
149                 return;
150
151         /* We don't check in the destentry if pmtu discovery is forbidden
152          * on this route. We just assume that no packet_to_big packets
153          * are send back when pmtu discovery is not active.
154          * There is a small race when the user changes this flag in the
155          * route, but I think that's acceptable.
156          */
157         if ((dst = __sk_dst_check(sk, 0)) == NULL)
158                 return;
159
160         dst->ops->update_pmtu(dst, mtu);
161
162         /* Something is about to be wrong... Remember soft error
163          * for the case, if this connection will not able to recover.
164          */
165         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
166                 sk->sk_err_soft = EMSGSIZE;
167
168         mtu = dst_mtu(dst);
169
170         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
171             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
172                 dccp_sync_mss(sk, mtu);
173
174                 /*
175                  * From RFC 4340, sec. 14.1:
176                  *
177                  *      DCCP-Sync packets are the best choice for upward
178                  *      probing, since DCCP-Sync probes do not risk application
179                  *      data loss.
180                  */
181                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
182         } /* else let the usual retransmit timer handle it */
183 }
184
185 /*
186  * This routine is called by the ICMP module when it gets some sort of error
187  * condition. If err < 0 then the socket should be closed and the error
188  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
189  * After adjustment header points to the first 8 bytes of the tcp header. We
190  * need to find the appropriate port.
191  *
192  * The locking strategy used here is very "optimistic". When someone else
193  * accesses the socket the ICMP is just dropped and for some paths there is no
194  * check at all. A more general error queue to queue errors for later handling
195  * is probably better.
196  */
197 static void dccp_v4_err(struct sk_buff *skb, u32 info)
198 {
199         const struct iphdr *iph = (struct iphdr *)skb->data;
200         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
201                                                         (iph->ihl << 2));
202         struct dccp_sock *dp;
203         struct inet_sock *inet;
204         const int type = icmp_hdr(skb)->type;
205         const int code = icmp_hdr(skb)->code;
206         struct sock *sk;
207         __u64 seq;
208         int err;
209
210         if (skb->len < (iph->ihl << 2) + 8) {
211                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
212                 return;
213         }
214
215         sk = inet_lookup(&init_net, &dccp_hashinfo, 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(&init_net, &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 sock *sk,
450                                            struct sk_buff *skb)
451 {
452         struct rtable *rt;
453         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->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(&init_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                                  struct dst_entry *dst)
476 {
477         int err = -1;
478         struct sk_buff *skb;
479
480         /* First, grab a route. */
481
482         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
483                 goto out;
484
485         skb = dccp_make_response(sk, dst, req);
486         if (skb != NULL) {
487                 const struct inet_request_sock *ireq = inet_rsk(req);
488                 struct dccp_hdr *dh = dccp_hdr(skb);
489
490                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
491                                                               ireq->rmt_addr);
492                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
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
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 (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
516                 return;
517
518         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
519         if (dst == NULL)
520                 return;
521
522         skb = dccp_ctl_make_reset(dccp_v4_ctl_socket, 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(dccp_v4_ctl_socket->sk);
532         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
533                                     rxiph->daddr, rxiph->saddr, NULL);
534         bh_unlock_sock(dccp_v4_ctl_socket->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 (((struct rtable *)skb->dst)->rt_flags &
568             (RTCF_BROADCAST | RTCF_MULTICAST))
569                 return 0;       /* discard, don't send a reset here */
570
571         if (dccp_bad_service_code(sk, service)) {
572                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
573                 goto drop;
574         }
575         /*
576          * TW buckets are converted to open requests without
577          * limitations, they conserve resources and peer is
578          * evidently real one.
579          */
580         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
581         if (inet_csk_reqsk_queue_is_full(sk))
582                 goto drop;
583
584         /*
585          * Accept backlog is full. If we have already queued enough
586          * of warm entries in syn queue, drop request. It is better than
587          * clogging syn queue with openreqs with exponentially increasing
588          * timeout.
589          */
590         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
591                 goto drop;
592
593         req = reqsk_alloc(&dccp_request_sock_ops);
594         if (req == NULL)
595                 goto drop;
596
597         dccp_reqsk_init(req, skb);
598
599         dreq = dccp_rsk(req);
600         if (dccp_parse_options(sk, dreq, skb))
601                 goto drop_and_free;
602
603         if (security_inet_conn_request(sk, skb, req))
604                 goto drop_and_free;
605
606         ireq = inet_rsk(req);
607         ireq->loc_addr = ip_hdr(skb)->daddr;
608         ireq->rmt_addr = ip_hdr(skb)->saddr;
609         ireq->opt       = NULL;
610
611         /*
612          * Step 3: Process LISTEN state
613          *
614          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
615          *
616          * In fact we defer setting S.GSR, S.SWL, S.SWH to
617          * dccp_create_openreq_child.
618          */
619         dreq->dreq_isr     = dcb->dccpd_seq;
620         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
621         dreq->dreq_service = service;
622
623         if (dccp_v4_send_response(sk, req, NULL))
624                 goto drop_and_free;
625
626         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
627         return 0;
628
629 drop_and_free:
630         reqsk_free(req);
631 drop:
632         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
633         return -1;
634 }
635
636 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
637
638 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
639 {
640         struct dccp_hdr *dh = dccp_hdr(skb);
641
642         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
643                 if (dccp_rcv_established(sk, skb, dh, skb->len))
644                         goto reset;
645                 return 0;
646         }
647
648         /*
649          *  Step 3: Process LISTEN state
650          *       If P.type == Request or P contains a valid Init Cookie option,
651          *            (* Must scan the packet's options to check for Init
652          *               Cookies.  Only Init Cookies are processed here,
653          *               however; other options are processed in Step 8.  This
654          *               scan need only be performed if the endpoint uses Init
655          *               Cookies *)
656          *            (* Generate a new socket and switch to that socket *)
657          *            Set S := new socket for this port pair
658          *            S.state = RESPOND
659          *            Choose S.ISS (initial seqno) or set from Init Cookies
660          *            Initialize S.GAR := S.ISS
661          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
662          *            Continue with S.state == RESPOND
663          *            (* A Response packet will be generated in Step 11 *)
664          *       Otherwise,
665          *            Generate Reset(No Connection) unless P.type == Reset
666          *            Drop packet and return
667          *
668          * NOTE: the check for the packet types is done in
669          *       dccp_rcv_state_process
670          */
671         if (sk->sk_state == DCCP_LISTEN) {
672                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
673
674                 if (nsk == NULL)
675                         goto discard;
676
677                 if (nsk != sk) {
678                         if (dccp_child_process(sk, nsk, skb))
679                                 goto reset;
680                         return 0;
681                 }
682         }
683
684         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
685                 goto reset;
686         return 0;
687
688 reset:
689         dccp_v4_ctl_send_reset(sk, skb);
690 discard:
691         kfree_skb(skb);
692         return 0;
693 }
694
695 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
696
697 /**
698  *      dccp_invalid_packet  -  check for malformed packets
699  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
700  *      Packets that fail these checks are ignored and do not receive Resets.
701  */
702 int dccp_invalid_packet(struct sk_buff *skb)
703 {
704         const struct dccp_hdr *dh;
705         unsigned int cscov;
706
707         if (skb->pkt_type != PACKET_HOST)
708                 return 1;
709
710         /* If the packet is shorter than 12 bytes, drop packet and return */
711         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
712                 DCCP_WARN("pskb_may_pull failed\n");
713                 return 1;
714         }
715
716         dh = dccp_hdr(skb);
717
718         /* If P.type is not understood, drop packet and return */
719         if (dh->dccph_type >= DCCP_PKT_INVALID) {
720                 DCCP_WARN("invalid packet type\n");
721                 return 1;
722         }
723
724         /*
725          * If P.Data Offset is too small for packet type, drop packet and return
726          */
727         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
728                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
729                 return 1;
730         }
731         /*
732          * If P.Data Offset is too too large for packet, drop packet and return
733          */
734         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
735                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
736                 return 1;
737         }
738
739         /*
740          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
741          * has short sequence numbers), drop packet and return
742          */
743         if (dh->dccph_type >= DCCP_PKT_DATA    &&
744             dh->dccph_type <= DCCP_PKT_DATAACK && dh->dccph_x == 0)  {
745                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
746                           dccp_packet_name(dh->dccph_type));
747                 return 1;
748         }
749
750         /*
751          * If P.CsCov is too large for the packet size, drop packet and return.
752          * This must come _before_ checksumming (not as RFC 4340 suggests).
753          */
754         cscov = dccp_csum_coverage(skb);
755         if (cscov > skb->len) {
756                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
757                           dh->dccph_cscov, skb->len);
758                 return 1;
759         }
760
761         /* If header checksum is incorrect, drop packet and return.
762          * (This step is completed in the AF-dependent functions.) */
763         skb->csum = skb_checksum(skb, 0, cscov, 0);
764
765         return 0;
766 }
767
768 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
769
770 /* this is called when real data arrives */
771 static int dccp_v4_rcv(struct sk_buff *skb)
772 {
773         const struct dccp_hdr *dh;
774         const struct iphdr *iph;
775         struct sock *sk;
776         int min_cov;
777
778         /* Step 1: Check header basics */
779
780         if (dccp_invalid_packet(skb))
781                 goto discard_it;
782
783         iph = ip_hdr(skb);
784         /* Step 1: If header checksum is incorrect, drop packet and return */
785         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
786                 DCCP_WARN("dropped packet with invalid checksum\n");
787                 goto discard_it;
788         }
789
790         dh = dccp_hdr(skb);
791
792         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
793         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
794
795         dccp_pr_debug("%8.8s "
796                       "src=%u.%u.%u.%u@%-5d "
797                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
798                       dccp_packet_name(dh->dccph_type),
799                       NIPQUAD(iph->saddr), ntohs(dh->dccph_sport),
800                       NIPQUAD(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(&init_net, &dccp_hashinfo,
815                            iph->saddr, dh->dccph_sport,
816                            iph->daddr, dh->dccph_dport, inet_iif(skb));
817         /*
818          * Step 2:
819          *      If no socket ...
820          */
821         if (sk == NULL) {
822                 dccp_pr_debug("failed to look up flow ID in table and "
823                               "get corresponding socket\n");
824                 goto no_dccp_socket;
825         }
826
827         /*
828          * Step 2:
829          *      ... or S.state == TIMEWAIT,
830          *              Generate Reset(No Connection) unless P.type == Reset
831          *              Drop packet and return
832          */
833         if (sk->sk_state == DCCP_TIME_WAIT) {
834                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
835                 inet_twsk_put(inet_twsk(sk));
836                 goto no_dccp_socket;
837         }
838
839         /*
840          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
841          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
842          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
843          */
844         min_cov = dccp_sk(sk)->dccps_pcrlen;
845         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
846                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
847                               dh->dccph_cscov, min_cov);
848                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
849                  *         options (Section 11.7) with Drop Code 0, Protocol
850                  *         Constraints."                                     */
851                 goto discard_and_relse;
852         }
853
854         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
855                 goto discard_and_relse;
856         nf_reset(skb);
857
858         return sk_receive_skb(sk, skb, 1);
859
860 no_dccp_socket:
861         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
862                 goto discard_it;
863         /*
864          * Step 2:
865          *      If no socket ...
866          *              Generate Reset(No Connection) unless P.type == Reset
867          *              Drop packet and return
868          */
869         if (dh->dccph_type != DCCP_PKT_RESET) {
870                 DCCP_SKB_CB(skb)->dccpd_reset_code =
871                                         DCCP_RESET_CODE_NO_CONNECTION;
872                 dccp_v4_ctl_send_reset(sk, skb);
873         }
874
875 discard_it:
876         kfree_skb(skb);
877         return 0;
878
879 discard_and_relse:
880         sock_put(sk);
881         goto discard_it;
882 }
883
884 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
885         .queue_xmit        = ip_queue_xmit,
886         .send_check        = dccp_v4_send_check,
887         .rebuild_header    = inet_sk_rebuild_header,
888         .conn_request      = dccp_v4_conn_request,
889         .syn_recv_sock     = dccp_v4_request_recv_sock,
890         .net_header_len    = sizeof(struct iphdr),
891         .setsockopt        = ip_setsockopt,
892         .getsockopt        = ip_getsockopt,
893         .addr2sockaddr     = inet_csk_addr2sockaddr,
894         .sockaddr_len      = sizeof(struct sockaddr_in),
895         .bind_conflict     = inet_csk_bind_conflict,
896 #ifdef CONFIG_COMPAT
897         .compat_setsockopt = compat_ip_setsockopt,
898         .compat_getsockopt = compat_ip_getsockopt,
899 #endif
900 };
901
902 static int dccp_v4_init_sock(struct sock *sk)
903 {
904         static __u8 dccp_v4_ctl_sock_initialized;
905         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
906
907         if (err == 0) {
908                 if (unlikely(!dccp_v4_ctl_sock_initialized))
909                         dccp_v4_ctl_sock_initialized = 1;
910                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
911         }
912
913         return err;
914 }
915
916 static struct timewait_sock_ops dccp_timewait_sock_ops = {
917         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
918 };
919
920 DEFINE_PROTO_INUSE(dccp_v4)
921
922 static struct proto dccp_v4_prot = {
923         .name                   = "DCCP",
924         .owner                  = THIS_MODULE,
925         .close                  = dccp_close,
926         .connect                = dccp_v4_connect,
927         .disconnect             = dccp_disconnect,
928         .ioctl                  = dccp_ioctl,
929         .init                   = dccp_v4_init_sock,
930         .setsockopt             = dccp_setsockopt,
931         .getsockopt             = dccp_getsockopt,
932         .sendmsg                = dccp_sendmsg,
933         .recvmsg                = dccp_recvmsg,
934         .backlog_rcv            = dccp_v4_do_rcv,
935         .hash                   = inet_hash,
936         .unhash                 = inet_unhash,
937         .accept                 = inet_csk_accept,
938         .get_port               = inet_csk_get_port,
939         .shutdown               = dccp_shutdown,
940         .destroy                = dccp_destroy_sock,
941         .orphan_count           = &dccp_orphan_count,
942         .max_header             = MAX_DCCP_HEADER,
943         .obj_size               = sizeof(struct dccp_sock),
944         .rsk_prot               = &dccp_request_sock_ops,
945         .twsk_prot              = &dccp_timewait_sock_ops,
946         .hashinfo               = &dccp_hashinfo,
947 #ifdef CONFIG_COMPAT
948         .compat_setsockopt      = compat_dccp_setsockopt,
949         .compat_getsockopt      = compat_dccp_getsockopt,
950 #endif
951         REF_PROTO_INUSE(dccp_v4)
952 };
953
954 static struct net_protocol dccp_v4_protocol = {
955         .handler        = dccp_v4_rcv,
956         .err_handler    = dccp_v4_err,
957         .no_policy      = 1,
958 };
959
960 static const struct proto_ops inet_dccp_ops = {
961         .family            = PF_INET,
962         .owner             = THIS_MODULE,
963         .release           = inet_release,
964         .bind              = inet_bind,
965         .connect           = inet_stream_connect,
966         .socketpair        = sock_no_socketpair,
967         .accept            = inet_accept,
968         .getname           = inet_getname,
969         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
970         .poll              = dccp_poll,
971         .ioctl             = inet_ioctl,
972         /* FIXME: work on inet_listen to rename it to sock_common_listen */
973         .listen            = inet_dccp_listen,
974         .shutdown          = inet_shutdown,
975         .setsockopt        = sock_common_setsockopt,
976         .getsockopt        = sock_common_getsockopt,
977         .sendmsg           = inet_sendmsg,
978         .recvmsg           = sock_common_recvmsg,
979         .mmap              = sock_no_mmap,
980         .sendpage          = sock_no_sendpage,
981 #ifdef CONFIG_COMPAT
982         .compat_setsockopt = compat_sock_common_setsockopt,
983         .compat_getsockopt = compat_sock_common_getsockopt,
984 #endif
985 };
986
987 static struct inet_protosw dccp_v4_protosw = {
988         .type           = SOCK_DCCP,
989         .protocol       = IPPROTO_DCCP,
990         .prot           = &dccp_v4_prot,
991         .ops            = &inet_dccp_ops,
992         .capability     = -1,
993         .no_check       = 0,
994         .flags          = INET_PROTOSW_ICSK,
995 };
996
997 static int __init dccp_v4_init(void)
998 {
999         int err = proto_register(&dccp_v4_prot, 1);
1000
1001         if (err != 0)
1002                 goto out;
1003
1004         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1005         if (err != 0)
1006                 goto out_proto_unregister;
1007
1008         inet_register_protosw(&dccp_v4_protosw);
1009
1010         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1011                                        SOCK_DCCP, IPPROTO_DCCP);
1012         if (err)
1013                 goto out_unregister_protosw;
1014 out:
1015         return err;
1016 out_unregister_protosw:
1017         inet_unregister_protosw(&dccp_v4_protosw);
1018         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1019 out_proto_unregister:
1020         proto_unregister(&dccp_v4_prot);
1021         goto out;
1022 }
1023
1024 static void __exit dccp_v4_exit(void)
1025 {
1026         inet_unregister_protosw(&dccp_v4_protosw);
1027         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1028         proto_unregister(&dccp_v4_prot);
1029 }
1030
1031 module_init(dccp_v4_init);
1032 module_exit(dccp_v4_exit);
1033
1034 /*
1035  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1036  * values directly, Also cover the case where the protocol is not specified,
1037  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1038  */
1039 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1040 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1041 MODULE_LICENSE("GPL");
1042 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1043 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");