Merge branch 'master' into upstream-fixes
[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 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         u32 daddr, nexthop;
54         int tmp;
55         int err;
56
57         dp->dccps_role = DCCP_ROLE_CLIENT;
58
59         if (dccp_service_not_initialized(sk))
60                 return -EPROTO;
61
62         if (addr_len < sizeof(struct sockaddr_in))
63                 return -EINVAL;
64
65         if (usin->sin_family != AF_INET)
66                 return -EAFNOSUPPORT;
67
68         nexthop = daddr = usin->sin_addr.s_addr;
69         if (inet->opt != NULL && inet->opt->srr) {
70                 if (daddr == 0)
71                         return -EINVAL;
72                 nexthop = inet->opt->faddr;
73         }
74
75         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
76                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
77                                IPPROTO_DCCP,
78                                inet->sport, usin->sin_port, sk);
79         if (tmp < 0)
80                 return tmp;
81
82         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
83                 ip_rt_put(rt);
84                 return -ENETUNREACH;
85         }
86
87         if (inet->opt == NULL || !inet->opt->srr)
88                 daddr = rt->rt_dst;
89
90         if (inet->saddr == 0)
91                 inet->saddr = rt->rt_src;
92         inet->rcv_saddr = inet->saddr;
93
94         inet->dport = usin->sin_port;
95         inet->daddr = daddr;
96
97         inet_csk(sk)->icsk_ext_hdr_len = 0;
98         if (inet->opt != NULL)
99                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
100         /*
101          * Socket identity is still unknown (sport may be zero).
102          * However we set state to DCCP_REQUESTING and not releasing socket
103          * lock select source port, enter ourselves into the hash tables and
104          * complete initialization after this.
105          */
106         dccp_set_state(sk, DCCP_REQUESTING);
107         err = inet_hash_connect(&dccp_death_row, sk);
108         if (err != 0)
109                 goto failure;
110
111         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
112                                 sk);
113         if (err != 0)
114                 goto failure;
115
116         /* OK, now commit destination to socket.  */
117         sk_setup_caps(sk, &rt->u.dst);
118
119         dp->dccps_gar =
120                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
121                                                             inet->daddr,
122                                                             inet->sport,
123                                                             usin->sin_port);
124         dccp_update_gss(sk, dp->dccps_iss);
125
126         inet->id = dp->dccps_iss ^ jiffies;
127
128         err = dccp_connect(sk);
129         rt = NULL;
130         if (err != 0)
131                 goto failure;
132 out:
133         return err;
134 failure:
135         /*
136          * This unhashes the socket and releases the local port, if necessary.
137          */
138         dccp_set_state(sk, DCCP_CLOSED);
139         ip_rt_put(rt);
140         sk->sk_route_caps = 0;
141         inet->dport = 0;
142         goto out;
143 }
144
145 EXPORT_SYMBOL_GPL(dccp_v4_connect);
146
147 /*
148  * This routine does path mtu discovery as defined in RFC1191.
149  */
150 static inline void dccp_do_pmtu_discovery(struct sock *sk,
151                                           const struct iphdr *iph,
152                                           u32 mtu)
153 {
154         struct dst_entry *dst;
155         const struct inet_sock *inet = inet_sk(sk);
156         const struct dccp_sock *dp = dccp_sk(sk);
157
158         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
159          * send out by Linux are always < 576bytes so they should go through
160          * unfragmented).
161          */
162         if (sk->sk_state == DCCP_LISTEN)
163                 return;
164
165         /* We don't check in the destentry if pmtu discovery is forbidden
166          * on this route. We just assume that no packet_to_big packets
167          * are send back when pmtu discovery is not active.
168          * There is a small race when the user changes this flag in the
169          * route, but I think that's acceptable.
170          */
171         if ((dst = __sk_dst_check(sk, 0)) == NULL)
172                 return;
173
174         dst->ops->update_pmtu(dst, mtu);
175
176         /* Something is about to be wrong... Remember soft error
177          * for the case, if this connection will not able to recover.
178          */
179         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
180                 sk->sk_err_soft = EMSGSIZE;
181
182         mtu = dst_mtu(dst);
183
184         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
185             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
186                 dccp_sync_mss(sk, mtu);
187
188                 /*
189                  * From: draft-ietf-dccp-spec-11.txt
190                  *
191                  *      DCCP-Sync packets are the best choice for upward
192                  *      probing, since DCCP-Sync probes do not risk application
193                  *      data loss.
194                  */
195                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
196         } /* else let the usual retransmit timer handle it */
197 }
198
199 static void dccp_v4_reqsk_send_ack(struct sk_buff *rxskb,
200                                    struct request_sock *req)
201 {
202         int err;
203         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
204         const u32 dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
205                                      sizeof(struct dccp_hdr_ext) +
206                                      sizeof(struct dccp_hdr_ack_bits);
207         struct sk_buff *skb;
208
209         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
210                 return;
211
212         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header, GFP_ATOMIC);
213         if (skb == NULL)
214                 return;
215
216         /* Reserve space for headers. */
217         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
218
219         skb->dst = dst_clone(rxskb->dst);
220
221         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
222         dh = dccp_hdr(skb);
223         memset(dh, 0, dccp_hdr_ack_len);
224
225         /* Build DCCP header and checksum it. */
226         dh->dccph_type     = DCCP_PKT_ACK;
227         dh->dccph_sport    = rxdh->dccph_dport;
228         dh->dccph_dport    = rxdh->dccph_sport;
229         dh->dccph_doff     = dccp_hdr_ack_len / 4;
230         dh->dccph_x        = 1;
231
232         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
233         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
234                          DCCP_SKB_CB(rxskb)->dccpd_seq);
235
236         bh_lock_sock(dccp_v4_ctl_socket->sk);
237         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
238                                     rxskb->nh.iph->daddr,
239                                     rxskb->nh.iph->saddr, NULL);
240         bh_unlock_sock(dccp_v4_ctl_socket->sk);
241
242         if (err == NET_XMIT_CN || err == 0) {
243                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
244                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
245         }
246 }
247
248 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
249                                  struct dst_entry *dst)
250 {
251         int err = -1;
252         struct sk_buff *skb;
253
254         /* First, grab a route. */
255         
256         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
257                 goto out;
258
259         skb = dccp_make_response(sk, dst, req);
260         if (skb != NULL) {
261                 const struct inet_request_sock *ireq = inet_rsk(req);
262                 struct dccp_hdr *dh = dccp_hdr(skb);
263
264                 dh->dccph_checksum = dccp_v4_checksum(skb, ireq->loc_addr,
265                                                       ireq->rmt_addr);
266                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
267                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
268                                             ireq->rmt_addr,
269                                             ireq->opt);
270                 if (err == NET_XMIT_CN)
271                         err = 0;
272         }
273
274 out:
275         dst_release(dst);
276         return err;
277 }
278
279 /*
280  * This routine is called by the ICMP module when it gets some sort of error
281  * condition. If err < 0 then the socket should be closed and the error
282  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
283  * After adjustment header points to the first 8 bytes of the tcp header. We
284  * need to find the appropriate port.
285  *
286  * The locking strategy used here is very "optimistic". When someone else
287  * accesses the socket the ICMP is just dropped and for some paths there is no
288  * check at all. A more general error queue to queue errors for later handling
289  * is probably better.
290  */
291 static void dccp_v4_err(struct sk_buff *skb, u32 info)
292 {
293         const struct iphdr *iph = (struct iphdr *)skb->data;
294         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
295                                                         (iph->ihl << 2));
296         struct dccp_sock *dp;
297         struct inet_sock *inet;
298         const int type = skb->h.icmph->type;
299         const int code = skb->h.icmph->code;
300         struct sock *sk;
301         __u64 seq;
302         int err;
303
304         if (skb->len < (iph->ihl << 2) + 8) {
305                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
306                 return;
307         }
308
309         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
310                          iph->saddr, dh->dccph_sport, inet_iif(skb));
311         if (sk == NULL) {
312                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
313                 return;
314         }
315
316         if (sk->sk_state == DCCP_TIME_WAIT) {
317                 inet_twsk_put((struct inet_timewait_sock *)sk);
318                 return;
319         }
320
321         bh_lock_sock(sk);
322         /* If too many ICMPs get dropped on busy
323          * servers this needs to be solved differently.
324          */
325         if (sock_owned_by_user(sk))
326                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
327
328         if (sk->sk_state == DCCP_CLOSED)
329                 goto out;
330
331         dp = dccp_sk(sk);
332         seq = dccp_hdr_seq(skb);
333         if (sk->sk_state != DCCP_LISTEN &&
334             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
335                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
336                 goto out;
337         }
338
339         switch (type) {
340         case ICMP_SOURCE_QUENCH:
341                 /* Just silently ignore these. */
342                 goto out;
343         case ICMP_PARAMETERPROB:
344                 err = EPROTO;
345                 break;
346         case ICMP_DEST_UNREACH:
347                 if (code > NR_ICMP_UNREACH)
348                         goto out;
349
350                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
351                         if (!sock_owned_by_user(sk))
352                                 dccp_do_pmtu_discovery(sk, iph, info);
353                         goto out;
354                 }
355
356                 err = icmp_err_convert[code].errno;
357                 break;
358         case ICMP_TIME_EXCEEDED:
359                 err = EHOSTUNREACH;
360                 break;
361         default:
362                 goto out;
363         }
364
365         switch (sk->sk_state) {
366                 struct request_sock *req , **prev;
367         case DCCP_LISTEN:
368                 if (sock_owned_by_user(sk))
369                         goto out;
370                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
371                                           iph->daddr, iph->saddr);
372                 if (!req)
373                         goto out;
374
375                 /*
376                  * ICMPs are not backlogged, hence we cannot get an established
377                  * socket here.
378                  */
379                 BUG_TRAP(!req->sk);
380
381                 if (seq != dccp_rsk(req)->dreq_iss) {
382                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
383                         goto out;
384                 }
385                 /*
386                  * Still in RESPOND, just remove it silently.
387                  * There is no good way to pass the error to the newly
388                  * created socket, and POSIX does not want network
389                  * errors returned from accept().
390                  */
391                 inet_csk_reqsk_queue_drop(sk, req, prev);
392                 goto out;
393
394         case DCCP_REQUESTING:
395         case DCCP_RESPOND:
396                 if (!sock_owned_by_user(sk)) {
397                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
398                         sk->sk_err = err;
399
400                         sk->sk_error_report(sk);
401
402                         dccp_done(sk);
403                 } else
404                         sk->sk_err_soft = err;
405                 goto out;
406         }
407
408         /* If we've already connected we will keep trying
409          * until we time out, or the user gives up.
410          *
411          * rfc1122 4.2.3.9 allows to consider as hard errors
412          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
413          * but it is obsoleted by pmtu discovery).
414          *
415          * Note, that in modern internet, where routing is unreliable
416          * and in each dark corner broken firewalls sit, sending random
417          * errors ordered by their masters even this two messages finally lose
418          * their original sense (even Linux sends invalid PORT_UNREACHs)
419          *
420          * Now we are in compliance with RFCs.
421          *                                                      --ANK (980905)
422          */
423
424         inet = inet_sk(sk);
425         if (!sock_owned_by_user(sk) && inet->recverr) {
426                 sk->sk_err = err;
427                 sk->sk_error_report(sk);
428         } else /* Only an error on timeout */
429                 sk->sk_err_soft = err;
430 out:
431         bh_unlock_sock(sk);
432         sock_put(sk);
433 }
434
435 /* This routine computes an IPv4 DCCP checksum. */
436 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
437 {
438         const struct inet_sock *inet = inet_sk(sk);
439         struct dccp_hdr *dh = dccp_hdr(skb);
440
441         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
442 }
443
444 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
445
446 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
447                                         const struct sk_buff *skb)
448 {
449         return secure_dccp_sequence_number(skb->nh.iph->daddr,
450                                            skb->nh.iph->saddr,
451                                            dccp_hdr(skb)->dccph_dport,
452                                            dccp_hdr(skb)->dccph_sport);
453 }
454
455 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
456 {
457         struct inet_request_sock *ireq;
458         struct dccp_sock dp;
459         struct request_sock *req;
460         struct dccp_request_sock *dreq;
461         const __be32 saddr = skb->nh.iph->saddr;
462         const __be32 daddr = skb->nh.iph->daddr;
463         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
464         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
465         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
466
467         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
468         if (((struct rtable *)skb->dst)->rt_flags &
469             (RTCF_BROADCAST | RTCF_MULTICAST)) {
470                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
471                 goto drop;
472         }
473
474         if (dccp_bad_service_code(sk, service)) {
475                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
476                 goto drop;
477         }
478         /*
479          * TW buckets are converted to open requests without
480          * limitations, they conserve resources and peer is
481          * evidently real one.
482          */
483         if (inet_csk_reqsk_queue_is_full(sk))
484                 goto drop;
485
486         /*
487          * Accept backlog is full. If we have already queued enough
488          * of warm entries in syn queue, drop request. It is better than
489          * clogging syn queue with openreqs with exponentially increasing
490          * timeout.
491          */
492         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
493                 goto drop;
494
495         req = reqsk_alloc(sk->sk_prot->rsk_prot);
496         if (req == NULL)
497                 goto drop;
498
499         if (dccp_parse_options(sk, skb))
500                 goto drop_and_free;
501
502         dccp_openreq_init(req, &dp, skb);
503
504         ireq = inet_rsk(req);
505         ireq->loc_addr = daddr;
506         ireq->rmt_addr = saddr;
507         req->rcv_wnd    = dccp_feat_default_sequence_window;
508         ireq->opt       = NULL;
509
510         /* 
511          * Step 3: Process LISTEN state
512          *
513          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
514          *
515          * In fact we defer setting S.GSR, S.SWL, S.SWH to
516          * dccp_create_openreq_child.
517          */
518         dreq = dccp_rsk(req);
519         dreq->dreq_isr     = dcb->dccpd_seq;
520         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
521         dreq->dreq_service = service;
522
523         if (dccp_v4_send_response(sk, req, NULL))
524                 goto drop_and_free;
525
526         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
527         return 0;
528
529 drop_and_free:
530         reqsk_free(req);
531 drop:
532         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
533         dcb->dccpd_reset_code = reset_code;
534         return -1;
535 }
536
537 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
538
539 /*
540  * The three way handshake has completed - we got a valid ACK or DATAACK -
541  * now create the new socket.
542  *
543  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
544  */
545 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
546                                        struct request_sock *req,
547                                        struct dst_entry *dst)
548 {
549         struct inet_request_sock *ireq;
550         struct inet_sock *newinet;
551         struct dccp_sock *newdp;
552         struct sock *newsk;
553
554         if (sk_acceptq_is_full(sk))
555                 goto exit_overflow;
556
557         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
558                 goto exit;
559
560         newsk = dccp_create_openreq_child(sk, req, skb);
561         if (newsk == NULL)
562                 goto exit;
563
564         sk_setup_caps(newsk, dst);
565
566         newdp              = dccp_sk(newsk);
567         newinet            = inet_sk(newsk);
568         ireq               = inet_rsk(req);
569         newinet->daddr     = ireq->rmt_addr;
570         newinet->rcv_saddr = ireq->loc_addr;
571         newinet->saddr     = ireq->loc_addr;
572         newinet->opt       = ireq->opt;
573         ireq->opt          = NULL;
574         newinet->mc_index  = inet_iif(skb);
575         newinet->mc_ttl    = skb->nh.iph->ttl;
576         newinet->id        = jiffies;
577
578         dccp_sync_mss(newsk, dst_mtu(dst));
579
580         __inet_hash(&dccp_hashinfo, newsk, 0);
581         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
582
583         return newsk;
584
585 exit_overflow:
586         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
587 exit:
588         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
589         dst_release(dst);
590         return NULL;
591 }
592
593 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
594
595 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
596 {
597         const struct dccp_hdr *dh = dccp_hdr(skb);
598         const struct iphdr *iph = skb->nh.iph;
599         struct sock *nsk;
600         struct request_sock **prev;
601         /* Find possible connection requests. */
602         struct request_sock *req = inet_csk_search_req(sk, &prev,
603                                                        dh->dccph_sport,
604                                                        iph->saddr, iph->daddr);
605         if (req != NULL)
606                 return dccp_check_req(sk, skb, req, prev);
607
608         nsk = __inet_lookup_established(&dccp_hashinfo,
609                                         iph->saddr, dh->dccph_sport,
610                                         iph->daddr, ntohs(dh->dccph_dport),
611                                         inet_iif(skb));
612         if (nsk != NULL) {
613                 if (nsk->sk_state != DCCP_TIME_WAIT) {
614                         bh_lock_sock(nsk);
615                         return nsk;
616                 }
617                 inet_twsk_put((struct inet_timewait_sock *)nsk);
618                 return NULL;
619         }
620
621         return sk;
622 }
623
624 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
625                      const __be32 daddr)
626 {
627         const struct dccp_hdr* dh = dccp_hdr(skb);
628         int checksum_len;
629         u32 tmp;
630
631         if (dh->dccph_cscov == 0)
632                 checksum_len = skb->len;
633         else {
634                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
635                 checksum_len = checksum_len < skb->len ? checksum_len :
636                                                          skb->len;
637         }
638
639         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
640         return csum_tcpudp_magic(saddr, daddr, checksum_len,
641                                  IPPROTO_DCCP, tmp);
642 }
643
644 EXPORT_SYMBOL_GPL(dccp_v4_checksum);
645
646 static int dccp_v4_verify_checksum(struct sk_buff *skb,
647                                    const __be32 saddr, const __be32 daddr)
648 {
649         struct dccp_hdr *dh = dccp_hdr(skb);
650         int checksum_len;
651         u32 tmp;
652
653         if (dh->dccph_cscov == 0)
654                 checksum_len = skb->len;
655         else {
656                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
657                 checksum_len = checksum_len < skb->len ? checksum_len :
658                                                          skb->len;
659         }
660         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
661         return csum_tcpudp_magic(saddr, daddr, checksum_len,
662                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
663 }
664
665 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
666                                            struct sk_buff *skb)
667 {
668         struct rtable *rt;
669         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
670                             .nl_u = { .ip4_u =
671                                       { .daddr = skb->nh.iph->saddr,
672                                         .saddr = skb->nh.iph->daddr,
673                                         .tos = RT_CONN_FLAGS(sk) } },
674                             .proto = sk->sk_protocol,
675                             .uli_u = { .ports =
676                                        { .sport = dccp_hdr(skb)->dccph_dport,
677                                          .dport = dccp_hdr(skb)->dccph_sport }
678                                      }
679                           };
680
681         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
682                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
683                 return NULL;
684         }
685
686         return &rt->u.dst;
687 }
688
689 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
690 {
691         int err;
692         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
693         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
694                                        sizeof(struct dccp_hdr_ext) +
695                                        sizeof(struct dccp_hdr_reset);
696         struct sk_buff *skb;
697         struct dst_entry *dst;
698         u64 seqno;
699
700         /* Never send a reset in response to a reset. */
701         if (rxdh->dccph_type == DCCP_PKT_RESET)
702                 return;
703
704         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
705                 return;
706
707         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
708         if (dst == NULL)
709                 return;
710
711         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header,
712                         GFP_ATOMIC);
713         if (skb == NULL)
714                 goto out;
715
716         /* Reserve space for headers. */
717         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
718         skb->dst = dst_clone(dst);
719
720         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
721         dh = dccp_hdr(skb);
722         memset(dh, 0, dccp_hdr_reset_len);
723
724         /* Build DCCP header and checksum it. */
725         dh->dccph_type     = DCCP_PKT_RESET;
726         dh->dccph_sport    = rxdh->dccph_dport;
727         dh->dccph_dport    = rxdh->dccph_sport;
728         dh->dccph_doff     = dccp_hdr_reset_len / 4;
729         dh->dccph_x        = 1;
730         dccp_hdr_reset(skb)->dccph_reset_code =
731                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
732
733         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
734         seqno = 0;
735         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
736                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
737
738         dccp_hdr_set_seq(dh, seqno);
739         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
740                          DCCP_SKB_CB(rxskb)->dccpd_seq);
741
742         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
743                                               rxskb->nh.iph->daddr);
744
745         bh_lock_sock(dccp_v4_ctl_socket->sk);
746         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
747                                     rxskb->nh.iph->daddr,
748                                     rxskb->nh.iph->saddr, NULL);
749         bh_unlock_sock(dccp_v4_ctl_socket->sk);
750
751         if (err == NET_XMIT_CN || err == 0) {
752                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
753                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
754         }
755 out:
756          dst_release(dst);
757 }
758
759 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
760 {
761         struct dccp_hdr *dh = dccp_hdr(skb);
762
763         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
764                 if (dccp_rcv_established(sk, skb, dh, skb->len))
765                         goto reset;
766                 return 0;
767         }
768
769         /*
770          *  Step 3: Process LISTEN state
771          *     If S.state == LISTEN,
772          *        If P.type == Request or P contains a valid Init Cookie
773          *              option,
774          *           * Must scan the packet's options to check for an Init
775          *              Cookie.  Only the Init Cookie is processed here,
776          *              however; other options are processed in Step 8.  This
777          *              scan need only be performed if the endpoint uses Init
778          *              Cookies *
779          *           * Generate a new socket and switch to that socket *
780          *           Set S := new socket for this port pair
781          *           S.state = RESPOND
782          *           Choose S.ISS (initial seqno) or set from Init Cookie
783          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
784          *           Continue with S.state == RESPOND
785          *           * A Response packet will be generated in Step 11 *
786          *        Otherwise,
787          *           Generate Reset(No Connection) unless P.type == Reset
788          *           Drop packet and return
789          *
790          * NOTE: the check for the packet types is done in
791          *       dccp_rcv_state_process
792          */
793         if (sk->sk_state == DCCP_LISTEN) {
794                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
795
796                 if (nsk == NULL)
797                         goto discard;
798
799                 if (nsk != sk) {
800                         if (dccp_child_process(sk, nsk, skb))
801                                 goto reset;
802                         return 0;
803                 }
804         }
805
806         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
807                 goto reset;
808         return 0;
809
810 reset:
811         dccp_v4_ctl_send_reset(skb);
812 discard:
813         kfree_skb(skb);
814         return 0;
815 }
816
817 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
818
819 int dccp_invalid_packet(struct sk_buff *skb)
820 {
821         const struct dccp_hdr *dh;
822
823         if (skb->pkt_type != PACKET_HOST)
824                 return 1;
825
826         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
827                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
828                 return 1;
829         }
830
831         dh = dccp_hdr(skb);
832
833         /* If the packet type is not understood, drop packet and return */
834         if (dh->dccph_type >= DCCP_PKT_INVALID) {
835                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
836                 return 1;
837         }
838
839         /*
840          * If P.Data Offset is too small for packet type, or too large for
841          * packet, drop packet and return
842          */
843         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
844                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
845                                             "too small 1\n",
846                                dh->dccph_doff);
847                 return 1;
848         }
849
850         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
851                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
852                                             "too small 2\n",
853                                dh->dccph_doff);
854                 return 1;
855         }
856
857         dh = dccp_hdr(skb);
858
859         /*
860          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
861          * has short sequence numbers), drop packet and return
862          */
863         if (dh->dccph_x == 0 &&
864             dh->dccph_type != DCCP_PKT_DATA &&
865             dh->dccph_type != DCCP_PKT_ACK &&
866             dh->dccph_type != DCCP_PKT_DATAACK) {
867                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
868                                             "nor DataAck and P.X == 0\n",
869                                dccp_packet_name(dh->dccph_type));
870                 return 1;
871         }
872
873         return 0;
874 }
875
876 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
877
878 /* this is called when real data arrives */
879 static int dccp_v4_rcv(struct sk_buff *skb)
880 {
881         const struct dccp_hdr *dh;
882         struct sock *sk;
883
884         /* Step 1: Check header basics: */
885
886         if (dccp_invalid_packet(skb))
887                 goto discard_it;
888
889         /* If the header checksum is incorrect, drop packet and return */
890         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
891                                     skb->nh.iph->daddr) < 0) {
892                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
893                                __FUNCTION__);
894                 goto discard_it;
895         }
896
897         dh = dccp_hdr(skb);
898
899         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
900         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
901
902         dccp_pr_debug("%8.8s "
903                       "src=%u.%u.%u.%u@%-5d "
904                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
905                       dccp_packet_name(dh->dccph_type),
906                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
907                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
908                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
909
910         if (dccp_packet_without_ack(skb)) {
911                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
912                 dccp_pr_debug_cat("\n");
913         } else {
914                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
915                 dccp_pr_debug_cat(", ack=%llu\n",
916                                   (unsigned long long)
917                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
918         }
919
920         /* Step 2:
921          *      Look up flow ID in table and get corresponding socket */
922         sk = __inet_lookup(&dccp_hashinfo,
923                            skb->nh.iph->saddr, dh->dccph_sport,
924                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
925                            inet_iif(skb));
926
927         /* 
928          * Step 2:
929          *      If no socket ...
930          *              Generate Reset(No Connection) unless P.type == Reset
931          *              Drop packet and return
932          */
933         if (sk == NULL) {
934                 dccp_pr_debug("failed to look up flow ID in table and "
935                               "get corresponding socket\n");
936                 goto no_dccp_socket;
937         }
938
939         /* 
940          * Step 2:
941          *      ... or S.state == TIMEWAIT,
942          *              Generate Reset(No Connection) unless P.type == Reset
943          *              Drop packet and return
944          */
945                
946         if (sk->sk_state == DCCP_TIME_WAIT) {
947                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
948                               "do_time_wait\n");
949                 goto do_time_wait;
950         }
951
952         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
953                 goto discard_and_relse;
954         nf_reset(skb);
955
956         return sk_receive_skb(sk, skb);
957
958 no_dccp_socket:
959         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
960                 goto discard_it;
961         /*
962          * Step 2:
963          *              Generate Reset(No Connection) unless P.type == Reset
964          *              Drop packet and return
965          */
966         if (dh->dccph_type != DCCP_PKT_RESET) {
967                 DCCP_SKB_CB(skb)->dccpd_reset_code =
968                                         DCCP_RESET_CODE_NO_CONNECTION;
969                 dccp_v4_ctl_send_reset(skb);
970         }
971
972 discard_it:
973         /* Discard frame. */
974         kfree_skb(skb);
975         return 0;
976
977 discard_and_relse:
978         sock_put(sk);
979         goto discard_it;
980
981 do_time_wait:
982         inet_twsk_put((struct inet_timewait_sock *)sk);
983         goto no_dccp_socket;
984 }
985
986 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
987         .queue_xmit        = ip_queue_xmit,
988         .send_check        = dccp_v4_send_check,
989         .rebuild_header    = inet_sk_rebuild_header,
990         .conn_request      = dccp_v4_conn_request,
991         .syn_recv_sock     = dccp_v4_request_recv_sock,
992         .net_header_len    = sizeof(struct iphdr),
993         .setsockopt        = ip_setsockopt,
994         .getsockopt        = ip_getsockopt,
995         .addr2sockaddr     = inet_csk_addr2sockaddr,
996         .sockaddr_len      = sizeof(struct sockaddr_in),
997 #ifdef CONFIG_COMPAT
998         .compat_setsockopt = compat_ip_setsockopt,
999         .compat_getsockopt = compat_ip_getsockopt,
1000 #endif
1001 };
1002
1003 static int dccp_v4_init_sock(struct sock *sk)
1004 {
1005         static __u8 dccp_v4_ctl_sock_initialized;
1006         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
1007
1008         if (err == 0) {
1009                 if (unlikely(!dccp_v4_ctl_sock_initialized))
1010                         dccp_v4_ctl_sock_initialized = 1;
1011                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1012         }
1013
1014         return err;
1015 }
1016
1017 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1018 {
1019         kfree(inet_rsk(req)->opt);
1020 }
1021
1022 static struct request_sock_ops dccp_request_sock_ops = {
1023         .family         = PF_INET,
1024         .obj_size       = sizeof(struct dccp_request_sock),
1025         .rtx_syn_ack    = dccp_v4_send_response,
1026         .send_ack       = dccp_v4_reqsk_send_ack,
1027         .destructor     = dccp_v4_reqsk_destructor,
1028         .send_reset     = dccp_v4_ctl_send_reset,
1029 };
1030
1031 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1032         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1033 };
1034
1035 static struct proto dccp_v4_prot = {
1036         .name                   = "DCCP",
1037         .owner                  = THIS_MODULE,
1038         .close                  = dccp_close,
1039         .connect                = dccp_v4_connect,
1040         .disconnect             = dccp_disconnect,
1041         .ioctl                  = dccp_ioctl,
1042         .init                   = dccp_v4_init_sock,
1043         .setsockopt             = dccp_setsockopt,
1044         .getsockopt             = dccp_getsockopt,
1045         .sendmsg                = dccp_sendmsg,
1046         .recvmsg                = dccp_recvmsg,
1047         .backlog_rcv            = dccp_v4_do_rcv,
1048         .hash                   = dccp_hash,
1049         .unhash                 = dccp_unhash,
1050         .accept                 = inet_csk_accept,
1051         .get_port               = dccp_v4_get_port,
1052         .shutdown               = dccp_shutdown,
1053         .destroy                = dccp_destroy_sock,
1054         .orphan_count           = &dccp_orphan_count,
1055         .max_header             = MAX_DCCP_HEADER,
1056         .obj_size               = sizeof(struct dccp_sock),
1057         .rsk_prot               = &dccp_request_sock_ops,
1058         .twsk_prot              = &dccp_timewait_sock_ops,
1059 #ifdef CONFIG_COMPAT
1060         .compat_setsockopt      = compat_dccp_setsockopt,
1061         .compat_getsockopt      = compat_dccp_getsockopt,
1062 #endif
1063 };
1064
1065 static struct net_protocol dccp_v4_protocol = {
1066         .handler        = dccp_v4_rcv,
1067         .err_handler    = dccp_v4_err,
1068         .no_policy      = 1,
1069 };
1070
1071 static const struct proto_ops inet_dccp_ops = {
1072         .family            = PF_INET,
1073         .owner             = THIS_MODULE,
1074         .release           = inet_release,
1075         .bind              = inet_bind,
1076         .connect           = inet_stream_connect,
1077         .socketpair        = sock_no_socketpair,
1078         .accept            = inet_accept,
1079         .getname           = inet_getname,
1080         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1081         .poll              = dccp_poll,
1082         .ioctl             = inet_ioctl,
1083         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1084         .listen            = inet_dccp_listen,
1085         .shutdown          = inet_shutdown,
1086         .setsockopt        = sock_common_setsockopt,
1087         .getsockopt        = sock_common_getsockopt,
1088         .sendmsg           = inet_sendmsg,
1089         .recvmsg           = sock_common_recvmsg,
1090         .mmap              = sock_no_mmap,
1091         .sendpage          = sock_no_sendpage,
1092 #ifdef CONFIG_COMPAT
1093         .compat_setsockopt = compat_sock_common_setsockopt,
1094         .compat_getsockopt = compat_sock_common_getsockopt,
1095 #endif
1096 };
1097
1098 static struct inet_protosw dccp_v4_protosw = {
1099         .type           = SOCK_DCCP,
1100         .protocol       = IPPROTO_DCCP,
1101         .prot           = &dccp_v4_prot,
1102         .ops            = &inet_dccp_ops,
1103         .capability     = -1,
1104         .no_check       = 0,
1105         .flags          = INET_PROTOSW_ICSK,
1106 };
1107
1108 static int __init dccp_v4_init(void)
1109 {
1110         int err = proto_register(&dccp_v4_prot, 1);
1111
1112         if (err != 0)
1113                 goto out;
1114
1115         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1116         if (err != 0)
1117                 goto out_proto_unregister;
1118
1119         inet_register_protosw(&dccp_v4_protosw);
1120
1121         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1122                                        SOCK_DCCP, IPPROTO_DCCP);
1123         if (err)
1124                 goto out_unregister_protosw;
1125 out:
1126         return err;
1127 out_unregister_protosw:
1128         inet_unregister_protosw(&dccp_v4_protosw);
1129         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1130 out_proto_unregister:
1131         proto_unregister(&dccp_v4_prot);
1132         goto out;
1133 }
1134
1135 static void __exit dccp_v4_exit(void)
1136 {
1137         inet_unregister_protosw(&dccp_v4_protosw);
1138         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1139         proto_unregister(&dccp_v4_prot);
1140 }
1141
1142 module_init(dccp_v4_init);
1143 module_exit(dccp_v4_exit);
1144
1145 /*
1146  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1147  * values directly, Also cover the case where the protocol is not specified,
1148  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1149  */
1150 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1151 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1152 MODULE_LICENSE("GPL");
1153 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1154 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");