9a1a76a7dc41f0122d0793a7504d50ad73b9ac8c
[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         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         if (security_inet_conn_request(sk, skb, req))
505                 goto drop_and_free;
506
507         ireq = inet_rsk(req);
508         ireq->loc_addr = daddr;
509         ireq->rmt_addr = saddr;
510         req->rcv_wnd    = dccp_feat_default_sequence_window;
511         ireq->opt       = NULL;
512
513         /* 
514          * Step 3: Process LISTEN state
515          *
516          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
517          *
518          * In fact we defer setting S.GSR, S.SWL, S.SWH to
519          * dccp_create_openreq_child.
520          */
521         dreq = dccp_rsk(req);
522         dreq->dreq_isr     = dcb->dccpd_seq;
523         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
524         dreq->dreq_service = service;
525
526         if (dccp_v4_send_response(sk, req, NULL))
527                 goto drop_and_free;
528
529         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
530         return 0;
531
532 drop_and_free:
533         reqsk_free(req);
534 drop:
535         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
536         dcb->dccpd_reset_code = reset_code;
537         return -1;
538 }
539
540 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
541
542 /*
543  * The three way handshake has completed - we got a valid ACK or DATAACK -
544  * now create the new socket.
545  *
546  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
547  */
548 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
549                                        struct request_sock *req,
550                                        struct dst_entry *dst)
551 {
552         struct inet_request_sock *ireq;
553         struct inet_sock *newinet;
554         struct dccp_sock *newdp;
555         struct sock *newsk;
556
557         if (sk_acceptq_is_full(sk))
558                 goto exit_overflow;
559
560         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
561                 goto exit;
562
563         newsk = dccp_create_openreq_child(sk, req, skb);
564         if (newsk == NULL)
565                 goto exit;
566
567         sk_setup_caps(newsk, dst);
568
569         newdp              = dccp_sk(newsk);
570         newinet            = inet_sk(newsk);
571         ireq               = inet_rsk(req);
572         newinet->daddr     = ireq->rmt_addr;
573         newinet->rcv_saddr = ireq->loc_addr;
574         newinet->saddr     = ireq->loc_addr;
575         newinet->opt       = ireq->opt;
576         ireq->opt          = NULL;
577         newinet->mc_index  = inet_iif(skb);
578         newinet->mc_ttl    = skb->nh.iph->ttl;
579         newinet->id        = jiffies;
580
581         dccp_sync_mss(newsk, dst_mtu(dst));
582
583         __inet_hash(&dccp_hashinfo, newsk, 0);
584         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
585
586         return newsk;
587
588 exit_overflow:
589         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
590 exit:
591         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
592         dst_release(dst);
593         return NULL;
594 }
595
596 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
597
598 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
599 {
600         const struct dccp_hdr *dh = dccp_hdr(skb);
601         const struct iphdr *iph = skb->nh.iph;
602         struct sock *nsk;
603         struct request_sock **prev;
604         /* Find possible connection requests. */
605         struct request_sock *req = inet_csk_search_req(sk, &prev,
606                                                        dh->dccph_sport,
607                                                        iph->saddr, iph->daddr);
608         if (req != NULL)
609                 return dccp_check_req(sk, skb, req, prev);
610
611         nsk = inet_lookup_established(&dccp_hashinfo,
612                                       iph->saddr, dh->dccph_sport,
613                                       iph->daddr, dh->dccph_dport,
614                                       inet_iif(skb));
615         if (nsk != NULL) {
616                 if (nsk->sk_state != DCCP_TIME_WAIT) {
617                         bh_lock_sock(nsk);
618                         return nsk;
619                 }
620                 inet_twsk_put((struct inet_timewait_sock *)nsk);
621                 return NULL;
622         }
623
624         return sk;
625 }
626
627 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
628                      const __be32 daddr)
629 {
630         const struct dccp_hdr* dh = dccp_hdr(skb);
631         int checksum_len;
632         u32 tmp;
633
634         if (dh->dccph_cscov == 0)
635                 checksum_len = skb->len;
636         else {
637                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
638                 checksum_len = checksum_len < skb->len ? checksum_len :
639                                                          skb->len;
640         }
641
642         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
643         return csum_tcpudp_magic(saddr, daddr, checksum_len,
644                                  IPPROTO_DCCP, tmp);
645 }
646
647 EXPORT_SYMBOL_GPL(dccp_v4_checksum);
648
649 static int dccp_v4_verify_checksum(struct sk_buff *skb,
650                                    const __be32 saddr, const __be32 daddr)
651 {
652         struct dccp_hdr *dh = dccp_hdr(skb);
653         int checksum_len;
654         u32 tmp;
655
656         if (dh->dccph_cscov == 0)
657                 checksum_len = skb->len;
658         else {
659                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
660                 checksum_len = checksum_len < skb->len ? checksum_len :
661                                                          skb->len;
662         }
663         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
664         return csum_tcpudp_magic(saddr, daddr, checksum_len,
665                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
666 }
667
668 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
669                                            struct sk_buff *skb)
670 {
671         struct rtable *rt;
672         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
673                             .nl_u = { .ip4_u =
674                                       { .daddr = skb->nh.iph->saddr,
675                                         .saddr = skb->nh.iph->daddr,
676                                         .tos = RT_CONN_FLAGS(sk) } },
677                             .proto = sk->sk_protocol,
678                             .uli_u = { .ports =
679                                        { .sport = dccp_hdr(skb)->dccph_dport,
680                                          .dport = dccp_hdr(skb)->dccph_sport }
681                                      }
682                           };
683
684         security_skb_classify_flow(skb, &fl);
685         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
686                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
687                 return NULL;
688         }
689
690         return &rt->u.dst;
691 }
692
693 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
694 {
695         int err;
696         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
697         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
698                                        sizeof(struct dccp_hdr_ext) +
699                                        sizeof(struct dccp_hdr_reset);
700         struct sk_buff *skb;
701         struct dst_entry *dst;
702         u64 seqno;
703
704         /* Never send a reset in response to a reset. */
705         if (rxdh->dccph_type == DCCP_PKT_RESET)
706                 return;
707
708         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
709                 return;
710
711         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
712         if (dst == NULL)
713                 return;
714
715         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header,
716                         GFP_ATOMIC);
717         if (skb == NULL)
718                 goto out;
719
720         /* Reserve space for headers. */
721         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
722         skb->dst = dst_clone(dst);
723
724         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
725         dh = dccp_hdr(skb);
726         memset(dh, 0, dccp_hdr_reset_len);
727
728         /* Build DCCP header and checksum it. */
729         dh->dccph_type     = DCCP_PKT_RESET;
730         dh->dccph_sport    = rxdh->dccph_dport;
731         dh->dccph_dport    = rxdh->dccph_sport;
732         dh->dccph_doff     = dccp_hdr_reset_len / 4;
733         dh->dccph_x        = 1;
734         dccp_hdr_reset(skb)->dccph_reset_code =
735                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
736
737         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
738         seqno = 0;
739         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
740                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
741
742         dccp_hdr_set_seq(dh, seqno);
743         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
744                          DCCP_SKB_CB(rxskb)->dccpd_seq);
745
746         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
747                                               rxskb->nh.iph->daddr);
748
749         bh_lock_sock(dccp_v4_ctl_socket->sk);
750         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
751                                     rxskb->nh.iph->daddr,
752                                     rxskb->nh.iph->saddr, NULL);
753         bh_unlock_sock(dccp_v4_ctl_socket->sk);
754
755         if (err == NET_XMIT_CN || err == 0) {
756                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
757                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
758         }
759 out:
760          dst_release(dst);
761 }
762
763 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
764 {
765         struct dccp_hdr *dh = dccp_hdr(skb);
766
767         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
768                 if (dccp_rcv_established(sk, skb, dh, skb->len))
769                         goto reset;
770                 return 0;
771         }
772
773         /*
774          *  Step 3: Process LISTEN state
775          *     If S.state == LISTEN,
776          *        If P.type == Request or P contains a valid Init Cookie
777          *              option,
778          *           * Must scan the packet's options to check for an Init
779          *              Cookie.  Only the Init Cookie is processed here,
780          *              however; other options are processed in Step 8.  This
781          *              scan need only be performed if the endpoint uses Init
782          *              Cookies *
783          *           * Generate a new socket and switch to that socket *
784          *           Set S := new socket for this port pair
785          *           S.state = RESPOND
786          *           Choose S.ISS (initial seqno) or set from Init Cookie
787          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
788          *           Continue with S.state == RESPOND
789          *           * A Response packet will be generated in Step 11 *
790          *        Otherwise,
791          *           Generate Reset(No Connection) unless P.type == Reset
792          *           Drop packet and return
793          *
794          * NOTE: the check for the packet types is done in
795          *       dccp_rcv_state_process
796          */
797         if (sk->sk_state == DCCP_LISTEN) {
798                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
799
800                 if (nsk == NULL)
801                         goto discard;
802
803                 if (nsk != sk) {
804                         if (dccp_child_process(sk, nsk, skb))
805                                 goto reset;
806                         return 0;
807                 }
808         }
809
810         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
811                 goto reset;
812         return 0;
813
814 reset:
815         dccp_v4_ctl_send_reset(skb);
816 discard:
817         kfree_skb(skb);
818         return 0;
819 }
820
821 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
822
823 int dccp_invalid_packet(struct sk_buff *skb)
824 {
825         const struct dccp_hdr *dh;
826
827         if (skb->pkt_type != PACKET_HOST)
828                 return 1;
829
830         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
831                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
832                 return 1;
833         }
834
835         dh = dccp_hdr(skb);
836
837         /* If the packet type is not understood, drop packet and return */
838         if (dh->dccph_type >= DCCP_PKT_INVALID) {
839                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
840                 return 1;
841         }
842
843         /*
844          * If P.Data Offset is too small for packet type, or too large for
845          * packet, drop packet and return
846          */
847         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
848                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
849                                             "too small 1\n",
850                                dh->dccph_doff);
851                 return 1;
852         }
853
854         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
855                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
856                                             "too small 2\n",
857                                dh->dccph_doff);
858                 return 1;
859         }
860
861         dh = dccp_hdr(skb);
862
863         /*
864          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
865          * has short sequence numbers), drop packet and return
866          */
867         if (dh->dccph_x == 0 &&
868             dh->dccph_type != DCCP_PKT_DATA &&
869             dh->dccph_type != DCCP_PKT_ACK &&
870             dh->dccph_type != DCCP_PKT_DATAACK) {
871                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
872                                             "nor DataAck and P.X == 0\n",
873                                dccp_packet_name(dh->dccph_type));
874                 return 1;
875         }
876
877         return 0;
878 }
879
880 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
881
882 /* this is called when real data arrives */
883 static int dccp_v4_rcv(struct sk_buff *skb)
884 {
885         const struct dccp_hdr *dh;
886         struct sock *sk;
887
888         /* Step 1: Check header basics: */
889
890         if (dccp_invalid_packet(skb))
891                 goto discard_it;
892
893         /* If the header checksum is incorrect, drop packet and return */
894         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
895                                     skb->nh.iph->daddr) < 0) {
896                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
897                                __FUNCTION__);
898                 goto discard_it;
899         }
900
901         dh = dccp_hdr(skb);
902
903         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
904         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
905
906         dccp_pr_debug("%8.8s "
907                       "src=%u.%u.%u.%u@%-5d "
908                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
909                       dccp_packet_name(dh->dccph_type),
910                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
911                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
912                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
913
914         if (dccp_packet_without_ack(skb)) {
915                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
916                 dccp_pr_debug_cat("\n");
917         } else {
918                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
919                 dccp_pr_debug_cat(", ack=%llu\n",
920                                   (unsigned long long)
921                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
922         }
923
924         /* Step 2:
925          *      Look up flow ID in table and get corresponding socket */
926         sk = __inet_lookup(&dccp_hashinfo,
927                            skb->nh.iph->saddr, dh->dccph_sport,
928                            skb->nh.iph->daddr, dh->dccph_dport,
929                            inet_iif(skb));
930
931         /* 
932          * Step 2:
933          *      If no socket ...
934          *              Generate Reset(No Connection) unless P.type == Reset
935          *              Drop packet and return
936          */
937         if (sk == NULL) {
938                 dccp_pr_debug("failed to look up flow ID in table and "
939                               "get corresponding socket\n");
940                 goto no_dccp_socket;
941         }
942
943         /* 
944          * Step 2:
945          *      ... or S.state == TIMEWAIT,
946          *              Generate Reset(No Connection) unless P.type == Reset
947          *              Drop packet and return
948          */
949                
950         if (sk->sk_state == DCCP_TIME_WAIT) {
951                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
952                               "do_time_wait\n");
953                 goto do_time_wait;
954         }
955
956         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
957                 goto discard_and_relse;
958         nf_reset(skb);
959
960         return sk_receive_skb(sk, skb);
961
962 no_dccp_socket:
963         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
964                 goto discard_it;
965         /*
966          * Step 2:
967          *              Generate Reset(No Connection) unless P.type == Reset
968          *              Drop packet and return
969          */
970         if (dh->dccph_type != DCCP_PKT_RESET) {
971                 DCCP_SKB_CB(skb)->dccpd_reset_code =
972                                         DCCP_RESET_CODE_NO_CONNECTION;
973                 dccp_v4_ctl_send_reset(skb);
974         }
975
976 discard_it:
977         /* Discard frame. */
978         kfree_skb(skb);
979         return 0;
980
981 discard_and_relse:
982         sock_put(sk);
983         goto discard_it;
984
985 do_time_wait:
986         inet_twsk_put((struct inet_timewait_sock *)sk);
987         goto no_dccp_socket;
988 }
989
990 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
991         .queue_xmit        = ip_queue_xmit,
992         .send_check        = dccp_v4_send_check,
993         .rebuild_header    = inet_sk_rebuild_header,
994         .conn_request      = dccp_v4_conn_request,
995         .syn_recv_sock     = dccp_v4_request_recv_sock,
996         .net_header_len    = sizeof(struct iphdr),
997         .setsockopt        = ip_setsockopt,
998         .getsockopt        = ip_getsockopt,
999         .addr2sockaddr     = inet_csk_addr2sockaddr,
1000         .sockaddr_len      = sizeof(struct sockaddr_in),
1001 #ifdef CONFIG_COMPAT
1002         .compat_setsockopt = compat_ip_setsockopt,
1003         .compat_getsockopt = compat_ip_getsockopt,
1004 #endif
1005 };
1006
1007 static int dccp_v4_init_sock(struct sock *sk)
1008 {
1009         static __u8 dccp_v4_ctl_sock_initialized;
1010         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
1011
1012         if (err == 0) {
1013                 if (unlikely(!dccp_v4_ctl_sock_initialized))
1014                         dccp_v4_ctl_sock_initialized = 1;
1015                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1016         }
1017
1018         return err;
1019 }
1020
1021 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1022 {
1023         kfree(inet_rsk(req)->opt);
1024 }
1025
1026 static struct request_sock_ops dccp_request_sock_ops = {
1027         .family         = PF_INET,
1028         .obj_size       = sizeof(struct dccp_request_sock),
1029         .rtx_syn_ack    = dccp_v4_send_response,
1030         .send_ack       = dccp_v4_reqsk_send_ack,
1031         .destructor     = dccp_v4_reqsk_destructor,
1032         .send_reset     = dccp_v4_ctl_send_reset,
1033 };
1034
1035 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1036         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1037 };
1038
1039 static struct proto dccp_v4_prot = {
1040         .name                   = "DCCP",
1041         .owner                  = THIS_MODULE,
1042         .close                  = dccp_close,
1043         .connect                = dccp_v4_connect,
1044         .disconnect             = dccp_disconnect,
1045         .ioctl                  = dccp_ioctl,
1046         .init                   = dccp_v4_init_sock,
1047         .setsockopt             = dccp_setsockopt,
1048         .getsockopt             = dccp_getsockopt,
1049         .sendmsg                = dccp_sendmsg,
1050         .recvmsg                = dccp_recvmsg,
1051         .backlog_rcv            = dccp_v4_do_rcv,
1052         .hash                   = dccp_hash,
1053         .unhash                 = dccp_unhash,
1054         .accept                 = inet_csk_accept,
1055         .get_port               = dccp_v4_get_port,
1056         .shutdown               = dccp_shutdown,
1057         .destroy                = dccp_destroy_sock,
1058         .orphan_count           = &dccp_orphan_count,
1059         .max_header             = MAX_DCCP_HEADER,
1060         .obj_size               = sizeof(struct dccp_sock),
1061         .rsk_prot               = &dccp_request_sock_ops,
1062         .twsk_prot              = &dccp_timewait_sock_ops,
1063 #ifdef CONFIG_COMPAT
1064         .compat_setsockopt      = compat_dccp_setsockopt,
1065         .compat_getsockopt      = compat_dccp_getsockopt,
1066 #endif
1067 };
1068
1069 static struct net_protocol dccp_v4_protocol = {
1070         .handler        = dccp_v4_rcv,
1071         .err_handler    = dccp_v4_err,
1072         .no_policy      = 1,
1073 };
1074
1075 static const struct proto_ops inet_dccp_ops = {
1076         .family            = PF_INET,
1077         .owner             = THIS_MODULE,
1078         .release           = inet_release,
1079         .bind              = inet_bind,
1080         .connect           = inet_stream_connect,
1081         .socketpair        = sock_no_socketpair,
1082         .accept            = inet_accept,
1083         .getname           = inet_getname,
1084         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1085         .poll              = dccp_poll,
1086         .ioctl             = inet_ioctl,
1087         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1088         .listen            = inet_dccp_listen,
1089         .shutdown          = inet_shutdown,
1090         .setsockopt        = sock_common_setsockopt,
1091         .getsockopt        = sock_common_getsockopt,
1092         .sendmsg           = inet_sendmsg,
1093         .recvmsg           = sock_common_recvmsg,
1094         .mmap              = sock_no_mmap,
1095         .sendpage          = sock_no_sendpage,
1096 #ifdef CONFIG_COMPAT
1097         .compat_setsockopt = compat_sock_common_setsockopt,
1098         .compat_getsockopt = compat_sock_common_getsockopt,
1099 #endif
1100 };
1101
1102 static struct inet_protosw dccp_v4_protosw = {
1103         .type           = SOCK_DCCP,
1104         .protocol       = IPPROTO_DCCP,
1105         .prot           = &dccp_v4_prot,
1106         .ops            = &inet_dccp_ops,
1107         .capability     = -1,
1108         .no_check       = 0,
1109         .flags          = INET_PROTOSW_ICSK,
1110 };
1111
1112 static int __init dccp_v4_init(void)
1113 {
1114         int err = proto_register(&dccp_v4_prot, 1);
1115
1116         if (err != 0)
1117                 goto out;
1118
1119         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1120         if (err != 0)
1121                 goto out_proto_unregister;
1122
1123         inet_register_protosw(&dccp_v4_protosw);
1124
1125         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1126                                        SOCK_DCCP, IPPROTO_DCCP);
1127         if (err)
1128                 goto out_unregister_protosw;
1129 out:
1130         return err;
1131 out_unregister_protosw:
1132         inet_unregister_protosw(&dccp_v4_protosw);
1133         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1134 out_proto_unregister:
1135         proto_unregister(&dccp_v4_prot);
1136         goto out;
1137 }
1138
1139 static void __exit dccp_v4_exit(void)
1140 {
1141         inet_unregister_protosw(&dccp_v4_protosw);
1142         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1143         proto_unregister(&dccp_v4_prot);
1144 }
1145
1146 module_init(dccp_v4_init);
1147 module_exit(dccp_v4_exit);
1148
1149 /*
1150  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1151  * values directly, Also cover the case where the protocol is not specified,
1152  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1153  */
1154 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1155 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1156 MODULE_LICENSE("GPL");
1157 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1158 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");