Merge branch 'master' of git://oak/home/sfr/kernels/iseries/work
[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         __be32 daddr, nexthop;
54         int tmp;
55         int err;
56
57         dp->dccps_role = DCCP_ROLE_CLIENT;
58
59         if (addr_len < sizeof(struct sockaddr_in))
60                 return -EINVAL;
61
62         if (usin->sin_family != AF_INET)
63                 return -EAFNOSUPPORT;
64
65         nexthop = daddr = usin->sin_addr.s_addr;
66         if (inet->opt != NULL && inet->opt->srr) {
67                 if (daddr == 0)
68                         return -EINVAL;
69                 nexthop = inet->opt->faddr;
70         }
71
72         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
73                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
74                                IPPROTO_DCCP,
75                                inet->sport, usin->sin_port, sk);
76         if (tmp < 0)
77                 return tmp;
78
79         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
80                 ip_rt_put(rt);
81                 return -ENETUNREACH;
82         }
83
84         if (inet->opt == NULL || !inet->opt->srr)
85                 daddr = rt->rt_dst;
86
87         if (inet->saddr == 0)
88                 inet->saddr = rt->rt_src;
89         inet->rcv_saddr = inet->saddr;
90
91         inet->dport = usin->sin_port;
92         inet->daddr = daddr;
93
94         inet_csk(sk)->icsk_ext_hdr_len = 0;
95         if (inet->opt != NULL)
96                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
97         /*
98          * Socket identity is still unknown (sport may be zero).
99          * However we set state to DCCP_REQUESTING and not releasing socket
100          * lock select source port, enter ourselves into the hash tables and
101          * complete initialization after this.
102          */
103         dccp_set_state(sk, DCCP_REQUESTING);
104         err = inet_hash_connect(&dccp_death_row, sk);
105         if (err != 0)
106                 goto failure;
107
108         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
109                                 sk);
110         if (err != 0)
111                 goto failure;
112
113         /* OK, now commit destination to socket.  */
114         sk_setup_caps(sk, &rt->u.dst);
115
116         dp->dccps_gar =
117                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
118                                                             inet->daddr,
119                                                             inet->sport,
120                                                             usin->sin_port);
121         dccp_update_gss(sk, dp->dccps_iss);
122
123         inet->id = dp->dccps_iss ^ jiffies;
124
125         err = dccp_connect(sk);
126         rt = NULL;
127         if (err != 0)
128                 goto failure;
129 out:
130         return err;
131 failure:
132         /*
133          * This unhashes the socket and releases the local port, if necessary.
134          */
135         dccp_set_state(sk, DCCP_CLOSED);
136         ip_rt_put(rt);
137         sk->sk_route_caps = 0;
138         inet->dport = 0;
139         goto out;
140 }
141
142 EXPORT_SYMBOL_GPL(dccp_v4_connect);
143
144 /*
145  * This routine does path mtu discovery as defined in RFC1191.
146  */
147 static inline void dccp_do_pmtu_discovery(struct sock *sk,
148                                           const struct iphdr *iph,
149                                           u32 mtu)
150 {
151         struct dst_entry *dst;
152         const struct inet_sock *inet = inet_sk(sk);
153         const struct dccp_sock *dp = dccp_sk(sk);
154
155         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
156          * send out by Linux are always < 576bytes so they should go through
157          * unfragmented).
158          */
159         if (sk->sk_state == DCCP_LISTEN)
160                 return;
161
162         /* We don't check in the destentry if pmtu discovery is forbidden
163          * on this route. We just assume that no packet_to_big packets
164          * are send back when pmtu discovery is not active.
165          * There is a small race when the user changes this flag in the
166          * route, but I think that's acceptable.
167          */
168         if ((dst = __sk_dst_check(sk, 0)) == NULL)
169                 return;
170
171         dst->ops->update_pmtu(dst, mtu);
172
173         /* Something is about to be wrong... Remember soft error
174          * for the case, if this connection will not able to recover.
175          */
176         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
177                 sk->sk_err_soft = EMSGSIZE;
178
179         mtu = dst_mtu(dst);
180
181         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
182             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
183                 dccp_sync_mss(sk, mtu);
184
185                 /*
186                  * From: draft-ietf-dccp-spec-11.txt
187                  *
188                  *      DCCP-Sync packets are the best choice for upward
189                  *      probing, since DCCP-Sync probes do not risk application
190                  *      data loss.
191                  */
192                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
193         } /* else let the usual retransmit timer handle it */
194 }
195
196 static void dccp_v4_reqsk_send_ack(struct sk_buff *rxskb,
197                                    struct request_sock *req)
198 {
199         int err;
200         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
201         const u32 dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
202                                      sizeof(struct dccp_hdr_ext) +
203                                      sizeof(struct dccp_hdr_ack_bits);
204         struct sk_buff *skb;
205
206         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
207                 return;
208
209         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header, GFP_ATOMIC);
210         if (skb == NULL)
211                 return;
212
213         /* Reserve space for headers. */
214         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
215
216         skb->dst = dst_clone(rxskb->dst);
217
218         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
219         dh = dccp_hdr(skb);
220         memset(dh, 0, dccp_hdr_ack_len);
221
222         /* Build DCCP header and checksum it. */
223         dh->dccph_type     = DCCP_PKT_ACK;
224         dh->dccph_sport    = rxdh->dccph_dport;
225         dh->dccph_dport    = rxdh->dccph_sport;
226         dh->dccph_doff     = dccp_hdr_ack_len / 4;
227         dh->dccph_x        = 1;
228
229         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
230         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
231                          DCCP_SKB_CB(rxskb)->dccpd_seq);
232
233         bh_lock_sock(dccp_v4_ctl_socket->sk);
234         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
235                                     rxskb->nh.iph->daddr,
236                                     rxskb->nh.iph->saddr, NULL);
237         bh_unlock_sock(dccp_v4_ctl_socket->sk);
238
239         if (err == NET_XMIT_CN || err == 0) {
240                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
241                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
242         }
243 }
244
245 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
246                                  struct dst_entry *dst)
247 {
248         int err = -1;
249         struct sk_buff *skb;
250
251         /* First, grab a route. */
252         
253         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
254                 goto out;
255
256         skb = dccp_make_response(sk, dst, req);
257         if (skb != NULL) {
258                 const struct inet_request_sock *ireq = inet_rsk(req);
259                 struct dccp_hdr *dh = dccp_hdr(skb);
260
261                 dh->dccph_checksum = dccp_v4_checksum(skb, ireq->loc_addr,
262                                                       ireq->rmt_addr);
263                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
264                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
265                                             ireq->rmt_addr,
266                                             ireq->opt);
267                 if (err == NET_XMIT_CN)
268                         err = 0;
269         }
270
271 out:
272         dst_release(dst);
273         return err;
274 }
275
276 /*
277  * This routine is called by the ICMP module when it gets some sort of error
278  * condition. If err < 0 then the socket should be closed and the error
279  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
280  * After adjustment header points to the first 8 bytes of the tcp header. We
281  * need to find the appropriate port.
282  *
283  * The locking strategy used here is very "optimistic". When someone else
284  * accesses the socket the ICMP is just dropped and for some paths there is no
285  * check at all. A more general error queue to queue errors for later handling
286  * is probably better.
287  */
288 static void dccp_v4_err(struct sk_buff *skb, u32 info)
289 {
290         const struct iphdr *iph = (struct iphdr *)skb->data;
291         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
292                                                         (iph->ihl << 2));
293         struct dccp_sock *dp;
294         struct inet_sock *inet;
295         const int type = skb->h.icmph->type;
296         const int code = skb->h.icmph->code;
297         struct sock *sk;
298         __u64 seq;
299         int err;
300
301         if (skb->len < (iph->ihl << 2) + 8) {
302                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
303                 return;
304         }
305
306         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
307                          iph->saddr, dh->dccph_sport, inet_iif(skb));
308         if (sk == NULL) {
309                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
310                 return;
311         }
312
313         if (sk->sk_state == DCCP_TIME_WAIT) {
314                 inet_twsk_put((struct inet_timewait_sock *)sk);
315                 return;
316         }
317
318         bh_lock_sock(sk);
319         /* If too many ICMPs get dropped on busy
320          * servers this needs to be solved differently.
321          */
322         if (sock_owned_by_user(sk))
323                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
324
325         if (sk->sk_state == DCCP_CLOSED)
326                 goto out;
327
328         dp = dccp_sk(sk);
329         seq = dccp_hdr_seq(skb);
330         if (sk->sk_state != DCCP_LISTEN &&
331             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
332                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
333                 goto out;
334         }
335
336         switch (type) {
337         case ICMP_SOURCE_QUENCH:
338                 /* Just silently ignore these. */
339                 goto out;
340         case ICMP_PARAMETERPROB:
341                 err = EPROTO;
342                 break;
343         case ICMP_DEST_UNREACH:
344                 if (code > NR_ICMP_UNREACH)
345                         goto out;
346
347                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
348                         if (!sock_owned_by_user(sk))
349                                 dccp_do_pmtu_discovery(sk, iph, info);
350                         goto out;
351                 }
352
353                 err = icmp_err_convert[code].errno;
354                 break;
355         case ICMP_TIME_EXCEEDED:
356                 err = EHOSTUNREACH;
357                 break;
358         default:
359                 goto out;
360         }
361
362         switch (sk->sk_state) {
363                 struct request_sock *req , **prev;
364         case DCCP_LISTEN:
365                 if (sock_owned_by_user(sk))
366                         goto out;
367                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
368                                           iph->daddr, iph->saddr);
369                 if (!req)
370                         goto out;
371
372                 /*
373                  * ICMPs are not backlogged, hence we cannot get an established
374                  * socket here.
375                  */
376                 BUG_TRAP(!req->sk);
377
378                 if (seq != dccp_rsk(req)->dreq_iss) {
379                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
380                         goto out;
381                 }
382                 /*
383                  * Still in RESPOND, just remove it silently.
384                  * There is no good way to pass the error to the newly
385                  * created socket, and POSIX does not want network
386                  * errors returned from accept().
387                  */
388                 inet_csk_reqsk_queue_drop(sk, req, prev);
389                 goto out;
390
391         case DCCP_REQUESTING:
392         case DCCP_RESPOND:
393                 if (!sock_owned_by_user(sk)) {
394                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
395                         sk->sk_err = err;
396
397                         sk->sk_error_report(sk);
398
399                         dccp_done(sk);
400                 } else
401                         sk->sk_err_soft = err;
402                 goto out;
403         }
404
405         /* If we've already connected we will keep trying
406          * until we time out, or the user gives up.
407          *
408          * rfc1122 4.2.3.9 allows to consider as hard errors
409          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
410          * but it is obsoleted by pmtu discovery).
411          *
412          * Note, that in modern internet, where routing is unreliable
413          * and in each dark corner broken firewalls sit, sending random
414          * errors ordered by their masters even this two messages finally lose
415          * their original sense (even Linux sends invalid PORT_UNREACHs)
416          *
417          * Now we are in compliance with RFCs.
418          *                                                      --ANK (980905)
419          */
420
421         inet = inet_sk(sk);
422         if (!sock_owned_by_user(sk) && inet->recverr) {
423                 sk->sk_err = err;
424                 sk->sk_error_report(sk);
425         } else /* Only an error on timeout */
426                 sk->sk_err_soft = err;
427 out:
428         bh_unlock_sock(sk);
429         sock_put(sk);
430 }
431
432 /* This routine computes an IPv4 DCCP checksum. */
433 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
434 {
435         const struct inet_sock *inet = inet_sk(sk);
436         struct dccp_hdr *dh = dccp_hdr(skb);
437
438         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
439 }
440
441 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
442
443 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
444                                         const struct sk_buff *skb)
445 {
446         return secure_dccp_sequence_number(skb->nh.iph->daddr,
447                                            skb->nh.iph->saddr,
448                                            dccp_hdr(skb)->dccph_dport,
449                                            dccp_hdr(skb)->dccph_sport);
450 }
451
452 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
453 {
454         struct inet_request_sock *ireq;
455         struct dccp_sock dp;
456         struct request_sock *req;
457         struct dccp_request_sock *dreq;
458         const __be32 saddr = skb->nh.iph->saddr;
459         const __be32 daddr = skb->nh.iph->daddr;
460         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
461         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
462         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
463
464         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
465         if (((struct rtable *)skb->dst)->rt_flags &
466             (RTCF_BROADCAST | RTCF_MULTICAST)) {
467                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
468                 goto drop;
469         }
470
471         if (dccp_bad_service_code(sk, service)) {
472                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
473                 goto drop;
474         }
475         /*
476          * TW buckets are converted to open requests without
477          * limitations, they conserve resources and peer is
478          * evidently real one.
479          */
480         if (inet_csk_reqsk_queue_is_full(sk))
481                 goto drop;
482
483         /*
484          * Accept backlog is full. If we have already queued enough
485          * of warm entries in syn queue, drop request. It is better than
486          * clogging syn queue with openreqs with exponentially increasing
487          * timeout.
488          */
489         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
490                 goto drop;
491
492         req = reqsk_alloc(sk->sk_prot->rsk_prot);
493         if (req == NULL)
494                 goto drop;
495
496         if (dccp_parse_options(sk, skb))
497                 goto drop_and_free;
498
499         dccp_openreq_init(req, &dp, skb);
500
501         if (security_inet_conn_request(sk, skb, req))
502                 goto drop_and_free;
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, 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         security_skb_classify_flow(skb, &fl);
682         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
683                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
684                 return NULL;
685         }
686
687         return &rt->u.dst;
688 }
689
690 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
691 {
692         int err;
693         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
694         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
695                                        sizeof(struct dccp_hdr_ext) +
696                                        sizeof(struct dccp_hdr_reset);
697         struct sk_buff *skb;
698         struct dst_entry *dst;
699         u64 seqno;
700
701         /* Never send a reset in response to a reset. */
702         if (rxdh->dccph_type == DCCP_PKT_RESET)
703                 return;
704
705         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
706                 return;
707
708         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
709         if (dst == NULL)
710                 return;
711
712         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header,
713                         GFP_ATOMIC);
714         if (skb == NULL)
715                 goto out;
716
717         /* Reserve space for headers. */
718         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
719         skb->dst = dst_clone(dst);
720
721         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
722         dh = dccp_hdr(skb);
723         memset(dh, 0, dccp_hdr_reset_len);
724
725         /* Build DCCP header and checksum it. */
726         dh->dccph_type     = DCCP_PKT_RESET;
727         dh->dccph_sport    = rxdh->dccph_dport;
728         dh->dccph_dport    = rxdh->dccph_sport;
729         dh->dccph_doff     = dccp_hdr_reset_len / 4;
730         dh->dccph_x        = 1;
731         dccp_hdr_reset(skb)->dccph_reset_code =
732                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
733
734         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
735         seqno = 0;
736         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
737                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
738
739         dccp_hdr_set_seq(dh, seqno);
740         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
741                          DCCP_SKB_CB(rxskb)->dccpd_seq);
742
743         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
744                                               rxskb->nh.iph->daddr);
745
746         bh_lock_sock(dccp_v4_ctl_socket->sk);
747         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
748                                     rxskb->nh.iph->daddr,
749                                     rxskb->nh.iph->saddr, NULL);
750         bh_unlock_sock(dccp_v4_ctl_socket->sk);
751
752         if (err == NET_XMIT_CN || err == 0) {
753                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
754                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
755         }
756 out:
757          dst_release(dst);
758 }
759
760 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
761 {
762         struct dccp_hdr *dh = dccp_hdr(skb);
763
764         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
765                 if (dccp_rcv_established(sk, skb, dh, skb->len))
766                         goto reset;
767                 return 0;
768         }
769
770         /*
771          *  Step 3: Process LISTEN state
772          *     If S.state == LISTEN,
773          *        If P.type == Request or P contains a valid Init Cookie
774          *              option,
775          *           * Must scan the packet's options to check for an Init
776          *              Cookie.  Only the Init Cookie is processed here,
777          *              however; other options are processed in Step 8.  This
778          *              scan need only be performed if the endpoint uses Init
779          *              Cookies *
780          *           * Generate a new socket and switch to that socket *
781          *           Set S := new socket for this port pair
782          *           S.state = RESPOND
783          *           Choose S.ISS (initial seqno) or set from Init Cookie
784          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
785          *           Continue with S.state == RESPOND
786          *           * A Response packet will be generated in Step 11 *
787          *        Otherwise,
788          *           Generate Reset(No Connection) unless P.type == Reset
789          *           Drop packet and return
790          *
791          * NOTE: the check for the packet types is done in
792          *       dccp_rcv_state_process
793          */
794         if (sk->sk_state == DCCP_LISTEN) {
795                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
796
797                 if (nsk == NULL)
798                         goto discard;
799
800                 if (nsk != sk) {
801                         if (dccp_child_process(sk, nsk, skb))
802                                 goto reset;
803                         return 0;
804                 }
805         }
806
807         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
808                 goto reset;
809         return 0;
810
811 reset:
812         dccp_v4_ctl_send_reset(skb);
813 discard:
814         kfree_skb(skb);
815         return 0;
816 }
817
818 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
819
820 int dccp_invalid_packet(struct sk_buff *skb)
821 {
822         const struct dccp_hdr *dh;
823
824         if (skb->pkt_type != PACKET_HOST)
825                 return 1;
826
827         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
828                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
829                 return 1;
830         }
831
832         dh = dccp_hdr(skb);
833
834         /* If the packet type is not understood, drop packet and return */
835         if (dh->dccph_type >= DCCP_PKT_INVALID) {
836                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
837                 return 1;
838         }
839
840         /*
841          * If P.Data Offset is too small for packet type, or too large for
842          * packet, drop packet and return
843          */
844         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
845                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
846                                             "too small 1\n",
847                                dh->dccph_doff);
848                 return 1;
849         }
850
851         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
852                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
853                                             "too small 2\n",
854                                dh->dccph_doff);
855                 return 1;
856         }
857
858         dh = dccp_hdr(skb);
859
860         /*
861          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
862          * has short sequence numbers), drop packet and return
863          */
864         if (dh->dccph_x == 0 &&
865             dh->dccph_type != DCCP_PKT_DATA &&
866             dh->dccph_type != DCCP_PKT_ACK &&
867             dh->dccph_type != DCCP_PKT_DATAACK) {
868                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
869                                             "nor DataAck and P.X == 0\n",
870                                dccp_packet_name(dh->dccph_type));
871                 return 1;
872         }
873
874         return 0;
875 }
876
877 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
878
879 /* this is called when real data arrives */
880 static int dccp_v4_rcv(struct sk_buff *skb)
881 {
882         const struct dccp_hdr *dh;
883         struct sock *sk;
884
885         /* Step 1: Check header basics: */
886
887         if (dccp_invalid_packet(skb))
888                 goto discard_it;
889
890         /* If the header checksum is incorrect, drop packet and return */
891         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
892                                     skb->nh.iph->daddr) < 0) {
893                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
894                                __FUNCTION__);
895                 goto discard_it;
896         }
897
898         dh = dccp_hdr(skb);
899
900         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
901         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
902
903         dccp_pr_debug("%8.8s "
904                       "src=%u.%u.%u.%u@%-5d "
905                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
906                       dccp_packet_name(dh->dccph_type),
907                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
908                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
909                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
910
911         if (dccp_packet_without_ack(skb)) {
912                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
913                 dccp_pr_debug_cat("\n");
914         } else {
915                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
916                 dccp_pr_debug_cat(", ack=%llu\n",
917                                   (unsigned long long)
918                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
919         }
920
921         /* Step 2:
922          *      Look up flow ID in table and get corresponding socket */
923         sk = __inet_lookup(&dccp_hashinfo,
924                            skb->nh.iph->saddr, dh->dccph_sport,
925                            skb->nh.iph->daddr, dh->dccph_dport,
926                            inet_iif(skb));
927
928         /* 
929          * Step 2:
930          *      If no socket ...
931          *              Generate Reset(No Connection) unless P.type == Reset
932          *              Drop packet and return
933          */
934         if (sk == NULL) {
935                 dccp_pr_debug("failed to look up flow ID in table and "
936                               "get corresponding socket\n");
937                 goto no_dccp_socket;
938         }
939
940         /* 
941          * Step 2:
942          *      ... or S.state == TIMEWAIT,
943          *              Generate Reset(No Connection) unless P.type == Reset
944          *              Drop packet and return
945          */
946                
947         if (sk->sk_state == DCCP_TIME_WAIT) {
948                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
949                               "do_time_wait\n");
950                 goto do_time_wait;
951         }
952
953         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
954                 goto discard_and_relse;
955         nf_reset(skb);
956
957         return sk_receive_skb(sk, skb);
958
959 no_dccp_socket:
960         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
961                 goto discard_it;
962         /*
963          * Step 2:
964          *              Generate Reset(No Connection) unless P.type == Reset
965          *              Drop packet and return
966          */
967         if (dh->dccph_type != DCCP_PKT_RESET) {
968                 DCCP_SKB_CB(skb)->dccpd_reset_code =
969                                         DCCP_RESET_CODE_NO_CONNECTION;
970                 dccp_v4_ctl_send_reset(skb);
971         }
972
973 discard_it:
974         /* Discard frame. */
975         kfree_skb(skb);
976         return 0;
977
978 discard_and_relse:
979         sock_put(sk);
980         goto discard_it;
981
982 do_time_wait:
983         inet_twsk_put((struct inet_timewait_sock *)sk);
984         goto no_dccp_socket;
985 }
986
987 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
988         .queue_xmit        = ip_queue_xmit,
989         .send_check        = dccp_v4_send_check,
990         .rebuild_header    = inet_sk_rebuild_header,
991         .conn_request      = dccp_v4_conn_request,
992         .syn_recv_sock     = dccp_v4_request_recv_sock,
993         .net_header_len    = sizeof(struct iphdr),
994         .setsockopt        = ip_setsockopt,
995         .getsockopt        = ip_getsockopt,
996         .addr2sockaddr     = inet_csk_addr2sockaddr,
997         .sockaddr_len      = sizeof(struct sockaddr_in),
998 #ifdef CONFIG_COMPAT
999         .compat_setsockopt = compat_ip_setsockopt,
1000         .compat_getsockopt = compat_ip_getsockopt,
1001 #endif
1002 };
1003
1004 static int dccp_v4_init_sock(struct sock *sk)
1005 {
1006         static __u8 dccp_v4_ctl_sock_initialized;
1007         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
1008
1009         if (err == 0) {
1010                 if (unlikely(!dccp_v4_ctl_sock_initialized))
1011                         dccp_v4_ctl_sock_initialized = 1;
1012                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1013         }
1014
1015         return err;
1016 }
1017
1018 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1019 {
1020         kfree(inet_rsk(req)->opt);
1021 }
1022
1023 static struct request_sock_ops dccp_request_sock_ops = {
1024         .family         = PF_INET,
1025         .obj_size       = sizeof(struct dccp_request_sock),
1026         .rtx_syn_ack    = dccp_v4_send_response,
1027         .send_ack       = dccp_v4_reqsk_send_ack,
1028         .destructor     = dccp_v4_reqsk_destructor,
1029         .send_reset     = dccp_v4_ctl_send_reset,
1030 };
1031
1032 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1033         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1034 };
1035
1036 static struct proto dccp_v4_prot = {
1037         .name                   = "DCCP",
1038         .owner                  = THIS_MODULE,
1039         .close                  = dccp_close,
1040         .connect                = dccp_v4_connect,
1041         .disconnect             = dccp_disconnect,
1042         .ioctl                  = dccp_ioctl,
1043         .init                   = dccp_v4_init_sock,
1044         .setsockopt             = dccp_setsockopt,
1045         .getsockopt             = dccp_getsockopt,
1046         .sendmsg                = dccp_sendmsg,
1047         .recvmsg                = dccp_recvmsg,
1048         .backlog_rcv            = dccp_v4_do_rcv,
1049         .hash                   = dccp_hash,
1050         .unhash                 = dccp_unhash,
1051         .accept                 = inet_csk_accept,
1052         .get_port               = dccp_v4_get_port,
1053         .shutdown               = dccp_shutdown,
1054         .destroy                = dccp_destroy_sock,
1055         .orphan_count           = &dccp_orphan_count,
1056         .max_header             = MAX_DCCP_HEADER,
1057         .obj_size               = sizeof(struct dccp_sock),
1058         .rsk_prot               = &dccp_request_sock_ops,
1059         .twsk_prot              = &dccp_timewait_sock_ops,
1060 #ifdef CONFIG_COMPAT
1061         .compat_setsockopt      = compat_dccp_setsockopt,
1062         .compat_getsockopt      = compat_dccp_getsockopt,
1063 #endif
1064 };
1065
1066 static struct net_protocol dccp_v4_protocol = {
1067         .handler        = dccp_v4_rcv,
1068         .err_handler    = dccp_v4_err,
1069         .no_policy      = 1,
1070 };
1071
1072 static const struct proto_ops inet_dccp_ops = {
1073         .family            = PF_INET,
1074         .owner             = THIS_MODULE,
1075         .release           = inet_release,
1076         .bind              = inet_bind,
1077         .connect           = inet_stream_connect,
1078         .socketpair        = sock_no_socketpair,
1079         .accept            = inet_accept,
1080         .getname           = inet_getname,
1081         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1082         .poll              = dccp_poll,
1083         .ioctl             = inet_ioctl,
1084         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1085         .listen            = inet_dccp_listen,
1086         .shutdown          = inet_shutdown,
1087         .setsockopt        = sock_common_setsockopt,
1088         .getsockopt        = sock_common_getsockopt,
1089         .sendmsg           = inet_sendmsg,
1090         .recvmsg           = sock_common_recvmsg,
1091         .mmap              = sock_no_mmap,
1092         .sendpage          = sock_no_sendpage,
1093 #ifdef CONFIG_COMPAT
1094         .compat_setsockopt = compat_sock_common_setsockopt,
1095         .compat_getsockopt = compat_sock_common_getsockopt,
1096 #endif
1097 };
1098
1099 static struct inet_protosw dccp_v4_protosw = {
1100         .type           = SOCK_DCCP,
1101         .protocol       = IPPROTO_DCCP,
1102         .prot           = &dccp_v4_prot,
1103         .ops            = &inet_dccp_ops,
1104         .capability     = -1,
1105         .no_check       = 0,
1106         .flags          = INET_PROTOSW_ICSK,
1107 };
1108
1109 static int __init dccp_v4_init(void)
1110 {
1111         int err = proto_register(&dccp_v4_prot, 1);
1112
1113         if (err != 0)
1114                 goto out;
1115
1116         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1117         if (err != 0)
1118                 goto out_proto_unregister;
1119
1120         inet_register_protosw(&dccp_v4_protosw);
1121
1122         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1123                                        SOCK_DCCP, IPPROTO_DCCP);
1124         if (err)
1125                 goto out_unregister_protosw;
1126 out:
1127         return err;
1128 out_unregister_protosw:
1129         inet_unregister_protosw(&dccp_v4_protosw);
1130         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1131 out_proto_unregister:
1132         proto_unregister(&dccp_v4_prot);
1133         goto out;
1134 }
1135
1136 static void __exit dccp_v4_exit(void)
1137 {
1138         inet_unregister_protosw(&dccp_v4_protosw);
1139         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1140         proto_unregister(&dccp_v4_prot);
1141 }
1142
1143 module_init(dccp_v4_init);
1144 module_exit(dccp_v4_exit);
1145
1146 /*
1147  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1148  * values directly, Also cover the case where the protocol is not specified,
1149  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1150  */
1151 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1152 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1153 MODULE_LICENSE("GPL");
1154 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1155 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");