[NET]: {get|set}sockopt compatibility layer
[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/config.h>
14 #include <linux/dccp.h>
15 #include <linux/icmp.h>
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/random.h>
19
20 #include <net/icmp.h>
21 #include <net/inet_common.h>
22 #include <net/inet_hashtables.h>
23 #include <net/inet_sock.h>
24 #include <net/protocol.h>
25 #include <net/sock.h>
26 #include <net/timewait_sock.h>
27 #include <net/tcp_states.h>
28 #include <net/xfrm.h>
29
30 #include "ackvec.h"
31 #include "ccid.h"
32 #include "dccp.h"
33 #include "feat.h"
34
35 /*
36  * This is the global socket data structure used for responding to
37  * the Out-of-the-blue (OOTB) packets. A control sock will be created
38  * for this socket at the initialization time.
39  */
40 static struct socket *dccp_v4_ctl_socket;
41
42 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
43 {
44         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
45                                  inet_csk_bind_conflict);
46 }
47
48 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
49 {
50         struct inet_sock *inet = inet_sk(sk);
51         struct dccp_sock *dp = dccp_sk(sk);
52         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
53         struct rtable *rt;
54         u32 daddr, nexthop;
55         int tmp;
56         int err;
57
58         dp->dccps_role = DCCP_ROLE_CLIENT;
59
60         if (dccp_service_not_initialized(sk))
61                 return -EPROTO;
62
63         if (addr_len < sizeof(struct sockaddr_in))
64                 return -EINVAL;
65
66         if (usin->sin_family != AF_INET)
67                 return -EAFNOSUPPORT;
68
69         nexthop = daddr = usin->sin_addr.s_addr;
70         if (inet->opt != NULL && inet->opt->srr) {
71                 if (daddr == 0)
72                         return -EINVAL;
73                 nexthop = inet->opt->faddr;
74         }
75
76         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
77                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
78                                IPPROTO_DCCP,
79                                inet->sport, usin->sin_port, sk);
80         if (tmp < 0)
81                 return tmp;
82
83         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
84                 ip_rt_put(rt);
85                 return -ENETUNREACH;
86         }
87
88         if (inet->opt == NULL || !inet->opt->srr)
89                 daddr = rt->rt_dst;
90
91         if (inet->saddr == 0)
92                 inet->saddr = rt->rt_src;
93         inet->rcv_saddr = inet->saddr;
94
95         inet->dport = usin->sin_port;
96         inet->daddr = daddr;
97
98         inet_csk(sk)->icsk_ext_hdr_len = 0;
99         if (inet->opt != NULL)
100                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
101         /*
102          * Socket identity is still unknown (sport may be zero).
103          * However we set state to DCCP_REQUESTING and not releasing socket
104          * lock select source port, enter ourselves into the hash tables and
105          * complete initialization after this.
106          */
107         dccp_set_state(sk, DCCP_REQUESTING);
108         err = inet_hash_connect(&dccp_death_row, sk);
109         if (err != 0)
110                 goto failure;
111
112         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
113                                 sk);
114         if (err != 0)
115                 goto failure;
116
117         /* OK, now commit destination to socket.  */
118         sk_setup_caps(sk, &rt->u.dst);
119
120         dp->dccps_gar =
121                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
122                                                             inet->daddr,
123                                                             inet->sport,
124                                                             usin->sin_port);
125         dccp_update_gss(sk, dp->dccps_iss);
126
127         inet->id = dp->dccps_iss ^ jiffies;
128
129         err = dccp_connect(sk);
130         rt = NULL;
131         if (err != 0)
132                 goto failure;
133 out:
134         return err;
135 failure:
136         /*
137          * This unhashes the socket and releases the local port, if necessary.
138          */
139         dccp_set_state(sk, DCCP_CLOSED);
140         ip_rt_put(rt);
141         sk->sk_route_caps = 0;
142         inet->dport = 0;
143         goto out;
144 }
145
146 EXPORT_SYMBOL_GPL(dccp_v4_connect);
147
148 /*
149  * This routine does path mtu discovery as defined in RFC1191.
150  */
151 static inline void dccp_do_pmtu_discovery(struct sock *sk,
152                                           const struct iphdr *iph,
153                                           u32 mtu)
154 {
155         struct dst_entry *dst;
156         const struct inet_sock *inet = inet_sk(sk);
157         const struct dccp_sock *dp = dccp_sk(sk);
158
159         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
160          * send out by Linux are always < 576bytes so they should go through
161          * unfragmented).
162          */
163         if (sk->sk_state == DCCP_LISTEN)
164                 return;
165
166         /* We don't check in the destentry if pmtu discovery is forbidden
167          * on this route. We just assume that no packet_to_big packets
168          * are send back when pmtu discovery is not active.
169          * There is a small race when the user changes this flag in the
170          * route, but I think that's acceptable.
171          */
172         if ((dst = __sk_dst_check(sk, 0)) == NULL)
173                 return;
174
175         dst->ops->update_pmtu(dst, mtu);
176
177         /* Something is about to be wrong... Remember soft error
178          * for the case, if this connection will not able to recover.
179          */
180         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
181                 sk->sk_err_soft = EMSGSIZE;
182
183         mtu = dst_mtu(dst);
184
185         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
186             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
187                 dccp_sync_mss(sk, mtu);
188
189                 /*
190                  * From: draft-ietf-dccp-spec-11.txt
191                  *
192                  *      DCCP-Sync packets are the best choice for upward
193                  *      probing, since DCCP-Sync probes do not risk application
194                  *      data loss.
195                  */
196                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
197         } /* else let the usual retransmit timer handle it */
198 }
199
200 static void dccp_v4_reqsk_send_ack(struct sk_buff *rxskb,
201                                    struct request_sock *req)
202 {
203         int err;
204         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
205         const u32 dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
206                                      sizeof(struct dccp_hdr_ext) +
207                                      sizeof(struct dccp_hdr_ack_bits);
208         struct sk_buff *skb;
209
210         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
211                 return;
212
213         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header, GFP_ATOMIC);
214         if (skb == NULL)
215                 return;
216
217         /* Reserve space for headers. */
218         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
219
220         skb->dst = dst_clone(rxskb->dst);
221
222         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
223         dh = dccp_hdr(skb);
224         memset(dh, 0, dccp_hdr_ack_len);
225
226         /* Build DCCP header and checksum it. */
227         dh->dccph_type     = DCCP_PKT_ACK;
228         dh->dccph_sport    = rxdh->dccph_dport;
229         dh->dccph_dport    = rxdh->dccph_sport;
230         dh->dccph_doff     = dccp_hdr_ack_len / 4;
231         dh->dccph_x        = 1;
232
233         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
234         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
235                          DCCP_SKB_CB(rxskb)->dccpd_seq);
236
237         bh_lock_sock(dccp_v4_ctl_socket->sk);
238         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
239                                     rxskb->nh.iph->daddr,
240                                     rxskb->nh.iph->saddr, NULL);
241         bh_unlock_sock(dccp_v4_ctl_socket->sk);
242
243         if (err == NET_XMIT_CN || err == 0) {
244                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
245                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
246         }
247 }
248
249 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
250                                  struct dst_entry *dst)
251 {
252         int err = -1;
253         struct sk_buff *skb;
254
255         /* First, grab a route. */
256         
257         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
258                 goto out;
259
260         skb = dccp_make_response(sk, dst, req);
261         if (skb != NULL) {
262                 const struct inet_request_sock *ireq = inet_rsk(req);
263                 struct dccp_hdr *dh = dccp_hdr(skb);
264
265                 dh->dccph_checksum = dccp_v4_checksum(skb, ireq->loc_addr,
266                                                       ireq->rmt_addr);
267                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
268                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
269                                             ireq->rmt_addr,
270                                             ireq->opt);
271                 if (err == NET_XMIT_CN)
272                         err = 0;
273         }
274
275 out:
276         dst_release(dst);
277         return err;
278 }
279
280 /*
281  * This routine is called by the ICMP module when it gets some sort of error
282  * condition. If err < 0 then the socket should be closed and the error
283  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
284  * After adjustment header points to the first 8 bytes of the tcp header. We
285  * need to find the appropriate port.
286  *
287  * The locking strategy used here is very "optimistic". When someone else
288  * accesses the socket the ICMP is just dropped and for some paths there is no
289  * check at all. A more general error queue to queue errors for later handling
290  * is probably better.
291  */
292 static void dccp_v4_err(struct sk_buff *skb, u32 info)
293 {
294         const struct iphdr *iph = (struct iphdr *)skb->data;
295         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
296                                                         (iph->ihl << 2));
297         struct dccp_sock *dp;
298         struct inet_sock *inet;
299         const int type = skb->h.icmph->type;
300         const int code = skb->h.icmph->code;
301         struct sock *sk;
302         __u64 seq;
303         int err;
304
305         if (skb->len < (iph->ihl << 2) + 8) {
306                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
307                 return;
308         }
309
310         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
311                          iph->saddr, dh->dccph_sport, inet_iif(skb));
312         if (sk == NULL) {
313                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
314                 return;
315         }
316
317         if (sk->sk_state == DCCP_TIME_WAIT) {
318                 inet_twsk_put((struct inet_timewait_sock *)sk);
319                 return;
320         }
321
322         bh_lock_sock(sk);
323         /* If too many ICMPs get dropped on busy
324          * servers this needs to be solved differently.
325          */
326         if (sock_owned_by_user(sk))
327                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
328
329         if (sk->sk_state == DCCP_CLOSED)
330                 goto out;
331
332         dp = dccp_sk(sk);
333         seq = dccp_hdr_seq(skb);
334         if (sk->sk_state != DCCP_LISTEN &&
335             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
336                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
337                 goto out;
338         }
339
340         switch (type) {
341         case ICMP_SOURCE_QUENCH:
342                 /* Just silently ignore these. */
343                 goto out;
344         case ICMP_PARAMETERPROB:
345                 err = EPROTO;
346                 break;
347         case ICMP_DEST_UNREACH:
348                 if (code > NR_ICMP_UNREACH)
349                         goto out;
350
351                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
352                         if (!sock_owned_by_user(sk))
353                                 dccp_do_pmtu_discovery(sk, iph, info);
354                         goto out;
355                 }
356
357                 err = icmp_err_convert[code].errno;
358                 break;
359         case ICMP_TIME_EXCEEDED:
360                 err = EHOSTUNREACH;
361                 break;
362         default:
363                 goto out;
364         }
365
366         switch (sk->sk_state) {
367                 struct request_sock *req , **prev;
368         case DCCP_LISTEN:
369                 if (sock_owned_by_user(sk))
370                         goto out;
371                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
372                                           iph->daddr, iph->saddr);
373                 if (!req)
374                         goto out;
375
376                 /*
377                  * ICMPs are not backlogged, hence we cannot get an established
378                  * socket here.
379                  */
380                 BUG_TRAP(!req->sk);
381
382                 if (seq != dccp_rsk(req)->dreq_iss) {
383                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
384                         goto out;
385                 }
386                 /*
387                  * Still in RESPOND, just remove it silently.
388                  * There is no good way to pass the error to the newly
389                  * created socket, and POSIX does not want network
390                  * errors returned from accept().
391                  */
392                 inet_csk_reqsk_queue_drop(sk, req, prev);
393                 goto out;
394
395         case DCCP_REQUESTING:
396         case DCCP_RESPOND:
397                 if (!sock_owned_by_user(sk)) {
398                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
399                         sk->sk_err = err;
400
401                         sk->sk_error_report(sk);
402
403                         dccp_done(sk);
404                 } else
405                         sk->sk_err_soft = err;
406                 goto out;
407         }
408
409         /* If we've already connected we will keep trying
410          * until we time out, or the user gives up.
411          *
412          * rfc1122 4.2.3.9 allows to consider as hard errors
413          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
414          * but it is obsoleted by pmtu discovery).
415          *
416          * Note, that in modern internet, where routing is unreliable
417          * and in each dark corner broken firewalls sit, sending random
418          * errors ordered by their masters even this two messages finally lose
419          * their original sense (even Linux sends invalid PORT_UNREACHs)
420          *
421          * Now we are in compliance with RFCs.
422          *                                                      --ANK (980905)
423          */
424
425         inet = inet_sk(sk);
426         if (!sock_owned_by_user(sk) && inet->recverr) {
427                 sk->sk_err = err;
428                 sk->sk_error_report(sk);
429         } else /* Only an error on timeout */
430                 sk->sk_err_soft = err;
431 out:
432         bh_unlock_sock(sk);
433         sock_put(sk);
434 }
435
436 /* This routine computes an IPv4 DCCP checksum. */
437 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
438 {
439         const struct inet_sock *inet = inet_sk(sk);
440         struct dccp_hdr *dh = dccp_hdr(skb);
441
442         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
443 }
444
445 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
446
447 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
448                                         const struct sk_buff *skb)
449 {
450         return secure_dccp_sequence_number(skb->nh.iph->daddr,
451                                            skb->nh.iph->saddr,
452                                            dccp_hdr(skb)->dccph_dport,
453                                            dccp_hdr(skb)->dccph_sport);
454 }
455
456 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
457 {
458         struct inet_request_sock *ireq;
459         struct dccp_sock dp;
460         struct request_sock *req;
461         struct dccp_request_sock *dreq;
462         const __be32 saddr = skb->nh.iph->saddr;
463         const __be32 daddr = skb->nh.iph->daddr;
464         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
465         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
466         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
467
468         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
469         if (((struct rtable *)skb->dst)->rt_flags &
470             (RTCF_BROADCAST | RTCF_MULTICAST)) {
471                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
472                 goto drop;
473         }
474
475         if (dccp_bad_service_code(sk, service)) {
476                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
477                 goto drop;
478         }
479         /*
480          * TW buckets are converted to open requests without
481          * limitations, they conserve resources and peer is
482          * evidently real one.
483          */
484         if (inet_csk_reqsk_queue_is_full(sk))
485                 goto drop;
486
487         /*
488          * Accept backlog is full. If we have already queued enough
489          * of warm entries in syn queue, drop request. It is better than
490          * clogging syn queue with openreqs with exponentially increasing
491          * timeout.
492          */
493         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
494                 goto drop;
495
496         req = reqsk_alloc(sk->sk_prot->rsk_prot);
497         if (req == NULL)
498                 goto drop;
499
500         if (dccp_parse_options(sk, skb))
501                 goto drop;
502
503         dccp_openreq_init(req, &dp, skb);
504
505         ireq = inet_rsk(req);
506         ireq->loc_addr = daddr;
507         ireq->rmt_addr = saddr;
508         req->rcv_wnd    = 100; /* Fake, option parsing will get the
509                                   right value */
510         ireq->opt       = NULL;
511
512         /* 
513          * Step 3: Process LISTEN state
514          *
515          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
516          *
517          * In fact we defer setting S.GSR, S.SWL, S.SWH to
518          * dccp_create_openreq_child.
519          */
520         dreq = dccp_rsk(req);
521         dreq->dreq_isr     = dcb->dccpd_seq;
522         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
523         dreq->dreq_service = service;
524
525         if (dccp_v4_send_response(sk, req, NULL))
526                 goto drop_and_free;
527
528         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
529         return 0;
530
531 drop_and_free:
532         reqsk_free(req);
533 drop:
534         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
535         dcb->dccpd_reset_code = reset_code;
536         return -1;
537 }
538
539 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
540
541 /*
542  * The three way handshake has completed - we got a valid ACK or DATAACK -
543  * now create the new socket.
544  *
545  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
546  */
547 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
548                                        struct request_sock *req,
549                                        struct dst_entry *dst)
550 {
551         struct inet_request_sock *ireq;
552         struct inet_sock *newinet;
553         struct dccp_sock *newdp;
554         struct sock *newsk;
555
556         if (sk_acceptq_is_full(sk))
557                 goto exit_overflow;
558
559         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
560                 goto exit;
561
562         newsk = dccp_create_openreq_child(sk, req, skb);
563         if (newsk == NULL)
564                 goto exit;
565
566         sk_setup_caps(newsk, dst);
567
568         newdp              = dccp_sk(newsk);
569         newinet            = inet_sk(newsk);
570         ireq               = inet_rsk(req);
571         newinet->daddr     = ireq->rmt_addr;
572         newinet->rcv_saddr = ireq->loc_addr;
573         newinet->saddr     = ireq->loc_addr;
574         newinet->opt       = ireq->opt;
575         ireq->opt          = NULL;
576         newinet->mc_index  = inet_iif(skb);
577         newinet->mc_ttl    = skb->nh.iph->ttl;
578         newinet->id        = jiffies;
579
580         dccp_sync_mss(newsk, dst_mtu(dst));
581
582         __inet_hash(&dccp_hashinfo, newsk, 0);
583         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
584
585         return newsk;
586
587 exit_overflow:
588         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
589 exit:
590         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
591         dst_release(dst);
592         return NULL;
593 }
594
595 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
596
597 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
598 {
599         const struct dccp_hdr *dh = dccp_hdr(skb);
600         const struct iphdr *iph = skb->nh.iph;
601         struct sock *nsk;
602         struct request_sock **prev;
603         /* Find possible connection requests. */
604         struct request_sock *req = inet_csk_search_req(sk, &prev,
605                                                        dh->dccph_sport,
606                                                        iph->saddr, iph->daddr);
607         if (req != NULL)
608                 return dccp_check_req(sk, skb, req, prev);
609
610         nsk = __inet_lookup_established(&dccp_hashinfo,
611                                         iph->saddr, dh->dccph_sport,
612                                         iph->daddr, ntohs(dh->dccph_dport),
613                                         inet_iif(skb));
614         if (nsk != NULL) {
615                 if (nsk->sk_state != DCCP_TIME_WAIT) {
616                         bh_lock_sock(nsk);
617                         return nsk;
618                 }
619                 inet_twsk_put((struct inet_timewait_sock *)nsk);
620                 return NULL;
621         }
622
623         return sk;
624 }
625
626 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
627                      const __be32 daddr)
628 {
629         const struct dccp_hdr* dh = dccp_hdr(skb);
630         int checksum_len;
631         u32 tmp;
632
633         if (dh->dccph_cscov == 0)
634                 checksum_len = skb->len;
635         else {
636                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
637                 checksum_len = checksum_len < skb->len ? checksum_len :
638                                                          skb->len;
639         }
640
641         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
642         return csum_tcpudp_magic(saddr, daddr, checksum_len,
643                                  IPPROTO_DCCP, tmp);
644 }
645
646 EXPORT_SYMBOL_GPL(dccp_v4_checksum);
647
648 static int dccp_v4_verify_checksum(struct sk_buff *skb,
649                                    const __be32 saddr, const __be32 daddr)
650 {
651         struct dccp_hdr *dh = dccp_hdr(skb);
652         int checksum_len;
653         u32 tmp;
654
655         if (dh->dccph_cscov == 0)
656                 checksum_len = skb->len;
657         else {
658                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
659                 checksum_len = checksum_len < skb->len ? checksum_len :
660                                                          skb->len;
661         }
662         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
663         return csum_tcpudp_magic(saddr, daddr, checksum_len,
664                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
665 }
666
667 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
668                                            struct sk_buff *skb)
669 {
670         struct rtable *rt;
671         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
672                             .nl_u = { .ip4_u =
673                                       { .daddr = skb->nh.iph->saddr,
674                                         .saddr = skb->nh.iph->daddr,
675                                         .tos = RT_CONN_FLAGS(sk) } },
676                             .proto = sk->sk_protocol,
677                             .uli_u = { .ports =
678                                        { .sport = dccp_hdr(skb)->dccph_dport,
679                                          .dport = dccp_hdr(skb)->dccph_sport }
680                                      }
681                           };
682
683         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
684                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
685                 return NULL;
686         }
687
688         return &rt->u.dst;
689 }
690
691 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
692 {
693         int err;
694         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
695         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
696                                        sizeof(struct dccp_hdr_ext) +
697                                        sizeof(struct dccp_hdr_reset);
698         struct sk_buff *skb;
699         struct dst_entry *dst;
700         u64 seqno;
701
702         /* Never send a reset in response to a reset. */
703         if (rxdh->dccph_type == DCCP_PKT_RESET)
704                 return;
705
706         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
707                 return;
708
709         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
710         if (dst == NULL)
711                 return;
712
713         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header,
714                         GFP_ATOMIC);
715         if (skb == NULL)
716                 goto out;
717
718         /* Reserve space for headers. */
719         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
720         skb->dst = dst_clone(dst);
721
722         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
723         dh = dccp_hdr(skb);
724         memset(dh, 0, dccp_hdr_reset_len);
725
726         /* Build DCCP header and checksum it. */
727         dh->dccph_type     = DCCP_PKT_RESET;
728         dh->dccph_sport    = rxdh->dccph_dport;
729         dh->dccph_dport    = rxdh->dccph_sport;
730         dh->dccph_doff     = dccp_hdr_reset_len / 4;
731         dh->dccph_x        = 1;
732         dccp_hdr_reset(skb)->dccph_reset_code =
733                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
734
735         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
736         seqno = 0;
737         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
738                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
739
740         dccp_hdr_set_seq(dh, seqno);
741         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
742                          DCCP_SKB_CB(rxskb)->dccpd_seq);
743
744         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
745                                               rxskb->nh.iph->daddr);
746
747         bh_lock_sock(dccp_v4_ctl_socket->sk);
748         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
749                                     rxskb->nh.iph->daddr,
750                                     rxskb->nh.iph->saddr, NULL);
751         bh_unlock_sock(dccp_v4_ctl_socket->sk);
752
753         if (err == NET_XMIT_CN || err == 0) {
754                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
755                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
756         }
757 out:
758          dst_release(dst);
759 }
760
761 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
762 {
763         struct dccp_hdr *dh = dccp_hdr(skb);
764
765         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
766                 if (dccp_rcv_established(sk, skb, dh, skb->len))
767                         goto reset;
768                 return 0;
769         }
770
771         /*
772          *  Step 3: Process LISTEN state
773          *     If S.state == LISTEN,
774          *        If P.type == Request or P contains a valid Init Cookie
775          *              option,
776          *           * Must scan the packet's options to check for an Init
777          *              Cookie.  Only the Init Cookie is processed here,
778          *              however; other options are processed in Step 8.  This
779          *              scan need only be performed if the endpoint uses Init
780          *              Cookies *
781          *           * Generate a new socket and switch to that socket *
782          *           Set S := new socket for this port pair
783          *           S.state = RESPOND
784          *           Choose S.ISS (initial seqno) or set from Init Cookie
785          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
786          *           Continue with S.state == RESPOND
787          *           * A Response packet will be generated in Step 11 *
788          *        Otherwise,
789          *           Generate Reset(No Connection) unless P.type == Reset
790          *           Drop packet and return
791          *
792          * NOTE: the check for the packet types is done in
793          *       dccp_rcv_state_process
794          */
795         if (sk->sk_state == DCCP_LISTEN) {
796                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
797
798                 if (nsk == NULL)
799                         goto discard;
800
801                 if (nsk != sk) {
802                         if (dccp_child_process(sk, nsk, skb))
803                                 goto reset;
804                         return 0;
805                 }
806         }
807
808         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
809                 goto reset;
810         return 0;
811
812 reset:
813         dccp_v4_ctl_send_reset(skb);
814 discard:
815         kfree_skb(skb);
816         return 0;
817 }
818
819 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
820
821 int dccp_invalid_packet(struct sk_buff *skb)
822 {
823         const struct dccp_hdr *dh;
824
825         if (skb->pkt_type != PACKET_HOST)
826                 return 1;
827
828         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
829                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
830                 return 1;
831         }
832
833         dh = dccp_hdr(skb);
834
835         /* If the packet type is not understood, drop packet and return */
836         if (dh->dccph_type >= DCCP_PKT_INVALID) {
837                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
838                 return 1;
839         }
840
841         /*
842          * If P.Data Offset is too small for packet type, or too large for
843          * packet, drop packet and return
844          */
845         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
846                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
847                                             "too small 1\n",
848                                dh->dccph_doff);
849                 return 1;
850         }
851
852         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
853                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
854                                             "too small 2\n",
855                                dh->dccph_doff);
856                 return 1;
857         }
858
859         dh = dccp_hdr(skb);
860
861         /*
862          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
863          * has short sequence numbers), drop packet and return
864          */
865         if (dh->dccph_x == 0 &&
866             dh->dccph_type != DCCP_PKT_DATA &&
867             dh->dccph_type != DCCP_PKT_ACK &&
868             dh->dccph_type != DCCP_PKT_DATAACK) {
869                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
870                                             "nor DataAck and P.X == 0\n",
871                                dccp_packet_name(dh->dccph_type));
872                 return 1;
873         }
874
875         return 0;
876 }
877
878 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
879
880 /* this is called when real data arrives */
881 static int dccp_v4_rcv(struct sk_buff *skb)
882 {
883         const struct dccp_hdr *dh;
884         struct sock *sk;
885
886         /* Step 1: Check header basics: */
887
888         if (dccp_invalid_packet(skb))
889                 goto discard_it;
890
891         /* If the header checksum is incorrect, drop packet and return */
892         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
893                                     skb->nh.iph->daddr) < 0) {
894                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
895                                __FUNCTION__);
896                 goto discard_it;
897         }
898
899         dh = dccp_hdr(skb);
900
901         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
902         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
903
904         dccp_pr_debug("%8.8s "
905                       "src=%u.%u.%u.%u@%-5d "
906                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
907                       dccp_packet_name(dh->dccph_type),
908                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
909                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
910                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
911
912         if (dccp_packet_without_ack(skb)) {
913                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
914                 dccp_pr_debug_cat("\n");
915         } else {
916                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
917                 dccp_pr_debug_cat(", ack=%llu\n",
918                                   (unsigned long long)
919                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
920         }
921
922         /* Step 2:
923          *      Look up flow ID in table and get corresponding socket */
924         sk = __inet_lookup(&dccp_hashinfo,
925                            skb->nh.iph->saddr, dh->dccph_sport,
926                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
927                            inet_iif(skb));
928
929         /* 
930          * Step 2:
931          *      If no socket ...
932          *              Generate Reset(No Connection) unless P.type == Reset
933          *              Drop packet and return
934          */
935         if (sk == NULL) {
936                 dccp_pr_debug("failed to look up flow ID in table and "
937                               "get corresponding socket\n");
938                 goto no_dccp_socket;
939         }
940
941         /* 
942          * Step 2:
943          *      ... or S.state == TIMEWAIT,
944          *              Generate Reset(No Connection) unless P.type == Reset
945          *              Drop packet and return
946          */
947                
948         if (sk->sk_state == DCCP_TIME_WAIT) {
949                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
950                               "do_time_wait\n");
951                 goto do_time_wait;
952         }
953
954         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
955                 goto discard_and_relse;
956         nf_reset(skb);
957
958         return sk_receive_skb(sk, skb);
959
960 no_dccp_socket:
961         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
962                 goto discard_it;
963         /*
964          * Step 2:
965          *              Generate Reset(No Connection) unless P.type == Reset
966          *              Drop packet and return
967          */
968         if (dh->dccph_type != DCCP_PKT_RESET) {
969                 DCCP_SKB_CB(skb)->dccpd_reset_code =
970                                         DCCP_RESET_CODE_NO_CONNECTION;
971                 dccp_v4_ctl_send_reset(skb);
972         }
973
974 discard_it:
975         /* Discard frame. */
976         kfree_skb(skb);
977         return 0;
978
979 discard_and_relse:
980         sock_put(sk);
981         goto discard_it;
982
983 do_time_wait:
984         inet_twsk_put((struct inet_timewait_sock *)sk);
985         goto no_dccp_socket;
986 }
987
988 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
989         .queue_xmit     = ip_queue_xmit,
990         .send_check     = dccp_v4_send_check,
991         .rebuild_header = inet_sk_rebuild_header,
992         .conn_request   = dccp_v4_conn_request,
993         .syn_recv_sock  = dccp_v4_request_recv_sock,
994         .net_header_len = sizeof(struct iphdr),
995         .setsockopt     = ip_setsockopt,
996         .getsockopt     = ip_getsockopt,
997 #ifdef CONFIG_COMPAT
998         .compat_setsockopt      = compat_ip_setsockopt,
999         .compat_getsockopt      = compat_ip_getsockopt,
1000 #endif
1001         .addr2sockaddr  = inet_csk_addr2sockaddr,
1002         .sockaddr_len   = sizeof(struct sockaddr_in),
1003 };
1004
1005 static int dccp_v4_init_sock(struct sock *sk)
1006 {
1007         static __u8 dccp_v4_ctl_sock_initialized;
1008         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
1009
1010         if (err == 0) {
1011                 if (unlikely(!dccp_v4_ctl_sock_initialized))
1012                         dccp_v4_ctl_sock_initialized = 1;
1013                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1014         }
1015
1016         return err;
1017 }
1018
1019 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1020 {
1021         kfree(inet_rsk(req)->opt);
1022 }
1023
1024 static struct request_sock_ops dccp_request_sock_ops = {
1025         .family         = PF_INET,
1026         .obj_size       = sizeof(struct dccp_request_sock),
1027         .rtx_syn_ack    = dccp_v4_send_response,
1028         .send_ack       = dccp_v4_reqsk_send_ack,
1029         .destructor     = dccp_v4_reqsk_destructor,
1030         .send_reset     = dccp_v4_ctl_send_reset,
1031 };
1032
1033 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1034         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1035 };
1036
1037 static struct proto dccp_v4_prot = {
1038         .name                   = "DCCP",
1039         .owner                  = THIS_MODULE,
1040         .close                  = dccp_close,
1041         .connect                = dccp_v4_connect,
1042         .disconnect             = dccp_disconnect,
1043         .ioctl                  = dccp_ioctl,
1044         .init                   = dccp_v4_init_sock,
1045         .setsockopt             = dccp_setsockopt,
1046         .getsockopt             = dccp_getsockopt,
1047 #ifdef CONFIG_COMPAT
1048         .compat_setsockopt      = compat_dccp_setsockopt,
1049         .compat_getsockopt      = compat_dccp_getsockopt,
1050 #endif
1051         .sendmsg                = dccp_sendmsg,
1052         .recvmsg                = dccp_recvmsg,
1053         .backlog_rcv            = dccp_v4_do_rcv,
1054         .hash                   = dccp_hash,
1055         .unhash                 = dccp_unhash,
1056         .accept                 = inet_csk_accept,
1057         .get_port               = dccp_v4_get_port,
1058         .shutdown               = dccp_shutdown,
1059         .destroy                = dccp_destroy_sock,
1060         .orphan_count           = &dccp_orphan_count,
1061         .max_header             = MAX_DCCP_HEADER,
1062         .obj_size               = sizeof(struct dccp_sock),
1063         .rsk_prot               = &dccp_request_sock_ops,
1064         .twsk_prot              = &dccp_timewait_sock_ops,
1065 };
1066
1067 static struct net_protocol dccp_v4_protocol = {
1068         .handler        = dccp_v4_rcv,
1069         .err_handler    = dccp_v4_err,
1070         .no_policy      = 1,
1071 };
1072
1073 static const struct proto_ops inet_dccp_ops = {
1074         .family         = PF_INET,
1075         .owner          = THIS_MODULE,
1076         .release        = inet_release,
1077         .bind           = inet_bind,
1078         .connect        = inet_stream_connect,
1079         .socketpair     = sock_no_socketpair,
1080         .accept         = inet_accept,
1081         .getname        = inet_getname,
1082         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1083         .poll           = dccp_poll,
1084         .ioctl          = inet_ioctl,
1085         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1086         .listen         = inet_dccp_listen,
1087         .shutdown       = inet_shutdown,
1088         .setsockopt     = sock_common_setsockopt,
1089         .getsockopt     = sock_common_getsockopt,
1090 #ifdef CONFIG_COMPAT
1091         .compat_setsockopt      = compat_sock_common_setsockopt,
1092         .compat_getsockopt      = compat_sock_common_getsockopt,
1093 #endif
1094         .sendmsg        = inet_sendmsg,
1095         .recvmsg        = sock_common_recvmsg,
1096         .mmap           = sock_no_mmap,
1097         .sendpage       = sock_no_sendpage,
1098 };
1099
1100 static struct inet_protosw dccp_v4_protosw = {
1101         .type           = SOCK_DCCP,
1102         .protocol       = IPPROTO_DCCP,
1103         .prot           = &dccp_v4_prot,
1104         .ops            = &inet_dccp_ops,
1105         .capability     = -1,
1106         .no_check       = 0,
1107         .flags          = INET_PROTOSW_ICSK,
1108 };
1109
1110 static int __init dccp_v4_init(void)
1111 {
1112         int err = proto_register(&dccp_v4_prot, 1);
1113
1114         if (err != 0)
1115                 goto out;
1116
1117         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1118         if (err != 0)
1119                 goto out_proto_unregister;
1120
1121         inet_register_protosw(&dccp_v4_protosw);
1122
1123         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1124                                        SOCK_DCCP, IPPROTO_DCCP);
1125         if (err)
1126                 goto out_unregister_protosw;
1127 out:
1128         return err;
1129 out_unregister_protosw:
1130         inet_unregister_protosw(&dccp_v4_protosw);
1131         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1132 out_proto_unregister:
1133         proto_unregister(&dccp_v4_prot);
1134         goto out;
1135 }
1136
1137 static void __exit dccp_v4_exit(void)
1138 {
1139         inet_unregister_protosw(&dccp_v4_protosw);
1140         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1141         proto_unregister(&dccp_v4_prot);
1142 }
1143
1144 module_init(dccp_v4_init);
1145 module_exit(dccp_v4_exit);
1146
1147 /*
1148  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1149  * values directly, Also cover the case where the protocol is not specified,
1150  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1151  */
1152 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1153 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1154 MODULE_LICENSE("GPL");
1155 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1156 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");