[NET]: Size listen hash tables using backlog hint
[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         __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 RFC 4340, sec. 14.1:
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(inet_twsk(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 static struct request_sock_ops dccp_request_sock_ops;
453
454 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
455 {
456         struct inet_request_sock *ireq;
457         struct dccp_sock dp;
458         struct request_sock *req;
459         struct dccp_request_sock *dreq;
460         const __be32 saddr = skb->nh.iph->saddr;
461         const __be32 daddr = skb->nh.iph->daddr;
462         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
463         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
464         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
465
466         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
467         if (((struct rtable *)skb->dst)->rt_flags &
468             (RTCF_BROADCAST | RTCF_MULTICAST)) {
469                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
470                 goto drop;
471         }
472
473         if (dccp_bad_service_code(sk, service)) {
474                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
475                 goto drop;
476         }
477         /*
478          * TW buckets are converted to open requests without
479          * limitations, they conserve resources and peer is
480          * evidently real one.
481          */
482         if (inet_csk_reqsk_queue_is_full(sk))
483                 goto drop;
484
485         /*
486          * Accept backlog is full. If we have already queued enough
487          * of warm entries in syn queue, drop request. It is better than
488          * clogging syn queue with openreqs with exponentially increasing
489          * timeout.
490          */
491         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
492                 goto drop;
493
494         req = reqsk_alloc(&dccp_request_sock_ops);
495         if (req == NULL)
496                 goto drop;
497
498         if (dccp_parse_options(sk, skb))
499                 goto drop_and_free;
500
501         dccp_openreq_init(req, &dp, skb);
502
503         if (security_inet_conn_request(sk, skb, req))
504                 goto drop_and_free;
505
506         ireq = inet_rsk(req);
507         ireq->loc_addr = daddr;
508         ireq->rmt_addr = saddr;
509         req->rcv_wnd    = dccp_feat_default_sequence_window;
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, 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(inet_twsk(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         security_skb_classify_flow(skb, &fl);
684         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
685                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
686                 return NULL;
687         }
688
689         return &rt->u.dst;
690 }
691
692 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
693 {
694         int err;
695         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
696         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
697                                        sizeof(struct dccp_hdr_ext) +
698                                        sizeof(struct dccp_hdr_reset);
699         struct sk_buff *skb;
700         struct dst_entry *dst;
701         u64 seqno;
702
703         /* Never send a reset in response to a reset. */
704         if (rxdh->dccph_type == DCCP_PKT_RESET)
705                 return;
706
707         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
708                 return;
709
710         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
711         if (dst == NULL)
712                 return;
713
714         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header,
715                         GFP_ATOMIC);
716         if (skb == NULL)
717                 goto out;
718
719         /* Reserve space for headers. */
720         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
721         skb->dst = dst_clone(dst);
722
723         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
724         dh = dccp_hdr(skb);
725         memset(dh, 0, dccp_hdr_reset_len);
726
727         /* Build DCCP header and checksum it. */
728         dh->dccph_type     = DCCP_PKT_RESET;
729         dh->dccph_sport    = rxdh->dccph_dport;
730         dh->dccph_dport    = rxdh->dccph_sport;
731         dh->dccph_doff     = dccp_hdr_reset_len / 4;
732         dh->dccph_x        = 1;
733         dccp_hdr_reset(skb)->dccph_reset_code =
734                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
735
736         /* See "8.3.1. Abnormal Termination" in RFC 4340 */
737         seqno = 0;
738         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
739                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
740
741         dccp_hdr_set_seq(dh, seqno);
742         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
743                          DCCP_SKB_CB(rxskb)->dccpd_seq);
744
745         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
746                                               rxskb->nh.iph->daddr);
747
748         bh_lock_sock(dccp_v4_ctl_socket->sk);
749         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
750                                     rxskb->nh.iph->daddr,
751                                     rxskb->nh.iph->saddr, NULL);
752         bh_unlock_sock(dccp_v4_ctl_socket->sk);
753
754         if (err == NET_XMIT_CN || err == 0) {
755                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
756                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
757         }
758 out:
759          dst_release(dst);
760 }
761
762 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
763 {
764         struct dccp_hdr *dh = dccp_hdr(skb);
765
766         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
767                 if (dccp_rcv_established(sk, skb, dh, skb->len))
768                         goto reset;
769                 return 0;
770         }
771
772         /*
773          *  Step 3: Process LISTEN state
774          *     If S.state == LISTEN,
775          *        If P.type == Request or P contains a valid Init Cookie
776          *              option,
777          *           * Must scan the packet's options to check for an Init
778          *              Cookie.  Only the Init Cookie is processed here,
779          *              however; other options are processed in Step 8.  This
780          *              scan need only be performed if the endpoint uses Init
781          *              Cookies *
782          *           * Generate a new socket and switch to that socket *
783          *           Set S := new socket for this port pair
784          *           S.state = RESPOND
785          *           Choose S.ISS (initial seqno) or set from Init Cookie
786          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
787          *           Continue with S.state == RESPOND
788          *           * A Response packet will be generated in Step 11 *
789          *        Otherwise,
790          *           Generate Reset(No Connection) unless P.type == Reset
791          *           Drop packet and return
792          *
793          * NOTE: the check for the packet types is done in
794          *       dccp_rcv_state_process
795          */
796         if (sk->sk_state == DCCP_LISTEN) {
797                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
798
799                 if (nsk == NULL)
800                         goto discard;
801
802                 if (nsk != sk) {
803                         if (dccp_child_process(sk, nsk, skb))
804                                 goto reset;
805                         return 0;
806                 }
807         }
808
809         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
810                 goto reset;
811         return 0;
812
813 reset:
814         dccp_v4_ctl_send_reset(skb);
815 discard:
816         kfree_skb(skb);
817         return 0;
818 }
819
820 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
821
822 int dccp_invalid_packet(struct sk_buff *skb)
823 {
824         const struct dccp_hdr *dh;
825
826         if (skb->pkt_type != PACKET_HOST)
827                 return 1;
828
829         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
830                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
831                 return 1;
832         }
833
834         dh = dccp_hdr(skb);
835
836         /* If the packet type is not understood, drop packet and return */
837         if (dh->dccph_type >= DCCP_PKT_INVALID) {
838                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
839                 return 1;
840         }
841
842         /*
843          * If P.Data Offset is too small for packet type, or too large for
844          * packet, drop packet and return
845          */
846         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
847                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
848                                             "too small 1\n",
849                                dh->dccph_doff);
850                 return 1;
851         }
852
853         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
854                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
855                                             "too small 2\n",
856                                dh->dccph_doff);
857                 return 1;
858         }
859
860         dh = dccp_hdr(skb);
861
862         /*
863          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
864          * has short sequence numbers), drop packet and return
865          */
866         if (dh->dccph_x == 0 &&
867             dh->dccph_type != DCCP_PKT_DATA &&
868             dh->dccph_type != DCCP_PKT_ACK &&
869             dh->dccph_type != DCCP_PKT_DATAACK) {
870                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
871                                             "nor DataAck and P.X == 0\n",
872                                dccp_packet_name(dh->dccph_type));
873                 return 1;
874         }
875
876         return 0;
877 }
878
879 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
880
881 /* this is called when real data arrives */
882 static int dccp_v4_rcv(struct sk_buff *skb)
883 {
884         const struct dccp_hdr *dh;
885         struct sock *sk;
886
887         /* Step 1: Check header basics: */
888
889         if (dccp_invalid_packet(skb))
890                 goto discard_it;
891
892         /* If the header checksum is incorrect, drop packet and return */
893         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
894                                     skb->nh.iph->daddr) < 0) {
895                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
896                                __FUNCTION__);
897                 goto discard_it;
898         }
899
900         dh = dccp_hdr(skb);
901
902         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
903         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
904
905         dccp_pr_debug("%8.8s "
906                       "src=%u.%u.%u.%u@%-5d "
907                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
908                       dccp_packet_name(dh->dccph_type),
909                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
910                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
911                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
912
913         if (dccp_packet_without_ack(skb)) {
914                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
915                 dccp_pr_debug_cat("\n");
916         } else {
917                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
918                 dccp_pr_debug_cat(", ack=%llu\n",
919                                   (unsigned long long)
920                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
921         }
922
923         /* Step 2:
924          *      Look up flow ID in table and get corresponding socket */
925         sk = __inet_lookup(&dccp_hashinfo,
926                            skb->nh.iph->saddr, dh->dccph_sport,
927                            skb->nh.iph->daddr, dh->dccph_dport,
928                            inet_iif(skb));
929
930         /* 
931          * Step 2:
932          *      If no socket ...
933          *              Generate Reset(No Connection) unless P.type == Reset
934          *              Drop packet and return
935          */
936         if (sk == NULL) {
937                 dccp_pr_debug("failed to look up flow ID in table and "
938                               "get corresponding socket\n");
939                 goto no_dccp_socket;
940         }
941
942         /* 
943          * Step 2:
944          *      ... or S.state == TIMEWAIT,
945          *              Generate Reset(No Connection) unless P.type == Reset
946          *              Drop packet and return
947          */
948                
949         if (sk->sk_state == DCCP_TIME_WAIT) {
950                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
951                               "do_time_wait\n");
952                 goto do_time_wait;
953         }
954
955         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
956                 goto discard_and_relse;
957         nf_reset(skb);
958
959         return sk_receive_skb(sk, skb);
960
961 no_dccp_socket:
962         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
963                 goto discard_it;
964         /*
965          * Step 2:
966          *              Generate Reset(No Connection) unless P.type == Reset
967          *              Drop packet and return
968          */
969         if (dh->dccph_type != DCCP_PKT_RESET) {
970                 DCCP_SKB_CB(skb)->dccpd_reset_code =
971                                         DCCP_RESET_CODE_NO_CONNECTION;
972                 dccp_v4_ctl_send_reset(skb);
973         }
974
975 discard_it:
976         /* Discard frame. */
977         kfree_skb(skb);
978         return 0;
979
980 discard_and_relse:
981         sock_put(sk);
982         goto discard_it;
983
984 do_time_wait:
985         inet_twsk_put(inet_twsk(sk));
986         goto no_dccp_socket;
987 }
988
989 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
990         .queue_xmit        = ip_queue_xmit,
991         .send_check        = dccp_v4_send_check,
992         .rebuild_header    = inet_sk_rebuild_header,
993         .conn_request      = dccp_v4_conn_request,
994         .syn_recv_sock     = dccp_v4_request_recv_sock,
995         .net_header_len    = sizeof(struct iphdr),
996         .setsockopt        = ip_setsockopt,
997         .getsockopt        = ip_getsockopt,
998         .addr2sockaddr     = inet_csk_addr2sockaddr,
999         .sockaddr_len      = sizeof(struct sockaddr_in),
1000 #ifdef CONFIG_COMPAT
1001         .compat_setsockopt = compat_ip_setsockopt,
1002         .compat_getsockopt = compat_ip_getsockopt,
1003 #endif
1004 };
1005
1006 static int dccp_v4_init_sock(struct sock *sk)
1007 {
1008         static __u8 dccp_v4_ctl_sock_initialized;
1009         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
1010
1011         if (err == 0) {
1012                 if (unlikely(!dccp_v4_ctl_sock_initialized))
1013                         dccp_v4_ctl_sock_initialized = 1;
1014                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1015         }
1016
1017         return err;
1018 }
1019
1020 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1021 {
1022         kfree(inet_rsk(req)->opt);
1023 }
1024
1025 static struct request_sock_ops dccp_request_sock_ops _read_mostly = {
1026         .family         = PF_INET,
1027         .obj_size       = sizeof(struct dccp_request_sock),
1028         .rtx_syn_ack    = dccp_v4_send_response,
1029         .send_ack       = dccp_v4_reqsk_send_ack,
1030         .destructor     = dccp_v4_reqsk_destructor,
1031         .send_reset     = dccp_v4_ctl_send_reset,
1032 };
1033
1034 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1035         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1036 };
1037
1038 static struct proto dccp_v4_prot = {
1039         .name                   = "DCCP",
1040         .owner                  = THIS_MODULE,
1041         .close                  = dccp_close,
1042         .connect                = dccp_v4_connect,
1043         .disconnect             = dccp_disconnect,
1044         .ioctl                  = dccp_ioctl,
1045         .init                   = dccp_v4_init_sock,
1046         .setsockopt             = dccp_setsockopt,
1047         .getsockopt             = dccp_getsockopt,
1048         .sendmsg                = dccp_sendmsg,
1049         .recvmsg                = dccp_recvmsg,
1050         .backlog_rcv            = dccp_v4_do_rcv,
1051         .hash                   = dccp_hash,
1052         .unhash                 = dccp_unhash,
1053         .accept                 = inet_csk_accept,
1054         .get_port               = dccp_v4_get_port,
1055         .shutdown               = dccp_shutdown,
1056         .destroy                = dccp_destroy_sock,
1057         .orphan_count           = &dccp_orphan_count,
1058         .max_header             = MAX_DCCP_HEADER,
1059         .obj_size               = sizeof(struct dccp_sock),
1060         .rsk_prot               = &dccp_request_sock_ops,
1061         .twsk_prot              = &dccp_timewait_sock_ops,
1062 #ifdef CONFIG_COMPAT
1063         .compat_setsockopt      = compat_dccp_setsockopt,
1064         .compat_getsockopt      = compat_dccp_getsockopt,
1065 #endif
1066 };
1067
1068 static struct net_protocol dccp_v4_protocol = {
1069         .handler        = dccp_v4_rcv,
1070         .err_handler    = dccp_v4_err,
1071         .no_policy      = 1,
1072 };
1073
1074 static const struct proto_ops inet_dccp_ops = {
1075         .family            = PF_INET,
1076         .owner             = THIS_MODULE,
1077         .release           = inet_release,
1078         .bind              = inet_bind,
1079         .connect           = inet_stream_connect,
1080         .socketpair        = sock_no_socketpair,
1081         .accept            = inet_accept,
1082         .getname           = inet_getname,
1083         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1084         .poll              = dccp_poll,
1085         .ioctl             = inet_ioctl,
1086         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1087         .listen            = inet_dccp_listen,
1088         .shutdown          = inet_shutdown,
1089         .setsockopt        = sock_common_setsockopt,
1090         .getsockopt        = sock_common_getsockopt,
1091         .sendmsg           = inet_sendmsg,
1092         .recvmsg           = sock_common_recvmsg,
1093         .mmap              = sock_no_mmap,
1094         .sendpage          = sock_no_sendpage,
1095 #ifdef CONFIG_COMPAT
1096         .compat_setsockopt = compat_sock_common_setsockopt,
1097         .compat_getsockopt = compat_sock_common_getsockopt,
1098 #endif
1099 };
1100
1101 static struct inet_protosw dccp_v4_protosw = {
1102         .type           = SOCK_DCCP,
1103         .protocol       = IPPROTO_DCCP,
1104         .prot           = &dccp_v4_prot,
1105         .ops            = &inet_dccp_ops,
1106         .capability     = -1,
1107         .no_check       = 0,
1108         .flags          = INET_PROTOSW_ICSK,
1109 };
1110
1111 static int __init dccp_v4_init(void)
1112 {
1113         int err = proto_register(&dccp_v4_prot, 1);
1114
1115         if (err != 0)
1116                 goto out;
1117
1118         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1119         if (err != 0)
1120                 goto out_proto_unregister;
1121
1122         inet_register_protosw(&dccp_v4_protosw);
1123
1124         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1125                                        SOCK_DCCP, IPPROTO_DCCP);
1126         if (err)
1127                 goto out_unregister_protosw;
1128 out:
1129         return err;
1130 out_unregister_protosw:
1131         inet_unregister_protosw(&dccp_v4_protosw);
1132         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1133 out_proto_unregister:
1134         proto_unregister(&dccp_v4_prot);
1135         goto out;
1136 }
1137
1138 static void __exit dccp_v4_exit(void)
1139 {
1140         inet_unregister_protosw(&dccp_v4_protosw);
1141         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1142         proto_unregister(&dccp_v4_prot);
1143 }
1144
1145 module_init(dccp_v4_init);
1146 module_exit(dccp_v4_exit);
1147
1148 /*
1149  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1150  * values directly, Also cover the case where the protocol is not specified,
1151  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1152  */
1153 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1154 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1155 MODULE_LICENSE("GPL");
1156 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1157 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");