[DCCP]: Combine allocating & zeroing header space on skb
[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         skb->dst = dst_clone(rxskb->dst);
216
217         dh = dccp_zeroed_hdr(skb, dccp_hdr_ack_len);
218
219         /* Build DCCP header and checksum it. */
220         dh->dccph_type     = DCCP_PKT_ACK;
221         dh->dccph_sport    = rxdh->dccph_dport;
222         dh->dccph_dport    = rxdh->dccph_sport;
223         dh->dccph_doff     = dccp_hdr_ack_len / 4;
224         dh->dccph_x        = 1;
225
226         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
227         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
228                          DCCP_SKB_CB(rxskb)->dccpd_seq);
229
230         bh_lock_sock(dccp_v4_ctl_socket->sk);
231         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
232                                     rxskb->nh.iph->daddr,
233                                     rxskb->nh.iph->saddr, NULL);
234         bh_unlock_sock(dccp_v4_ctl_socket->sk);
235
236         if (err == NET_XMIT_CN || err == 0) {
237                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
238                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
239         }
240 }
241
242 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
243                                  struct dst_entry *dst)
244 {
245         int err = -1;
246         struct sk_buff *skb;
247
248         /* First, grab a route. */
249         
250         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
251                 goto out;
252
253         skb = dccp_make_response(sk, dst, req);
254         if (skb != NULL) {
255                 const struct inet_request_sock *ireq = inet_rsk(req);
256                 struct dccp_hdr *dh = dccp_hdr(skb);
257
258                 dh->dccph_checksum = dccp_v4_checksum(skb, ireq->loc_addr,
259                                                       ireq->rmt_addr);
260                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
261                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
262                                             ireq->rmt_addr,
263                                             ireq->opt);
264                 if (err == NET_XMIT_CN)
265                         err = 0;
266         }
267
268 out:
269         dst_release(dst);
270         return err;
271 }
272
273 /*
274  * This routine is called by the ICMP module when it gets some sort of error
275  * condition. If err < 0 then the socket should be closed and the error
276  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
277  * After adjustment header points to the first 8 bytes of the tcp header. We
278  * need to find the appropriate port.
279  *
280  * The locking strategy used here is very "optimistic". When someone else
281  * accesses the socket the ICMP is just dropped and for some paths there is no
282  * check at all. A more general error queue to queue errors for later handling
283  * is probably better.
284  */
285 static void dccp_v4_err(struct sk_buff *skb, u32 info)
286 {
287         const struct iphdr *iph = (struct iphdr *)skb->data;
288         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
289                                                         (iph->ihl << 2));
290         struct dccp_sock *dp;
291         struct inet_sock *inet;
292         const int type = skb->h.icmph->type;
293         const int code = skb->h.icmph->code;
294         struct sock *sk;
295         __u64 seq;
296         int err;
297
298         if (skb->len < (iph->ihl << 2) + 8) {
299                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
300                 return;
301         }
302
303         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
304                          iph->saddr, dh->dccph_sport, inet_iif(skb));
305         if (sk == NULL) {
306                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
307                 return;
308         }
309
310         if (sk->sk_state == DCCP_TIME_WAIT) {
311                 inet_twsk_put(inet_twsk(sk));
312                 return;
313         }
314
315         bh_lock_sock(sk);
316         /* If too many ICMPs get dropped on busy
317          * servers this needs to be solved differently.
318          */
319         if (sock_owned_by_user(sk))
320                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
321
322         if (sk->sk_state == DCCP_CLOSED)
323                 goto out;
324
325         dp = dccp_sk(sk);
326         seq = dccp_hdr_seq(skb);
327         if (sk->sk_state != DCCP_LISTEN &&
328             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
329                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
330                 goto out;
331         }
332
333         switch (type) {
334         case ICMP_SOURCE_QUENCH:
335                 /* Just silently ignore these. */
336                 goto out;
337         case ICMP_PARAMETERPROB:
338                 err = EPROTO;
339                 break;
340         case ICMP_DEST_UNREACH:
341                 if (code > NR_ICMP_UNREACH)
342                         goto out;
343
344                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
345                         if (!sock_owned_by_user(sk))
346                                 dccp_do_pmtu_discovery(sk, iph, info);
347                         goto out;
348                 }
349
350                 err = icmp_err_convert[code].errno;
351                 break;
352         case ICMP_TIME_EXCEEDED:
353                 err = EHOSTUNREACH;
354                 break;
355         default:
356                 goto out;
357         }
358
359         switch (sk->sk_state) {
360                 struct request_sock *req , **prev;
361         case DCCP_LISTEN:
362                 if (sock_owned_by_user(sk))
363                         goto out;
364                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
365                                           iph->daddr, iph->saddr);
366                 if (!req)
367                         goto out;
368
369                 /*
370                  * ICMPs are not backlogged, hence we cannot get an established
371                  * socket here.
372                  */
373                 BUG_TRAP(!req->sk);
374
375                 if (seq != dccp_rsk(req)->dreq_iss) {
376                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
377                         goto out;
378                 }
379                 /*
380                  * Still in RESPOND, just remove it silently.
381                  * There is no good way to pass the error to the newly
382                  * created socket, and POSIX does not want network
383                  * errors returned from accept().
384                  */
385                 inet_csk_reqsk_queue_drop(sk, req, prev);
386                 goto out;
387
388         case DCCP_REQUESTING:
389         case DCCP_RESPOND:
390                 if (!sock_owned_by_user(sk)) {
391                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
392                         sk->sk_err = err;
393
394                         sk->sk_error_report(sk);
395
396                         dccp_done(sk);
397                 } else
398                         sk->sk_err_soft = err;
399                 goto out;
400         }
401
402         /* If we've already connected we will keep trying
403          * until we time out, or the user gives up.
404          *
405          * rfc1122 4.2.3.9 allows to consider as hard errors
406          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
407          * but it is obsoleted by pmtu discovery).
408          *
409          * Note, that in modern internet, where routing is unreliable
410          * and in each dark corner broken firewalls sit, sending random
411          * errors ordered by their masters even this two messages finally lose
412          * their original sense (even Linux sends invalid PORT_UNREACHs)
413          *
414          * Now we are in compliance with RFCs.
415          *                                                      --ANK (980905)
416          */
417
418         inet = inet_sk(sk);
419         if (!sock_owned_by_user(sk) && inet->recverr) {
420                 sk->sk_err = err;
421                 sk->sk_error_report(sk);
422         } else /* Only an error on timeout */
423                 sk->sk_err_soft = err;
424 out:
425         bh_unlock_sock(sk);
426         sock_put(sk);
427 }
428
429 /* This routine computes an IPv4 DCCP checksum. */
430 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
431 {
432         const struct inet_sock *inet = inet_sk(sk);
433         struct dccp_hdr *dh = dccp_hdr(skb);
434
435         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
436 }
437
438 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
439
440 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
441                                         const struct sk_buff *skb)
442 {
443         return secure_dccp_sequence_number(skb->nh.iph->daddr,
444                                            skb->nh.iph->saddr,
445                                            dccp_hdr(skb)->dccph_dport,
446                                            dccp_hdr(skb)->dccph_sport);
447 }
448
449 static struct request_sock_ops dccp_request_sock_ops;
450
451 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
452 {
453         struct inet_request_sock *ireq;
454         struct dccp_sock dp;
455         struct request_sock *req;
456         struct dccp_request_sock *dreq;
457         const __be32 saddr = skb->nh.iph->saddr;
458         const __be32 daddr = skb->nh.iph->daddr;
459         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
460         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
461         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
462
463         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
464         if (((struct rtable *)skb->dst)->rt_flags &
465             (RTCF_BROADCAST | RTCF_MULTICAST)) {
466                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
467                 goto drop;
468         }
469
470         if (dccp_bad_service_code(sk, service)) {
471                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
472                 goto drop;
473         }
474         /*
475          * TW buckets are converted to open requests without
476          * limitations, they conserve resources and peer is
477          * evidently real one.
478          */
479         if (inet_csk_reqsk_queue_is_full(sk))
480                 goto drop;
481
482         /*
483          * Accept backlog is full. If we have already queued enough
484          * of warm entries in syn queue, drop request. It is better than
485          * clogging syn queue with openreqs with exponentially increasing
486          * timeout.
487          */
488         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
489                 goto drop;
490
491         req = reqsk_alloc(&dccp_request_sock_ops);
492         if (req == NULL)
493                 goto drop;
494
495         if (dccp_parse_options(sk, skb))
496                 goto drop_and_free;
497
498         dccp_openreq_init(req, &dp, skb);
499
500         if (security_inet_conn_request(sk, skb, req))
501                 goto drop_and_free;
502
503         ireq = inet_rsk(req);
504         ireq->loc_addr = daddr;
505         ireq->rmt_addr = saddr;
506         req->rcv_wnd    = dccp_feat_default_sequence_window;
507         ireq->opt       = NULL;
508
509         /* 
510          * Step 3: Process LISTEN state
511          *
512          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
513          *
514          * In fact we defer setting S.GSR, S.SWL, S.SWH to
515          * dccp_create_openreq_child.
516          */
517         dreq = dccp_rsk(req);
518         dreq->dreq_isr     = dcb->dccpd_seq;
519         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
520         dreq->dreq_service = service;
521
522         if (dccp_v4_send_response(sk, req, NULL))
523                 goto drop_and_free;
524
525         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
526         return 0;
527
528 drop_and_free:
529         reqsk_free(req);
530 drop:
531         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
532         dcb->dccpd_reset_code = reset_code;
533         return -1;
534 }
535
536 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
537
538 /*
539  * The three way handshake has completed - we got a valid ACK or DATAACK -
540  * now create the new socket.
541  *
542  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
543  */
544 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
545                                        struct request_sock *req,
546                                        struct dst_entry *dst)
547 {
548         struct inet_request_sock *ireq;
549         struct inet_sock *newinet;
550         struct dccp_sock *newdp;
551         struct sock *newsk;
552
553         if (sk_acceptq_is_full(sk))
554                 goto exit_overflow;
555
556         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
557                 goto exit;
558
559         newsk = dccp_create_openreq_child(sk, req, skb);
560         if (newsk == NULL)
561                 goto exit;
562
563         sk_setup_caps(newsk, dst);
564
565         newdp              = dccp_sk(newsk);
566         newinet            = inet_sk(newsk);
567         ireq               = inet_rsk(req);
568         newinet->daddr     = ireq->rmt_addr;
569         newinet->rcv_saddr = ireq->loc_addr;
570         newinet->saddr     = ireq->loc_addr;
571         newinet->opt       = ireq->opt;
572         ireq->opt          = NULL;
573         newinet->mc_index  = inet_iif(skb);
574         newinet->mc_ttl    = skb->nh.iph->ttl;
575         newinet->id        = jiffies;
576
577         dccp_sync_mss(newsk, dst_mtu(dst));
578
579         __inet_hash(&dccp_hashinfo, newsk, 0);
580         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
581
582         return newsk;
583
584 exit_overflow:
585         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
586 exit:
587         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
588         dst_release(dst);
589         return NULL;
590 }
591
592 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
593
594 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
595 {
596         const struct dccp_hdr *dh = dccp_hdr(skb);
597         const struct iphdr *iph = skb->nh.iph;
598         struct sock *nsk;
599         struct request_sock **prev;
600         /* Find possible connection requests. */
601         struct request_sock *req = inet_csk_search_req(sk, &prev,
602                                                        dh->dccph_sport,
603                                                        iph->saddr, iph->daddr);
604         if (req != NULL)
605                 return dccp_check_req(sk, skb, req, prev);
606
607         nsk = inet_lookup_established(&dccp_hashinfo,
608                                       iph->saddr, dh->dccph_sport,
609                                       iph->daddr, dh->dccph_dport,
610                                       inet_iif(skb));
611         if (nsk != NULL) {
612                 if (nsk->sk_state != DCCP_TIME_WAIT) {
613                         bh_lock_sock(nsk);
614                         return nsk;
615                 }
616                 inet_twsk_put(inet_twsk(nsk));
617                 return NULL;
618         }
619
620         return sk;
621 }
622
623 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
624                      const __be32 daddr)
625 {
626         const struct dccp_hdr* dh = dccp_hdr(skb);
627         int checksum_len;
628         u32 tmp;
629
630         if (dh->dccph_cscov == 0)
631                 checksum_len = skb->len;
632         else {
633                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
634                 checksum_len = checksum_len < skb->len ? checksum_len :
635                                                          skb->len;
636         }
637
638         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
639         return csum_tcpudp_magic(saddr, daddr, checksum_len,
640                                  IPPROTO_DCCP, tmp);
641 }
642
643 EXPORT_SYMBOL_GPL(dccp_v4_checksum);
644
645 static int dccp_v4_verify_checksum(struct sk_buff *skb,
646                                    const __be32 saddr, const __be32 daddr)
647 {
648         struct dccp_hdr *dh = dccp_hdr(skb);
649         int checksum_len;
650         u32 tmp;
651
652         if (dh->dccph_cscov == 0)
653                 checksum_len = skb->len;
654         else {
655                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
656                 checksum_len = checksum_len < skb->len ? checksum_len :
657                                                          skb->len;
658         }
659         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
660         return csum_tcpudp_magic(saddr, daddr, checksum_len,
661                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
662 }
663
664 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
665                                            struct sk_buff *skb)
666 {
667         struct rtable *rt;
668         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
669                             .nl_u = { .ip4_u =
670                                       { .daddr = skb->nh.iph->saddr,
671                                         .saddr = skb->nh.iph->daddr,
672                                         .tos = RT_CONN_FLAGS(sk) } },
673                             .proto = sk->sk_protocol,
674                             .uli_u = { .ports =
675                                        { .sport = dccp_hdr(skb)->dccph_dport,
676                                          .dport = dccp_hdr(skb)->dccph_sport }
677                                      }
678                           };
679
680         security_skb_classify_flow(skb, &fl);
681         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
682                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
683                 return NULL;
684         }
685
686         return &rt->u.dst;
687 }
688
689 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
690 {
691         int err;
692         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
693         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
694                                        sizeof(struct dccp_hdr_ext) +
695                                        sizeof(struct dccp_hdr_reset);
696         struct sk_buff *skb;
697         struct dst_entry *dst;
698         u64 seqno;
699
700         /* Never send a reset in response to a reset. */
701         if (rxdh->dccph_type == DCCP_PKT_RESET)
702                 return;
703
704         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
705                 return;
706
707         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
708         if (dst == NULL)
709                 return;
710
711         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header,
712                         GFP_ATOMIC);
713         if (skb == NULL)
714                 goto out;
715
716         /* Reserve space for headers. */
717         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
718         skb->dst = dst_clone(dst);
719
720         dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
721
722         /* Build DCCP header and checksum it. */
723         dh->dccph_type     = DCCP_PKT_RESET;
724         dh->dccph_sport    = rxdh->dccph_dport;
725         dh->dccph_dport    = rxdh->dccph_sport;
726         dh->dccph_doff     = dccp_hdr_reset_len / 4;
727         dh->dccph_x        = 1;
728         dccp_hdr_reset(skb)->dccph_reset_code =
729                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
730
731         /* See "8.3.1. Abnormal Termination" in RFC 4340 */
732         seqno = 0;
733         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
734                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
735
736         dccp_hdr_set_seq(dh, seqno);
737         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
738                          DCCP_SKB_CB(rxskb)->dccpd_seq);
739
740         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
741                                               rxskb->nh.iph->daddr);
742
743         bh_lock_sock(dccp_v4_ctl_socket->sk);
744         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
745                                     rxskb->nh.iph->daddr,
746                                     rxskb->nh.iph->saddr, NULL);
747         bh_unlock_sock(dccp_v4_ctl_socket->sk);
748
749         if (err == NET_XMIT_CN || err == 0) {
750                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
751                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
752         }
753 out:
754          dst_release(dst);
755 }
756
757 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
758 {
759         struct dccp_hdr *dh = dccp_hdr(skb);
760
761         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
762                 if (dccp_rcv_established(sk, skb, dh, skb->len))
763                         goto reset;
764                 return 0;
765         }
766
767         /*
768          *  Step 3: Process LISTEN state
769          *     If S.state == LISTEN,
770          *        If P.type == Request or P contains a valid Init Cookie
771          *              option,
772          *           * Must scan the packet's options to check for an Init
773          *              Cookie.  Only the Init Cookie is processed here,
774          *              however; other options are processed in Step 8.  This
775          *              scan need only be performed if the endpoint uses Init
776          *              Cookies *
777          *           * Generate a new socket and switch to that socket *
778          *           Set S := new socket for this port pair
779          *           S.state = RESPOND
780          *           Choose S.ISS (initial seqno) or set from Init Cookie
781          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
782          *           Continue with S.state == RESPOND
783          *           * A Response packet will be generated in Step 11 *
784          *        Otherwise,
785          *           Generate Reset(No Connection) unless P.type == Reset
786          *           Drop packet and return
787          *
788          * NOTE: the check for the packet types is done in
789          *       dccp_rcv_state_process
790          */
791         if (sk->sk_state == DCCP_LISTEN) {
792                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
793
794                 if (nsk == NULL)
795                         goto discard;
796
797                 if (nsk != sk) {
798                         if (dccp_child_process(sk, nsk, skb))
799                                 goto reset;
800                         return 0;
801                 }
802         }
803
804         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
805                 goto reset;
806         return 0;
807
808 reset:
809         dccp_v4_ctl_send_reset(skb);
810 discard:
811         kfree_skb(skb);
812         return 0;
813 }
814
815 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
816
817 int dccp_invalid_packet(struct sk_buff *skb)
818 {
819         const struct dccp_hdr *dh;
820
821         if (skb->pkt_type != PACKET_HOST)
822                 return 1;
823
824         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
825                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
826                 return 1;
827         }
828
829         dh = dccp_hdr(skb);
830
831         /* If the packet type is not understood, drop packet and return */
832         if (dh->dccph_type >= DCCP_PKT_INVALID) {
833                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
834                 return 1;
835         }
836
837         /*
838          * If P.Data Offset is too small for packet type, or too large for
839          * packet, drop packet and return
840          */
841         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
842                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
843                                             "too small 1\n",
844                                dh->dccph_doff);
845                 return 1;
846         }
847
848         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
849                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
850                                             "too small 2\n",
851                                dh->dccph_doff);
852                 return 1;
853         }
854
855         dh = dccp_hdr(skb);
856
857         /*
858          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
859          * has short sequence numbers), drop packet and return
860          */
861         if (dh->dccph_x == 0 &&
862             dh->dccph_type != DCCP_PKT_DATA &&
863             dh->dccph_type != DCCP_PKT_ACK &&
864             dh->dccph_type != DCCP_PKT_DATAACK) {
865                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
866                                             "nor DataAck and P.X == 0\n",
867                                dccp_packet_name(dh->dccph_type));
868                 return 1;
869         }
870
871         return 0;
872 }
873
874 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
875
876 /* this is called when real data arrives */
877 static int dccp_v4_rcv(struct sk_buff *skb)
878 {
879         const struct dccp_hdr *dh;
880         struct sock *sk;
881
882         /* Step 1: Check header basics: */
883
884         if (dccp_invalid_packet(skb))
885                 goto discard_it;
886
887         /* If the header checksum is incorrect, drop packet and return */
888         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
889                                     skb->nh.iph->daddr) < 0) {
890                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
891                                __FUNCTION__);
892                 goto discard_it;
893         }
894
895         dh = dccp_hdr(skb);
896
897         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
898         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
899
900         dccp_pr_debug("%8.8s "
901                       "src=%u.%u.%u.%u@%-5d "
902                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
903                       dccp_packet_name(dh->dccph_type),
904                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
905                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
906                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
907
908         if (dccp_packet_without_ack(skb)) {
909                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
910                 dccp_pr_debug_cat("\n");
911         } else {
912                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
913                 dccp_pr_debug_cat(", ack=%llu\n",
914                                   (unsigned long long)
915                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
916         }
917
918         /* Step 2:
919          *      Look up flow ID in table and get corresponding socket */
920         sk = __inet_lookup(&dccp_hashinfo,
921                            skb->nh.iph->saddr, dh->dccph_sport,
922                            skb->nh.iph->daddr, dh->dccph_dport,
923                            inet_iif(skb));
924
925         /* 
926          * Step 2:
927          *      If no socket ...
928          *              Generate Reset(No Connection) unless P.type == Reset
929          *              Drop packet and return
930          */
931         if (sk == NULL) {
932                 dccp_pr_debug("failed to look up flow ID in table and "
933                               "get corresponding socket\n");
934                 goto no_dccp_socket;
935         }
936
937         /* 
938          * Step 2:
939          *      ... or S.state == TIMEWAIT,
940          *              Generate Reset(No Connection) unless P.type == Reset
941          *              Drop packet and return
942          */
943                
944         if (sk->sk_state == DCCP_TIME_WAIT) {
945                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
946                               "do_time_wait\n");
947                 goto do_time_wait;
948         }
949
950         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
951                 goto discard_and_relse;
952         nf_reset(skb);
953
954         return sk_receive_skb(sk, skb);
955
956 no_dccp_socket:
957         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
958                 goto discard_it;
959         /*
960          * Step 2:
961          *              Generate Reset(No Connection) unless P.type == Reset
962          *              Drop packet and return
963          */
964         if (dh->dccph_type != DCCP_PKT_RESET) {
965                 DCCP_SKB_CB(skb)->dccpd_reset_code =
966                                         DCCP_RESET_CODE_NO_CONNECTION;
967                 dccp_v4_ctl_send_reset(skb);
968         }
969
970 discard_it:
971         /* Discard frame. */
972         kfree_skb(skb);
973         return 0;
974
975 discard_and_relse:
976         sock_put(sk);
977         goto discard_it;
978
979 do_time_wait:
980         inet_twsk_put(inet_twsk(sk));
981         goto no_dccp_socket;
982 }
983
984 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
985         .queue_xmit        = ip_queue_xmit,
986         .send_check        = dccp_v4_send_check,
987         .rebuild_header    = inet_sk_rebuild_header,
988         .conn_request      = dccp_v4_conn_request,
989         .syn_recv_sock     = dccp_v4_request_recv_sock,
990         .net_header_len    = sizeof(struct iphdr),
991         .setsockopt        = ip_setsockopt,
992         .getsockopt        = ip_getsockopt,
993         .addr2sockaddr     = inet_csk_addr2sockaddr,
994         .sockaddr_len      = sizeof(struct sockaddr_in),
995 #ifdef CONFIG_COMPAT
996         .compat_setsockopt = compat_ip_setsockopt,
997         .compat_getsockopt = compat_ip_getsockopt,
998 #endif
999 };
1000
1001 static int dccp_v4_init_sock(struct sock *sk)
1002 {
1003         static __u8 dccp_v4_ctl_sock_initialized;
1004         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
1005
1006         if (err == 0) {
1007                 if (unlikely(!dccp_v4_ctl_sock_initialized))
1008                         dccp_v4_ctl_sock_initialized = 1;
1009                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1010         }
1011
1012         return err;
1013 }
1014
1015 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1016 {
1017         kfree(inet_rsk(req)->opt);
1018 }
1019
1020 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
1021         .family         = PF_INET,
1022         .obj_size       = sizeof(struct dccp_request_sock),
1023         .rtx_syn_ack    = dccp_v4_send_response,
1024         .send_ack       = dccp_v4_reqsk_send_ack,
1025         .destructor     = dccp_v4_reqsk_destructor,
1026         .send_reset     = dccp_v4_ctl_send_reset,
1027 };
1028
1029 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1030         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1031 };
1032
1033 static struct proto dccp_v4_prot = {
1034         .name                   = "DCCP",
1035         .owner                  = THIS_MODULE,
1036         .close                  = dccp_close,
1037         .connect                = dccp_v4_connect,
1038         .disconnect             = dccp_disconnect,
1039         .ioctl                  = dccp_ioctl,
1040         .init                   = dccp_v4_init_sock,
1041         .setsockopt             = dccp_setsockopt,
1042         .getsockopt             = dccp_getsockopt,
1043         .sendmsg                = dccp_sendmsg,
1044         .recvmsg                = dccp_recvmsg,
1045         .backlog_rcv            = dccp_v4_do_rcv,
1046         .hash                   = dccp_hash,
1047         .unhash                 = dccp_unhash,
1048         .accept                 = inet_csk_accept,
1049         .get_port               = dccp_v4_get_port,
1050         .shutdown               = dccp_shutdown,
1051         .destroy                = dccp_destroy_sock,
1052         .orphan_count           = &dccp_orphan_count,
1053         .max_header             = MAX_DCCP_HEADER,
1054         .obj_size               = sizeof(struct dccp_sock),
1055         .rsk_prot               = &dccp_request_sock_ops,
1056         .twsk_prot              = &dccp_timewait_sock_ops,
1057 #ifdef CONFIG_COMPAT
1058         .compat_setsockopt      = compat_dccp_setsockopt,
1059         .compat_getsockopt      = compat_dccp_getsockopt,
1060 #endif
1061 };
1062
1063 static struct net_protocol dccp_v4_protocol = {
1064         .handler        = dccp_v4_rcv,
1065         .err_handler    = dccp_v4_err,
1066         .no_policy      = 1,
1067 };
1068
1069 static const struct proto_ops inet_dccp_ops = {
1070         .family            = PF_INET,
1071         .owner             = THIS_MODULE,
1072         .release           = inet_release,
1073         .bind              = inet_bind,
1074         .connect           = inet_stream_connect,
1075         .socketpair        = sock_no_socketpair,
1076         .accept            = inet_accept,
1077         .getname           = inet_getname,
1078         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1079         .poll              = dccp_poll,
1080         .ioctl             = inet_ioctl,
1081         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1082         .listen            = inet_dccp_listen,
1083         .shutdown          = inet_shutdown,
1084         .setsockopt        = sock_common_setsockopt,
1085         .getsockopt        = sock_common_getsockopt,
1086         .sendmsg           = inet_sendmsg,
1087         .recvmsg           = sock_common_recvmsg,
1088         .mmap              = sock_no_mmap,
1089         .sendpage          = sock_no_sendpage,
1090 #ifdef CONFIG_COMPAT
1091         .compat_setsockopt = compat_sock_common_setsockopt,
1092         .compat_getsockopt = compat_sock_common_getsockopt,
1093 #endif
1094 };
1095
1096 static struct inet_protosw dccp_v4_protosw = {
1097         .type           = SOCK_DCCP,
1098         .protocol       = IPPROTO_DCCP,
1099         .prot           = &dccp_v4_prot,
1100         .ops            = &inet_dccp_ops,
1101         .capability     = -1,
1102         .no_check       = 0,
1103         .flags          = INET_PROTOSW_ICSK,
1104 };
1105
1106 static int __init dccp_v4_init(void)
1107 {
1108         int err = proto_register(&dccp_v4_prot, 1);
1109
1110         if (err != 0)
1111                 goto out;
1112
1113         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1114         if (err != 0)
1115                 goto out_proto_unregister;
1116
1117         inet_register_protosw(&dccp_v4_protosw);
1118
1119         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1120                                        SOCK_DCCP, IPPROTO_DCCP);
1121         if (err)
1122                 goto out_unregister_protosw;
1123 out:
1124         return err;
1125 out_unregister_protosw:
1126         inet_unregister_protosw(&dccp_v4_protosw);
1127         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1128 out_proto_unregister:
1129         proto_unregister(&dccp_v4_prot);
1130         goto out;
1131 }
1132
1133 static void __exit dccp_v4_exit(void)
1134 {
1135         inet_unregister_protosw(&dccp_v4_protosw);
1136         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1137         proto_unregister(&dccp_v4_prot);
1138 }
1139
1140 module_init(dccp_v4_init);
1141 module_exit(dccp_v4_exit);
1142
1143 /*
1144  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1145  * values directly, Also cover the case where the protocol is not specified,
1146  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1147  */
1148 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1149 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1150 MODULE_LICENSE("GPL");
1151 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1152 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");