Merge branch 'upstream-fixes'
[linux-2.6] / net / dccp / ipv4.c
1 /*
2  *  net/dccp/ipv4.c
3  *
4  *  An implementation of the DCCP protocol
5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
6  *
7  *      This program is free software; you can redistribute it and/or
8  *      modify it under the terms of the GNU General Public License
9  *      as published by the Free Software Foundation; either version
10  *      2 of the License, or (at your option) any later version.
11  */
12
13 #include <linux/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_hashtables.h>
22 #include <net/inet_sock.h>
23 #include <net/sock.h>
24 #include <net/timewait_sock.h>
25 #include <net/tcp_states.h>
26 #include <net/xfrm.h>
27
28 #include "ackvec.h"
29 #include "ccid.h"
30 #include "dccp.h"
31
32 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
33         .lhash_lock     = RW_LOCK_UNLOCKED,
34         .lhash_users    = ATOMIC_INIT(0),
35         .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
36 };
37
38 EXPORT_SYMBOL_GPL(dccp_hashinfo);
39
40 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
41 {
42         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
43                                  inet_csk_bind_conflict);
44 }
45
46 static void dccp_v4_hash(struct sock *sk)
47 {
48         inet_hash(&dccp_hashinfo, sk);
49 }
50
51 void dccp_unhash(struct sock *sk)
52 {
53         inet_unhash(&dccp_hashinfo, sk);
54 }
55
56 EXPORT_SYMBOL_GPL(dccp_unhash);
57
58 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
59 {
60         struct inet_sock *inet = inet_sk(sk);
61         struct dccp_sock *dp = dccp_sk(sk);
62         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
63         struct rtable *rt;
64         u32 daddr, nexthop;
65         int tmp;
66         int err;
67
68         dp->dccps_role = DCCP_ROLE_CLIENT;
69
70         if (dccp_service_not_initialized(sk))
71                 return -EPROTO;
72
73         if (addr_len < sizeof(struct sockaddr_in))
74                 return -EINVAL;
75
76         if (usin->sin_family != AF_INET)
77                 return -EAFNOSUPPORT;
78
79         nexthop = daddr = usin->sin_addr.s_addr;
80         if (inet->opt != NULL && inet->opt->srr) {
81                 if (daddr == 0)
82                         return -EINVAL;
83                 nexthop = inet->opt->faddr;
84         }
85
86         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
87                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
88                                IPPROTO_DCCP,
89                                inet->sport, usin->sin_port, sk);
90         if (tmp < 0)
91                 return tmp;
92
93         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
94                 ip_rt_put(rt);
95                 return -ENETUNREACH;
96         }
97
98         if (inet->opt == NULL || !inet->opt->srr)
99                 daddr = rt->rt_dst;
100
101         if (inet->saddr == 0)
102                 inet->saddr = rt->rt_src;
103         inet->rcv_saddr = inet->saddr;
104
105         inet->dport = usin->sin_port;
106         inet->daddr = daddr;
107
108         inet_csk(sk)->icsk_ext_hdr_len = 0;
109         if (inet->opt != NULL)
110                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
111         /*
112          * Socket identity is still unknown (sport may be zero).
113          * However we set state to DCCP_REQUESTING and not releasing socket
114          * lock select source port, enter ourselves into the hash tables and
115          * complete initialization after this.
116          */
117         dccp_set_state(sk, DCCP_REQUESTING);
118         err = inet_hash_connect(&dccp_death_row, sk);
119         if (err != 0)
120                 goto failure;
121
122         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
123                                 sk);
124         if (err != 0)
125                 goto failure;
126
127         /* OK, now commit destination to socket.  */
128         sk_setup_caps(sk, &rt->u.dst);
129
130         dp->dccps_gar =
131                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
132                                                             inet->daddr,
133                                                             inet->sport,
134                                                             usin->sin_port);
135         dccp_update_gss(sk, dp->dccps_iss);
136
137         inet->id = dp->dccps_iss ^ jiffies;
138
139         err = dccp_connect(sk);
140         rt = NULL;
141         if (err != 0)
142                 goto failure;
143 out:
144         return err;
145 failure:
146         /*
147          * This unhashes the socket and releases the local port, if necessary.
148          */
149         dccp_set_state(sk, DCCP_CLOSED);
150         ip_rt_put(rt);
151         sk->sk_route_caps = 0;
152         inet->dport = 0;
153         goto out;
154 }
155
156 EXPORT_SYMBOL_GPL(dccp_v4_connect);
157
158 /*
159  * This routine does path mtu discovery as defined in RFC1191.
160  */
161 static inline void dccp_do_pmtu_discovery(struct sock *sk,
162                                           const struct iphdr *iph,
163                                           u32 mtu)
164 {
165         struct dst_entry *dst;
166         const struct inet_sock *inet = inet_sk(sk);
167         const struct dccp_sock *dp = dccp_sk(sk);
168
169         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
170          * send out by Linux are always < 576bytes so they should go through
171          * unfragmented).
172          */
173         if (sk->sk_state == DCCP_LISTEN)
174                 return;
175
176         /* We don't check in the destentry if pmtu discovery is forbidden
177          * on this route. We just assume that no packet_to_big packets
178          * are send back when pmtu discovery is not active.
179          * There is a small race when the user changes this flag in the
180          * route, but I think that's acceptable.
181          */
182         if ((dst = __sk_dst_check(sk, 0)) == NULL)
183                 return;
184
185         dst->ops->update_pmtu(dst, mtu);
186
187         /* Something is about to be wrong... Remember soft error
188          * for the case, if this connection will not able to recover.
189          */
190         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
191                 sk->sk_err_soft = EMSGSIZE;
192
193         mtu = dst_mtu(dst);
194
195         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
196             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
197                 dccp_sync_mss(sk, mtu);
198
199                 /*
200                  * From: draft-ietf-dccp-spec-11.txt
201                  *
202                  *      DCCP-Sync packets are the best choice for upward
203                  *      probing, since DCCP-Sync probes do not risk application
204                  *      data loss.
205                  */
206                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
207         } /* else let the usual retransmit timer handle it */
208 }
209
210 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
211 {
212         int err;
213         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
214         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
215                                      sizeof(struct dccp_hdr_ext) +
216                                      sizeof(struct dccp_hdr_ack_bits);
217         struct sk_buff *skb;
218
219         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
220                 return;
221
222         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
223         if (skb == NULL)
224                 return;
225
226         /* Reserve space for headers. */
227         skb_reserve(skb, MAX_DCCP_HEADER);
228
229         skb->dst = dst_clone(rxskb->dst);
230
231         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
232         dh = dccp_hdr(skb);
233         memset(dh, 0, dccp_hdr_ack_len);
234
235         /* Build DCCP header and checksum it. */
236         dh->dccph_type     = DCCP_PKT_ACK;
237         dh->dccph_sport    = rxdh->dccph_dport;
238         dh->dccph_dport    = rxdh->dccph_sport;
239         dh->dccph_doff     = dccp_hdr_ack_len / 4;
240         dh->dccph_x        = 1;
241
242         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
243         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
244                          DCCP_SKB_CB(rxskb)->dccpd_seq);
245
246         bh_lock_sock(dccp_ctl_socket->sk);
247         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
248                                     rxskb->nh.iph->daddr,
249                                     rxskb->nh.iph->saddr, NULL);
250         bh_unlock_sock(dccp_ctl_socket->sk);
251
252         if (err == NET_XMIT_CN || err == 0) {
253                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
254                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
255         }
256 }
257
258 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
259                                    struct request_sock *req)
260 {
261         dccp_v4_ctl_send_ack(skb);
262 }
263
264 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
265                                  struct dst_entry *dst)
266 {
267         int err = -1;
268         struct sk_buff *skb;
269
270         /* First, grab a route. */
271         
272         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
273                 goto out;
274
275         skb = dccp_make_response(sk, dst, req);
276         if (skb != NULL) {
277                 const struct inet_request_sock *ireq = inet_rsk(req);
278
279                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
280                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
281                                             ireq->rmt_addr,
282                                             ireq->opt);
283                 if (err == NET_XMIT_CN)
284                         err = 0;
285         }
286
287 out:
288         dst_release(dst);
289         return err;
290 }
291
292 /*
293  * This routine is called by the ICMP module when it gets some sort of error
294  * condition. If err < 0 then the socket should be closed and the error
295  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
296  * After adjustment header points to the first 8 bytes of the tcp header. We
297  * need to find the appropriate port.
298  *
299  * The locking strategy used here is very "optimistic". When someone else
300  * accesses the socket the ICMP is just dropped and for some paths there is no
301  * check at all. A more general error queue to queue errors for later handling
302  * is probably better.
303  */
304 void dccp_v4_err(struct sk_buff *skb, u32 info)
305 {
306         const struct iphdr *iph = (struct iphdr *)skb->data;
307         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
308                                                         (iph->ihl << 2));
309         struct dccp_sock *dp;
310         struct inet_sock *inet;
311         const int type = skb->h.icmph->type;
312         const int code = skb->h.icmph->code;
313         struct sock *sk;
314         __u64 seq;
315         int err;
316
317         if (skb->len < (iph->ihl << 2) + 8) {
318                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
319                 return;
320         }
321
322         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
323                          iph->saddr, dh->dccph_sport, inet_iif(skb));
324         if (sk == NULL) {
325                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
326                 return;
327         }
328
329         if (sk->sk_state == DCCP_TIME_WAIT) {
330                 inet_twsk_put((struct inet_timewait_sock *)sk);
331                 return;
332         }
333
334         bh_lock_sock(sk);
335         /* If too many ICMPs get dropped on busy
336          * servers this needs to be solved differently.
337          */
338         if (sock_owned_by_user(sk))
339                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
340
341         if (sk->sk_state == DCCP_CLOSED)
342                 goto out;
343
344         dp = dccp_sk(sk);
345         seq = dccp_hdr_seq(skb);
346         if (sk->sk_state != DCCP_LISTEN &&
347             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
348                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
349                 goto out;
350         }
351
352         switch (type) {
353         case ICMP_SOURCE_QUENCH:
354                 /* Just silently ignore these. */
355                 goto out;
356         case ICMP_PARAMETERPROB:
357                 err = EPROTO;
358                 break;
359         case ICMP_DEST_UNREACH:
360                 if (code > NR_ICMP_UNREACH)
361                         goto out;
362
363                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
364                         if (!sock_owned_by_user(sk))
365                                 dccp_do_pmtu_discovery(sk, iph, info);
366                         goto out;
367                 }
368
369                 err = icmp_err_convert[code].errno;
370                 break;
371         case ICMP_TIME_EXCEEDED:
372                 err = EHOSTUNREACH;
373                 break;
374         default:
375                 goto out;
376         }
377
378         switch (sk->sk_state) {
379                 struct request_sock *req , **prev;
380         case DCCP_LISTEN:
381                 if (sock_owned_by_user(sk))
382                         goto out;
383                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
384                                           iph->daddr, iph->saddr);
385                 if (!req)
386                         goto out;
387
388                 /*
389                  * ICMPs are not backlogged, hence we cannot get an established
390                  * socket here.
391                  */
392                 BUG_TRAP(!req->sk);
393
394                 if (seq != dccp_rsk(req)->dreq_iss) {
395                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
396                         goto out;
397                 }
398                 /*
399                  * Still in RESPOND, just remove it silently.
400                  * There is no good way to pass the error to the newly
401                  * created socket, and POSIX does not want network
402                  * errors returned from accept().
403                  */
404                 inet_csk_reqsk_queue_drop(sk, req, prev);
405                 goto out;
406
407         case DCCP_REQUESTING:
408         case DCCP_RESPOND:
409                 if (!sock_owned_by_user(sk)) {
410                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
411                         sk->sk_err = err;
412
413                         sk->sk_error_report(sk);
414
415                         dccp_done(sk);
416                 } else
417                         sk->sk_err_soft = err;
418                 goto out;
419         }
420
421         /* If we've already connected we will keep trying
422          * until we time out, or the user gives up.
423          *
424          * rfc1122 4.2.3.9 allows to consider as hard errors
425          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
426          * but it is obsoleted by pmtu discovery).
427          *
428          * Note, that in modern internet, where routing is unreliable
429          * and in each dark corner broken firewalls sit, sending random
430          * errors ordered by their masters even this two messages finally lose
431          * their original sense (even Linux sends invalid PORT_UNREACHs)
432          *
433          * Now we are in compliance with RFCs.
434          *                                                      --ANK (980905)
435          */
436
437         inet = inet_sk(sk);
438         if (!sock_owned_by_user(sk) && inet->recverr) {
439                 sk->sk_err = err;
440                 sk->sk_error_report(sk);
441         } else /* Only an error on timeout */
442                 sk->sk_err_soft = err;
443 out:
444         bh_unlock_sock(sk);
445         sock_put(sk);
446 }
447
448 /* This routine computes an IPv4 DCCP checksum. */
449 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
450 {
451         const struct inet_sock *inet = inet_sk(sk);
452         struct dccp_hdr *dh = dccp_hdr(skb);
453
454         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
455 }
456
457 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
458
459 int dccp_v4_send_reset(struct sock *sk, enum dccp_reset_codes code)
460 {
461         struct sk_buff *skb;
462         /*
463          * FIXME: what if rebuild_header fails?
464          * Should we be doing a rebuild_header here?
465          */
466         int err = inet_sk_rebuild_header(sk);
467
468         if (err != 0)
469                 return err;
470
471         skb = dccp_make_reset(sk, sk->sk_dst_cache, code);
472         if (skb != NULL) {
473                 const struct inet_sock *inet = inet_sk(sk);
474
475                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
476                 err = ip_build_and_send_pkt(skb, sk,
477                                             inet->saddr, inet->daddr, NULL);
478                 if (err == NET_XMIT_CN)
479                         err = 0;
480         }
481
482         return err;
483 }
484
485 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
486                                         const struct sk_buff *skb)
487 {
488         return secure_dccp_sequence_number(skb->nh.iph->daddr,
489                                            skb->nh.iph->saddr,
490                                            dccp_hdr(skb)->dccph_dport,
491                                            dccp_hdr(skb)->dccph_sport);
492 }
493
494 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
495 {
496         struct inet_request_sock *ireq;
497         struct dccp_sock dp;
498         struct request_sock *req;
499         struct dccp_request_sock *dreq;
500         const __u32 saddr = skb->nh.iph->saddr;
501         const __u32 daddr = skb->nh.iph->daddr;
502         const __u32 service = dccp_hdr_request(skb)->dccph_req_service;
503         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
504         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
505
506         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
507         if (((struct rtable *)skb->dst)->rt_flags &
508             (RTCF_BROADCAST | RTCF_MULTICAST)) {
509                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
510                 goto drop;
511         }
512
513         if (dccp_bad_service_code(sk, service)) {
514                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
515                 goto drop;
516         }
517         /*
518          * TW buckets are converted to open requests without
519          * limitations, they conserve resources and peer is
520          * evidently real one.
521          */
522         if (inet_csk_reqsk_queue_is_full(sk))
523                 goto drop;
524
525         /*
526          * Accept backlog is full. If we have already queued enough
527          * of warm entries in syn queue, drop request. It is better than
528          * clogging syn queue with openreqs with exponentially increasing
529          * timeout.
530          */
531         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
532                 goto drop;
533
534         req = reqsk_alloc(sk->sk_prot->rsk_prot);
535         if (req == NULL)
536                 goto drop;
537
538         /* FIXME: process options */
539
540         dccp_openreq_init(req, &dp, skb);
541
542         ireq = inet_rsk(req);
543         ireq->loc_addr = daddr;
544         ireq->rmt_addr = saddr;
545         req->rcv_wnd    = 100; /* Fake, option parsing will get the
546                                   right value */
547         ireq->opt       = NULL;
548
549         /* 
550          * Step 3: Process LISTEN state
551          *
552          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
553          *
554          * In fact we defer setting S.GSR, S.SWL, S.SWH to
555          * dccp_create_openreq_child.
556          */
557         dreq = dccp_rsk(req);
558         dreq->dreq_isr     = dcb->dccpd_seq;
559         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
560         dreq->dreq_service = service;
561
562         if (dccp_v4_send_response(sk, req, NULL))
563                 goto drop_and_free;
564
565         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
566         return 0;
567
568 drop_and_free:
569         reqsk_free(req);
570 drop:
571         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
572         dcb->dccpd_reset_code = reset_code;
573         return -1;
574 }
575
576 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
577
578 /*
579  * The three way handshake has completed - we got a valid ACK or DATAACK -
580  * now create the new socket.
581  *
582  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
583  */
584 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
585                                        struct request_sock *req,
586                                        struct dst_entry *dst)
587 {
588         struct inet_request_sock *ireq;
589         struct inet_sock *newinet;
590         struct dccp_sock *newdp;
591         struct sock *newsk;
592
593         if (sk_acceptq_is_full(sk))
594                 goto exit_overflow;
595
596         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
597                 goto exit;
598
599         newsk = dccp_create_openreq_child(sk, req, skb);
600         if (newsk == NULL)
601                 goto exit;
602
603         sk_setup_caps(newsk, dst);
604
605         newdp              = dccp_sk(newsk);
606         newinet            = inet_sk(newsk);
607         ireq               = inet_rsk(req);
608         newinet->daddr     = ireq->rmt_addr;
609         newinet->rcv_saddr = ireq->loc_addr;
610         newinet->saddr     = ireq->loc_addr;
611         newinet->opt       = ireq->opt;
612         ireq->opt          = NULL;
613         newinet->mc_index  = inet_iif(skb);
614         newinet->mc_ttl    = skb->nh.iph->ttl;
615         newinet->id        = jiffies;
616
617         dccp_sync_mss(newsk, dst_mtu(dst));
618
619         __inet_hash(&dccp_hashinfo, newsk, 0);
620         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
621
622         return newsk;
623
624 exit_overflow:
625         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
626 exit:
627         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
628         dst_release(dst);
629         return NULL;
630 }
631
632 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
633
634 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
635 {
636         const struct dccp_hdr *dh = dccp_hdr(skb);
637         const struct iphdr *iph = skb->nh.iph;
638         struct sock *nsk;
639         struct request_sock **prev;
640         /* Find possible connection requests. */
641         struct request_sock *req = inet_csk_search_req(sk, &prev,
642                                                        dh->dccph_sport,
643                                                        iph->saddr, iph->daddr);
644         if (req != NULL)
645                 return dccp_check_req(sk, skb, req, prev);
646
647         nsk = __inet_lookup_established(&dccp_hashinfo,
648                                         iph->saddr, dh->dccph_sport,
649                                         iph->daddr, ntohs(dh->dccph_dport),
650                                         inet_iif(skb));
651         if (nsk != NULL) {
652                 if (nsk->sk_state != DCCP_TIME_WAIT) {
653                         bh_lock_sock(nsk);
654                         return nsk;
655                 }
656                 inet_twsk_put((struct inet_timewait_sock *)nsk);
657                 return NULL;
658         }
659
660         return sk;
661 }
662
663 int dccp_v4_checksum(const struct sk_buff *skb, const u32 saddr,
664                      const u32 daddr)
665 {
666         const struct dccp_hdr* dh = dccp_hdr(skb);
667         int checksum_len;
668         u32 tmp;
669
670         if (dh->dccph_cscov == 0)
671                 checksum_len = skb->len;
672         else {
673                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
674                 checksum_len = checksum_len < skb->len ? checksum_len :
675                                                          skb->len;
676         }
677
678         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
679         return csum_tcpudp_magic(saddr, daddr, checksum_len,
680                                  IPPROTO_DCCP, tmp);
681 }
682
683 static int dccp_v4_verify_checksum(struct sk_buff *skb,
684                                    const u32 saddr, const u32 daddr)
685 {
686         struct dccp_hdr *dh = dccp_hdr(skb);
687         int checksum_len;
688         u32 tmp;
689
690         if (dh->dccph_cscov == 0)
691                 checksum_len = skb->len;
692         else {
693                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
694                 checksum_len = checksum_len < skb->len ? checksum_len :
695                                                          skb->len;
696         }
697         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
698         return csum_tcpudp_magic(saddr, daddr, checksum_len,
699                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
700 }
701
702 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
703                                            struct sk_buff *skb)
704 {
705         struct rtable *rt;
706         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
707                             .nl_u = { .ip4_u =
708                                       { .daddr = skb->nh.iph->saddr,
709                                         .saddr = skb->nh.iph->daddr,
710                                         .tos = RT_CONN_FLAGS(sk) } },
711                             .proto = sk->sk_protocol,
712                             .uli_u = { .ports =
713                                        { .sport = dccp_hdr(skb)->dccph_dport,
714                                          .dport = dccp_hdr(skb)->dccph_sport }
715                                      }
716                           };
717
718         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
719                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
720                 return NULL;
721         }
722
723         return &rt->u.dst;
724 }
725
726 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
727 {
728         int err;
729         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
730         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
731                                        sizeof(struct dccp_hdr_ext) +
732                                        sizeof(struct dccp_hdr_reset);
733         struct sk_buff *skb;
734         struct dst_entry *dst;
735         u64 seqno;
736
737         /* Never send a reset in response to a reset. */
738         if (rxdh->dccph_type == DCCP_PKT_RESET)
739                 return;
740
741         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
742                 return;
743
744         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
745         if (dst == NULL)
746                 return;
747
748         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
749         if (skb == NULL)
750                 goto out;
751
752         /* Reserve space for headers. */
753         skb_reserve(skb, MAX_DCCP_HEADER);
754         skb->dst = dst_clone(dst);
755
756         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
757         dh = dccp_hdr(skb);
758         memset(dh, 0, dccp_hdr_reset_len);
759
760         /* Build DCCP header and checksum it. */
761         dh->dccph_type     = DCCP_PKT_RESET;
762         dh->dccph_sport    = rxdh->dccph_dport;
763         dh->dccph_dport    = rxdh->dccph_sport;
764         dh->dccph_doff     = dccp_hdr_reset_len / 4;
765         dh->dccph_x        = 1;
766         dccp_hdr_reset(skb)->dccph_reset_code =
767                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
768
769         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
770         seqno = 0;
771         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
772                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
773
774         dccp_hdr_set_seq(dh, seqno);
775         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
776                          DCCP_SKB_CB(rxskb)->dccpd_seq);
777
778         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
779                                               rxskb->nh.iph->daddr);
780
781         bh_lock_sock(dccp_ctl_socket->sk);
782         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
783                                     rxskb->nh.iph->daddr,
784                                     rxskb->nh.iph->saddr, NULL);
785         bh_unlock_sock(dccp_ctl_socket->sk);
786
787         if (err == NET_XMIT_CN || err == 0) {
788                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
789                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
790         }
791 out:
792          dst_release(dst);
793 }
794
795 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
796 {
797         struct dccp_hdr *dh = dccp_hdr(skb);
798
799         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
800                 if (dccp_rcv_established(sk, skb, dh, skb->len))
801                         goto reset;
802                 return 0;
803         }
804
805         /*
806          *  Step 3: Process LISTEN state
807          *     If S.state == LISTEN,
808          *        If P.type == Request or P contains a valid Init Cookie
809          *              option,
810          *           * Must scan the packet's options to check for an Init
811          *              Cookie.  Only the Init Cookie is processed here,
812          *              however; other options are processed in Step 8.  This
813          *              scan need only be performed if the endpoint uses Init
814          *              Cookies *
815          *           * Generate a new socket and switch to that socket *
816          *           Set S := new socket for this port pair
817          *           S.state = RESPOND
818          *           Choose S.ISS (initial seqno) or set from Init Cookie
819          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
820          *           Continue with S.state == RESPOND
821          *           * A Response packet will be generated in Step 11 *
822          *        Otherwise,
823          *           Generate Reset(No Connection) unless P.type == Reset
824          *           Drop packet and return
825          *
826          * NOTE: the check for the packet types is done in
827          *       dccp_rcv_state_process
828          */
829         if (sk->sk_state == DCCP_LISTEN) {
830                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
831
832                 if (nsk == NULL)
833                         goto discard;
834
835                 if (nsk != sk) {
836                         if (dccp_child_process(sk, nsk, skb))
837                                 goto reset;
838                         return 0;
839                 }
840         }
841
842         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
843                 goto reset;
844         return 0;
845
846 reset:
847         dccp_v4_ctl_send_reset(skb);
848 discard:
849         kfree_skb(skb);
850         return 0;
851 }
852
853 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
854
855 int dccp_invalid_packet(struct sk_buff *skb)
856 {
857         const struct dccp_hdr *dh;
858
859         if (skb->pkt_type != PACKET_HOST)
860                 return 1;
861
862         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
863                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
864                 return 1;
865         }
866
867         dh = dccp_hdr(skb);
868
869         /* If the packet type is not understood, drop packet and return */
870         if (dh->dccph_type >= DCCP_PKT_INVALID) {
871                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
872                 return 1;
873         }
874
875         /*
876          * If P.Data Offset is too small for packet type, or too large for
877          * packet, drop packet and return
878          */
879         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
880                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
881                                             "too small 1\n",
882                                dh->dccph_doff);
883                 return 1;
884         }
885
886         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
887                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
888                                             "too small 2\n",
889                                dh->dccph_doff);
890                 return 1;
891         }
892
893         dh = dccp_hdr(skb);
894
895         /*
896          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
897          * has short sequence numbers), drop packet and return
898          */
899         if (dh->dccph_x == 0 &&
900             dh->dccph_type != DCCP_PKT_DATA &&
901             dh->dccph_type != DCCP_PKT_ACK &&
902             dh->dccph_type != DCCP_PKT_DATAACK) {
903                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
904                                             "nor DataAck and P.X == 0\n",
905                                dccp_packet_name(dh->dccph_type));
906                 return 1;
907         }
908
909         return 0;
910 }
911
912 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
913
914 /* this is called when real data arrives */
915 int dccp_v4_rcv(struct sk_buff *skb)
916 {
917         const struct dccp_hdr *dh;
918         struct sock *sk;
919
920         /* Step 1: Check header basics: */
921
922         if (dccp_invalid_packet(skb))
923                 goto discard_it;
924
925         /* If the header checksum is incorrect, drop packet and return */
926         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
927                                     skb->nh.iph->daddr) < 0) {
928                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
929                                __FUNCTION__);
930                 goto discard_it;
931         }
932
933         dh = dccp_hdr(skb);
934
935         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
936         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
937
938         dccp_pr_debug("%8.8s "
939                       "src=%u.%u.%u.%u@%-5d "
940                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
941                       dccp_packet_name(dh->dccph_type),
942                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
943                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
944                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
945
946         if (dccp_packet_without_ack(skb)) {
947                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
948                 dccp_pr_debug_cat("\n");
949         } else {
950                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
951                 dccp_pr_debug_cat(", ack=%llu\n",
952                                   (unsigned long long)
953                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
954         }
955
956         /* Step 2:
957          *      Look up flow ID in table and get corresponding socket */
958         sk = __inet_lookup(&dccp_hashinfo,
959                            skb->nh.iph->saddr, dh->dccph_sport,
960                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
961                            inet_iif(skb));
962
963         /* 
964          * Step 2:
965          *      If no socket ...
966          *              Generate Reset(No Connection) unless P.type == Reset
967          *              Drop packet and return
968          */
969         if (sk == NULL) {
970                 dccp_pr_debug("failed to look up flow ID in table and "
971                               "get corresponding socket\n");
972                 goto no_dccp_socket;
973         }
974
975         /* 
976          * Step 2:
977          *      ... or S.state == TIMEWAIT,
978          *              Generate Reset(No Connection) unless P.type == Reset
979          *              Drop packet and return
980          */
981                
982         if (sk->sk_state == DCCP_TIME_WAIT) {
983                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
984                               "do_time_wait\n");
985                 goto do_time_wait;
986         }
987
988         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
989                 goto discard_and_relse;
990         nf_reset(skb);
991
992         return sk_receive_skb(sk, skb);
993
994 no_dccp_socket:
995         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
996                 goto discard_it;
997         /*
998          * Step 2:
999          *              Generate Reset(No Connection) unless P.type == Reset
1000          *              Drop packet and return
1001          */
1002         if (dh->dccph_type != DCCP_PKT_RESET) {
1003                 DCCP_SKB_CB(skb)->dccpd_reset_code =
1004                                         DCCP_RESET_CODE_NO_CONNECTION;
1005                 dccp_v4_ctl_send_reset(skb);
1006         }
1007
1008 discard_it:
1009         /* Discard frame. */
1010         kfree_skb(skb);
1011         return 0;
1012
1013 discard_and_relse:
1014         sock_put(sk);
1015         goto discard_it;
1016
1017 do_time_wait:
1018         inet_twsk_put((struct inet_timewait_sock *)sk);
1019         goto no_dccp_socket;
1020 }
1021
1022 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
1023         .queue_xmit     = ip_queue_xmit,
1024         .send_check     = dccp_v4_send_check,
1025         .rebuild_header = inet_sk_rebuild_header,
1026         .conn_request   = dccp_v4_conn_request,
1027         .syn_recv_sock  = dccp_v4_request_recv_sock,
1028         .net_header_len = sizeof(struct iphdr),
1029         .setsockopt     = ip_setsockopt,
1030         .getsockopt     = ip_getsockopt,
1031         .addr2sockaddr  = inet_csk_addr2sockaddr,
1032         .sockaddr_len   = sizeof(struct sockaddr_in),
1033 };
1034
1035 int dccp_v4_init_sock(struct sock *sk)
1036 {
1037         struct dccp_sock *dp = dccp_sk(sk);
1038         struct inet_connection_sock *icsk = inet_csk(sk);
1039         static int dccp_ctl_socket_init = 1;
1040
1041         dccp_options_init(&dp->dccps_options);
1042         do_gettimeofday(&dp->dccps_epoch);
1043
1044         if (dp->dccps_options.dccpo_send_ack_vector) {
1045                 dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(DCCP_MAX_ACKVEC_LEN,
1046                                                            GFP_KERNEL);
1047                 if (dp->dccps_hc_rx_ackvec == NULL)
1048                         return -ENOMEM;
1049         }
1050
1051         /*
1052          * FIXME: We're hardcoding the CCID, and doing this at this point makes
1053          * the listening (master) sock get CCID control blocks, which is not
1054          * necessary, but for now, to not mess with the test userspace apps,
1055          * lets leave it here, later the real solution is to do this in a
1056          * setsockopt(CCIDs-I-want/accept). -acme
1057          */
1058         if (likely(!dccp_ctl_socket_init)) {
1059                 dp->dccps_hc_rx_ccid = ccid_init(dp->dccps_options.dccpo_rx_ccid,
1060                                                  sk);
1061                 dp->dccps_hc_tx_ccid = ccid_init(dp->dccps_options.dccpo_tx_ccid,
1062                                                  sk);
1063                 if (dp->dccps_hc_rx_ccid == NULL ||
1064                     dp->dccps_hc_tx_ccid == NULL) {
1065                         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1066                         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1067                         if (dp->dccps_options.dccpo_send_ack_vector) {
1068                                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1069                                 dp->dccps_hc_rx_ackvec = NULL;
1070                         }
1071                         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1072                         return -ENOMEM;
1073                 }
1074         } else
1075                 dccp_ctl_socket_init = 0;
1076
1077         dccp_init_xmit_timers(sk);
1078         icsk->icsk_rto = DCCP_TIMEOUT_INIT;
1079         sk->sk_state = DCCP_CLOSED;
1080         sk->sk_write_space = dccp_write_space;
1081         icsk->icsk_af_ops = &dccp_ipv4_af_ops;
1082         icsk->icsk_sync_mss = dccp_sync_mss;
1083         dp->dccps_mss_cache = 536;
1084         dp->dccps_role = DCCP_ROLE_UNDEFINED;
1085         dp->dccps_service = DCCP_SERVICE_INVALID_VALUE;
1086
1087         return 0;
1088 }
1089
1090 EXPORT_SYMBOL_GPL(dccp_v4_init_sock);
1091
1092 int dccp_v4_destroy_sock(struct sock *sk)
1093 {
1094         struct dccp_sock *dp = dccp_sk(sk);
1095
1096         /*
1097          * DCCP doesn't use sk_write_queue, just sk_send_head
1098          * for retransmissions
1099          */
1100         if (sk->sk_send_head != NULL) {
1101                 kfree_skb(sk->sk_send_head);
1102                 sk->sk_send_head = NULL;
1103         }
1104
1105         /* Clean up a referenced DCCP bind bucket. */
1106         if (inet_csk(sk)->icsk_bind_hash != NULL)
1107                 inet_put_port(&dccp_hashinfo, sk);
1108
1109         kfree(dp->dccps_service_list);
1110         dp->dccps_service_list = NULL;
1111
1112         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
1113         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
1114         if (dp->dccps_options.dccpo_send_ack_vector) {
1115                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1116                 dp->dccps_hc_rx_ackvec = NULL;
1117         }
1118         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1119         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1120         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1121
1122         return 0;
1123 }
1124
1125 EXPORT_SYMBOL_GPL(dccp_v4_destroy_sock);
1126
1127 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1128 {
1129         kfree(inet_rsk(req)->opt);
1130 }
1131
1132 static struct request_sock_ops dccp_request_sock_ops = {
1133         .family         = PF_INET,
1134         .obj_size       = sizeof(struct dccp_request_sock),
1135         .rtx_syn_ack    = dccp_v4_send_response,
1136         .send_ack       = dccp_v4_reqsk_send_ack,
1137         .destructor     = dccp_v4_reqsk_destructor,
1138         .send_reset     = dccp_v4_ctl_send_reset,
1139 };
1140
1141 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1142         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1143 };
1144
1145 struct proto dccp_prot = {
1146         .name                   = "DCCP",
1147         .owner                  = THIS_MODULE,
1148         .close                  = dccp_close,
1149         .connect                = dccp_v4_connect,
1150         .disconnect             = dccp_disconnect,
1151         .ioctl                  = dccp_ioctl,
1152         .init                   = dccp_v4_init_sock,
1153         .setsockopt             = dccp_setsockopt,
1154         .getsockopt             = dccp_getsockopt,
1155         .sendmsg                = dccp_sendmsg,
1156         .recvmsg                = dccp_recvmsg,
1157         .backlog_rcv            = dccp_v4_do_rcv,
1158         .hash                   = dccp_v4_hash,
1159         .unhash                 = dccp_unhash,
1160         .accept                 = inet_csk_accept,
1161         .get_port               = dccp_v4_get_port,
1162         .shutdown               = dccp_shutdown,
1163         .destroy                = dccp_v4_destroy_sock,
1164         .orphan_count           = &dccp_orphan_count,
1165         .max_header             = MAX_DCCP_HEADER,
1166         .obj_size               = sizeof(struct dccp_sock),
1167         .rsk_prot               = &dccp_request_sock_ops,
1168         .twsk_prot              = &dccp_timewait_sock_ops,
1169 };
1170
1171 EXPORT_SYMBOL_GPL(dccp_prot);