[NETNS]: Add namespace parameter to ip_route_output_flow.
[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, 1);
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_iss = secure_dccp_sequence_number(inet->saddr, inet->daddr,
117                                                     inet->sport, inet->dport);
118         inet->id = dp->dccps_iss ^ jiffies;
119
120         err = dccp_connect(sk);
121         rt = NULL;
122         if (err != 0)
123                 goto failure;
124 out:
125         return err;
126 failure:
127         /*
128          * This unhashes the socket and releases the local port, if necessary.
129          */
130         dccp_set_state(sk, DCCP_CLOSED);
131         ip_rt_put(rt);
132         sk->sk_route_caps = 0;
133         inet->dport = 0;
134         goto out;
135 }
136
137 EXPORT_SYMBOL_GPL(dccp_v4_connect);
138
139 /*
140  * This routine does path mtu discovery as defined in RFC1191.
141  */
142 static inline void dccp_do_pmtu_discovery(struct sock *sk,
143                                           const struct iphdr *iph,
144                                           u32 mtu)
145 {
146         struct dst_entry *dst;
147         const struct inet_sock *inet = inet_sk(sk);
148         const struct dccp_sock *dp = dccp_sk(sk);
149
150         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
151          * send out by Linux are always < 576bytes so they should go through
152          * unfragmented).
153          */
154         if (sk->sk_state == DCCP_LISTEN)
155                 return;
156
157         /* We don't check in the destentry if pmtu discovery is forbidden
158          * on this route. We just assume that no packet_to_big packets
159          * are send back when pmtu discovery is not active.
160          * There is a small race when the user changes this flag in the
161          * route, but I think that's acceptable.
162          */
163         if ((dst = __sk_dst_check(sk, 0)) == NULL)
164                 return;
165
166         dst->ops->update_pmtu(dst, mtu);
167
168         /* Something is about to be wrong... Remember soft error
169          * for the case, if this connection will not able to recover.
170          */
171         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
172                 sk->sk_err_soft = EMSGSIZE;
173
174         mtu = dst_mtu(dst);
175
176         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
177             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
178                 dccp_sync_mss(sk, mtu);
179
180                 /*
181                  * From RFC 4340, sec. 14.1:
182                  *
183                  *      DCCP-Sync packets are the best choice for upward
184                  *      probing, since DCCP-Sync probes do not risk application
185                  *      data loss.
186                  */
187                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
188         } /* else let the usual retransmit timer handle it */
189 }
190
191 /*
192  * This routine is called by the ICMP module when it gets some sort of error
193  * condition. If err < 0 then the socket should be closed and the error
194  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
195  * After adjustment header points to the first 8 bytes of the tcp header. We
196  * need to find the appropriate port.
197  *
198  * The locking strategy used here is very "optimistic". When someone else
199  * accesses the socket the ICMP is just dropped and for some paths there is no
200  * check at all. A more general error queue to queue errors for later handling
201  * is probably better.
202  */
203 static void dccp_v4_err(struct sk_buff *skb, u32 info)
204 {
205         const struct iphdr *iph = (struct iphdr *)skb->data;
206         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
207                                                         (iph->ihl << 2));
208         struct dccp_sock *dp;
209         struct inet_sock *inet;
210         const int type = icmp_hdr(skb)->type;
211         const int code = icmp_hdr(skb)->code;
212         struct sock *sk;
213         __u64 seq;
214         int err;
215
216         if (skb->len < (iph->ihl << 2) + 8) {
217                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
218                 return;
219         }
220
221         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
222                          iph->saddr, dh->dccph_sport, inet_iif(skb));
223         if (sk == NULL) {
224                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
225                 return;
226         }
227
228         if (sk->sk_state == DCCP_TIME_WAIT) {
229                 inet_twsk_put(inet_twsk(sk));
230                 return;
231         }
232
233         bh_lock_sock(sk);
234         /* If too many ICMPs get dropped on busy
235          * servers this needs to be solved differently.
236          */
237         if (sock_owned_by_user(sk))
238                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
239
240         if (sk->sk_state == DCCP_CLOSED)
241                 goto out;
242
243         dp = dccp_sk(sk);
244         seq = dccp_hdr_seq(dh);
245         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
246             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
247                 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
248                 goto out;
249         }
250
251         switch (type) {
252         case ICMP_SOURCE_QUENCH:
253                 /* Just silently ignore these. */
254                 goto out;
255         case ICMP_PARAMETERPROB:
256                 err = EPROTO;
257                 break;
258         case ICMP_DEST_UNREACH:
259                 if (code > NR_ICMP_UNREACH)
260                         goto out;
261
262                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
263                         if (!sock_owned_by_user(sk))
264                                 dccp_do_pmtu_discovery(sk, iph, info);
265                         goto out;
266                 }
267
268                 err = icmp_err_convert[code].errno;
269                 break;
270         case ICMP_TIME_EXCEEDED:
271                 err = EHOSTUNREACH;
272                 break;
273         default:
274                 goto out;
275         }
276
277         switch (sk->sk_state) {
278                 struct request_sock *req , **prev;
279         case DCCP_LISTEN:
280                 if (sock_owned_by_user(sk))
281                         goto out;
282                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
283                                           iph->daddr, iph->saddr);
284                 if (!req)
285                         goto out;
286
287                 /*
288                  * ICMPs are not backlogged, hence we cannot get an established
289                  * socket here.
290                  */
291                 BUG_TRAP(!req->sk);
292
293                 if (seq != dccp_rsk(req)->dreq_iss) {
294                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
295                         goto out;
296                 }
297                 /*
298                  * Still in RESPOND, just remove it silently.
299                  * There is no good way to pass the error to the newly
300                  * created socket, and POSIX does not want network
301                  * errors returned from accept().
302                  */
303                 inet_csk_reqsk_queue_drop(sk, req, prev);
304                 goto out;
305
306         case DCCP_REQUESTING:
307         case DCCP_RESPOND:
308                 if (!sock_owned_by_user(sk)) {
309                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
310                         sk->sk_err = err;
311
312                         sk->sk_error_report(sk);
313
314                         dccp_done(sk);
315                 } else
316                         sk->sk_err_soft = err;
317                 goto out;
318         }
319
320         /* If we've already connected we will keep trying
321          * until we time out, or the user gives up.
322          *
323          * rfc1122 4.2.3.9 allows to consider as hard errors
324          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
325          * but it is obsoleted by pmtu discovery).
326          *
327          * Note, that in modern internet, where routing is unreliable
328          * and in each dark corner broken firewalls sit, sending random
329          * errors ordered by their masters even this two messages finally lose
330          * their original sense (even Linux sends invalid PORT_UNREACHs)
331          *
332          * Now we are in compliance with RFCs.
333          *                                                      --ANK (980905)
334          */
335
336         inet = inet_sk(sk);
337         if (!sock_owned_by_user(sk) && inet->recverr) {
338                 sk->sk_err = err;
339                 sk->sk_error_report(sk);
340         } else /* Only an error on timeout */
341                 sk->sk_err_soft = err;
342 out:
343         bh_unlock_sock(sk);
344         sock_put(sk);
345 }
346
347 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
348                                       __be32 src, __be32 dst)
349 {
350         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
351 }
352
353 void dccp_v4_send_check(struct sock *sk, int unused, struct sk_buff *skb)
354 {
355         const struct inet_sock *inet = inet_sk(sk);
356         struct dccp_hdr *dh = dccp_hdr(skb);
357
358         dccp_csum_outgoing(skb);
359         dh->dccph_checksum = dccp_v4_csum_finish(skb, inet->saddr, inet->daddr);
360 }
361
362 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
363
364 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
365 {
366         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
367                                            ip_hdr(skb)->saddr,
368                                            dccp_hdr(skb)->dccph_dport,
369                                            dccp_hdr(skb)->dccph_sport);
370 }
371
372 /*
373  * The three way handshake has completed - we got a valid ACK or DATAACK -
374  * now create the new socket.
375  *
376  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
377  */
378 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
379                                        struct request_sock *req,
380                                        struct dst_entry *dst)
381 {
382         struct inet_request_sock *ireq;
383         struct inet_sock *newinet;
384         struct sock *newsk;
385
386         if (sk_acceptq_is_full(sk))
387                 goto exit_overflow;
388
389         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
390                 goto exit;
391
392         newsk = dccp_create_openreq_child(sk, req, skb);
393         if (newsk == NULL)
394                 goto exit;
395
396         sk_setup_caps(newsk, dst);
397
398         newinet            = inet_sk(newsk);
399         ireq               = inet_rsk(req);
400         newinet->daddr     = ireq->rmt_addr;
401         newinet->rcv_saddr = ireq->loc_addr;
402         newinet->saddr     = ireq->loc_addr;
403         newinet->opt       = ireq->opt;
404         ireq->opt          = NULL;
405         newinet->mc_index  = inet_iif(skb);
406         newinet->mc_ttl    = ip_hdr(skb)->ttl;
407         newinet->id        = jiffies;
408
409         dccp_sync_mss(newsk, dst_mtu(dst));
410
411         __inet_hash_nolisten(&dccp_hashinfo, newsk);
412         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
413
414         return newsk;
415
416 exit_overflow:
417         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
418 exit:
419         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
420         dst_release(dst);
421         return NULL;
422 }
423
424 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
425
426 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
427 {
428         const struct dccp_hdr *dh = dccp_hdr(skb);
429         const struct iphdr *iph = ip_hdr(skb);
430         struct sock *nsk;
431         struct request_sock **prev;
432         /* Find possible connection requests. */
433         struct request_sock *req = inet_csk_search_req(sk, &prev,
434                                                        dh->dccph_sport,
435                                                        iph->saddr, iph->daddr);
436         if (req != NULL)
437                 return dccp_check_req(sk, skb, req, prev);
438
439         nsk = inet_lookup_established(&dccp_hashinfo,
440                                       iph->saddr, dh->dccph_sport,
441                                       iph->daddr, dh->dccph_dport,
442                                       inet_iif(skb));
443         if (nsk != NULL) {
444                 if (nsk->sk_state != DCCP_TIME_WAIT) {
445                         bh_lock_sock(nsk);
446                         return nsk;
447                 }
448                 inet_twsk_put(inet_twsk(nsk));
449                 return NULL;
450         }
451
452         return sk;
453 }
454
455 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
456                                            struct sk_buff *skb)
457 {
458         struct rtable *rt;
459         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
460                             .nl_u = { .ip4_u =
461                                       { .daddr = ip_hdr(skb)->saddr,
462                                         .saddr = ip_hdr(skb)->daddr,
463                                         .tos = RT_CONN_FLAGS(sk) } },
464                             .proto = sk->sk_protocol,
465                             .uli_u = { .ports =
466                                        { .sport = dccp_hdr(skb)->dccph_dport,
467                                          .dport = dccp_hdr(skb)->dccph_sport }
468                                      }
469                           };
470
471         security_skb_classify_flow(skb, &fl);
472         if (ip_route_output_flow(&init_net, &rt, &fl, sk, 0)) {
473                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
474                 return NULL;
475         }
476
477         return &rt->u.dst;
478 }
479
480 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
481                                  struct dst_entry *dst)
482 {
483         int err = -1;
484         struct sk_buff *skb;
485
486         /* First, grab a route. */
487
488         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
489                 goto out;
490
491         skb = dccp_make_response(sk, dst, req);
492         if (skb != NULL) {
493                 const struct inet_request_sock *ireq = inet_rsk(req);
494                 struct dccp_hdr *dh = dccp_hdr(skb);
495
496                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
497                                                               ireq->rmt_addr);
498                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
499                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
500                                             ireq->rmt_addr,
501                                             ireq->opt);
502                 err = net_xmit_eval(err);
503         }
504
505 out:
506         dst_release(dst);
507         return err;
508 }
509
510 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
511 {
512         int err;
513         const struct iphdr *rxiph;
514         struct sk_buff *skb;
515         struct dst_entry *dst;
516
517         /* Never send a reset in response to a reset. */
518         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
519                 return;
520
521         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
522                 return;
523
524         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
525         if (dst == NULL)
526                 return;
527
528         skb = dccp_ctl_make_reset(dccp_v4_ctl_socket, rxskb);
529         if (skb == NULL)
530                 goto out;
531
532         rxiph = ip_hdr(rxskb);
533         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
534                                                                  rxiph->daddr);
535         skb->dst = dst_clone(dst);
536
537         bh_lock_sock(dccp_v4_ctl_socket->sk);
538         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
539                                     rxiph->daddr, rxiph->saddr, NULL);
540         bh_unlock_sock(dccp_v4_ctl_socket->sk);
541
542         if (net_xmit_eval(err) == 0) {
543                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
544                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
545         }
546 out:
547          dst_release(dst);
548 }
549
550 static void dccp_v4_reqsk_destructor(struct request_sock *req)
551 {
552         kfree(inet_rsk(req)->opt);
553 }
554
555 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
556         .family         = PF_INET,
557         .obj_size       = sizeof(struct dccp_request_sock),
558         .rtx_syn_ack    = dccp_v4_send_response,
559         .send_ack       = dccp_reqsk_send_ack,
560         .destructor     = dccp_v4_reqsk_destructor,
561         .send_reset     = dccp_v4_ctl_send_reset,
562 };
563
564 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
565 {
566         struct inet_request_sock *ireq;
567         struct request_sock *req;
568         struct dccp_request_sock *dreq;
569         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
570         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
571
572         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
573         if (((struct rtable *)skb->dst)->rt_flags &
574             (RTCF_BROADCAST | RTCF_MULTICAST))
575                 return 0;       /* discard, don't send a reset here */
576
577         if (dccp_bad_service_code(sk, service)) {
578                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
579                 goto drop;
580         }
581         /*
582          * TW buckets are converted to open requests without
583          * limitations, they conserve resources and peer is
584          * evidently real one.
585          */
586         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
587         if (inet_csk_reqsk_queue_is_full(sk))
588                 goto drop;
589
590         /*
591          * Accept backlog is full. If we have already queued enough
592          * of warm entries in syn queue, drop request. It is better than
593          * clogging syn queue with openreqs with exponentially increasing
594          * timeout.
595          */
596         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
597                 goto drop;
598
599         req = reqsk_alloc(&dccp_request_sock_ops);
600         if (req == NULL)
601                 goto drop;
602
603         dccp_reqsk_init(req, skb);
604
605         dreq = dccp_rsk(req);
606         if (dccp_parse_options(sk, dreq, skb))
607                 goto drop_and_free;
608
609         if (security_inet_conn_request(sk, skb, req))
610                 goto drop_and_free;
611
612         ireq = inet_rsk(req);
613         ireq->loc_addr = ip_hdr(skb)->daddr;
614         ireq->rmt_addr = ip_hdr(skb)->saddr;
615         ireq->opt       = NULL;
616
617         /*
618          * Step 3: Process LISTEN state
619          *
620          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
621          *
622          * In fact we defer setting S.GSR, S.SWL, S.SWH to
623          * dccp_create_openreq_child.
624          */
625         dreq->dreq_isr     = dcb->dccpd_seq;
626         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
627         dreq->dreq_service = service;
628
629         if (dccp_v4_send_response(sk, req, NULL))
630                 goto drop_and_free;
631
632         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
633         return 0;
634
635 drop_and_free:
636         reqsk_free(req);
637 drop:
638         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
639         return -1;
640 }
641
642 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
643
644 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
645 {
646         struct dccp_hdr *dh = dccp_hdr(skb);
647
648         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
649                 if (dccp_rcv_established(sk, skb, dh, skb->len))
650                         goto reset;
651                 return 0;
652         }
653
654         /*
655          *  Step 3: Process LISTEN state
656          *       If P.type == Request or P contains a valid Init Cookie option,
657          *            (* Must scan the packet's options to check for Init
658          *               Cookies.  Only Init Cookies are processed here,
659          *               however; other options are processed in Step 8.  This
660          *               scan need only be performed if the endpoint uses Init
661          *               Cookies *)
662          *            (* Generate a new socket and switch to that socket *)
663          *            Set S := new socket for this port pair
664          *            S.state = RESPOND
665          *            Choose S.ISS (initial seqno) or set from Init Cookies
666          *            Initialize S.GAR := S.ISS
667          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
668          *            Continue with S.state == RESPOND
669          *            (* A Response packet will be generated in Step 11 *)
670          *       Otherwise,
671          *            Generate Reset(No Connection) unless P.type == Reset
672          *            Drop packet and return
673          *
674          * NOTE: the check for the packet types is done in
675          *       dccp_rcv_state_process
676          */
677         if (sk->sk_state == DCCP_LISTEN) {
678                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
679
680                 if (nsk == NULL)
681                         goto discard;
682
683                 if (nsk != sk) {
684                         if (dccp_child_process(sk, nsk, skb))
685                                 goto reset;
686                         return 0;
687                 }
688         }
689
690         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
691                 goto reset;
692         return 0;
693
694 reset:
695         dccp_v4_ctl_send_reset(sk, skb);
696 discard:
697         kfree_skb(skb);
698         return 0;
699 }
700
701 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
702
703 /**
704  *      dccp_invalid_packet  -  check for malformed packets
705  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
706  *      Packets that fail these checks are ignored and do not receive Resets.
707  */
708 int dccp_invalid_packet(struct sk_buff *skb)
709 {
710         const struct dccp_hdr *dh;
711         unsigned int cscov;
712
713         if (skb->pkt_type != PACKET_HOST)
714                 return 1;
715
716         /* If the packet is shorter than 12 bytes, drop packet and return */
717         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
718                 DCCP_WARN("pskb_may_pull failed\n");
719                 return 1;
720         }
721
722         dh = dccp_hdr(skb);
723
724         /* If P.type is not understood, drop packet and return */
725         if (dh->dccph_type >= DCCP_PKT_INVALID) {
726                 DCCP_WARN("invalid packet type\n");
727                 return 1;
728         }
729
730         /*
731          * If P.Data Offset is too small for packet type, drop packet and return
732          */
733         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
734                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
735                 return 1;
736         }
737         /*
738          * If P.Data Offset is too too large for packet, drop packet and return
739          */
740         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
741                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
742                 return 1;
743         }
744
745         /*
746          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
747          * has short sequence numbers), drop packet and return
748          */
749         if (dh->dccph_type >= DCCP_PKT_DATA    &&
750             dh->dccph_type <= DCCP_PKT_DATAACK && dh->dccph_x == 0)  {
751                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
752                           dccp_packet_name(dh->dccph_type));
753                 return 1;
754         }
755
756         /*
757          * If P.CsCov is too large for the packet size, drop packet and return.
758          * This must come _before_ checksumming (not as RFC 4340 suggests).
759          */
760         cscov = dccp_csum_coverage(skb);
761         if (cscov > skb->len) {
762                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
763                           dh->dccph_cscov, skb->len);
764                 return 1;
765         }
766
767         /* If header checksum is incorrect, drop packet and return.
768          * (This step is completed in the AF-dependent functions.) */
769         skb->csum = skb_checksum(skb, 0, cscov, 0);
770
771         return 0;
772 }
773
774 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
775
776 /* this is called when real data arrives */
777 static int dccp_v4_rcv(struct sk_buff *skb)
778 {
779         const struct dccp_hdr *dh;
780         const struct iphdr *iph;
781         struct sock *sk;
782         int min_cov;
783
784         /* Step 1: Check header basics */
785
786         if (dccp_invalid_packet(skb))
787                 goto discard_it;
788
789         iph = ip_hdr(skb);
790         /* Step 1: If header checksum is incorrect, drop packet and return */
791         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
792                 DCCP_WARN("dropped packet with invalid checksum\n");
793                 goto discard_it;
794         }
795
796         dh = dccp_hdr(skb);
797
798         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
799         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
800
801         dccp_pr_debug("%8.8s "
802                       "src=%u.%u.%u.%u@%-5d "
803                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
804                       dccp_packet_name(dh->dccph_type),
805                       NIPQUAD(iph->saddr), ntohs(dh->dccph_sport),
806                       NIPQUAD(iph->daddr), ntohs(dh->dccph_dport),
807                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
808
809         if (dccp_packet_without_ack(skb)) {
810                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
811                 dccp_pr_debug_cat("\n");
812         } else {
813                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
814                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
815                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
816         }
817
818         /* Step 2:
819          *      Look up flow ID in table and get corresponding socket */
820         sk = __inet_lookup(&dccp_hashinfo,
821                            iph->saddr, dh->dccph_sport,
822                            iph->daddr, dh->dccph_dport, inet_iif(skb));
823         /*
824          * Step 2:
825          *      If no socket ...
826          */
827         if (sk == NULL) {
828                 dccp_pr_debug("failed to look up flow ID in table and "
829                               "get corresponding socket\n");
830                 goto no_dccp_socket;
831         }
832
833         /*
834          * Step 2:
835          *      ... or S.state == TIMEWAIT,
836          *              Generate Reset(No Connection) unless P.type == Reset
837          *              Drop packet and return
838          */
839         if (sk->sk_state == DCCP_TIME_WAIT) {
840                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
841                 inet_twsk_put(inet_twsk(sk));
842                 goto no_dccp_socket;
843         }
844
845         /*
846          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
847          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
848          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
849          */
850         min_cov = dccp_sk(sk)->dccps_pcrlen;
851         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
852                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
853                               dh->dccph_cscov, min_cov);
854                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
855                  *         options (Section 11.7) with Drop Code 0, Protocol
856                  *         Constraints."                                     */
857                 goto discard_and_relse;
858         }
859
860         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
861                 goto discard_and_relse;
862         nf_reset(skb);
863
864         return sk_receive_skb(sk, skb, 1);
865
866 no_dccp_socket:
867         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
868                 goto discard_it;
869         /*
870          * Step 2:
871          *      If no socket ...
872          *              Generate Reset(No Connection) unless P.type == Reset
873          *              Drop packet and return
874          */
875         if (dh->dccph_type != DCCP_PKT_RESET) {
876                 DCCP_SKB_CB(skb)->dccpd_reset_code =
877                                         DCCP_RESET_CODE_NO_CONNECTION;
878                 dccp_v4_ctl_send_reset(sk, skb);
879         }
880
881 discard_it:
882         kfree_skb(skb);
883         return 0;
884
885 discard_and_relse:
886         sock_put(sk);
887         goto discard_it;
888 }
889
890 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
891         .queue_xmit        = ip_queue_xmit,
892         .send_check        = dccp_v4_send_check,
893         .rebuild_header    = inet_sk_rebuild_header,
894         .conn_request      = dccp_v4_conn_request,
895         .syn_recv_sock     = dccp_v4_request_recv_sock,
896         .net_header_len    = sizeof(struct iphdr),
897         .setsockopt        = ip_setsockopt,
898         .getsockopt        = ip_getsockopt,
899         .addr2sockaddr     = inet_csk_addr2sockaddr,
900         .sockaddr_len      = sizeof(struct sockaddr_in),
901 #ifdef CONFIG_COMPAT
902         .compat_setsockopt = compat_ip_setsockopt,
903         .compat_getsockopt = compat_ip_getsockopt,
904 #endif
905 };
906
907 static int dccp_v4_init_sock(struct sock *sk)
908 {
909         static __u8 dccp_v4_ctl_sock_initialized;
910         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
911
912         if (err == 0) {
913                 if (unlikely(!dccp_v4_ctl_sock_initialized))
914                         dccp_v4_ctl_sock_initialized = 1;
915                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
916         }
917
918         return err;
919 }
920
921 static struct timewait_sock_ops dccp_timewait_sock_ops = {
922         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
923 };
924
925 DEFINE_PROTO_INUSE(dccp_v4)
926
927 static struct proto dccp_v4_prot = {
928         .name                   = "DCCP",
929         .owner                  = THIS_MODULE,
930         .close                  = dccp_close,
931         .connect                = dccp_v4_connect,
932         .disconnect             = dccp_disconnect,
933         .ioctl                  = dccp_ioctl,
934         .init                   = dccp_v4_init_sock,
935         .setsockopt             = dccp_setsockopt,
936         .getsockopt             = dccp_getsockopt,
937         .sendmsg                = dccp_sendmsg,
938         .recvmsg                = dccp_recvmsg,
939         .backlog_rcv            = dccp_v4_do_rcv,
940         .hash                   = dccp_hash,
941         .unhash                 = dccp_unhash,
942         .accept                 = inet_csk_accept,
943         .get_port               = dccp_v4_get_port,
944         .shutdown               = dccp_shutdown,
945         .destroy                = dccp_destroy_sock,
946         .orphan_count           = &dccp_orphan_count,
947         .max_header             = MAX_DCCP_HEADER,
948         .obj_size               = sizeof(struct dccp_sock),
949         .rsk_prot               = &dccp_request_sock_ops,
950         .twsk_prot              = &dccp_timewait_sock_ops,
951 #ifdef CONFIG_COMPAT
952         .compat_setsockopt      = compat_dccp_setsockopt,
953         .compat_getsockopt      = compat_dccp_getsockopt,
954 #endif
955         REF_PROTO_INUSE(dccp_v4)
956 };
957
958 static struct net_protocol dccp_v4_protocol = {
959         .handler        = dccp_v4_rcv,
960         .err_handler    = dccp_v4_err,
961         .no_policy      = 1,
962 };
963
964 static const struct proto_ops inet_dccp_ops = {
965         .family            = PF_INET,
966         .owner             = THIS_MODULE,
967         .release           = inet_release,
968         .bind              = inet_bind,
969         .connect           = inet_stream_connect,
970         .socketpair        = sock_no_socketpair,
971         .accept            = inet_accept,
972         .getname           = inet_getname,
973         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
974         .poll              = dccp_poll,
975         .ioctl             = inet_ioctl,
976         /* FIXME: work on inet_listen to rename it to sock_common_listen */
977         .listen            = inet_dccp_listen,
978         .shutdown          = inet_shutdown,
979         .setsockopt        = sock_common_setsockopt,
980         .getsockopt        = sock_common_getsockopt,
981         .sendmsg           = inet_sendmsg,
982         .recvmsg           = sock_common_recvmsg,
983         .mmap              = sock_no_mmap,
984         .sendpage          = sock_no_sendpage,
985 #ifdef CONFIG_COMPAT
986         .compat_setsockopt = compat_sock_common_setsockopt,
987         .compat_getsockopt = compat_sock_common_getsockopt,
988 #endif
989 };
990
991 static struct inet_protosw dccp_v4_protosw = {
992         .type           = SOCK_DCCP,
993         .protocol       = IPPROTO_DCCP,
994         .prot           = &dccp_v4_prot,
995         .ops            = &inet_dccp_ops,
996         .capability     = -1,
997         .no_check       = 0,
998         .flags          = INET_PROTOSW_ICSK,
999 };
1000
1001 static int __init dccp_v4_init(void)
1002 {
1003         int err = proto_register(&dccp_v4_prot, 1);
1004
1005         if (err != 0)
1006                 goto out;
1007
1008         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1009         if (err != 0)
1010                 goto out_proto_unregister;
1011
1012         inet_register_protosw(&dccp_v4_protosw);
1013
1014         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1015                                        SOCK_DCCP, IPPROTO_DCCP);
1016         if (err)
1017                 goto out_unregister_protosw;
1018 out:
1019         return err;
1020 out_unregister_protosw:
1021         inet_unregister_protosw(&dccp_v4_protosw);
1022         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1023 out_proto_unregister:
1024         proto_unregister(&dccp_v4_prot);
1025         goto out;
1026 }
1027
1028 static void __exit dccp_v4_exit(void)
1029 {
1030         inet_unregister_protosw(&dccp_v4_protosw);
1031         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1032         proto_unregister(&dccp_v4_prot);
1033 }
1034
1035 module_init(dccp_v4_init);
1036 module_exit(dccp_v4_exit);
1037
1038 /*
1039  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1040  * values directly, Also cover the case where the protocol is not specified,
1041  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1042  */
1043 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1044 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1045 MODULE_LICENSE("GPL");
1046 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1047 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");