tcp: account SYN-ACK timeouts & retransmissions
[linux-3.10.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  * The per-net dccp.v4_ctl_sk socket is 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
40 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
41 {
42         struct inet_sock *inet = inet_sk(sk);
43         struct dccp_sock *dp = dccp_sk(sk);
44         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
45         struct rtable *rt;
46         __be32 daddr, nexthop;
47         int tmp;
48         int err;
49
50         dp->dccps_role = DCCP_ROLE_CLIENT;
51
52         if (addr_len < sizeof(struct sockaddr_in))
53                 return -EINVAL;
54
55         if (usin->sin_family != AF_INET)
56                 return -EAFNOSUPPORT;
57
58         nexthop = daddr = usin->sin_addr.s_addr;
59         if (inet->opt != NULL && inet->opt->srr) {
60                 if (daddr == 0)
61                         return -EINVAL;
62                 nexthop = inet->opt->faddr;
63         }
64
65         tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr,
66                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
67                                IPPROTO_DCCP,
68                                inet->inet_sport, usin->sin_port, sk, 1);
69         if (tmp < 0)
70                 return tmp;
71
72         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
73                 ip_rt_put(rt);
74                 return -ENETUNREACH;
75         }
76
77         if (inet->opt == NULL || !inet->opt->srr)
78                 daddr = rt->rt_dst;
79
80         if (inet->inet_saddr == 0)
81                 inet->inet_saddr = rt->rt_src;
82         inet->inet_rcv_saddr = inet->inet_saddr;
83
84         inet->inet_dport = usin->sin_port;
85         inet->inet_daddr = daddr;
86
87         inet_csk(sk)->icsk_ext_hdr_len = 0;
88         if (inet->opt != NULL)
89                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
90         /*
91          * Socket identity is still unknown (sport may be zero).
92          * However we set state to DCCP_REQUESTING and not releasing socket
93          * lock select source port, enter ourselves into the hash tables and
94          * complete initialization after this.
95          */
96         dccp_set_state(sk, DCCP_REQUESTING);
97         err = inet_hash_connect(&dccp_death_row, sk);
98         if (err != 0)
99                 goto failure;
100
101         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->inet_sport,
102                                 inet->inet_dport, sk);
103         if (err != 0)
104                 goto failure;
105
106         /* OK, now commit destination to socket.  */
107         sk_setup_caps(sk, &rt->u.dst);
108
109         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
110                                                     inet->inet_daddr,
111                                                     inet->inet_sport,
112                                                     inet->inet_dport);
113         inet->inet_id = dp->dccps_iss ^ jiffies;
114
115         err = dccp_connect(sk);
116         rt = NULL;
117         if (err != 0)
118                 goto failure;
119 out:
120         return err;
121 failure:
122         /*
123          * This unhashes the socket and releases the local port, if necessary.
124          */
125         dccp_set_state(sk, DCCP_CLOSED);
126         ip_rt_put(rt);
127         sk->sk_route_caps = 0;
128         inet->inet_dport = 0;
129         goto out;
130 }
131
132 EXPORT_SYMBOL_GPL(dccp_v4_connect);
133
134 /*
135  * This routine does path mtu discovery as defined in RFC1191.
136  */
137 static inline void dccp_do_pmtu_discovery(struct sock *sk,
138                                           const struct iphdr *iph,
139                                           u32 mtu)
140 {
141         struct dst_entry *dst;
142         const struct inet_sock *inet = inet_sk(sk);
143         const struct dccp_sock *dp = dccp_sk(sk);
144
145         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
146          * send out by Linux are always < 576bytes so they should go through
147          * unfragmented).
148          */
149         if (sk->sk_state == DCCP_LISTEN)
150                 return;
151
152         /* We don't check in the destentry if pmtu discovery is forbidden
153          * on this route. We just assume that no packet_to_big packets
154          * are send back when pmtu discovery is not active.
155          * There is a small race when the user changes this flag in the
156          * route, but I think that's acceptable.
157          */
158         if ((dst = __sk_dst_check(sk, 0)) == NULL)
159                 return;
160
161         dst->ops->update_pmtu(dst, mtu);
162
163         /* Something is about to be wrong... Remember soft error
164          * for the case, if this connection will not able to recover.
165          */
166         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
167                 sk->sk_err_soft = EMSGSIZE;
168
169         mtu = dst_mtu(dst);
170
171         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
172             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
173                 dccp_sync_mss(sk, mtu);
174
175                 /*
176                  * From RFC 4340, sec. 14.1:
177                  *
178                  *      DCCP-Sync packets are the best choice for upward
179                  *      probing, since DCCP-Sync probes do not risk application
180                  *      data loss.
181                  */
182                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
183         } /* else let the usual retransmit timer handle it */
184 }
185
186 /*
187  * This routine is called by the ICMP module when it gets some sort of error
188  * condition. If err < 0 then the socket should be closed and the error
189  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
190  * After adjustment header points to the first 8 bytes of the tcp header. We
191  * need to find the appropriate port.
192  *
193  * The locking strategy used here is very "optimistic". When someone else
194  * accesses the socket the ICMP is just dropped and for some paths there is no
195  * check at all. A more general error queue to queue errors for later handling
196  * is probably better.
197  */
198 static void dccp_v4_err(struct sk_buff *skb, u32 info)
199 {
200         const struct iphdr *iph = (struct iphdr *)skb->data;
201         const u8 offset = iph->ihl << 2;
202         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
203         struct dccp_sock *dp;
204         struct inet_sock *inet;
205         const int type = icmp_hdr(skb)->type;
206         const int code = icmp_hdr(skb)->code;
207         struct sock *sk;
208         __u64 seq;
209         int err;
210         struct net *net = dev_net(skb->dev);
211
212         if (skb->len < offset + sizeof(*dh) ||
213             skb->len < offset + __dccp_basic_hdr_len(dh)) {
214                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
215                 return;
216         }
217
218         sk = inet_lookup(net, &dccp_hashinfo,
219                         iph->daddr, dh->dccph_dport,
220                         iph->saddr, dh->dccph_sport, inet_iif(skb));
221         if (sk == NULL) {
222                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
223                 return;
224         }
225
226         if (sk->sk_state == DCCP_TIME_WAIT) {
227                 inet_twsk_put(inet_twsk(sk));
228                 return;
229         }
230
231         bh_lock_sock(sk);
232         /* If too many ICMPs get dropped on busy
233          * servers this needs to be solved differently.
234          */
235         if (sock_owned_by_user(sk))
236                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
237
238         if (sk->sk_state == DCCP_CLOSED)
239                 goto out;
240
241         dp = dccp_sk(sk);
242         seq = dccp_hdr_seq(dh);
243         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
244             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
245                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
246                 goto out;
247         }
248
249         switch (type) {
250         case ICMP_SOURCE_QUENCH:
251                 /* Just silently ignore these. */
252                 goto out;
253         case ICMP_PARAMETERPROB:
254                 err = EPROTO;
255                 break;
256         case ICMP_DEST_UNREACH:
257                 if (code > NR_ICMP_UNREACH)
258                         goto out;
259
260                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
261                         if (!sock_owned_by_user(sk))
262                                 dccp_do_pmtu_discovery(sk, iph, info);
263                         goto out;
264                 }
265
266                 err = icmp_err_convert[code].errno;
267                 break;
268         case ICMP_TIME_EXCEEDED:
269                 err = EHOSTUNREACH;
270                 break;
271         default:
272                 goto out;
273         }
274
275         switch (sk->sk_state) {
276                 struct request_sock *req , **prev;
277         case DCCP_LISTEN:
278                 if (sock_owned_by_user(sk))
279                         goto out;
280                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
281                                           iph->daddr, iph->saddr);
282                 if (!req)
283                         goto out;
284
285                 /*
286                  * ICMPs are not backlogged, hence we cannot get an established
287                  * socket here.
288                  */
289                 WARN_ON(req->sk);
290
291                 if (seq != dccp_rsk(req)->dreq_iss) {
292                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
293                         goto out;
294                 }
295                 /*
296                  * Still in RESPOND, just remove it silently.
297                  * There is no good way to pass the error to the newly
298                  * created socket, and POSIX does not want network
299                  * errors returned from accept().
300                  */
301                 inet_csk_reqsk_queue_drop(sk, req, prev);
302                 goto out;
303
304         case DCCP_REQUESTING:
305         case DCCP_RESPOND:
306                 if (!sock_owned_by_user(sk)) {
307                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
308                         sk->sk_err = err;
309
310                         sk->sk_error_report(sk);
311
312                         dccp_done(sk);
313                 } else
314                         sk->sk_err_soft = err;
315                 goto out;
316         }
317
318         /* If we've already connected we will keep trying
319          * until we time out, or the user gives up.
320          *
321          * rfc1122 4.2.3.9 allows to consider as hard errors
322          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
323          * but it is obsoleted by pmtu discovery).
324          *
325          * Note, that in modern internet, where routing is unreliable
326          * and in each dark corner broken firewalls sit, sending random
327          * errors ordered by their masters even this two messages finally lose
328          * their original sense (even Linux sends invalid PORT_UNREACHs)
329          *
330          * Now we are in compliance with RFCs.
331          *                                                      --ANK (980905)
332          */
333
334         inet = inet_sk(sk);
335         if (!sock_owned_by_user(sk) && inet->recverr) {
336                 sk->sk_err = err;
337                 sk->sk_error_report(sk);
338         } else /* Only an error on timeout */
339                 sk->sk_err_soft = err;
340 out:
341         bh_unlock_sock(sk);
342         sock_put(sk);
343 }
344
345 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
346                                       __be32 src, __be32 dst)
347 {
348         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
349 }
350
351 void dccp_v4_send_check(struct sock *sk, int unused, struct sk_buff *skb)
352 {
353         const struct inet_sock *inet = inet_sk(sk);
354         struct dccp_hdr *dh = dccp_hdr(skb);
355
356         dccp_csum_outgoing(skb);
357         dh->dccph_checksum = dccp_v4_csum_finish(skb,
358                                                  inet->inet_saddr,
359                                                  inet->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->inet_daddr     = ireq->rmt_addr;
401         newinet->inet_rcv_saddr = ireq->loc_addr;
402         newinet->inet_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->inet_id   = jiffies;
408
409         dccp_sync_mss(newsk, dst_mtu(dst));
410
411         __inet_hash_nolisten(newsk, NULL);
412         __inet_inherit_port(sk, newsk);
413
414         return newsk;
415
416 exit_overflow:
417         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
418 exit:
419         NET_INC_STATS_BH(sock_net(sk), 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(sock_net(sk), &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 net *net, struct sock *sk,
456                                            struct sk_buff *skb)
457 {
458         struct rtable *rt;
459         struct flowi fl = { .oif = skb_rtable(skb)->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(net, &rt, &fl, sk, 0)) {
473                 IP_INC_STATS_BH(net, 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 request_values *rv_unused)
482 {
483         int err = -1;
484         struct sk_buff *skb;
485         struct dst_entry *dst;
486
487         dst = inet_csk_route_req(sk, req);
488         if (dst == 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                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
499                                             ireq->rmt_addr,
500                                             ireq->opt);
501                 err = net_xmit_eval(err);
502         }
503
504 out:
505         dst_release(dst);
506         return err;
507 }
508
509 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
510 {
511         int err;
512         const struct iphdr *rxiph;
513         struct sk_buff *skb;
514         struct dst_entry *dst;
515         struct net *net = dev_net(skb_dst(rxskb)->dev);
516         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
517
518         /* Never send a reset in response to a reset. */
519         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
520                 return;
521
522         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
523                 return;
524
525         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
526         if (dst == NULL)
527                 return;
528
529         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
530         if (skb == NULL)
531                 goto out;
532
533         rxiph = ip_hdr(rxskb);
534         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
535                                                                  rxiph->daddr);
536         skb_dst_set(skb, dst_clone(dst));
537
538         bh_lock_sock(ctl_sk);
539         err = ip_build_and_send_pkt(skb, ctl_sk,
540                                     rxiph->daddr, rxiph->saddr, NULL);
541         bh_unlock_sock(ctl_sk);
542
543         if (net_xmit_eval(err) == 0) {
544                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
545                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
546         }
547 out:
548          dst_release(dst);
549 }
550
551 static void dccp_v4_reqsk_destructor(struct request_sock *req)
552 {
553         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
554         kfree(inet_rsk(req)->opt);
555 }
556
557 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
558         .family         = PF_INET,
559         .obj_size       = sizeof(struct dccp_request_sock),
560         .rtx_syn_ack    = dccp_v4_send_response,
561         .send_ack       = dccp_reqsk_send_ack,
562         .destructor     = dccp_v4_reqsk_destructor,
563         .send_reset     = dccp_v4_ctl_send_reset,
564 };
565
566 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
567 {
568         struct inet_request_sock *ireq;
569         struct request_sock *req;
570         struct dccp_request_sock *dreq;
571         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
572         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
573
574         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
575         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
576                 return 0;       /* discard, don't send a reset here */
577
578         if (dccp_bad_service_code(sk, service)) {
579                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
580                 goto drop;
581         }
582         /*
583          * TW buckets are converted to open requests without
584          * limitations, they conserve resources and peer is
585          * evidently real one.
586          */
587         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
588         if (inet_csk_reqsk_queue_is_full(sk))
589                 goto drop;
590
591         /*
592          * Accept backlog is full. If we have already queued enough
593          * of warm entries in syn queue, drop request. It is better than
594          * clogging syn queue with openreqs with exponentially increasing
595          * timeout.
596          */
597         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
598                 goto drop;
599
600         req = inet_reqsk_alloc(&dccp_request_sock_ops);
601         if (req == NULL)
602                 goto drop;
603
604         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
605                 goto drop_and_free;
606
607         dreq = dccp_rsk(req);
608         if (dccp_parse_options(sk, dreq, skb))
609                 goto drop_and_free;
610
611         if (security_inet_conn_request(sk, skb, req))
612                 goto drop_and_free;
613
614         ireq = inet_rsk(req);
615         ireq->loc_addr = ip_hdr(skb)->daddr;
616         ireq->rmt_addr = ip_hdr(skb)->saddr;
617
618         /*
619          * Step 3: Process LISTEN state
620          *
621          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
622          *
623          * In fact we defer setting S.GSR, S.SWL, S.SWH to
624          * dccp_create_openreq_child.
625          */
626         dreq->dreq_isr     = dcb->dccpd_seq;
627         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
628         dreq->dreq_service = service;
629
630         if (dccp_v4_send_response(sk, req, NULL))
631                 goto drop_and_free;
632
633         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
634         return 0;
635
636 drop_and_free:
637         reqsk_free(req);
638 drop:
639         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
640         return -1;
641 }
642
643 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
644
645 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
646 {
647         struct dccp_hdr *dh = dccp_hdr(skb);
648
649         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
650                 if (dccp_rcv_established(sk, skb, dh, skb->len))
651                         goto reset;
652                 return 0;
653         }
654
655         /*
656          *  Step 3: Process LISTEN state
657          *       If P.type == Request or P contains a valid Init Cookie option,
658          *            (* Must scan the packet's options to check for Init
659          *               Cookies.  Only Init Cookies are processed here,
660          *               however; other options are processed in Step 8.  This
661          *               scan need only be performed if the endpoint uses Init
662          *               Cookies *)
663          *            (* Generate a new socket and switch to that socket *)
664          *            Set S := new socket for this port pair
665          *            S.state = RESPOND
666          *            Choose S.ISS (initial seqno) or set from Init Cookies
667          *            Initialize S.GAR := S.ISS
668          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
669          *            Continue with S.state == RESPOND
670          *            (* A Response packet will be generated in Step 11 *)
671          *       Otherwise,
672          *            Generate Reset(No Connection) unless P.type == Reset
673          *            Drop packet and return
674          *
675          * NOTE: the check for the packet types is done in
676          *       dccp_rcv_state_process
677          */
678         if (sk->sk_state == DCCP_LISTEN) {
679                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
680
681                 if (nsk == NULL)
682                         goto discard;
683
684                 if (nsk != sk) {
685                         if (dccp_child_process(sk, nsk, skb))
686                                 goto reset;
687                         return 0;
688                 }
689         }
690
691         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
692                 goto reset;
693         return 0;
694
695 reset:
696         dccp_v4_ctl_send_reset(sk, skb);
697 discard:
698         kfree_skb(skb);
699         return 0;
700 }
701
702 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
703
704 /**
705  *      dccp_invalid_packet  -  check for malformed packets
706  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
707  *      Packets that fail these checks are ignored and do not receive Resets.
708  */
709 int dccp_invalid_packet(struct sk_buff *skb)
710 {
711         const struct dccp_hdr *dh;
712         unsigned int cscov;
713
714         if (skb->pkt_type != PACKET_HOST)
715                 return 1;
716
717         /* If the packet is shorter than 12 bytes, drop packet and return */
718         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
719                 DCCP_WARN("pskb_may_pull failed\n");
720                 return 1;
721         }
722
723         dh = dccp_hdr(skb);
724
725         /* If P.type is not understood, drop packet and return */
726         if (dh->dccph_type >= DCCP_PKT_INVALID) {
727                 DCCP_WARN("invalid packet type\n");
728                 return 1;
729         }
730
731         /*
732          * If P.Data Offset is too small for packet type, drop packet and return
733          */
734         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
735                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
736                 return 1;
737         }
738         /*
739          * If P.Data Offset is too too large for packet, drop packet and return
740          */
741         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
742                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
743                 return 1;
744         }
745
746         /*
747          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
748          * has short sequence numbers), drop packet and return
749          */
750         if ((dh->dccph_type < DCCP_PKT_DATA    ||
751             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
752                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
753                           dccp_packet_name(dh->dccph_type));
754                 return 1;
755         }
756
757         /*
758          * If P.CsCov is too large for the packet size, drop packet and return.
759          * This must come _before_ checksumming (not as RFC 4340 suggests).
760          */
761         cscov = dccp_csum_coverage(skb);
762         if (cscov > skb->len) {
763                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
764                           dh->dccph_cscov, skb->len);
765                 return 1;
766         }
767
768         /* If header checksum is incorrect, drop packet and return.
769          * (This step is completed in the AF-dependent functions.) */
770         skb->csum = skb_checksum(skb, 0, cscov, 0);
771
772         return 0;
773 }
774
775 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
776
777 /* this is called when real data arrives */
778 static int dccp_v4_rcv(struct sk_buff *skb)
779 {
780         const struct dccp_hdr *dh;
781         const struct iphdr *iph;
782         struct sock *sk;
783         int min_cov;
784
785         /* Step 1: Check header basics */
786
787         if (dccp_invalid_packet(skb))
788                 goto discard_it;
789
790         iph = ip_hdr(skb);
791         /* Step 1: If header checksum is incorrect, drop packet and return */
792         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
793                 DCCP_WARN("dropped packet with invalid checksum\n");
794                 goto discard_it;
795         }
796
797         dh = dccp_hdr(skb);
798
799         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
800         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
801
802         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
803                       dccp_packet_name(dh->dccph_type),
804                       &iph->saddr, ntohs(dh->dccph_sport),
805                       &iph->daddr, ntohs(dh->dccph_dport),
806                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
807
808         if (dccp_packet_without_ack(skb)) {
809                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
810                 dccp_pr_debug_cat("\n");
811         } else {
812                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
813                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
814                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
815         }
816
817         /* Step 2:
818          *      Look up flow ID in table and get corresponding socket */
819         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
820                                dh->dccph_sport, dh->dccph_dport);
821         /*
822          * Step 2:
823          *      If no socket ...
824          */
825         if (sk == NULL) {
826                 dccp_pr_debug("failed to look up flow ID in table and "
827                               "get corresponding socket\n");
828                 goto no_dccp_socket;
829         }
830
831         /*
832          * Step 2:
833          *      ... or S.state == TIMEWAIT,
834          *              Generate Reset(No Connection) unless P.type == Reset
835          *              Drop packet and return
836          */
837         if (sk->sk_state == DCCP_TIME_WAIT) {
838                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
839                 inet_twsk_put(inet_twsk(sk));
840                 goto no_dccp_socket;
841         }
842
843         /*
844          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
845          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
846          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
847          */
848         min_cov = dccp_sk(sk)->dccps_pcrlen;
849         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
850                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
851                               dh->dccph_cscov, min_cov);
852                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
853                  *         options (Section 11.7) with Drop Code 0, Protocol
854                  *         Constraints."                                     */
855                 goto discard_and_relse;
856         }
857
858         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
859                 goto discard_and_relse;
860         nf_reset(skb);
861
862         return sk_receive_skb(sk, skb, 1);
863
864 no_dccp_socket:
865         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
866                 goto discard_it;
867         /*
868          * Step 2:
869          *      If no socket ...
870          *              Generate Reset(No Connection) unless P.type == Reset
871          *              Drop packet and return
872          */
873         if (dh->dccph_type != DCCP_PKT_RESET) {
874                 DCCP_SKB_CB(skb)->dccpd_reset_code =
875                                         DCCP_RESET_CODE_NO_CONNECTION;
876                 dccp_v4_ctl_send_reset(sk, skb);
877         }
878
879 discard_it:
880         kfree_skb(skb);
881         return 0;
882
883 discard_and_relse:
884         sock_put(sk);
885         goto discard_it;
886 }
887
888 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
889         .queue_xmit        = ip_queue_xmit,
890         .send_check        = dccp_v4_send_check,
891         .rebuild_header    = inet_sk_rebuild_header,
892         .conn_request      = dccp_v4_conn_request,
893         .syn_recv_sock     = dccp_v4_request_recv_sock,
894         .net_header_len    = sizeof(struct iphdr),
895         .setsockopt        = ip_setsockopt,
896         .getsockopt        = ip_getsockopt,
897         .addr2sockaddr     = inet_csk_addr2sockaddr,
898         .sockaddr_len      = sizeof(struct sockaddr_in),
899         .bind_conflict     = inet_csk_bind_conflict,
900 #ifdef CONFIG_COMPAT
901         .compat_setsockopt = compat_ip_setsockopt,
902         .compat_getsockopt = compat_ip_getsockopt,
903 #endif
904 };
905
906 static int dccp_v4_init_sock(struct sock *sk)
907 {
908         static __u8 dccp_v4_ctl_sock_initialized;
909         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
910
911         if (err == 0) {
912                 if (unlikely(!dccp_v4_ctl_sock_initialized))
913                         dccp_v4_ctl_sock_initialized = 1;
914                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
915         }
916
917         return err;
918 }
919
920 static struct timewait_sock_ops dccp_timewait_sock_ops = {
921         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
922 };
923
924 static struct proto dccp_v4_prot = {
925         .name                   = "DCCP",
926         .owner                  = THIS_MODULE,
927         .close                  = dccp_close,
928         .connect                = dccp_v4_connect,
929         .disconnect             = dccp_disconnect,
930         .ioctl                  = dccp_ioctl,
931         .init                   = dccp_v4_init_sock,
932         .setsockopt             = dccp_setsockopt,
933         .getsockopt             = dccp_getsockopt,
934         .sendmsg                = dccp_sendmsg,
935         .recvmsg                = dccp_recvmsg,
936         .backlog_rcv            = dccp_v4_do_rcv,
937         .hash                   = inet_hash,
938         .unhash                 = inet_unhash,
939         .accept                 = inet_csk_accept,
940         .get_port               = inet_csk_get_port,
941         .shutdown               = dccp_shutdown,
942         .destroy                = dccp_destroy_sock,
943         .orphan_count           = &dccp_orphan_count,
944         .max_header             = MAX_DCCP_HEADER,
945         .obj_size               = sizeof(struct dccp_sock),
946         .slab_flags             = SLAB_DESTROY_BY_RCU,
947         .rsk_prot               = &dccp_request_sock_ops,
948         .twsk_prot              = &dccp_timewait_sock_ops,
949         .h.hashinfo             = &dccp_hashinfo,
950 #ifdef CONFIG_COMPAT
951         .compat_setsockopt      = compat_dccp_setsockopt,
952         .compat_getsockopt      = compat_dccp_getsockopt,
953 #endif
954 };
955
956 static const struct net_protocol dccp_v4_protocol = {
957         .handler        = dccp_v4_rcv,
958         .err_handler    = dccp_v4_err,
959         .no_policy      = 1,
960         .netns_ok       = 1,
961 };
962
963 static const struct proto_ops inet_dccp_ops = {
964         .family            = PF_INET,
965         .owner             = THIS_MODULE,
966         .release           = inet_release,
967         .bind              = inet_bind,
968         .connect           = inet_stream_connect,
969         .socketpair        = sock_no_socketpair,
970         .accept            = inet_accept,
971         .getname           = inet_getname,
972         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
973         .poll              = dccp_poll,
974         .ioctl             = inet_ioctl,
975         /* FIXME: work on inet_listen to rename it to sock_common_listen */
976         .listen            = inet_dccp_listen,
977         .shutdown          = inet_shutdown,
978         .setsockopt        = sock_common_setsockopt,
979         .getsockopt        = sock_common_getsockopt,
980         .sendmsg           = inet_sendmsg,
981         .recvmsg           = sock_common_recvmsg,
982         .mmap              = sock_no_mmap,
983         .sendpage          = sock_no_sendpage,
984 #ifdef CONFIG_COMPAT
985         .compat_setsockopt = compat_sock_common_setsockopt,
986         .compat_getsockopt = compat_sock_common_getsockopt,
987 #endif
988 };
989
990 static struct inet_protosw dccp_v4_protosw = {
991         .type           = SOCK_DCCP,
992         .protocol       = IPPROTO_DCCP,
993         .prot           = &dccp_v4_prot,
994         .ops            = &inet_dccp_ops,
995         .no_check       = 0,
996         .flags          = INET_PROTOSW_ICSK,
997 };
998
999 static int dccp_v4_init_net(struct net *net)
1000 {
1001         int err;
1002
1003         err = inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1004                                    SOCK_DCCP, IPPROTO_DCCP, net);
1005         return err;
1006 }
1007
1008 static void dccp_v4_exit_net(struct net *net)
1009 {
1010         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1011 }
1012
1013 static struct pernet_operations dccp_v4_ops = {
1014         .init   = dccp_v4_init_net,
1015         .exit   = dccp_v4_exit_net,
1016 };
1017
1018 static int __init dccp_v4_init(void)
1019 {
1020         int err = proto_register(&dccp_v4_prot, 1);
1021
1022         if (err != 0)
1023                 goto out;
1024
1025         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1026         if (err != 0)
1027                 goto out_proto_unregister;
1028
1029         inet_register_protosw(&dccp_v4_protosw);
1030
1031         err = register_pernet_subsys(&dccp_v4_ops);
1032         if (err)
1033                 goto out_destroy_ctl_sock;
1034 out:
1035         return err;
1036 out_destroy_ctl_sock:
1037         inet_unregister_protosw(&dccp_v4_protosw);
1038         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1039 out_proto_unregister:
1040         proto_unregister(&dccp_v4_prot);
1041         goto out;
1042 }
1043
1044 static void __exit dccp_v4_exit(void)
1045 {
1046         unregister_pernet_subsys(&dccp_v4_ops);
1047         inet_unregister_protosw(&dccp_v4_protosw);
1048         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1049         proto_unregister(&dccp_v4_prot);
1050 }
1051
1052 module_init(dccp_v4_init);
1053 module_exit(dccp_v4_exit);
1054
1055 /*
1056  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1057  * values directly, Also cover the case where the protocol is not specified,
1058  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1059  */
1060 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1061 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1062 MODULE_LICENSE("GPL");
1063 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1064 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");