GNU Linux-libre 4.14.266-gnu1
[releases.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/slab.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_common.h>
22 #include <net/inet_hashtables.h>
23 #include <net/inet_sock.h>
24 #include <net/protocol.h>
25 #include <net/sock.h>
26 #include <net/timewait_sock.h>
27 #include <net/tcp_states.h>
28 #include <net/xfrm.h>
29 #include <net/secure_seq.h>
30
31 #include "ackvec.h"
32 #include "ccid.h"
33 #include "dccp.h"
34 #include "feat.h"
35
36 /*
37  * The per-net dccp.v4_ctl_sk socket is used for responding to
38  * the Out-of-the-blue (OOTB) packets. A control sock will be created
39  * for this socket at the initialization time.
40  */
41
42 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
43 {
44         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
45         struct inet_sock *inet = inet_sk(sk);
46         struct dccp_sock *dp = dccp_sk(sk);
47         __be16 orig_sport, orig_dport;
48         __be32 daddr, nexthop;
49         struct flowi4 *fl4;
50         struct rtable *rt;
51         int err;
52         struct ip_options_rcu *inet_opt;
53
54         dp->dccps_role = DCCP_ROLE_CLIENT;
55
56         if (addr_len < sizeof(struct sockaddr_in))
57                 return -EINVAL;
58
59         if (usin->sin_family != AF_INET)
60                 return -EAFNOSUPPORT;
61
62         nexthop = daddr = usin->sin_addr.s_addr;
63
64         inet_opt = rcu_dereference_protected(inet->inet_opt,
65                                              lockdep_sock_is_held(sk));
66         if (inet_opt != NULL && inet_opt->opt.srr) {
67                 if (daddr == 0)
68                         return -EINVAL;
69                 nexthop = inet_opt->opt.faddr;
70         }
71
72         orig_sport = inet->inet_sport;
73         orig_dport = usin->sin_port;
74         fl4 = &inet->cork.fl.u.ip4;
75         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
76                               RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
77                               IPPROTO_DCCP,
78                               orig_sport, orig_dport, sk);
79         if (IS_ERR(rt))
80                 return PTR_ERR(rt);
81
82         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
83                 ip_rt_put(rt);
84                 return -ENETUNREACH;
85         }
86
87         if (inet_opt == NULL || !inet_opt->opt.srr)
88                 daddr = fl4->daddr;
89
90         if (inet->inet_saddr == 0)
91                 inet->inet_saddr = fl4->saddr;
92         sk_rcv_saddr_set(sk, inet->inet_saddr);
93         inet->inet_dport = usin->sin_port;
94         sk_daddr_set(sk, daddr);
95
96         inet_csk(sk)->icsk_ext_hdr_len = 0;
97         if (inet_opt)
98                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
99         /*
100          * Socket identity is still unknown (sport may be zero).
101          * However we set state to DCCP_REQUESTING and not releasing socket
102          * lock select source port, enter ourselves into the hash tables and
103          * complete initialization after this.
104          */
105         dccp_set_state(sk, DCCP_REQUESTING);
106         err = inet_hash_connect(&dccp_death_row, sk);
107         if (err != 0)
108                 goto failure;
109
110         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
111                                inet->inet_sport, inet->inet_dport, sk);
112         if (IS_ERR(rt)) {
113                 err = PTR_ERR(rt);
114                 rt = NULL;
115                 goto failure;
116         }
117         /* OK, now commit destination to socket.  */
118         sk_setup_caps(sk, &rt->dst);
119
120         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
121                                                     inet->inet_daddr,
122                                                     inet->inet_sport,
123                                                     inet->inet_dport);
124         inet->inet_id = prandom_u32();
125
126         err = dccp_connect(sk);
127         rt = NULL;
128         if (err != 0)
129                 goto failure;
130 out:
131         return err;
132 failure:
133         /*
134          * This unhashes the socket and releases the local port, if necessary.
135          */
136         dccp_set_state(sk, DCCP_CLOSED);
137         ip_rt_put(rt);
138         sk->sk_route_caps = 0;
139         inet->inet_dport = 0;
140         goto out;
141 }
142 EXPORT_SYMBOL_GPL(dccp_v4_connect);
143
144 /*
145  * This routine does path mtu discovery as defined in RFC1191.
146  */
147 static inline void dccp_do_pmtu_discovery(struct sock *sk,
148                                           const struct iphdr *iph,
149                                           u32 mtu)
150 {
151         struct dst_entry *dst;
152         const struct inet_sock *inet = inet_sk(sk);
153         const struct dccp_sock *dp = dccp_sk(sk);
154
155         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
156          * send out by Linux are always < 576bytes so they should go through
157          * unfragmented).
158          */
159         if (sk->sk_state == DCCP_LISTEN)
160                 return;
161
162         dst = inet_csk_update_pmtu(sk, mtu);
163         if (!dst)
164                 return;
165
166         /* Something is about to be wrong... Remember soft error
167          * for the case, if this connection will not able to recover.
168          */
169         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
170                 sk->sk_err_soft = EMSGSIZE;
171
172         mtu = dst_mtu(dst);
173
174         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
175             ip_sk_accept_pmtu(sk) &&
176             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
177                 dccp_sync_mss(sk, mtu);
178
179                 /*
180                  * From RFC 4340, sec. 14.1:
181                  *
182                  *      DCCP-Sync packets are the best choice for upward
183                  *      probing, since DCCP-Sync probes do not risk application
184                  *      data loss.
185                  */
186                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
187         } /* else let the usual retransmit timer handle it */
188 }
189
190 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
191 {
192         struct dst_entry *dst = __sk_dst_check(sk, 0);
193
194         if (dst)
195                 dst->ops->redirect(dst, sk, skb);
196 }
197
198 void dccp_req_err(struct sock *sk, u64 seq)
199         {
200         struct request_sock *req = inet_reqsk(sk);
201         struct net *net = sock_net(sk);
202
203         /*
204          * ICMPs are not backlogged, hence we cannot get an established
205          * socket here.
206          */
207         if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
208                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
209         } else {
210                 /*
211                  * Still in RESPOND, just remove it silently.
212                  * There is no good way to pass the error to the newly
213                  * created socket, and POSIX does not want network
214                  * errors returned from accept().
215                  */
216                 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
217         }
218         reqsk_put(req);
219 }
220 EXPORT_SYMBOL(dccp_req_err);
221
222 /*
223  * This routine is called by the ICMP module when it gets some sort of error
224  * condition. If err < 0 then the socket should be closed and the error
225  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
226  * After adjustment header points to the first 8 bytes of the tcp header. We
227  * need to find the appropriate port.
228  *
229  * The locking strategy used here is very "optimistic". When someone else
230  * accesses the socket the ICMP is just dropped and for some paths there is no
231  * check at all. A more general error queue to queue errors for later handling
232  * is probably better.
233  */
234 static void dccp_v4_err(struct sk_buff *skb, u32 info)
235 {
236         const struct iphdr *iph = (struct iphdr *)skb->data;
237         const u8 offset = iph->ihl << 2;
238         const struct dccp_hdr *dh;
239         struct dccp_sock *dp;
240         struct inet_sock *inet;
241         const int type = icmp_hdr(skb)->type;
242         const int code = icmp_hdr(skb)->code;
243         struct sock *sk;
244         __u64 seq;
245         int err;
246         struct net *net = dev_net(skb->dev);
247
248         /* Only need dccph_dport & dccph_sport which are the first
249          * 4 bytes in dccp header.
250          * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
251          */
252         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
253         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
254         dh = (struct dccp_hdr *)(skb->data + offset);
255
256         sk = __inet_lookup_established(net, &dccp_hashinfo,
257                                        iph->daddr, dh->dccph_dport,
258                                        iph->saddr, ntohs(dh->dccph_sport),
259                                        inet_iif(skb), 0);
260         if (!sk) {
261                 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
262                 return;
263         }
264
265         if (sk->sk_state == DCCP_TIME_WAIT) {
266                 inet_twsk_put(inet_twsk(sk));
267                 return;
268         }
269         seq = dccp_hdr_seq(dh);
270         if (sk->sk_state == DCCP_NEW_SYN_RECV)
271                 return dccp_req_err(sk, seq);
272
273         bh_lock_sock(sk);
274         /* If too many ICMPs get dropped on busy
275          * servers this needs to be solved differently.
276          */
277         if (sock_owned_by_user(sk))
278                 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
279
280         if (sk->sk_state == DCCP_CLOSED)
281                 goto out;
282
283         dp = dccp_sk(sk);
284         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
285             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
286                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
287                 goto out;
288         }
289
290         switch (type) {
291         case ICMP_REDIRECT:
292                 if (!sock_owned_by_user(sk))
293                         dccp_do_redirect(skb, sk);
294                 goto out;
295         case ICMP_SOURCE_QUENCH:
296                 /* Just silently ignore these. */
297                 goto out;
298         case ICMP_PARAMETERPROB:
299                 err = EPROTO;
300                 break;
301         case ICMP_DEST_UNREACH:
302                 if (code > NR_ICMP_UNREACH)
303                         goto out;
304
305                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
306                         if (!sock_owned_by_user(sk))
307                                 dccp_do_pmtu_discovery(sk, iph, info);
308                         goto out;
309                 }
310
311                 err = icmp_err_convert[code].errno;
312                 break;
313         case ICMP_TIME_EXCEEDED:
314                 err = EHOSTUNREACH;
315                 break;
316         default:
317                 goto out;
318         }
319
320         switch (sk->sk_state) {
321         case DCCP_REQUESTING:
322         case DCCP_RESPOND:
323                 if (!sock_owned_by_user(sk)) {
324                         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
325                         sk->sk_err = err;
326
327                         sk->sk_error_report(sk);
328
329                         dccp_done(sk);
330                 } else
331                         sk->sk_err_soft = err;
332                 goto out;
333         }
334
335         /* If we've already connected we will keep trying
336          * until we time out, or the user gives up.
337          *
338          * rfc1122 4.2.3.9 allows to consider as hard errors
339          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
340          * but it is obsoleted by pmtu discovery).
341          *
342          * Note, that in modern internet, where routing is unreliable
343          * and in each dark corner broken firewalls sit, sending random
344          * errors ordered by their masters even this two messages finally lose
345          * their original sense (even Linux sends invalid PORT_UNREACHs)
346          *
347          * Now we are in compliance with RFCs.
348          *                                                      --ANK (980905)
349          */
350
351         inet = inet_sk(sk);
352         if (!sock_owned_by_user(sk) && inet->recverr) {
353                 sk->sk_err = err;
354                 sk->sk_error_report(sk);
355         } else /* Only an error on timeout */
356                 sk->sk_err_soft = err;
357 out:
358         bh_unlock_sock(sk);
359         sock_put(sk);
360 }
361
362 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
363                                       __be32 src, __be32 dst)
364 {
365         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
366 }
367
368 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
369 {
370         const struct inet_sock *inet = inet_sk(sk);
371         struct dccp_hdr *dh = dccp_hdr(skb);
372
373         dccp_csum_outgoing(skb);
374         dh->dccph_checksum = dccp_v4_csum_finish(skb,
375                                                  inet->inet_saddr,
376                                                  inet->inet_daddr);
377 }
378 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
379
380 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
381 {
382         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
383                                            ip_hdr(skb)->saddr,
384                                            dccp_hdr(skb)->dccph_dport,
385                                            dccp_hdr(skb)->dccph_sport);
386 }
387
388 /*
389  * The three way handshake has completed - we got a valid ACK or DATAACK -
390  * now create the new socket.
391  *
392  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
393  */
394 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
395                                        struct sk_buff *skb,
396                                        struct request_sock *req,
397                                        struct dst_entry *dst,
398                                        struct request_sock *req_unhash,
399                                        bool *own_req)
400 {
401         struct inet_request_sock *ireq;
402         struct inet_sock *newinet;
403         struct sock *newsk;
404
405         if (sk_acceptq_is_full(sk))
406                 goto exit_overflow;
407
408         newsk = dccp_create_openreq_child(sk, req, skb);
409         if (newsk == NULL)
410                 goto exit_nonewsk;
411
412         newinet            = inet_sk(newsk);
413         ireq               = inet_rsk(req);
414         sk_daddr_set(newsk, ireq->ir_rmt_addr);
415         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
416         newinet->inet_saddr     = ireq->ir_loc_addr;
417         RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
418         newinet->mc_index  = inet_iif(skb);
419         newinet->mc_ttl    = ip_hdr(skb)->ttl;
420         newinet->inet_id   = prandom_u32();
421
422         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
423                 goto put_and_exit;
424
425         sk_setup_caps(newsk, dst);
426
427         dccp_sync_mss(newsk, dst_mtu(dst));
428
429         if (__inet_inherit_port(sk, newsk) < 0)
430                 goto put_and_exit;
431         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
432         if (*own_req)
433                 ireq->ireq_opt = NULL;
434         else
435                 newinet->inet_opt = NULL;
436         return newsk;
437
438 exit_overflow:
439         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
440 exit_nonewsk:
441         dst_release(dst);
442 exit:
443         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
444         return NULL;
445 put_and_exit:
446         newinet->inet_opt = NULL;
447         inet_csk_prepare_forced_close(newsk);
448         dccp_done(newsk);
449         goto exit;
450 }
451 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
452
453 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
454                                            struct sk_buff *skb)
455 {
456         struct rtable *rt;
457         const struct iphdr *iph = ip_hdr(skb);
458         struct flowi4 fl4 = {
459                 .flowi4_oif = inet_iif(skb),
460                 .daddr = iph->saddr,
461                 .saddr = iph->daddr,
462                 .flowi4_tos = RT_CONN_FLAGS(sk),
463                 .flowi4_proto = sk->sk_protocol,
464                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
465                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
466         };
467
468         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
469         rt = ip_route_output_flow(net, &fl4, sk);
470         if (IS_ERR(rt)) {
471                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
472                 return NULL;
473         }
474
475         return &rt->dst;
476 }
477
478 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
479 {
480         int err = -1;
481         struct sk_buff *skb;
482         struct dst_entry *dst;
483         struct flowi4 fl4;
484
485         dst = inet_csk_route_req(sk, &fl4, req);
486         if (dst == NULL)
487                 goto out;
488
489         skb = dccp_make_response(sk, dst, req);
490         if (skb != NULL) {
491                 const struct inet_request_sock *ireq = inet_rsk(req);
492                 struct dccp_hdr *dh = dccp_hdr(skb);
493
494                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
495                                                               ireq->ir_rmt_addr);
496                 rcu_read_lock();
497                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
498                                             ireq->ir_rmt_addr,
499                                             rcu_dereference(ireq->ireq_opt));
500                 rcu_read_unlock();
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(const 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         local_bh_disable();
539         bh_lock_sock(ctl_sk);
540         err = ip_build_and_send_pkt(skb, ctl_sk,
541                                     rxiph->daddr, rxiph->saddr, NULL);
542         bh_unlock_sock(ctl_sk);
543
544         if (net_xmit_eval(err) == 0) {
545                 __DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
546                 __DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
547         }
548         local_bh_enable();
549 out:
550         dst_release(dst);
551 }
552
553 static void dccp_v4_reqsk_destructor(struct request_sock *req)
554 {
555         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
556         kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
557 }
558
559 void dccp_syn_ack_timeout(const struct request_sock *req)
560 {
561 }
562 EXPORT_SYMBOL(dccp_syn_ack_timeout);
563
564 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
565         .family         = PF_INET,
566         .obj_size       = sizeof(struct dccp_request_sock),
567         .rtx_syn_ack    = dccp_v4_send_response,
568         .send_ack       = dccp_reqsk_send_ack,
569         .destructor     = dccp_v4_reqsk_destructor,
570         .send_reset     = dccp_v4_ctl_send_reset,
571         .syn_ack_timeout = dccp_syn_ack_timeout,
572 };
573
574 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
575 {
576         struct inet_request_sock *ireq;
577         struct request_sock *req;
578         struct dccp_request_sock *dreq;
579         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
580         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
581
582         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
583         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
584                 return 0;       /* discard, don't send a reset here */
585
586         if (dccp_bad_service_code(sk, service)) {
587                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
588                 goto drop;
589         }
590         /*
591          * TW buckets are converted to open requests without
592          * limitations, they conserve resources and peer is
593          * evidently real one.
594          */
595         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
596         if (inet_csk_reqsk_queue_is_full(sk))
597                 goto drop;
598
599         if (sk_acceptq_is_full(sk))
600                 goto drop;
601
602         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
603         if (req == NULL)
604                 goto drop;
605
606         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
607                 goto drop_and_free;
608
609         dreq = dccp_rsk(req);
610         if (dccp_parse_options(sk, dreq, skb))
611                 goto drop_and_free;
612
613         if (security_inet_conn_request(sk, skb, req))
614                 goto drop_and_free;
615
616         ireq = inet_rsk(req);
617         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
618         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
619         ireq->ir_mark = inet_request_mark(sk, skb);
620         ireq->ireq_family = AF_INET;
621         ireq->ir_iif = sk->sk_bound_dev_if;
622
623         /*
624          * Step 3: Process LISTEN state
625          *
626          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
627          *
628          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
629          */
630         dreq->dreq_isr     = dcb->dccpd_seq;
631         dreq->dreq_gsr     = dreq->dreq_isr;
632         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
633         dreq->dreq_gss     = dreq->dreq_iss;
634         dreq->dreq_service = service;
635
636         if (dccp_v4_send_response(sk, req))
637                 goto drop_and_free;
638
639         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
640         reqsk_put(req);
641         return 0;
642
643 drop_and_free:
644         reqsk_free(req);
645 drop:
646         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
647         return -1;
648 }
649 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
650
651 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
652 {
653         struct dccp_hdr *dh = dccp_hdr(skb);
654
655         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
656                 if (dccp_rcv_established(sk, skb, dh, skb->len))
657                         goto reset;
658                 return 0;
659         }
660
661         /*
662          *  Step 3: Process LISTEN state
663          *       If P.type == Request or P contains a valid Init Cookie option,
664          *            (* Must scan the packet's options to check for Init
665          *               Cookies.  Only Init Cookies are processed here,
666          *               however; other options are processed in Step 8.  This
667          *               scan need only be performed if the endpoint uses Init
668          *               Cookies *)
669          *            (* Generate a new socket and switch to that socket *)
670          *            Set S := new socket for this port pair
671          *            S.state = RESPOND
672          *            Choose S.ISS (initial seqno) or set from Init Cookies
673          *            Initialize S.GAR := S.ISS
674          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
675          *            Continue with S.state == RESPOND
676          *            (* A Response packet will be generated in Step 11 *)
677          *       Otherwise,
678          *            Generate Reset(No Connection) unless P.type == Reset
679          *            Drop packet and return
680          *
681          * NOTE: the check for the packet types is done in
682          *       dccp_rcv_state_process
683          */
684
685         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
686                 goto reset;
687         return 0;
688
689 reset:
690         dccp_v4_ctl_send_reset(sk, skb);
691         kfree_skb(skb);
692         return 0;
693 }
694 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
695
696 /**
697  *      dccp_invalid_packet  -  check for malformed packets
698  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
699  *      Packets that fail these checks are ignored and do not receive Resets.
700  */
701 int dccp_invalid_packet(struct sk_buff *skb)
702 {
703         const struct dccp_hdr *dh;
704         unsigned int cscov;
705         u8 dccph_doff;
706
707         if (skb->pkt_type != PACKET_HOST)
708                 return 1;
709
710         /* If the packet is shorter than 12 bytes, drop packet and return */
711         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
712                 DCCP_WARN("pskb_may_pull failed\n");
713                 return 1;
714         }
715
716         dh = dccp_hdr(skb);
717
718         /* If P.type is not understood, drop packet and return */
719         if (dh->dccph_type >= DCCP_PKT_INVALID) {
720                 DCCP_WARN("invalid packet type\n");
721                 return 1;
722         }
723
724         /*
725          * If P.Data Offset is too small for packet type, drop packet and return
726          */
727         dccph_doff = dh->dccph_doff;
728         if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
729                 DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
730                 return 1;
731         }
732         /*
733          * If P.Data Offset is too too large for packet, drop packet and return
734          */
735         if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
736                 DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
737                 return 1;
738         }
739         dh = dccp_hdr(skb);
740         /*
741          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
742          * has short sequence numbers), drop packet and return
743          */
744         if ((dh->dccph_type < DCCP_PKT_DATA    ||
745             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
746                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
747                           dccp_packet_name(dh->dccph_type));
748                 return 1;
749         }
750
751         /*
752          * If P.CsCov is too large for the packet size, drop packet and return.
753          * This must come _before_ checksumming (not as RFC 4340 suggests).
754          */
755         cscov = dccp_csum_coverage(skb);
756         if (cscov > skb->len) {
757                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
758                           dh->dccph_cscov, skb->len);
759                 return 1;
760         }
761
762         /* If header checksum is incorrect, drop packet and return.
763          * (This step is completed in the AF-dependent functions.) */
764         skb->csum = skb_checksum(skb, 0, cscov, 0);
765
766         return 0;
767 }
768 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
769
770 /* this is called when real data arrives */
771 static int dccp_v4_rcv(struct sk_buff *skb)
772 {
773         const struct dccp_hdr *dh;
774         const struct iphdr *iph;
775         bool refcounted;
776         struct sock *sk;
777         int min_cov;
778
779         /* Step 1: Check header basics */
780
781         if (dccp_invalid_packet(skb))
782                 goto discard_it;
783
784         iph = ip_hdr(skb);
785         /* Step 1: If header checksum is incorrect, drop packet and return */
786         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
787                 DCCP_WARN("dropped packet with invalid checksum\n");
788                 goto discard_it;
789         }
790
791         dh = dccp_hdr(skb);
792
793         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
794         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
795
796         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
797                       dccp_packet_name(dh->dccph_type),
798                       &iph->saddr, ntohs(dh->dccph_sport),
799                       &iph->daddr, ntohs(dh->dccph_dport),
800                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
801
802         if (dccp_packet_without_ack(skb)) {
803                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
804                 dccp_pr_debug_cat("\n");
805         } else {
806                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
807                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
808                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
809         }
810
811 lookup:
812         sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
813                                dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
814         if (!sk) {
815                 dccp_pr_debug("failed to look up flow ID in table and "
816                               "get corresponding socket\n");
817                 goto no_dccp_socket;
818         }
819
820         /*
821          * Step 2:
822          *      ... or S.state == TIMEWAIT,
823          *              Generate Reset(No Connection) unless P.type == Reset
824          *              Drop packet and return
825          */
826         if (sk->sk_state == DCCP_TIME_WAIT) {
827                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
828                 inet_twsk_put(inet_twsk(sk));
829                 goto no_dccp_socket;
830         }
831
832         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
833                 struct request_sock *req = inet_reqsk(sk);
834                 struct sock *nsk;
835
836                 sk = req->rsk_listener;
837                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
838                         inet_csk_reqsk_queue_drop_and_put(sk, req);
839                         goto lookup;
840                 }
841                 sock_hold(sk);
842                 refcounted = true;
843                 nsk = dccp_check_req(sk, skb, req);
844                 if (!nsk) {
845                         reqsk_put(req);
846                         goto discard_and_relse;
847                 }
848                 if (nsk == sk) {
849                         reqsk_put(req);
850                 } else if (dccp_child_process(sk, nsk, skb)) {
851                         dccp_v4_ctl_send_reset(sk, skb);
852                         goto discard_and_relse;
853                 } else {
854                         sock_put(sk);
855                         return 0;
856                 }
857         }
858         /*
859          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
860          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
861          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
862          */
863         min_cov = dccp_sk(sk)->dccps_pcrlen;
864         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
865                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
866                               dh->dccph_cscov, min_cov);
867                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
868                  *         options (Section 11.7) with Drop Code 0, Protocol
869                  *         Constraints."                                     */
870                 goto discard_and_relse;
871         }
872
873         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
874                 goto discard_and_relse;
875         nf_reset(skb);
876
877         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
878
879 no_dccp_socket:
880         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
881                 goto discard_it;
882         /*
883          * Step 2:
884          *      If no socket ...
885          *              Generate Reset(No Connection) unless P.type == Reset
886          *              Drop packet and return
887          */
888         if (dh->dccph_type != DCCP_PKT_RESET) {
889                 DCCP_SKB_CB(skb)->dccpd_reset_code =
890                                         DCCP_RESET_CODE_NO_CONNECTION;
891                 dccp_v4_ctl_send_reset(sk, skb);
892         }
893
894 discard_it:
895         kfree_skb(skb);
896         return 0;
897
898 discard_and_relse:
899         if (refcounted)
900                 sock_put(sk);
901         goto discard_it;
902 }
903
904 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
905         .queue_xmit        = ip_queue_xmit,
906         .send_check        = dccp_v4_send_check,
907         .rebuild_header    = inet_sk_rebuild_header,
908         .conn_request      = dccp_v4_conn_request,
909         .syn_recv_sock     = dccp_v4_request_recv_sock,
910         .net_header_len    = sizeof(struct iphdr),
911         .setsockopt        = ip_setsockopt,
912         .getsockopt        = ip_getsockopt,
913         .addr2sockaddr     = inet_csk_addr2sockaddr,
914         .sockaddr_len      = sizeof(struct sockaddr_in),
915 #ifdef CONFIG_COMPAT
916         .compat_setsockopt = compat_ip_setsockopt,
917         .compat_getsockopt = compat_ip_getsockopt,
918 #endif
919 };
920
921 static int dccp_v4_init_sock(struct sock *sk)
922 {
923         static __u8 dccp_v4_ctl_sock_initialized;
924         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
925
926         if (err == 0) {
927                 if (unlikely(!dccp_v4_ctl_sock_initialized))
928                         dccp_v4_ctl_sock_initialized = 1;
929                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
930         }
931
932         return err;
933 }
934
935 static struct timewait_sock_ops dccp_timewait_sock_ops = {
936         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
937 };
938
939 static struct proto dccp_v4_prot = {
940         .name                   = "DCCP",
941         .owner                  = THIS_MODULE,
942         .close                  = dccp_close,
943         .connect                = dccp_v4_connect,
944         .disconnect             = dccp_disconnect,
945         .ioctl                  = dccp_ioctl,
946         .init                   = dccp_v4_init_sock,
947         .setsockopt             = dccp_setsockopt,
948         .getsockopt             = dccp_getsockopt,
949         .sendmsg                = dccp_sendmsg,
950         .recvmsg                = dccp_recvmsg,
951         .backlog_rcv            = dccp_v4_do_rcv,
952         .hash                   = inet_hash,
953         .unhash                 = inet_unhash,
954         .accept                 = inet_csk_accept,
955         .get_port               = inet_csk_get_port,
956         .shutdown               = dccp_shutdown,
957         .destroy                = dccp_destroy_sock,
958         .orphan_count           = &dccp_orphan_count,
959         .max_header             = MAX_DCCP_HEADER,
960         .obj_size               = sizeof(struct dccp_sock),
961         .slab_flags             = SLAB_TYPESAFE_BY_RCU,
962         .rsk_prot               = &dccp_request_sock_ops,
963         .twsk_prot              = &dccp_timewait_sock_ops,
964         .h.hashinfo             = &dccp_hashinfo,
965 #ifdef CONFIG_COMPAT
966         .compat_setsockopt      = compat_dccp_setsockopt,
967         .compat_getsockopt      = compat_dccp_getsockopt,
968 #endif
969 };
970
971 static const struct net_protocol dccp_v4_protocol = {
972         .handler        = dccp_v4_rcv,
973         .err_handler    = dccp_v4_err,
974         .no_policy      = 1,
975         .netns_ok       = 1,
976         .icmp_strict_tag_validation = 1,
977 };
978
979 static const struct proto_ops inet_dccp_ops = {
980         .family            = PF_INET,
981         .owner             = THIS_MODULE,
982         .release           = inet_release,
983         .bind              = inet_bind,
984         .connect           = inet_stream_connect,
985         .socketpair        = sock_no_socketpair,
986         .accept            = inet_accept,
987         .getname           = inet_getname,
988         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
989         .poll              = dccp_poll,
990         .ioctl             = inet_ioctl,
991         /* FIXME: work on inet_listen to rename it to sock_common_listen */
992         .listen            = inet_dccp_listen,
993         .shutdown          = inet_shutdown,
994         .setsockopt        = sock_common_setsockopt,
995         .getsockopt        = sock_common_getsockopt,
996         .sendmsg           = inet_sendmsg,
997         .recvmsg           = sock_common_recvmsg,
998         .mmap              = sock_no_mmap,
999         .sendpage          = sock_no_sendpage,
1000 #ifdef CONFIG_COMPAT
1001         .compat_setsockopt = compat_sock_common_setsockopt,
1002         .compat_getsockopt = compat_sock_common_getsockopt,
1003 #endif
1004 };
1005
1006 static struct inet_protosw dccp_v4_protosw = {
1007         .type           = SOCK_DCCP,
1008         .protocol       = IPPROTO_DCCP,
1009         .prot           = &dccp_v4_prot,
1010         .ops            = &inet_dccp_ops,
1011         .flags          = INET_PROTOSW_ICSK,
1012 };
1013
1014 static int __net_init dccp_v4_init_net(struct net *net)
1015 {
1016         if (dccp_hashinfo.bhash == NULL)
1017                 return -ESOCKTNOSUPPORT;
1018
1019         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1020                                     SOCK_DCCP, IPPROTO_DCCP, net);
1021 }
1022
1023 static void __net_exit dccp_v4_exit_net(struct net *net)
1024 {
1025         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1026 }
1027
1028 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1029 {
1030         inet_twsk_purge(&dccp_hashinfo, AF_INET);
1031 }
1032
1033 static struct pernet_operations dccp_v4_ops = {
1034         .init   = dccp_v4_init_net,
1035         .exit   = dccp_v4_exit_net,
1036         .exit_batch = dccp_v4_exit_batch,
1037 };
1038
1039 static int __init dccp_v4_init(void)
1040 {
1041         int err = proto_register(&dccp_v4_prot, 1);
1042
1043         if (err)
1044                 goto out;
1045
1046         inet_register_protosw(&dccp_v4_protosw);
1047
1048         err = register_pernet_subsys(&dccp_v4_ops);
1049         if (err)
1050                 goto out_destroy_ctl_sock;
1051
1052         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1053         if (err)
1054                 goto out_proto_unregister;
1055
1056 out:
1057         return err;
1058 out_proto_unregister:
1059         unregister_pernet_subsys(&dccp_v4_ops);
1060 out_destroy_ctl_sock:
1061         inet_unregister_protosw(&dccp_v4_protosw);
1062         proto_unregister(&dccp_v4_prot);
1063         goto out;
1064 }
1065
1066 static void __exit dccp_v4_exit(void)
1067 {
1068         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1069         unregister_pernet_subsys(&dccp_v4_ops);
1070         inet_unregister_protosw(&dccp_v4_protosw);
1071         proto_unregister(&dccp_v4_prot);
1072 }
1073
1074 module_init(dccp_v4_init);
1075 module_exit(dccp_v4_exit);
1076
1077 /*
1078  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1079  * values directly, Also cover the case where the protocol is not specified,
1080  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1081  */
1082 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1083 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1084 MODULE_LICENSE("GPL");
1085 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1086 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");