inet: stop leaking jiffies on the wire
[muen/linux.git] / net / dccp / ipv4.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  net/dccp/ipv4.c
4  *
5  *  An implementation of the DCCP protocol
6  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7  */
8
9 #include <linux/dccp.h>
10 #include <linux/icmp.h>
11 #include <linux/slab.h>
12 #include <linux/module.h>
13 #include <linux/skbuff.h>
14 #include <linux/random.h>
15
16 #include <net/icmp.h>
17 #include <net/inet_common.h>
18 #include <net/inet_hashtables.h>
19 #include <net/inet_sock.h>
20 #include <net/protocol.h>
21 #include <net/sock.h>
22 #include <net/timewait_sock.h>
23 #include <net/tcp_states.h>
24 #include <net/xfrm.h>
25 #include <net/secure_seq.h>
26
27 #include "ackvec.h"
28 #include "ccid.h"
29 #include "dccp.h"
30 #include "feat.h"
31
32 /*
33  * The per-net dccp.v4_ctl_sk socket is used for responding to
34  * the Out-of-the-blue (OOTB) packets. A control sock will be created
35  * for this socket at the initialization time.
36  */
37
38 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
39 {
40         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
41         struct inet_sock *inet = inet_sk(sk);
42         struct dccp_sock *dp = dccp_sk(sk);
43         __be16 orig_sport, orig_dport;
44         __be32 daddr, nexthop;
45         struct flowi4 *fl4;
46         struct rtable *rt;
47         int err;
48         struct ip_options_rcu *inet_opt;
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
60         inet_opt = rcu_dereference_protected(inet->inet_opt,
61                                              lockdep_sock_is_held(sk));
62         if (inet_opt != NULL && inet_opt->opt.srr) {
63                 if (daddr == 0)
64                         return -EINVAL;
65                 nexthop = inet_opt->opt.faddr;
66         }
67
68         orig_sport = inet->inet_sport;
69         orig_dport = usin->sin_port;
70         fl4 = &inet->cork.fl.u.ip4;
71         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
72                               RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
73                               IPPROTO_DCCP,
74                               orig_sport, orig_dport, sk);
75         if (IS_ERR(rt))
76                 return PTR_ERR(rt);
77
78         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
79                 ip_rt_put(rt);
80                 return -ENETUNREACH;
81         }
82
83         if (inet_opt == NULL || !inet_opt->opt.srr)
84                 daddr = fl4->daddr;
85
86         if (inet->inet_saddr == 0)
87                 inet->inet_saddr = fl4->saddr;
88         sk_rcv_saddr_set(sk, inet->inet_saddr);
89         inet->inet_dport = usin->sin_port;
90         sk_daddr_set(sk, daddr);
91
92         inet_csk(sk)->icsk_ext_hdr_len = 0;
93         if (inet_opt)
94                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
95         /*
96          * Socket identity is still unknown (sport may be zero).
97          * However we set state to DCCP_REQUESTING and not releasing socket
98          * lock select source port, enter ourselves into the hash tables and
99          * complete initialization after this.
100          */
101         dccp_set_state(sk, DCCP_REQUESTING);
102         err = inet_hash_connect(&dccp_death_row, sk);
103         if (err != 0)
104                 goto failure;
105
106         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
107                                inet->inet_sport, inet->inet_dport, sk);
108         if (IS_ERR(rt)) {
109                 err = PTR_ERR(rt);
110                 rt = NULL;
111                 goto failure;
112         }
113         /* OK, now commit destination to socket.  */
114         sk_setup_caps(sk, &rt->dst);
115
116         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
117                                                     inet->inet_daddr,
118                                                     inet->inet_sport,
119                                                     inet->inet_dport);
120         inet->inet_id = prandom_u32();
121
122         err = dccp_connect(sk);
123         rt = NULL;
124         if (err != 0)
125                 goto failure;
126 out:
127         return err;
128 failure:
129         /*
130          * This unhashes the socket and releases the local port, if necessary.
131          */
132         dccp_set_state(sk, DCCP_CLOSED);
133         ip_rt_put(rt);
134         sk->sk_route_caps = 0;
135         inet->inet_dport = 0;
136         goto out;
137 }
138 EXPORT_SYMBOL_GPL(dccp_v4_connect);
139
140 /*
141  * This routine does path mtu discovery as defined in RFC1191.
142  */
143 static inline void dccp_do_pmtu_discovery(struct sock *sk,
144                                           const struct iphdr *iph,
145                                           u32 mtu)
146 {
147         struct dst_entry *dst;
148         const struct inet_sock *inet = inet_sk(sk);
149         const struct dccp_sock *dp = dccp_sk(sk);
150
151         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
152          * send out by Linux are always < 576bytes so they should go through
153          * unfragmented).
154          */
155         if (sk->sk_state == DCCP_LISTEN)
156                 return;
157
158         dst = inet_csk_update_pmtu(sk, mtu);
159         if (!dst)
160                 return;
161
162         /* Something is about to be wrong... Remember soft error
163          * for the case, if this connection will not able to recover.
164          */
165         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
166                 sk->sk_err_soft = EMSGSIZE;
167
168         mtu = dst_mtu(dst);
169
170         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
171             ip_sk_accept_pmtu(sk) &&
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 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
187 {
188         struct dst_entry *dst = __sk_dst_check(sk, 0);
189
190         if (dst)
191                 dst->ops->redirect(dst, sk, skb);
192 }
193
194 void dccp_req_err(struct sock *sk, u64 seq)
195         {
196         struct request_sock *req = inet_reqsk(sk);
197         struct net *net = sock_net(sk);
198
199         /*
200          * ICMPs are not backlogged, hence we cannot get an established
201          * socket here.
202          */
203         if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
204                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
205         } else {
206                 /*
207                  * Still in RESPOND, just remove it silently.
208                  * There is no good way to pass the error to the newly
209                  * created socket, and POSIX does not want network
210                  * errors returned from accept().
211                  */
212                 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
213         }
214         reqsk_put(req);
215 }
216 EXPORT_SYMBOL(dccp_req_err);
217
218 /*
219  * This routine is called by the ICMP module when it gets some sort of error
220  * condition. If err < 0 then the socket should be closed and the error
221  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
222  * After adjustment header points to the first 8 bytes of the tcp header. We
223  * need to find the appropriate port.
224  *
225  * The locking strategy used here is very "optimistic". When someone else
226  * accesses the socket the ICMP is just dropped and for some paths there is no
227  * check at all. A more general error queue to queue errors for later handling
228  * is probably better.
229  */
230 static int dccp_v4_err(struct sk_buff *skb, u32 info)
231 {
232         const struct iphdr *iph = (struct iphdr *)skb->data;
233         const u8 offset = iph->ihl << 2;
234         const struct dccp_hdr *dh;
235         struct dccp_sock *dp;
236         struct inet_sock *inet;
237         const int type = icmp_hdr(skb)->type;
238         const int code = icmp_hdr(skb)->code;
239         struct sock *sk;
240         __u64 seq;
241         int err;
242         struct net *net = dev_net(skb->dev);
243
244         /* Only need dccph_dport & dccph_sport which are the first
245          * 4 bytes in dccp header.
246          * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
247          */
248         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
249         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
250         dh = (struct dccp_hdr *)(skb->data + offset);
251
252         sk = __inet_lookup_established(net, &dccp_hashinfo,
253                                        iph->daddr, dh->dccph_dport,
254                                        iph->saddr, ntohs(dh->dccph_sport),
255                                        inet_iif(skb), 0);
256         if (!sk) {
257                 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
258                 return -ENOENT;
259         }
260
261         if (sk->sk_state == DCCP_TIME_WAIT) {
262                 inet_twsk_put(inet_twsk(sk));
263                 return 0;
264         }
265         seq = dccp_hdr_seq(dh);
266         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
267                 dccp_req_err(sk, seq);
268                 return 0;
269         }
270
271         bh_lock_sock(sk);
272         /* If too many ICMPs get dropped on busy
273          * servers this needs to be solved differently.
274          */
275         if (sock_owned_by_user(sk))
276                 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
277
278         if (sk->sk_state == DCCP_CLOSED)
279                 goto out;
280
281         dp = dccp_sk(sk);
282         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
283             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
284                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
285                 goto out;
286         }
287
288         switch (type) {
289         case ICMP_REDIRECT:
290                 if (!sock_owned_by_user(sk))
291                         dccp_do_redirect(skb, sk);
292                 goto out;
293         case ICMP_SOURCE_QUENCH:
294                 /* Just silently ignore these. */
295                 goto out;
296         case ICMP_PARAMETERPROB:
297                 err = EPROTO;
298                 break;
299         case ICMP_DEST_UNREACH:
300                 if (code > NR_ICMP_UNREACH)
301                         goto out;
302
303                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
304                         if (!sock_owned_by_user(sk))
305                                 dccp_do_pmtu_discovery(sk, iph, info);
306                         goto out;
307                 }
308
309                 err = icmp_err_convert[code].errno;
310                 break;
311         case ICMP_TIME_EXCEEDED:
312                 err = EHOSTUNREACH;
313                 break;
314         default:
315                 goto out;
316         }
317
318         switch (sk->sk_state) {
319         case DCCP_REQUESTING:
320         case DCCP_RESPOND:
321                 if (!sock_owned_by_user(sk)) {
322                         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
323                         sk->sk_err = err;
324
325                         sk->sk_error_report(sk);
326
327                         dccp_done(sk);
328                 } else
329                         sk->sk_err_soft = err;
330                 goto out;
331         }
332
333         /* If we've already connected we will keep trying
334          * until we time out, or the user gives up.
335          *
336          * rfc1122 4.2.3.9 allows to consider as hard errors
337          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
338          * but it is obsoleted by pmtu discovery).
339          *
340          * Note, that in modern internet, where routing is unreliable
341          * and in each dark corner broken firewalls sit, sending random
342          * errors ordered by their masters even this two messages finally lose
343          * their original sense (even Linux sends invalid PORT_UNREACHs)
344          *
345          * Now we are in compliance with RFCs.
346          *                                                      --ANK (980905)
347          */
348
349         inet = inet_sk(sk);
350         if (!sock_owned_by_user(sk) && inet->recverr) {
351                 sk->sk_err = err;
352                 sk->sk_error_report(sk);
353         } else /* Only an error on timeout */
354                 sk->sk_err_soft = err;
355 out:
356         bh_unlock_sock(sk);
357         sock_put(sk);
358         return 0;
359 }
360
361 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
362                                       __be32 src, __be32 dst)
363 {
364         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
365 }
366
367 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
368 {
369         const struct inet_sock *inet = inet_sk(sk);
370         struct dccp_hdr *dh = dccp_hdr(skb);
371
372         dccp_csum_outgoing(skb);
373         dh->dccph_checksum = dccp_v4_csum_finish(skb,
374                                                  inet->inet_saddr,
375                                                  inet->inet_daddr);
376 }
377 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
378
379 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
380 {
381         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
382                                            ip_hdr(skb)->saddr,
383                                            dccp_hdr(skb)->dccph_dport,
384                                            dccp_hdr(skb)->dccph_sport);
385 }
386
387 /*
388  * The three way handshake has completed - we got a valid ACK or DATAACK -
389  * now create the new socket.
390  *
391  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
392  */
393 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
394                                        struct sk_buff *skb,
395                                        struct request_sock *req,
396                                        struct dst_entry *dst,
397                                        struct request_sock *req_unhash,
398                                        bool *own_req)
399 {
400         struct inet_request_sock *ireq;
401         struct inet_sock *newinet;
402         struct sock *newsk;
403
404         if (sk_acceptq_is_full(sk))
405                 goto exit_overflow;
406
407         newsk = dccp_create_openreq_child(sk, req, skb);
408         if (newsk == NULL)
409                 goto exit_nonewsk;
410
411         newinet            = inet_sk(newsk);
412         ireq               = inet_rsk(req);
413         sk_daddr_set(newsk, ireq->ir_rmt_addr);
414         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
415         newinet->inet_saddr     = ireq->ir_loc_addr;
416         RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
417         newinet->mc_index  = inet_iif(skb);
418         newinet->mc_ttl    = ip_hdr(skb)->ttl;
419         newinet->inet_id   = jiffies;
420
421         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
422                 goto put_and_exit;
423
424         sk_setup_caps(newsk, dst);
425
426         dccp_sync_mss(newsk, dst_mtu(dst));
427
428         if (__inet_inherit_port(sk, newsk) < 0)
429                 goto put_and_exit;
430         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
431         if (*own_req)
432                 ireq->ireq_opt = NULL;
433         else
434                 newinet->inet_opt = NULL;
435         return newsk;
436
437 exit_overflow:
438         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
439 exit_nonewsk:
440         dst_release(dst);
441 exit:
442         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
443         return NULL;
444 put_and_exit:
445         newinet->inet_opt = NULL;
446         inet_csk_prepare_forced_close(newsk);
447         dccp_done(newsk);
448         goto exit;
449 }
450 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
451
452 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
453                                            struct sk_buff *skb)
454 {
455         struct rtable *rt;
456         const struct iphdr *iph = ip_hdr(skb);
457         struct flowi4 fl4 = {
458                 .flowi4_oif = inet_iif(skb),
459                 .daddr = iph->saddr,
460                 .saddr = iph->daddr,
461                 .flowi4_tos = RT_CONN_FLAGS(sk),
462                 .flowi4_proto = sk->sk_protocol,
463                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
464                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
465         };
466
467         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
468         rt = ip_route_output_flow(net, &fl4, sk);
469         if (IS_ERR(rt)) {
470                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
471                 return NULL;
472         }
473
474         return &rt->dst;
475 }
476
477 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
478 {
479         int err = -1;
480         struct sk_buff *skb;
481         struct dst_entry *dst;
482         struct flowi4 fl4;
483
484         dst = inet_csk_route_req(sk, &fl4, req);
485         if (dst == NULL)
486                 goto out;
487
488         skb = dccp_make_response(sk, dst, req);
489         if (skb != NULL) {
490                 const struct inet_request_sock *ireq = inet_rsk(req);
491                 struct dccp_hdr *dh = dccp_hdr(skb);
492
493                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
494                                                               ireq->ir_rmt_addr);
495                 rcu_read_lock();
496                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
497                                             ireq->ir_rmt_addr,
498                                             rcu_dereference(ireq->ireq_opt));
499                 rcu_read_unlock();
500                 err = net_xmit_eval(err);
501         }
502
503 out:
504         dst_release(dst);
505         return err;
506 }
507
508 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
509 {
510         int err;
511         const struct iphdr *rxiph;
512         struct sk_buff *skb;
513         struct dst_entry *dst;
514         struct net *net = dev_net(skb_dst(rxskb)->dev);
515         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
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 (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
522                 return;
523
524         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
525         if (dst == NULL)
526                 return;
527
528         skb = dccp_ctl_make_reset(ctl_sk, 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_set(skb, dst_clone(dst));
536
537         local_bh_disable();
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(DCCP_MIB_OUTSEGS);
545                 __DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
546         }
547         local_bh_enable();
548 out:
549         dst_release(dst);
550 }
551
552 static void dccp_v4_reqsk_destructor(struct request_sock *req)
553 {
554         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
555         kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
556 }
557
558 void dccp_syn_ack_timeout(const struct request_sock *req)
559 {
560 }
561 EXPORT_SYMBOL(dccp_syn_ack_timeout);
562
563 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
564         .family         = PF_INET,
565         .obj_size       = sizeof(struct dccp_request_sock),
566         .rtx_syn_ack    = dccp_v4_send_response,
567         .send_ack       = dccp_reqsk_send_ack,
568         .destructor     = dccp_v4_reqsk_destructor,
569         .send_reset     = dccp_v4_ctl_send_reset,
570         .syn_ack_timeout = dccp_syn_ack_timeout,
571 };
572
573 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
574 {
575         struct inet_request_sock *ireq;
576         struct request_sock *req;
577         struct dccp_request_sock *dreq;
578         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
579         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
580
581         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
582         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
583                 return 0;       /* discard, don't send a reset here */
584
585         if (dccp_bad_service_code(sk, service)) {
586                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
587                 goto drop;
588         }
589         /*
590          * TW buckets are converted to open requests without
591          * limitations, they conserve resources and peer is
592          * evidently real one.
593          */
594         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
595         if (inet_csk_reqsk_queue_is_full(sk))
596                 goto drop;
597
598         if (sk_acceptq_is_full(sk))
599                 goto drop;
600
601         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
602         if (req == NULL)
603                 goto drop;
604
605         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
606                 goto drop_and_free;
607
608         dreq = dccp_rsk(req);
609         if (dccp_parse_options(sk, dreq, skb))
610                 goto drop_and_free;
611
612         if (security_inet_conn_request(sk, skb, req))
613                 goto drop_and_free;
614
615         ireq = inet_rsk(req);
616         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
617         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
618         ireq->ir_mark = inet_request_mark(sk, skb);
619         ireq->ireq_family = AF_INET;
620         ireq->ir_iif = sk->sk_bound_dev_if;
621
622         /*
623          * Step 3: Process LISTEN state
624          *
625          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
626          *
627          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
628          */
629         dreq->dreq_isr     = dcb->dccpd_seq;
630         dreq->dreq_gsr     = dreq->dreq_isr;
631         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
632         dreq->dreq_gss     = dreq->dreq_iss;
633         dreq->dreq_service = service;
634
635         if (dccp_v4_send_response(sk, req))
636                 goto drop_and_free;
637
638         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
639         reqsk_put(req);
640         return 0;
641
642 drop_and_free:
643         reqsk_free(req);
644 drop:
645         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
646         return -1;
647 }
648 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
649
650 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
651 {
652         struct dccp_hdr *dh = dccp_hdr(skb);
653
654         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
655                 if (dccp_rcv_established(sk, skb, dh, skb->len))
656                         goto reset;
657                 return 0;
658         }
659
660         /*
661          *  Step 3: Process LISTEN state
662          *       If P.type == Request or P contains a valid Init Cookie option,
663          *            (* Must scan the packet's options to check for Init
664          *               Cookies.  Only Init Cookies are processed here,
665          *               however; other options are processed in Step 8.  This
666          *               scan need only be performed if the endpoint uses Init
667          *               Cookies *)
668          *            (* Generate a new socket and switch to that socket *)
669          *            Set S := new socket for this port pair
670          *            S.state = RESPOND
671          *            Choose S.ISS (initial seqno) or set from Init Cookies
672          *            Initialize S.GAR := S.ISS
673          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
674          *            Continue with S.state == RESPOND
675          *            (* A Response packet will be generated in Step 11 *)
676          *       Otherwise,
677          *            Generate Reset(No Connection) unless P.type == Reset
678          *            Drop packet and return
679          *
680          * NOTE: the check for the packet types is done in
681          *       dccp_rcv_state_process
682          */
683
684         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
685                 goto reset;
686         return 0;
687
688 reset:
689         dccp_v4_ctl_send_reset(sk, skb);
690         kfree_skb(skb);
691         return 0;
692 }
693 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
694
695 /**
696  *      dccp_invalid_packet  -  check for malformed packets
697  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
698  *      Packets that fail these checks are ignored and do not receive Resets.
699  */
700 int dccp_invalid_packet(struct sk_buff *skb)
701 {
702         const struct dccp_hdr *dh;
703         unsigned int cscov;
704         u8 dccph_doff;
705
706         if (skb->pkt_type != PACKET_HOST)
707                 return 1;
708
709         /* If the packet is shorter than 12 bytes, drop packet and return */
710         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
711                 DCCP_WARN("pskb_may_pull failed\n");
712                 return 1;
713         }
714
715         dh = dccp_hdr(skb);
716
717         /* If P.type is not understood, drop packet and return */
718         if (dh->dccph_type >= DCCP_PKT_INVALID) {
719                 DCCP_WARN("invalid packet type\n");
720                 return 1;
721         }
722
723         /*
724          * If P.Data Offset is too small for packet type, drop packet and return
725          */
726         dccph_doff = dh->dccph_doff;
727         if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
728                 DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
729                 return 1;
730         }
731         /*
732          * If P.Data Offset is too too large for packet, drop packet and return
733          */
734         if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
735                 DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
736                 return 1;
737         }
738         dh = dccp_hdr(skb);
739         /*
740          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
741          * has short sequence numbers), drop packet and return
742          */
743         if ((dh->dccph_type < DCCP_PKT_DATA    ||
744             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
745                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
746                           dccp_packet_name(dh->dccph_type));
747                 return 1;
748         }
749
750         /*
751          * If P.CsCov is too large for the packet size, drop packet and return.
752          * This must come _before_ checksumming (not as RFC 4340 suggests).
753          */
754         cscov = dccp_csum_coverage(skb);
755         if (cscov > skb->len) {
756                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
757                           dh->dccph_cscov, skb->len);
758                 return 1;
759         }
760
761         /* If header checksum is incorrect, drop packet and return.
762          * (This step is completed in the AF-dependent functions.) */
763         skb->csum = skb_checksum(skb, 0, cscov, 0);
764
765         return 0;
766 }
767 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
768
769 /* this is called when real data arrives */
770 static int dccp_v4_rcv(struct sk_buff *skb)
771 {
772         const struct dccp_hdr *dh;
773         const struct iphdr *iph;
774         bool refcounted;
775         struct sock *sk;
776         int min_cov;
777
778         /* Step 1: Check header basics */
779
780         if (dccp_invalid_packet(skb))
781                 goto discard_it;
782
783         iph = ip_hdr(skb);
784         /* Step 1: If header checksum is incorrect, drop packet and return */
785         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
786                 DCCP_WARN("dropped packet with invalid checksum\n");
787                 goto discard_it;
788         }
789
790         dh = dccp_hdr(skb);
791
792         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
793         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
794
795         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
796                       dccp_packet_name(dh->dccph_type),
797                       &iph->saddr, ntohs(dh->dccph_sport),
798                       &iph->daddr, ntohs(dh->dccph_dport),
799                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
800
801         if (dccp_packet_without_ack(skb)) {
802                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
803                 dccp_pr_debug_cat("\n");
804         } else {
805                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
806                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
807                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
808         }
809
810 lookup:
811         sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
812                                dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
813         if (!sk) {
814                 dccp_pr_debug("failed to look up flow ID in table and "
815                               "get corresponding socket\n");
816                 goto no_dccp_socket;
817         }
818
819         /*
820          * Step 2:
821          *      ... or S.state == TIMEWAIT,
822          *              Generate Reset(No Connection) unless P.type == Reset
823          *              Drop packet and return
824          */
825         if (sk->sk_state == DCCP_TIME_WAIT) {
826                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
827                 inet_twsk_put(inet_twsk(sk));
828                 goto no_dccp_socket;
829         }
830
831         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
832                 struct request_sock *req = inet_reqsk(sk);
833                 struct sock *nsk;
834
835                 sk = req->rsk_listener;
836                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
837                         inet_csk_reqsk_queue_drop_and_put(sk, req);
838                         goto lookup;
839                 }
840                 sock_hold(sk);
841                 refcounted = true;
842                 nsk = dccp_check_req(sk, skb, req);
843                 if (!nsk) {
844                         reqsk_put(req);
845                         goto discard_and_relse;
846                 }
847                 if (nsk == sk) {
848                         reqsk_put(req);
849                 } else if (dccp_child_process(sk, nsk, skb)) {
850                         dccp_v4_ctl_send_reset(sk, skb);
851                         goto discard_and_relse;
852                 } else {
853                         sock_put(sk);
854                         return 0;
855                 }
856         }
857         /*
858          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
859          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
860          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
861          */
862         min_cov = dccp_sk(sk)->dccps_pcrlen;
863         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
864                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
865                               dh->dccph_cscov, min_cov);
866                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
867                  *         options (Section 11.7) with Drop Code 0, Protocol
868                  *         Constraints."                                     */
869                 goto discard_and_relse;
870         }
871
872         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
873                 goto discard_and_relse;
874         nf_reset_ct(skb);
875
876         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
877
878 no_dccp_socket:
879         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
880                 goto discard_it;
881         /*
882          * Step 2:
883          *      If no socket ...
884          *              Generate Reset(No Connection) unless P.type == Reset
885          *              Drop packet and return
886          */
887         if (dh->dccph_type != DCCP_PKT_RESET) {
888                 DCCP_SKB_CB(skb)->dccpd_reset_code =
889                                         DCCP_RESET_CODE_NO_CONNECTION;
890                 dccp_v4_ctl_send_reset(sk, skb);
891         }
892
893 discard_it:
894         kfree_skb(skb);
895         return 0;
896
897 discard_and_relse:
898         if (refcounted)
899                 sock_put(sk);
900         goto discard_it;
901 }
902
903 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
904         .queue_xmit        = ip_queue_xmit,
905         .send_check        = dccp_v4_send_check,
906         .rebuild_header    = inet_sk_rebuild_header,
907         .conn_request      = dccp_v4_conn_request,
908         .syn_recv_sock     = dccp_v4_request_recv_sock,
909         .net_header_len    = sizeof(struct iphdr),
910         .setsockopt        = ip_setsockopt,
911         .getsockopt        = ip_getsockopt,
912         .addr2sockaddr     = inet_csk_addr2sockaddr,
913         .sockaddr_len      = sizeof(struct sockaddr_in),
914 #ifdef CONFIG_COMPAT
915         .compat_setsockopt = compat_ip_setsockopt,
916         .compat_getsockopt = compat_ip_getsockopt,
917 #endif
918 };
919
920 static int dccp_v4_init_sock(struct sock *sk)
921 {
922         static __u8 dccp_v4_ctl_sock_initialized;
923         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
924
925         if (err == 0) {
926                 if (unlikely(!dccp_v4_ctl_sock_initialized))
927                         dccp_v4_ctl_sock_initialized = 1;
928                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
929         }
930
931         return err;
932 }
933
934 static struct timewait_sock_ops dccp_timewait_sock_ops = {
935         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
936 };
937
938 static struct proto dccp_v4_prot = {
939         .name                   = "DCCP",
940         .owner                  = THIS_MODULE,
941         .close                  = dccp_close,
942         .connect                = dccp_v4_connect,
943         .disconnect             = dccp_disconnect,
944         .ioctl                  = dccp_ioctl,
945         .init                   = dccp_v4_init_sock,
946         .setsockopt             = dccp_setsockopt,
947         .getsockopt             = dccp_getsockopt,
948         .sendmsg                = dccp_sendmsg,
949         .recvmsg                = dccp_recvmsg,
950         .backlog_rcv            = dccp_v4_do_rcv,
951         .hash                   = inet_hash,
952         .unhash                 = inet_unhash,
953         .accept                 = inet_csk_accept,
954         .get_port               = inet_csk_get_port,
955         .shutdown               = dccp_shutdown,
956         .destroy                = dccp_destroy_sock,
957         .orphan_count           = &dccp_orphan_count,
958         .max_header             = MAX_DCCP_HEADER,
959         .obj_size               = sizeof(struct dccp_sock),
960         .slab_flags             = SLAB_TYPESAFE_BY_RCU,
961         .rsk_prot               = &dccp_request_sock_ops,
962         .twsk_prot              = &dccp_timewait_sock_ops,
963         .h.hashinfo             = &dccp_hashinfo,
964 #ifdef CONFIG_COMPAT
965         .compat_setsockopt      = compat_dccp_setsockopt,
966         .compat_getsockopt      = compat_dccp_getsockopt,
967 #endif
968 };
969
970 static const struct net_protocol dccp_v4_protocol = {
971         .handler        = dccp_v4_rcv,
972         .err_handler    = dccp_v4_err,
973         .no_policy      = 1,
974         .netns_ok       = 1,
975         .icmp_strict_tag_validation = 1,
976 };
977
978 static const struct proto_ops inet_dccp_ops = {
979         .family            = PF_INET,
980         .owner             = THIS_MODULE,
981         .release           = inet_release,
982         .bind              = inet_bind,
983         .connect           = inet_stream_connect,
984         .socketpair        = sock_no_socketpair,
985         .accept            = inet_accept,
986         .getname           = inet_getname,
987         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
988         .poll              = dccp_poll,
989         .ioctl             = inet_ioctl,
990         .gettstamp         = sock_gettstamp,
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");