[TCP]: Reduce sacked_out with reno when purging write_queue
[linux-2.6] / net / dccp / ipv4.c
1 /*
2  *  net/dccp/ipv4.c
3  *
4  *  An implementation of the DCCP protocol
5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
6  *
7  *      This program is free software; you can redistribute it and/or
8  *      modify it under the terms of the GNU General Public License
9  *      as published by the Free Software Foundation; either version
10  *      2 of the License, or (at your option) any later version.
11  */
12
13 #include <linux/dccp.h>
14 #include <linux/icmp.h>
15 #include <linux/module.h>
16 #include <linux/skbuff.h>
17 #include <linux/random.h>
18
19 #include <net/icmp.h>
20 #include <net/inet_common.h>
21 #include <net/inet_hashtables.h>
22 #include <net/inet_sock.h>
23 #include <net/protocol.h>
24 #include <net/sock.h>
25 #include <net/timewait_sock.h>
26 #include <net/tcp_states.h>
27 #include <net/xfrm.h>
28
29 #include "ackvec.h"
30 #include "ccid.h"
31 #include "dccp.h"
32 #include "feat.h"
33
34 /*
35  * This is the global socket data structure used for responding to
36  * the Out-of-the-blue (OOTB) packets. A control sock will be created
37  * for this socket at the initialization time.
38  */
39 static struct socket *dccp_v4_ctl_socket;
40
41 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
42 {
43         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
44                                  inet_csk_bind_conflict);
45 }
46
47 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
48 {
49         struct inet_sock *inet = inet_sk(sk);
50         struct dccp_sock *dp = dccp_sk(sk);
51         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
52         struct rtable *rt;
53         __be32 daddr, nexthop;
54         int tmp;
55         int err;
56
57         dp->dccps_role = DCCP_ROLE_CLIENT;
58
59         if (addr_len < sizeof(struct sockaddr_in))
60                 return -EINVAL;
61
62         if (usin->sin_family != AF_INET)
63                 return -EAFNOSUPPORT;
64
65         nexthop = daddr = usin->sin_addr.s_addr;
66         if (inet->opt != NULL && inet->opt->srr) {
67                 if (daddr == 0)
68                         return -EINVAL;
69                 nexthop = inet->opt->faddr;
70         }
71
72         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
73                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
74                                IPPROTO_DCCP,
75                                inet->sport, usin->sin_port, sk, 1);
76         if (tmp < 0)
77                 return tmp;
78
79         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
80                 ip_rt_put(rt);
81                 return -ENETUNREACH;
82         }
83
84         if (inet->opt == NULL || !inet->opt->srr)
85                 daddr = rt->rt_dst;
86
87         if (inet->saddr == 0)
88                 inet->saddr = rt->rt_src;
89         inet->rcv_saddr = inet->saddr;
90
91         inet->dport = usin->sin_port;
92         inet->daddr = daddr;
93
94         inet_csk(sk)->icsk_ext_hdr_len = 0;
95         if (inet->opt != NULL)
96                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
97         /*
98          * Socket identity is still unknown (sport may be zero).
99          * However we set state to DCCP_REQUESTING and not releasing socket
100          * lock select source port, enter ourselves into the hash tables and
101          * complete initialization after this.
102          */
103         dccp_set_state(sk, DCCP_REQUESTING);
104         err = inet_hash_connect(&dccp_death_row, sk);
105         if (err != 0)
106                 goto failure;
107
108         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
109                                 sk);
110         if (err != 0)
111                 goto failure;
112
113         /* OK, now commit destination to socket.  */
114         sk_setup_caps(sk, &rt->u.dst);
115
116         dp->dccps_iss = secure_dccp_sequence_number(inet->saddr, inet->daddr,
117                                                     inet->sport, inet->dport);
118         inet->id = dp->dccps_iss ^ jiffies;
119
120         err = dccp_connect(sk);
121         rt = NULL;
122         if (err != 0)
123                 goto failure;
124 out:
125         return err;
126 failure:
127         /*
128          * This unhashes the socket and releases the local port, if necessary.
129          */
130         dccp_set_state(sk, DCCP_CLOSED);
131         ip_rt_put(rt);
132         sk->sk_route_caps = 0;
133         inet->dport = 0;
134         goto out;
135 }
136
137 EXPORT_SYMBOL_GPL(dccp_v4_connect);
138
139 /*
140  * This routine does path mtu discovery as defined in RFC1191.
141  */
142 static inline void dccp_do_pmtu_discovery(struct sock *sk,
143                                           const struct iphdr *iph,
144                                           u32 mtu)
145 {
146         struct dst_entry *dst;
147         const struct inet_sock *inet = inet_sk(sk);
148         const struct dccp_sock *dp = dccp_sk(sk);
149
150         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
151          * send out by Linux are always < 576bytes so they should go through
152          * unfragmented).
153          */
154         if (sk->sk_state == DCCP_LISTEN)
155                 return;
156
157         /* We don't check in the destentry if pmtu discovery is forbidden
158          * on this route. We just assume that no packet_to_big packets
159          * are send back when pmtu discovery is not active.
160          * There is a small race when the user changes this flag in the
161          * route, but I think that's acceptable.
162          */
163         if ((dst = __sk_dst_check(sk, 0)) == NULL)
164                 return;
165
166         dst->ops->update_pmtu(dst, mtu);
167
168         /* Something is about to be wrong... Remember soft error
169          * for the case, if this connection will not able to recover.
170          */
171         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
172                 sk->sk_err_soft = EMSGSIZE;
173
174         mtu = dst_mtu(dst);
175
176         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
177             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
178                 dccp_sync_mss(sk, mtu);
179
180                 /*
181                  * From RFC 4340, sec. 14.1:
182                  *
183                  *      DCCP-Sync packets are the best choice for upward
184                  *      probing, since DCCP-Sync probes do not risk application
185                  *      data loss.
186                  */
187                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
188         } /* else let the usual retransmit timer handle it */
189 }
190
191 /*
192  * This routine is called by the ICMP module when it gets some sort of error
193  * condition. If err < 0 then the socket should be closed and the error
194  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
195  * After adjustment header points to the first 8 bytes of the tcp header. We
196  * need to find the appropriate port.
197  *
198  * The locking strategy used here is very "optimistic". When someone else
199  * accesses the socket the ICMP is just dropped and for some paths there is no
200  * check at all. A more general error queue to queue errors for later handling
201  * is probably better.
202  */
203 static void dccp_v4_err(struct sk_buff *skb, u32 info)
204 {
205         const struct iphdr *iph = (struct iphdr *)skb->data;
206         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
207                                                         (iph->ihl << 2));
208         struct dccp_sock *dp;
209         struct inet_sock *inet;
210         const int type = icmp_hdr(skb)->type;
211         const int code = icmp_hdr(skb)->code;
212         struct sock *sk;
213         __u64 seq;
214         int err;
215
216         if (skb->len < (iph->ihl << 2) + 8) {
217                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
218                 return;
219         }
220
221         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
222                          iph->saddr, dh->dccph_sport, inet_iif(skb));
223         if (sk == NULL) {
224                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
225                 return;
226         }
227
228         if (sk->sk_state == DCCP_TIME_WAIT) {
229                 inet_twsk_put(inet_twsk(sk));
230                 return;
231         }
232
233         bh_lock_sock(sk);
234         /* If too many ICMPs get dropped on busy
235          * servers this needs to be solved differently.
236          */
237         if (sock_owned_by_user(sk))
238                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
239
240         if (sk->sk_state == DCCP_CLOSED)
241                 goto out;
242
243         dp = dccp_sk(sk);
244         seq = dccp_hdr_seq(skb);
245         if (sk->sk_state != DCCP_LISTEN &&
246             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
247                 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
248                 goto out;
249         }
250
251         switch (type) {
252         case ICMP_SOURCE_QUENCH:
253                 /* Just silently ignore these. */
254                 goto out;
255         case ICMP_PARAMETERPROB:
256                 err = EPROTO;
257                 break;
258         case ICMP_DEST_UNREACH:
259                 if (code > NR_ICMP_UNREACH)
260                         goto out;
261
262                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
263                         if (!sock_owned_by_user(sk))
264                                 dccp_do_pmtu_discovery(sk, iph, info);
265                         goto out;
266                 }
267
268                 err = icmp_err_convert[code].errno;
269                 break;
270         case ICMP_TIME_EXCEEDED:
271                 err = EHOSTUNREACH;
272                 break;
273         default:
274                 goto out;
275         }
276
277         switch (sk->sk_state) {
278                 struct request_sock *req , **prev;
279         case DCCP_LISTEN:
280                 if (sock_owned_by_user(sk))
281                         goto out;
282                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
283                                           iph->daddr, iph->saddr);
284                 if (!req)
285                         goto out;
286
287                 /*
288                  * ICMPs are not backlogged, hence we cannot get an established
289                  * socket here.
290                  */
291                 BUG_TRAP(!req->sk);
292
293                 if (seq != dccp_rsk(req)->dreq_iss) {
294                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
295                         goto out;
296                 }
297                 /*
298                  * Still in RESPOND, just remove it silently.
299                  * There is no good way to pass the error to the newly
300                  * created socket, and POSIX does not want network
301                  * errors returned from accept().
302                  */
303                 inet_csk_reqsk_queue_drop(sk, req, prev);
304                 goto out;
305
306         case DCCP_REQUESTING:
307         case DCCP_RESPOND:
308                 if (!sock_owned_by_user(sk)) {
309                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
310                         sk->sk_err = err;
311
312                         sk->sk_error_report(sk);
313
314                         dccp_done(sk);
315                 } else
316                         sk->sk_err_soft = err;
317                 goto out;
318         }
319
320         /* If we've already connected we will keep trying
321          * until we time out, or the user gives up.
322          *
323          * rfc1122 4.2.3.9 allows to consider as hard errors
324          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
325          * but it is obsoleted by pmtu discovery).
326          *
327          * Note, that in modern internet, where routing is unreliable
328          * and in each dark corner broken firewalls sit, sending random
329          * errors ordered by their masters even this two messages finally lose
330          * their original sense (even Linux sends invalid PORT_UNREACHs)
331          *
332          * Now we are in compliance with RFCs.
333          *                                                      --ANK (980905)
334          */
335
336         inet = inet_sk(sk);
337         if (!sock_owned_by_user(sk) && inet->recverr) {
338                 sk->sk_err = err;
339                 sk->sk_error_report(sk);
340         } else /* Only an error on timeout */
341                 sk->sk_err_soft = err;
342 out:
343         bh_unlock_sock(sk);
344         sock_put(sk);
345 }
346
347 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
348                                       __be32 src, __be32 dst)
349 {
350         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
351 }
352
353 void dccp_v4_send_check(struct sock *sk, int unused, struct sk_buff *skb)
354 {
355         const struct inet_sock *inet = inet_sk(sk);
356         struct dccp_hdr *dh = dccp_hdr(skb);
357
358         dccp_csum_outgoing(skb);
359         dh->dccph_checksum = dccp_v4_csum_finish(skb, inet->saddr, inet->daddr);
360 }
361
362 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
363
364 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
365 {
366         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
367                                            ip_hdr(skb)->saddr,
368                                            dccp_hdr(skb)->dccph_dport,
369                                            dccp_hdr(skb)->dccph_sport);
370 }
371
372 /*
373  * The three way handshake has completed - we got a valid ACK or DATAACK -
374  * now create the new socket.
375  *
376  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
377  */
378 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
379                                        struct request_sock *req,
380                                        struct dst_entry *dst)
381 {
382         struct inet_request_sock *ireq;
383         struct inet_sock *newinet;
384         struct dccp_sock *newdp;
385         struct sock *newsk;
386
387         if (sk_acceptq_is_full(sk))
388                 goto exit_overflow;
389
390         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
391                 goto exit;
392
393         newsk = dccp_create_openreq_child(sk, req, skb);
394         if (newsk == NULL)
395                 goto exit;
396
397         sk_setup_caps(newsk, dst);
398
399         newdp              = dccp_sk(newsk);
400         newinet            = inet_sk(newsk);
401         ireq               = inet_rsk(req);
402         newinet->daddr     = ireq->rmt_addr;
403         newinet->rcv_saddr = ireq->loc_addr;
404         newinet->saddr     = ireq->loc_addr;
405         newinet->opt       = ireq->opt;
406         ireq->opt          = NULL;
407         newinet->mc_index  = inet_iif(skb);
408         newinet->mc_ttl    = ip_hdr(skb)->ttl;
409         newinet->id        = jiffies;
410
411         dccp_sync_mss(newsk, dst_mtu(dst));
412
413         __inet_hash(&dccp_hashinfo, newsk, 0);
414         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
415
416         return newsk;
417
418 exit_overflow:
419         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
420 exit:
421         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
422         dst_release(dst);
423         return NULL;
424 }
425
426 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
427
428 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
429 {
430         const struct dccp_hdr *dh = dccp_hdr(skb);
431         const struct iphdr *iph = ip_hdr(skb);
432         struct sock *nsk;
433         struct request_sock **prev;
434         /* Find possible connection requests. */
435         struct request_sock *req = inet_csk_search_req(sk, &prev,
436                                                        dh->dccph_sport,
437                                                        iph->saddr, iph->daddr);
438         if (req != NULL)
439                 return dccp_check_req(sk, skb, req, prev);
440
441         nsk = inet_lookup_established(&dccp_hashinfo,
442                                       iph->saddr, dh->dccph_sport,
443                                       iph->daddr, dh->dccph_dport,
444                                       inet_iif(skb));
445         if (nsk != NULL) {
446                 if (nsk->sk_state != DCCP_TIME_WAIT) {
447                         bh_lock_sock(nsk);
448                         return nsk;
449                 }
450                 inet_twsk_put(inet_twsk(nsk));
451                 return NULL;
452         }
453
454         return sk;
455 }
456
457 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
458                                            struct sk_buff *skb)
459 {
460         struct rtable *rt;
461         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
462                             .nl_u = { .ip4_u =
463                                       { .daddr = ip_hdr(skb)->saddr,
464                                         .saddr = ip_hdr(skb)->daddr,
465                                         .tos = RT_CONN_FLAGS(sk) } },
466                             .proto = sk->sk_protocol,
467                             .uli_u = { .ports =
468                                        { .sport = dccp_hdr(skb)->dccph_dport,
469                                          .dport = dccp_hdr(skb)->dccph_sport }
470                                      }
471                           };
472
473         security_skb_classify_flow(skb, &fl);
474         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
475                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
476                 return NULL;
477         }
478
479         return &rt->u.dst;
480 }
481
482 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
483                                  struct dst_entry *dst)
484 {
485         int err = -1;
486         struct sk_buff *skb;
487
488         /* First, grab a route. */
489
490         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
491                 goto out;
492
493         skb = dccp_make_response(sk, dst, req);
494         if (skb != NULL) {
495                 const struct inet_request_sock *ireq = inet_rsk(req);
496                 struct dccp_hdr *dh = dccp_hdr(skb);
497
498                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
499                                                               ireq->rmt_addr);
500                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
501                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
502                                             ireq->rmt_addr,
503                                             ireq->opt);
504                 err = net_xmit_eval(err);
505         }
506
507 out:
508         dst_release(dst);
509         return err;
510 }
511
512 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
513 {
514         int err;
515         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
516         const struct iphdr *rxiph;
517         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
518                                        sizeof(struct dccp_hdr_ext) +
519                                        sizeof(struct dccp_hdr_reset);
520         struct sk_buff *skb;
521         struct dst_entry *dst;
522         u64 seqno = 0;
523
524         /* Never send a reset in response to a reset. */
525         if (rxdh->dccph_type == DCCP_PKT_RESET)
526                 return;
527
528         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
529                 return;
530
531         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
532         if (dst == NULL)
533                 return;
534
535         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header,
536                         GFP_ATOMIC);
537         if (skb == NULL)
538                 goto out;
539
540         /* Reserve space for headers. */
541         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
542         skb->dst = dst_clone(dst);
543
544         dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
545
546         /* Build DCCP header and checksum it. */
547         dh->dccph_type     = DCCP_PKT_RESET;
548         dh->dccph_sport    = rxdh->dccph_dport;
549         dh->dccph_dport    = rxdh->dccph_sport;
550         dh->dccph_doff     = dccp_hdr_reset_len / 4;
551         dh->dccph_x        = 1;
552         dccp_hdr_reset(skb)->dccph_reset_code =
553                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
554
555         /* See "8.3.1. Abnormal Termination" in RFC 4340 */
556         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
557                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
558
559         dccp_hdr_set_seq(dh, seqno);
560         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), DCCP_SKB_CB(rxskb)->dccpd_seq);
561
562         dccp_csum_outgoing(skb);
563         rxiph = ip_hdr(rxskb);
564         dh->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
565                                                  rxiph->daddr);
566
567         bh_lock_sock(dccp_v4_ctl_socket->sk);
568         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
569                                     rxiph->daddr, rxiph->saddr, NULL);
570         bh_unlock_sock(dccp_v4_ctl_socket->sk);
571
572         if (net_xmit_eval(err) == 0) {
573                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
574                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
575         }
576 out:
577          dst_release(dst);
578 }
579
580 static void dccp_v4_reqsk_destructor(struct request_sock *req)
581 {
582         kfree(inet_rsk(req)->opt);
583 }
584
585 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
586         .family         = PF_INET,
587         .obj_size       = sizeof(struct dccp_request_sock),
588         .rtx_syn_ack    = dccp_v4_send_response,
589         .send_ack       = dccp_reqsk_send_ack,
590         .destructor     = dccp_v4_reqsk_destructor,
591         .send_reset     = dccp_v4_ctl_send_reset,
592 };
593
594 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
595 {
596         struct inet_request_sock *ireq;
597         struct request_sock *req;
598         struct dccp_request_sock *dreq;
599         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
600         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
601         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
602
603         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
604         if (((struct rtable *)skb->dst)->rt_flags &
605             (RTCF_BROADCAST | RTCF_MULTICAST)) {
606                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
607                 goto drop;
608         }
609
610         if (dccp_bad_service_code(sk, service)) {
611                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
612                 goto drop;
613         }
614         /*
615          * TW buckets are converted to open requests without
616          * limitations, they conserve resources and peer is
617          * evidently real one.
618          */
619         if (inet_csk_reqsk_queue_is_full(sk))
620                 goto drop;
621
622         /*
623          * Accept backlog is full. If we have already queued enough
624          * of warm entries in syn queue, drop request. It is better than
625          * clogging syn queue with openreqs with exponentially increasing
626          * timeout.
627          */
628         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
629                 goto drop;
630
631         req = reqsk_alloc(&dccp_request_sock_ops);
632         if (req == NULL)
633                 goto drop;
634
635         if (dccp_parse_options(sk, skb))
636                 goto drop_and_free;
637
638         dccp_reqsk_init(req, skb);
639
640         if (security_inet_conn_request(sk, skb, req))
641                 goto drop_and_free;
642
643         ireq = inet_rsk(req);
644         ireq->loc_addr = ip_hdr(skb)->daddr;
645         ireq->rmt_addr = ip_hdr(skb)->saddr;
646         ireq->opt       = NULL;
647
648         /*
649          * Step 3: Process LISTEN state
650          *
651          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
652          *
653          * In fact we defer setting S.GSR, S.SWL, S.SWH to
654          * dccp_create_openreq_child.
655          */
656         dreq = dccp_rsk(req);
657         dreq->dreq_isr     = dcb->dccpd_seq;
658         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
659         dreq->dreq_service = service;
660
661         if (dccp_v4_send_response(sk, req, NULL))
662                 goto drop_and_free;
663
664         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
665         return 0;
666
667 drop_and_free:
668         reqsk_free(req);
669 drop:
670         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
671         dcb->dccpd_reset_code = reset_code;
672         return -1;
673 }
674
675 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
676
677 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
678 {
679         struct dccp_hdr *dh = dccp_hdr(skb);
680
681         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
682                 if (dccp_rcv_established(sk, skb, dh, skb->len))
683                         goto reset;
684                 return 0;
685         }
686
687         /*
688          *  Step 3: Process LISTEN state
689          *       If P.type == Request or P contains a valid Init Cookie option,
690          *            (* Must scan the packet's options to check for Init
691          *               Cookies.  Only Init Cookies are processed here,
692          *               however; other options are processed in Step 8.  This
693          *               scan need only be performed if the endpoint uses Init
694          *               Cookies *)
695          *            (* Generate a new socket and switch to that socket *)
696          *            Set S := new socket for this port pair
697          *            S.state = RESPOND
698          *            Choose S.ISS (initial seqno) or set from Init Cookies
699          *            Initialize S.GAR := S.ISS
700          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
701          *            Continue with S.state == RESPOND
702          *            (* A Response packet will be generated in Step 11 *)
703          *       Otherwise,
704          *            Generate Reset(No Connection) unless P.type == Reset
705          *            Drop packet and return
706          *
707          * NOTE: the check for the packet types is done in
708          *       dccp_rcv_state_process
709          */
710         if (sk->sk_state == DCCP_LISTEN) {
711                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
712
713                 if (nsk == NULL)
714                         goto discard;
715
716                 if (nsk != sk) {
717                         if (dccp_child_process(sk, nsk, skb))
718                                 goto reset;
719                         return 0;
720                 }
721         }
722
723         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
724                 goto reset;
725         return 0;
726
727 reset:
728         dccp_v4_ctl_send_reset(sk, skb);
729 discard:
730         kfree_skb(skb);
731         return 0;
732 }
733
734 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
735
736 /**
737  *      dccp_invalid_packet  -  check for malformed packets
738  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
739  *      Packets that fail these checks are ignored and do not receive Resets.
740  */
741 int dccp_invalid_packet(struct sk_buff *skb)
742 {
743         const struct dccp_hdr *dh;
744         unsigned int cscov;
745
746         if (skb->pkt_type != PACKET_HOST)
747                 return 1;
748
749         /* If the packet is shorter than 12 bytes, drop packet and return */
750         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
751                 DCCP_WARN("pskb_may_pull failed\n");
752                 return 1;
753         }
754
755         dh = dccp_hdr(skb);
756
757         /* If P.type is not understood, drop packet and return */
758         if (dh->dccph_type >= DCCP_PKT_INVALID) {
759                 DCCP_WARN("invalid packet type\n");
760                 return 1;
761         }
762
763         /*
764          * If P.Data Offset is too small for packet type, drop packet and return
765          */
766         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
767                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
768                 return 1;
769         }
770         /*
771          * If P.Data Offset is too too large for packet, drop packet and return
772          */
773         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
774                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
775                 return 1;
776         }
777
778         /*
779          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
780          * has short sequence numbers), drop packet and return
781          */
782         if (dh->dccph_type >= DCCP_PKT_DATA    &&
783             dh->dccph_type <= DCCP_PKT_DATAACK && dh->dccph_x == 0)  {
784                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
785                           dccp_packet_name(dh->dccph_type));
786                 return 1;
787         }
788
789         /*
790          * If P.CsCov is too large for the packet size, drop packet and return.
791          * This must come _before_ checksumming (not as RFC 4340 suggests).
792          */
793         cscov = dccp_csum_coverage(skb);
794         if (cscov > skb->len) {
795                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
796                           dh->dccph_cscov, skb->len);
797                 return 1;
798         }
799
800         /* If header checksum is incorrect, drop packet and return.
801          * (This step is completed in the AF-dependent functions.) */
802         skb->csum = skb_checksum(skb, 0, cscov, 0);
803
804         return 0;
805 }
806
807 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
808
809 /* this is called when real data arrives */
810 static int dccp_v4_rcv(struct sk_buff *skb)
811 {
812         const struct dccp_hdr *dh;
813         const struct iphdr *iph;
814         struct sock *sk;
815         int min_cov;
816
817         /* Step 1: Check header basics */
818
819         if (dccp_invalid_packet(skb))
820                 goto discard_it;
821
822         iph = ip_hdr(skb);
823         /* Step 1: If header checksum is incorrect, drop packet and return */
824         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
825                 DCCP_WARN("dropped packet with invalid checksum\n");
826                 goto discard_it;
827         }
828
829         dh = dccp_hdr(skb);
830
831         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
832         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
833
834         dccp_pr_debug("%8.8s "
835                       "src=%u.%u.%u.%u@%-5d "
836                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
837                       dccp_packet_name(dh->dccph_type),
838                       NIPQUAD(iph->saddr), ntohs(dh->dccph_sport),
839                       NIPQUAD(iph->daddr), ntohs(dh->dccph_dport),
840                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
841
842         if (dccp_packet_without_ack(skb)) {
843                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
844                 dccp_pr_debug_cat("\n");
845         } else {
846                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
847                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
848                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
849         }
850
851         /* Step 2:
852          *      Look up flow ID in table and get corresponding socket */
853         sk = __inet_lookup(&dccp_hashinfo,
854                            iph->saddr, dh->dccph_sport,
855                            iph->daddr, dh->dccph_dport, inet_iif(skb));
856         /*
857          * Step 2:
858          *      If no socket ...
859          */
860         if (sk == NULL) {
861                 dccp_pr_debug("failed to look up flow ID in table and "
862                               "get corresponding socket\n");
863                 goto no_dccp_socket;
864         }
865
866         /*
867          * Step 2:
868          *      ... or S.state == TIMEWAIT,
869          *              Generate Reset(No Connection) unless P.type == Reset
870          *              Drop packet and return
871          */
872         if (sk->sk_state == DCCP_TIME_WAIT) {
873                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
874                 inet_twsk_put(inet_twsk(sk));
875                 goto no_dccp_socket;
876         }
877
878         /*
879          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
880          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
881          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
882          */
883         min_cov = dccp_sk(sk)->dccps_pcrlen;
884         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
885                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
886                               dh->dccph_cscov, min_cov);
887                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
888                  *         options (Section 11.7) with Drop Code 0, Protocol
889                  *         Constraints."                                     */
890                 goto discard_and_relse;
891         }
892
893         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
894                 goto discard_and_relse;
895         nf_reset(skb);
896
897         return sk_receive_skb(sk, skb, 1);
898
899 no_dccp_socket:
900         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
901                 goto discard_it;
902         /*
903          * Step 2:
904          *      If no socket ...
905          *              Generate Reset(No Connection) unless P.type == Reset
906          *              Drop packet and return
907          */
908         if (dh->dccph_type != DCCP_PKT_RESET) {
909                 DCCP_SKB_CB(skb)->dccpd_reset_code =
910                                         DCCP_RESET_CODE_NO_CONNECTION;
911                 dccp_v4_ctl_send_reset(sk, skb);
912         }
913
914 discard_it:
915         kfree_skb(skb);
916         return 0;
917
918 discard_and_relse:
919         sock_put(sk);
920         goto discard_it;
921 }
922
923 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
924         .queue_xmit        = ip_queue_xmit,
925         .send_check        = dccp_v4_send_check,
926         .rebuild_header    = inet_sk_rebuild_header,
927         .conn_request      = dccp_v4_conn_request,
928         .syn_recv_sock     = dccp_v4_request_recv_sock,
929         .net_header_len    = sizeof(struct iphdr),
930         .setsockopt        = ip_setsockopt,
931         .getsockopt        = ip_getsockopt,
932         .addr2sockaddr     = inet_csk_addr2sockaddr,
933         .sockaddr_len      = sizeof(struct sockaddr_in),
934 #ifdef CONFIG_COMPAT
935         .compat_setsockopt = compat_ip_setsockopt,
936         .compat_getsockopt = compat_ip_getsockopt,
937 #endif
938 };
939
940 static int dccp_v4_init_sock(struct sock *sk)
941 {
942         static __u8 dccp_v4_ctl_sock_initialized;
943         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
944
945         if (err == 0) {
946                 if (unlikely(!dccp_v4_ctl_sock_initialized))
947                         dccp_v4_ctl_sock_initialized = 1;
948                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
949         }
950
951         return err;
952 }
953
954 static struct timewait_sock_ops dccp_timewait_sock_ops = {
955         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
956 };
957
958 static struct proto dccp_v4_prot = {
959         .name                   = "DCCP",
960         .owner                  = THIS_MODULE,
961         .close                  = dccp_close,
962         .connect                = dccp_v4_connect,
963         .disconnect             = dccp_disconnect,
964         .ioctl                  = dccp_ioctl,
965         .init                   = dccp_v4_init_sock,
966         .setsockopt             = dccp_setsockopt,
967         .getsockopt             = dccp_getsockopt,
968         .sendmsg                = dccp_sendmsg,
969         .recvmsg                = dccp_recvmsg,
970         .backlog_rcv            = dccp_v4_do_rcv,
971         .hash                   = dccp_hash,
972         .unhash                 = dccp_unhash,
973         .accept                 = inet_csk_accept,
974         .get_port               = dccp_v4_get_port,
975         .shutdown               = dccp_shutdown,
976         .destroy                = dccp_destroy_sock,
977         .orphan_count           = &dccp_orphan_count,
978         .max_header             = MAX_DCCP_HEADER,
979         .obj_size               = sizeof(struct dccp_sock),
980         .rsk_prot               = &dccp_request_sock_ops,
981         .twsk_prot              = &dccp_timewait_sock_ops,
982 #ifdef CONFIG_COMPAT
983         .compat_setsockopt      = compat_dccp_setsockopt,
984         .compat_getsockopt      = compat_dccp_getsockopt,
985 #endif
986 };
987
988 static struct net_protocol dccp_v4_protocol = {
989         .handler        = dccp_v4_rcv,
990         .err_handler    = dccp_v4_err,
991         .no_policy      = 1,
992 };
993
994 static const struct proto_ops inet_dccp_ops = {
995         .family            = PF_INET,
996         .owner             = THIS_MODULE,
997         .release           = inet_release,
998         .bind              = inet_bind,
999         .connect           = inet_stream_connect,
1000         .socketpair        = sock_no_socketpair,
1001         .accept            = inet_accept,
1002         .getname           = inet_getname,
1003         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1004         .poll              = dccp_poll,
1005         .ioctl             = inet_ioctl,
1006         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1007         .listen            = inet_dccp_listen,
1008         .shutdown          = inet_shutdown,
1009         .setsockopt        = sock_common_setsockopt,
1010         .getsockopt        = sock_common_getsockopt,
1011         .sendmsg           = inet_sendmsg,
1012         .recvmsg           = sock_common_recvmsg,
1013         .mmap              = sock_no_mmap,
1014         .sendpage          = sock_no_sendpage,
1015 #ifdef CONFIG_COMPAT
1016         .compat_setsockopt = compat_sock_common_setsockopt,
1017         .compat_getsockopt = compat_sock_common_getsockopt,
1018 #endif
1019 };
1020
1021 static struct inet_protosw dccp_v4_protosw = {
1022         .type           = SOCK_DCCP,
1023         .protocol       = IPPROTO_DCCP,
1024         .prot           = &dccp_v4_prot,
1025         .ops            = &inet_dccp_ops,
1026         .capability     = -1,
1027         .no_check       = 0,
1028         .flags          = INET_PROTOSW_ICSK,
1029 };
1030
1031 static int __init dccp_v4_init(void)
1032 {
1033         int err = proto_register(&dccp_v4_prot, 1);
1034
1035         if (err != 0)
1036                 goto out;
1037
1038         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1039         if (err != 0)
1040                 goto out_proto_unregister;
1041
1042         inet_register_protosw(&dccp_v4_protosw);
1043
1044         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1045                                        SOCK_DCCP, IPPROTO_DCCP);
1046         if (err)
1047                 goto out_unregister_protosw;
1048 out:
1049         return err;
1050 out_unregister_protosw:
1051         inet_unregister_protosw(&dccp_v4_protosw);
1052         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1053 out_proto_unregister:
1054         proto_unregister(&dccp_v4_prot);
1055         goto out;
1056 }
1057
1058 static void __exit dccp_v4_exit(void)
1059 {
1060         inet_unregister_protosw(&dccp_v4_protosw);
1061         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1062         proto_unregister(&dccp_v4_prot);
1063 }
1064
1065 module_init(dccp_v4_init);
1066 module_exit(dccp_v4_exit);
1067
1068 /*
1069  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1070  * values directly, Also cover the case where the protocol is not specified,
1071  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1072  */
1073 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1074 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1075 MODULE_LICENSE("GPL");
1076 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1077 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");