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/config.h>
  14 #include <linux/dccp.h>
  15 #include <linux/icmp.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
  30 #include "ackvec.h"
  31 #include "ccid.h"
  32 #include "dccp.h"
  33 #include "feat.h"
  34
  35 /*
  36  * This is the global socket data structure used for responding to
  37  * the Out-of-the-blue (OOTB) packets. A control sock will be created
  38  * for this socket at the initialization time.
  39  */
  40 static struct socket *dccp_v4_ctl_socket;
  41
  42 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  43 {
  44         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
  45                                  inet_csk_bind_conflict);
  46 }
  47
  48 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  49 {
  50         struct inet_sock *inet = inet_sk(sk);
  51         struct dccp_sock *dp = dccp_sk(sk);
  52         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  53         struct rtable *rt;
  54         u32 daddr, nexthop;
  55         int tmp;
  56         int err;
  57
  58         dp->dccps_role = DCCP_ROLE_CLIENT;
  59
  60         if (dccp_service_not_initialized(sk))
  61                 return -EPROTO;
  62
  63         if (addr_len < sizeof(struct sockaddr_in))
  64                 return -EINVAL;
  65
  66         if (usin->sin_family != AF_INET)
  67                 return -EAFNOSUPPORT;
  68
  69         nexthop = daddr = usin->sin_addr.s_addr;
  70         if (inet->opt != NULL && inet->opt->srr) {
  71                 if (daddr == 0)
  72                         return -EINVAL;
  73                 nexthop = inet->opt->faddr;
  74         }
  75
  76         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
  77                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  78                                IPPROTO_DCCP,
  79                                inet->sport, usin->sin_port, sk);
  80         if (tmp < 0)
  81                 return tmp;
  82
  83         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  84                 ip_rt_put(rt);
  85                 return -ENETUNREACH;
  86         }
  87
  88         if (inet->opt == NULL || !inet->opt->srr)
  89                 daddr = rt->rt_dst;
  90
  91         if (inet->saddr == 0)
  92                 inet->saddr = rt->rt_src;
  93         inet->rcv_saddr = inet->saddr;
  94
  95         inet->dport = usin->sin_port;
  96         inet->daddr = daddr;
  97
  98         inet_csk(sk)->icsk_ext_hdr_len = 0;
  99         if (inet->opt != NULL)
 100                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
 101         /*
 102          * Socket identity is still unknown (sport may be zero).
 103          * However we set state to DCCP_REQUESTING and not releasing socket
 104          * lock select source port, enter ourselves into the hash tables and
 105          * complete initialization after this.
 106          */
 107         dccp_set_state(sk, DCCP_REQUESTING);
 108         err = inet_hash_connect(&dccp_death_row, sk);
 109         if (err != 0)
 110                 goto failure;
 111
 112         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
 113                                 sk);
 114         if (err != 0)
 115                 goto failure;
 116
 117         /* OK, now commit destination to socket.  */
 118         sk_setup_caps(sk, &rt->u.dst);
 119
 120         dp->dccps_gar =
 121                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 122                                                             inet->daddr,
 123                                                             inet->sport,
 124                                                             usin->sin_port);
 125         dccp_update_gss(sk, dp->dccps_iss);
 126
 127         inet->id = dp->dccps_iss ^ jiffies;
 128
 129         err = dccp_connect(sk);
 130         rt = NULL;
 131         if (err != 0)
 132                 goto failure;
 133 out:
 134         return err;
 135 failure:
 136         /*
 137          * This unhashes the socket and releases the local port, if necessary.
 138          */
 139         dccp_set_state(sk, DCCP_CLOSED);
 140         ip_rt_put(rt);
 141         sk->sk_route_caps = 0;
 142         inet->dport = 0;
 143         goto out;
 144 }
 145
 146 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 147
 148 /*
 149  * This routine does path mtu discovery as defined in RFC1191.
 150  */
 151 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 152                                           const struct iphdr *iph,
 153                                           u32 mtu)
 154 {
 155         struct dst_entry *dst;
 156         const struct inet_sock *inet = inet_sk(sk);
 157         const struct dccp_sock *dp = dccp_sk(sk);
 158
 159         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 160          * send out by Linux are always < 576bytes so they should go through
 161          * unfragmented).
 162          */
 163         if (sk->sk_state == DCCP_LISTEN)
 164                 return;
 165
 166         /* We don't check in the destentry if pmtu discovery is forbidden
 167          * on this route. We just assume that no packet_to_big packets
 168          * are send back when pmtu discovery is not active.
 169          * There is a small race when the user changes this flag in the
 170          * route, but I think that's acceptable.
 171          */
 172         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 173                 return;
 174
 175         dst->ops->update_pmtu(dst, mtu);
 176
 177         /* Something is about to be wrong... Remember soft error
 178          * for the case, if this connection will not able to recover.
 179          */
 180         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 181                 sk->sk_err_soft = EMSGSIZE;
 182
 183         mtu = dst_mtu(dst);
 184
 185         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 186             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 187                 dccp_sync_mss(sk, mtu);
 188
 189                 /*
 190                  * From: draft-ietf-dccp-spec-11.txt
 191                  *
 192                  *      DCCP-Sync packets are the best choice for upward
 193                  *      probing, since DCCP-Sync probes do not risk application
 194                  *      data loss.
 195                  */
 196                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 197         } /* else let the usual retransmit timer handle it */
 198 }
 199
 200 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 201 {
 202         int err;
 203         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 204         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 205                                      sizeof(struct dccp_hdr_ext) +
 206                                      sizeof(struct dccp_hdr_ack_bits);
 207         struct sk_buff *skb;
 208
 209         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 210                 return;
 211
 212         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 213         if (skb == NULL)
 214                 return;
 215
 216         /* Reserve space for headers. */
 217         skb_reserve(skb, MAX_DCCP_HEADER);
 218
 219         skb->dst = dst_clone(rxskb->dst);
 220
 221         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 222         dh = dccp_hdr(skb);
 223         memset(dh, 0, dccp_hdr_ack_len);
 224
 225         /* Build DCCP header and checksum it. */
 226         dh->dccph_type     = DCCP_PKT_ACK;
 227         dh->dccph_sport    = rxdh->dccph_dport;
 228         dh->dccph_dport    = rxdh->dccph_sport;
 229         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 230         dh->dccph_x        = 1;
 231
 232         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 233         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 234                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 235
 236         bh_lock_sock(dccp_v4_ctl_socket->sk);
 237         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
 238                                     rxskb->nh.iph->daddr,
 239                                     rxskb->nh.iph->saddr, NULL);
 240         bh_unlock_sock(dccp_v4_ctl_socket->sk);
 241
 242         if (err == NET_XMIT_CN || err == 0) {
 243                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 244                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 245         }
 246 }
 247
 248 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 249                                    struct request_sock *req)
 250 {
 251         dccp_v4_ctl_send_ack(skb);
 252 }
 253
 254 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 255                                  struct dst_entry *dst)
 256 {
 257         int err = -1;
 258         struct sk_buff *skb;
 259
 260         /* First, grab a route. */
 261
 262         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 263                 goto out;
 264
 265         skb = dccp_make_response(sk, dst, req);
 266         if (skb != NULL) {
 267                 const struct inet_request_sock *ireq = inet_rsk(req);
 268                 struct dccp_hdr *dh = dccp_hdr(skb);
 269
 270                 dh->dccph_checksum = dccp_v4_checksum(skb, ireq->loc_addr,
 271                                                       ireq->rmt_addr);
 272                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 273                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 274                                             ireq->rmt_addr,
 275                                             ireq->opt);
 276                 if (err == NET_XMIT_CN)
 277                         err = 0;
 278         }
 279
 280 out:
 281         dst_release(dst);
 282         return err;
 283 }
 284
 285 /*
 286  * This routine is called by the ICMP module when it gets some sort of error
 287  * condition. If err < 0 then the socket should be closed and the error
 288  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 289  * After adjustment header points to the first 8 bytes of the tcp header. We
 290  * need to find the appropriate port.
 291  *
 292  * The locking strategy used here is very "optimistic". When someone else
 293  * accesses the socket the ICMP is just dropped and for some paths there is no
 294  * check at all. A more general error queue to queue errors for later handling
 295  * is probably better.
 296  */
 297 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 298 {
 299         const struct iphdr *iph = (struct iphdr *)skb->data;
 300         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 301                                                         (iph->ihl << 2));
 302         struct dccp_sock *dp;
 303         struct inet_sock *inet;
 304         const int type = skb->h.icmph->type;
 305         const int code = skb->h.icmph->code;
 306         struct sock *sk;
 307         __u64 seq;
 308         int err;
 309
 310         if (skb->len < (iph->ihl << 2) + 8) {
 311                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 312                 return;
 313         }
 314
 315         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 316                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 317         if (sk == NULL) {
 318                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 319                 return;
 320         }
 321
 322         if (sk->sk_state == DCCP_TIME_WAIT) {
 323                 inet_twsk_put((struct inet_timewait_sock *)sk);
 324                 return;
 325         }
 326
 327         bh_lock_sock(sk);
 328         /* If too many ICMPs get dropped on busy
 329          * servers this needs to be solved differently.
 330          */
 331         if (sock_owned_by_user(sk))
 332                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 333
 334         if (sk->sk_state == DCCP_CLOSED)
 335                 goto out;
 336
 337         dp = dccp_sk(sk);
 338         seq = dccp_hdr_seq(skb);
 339         if (sk->sk_state != DCCP_LISTEN &&
 340             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 341                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 342                 goto out;
 343         }
 344
 345         switch (type) {
 346         case ICMP_SOURCE_QUENCH:
 347                 /* Just silently ignore these. */
 348                 goto out;
 349         case ICMP_PARAMETERPROB:
 350                 err = EPROTO;
 351                 break;
 352         case ICMP_DEST_UNREACH:
 353                 if (code > NR_ICMP_UNREACH)
 354                         goto out;
 355
 356                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 357                         if (!sock_owned_by_user(sk))
 358                                 dccp_do_pmtu_discovery(sk, iph, info);
 359                         goto out;
 360                 }
 361
 362                 err = icmp_err_convert[code].errno;
 363                 break;
 364         case ICMP_TIME_EXCEEDED:
 365                 err = EHOSTUNREACH;
 366                 break;
 367         default:
 368                 goto out;
 369         }
 370
 371         switch (sk->sk_state) {
 372                 struct request_sock *req , **prev;
 373         case DCCP_LISTEN:
 374                 if (sock_owned_by_user(sk))
 375                         goto out;
 376                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 377                                           iph->daddr, iph->saddr);
 378                 if (!req)
 379                         goto out;
 380
 381                 /*
 382                  * ICMPs are not backlogged, hence we cannot get an established
 383                  * socket here.
 384                  */
 385                 BUG_TRAP(!req->sk);
 386
 387                 if (seq != dccp_rsk(req)->dreq_iss) {
 388                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 389                         goto out;
 390                 }
 391                 /*
 392                  * Still in RESPOND, just remove it silently.
 393                  * There is no good way to pass the error to the newly
 394                  * created socket, and POSIX does not want network
 395                  * errors returned from accept().
 396                  */
 397                 inet_csk_reqsk_queue_drop(sk, req, prev);
 398                 goto out;
 399
 400         case DCCP_REQUESTING:
 401         case DCCP_RESPOND:
 402                 if (!sock_owned_by_user(sk)) {
 403                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 404                         sk->sk_err = err;
 405
 406                         sk->sk_error_report(sk);
 407
 408                         dccp_done(sk);
 409                 } else
 410                         sk->sk_err_soft = err;
 411                 goto out;
 412         }
 413
 414         /* If we've already connected we will keep trying
 415          * until we time out, or the user gives up.
 416          *
 417          * rfc1122 4.2.3.9 allows to consider as hard errors
 418          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 419          * but it is obsoleted by pmtu discovery).
 420          *
 421          * Note, that in modern internet, where routing is unreliable
 422          * and in each dark corner broken firewalls sit, sending random
 423          * errors ordered by their masters even this two messages finally lose
 424          * their original sense (even Linux sends invalid PORT_UNREACHs)
 425          *
 426          * Now we are in compliance with RFCs.
 427          *                                                      --ANK (980905)
 428          */
 429
 430         inet = inet_sk(sk);
 431         if (!sock_owned_by_user(sk) && inet->recverr) {
 432                 sk->sk_err = err;
 433                 sk->sk_error_report(sk);
 434         } else /* Only an error on timeout */
 435                 sk->sk_err_soft = err;
 436 out:
 437         bh_unlock_sock(sk);
 438         sock_put(sk);
 439 }
 440
 441 /* This routine computes an IPv4 DCCP checksum. */
 442 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 443 {
 444         const struct inet_sock *inet = inet_sk(sk);
 445         struct dccp_hdr *dh = dccp_hdr(skb);
 446
 447         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 448 }
 449
 450 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 451
 452 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 453                                         const struct sk_buff *skb)
 454 {
 455         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 456                                            skb->nh.iph->saddr,
 457                                            dccp_hdr(skb)->dccph_dport,
 458                                            dccp_hdr(skb)->dccph_sport);
 459 }
 460
 461 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 462 {
 463         struct inet_request_sock *ireq;
 464         struct dccp_sock dp;
 465         struct request_sock *req;
 466         struct dccp_request_sock *dreq;
 467         const __be32 saddr = skb->nh.iph->saddr;
 468         const __be32 daddr = skb->nh.iph->daddr;
 469         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 470         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 471         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 472
 473         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 474         if (((struct rtable *)skb->dst)->rt_flags &
 475             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 476                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 477                 goto drop;
 478         }
 479
 480         if (dccp_bad_service_code(sk, service)) {
 481                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 482                 goto drop;
 483         }
 484         /*
 485          * TW buckets are converted to open requests without
 486          * limitations, they conserve resources and peer is
 487          * evidently real one.
 488          */
 489         if (inet_csk_reqsk_queue_is_full(sk))
 490                 goto drop;
 491
 492         /*
 493          * Accept backlog is full. If we have already queued enough
 494          * of warm entries in syn queue, drop request. It is better than
 495          * clogging syn queue with openreqs with exponentially increasing
 496          * timeout.
 497          */
 498         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 499                 goto drop;
 500
 501         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 502         if (req == NULL)
 503                 goto drop;
 504
 505         if (dccp_parse_options(sk, skb))
 506                 goto drop;
 507
 508         dccp_openreq_init(req, &dp, skb);
 509
 510         ireq = inet_rsk(req);
 511         ireq->loc_addr = daddr;
 512         ireq->rmt_addr = saddr;
 513         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 514                                   right value */
 515         ireq->opt       = NULL;
 516
 517         /*
 518          * Step 3: Process LISTEN state
 519          *
 520          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 521          *
 522          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 523          * dccp_create_openreq_child.
 524          */
 525         dreq = dccp_rsk(req);
 526         dreq->dreq_isr     = dcb->dccpd_seq;
 527         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 528         dreq->dreq_service = service;
 529
 530         if (dccp_v4_send_response(sk, req, NULL))
 531                 goto drop_and_free;
 532
 533         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 534         return 0;
 535
 536 drop_and_free:
 537         reqsk_free(req);
 538 drop:
 539         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 540         dcb->dccpd_reset_code = reset_code;
 541         return -1;
 542 }
 543
 544 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 545
 546 /*
 547  * The three way handshake has completed - we got a valid ACK or DATAACK -
 548  * now create the new socket.
 549  *
 550  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 551  */
 552 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 553                                        struct request_sock *req,
 554                                        struct dst_entry *dst)
 555 {
 556         struct inet_request_sock *ireq;
 557         struct inet_sock *newinet;
 558         struct dccp_sock *newdp;
 559         struct sock *newsk;
 560
 561         if (sk_acceptq_is_full(sk))
 562                 goto exit_overflow;
 563
 564         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 565                 goto exit;
 566
 567         newsk = dccp_create_openreq_child(sk, req, skb);
 568         if (newsk == NULL)
 569                 goto exit;
 570
 571         sk_setup_caps(newsk, dst);
 572
 573         newdp              = dccp_sk(newsk);
 574         newinet            = inet_sk(newsk);
 575         ireq               = inet_rsk(req);
 576         newinet->daddr     = ireq->rmt_addr;
 577         newinet->rcv_saddr = ireq->loc_addr;
 578         newinet->saddr     = ireq->loc_addr;
 579         newinet->opt       = ireq->opt;
 580         ireq->opt          = NULL;
 581         newinet->mc_index  = inet_iif(skb);
 582         newinet->mc_ttl    = skb->nh.iph->ttl;
 583         newinet->id        = jiffies;
 584
 585         dccp_sync_mss(newsk, dst_mtu(dst));
 586
 587         __inet_hash(&dccp_hashinfo, newsk, 0);
 588         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 589
 590         return newsk;
 591
 592 exit_overflow:
 593         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 594 exit:
 595         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 596         dst_release(dst);
 597         return NULL;
 598 }
 599
 600 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 601
 602 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 603 {
 604         const struct dccp_hdr *dh = dccp_hdr(skb);
 605         const struct iphdr *iph = skb->nh.iph;
 606         struct sock *nsk;
 607         struct request_sock **prev;
 608         /* Find possible connection requests. */
 609         struct request_sock *req = inet_csk_search_req(sk, &prev,
 610                                                        dh->dccph_sport,
 611                                                        iph->saddr, iph->daddr);
 612         if (req != NULL)
 613                 return dccp_check_req(sk, skb, req, prev);
 614
 615         nsk = __inet_lookup_established(&dccp_hashinfo,
 616                                         iph->saddr, dh->dccph_sport,
 617                                         iph->daddr, ntohs(dh->dccph_dport),
 618                                         inet_iif(skb));
 619         if (nsk != NULL) {
 620                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 621                         bh_lock_sock(nsk);
 622                         return nsk;
 623                 }
 624                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 625                 return NULL;
 626         }
 627
 628         return sk;
 629 }
 630
 631 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
 632                      const __be32 daddr)
 633 {
 634         const struct dccp_hdr* dh = dccp_hdr(skb);
 635         int checksum_len;
 636         u32 tmp;
 637
 638         if (dh->dccph_cscov == 0)
 639                 checksum_len = skb->len;
 640         else {
 641                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 642                 checksum_len = checksum_len < skb->len ? checksum_len :
 643                                                          skb->len;
 644         }
 645
 646         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 647         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 648                                  IPPROTO_DCCP, tmp);
 649 }
 650
 651 EXPORT_SYMBOL_GPL(dccp_v4_checksum);
 652
 653 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 654                                    const __be32 saddr, const __be32 daddr)
 655 {
 656         struct dccp_hdr *dh = dccp_hdr(skb);
 657         int checksum_len;
 658         u32 tmp;
 659
 660         if (dh->dccph_cscov == 0)
 661                 checksum_len = skb->len;
 662         else {
 663                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 664                 checksum_len = checksum_len < skb->len ? checksum_len :
 665                                                          skb->len;
 666         }
 667         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 668         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 669                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 670 }
 671
 672 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 673                                            struct sk_buff *skb)
 674 {
 675         struct rtable *rt;
 676         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 677                             .nl_u = { .ip4_u =
 678                                       { .daddr = skb->nh.iph->saddr,
 679                                         .saddr = skb->nh.iph->daddr,
 680                                         .tos = RT_CONN_FLAGS(sk) } },
 681                             .proto = sk->sk_protocol,
 682                             .uli_u = { .ports =
 683                                        { .sport = dccp_hdr(skb)->dccph_dport,
 684                                          .dport = dccp_hdr(skb)->dccph_sport }
 685                                      }
 686                           };
 687
 688         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 689                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 690                 return NULL;
 691         }
 692
 693         return &rt->u.dst;
 694 }
 695
 696 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 697 {
 698         int err;
 699         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 700         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 701                                        sizeof(struct dccp_hdr_ext) +
 702                                        sizeof(struct dccp_hdr_reset);
 703         struct sk_buff *skb;
 704         struct dst_entry *dst;
 705         u64 seqno;
 706
 707         /* Never send a reset in response to a reset. */
 708         if (rxdh->dccph_type == DCCP_PKT_RESET)
 709                 return;
 710
 711         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 712                 return;
 713
 714         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
 715         if (dst == NULL)
 716                 return;
 717
 718         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 719         if (skb == NULL)
 720                 goto out;
 721
 722         /* Reserve space for headers. */
 723         skb_reserve(skb, MAX_DCCP_HEADER);
 724         skb->dst = dst_clone(dst);
 725
 726         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 727         dh = dccp_hdr(skb);
 728         memset(dh, 0, dccp_hdr_reset_len);
 729
 730         /* Build DCCP header and checksum it. */
 731         dh->dccph_type     = DCCP_PKT_RESET;
 732         dh->dccph_sport    = rxdh->dccph_dport;
 733         dh->dccph_dport    = rxdh->dccph_sport;
 734         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 735         dh->dccph_x        = 1;
 736         dccp_hdr_reset(skb)->dccph_reset_code =
 737                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 738
 739         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 740         seqno = 0;
 741         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 742                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 743
 744         dccp_hdr_set_seq(dh, seqno);
 745         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 746                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 747
 748         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 749                                               rxskb->nh.iph->daddr);
 750
 751         bh_lock_sock(dccp_v4_ctl_socket->sk);
 752         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
 753                                     rxskb->nh.iph->daddr,
 754                                     rxskb->nh.iph->saddr, NULL);
 755         bh_unlock_sock(dccp_v4_ctl_socket->sk);
 756
 757         if (err == NET_XMIT_CN || err == 0) {
 758                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 759                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 760         }
 761 out:
 762          dst_release(dst);
 763 }
 764
 765 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 766 {
 767         struct dccp_hdr *dh = dccp_hdr(skb);
 768
 769         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 770                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 771                         goto reset;
 772                 return 0;
 773         }
 774
 775         /*
 776          *  Step 3: Process LISTEN state
 777          *     If S.state == LISTEN,
 778          *        If P.type == Request or P contains a valid Init Cookie
 779          *              option,
 780          *           * Must scan the packet's options to check for an Init
 781          *              Cookie.  Only the Init Cookie is processed here,
 782          *              however; other options are processed in Step 8.  This
 783          *              scan need only be performed if the endpoint uses Init
 784          *              Cookies *
 785          *           * Generate a new socket and switch to that socket *
 786          *           Set S := new socket for this port pair
 787          *           S.state = RESPOND
 788          *           Choose S.ISS (initial seqno) or set from Init Cookie
 789          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 790          *           Continue with S.state == RESPOND
 791          *           * A Response packet will be generated in Step 11 *
 792          *        Otherwise,
 793          *           Generate Reset(No Connection) unless P.type == Reset
 794          *           Drop packet and return
 795          *
 796          * NOTE: the check for the packet types is done in
 797          *       dccp_rcv_state_process
 798          */
 799         if (sk->sk_state == DCCP_LISTEN) {
 800                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 801
 802                 if (nsk == NULL)
 803                         goto discard;
 804
 805                 if (nsk != sk) {
 806                         if (dccp_child_process(sk, nsk, skb))
 807                                 goto reset;
 808                         return 0;
 809                 }
 810         }
 811
 812         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 813                 goto reset;
 814         return 0;
 815
 816 reset:
 817         dccp_v4_ctl_send_reset(skb);
 818 discard:
 819         kfree_skb(skb);
 820         return 0;
 821 }
 822
 823 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 824
 825 int dccp_invalid_packet(struct sk_buff *skb)
 826 {
 827         const struct dccp_hdr *dh;
 828
 829         if (skb->pkt_type != PACKET_HOST)
 830                 return 1;
 831
 832         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 833                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 834                 return 1;
 835         }
 836
 837         dh = dccp_hdr(skb);
 838
 839         /* If the packet type is not understood, drop packet and return */
 840         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 841                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 842                 return 1;
 843         }
 844
 845         /*
 846          * If P.Data Offset is too small for packet type, or too large for
 847          * packet, drop packet and return
 848          */
 849         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 850                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 851                                             "too small 1\n",
 852                                dh->dccph_doff);
 853                 return 1;
 854         }
 855
 856         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 857                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 858                                             "too small 2\n",
 859                                dh->dccph_doff);
 860                 return 1;
 861         }
 862
 863         dh = dccp_hdr(skb);
 864
 865         /*
 866          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 867          * has short sequence numbers), drop packet and return
 868          */
 869         if (dh->dccph_x == 0 &&
 870             dh->dccph_type != DCCP_PKT_DATA &&
 871             dh->dccph_type != DCCP_PKT_ACK &&
 872             dh->dccph_type != DCCP_PKT_DATAACK) {
 873                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 874                                             "nor DataAck and P.X == 0\n",
 875                                dccp_packet_name(dh->dccph_type));
 876                 return 1;
 877         }
 878
 879         return 0;
 880 }
 881
 882 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 883
 884 /* this is called when real data arrives */
 885 static int dccp_v4_rcv(struct sk_buff *skb)
 886 {
 887         const struct dccp_hdr *dh;
 888         struct sock *sk;
 889
 890         /* Step 1: Check header basics: */
 891
 892         if (dccp_invalid_packet(skb))
 893                 goto discard_it;
 894
 895         /* If the header checksum is incorrect, drop packet and return */
 896         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
 897                                     skb->nh.iph->daddr) < 0) {
 898                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
 899                                __FUNCTION__);
 900                 goto discard_it;
 901         }
 902
 903         dh = dccp_hdr(skb);
 904
 905         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 906         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 907
 908         dccp_pr_debug("%8.8s "
 909                       "src=%u.%u.%u.%u@%-5d "
 910                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 911                       dccp_packet_name(dh->dccph_type),
 912                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 913                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 914                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 915
 916         if (dccp_packet_without_ack(skb)) {
 917                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 918                 dccp_pr_debug_cat("\n");
 919         } else {
 920                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 921                 dccp_pr_debug_cat(", ack=%llu\n",
 922                                   (unsigned long long)
 923                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 924         }
 925
 926         /* Step 2:
 927          *      Look up flow ID in table and get corresponding socket */
 928         sk = __inet_lookup(&dccp_hashinfo,
 929                            skb->nh.iph->saddr, dh->dccph_sport,
 930                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
 931                            inet_iif(skb));
 932
 933         /*
 934          * Step 2:
 935          *      If no socket ...
 936          *              Generate Reset(No Connection) unless P.type == Reset
 937          *              Drop packet and return
 938          */
 939         if (sk == NULL) {
 940                 dccp_pr_debug("failed to look up flow ID in table and "
 941                               "get corresponding socket\n");
 942                 goto no_dccp_socket;
 943         }
 944
 945         /*
 946          * Step 2:
 947          *      ... or S.state == TIMEWAIT,
 948          *              Generate Reset(No Connection) unless P.type == Reset
 949          *              Drop packet and return
 950          */
 951
 952         if (sk->sk_state == DCCP_TIME_WAIT) {
 953                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
 954                               "do_time_wait\n");
 955                 goto do_time_wait;
 956         }
 957
 958         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 959                 goto discard_and_relse;
 960         nf_reset(skb);
 961
 962         return sk_receive_skb(sk, skb);
 963
 964 no_dccp_socket:
 965         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 966                 goto discard_it;
 967         /*
 968          * Step 2:
 969          *              Generate Reset(No Connection) unless P.type == Reset
 970          *              Drop packet and return
 971          */
 972         if (dh->dccph_type != DCCP_PKT_RESET) {
 973                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 974                                         DCCP_RESET_CODE_NO_CONNECTION;
 975                 dccp_v4_ctl_send_reset(skb);
 976         }
 977
 978 discard_it:
 979         /* Discard frame. */
 980         kfree_skb(skb);
 981         return 0;
 982
 983 discard_and_relse:
 984         sock_put(sk);
 985         goto discard_it;
 986
 987 do_time_wait:
 988         inet_twsk_put((struct inet_timewait_sock *)sk);
 989         goto no_dccp_socket;
 990 }
 991
 992 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 993         .queue_xmit     = ip_queue_xmit,
 994         .send_check     = dccp_v4_send_check,
 995         .rebuild_header = inet_sk_rebuild_header,
 996         .conn_request   = dccp_v4_conn_request,
 997         .syn_recv_sock  = dccp_v4_request_recv_sock,
 998         .net_header_len = sizeof(struct iphdr),
 999         .setsockopt     = ip_setsockopt,
1000         .getsockopt     = ip_getsockopt,
1001         .addr2sockaddr  = inet_csk_addr2sockaddr,
1002         .sockaddr_len   = sizeof(struct sockaddr_in),
1003 };
1004
1005 static int dccp_v4_init_sock(struct sock *sk)
1006 {
1007         static __u8 dccp_v4_ctl_sock_initialized;
1008         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
1009
1010         if (err == 0) {
1011                 if (unlikely(!dccp_v4_ctl_sock_initialized))
1012                         dccp_v4_ctl_sock_initialized = 1;
1013                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1014         }
1015
1016         return err;
1017 }
1018
1019 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1020 {
1021         kfree(inet_rsk(req)->opt);
1022 }
1023
1024 static struct request_sock_ops dccp_request_sock_ops = {
1025         .family         = PF_INET,
1026         .obj_size       = sizeof(struct dccp_request_sock),
1027         .rtx_syn_ack    = dccp_v4_send_response,
1028         .send_ack       = dccp_v4_reqsk_send_ack,
1029         .destructor     = dccp_v4_reqsk_destructor,
1030         .send_reset     = dccp_v4_ctl_send_reset,
1031 };
1032
1033 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1034         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1035 };
1036
1037 static struct proto dccp_v4_prot = {
1038         .name                   = "DCCP",
1039         .owner                  = THIS_MODULE,
1040         .close                  = dccp_close,
1041         .connect                = dccp_v4_connect,
1042         .disconnect             = dccp_disconnect,
1043         .ioctl                  = dccp_ioctl,
1044         .init                   = dccp_v4_init_sock,
1045         .setsockopt             = dccp_setsockopt,
1046         .getsockopt             = dccp_getsockopt,
1047         .sendmsg                = dccp_sendmsg,
1048         .recvmsg                = dccp_recvmsg,
1049         .backlog_rcv            = dccp_v4_do_rcv,
1050         .hash                   = dccp_hash,
1051         .unhash                 = dccp_unhash,
1052         .accept                 = inet_csk_accept,
1053         .get_port               = dccp_v4_get_port,
1054         .shutdown               = dccp_shutdown,
1055         .destroy                = dccp_destroy_sock,
1056         .orphan_count           = &dccp_orphan_count,
1057         .max_header             = MAX_DCCP_HEADER,
1058         .obj_size               = sizeof(struct dccp_sock),
1059         .rsk_prot               = &dccp_request_sock_ops,
1060         .twsk_prot              = &dccp_timewait_sock_ops,
1061 };
1062
1063 static struct net_protocol dccp_v4_protocol = {
1064         .handler        = dccp_v4_rcv,
1065         .err_handler    = dccp_v4_err,
1066         .no_policy      = 1,
1067 };
1068
1069 static const struct proto_ops inet_dccp_ops = {
1070         .family         = PF_INET,
1071         .owner          = THIS_MODULE,
1072         .release        = inet_release,
1073         .bind           = inet_bind,
1074         .connect        = inet_stream_connect,
1075         .socketpair     = sock_no_socketpair,
1076         .accept         = inet_accept,
1077         .getname        = inet_getname,
1078         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1079         .poll           = dccp_poll,
1080         .ioctl          = inet_ioctl,
1081         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1082         .listen         = inet_dccp_listen,
1083         .shutdown       = inet_shutdown,
1084         .setsockopt     = sock_common_setsockopt,
1085         .getsockopt     = sock_common_getsockopt,
1086         .sendmsg        = inet_sendmsg,
1087         .recvmsg        = sock_common_recvmsg,
1088         .mmap           = sock_no_mmap,
1089         .sendpage       = sock_no_sendpage,
1090 };
1091
1092 static struct inet_protosw dccp_v4_protosw = {
1093         .type           = SOCK_DCCP,
1094         .protocol       = IPPROTO_DCCP,
1095         .prot           = &dccp_v4_prot,
1096         .ops            = &inet_dccp_ops,
1097         .capability     = -1,
1098         .no_check       = 0,
1099         .flags          = INET_PROTOSW_ICSK,
1100 };
1101
1102 static char dccp_v4_ctl_socket_err_msg[] __initdata =
1103         KERN_ERR "DCCP: Failed to create the control socket.\n";
1104
1105 static int __init dccp_v4_ctl_sock_init(void)
1106 {
1107         int rc = sock_create_kern(PF_INET, SOCK_DCCP, IPPROTO_DCCP,
1108                                   &dccp_v4_ctl_socket);
1109         if (rc < 0)
1110                 printk(dccp_v4_ctl_socket_err_msg);
1111         else {
1112                 dccp_v4_ctl_socket->sk->sk_allocation = GFP_ATOMIC;
1113                 inet_sk(dccp_v4_ctl_socket->sk)->uc_ttl = -1;
1114
1115                 /* Unhash it so that IP input processing does not even
1116                  * see it, we do not wish this socket to see incoming
1117                  * packets.
1118                  */
1119                 dccp_v4_ctl_socket->sk->sk_prot->unhash(dccp_v4_ctl_socket->sk);
1120         }
1121
1122         return rc;
1123 }
1124
1125 static int __init dccp_v4_init(void)
1126 {
1127         int err = proto_register(&dccp_v4_prot, 1);
1128
1129         if (err != 0)
1130                 goto out;
1131
1132         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1133         if (err != 0)
1134                 goto out_proto_unregister;
1135
1136         inet_register_protosw(&dccp_v4_protosw);
1137
1138         err = dccp_v4_ctl_sock_init();
1139         if (err)
1140                 goto out_unregister_protosw;
1141 out:
1142         return err;
1143 out_unregister_protosw:
1144         inet_unregister_protosw(&dccp_v4_protosw);
1145         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1146 out_proto_unregister:
1147         proto_unregister(&dccp_v4_prot);
1148         goto out;
1149 }
1150
1151 static void __exit dccp_v4_exit(void)
1152 {
1153         inet_unregister_protosw(&dccp_v4_protosw);
1154         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1155         proto_unregister(&dccp_v4_prot);
1156 }
1157
1158 module_init(dccp_v4_init);
1159 module_exit(dccp_v4_exit);
1160
1161 /*
1162  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1163  * values directly, Also cover the case where the protocol is not specified,
1164  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1165  */
1166 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1167 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1168 MODULE_LICENSE("GPL");
1169 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1170 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");