net/dccp/ipv4.c

   1 /*
   2  *  net/dccp/ipv4.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/dccp.h>
  14 #include <linux/icmp.h>
  15 #include <linux/slab.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/random.h>
  19
  20 #include <net/icmp.h>
  21 #include <net/inet_common.h>
  22 #include <net/inet_hashtables.h>
  23 #include <net/inet_sock.h>
  24 #include <net/protocol.h>
  25 #include <net/sock.h>
  26 #include <net/timewait_sock.h>
  27 #include <net/tcp_states.h>
  28 #include <net/xfrm.h>
  29
  30 #include "ackvec.h"
  31 #include "ccid.h"
  32 #include "dccp.h"
  33 #include "feat.h"
  34
  35 /*
  36  * The per-net dccp.v4_ctl_sk socket is 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
  41 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  42 {
  43         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  44         struct inet_sock *inet = inet_sk(sk);
  45         struct dccp_sock *dp = dccp_sk(sk);
  46         __be16 orig_sport, orig_dport;
  47         __be32 daddr, nexthop;
  48         struct flowi4 *fl4;
  49         struct rtable *rt;
  50         int err;
  51         struct ip_options_rcu *inet_opt;
  52
  53         dp->dccps_role = DCCP_ROLE_CLIENT;
  54
  55         if (addr_len < sizeof(struct sockaddr_in))
  56                 return -EINVAL;
  57
  58         if (usin->sin_family != AF_INET)
  59                 return -EAFNOSUPPORT;
  60
  61         nexthop = daddr = usin->sin_addr.s_addr;
  62
  63         inet_opt = rcu_dereference_protected(inet->inet_opt,
  64                                              sock_owned_by_user(sk));
  65         if (inet_opt != NULL && inet_opt->opt.srr) {
  66                 if (daddr == 0)
  67                         return -EINVAL;
  68                 nexthop = inet_opt->opt.faddr;
  69         }
  70
  71         orig_sport = inet->inet_sport;
  72         orig_dport = usin->sin_port;
  73         fl4 = &inet->cork.fl.u.ip4;
  74         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
  75                               RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  76                               IPPROTO_DCCP,
  77                               orig_sport, orig_dport, sk, true);
  78         if (IS_ERR(rt))
  79                 return PTR_ERR(rt);
  80
  81         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  82                 ip_rt_put(rt);
  83                 return -ENETUNREACH;
  84         }
  85
  86         if (inet_opt == NULL || !inet_opt->opt.srr)
  87                 daddr = fl4->daddr;
  88
  89         if (inet->inet_saddr == 0)
  90                 inet->inet_saddr = fl4->saddr;
  91         inet->inet_rcv_saddr = inet->inet_saddr;
  92
  93         inet->inet_dport = usin->sin_port;
  94         inet->inet_daddr = daddr;
  95
  96         inet_csk(sk)->icsk_ext_hdr_len = 0;
  97         if (inet_opt)
  98                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
  99         /*
 100          * Socket identity is still unknown (sport may be zero).
 101          * However we set state to DCCP_REQUESTING and not releasing socket
 102          * lock select source port, enter ourselves into the hash tables and
 103          * complete initialization after this.
 104          */
 105         dccp_set_state(sk, DCCP_REQUESTING);
 106         err = inet_hash_connect(&dccp_death_row, sk);
 107         if (err != 0)
 108                 goto failure;
 109
 110         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 111                                inet->inet_sport, inet->inet_dport, sk);
 112         if (IS_ERR(rt)) {
 113                 rt = NULL;
 114                 goto failure;
 115         }
 116         /* OK, now commit destination to socket.  */
 117         sk_setup_caps(sk, &rt->dst);
 118
 119         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 120                                                     inet->inet_daddr,
 121                                                     inet->inet_sport,
 122                                                     inet->inet_dport);
 123         inet->inet_id = dp->dccps_iss ^ jiffies;
 124
 125         err = dccp_connect(sk);
 126         rt = NULL;
 127         if (err != 0)
 128                 goto failure;
 129 out:
 130         return err;
 131 failure:
 132         /*
 133          * This unhashes the socket and releases the local port, if necessary.
 134          */
 135         dccp_set_state(sk, DCCP_CLOSED);
 136         ip_rt_put(rt);
 137         sk->sk_route_caps = 0;
 138         inet->inet_dport = 0;
 139         goto out;
 140 }
 141
 142 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 143
 144 /*
 145  * This routine does path mtu discovery as defined in RFC1191.
 146  */
 147 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 148                                           const struct iphdr *iph,
 149                                           u32 mtu)
 150 {
 151         struct dst_entry *dst;
 152         const struct inet_sock *inet = inet_sk(sk);
 153         const struct dccp_sock *dp = dccp_sk(sk);
 154
 155         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 156          * send out by Linux are always < 576bytes so they should go through
 157          * unfragmented).
 158          */
 159         if (sk->sk_state == DCCP_LISTEN)
 160                 return;
 161
 162         /* We don't check in the destentry if pmtu discovery is forbidden
 163          * on this route. We just assume that no packet_to_big packets
 164          * are send back when pmtu discovery is not active.
 165          * There is a small race when the user changes this flag in the
 166          * route, but I think that's acceptable.
 167          */
 168         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 169                 return;
 170
 171         dst->ops->update_pmtu(dst, mtu);
 172
 173         /* Something is about to be wrong... Remember soft error
 174          * for the case, if this connection will not able to recover.
 175          */
 176         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 177                 sk->sk_err_soft = EMSGSIZE;
 178
 179         mtu = dst_mtu(dst);
 180
 181         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 182             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 183                 dccp_sync_mss(sk, mtu);
 184
 185                 /*
 186                  * From RFC 4340, sec. 14.1:
 187                  *
 188                  *      DCCP-Sync packets are the best choice for upward
 189                  *      probing, since DCCP-Sync probes do not risk application
 190                  *      data loss.
 191                  */
 192                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 193         } /* else let the usual retransmit timer handle it */
 194 }
 195
 196 /*
 197  * This routine is called by the ICMP module when it gets some sort of error
 198  * condition. If err < 0 then the socket should be closed and the error
 199  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 200  * After adjustment header points to the first 8 bytes of the tcp header. We
 201  * need to find the appropriate port.
 202  *
 203  * The locking strategy used here is very "optimistic". When someone else
 204  * accesses the socket the ICMP is just dropped and for some paths there is no
 205  * check at all. A more general error queue to queue errors for later handling
 206  * is probably better.
 207  */
 208 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 209 {
 210         const struct iphdr *iph = (struct iphdr *)skb->data;
 211         const u8 offset = iph->ihl << 2;
 212         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
 213         struct dccp_sock *dp;
 214         struct inet_sock *inet;
 215         const int type = icmp_hdr(skb)->type;
 216         const int code = icmp_hdr(skb)->code;
 217         struct sock *sk;
 218         __u64 seq;
 219         int err;
 220         struct net *net = dev_net(skb->dev);
 221
 222         if (skb->len < offset + sizeof(*dh) ||
 223             skb->len < offset + __dccp_basic_hdr_len(dh)) {
 224                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 225                 return;
 226         }
 227
 228         sk = inet_lookup(net, &dccp_hashinfo,
 229                         iph->daddr, dh->dccph_dport,
 230                         iph->saddr, dh->dccph_sport, inet_iif(skb));
 231         if (sk == NULL) {
 232                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 233                 return;
 234         }
 235
 236         if (sk->sk_state == DCCP_TIME_WAIT) {
 237                 inet_twsk_put(inet_twsk(sk));
 238                 return;
 239         }
 240
 241         bh_lock_sock(sk);
 242         /* If too many ICMPs get dropped on busy
 243          * servers this needs to be solved differently.
 244          */
 245         if (sock_owned_by_user(sk))
 246                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 247
 248         if (sk->sk_state == DCCP_CLOSED)
 249                 goto out;
 250
 251         dp = dccp_sk(sk);
 252         seq = dccp_hdr_seq(dh);
 253         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 254             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 255                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 256                 goto out;
 257         }
 258
 259         switch (type) {
 260         case ICMP_SOURCE_QUENCH:
 261                 /* Just silently ignore these. */
 262                 goto out;
 263         case ICMP_PARAMETERPROB:
 264                 err = EPROTO;
 265                 break;
 266         case ICMP_DEST_UNREACH:
 267                 if (code > NR_ICMP_UNREACH)
 268                         goto out;
 269
 270                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 271                         if (!sock_owned_by_user(sk))
 272                                 dccp_do_pmtu_discovery(sk, iph, info);
 273                         goto out;
 274                 }
 275
 276                 err = icmp_err_convert[code].errno;
 277                 break;
 278         case ICMP_TIME_EXCEEDED:
 279                 err = EHOSTUNREACH;
 280                 break;
 281         default:
 282                 goto out;
 283         }
 284
 285         switch (sk->sk_state) {
 286                 struct request_sock *req , **prev;
 287         case DCCP_LISTEN:
 288                 if (sock_owned_by_user(sk))
 289                         goto out;
 290                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 291                                           iph->daddr, iph->saddr);
 292                 if (!req)
 293                         goto out;
 294
 295                 /*
 296                  * ICMPs are not backlogged, hence we cannot get an established
 297                  * socket here.
 298                  */
 299                 WARN_ON(req->sk);
 300
 301                 if (seq != dccp_rsk(req)->dreq_iss) {
 302                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 303                         goto out;
 304                 }
 305                 /*
 306                  * Still in RESPOND, just remove it silently.
 307                  * There is no good way to pass the error to the newly
 308                  * created socket, and POSIX does not want network
 309                  * errors returned from accept().
 310                  */
 311                 inet_csk_reqsk_queue_drop(sk, req, prev);
 312                 goto out;
 313
 314         case DCCP_REQUESTING:
 315         case DCCP_RESPOND:
 316                 if (!sock_owned_by_user(sk)) {
 317                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 318                         sk->sk_err = err;
 319
 320                         sk->sk_error_report(sk);
 321
 322                         dccp_done(sk);
 323                 } else
 324                         sk->sk_err_soft = err;
 325                 goto out;
 326         }
 327
 328         /* If we've already connected we will keep trying
 329          * until we time out, or the user gives up.
 330          *
 331          * rfc1122 4.2.3.9 allows to consider as hard errors
 332          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 333          * but it is obsoleted by pmtu discovery).
 334          *
 335          * Note, that in modern internet, where routing is unreliable
 336          * and in each dark corner broken firewalls sit, sending random
 337          * errors ordered by their masters even this two messages finally lose
 338          * their original sense (even Linux sends invalid PORT_UNREACHs)
 339          *
 340          * Now we are in compliance with RFCs.
 341          *                                                      --ANK (980905)
 342          */
 343
 344         inet = inet_sk(sk);
 345         if (!sock_owned_by_user(sk) && inet->recverr) {
 346                 sk->sk_err = err;
 347                 sk->sk_error_report(sk);
 348         } else /* Only an error on timeout */
 349                 sk->sk_err_soft = err;
 350 out:
 351         bh_unlock_sock(sk);
 352         sock_put(sk);
 353 }
 354
 355 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 356                                       __be32 src, __be32 dst)
 357 {
 358         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 359 }
 360
 361 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 362 {
 363         const struct inet_sock *inet = inet_sk(sk);
 364         struct dccp_hdr *dh = dccp_hdr(skb);
 365
 366         dccp_csum_outgoing(skb);
 367         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 368                                                  inet->inet_saddr,
 369                                                  inet->inet_daddr);
 370 }
 371
 372 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 373
 374 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 375 {
 376         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 377                                            ip_hdr(skb)->saddr,
 378                                            dccp_hdr(skb)->dccph_dport,
 379                                            dccp_hdr(skb)->dccph_sport);
 380 }
 381
 382 /*
 383  * The three way handshake has completed - we got a valid ACK or DATAACK -
 384  * now create the new socket.
 385  *
 386  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 387  */
 388 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 389                                        struct request_sock *req,
 390                                        struct dst_entry *dst)
 391 {
 392         struct inet_request_sock *ireq;
 393         struct inet_sock *newinet;
 394         struct sock *newsk;
 395
 396         if (sk_acceptq_is_full(sk))
 397                 goto exit_overflow;
 398
 399         newsk = dccp_create_openreq_child(sk, req, skb);
 400         if (newsk == NULL)
 401                 goto exit_nonewsk;
 402
 403         newinet            = inet_sk(newsk);
 404         ireq               = inet_rsk(req);
 405         newinet->inet_daddr     = ireq->rmt_addr;
 406         newinet->inet_rcv_saddr = ireq->loc_addr;
 407         newinet->inet_saddr     = ireq->loc_addr;
 408         newinet->inet_opt       = ireq->opt;
 409         ireq->opt          = NULL;
 410         newinet->mc_index  = inet_iif(skb);
 411         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 412         newinet->inet_id   = jiffies;
 413
 414         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 415                 goto put_and_exit;
 416
 417         sk_setup_caps(newsk, dst);
 418
 419         dccp_sync_mss(newsk, dst_mtu(dst));
 420
 421         if (__inet_inherit_port(sk, newsk) < 0)
 422                 goto put_and_exit;
 423         __inet_hash_nolisten(newsk, NULL);
 424
 425         return newsk;
 426
 427 exit_overflow:
 428         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 429 exit_nonewsk:
 430         dst_release(dst);
 431 exit:
 432         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 433         return NULL;
 434 put_and_exit:
 435         sock_put(newsk);
 436         goto exit;
 437 }
 438
 439 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 440
 441 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 442 {
 443         const struct dccp_hdr *dh = dccp_hdr(skb);
 444         const struct iphdr *iph = ip_hdr(skb);
 445         struct sock *nsk;
 446         struct request_sock **prev;
 447         /* Find possible connection requests. */
 448         struct request_sock *req = inet_csk_search_req(sk, &prev,
 449                                                        dh->dccph_sport,
 450                                                        iph->saddr, iph->daddr);
 451         if (req != NULL)
 452                 return dccp_check_req(sk, skb, req, prev);
 453
 454         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 455                                       iph->saddr, dh->dccph_sport,
 456                                       iph->daddr, dh->dccph_dport,
 457                                       inet_iif(skb));
 458         if (nsk != NULL) {
 459                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 460                         bh_lock_sock(nsk);
 461                         return nsk;
 462                 }
 463                 inet_twsk_put(inet_twsk(nsk));
 464                 return NULL;
 465         }
 466
 467         return sk;
 468 }
 469
 470 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 471                                            struct sk_buff *skb)
 472 {
 473         struct rtable *rt;
 474         struct flowi4 fl4 = {
 475                 .flowi4_oif = skb_rtable(skb)->rt_iif,
 476                 .daddr = ip_hdr(skb)->saddr,
 477                 .saddr = ip_hdr(skb)->daddr,
 478                 .flowi4_tos = RT_CONN_FLAGS(sk),
 479                 .flowi4_proto = sk->sk_protocol,
 480                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 481                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 482         };
 483
 484         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 485         rt = ip_route_output_flow(net, &fl4, sk);
 486         if (IS_ERR(rt)) {
 487                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 488                 return NULL;
 489         }
 490
 491         return &rt->dst;
 492 }
 493
 494 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 495                                  struct request_values *rv_unused)
 496 {
 497         int err = -1;
 498         struct sk_buff *skb;
 499         struct dst_entry *dst;
 500         struct flowi4 fl4;
 501
 502         dst = inet_csk_route_req(sk, &fl4, req);
 503         if (dst == NULL)
 504                 goto out;
 505
 506         skb = dccp_make_response(sk, dst, req);
 507         if (skb != NULL) {
 508                 const struct inet_request_sock *ireq = inet_rsk(req);
 509                 struct dccp_hdr *dh = dccp_hdr(skb);
 510
 511                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
 512                                                               ireq->rmt_addr);
 513                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 514                                             ireq->rmt_addr,
 515                                             ireq->opt);
 516                 err = net_xmit_eval(err);
 517         }
 518
 519 out:
 520         dst_release(dst);
 521         return err;
 522 }
 523
 524 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
 525 {
 526         int err;
 527         const struct iphdr *rxiph;
 528         struct sk_buff *skb;
 529         struct dst_entry *dst;
 530         struct net *net = dev_net(skb_dst(rxskb)->dev);
 531         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 532
 533         /* Never send a reset in response to a reset. */
 534         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 535                 return;
 536
 537         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 538                 return;
 539
 540         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 541         if (dst == NULL)
 542                 return;
 543
 544         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 545         if (skb == NULL)
 546                 goto out;
 547
 548         rxiph = ip_hdr(rxskb);
 549         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 550                                                                  rxiph->daddr);
 551         skb_dst_set(skb, dst_clone(dst));
 552
 553         bh_lock_sock(ctl_sk);
 554         err = ip_build_and_send_pkt(skb, ctl_sk,
 555                                     rxiph->daddr, rxiph->saddr, NULL);
 556         bh_unlock_sock(ctl_sk);
 557
 558         if (net_xmit_eval(err) == 0) {
 559                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 560                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 561         }
 562 out:
 563          dst_release(dst);
 564 }
 565
 566 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 567 {
 568         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 569         kfree(inet_rsk(req)->opt);
 570 }
 571
 572 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 573         .family         = PF_INET,
 574         .obj_size       = sizeof(struct dccp_request_sock),
 575         .rtx_syn_ack    = dccp_v4_send_response,
 576         .send_ack       = dccp_reqsk_send_ack,
 577         .destructor     = dccp_v4_reqsk_destructor,
 578         .send_reset     = dccp_v4_ctl_send_reset,
 579 };
 580
 581 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 582 {
 583         struct inet_request_sock *ireq;
 584         struct request_sock *req;
 585         struct dccp_request_sock *dreq;
 586         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 587         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 588
 589         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 590         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 591                 return 0;       /* discard, don't send a reset here */
 592
 593         if (dccp_bad_service_code(sk, service)) {
 594                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 595                 goto drop;
 596         }
 597         /*
 598          * TW buckets are converted to open requests without
 599          * limitations, they conserve resources and peer is
 600          * evidently real one.
 601          */
 602         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 603         if (inet_csk_reqsk_queue_is_full(sk))
 604                 goto drop;
 605
 606         /*
 607          * Accept backlog is full. If we have already queued enough
 608          * of warm entries in syn queue, drop request. It is better than
 609          * clogging syn queue with openreqs with exponentially increasing
 610          * timeout.
 611          */
 612         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 613                 goto drop;
 614
 615         req = inet_reqsk_alloc(&dccp_request_sock_ops);
 616         if (req == NULL)
 617                 goto drop;
 618
 619         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 620                 goto drop_and_free;
 621
 622         dreq = dccp_rsk(req);
 623         if (dccp_parse_options(sk, dreq, skb))
 624                 goto drop_and_free;
 625
 626         if (security_inet_conn_request(sk, skb, req))
 627                 goto drop_and_free;
 628
 629         ireq = inet_rsk(req);
 630         ireq->loc_addr = ip_hdr(skb)->daddr;
 631         ireq->rmt_addr = ip_hdr(skb)->saddr;
 632
 633         /*
 634          * Step 3: Process LISTEN state
 635          *
 636          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 637          *
 638          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 639          * dccp_create_openreq_child.
 640          */
 641         dreq->dreq_isr     = dcb->dccpd_seq;
 642         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 643         dreq->dreq_service = service;
 644
 645         if (dccp_v4_send_response(sk, req, NULL))
 646                 goto drop_and_free;
 647
 648         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 649         return 0;
 650
 651 drop_and_free:
 652         reqsk_free(req);
 653 drop:
 654         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 655         return -1;
 656 }
 657
 658 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 659
 660 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 661 {
 662         struct dccp_hdr *dh = dccp_hdr(skb);
 663
 664         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 665                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 666                         goto reset;
 667                 return 0;
 668         }
 669
 670         /*
 671          *  Step 3: Process LISTEN state
 672          *       If P.type == Request or P contains a valid Init Cookie option,
 673          *            (* Must scan the packet's options to check for Init
 674          *               Cookies.  Only Init Cookies are processed here,
 675          *               however; other options are processed in Step 8.  This
 676          *               scan need only be performed if the endpoint uses Init
 677          *               Cookies *)
 678          *            (* Generate a new socket and switch to that socket *)
 679          *            Set S := new socket for this port pair
 680          *            S.state = RESPOND
 681          *            Choose S.ISS (initial seqno) or set from Init Cookies
 682          *            Initialize S.GAR := S.ISS
 683          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 684          *            Continue with S.state == RESPOND
 685          *            (* A Response packet will be generated in Step 11 *)
 686          *       Otherwise,
 687          *            Generate Reset(No Connection) unless P.type == Reset
 688          *            Drop packet and return
 689          *
 690          * NOTE: the check for the packet types is done in
 691          *       dccp_rcv_state_process
 692          */
 693         if (sk->sk_state == DCCP_LISTEN) {
 694                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 695
 696                 if (nsk == NULL)
 697                         goto discard;
 698
 699                 if (nsk != sk) {
 700                         if (dccp_child_process(sk, nsk, skb))
 701                                 goto reset;
 702                         return 0;
 703                 }
 704         }
 705
 706         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 707                 goto reset;
 708         return 0;
 709
 710 reset:
 711         dccp_v4_ctl_send_reset(sk, skb);
 712 discard:
 713         kfree_skb(skb);
 714         return 0;
 715 }
 716
 717 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 718
 719 /**
 720  *      dccp_invalid_packet  -  check for malformed packets
 721  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 722  *      Packets that fail these checks are ignored and do not receive Resets.
 723  */
 724 int dccp_invalid_packet(struct sk_buff *skb)
 725 {
 726         const struct dccp_hdr *dh;
 727         unsigned int cscov;
 728
 729         if (skb->pkt_type != PACKET_HOST)
 730                 return 1;
 731
 732         /* If the packet is shorter than 12 bytes, drop packet and return */
 733         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 734                 DCCP_WARN("pskb_may_pull failed\n");
 735                 return 1;
 736         }
 737
 738         dh = dccp_hdr(skb);
 739
 740         /* If P.type is not understood, drop packet and return */
 741         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 742                 DCCP_WARN("invalid packet type\n");
 743                 return 1;
 744         }
 745
 746         /*
 747          * If P.Data Offset is too small for packet type, drop packet and return
 748          */
 749         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 750                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 751                 return 1;
 752         }
 753         /*
 754          * If P.Data Offset is too too large for packet, drop packet and return
 755          */
 756         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 757                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 758                 return 1;
 759         }
 760
 761         /*
 762          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 763          * has short sequence numbers), drop packet and return
 764          */
 765         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 766             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 767                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 768                           dccp_packet_name(dh->dccph_type));
 769                 return 1;
 770         }
 771
 772         /*
 773          * If P.CsCov is too large for the packet size, drop packet and return.
 774          * This must come _before_ checksumming (not as RFC 4340 suggests).
 775          */
 776         cscov = dccp_csum_coverage(skb);
 777         if (cscov > skb->len) {
 778                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 779                           dh->dccph_cscov, skb->len);
 780                 return 1;
 781         }
 782
 783         /* If header checksum is incorrect, drop packet and return.
 784          * (This step is completed in the AF-dependent functions.) */
 785         skb->csum = skb_checksum(skb, 0, cscov, 0);
 786
 787         return 0;
 788 }
 789
 790 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 791
 792 /* this is called when real data arrives */
 793 static int dccp_v4_rcv(struct sk_buff *skb)
 794 {
 795         const struct dccp_hdr *dh;
 796         const struct iphdr *iph;
 797         struct sock *sk;
 798         int min_cov;
 799
 800         /* Step 1: Check header basics */
 801
 802         if (dccp_invalid_packet(skb))
 803                 goto discard_it;
 804
 805         iph = ip_hdr(skb);
 806         /* Step 1: If header checksum is incorrect, drop packet and return */
 807         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 808                 DCCP_WARN("dropped packet with invalid checksum\n");
 809                 goto discard_it;
 810         }
 811
 812         dh = dccp_hdr(skb);
 813
 814         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 815         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 816
 817         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 818                       dccp_packet_name(dh->dccph_type),
 819                       &iph->saddr, ntohs(dh->dccph_sport),
 820                       &iph->daddr, ntohs(dh->dccph_dport),
 821                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 822
 823         if (dccp_packet_without_ack(skb)) {
 824                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 825                 dccp_pr_debug_cat("\n");
 826         } else {
 827                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 828                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 829                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 830         }
 831
 832         /* Step 2:
 833          *      Look up flow ID in table and get corresponding socket */
 834         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 835                                dh->dccph_sport, dh->dccph_dport);
 836         /*
 837          * Step 2:
 838          *      If no socket ...
 839          */
 840         if (sk == NULL) {
 841                 dccp_pr_debug("failed to look up flow ID in table and "
 842                               "get corresponding socket\n");
 843                 goto no_dccp_socket;
 844         }
 845
 846         /*
 847          * Step 2:
 848          *      ... or S.state == TIMEWAIT,
 849          *              Generate Reset(No Connection) unless P.type == Reset
 850          *              Drop packet and return
 851          */
 852         if (sk->sk_state == DCCP_TIME_WAIT) {
 853                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 854                 inet_twsk_put(inet_twsk(sk));
 855                 goto no_dccp_socket;
 856         }
 857
 858         /*
 859          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 860          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 861          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 862          */
 863         min_cov = dccp_sk(sk)->dccps_pcrlen;
 864         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 865                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 866                               dh->dccph_cscov, min_cov);
 867                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 868                  *         options (Section 11.7) with Drop Code 0, Protocol
 869                  *         Constraints."                                     */
 870                 goto discard_and_relse;
 871         }
 872
 873         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 874                 goto discard_and_relse;
 875         nf_reset(skb);
 876
 877         return sk_receive_skb(sk, skb, 1);
 878
 879 no_dccp_socket:
 880         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 881                 goto discard_it;
 882         /*
 883          * Step 2:
 884          *      If no socket ...
 885          *              Generate Reset(No Connection) unless P.type == Reset
 886          *              Drop packet and return
 887          */
 888         if (dh->dccph_type != DCCP_PKT_RESET) {
 889                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 890                                         DCCP_RESET_CODE_NO_CONNECTION;
 891                 dccp_v4_ctl_send_reset(sk, skb);
 892         }
 893
 894 discard_it:
 895         kfree_skb(skb);
 896         return 0;
 897
 898 discard_and_relse:
 899         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         .bind_conflict     = inet_csk_bind_conflict,
 915 #ifdef CONFIG_COMPAT
 916         .compat_setsockopt = compat_ip_setsockopt,
 917         .compat_getsockopt = compat_ip_getsockopt,
 918 #endif
 919 };
 920
 921 static int dccp_v4_init_sock(struct sock *sk)
 922 {
 923         static __u8 dccp_v4_ctl_sock_initialized;
 924         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 925
 926         if (err == 0) {
 927                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 928                         dccp_v4_ctl_sock_initialized = 1;
 929                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 930         }
 931
 932         return err;
 933 }
 934
 935 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 936         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 937 };
 938
 939 static struct proto dccp_v4_prot = {
 940         .name                   = "DCCP",
 941         .owner                  = THIS_MODULE,
 942         .close                  = dccp_close,
 943         .connect                = dccp_v4_connect,
 944         .disconnect             = dccp_disconnect,
 945         .ioctl                  = dccp_ioctl,
 946         .init                   = dccp_v4_init_sock,
 947         .setsockopt             = dccp_setsockopt,
 948         .getsockopt             = dccp_getsockopt,
 949         .sendmsg                = dccp_sendmsg,
 950         .recvmsg                = dccp_recvmsg,
 951         .backlog_rcv            = dccp_v4_do_rcv,
 952         .hash                   = inet_hash,
 953         .unhash                 = inet_unhash,
 954         .accept                 = inet_csk_accept,
 955         .get_port               = inet_csk_get_port,
 956         .shutdown               = dccp_shutdown,
 957         .destroy                = dccp_destroy_sock,
 958         .orphan_count           = &dccp_orphan_count,
 959         .max_header             = MAX_DCCP_HEADER,
 960         .obj_size               = sizeof(struct dccp_sock),
 961         .slab_flags             = SLAB_DESTROY_BY_RCU,
 962         .rsk_prot               = &dccp_request_sock_ops,
 963         .twsk_prot              = &dccp_timewait_sock_ops,
 964         .h.hashinfo             = &dccp_hashinfo,
 965 #ifdef CONFIG_COMPAT
 966         .compat_setsockopt      = compat_dccp_setsockopt,
 967         .compat_getsockopt      = compat_dccp_getsockopt,
 968 #endif
 969 };
 970
 971 static const struct net_protocol dccp_v4_protocol = {
 972         .handler        = dccp_v4_rcv,
 973         .err_handler    = dccp_v4_err,
 974         .no_policy      = 1,
 975         .netns_ok       = 1,
 976 };
 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         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 991         .listen            = inet_dccp_listen,
 992         .shutdown          = inet_shutdown,
 993         .setsockopt        = sock_common_setsockopt,
 994         .getsockopt        = sock_common_getsockopt,
 995         .sendmsg           = inet_sendmsg,
 996         .recvmsg           = sock_common_recvmsg,
 997         .mmap              = sock_no_mmap,
 998         .sendpage          = sock_no_sendpage,
 999 #ifdef CONFIG_COMPAT
1000         .compat_setsockopt = compat_sock_common_setsockopt,
1001         .compat_getsockopt = compat_sock_common_getsockopt,
1002 #endif
1003 };
1004
1005 static struct inet_protosw dccp_v4_protosw = {
1006         .type           = SOCK_DCCP,
1007         .protocol       = IPPROTO_DCCP,
1008         .prot           = &dccp_v4_prot,
1009         .ops            = &inet_dccp_ops,
1010         .no_check       = 0,
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 struct pernet_operations dccp_v4_ops = {
1029         .init   = dccp_v4_init_net,
1030         .exit   = dccp_v4_exit_net,
1031 };
1032
1033 static int __init dccp_v4_init(void)
1034 {
1035         int err = proto_register(&dccp_v4_prot, 1);
1036
1037         if (err != 0)
1038                 goto out;
1039
1040         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1041         if (err != 0)
1042                 goto out_proto_unregister;
1043
1044         inet_register_protosw(&dccp_v4_protosw);
1045
1046         err = register_pernet_subsys(&dccp_v4_ops);
1047         if (err)
1048                 goto out_destroy_ctl_sock;
1049 out:
1050         return err;
1051 out_destroy_ctl_sock:
1052         inet_unregister_protosw(&dccp_v4_protosw);
1053         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1054 out_proto_unregister:
1055         proto_unregister(&dccp_v4_prot);
1056         goto out;
1057 }
1058
1059 static void __exit dccp_v4_exit(void)
1060 {
1061         unregister_pernet_subsys(&dccp_v4_ops);
1062         inet_unregister_protosw(&dccp_v4_protosw);
1063         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1064         proto_unregister(&dccp_v4_prot);
1065 }
1066
1067 module_init(dccp_v4_init);
1068 module_exit(dccp_v4_exit);
1069
1070 /*
1071  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1072  * values directly, Also cover the case where the protocol is not specified,
1073  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1074  */
1075 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1076 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1077 MODULE_LICENSE("GPL");
1078 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1079 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");