net/dccp/ipv4.c

   1 /*
   2  *  net/dccp/ipv4.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/dccp.h>
  14 #include <linux/icmp.h>
  15 #include <linux/slab.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/random.h>
  19
  20 #include <net/icmp.h>
  21 #include <net/inet_common.h>
  22 #include <net/inet_hashtables.h>
  23 #include <net/inet_sock.h>
  24 #include <net/protocol.h>
  25 #include <net/sock.h>
  26 #include <net/timewait_sock.h>
  27 #include <net/tcp_states.h>
  28 #include <net/xfrm.h>
  29 #include <net/secure_seq.h>
  30
  31 #include "ackvec.h"
  32 #include "ccid.h"
  33 #include "dccp.h"
  34 #include "feat.h"
  35
  36 /*
  37  * The per-net dccp.v4_ctl_sk socket is used for responding to
  38  * the Out-of-the-blue (OOTB) packets. A control sock will be created
  39  * for this socket at the initialization time.
  40  */
  41
  42 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  43 {
  44         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  45         struct inet_sock *inet = inet_sk(sk);
  46         struct dccp_sock *dp = dccp_sk(sk);
  47         __be16 orig_sport, orig_dport;
  48         __be32 daddr, nexthop;
  49         struct flowi4 *fl4;
  50         struct rtable *rt;
  51         int err;
  52         struct ip_options_rcu *inet_opt;
  53
  54         dp->dccps_role = DCCP_ROLE_CLIENT;
  55
  56         if (addr_len < sizeof(struct sockaddr_in))
  57                 return -EINVAL;
  58
  59         if (usin->sin_family != AF_INET)
  60                 return -EAFNOSUPPORT;
  61
  62         nexthop = daddr = usin->sin_addr.s_addr;
  63
  64         inet_opt = rcu_dereference_protected(inet->inet_opt,
  65                                              sock_owned_by_user(sk));
  66         if (inet_opt != NULL && inet_opt->opt.srr) {
  67                 if (daddr == 0)
  68                         return -EINVAL;
  69                 nexthop = inet_opt->opt.faddr;
  70         }
  71
  72         orig_sport = inet->inet_sport;
  73         orig_dport = usin->sin_port;
  74         fl4 = &inet->cork.fl.u.ip4;
  75         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
  76                               RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  77                               IPPROTO_DCCP,
  78                               orig_sport, orig_dport, sk, true);
  79         if (IS_ERR(rt))
  80                 return PTR_ERR(rt);
  81
  82         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  83                 ip_rt_put(rt);
  84                 return -ENETUNREACH;
  85         }
  86
  87         if (inet_opt == NULL || !inet_opt->opt.srr)
  88                 daddr = fl4->daddr;
  89
  90         if (inet->inet_saddr == 0)
  91                 inet->inet_saddr = fl4->saddr;
  92         inet->inet_rcv_saddr = inet->inet_saddr;
  93
  94         inet->inet_dport = usin->sin_port;
  95         inet->inet_daddr = daddr;
  96
  97         inet_csk(sk)->icsk_ext_hdr_len = 0;
  98         if (inet_opt)
  99                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 100         /*
 101          * Socket identity is still unknown (sport may be zero).
 102          * However we set state to DCCP_REQUESTING and not releasing socket
 103          * lock select source port, enter ourselves into the hash tables and
 104          * complete initialization after this.
 105          */
 106         dccp_set_state(sk, DCCP_REQUESTING);
 107         err = inet_hash_connect(&dccp_death_row, sk);
 108         if (err != 0)
 109                 goto failure;
 110
 111         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 112                                inet->inet_sport, inet->inet_dport, sk);
 113         if (IS_ERR(rt)) {
 114                 err = PTR_ERR(rt);
 115                 rt = NULL;
 116                 goto failure;
 117         }
 118         /* OK, now commit destination to socket.  */
 119         sk_setup_caps(sk, &rt->dst);
 120
 121         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 122                                                     inet->inet_daddr,
 123                                                     inet->inet_sport,
 124                                                     inet->inet_dport);
 125         inet->inet_id = dp->dccps_iss ^ jiffies;
 126
 127         err = dccp_connect(sk);
 128         rt = NULL;
 129         if (err != 0)
 130                 goto failure;
 131 out:
 132         return err;
 133 failure:
 134         /*
 135          * This unhashes the socket and releases the local port, if necessary.
 136          */
 137         dccp_set_state(sk, DCCP_CLOSED);
 138         ip_rt_put(rt);
 139         sk->sk_route_caps = 0;
 140         inet->inet_dport = 0;
 141         goto out;
 142 }
 143
 144 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 145
 146 /*
 147  * This routine does path mtu discovery as defined in RFC1191.
 148  */
 149 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 150                                           const struct iphdr *iph,
 151                                           u32 mtu)
 152 {
 153         struct dst_entry *dst;
 154         const struct inet_sock *inet = inet_sk(sk);
 155         const struct dccp_sock *dp = dccp_sk(sk);
 156
 157         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 158          * send out by Linux are always < 576bytes so they should go through
 159          * unfragmented).
 160          */
 161         if (sk->sk_state == DCCP_LISTEN)
 162                 return;
 163
 164         /* We don't check in the destentry if pmtu discovery is forbidden
 165          * on this route. We just assume that no packet_to_big packets
 166          * are send back when pmtu discovery is not active.
 167          * There is a small race when the user changes this flag in the
 168          * route, but I think that's acceptable.
 169          */
 170         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 171                 return;
 172
 173         dst->ops->update_pmtu(dst, mtu);
 174
 175         /* Something is about to be wrong... Remember soft error
 176          * for the case, if this connection will not able to recover.
 177          */
 178         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 179                 sk->sk_err_soft = EMSGSIZE;
 180
 181         mtu = dst_mtu(dst);
 182
 183         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 184             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 185                 dccp_sync_mss(sk, mtu);
 186
 187                 /*
 188                  * From RFC 4340, sec. 14.1:
 189                  *
 190                  *      DCCP-Sync packets are the best choice for upward
 191                  *      probing, since DCCP-Sync probes do not risk application
 192                  *      data loss.
 193                  */
 194                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 195         } /* else let the usual retransmit timer handle it */
 196 }
 197
 198 /*
 199  * This routine is called by the ICMP module when it gets some sort of error
 200  * condition. If err < 0 then the socket should be closed and the error
 201  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 202  * After adjustment header points to the first 8 bytes of the tcp header. We
 203  * need to find the appropriate port.
 204  *
 205  * The locking strategy used here is very "optimistic". When someone else
 206  * accesses the socket the ICMP is just dropped and for some paths there is no
 207  * check at all. A more general error queue to queue errors for later handling
 208  * is probably better.
 209  */
 210 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 211 {
 212         const struct iphdr *iph = (struct iphdr *)skb->data;
 213         const u8 offset = iph->ihl << 2;
 214         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
 215         struct dccp_sock *dp;
 216         struct inet_sock *inet;
 217         const int type = icmp_hdr(skb)->type;
 218         const int code = icmp_hdr(skb)->code;
 219         struct sock *sk;
 220         __u64 seq;
 221         int err;
 222         struct net *net = dev_net(skb->dev);
 223
 224         if (skb->len < offset + sizeof(*dh) ||
 225             skb->len < offset + __dccp_basic_hdr_len(dh)) {
 226                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 227                 return;
 228         }
 229
 230         sk = inet_lookup(net, &dccp_hashinfo,
 231                         iph->daddr, dh->dccph_dport,
 232                         iph->saddr, dh->dccph_sport, inet_iif(skb));
 233         if (sk == NULL) {
 234                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 235                 return;
 236         }
 237
 238         if (sk->sk_state == DCCP_TIME_WAIT) {
 239                 inet_twsk_put(inet_twsk(sk));
 240                 return;
 241         }
 242
 243         bh_lock_sock(sk);
 244         /* If too many ICMPs get dropped on busy
 245          * servers this needs to be solved differently.
 246          */
 247         if (sock_owned_by_user(sk))
 248                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 249
 250         if (sk->sk_state == DCCP_CLOSED)
 251                 goto out;
 252
 253         dp = dccp_sk(sk);
 254         seq = dccp_hdr_seq(dh);
 255         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 256             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 257                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 258                 goto out;
 259         }
 260
 261         switch (type) {
 262         case ICMP_SOURCE_QUENCH:
 263                 /* Just silently ignore these. */
 264                 goto out;
 265         case ICMP_PARAMETERPROB:
 266                 err = EPROTO;
 267                 break;
 268         case ICMP_DEST_UNREACH:
 269                 if (code > NR_ICMP_UNREACH)
 270                         goto out;
 271
 272                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 273                         if (!sock_owned_by_user(sk))
 274                                 dccp_do_pmtu_discovery(sk, iph, info);
 275                         goto out;
 276                 }
 277
 278                 err = icmp_err_convert[code].errno;
 279                 break;
 280         case ICMP_TIME_EXCEEDED:
 281                 err = EHOSTUNREACH;
 282                 break;
 283         default:
 284                 goto out;
 285         }
 286
 287         switch (sk->sk_state) {
 288                 struct request_sock *req , **prev;
 289         case DCCP_LISTEN:
 290                 if (sock_owned_by_user(sk))
 291                         goto out;
 292                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 293                                           iph->daddr, iph->saddr);
 294                 if (!req)
 295                         goto out;
 296
 297                 /*
 298                  * ICMPs are not backlogged, hence we cannot get an established
 299                  * socket here.
 300                  */
 301                 WARN_ON(req->sk);
 302
 303                 if (!between48(seq, dccp_rsk(req)->dreq_iss,
 304                                     dccp_rsk(req)->dreq_gss)) {
 305                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 306                         goto out;
 307                 }
 308                 /*
 309                  * Still in RESPOND, just remove it silently.
 310                  * There is no good way to pass the error to the newly
 311                  * created socket, and POSIX does not want network
 312                  * errors returned from accept().
 313                  */
 314                 inet_csk_reqsk_queue_drop(sk, req, prev);
 315                 goto out;
 316
 317         case DCCP_REQUESTING:
 318         case DCCP_RESPOND:
 319                 if (!sock_owned_by_user(sk)) {
 320                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 321                         sk->sk_err = err;
 322
 323                         sk->sk_error_report(sk);
 324
 325                         dccp_done(sk);
 326                 } else
 327                         sk->sk_err_soft = err;
 328                 goto out;
 329         }
 330
 331         /* If we've already connected we will keep trying
 332          * until we time out, or the user gives up.
 333          *
 334          * rfc1122 4.2.3.9 allows to consider as hard errors
 335          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 336          * but it is obsoleted by pmtu discovery).
 337          *
 338          * Note, that in modern internet, where routing is unreliable
 339          * and in each dark corner broken firewalls sit, sending random
 340          * errors ordered by their masters even this two messages finally lose
 341          * their original sense (even Linux sends invalid PORT_UNREACHs)
 342          *
 343          * Now we are in compliance with RFCs.
 344          *                                                      --ANK (980905)
 345          */
 346
 347         inet = inet_sk(sk);
 348         if (!sock_owned_by_user(sk) && inet->recverr) {
 349                 sk->sk_err = err;
 350                 sk->sk_error_report(sk);
 351         } else /* Only an error on timeout */
 352                 sk->sk_err_soft = err;
 353 out:
 354         bh_unlock_sock(sk);
 355         sock_put(sk);
 356 }
 357
 358 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 359                                       __be32 src, __be32 dst)
 360 {
 361         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 362 }
 363
 364 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 365 {
 366         const struct inet_sock *inet = inet_sk(sk);
 367         struct dccp_hdr *dh = dccp_hdr(skb);
 368
 369         dccp_csum_outgoing(skb);
 370         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 371                                                  inet->inet_saddr,
 372                                                  inet->inet_daddr);
 373 }
 374
 375 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 376
 377 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 378 {
 379         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 380                                            ip_hdr(skb)->saddr,
 381                                            dccp_hdr(skb)->dccph_dport,
 382                                            dccp_hdr(skb)->dccph_sport);
 383 }
 384
 385 /*
 386  * The three way handshake has completed - we got a valid ACK or DATAACK -
 387  * now create the new socket.
 388  *
 389  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 390  */
 391 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 392                                        struct request_sock *req,
 393                                        struct dst_entry *dst)
 394 {
 395         struct inet_request_sock *ireq;
 396         struct inet_sock *newinet;
 397         struct sock *newsk;
 398
 399         if (sk_acceptq_is_full(sk))
 400                 goto exit_overflow;
 401
 402         newsk = dccp_create_openreq_child(sk, req, skb);
 403         if (newsk == NULL)
 404                 goto exit_nonewsk;
 405
 406         newinet            = inet_sk(newsk);
 407         ireq               = inet_rsk(req);
 408         newinet->inet_daddr     = ireq->rmt_addr;
 409         newinet->inet_rcv_saddr = ireq->loc_addr;
 410         newinet->inet_saddr     = ireq->loc_addr;
 411         newinet->inet_opt       = ireq->opt;
 412         ireq->opt          = NULL;
 413         newinet->mc_index  = inet_iif(skb);
 414         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 415         newinet->inet_id   = jiffies;
 416
 417         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 418                 goto put_and_exit;
 419
 420         sk_setup_caps(newsk, dst);
 421
 422         dccp_sync_mss(newsk, dst_mtu(dst));
 423
 424         if (__inet_inherit_port(sk, newsk) < 0)
 425                 goto put_and_exit;
 426         __inet_hash_nolisten(newsk, NULL);
 427
 428         return newsk;
 429
 430 exit_overflow:
 431         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 432 exit_nonewsk:
 433         dst_release(dst);
 434 exit:
 435         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 436         return NULL;
 437 put_and_exit:
 438         bh_unlock_sock(newsk);
 439         sock_put(newsk);
 440         goto exit;
 441 }
 442
 443 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 444
 445 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 446 {
 447         const struct dccp_hdr *dh = dccp_hdr(skb);
 448         const struct iphdr *iph = ip_hdr(skb);
 449         struct sock *nsk;
 450         struct request_sock **prev;
 451         /* Find possible connection requests. */
 452         struct request_sock *req = inet_csk_search_req(sk, &prev,
 453                                                        dh->dccph_sport,
 454                                                        iph->saddr, iph->daddr);
 455         if (req != NULL)
 456                 return dccp_check_req(sk, skb, req, prev);
 457
 458         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 459                                       iph->saddr, dh->dccph_sport,
 460                                       iph->daddr, dh->dccph_dport,
 461                                       inet_iif(skb));
 462         if (nsk != NULL) {
 463                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 464                         bh_lock_sock(nsk);
 465                         return nsk;
 466                 }
 467                 inet_twsk_put(inet_twsk(nsk));
 468                 return NULL;
 469         }
 470
 471         return sk;
 472 }
 473
 474 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 475                                            struct sk_buff *skb)
 476 {
 477         struct rtable *rt;
 478         const struct iphdr *iph = ip_hdr(skb);
 479         struct flowi4 fl4 = {
 480                 .flowi4_oif = skb_rtable(skb)->rt_iif,
 481                 .daddr = iph->saddr,
 482                 .saddr = iph->daddr,
 483                 .flowi4_tos = RT_CONN_FLAGS(sk),
 484                 .flowi4_proto = sk->sk_protocol,
 485                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 486                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 487         };
 488
 489         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 490         rt = ip_route_output_flow(net, &fl4, sk);
 491         if (IS_ERR(rt)) {
 492                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 493                 return NULL;
 494         }
 495
 496         return &rt->dst;
 497 }
 498
 499 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 500                                  struct request_values *rv_unused)
 501 {
 502         int err = -1;
 503         struct sk_buff *skb;
 504         struct dst_entry *dst;
 505         struct flowi4 fl4;
 506
 507         dst = inet_csk_route_req(sk, &fl4, req);
 508         if (dst == NULL)
 509                 goto out;
 510
 511         skb = dccp_make_response(sk, dst, req);
 512         if (skb != NULL) {
 513                 const struct inet_request_sock *ireq = inet_rsk(req);
 514                 struct dccp_hdr *dh = dccp_hdr(skb);
 515
 516                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
 517                                                               ireq->rmt_addr);
 518                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 519                                             ireq->rmt_addr,
 520                                             ireq->opt);
 521                 err = net_xmit_eval(err);
 522         }
 523
 524 out:
 525         dst_release(dst);
 526         return err;
 527 }
 528
 529 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
 530 {
 531         int err;
 532         const struct iphdr *rxiph;
 533         struct sk_buff *skb;
 534         struct dst_entry *dst;
 535         struct net *net = dev_net(skb_dst(rxskb)->dev);
 536         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 537
 538         /* Never send a reset in response to a reset. */
 539         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 540                 return;
 541
 542         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 543                 return;
 544
 545         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 546         if (dst == NULL)
 547                 return;
 548
 549         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 550         if (skb == NULL)
 551                 goto out;
 552
 553         rxiph = ip_hdr(rxskb);
 554         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 555                                                                  rxiph->daddr);
 556         skb_dst_set(skb, dst_clone(dst));
 557
 558         bh_lock_sock(ctl_sk);
 559         err = ip_build_and_send_pkt(skb, ctl_sk,
 560                                     rxiph->daddr, rxiph->saddr, NULL);
 561         bh_unlock_sock(ctl_sk);
 562
 563         if (net_xmit_eval(err) == 0) {
 564                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 565                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 566         }
 567 out:
 568          dst_release(dst);
 569 }
 570
 571 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 572 {
 573         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 574         kfree(inet_rsk(req)->opt);
 575 }
 576
 577 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 578         .family         = PF_INET,
 579         .obj_size       = sizeof(struct dccp_request_sock),
 580         .rtx_syn_ack    = dccp_v4_send_response,
 581         .send_ack       = dccp_reqsk_send_ack,
 582         .destructor     = dccp_v4_reqsk_destructor,
 583         .send_reset     = dccp_v4_ctl_send_reset,
 584 };
 585
 586 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 587 {
 588         struct inet_request_sock *ireq;
 589         struct request_sock *req;
 590         struct dccp_request_sock *dreq;
 591         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 592         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 593
 594         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 595         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 596                 return 0;       /* discard, don't send a reset here */
 597
 598         if (dccp_bad_service_code(sk, service)) {
 599                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 600                 goto drop;
 601         }
 602         /*
 603          * TW buckets are converted to open requests without
 604          * limitations, they conserve resources and peer is
 605          * evidently real one.
 606          */
 607         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 608         if (inet_csk_reqsk_queue_is_full(sk))
 609                 goto drop;
 610
 611         /*
 612          * Accept backlog is full. If we have already queued enough
 613          * of warm entries in syn queue, drop request. It is better than
 614          * clogging syn queue with openreqs with exponentially increasing
 615          * timeout.
 616          */
 617         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 618                 goto drop;
 619
 620         req = inet_reqsk_alloc(&dccp_request_sock_ops);
 621         if (req == NULL)
 622                 goto drop;
 623
 624         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 625                 goto drop_and_free;
 626
 627         dreq = dccp_rsk(req);
 628         if (dccp_parse_options(sk, dreq, skb))
 629                 goto drop_and_free;
 630
 631         if (security_inet_conn_request(sk, skb, req))
 632                 goto drop_and_free;
 633
 634         ireq = inet_rsk(req);
 635         ireq->loc_addr = ip_hdr(skb)->daddr;
 636         ireq->rmt_addr = ip_hdr(skb)->saddr;
 637
 638         /*
 639          * Step 3: Process LISTEN state
 640          *
 641          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 642          *
 643          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 644          */
 645         dreq->dreq_isr     = dcb->dccpd_seq;
 646         dreq->dreq_gsr     = dreq->dreq_isr;
 647         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 648         dreq->dreq_gss     = dreq->dreq_iss;
 649         dreq->dreq_service = service;
 650
 651         if (dccp_v4_send_response(sk, req, NULL))
 652                 goto drop_and_free;
 653
 654         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 655         return 0;
 656
 657 drop_and_free:
 658         reqsk_free(req);
 659 drop:
 660         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 661         return -1;
 662 }
 663
 664 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 665
 666 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 667 {
 668         struct dccp_hdr *dh = dccp_hdr(skb);
 669
 670         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 671                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 672                         goto reset;
 673                 return 0;
 674         }
 675
 676         /*
 677          *  Step 3: Process LISTEN state
 678          *       If P.type == Request or P contains a valid Init Cookie option,
 679          *            (* Must scan the packet's options to check for Init
 680          *               Cookies.  Only Init Cookies are processed here,
 681          *               however; other options are processed in Step 8.  This
 682          *               scan need only be performed if the endpoint uses Init
 683          *               Cookies *)
 684          *            (* Generate a new socket and switch to that socket *)
 685          *            Set S := new socket for this port pair
 686          *            S.state = RESPOND
 687          *            Choose S.ISS (initial seqno) or set from Init Cookies
 688          *            Initialize S.GAR := S.ISS
 689          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 690          *            Continue with S.state == RESPOND
 691          *            (* A Response packet will be generated in Step 11 *)
 692          *       Otherwise,
 693          *            Generate Reset(No Connection) unless P.type == Reset
 694          *            Drop packet and return
 695          *
 696          * NOTE: the check for the packet types is done in
 697          *       dccp_rcv_state_process
 698          */
 699         if (sk->sk_state == DCCP_LISTEN) {
 700                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 701
 702                 if (nsk == NULL)
 703                         goto discard;
 704
 705                 if (nsk != sk) {
 706                         if (dccp_child_process(sk, nsk, skb))
 707                                 goto reset;
 708                         return 0;
 709                 }
 710         }
 711
 712         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 713                 goto reset;
 714         return 0;
 715
 716 reset:
 717         dccp_v4_ctl_send_reset(sk, skb);
 718 discard:
 719         kfree_skb(skb);
 720         return 0;
 721 }
 722
 723 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 724
 725 /**
 726  *      dccp_invalid_packet  -  check for malformed packets
 727  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 728  *      Packets that fail these checks are ignored and do not receive Resets.
 729  */
 730 int dccp_invalid_packet(struct sk_buff *skb)
 731 {
 732         const struct dccp_hdr *dh;
 733         unsigned int cscov;
 734
 735         if (skb->pkt_type != PACKET_HOST)
 736                 return 1;
 737
 738         /* If the packet is shorter than 12 bytes, drop packet and return */
 739         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 740                 DCCP_WARN("pskb_may_pull failed\n");
 741                 return 1;
 742         }
 743
 744         dh = dccp_hdr(skb);
 745
 746         /* If P.type is not understood, drop packet and return */
 747         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 748                 DCCP_WARN("invalid packet type\n");
 749                 return 1;
 750         }
 751
 752         /*
 753          * If P.Data Offset is too small for packet type, drop packet and return
 754          */
 755         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 756                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 757                 return 1;
 758         }
 759         /*
 760          * If P.Data Offset is too too large for packet, drop packet and return
 761          */
 762         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 763                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 764                 return 1;
 765         }
 766
 767         /*
 768          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 769          * has short sequence numbers), drop packet and return
 770          */
 771         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 772             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 773                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 774                           dccp_packet_name(dh->dccph_type));
 775                 return 1;
 776         }
 777
 778         /*
 779          * If P.CsCov is too large for the packet size, drop packet and return.
 780          * This must come _before_ checksumming (not as RFC 4340 suggests).
 781          */
 782         cscov = dccp_csum_coverage(skb);
 783         if (cscov > skb->len) {
 784                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 785                           dh->dccph_cscov, skb->len);
 786                 return 1;
 787         }
 788
 789         /* If header checksum is incorrect, drop packet and return.
 790          * (This step is completed in the AF-dependent functions.) */
 791         skb->csum = skb_checksum(skb, 0, cscov, 0);
 792
 793         return 0;
 794 }
 795
 796 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 797
 798 /* this is called when real data arrives */
 799 static int dccp_v4_rcv(struct sk_buff *skb)
 800 {
 801         const struct dccp_hdr *dh;
 802         const struct iphdr *iph;
 803         struct sock *sk;
 804         int min_cov;
 805
 806         /* Step 1: Check header basics */
 807
 808         if (dccp_invalid_packet(skb))
 809                 goto discard_it;
 810
 811         iph = ip_hdr(skb);
 812         /* Step 1: If header checksum is incorrect, drop packet and return */
 813         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 814                 DCCP_WARN("dropped packet with invalid checksum\n");
 815                 goto discard_it;
 816         }
 817
 818         dh = dccp_hdr(skb);
 819
 820         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 821         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 822
 823         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 824                       dccp_packet_name(dh->dccph_type),
 825                       &iph->saddr, ntohs(dh->dccph_sport),
 826                       &iph->daddr, ntohs(dh->dccph_dport),
 827                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 828
 829         if (dccp_packet_without_ack(skb)) {
 830                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 831                 dccp_pr_debug_cat("\n");
 832         } else {
 833                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 834                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 835                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 836         }
 837
 838         /* Step 2:
 839          *      Look up flow ID in table and get corresponding socket */
 840         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 841                                dh->dccph_sport, dh->dccph_dport);
 842         /*
 843          * Step 2:
 844          *      If no socket ...
 845          */
 846         if (sk == NULL) {
 847                 dccp_pr_debug("failed to look up flow ID in table and "
 848                               "get corresponding socket\n");
 849                 goto no_dccp_socket;
 850         }
 851
 852         /*
 853          * Step 2:
 854          *      ... or S.state == TIMEWAIT,
 855          *              Generate Reset(No Connection) unless P.type == Reset
 856          *              Drop packet and return
 857          */
 858         if (sk->sk_state == DCCP_TIME_WAIT) {
 859                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 860                 inet_twsk_put(inet_twsk(sk));
 861                 goto no_dccp_socket;
 862         }
 863
 864         /*
 865          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 866          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 867          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 868          */
 869         min_cov = dccp_sk(sk)->dccps_pcrlen;
 870         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 871                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 872                               dh->dccph_cscov, min_cov);
 873                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 874                  *         options (Section 11.7) with Drop Code 0, Protocol
 875                  *         Constraints."                                     */
 876                 goto discard_and_relse;
 877         }
 878
 879         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 880                 goto discard_and_relse;
 881         nf_reset(skb);
 882
 883         return sk_receive_skb(sk, skb, 1);
 884
 885 no_dccp_socket:
 886         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 887                 goto discard_it;
 888         /*
 889          * Step 2:
 890          *      If no socket ...
 891          *              Generate Reset(No Connection) unless P.type == Reset
 892          *              Drop packet and return
 893          */
 894         if (dh->dccph_type != DCCP_PKT_RESET) {
 895                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 896                                         DCCP_RESET_CODE_NO_CONNECTION;
 897                 dccp_v4_ctl_send_reset(sk, skb);
 898         }
 899
 900 discard_it:
 901         kfree_skb(skb);
 902         return 0;
 903
 904 discard_and_relse:
 905         sock_put(sk);
 906         goto discard_it;
 907 }
 908
 909 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 910         .queue_xmit        = ip_queue_xmit,
 911         .send_check        = dccp_v4_send_check,
 912         .rebuild_header    = inet_sk_rebuild_header,
 913         .conn_request      = dccp_v4_conn_request,
 914         .syn_recv_sock     = dccp_v4_request_recv_sock,
 915         .net_header_len    = sizeof(struct iphdr),
 916         .setsockopt        = ip_setsockopt,
 917         .getsockopt        = ip_getsockopt,
 918         .addr2sockaddr     = inet_csk_addr2sockaddr,
 919         .sockaddr_len      = sizeof(struct sockaddr_in),
 920         .bind_conflict     = inet_csk_bind_conflict,
 921 #ifdef CONFIG_COMPAT
 922         .compat_setsockopt = compat_ip_setsockopt,
 923         .compat_getsockopt = compat_ip_getsockopt,
 924 #endif
 925 };
 926
 927 static int dccp_v4_init_sock(struct sock *sk)
 928 {
 929         static __u8 dccp_v4_ctl_sock_initialized;
 930         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 931
 932         if (err == 0) {
 933                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 934                         dccp_v4_ctl_sock_initialized = 1;
 935                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 936         }
 937
 938         return err;
 939 }
 940
 941 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 942         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 943 };
 944
 945 static struct proto dccp_v4_prot = {
 946         .name                   = "DCCP",
 947         .owner                  = THIS_MODULE,
 948         .close                  = dccp_close,
 949         .connect                = dccp_v4_connect,
 950         .disconnect             = dccp_disconnect,
 951         .ioctl                  = dccp_ioctl,
 952         .init                   = dccp_v4_init_sock,
 953         .setsockopt             = dccp_setsockopt,
 954         .getsockopt             = dccp_getsockopt,
 955         .sendmsg                = dccp_sendmsg,
 956         .recvmsg                = dccp_recvmsg,
 957         .backlog_rcv            = dccp_v4_do_rcv,
 958         .hash                   = inet_hash,
 959         .unhash                 = inet_unhash,
 960         .accept                 = inet_csk_accept,
 961         .get_port               = inet_csk_get_port,
 962         .shutdown               = dccp_shutdown,
 963         .destroy                = dccp_destroy_sock,
 964         .orphan_count           = &dccp_orphan_count,
 965         .max_header             = MAX_DCCP_HEADER,
 966         .obj_size               = sizeof(struct dccp_sock),
 967         .slab_flags             = SLAB_DESTROY_BY_RCU,
 968         .rsk_prot               = &dccp_request_sock_ops,
 969         .twsk_prot              = &dccp_timewait_sock_ops,
 970         .h.hashinfo             = &dccp_hashinfo,
 971 #ifdef CONFIG_COMPAT
 972         .compat_setsockopt      = compat_dccp_setsockopt,
 973         .compat_getsockopt      = compat_dccp_getsockopt,
 974 #endif
 975 };
 976
 977 static const struct net_protocol dccp_v4_protocol = {
 978         .handler        = dccp_v4_rcv,
 979         .err_handler    = dccp_v4_err,
 980         .no_policy      = 1,
 981         .netns_ok       = 1,
 982 };
 983
 984 static const struct proto_ops inet_dccp_ops = {
 985         .family            = PF_INET,
 986         .owner             = THIS_MODULE,
 987         .release           = inet_release,
 988         .bind              = inet_bind,
 989         .connect           = inet_stream_connect,
 990         .socketpair        = sock_no_socketpair,
 991         .accept            = inet_accept,
 992         .getname           = inet_getname,
 993         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 994         .poll              = dccp_poll,
 995         .ioctl             = inet_ioctl,
 996         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 997         .listen            = inet_dccp_listen,
 998         .shutdown          = inet_shutdown,
 999         .setsockopt        = sock_common_setsockopt,
1000         .getsockopt        = sock_common_getsockopt,
1001         .sendmsg           = inet_sendmsg,
1002         .recvmsg           = sock_common_recvmsg,
1003         .mmap              = sock_no_mmap,
1004         .sendpage          = sock_no_sendpage,
1005 #ifdef CONFIG_COMPAT
1006         .compat_setsockopt = compat_sock_common_setsockopt,
1007         .compat_getsockopt = compat_sock_common_getsockopt,
1008 #endif
1009 };
1010
1011 static struct inet_protosw dccp_v4_protosw = {
1012         .type           = SOCK_DCCP,
1013         .protocol       = IPPROTO_DCCP,
1014         .prot           = &dccp_v4_prot,
1015         .ops            = &inet_dccp_ops,
1016         .no_check       = 0,
1017         .flags          = INET_PROTOSW_ICSK,
1018 };
1019
1020 static int __net_init dccp_v4_init_net(struct net *net)
1021 {
1022         if (dccp_hashinfo.bhash == NULL)
1023                 return -ESOCKTNOSUPPORT;
1024
1025         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1026                                     SOCK_DCCP, IPPROTO_DCCP, net);
1027 }
1028
1029 static void __net_exit dccp_v4_exit_net(struct net *net)
1030 {
1031         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1032 }
1033
1034 static struct pernet_operations dccp_v4_ops = {
1035         .init   = dccp_v4_init_net,
1036         .exit   = dccp_v4_exit_net,
1037 };
1038
1039 static int __init dccp_v4_init(void)
1040 {
1041         int err = proto_register(&dccp_v4_prot, 1);
1042
1043         if (err != 0)
1044                 goto out;
1045
1046         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1047         if (err != 0)
1048                 goto out_proto_unregister;
1049
1050         inet_register_protosw(&dccp_v4_protosw);
1051
1052         err = register_pernet_subsys(&dccp_v4_ops);
1053         if (err)
1054                 goto out_destroy_ctl_sock;
1055 out:
1056         return err;
1057 out_destroy_ctl_sock:
1058         inet_unregister_protosw(&dccp_v4_protosw);
1059         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1060 out_proto_unregister:
1061         proto_unregister(&dccp_v4_prot);
1062         goto out;
1063 }
1064
1065 static void __exit dccp_v4_exit(void)
1066 {
1067         unregister_pernet_subsys(&dccp_v4_ops);
1068         inet_unregister_protosw(&dccp_v4_protosw);
1069         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1070         proto_unregister(&dccp_v4_prot);
1071 }
1072
1073 module_init(dccp_v4_init);
1074 module_exit(dccp_v4_exit);
1075
1076 /*
1077  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1078  * values directly, Also cover the case where the protocol is not specified,
1079  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1080  */
1081 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1082 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1083 MODULE_LICENSE("GPL");
1084 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1085 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");