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;
 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         /* Only need dccph_dport & dccph_sport which are the first
 225          * 4 bytes in dccp header.
 226          * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
 227          */
 228         BUILD_BUG_ON(offsetof(struct dccp_hdr, dccph_sport) + sizeof(dh->dccph_sport) > 8);
 229         BUILD_BUG_ON(offsetof(struct dccp_hdr, dccph_dport) + sizeof(dh->dccph_dport) > 8);
 230         dh = (struct dccp_hdr *)(skb->data + offset);
 231
 232         sk = inet_lookup(net, &dccp_hashinfo,
 233                         iph->daddr, dh->dccph_dport,
 234                         iph->saddr, dh->dccph_sport, inet_iif(skb));
 235         if (sk == NULL) {
 236                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 237                 return;
 238         }
 239
 240         if (sk->sk_state == DCCP_TIME_WAIT) {
 241                 inet_twsk_put(inet_twsk(sk));
 242                 return;
 243         }
 244
 245         bh_lock_sock(sk);
 246         /* If too many ICMPs get dropped on busy
 247          * servers this needs to be solved differently.
 248          */
 249         if (sock_owned_by_user(sk))
 250                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 251
 252         if (sk->sk_state == DCCP_CLOSED)
 253                 goto out;
 254
 255         dp = dccp_sk(sk);
 256         seq = dccp_hdr_seq(dh);
 257         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 258             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 259                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 260                 goto out;
 261         }
 262
 263         switch (type) {
 264         case ICMP_SOURCE_QUENCH:
 265                 /* Just silently ignore these. */
 266                 goto out;
 267         case ICMP_PARAMETERPROB:
 268                 err = EPROTO;
 269                 break;
 270         case ICMP_DEST_UNREACH:
 271                 if (code > NR_ICMP_UNREACH)
 272                         goto out;
 273
 274                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 275                         if (!sock_owned_by_user(sk))
 276                                 dccp_do_pmtu_discovery(sk, iph, info);
 277                         goto out;
 278                 }
 279
 280                 err = icmp_err_convert[code].errno;
 281                 break;
 282         case ICMP_TIME_EXCEEDED:
 283                 err = EHOSTUNREACH;
 284                 break;
 285         default:
 286                 goto out;
 287         }
 288
 289         switch (sk->sk_state) {
 290                 struct request_sock *req , **prev;
 291         case DCCP_LISTEN:
 292                 if (sock_owned_by_user(sk))
 293                         goto out;
 294                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 295                                           iph->daddr, iph->saddr);
 296                 if (!req)
 297                         goto out;
 298
 299                 /*
 300                  * ICMPs are not backlogged, hence we cannot get an established
 301                  * socket here.
 302                  */
 303                 WARN_ON(req->sk);
 304
 305                 if (seq != dccp_rsk(req)->dreq_iss) {
 306                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 307                         goto out;
 308                 }
 309                 /*
 310                  * Still in RESPOND, just remove it silently.
 311                  * There is no good way to pass the error to the newly
 312                  * created socket, and POSIX does not want network
 313                  * errors returned from accept().
 314                  */
 315                 inet_csk_reqsk_queue_drop(sk, req, prev);
 316                 goto out;
 317
 318         case DCCP_REQUESTING:
 319         case DCCP_RESPOND:
 320                 if (!sock_owned_by_user(sk)) {
 321                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 322                         sk->sk_err = err;
 323
 324                         sk->sk_error_report(sk);
 325
 326                         dccp_done(sk);
 327                 } else
 328                         sk->sk_err_soft = err;
 329                 goto out;
 330         }
 331
 332         /* If we've already connected we will keep trying
 333          * until we time out, or the user gives up.
 334          *
 335          * rfc1122 4.2.3.9 allows to consider as hard errors
 336          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 337          * but it is obsoleted by pmtu discovery).
 338          *
 339          * Note, that in modern internet, where routing is unreliable
 340          * and in each dark corner broken firewalls sit, sending random
 341          * errors ordered by their masters even this two messages finally lose
 342          * their original sense (even Linux sends invalid PORT_UNREACHs)
 343          *
 344          * Now we are in compliance with RFCs.
 345          *                                                      --ANK (980905)
 346          */
 347
 348         inet = inet_sk(sk);
 349         if (!sock_owned_by_user(sk) && inet->recverr) {
 350                 sk->sk_err = err;
 351                 sk->sk_error_report(sk);
 352         } else /* Only an error on timeout */
 353                 sk->sk_err_soft = err;
 354 out:
 355         bh_unlock_sock(sk);
 356         sock_put(sk);
 357 }
 358
 359 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 360                                       __be32 src, __be32 dst)
 361 {
 362         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 363 }
 364
 365 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 366 {
 367         const struct inet_sock *inet = inet_sk(sk);
 368         struct dccp_hdr *dh = dccp_hdr(skb);
 369
 370         dccp_csum_outgoing(skb);
 371         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 372                                                  inet->inet_saddr,
 373                                                  inet->inet_daddr);
 374 }
 375
 376 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 377
 378 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 379 {
 380         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 381                                            ip_hdr(skb)->saddr,
 382                                            dccp_hdr(skb)->dccph_dport,
 383                                            dccp_hdr(skb)->dccph_sport);
 384 }
 385
 386 /*
 387  * The three way handshake has completed - we got a valid ACK or DATAACK -
 388  * now create the new socket.
 389  *
 390  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 391  */
 392 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 393                                        struct request_sock *req,
 394                                        struct dst_entry *dst)
 395 {
 396         struct inet_request_sock *ireq;
 397         struct inet_sock *newinet;
 398         struct sock *newsk;
 399
 400         if (sk_acceptq_is_full(sk))
 401                 goto exit_overflow;
 402
 403         newsk = dccp_create_openreq_child(sk, req, skb);
 404         if (newsk == NULL)
 405                 goto exit_nonewsk;
 406
 407         newinet            = inet_sk(newsk);
 408         ireq               = inet_rsk(req);
 409         newinet->inet_daddr     = ireq->rmt_addr;
 410         newinet->inet_rcv_saddr = ireq->loc_addr;
 411         newinet->inet_saddr     = ireq->loc_addr;
 412         newinet->inet_opt       = ireq->opt;
 413         ireq->opt          = NULL;
 414         newinet->mc_index  = inet_iif(skb);
 415         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 416         newinet->inet_id   = jiffies;
 417
 418         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 419                 goto put_and_exit;
 420
 421         sk_setup_caps(newsk, dst);
 422
 423         dccp_sync_mss(newsk, dst_mtu(dst));
 424
 425         if (__inet_inherit_port(sk, newsk) < 0)
 426                 goto put_and_exit;
 427         __inet_hash_nolisten(newsk, NULL);
 428
 429         return newsk;
 430
 431 exit_overflow:
 432         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 433 exit_nonewsk:
 434         dst_release(dst);
 435 exit:
 436         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 437         return NULL;
 438 put_and_exit:
 439         inet_csk_prepare_forced_close(newsk);
 440         dccp_done(newsk);
 441         goto exit;
 442 }
 443
 444 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 445
 446 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 447 {
 448         const struct dccp_hdr *dh = dccp_hdr(skb);
 449         const struct iphdr *iph = ip_hdr(skb);
 450         struct sock *nsk;
 451         struct request_sock **prev;
 452         /* Find possible connection requests. */
 453         struct request_sock *req = inet_csk_search_req(sk, &prev,
 454                                                        dh->dccph_sport,
 455                                                        iph->saddr, iph->daddr);
 456         if (req != NULL)
 457                 return dccp_check_req(sk, skb, req, prev);
 458
 459         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 460                                       iph->saddr, dh->dccph_sport,
 461                                       iph->daddr, dh->dccph_dport,
 462                                       inet_iif(skb));
 463         if (nsk != NULL) {
 464                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 465                         bh_lock_sock(nsk);
 466                         return nsk;
 467                 }
 468                 inet_twsk_put(inet_twsk(nsk));
 469                 return NULL;
 470         }
 471
 472         return sk;
 473 }
 474
 475 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 476                                            struct sk_buff *skb)
 477 {
 478         struct rtable *rt;
 479         struct flowi4 fl4 = {
 480                 .flowi4_oif = skb_rtable(skb)->rt_iif,
 481                 .daddr = ip_hdr(skb)->saddr,
 482                 .saddr = ip_hdr(skb)->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          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 644          * dccp_create_openreq_child.
 645          */
 646         dreq->dreq_isr     = dcb->dccpd_seq;
 647         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 648         dreq->dreq_service = service;
 649
 650         if (dccp_v4_send_response(sk, req, NULL))
 651                 goto drop_and_free;
 652
 653         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 654         return 0;
 655
 656 drop_and_free:
 657         reqsk_free(req);
 658 drop:
 659         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 660         return -1;
 661 }
 662
 663 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 664
 665 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 666 {
 667         struct dccp_hdr *dh = dccp_hdr(skb);
 668
 669         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 670                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 671                         goto reset;
 672                 return 0;
 673         }
 674
 675         /*
 676          *  Step 3: Process LISTEN state
 677          *       If P.type == Request or P contains a valid Init Cookie option,
 678          *            (* Must scan the packet's options to check for Init
 679          *               Cookies.  Only Init Cookies are processed here,
 680          *               however; other options are processed in Step 8.  This
 681          *               scan need only be performed if the endpoint uses Init
 682          *               Cookies *)
 683          *            (* Generate a new socket and switch to that socket *)
 684          *            Set S := new socket for this port pair
 685          *            S.state = RESPOND
 686          *            Choose S.ISS (initial seqno) or set from Init Cookies
 687          *            Initialize S.GAR := S.ISS
 688          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 689          *            Continue with S.state == RESPOND
 690          *            (* A Response packet will be generated in Step 11 *)
 691          *       Otherwise,
 692          *            Generate Reset(No Connection) unless P.type == Reset
 693          *            Drop packet and return
 694          *
 695          * NOTE: the check for the packet types is done in
 696          *       dccp_rcv_state_process
 697          */
 698         if (sk->sk_state == DCCP_LISTEN) {
 699                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 700
 701                 if (nsk == NULL)
 702                         goto discard;
 703
 704                 if (nsk != sk) {
 705                         if (dccp_child_process(sk, nsk, skb))
 706                                 goto reset;
 707                         return 0;
 708                 }
 709         }
 710
 711         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 712                 goto reset;
 713         return 0;
 714
 715 reset:
 716         dccp_v4_ctl_send_reset(sk, skb);
 717 discard:
 718         kfree_skb(skb);
 719         return 0;
 720 }
 721
 722 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 723
 724 /**
 725  *      dccp_invalid_packet  -  check for malformed packets
 726  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 727  *      Packets that fail these checks are ignored and do not receive Resets.
 728  */
 729 int dccp_invalid_packet(struct sk_buff *skb)
 730 {
 731         const struct dccp_hdr *dh;
 732         unsigned int cscov;
 733
 734         if (skb->pkt_type != PACKET_HOST)
 735                 return 1;
 736
 737         /* If the packet is shorter than 12 bytes, drop packet and return */
 738         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 739                 DCCP_WARN("pskb_may_pull failed\n");
 740                 return 1;
 741         }
 742
 743         dh = dccp_hdr(skb);
 744
 745         /* If P.type is not understood, drop packet and return */
 746         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 747                 DCCP_WARN("invalid packet type\n");
 748                 return 1;
 749         }
 750
 751         /*
 752          * If P.Data Offset is too small for packet type, drop packet and return
 753          */
 754         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 755                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 756                 return 1;
 757         }
 758         /*
 759          * If P.Data Offset is too too large for packet, drop packet and return
 760          */
 761         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 762                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 763                 return 1;
 764         }
 765
 766         /*
 767          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 768          * has short sequence numbers), drop packet and return
 769          */
 770         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 771             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 772                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 773                           dccp_packet_name(dh->dccph_type));
 774                 return 1;
 775         }
 776
 777         /*
 778          * If P.CsCov is too large for the packet size, drop packet and return.
 779          * This must come _before_ checksumming (not as RFC 4340 suggests).
 780          */
 781         cscov = dccp_csum_coverage(skb);
 782         if (cscov > skb->len) {
 783                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 784                           dh->dccph_cscov, skb->len);
 785                 return 1;
 786         }
 787
 788         /* If header checksum is incorrect, drop packet and return.
 789          * (This step is completed in the AF-dependent functions.) */
 790         skb->csum = skb_checksum(skb, 0, cscov, 0);
 791
 792         return 0;
 793 }
 794
 795 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 796
 797 /* this is called when real data arrives */
 798 static int dccp_v4_rcv(struct sk_buff *skb)
 799 {
 800         const struct dccp_hdr *dh;
 801         const struct iphdr *iph;
 802         struct sock *sk;
 803         int min_cov;
 804
 805         /* Step 1: Check header basics */
 806
 807         if (dccp_invalid_packet(skb))
 808                 goto discard_it;
 809
 810         iph = ip_hdr(skb);
 811         /* Step 1: If header checksum is incorrect, drop packet and return */
 812         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 813                 DCCP_WARN("dropped packet with invalid checksum\n");
 814                 goto discard_it;
 815         }
 816
 817         dh = dccp_hdr(skb);
 818
 819         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 820         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 821
 822         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 823                       dccp_packet_name(dh->dccph_type),
 824                       &iph->saddr, ntohs(dh->dccph_sport),
 825                       &iph->daddr, ntohs(dh->dccph_dport),
 826                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 827
 828         if (dccp_packet_without_ack(skb)) {
 829                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 830                 dccp_pr_debug_cat("\n");
 831         } else {
 832                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 833                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 834                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 835         }
 836
 837         /* Step 2:
 838          *      Look up flow ID in table and get corresponding socket */
 839         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 840                                dh->dccph_sport, dh->dccph_dport);
 841         /*
 842          * Step 2:
 843          *      If no socket ...
 844          */
 845         if (sk == NULL) {
 846                 dccp_pr_debug("failed to look up flow ID in table and "
 847                               "get corresponding socket\n");
 848                 goto no_dccp_socket;
 849         }
 850
 851         /*
 852          * Step 2:
 853          *      ... or S.state == TIMEWAIT,
 854          *              Generate Reset(No Connection) unless P.type == Reset
 855          *              Drop packet and return
 856          */
 857         if (sk->sk_state == DCCP_TIME_WAIT) {
 858                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 859                 inet_twsk_put(inet_twsk(sk));
 860                 goto no_dccp_socket;
 861         }
 862
 863         /*
 864          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 865          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 866          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 867          */
 868         min_cov = dccp_sk(sk)->dccps_pcrlen;
 869         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 870                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 871                               dh->dccph_cscov, min_cov);
 872                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 873                  *         options (Section 11.7) with Drop Code 0, Protocol
 874                  *         Constraints."                                     */
 875                 goto discard_and_relse;
 876         }
 877
 878         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 879                 goto discard_and_relse;
 880         nf_reset(skb);
 881
 882         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4);
 883
 884 no_dccp_socket:
 885         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 886                 goto discard_it;
 887         /*
 888          * Step 2:
 889          *      If no socket ...
 890          *              Generate Reset(No Connection) unless P.type == Reset
 891          *              Drop packet and return
 892          */
 893         if (dh->dccph_type != DCCP_PKT_RESET) {
 894                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 895                                         DCCP_RESET_CODE_NO_CONNECTION;
 896                 dccp_v4_ctl_send_reset(sk, skb);
 897         }
 898
 899 discard_it:
 900         kfree_skb(skb);
 901         return 0;
 902
 903 discard_and_relse:
 904         sock_put(sk);
 905         goto discard_it;
 906 }
 907
 908 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 909         .queue_xmit        = ip_queue_xmit,
 910         .send_check        = dccp_v4_send_check,
 911         .rebuild_header    = inet_sk_rebuild_header,
 912         .conn_request      = dccp_v4_conn_request,
 913         .syn_recv_sock     = dccp_v4_request_recv_sock,
 914         .net_header_len    = sizeof(struct iphdr),
 915         .setsockopt        = ip_setsockopt,
 916         .getsockopt        = ip_getsockopt,
 917         .addr2sockaddr     = inet_csk_addr2sockaddr,
 918         .sockaddr_len      = sizeof(struct sockaddr_in),
 919         .bind_conflict     = inet_csk_bind_conflict,
 920 #ifdef CONFIG_COMPAT
 921         .compat_setsockopt = compat_ip_setsockopt,
 922         .compat_getsockopt = compat_ip_getsockopt,
 923 #endif
 924 };
 925
 926 static int dccp_v4_init_sock(struct sock *sk)
 927 {
 928         static __u8 dccp_v4_ctl_sock_initialized;
 929         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 930
 931         if (err == 0) {
 932                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 933                         dccp_v4_ctl_sock_initialized = 1;
 934                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 935         }
 936
 937         return err;
 938 }
 939
 940 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 941         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 942 };
 943
 944 static struct proto dccp_v4_prot = {
 945         .name                   = "DCCP",
 946         .owner                  = THIS_MODULE,
 947         .close                  = dccp_close,
 948         .connect                = dccp_v4_connect,
 949         .disconnect             = dccp_disconnect,
 950         .ioctl                  = dccp_ioctl,
 951         .init                   = dccp_v4_init_sock,
 952         .setsockopt             = dccp_setsockopt,
 953         .getsockopt             = dccp_getsockopt,
 954         .sendmsg                = dccp_sendmsg,
 955         .recvmsg                = dccp_recvmsg,
 956         .backlog_rcv            = dccp_v4_do_rcv,
 957         .hash                   = inet_hash,
 958         .unhash                 = inet_unhash,
 959         .accept                 = inet_csk_accept,
 960         .get_port               = inet_csk_get_port,
 961         .shutdown               = dccp_shutdown,
 962         .destroy                = dccp_destroy_sock,
 963         .orphan_count           = &dccp_orphan_count,
 964         .max_header             = MAX_DCCP_HEADER,
 965         .obj_size               = sizeof(struct dccp_sock),
 966         .slab_flags             = SLAB_DESTROY_BY_RCU,
 967         .rsk_prot               = &dccp_request_sock_ops,
 968         .twsk_prot              = &dccp_timewait_sock_ops,
 969         .h.hashinfo             = &dccp_hashinfo,
 970 #ifdef CONFIG_COMPAT
 971         .compat_setsockopt      = compat_dccp_setsockopt,
 972         .compat_getsockopt      = compat_dccp_getsockopt,
 973 #endif
 974 };
 975
 976 static const struct net_protocol dccp_v4_protocol = {
 977         .handler        = dccp_v4_rcv,
 978         .err_handler    = dccp_v4_err,
 979         .no_policy      = 1,
 980         .netns_ok       = 1,
 981 };
 982
 983 static const struct proto_ops inet_dccp_ops = {
 984         .family            = PF_INET,
 985         .owner             = THIS_MODULE,
 986         .release           = inet_release,
 987         .bind              = inet_bind,
 988         .connect           = inet_stream_connect,
 989         .socketpair        = sock_no_socketpair,
 990         .accept            = inet_accept,
 991         .getname           = inet_getname,
 992         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 993         .poll              = dccp_poll,
 994         .ioctl             = inet_ioctl,
 995         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 996         .listen            = inet_dccp_listen,
 997         .shutdown          = inet_shutdown,
 998         .setsockopt        = sock_common_setsockopt,
 999         .getsockopt        = sock_common_getsockopt,
1000         .sendmsg           = inet_sendmsg,
1001         .recvmsg           = sock_common_recvmsg,
1002         .mmap              = sock_no_mmap,
1003         .sendpage          = sock_no_sendpage,
1004 #ifdef CONFIG_COMPAT
1005         .compat_setsockopt = compat_sock_common_setsockopt,
1006         .compat_getsockopt = compat_sock_common_getsockopt,
1007 #endif
1008 };
1009
1010 static struct inet_protosw dccp_v4_protosw = {
1011         .type           = SOCK_DCCP,
1012         .protocol       = IPPROTO_DCCP,
1013         .prot           = &dccp_v4_prot,
1014         .ops            = &inet_dccp_ops,
1015         .no_check       = 0,
1016         .flags          = INET_PROTOSW_ICSK,
1017 };
1018
1019 static int __net_init dccp_v4_init_net(struct net *net)
1020 {
1021         if (dccp_hashinfo.bhash == NULL)
1022                 return -ESOCKTNOSUPPORT;
1023
1024         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1025                                     SOCK_DCCP, IPPROTO_DCCP, net);
1026 }
1027
1028 static void __net_exit dccp_v4_exit_net(struct net *net)
1029 {
1030         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1031 }
1032
1033 static struct pernet_operations dccp_v4_ops = {
1034         .init   = dccp_v4_init_net,
1035         .exit   = dccp_v4_exit_net,
1036 };
1037
1038 static int __init dccp_v4_init(void)
1039 {
1040         int err = proto_register(&dccp_v4_prot, 1);
1041
1042         if (err != 0)
1043                 goto out;
1044
1045         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1046         if (err != 0)
1047                 goto out_proto_unregister;
1048
1049         inet_register_protosw(&dccp_v4_protosw);
1050
1051         err = register_pernet_subsys(&dccp_v4_ops);
1052         if (err)
1053                 goto out_destroy_ctl_sock;
1054 out:
1055         return err;
1056 out_destroy_ctl_sock:
1057         inet_unregister_protosw(&dccp_v4_protosw);
1058         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1059 out_proto_unregister:
1060         proto_unregister(&dccp_v4_prot);
1061         goto out;
1062 }
1063
1064 static void __exit dccp_v4_exit(void)
1065 {
1066         unregister_pernet_subsys(&dccp_v4_ops);
1067         inet_unregister_protosw(&dccp_v4_protosw);
1068         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1069         proto_unregister(&dccp_v4_prot);
1070 }
1071
1072 module_init(dccp_v4_init);
1073 module_exit(dccp_v4_exit);
1074
1075 /*
1076  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1077  * values directly, Also cover the case where the protocol is not specified,
1078  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1079  */
1080 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1081 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1082 MODULE_LICENSE("GPL");
1083 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1084 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");