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