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/module.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/random.h>
  18
  19 #include <net/icmp.h>
  20 #include <net/inet_common.h>
  21 #include <net/inet_hashtables.h>
  22 #include <net/inet_sock.h>
  23 #include <net/protocol.h>
  24 #include <net/sock.h>
  25 #include <net/timewait_sock.h>
  26 #include <net/tcp_states.h>
  27 #include <net/xfrm.h>
  28
  29 #include "ackvec.h"
  30 #include "ccid.h"
  31 #include "dccp.h"
  32 #include "feat.h"
  33
  34 /*
  35  * This is the global socket data structure used for responding to
  36  * the Out-of-the-blue (OOTB) packets. A control sock will be created
  37  * for this socket at the initialization time.
  38  */
  39 static struct socket *dccp_v4_ctl_socket;
  40
  41 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  42 {
  43         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
  44                                  inet_csk_bind_conflict);
  45 }
  46
  47 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  48 {
  49         struct inet_sock *inet = inet_sk(sk);
  50         struct dccp_sock *dp = dccp_sk(sk);
  51         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  52         struct rtable *rt;
  53         __be32 daddr, nexthop;
  54         int tmp;
  55         int err;
  56
  57         dp->dccps_role = DCCP_ROLE_CLIENT;
  58
  59         if (addr_len < sizeof(struct sockaddr_in))
  60                 return -EINVAL;
  61
  62         if (usin->sin_family != AF_INET)
  63                 return -EAFNOSUPPORT;
  64
  65         nexthop = daddr = usin->sin_addr.s_addr;
  66         if (inet->opt != NULL && inet->opt->srr) {
  67                 if (daddr == 0)
  68                         return -EINVAL;
  69                 nexthop = inet->opt->faddr;
  70         }
  71
  72         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
  73                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  74                                IPPROTO_DCCP,
  75                                inet->sport, usin->sin_port, sk);
  76         if (tmp < 0)
  77                 return tmp;
  78
  79         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  80                 ip_rt_put(rt);
  81                 return -ENETUNREACH;
  82         }
  83
  84         if (inet->opt == NULL || !inet->opt->srr)
  85                 daddr = rt->rt_dst;
  86
  87         if (inet->saddr == 0)
  88                 inet->saddr = rt->rt_src;
  89         inet->rcv_saddr = inet->saddr;
  90
  91         inet->dport = usin->sin_port;
  92         inet->daddr = daddr;
  93
  94         inet_csk(sk)->icsk_ext_hdr_len = 0;
  95         if (inet->opt != NULL)
  96                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
  97         /*
  98          * Socket identity is still unknown (sport may be zero).
  99          * However we set state to DCCP_REQUESTING and not releasing socket
 100          * lock select source port, enter ourselves into the hash tables and
 101          * complete initialization after this.
 102          */
 103         dccp_set_state(sk, DCCP_REQUESTING);
 104         err = inet_hash_connect(&dccp_death_row, sk);
 105         if (err != 0)
 106                 goto failure;
 107
 108         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
 109                                 sk);
 110         if (err != 0)
 111                 goto failure;
 112
 113         /* OK, now commit destination to socket.  */
 114         sk_setup_caps(sk, &rt->u.dst);
 115
 116         dp->dccps_gar =
 117                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 118                                                             inet->daddr,
 119                                                             inet->sport,
 120                                                             usin->sin_port);
 121         dccp_update_gss(sk, dp->dccps_iss);
 122
 123         inet->id = dp->dccps_iss ^ jiffies;
 124
 125         err = dccp_connect(sk);
 126         rt = NULL;
 127         if (err != 0)
 128                 goto failure;
 129 out:
 130         return err;
 131 failure:
 132         /*
 133          * This unhashes the socket and releases the local port, if necessary.
 134          */
 135         dccp_set_state(sk, DCCP_CLOSED);
 136         ip_rt_put(rt);
 137         sk->sk_route_caps = 0;
 138         inet->dport = 0;
 139         goto out;
 140 }
 141
 142 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 143
 144 /*
 145  * This routine does path mtu discovery as defined in RFC1191.
 146  */
 147 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 148                                           const struct iphdr *iph,
 149                                           u32 mtu)
 150 {
 151         struct dst_entry *dst;
 152         const struct inet_sock *inet = inet_sk(sk);
 153         const struct dccp_sock *dp = dccp_sk(sk);
 154
 155         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 156          * send out by Linux are always < 576bytes so they should go through
 157          * unfragmented).
 158          */
 159         if (sk->sk_state == DCCP_LISTEN)
 160                 return;
 161
 162         /* We don't check in the destentry if pmtu discovery is forbidden
 163          * on this route. We just assume that no packet_to_big packets
 164          * are send back when pmtu discovery is not active.
 165          * There is a small race when the user changes this flag in the
 166          * route, but I think that's acceptable.
 167          */
 168         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 169                 return;
 170
 171         dst->ops->update_pmtu(dst, mtu);
 172
 173         /* Something is about to be wrong... Remember soft error
 174          * for the case, if this connection will not able to recover.
 175          */
 176         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 177                 sk->sk_err_soft = EMSGSIZE;
 178
 179         mtu = dst_mtu(dst);
 180
 181         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 182             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 183                 dccp_sync_mss(sk, mtu);
 184
 185                 /*
 186                  * From RFC 4340, sec. 14.1:
 187                  *
 188                  *      DCCP-Sync packets are the best choice for upward
 189                  *      probing, since DCCP-Sync probes do not risk application
 190                  *      data loss.
 191                  */
 192                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 193         } /* else let the usual retransmit timer handle it */
 194 }
 195
 196 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 197                                  struct dst_entry *dst)
 198 {
 199         int err = -1;
 200         struct sk_buff *skb;
 201
 202         /* First, grab a route. */
 203
 204         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 205                 goto out;
 206
 207         skb = dccp_make_response(sk, dst, req);
 208         if (skb != NULL) {
 209                 const struct inet_request_sock *ireq = inet_rsk(req);
 210                 struct dccp_hdr *dh = dccp_hdr(skb);
 211
 212                 dh->dccph_checksum = dccp_v4_checksum(skb, ireq->loc_addr,
 213                                                       ireq->rmt_addr);
 214                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 215                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 216                                             ireq->rmt_addr,
 217                                             ireq->opt);
 218                 if (err == NET_XMIT_CN)
 219                         err = 0;
 220         }
 221
 222 out:
 223         dst_release(dst);
 224         return err;
 225 }
 226
 227 /*
 228  * This routine is called by the ICMP module when it gets some sort of error
 229  * condition. If err < 0 then the socket should be closed and the error
 230  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 231  * After adjustment header points to the first 8 bytes of the tcp header. We
 232  * need to find the appropriate port.
 233  *
 234  * The locking strategy used here is very "optimistic". When someone else
 235  * accesses the socket the ICMP is just dropped and for some paths there is no
 236  * check at all. A more general error queue to queue errors for later handling
 237  * is probably better.
 238  */
 239 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 240 {
 241         const struct iphdr *iph = (struct iphdr *)skb->data;
 242         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 243                                                         (iph->ihl << 2));
 244         struct dccp_sock *dp;
 245         struct inet_sock *inet;
 246         const int type = skb->h.icmph->type;
 247         const int code = skb->h.icmph->code;
 248         struct sock *sk;
 249         __u64 seq;
 250         int err;
 251
 252         if (skb->len < (iph->ihl << 2) + 8) {
 253                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 254                 return;
 255         }
 256
 257         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 258                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 259         if (sk == NULL) {
 260                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 261                 return;
 262         }
 263
 264         if (sk->sk_state == DCCP_TIME_WAIT) {
 265                 inet_twsk_put(inet_twsk(sk));
 266                 return;
 267         }
 268
 269         bh_lock_sock(sk);
 270         /* If too many ICMPs get dropped on busy
 271          * servers this needs to be solved differently.
 272          */
 273         if (sock_owned_by_user(sk))
 274                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 275
 276         if (sk->sk_state == DCCP_CLOSED)
 277                 goto out;
 278
 279         dp = dccp_sk(sk);
 280         seq = dccp_hdr_seq(skb);
 281         if (sk->sk_state != DCCP_LISTEN &&
 282             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 283                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 284                 goto out;
 285         }
 286
 287         switch (type) {
 288         case ICMP_SOURCE_QUENCH:
 289                 /* Just silently ignore these. */
 290                 goto out;
 291         case ICMP_PARAMETERPROB:
 292                 err = EPROTO;
 293                 break;
 294         case ICMP_DEST_UNREACH:
 295                 if (code > NR_ICMP_UNREACH)
 296                         goto out;
 297
 298                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 299                         if (!sock_owned_by_user(sk))
 300                                 dccp_do_pmtu_discovery(sk, iph, info);
 301                         goto out;
 302                 }
 303
 304                 err = icmp_err_convert[code].errno;
 305                 break;
 306         case ICMP_TIME_EXCEEDED:
 307                 err = EHOSTUNREACH;
 308                 break;
 309         default:
 310                 goto out;
 311         }
 312
 313         switch (sk->sk_state) {
 314                 struct request_sock *req , **prev;
 315         case DCCP_LISTEN:
 316                 if (sock_owned_by_user(sk))
 317                         goto out;
 318                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 319                                           iph->daddr, iph->saddr);
 320                 if (!req)
 321                         goto out;
 322
 323                 /*
 324                  * ICMPs are not backlogged, hence we cannot get an established
 325                  * socket here.
 326                  */
 327                 BUG_TRAP(!req->sk);
 328
 329                 if (seq != dccp_rsk(req)->dreq_iss) {
 330                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 331                         goto out;
 332                 }
 333                 /*
 334                  * Still in RESPOND, just remove it silently.
 335                  * There is no good way to pass the error to the newly
 336                  * created socket, and POSIX does not want network
 337                  * errors returned from accept().
 338                  */
 339                 inet_csk_reqsk_queue_drop(sk, req, prev);
 340                 goto out;
 341
 342         case DCCP_REQUESTING:
 343         case DCCP_RESPOND:
 344                 if (!sock_owned_by_user(sk)) {
 345                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 346                         sk->sk_err = err;
 347
 348                         sk->sk_error_report(sk);
 349
 350                         dccp_done(sk);
 351                 } else
 352                         sk->sk_err_soft = err;
 353                 goto out;
 354         }
 355
 356         /* If we've already connected we will keep trying
 357          * until we time out, or the user gives up.
 358          *
 359          * rfc1122 4.2.3.9 allows to consider as hard errors
 360          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 361          * but it is obsoleted by pmtu discovery).
 362          *
 363          * Note, that in modern internet, where routing is unreliable
 364          * and in each dark corner broken firewalls sit, sending random
 365          * errors ordered by their masters even this two messages finally lose
 366          * their original sense (even Linux sends invalid PORT_UNREACHs)
 367          *
 368          * Now we are in compliance with RFCs.
 369          *                                                      --ANK (980905)
 370          */
 371
 372         inet = inet_sk(sk);
 373         if (!sock_owned_by_user(sk) && inet->recverr) {
 374                 sk->sk_err = err;
 375                 sk->sk_error_report(sk);
 376         } else /* Only an error on timeout */
 377                 sk->sk_err_soft = err;
 378 out:
 379         bh_unlock_sock(sk);
 380         sock_put(sk);
 381 }
 382
 383 /* This routine computes an IPv4 DCCP checksum. */
 384 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 385 {
 386         const struct inet_sock *inet = inet_sk(sk);
 387         struct dccp_hdr *dh = dccp_hdr(skb);
 388
 389         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 390 }
 391
 392 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 393
 394 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 395                                         const struct sk_buff *skb)
 396 {
 397         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 398                                            skb->nh.iph->saddr,
 399                                            dccp_hdr(skb)->dccph_dport,
 400                                            dccp_hdr(skb)->dccph_sport);
 401 }
 402
 403 static struct request_sock_ops dccp_request_sock_ops;
 404
 405 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 406 {
 407         struct inet_request_sock *ireq;
 408         struct dccp_sock dp;
 409         struct request_sock *req;
 410         struct dccp_request_sock *dreq;
 411         const __be32 saddr = skb->nh.iph->saddr;
 412         const __be32 daddr = skb->nh.iph->daddr;
 413         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 414         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 415         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 416
 417         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 418         if (((struct rtable *)skb->dst)->rt_flags &
 419             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 420                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 421                 goto drop;
 422         }
 423
 424         if (dccp_bad_service_code(sk, service)) {
 425                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 426                 goto drop;
 427         }
 428         /*
 429          * TW buckets are converted to open requests without
 430          * limitations, they conserve resources and peer is
 431          * evidently real one.
 432          */
 433         if (inet_csk_reqsk_queue_is_full(sk))
 434                 goto drop;
 435
 436         /*
 437          * Accept backlog is full. If we have already queued enough
 438          * of warm entries in syn queue, drop request. It is better than
 439          * clogging syn queue with openreqs with exponentially increasing
 440          * timeout.
 441          */
 442         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 443                 goto drop;
 444
 445         req = reqsk_alloc(&dccp_request_sock_ops);
 446         if (req == NULL)
 447                 goto drop;
 448
 449         if (dccp_parse_options(sk, skb))
 450                 goto drop_and_free;
 451
 452         dccp_openreq_init(req, &dp, skb);
 453
 454         if (security_inet_conn_request(sk, skb, req))
 455                 goto drop_and_free;
 456
 457         ireq = inet_rsk(req);
 458         ireq->loc_addr = daddr;
 459         ireq->rmt_addr = saddr;
 460         req->rcv_wnd    = dccp_feat_default_sequence_window;
 461         ireq->opt       = NULL;
 462
 463         /*
 464          * Step 3: Process LISTEN state
 465          *
 466          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 467          *
 468          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 469          * dccp_create_openreq_child.
 470          */
 471         dreq = dccp_rsk(req);
 472         dreq->dreq_isr     = dcb->dccpd_seq;
 473         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 474         dreq->dreq_service = service;
 475
 476         if (dccp_v4_send_response(sk, req, NULL))
 477                 goto drop_and_free;
 478
 479         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 480         return 0;
 481
 482 drop_and_free:
 483         reqsk_free(req);
 484 drop:
 485         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 486         dcb->dccpd_reset_code = reset_code;
 487         return -1;
 488 }
 489
 490 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 491
 492 /*
 493  * The three way handshake has completed - we got a valid ACK or DATAACK -
 494  * now create the new socket.
 495  *
 496  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 497  */
 498 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 499                                        struct request_sock *req,
 500                                        struct dst_entry *dst)
 501 {
 502         struct inet_request_sock *ireq;
 503         struct inet_sock *newinet;
 504         struct dccp_sock *newdp;
 505         struct sock *newsk;
 506
 507         if (sk_acceptq_is_full(sk))
 508                 goto exit_overflow;
 509
 510         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 511                 goto exit;
 512
 513         newsk = dccp_create_openreq_child(sk, req, skb);
 514         if (newsk == NULL)
 515                 goto exit;
 516
 517         sk_setup_caps(newsk, dst);
 518
 519         newdp              = dccp_sk(newsk);
 520         newinet            = inet_sk(newsk);
 521         ireq               = inet_rsk(req);
 522         newinet->daddr     = ireq->rmt_addr;
 523         newinet->rcv_saddr = ireq->loc_addr;
 524         newinet->saddr     = ireq->loc_addr;
 525         newinet->opt       = ireq->opt;
 526         ireq->opt          = NULL;
 527         newinet->mc_index  = inet_iif(skb);
 528         newinet->mc_ttl    = skb->nh.iph->ttl;
 529         newinet->id        = jiffies;
 530
 531         dccp_sync_mss(newsk, dst_mtu(dst));
 532
 533         __inet_hash(&dccp_hashinfo, newsk, 0);
 534         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 535
 536         return newsk;
 537
 538 exit_overflow:
 539         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 540 exit:
 541         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 542         dst_release(dst);
 543         return NULL;
 544 }
 545
 546 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 547
 548 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 549 {
 550         const struct dccp_hdr *dh = dccp_hdr(skb);
 551         const struct iphdr *iph = skb->nh.iph;
 552         struct sock *nsk;
 553         struct request_sock **prev;
 554         /* Find possible connection requests. */
 555         struct request_sock *req = inet_csk_search_req(sk, &prev,
 556                                                        dh->dccph_sport,
 557                                                        iph->saddr, iph->daddr);
 558         if (req != NULL)
 559                 return dccp_check_req(sk, skb, req, prev);
 560
 561         nsk = inet_lookup_established(&dccp_hashinfo,
 562                                       iph->saddr, dh->dccph_sport,
 563                                       iph->daddr, dh->dccph_dport,
 564                                       inet_iif(skb));
 565         if (nsk != NULL) {
 566                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 567                         bh_lock_sock(nsk);
 568                         return nsk;
 569                 }
 570                 inet_twsk_put(inet_twsk(nsk));
 571                 return NULL;
 572         }
 573
 574         return sk;
 575 }
 576
 577 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
 578                      const __be32 daddr)
 579 {
 580         const struct dccp_hdr* dh = dccp_hdr(skb);
 581         int checksum_len;
 582         u32 tmp;
 583
 584         if (dh->dccph_cscov == 0)
 585                 checksum_len = skb->len;
 586         else {
 587                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 588                 checksum_len = checksum_len < skb->len ? checksum_len :
 589                                                          skb->len;
 590         }
 591
 592         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 593         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 594                                  IPPROTO_DCCP, tmp);
 595 }
 596
 597 EXPORT_SYMBOL_GPL(dccp_v4_checksum);
 598
 599 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 600                                    const __be32 saddr, const __be32 daddr)
 601 {
 602         struct dccp_hdr *dh = dccp_hdr(skb);
 603         int checksum_len;
 604         u32 tmp;
 605
 606         if (dh->dccph_cscov == 0)
 607                 checksum_len = skb->len;
 608         else {
 609                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 610                 checksum_len = checksum_len < skb->len ? checksum_len :
 611                                                          skb->len;
 612         }
 613         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 614         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 615                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 616 }
 617
 618 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 619                                            struct sk_buff *skb)
 620 {
 621         struct rtable *rt;
 622         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 623                             .nl_u = { .ip4_u =
 624                                       { .daddr = skb->nh.iph->saddr,
 625                                         .saddr = skb->nh.iph->daddr,
 626                                         .tos = RT_CONN_FLAGS(sk) } },
 627                             .proto = sk->sk_protocol,
 628                             .uli_u = { .ports =
 629                                        { .sport = dccp_hdr(skb)->dccph_dport,
 630                                          .dport = dccp_hdr(skb)->dccph_sport }
 631                                      }
 632                           };
 633
 634         security_skb_classify_flow(skb, &fl);
 635         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 636                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 637                 return NULL;
 638         }
 639
 640         return &rt->u.dst;
 641 }
 642
 643 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 644 {
 645         int err;
 646         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 647         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 648                                        sizeof(struct dccp_hdr_ext) +
 649                                        sizeof(struct dccp_hdr_reset);
 650         struct sk_buff *skb;
 651         struct dst_entry *dst;
 652         u64 seqno;
 653
 654         /* Never send a reset in response to a reset. */
 655         if (rxdh->dccph_type == DCCP_PKT_RESET)
 656                 return;
 657
 658         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 659                 return;
 660
 661         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
 662         if (dst == NULL)
 663                 return;
 664
 665         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header,
 666                         GFP_ATOMIC);
 667         if (skb == NULL)
 668                 goto out;
 669
 670         /* Reserve space for headers. */
 671         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
 672         skb->dst = dst_clone(dst);
 673
 674         dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
 675
 676         /* Build DCCP header and checksum it. */
 677         dh->dccph_type     = DCCP_PKT_RESET;
 678         dh->dccph_sport    = rxdh->dccph_dport;
 679         dh->dccph_dport    = rxdh->dccph_sport;
 680         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 681         dh->dccph_x        = 1;
 682         dccp_hdr_reset(skb)->dccph_reset_code =
 683                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 684
 685         /* See "8.3.1. Abnormal Termination" in RFC 4340 */
 686         seqno = 0;
 687         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 688                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 689
 690         dccp_hdr_set_seq(dh, seqno);
 691         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 692                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 693
 694         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 695                                               rxskb->nh.iph->daddr);
 696
 697         bh_lock_sock(dccp_v4_ctl_socket->sk);
 698         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
 699                                     rxskb->nh.iph->daddr,
 700                                     rxskb->nh.iph->saddr, NULL);
 701         bh_unlock_sock(dccp_v4_ctl_socket->sk);
 702
 703         if (err == NET_XMIT_CN || err == 0) {
 704                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 705                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 706         }
 707 out:
 708          dst_release(dst);
 709 }
 710
 711 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 712 {
 713         struct dccp_hdr *dh = dccp_hdr(skb);
 714
 715         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 716                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 717                         goto reset;
 718                 return 0;
 719         }
 720
 721         /*
 722          *  Step 3: Process LISTEN state
 723          *     If S.state == LISTEN,
 724          *        If P.type == Request or P contains a valid Init Cookie
 725          *              option,
 726          *           * Must scan the packet's options to check for an Init
 727          *              Cookie.  Only the Init Cookie is processed here,
 728          *              however; other options are processed in Step 8.  This
 729          *              scan need only be performed if the endpoint uses Init
 730          *              Cookies *
 731          *           * Generate a new socket and switch to that socket *
 732          *           Set S := new socket for this port pair
 733          *           S.state = RESPOND
 734          *           Choose S.ISS (initial seqno) or set from Init Cookie
 735          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 736          *           Continue with S.state == RESPOND
 737          *           * A Response packet will be generated in Step 11 *
 738          *        Otherwise,
 739          *           Generate Reset(No Connection) unless P.type == Reset
 740          *           Drop packet and return
 741          *
 742          * NOTE: the check for the packet types is done in
 743          *       dccp_rcv_state_process
 744          */
 745         if (sk->sk_state == DCCP_LISTEN) {
 746                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 747
 748                 if (nsk == NULL)
 749                         goto discard;
 750
 751                 if (nsk != sk) {
 752                         if (dccp_child_process(sk, nsk, skb))
 753                                 goto reset;
 754                         return 0;
 755                 }
 756         }
 757
 758         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 759                 goto reset;
 760         return 0;
 761
 762 reset:
 763         dccp_v4_ctl_send_reset(skb);
 764 discard:
 765         kfree_skb(skb);
 766         return 0;
 767 }
 768
 769 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 770
 771 int dccp_invalid_packet(struct sk_buff *skb)
 772 {
 773         const struct dccp_hdr *dh;
 774
 775         if (skb->pkt_type != PACKET_HOST)
 776                 return 1;
 777
 778         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 779                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 780                 return 1;
 781         }
 782
 783         dh = dccp_hdr(skb);
 784
 785         /* If the packet type is not understood, drop packet and return */
 786         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 787                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 788                 return 1;
 789         }
 790
 791         /*
 792          * If P.Data Offset is too small for packet type, or too large for
 793          * packet, drop packet and return
 794          */
 795         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 796                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 797                                             "too small 1\n",
 798                                dh->dccph_doff);
 799                 return 1;
 800         }
 801
 802         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 803                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 804                                             "too small 2\n",
 805                                dh->dccph_doff);
 806                 return 1;
 807         }
 808
 809         dh = dccp_hdr(skb);
 810
 811         /*
 812          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 813          * has short sequence numbers), drop packet and return
 814          */
 815         if (dh->dccph_x == 0 &&
 816             dh->dccph_type != DCCP_PKT_DATA &&
 817             dh->dccph_type != DCCP_PKT_ACK &&
 818             dh->dccph_type != DCCP_PKT_DATAACK) {
 819                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 820                                             "nor DataAck and P.X == 0\n",
 821                                dccp_packet_name(dh->dccph_type));
 822                 return 1;
 823         }
 824
 825         return 0;
 826 }
 827
 828 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 829
 830 /* this is called when real data arrives */
 831 static int dccp_v4_rcv(struct sk_buff *skb)
 832 {
 833         const struct dccp_hdr *dh;
 834         struct sock *sk;
 835
 836         /* Step 1: Check header basics: */
 837
 838         if (dccp_invalid_packet(skb))
 839                 goto discard_it;
 840
 841         /* If the header checksum is incorrect, drop packet and return */
 842         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
 843                                     skb->nh.iph->daddr) < 0) {
 844                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
 845                                __FUNCTION__);
 846                 goto discard_it;
 847         }
 848
 849         dh = dccp_hdr(skb);
 850
 851         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 852         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 853
 854         dccp_pr_debug("%8.8s "
 855                       "src=%u.%u.%u.%u@%-5d "
 856                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 857                       dccp_packet_name(dh->dccph_type),
 858                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 859                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 860                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 861
 862         if (dccp_packet_without_ack(skb)) {
 863                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 864                 dccp_pr_debug_cat("\n");
 865         } else {
 866                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 867                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 868                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 869         }
 870
 871         /* Step 2:
 872          *      Look up flow ID in table and get corresponding socket */
 873         sk = __inet_lookup(&dccp_hashinfo,
 874                            skb->nh.iph->saddr, dh->dccph_sport,
 875                            skb->nh.iph->daddr, dh->dccph_dport,
 876                            inet_iif(skb));
 877
 878         /*
 879          * Step 2:
 880          *      If no socket ...
 881          *              Generate Reset(No Connection) unless P.type == Reset
 882          *              Drop packet and return
 883          */
 884         if (sk == NULL) {
 885                 dccp_pr_debug("failed to look up flow ID in table and "
 886                               "get corresponding socket\n");
 887                 goto no_dccp_socket;
 888         }
 889
 890         /*
 891          * Step 2:
 892          *      ... or S.state == TIMEWAIT,
 893          *              Generate Reset(No Connection) unless P.type == Reset
 894          *              Drop packet and return
 895          */
 896         if (sk->sk_state == DCCP_TIME_WAIT) {
 897                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 898                 inet_twsk_put(inet_twsk(sk));
 899                 goto no_dccp_socket;
 900         }
 901
 902         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 903                 goto discard_and_relse;
 904         nf_reset(skb);
 905
 906         return sk_receive_skb(sk, skb);
 907
 908 no_dccp_socket:
 909         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 910                 goto discard_it;
 911         /*
 912          * Step 2:
 913          *              Generate Reset(No Connection) unless P.type == Reset
 914          *              Drop packet and return
 915          */
 916         if (dh->dccph_type != DCCP_PKT_RESET) {
 917                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 918                                         DCCP_RESET_CODE_NO_CONNECTION;
 919                 dccp_v4_ctl_send_reset(skb);
 920         }
 921
 922 discard_it:
 923         kfree_skb(skb);
 924         return 0;
 925
 926 discard_and_relse:
 927         sock_put(sk);
 928         goto discard_it;
 929 }
 930
 931 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 932         .queue_xmit        = ip_queue_xmit,
 933         .send_check        = dccp_v4_send_check,
 934         .rebuild_header    = inet_sk_rebuild_header,
 935         .conn_request      = dccp_v4_conn_request,
 936         .syn_recv_sock     = dccp_v4_request_recv_sock,
 937         .net_header_len    = sizeof(struct iphdr),
 938         .setsockopt        = ip_setsockopt,
 939         .getsockopt        = ip_getsockopt,
 940         .addr2sockaddr     = inet_csk_addr2sockaddr,
 941         .sockaddr_len      = sizeof(struct sockaddr_in),
 942 #ifdef CONFIG_COMPAT
 943         .compat_setsockopt = compat_ip_setsockopt,
 944         .compat_getsockopt = compat_ip_getsockopt,
 945 #endif
 946 };
 947
 948 static int dccp_v4_init_sock(struct sock *sk)
 949 {
 950         static __u8 dccp_v4_ctl_sock_initialized;
 951         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 952
 953         if (err == 0) {
 954                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 955                         dccp_v4_ctl_sock_initialized = 1;
 956                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 957         }
 958
 959         return err;
 960 }
 961
 962 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 963 {
 964         kfree(inet_rsk(req)->opt);
 965 }
 966
 967 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 968         .family         = PF_INET,
 969         .obj_size       = sizeof(struct dccp_request_sock),
 970         .rtx_syn_ack    = dccp_v4_send_response,
 971         .send_ack       = dccp_reqsk_send_ack,
 972         .destructor     = dccp_v4_reqsk_destructor,
 973         .send_reset     = dccp_v4_ctl_send_reset,
 974 };
 975
 976 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 977         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 978 };
 979
 980 static struct proto dccp_v4_prot = {
 981         .name                   = "DCCP",
 982         .owner                  = THIS_MODULE,
 983         .close                  = dccp_close,
 984         .connect                = dccp_v4_connect,
 985         .disconnect             = dccp_disconnect,
 986         .ioctl                  = dccp_ioctl,
 987         .init                   = dccp_v4_init_sock,
 988         .setsockopt             = dccp_setsockopt,
 989         .getsockopt             = dccp_getsockopt,
 990         .sendmsg                = dccp_sendmsg,
 991         .recvmsg                = dccp_recvmsg,
 992         .backlog_rcv            = dccp_v4_do_rcv,
 993         .hash                   = dccp_hash,
 994         .unhash                 = dccp_unhash,
 995         .accept                 = inet_csk_accept,
 996         .get_port               = dccp_v4_get_port,
 997         .shutdown               = dccp_shutdown,
 998         .destroy                = dccp_destroy_sock,
 999         .orphan_count           = &dccp_orphan_count,
1000         .max_header             = MAX_DCCP_HEADER,
1001         .obj_size               = sizeof(struct dccp_sock),
1002         .rsk_prot               = &dccp_request_sock_ops,
1003         .twsk_prot              = &dccp_timewait_sock_ops,
1004 #ifdef CONFIG_COMPAT
1005         .compat_setsockopt      = compat_dccp_setsockopt,
1006         .compat_getsockopt      = compat_dccp_getsockopt,
1007 #endif
1008 };
1009
1010 static struct net_protocol dccp_v4_protocol = {
1011         .handler        = dccp_v4_rcv,
1012         .err_handler    = dccp_v4_err,
1013         .no_policy      = 1,
1014 };
1015
1016 static const struct proto_ops inet_dccp_ops = {
1017         .family            = PF_INET,
1018         .owner             = THIS_MODULE,
1019         .release           = inet_release,
1020         .bind              = inet_bind,
1021         .connect           = inet_stream_connect,
1022         .socketpair        = sock_no_socketpair,
1023         .accept            = inet_accept,
1024         .getname           = inet_getname,
1025         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1026         .poll              = dccp_poll,
1027         .ioctl             = inet_ioctl,
1028         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1029         .listen            = inet_dccp_listen,
1030         .shutdown          = inet_shutdown,
1031         .setsockopt        = sock_common_setsockopt,
1032         .getsockopt        = sock_common_getsockopt,
1033         .sendmsg           = inet_sendmsg,
1034         .recvmsg           = sock_common_recvmsg,
1035         .mmap              = sock_no_mmap,
1036         .sendpage          = sock_no_sendpage,
1037 #ifdef CONFIG_COMPAT
1038         .compat_setsockopt = compat_sock_common_setsockopt,
1039         .compat_getsockopt = compat_sock_common_getsockopt,
1040 #endif
1041 };
1042
1043 static struct inet_protosw dccp_v4_protosw = {
1044         .type           = SOCK_DCCP,
1045         .protocol       = IPPROTO_DCCP,
1046         .prot           = &dccp_v4_prot,
1047         .ops            = &inet_dccp_ops,
1048         .capability     = -1,
1049         .no_check       = 0,
1050         .flags          = INET_PROTOSW_ICSK,
1051 };
1052
1053 static int __init dccp_v4_init(void)
1054 {
1055         int err = proto_register(&dccp_v4_prot, 1);
1056
1057         if (err != 0)
1058                 goto out;
1059
1060         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1061         if (err != 0)
1062                 goto out_proto_unregister;
1063
1064         inet_register_protosw(&dccp_v4_protosw);
1065
1066         err = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1067                                        SOCK_DCCP, IPPROTO_DCCP);
1068         if (err)
1069                 goto out_unregister_protosw;
1070 out:
1071         return err;
1072 out_unregister_protosw:
1073         inet_unregister_protosw(&dccp_v4_protosw);
1074         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1075 out_proto_unregister:
1076         proto_unregister(&dccp_v4_prot);
1077         goto out;
1078 }
1079
1080 static void __exit dccp_v4_exit(void)
1081 {
1082         inet_unregister_protosw(&dccp_v4_protosw);
1083         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1084         proto_unregister(&dccp_v4_prot);
1085 }
1086
1087 module_init(dccp_v4_init);
1088 module_exit(dccp_v4_exit);
1089
1090 /*
1091  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1092  * values directly, Also cover the case where the protocol is not specified,
1093  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1094  */
1095 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1096 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1097 MODULE_LICENSE("GPL");
1098 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1099 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");