SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/config.h>
  14 #include <linux/dccp.h>
  15 #include <linux/icmp.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_hashtables.h>
  22 #include <net/inet_sock.h>
  23 #include <net/sock.h>
  24 #include <net/timewait_sock.h>
  25 #include <net/tcp_states.h>
  26 #include <net/xfrm.h>
  27
  28 #include "ackvec.h"
  29 #include "ccid.h"
  30 #include "dccp.h"
  31 #include "feat.h"
  32
  33 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
  34         .lhash_lock     = RW_LOCK_UNLOCKED,
  35         .lhash_users    = ATOMIC_INIT(0),
  36         .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
  37 };
  38
  39 EXPORT_SYMBOL_GPL(dccp_hashinfo);
  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         u32 daddr, nexthop;
  54         int tmp;
  55         int err;
  56
  57         dp->dccps_role = DCCP_ROLE_CLIENT;
  58
  59         if (dccp_service_not_initialized(sk))
  60                 return -EPROTO;
  61
  62         if (addr_len < sizeof(struct sockaddr_in))
  63                 return -EINVAL;
  64
  65         if (usin->sin_family != AF_INET)
  66                 return -EAFNOSUPPORT;
  67
  68         nexthop = daddr = usin->sin_addr.s_addr;
  69         if (inet->opt != NULL && inet->opt->srr) {
  70                 if (daddr == 0)
  71                         return -EINVAL;
  72                 nexthop = inet->opt->faddr;
  73         }
  74
  75         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
  76                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  77                                IPPROTO_DCCP,
  78                                inet->sport, usin->sin_port, sk);
  79         if (tmp < 0)
  80                 return tmp;
  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->srr)
  88                 daddr = rt->rt_dst;
  89
  90         if (inet->saddr == 0)
  91                 inet->saddr = rt->rt_src;
  92         inet->rcv_saddr = inet->saddr;
  93
  94         inet->dport = usin->sin_port;
  95         inet->daddr = daddr;
  96
  97         inet_csk(sk)->icsk_ext_hdr_len = 0;
  98         if (inet->opt != NULL)
  99                 inet_csk(sk)->icsk_ext_hdr_len = inet->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         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
 112                                 sk);
 113         if (err != 0)
 114                 goto failure;
 115
 116         /* OK, now commit destination to socket.  */
 117         sk_setup_caps(sk, &rt->u.dst);
 118
 119         dp->dccps_gar =
 120                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 121                                                             inet->daddr,
 122                                                             inet->sport,
 123                                                             usin->sin_port);
 124         dccp_update_gss(sk, dp->dccps_iss);
 125
 126         inet->id = dp->dccps_iss ^ jiffies;
 127
 128         err = dccp_connect(sk);
 129         rt = NULL;
 130         if (err != 0)
 131                 goto failure;
 132 out:
 133         return err;
 134 failure:
 135         /*
 136          * This unhashes the socket and releases the local port, if necessary.
 137          */
 138         dccp_set_state(sk, DCCP_CLOSED);
 139         ip_rt_put(rt);
 140         sk->sk_route_caps = 0;
 141         inet->dport = 0;
 142         goto out;
 143 }
 144
 145 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 146
 147 /*
 148  * This routine does path mtu discovery as defined in RFC1191.
 149  */
 150 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 151                                           const struct iphdr *iph,
 152                                           u32 mtu)
 153 {
 154         struct dst_entry *dst;
 155         const struct inet_sock *inet = inet_sk(sk);
 156         const struct dccp_sock *dp = dccp_sk(sk);
 157
 158         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 159          * send out by Linux are always < 576bytes so they should go through
 160          * unfragmented).
 161          */
 162         if (sk->sk_state == DCCP_LISTEN)
 163                 return;
 164
 165         /* We don't check in the destentry if pmtu discovery is forbidden
 166          * on this route. We just assume that no packet_to_big packets
 167          * are send back when pmtu discovery is not active.
 168          * There is a small race when the user changes this flag in the
 169          * route, but I think that's acceptable.
 170          */
 171         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 172                 return;
 173
 174         dst->ops->update_pmtu(dst, mtu);
 175
 176         /* Something is about to be wrong... Remember soft error
 177          * for the case, if this connection will not able to recover.
 178          */
 179         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 180                 sk->sk_err_soft = EMSGSIZE;
 181
 182         mtu = dst_mtu(dst);
 183
 184         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 185             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 186                 dccp_sync_mss(sk, mtu);
 187
 188                 /*
 189                  * From: draft-ietf-dccp-spec-11.txt
 190                  *
 191                  *      DCCP-Sync packets are the best choice for upward
 192                  *      probing, since DCCP-Sync probes do not risk application
 193                  *      data loss.
 194                  */
 195                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 196         } /* else let the usual retransmit timer handle it */
 197 }
 198
 199 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 200 {
 201         int err;
 202         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 203         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 204                                      sizeof(struct dccp_hdr_ext) +
 205                                      sizeof(struct dccp_hdr_ack_bits);
 206         struct sk_buff *skb;
 207
 208         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 209                 return;
 210
 211         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 212         if (skb == NULL)
 213                 return;
 214
 215         /* Reserve space for headers. */
 216         skb_reserve(skb, MAX_DCCP_HEADER);
 217
 218         skb->dst = dst_clone(rxskb->dst);
 219
 220         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 221         dh = dccp_hdr(skb);
 222         memset(dh, 0, dccp_hdr_ack_len);
 223
 224         /* Build DCCP header and checksum it. */
 225         dh->dccph_type     = DCCP_PKT_ACK;
 226         dh->dccph_sport    = rxdh->dccph_dport;
 227         dh->dccph_dport    = rxdh->dccph_sport;
 228         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 229         dh->dccph_x        = 1;
 230
 231         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 232         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 233                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 234
 235         bh_lock_sock(dccp_ctl_socket->sk);
 236         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 237                                     rxskb->nh.iph->daddr,
 238                                     rxskb->nh.iph->saddr, NULL);
 239         bh_unlock_sock(dccp_ctl_socket->sk);
 240
 241         if (err == NET_XMIT_CN || err == 0) {
 242                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 243                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 244         }
 245 }
 246
 247 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 248                                    struct request_sock *req)
 249 {
 250         dccp_v4_ctl_send_ack(skb);
 251 }
 252
 253 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 254                                  struct dst_entry *dst)
 255 {
 256         int err = -1;
 257         struct sk_buff *skb;
 258
 259         /* First, grab a route. */
 260
 261         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 262                 goto out;
 263
 264         skb = dccp_make_response(sk, dst, req);
 265         if (skb != NULL) {
 266                 const struct inet_request_sock *ireq = inet_rsk(req);
 267
 268                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 269                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 270                                             ireq->rmt_addr,
 271                                             ireq->opt);
 272                 if (err == NET_XMIT_CN)
 273                         err = 0;
 274         }
 275
 276 out:
 277         dst_release(dst);
 278         return err;
 279 }
 280
 281 /*
 282  * This routine is called by the ICMP module when it gets some sort of error
 283  * condition. If err < 0 then the socket should be closed and the error
 284  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 285  * After adjustment header points to the first 8 bytes of the tcp header. We
 286  * need to find the appropriate port.
 287  *
 288  * The locking strategy used here is very "optimistic". When someone else
 289  * accesses the socket the ICMP is just dropped and for some paths there is no
 290  * check at all. A more general error queue to queue errors for later handling
 291  * is probably better.
 292  */
 293 void dccp_v4_err(struct sk_buff *skb, u32 info)
 294 {
 295         const struct iphdr *iph = (struct iphdr *)skb->data;
 296         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 297                                                         (iph->ihl << 2));
 298         struct dccp_sock *dp;
 299         struct inet_sock *inet;
 300         const int type = skb->h.icmph->type;
 301         const int code = skb->h.icmph->code;
 302         struct sock *sk;
 303         __u64 seq;
 304         int err;
 305
 306         if (skb->len < (iph->ihl << 2) + 8) {
 307                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 308                 return;
 309         }
 310
 311         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 312                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 313         if (sk == NULL) {
 314                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 315                 return;
 316         }
 317
 318         if (sk->sk_state == DCCP_TIME_WAIT) {
 319                 inet_twsk_put((struct inet_timewait_sock *)sk);
 320                 return;
 321         }
 322
 323         bh_lock_sock(sk);
 324         /* If too many ICMPs get dropped on busy
 325          * servers this needs to be solved differently.
 326          */
 327         if (sock_owned_by_user(sk))
 328                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 329
 330         if (sk->sk_state == DCCP_CLOSED)
 331                 goto out;
 332
 333         dp = dccp_sk(sk);
 334         seq = dccp_hdr_seq(skb);
 335         if (sk->sk_state != DCCP_LISTEN &&
 336             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 337                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 338                 goto out;
 339         }
 340
 341         switch (type) {
 342         case ICMP_SOURCE_QUENCH:
 343                 /* Just silently ignore these. */
 344                 goto out;
 345         case ICMP_PARAMETERPROB:
 346                 err = EPROTO;
 347                 break;
 348         case ICMP_DEST_UNREACH:
 349                 if (code > NR_ICMP_UNREACH)
 350                         goto out;
 351
 352                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 353                         if (!sock_owned_by_user(sk))
 354                                 dccp_do_pmtu_discovery(sk, iph, info);
 355                         goto out;
 356                 }
 357
 358                 err = icmp_err_convert[code].errno;
 359                 break;
 360         case ICMP_TIME_EXCEEDED:
 361                 err = EHOSTUNREACH;
 362                 break;
 363         default:
 364                 goto out;
 365         }
 366
 367         switch (sk->sk_state) {
 368                 struct request_sock *req , **prev;
 369         case DCCP_LISTEN:
 370                 if (sock_owned_by_user(sk))
 371                         goto out;
 372                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 373                                           iph->daddr, iph->saddr);
 374                 if (!req)
 375                         goto out;
 376
 377                 /*
 378                  * ICMPs are not backlogged, hence we cannot get an established
 379                  * socket here.
 380                  */
 381                 BUG_TRAP(!req->sk);
 382
 383                 if (seq != dccp_rsk(req)->dreq_iss) {
 384                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 385                         goto out;
 386                 }
 387                 /*
 388                  * Still in RESPOND, just remove it silently.
 389                  * There is no good way to pass the error to the newly
 390                  * created socket, and POSIX does not want network
 391                  * errors returned from accept().
 392                  */
 393                 inet_csk_reqsk_queue_drop(sk, req, prev);
 394                 goto out;
 395
 396         case DCCP_REQUESTING:
 397         case DCCP_RESPOND:
 398                 if (!sock_owned_by_user(sk)) {
 399                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 400                         sk->sk_err = err;
 401
 402                         sk->sk_error_report(sk);
 403
 404                         dccp_done(sk);
 405                 } else
 406                         sk->sk_err_soft = err;
 407                 goto out;
 408         }
 409
 410         /* If we've already connected we will keep trying
 411          * until we time out, or the user gives up.
 412          *
 413          * rfc1122 4.2.3.9 allows to consider as hard errors
 414          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 415          * but it is obsoleted by pmtu discovery).
 416          *
 417          * Note, that in modern internet, where routing is unreliable
 418          * and in each dark corner broken firewalls sit, sending random
 419          * errors ordered by their masters even this two messages finally lose
 420          * their original sense (even Linux sends invalid PORT_UNREACHs)
 421          *
 422          * Now we are in compliance with RFCs.
 423          *                                                      --ANK (980905)
 424          */
 425
 426         inet = inet_sk(sk);
 427         if (!sock_owned_by_user(sk) && inet->recverr) {
 428                 sk->sk_err = err;
 429                 sk->sk_error_report(sk);
 430         } else /* Only an error on timeout */
 431                 sk->sk_err_soft = err;
 432 out:
 433         bh_unlock_sock(sk);
 434         sock_put(sk);
 435 }
 436
 437 /* This routine computes an IPv4 DCCP checksum. */
 438 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 439 {
 440         const struct inet_sock *inet = inet_sk(sk);
 441         struct dccp_hdr *dh = dccp_hdr(skb);
 442
 443         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 444 }
 445
 446 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 447
 448 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 449                                         const struct sk_buff *skb)
 450 {
 451         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 452                                            skb->nh.iph->saddr,
 453                                            dccp_hdr(skb)->dccph_dport,
 454                                            dccp_hdr(skb)->dccph_sport);
 455 }
 456
 457 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 458 {
 459         struct inet_request_sock *ireq;
 460         struct dccp_sock dp;
 461         struct request_sock *req;
 462         struct dccp_request_sock *dreq;
 463         const __be32 saddr = skb->nh.iph->saddr;
 464         const __be32 daddr = skb->nh.iph->daddr;
 465         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 466         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 467         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 468
 469         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 470         if (((struct rtable *)skb->dst)->rt_flags &
 471             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 472                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 473                 goto drop;
 474         }
 475
 476         if (dccp_bad_service_code(sk, service)) {
 477                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 478                 goto drop;
 479         }
 480         /*
 481          * TW buckets are converted to open requests without
 482          * limitations, they conserve resources and peer is
 483          * evidently real one.
 484          */
 485         if (inet_csk_reqsk_queue_is_full(sk))
 486                 goto drop;
 487
 488         /*
 489          * Accept backlog is full. If we have already queued enough
 490          * of warm entries in syn queue, drop request. It is better than
 491          * clogging syn queue with openreqs with exponentially increasing
 492          * timeout.
 493          */
 494         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 495                 goto drop;
 496
 497         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 498         if (req == NULL)
 499                 goto drop;
 500
 501         if (dccp_parse_options(sk, skb))
 502                 goto drop;
 503
 504         dccp_openreq_init(req, &dp, skb);
 505
 506         ireq = inet_rsk(req);
 507         ireq->loc_addr = daddr;
 508         ireq->rmt_addr = saddr;
 509         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 510                                   right value */
 511         ireq->opt       = NULL;
 512
 513         /*
 514          * Step 3: Process LISTEN state
 515          *
 516          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 517          *
 518          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 519          * dccp_create_openreq_child.
 520          */
 521         dreq = dccp_rsk(req);
 522         dreq->dreq_isr     = dcb->dccpd_seq;
 523         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 524         dreq->dreq_service = service;
 525
 526         if (dccp_v4_send_response(sk, req, NULL))
 527                 goto drop_and_free;
 528
 529         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 530         return 0;
 531
 532 drop_and_free:
 533         reqsk_free(req);
 534 drop:
 535         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 536         dcb->dccpd_reset_code = reset_code;
 537         return -1;
 538 }
 539
 540 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 541
 542 /*
 543  * The three way handshake has completed - we got a valid ACK or DATAACK -
 544  * now create the new socket.
 545  *
 546  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 547  */
 548 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 549                                        struct request_sock *req,
 550                                        struct dst_entry *dst)
 551 {
 552         struct inet_request_sock *ireq;
 553         struct inet_sock *newinet;
 554         struct dccp_sock *newdp;
 555         struct sock *newsk;
 556
 557         if (sk_acceptq_is_full(sk))
 558                 goto exit_overflow;
 559
 560         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 561                 goto exit;
 562
 563         newsk = dccp_create_openreq_child(sk, req, skb);
 564         if (newsk == NULL)
 565                 goto exit;
 566
 567         sk_setup_caps(newsk, dst);
 568
 569         newdp              = dccp_sk(newsk);
 570         newinet            = inet_sk(newsk);
 571         ireq               = inet_rsk(req);
 572         newinet->daddr     = ireq->rmt_addr;
 573         newinet->rcv_saddr = ireq->loc_addr;
 574         newinet->saddr     = ireq->loc_addr;
 575         newinet->opt       = ireq->opt;
 576         ireq->opt          = NULL;
 577         newinet->mc_index  = inet_iif(skb);
 578         newinet->mc_ttl    = skb->nh.iph->ttl;
 579         newinet->id        = jiffies;
 580
 581         dccp_sync_mss(newsk, dst_mtu(dst));
 582
 583         __inet_hash(&dccp_hashinfo, newsk, 0);
 584         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 585
 586         return newsk;
 587
 588 exit_overflow:
 589         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 590 exit:
 591         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 592         dst_release(dst);
 593         return NULL;
 594 }
 595
 596 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 597
 598 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 599 {
 600         const struct dccp_hdr *dh = dccp_hdr(skb);
 601         const struct iphdr *iph = skb->nh.iph;
 602         struct sock *nsk;
 603         struct request_sock **prev;
 604         /* Find possible connection requests. */
 605         struct request_sock *req = inet_csk_search_req(sk, &prev,
 606                                                        dh->dccph_sport,
 607                                                        iph->saddr, iph->daddr);
 608         if (req != NULL)
 609                 return dccp_check_req(sk, skb, req, prev);
 610
 611         nsk = __inet_lookup_established(&dccp_hashinfo,
 612                                         iph->saddr, dh->dccph_sport,
 613                                         iph->daddr, ntohs(dh->dccph_dport),
 614                                         inet_iif(skb));
 615         if (nsk != NULL) {
 616                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 617                         bh_lock_sock(nsk);
 618                         return nsk;
 619                 }
 620                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 621                 return NULL;
 622         }
 623
 624         return sk;
 625 }
 626
 627 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
 628                      const __be32 daddr)
 629 {
 630         const struct dccp_hdr* dh = dccp_hdr(skb);
 631         int checksum_len;
 632         u32 tmp;
 633
 634         if (dh->dccph_cscov == 0)
 635                 checksum_len = skb->len;
 636         else {
 637                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 638                 checksum_len = checksum_len < skb->len ? checksum_len :
 639                                                          skb->len;
 640         }
 641
 642         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 643         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 644                                  IPPROTO_DCCP, tmp);
 645 }
 646
 647 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 648                                    const __be32 saddr, const __be32 daddr)
 649 {
 650         struct dccp_hdr *dh = dccp_hdr(skb);
 651         int checksum_len;
 652         u32 tmp;
 653
 654         if (dh->dccph_cscov == 0)
 655                 checksum_len = skb->len;
 656         else {
 657                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 658                 checksum_len = checksum_len < skb->len ? checksum_len :
 659                                                          skb->len;
 660         }
 661         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 662         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 663                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 664 }
 665
 666 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 667                                            struct sk_buff *skb)
 668 {
 669         struct rtable *rt;
 670         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 671                             .nl_u = { .ip4_u =
 672                                       { .daddr = skb->nh.iph->saddr,
 673                                         .saddr = skb->nh.iph->daddr,
 674                                         .tos = RT_CONN_FLAGS(sk) } },
 675                             .proto = sk->sk_protocol,
 676                             .uli_u = { .ports =
 677                                        { .sport = dccp_hdr(skb)->dccph_dport,
 678                                          .dport = dccp_hdr(skb)->dccph_sport }
 679                                      }
 680                           };
 681
 682         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 683                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 684                 return NULL;
 685         }
 686
 687         return &rt->u.dst;
 688 }
 689
 690 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 691 {
 692         int err;
 693         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 694         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 695                                        sizeof(struct dccp_hdr_ext) +
 696                                        sizeof(struct dccp_hdr_reset);
 697         struct sk_buff *skb;
 698         struct dst_entry *dst;
 699         u64 seqno;
 700
 701         /* Never send a reset in response to a reset. */
 702         if (rxdh->dccph_type == DCCP_PKT_RESET)
 703                 return;
 704
 705         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 706                 return;
 707
 708         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 709         if (dst == NULL)
 710                 return;
 711
 712         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 713         if (skb == NULL)
 714                 goto out;
 715
 716         /* Reserve space for headers. */
 717         skb_reserve(skb, MAX_DCCP_HEADER);
 718         skb->dst = dst_clone(dst);
 719
 720         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 721         dh = dccp_hdr(skb);
 722         memset(dh, 0, dccp_hdr_reset_len);
 723
 724         /* Build DCCP header and checksum it. */
 725         dh->dccph_type     = DCCP_PKT_RESET;
 726         dh->dccph_sport    = rxdh->dccph_dport;
 727         dh->dccph_dport    = rxdh->dccph_sport;
 728         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 729         dh->dccph_x        = 1;
 730         dccp_hdr_reset(skb)->dccph_reset_code =
 731                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 732
 733         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 734         seqno = 0;
 735         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 736                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 737
 738         dccp_hdr_set_seq(dh, seqno);
 739         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 740                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 741
 742         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 743                                               rxskb->nh.iph->daddr);
 744
 745         bh_lock_sock(dccp_ctl_socket->sk);
 746         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 747                                     rxskb->nh.iph->daddr,
 748                                     rxskb->nh.iph->saddr, NULL);
 749         bh_unlock_sock(dccp_ctl_socket->sk);
 750
 751         if (err == NET_XMIT_CN || err == 0) {
 752                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 753                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 754         }
 755 out:
 756          dst_release(dst);
 757 }
 758
 759 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 760 {
 761         struct dccp_hdr *dh = dccp_hdr(skb);
 762
 763         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 764                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 765                         goto reset;
 766                 return 0;
 767         }
 768
 769         /*
 770          *  Step 3: Process LISTEN state
 771          *     If S.state == LISTEN,
 772          *        If P.type == Request or P contains a valid Init Cookie
 773          *              option,
 774          *           * Must scan the packet's options to check for an Init
 775          *              Cookie.  Only the Init Cookie is processed here,
 776          *              however; other options are processed in Step 8.  This
 777          *              scan need only be performed if the endpoint uses Init
 778          *              Cookies *
 779          *           * Generate a new socket and switch to that socket *
 780          *           Set S := new socket for this port pair
 781          *           S.state = RESPOND
 782          *           Choose S.ISS (initial seqno) or set from Init Cookie
 783          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 784          *           Continue with S.state == RESPOND
 785          *           * A Response packet will be generated in Step 11 *
 786          *        Otherwise,
 787          *           Generate Reset(No Connection) unless P.type == Reset
 788          *           Drop packet and return
 789          *
 790          * NOTE: the check for the packet types is done in
 791          *       dccp_rcv_state_process
 792          */
 793         if (sk->sk_state == DCCP_LISTEN) {
 794                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 795
 796                 if (nsk == NULL)
 797                         goto discard;
 798
 799                 if (nsk != sk) {
 800                         if (dccp_child_process(sk, nsk, skb))
 801                                 goto reset;
 802                         return 0;
 803                 }
 804         }
 805
 806         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 807                 goto reset;
 808         return 0;
 809
 810 reset:
 811         dccp_v4_ctl_send_reset(skb);
 812 discard:
 813         kfree_skb(skb);
 814         return 0;
 815 }
 816
 817 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 818
 819 int dccp_invalid_packet(struct sk_buff *skb)
 820 {
 821         const struct dccp_hdr *dh;
 822
 823         if (skb->pkt_type != PACKET_HOST)
 824                 return 1;
 825
 826         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 827                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 828                 return 1;
 829         }
 830
 831         dh = dccp_hdr(skb);
 832
 833         /* If the packet type is not understood, drop packet and return */
 834         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 835                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 836                 return 1;
 837         }
 838
 839         /*
 840          * If P.Data Offset is too small for packet type, or too large for
 841          * packet, drop packet and return
 842          */
 843         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 844                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 845                                             "too small 1\n",
 846                                dh->dccph_doff);
 847                 return 1;
 848         }
 849
 850         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 851                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 852                                             "too small 2\n",
 853                                dh->dccph_doff);
 854                 return 1;
 855         }
 856
 857         dh = dccp_hdr(skb);
 858
 859         /*
 860          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 861          * has short sequence numbers), drop packet and return
 862          */
 863         if (dh->dccph_x == 0 &&
 864             dh->dccph_type != DCCP_PKT_DATA &&
 865             dh->dccph_type != DCCP_PKT_ACK &&
 866             dh->dccph_type != DCCP_PKT_DATAACK) {
 867                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 868                                             "nor DataAck and P.X == 0\n",
 869                                dccp_packet_name(dh->dccph_type));
 870                 return 1;
 871         }
 872
 873         return 0;
 874 }
 875
 876 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 877
 878 /* this is called when real data arrives */
 879 int dccp_v4_rcv(struct sk_buff *skb)
 880 {
 881         const struct dccp_hdr *dh;
 882         struct sock *sk;
 883
 884         /* Step 1: Check header basics: */
 885
 886         if (dccp_invalid_packet(skb))
 887                 goto discard_it;
 888
 889         /* If the header checksum is incorrect, drop packet and return */
 890         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
 891                                     skb->nh.iph->daddr) < 0) {
 892                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
 893                                __FUNCTION__);
 894                 goto discard_it;
 895         }
 896
 897         dh = dccp_hdr(skb);
 898
 899         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 900         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 901
 902         dccp_pr_debug("%8.8s "
 903                       "src=%u.%u.%u.%u@%-5d "
 904                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 905                       dccp_packet_name(dh->dccph_type),
 906                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 907                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 908                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 909
 910         if (dccp_packet_without_ack(skb)) {
 911                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 912                 dccp_pr_debug_cat("\n");
 913         } else {
 914                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 915                 dccp_pr_debug_cat(", ack=%llu\n",
 916                                   (unsigned long long)
 917                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 918         }
 919
 920         /* Step 2:
 921          *      Look up flow ID in table and get corresponding socket */
 922         sk = __inet_lookup(&dccp_hashinfo,
 923                            skb->nh.iph->saddr, dh->dccph_sport,
 924                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
 925                            inet_iif(skb));
 926
 927         /*
 928          * Step 2:
 929          *      If no socket ...
 930          *              Generate Reset(No Connection) unless P.type == Reset
 931          *              Drop packet and return
 932          */
 933         if (sk == NULL) {
 934                 dccp_pr_debug("failed to look up flow ID in table and "
 935                               "get corresponding socket\n");
 936                 goto no_dccp_socket;
 937         }
 938
 939         /*
 940          * Step 2:
 941          *      ... or S.state == TIMEWAIT,
 942          *              Generate Reset(No Connection) unless P.type == Reset
 943          *              Drop packet and return
 944          */
 945
 946         if (sk->sk_state == DCCP_TIME_WAIT) {
 947                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
 948                               "do_time_wait\n");
 949                 goto do_time_wait;
 950         }
 951
 952         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 953                 goto discard_and_relse;
 954         nf_reset(skb);
 955
 956         return sk_receive_skb(sk, skb);
 957
 958 no_dccp_socket:
 959         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 960                 goto discard_it;
 961         /*
 962          * Step 2:
 963          *              Generate Reset(No Connection) unless P.type == Reset
 964          *              Drop packet and return
 965          */
 966         if (dh->dccph_type != DCCP_PKT_RESET) {
 967                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 968                                         DCCP_RESET_CODE_NO_CONNECTION;
 969                 dccp_v4_ctl_send_reset(skb);
 970         }
 971
 972 discard_it:
 973         /* Discard frame. */
 974         kfree_skb(skb);
 975         return 0;
 976
 977 discard_and_relse:
 978         sock_put(sk);
 979         goto discard_it;
 980
 981 do_time_wait:
 982         inet_twsk_put((struct inet_timewait_sock *)sk);
 983         goto no_dccp_socket;
 984 }
 985
 986 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 987         .queue_xmit     = ip_queue_xmit,
 988         .send_check     = dccp_v4_send_check,
 989         .rebuild_header = inet_sk_rebuild_header,
 990         .conn_request   = dccp_v4_conn_request,
 991         .syn_recv_sock  = dccp_v4_request_recv_sock,
 992         .net_header_len = sizeof(struct iphdr),
 993         .setsockopt     = ip_setsockopt,
 994         .getsockopt     = ip_getsockopt,
 995         .addr2sockaddr  = inet_csk_addr2sockaddr,
 996         .sockaddr_len   = sizeof(struct sockaddr_in),
 997 };
 998
 999 static int dccp_v4_init_sock(struct sock *sk)
1000 {
1001         const int err = dccp_init_sock(sk);
1002
1003         if (err == 0)
1004                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1005         return err;
1006 }
1007
1008 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1009 {
1010         kfree(inet_rsk(req)->opt);
1011 }
1012
1013 static struct request_sock_ops dccp_request_sock_ops = {
1014         .family         = PF_INET,
1015         .obj_size       = sizeof(struct dccp_request_sock),
1016         .rtx_syn_ack    = dccp_v4_send_response,
1017         .send_ack       = dccp_v4_reqsk_send_ack,
1018         .destructor     = dccp_v4_reqsk_destructor,
1019         .send_reset     = dccp_v4_ctl_send_reset,
1020 };
1021
1022 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1023         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1024 };
1025
1026 struct proto dccp_prot = {
1027         .name                   = "DCCP",
1028         .owner                  = THIS_MODULE,
1029         .close                  = dccp_close,
1030         .connect                = dccp_v4_connect,
1031         .disconnect             = dccp_disconnect,
1032         .ioctl                  = dccp_ioctl,
1033         .init                   = dccp_v4_init_sock,
1034         .setsockopt             = dccp_setsockopt,
1035         .getsockopt             = dccp_getsockopt,
1036         .sendmsg                = dccp_sendmsg,
1037         .recvmsg                = dccp_recvmsg,
1038         .backlog_rcv            = dccp_v4_do_rcv,
1039         .hash                   = dccp_hash,
1040         .unhash                 = dccp_unhash,
1041         .accept                 = inet_csk_accept,
1042         .get_port               = dccp_v4_get_port,
1043         .shutdown               = dccp_shutdown,
1044         .destroy                = dccp_destroy_sock,
1045         .orphan_count           = &dccp_orphan_count,
1046         .max_header             = MAX_DCCP_HEADER,
1047         .obj_size               = sizeof(struct dccp_sock),
1048         .rsk_prot               = &dccp_request_sock_ops,
1049         .twsk_prot              = &dccp_timewait_sock_ops,
1050 };
1051
1052 EXPORT_SYMBOL_GPL(dccp_prot);