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                 struct dccp_hdr *dh = dccp_hdr(skb);
 268
 269                 dh->dccph_checksum = dccp_v4_checksum(skb, ireq->loc_addr,
 270                                                       ireq->rmt_addr);
 271                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 272                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 273                                             ireq->rmt_addr,
 274                                             ireq->opt);
 275                 if (err == NET_XMIT_CN)
 276                         err = 0;
 277         }
 278
 279 out:
 280         dst_release(dst);
 281         return err;
 282 }
 283
 284 /*
 285  * This routine is called by the ICMP module when it gets some sort of error
 286  * condition. If err < 0 then the socket should be closed and the error
 287  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 288  * After adjustment header points to the first 8 bytes of the tcp header. We
 289  * need to find the appropriate port.
 290  *
 291  * The locking strategy used here is very "optimistic". When someone else
 292  * accesses the socket the ICMP is just dropped and for some paths there is no
 293  * check at all. A more general error queue to queue errors for later handling
 294  * is probably better.
 295  */
 296 void dccp_v4_err(struct sk_buff *skb, u32 info)
 297 {
 298         const struct iphdr *iph = (struct iphdr *)skb->data;
 299         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 300                                                         (iph->ihl << 2));
 301         struct dccp_sock *dp;
 302         struct inet_sock *inet;
 303         const int type = skb->h.icmph->type;
 304         const int code = skb->h.icmph->code;
 305         struct sock *sk;
 306         __u64 seq;
 307         int err;
 308
 309         if (skb->len < (iph->ihl << 2) + 8) {
 310                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 311                 return;
 312         }
 313
 314         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 315                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 316         if (sk == NULL) {
 317                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 318                 return;
 319         }
 320
 321         if (sk->sk_state == DCCP_TIME_WAIT) {
 322                 inet_twsk_put((struct inet_timewait_sock *)sk);
 323                 return;
 324         }
 325
 326         bh_lock_sock(sk);
 327         /* If too many ICMPs get dropped on busy
 328          * servers this needs to be solved differently.
 329          */
 330         if (sock_owned_by_user(sk))
 331                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 332
 333         if (sk->sk_state == DCCP_CLOSED)
 334                 goto out;
 335
 336         dp = dccp_sk(sk);
 337         seq = dccp_hdr_seq(skb);
 338         if (sk->sk_state != DCCP_LISTEN &&
 339             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 340                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 341                 goto out;
 342         }
 343
 344         switch (type) {
 345         case ICMP_SOURCE_QUENCH:
 346                 /* Just silently ignore these. */
 347                 goto out;
 348         case ICMP_PARAMETERPROB:
 349                 err = EPROTO;
 350                 break;
 351         case ICMP_DEST_UNREACH:
 352                 if (code > NR_ICMP_UNREACH)
 353                         goto out;
 354
 355                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 356                         if (!sock_owned_by_user(sk))
 357                                 dccp_do_pmtu_discovery(sk, iph, info);
 358                         goto out;
 359                 }
 360
 361                 err = icmp_err_convert[code].errno;
 362                 break;
 363         case ICMP_TIME_EXCEEDED:
 364                 err = EHOSTUNREACH;
 365                 break;
 366         default:
 367                 goto out;
 368         }
 369
 370         switch (sk->sk_state) {
 371                 struct request_sock *req , **prev;
 372         case DCCP_LISTEN:
 373                 if (sock_owned_by_user(sk))
 374                         goto out;
 375                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 376                                           iph->daddr, iph->saddr);
 377                 if (!req)
 378                         goto out;
 379
 380                 /*
 381                  * ICMPs are not backlogged, hence we cannot get an established
 382                  * socket here.
 383                  */
 384                 BUG_TRAP(!req->sk);
 385
 386                 if (seq != dccp_rsk(req)->dreq_iss) {
 387                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 388                         goto out;
 389                 }
 390                 /*
 391                  * Still in RESPOND, just remove it silently.
 392                  * There is no good way to pass the error to the newly
 393                  * created socket, and POSIX does not want network
 394                  * errors returned from accept().
 395                  */
 396                 inet_csk_reqsk_queue_drop(sk, req, prev);
 397                 goto out;
 398
 399         case DCCP_REQUESTING:
 400         case DCCP_RESPOND:
 401                 if (!sock_owned_by_user(sk)) {
 402                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 403                         sk->sk_err = err;
 404
 405                         sk->sk_error_report(sk);
 406
 407                         dccp_done(sk);
 408                 } else
 409                         sk->sk_err_soft = err;
 410                 goto out;
 411         }
 412
 413         /* If we've already connected we will keep trying
 414          * until we time out, or the user gives up.
 415          *
 416          * rfc1122 4.2.3.9 allows to consider as hard errors
 417          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 418          * but it is obsoleted by pmtu discovery).
 419          *
 420          * Note, that in modern internet, where routing is unreliable
 421          * and in each dark corner broken firewalls sit, sending random
 422          * errors ordered by their masters even this two messages finally lose
 423          * their original sense (even Linux sends invalid PORT_UNREACHs)
 424          *
 425          * Now we are in compliance with RFCs.
 426          *                                                      --ANK (980905)
 427          */
 428
 429         inet = inet_sk(sk);
 430         if (!sock_owned_by_user(sk) && inet->recverr) {
 431                 sk->sk_err = err;
 432                 sk->sk_error_report(sk);
 433         } else /* Only an error on timeout */
 434                 sk->sk_err_soft = err;
 435 out:
 436         bh_unlock_sock(sk);
 437         sock_put(sk);
 438 }
 439
 440 /* This routine computes an IPv4 DCCP checksum. */
 441 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 442 {
 443         const struct inet_sock *inet = inet_sk(sk);
 444         struct dccp_hdr *dh = dccp_hdr(skb);
 445
 446         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 447 }
 448
 449 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 450
 451 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 452                                         const struct sk_buff *skb)
 453 {
 454         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 455                                            skb->nh.iph->saddr,
 456                                            dccp_hdr(skb)->dccph_dport,
 457                                            dccp_hdr(skb)->dccph_sport);
 458 }
 459
 460 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 461 {
 462         struct inet_request_sock *ireq;
 463         struct dccp_sock dp;
 464         struct request_sock *req;
 465         struct dccp_request_sock *dreq;
 466         const __be32 saddr = skb->nh.iph->saddr;
 467         const __be32 daddr = skb->nh.iph->daddr;
 468         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 469         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 470         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 471
 472         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 473         if (((struct rtable *)skb->dst)->rt_flags &
 474             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 475                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 476                 goto drop;
 477         }
 478
 479         if (dccp_bad_service_code(sk, service)) {
 480                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 481                 goto drop;
 482         }
 483         /*
 484          * TW buckets are converted to open requests without
 485          * limitations, they conserve resources and peer is
 486          * evidently real one.
 487          */
 488         if (inet_csk_reqsk_queue_is_full(sk))
 489                 goto drop;
 490
 491         /*
 492          * Accept backlog is full. If we have already queued enough
 493          * of warm entries in syn queue, drop request. It is better than
 494          * clogging syn queue with openreqs with exponentially increasing
 495          * timeout.
 496          */
 497         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 498                 goto drop;
 499
 500         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 501         if (req == NULL)
 502                 goto drop;
 503
 504         if (dccp_parse_options(sk, skb))
 505                 goto drop;
 506
 507         dccp_openreq_init(req, &dp, skb);
 508
 509         ireq = inet_rsk(req);
 510         ireq->loc_addr = daddr;
 511         ireq->rmt_addr = saddr;
 512         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 513                                   right value */
 514         ireq->opt       = NULL;
 515
 516         /*
 517          * Step 3: Process LISTEN state
 518          *
 519          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 520          *
 521          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 522          * dccp_create_openreq_child.
 523          */
 524         dreq = dccp_rsk(req);
 525         dreq->dreq_isr     = dcb->dccpd_seq;
 526         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 527         dreq->dreq_service = service;
 528
 529         if (dccp_v4_send_response(sk, req, NULL))
 530                 goto drop_and_free;
 531
 532         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 533         return 0;
 534
 535 drop_and_free:
 536         reqsk_free(req);
 537 drop:
 538         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 539         dcb->dccpd_reset_code = reset_code;
 540         return -1;
 541 }
 542
 543 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 544
 545 /*
 546  * The three way handshake has completed - we got a valid ACK or DATAACK -
 547  * now create the new socket.
 548  *
 549  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 550  */
 551 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 552                                        struct request_sock *req,
 553                                        struct dst_entry *dst)
 554 {
 555         struct inet_request_sock *ireq;
 556         struct inet_sock *newinet;
 557         struct dccp_sock *newdp;
 558         struct sock *newsk;
 559
 560         if (sk_acceptq_is_full(sk))
 561                 goto exit_overflow;
 562
 563         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 564                 goto exit;
 565
 566         newsk = dccp_create_openreq_child(sk, req, skb);
 567         if (newsk == NULL)
 568                 goto exit;
 569
 570         sk_setup_caps(newsk, dst);
 571
 572         newdp              = dccp_sk(newsk);
 573         newinet            = inet_sk(newsk);
 574         ireq               = inet_rsk(req);
 575         newinet->daddr     = ireq->rmt_addr;
 576         newinet->rcv_saddr = ireq->loc_addr;
 577         newinet->saddr     = ireq->loc_addr;
 578         newinet->opt       = ireq->opt;
 579         ireq->opt          = NULL;
 580         newinet->mc_index  = inet_iif(skb);
 581         newinet->mc_ttl    = skb->nh.iph->ttl;
 582         newinet->id        = jiffies;
 583
 584         dccp_sync_mss(newsk, dst_mtu(dst));
 585
 586         __inet_hash(&dccp_hashinfo, newsk, 0);
 587         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 588
 589         return newsk;
 590
 591 exit_overflow:
 592         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 593 exit:
 594         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 595         dst_release(dst);
 596         return NULL;
 597 }
 598
 599 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 600
 601 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 602 {
 603         const struct dccp_hdr *dh = dccp_hdr(skb);
 604         const struct iphdr *iph = skb->nh.iph;
 605         struct sock *nsk;
 606         struct request_sock **prev;
 607         /* Find possible connection requests. */
 608         struct request_sock *req = inet_csk_search_req(sk, &prev,
 609                                                        dh->dccph_sport,
 610                                                        iph->saddr, iph->daddr);
 611         if (req != NULL)
 612                 return dccp_check_req(sk, skb, req, prev);
 613
 614         nsk = __inet_lookup_established(&dccp_hashinfo,
 615                                         iph->saddr, dh->dccph_sport,
 616                                         iph->daddr, ntohs(dh->dccph_dport),
 617                                         inet_iif(skb));
 618         if (nsk != NULL) {
 619                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 620                         bh_lock_sock(nsk);
 621                         return nsk;
 622                 }
 623                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 624                 return NULL;
 625         }
 626
 627         return sk;
 628 }
 629
 630 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
 631                      const __be32 daddr)
 632 {
 633         const struct dccp_hdr* dh = dccp_hdr(skb);
 634         int checksum_len;
 635         u32 tmp;
 636
 637         if (dh->dccph_cscov == 0)
 638                 checksum_len = skb->len;
 639         else {
 640                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 641                 checksum_len = checksum_len < skb->len ? checksum_len :
 642                                                          skb->len;
 643         }
 644
 645         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 646         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 647                                  IPPROTO_DCCP, tmp);
 648 }
 649
 650 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 651                                    const __be32 saddr, const __be32 daddr)
 652 {
 653         struct dccp_hdr *dh = dccp_hdr(skb);
 654         int checksum_len;
 655         u32 tmp;
 656
 657         if (dh->dccph_cscov == 0)
 658                 checksum_len = skb->len;
 659         else {
 660                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 661                 checksum_len = checksum_len < skb->len ? checksum_len :
 662                                                          skb->len;
 663         }
 664         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 665         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 666                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 667 }
 668
 669 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 670                                            struct sk_buff *skb)
 671 {
 672         struct rtable *rt;
 673         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 674                             .nl_u = { .ip4_u =
 675                                       { .daddr = skb->nh.iph->saddr,
 676                                         .saddr = skb->nh.iph->daddr,
 677                                         .tos = RT_CONN_FLAGS(sk) } },
 678                             .proto = sk->sk_protocol,
 679                             .uli_u = { .ports =
 680                                        { .sport = dccp_hdr(skb)->dccph_dport,
 681                                          .dport = dccp_hdr(skb)->dccph_sport }
 682                                      }
 683                           };
 684
 685         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 686                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 687                 return NULL;
 688         }
 689
 690         return &rt->u.dst;
 691 }
 692
 693 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 694 {
 695         int err;
 696         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 697         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 698                                        sizeof(struct dccp_hdr_ext) +
 699                                        sizeof(struct dccp_hdr_reset);
 700         struct sk_buff *skb;
 701         struct dst_entry *dst;
 702         u64 seqno;
 703
 704         /* Never send a reset in response to a reset. */
 705         if (rxdh->dccph_type == DCCP_PKT_RESET)
 706                 return;
 707
 708         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 709                 return;
 710
 711         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 712         if (dst == NULL)
 713                 return;
 714
 715         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 716         if (skb == NULL)
 717                 goto out;
 718
 719         /* Reserve space for headers. */
 720         skb_reserve(skb, MAX_DCCP_HEADER);
 721         skb->dst = dst_clone(dst);
 722
 723         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 724         dh = dccp_hdr(skb);
 725         memset(dh, 0, dccp_hdr_reset_len);
 726
 727         /* Build DCCP header and checksum it. */
 728         dh->dccph_type     = DCCP_PKT_RESET;
 729         dh->dccph_sport    = rxdh->dccph_dport;
 730         dh->dccph_dport    = rxdh->dccph_sport;
 731         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 732         dh->dccph_x        = 1;
 733         dccp_hdr_reset(skb)->dccph_reset_code =
 734                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 735
 736         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 737         seqno = 0;
 738         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 739                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 740
 741         dccp_hdr_set_seq(dh, seqno);
 742         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 743                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 744
 745         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 746                                               rxskb->nh.iph->daddr);
 747
 748         bh_lock_sock(dccp_ctl_socket->sk);
 749         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 750                                     rxskb->nh.iph->daddr,
 751                                     rxskb->nh.iph->saddr, NULL);
 752         bh_unlock_sock(dccp_ctl_socket->sk);
 753
 754         if (err == NET_XMIT_CN || err == 0) {
 755                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 756                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 757         }
 758 out:
 759          dst_release(dst);
 760 }
 761
 762 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 763 {
 764         struct dccp_hdr *dh = dccp_hdr(skb);
 765
 766         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 767                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 768                         goto reset;
 769                 return 0;
 770         }
 771
 772         /*
 773          *  Step 3: Process LISTEN state
 774          *     If S.state == LISTEN,
 775          *        If P.type == Request or P contains a valid Init Cookie
 776          *              option,
 777          *           * Must scan the packet's options to check for an Init
 778          *              Cookie.  Only the Init Cookie is processed here,
 779          *              however; other options are processed in Step 8.  This
 780          *              scan need only be performed if the endpoint uses Init
 781          *              Cookies *
 782          *           * Generate a new socket and switch to that socket *
 783          *           Set S := new socket for this port pair
 784          *           S.state = RESPOND
 785          *           Choose S.ISS (initial seqno) or set from Init Cookie
 786          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 787          *           Continue with S.state == RESPOND
 788          *           * A Response packet will be generated in Step 11 *
 789          *        Otherwise,
 790          *           Generate Reset(No Connection) unless P.type == Reset
 791          *           Drop packet and return
 792          *
 793          * NOTE: the check for the packet types is done in
 794          *       dccp_rcv_state_process
 795          */
 796         if (sk->sk_state == DCCP_LISTEN) {
 797                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 798
 799                 if (nsk == NULL)
 800                         goto discard;
 801
 802                 if (nsk != sk) {
 803                         if (dccp_child_process(sk, nsk, skb))
 804                                 goto reset;
 805                         return 0;
 806                 }
 807         }
 808
 809         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 810                 goto reset;
 811         return 0;
 812
 813 reset:
 814         dccp_v4_ctl_send_reset(skb);
 815 discard:
 816         kfree_skb(skb);
 817         return 0;
 818 }
 819
 820 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 821
 822 int dccp_invalid_packet(struct sk_buff *skb)
 823 {
 824         const struct dccp_hdr *dh;
 825
 826         if (skb->pkt_type != PACKET_HOST)
 827                 return 1;
 828
 829         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 830                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 831                 return 1;
 832         }
 833
 834         dh = dccp_hdr(skb);
 835
 836         /* If the packet type is not understood, drop packet and return */
 837         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 838                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 839                 return 1;
 840         }
 841
 842         /*
 843          * If P.Data Offset is too small for packet type, or too large for
 844          * packet, drop packet and return
 845          */
 846         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 847                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 848                                             "too small 1\n",
 849                                dh->dccph_doff);
 850                 return 1;
 851         }
 852
 853         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 854                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 855                                             "too small 2\n",
 856                                dh->dccph_doff);
 857                 return 1;
 858         }
 859
 860         dh = dccp_hdr(skb);
 861
 862         /*
 863          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 864          * has short sequence numbers), drop packet and return
 865          */
 866         if (dh->dccph_x == 0 &&
 867             dh->dccph_type != DCCP_PKT_DATA &&
 868             dh->dccph_type != DCCP_PKT_ACK &&
 869             dh->dccph_type != DCCP_PKT_DATAACK) {
 870                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 871                                             "nor DataAck and P.X == 0\n",
 872                                dccp_packet_name(dh->dccph_type));
 873                 return 1;
 874         }
 875
 876         return 0;
 877 }
 878
 879 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 880
 881 /* this is called when real data arrives */
 882 int dccp_v4_rcv(struct sk_buff *skb)
 883 {
 884         const struct dccp_hdr *dh;
 885         struct sock *sk;
 886
 887         /* Step 1: Check header basics: */
 888
 889         if (dccp_invalid_packet(skb))
 890                 goto discard_it;
 891
 892         /* If the header checksum is incorrect, drop packet and return */
 893         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
 894                                     skb->nh.iph->daddr) < 0) {
 895                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
 896                                __FUNCTION__);
 897                 goto discard_it;
 898         }
 899
 900         dh = dccp_hdr(skb);
 901
 902         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 903         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 904
 905         dccp_pr_debug("%8.8s "
 906                       "src=%u.%u.%u.%u@%-5d "
 907                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 908                       dccp_packet_name(dh->dccph_type),
 909                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 910                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 911                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 912
 913         if (dccp_packet_without_ack(skb)) {
 914                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 915                 dccp_pr_debug_cat("\n");
 916         } else {
 917                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 918                 dccp_pr_debug_cat(", ack=%llu\n",
 919                                   (unsigned long long)
 920                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 921         }
 922
 923         /* Step 2:
 924          *      Look up flow ID in table and get corresponding socket */
 925         sk = __inet_lookup(&dccp_hashinfo,
 926                            skb->nh.iph->saddr, dh->dccph_sport,
 927                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
 928                            inet_iif(skb));
 929
 930         /*
 931          * Step 2:
 932          *      If no socket ...
 933          *              Generate Reset(No Connection) unless P.type == Reset
 934          *              Drop packet and return
 935          */
 936         if (sk == NULL) {
 937                 dccp_pr_debug("failed to look up flow ID in table and "
 938                               "get corresponding socket\n");
 939                 goto no_dccp_socket;
 940         }
 941
 942         /*
 943          * Step 2:
 944          *      ... or S.state == TIMEWAIT,
 945          *              Generate Reset(No Connection) unless P.type == Reset
 946          *              Drop packet and return
 947          */
 948
 949         if (sk->sk_state == DCCP_TIME_WAIT) {
 950                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
 951                               "do_time_wait\n");
 952                 goto do_time_wait;
 953         }
 954
 955         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 956                 goto discard_and_relse;
 957         nf_reset(skb);
 958
 959         return sk_receive_skb(sk, skb);
 960
 961 no_dccp_socket:
 962         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 963                 goto discard_it;
 964         /*
 965          * Step 2:
 966          *              Generate Reset(No Connection) unless P.type == Reset
 967          *              Drop packet and return
 968          */
 969         if (dh->dccph_type != DCCP_PKT_RESET) {
 970                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 971                                         DCCP_RESET_CODE_NO_CONNECTION;
 972                 dccp_v4_ctl_send_reset(skb);
 973         }
 974
 975 discard_it:
 976         /* Discard frame. */
 977         kfree_skb(skb);
 978         return 0;
 979
 980 discard_and_relse:
 981         sock_put(sk);
 982         goto discard_it;
 983
 984 do_time_wait:
 985         inet_twsk_put((struct inet_timewait_sock *)sk);
 986         goto no_dccp_socket;
 987 }
 988
 989 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 990         .queue_xmit     = ip_queue_xmit,
 991         .send_check     = dccp_v4_send_check,
 992         .rebuild_header = inet_sk_rebuild_header,
 993         .conn_request   = dccp_v4_conn_request,
 994         .syn_recv_sock  = dccp_v4_request_recv_sock,
 995         .net_header_len = sizeof(struct iphdr),
 996         .setsockopt     = ip_setsockopt,
 997         .getsockopt     = ip_getsockopt,
 998         .addr2sockaddr  = inet_csk_addr2sockaddr,
 999         .sockaddr_len   = sizeof(struct sockaddr_in),
1000 };
1001
1002 static int dccp_v4_init_sock(struct sock *sk)
1003 {
1004         const int err = dccp_init_sock(sk);
1005
1006         if (err == 0)
1007                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1008         return err;
1009 }
1010
1011 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1012 {
1013         kfree(inet_rsk(req)->opt);
1014 }
1015
1016 static struct request_sock_ops dccp_request_sock_ops = {
1017         .family         = PF_INET,
1018         .obj_size       = sizeof(struct dccp_request_sock),
1019         .rtx_syn_ack    = dccp_v4_send_response,
1020         .send_ack       = dccp_v4_reqsk_send_ack,
1021         .destructor     = dccp_v4_reqsk_destructor,
1022         .send_reset     = dccp_v4_ctl_send_reset,
1023 };
1024
1025 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1026         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1027 };
1028
1029 struct proto dccp_prot = {
1030         .name                   = "DCCP",
1031         .owner                  = THIS_MODULE,
1032         .close                  = dccp_close,
1033         .connect                = dccp_v4_connect,
1034         .disconnect             = dccp_disconnect,
1035         .ioctl                  = dccp_ioctl,
1036         .init                   = dccp_v4_init_sock,
1037         .setsockopt             = dccp_setsockopt,
1038         .getsockopt             = dccp_getsockopt,
1039         .sendmsg                = dccp_sendmsg,
1040         .recvmsg                = dccp_recvmsg,
1041         .backlog_rcv            = dccp_v4_do_rcv,
1042         .hash                   = dccp_hash,
1043         .unhash                 = dccp_unhash,
1044         .accept                 = inet_csk_accept,
1045         .get_port               = dccp_v4_get_port,
1046         .shutdown               = dccp_shutdown,
1047         .destroy                = dccp_destroy_sock,
1048         .orphan_count           = &dccp_orphan_count,
1049         .max_header             = MAX_DCCP_HEADER,
1050         .obj_size               = sizeof(struct dccp_sock),
1051         .rsk_prot               = &dccp_request_sock_ops,
1052         .twsk_prot              = &dccp_timewait_sock_ops,
1053 };
1054
1055 EXPORT_SYMBOL_GPL(dccp_prot);