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
  32 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
  33         .lhash_lock     = RW_LOCK_UNLOCKED,
  34         .lhash_users    = ATOMIC_INIT(0),
  35         .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
  36 };
  37
  38 EXPORT_SYMBOL_GPL(dccp_hashinfo);
  39
  40 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  41 {
  42         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
  43                                  inet_csk_bind_conflict);
  44 }
  45
  46 static void dccp_v4_hash(struct sock *sk)
  47 {
  48         inet_hash(&dccp_hashinfo, sk);
  49 }
  50
  51 void dccp_unhash(struct sock *sk)
  52 {
  53         inet_unhash(&dccp_hashinfo, sk);
  54 }
  55
  56 EXPORT_SYMBOL_GPL(dccp_unhash);
  57
  58 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  59 {
  60         struct inet_sock *inet = inet_sk(sk);
  61         struct dccp_sock *dp = dccp_sk(sk);
  62         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  63         struct rtable *rt;
  64         u32 daddr, nexthop;
  65         int tmp;
  66         int err;
  67
  68         dp->dccps_role = DCCP_ROLE_CLIENT;
  69
  70         if (dccp_service_not_initialized(sk))
  71                 return -EPROTO;
  72
  73         if (addr_len < sizeof(struct sockaddr_in))
  74                 return -EINVAL;
  75
  76         if (usin->sin_family != AF_INET)
  77                 return -EAFNOSUPPORT;
  78
  79         nexthop = daddr = usin->sin_addr.s_addr;
  80         if (inet->opt != NULL && inet->opt->srr) {
  81                 if (daddr == 0)
  82                         return -EINVAL;
  83                 nexthop = inet->opt->faddr;
  84         }
  85
  86         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
  87                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  88                                IPPROTO_DCCP,
  89                                inet->sport, usin->sin_port, sk);
  90         if (tmp < 0)
  91                 return tmp;
  92
  93         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  94                 ip_rt_put(rt);
  95                 return -ENETUNREACH;
  96         }
  97
  98         if (inet->opt == NULL || !inet->opt->srr)
  99                 daddr = rt->rt_dst;
 100
 101         if (inet->saddr == 0)
 102                 inet->saddr = rt->rt_src;
 103         inet->rcv_saddr = inet->saddr;
 104
 105         inet->dport = usin->sin_port;
 106         inet->daddr = daddr;
 107
 108         inet_csk(sk)->icsk_ext_hdr_len = 0;
 109         if (inet->opt != NULL)
 110                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
 111         /*
 112          * Socket identity is still unknown (sport may be zero).
 113          * However we set state to DCCP_REQUESTING and not releasing socket
 114          * lock select source port, enter ourselves into the hash tables and
 115          * complete initialization after this.
 116          */
 117         dccp_set_state(sk, DCCP_REQUESTING);
 118         err = inet_hash_connect(&dccp_death_row, sk);
 119         if (err != 0)
 120                 goto failure;
 121
 122         err = ip_route_newports(&rt, inet->sport, inet->dport, sk);
 123         if (err != 0)
 124                 goto failure;
 125
 126         /* OK, now commit destination to socket.  */
 127         sk_setup_caps(sk, &rt->u.dst);
 128
 129         dp->dccps_gar =
 130                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 131                                                             inet->daddr,
 132                                                             inet->sport,
 133                                                             usin->sin_port);
 134         dccp_update_gss(sk, dp->dccps_iss);
 135
 136         inet->id = dp->dccps_iss ^ jiffies;
 137
 138         err = dccp_connect(sk);
 139         rt = NULL;
 140         if (err != 0)
 141                 goto failure;
 142 out:
 143         return err;
 144 failure:
 145         /*
 146          * This unhashes the socket and releases the local port, if necessary.
 147          */
 148         dccp_set_state(sk, DCCP_CLOSED);
 149         ip_rt_put(rt);
 150         sk->sk_route_caps = 0;
 151         inet->dport = 0;
 152         goto out;
 153 }
 154
 155 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 156
 157 /*
 158  * This routine does path mtu discovery as defined in RFC1191.
 159  */
 160 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 161                                           const struct iphdr *iph,
 162                                           u32 mtu)
 163 {
 164         struct dst_entry *dst;
 165         const struct inet_sock *inet = inet_sk(sk);
 166         const struct dccp_sock *dp = dccp_sk(sk);
 167
 168         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 169          * send out by Linux are always < 576bytes so they should go through
 170          * unfragmented).
 171          */
 172         if (sk->sk_state == DCCP_LISTEN)
 173                 return;
 174
 175         /* We don't check in the destentry if pmtu discovery is forbidden
 176          * on this route. We just assume that no packet_to_big packets
 177          * are send back when pmtu discovery is not active.
 178          * There is a small race when the user changes this flag in the
 179          * route, but I think that's acceptable.
 180          */
 181         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 182                 return;
 183
 184         dst->ops->update_pmtu(dst, mtu);
 185
 186         /* Something is about to be wrong... Remember soft error
 187          * for the case, if this connection will not able to recover.
 188          */
 189         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 190                 sk->sk_err_soft = EMSGSIZE;
 191
 192         mtu = dst_mtu(dst);
 193
 194         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 195             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 196                 dccp_sync_mss(sk, mtu);
 197
 198                 /*
 199                  * From: draft-ietf-dccp-spec-11.txt
 200                  *
 201                  *      DCCP-Sync packets are the best choice for upward
 202                  *      probing, since DCCP-Sync probes do not risk application
 203                  *      data loss.
 204                  */
 205                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 206         } /* else let the usual retransmit timer handle it */
 207 }
 208
 209 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 210 {
 211         int err;
 212         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 213         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 214                                      sizeof(struct dccp_hdr_ext) +
 215                                      sizeof(struct dccp_hdr_ack_bits);
 216         struct sk_buff *skb;
 217
 218         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 219                 return;
 220
 221         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 222         if (skb == NULL)
 223                 return;
 224
 225         /* Reserve space for headers. */
 226         skb_reserve(skb, MAX_DCCP_HEADER);
 227
 228         skb->dst = dst_clone(rxskb->dst);
 229
 230         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 231         dh = dccp_hdr(skb);
 232         memset(dh, 0, dccp_hdr_ack_len);
 233
 234         /* Build DCCP header and checksum it. */
 235         dh->dccph_type     = DCCP_PKT_ACK;
 236         dh->dccph_sport    = rxdh->dccph_dport;
 237         dh->dccph_dport    = rxdh->dccph_sport;
 238         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 239         dh->dccph_x        = 1;
 240
 241         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 242         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 243                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 244
 245         bh_lock_sock(dccp_ctl_socket->sk);
 246         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 247                                     rxskb->nh.iph->daddr,
 248                                     rxskb->nh.iph->saddr, NULL);
 249         bh_unlock_sock(dccp_ctl_socket->sk);
 250
 251         if (err == NET_XMIT_CN || err == 0) {
 252                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 253                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 254         }
 255 }
 256
 257 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 258                                    struct request_sock *req)
 259 {
 260         dccp_v4_ctl_send_ack(skb);
 261 }
 262
 263 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 264                                  struct dst_entry *dst)
 265 {
 266         int err = -1;
 267         struct sk_buff *skb;
 268
 269         /* First, grab a route. */
 270
 271         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 272                 goto out;
 273
 274         skb = dccp_make_response(sk, dst, req);
 275         if (skb != NULL) {
 276                 const struct inet_request_sock *ireq = inet_rsk(req);
 277
 278                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 279                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 280                                             ireq->rmt_addr,
 281                                             ireq->opt);
 282                 if (err == NET_XMIT_CN)
 283                         err = 0;
 284         }
 285
 286 out:
 287         dst_release(dst);
 288         return err;
 289 }
 290
 291 /*
 292  * This routine is called by the ICMP module when it gets some sort of error
 293  * condition. If err < 0 then the socket should be closed and the error
 294  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 295  * After adjustment header points to the first 8 bytes of the tcp header. We
 296  * need to find the appropriate port.
 297  *
 298  * The locking strategy used here is very "optimistic". When someone else
 299  * accesses the socket the ICMP is just dropped and for some paths there is no
 300  * check at all. A more general error queue to queue errors for later handling
 301  * is probably better.
 302  */
 303 void dccp_v4_err(struct sk_buff *skb, u32 info)
 304 {
 305         const struct iphdr *iph = (struct iphdr *)skb->data;
 306         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 307                                                         (iph->ihl << 2));
 308         struct dccp_sock *dp;
 309         struct inet_sock *inet;
 310         const int type = skb->h.icmph->type;
 311         const int code = skb->h.icmph->code;
 312         struct sock *sk;
 313         __u64 seq;
 314         int err;
 315
 316         if (skb->len < (iph->ihl << 2) + 8) {
 317                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 318                 return;
 319         }
 320
 321         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 322                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 323         if (sk == NULL) {
 324                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 325                 return;
 326         }
 327
 328         if (sk->sk_state == DCCP_TIME_WAIT) {
 329                 inet_twsk_put((struct inet_timewait_sock *)sk);
 330                 return;
 331         }
 332
 333         bh_lock_sock(sk);
 334         /* If too many ICMPs get dropped on busy
 335          * servers this needs to be solved differently.
 336          */
 337         if (sock_owned_by_user(sk))
 338                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 339
 340         if (sk->sk_state == DCCP_CLOSED)
 341                 goto out;
 342
 343         dp = dccp_sk(sk);
 344         seq = dccp_hdr_seq(skb);
 345         if (sk->sk_state != DCCP_LISTEN &&
 346             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 347                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 348                 goto out;
 349         }
 350
 351         switch (type) {
 352         case ICMP_SOURCE_QUENCH:
 353                 /* Just silently ignore these. */
 354                 goto out;
 355         case ICMP_PARAMETERPROB:
 356                 err = EPROTO;
 357                 break;
 358         case ICMP_DEST_UNREACH:
 359                 if (code > NR_ICMP_UNREACH)
 360                         goto out;
 361
 362                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 363                         if (!sock_owned_by_user(sk))
 364                                 dccp_do_pmtu_discovery(sk, iph, info);
 365                         goto out;
 366                 }
 367
 368                 err = icmp_err_convert[code].errno;
 369                 break;
 370         case ICMP_TIME_EXCEEDED:
 371                 err = EHOSTUNREACH;
 372                 break;
 373         default:
 374                 goto out;
 375         }
 376
 377         switch (sk->sk_state) {
 378                 struct request_sock *req , **prev;
 379         case DCCP_LISTEN:
 380                 if (sock_owned_by_user(sk))
 381                         goto out;
 382                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 383                                           iph->daddr, iph->saddr);
 384                 if (!req)
 385                         goto out;
 386
 387                 /*
 388                  * ICMPs are not backlogged, hence we cannot get an established
 389                  * socket here.
 390                  */
 391                 BUG_TRAP(!req->sk);
 392
 393                 if (seq != dccp_rsk(req)->dreq_iss) {
 394                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 395                         goto out;
 396                 }
 397                 /*
 398                  * Still in RESPOND, just remove it silently.
 399                  * There is no good way to pass the error to the newly
 400                  * created socket, and POSIX does not want network
 401                  * errors returned from accept().
 402                  */
 403                 inet_csk_reqsk_queue_drop(sk, req, prev);
 404                 goto out;
 405
 406         case DCCP_REQUESTING:
 407         case DCCP_RESPOND:
 408                 if (!sock_owned_by_user(sk)) {
 409                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 410                         sk->sk_err = err;
 411
 412                         sk->sk_error_report(sk);
 413
 414                         dccp_done(sk);
 415                 } else
 416                         sk->sk_err_soft = err;
 417                 goto out;
 418         }
 419
 420         /* If we've already connected we will keep trying
 421          * until we time out, or the user gives up.
 422          *
 423          * rfc1122 4.2.3.9 allows to consider as hard errors
 424          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 425          * but it is obsoleted by pmtu discovery).
 426          *
 427          * Note, that in modern internet, where routing is unreliable
 428          * and in each dark corner broken firewalls sit, sending random
 429          * errors ordered by their masters even this two messages finally lose
 430          * their original sense (even Linux sends invalid PORT_UNREACHs)
 431          *
 432          * Now we are in compliance with RFCs.
 433          *                                                      --ANK (980905)
 434          */
 435
 436         inet = inet_sk(sk);
 437         if (!sock_owned_by_user(sk) && inet->recverr) {
 438                 sk->sk_err = err;
 439                 sk->sk_error_report(sk);
 440         } else /* Only an error on timeout */
 441                 sk->sk_err_soft = err;
 442 out:
 443         bh_unlock_sock(sk);
 444         sock_put(sk);
 445 }
 446
 447 /* This routine computes an IPv4 DCCP checksum. */
 448 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 449 {
 450         const struct inet_sock *inet = inet_sk(sk);
 451         struct dccp_hdr *dh = dccp_hdr(skb);
 452
 453         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 454 }
 455
 456 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 457
 458 int dccp_v4_send_reset(struct sock *sk, enum dccp_reset_codes code)
 459 {
 460         struct sk_buff *skb;
 461         /*
 462          * FIXME: what if rebuild_header fails?
 463          * Should we be doing a rebuild_header here?
 464          */
 465         int err = inet_sk_rebuild_header(sk);
 466
 467         if (err != 0)
 468                 return err;
 469
 470         skb = dccp_make_reset(sk, sk->sk_dst_cache, code);
 471         if (skb != NULL) {
 472                 const struct inet_sock *inet = inet_sk(sk);
 473
 474                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 475                 err = ip_build_and_send_pkt(skb, sk,
 476                                             inet->saddr, inet->daddr, NULL);
 477                 if (err == NET_XMIT_CN)
 478                         err = 0;
 479         }
 480
 481         return err;
 482 }
 483
 484 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 485                                         const struct sk_buff *skb)
 486 {
 487         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 488                                            skb->nh.iph->saddr,
 489                                            dccp_hdr(skb)->dccph_dport,
 490                                            dccp_hdr(skb)->dccph_sport);
 491 }
 492
 493 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 494 {
 495         struct inet_request_sock *ireq;
 496         struct dccp_sock dp;
 497         struct request_sock *req;
 498         struct dccp_request_sock *dreq;
 499         const __u32 saddr = skb->nh.iph->saddr;
 500         const __u32 daddr = skb->nh.iph->daddr;
 501         const __u32 service = dccp_hdr_request(skb)->dccph_req_service;
 502         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 503         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 504
 505         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 506         if (((struct rtable *)skb->dst)->rt_flags &
 507             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 508                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 509                 goto drop;
 510         }
 511
 512         if (dccp_bad_service_code(sk, service)) {
 513                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 514                 goto drop;
 515         }
 516         /*
 517          * TW buckets are converted to open requests without
 518          * limitations, they conserve resources and peer is
 519          * evidently real one.
 520          */
 521         if (inet_csk_reqsk_queue_is_full(sk))
 522                 goto drop;
 523
 524         /*
 525          * Accept backlog is full. If we have already queued enough
 526          * of warm entries in syn queue, drop request. It is better than
 527          * clogging syn queue with openreqs with exponentially increasing
 528          * timeout.
 529          */
 530         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 531                 goto drop;
 532
 533         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 534         if (req == NULL)
 535                 goto drop;
 536
 537         /* FIXME: process options */
 538
 539         dccp_openreq_init(req, &dp, skb);
 540
 541         ireq = inet_rsk(req);
 542         ireq->loc_addr = daddr;
 543         ireq->rmt_addr = saddr;
 544         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 545                                   right value */
 546         ireq->opt       = NULL;
 547
 548         /*
 549          * Step 3: Process LISTEN state
 550          *
 551          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 552          *
 553          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 554          * dccp_create_openreq_child.
 555          */
 556         dreq = dccp_rsk(req);
 557         dreq->dreq_isr     = dcb->dccpd_seq;
 558         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 559         dreq->dreq_service = service;
 560
 561         if (dccp_v4_send_response(sk, req, NULL))
 562                 goto drop_and_free;
 563
 564         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 565         return 0;
 566
 567 drop_and_free:
 568         reqsk_free(req);
 569 drop:
 570         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 571         dcb->dccpd_reset_code = reset_code;
 572         return -1;
 573 }
 574
 575 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 576
 577 /*
 578  * The three way handshake has completed - we got a valid ACK or DATAACK -
 579  * now create the new socket.
 580  *
 581  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 582  */
 583 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 584                                        struct request_sock *req,
 585                                        struct dst_entry *dst)
 586 {
 587         struct inet_request_sock *ireq;
 588         struct inet_sock *newinet;
 589         struct dccp_sock *newdp;
 590         struct sock *newsk;
 591
 592         if (sk_acceptq_is_full(sk))
 593                 goto exit_overflow;
 594
 595         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 596                 goto exit;
 597
 598         newsk = dccp_create_openreq_child(sk, req, skb);
 599         if (newsk == NULL)
 600                 goto exit;
 601
 602         sk_setup_caps(newsk, dst);
 603
 604         newdp              = dccp_sk(newsk);
 605         newinet            = inet_sk(newsk);
 606         ireq               = inet_rsk(req);
 607         newinet->daddr     = ireq->rmt_addr;
 608         newinet->rcv_saddr = ireq->loc_addr;
 609         newinet->saddr     = ireq->loc_addr;
 610         newinet->opt       = ireq->opt;
 611         ireq->opt          = NULL;
 612         newinet->mc_index  = inet_iif(skb);
 613         newinet->mc_ttl    = skb->nh.iph->ttl;
 614         newinet->id        = jiffies;
 615
 616         dccp_sync_mss(newsk, dst_mtu(dst));
 617
 618         __inet_hash(&dccp_hashinfo, newsk, 0);
 619         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 620
 621         return newsk;
 622
 623 exit_overflow:
 624         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 625 exit:
 626         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 627         dst_release(dst);
 628         return NULL;
 629 }
 630
 631 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 632
 633 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 634 {
 635         const struct dccp_hdr *dh = dccp_hdr(skb);
 636         const struct iphdr *iph = skb->nh.iph;
 637         struct sock *nsk;
 638         struct request_sock **prev;
 639         /* Find possible connection requests. */
 640         struct request_sock *req = inet_csk_search_req(sk, &prev,
 641                                                        dh->dccph_sport,
 642                                                        iph->saddr, iph->daddr);
 643         if (req != NULL)
 644                 return dccp_check_req(sk, skb, req, prev);
 645
 646         nsk = __inet_lookup_established(&dccp_hashinfo,
 647                                         iph->saddr, dh->dccph_sport,
 648                                         iph->daddr, ntohs(dh->dccph_dport),
 649                                         inet_iif(skb));
 650         if (nsk != NULL) {
 651                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 652                         bh_lock_sock(nsk);
 653                         return nsk;
 654                 }
 655                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 656                 return NULL;
 657         }
 658
 659         return sk;
 660 }
 661
 662 int dccp_v4_checksum(const struct sk_buff *skb, const u32 saddr,
 663                      const u32 daddr)
 664 {
 665         const struct dccp_hdr* dh = dccp_hdr(skb);
 666         int checksum_len;
 667         u32 tmp;
 668
 669         if (dh->dccph_cscov == 0)
 670                 checksum_len = skb->len;
 671         else {
 672                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 673                 checksum_len = checksum_len < skb->len ? checksum_len :
 674                                                          skb->len;
 675         }
 676
 677         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 678         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 679                                  IPPROTO_DCCP, tmp);
 680 }
 681
 682 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 683                                    const u32 saddr, const u32 daddr)
 684 {
 685         struct dccp_hdr *dh = dccp_hdr(skb);
 686         int checksum_len;
 687         u32 tmp;
 688
 689         if (dh->dccph_cscov == 0)
 690                 checksum_len = skb->len;
 691         else {
 692                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 693                 checksum_len = checksum_len < skb->len ? checksum_len :
 694                                                          skb->len;
 695         }
 696         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 697         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 698                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 699 }
 700
 701 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 702                                            struct sk_buff *skb)
 703 {
 704         struct rtable *rt;
 705         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 706                             .nl_u = { .ip4_u =
 707                                       { .daddr = skb->nh.iph->saddr,
 708                                         .saddr = skb->nh.iph->daddr,
 709                                         .tos = RT_CONN_FLAGS(sk) } },
 710                             .proto = sk->sk_protocol,
 711                             .uli_u = { .ports =
 712                                        { .sport = dccp_hdr(skb)->dccph_dport,
 713                                          .dport = dccp_hdr(skb)->dccph_sport }
 714                                      }
 715                           };
 716
 717         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 718                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 719                 return NULL;
 720         }
 721
 722         return &rt->u.dst;
 723 }
 724
 725 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 726 {
 727         int err;
 728         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 729         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 730                                        sizeof(struct dccp_hdr_ext) +
 731                                        sizeof(struct dccp_hdr_reset);
 732         struct sk_buff *skb;
 733         struct dst_entry *dst;
 734         u64 seqno;
 735
 736         /* Never send a reset in response to a reset. */
 737         if (rxdh->dccph_type == DCCP_PKT_RESET)
 738                 return;
 739
 740         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 741                 return;
 742
 743         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 744         if (dst == NULL)
 745                 return;
 746
 747         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 748         if (skb == NULL)
 749                 goto out;
 750
 751         /* Reserve space for headers. */
 752         skb_reserve(skb, MAX_DCCP_HEADER);
 753         skb->dst = dst_clone(dst);
 754
 755         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 756         dh = dccp_hdr(skb);
 757         memset(dh, 0, dccp_hdr_reset_len);
 758
 759         /* Build DCCP header and checksum it. */
 760         dh->dccph_type     = DCCP_PKT_RESET;
 761         dh->dccph_sport    = rxdh->dccph_dport;
 762         dh->dccph_dport    = rxdh->dccph_sport;
 763         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 764         dh->dccph_x        = 1;
 765         dccp_hdr_reset(skb)->dccph_reset_code =
 766                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 767
 768         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 769         seqno = 0;
 770         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 771                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 772
 773         dccp_hdr_set_seq(dh, seqno);
 774         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 775                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 776
 777         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 778                                               rxskb->nh.iph->daddr);
 779
 780         bh_lock_sock(dccp_ctl_socket->sk);
 781         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 782                                     rxskb->nh.iph->daddr,
 783                                     rxskb->nh.iph->saddr, NULL);
 784         bh_unlock_sock(dccp_ctl_socket->sk);
 785
 786         if (err == NET_XMIT_CN || err == 0) {
 787                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 788                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 789         }
 790 out:
 791          dst_release(dst);
 792 }
 793
 794 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 795 {
 796         struct dccp_hdr *dh = dccp_hdr(skb);
 797
 798         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 799                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 800                         goto reset;
 801                 return 0;
 802         }
 803
 804         /*
 805          *  Step 3: Process LISTEN state
 806          *     If S.state == LISTEN,
 807          *        If P.type == Request or P contains a valid Init Cookie
 808          *              option,
 809          *           * Must scan the packet's options to check for an Init
 810          *              Cookie.  Only the Init Cookie is processed here,
 811          *              however; other options are processed in Step 8.  This
 812          *              scan need only be performed if the endpoint uses Init
 813          *              Cookies *
 814          *           * Generate a new socket and switch to that socket *
 815          *           Set S := new socket for this port pair
 816          *           S.state = RESPOND
 817          *           Choose S.ISS (initial seqno) or set from Init Cookie
 818          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 819          *           Continue with S.state == RESPOND
 820          *           * A Response packet will be generated in Step 11 *
 821          *        Otherwise,
 822          *           Generate Reset(No Connection) unless P.type == Reset
 823          *           Drop packet and return
 824          *
 825          * NOTE: the check for the packet types is done in
 826          *       dccp_rcv_state_process
 827          */
 828         if (sk->sk_state == DCCP_LISTEN) {
 829                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 830
 831                 if (nsk == NULL)
 832                         goto discard;
 833
 834                 if (nsk != sk) {
 835                         if (dccp_child_process(sk, nsk, skb))
 836                                 goto reset;
 837                         return 0;
 838                 }
 839         }
 840
 841         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 842                 goto reset;
 843         return 0;
 844
 845 reset:
 846         dccp_v4_ctl_send_reset(skb);
 847 discard:
 848         kfree_skb(skb);
 849         return 0;
 850 }
 851
 852 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 853
 854 int dccp_invalid_packet(struct sk_buff *skb)
 855 {
 856         const struct dccp_hdr *dh;
 857
 858         if (skb->pkt_type != PACKET_HOST)
 859                 return 1;
 860
 861         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 862                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 863                 return 1;
 864         }
 865
 866         dh = dccp_hdr(skb);
 867
 868         /* If the packet type is not understood, drop packet and return */
 869         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 870                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 871                 return 1;
 872         }
 873
 874         /*
 875          * If P.Data Offset is too small for packet type, or too large for
 876          * packet, drop packet and return
 877          */
 878         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 879                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 880                                             "too small 1\n",
 881                                dh->dccph_doff);
 882                 return 1;
 883         }
 884
 885         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 886                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 887                                             "too small 2\n",
 888                                dh->dccph_doff);
 889                 return 1;
 890         }
 891
 892         dh = dccp_hdr(skb);
 893
 894         /*
 895          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 896          * has short sequence numbers), drop packet and return
 897          */
 898         if (dh->dccph_x == 0 &&
 899             dh->dccph_type != DCCP_PKT_DATA &&
 900             dh->dccph_type != DCCP_PKT_ACK &&
 901             dh->dccph_type != DCCP_PKT_DATAACK) {
 902                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 903                                             "nor DataAck and P.X == 0\n",
 904                                dccp_packet_name(dh->dccph_type));
 905                 return 1;
 906         }
 907
 908         return 0;
 909 }
 910
 911 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 912
 913 /* this is called when real data arrives */
 914 int dccp_v4_rcv(struct sk_buff *skb)
 915 {
 916         const struct dccp_hdr *dh;
 917         struct sock *sk;
 918
 919         /* Step 1: Check header basics: */
 920
 921         if (dccp_invalid_packet(skb))
 922                 goto discard_it;
 923
 924         /* If the header checksum is incorrect, drop packet and return */
 925         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
 926                                     skb->nh.iph->daddr) < 0) {
 927                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
 928                                __FUNCTION__);
 929                 goto discard_it;
 930         }
 931
 932         dh = dccp_hdr(skb);
 933
 934         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 935         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 936
 937         dccp_pr_debug("%8.8s "
 938                       "src=%u.%u.%u.%u@%-5d "
 939                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 940                       dccp_packet_name(dh->dccph_type),
 941                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 942                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 943                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 944
 945         if (dccp_packet_without_ack(skb)) {
 946                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 947                 dccp_pr_debug_cat("\n");
 948         } else {
 949                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 950                 dccp_pr_debug_cat(", ack=%llu\n",
 951                                   (unsigned long long)
 952                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 953         }
 954
 955         /* Step 2:
 956          *      Look up flow ID in table and get corresponding socket */
 957         sk = __inet_lookup(&dccp_hashinfo,
 958                            skb->nh.iph->saddr, dh->dccph_sport,
 959                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
 960                            inet_iif(skb));
 961
 962         /*
 963          * Step 2:
 964          *      If no socket ...
 965          *              Generate Reset(No Connection) unless P.type == Reset
 966          *              Drop packet and return
 967          */
 968         if (sk == NULL) {
 969                 dccp_pr_debug("failed to look up flow ID in table and "
 970                               "get corresponding socket\n");
 971                 goto no_dccp_socket;
 972         }
 973
 974         /*
 975          * Step 2:
 976          *      ... or S.state == TIMEWAIT,
 977          *              Generate Reset(No Connection) unless P.type == Reset
 978          *              Drop packet and return
 979          */
 980
 981         if (sk->sk_state == DCCP_TIME_WAIT) {
 982                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
 983                               "do_time_wait\n");
 984                 goto do_time_wait;
 985         }
 986
 987         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 988                 goto discard_and_relse;
 989         nf_reset(skb);
 990
 991         return sk_receive_skb(sk, skb);
 992
 993 no_dccp_socket:
 994         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 995                 goto discard_it;
 996         /*
 997          * Step 2:
 998          *              Generate Reset(No Connection) unless P.type == Reset
 999          *              Drop packet and return
1000          */
1001         if (dh->dccph_type != DCCP_PKT_RESET) {
1002                 DCCP_SKB_CB(skb)->dccpd_reset_code =
1003                                         DCCP_RESET_CODE_NO_CONNECTION;
1004                 dccp_v4_ctl_send_reset(skb);
1005         }
1006
1007 discard_it:
1008         /* Discard frame. */
1009         kfree_skb(skb);
1010         return 0;
1011
1012 discard_and_relse:
1013         sock_put(sk);
1014         goto discard_it;
1015
1016 do_time_wait:
1017         inet_twsk_put((struct inet_timewait_sock *)sk);
1018         goto no_dccp_socket;
1019 }
1020
1021 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
1022         .queue_xmit     = ip_queue_xmit,
1023         .send_check     = dccp_v4_send_check,
1024         .rebuild_header = inet_sk_rebuild_header,
1025         .conn_request   = dccp_v4_conn_request,
1026         .syn_recv_sock  = dccp_v4_request_recv_sock,
1027         .net_header_len = sizeof(struct iphdr),
1028         .setsockopt     = ip_setsockopt,
1029         .getsockopt     = ip_getsockopt,
1030         .addr2sockaddr  = inet_csk_addr2sockaddr,
1031         .sockaddr_len   = sizeof(struct sockaddr_in),
1032 };
1033
1034 int dccp_v4_init_sock(struct sock *sk)
1035 {
1036         struct dccp_sock *dp = dccp_sk(sk);
1037         struct inet_connection_sock *icsk = inet_csk(sk);
1038         static int dccp_ctl_socket_init = 1;
1039
1040         dccp_options_init(&dp->dccps_options);
1041         do_gettimeofday(&dp->dccps_epoch);
1042
1043         if (dp->dccps_options.dccpo_send_ack_vector) {
1044                 dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(DCCP_MAX_ACKVEC_LEN,
1045                                                            GFP_KERNEL);
1046                 if (dp->dccps_hc_rx_ackvec == NULL)
1047                         return -ENOMEM;
1048         }
1049
1050         /*
1051          * FIXME: We're hardcoding the CCID, and doing this at this point makes
1052          * the listening (master) sock get CCID control blocks, which is not
1053          * necessary, but for now, to not mess with the test userspace apps,
1054          * lets leave it here, later the real solution is to do this in a
1055          * setsockopt(CCIDs-I-want/accept). -acme
1056          */
1057         if (likely(!dccp_ctl_socket_init)) {
1058                 dp->dccps_hc_rx_ccid = ccid_init(dp->dccps_options.dccpo_rx_ccid,
1059                                                  sk);
1060                 dp->dccps_hc_tx_ccid = ccid_init(dp->dccps_options.dccpo_tx_ccid,
1061                                                  sk);
1062                 if (dp->dccps_hc_rx_ccid == NULL ||
1063                     dp->dccps_hc_tx_ccid == NULL) {
1064                         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1065                         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1066                         if (dp->dccps_options.dccpo_send_ack_vector) {
1067                                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1068                                 dp->dccps_hc_rx_ackvec = NULL;
1069                         }
1070                         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1071                         return -ENOMEM;
1072                 }
1073         } else
1074                 dccp_ctl_socket_init = 0;
1075
1076         dccp_init_xmit_timers(sk);
1077         icsk->icsk_rto = DCCP_TIMEOUT_INIT;
1078         sk->sk_state = DCCP_CLOSED;
1079         sk->sk_write_space = dccp_write_space;
1080         icsk->icsk_af_ops = &dccp_ipv4_af_ops;
1081         icsk->icsk_sync_mss = dccp_sync_mss;
1082         dp->dccps_mss_cache = 536;
1083         dp->dccps_role = DCCP_ROLE_UNDEFINED;
1084         dp->dccps_service = DCCP_SERVICE_INVALID_VALUE;
1085
1086         return 0;
1087 }
1088
1089 EXPORT_SYMBOL_GPL(dccp_v4_init_sock);
1090
1091 int dccp_v4_destroy_sock(struct sock *sk)
1092 {
1093         struct dccp_sock *dp = dccp_sk(sk);
1094
1095         /*
1096          * DCCP doesn't use sk_write_queue, just sk_send_head
1097          * for retransmissions
1098          */
1099         if (sk->sk_send_head != NULL) {
1100                 kfree_skb(sk->sk_send_head);
1101                 sk->sk_send_head = NULL;
1102         }
1103
1104         /* Clean up a referenced DCCP bind bucket. */
1105         if (inet_csk(sk)->icsk_bind_hash != NULL)
1106                 inet_put_port(&dccp_hashinfo, sk);
1107
1108         kfree(dp->dccps_service_list);
1109         dp->dccps_service_list = NULL;
1110
1111         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
1112         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
1113         if (dp->dccps_options.dccpo_send_ack_vector) {
1114                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1115                 dp->dccps_hc_rx_ackvec = NULL;
1116         }
1117         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1118         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1119         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1120
1121         return 0;
1122 }
1123
1124 EXPORT_SYMBOL_GPL(dccp_v4_destroy_sock);
1125
1126 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1127 {
1128         kfree(inet_rsk(req)->opt);
1129 }
1130
1131 static struct request_sock_ops dccp_request_sock_ops = {
1132         .family         = PF_INET,
1133         .obj_size       = sizeof(struct dccp_request_sock),
1134         .rtx_syn_ack    = dccp_v4_send_response,
1135         .send_ack       = dccp_v4_reqsk_send_ack,
1136         .destructor     = dccp_v4_reqsk_destructor,
1137         .send_reset     = dccp_v4_ctl_send_reset,
1138 };
1139
1140 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1141         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1142 };
1143
1144 struct proto dccp_prot = {
1145         .name                   = "DCCP",
1146         .owner                  = THIS_MODULE,
1147         .close                  = dccp_close,
1148         .connect                = dccp_v4_connect,
1149         .disconnect             = dccp_disconnect,
1150         .ioctl                  = dccp_ioctl,
1151         .init                   = dccp_v4_init_sock,
1152         .setsockopt             = dccp_setsockopt,
1153         .getsockopt             = dccp_getsockopt,
1154         .sendmsg                = dccp_sendmsg,
1155         .recvmsg                = dccp_recvmsg,
1156         .backlog_rcv            = dccp_v4_do_rcv,
1157         .hash                   = dccp_v4_hash,
1158         .unhash                 = dccp_unhash,
1159         .accept                 = inet_csk_accept,
1160         .get_port               = dccp_v4_get_port,
1161         .shutdown               = dccp_shutdown,
1162         .destroy                = dccp_v4_destroy_sock,
1163         .orphan_count           = &dccp_orphan_count,
1164         .max_header             = MAX_DCCP_HEADER,
1165         .obj_size               = sizeof(struct dccp_sock),
1166         .rsk_prot               = &dccp_request_sock_ops,
1167         .twsk_prot              = &dccp_timewait_sock_ops,
1168 };
1169
1170 EXPORT_SYMBOL_GPL(dccp_prot);