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 static void dccp_v4_hash(struct sock *sk)
  48 {
  49         inet_hash(&dccp_hashinfo, sk);
  50 }
  51
  52 void dccp_unhash(struct sock *sk)
  53 {
  54         inet_unhash(&dccp_hashinfo, sk);
  55 }
  56
  57 EXPORT_SYMBOL_GPL(dccp_unhash);
  58
  59 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  60 {
  61         struct inet_sock *inet = inet_sk(sk);
  62         struct dccp_sock *dp = dccp_sk(sk);
  63         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  64         struct rtable *rt;
  65         u32 daddr, nexthop;
  66         int tmp;
  67         int err;
  68
  69         dp->dccps_role = DCCP_ROLE_CLIENT;
  70
  71         if (dccp_service_not_initialized(sk))
  72                 return -EPROTO;
  73
  74         if (addr_len < sizeof(struct sockaddr_in))
  75                 return -EINVAL;
  76
  77         if (usin->sin_family != AF_INET)
  78                 return -EAFNOSUPPORT;
  79
  80         nexthop = daddr = usin->sin_addr.s_addr;
  81         if (inet->opt != NULL && inet->opt->srr) {
  82                 if (daddr == 0)
  83                         return -EINVAL;
  84                 nexthop = inet->opt->faddr;
  85         }
  86
  87         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
  88                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  89                                IPPROTO_DCCP,
  90                                inet->sport, usin->sin_port, sk);
  91         if (tmp < 0)
  92                 return tmp;
  93
  94         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  95                 ip_rt_put(rt);
  96                 return -ENETUNREACH;
  97         }
  98
  99         if (inet->opt == NULL || !inet->opt->srr)
 100                 daddr = rt->rt_dst;
 101
 102         if (inet->saddr == 0)
 103                 inet->saddr = rt->rt_src;
 104         inet->rcv_saddr = inet->saddr;
 105
 106         inet->dport = usin->sin_port;
 107         inet->daddr = daddr;
 108
 109         inet_csk(sk)->icsk_ext_hdr_len = 0;
 110         if (inet->opt != NULL)
 111                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
 112         /*
 113          * Socket identity is still unknown (sport may be zero).
 114          * However we set state to DCCP_REQUESTING and not releasing socket
 115          * lock select source port, enter ourselves into the hash tables and
 116          * complete initialization after this.
 117          */
 118         dccp_set_state(sk, DCCP_REQUESTING);
 119         err = inet_hash_connect(&dccp_death_row, sk);
 120         if (err != 0)
 121                 goto failure;
 122
 123         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
 124                                 sk);
 125         if (err != 0)
 126                 goto failure;
 127
 128         /* OK, now commit destination to socket.  */
 129         sk_setup_caps(sk, &rt->u.dst);
 130
 131         dp->dccps_gar =
 132                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 133                                                             inet->daddr,
 134                                                             inet->sport,
 135                                                             usin->sin_port);
 136         dccp_update_gss(sk, dp->dccps_iss);
 137
 138         inet->id = dp->dccps_iss ^ jiffies;
 139
 140         err = dccp_connect(sk);
 141         rt = NULL;
 142         if (err != 0)
 143                 goto failure;
 144 out:
 145         return err;
 146 failure:
 147         /*
 148          * This unhashes the socket and releases the local port, if necessary.
 149          */
 150         dccp_set_state(sk, DCCP_CLOSED);
 151         ip_rt_put(rt);
 152         sk->sk_route_caps = 0;
 153         inet->dport = 0;
 154         goto out;
 155 }
 156
 157 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 158
 159 /*
 160  * This routine does path mtu discovery as defined in RFC1191.
 161  */
 162 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 163                                           const struct iphdr *iph,
 164                                           u32 mtu)
 165 {
 166         struct dst_entry *dst;
 167         const struct inet_sock *inet = inet_sk(sk);
 168         const struct dccp_sock *dp = dccp_sk(sk);
 169
 170         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 171          * send out by Linux are always < 576bytes so they should go through
 172          * unfragmented).
 173          */
 174         if (sk->sk_state == DCCP_LISTEN)
 175                 return;
 176
 177         /* We don't check in the destentry if pmtu discovery is forbidden
 178          * on this route. We just assume that no packet_to_big packets
 179          * are send back when pmtu discovery is not active.
 180          * There is a small race when the user changes this flag in the
 181          * route, but I think that's acceptable.
 182          */
 183         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 184                 return;
 185
 186         dst->ops->update_pmtu(dst, mtu);
 187
 188         /* Something is about to be wrong... Remember soft error
 189          * for the case, if this connection will not able to recover.
 190          */
 191         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 192                 sk->sk_err_soft = EMSGSIZE;
 193
 194         mtu = dst_mtu(dst);
 195
 196         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 197             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 198                 dccp_sync_mss(sk, mtu);
 199
 200                 /*
 201                  * From: draft-ietf-dccp-spec-11.txt
 202                  *
 203                  *      DCCP-Sync packets are the best choice for upward
 204                  *      probing, since DCCP-Sync probes do not risk application
 205                  *      data loss.
 206                  */
 207                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 208         } /* else let the usual retransmit timer handle it */
 209 }
 210
 211 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 212 {
 213         int err;
 214         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 215         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 216                                      sizeof(struct dccp_hdr_ext) +
 217                                      sizeof(struct dccp_hdr_ack_bits);
 218         struct sk_buff *skb;
 219
 220         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 221                 return;
 222
 223         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 224         if (skb == NULL)
 225                 return;
 226
 227         /* Reserve space for headers. */
 228         skb_reserve(skb, MAX_DCCP_HEADER);
 229
 230         skb->dst = dst_clone(rxskb->dst);
 231
 232         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 233         dh = dccp_hdr(skb);
 234         memset(dh, 0, dccp_hdr_ack_len);
 235
 236         /* Build DCCP header and checksum it. */
 237         dh->dccph_type     = DCCP_PKT_ACK;
 238         dh->dccph_sport    = rxdh->dccph_dport;
 239         dh->dccph_dport    = rxdh->dccph_sport;
 240         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 241         dh->dccph_x        = 1;
 242
 243         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 244         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 245                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 246
 247         bh_lock_sock(dccp_ctl_socket->sk);
 248         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 249                                     rxskb->nh.iph->daddr,
 250                                     rxskb->nh.iph->saddr, NULL);
 251         bh_unlock_sock(dccp_ctl_socket->sk);
 252
 253         if (err == NET_XMIT_CN || err == 0) {
 254                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 255                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 256         }
 257 }
 258
 259 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 260                                    struct request_sock *req)
 261 {
 262         dccp_v4_ctl_send_ack(skb);
 263 }
 264
 265 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 266                                  struct dst_entry *dst)
 267 {
 268         int err = -1;
 269         struct sk_buff *skb;
 270
 271         /* First, grab a route. */
 272
 273         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 274                 goto out;
 275
 276         skb = dccp_make_response(sk, dst, req);
 277         if (skb != NULL) {
 278                 const struct inet_request_sock *ireq = inet_rsk(req);
 279
 280                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 281                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 282                                             ireq->rmt_addr,
 283                                             ireq->opt);
 284                 if (err == NET_XMIT_CN)
 285                         err = 0;
 286         }
 287
 288 out:
 289         dst_release(dst);
 290         return err;
 291 }
 292
 293 /*
 294  * This routine is called by the ICMP module when it gets some sort of error
 295  * condition. If err < 0 then the socket should be closed and the error
 296  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 297  * After adjustment header points to the first 8 bytes of the tcp header. We
 298  * need to find the appropriate port.
 299  *
 300  * The locking strategy used here is very "optimistic". When someone else
 301  * accesses the socket the ICMP is just dropped and for some paths there is no
 302  * check at all. A more general error queue to queue errors for later handling
 303  * is probably better.
 304  */
 305 void dccp_v4_err(struct sk_buff *skb, u32 info)
 306 {
 307         const struct iphdr *iph = (struct iphdr *)skb->data;
 308         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 309                                                         (iph->ihl << 2));
 310         struct dccp_sock *dp;
 311         struct inet_sock *inet;
 312         const int type = skb->h.icmph->type;
 313         const int code = skb->h.icmph->code;
 314         struct sock *sk;
 315         __u64 seq;
 316         int err;
 317
 318         if (skb->len < (iph->ihl << 2) + 8) {
 319                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 320                 return;
 321         }
 322
 323         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 324                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 325         if (sk == NULL) {
 326                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 327                 return;
 328         }
 329
 330         if (sk->sk_state == DCCP_TIME_WAIT) {
 331                 inet_twsk_put((struct inet_timewait_sock *)sk);
 332                 return;
 333         }
 334
 335         bh_lock_sock(sk);
 336         /* If too many ICMPs get dropped on busy
 337          * servers this needs to be solved differently.
 338          */
 339         if (sock_owned_by_user(sk))
 340                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 341
 342         if (sk->sk_state == DCCP_CLOSED)
 343                 goto out;
 344
 345         dp = dccp_sk(sk);
 346         seq = dccp_hdr_seq(skb);
 347         if (sk->sk_state != DCCP_LISTEN &&
 348             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 349                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 350                 goto out;
 351         }
 352
 353         switch (type) {
 354         case ICMP_SOURCE_QUENCH:
 355                 /* Just silently ignore these. */
 356                 goto out;
 357         case ICMP_PARAMETERPROB:
 358                 err = EPROTO;
 359                 break;
 360         case ICMP_DEST_UNREACH:
 361                 if (code > NR_ICMP_UNREACH)
 362                         goto out;
 363
 364                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 365                         if (!sock_owned_by_user(sk))
 366                                 dccp_do_pmtu_discovery(sk, iph, info);
 367                         goto out;
 368                 }
 369
 370                 err = icmp_err_convert[code].errno;
 371                 break;
 372         case ICMP_TIME_EXCEEDED:
 373                 err = EHOSTUNREACH;
 374                 break;
 375         default:
 376                 goto out;
 377         }
 378
 379         switch (sk->sk_state) {
 380                 struct request_sock *req , **prev;
 381         case DCCP_LISTEN:
 382                 if (sock_owned_by_user(sk))
 383                         goto out;
 384                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 385                                           iph->daddr, iph->saddr);
 386                 if (!req)
 387                         goto out;
 388
 389                 /*
 390                  * ICMPs are not backlogged, hence we cannot get an established
 391                  * socket here.
 392                  */
 393                 BUG_TRAP(!req->sk);
 394
 395                 if (seq != dccp_rsk(req)->dreq_iss) {
 396                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 397                         goto out;
 398                 }
 399                 /*
 400                  * Still in RESPOND, just remove it silently.
 401                  * There is no good way to pass the error to the newly
 402                  * created socket, and POSIX does not want network
 403                  * errors returned from accept().
 404                  */
 405                 inet_csk_reqsk_queue_drop(sk, req, prev);
 406                 goto out;
 407
 408         case DCCP_REQUESTING:
 409         case DCCP_RESPOND:
 410                 if (!sock_owned_by_user(sk)) {
 411                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 412                         sk->sk_err = err;
 413
 414                         sk->sk_error_report(sk);
 415
 416                         dccp_done(sk);
 417                 } else
 418                         sk->sk_err_soft = err;
 419                 goto out;
 420         }
 421
 422         /* If we've already connected we will keep trying
 423          * until we time out, or the user gives up.
 424          *
 425          * rfc1122 4.2.3.9 allows to consider as hard errors
 426          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 427          * but it is obsoleted by pmtu discovery).
 428          *
 429          * Note, that in modern internet, where routing is unreliable
 430          * and in each dark corner broken firewalls sit, sending random
 431          * errors ordered by their masters even this two messages finally lose
 432          * their original sense (even Linux sends invalid PORT_UNREACHs)
 433          *
 434          * Now we are in compliance with RFCs.
 435          *                                                      --ANK (980905)
 436          */
 437
 438         inet = inet_sk(sk);
 439         if (!sock_owned_by_user(sk) && inet->recverr) {
 440                 sk->sk_err = err;
 441                 sk->sk_error_report(sk);
 442         } else /* Only an error on timeout */
 443                 sk->sk_err_soft = err;
 444 out:
 445         bh_unlock_sock(sk);
 446         sock_put(sk);
 447 }
 448
 449 /* This routine computes an IPv4 DCCP checksum. */
 450 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 451 {
 452         const struct inet_sock *inet = inet_sk(sk);
 453         struct dccp_hdr *dh = dccp_hdr(skb);
 454
 455         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 456 }
 457
 458 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 459
 460 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 461                                         const struct sk_buff *skb)
 462 {
 463         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 464                                            skb->nh.iph->saddr,
 465                                            dccp_hdr(skb)->dccph_dport,
 466                                            dccp_hdr(skb)->dccph_sport);
 467 }
 468
 469 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 470 {
 471         struct inet_request_sock *ireq;
 472         struct dccp_sock dp;
 473         struct request_sock *req;
 474         struct dccp_request_sock *dreq;
 475         const __be32 saddr = skb->nh.iph->saddr;
 476         const __be32 daddr = skb->nh.iph->daddr;
 477         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 478         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 479         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 480
 481         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 482         if (((struct rtable *)skb->dst)->rt_flags &
 483             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 484                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 485                 goto drop;
 486         }
 487
 488         if (dccp_bad_service_code(sk, service)) {
 489                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 490                 goto drop;
 491         }
 492         /*
 493          * TW buckets are converted to open requests without
 494          * limitations, they conserve resources and peer is
 495          * evidently real one.
 496          */
 497         if (inet_csk_reqsk_queue_is_full(sk))
 498                 goto drop;
 499
 500         /*
 501          * Accept backlog is full. If we have already queued enough
 502          * of warm entries in syn queue, drop request. It is better than
 503          * clogging syn queue with openreqs with exponentially increasing
 504          * timeout.
 505          */
 506         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 507                 goto drop;
 508
 509         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 510         if (req == NULL)
 511                 goto drop;
 512
 513         if (dccp_parse_options(sk, skb))
 514                 goto drop;
 515
 516         dccp_openreq_init(req, &dp, skb);
 517
 518         ireq = inet_rsk(req);
 519         ireq->loc_addr = daddr;
 520         ireq->rmt_addr = saddr;
 521         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 522                                   right value */
 523         ireq->opt       = NULL;
 524
 525         /*
 526          * Step 3: Process LISTEN state
 527          *
 528          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 529          *
 530          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 531          * dccp_create_openreq_child.
 532          */
 533         dreq = dccp_rsk(req);
 534         dreq->dreq_isr     = dcb->dccpd_seq;
 535         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 536         dreq->dreq_service = service;
 537
 538         if (dccp_v4_send_response(sk, req, NULL))
 539                 goto drop_and_free;
 540
 541         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 542         return 0;
 543
 544 drop_and_free:
 545         reqsk_free(req);
 546 drop:
 547         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 548         dcb->dccpd_reset_code = reset_code;
 549         return -1;
 550 }
 551
 552 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 553
 554 /*
 555  * The three way handshake has completed - we got a valid ACK or DATAACK -
 556  * now create the new socket.
 557  *
 558  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 559  */
 560 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 561                                        struct request_sock *req,
 562                                        struct dst_entry *dst)
 563 {
 564         struct inet_request_sock *ireq;
 565         struct inet_sock *newinet;
 566         struct dccp_sock *newdp;
 567         struct sock *newsk;
 568
 569         if (sk_acceptq_is_full(sk))
 570                 goto exit_overflow;
 571
 572         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 573                 goto exit;
 574
 575         newsk = dccp_create_openreq_child(sk, req, skb);
 576         if (newsk == NULL)
 577                 goto exit;
 578
 579         sk_setup_caps(newsk, dst);
 580
 581         newdp              = dccp_sk(newsk);
 582         newinet            = inet_sk(newsk);
 583         ireq               = inet_rsk(req);
 584         newinet->daddr     = ireq->rmt_addr;
 585         newinet->rcv_saddr = ireq->loc_addr;
 586         newinet->saddr     = ireq->loc_addr;
 587         newinet->opt       = ireq->opt;
 588         ireq->opt          = NULL;
 589         newinet->mc_index  = inet_iif(skb);
 590         newinet->mc_ttl    = skb->nh.iph->ttl;
 591         newinet->id        = jiffies;
 592
 593         dccp_sync_mss(newsk, dst_mtu(dst));
 594
 595         __inet_hash(&dccp_hashinfo, newsk, 0);
 596         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 597
 598         return newsk;
 599
 600 exit_overflow:
 601         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 602 exit:
 603         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 604         dst_release(dst);
 605         return NULL;
 606 }
 607
 608 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 609
 610 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 611 {
 612         const struct dccp_hdr *dh = dccp_hdr(skb);
 613         const struct iphdr *iph = skb->nh.iph;
 614         struct sock *nsk;
 615         struct request_sock **prev;
 616         /* Find possible connection requests. */
 617         struct request_sock *req = inet_csk_search_req(sk, &prev,
 618                                                        dh->dccph_sport,
 619                                                        iph->saddr, iph->daddr);
 620         if (req != NULL)
 621                 return dccp_check_req(sk, skb, req, prev);
 622
 623         nsk = __inet_lookup_established(&dccp_hashinfo,
 624                                         iph->saddr, dh->dccph_sport,
 625                                         iph->daddr, ntohs(dh->dccph_dport),
 626                                         inet_iif(skb));
 627         if (nsk != NULL) {
 628                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 629                         bh_lock_sock(nsk);
 630                         return nsk;
 631                 }
 632                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 633                 return NULL;
 634         }
 635
 636         return sk;
 637 }
 638
 639 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
 640                      const __be32 daddr)
 641 {
 642         const struct dccp_hdr* dh = dccp_hdr(skb);
 643         int checksum_len;
 644         u32 tmp;
 645
 646         if (dh->dccph_cscov == 0)
 647                 checksum_len = skb->len;
 648         else {
 649                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 650                 checksum_len = checksum_len < skb->len ? checksum_len :
 651                                                          skb->len;
 652         }
 653
 654         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 655         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 656                                  IPPROTO_DCCP, tmp);
 657 }
 658
 659 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 660                                    const __be32 saddr, const __be32 daddr)
 661 {
 662         struct dccp_hdr *dh = dccp_hdr(skb);
 663         int checksum_len;
 664         u32 tmp;
 665
 666         if (dh->dccph_cscov == 0)
 667                 checksum_len = skb->len;
 668         else {
 669                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 670                 checksum_len = checksum_len < skb->len ? checksum_len :
 671                                                          skb->len;
 672         }
 673         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 674         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 675                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 676 }
 677
 678 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 679                                            struct sk_buff *skb)
 680 {
 681         struct rtable *rt;
 682         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 683                             .nl_u = { .ip4_u =
 684                                       { .daddr = skb->nh.iph->saddr,
 685                                         .saddr = skb->nh.iph->daddr,
 686                                         .tos = RT_CONN_FLAGS(sk) } },
 687                             .proto = sk->sk_protocol,
 688                             .uli_u = { .ports =
 689                                        { .sport = dccp_hdr(skb)->dccph_dport,
 690                                          .dport = dccp_hdr(skb)->dccph_sport }
 691                                      }
 692                           };
 693
 694         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 695                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 696                 return NULL;
 697         }
 698
 699         return &rt->u.dst;
 700 }
 701
 702 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 703 {
 704         int err;
 705         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 706         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 707                                        sizeof(struct dccp_hdr_ext) +
 708                                        sizeof(struct dccp_hdr_reset);
 709         struct sk_buff *skb;
 710         struct dst_entry *dst;
 711         u64 seqno;
 712
 713         /* Never send a reset in response to a reset. */
 714         if (rxdh->dccph_type == DCCP_PKT_RESET)
 715                 return;
 716
 717         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 718                 return;
 719
 720         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 721         if (dst == NULL)
 722                 return;
 723
 724         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 725         if (skb == NULL)
 726                 goto out;
 727
 728         /* Reserve space for headers. */
 729         skb_reserve(skb, MAX_DCCP_HEADER);
 730         skb->dst = dst_clone(dst);
 731
 732         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 733         dh = dccp_hdr(skb);
 734         memset(dh, 0, dccp_hdr_reset_len);
 735
 736         /* Build DCCP header and checksum it. */
 737         dh->dccph_type     = DCCP_PKT_RESET;
 738         dh->dccph_sport    = rxdh->dccph_dport;
 739         dh->dccph_dport    = rxdh->dccph_sport;
 740         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 741         dh->dccph_x        = 1;
 742         dccp_hdr_reset(skb)->dccph_reset_code =
 743                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 744
 745         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 746         seqno = 0;
 747         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 748                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 749
 750         dccp_hdr_set_seq(dh, seqno);
 751         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 752                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 753
 754         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 755                                               rxskb->nh.iph->daddr);
 756
 757         bh_lock_sock(dccp_ctl_socket->sk);
 758         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 759                                     rxskb->nh.iph->daddr,
 760                                     rxskb->nh.iph->saddr, NULL);
 761         bh_unlock_sock(dccp_ctl_socket->sk);
 762
 763         if (err == NET_XMIT_CN || err == 0) {
 764                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 765                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 766         }
 767 out:
 768          dst_release(dst);
 769 }
 770
 771 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 772 {
 773         struct dccp_hdr *dh = dccp_hdr(skb);
 774
 775         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 776                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 777                         goto reset;
 778                 return 0;
 779         }
 780
 781         /*
 782          *  Step 3: Process LISTEN state
 783          *     If S.state == LISTEN,
 784          *        If P.type == Request or P contains a valid Init Cookie
 785          *              option,
 786          *           * Must scan the packet's options to check for an Init
 787          *              Cookie.  Only the Init Cookie is processed here,
 788          *              however; other options are processed in Step 8.  This
 789          *              scan need only be performed if the endpoint uses Init
 790          *              Cookies *
 791          *           * Generate a new socket and switch to that socket *
 792          *           Set S := new socket for this port pair
 793          *           S.state = RESPOND
 794          *           Choose S.ISS (initial seqno) or set from Init Cookie
 795          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 796          *           Continue with S.state == RESPOND
 797          *           * A Response packet will be generated in Step 11 *
 798          *        Otherwise,
 799          *           Generate Reset(No Connection) unless P.type == Reset
 800          *           Drop packet and return
 801          *
 802          * NOTE: the check for the packet types is done in
 803          *       dccp_rcv_state_process
 804          */
 805         if (sk->sk_state == DCCP_LISTEN) {
 806                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 807
 808                 if (nsk == NULL)
 809                         goto discard;
 810
 811                 if (nsk != sk) {
 812                         if (dccp_child_process(sk, nsk, skb))
 813                                 goto reset;
 814                         return 0;
 815                 }
 816         }
 817
 818         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 819                 goto reset;
 820         return 0;
 821
 822 reset:
 823         dccp_v4_ctl_send_reset(skb);
 824 discard:
 825         kfree_skb(skb);
 826         return 0;
 827 }
 828
 829 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 830
 831 int dccp_invalid_packet(struct sk_buff *skb)
 832 {
 833         const struct dccp_hdr *dh;
 834
 835         if (skb->pkt_type != PACKET_HOST)
 836                 return 1;
 837
 838         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 839                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 840                 return 1;
 841         }
 842
 843         dh = dccp_hdr(skb);
 844
 845         /* If the packet type is not understood, drop packet and return */
 846         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 847                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 848                 return 1;
 849         }
 850
 851         /*
 852          * If P.Data Offset is too small for packet type, or too large for
 853          * packet, drop packet and return
 854          */
 855         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 856                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 857                                             "too small 1\n",
 858                                dh->dccph_doff);
 859                 return 1;
 860         }
 861
 862         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 863                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 864                                             "too small 2\n",
 865                                dh->dccph_doff);
 866                 return 1;
 867         }
 868
 869         dh = dccp_hdr(skb);
 870
 871         /*
 872          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 873          * has short sequence numbers), drop packet and return
 874          */
 875         if (dh->dccph_x == 0 &&
 876             dh->dccph_type != DCCP_PKT_DATA &&
 877             dh->dccph_type != DCCP_PKT_ACK &&
 878             dh->dccph_type != DCCP_PKT_DATAACK) {
 879                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 880                                             "nor DataAck and P.X == 0\n",
 881                                dccp_packet_name(dh->dccph_type));
 882                 return 1;
 883         }
 884
 885         return 0;
 886 }
 887
 888 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 889
 890 /* this is called when real data arrives */
 891 int dccp_v4_rcv(struct sk_buff *skb)
 892 {
 893         const struct dccp_hdr *dh;
 894         struct sock *sk;
 895
 896         /* Step 1: Check header basics: */
 897
 898         if (dccp_invalid_packet(skb))
 899                 goto discard_it;
 900
 901         /* If the header checksum is incorrect, drop packet and return */
 902         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
 903                                     skb->nh.iph->daddr) < 0) {
 904                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
 905                                __FUNCTION__);
 906                 goto discard_it;
 907         }
 908
 909         dh = dccp_hdr(skb);
 910
 911         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 912         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 913
 914         dccp_pr_debug("%8.8s "
 915                       "src=%u.%u.%u.%u@%-5d "
 916                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 917                       dccp_packet_name(dh->dccph_type),
 918                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 919                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 920                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 921
 922         if (dccp_packet_without_ack(skb)) {
 923                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 924                 dccp_pr_debug_cat("\n");
 925         } else {
 926                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 927                 dccp_pr_debug_cat(", ack=%llu\n",
 928                                   (unsigned long long)
 929                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 930         }
 931
 932         /* Step 2:
 933          *      Look up flow ID in table and get corresponding socket */
 934         sk = __inet_lookup(&dccp_hashinfo,
 935                            skb->nh.iph->saddr, dh->dccph_sport,
 936                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
 937                            inet_iif(skb));
 938
 939         /*
 940          * Step 2:
 941          *      If no socket ...
 942          *              Generate Reset(No Connection) unless P.type == Reset
 943          *              Drop packet and return
 944          */
 945         if (sk == NULL) {
 946                 dccp_pr_debug("failed to look up flow ID in table and "
 947                               "get corresponding socket\n");
 948                 goto no_dccp_socket;
 949         }
 950
 951         /*
 952          * Step 2:
 953          *      ... or S.state == TIMEWAIT,
 954          *              Generate Reset(No Connection) unless P.type == Reset
 955          *              Drop packet and return
 956          */
 957
 958         if (sk->sk_state == DCCP_TIME_WAIT) {
 959                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
 960                               "do_time_wait\n");
 961                 goto do_time_wait;
 962         }
 963
 964         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 965                 goto discard_and_relse;
 966         nf_reset(skb);
 967
 968         return sk_receive_skb(sk, skb);
 969
 970 no_dccp_socket:
 971         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 972                 goto discard_it;
 973         /*
 974          * Step 2:
 975          *              Generate Reset(No Connection) unless P.type == Reset
 976          *              Drop packet and return
 977          */
 978         if (dh->dccph_type != DCCP_PKT_RESET) {
 979                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 980                                         DCCP_RESET_CODE_NO_CONNECTION;
 981                 dccp_v4_ctl_send_reset(skb);
 982         }
 983
 984 discard_it:
 985         /* Discard frame. */
 986         kfree_skb(skb);
 987         return 0;
 988
 989 discard_and_relse:
 990         sock_put(sk);
 991         goto discard_it;
 992
 993 do_time_wait:
 994         inet_twsk_put((struct inet_timewait_sock *)sk);
 995         goto no_dccp_socket;
 996 }
 997
 998 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 999         .queue_xmit     = ip_queue_xmit,
1000         .send_check     = dccp_v4_send_check,
1001         .rebuild_header = inet_sk_rebuild_header,
1002         .conn_request   = dccp_v4_conn_request,
1003         .syn_recv_sock  = dccp_v4_request_recv_sock,
1004         .net_header_len = sizeof(struct iphdr),
1005         .setsockopt     = ip_setsockopt,
1006         .getsockopt     = ip_getsockopt,
1007         .addr2sockaddr  = inet_csk_addr2sockaddr,
1008         .sockaddr_len   = sizeof(struct sockaddr_in),
1009 };
1010
1011 int dccp_v4_init_sock(struct sock *sk)
1012 {
1013         struct dccp_sock *dp = dccp_sk(sk);
1014         struct inet_connection_sock *icsk = inet_csk(sk);
1015         static int dccp_ctl_socket_init = 1;
1016
1017         dccp_options_init(&dp->dccps_options);
1018         do_gettimeofday(&dp->dccps_epoch);
1019
1020         /*
1021          * FIXME: We're hardcoding the CCID, and doing this at this point makes
1022          * the listening (master) sock get CCID control blocks, which is not
1023          * necessary, but for now, to not mess with the test userspace apps,
1024          * lets leave it here, later the real solution is to do this in a
1025          * setsockopt(CCIDs-I-want/accept). -acme
1026          */
1027         if (likely(!dccp_ctl_socket_init)) {
1028                 int rc = dccp_feat_init(sk);
1029
1030                 if (rc)
1031                         return rc;
1032
1033                 if (dp->dccps_options.dccpo_send_ack_vector) {
1034                         dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(GFP_KERNEL);
1035                         if (dp->dccps_hc_rx_ackvec == NULL)
1036                                 return -ENOMEM;
1037                 }
1038                 dp->dccps_hc_rx_ccid =
1039                                 ccid_hc_rx_new(dp->dccps_options.dccpo_rx_ccid,
1040                                                sk, GFP_KERNEL);
1041                 dp->dccps_hc_tx_ccid =
1042                                 ccid_hc_tx_new(dp->dccps_options.dccpo_tx_ccid,
1043                                                sk, GFP_KERNEL);
1044                 if (unlikely(dp->dccps_hc_rx_ccid == NULL ||
1045                              dp->dccps_hc_tx_ccid == NULL)) {
1046                         ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
1047                         ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
1048                         if (dp->dccps_options.dccpo_send_ack_vector) {
1049                                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1050                                 dp->dccps_hc_rx_ackvec = NULL;
1051                         }
1052                         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1053                         return -ENOMEM;
1054                 }
1055         } else {
1056                 /* control socket doesn't need feat nego */
1057                 INIT_LIST_HEAD(&dp->dccps_options.dccpo_pending);
1058                 INIT_LIST_HEAD(&dp->dccps_options.dccpo_conf);
1059                 dccp_ctl_socket_init = 0;
1060         }
1061
1062         dccp_init_xmit_timers(sk);
1063         icsk->icsk_rto = DCCP_TIMEOUT_INIT;
1064         sk->sk_state = DCCP_CLOSED;
1065         sk->sk_write_space = dccp_write_space;
1066         icsk->icsk_af_ops = &dccp_ipv4_af_ops;
1067         icsk->icsk_sync_mss = dccp_sync_mss;
1068         dp->dccps_mss_cache = 536;
1069         dp->dccps_role = DCCP_ROLE_UNDEFINED;
1070         dp->dccps_service = DCCP_SERVICE_INVALID_VALUE;
1071         dp->dccps_l_ack_ratio = dp->dccps_r_ack_ratio = 1;
1072
1073         return 0;
1074 }
1075
1076 EXPORT_SYMBOL_GPL(dccp_v4_init_sock);
1077
1078 int dccp_v4_destroy_sock(struct sock *sk)
1079 {
1080         struct dccp_sock *dp = dccp_sk(sk);
1081
1082         /*
1083          * DCCP doesn't use sk_write_queue, just sk_send_head
1084          * for retransmissions
1085          */
1086         if (sk->sk_send_head != NULL) {
1087                 kfree_skb(sk->sk_send_head);
1088                 sk->sk_send_head = NULL;
1089         }
1090
1091         /* Clean up a referenced DCCP bind bucket. */
1092         if (inet_csk(sk)->icsk_bind_hash != NULL)
1093                 inet_put_port(&dccp_hashinfo, sk);
1094
1095         kfree(dp->dccps_service_list);
1096         dp->dccps_service_list = NULL;
1097
1098         if (dp->dccps_options.dccpo_send_ack_vector) {
1099                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1100                 dp->dccps_hc_rx_ackvec = NULL;
1101         }
1102         ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
1103         ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
1104         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1105
1106         /* clean up feature negotiation state */
1107         dccp_feat_clean(sk);
1108
1109         return 0;
1110 }
1111
1112 EXPORT_SYMBOL_GPL(dccp_v4_destroy_sock);
1113
1114 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1115 {
1116         kfree(inet_rsk(req)->opt);
1117 }
1118
1119 static struct request_sock_ops dccp_request_sock_ops = {
1120         .family         = PF_INET,
1121         .obj_size       = sizeof(struct dccp_request_sock),
1122         .rtx_syn_ack    = dccp_v4_send_response,
1123         .send_ack       = dccp_v4_reqsk_send_ack,
1124         .destructor     = dccp_v4_reqsk_destructor,
1125         .send_reset     = dccp_v4_ctl_send_reset,
1126 };
1127
1128 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1129         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1130 };
1131
1132 struct proto dccp_prot = {
1133         .name                   = "DCCP",
1134         .owner                  = THIS_MODULE,
1135         .close                  = dccp_close,
1136         .connect                = dccp_v4_connect,
1137         .disconnect             = dccp_disconnect,
1138         .ioctl                  = dccp_ioctl,
1139         .init                   = dccp_v4_init_sock,
1140         .setsockopt             = dccp_setsockopt,
1141         .getsockopt             = dccp_getsockopt,
1142         .sendmsg                = dccp_sendmsg,
1143         .recvmsg                = dccp_recvmsg,
1144         .backlog_rcv            = dccp_v4_do_rcv,
1145         .hash                   = dccp_v4_hash,
1146         .unhash                 = dccp_unhash,
1147         .accept                 = inet_csk_accept,
1148         .get_port               = dccp_v4_get_port,
1149         .shutdown               = dccp_shutdown,
1150         .destroy                = dccp_v4_destroy_sock,
1151         .orphan_count           = &dccp_orphan_count,
1152         .max_header             = MAX_DCCP_HEADER,
1153         .obj_size               = sizeof(struct dccp_sock),
1154         .rsk_prot               = &dccp_request_sock_ops,
1155         .twsk_prot              = &dccp_timewait_sock_ops,
1156 };
1157
1158 EXPORT_SYMBOL_GPL(dccp_prot);