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