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/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 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  34 {
  35         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
  36                                  inet_csk_bind_conflict);
  37 }
  38
  39 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  40 {
  41         struct inet_sock *inet = inet_sk(sk);
  42         struct dccp_sock *dp = dccp_sk(sk);
  43         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  44         struct rtable *rt;
  45         u32 daddr, nexthop;
  46         int tmp;
  47         int err;
  48
  49         dp->dccps_role = DCCP_ROLE_CLIENT;
  50
  51         if (dccp_service_not_initialized(sk))
  52                 return -EPROTO;
  53
  54         if (addr_len < sizeof(struct sockaddr_in))
  55                 return -EINVAL;
  56
  57         if (usin->sin_family != AF_INET)
  58                 return -EAFNOSUPPORT;
  59
  60         nexthop = daddr = usin->sin_addr.s_addr;
  61         if (inet->opt != NULL && inet->opt->srr) {
  62                 if (daddr == 0)
  63                         return -EINVAL;
  64                 nexthop = inet->opt->faddr;
  65         }
  66
  67         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
  68                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  69                                IPPROTO_DCCP,
  70                                inet->sport, usin->sin_port, sk);
  71         if (tmp < 0)
  72                 return tmp;
  73
  74         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  75                 ip_rt_put(rt);
  76                 return -ENETUNREACH;
  77         }
  78
  79         if (inet->opt == NULL || !inet->opt->srr)
  80                 daddr = rt->rt_dst;
  81
  82         if (inet->saddr == 0)
  83                 inet->saddr = rt->rt_src;
  84         inet->rcv_saddr = inet->saddr;
  85
  86         inet->dport = usin->sin_port;
  87         inet->daddr = daddr;
  88
  89         inet_csk(sk)->icsk_ext_hdr_len = 0;
  90         if (inet->opt != NULL)
  91                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
  92         /*
  93          * Socket identity is still unknown (sport may be zero).
  94          * However we set state to DCCP_REQUESTING and not releasing socket
  95          * lock select source port, enter ourselves into the hash tables and
  96          * complete initialization after this.
  97          */
  98         dccp_set_state(sk, DCCP_REQUESTING);
  99         err = inet_hash_connect(&dccp_death_row, sk);
 100         if (err != 0)
 101                 goto failure;
 102
 103         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
 104                                 sk);
 105         if (err != 0)
 106                 goto failure;
 107
 108         /* OK, now commit destination to socket.  */
 109         sk_setup_caps(sk, &rt->u.dst);
 110
 111         dp->dccps_gar =
 112                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 113                                                             inet->daddr,
 114                                                             inet->sport,
 115                                                             usin->sin_port);
 116         dccp_update_gss(sk, dp->dccps_iss);
 117
 118         inet->id = dp->dccps_iss ^ jiffies;
 119
 120         err = dccp_connect(sk);
 121         rt = NULL;
 122         if (err != 0)
 123                 goto failure;
 124 out:
 125         return err;
 126 failure:
 127         /*
 128          * This unhashes the socket and releases the local port, if necessary.
 129          */
 130         dccp_set_state(sk, DCCP_CLOSED);
 131         ip_rt_put(rt);
 132         sk->sk_route_caps = 0;
 133         inet->dport = 0;
 134         goto out;
 135 }
 136
 137 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 138
 139 /*
 140  * This routine does path mtu discovery as defined in RFC1191.
 141  */
 142 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 143                                           const struct iphdr *iph,
 144                                           u32 mtu)
 145 {
 146         struct dst_entry *dst;
 147         const struct inet_sock *inet = inet_sk(sk);
 148         const struct dccp_sock *dp = dccp_sk(sk);
 149
 150         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 151          * send out by Linux are always < 576bytes so they should go through
 152          * unfragmented).
 153          */
 154         if (sk->sk_state == DCCP_LISTEN)
 155                 return;
 156
 157         /* We don't check in the destentry if pmtu discovery is forbidden
 158          * on this route. We just assume that no packet_to_big packets
 159          * are send back when pmtu discovery is not active.
 160          * There is a small race when the user changes this flag in the
 161          * route, but I think that's acceptable.
 162          */
 163         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 164                 return;
 165
 166         dst->ops->update_pmtu(dst, mtu);
 167
 168         /* Something is about to be wrong... Remember soft error
 169          * for the case, if this connection will not able to recover.
 170          */
 171         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 172                 sk->sk_err_soft = EMSGSIZE;
 173
 174         mtu = dst_mtu(dst);
 175
 176         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 177             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 178                 dccp_sync_mss(sk, mtu);
 179
 180                 /*
 181                  * From: draft-ietf-dccp-spec-11.txt
 182                  *
 183                  *      DCCP-Sync packets are the best choice for upward
 184                  *      probing, since DCCP-Sync probes do not risk application
 185                  *      data loss.
 186                  */
 187                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 188         } /* else let the usual retransmit timer handle it */
 189 }
 190
 191 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 192 {
 193         int err;
 194         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 195         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 196                                      sizeof(struct dccp_hdr_ext) +
 197                                      sizeof(struct dccp_hdr_ack_bits);
 198         struct sk_buff *skb;
 199
 200         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 201                 return;
 202
 203         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 204         if (skb == NULL)
 205                 return;
 206
 207         /* Reserve space for headers. */
 208         skb_reserve(skb, MAX_DCCP_HEADER);
 209
 210         skb->dst = dst_clone(rxskb->dst);
 211
 212         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 213         dh = dccp_hdr(skb);
 214         memset(dh, 0, dccp_hdr_ack_len);
 215
 216         /* Build DCCP header and checksum it. */
 217         dh->dccph_type     = DCCP_PKT_ACK;
 218         dh->dccph_sport    = rxdh->dccph_dport;
 219         dh->dccph_dport    = rxdh->dccph_sport;
 220         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 221         dh->dccph_x        = 1;
 222
 223         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 224         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 225                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 226
 227         bh_lock_sock(dccp_ctl_socket->sk);
 228         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 229                                     rxskb->nh.iph->daddr,
 230                                     rxskb->nh.iph->saddr, NULL);
 231         bh_unlock_sock(dccp_ctl_socket->sk);
 232
 233         if (err == NET_XMIT_CN || err == 0) {
 234                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 235                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 236         }
 237 }
 238
 239 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 240                                    struct request_sock *req)
 241 {
 242         dccp_v4_ctl_send_ack(skb);
 243 }
 244
 245 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 246                                  struct dst_entry *dst)
 247 {
 248         int err = -1;
 249         struct sk_buff *skb;
 250
 251         /* First, grab a route. */
 252
 253         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 254                 goto out;
 255
 256         skb = dccp_make_response(sk, dst, req);
 257         if (skb != NULL) {
 258                 const struct inet_request_sock *ireq = inet_rsk(req);
 259                 struct dccp_hdr *dh = dccp_hdr(skb);
 260
 261                 dh->dccph_checksum = dccp_v4_checksum(skb, ireq->loc_addr,
 262                                                       ireq->rmt_addr);
 263                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 264                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 265                                             ireq->rmt_addr,
 266                                             ireq->opt);
 267                 if (err == NET_XMIT_CN)
 268                         err = 0;
 269         }
 270
 271 out:
 272         dst_release(dst);
 273         return err;
 274 }
 275
 276 /*
 277  * This routine is called by the ICMP module when it gets some sort of error
 278  * condition. If err < 0 then the socket should be closed and the error
 279  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 280  * After adjustment header points to the first 8 bytes of the tcp header. We
 281  * need to find the appropriate port.
 282  *
 283  * The locking strategy used here is very "optimistic". When someone else
 284  * accesses the socket the ICMP is just dropped and for some paths there is no
 285  * check at all. A more general error queue to queue errors for later handling
 286  * is probably better.
 287  */
 288 void dccp_v4_err(struct sk_buff *skb, u32 info)
 289 {
 290         const struct iphdr *iph = (struct iphdr *)skb->data;
 291         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 292                                                         (iph->ihl << 2));
 293         struct dccp_sock *dp;
 294         struct inet_sock *inet;
 295         const int type = skb->h.icmph->type;
 296         const int code = skb->h.icmph->code;
 297         struct sock *sk;
 298         __u64 seq;
 299         int err;
 300
 301         if (skb->len < (iph->ihl << 2) + 8) {
 302                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 303                 return;
 304         }
 305
 306         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 307                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 308         if (sk == NULL) {
 309                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 310                 return;
 311         }
 312
 313         if (sk->sk_state == DCCP_TIME_WAIT) {
 314                 inet_twsk_put((struct inet_timewait_sock *)sk);
 315                 return;
 316         }
 317
 318         bh_lock_sock(sk);
 319         /* If too many ICMPs get dropped on busy
 320          * servers this needs to be solved differently.
 321          */
 322         if (sock_owned_by_user(sk))
 323                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 324
 325         if (sk->sk_state == DCCP_CLOSED)
 326                 goto out;
 327
 328         dp = dccp_sk(sk);
 329         seq = dccp_hdr_seq(skb);
 330         if (sk->sk_state != DCCP_LISTEN &&
 331             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 332                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 333                 goto out;
 334         }
 335
 336         switch (type) {
 337         case ICMP_SOURCE_QUENCH:
 338                 /* Just silently ignore these. */
 339                 goto out;
 340         case ICMP_PARAMETERPROB:
 341                 err = EPROTO;
 342                 break;
 343         case ICMP_DEST_UNREACH:
 344                 if (code > NR_ICMP_UNREACH)
 345                         goto out;
 346
 347                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 348                         if (!sock_owned_by_user(sk))
 349                                 dccp_do_pmtu_discovery(sk, iph, info);
 350                         goto out;
 351                 }
 352
 353                 err = icmp_err_convert[code].errno;
 354                 break;
 355         case ICMP_TIME_EXCEEDED:
 356                 err = EHOSTUNREACH;
 357                 break;
 358         default:
 359                 goto out;
 360         }
 361
 362         switch (sk->sk_state) {
 363                 struct request_sock *req , **prev;
 364         case DCCP_LISTEN:
 365                 if (sock_owned_by_user(sk))
 366                         goto out;
 367                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 368                                           iph->daddr, iph->saddr);
 369                 if (!req)
 370                         goto out;
 371
 372                 /*
 373                  * ICMPs are not backlogged, hence we cannot get an established
 374                  * socket here.
 375                  */
 376                 BUG_TRAP(!req->sk);
 377
 378                 if (seq != dccp_rsk(req)->dreq_iss) {
 379                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 380                         goto out;
 381                 }
 382                 /*
 383                  * Still in RESPOND, just remove it silently.
 384                  * There is no good way to pass the error to the newly
 385                  * created socket, and POSIX does not want network
 386                  * errors returned from accept().
 387                  */
 388                 inet_csk_reqsk_queue_drop(sk, req, prev);
 389                 goto out;
 390
 391         case DCCP_REQUESTING:
 392         case DCCP_RESPOND:
 393                 if (!sock_owned_by_user(sk)) {
 394                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 395                         sk->sk_err = err;
 396
 397                         sk->sk_error_report(sk);
 398
 399                         dccp_done(sk);
 400                 } else
 401                         sk->sk_err_soft = err;
 402                 goto out;
 403         }
 404
 405         /* If we've already connected we will keep trying
 406          * until we time out, or the user gives up.
 407          *
 408          * rfc1122 4.2.3.9 allows to consider as hard errors
 409          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 410          * but it is obsoleted by pmtu discovery).
 411          *
 412          * Note, that in modern internet, where routing is unreliable
 413          * and in each dark corner broken firewalls sit, sending random
 414          * errors ordered by their masters even this two messages finally lose
 415          * their original sense (even Linux sends invalid PORT_UNREACHs)
 416          *
 417          * Now we are in compliance with RFCs.
 418          *                                                      --ANK (980905)
 419          */
 420
 421         inet = inet_sk(sk);
 422         if (!sock_owned_by_user(sk) && inet->recverr) {
 423                 sk->sk_err = err;
 424                 sk->sk_error_report(sk);
 425         } else /* Only an error on timeout */
 426                 sk->sk_err_soft = err;
 427 out:
 428         bh_unlock_sock(sk);
 429         sock_put(sk);
 430 }
 431
 432 /* This routine computes an IPv4 DCCP checksum. */
 433 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 434 {
 435         const struct inet_sock *inet = inet_sk(sk);
 436         struct dccp_hdr *dh = dccp_hdr(skb);
 437
 438         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 439 }
 440
 441 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 442
 443 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 444                                         const struct sk_buff *skb)
 445 {
 446         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 447                                            skb->nh.iph->saddr,
 448                                            dccp_hdr(skb)->dccph_dport,
 449                                            dccp_hdr(skb)->dccph_sport);
 450 }
 451
 452 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 453 {
 454         struct inet_request_sock *ireq;
 455         struct dccp_sock dp;
 456         struct request_sock *req;
 457         struct dccp_request_sock *dreq;
 458         const __be32 saddr = skb->nh.iph->saddr;
 459         const __be32 daddr = skb->nh.iph->daddr;
 460         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 461         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 462         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 463
 464         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 465         if (((struct rtable *)skb->dst)->rt_flags &
 466             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 467                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 468                 goto drop;
 469         }
 470
 471         if (dccp_bad_service_code(sk, service)) {
 472                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 473                 goto drop;
 474         }
 475         /*
 476          * TW buckets are converted to open requests without
 477          * limitations, they conserve resources and peer is
 478          * evidently real one.
 479          */
 480         if (inet_csk_reqsk_queue_is_full(sk))
 481                 goto drop;
 482
 483         /*
 484          * Accept backlog is full. If we have already queued enough
 485          * of warm entries in syn queue, drop request. It is better than
 486          * clogging syn queue with openreqs with exponentially increasing
 487          * timeout.
 488          */
 489         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 490                 goto drop;
 491
 492         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 493         if (req == NULL)
 494                 goto drop;
 495
 496         if (dccp_parse_options(sk, skb))
 497                 goto drop;
 498
 499         dccp_openreq_init(req, &dp, skb);
 500
 501         ireq = inet_rsk(req);
 502         ireq->loc_addr = daddr;
 503         ireq->rmt_addr = saddr;
 504         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 505                                   right value */
 506         ireq->opt       = NULL;
 507
 508         /*
 509          * Step 3: Process LISTEN state
 510          *
 511          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 512          *
 513          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 514          * dccp_create_openreq_child.
 515          */
 516         dreq = dccp_rsk(req);
 517         dreq->dreq_isr     = dcb->dccpd_seq;
 518         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 519         dreq->dreq_service = service;
 520
 521         if (dccp_v4_send_response(sk, req, NULL))
 522                 goto drop_and_free;
 523
 524         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 525         return 0;
 526
 527 drop_and_free:
 528         reqsk_free(req);
 529 drop:
 530         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 531         dcb->dccpd_reset_code = reset_code;
 532         return -1;
 533 }
 534
 535 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 536
 537 /*
 538  * The three way handshake has completed - we got a valid ACK or DATAACK -
 539  * now create the new socket.
 540  *
 541  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 542  */
 543 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 544                                        struct request_sock *req,
 545                                        struct dst_entry *dst)
 546 {
 547         struct inet_request_sock *ireq;
 548         struct inet_sock *newinet;
 549         struct dccp_sock *newdp;
 550         struct sock *newsk;
 551
 552         if (sk_acceptq_is_full(sk))
 553                 goto exit_overflow;
 554
 555         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 556                 goto exit;
 557
 558         newsk = dccp_create_openreq_child(sk, req, skb);
 559         if (newsk == NULL)
 560                 goto exit;
 561
 562         sk_setup_caps(newsk, dst);
 563
 564         newdp              = dccp_sk(newsk);
 565         newinet            = inet_sk(newsk);
 566         ireq               = inet_rsk(req);
 567         newinet->daddr     = ireq->rmt_addr;
 568         newinet->rcv_saddr = ireq->loc_addr;
 569         newinet->saddr     = ireq->loc_addr;
 570         newinet->opt       = ireq->opt;
 571         ireq->opt          = NULL;
 572         newinet->mc_index  = inet_iif(skb);
 573         newinet->mc_ttl    = skb->nh.iph->ttl;
 574         newinet->id        = jiffies;
 575
 576         dccp_sync_mss(newsk, dst_mtu(dst));
 577
 578         __inet_hash(&dccp_hashinfo, newsk, 0);
 579         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 580
 581         return newsk;
 582
 583 exit_overflow:
 584         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 585 exit:
 586         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 587         dst_release(dst);
 588         return NULL;
 589 }
 590
 591 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 592
 593 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 594 {
 595         const struct dccp_hdr *dh = dccp_hdr(skb);
 596         const struct iphdr *iph = skb->nh.iph;
 597         struct sock *nsk;
 598         struct request_sock **prev;
 599         /* Find possible connection requests. */
 600         struct request_sock *req = inet_csk_search_req(sk, &prev,
 601                                                        dh->dccph_sport,
 602                                                        iph->saddr, iph->daddr);
 603         if (req != NULL)
 604                 return dccp_check_req(sk, skb, req, prev);
 605
 606         nsk = __inet_lookup_established(&dccp_hashinfo,
 607                                         iph->saddr, dh->dccph_sport,
 608                                         iph->daddr, ntohs(dh->dccph_dport),
 609                                         inet_iif(skb));
 610         if (nsk != NULL) {
 611                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 612                         bh_lock_sock(nsk);
 613                         return nsk;
 614                 }
 615                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 616                 return NULL;
 617         }
 618
 619         return sk;
 620 }
 621
 622 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
 623                      const __be32 daddr)
 624 {
 625         const struct dccp_hdr* dh = dccp_hdr(skb);
 626         int checksum_len;
 627         u32 tmp;
 628
 629         if (dh->dccph_cscov == 0)
 630                 checksum_len = skb->len;
 631         else {
 632                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 633                 checksum_len = checksum_len < skb->len ? checksum_len :
 634                                                          skb->len;
 635         }
 636
 637         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 638         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 639                                  IPPROTO_DCCP, tmp);
 640 }
 641
 642 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 643                                    const __be32 saddr, const __be32 daddr)
 644 {
 645         struct dccp_hdr *dh = dccp_hdr(skb);
 646         int checksum_len;
 647         u32 tmp;
 648
 649         if (dh->dccph_cscov == 0)
 650                 checksum_len = skb->len;
 651         else {
 652                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 653                 checksum_len = checksum_len < skb->len ? checksum_len :
 654                                                          skb->len;
 655         }
 656         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 657         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 658                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 659 }
 660
 661 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 662                                            struct sk_buff *skb)
 663 {
 664         struct rtable *rt;
 665         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 666                             .nl_u = { .ip4_u =
 667                                       { .daddr = skb->nh.iph->saddr,
 668                                         .saddr = skb->nh.iph->daddr,
 669                                         .tos = RT_CONN_FLAGS(sk) } },
 670                             .proto = sk->sk_protocol,
 671                             .uli_u = { .ports =
 672                                        { .sport = dccp_hdr(skb)->dccph_dport,
 673                                          .dport = dccp_hdr(skb)->dccph_sport }
 674                                      }
 675                           };
 676
 677         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 678                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 679                 return NULL;
 680         }
 681
 682         return &rt->u.dst;
 683 }
 684
 685 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 686 {
 687         int err;
 688         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 689         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 690                                        sizeof(struct dccp_hdr_ext) +
 691                                        sizeof(struct dccp_hdr_reset);
 692         struct sk_buff *skb;
 693         struct dst_entry *dst;
 694         u64 seqno;
 695
 696         /* Never send a reset in response to a reset. */
 697         if (rxdh->dccph_type == DCCP_PKT_RESET)
 698                 return;
 699
 700         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 701                 return;
 702
 703         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 704         if (dst == NULL)
 705                 return;
 706
 707         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 708         if (skb == NULL)
 709                 goto out;
 710
 711         /* Reserve space for headers. */
 712         skb_reserve(skb, MAX_DCCP_HEADER);
 713         skb->dst = dst_clone(dst);
 714
 715         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 716         dh = dccp_hdr(skb);
 717         memset(dh, 0, dccp_hdr_reset_len);
 718
 719         /* Build DCCP header and checksum it. */
 720         dh->dccph_type     = DCCP_PKT_RESET;
 721         dh->dccph_sport    = rxdh->dccph_dport;
 722         dh->dccph_dport    = rxdh->dccph_sport;
 723         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 724         dh->dccph_x        = 1;
 725         dccp_hdr_reset(skb)->dccph_reset_code =
 726                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 727
 728         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 729         seqno = 0;
 730         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 731                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 732
 733         dccp_hdr_set_seq(dh, seqno);
 734         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 735                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 736
 737         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 738                                               rxskb->nh.iph->daddr);
 739
 740         bh_lock_sock(dccp_ctl_socket->sk);
 741         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 742                                     rxskb->nh.iph->daddr,
 743                                     rxskb->nh.iph->saddr, NULL);
 744         bh_unlock_sock(dccp_ctl_socket->sk);
 745
 746         if (err == NET_XMIT_CN || err == 0) {
 747                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 748                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 749         }
 750 out:
 751          dst_release(dst);
 752 }
 753
 754 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 755 {
 756         struct dccp_hdr *dh = dccp_hdr(skb);
 757
 758         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 759                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 760                         goto reset;
 761                 return 0;
 762         }
 763
 764         /*
 765          *  Step 3: Process LISTEN state
 766          *     If S.state == LISTEN,
 767          *        If P.type == Request or P contains a valid Init Cookie
 768          *              option,
 769          *           * Must scan the packet's options to check for an Init
 770          *              Cookie.  Only the Init Cookie is processed here,
 771          *              however; other options are processed in Step 8.  This
 772          *              scan need only be performed if the endpoint uses Init
 773          *              Cookies *
 774          *           * Generate a new socket and switch to that socket *
 775          *           Set S := new socket for this port pair
 776          *           S.state = RESPOND
 777          *           Choose S.ISS (initial seqno) or set from Init Cookie
 778          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 779          *           Continue with S.state == RESPOND
 780          *           * A Response packet will be generated in Step 11 *
 781          *        Otherwise,
 782          *           Generate Reset(No Connection) unless P.type == Reset
 783          *           Drop packet and return
 784          *
 785          * NOTE: the check for the packet types is done in
 786          *       dccp_rcv_state_process
 787          */
 788         if (sk->sk_state == DCCP_LISTEN) {
 789                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 790
 791                 if (nsk == NULL)
 792                         goto discard;
 793
 794                 if (nsk != sk) {
 795                         if (dccp_child_process(sk, nsk, skb))
 796                                 goto reset;
 797                         return 0;
 798                 }
 799         }
 800
 801         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 802                 goto reset;
 803         return 0;
 804
 805 reset:
 806         dccp_v4_ctl_send_reset(skb);
 807 discard:
 808         kfree_skb(skb);
 809         return 0;
 810 }
 811
 812 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 813
 814 int dccp_invalid_packet(struct sk_buff *skb)
 815 {
 816         const struct dccp_hdr *dh;
 817
 818         if (skb->pkt_type != PACKET_HOST)
 819                 return 1;
 820
 821         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 822                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 823                 return 1;
 824         }
 825
 826         dh = dccp_hdr(skb);
 827
 828         /* If the packet type is not understood, drop packet and return */
 829         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 830                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 831                 return 1;
 832         }
 833
 834         /*
 835          * If P.Data Offset is too small for packet type, or too large for
 836          * packet, drop packet and return
 837          */
 838         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 839                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 840                                             "too small 1\n",
 841                                dh->dccph_doff);
 842                 return 1;
 843         }
 844
 845         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 846                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 847                                             "too small 2\n",
 848                                dh->dccph_doff);
 849                 return 1;
 850         }
 851
 852         dh = dccp_hdr(skb);
 853
 854         /*
 855          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 856          * has short sequence numbers), drop packet and return
 857          */
 858         if (dh->dccph_x == 0 &&
 859             dh->dccph_type != DCCP_PKT_DATA &&
 860             dh->dccph_type != DCCP_PKT_ACK &&
 861             dh->dccph_type != DCCP_PKT_DATAACK) {
 862                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 863                                             "nor DataAck and P.X == 0\n",
 864                                dccp_packet_name(dh->dccph_type));
 865                 return 1;
 866         }
 867
 868         return 0;
 869 }
 870
 871 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 872
 873 /* this is called when real data arrives */
 874 int dccp_v4_rcv(struct sk_buff *skb)
 875 {
 876         const struct dccp_hdr *dh;
 877         struct sock *sk;
 878
 879         /* Step 1: Check header basics: */
 880
 881         if (dccp_invalid_packet(skb))
 882                 goto discard_it;
 883
 884         /* If the header checksum is incorrect, drop packet and return */
 885         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
 886                                     skb->nh.iph->daddr) < 0) {
 887                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
 888                                __FUNCTION__);
 889                 goto discard_it;
 890         }
 891
 892         dh = dccp_hdr(skb);
 893
 894         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 895         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 896
 897         dccp_pr_debug("%8.8s "
 898                       "src=%u.%u.%u.%u@%-5d "
 899                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 900                       dccp_packet_name(dh->dccph_type),
 901                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 902                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 903                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 904
 905         if (dccp_packet_without_ack(skb)) {
 906                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 907                 dccp_pr_debug_cat("\n");
 908         } else {
 909                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 910                 dccp_pr_debug_cat(", ack=%llu\n",
 911                                   (unsigned long long)
 912                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 913         }
 914
 915         /* Step 2:
 916          *      Look up flow ID in table and get corresponding socket */
 917         sk = __inet_lookup(&dccp_hashinfo,
 918                            skb->nh.iph->saddr, dh->dccph_sport,
 919                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
 920                            inet_iif(skb));
 921
 922         /*
 923          * Step 2:
 924          *      If no socket ...
 925          *              Generate Reset(No Connection) unless P.type == Reset
 926          *              Drop packet and return
 927          */
 928         if (sk == NULL) {
 929                 dccp_pr_debug("failed to look up flow ID in table and "
 930                               "get corresponding socket\n");
 931                 goto no_dccp_socket;
 932         }
 933
 934         /*
 935          * Step 2:
 936          *      ... or S.state == TIMEWAIT,
 937          *              Generate Reset(No Connection) unless P.type == Reset
 938          *              Drop packet and return
 939          */
 940
 941         if (sk->sk_state == DCCP_TIME_WAIT) {
 942                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
 943                               "do_time_wait\n");
 944                 goto do_time_wait;
 945         }
 946
 947         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 948                 goto discard_and_relse;
 949         nf_reset(skb);
 950
 951         return sk_receive_skb(sk, skb);
 952
 953 no_dccp_socket:
 954         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 955                 goto discard_it;
 956         /*
 957          * Step 2:
 958          *              Generate Reset(No Connection) unless P.type == Reset
 959          *              Drop packet and return
 960          */
 961         if (dh->dccph_type != DCCP_PKT_RESET) {
 962                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 963                                         DCCP_RESET_CODE_NO_CONNECTION;
 964                 dccp_v4_ctl_send_reset(skb);
 965         }
 966
 967 discard_it:
 968         /* Discard frame. */
 969         kfree_skb(skb);
 970         return 0;
 971
 972 discard_and_relse:
 973         sock_put(sk);
 974         goto discard_it;
 975
 976 do_time_wait:
 977         inet_twsk_put((struct inet_timewait_sock *)sk);
 978         goto no_dccp_socket;
 979 }
 980
 981 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 982         .queue_xmit     = ip_queue_xmit,
 983         .send_check     = dccp_v4_send_check,
 984         .rebuild_header = inet_sk_rebuild_header,
 985         .conn_request   = dccp_v4_conn_request,
 986         .syn_recv_sock  = dccp_v4_request_recv_sock,
 987         .net_header_len = sizeof(struct iphdr),
 988         .setsockopt     = ip_setsockopt,
 989         .getsockopt     = ip_getsockopt,
 990         .addr2sockaddr  = inet_csk_addr2sockaddr,
 991         .sockaddr_len   = sizeof(struct sockaddr_in),
 992 };
 993
 994 static int dccp_v4_init_sock(struct sock *sk)
 995 {
 996         const int err = dccp_init_sock(sk);
 997
 998         if (err == 0)
 999                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1000         return err;
1001 }
1002
1003 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1004 {
1005         kfree(inet_rsk(req)->opt);
1006 }
1007
1008 static struct request_sock_ops dccp_request_sock_ops = {
1009         .family         = PF_INET,
1010         .obj_size       = sizeof(struct dccp_request_sock),
1011         .rtx_syn_ack    = dccp_v4_send_response,
1012         .send_ack       = dccp_v4_reqsk_send_ack,
1013         .destructor     = dccp_v4_reqsk_destructor,
1014         .send_reset     = dccp_v4_ctl_send_reset,
1015 };
1016
1017 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1018         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1019 };
1020
1021 struct proto dccp_prot = {
1022         .name                   = "DCCP",
1023         .owner                  = THIS_MODULE,
1024         .close                  = dccp_close,
1025         .connect                = dccp_v4_connect,
1026         .disconnect             = dccp_disconnect,
1027         .ioctl                  = dccp_ioctl,
1028         .init                   = dccp_v4_init_sock,
1029         .setsockopt             = dccp_setsockopt,
1030         .getsockopt             = dccp_getsockopt,
1031         .sendmsg                = dccp_sendmsg,
1032         .recvmsg                = dccp_recvmsg,
1033         .backlog_rcv            = dccp_v4_do_rcv,
1034         .hash                   = dccp_hash,
1035         .unhash                 = dccp_unhash,
1036         .accept                 = inet_csk_accept,
1037         .get_port               = dccp_v4_get_port,
1038         .shutdown               = dccp_shutdown,
1039         .destroy                = dccp_destroy_sock,
1040         .orphan_count           = &dccp_orphan_count,
1041         .max_header             = MAX_DCCP_HEADER,
1042         .obj_size               = sizeof(struct dccp_sock),
1043         .rsk_prot               = &dccp_request_sock_ops,
1044         .twsk_prot              = &dccp_timewait_sock_ops,
1045 };
1046
1047 EXPORT_SYMBOL_GPL(dccp_prot);