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_common.h>
  22 #include <net/inet_hashtables.h>
  23 #include <net/inet_sock.h>
  24 #include <net/protocol.h>
  25 #include <net/sock.h>
  26 #include <net/timewait_sock.h>
  27 #include <net/tcp_states.h>
  28 #include <net/xfrm.h>
  29
  30 #include "ackvec.h"
  31 #include "ccid.h"
  32 #include "dccp.h"
  33 #include "feat.h"
  34
  35 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  36 {
  37         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
  38                                  inet_csk_bind_conflict);
  39 }
  40
  41 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  42 {
  43         struct inet_sock *inet = inet_sk(sk);
  44         struct dccp_sock *dp = dccp_sk(sk);
  45         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  46         struct rtable *rt;
  47         u32 daddr, nexthop;
  48         int tmp;
  49         int err;
  50
  51         dp->dccps_role = DCCP_ROLE_CLIENT;
  52
  53         if (dccp_service_not_initialized(sk))
  54                 return -EPROTO;
  55
  56         if (addr_len < sizeof(struct sockaddr_in))
  57                 return -EINVAL;
  58
  59         if (usin->sin_family != AF_INET)
  60                 return -EAFNOSUPPORT;
  61
  62         nexthop = daddr = usin->sin_addr.s_addr;
  63         if (inet->opt != NULL && inet->opt->srr) {
  64                 if (daddr == 0)
  65                         return -EINVAL;
  66                 nexthop = inet->opt->faddr;
  67         }
  68
  69         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
  70                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  71                                IPPROTO_DCCP,
  72                                inet->sport, usin->sin_port, sk);
  73         if (tmp < 0)
  74                 return tmp;
  75
  76         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  77                 ip_rt_put(rt);
  78                 return -ENETUNREACH;
  79         }
  80
  81         if (inet->opt == NULL || !inet->opt->srr)
  82                 daddr = rt->rt_dst;
  83
  84         if (inet->saddr == 0)
  85                 inet->saddr = rt->rt_src;
  86         inet->rcv_saddr = inet->saddr;
  87
  88         inet->dport = usin->sin_port;
  89         inet->daddr = daddr;
  90
  91         inet_csk(sk)->icsk_ext_hdr_len = 0;
  92         if (inet->opt != NULL)
  93                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
  94         /*
  95          * Socket identity is still unknown (sport may be zero).
  96          * However we set state to DCCP_REQUESTING and not releasing socket
  97          * lock select source port, enter ourselves into the hash tables and
  98          * complete initialization after this.
  99          */
 100         dccp_set_state(sk, DCCP_REQUESTING);
 101         err = inet_hash_connect(&dccp_death_row, sk);
 102         if (err != 0)
 103                 goto failure;
 104
 105         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
 106                                 sk);
 107         if (err != 0)
 108                 goto failure;
 109
 110         /* OK, now commit destination to socket.  */
 111         sk_setup_caps(sk, &rt->u.dst);
 112
 113         dp->dccps_gar =
 114                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 115                                                             inet->daddr,
 116                                                             inet->sport,
 117                                                             usin->sin_port);
 118         dccp_update_gss(sk, dp->dccps_iss);
 119
 120         inet->id = dp->dccps_iss ^ jiffies;
 121
 122         err = dccp_connect(sk);
 123         rt = NULL;
 124         if (err != 0)
 125                 goto failure;
 126 out:
 127         return err;
 128 failure:
 129         /*
 130          * This unhashes the socket and releases the local port, if necessary.
 131          */
 132         dccp_set_state(sk, DCCP_CLOSED);
 133         ip_rt_put(rt);
 134         sk->sk_route_caps = 0;
 135         inet->dport = 0;
 136         goto out;
 137 }
 138
 139 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 140
 141 /*
 142  * This routine does path mtu discovery as defined in RFC1191.
 143  */
 144 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 145                                           const struct iphdr *iph,
 146                                           u32 mtu)
 147 {
 148         struct dst_entry *dst;
 149         const struct inet_sock *inet = inet_sk(sk);
 150         const struct dccp_sock *dp = dccp_sk(sk);
 151
 152         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 153          * send out by Linux are always < 576bytes so they should go through
 154          * unfragmented).
 155          */
 156         if (sk->sk_state == DCCP_LISTEN)
 157                 return;
 158
 159         /* We don't check in the destentry if pmtu discovery is forbidden
 160          * on this route. We just assume that no packet_to_big packets
 161          * are send back when pmtu discovery is not active.
 162          * There is a small race when the user changes this flag in the
 163          * route, but I think that's acceptable.
 164          */
 165         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 166                 return;
 167
 168         dst->ops->update_pmtu(dst, mtu);
 169
 170         /* Something is about to be wrong... Remember soft error
 171          * for the case, if this connection will not able to recover.
 172          */
 173         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 174                 sk->sk_err_soft = EMSGSIZE;
 175
 176         mtu = dst_mtu(dst);
 177
 178         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 179             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 180                 dccp_sync_mss(sk, mtu);
 181
 182                 /*
 183                  * From: draft-ietf-dccp-spec-11.txt
 184                  *
 185                  *      DCCP-Sync packets are the best choice for upward
 186                  *      probing, since DCCP-Sync probes do not risk application
 187                  *      data loss.
 188                  */
 189                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 190         } /* else let the usual retransmit timer handle it */
 191 }
 192
 193 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 194 {
 195         int err;
 196         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 197         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 198                                      sizeof(struct dccp_hdr_ext) +
 199                                      sizeof(struct dccp_hdr_ack_bits);
 200         struct sk_buff *skb;
 201
 202         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 203                 return;
 204
 205         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 206         if (skb == NULL)
 207                 return;
 208
 209         /* Reserve space for headers. */
 210         skb_reserve(skb, MAX_DCCP_HEADER);
 211
 212         skb->dst = dst_clone(rxskb->dst);
 213
 214         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 215         dh = dccp_hdr(skb);
 216         memset(dh, 0, dccp_hdr_ack_len);
 217
 218         /* Build DCCP header and checksum it. */
 219         dh->dccph_type     = DCCP_PKT_ACK;
 220         dh->dccph_sport    = rxdh->dccph_dport;
 221         dh->dccph_dport    = rxdh->dccph_sport;
 222         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 223         dh->dccph_x        = 1;
 224
 225         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 226         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 227                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 228
 229         bh_lock_sock(dccp_ctl_socket->sk);
 230         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 231                                     rxskb->nh.iph->daddr,
 232                                     rxskb->nh.iph->saddr, NULL);
 233         bh_unlock_sock(dccp_ctl_socket->sk);
 234
 235         if (err == NET_XMIT_CN || err == 0) {
 236                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 237                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 238         }
 239 }
 240
 241 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 242                                    struct request_sock *req)
 243 {
 244         dccp_v4_ctl_send_ack(skb);
 245 }
 246
 247 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 248                                  struct dst_entry *dst)
 249 {
 250         int err = -1;
 251         struct sk_buff *skb;
 252
 253         /* First, grab a route. */
 254
 255         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 256                 goto out;
 257
 258         skb = dccp_make_response(sk, dst, req);
 259         if (skb != NULL) {
 260                 const struct inet_request_sock *ireq = inet_rsk(req);
 261                 struct dccp_hdr *dh = dccp_hdr(skb);
 262
 263                 dh->dccph_checksum = dccp_v4_checksum(skb, ireq->loc_addr,
 264                                                       ireq->rmt_addr);
 265                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 266                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 267                                             ireq->rmt_addr,
 268                                             ireq->opt);
 269                 if (err == NET_XMIT_CN)
 270                         err = 0;
 271         }
 272
 273 out:
 274         dst_release(dst);
 275         return err;
 276 }
 277
 278 /*
 279  * This routine is called by the ICMP module when it gets some sort of error
 280  * condition. If err < 0 then the socket should be closed and the error
 281  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 282  * After adjustment header points to the first 8 bytes of the tcp header. We
 283  * need to find the appropriate port.
 284  *
 285  * The locking strategy used here is very "optimistic". When someone else
 286  * accesses the socket the ICMP is just dropped and for some paths there is no
 287  * check at all. A more general error queue to queue errors for later handling
 288  * is probably better.
 289  */
 290 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 291 {
 292         const struct iphdr *iph = (struct iphdr *)skb->data;
 293         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 294                                                         (iph->ihl << 2));
 295         struct dccp_sock *dp;
 296         struct inet_sock *inet;
 297         const int type = skb->h.icmph->type;
 298         const int code = skb->h.icmph->code;
 299         struct sock *sk;
 300         __u64 seq;
 301         int err;
 302
 303         if (skb->len < (iph->ihl << 2) + 8) {
 304                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 305                 return;
 306         }
 307
 308         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 309                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 310         if (sk == NULL) {
 311                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 312                 return;
 313         }
 314
 315         if (sk->sk_state == DCCP_TIME_WAIT) {
 316                 inet_twsk_put((struct inet_timewait_sock *)sk);
 317                 return;
 318         }
 319
 320         bh_lock_sock(sk);
 321         /* If too many ICMPs get dropped on busy
 322          * servers this needs to be solved differently.
 323          */
 324         if (sock_owned_by_user(sk))
 325                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 326
 327         if (sk->sk_state == DCCP_CLOSED)
 328                 goto out;
 329
 330         dp = dccp_sk(sk);
 331         seq = dccp_hdr_seq(skb);
 332         if (sk->sk_state != DCCP_LISTEN &&
 333             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 334                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 335                 goto out;
 336         }
 337
 338         switch (type) {
 339         case ICMP_SOURCE_QUENCH:
 340                 /* Just silently ignore these. */
 341                 goto out;
 342         case ICMP_PARAMETERPROB:
 343                 err = EPROTO;
 344                 break;
 345         case ICMP_DEST_UNREACH:
 346                 if (code > NR_ICMP_UNREACH)
 347                         goto out;
 348
 349                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 350                         if (!sock_owned_by_user(sk))
 351                                 dccp_do_pmtu_discovery(sk, iph, info);
 352                         goto out;
 353                 }
 354
 355                 err = icmp_err_convert[code].errno;
 356                 break;
 357         case ICMP_TIME_EXCEEDED:
 358                 err = EHOSTUNREACH;
 359                 break;
 360         default:
 361                 goto out;
 362         }
 363
 364         switch (sk->sk_state) {
 365                 struct request_sock *req , **prev;
 366         case DCCP_LISTEN:
 367                 if (sock_owned_by_user(sk))
 368                         goto out;
 369                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 370                                           iph->daddr, iph->saddr);
 371                 if (!req)
 372                         goto out;
 373
 374                 /*
 375                  * ICMPs are not backlogged, hence we cannot get an established
 376                  * socket here.
 377                  */
 378                 BUG_TRAP(!req->sk);
 379
 380                 if (seq != dccp_rsk(req)->dreq_iss) {
 381                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 382                         goto out;
 383                 }
 384                 /*
 385                  * Still in RESPOND, just remove it silently.
 386                  * There is no good way to pass the error to the newly
 387                  * created socket, and POSIX does not want network
 388                  * errors returned from accept().
 389                  */
 390                 inet_csk_reqsk_queue_drop(sk, req, prev);
 391                 goto out;
 392
 393         case DCCP_REQUESTING:
 394         case DCCP_RESPOND:
 395                 if (!sock_owned_by_user(sk)) {
 396                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 397                         sk->sk_err = err;
 398
 399                         sk->sk_error_report(sk);
 400
 401                         dccp_done(sk);
 402                 } else
 403                         sk->sk_err_soft = err;
 404                 goto out;
 405         }
 406
 407         /* If we've already connected we will keep trying
 408          * until we time out, or the user gives up.
 409          *
 410          * rfc1122 4.2.3.9 allows to consider as hard errors
 411          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 412          * but it is obsoleted by pmtu discovery).
 413          *
 414          * Note, that in modern internet, where routing is unreliable
 415          * and in each dark corner broken firewalls sit, sending random
 416          * errors ordered by their masters even this two messages finally lose
 417          * their original sense (even Linux sends invalid PORT_UNREACHs)
 418          *
 419          * Now we are in compliance with RFCs.
 420          *                                                      --ANK (980905)
 421          */
 422
 423         inet = inet_sk(sk);
 424         if (!sock_owned_by_user(sk) && inet->recverr) {
 425                 sk->sk_err = err;
 426                 sk->sk_error_report(sk);
 427         } else /* Only an error on timeout */
 428                 sk->sk_err_soft = err;
 429 out:
 430         bh_unlock_sock(sk);
 431         sock_put(sk);
 432 }
 433
 434 /* This routine computes an IPv4 DCCP checksum. */
 435 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 436 {
 437         const struct inet_sock *inet = inet_sk(sk);
 438         struct dccp_hdr *dh = dccp_hdr(skb);
 439
 440         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 441 }
 442
 443 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 444
 445 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 446                                         const struct sk_buff *skb)
 447 {
 448         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 449                                            skb->nh.iph->saddr,
 450                                            dccp_hdr(skb)->dccph_dport,
 451                                            dccp_hdr(skb)->dccph_sport);
 452 }
 453
 454 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 455 {
 456         struct inet_request_sock *ireq;
 457         struct dccp_sock dp;
 458         struct request_sock *req;
 459         struct dccp_request_sock *dreq;
 460         const __be32 saddr = skb->nh.iph->saddr;
 461         const __be32 daddr = skb->nh.iph->daddr;
 462         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 463         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 464         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 465
 466         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 467         if (((struct rtable *)skb->dst)->rt_flags &
 468             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 469                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 470                 goto drop;
 471         }
 472
 473         if (dccp_bad_service_code(sk, service)) {
 474                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 475                 goto drop;
 476         }
 477         /*
 478          * TW buckets are converted to open requests without
 479          * limitations, they conserve resources and peer is
 480          * evidently real one.
 481          */
 482         if (inet_csk_reqsk_queue_is_full(sk))
 483                 goto drop;
 484
 485         /*
 486          * Accept backlog is full. If we have already queued enough
 487          * of warm entries in syn queue, drop request. It is better than
 488          * clogging syn queue with openreqs with exponentially increasing
 489          * timeout.
 490          */
 491         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 492                 goto drop;
 493
 494         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 495         if (req == NULL)
 496                 goto drop;
 497
 498         if (dccp_parse_options(sk, skb))
 499                 goto drop;
 500
 501         dccp_openreq_init(req, &dp, skb);
 502
 503         ireq = inet_rsk(req);
 504         ireq->loc_addr = daddr;
 505         ireq->rmt_addr = saddr;
 506         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 507                                   right value */
 508         ireq->opt       = NULL;
 509
 510         /*
 511          * Step 3: Process LISTEN state
 512          *
 513          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 514          *
 515          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 516          * dccp_create_openreq_child.
 517          */
 518         dreq = dccp_rsk(req);
 519         dreq->dreq_isr     = dcb->dccpd_seq;
 520         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 521         dreq->dreq_service = service;
 522
 523         if (dccp_v4_send_response(sk, req, NULL))
 524                 goto drop_and_free;
 525
 526         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 527         return 0;
 528
 529 drop_and_free:
 530         reqsk_free(req);
 531 drop:
 532         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 533         dcb->dccpd_reset_code = reset_code;
 534         return -1;
 535 }
 536
 537 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 538
 539 /*
 540  * The three way handshake has completed - we got a valid ACK or DATAACK -
 541  * now create the new socket.
 542  *
 543  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 544  */
 545 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 546                                        struct request_sock *req,
 547                                        struct dst_entry *dst)
 548 {
 549         struct inet_request_sock *ireq;
 550         struct inet_sock *newinet;
 551         struct dccp_sock *newdp;
 552         struct sock *newsk;
 553
 554         if (sk_acceptq_is_full(sk))
 555                 goto exit_overflow;
 556
 557         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 558                 goto exit;
 559
 560         newsk = dccp_create_openreq_child(sk, req, skb);
 561         if (newsk == NULL)
 562                 goto exit;
 563
 564         sk_setup_caps(newsk, dst);
 565
 566         newdp              = dccp_sk(newsk);
 567         newinet            = inet_sk(newsk);
 568         ireq               = inet_rsk(req);
 569         newinet->daddr     = ireq->rmt_addr;
 570         newinet->rcv_saddr = ireq->loc_addr;
 571         newinet->saddr     = ireq->loc_addr;
 572         newinet->opt       = ireq->opt;
 573         ireq->opt          = NULL;
 574         newinet->mc_index  = inet_iif(skb);
 575         newinet->mc_ttl    = skb->nh.iph->ttl;
 576         newinet->id        = jiffies;
 577
 578         dccp_sync_mss(newsk, dst_mtu(dst));
 579
 580         __inet_hash(&dccp_hashinfo, newsk, 0);
 581         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 582
 583         return newsk;
 584
 585 exit_overflow:
 586         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 587 exit:
 588         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 589         dst_release(dst);
 590         return NULL;
 591 }
 592
 593 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 594
 595 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 596 {
 597         const struct dccp_hdr *dh = dccp_hdr(skb);
 598         const struct iphdr *iph = skb->nh.iph;
 599         struct sock *nsk;
 600         struct request_sock **prev;
 601         /* Find possible connection requests. */
 602         struct request_sock *req = inet_csk_search_req(sk, &prev,
 603                                                        dh->dccph_sport,
 604                                                        iph->saddr, iph->daddr);
 605         if (req != NULL)
 606                 return dccp_check_req(sk, skb, req, prev);
 607
 608         nsk = __inet_lookup_established(&dccp_hashinfo,
 609                                         iph->saddr, dh->dccph_sport,
 610                                         iph->daddr, ntohs(dh->dccph_dport),
 611                                         inet_iif(skb));
 612         if (nsk != NULL) {
 613                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 614                         bh_lock_sock(nsk);
 615                         return nsk;
 616                 }
 617                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 618                 return NULL;
 619         }
 620
 621         return sk;
 622 }
 623
 624 int dccp_v4_checksum(const struct sk_buff *skb, const __be32 saddr,
 625                      const __be32 daddr)
 626 {
 627         const struct dccp_hdr* dh = dccp_hdr(skb);
 628         int checksum_len;
 629         u32 tmp;
 630
 631         if (dh->dccph_cscov == 0)
 632                 checksum_len = skb->len;
 633         else {
 634                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 635                 checksum_len = checksum_len < skb->len ? checksum_len :
 636                                                          skb->len;
 637         }
 638
 639         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 640         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 641                                  IPPROTO_DCCP, tmp);
 642 }
 643
 644 EXPORT_SYMBOL_GPL(dccp_v4_checksum);
 645
 646 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 647                                    const __be32 saddr, const __be32 daddr)
 648 {
 649         struct dccp_hdr *dh = dccp_hdr(skb);
 650         int checksum_len;
 651         u32 tmp;
 652
 653         if (dh->dccph_cscov == 0)
 654                 checksum_len = skb->len;
 655         else {
 656                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 657                 checksum_len = checksum_len < skb->len ? checksum_len :
 658                                                          skb->len;
 659         }
 660         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 661         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 662                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 663 }
 664
 665 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 666                                            struct sk_buff *skb)
 667 {
 668         struct rtable *rt;
 669         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 670                             .nl_u = { .ip4_u =
 671                                       { .daddr = skb->nh.iph->saddr,
 672                                         .saddr = skb->nh.iph->daddr,
 673                                         .tos = RT_CONN_FLAGS(sk) } },
 674                             .proto = sk->sk_protocol,
 675                             .uli_u = { .ports =
 676                                        { .sport = dccp_hdr(skb)->dccph_dport,
 677                                          .dport = dccp_hdr(skb)->dccph_sport }
 678                                      }
 679                           };
 680
 681         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 682                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 683                 return NULL;
 684         }
 685
 686         return &rt->u.dst;
 687 }
 688
 689 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 690 {
 691         int err;
 692         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 693         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 694                                        sizeof(struct dccp_hdr_ext) +
 695                                        sizeof(struct dccp_hdr_reset);
 696         struct sk_buff *skb;
 697         struct dst_entry *dst;
 698         u64 seqno;
 699
 700         /* Never send a reset in response to a reset. */
 701         if (rxdh->dccph_type == DCCP_PKT_RESET)
 702                 return;
 703
 704         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 705                 return;
 706
 707         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 708         if (dst == NULL)
 709                 return;
 710
 711         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 712         if (skb == NULL)
 713                 goto out;
 714
 715         /* Reserve space for headers. */
 716         skb_reserve(skb, MAX_DCCP_HEADER);
 717         skb->dst = dst_clone(dst);
 718
 719         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 720         dh = dccp_hdr(skb);
 721         memset(dh, 0, dccp_hdr_reset_len);
 722
 723         /* Build DCCP header and checksum it. */
 724         dh->dccph_type     = DCCP_PKT_RESET;
 725         dh->dccph_sport    = rxdh->dccph_dport;
 726         dh->dccph_dport    = rxdh->dccph_sport;
 727         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 728         dh->dccph_x        = 1;
 729         dccp_hdr_reset(skb)->dccph_reset_code =
 730                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 731
 732         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 733         seqno = 0;
 734         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 735                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 736
 737         dccp_hdr_set_seq(dh, seqno);
 738         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 739                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 740
 741         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 742                                               rxskb->nh.iph->daddr);
 743
 744         bh_lock_sock(dccp_ctl_socket->sk);
 745         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 746                                     rxskb->nh.iph->daddr,
 747                                     rxskb->nh.iph->saddr, NULL);
 748         bh_unlock_sock(dccp_ctl_socket->sk);
 749
 750         if (err == NET_XMIT_CN || err == 0) {
 751                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 752                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 753         }
 754 out:
 755          dst_release(dst);
 756 }
 757
 758 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 759 {
 760         struct dccp_hdr *dh = dccp_hdr(skb);
 761
 762         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 763                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 764                         goto reset;
 765                 return 0;
 766         }
 767
 768         /*
 769          *  Step 3: Process LISTEN state
 770          *     If S.state == LISTEN,
 771          *        If P.type == Request or P contains a valid Init Cookie
 772          *              option,
 773          *           * Must scan the packet's options to check for an Init
 774          *              Cookie.  Only the Init Cookie is processed here,
 775          *              however; other options are processed in Step 8.  This
 776          *              scan need only be performed if the endpoint uses Init
 777          *              Cookies *
 778          *           * Generate a new socket and switch to that socket *
 779          *           Set S := new socket for this port pair
 780          *           S.state = RESPOND
 781          *           Choose S.ISS (initial seqno) or set from Init Cookie
 782          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 783          *           Continue with S.state == RESPOND
 784          *           * A Response packet will be generated in Step 11 *
 785          *        Otherwise,
 786          *           Generate Reset(No Connection) unless P.type == Reset
 787          *           Drop packet and return
 788          *
 789          * NOTE: the check for the packet types is done in
 790          *       dccp_rcv_state_process
 791          */
 792         if (sk->sk_state == DCCP_LISTEN) {
 793                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 794
 795                 if (nsk == NULL)
 796                         goto discard;
 797
 798                 if (nsk != sk) {
 799                         if (dccp_child_process(sk, nsk, skb))
 800                                 goto reset;
 801                         return 0;
 802                 }
 803         }
 804
 805         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 806                 goto reset;
 807         return 0;
 808
 809 reset:
 810         dccp_v4_ctl_send_reset(skb);
 811 discard:
 812         kfree_skb(skb);
 813         return 0;
 814 }
 815
 816 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 817
 818 int dccp_invalid_packet(struct sk_buff *skb)
 819 {
 820         const struct dccp_hdr *dh;
 821
 822         if (skb->pkt_type != PACKET_HOST)
 823                 return 1;
 824
 825         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 826                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 827                 return 1;
 828         }
 829
 830         dh = dccp_hdr(skb);
 831
 832         /* If the packet type is not understood, drop packet and return */
 833         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 834                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 835                 return 1;
 836         }
 837
 838         /*
 839          * If P.Data Offset is too small for packet type, or too large for
 840          * packet, drop packet and return
 841          */
 842         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 843                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 844                                             "too small 1\n",
 845                                dh->dccph_doff);
 846                 return 1;
 847         }
 848
 849         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 850                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 851                                             "too small 2\n",
 852                                dh->dccph_doff);
 853                 return 1;
 854         }
 855
 856         dh = dccp_hdr(skb);
 857
 858         /*
 859          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 860          * has short sequence numbers), drop packet and return
 861          */
 862         if (dh->dccph_x == 0 &&
 863             dh->dccph_type != DCCP_PKT_DATA &&
 864             dh->dccph_type != DCCP_PKT_ACK &&
 865             dh->dccph_type != DCCP_PKT_DATAACK) {
 866                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 867                                             "nor DataAck and P.X == 0\n",
 868                                dccp_packet_name(dh->dccph_type));
 869                 return 1;
 870         }
 871
 872         return 0;
 873 }
 874
 875 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 876
 877 /* this is called when real data arrives */
 878 static int dccp_v4_rcv(struct sk_buff *skb)
 879 {
 880         const struct dccp_hdr *dh;
 881         struct sock *sk;
 882
 883         /* Step 1: Check header basics: */
 884
 885         if (dccp_invalid_packet(skb))
 886                 goto discard_it;
 887
 888         /* If the header checksum is incorrect, drop packet and return */
 889         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
 890                                     skb->nh.iph->daddr) < 0) {
 891                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
 892                                __FUNCTION__);
 893                 goto discard_it;
 894         }
 895
 896         dh = dccp_hdr(skb);
 897
 898         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 899         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 900
 901         dccp_pr_debug("%8.8s "
 902                       "src=%u.%u.%u.%u@%-5d "
 903                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 904                       dccp_packet_name(dh->dccph_type),
 905                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 906                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 907                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 908
 909         if (dccp_packet_without_ack(skb)) {
 910                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 911                 dccp_pr_debug_cat("\n");
 912         } else {
 913                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 914                 dccp_pr_debug_cat(", ack=%llu\n",
 915                                   (unsigned long long)
 916                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 917         }
 918
 919         /* Step 2:
 920          *      Look up flow ID in table and get corresponding socket */
 921         sk = __inet_lookup(&dccp_hashinfo,
 922                            skb->nh.iph->saddr, dh->dccph_sport,
 923                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
 924                            inet_iif(skb));
 925
 926         /*
 927          * Step 2:
 928          *      If no socket ...
 929          *              Generate Reset(No Connection) unless P.type == Reset
 930          *              Drop packet and return
 931          */
 932         if (sk == NULL) {
 933                 dccp_pr_debug("failed to look up flow ID in table and "
 934                               "get corresponding socket\n");
 935                 goto no_dccp_socket;
 936         }
 937
 938         /*
 939          * Step 2:
 940          *      ... or S.state == TIMEWAIT,
 941          *              Generate Reset(No Connection) unless P.type == Reset
 942          *              Drop packet and return
 943          */
 944
 945         if (sk->sk_state == DCCP_TIME_WAIT) {
 946                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
 947                               "do_time_wait\n");
 948                 goto do_time_wait;
 949         }
 950
 951         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 952                 goto discard_and_relse;
 953         nf_reset(skb);
 954
 955         return sk_receive_skb(sk, skb);
 956
 957 no_dccp_socket:
 958         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 959                 goto discard_it;
 960         /*
 961          * Step 2:
 962          *              Generate Reset(No Connection) unless P.type == Reset
 963          *              Drop packet and return
 964          */
 965         if (dh->dccph_type != DCCP_PKT_RESET) {
 966                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 967                                         DCCP_RESET_CODE_NO_CONNECTION;
 968                 dccp_v4_ctl_send_reset(skb);
 969         }
 970
 971 discard_it:
 972         /* Discard frame. */
 973         kfree_skb(skb);
 974         return 0;
 975
 976 discard_and_relse:
 977         sock_put(sk);
 978         goto discard_it;
 979
 980 do_time_wait:
 981         inet_twsk_put((struct inet_timewait_sock *)sk);
 982         goto no_dccp_socket;
 983 }
 984
 985 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 986         .queue_xmit     = ip_queue_xmit,
 987         .send_check     = dccp_v4_send_check,
 988         .rebuild_header = inet_sk_rebuild_header,
 989         .conn_request   = dccp_v4_conn_request,
 990         .syn_recv_sock  = dccp_v4_request_recv_sock,
 991         .net_header_len = sizeof(struct iphdr),
 992         .setsockopt     = ip_setsockopt,
 993         .getsockopt     = ip_getsockopt,
 994         .addr2sockaddr  = inet_csk_addr2sockaddr,
 995         .sockaddr_len   = sizeof(struct sockaddr_in),
 996 };
 997
 998 static int dccp_v4_init_sock(struct sock *sk)
 999 {
1000         const int err = dccp_init_sock(sk);
1001
1002         if (err == 0)
1003                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1004         return err;
1005 }
1006
1007 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1008 {
1009         kfree(inet_rsk(req)->opt);
1010 }
1011
1012 static struct request_sock_ops dccp_request_sock_ops = {
1013         .family         = PF_INET,
1014         .obj_size       = sizeof(struct dccp_request_sock),
1015         .rtx_syn_ack    = dccp_v4_send_response,
1016         .send_ack       = dccp_v4_reqsk_send_ack,
1017         .destructor     = dccp_v4_reqsk_destructor,
1018         .send_reset     = dccp_v4_ctl_send_reset,
1019 };
1020
1021 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1022         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1023 };
1024
1025 struct proto dccp_v4_prot = {
1026         .name                   = "DCCP",
1027         .owner                  = THIS_MODULE,
1028         .close                  = dccp_close,
1029         .connect                = dccp_v4_connect,
1030         .disconnect             = dccp_disconnect,
1031         .ioctl                  = dccp_ioctl,
1032         .init                   = dccp_v4_init_sock,
1033         .setsockopt             = dccp_setsockopt,
1034         .getsockopt             = dccp_getsockopt,
1035         .sendmsg                = dccp_sendmsg,
1036         .recvmsg                = dccp_recvmsg,
1037         .backlog_rcv            = dccp_v4_do_rcv,
1038         .hash                   = dccp_hash,
1039         .unhash                 = dccp_unhash,
1040         .accept                 = inet_csk_accept,
1041         .get_port               = dccp_v4_get_port,
1042         .shutdown               = dccp_shutdown,
1043         .destroy                = dccp_destroy_sock,
1044         .orphan_count           = &dccp_orphan_count,
1045         .max_header             = MAX_DCCP_HEADER,
1046         .obj_size               = sizeof(struct dccp_sock),
1047         .rsk_prot               = &dccp_request_sock_ops,
1048         .twsk_prot              = &dccp_timewait_sock_ops,
1049 };
1050
1051 static struct net_protocol dccp_v4_protocol = {
1052         .handler        = dccp_v4_rcv,
1053         .err_handler    = dccp_v4_err,
1054         .no_policy      = 1,
1055 };
1056
1057 static const struct proto_ops inet_dccp_ops = {
1058         .family         = PF_INET,
1059         .owner          = THIS_MODULE,
1060         .release        = inet_release,
1061         .bind           = inet_bind,
1062         .connect        = inet_stream_connect,
1063         .socketpair     = sock_no_socketpair,
1064         .accept         = inet_accept,
1065         .getname        = inet_getname,
1066         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1067         .poll           = dccp_poll,
1068         .ioctl          = inet_ioctl,
1069         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1070         .listen         = inet_dccp_listen,
1071         .shutdown       = inet_shutdown,
1072         .setsockopt     = sock_common_setsockopt,
1073         .getsockopt     = sock_common_getsockopt,
1074         .sendmsg        = inet_sendmsg,
1075         .recvmsg        = sock_common_recvmsg,
1076         .mmap           = sock_no_mmap,
1077         .sendpage       = sock_no_sendpage,
1078 };
1079
1080 static struct inet_protosw dccp_v4_protosw = {
1081         .type           = SOCK_DCCP,
1082         .protocol       = IPPROTO_DCCP,
1083         .prot           = &dccp_v4_prot,
1084         .ops            = &inet_dccp_ops,
1085         .capability     = -1,
1086         .no_check       = 0,
1087         .flags          = INET_PROTOSW_ICSK,
1088 };
1089
1090 /*
1091  * This is the global socket data structure used for responding to
1092  * the Out-of-the-blue (OOTB) packets. A control sock will be created
1093  * for this socket at the initialization time.
1094  */
1095 struct socket *dccp_ctl_socket;
1096
1097 static char dccp_ctl_socket_err_msg[] __initdata =
1098         KERN_ERR "DCCP: Failed to create the control socket.\n";
1099
1100 static int __init dccp_ctl_sock_init(void)
1101 {
1102         int rc = sock_create_kern(PF_INET, SOCK_DCCP, IPPROTO_DCCP,
1103                                   &dccp_ctl_socket);
1104         if (rc < 0)
1105                 printk(dccp_ctl_socket_err_msg);
1106         else {
1107                 dccp_ctl_socket->sk->sk_allocation = GFP_ATOMIC;
1108                 inet_sk(dccp_ctl_socket->sk)->uc_ttl = -1;
1109
1110                 /* Unhash it so that IP input processing does not even
1111                  * see it, we do not wish this socket to see incoming
1112                  * packets.
1113                  */
1114                 dccp_ctl_socket->sk->sk_prot->unhash(dccp_ctl_socket->sk);
1115         }
1116
1117         return rc;
1118 }
1119
1120 #ifdef CONFIG_IP_DCCP_UNLOAD_HACK
1121 void dccp_ctl_sock_exit(void)
1122 {
1123         if (dccp_ctl_socket != NULL) {
1124                 sock_release(dccp_ctl_socket);
1125                 dccp_ctl_socket = NULL;
1126         }
1127 }
1128
1129 EXPORT_SYMBOL_GPL(dccp_ctl_sock_exit);
1130 #endif
1131
1132 static int __init dccp_v4_init(void)
1133 {
1134         int err = proto_register(&dccp_v4_prot, 1);
1135
1136         if (err != 0)
1137                 goto out;
1138
1139         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1140         if (err != 0)
1141                 goto out_proto_unregister;
1142
1143         inet_register_protosw(&dccp_v4_protosw);
1144
1145         err = dccp_ctl_sock_init();
1146         if (err)
1147                 goto out_unregister_protosw;
1148 out:
1149         return err;
1150 out_unregister_protosw:
1151         inet_unregister_protosw(&dccp_v4_protosw);
1152         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1153 out_proto_unregister:
1154         proto_unregister(&dccp_v4_prot);
1155         goto out;
1156 }
1157
1158 static void __exit dccp_v4_exit(void)
1159 {
1160         inet_unregister_protosw(&dccp_v4_protosw);
1161         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1162         proto_unregister(&dccp_v4_prot);
1163 }
1164
1165 module_init(dccp_v4_init);
1166 module_exit(dccp_v4_exit);
1167
1168 /*
1169  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1170  * values directly, Also cover the case where the protocol is not specified,
1171  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1172  */
1173 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
1174 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1175 MODULE_LICENSE("GPL");
1176 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1177 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");