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/sock.h>
  23 #include <net/timewait_sock.h>
  24 #include <net/tcp_states.h>
  25 #include <net/xfrm.h>
  26
  27 #include "ackvec.h"
  28 #include "ccid.h"
  29 #include "dccp.h"
  30
  31 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
  32         .lhash_lock     = RW_LOCK_UNLOCKED,
  33         .lhash_users    = ATOMIC_INIT(0),
  34         .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
  35 };
  36
  37 EXPORT_SYMBOL_GPL(dccp_hashinfo);
  38
  39 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  40 {
  41         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
  42                                  inet_csk_bind_conflict);
  43 }
  44
  45 static void dccp_v4_hash(struct sock *sk)
  46 {
  47         inet_hash(&dccp_hashinfo, sk);
  48 }
  49
  50 void dccp_unhash(struct sock *sk)
  51 {
  52         inet_unhash(&dccp_hashinfo, sk);
  53 }
  54
  55 EXPORT_SYMBOL_GPL(dccp_unhash);
  56
  57 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  58 {
  59         struct inet_sock *inet = inet_sk(sk);
  60         struct dccp_sock *dp = dccp_sk(sk);
  61         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  62         struct rtable *rt;
  63         u32 daddr, nexthop;
  64         int tmp;
  65         int err;
  66
  67         dp->dccps_role = DCCP_ROLE_CLIENT;
  68
  69         if (dccp_service_not_initialized(sk))
  70                 return -EPROTO;
  71
  72         if (addr_len < sizeof(struct sockaddr_in))
  73                 return -EINVAL;
  74
  75         if (usin->sin_family != AF_INET)
  76                 return -EAFNOSUPPORT;
  77
  78         nexthop = daddr = usin->sin_addr.s_addr;
  79         if (inet->opt != NULL && inet->opt->srr) {
  80                 if (daddr == 0)
  81                         return -EINVAL;
  82                 nexthop = inet->opt->faddr;
  83         }
  84
  85         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
  86                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  87                                IPPROTO_DCCP,
  88                                inet->sport, usin->sin_port, sk);
  89         if (tmp < 0)
  90                 return tmp;
  91
  92         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  93                 ip_rt_put(rt);
  94                 return -ENETUNREACH;
  95         }
  96
  97         if (inet->opt == NULL || !inet->opt->srr)
  98                 daddr = rt->rt_dst;
  99
 100         if (inet->saddr == 0)
 101                 inet->saddr = rt->rt_src;
 102         inet->rcv_saddr = inet->saddr;
 103
 104         inet->dport = usin->sin_port;
 105         inet->daddr = daddr;
 106
 107         dp->dccps_ext_header_len = 0;
 108         if (inet->opt != NULL)
 109                 dp->dccps_ext_header_len = inet->opt->optlen;
 110         /*
 111          * Socket identity is still unknown (sport may be zero).
 112          * However we set state to DCCP_REQUESTING and not releasing socket
 113          * lock select source port, enter ourselves into the hash tables and
 114          * complete initialization after this.
 115          */
 116         dccp_set_state(sk, DCCP_REQUESTING);
 117         err = inet_hash_connect(&dccp_death_row, sk);
 118         if (err != 0)
 119                 goto failure;
 120
 121         err = ip_route_newports(&rt, inet->sport, inet->dport, sk);
 122         if (err != 0)
 123                 goto failure;
 124
 125         /* OK, now commit destination to socket.  */
 126         sk_setup_caps(sk, &rt->u.dst);
 127
 128         dp->dccps_gar =
 129                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 130                                                             inet->daddr,
 131                                                             inet->sport,
 132                                                             usin->sin_port);
 133         dccp_update_gss(sk, dp->dccps_iss);
 134
 135         inet->id = dp->dccps_iss ^ jiffies;
 136
 137         err = dccp_connect(sk);
 138         rt = NULL;
 139         if (err != 0)
 140                 goto failure;
 141 out:
 142         return err;
 143 failure:
 144         /*
 145          * This unhashes the socket and releases the local port, if necessary.
 146          */
 147         dccp_set_state(sk, DCCP_CLOSED);
 148         ip_rt_put(rt);
 149         sk->sk_route_caps = 0;
 150         inet->dport = 0;
 151         goto out;
 152 }
 153
 154 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 155
 156 /*
 157  * This routine does path mtu discovery as defined in RFC1191.
 158  */
 159 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 160                                           const struct iphdr *iph,
 161                                           u32 mtu)
 162 {
 163         struct dst_entry *dst;
 164         const struct inet_sock *inet = inet_sk(sk);
 165         const struct dccp_sock *dp = dccp_sk(sk);
 166
 167         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 168          * send out by Linux are always < 576bytes so they should go through
 169          * unfragmented).
 170          */
 171         if (sk->sk_state == DCCP_LISTEN)
 172                 return;
 173
 174         /* We don't check in the destentry if pmtu discovery is forbidden
 175          * on this route. We just assume that no packet_to_big packets
 176          * are send back when pmtu discovery is not active.
 177          * There is a small race when the user changes this flag in the
 178          * route, but I think that's acceptable.
 179          */
 180         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 181                 return;
 182
 183         dst->ops->update_pmtu(dst, mtu);
 184
 185         /* Something is about to be wrong... Remember soft error
 186          * for the case, if this connection will not able to recover.
 187          */
 188         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 189                 sk->sk_err_soft = EMSGSIZE;
 190
 191         mtu = dst_mtu(dst);
 192
 193         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 194             dp->dccps_pmtu_cookie > mtu) {
 195                 dccp_sync_mss(sk, mtu);
 196
 197                 /*
 198                  * From: draft-ietf-dccp-spec-11.txt
 199                  *
 200                  *      DCCP-Sync packets are the best choice for upward
 201                  *      probing, since DCCP-Sync probes do not risk application
 202                  *      data loss.
 203                  */
 204                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 205         } /* else let the usual retransmit timer handle it */
 206 }
 207
 208 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 209 {
 210         int err;
 211         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 212         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 213                                      sizeof(struct dccp_hdr_ext) +
 214                                      sizeof(struct dccp_hdr_ack_bits);
 215         struct sk_buff *skb;
 216
 217         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 218                 return;
 219
 220         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 221         if (skb == NULL)
 222                 return;
 223
 224         /* Reserve space for headers. */
 225         skb_reserve(skb, MAX_DCCP_HEADER);
 226
 227         skb->dst = dst_clone(rxskb->dst);
 228
 229         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 230         dh = dccp_hdr(skb);
 231         memset(dh, 0, dccp_hdr_ack_len);
 232
 233         /* Build DCCP header and checksum it. */
 234         dh->dccph_type     = DCCP_PKT_ACK;
 235         dh->dccph_sport    = rxdh->dccph_dport;
 236         dh->dccph_dport    = rxdh->dccph_sport;
 237         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 238         dh->dccph_x        = 1;
 239
 240         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 241         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 242                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 243
 244         bh_lock_sock(dccp_ctl_socket->sk);
 245         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 246                                     rxskb->nh.iph->daddr,
 247                                     rxskb->nh.iph->saddr, NULL);
 248         bh_unlock_sock(dccp_ctl_socket->sk);
 249
 250         if (err == NET_XMIT_CN || err == 0) {
 251                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 252                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 253         }
 254 }
 255
 256 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 257                                    struct request_sock *req)
 258 {
 259         dccp_v4_ctl_send_ack(skb);
 260 }
 261
 262 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 263                                  struct dst_entry *dst)
 264 {
 265         int err = -1;
 266         struct sk_buff *skb;
 267
 268         /* First, grab a route. */
 269
 270         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 271                 goto out;
 272
 273         skb = dccp_make_response(sk, dst, req);
 274         if (skb != NULL) {
 275                 const struct inet_request_sock *ireq = inet_rsk(req);
 276
 277                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 278                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 279                                             ireq->rmt_addr,
 280                                             ireq->opt);
 281                 if (err == NET_XMIT_CN)
 282                         err = 0;
 283         }
 284
 285 out:
 286         dst_release(dst);
 287         return err;
 288 }
 289
 290 /*
 291  * This routine is called by the ICMP module when it gets some sort of error
 292  * condition. If err < 0 then the socket should be closed and the error
 293  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 294  * After adjustment header points to the first 8 bytes of the tcp header. We
 295  * need to find the appropriate port.
 296  *
 297  * The locking strategy used here is very "optimistic". When someone else
 298  * accesses the socket the ICMP is just dropped and for some paths there is no
 299  * check at all. A more general error queue to queue errors for later handling
 300  * is probably better.
 301  */
 302 void dccp_v4_err(struct sk_buff *skb, u32 info)
 303 {
 304         const struct iphdr *iph = (struct iphdr *)skb->data;
 305         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 306                                                         (iph->ihl << 2));
 307         struct dccp_sock *dp;
 308         struct inet_sock *inet;
 309         const int type = skb->h.icmph->type;
 310         const int code = skb->h.icmph->code;
 311         struct sock *sk;
 312         __u64 seq;
 313         int err;
 314
 315         if (skb->len < (iph->ihl << 2) + 8) {
 316                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 317                 return;
 318         }
 319
 320         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 321                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 322         if (sk == NULL) {
 323                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 324                 return;
 325         }
 326
 327         if (sk->sk_state == DCCP_TIME_WAIT) {
 328                 inet_twsk_put((struct inet_timewait_sock *)sk);
 329                 return;
 330         }
 331
 332         bh_lock_sock(sk);
 333         /* If too many ICMPs get dropped on busy
 334          * servers this needs to be solved differently.
 335          */
 336         if (sock_owned_by_user(sk))
 337                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 338
 339         if (sk->sk_state == DCCP_CLOSED)
 340                 goto out;
 341
 342         dp = dccp_sk(sk);
 343         seq = dccp_hdr_seq(skb);
 344         if (sk->sk_state != DCCP_LISTEN &&
 345             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 346                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 347                 goto out;
 348         }
 349
 350         switch (type) {
 351         case ICMP_SOURCE_QUENCH:
 352                 /* Just silently ignore these. */
 353                 goto out;
 354         case ICMP_PARAMETERPROB:
 355                 err = EPROTO;
 356                 break;
 357         case ICMP_DEST_UNREACH:
 358                 if (code > NR_ICMP_UNREACH)
 359                         goto out;
 360
 361                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 362                         if (!sock_owned_by_user(sk))
 363                                 dccp_do_pmtu_discovery(sk, iph, info);
 364                         goto out;
 365                 }
 366
 367                 err = icmp_err_convert[code].errno;
 368                 break;
 369         case ICMP_TIME_EXCEEDED:
 370                 err = EHOSTUNREACH;
 371                 break;
 372         default:
 373                 goto out;
 374         }
 375
 376         switch (sk->sk_state) {
 377                 struct request_sock *req , **prev;
 378         case DCCP_LISTEN:
 379                 if (sock_owned_by_user(sk))
 380                         goto out;
 381                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 382                                           iph->daddr, iph->saddr);
 383                 if (!req)
 384                         goto out;
 385
 386                 /*
 387                  * ICMPs are not backlogged, hence we cannot get an established
 388                  * socket here.
 389                  */
 390                 BUG_TRAP(!req->sk);
 391
 392                 if (seq != dccp_rsk(req)->dreq_iss) {
 393                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 394                         goto out;
 395                 }
 396                 /*
 397                  * Still in RESPOND, just remove it silently.
 398                  * There is no good way to pass the error to the newly
 399                  * created socket, and POSIX does not want network
 400                  * errors returned from accept().
 401                  */
 402                 inet_csk_reqsk_queue_drop(sk, req, prev);
 403                 goto out;
 404
 405         case DCCP_REQUESTING:
 406         case DCCP_RESPOND:
 407                 if (!sock_owned_by_user(sk)) {
 408                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 409                         sk->sk_err = err;
 410
 411                         sk->sk_error_report(sk);
 412
 413                         dccp_done(sk);
 414                 } else
 415                         sk->sk_err_soft = err;
 416                 goto out;
 417         }
 418
 419         /* If we've already connected we will keep trying
 420          * until we time out, or the user gives up.
 421          *
 422          * rfc1122 4.2.3.9 allows to consider as hard errors
 423          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 424          * but it is obsoleted by pmtu discovery).
 425          *
 426          * Note, that in modern internet, where routing is unreliable
 427          * and in each dark corner broken firewalls sit, sending random
 428          * errors ordered by their masters even this two messages finally lose
 429          * their original sense (even Linux sends invalid PORT_UNREACHs)
 430          *
 431          * Now we are in compliance with RFCs.
 432          *                                                      --ANK (980905)
 433          */
 434
 435         inet = inet_sk(sk);
 436         if (!sock_owned_by_user(sk) && inet->recverr) {
 437                 sk->sk_err = err;
 438                 sk->sk_error_report(sk);
 439         } else /* Only an error on timeout */
 440                 sk->sk_err_soft = err;
 441 out:
 442         bh_unlock_sock(sk);
 443         sock_put(sk);
 444 }
 445
 446 /* This routine computes an IPv4 DCCP checksum. */
 447 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 448 {
 449         const struct inet_sock *inet = inet_sk(sk);
 450         struct dccp_hdr *dh = dccp_hdr(skb);
 451
 452         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 453 }
 454
 455 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 456
 457 int dccp_v4_send_reset(struct sock *sk, enum dccp_reset_codes code)
 458 {
 459         struct sk_buff *skb;
 460         /*
 461          * FIXME: what if rebuild_header fails?
 462          * Should we be doing a rebuild_header here?
 463          */
 464         int err = inet_sk_rebuild_header(sk);
 465
 466         if (err != 0)
 467                 return err;
 468
 469         skb = dccp_make_reset(sk, sk->sk_dst_cache, code);
 470         if (skb != NULL) {
 471                 const struct inet_sock *inet = inet_sk(sk);
 472
 473                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 474                 err = ip_build_and_send_pkt(skb, sk,
 475                                             inet->saddr, inet->daddr, NULL);
 476                 if (err == NET_XMIT_CN)
 477                         err = 0;
 478         }
 479
 480         return err;
 481 }
 482
 483 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 484                                         const struct sk_buff *skb)
 485 {
 486         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 487                                            skb->nh.iph->saddr,
 488                                            dccp_hdr(skb)->dccph_dport,
 489                                            dccp_hdr(skb)->dccph_sport);
 490 }
 491
 492 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 493 {
 494         struct inet_request_sock *ireq;
 495         struct dccp_sock dp;
 496         struct request_sock *req;
 497         struct dccp_request_sock *dreq;
 498         const __u32 saddr = skb->nh.iph->saddr;
 499         const __u32 daddr = skb->nh.iph->daddr;
 500         const __u32 service = dccp_hdr_request(skb)->dccph_req_service;
 501         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 502         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 503
 504         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 505         if (((struct rtable *)skb->dst)->rt_flags &
 506             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 507                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 508                 goto drop;
 509         }
 510
 511         if (dccp_bad_service_code(sk, service)) {
 512                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 513                 goto drop;
 514         }
 515         /*
 516          * TW buckets are converted to open requests without
 517          * limitations, they conserve resources and peer is
 518          * evidently real one.
 519          */
 520         if (inet_csk_reqsk_queue_is_full(sk))
 521                 goto drop;
 522
 523         /*
 524          * Accept backlog is full. If we have already queued enough
 525          * of warm entries in syn queue, drop request. It is better than
 526          * clogging syn queue with openreqs with exponentially increasing
 527          * timeout.
 528          */
 529         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 530                 goto drop;
 531
 532         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 533         if (req == NULL)
 534                 goto drop;
 535
 536         /* FIXME: process options */
 537
 538         dccp_openreq_init(req, &dp, skb);
 539
 540         ireq = inet_rsk(req);
 541         ireq->loc_addr = daddr;
 542         ireq->rmt_addr = saddr;
 543         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 544                                   right value */
 545         ireq->opt       = NULL;
 546
 547         /*
 548          * Step 3: Process LISTEN state
 549          *
 550          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 551          *
 552          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 553          * dccp_create_openreq_child.
 554          */
 555         dreq = dccp_rsk(req);
 556         dreq->dreq_isr     = dcb->dccpd_seq;
 557         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 558         dreq->dreq_service = service;
 559
 560         if (dccp_v4_send_response(sk, req, NULL))
 561                 goto drop_and_free;
 562
 563         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 564         return 0;
 565
 566 drop_and_free:
 567         reqsk_free(req);
 568 drop:
 569         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 570         dcb->dccpd_reset_code = reset_code;
 571         return -1;
 572 }
 573
 574 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 575
 576 /*
 577  * The three way handshake has completed - we got a valid ACK or DATAACK -
 578  * now create the new socket.
 579  *
 580  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 581  */
 582 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 583                                        struct request_sock *req,
 584                                        struct dst_entry *dst)
 585 {
 586         struct inet_request_sock *ireq;
 587         struct inet_sock *newinet;
 588         struct dccp_sock *newdp;
 589         struct sock *newsk;
 590
 591         if (sk_acceptq_is_full(sk))
 592                 goto exit_overflow;
 593
 594         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 595                 goto exit;
 596
 597         newsk = dccp_create_openreq_child(sk, req, skb);
 598         if (newsk == NULL)
 599                 goto exit;
 600
 601         sk_setup_caps(newsk, dst);
 602
 603         newdp              = dccp_sk(newsk);
 604         newinet            = inet_sk(newsk);
 605         ireq               = inet_rsk(req);
 606         newinet->daddr     = ireq->rmt_addr;
 607         newinet->rcv_saddr = ireq->loc_addr;
 608         newinet->saddr     = ireq->loc_addr;
 609         newinet->opt       = ireq->opt;
 610         ireq->opt          = NULL;
 611         newinet->mc_index  = inet_iif(skb);
 612         newinet->mc_ttl    = skb->nh.iph->ttl;
 613         newinet->id        = jiffies;
 614
 615         dccp_sync_mss(newsk, dst_mtu(dst));
 616
 617         __inet_hash(&dccp_hashinfo, newsk, 0);
 618         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 619
 620         return newsk;
 621
 622 exit_overflow:
 623         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 624 exit:
 625         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 626         dst_release(dst);
 627         return NULL;
 628 }
 629
 630 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 631
 632 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 633 {
 634         const struct dccp_hdr *dh = dccp_hdr(skb);
 635         const struct iphdr *iph = skb->nh.iph;
 636         struct sock *nsk;
 637         struct request_sock **prev;
 638         /* Find possible connection requests. */
 639         struct request_sock *req = inet_csk_search_req(sk, &prev,
 640                                                        dh->dccph_sport,
 641                                                        iph->saddr, iph->daddr);
 642         if (req != NULL)
 643                 return dccp_check_req(sk, skb, req, prev);
 644
 645         nsk = __inet_lookup_established(&dccp_hashinfo,
 646                                         iph->saddr, dh->dccph_sport,
 647                                         iph->daddr, ntohs(dh->dccph_dport),
 648                                         inet_iif(skb));
 649         if (nsk != NULL) {
 650                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 651                         bh_lock_sock(nsk);
 652                         return nsk;
 653                 }
 654                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 655                 return NULL;
 656         }
 657
 658         return sk;
 659 }
 660
 661 int dccp_v4_checksum(const struct sk_buff *skb, const u32 saddr,
 662                      const u32 daddr)
 663 {
 664         const struct dccp_hdr* dh = dccp_hdr(skb);
 665         int checksum_len;
 666         u32 tmp;
 667
 668         if (dh->dccph_cscov == 0)
 669                 checksum_len = skb->len;
 670         else {
 671                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 672                 checksum_len = checksum_len < skb->len ? checksum_len :
 673                                                          skb->len;
 674         }
 675
 676         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 677         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 678                                  IPPROTO_DCCP, tmp);
 679 }
 680
 681 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 682                                    const u32 saddr, const u32 daddr)
 683 {
 684         struct dccp_hdr *dh = dccp_hdr(skb);
 685         int checksum_len;
 686         u32 tmp;
 687
 688         if (dh->dccph_cscov == 0)
 689                 checksum_len = skb->len;
 690         else {
 691                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 692                 checksum_len = checksum_len < skb->len ? checksum_len :
 693                                                          skb->len;
 694         }
 695         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 696         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 697                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 698 }
 699
 700 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 701                                            struct sk_buff *skb)
 702 {
 703         struct rtable *rt;
 704         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 705                             .nl_u = { .ip4_u =
 706                                       { .daddr = skb->nh.iph->saddr,
 707                                         .saddr = skb->nh.iph->daddr,
 708                                         .tos = RT_CONN_FLAGS(sk) } },
 709                             .proto = sk->sk_protocol,
 710                             .uli_u = { .ports =
 711                                        { .sport = dccp_hdr(skb)->dccph_dport,
 712                                          .dport = dccp_hdr(skb)->dccph_sport }
 713                                      }
 714                           };
 715
 716         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 717                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 718                 return NULL;
 719         }
 720
 721         return &rt->u.dst;
 722 }
 723
 724 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 725 {
 726         int err;
 727         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 728         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 729                                        sizeof(struct dccp_hdr_ext) +
 730                                        sizeof(struct dccp_hdr_reset);
 731         struct sk_buff *skb;
 732         struct dst_entry *dst;
 733         u64 seqno;
 734
 735         /* Never send a reset in response to a reset. */
 736         if (rxdh->dccph_type == DCCP_PKT_RESET)
 737                 return;
 738
 739         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 740                 return;
 741
 742         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 743         if (dst == NULL)
 744                 return;
 745
 746         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 747         if (skb == NULL)
 748                 goto out;
 749
 750         /* Reserve space for headers. */
 751         skb_reserve(skb, MAX_DCCP_HEADER);
 752         skb->dst = dst_clone(dst);
 753
 754         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 755         dh = dccp_hdr(skb);
 756         memset(dh, 0, dccp_hdr_reset_len);
 757
 758         /* Build DCCP header and checksum it. */
 759         dh->dccph_type     = DCCP_PKT_RESET;
 760         dh->dccph_sport    = rxdh->dccph_dport;
 761         dh->dccph_dport    = rxdh->dccph_sport;
 762         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 763         dh->dccph_x        = 1;
 764         dccp_hdr_reset(skb)->dccph_reset_code =
 765                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 766
 767         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 768         seqno = 0;
 769         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 770                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 771
 772         dccp_hdr_set_seq(dh, seqno);
 773         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 774                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 775
 776         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 777                                               rxskb->nh.iph->daddr);
 778
 779         bh_lock_sock(dccp_ctl_socket->sk);
 780         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 781                                     rxskb->nh.iph->daddr,
 782                                     rxskb->nh.iph->saddr, NULL);
 783         bh_unlock_sock(dccp_ctl_socket->sk);
 784
 785         if (err == NET_XMIT_CN || err == 0) {
 786                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 787                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 788         }
 789 out:
 790          dst_release(dst);
 791 }
 792
 793 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 794 {
 795         struct dccp_hdr *dh = dccp_hdr(skb);
 796
 797         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 798                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 799                         goto reset;
 800                 return 0;
 801         }
 802
 803         /*
 804          *  Step 3: Process LISTEN state
 805          *     If S.state == LISTEN,
 806          *        If P.type == Request or P contains a valid Init Cookie
 807          *              option,
 808          *           * Must scan the packet's options to check for an Init
 809          *              Cookie.  Only the Init Cookie is processed here,
 810          *              however; other options are processed in Step 8.  This
 811          *              scan need only be performed if the endpoint uses Init
 812          *              Cookies *
 813          *           * Generate a new socket and switch to that socket *
 814          *           Set S := new socket for this port pair
 815          *           S.state = RESPOND
 816          *           Choose S.ISS (initial seqno) or set from Init Cookie
 817          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 818          *           Continue with S.state == RESPOND
 819          *           * A Response packet will be generated in Step 11 *
 820          *        Otherwise,
 821          *           Generate Reset(No Connection) unless P.type == Reset
 822          *           Drop packet and return
 823          *
 824          * NOTE: the check for the packet types is done in
 825          *       dccp_rcv_state_process
 826          */
 827         if (sk->sk_state == DCCP_LISTEN) {
 828                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 829
 830                 if (nsk == NULL)
 831                         goto discard;
 832
 833                 if (nsk != sk) {
 834                         if (dccp_child_process(sk, nsk, skb))
 835                                 goto reset;
 836                         return 0;
 837                 }
 838         }
 839
 840         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 841                 goto reset;
 842         return 0;
 843
 844 reset:
 845         dccp_v4_ctl_send_reset(skb);
 846 discard:
 847         kfree_skb(skb);
 848         return 0;
 849 }
 850
 851 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 852
 853 int dccp_invalid_packet(struct sk_buff *skb)
 854 {
 855         const struct dccp_hdr *dh;
 856
 857         if (skb->pkt_type != PACKET_HOST)
 858                 return 1;
 859
 860         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 861                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 862                 return 1;
 863         }
 864
 865         dh = dccp_hdr(skb);
 866
 867         /* If the packet type is not understood, drop packet and return */
 868         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 869                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 870                 return 1;
 871         }
 872
 873         /*
 874          * If P.Data Offset is too small for packet type, or too large for
 875          * packet, drop packet and return
 876          */
 877         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 878                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 879                                             "too small 1\n",
 880                                dh->dccph_doff);
 881                 return 1;
 882         }
 883
 884         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 885                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 886                                             "too small 2\n",
 887                                dh->dccph_doff);
 888                 return 1;
 889         }
 890
 891         dh = dccp_hdr(skb);
 892
 893         /*
 894          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 895          * has short sequence numbers), drop packet and return
 896          */
 897         if (dh->dccph_x == 0 &&
 898             dh->dccph_type != DCCP_PKT_DATA &&
 899             dh->dccph_type != DCCP_PKT_ACK &&
 900             dh->dccph_type != DCCP_PKT_DATAACK) {
 901                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 902                                             "nor DataAck and P.X == 0\n",
 903                                dccp_packet_name(dh->dccph_type));
 904                 return 1;
 905         }
 906
 907         return 0;
 908 }
 909
 910 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 911
 912 /* this is called when real data arrives */
 913 int dccp_v4_rcv(struct sk_buff *skb)
 914 {
 915         const struct dccp_hdr *dh;
 916         struct sock *sk;
 917         int rc;
 918
 919         /* Step 1: Check header basics: */
 920
 921         if (dccp_invalid_packet(skb))
 922                 goto discard_it;
 923
 924         /* If the header checksum is incorrect, drop packet and return */
 925         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
 926                                     skb->nh.iph->daddr) < 0) {
 927                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
 928                                __FUNCTION__);
 929                 goto discard_it;
 930         }
 931
 932         dh = dccp_hdr(skb);
 933
 934         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 935         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 936
 937         dccp_pr_debug("%8.8s "
 938                       "src=%u.%u.%u.%u@%-5d "
 939                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 940                       dccp_packet_name(dh->dccph_type),
 941                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 942                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 943                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 944
 945         if (dccp_packet_without_ack(skb)) {
 946                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 947                 dccp_pr_debug_cat("\n");
 948         } else {
 949                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 950                 dccp_pr_debug_cat(", ack=%llu\n",
 951                                   (unsigned long long)
 952                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 953         }
 954
 955         /* Step 2:
 956          *      Look up flow ID in table and get corresponding socket */
 957         sk = __inet_lookup(&dccp_hashinfo,
 958                            skb->nh.iph->saddr, dh->dccph_sport,
 959                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
 960                            inet_iif(skb));
 961
 962         /*
 963          * Step 2:
 964          *      If no socket ...
 965          *              Generate Reset(No Connection) unless P.type == Reset
 966          *              Drop packet and return
 967          */
 968         if (sk == NULL) {
 969                 dccp_pr_debug("failed to look up flow ID in table and "
 970                               "get corresponding socket\n");
 971                 goto no_dccp_socket;
 972         }
 973
 974         /*
 975          * Step 2:
 976          *      ... or S.state == TIMEWAIT,
 977          *              Generate Reset(No Connection) unless P.type == Reset
 978          *              Drop packet and return
 979          */
 980
 981         if (sk->sk_state == DCCP_TIME_WAIT) {
 982                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
 983                               "do_time_wait\n");
 984                 goto do_time_wait;
 985         }
 986
 987         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
 988                 dccp_pr_debug("xfrm4_policy_check failed\n");
 989                 goto discard_and_relse;
 990         }
 991
 992         if (sk_filter(sk, skb, 0)) {
 993                 dccp_pr_debug("sk_filter failed\n");
 994                 goto discard_and_relse;
 995         }
 996
 997         skb->dev = NULL;
 998
 999         bh_lock_sock(sk);
1000         rc = 0;
1001         if (!sock_owned_by_user(sk))
1002                 rc = dccp_v4_do_rcv(sk, skb);
1003         else
1004                 sk_add_backlog(sk, skb);
1005         bh_unlock_sock(sk);
1006
1007         sock_put(sk);
1008         return rc;
1009
1010 no_dccp_socket:
1011         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1012                 goto discard_it;
1013         /*
1014          * Step 2:
1015          *              Generate Reset(No Connection) unless P.type == Reset
1016          *              Drop packet and return
1017          */
1018         if (dh->dccph_type != DCCP_PKT_RESET) {
1019                 DCCP_SKB_CB(skb)->dccpd_reset_code =
1020                                         DCCP_RESET_CODE_NO_CONNECTION;
1021                 dccp_v4_ctl_send_reset(skb);
1022         }
1023
1024 discard_it:
1025         /* Discard frame. */
1026         kfree_skb(skb);
1027         return 0;
1028
1029 discard_and_relse:
1030         sock_put(sk);
1031         goto discard_it;
1032
1033 do_time_wait:
1034         inet_twsk_put((struct inet_timewait_sock *)sk);
1035         goto no_dccp_socket;
1036 }
1037
1038 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
1039         .queue_xmit     = ip_queue_xmit,
1040         .send_check     = dccp_v4_send_check,
1041         .rebuild_header = inet_sk_rebuild_header,
1042         .conn_request   = dccp_v4_conn_request,
1043         .syn_recv_sock  = dccp_v4_request_recv_sock,
1044         .net_header_len = sizeof(struct iphdr),
1045         .setsockopt     = ip_setsockopt,
1046         .getsockopt     = ip_getsockopt,
1047         .addr2sockaddr  = inet_csk_addr2sockaddr,
1048         .sockaddr_len   = sizeof(struct sockaddr_in),
1049 };
1050
1051 int dccp_v4_init_sock(struct sock *sk)
1052 {
1053         struct dccp_sock *dp = dccp_sk(sk);
1054         static int dccp_ctl_socket_init = 1;
1055
1056         dccp_options_init(&dp->dccps_options);
1057         do_gettimeofday(&dp->dccps_epoch);
1058
1059         if (dp->dccps_options.dccpo_send_ack_vector) {
1060                 dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(DCCP_MAX_ACKVEC_LEN,
1061                                                            GFP_KERNEL);
1062                 if (dp->dccps_hc_rx_ackvec == NULL)
1063                         return -ENOMEM;
1064         }
1065
1066         /*
1067          * FIXME: We're hardcoding the CCID, and doing this at this point makes
1068          * the listening (master) sock get CCID control blocks, which is not
1069          * necessary, but for now, to not mess with the test userspace apps,
1070          * lets leave it here, later the real solution is to do this in a
1071          * setsockopt(CCIDs-I-want/accept). -acme
1072          */
1073         if (likely(!dccp_ctl_socket_init)) {
1074                 dp->dccps_hc_rx_ccid = ccid_init(dp->dccps_options.dccpo_rx_ccid,
1075                                                  sk);
1076                 dp->dccps_hc_tx_ccid = ccid_init(dp->dccps_options.dccpo_tx_ccid,
1077                                                  sk);
1078                 if (dp->dccps_hc_rx_ccid == NULL ||
1079                     dp->dccps_hc_tx_ccid == NULL) {
1080                         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1081                         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1082                         if (dp->dccps_options.dccpo_send_ack_vector) {
1083                                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1084                                 dp->dccps_hc_rx_ackvec = NULL;
1085                         }
1086                         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1087                         return -ENOMEM;
1088                 }
1089         } else
1090                 dccp_ctl_socket_init = 0;
1091
1092         dccp_init_xmit_timers(sk);
1093         inet_csk(sk)->icsk_rto = DCCP_TIMEOUT_INIT;
1094         sk->sk_state = DCCP_CLOSED;
1095         sk->sk_write_space = dccp_write_space;
1096         inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1097         dp->dccps_mss_cache = 536;
1098         dp->dccps_role = DCCP_ROLE_UNDEFINED;
1099         dp->dccps_service = DCCP_SERVICE_INVALID_VALUE;
1100
1101         return 0;
1102 }
1103
1104 EXPORT_SYMBOL_GPL(dccp_v4_init_sock);
1105
1106 int dccp_v4_destroy_sock(struct sock *sk)
1107 {
1108         struct dccp_sock *dp = dccp_sk(sk);
1109
1110         /*
1111          * DCCP doesn't use sk_write_queue, just sk_send_head
1112          * for retransmissions
1113          */
1114         if (sk->sk_send_head != NULL) {
1115                 kfree_skb(sk->sk_send_head);
1116                 sk->sk_send_head = NULL;
1117         }
1118
1119         /* Clean up a referenced DCCP bind bucket. */
1120         if (inet_csk(sk)->icsk_bind_hash != NULL)
1121                 inet_put_port(&dccp_hashinfo, sk);
1122
1123         kfree(dp->dccps_service_list);
1124         dp->dccps_service_list = NULL;
1125
1126         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
1127         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
1128         if (dp->dccps_options.dccpo_send_ack_vector) {
1129                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1130                 dp->dccps_hc_rx_ackvec = NULL;
1131         }
1132         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1133         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1134         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1135
1136         return 0;
1137 }
1138
1139 EXPORT_SYMBOL_GPL(dccp_v4_destroy_sock);
1140
1141 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1142 {
1143         kfree(inet_rsk(req)->opt);
1144 }
1145
1146 static struct request_sock_ops dccp_request_sock_ops = {
1147         .family         = PF_INET,
1148         .obj_size       = sizeof(struct dccp_request_sock),
1149         .rtx_syn_ack    = dccp_v4_send_response,
1150         .send_ack       = dccp_v4_reqsk_send_ack,
1151         .destructor     = dccp_v4_reqsk_destructor,
1152         .send_reset     = dccp_v4_ctl_send_reset,
1153 };
1154
1155 static struct timewait_sock_ops dccp_timewait_sock_ops = {
1156         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
1157 };
1158
1159 struct proto dccp_prot = {
1160         .name                   = "DCCP",
1161         .owner                  = THIS_MODULE,
1162         .close                  = dccp_close,
1163         .connect                = dccp_v4_connect,
1164         .disconnect             = dccp_disconnect,
1165         .ioctl                  = dccp_ioctl,
1166         .init                   = dccp_v4_init_sock,
1167         .setsockopt             = dccp_setsockopt,
1168         .getsockopt             = dccp_getsockopt,
1169         .sendmsg                = dccp_sendmsg,
1170         .recvmsg                = dccp_recvmsg,
1171         .backlog_rcv            = dccp_v4_do_rcv,
1172         .hash                   = dccp_v4_hash,
1173         .unhash                 = dccp_unhash,
1174         .accept                 = inet_csk_accept,
1175         .get_port               = dccp_v4_get_port,
1176         .shutdown               = dccp_shutdown,
1177         .destroy                = dccp_v4_destroy_sock,
1178         .orphan_count           = &dccp_orphan_count,
1179         .max_header             = MAX_DCCP_HEADER,
1180         .obj_size               = sizeof(struct dccp_sock),
1181         .rsk_prot               = &dccp_request_sock_ops,
1182         .twsk_prot              = &dccp_timewait_sock_ops,
1183 };
1184
1185 EXPORT_SYMBOL_GPL(dccp_prot);