net/dccp/ipv4.c

   1 /*
   2  *  net/dccp/ipv4.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/config.h>
  14 #include <linux/dccp.h>
  15 #include <linux/icmp.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/random.h>
  19
  20 #include <net/icmp.h>
  21 #include <net/inet_hashtables.h>
  22 #include <net/inet_sock.h>
  23 #include <net/sock.h>
  24 #include <net/timewait_sock.h>
  25 #include <net/tcp_states.h>
  26 #include <net/xfrm.h>
  27
  28 #include "ackvec.h"
  29 #include "ccid.h"
  30 #include "dccp.h"
  31 #include "feat.h"
  32
  33 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
  34         .lhash_lock     = RW_LOCK_UNLOCKED,
  35         .lhash_users    = ATOMIC_INIT(0),
  36         .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
  37 };
  38
  39 EXPORT_SYMBOL_GPL(dccp_hashinfo);
  40
  41 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  42 {
  43         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
  44                                  inet_csk_bind_conflict);
  45 }
  46
  47 static void dccp_v4_hash(struct sock *sk)
  48 {
  49         inet_hash(&dccp_hashinfo, sk);
  50 }
  51
  52 void dccp_unhash(struct sock *sk)
  53 {
  54         inet_unhash(&dccp_hashinfo, sk);
  55 }
  56
  57 EXPORT_SYMBOL_GPL(dccp_unhash);
  58
  59 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  60 {
  61         struct inet_sock *inet = inet_sk(sk);
  62         struct dccp_sock *dp = dccp_sk(sk);
  63         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  64         struct rtable *rt;
  65         u32 daddr, nexthop;
  66         int tmp;
  67         int err;
  68
  69         dp->dccps_role = DCCP_ROLE_CLIENT;
  70
  71         if (dccp_service_not_initialized(sk))
  72                 return -EPROTO;
  73
  74         if (addr_len < sizeof(struct sockaddr_in))
  75                 return -EINVAL;
  76
  77         if (usin->sin_family != AF_INET)
  78                 return -EAFNOSUPPORT;
  79
  80         nexthop = daddr = usin->sin_addr.s_addr;
  81         if (inet->opt != NULL && inet->opt->srr) {
  82                 if (daddr == 0)
  83                         return -EINVAL;
  84                 nexthop = inet->opt->faddr;
  85         }
  86
  87         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
  88                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  89                                IPPROTO_DCCP,
  90                                inet->sport, usin->sin_port, sk);
  91         if (tmp < 0)
  92                 return tmp;
  93
  94         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  95                 ip_rt_put(rt);
  96                 return -ENETUNREACH;
  97         }
  98
  99         if (inet->opt == NULL || !inet->opt->srr)
 100                 daddr = rt->rt_dst;
 101
 102         if (inet->saddr == 0)
 103                 inet->saddr = rt->rt_src;
 104         inet->rcv_saddr = inet->saddr;
 105
 106         inet->dport = usin->sin_port;
 107         inet->daddr = daddr;
 108
 109         inet_csk(sk)->icsk_ext_hdr_len = 0;
 110         if (inet->opt != NULL)
 111                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
 112         /*
 113          * Socket identity is still unknown (sport may be zero).
 114          * However we set state to DCCP_REQUESTING and not releasing socket
 115          * lock select source port, enter ourselves into the hash tables and
 116          * complete initialization after this.
 117          */
 118         dccp_set_state(sk, DCCP_REQUESTING);
 119         err = inet_hash_connect(&dccp_death_row, sk);
 120         if (err != 0)
 121                 goto failure;
 122
 123         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
 124                                 sk);
 125         if (err != 0)
 126                 goto failure;
 127
 128         /* OK, now commit destination to socket.  */
 129         sk_setup_caps(sk, &rt->u.dst);
 130
 131         dp->dccps_gar =
 132                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 133                                                             inet->daddr,
 134                                                             inet->sport,
 135                                                             usin->sin_port);
 136         dccp_update_gss(sk, dp->dccps_iss);
 137
 138         inet->id = dp->dccps_iss ^ jiffies;
 139
 140         err = dccp_connect(sk);
 141         rt = NULL;
 142         if (err != 0)
 143                 goto failure;
 144 out:
 145         return err;
 146 failure:
 147         /*
 148          * This unhashes the socket and releases the local port, if necessary.
 149          */
 150         dccp_set_state(sk, DCCP_CLOSED);
 151         ip_rt_put(rt);
 152         sk->sk_route_caps = 0;
 153         inet->dport = 0;
 154         goto out;
 155 }
 156
 157 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 158
 159 /*
 160  * This routine does path mtu discovery as defined in RFC1191.
 161  */
 162 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 163                                           const struct iphdr *iph,
 164                                           u32 mtu)
 165 {
 166         struct dst_entry *dst;
 167         const struct inet_sock *inet = inet_sk(sk);
 168         const struct dccp_sock *dp = dccp_sk(sk);
 169
 170         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 171          * send out by Linux are always < 576bytes so they should go through
 172          * unfragmented).
 173          */
 174         if (sk->sk_state == DCCP_LISTEN)
 175                 return;
 176
 177         /* We don't check in the destentry if pmtu discovery is forbidden
 178          * on this route. We just assume that no packet_to_big packets
 179          * are send back when pmtu discovery is not active.
 180          * There is a small race when the user changes this flag in the
 181          * route, but I think that's acceptable.
 182          */
 183         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 184                 return;
 185
 186         dst->ops->update_pmtu(dst, mtu);
 187
 188         /* Something is about to be wrong... Remember soft error
 189          * for the case, if this connection will not able to recover.
 190          */
 191         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 192                 sk->sk_err_soft = EMSGSIZE;
 193
 194         mtu = dst_mtu(dst);
 195
 196         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 197             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 198                 dccp_sync_mss(sk, mtu);
 199
 200                 /*
 201                  * From: draft-ietf-dccp-spec-11.txt
 202                  *
 203                  *      DCCP-Sync packets are the best choice for upward
 204                  *      probing, since DCCP-Sync probes do not risk application
 205                  *      data loss.
 206                  */
 207                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 208         } /* else let the usual retransmit timer handle it */
 209 }
 210
 211 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 212 {
 213         int err;
 214         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 215         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 216                                      sizeof(struct dccp_hdr_ext) +
 217                                      sizeof(struct dccp_hdr_ack_bits);
 218         struct sk_buff *skb;
 219
 220         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 221                 return;
 222
 223         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 224         if (skb == NULL)
 225                 return;
 226
 227         /* Reserve space for headers. */
 228         skb_reserve(skb, MAX_DCCP_HEADER);
 229
 230         skb->dst = dst_clone(rxskb->dst);
 231
 232         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 233         dh = dccp_hdr(skb);
 234         memset(dh, 0, dccp_hdr_ack_len);
 235
 236         /* Build DCCP header and checksum it. */
 237         dh->dccph_type     = DCCP_PKT_ACK;
 238         dh->dccph_sport    = rxdh->dccph_dport;
 239         dh->dccph_dport    = rxdh->dccph_sport;
 240         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 241         dh->dccph_x        = 1;
 242
 243         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 244         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 245                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 246
 247         bh_lock_sock(dccp_ctl_socket->sk);
 248         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 249                                     rxskb->nh.iph->daddr,
 250                                     rxskb->nh.iph->saddr, NULL);
 251         bh_unlock_sock(dccp_ctl_socket->sk);
 252
 253         if (err == NET_XMIT_CN || err == 0) {
 254                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 255                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 256         }
 257 }
 258
 259 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 260                                    struct request_sock *req)
 261 {
 262         dccp_v4_ctl_send_ack(skb);
 263 }
 264
 265 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 266                                  struct dst_entry *dst)
 267 {
 268         int err = -1;
 269         struct sk_buff *skb;
 270
 271         /* First, grab a route. */
 272
 273         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 274                 goto out;
 275
 276         skb = dccp_make_response(sk, dst, req);
 277         if (skb != NULL) {
 278                 const struct inet_request_sock *ireq = inet_rsk(req);
 279
 280                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 281                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 282                                             ireq->rmt_addr,
 283                                             ireq->opt);
 284                 if (err == NET_XMIT_CN)
 285                         err = 0;
 286         }
 287
 288 out:
 289         dst_release(dst);
 290         return err;
 291 }
 292
 293 /*
 294  * This routine is called by the ICMP module when it gets some sort of error
 295  * condition. If err < 0 then the socket should be closed and the error
 296  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 297  * After adjustment header points to the first 8 bytes of the tcp header. We
 298  * need to find the appropriate port.
 299  *
 300  * The locking strategy used here is very "optimistic". When someone else
 301  * accesses the socket the ICMP is just dropped and for some paths there is no
 302  * check at all. A more general error queue to queue errors for later handling
 303  * is probably better.
 304  */
 305 void dccp_v4_err(struct sk_buff *skb, u32 info)
 306 {
 307         const struct iphdr *iph = (struct iphdr *)skb->data;
 308         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 309                                                         (iph->ihl << 2));
 310         struct dccp_sock *dp;
 311         struct inet_sock *inet;
 312         const int type = skb->h.icmph->type;
 313         const int code = skb->h.icmph->code;
 314         struct sock *sk;
 315         __u64 seq;
 316         int err;
 317
 318         if (skb->len < (iph->ihl << 2) + 8) {
 319                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 320                 return;
 321         }
 322
 323         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 324                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 325         if (sk == NULL) {
 326                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 327                 return;
 328         }
 329
 330         if (sk->sk_state == DCCP_TIME_WAIT) {
 331                 inet_twsk_put((struct inet_timewait_sock *)sk);
 332                 return;
 333         }
 334
 335         bh_lock_sock(sk);
 336         /* If too many ICMPs get dropped on busy
 337          * servers this needs to be solved differently.
 338          */
 339         if (sock_owned_by_user(sk))
 340                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 341
 342         if (sk->sk_state == DCCP_CLOSED)
 343                 goto out;
 344
 345         dp = dccp_sk(sk);
 346         seq = dccp_hdr_seq(skb);
 347         if (sk->sk_state != DCCP_LISTEN &&
 348             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 349                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 350                 goto out;
 351         }
 352
 353         switch (type) {
 354         case ICMP_SOURCE_QUENCH:
 355                 /* Just silently ignore these. */
 356                 goto out;
 357         case ICMP_PARAMETERPROB:
 358                 err = EPROTO;
 359                 break;
 360         case ICMP_DEST_UNREACH:
 361                 if (code > NR_ICMP_UNREACH)
 362                         goto out;
 363
 364                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 365                         if (!sock_owned_by_user(sk))
 366                                 dccp_do_pmtu_discovery(sk, iph, info);
 367                         goto out;
 368                 }
 369
 370                 err = icmp_err_convert[code].errno;
 371                 break;
 372         case ICMP_TIME_EXCEEDED:
 373                 err = EHOSTUNREACH;
 374                 break;
 375         default:
 376                 goto out;
 377         }
 378
 379         switch (sk->sk_state) {
 380                 struct request_sock *req , **prev;
 381         case DCCP_LISTEN:
 382                 if (sock_owned_by_user(sk))
 383                         goto out;
 384                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 385                                           iph->daddr, iph->saddr);
 386                 if (!req)
 387                         goto out;
 388
 389                 /*
 390                  * ICMPs are not backlogged, hence we cannot get an established
 391                  * socket here.
 392                  */
 393                 BUG_TRAP(!req->sk);
 394
 395                 if (seq != dccp_rsk(req)->dreq_iss) {
 396                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 397                         goto out;
 398                 }
 399                 /*
 400                  * Still in RESPOND, just remove it silently.
 401                  * There is no good way to pass the error to the newly
 402                  * created socket, and POSIX does not want network
 403                  * errors returned from accept().
 404                  */
 405                 inet_csk_reqsk_queue_drop(sk, req, prev);
 406                 goto out;
 407
 408         case DCCP_REQUESTING:
 409         case DCCP_RESPOND:
 410                 if (!sock_owned_by_user(sk)) {
 411                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 412                         sk->sk_err = err;
 413
 414                         sk->sk_error_report(sk);
 415
 416                         dccp_done(sk);
 417                 } else
 418                         sk->sk_err_soft = err;
 419                 goto out;
 420         }
 421
 422         /* If we've already connected we will keep trying
 423          * until we time out, or the user gives up.
 424          *
 425          * rfc1122 4.2.3.9 allows to consider as hard errors
 426          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 427          * but it is obsoleted by pmtu discovery).
 428          *
 429          * Note, that in modern internet, where routing is unreliable
 430          * and in each dark corner broken firewalls sit, sending random
 431          * errors ordered by their masters even this two messages finally lose
 432          * their original sense (even Linux sends invalid PORT_UNREACHs)
 433          *
 434          * Now we are in compliance with RFCs.
 435          *                                                      --ANK (980905)
 436          */
 437
 438         inet = inet_sk(sk);
 439         if (!sock_owned_by_user(sk) && inet->recverr) {
 440                 sk->sk_err = err;
 441                 sk->sk_error_report(sk);
 442         } else /* Only an error on timeout */
 443                 sk->sk_err_soft = err;
 444 out:
 445         bh_unlock_sock(sk);
 446         sock_put(sk);
 447 }
 448
 449 /* This routine computes an IPv4 DCCP checksum. */
 450 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 451 {
 452         const struct inet_sock *inet = inet_sk(sk);
 453         struct dccp_hdr *dh = dccp_hdr(skb);
 454
 455         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 456 }
 457
 458 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 459
 460 int dccp_v4_send_reset(struct sock *sk, enum dccp_reset_codes code)
 461 {
 462         struct sk_buff *skb;
 463         /*
 464          * FIXME: what if rebuild_header fails?
 465          * Should we be doing a rebuild_header here?
 466          */
 467         int err = inet_sk_rebuild_header(sk);
 468
 469         if (err != 0)
 470                 return err;
 471
 472         skb = dccp_make_reset(sk, sk->sk_dst_cache, code);
 473         if (skb != NULL) {
 474                 const struct inet_sock *inet = inet_sk(sk);
 475
 476                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 477                 err = ip_build_and_send_pkt(skb, sk,
 478                                             inet->saddr, inet->daddr, NULL);
 479                 if (err == NET_XMIT_CN)
 480                         err = 0;
 481         }
 482
 483         return err;
 484 }
 485
 486 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 487                                         const struct sk_buff *skb)
 488 {
 489         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 490                                            skb->nh.iph->saddr,
 491                                            dccp_hdr(skb)->dccph_dport,
 492                                            dccp_hdr(skb)->dccph_sport);
 493 }
 494
 495 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 496 {
 497         struct inet_request_sock *ireq;
 498         struct dccp_sock dp;
 499         struct request_sock *req;
 500         struct dccp_request_sock *dreq;
 501         const __u32 saddr = skb->nh.iph->saddr;
 502         const __u32 daddr = skb->nh.iph->daddr;
 503         const __u32 service = dccp_hdr_request(skb)->dccph_req_service;
 504         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 505         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 506
 507         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 508         if (((struct rtable *)skb->dst)->rt_flags &
 509             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 510                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 511                 goto drop;
 512         }
 513
 514         if (dccp_bad_service_code(sk, service)) {
 515                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 516                 goto drop;
 517         }
 518         /*
 519          * TW buckets are converted to open requests without
 520          * limitations, they conserve resources and peer is
 521          * evidently real one.
 522          */
 523         if (inet_csk_reqsk_queue_is_full(sk))
 524                 goto drop;
 525
 526         /*
 527          * Accept backlog is full. If we have already queued enough
 528          * of warm entries in syn queue, drop request. It is better than
 529          * clogging syn queue with openreqs with exponentially increasing
 530          * timeout.
 531          */
 532         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 533                 goto drop;
 534
 535         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 536         if (req == NULL)
 537                 goto drop;
 538
 539         if (dccp_parse_options(sk, skb))
 540                 goto drop;
 541
 542         dccp_openreq_init(req, &dp, skb);
 543
 544         ireq = inet_rsk(req);
 545         ireq->loc_addr = daddr;
 546         ireq->rmt_addr = saddr;
 547         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 548                                   right value */
 549         ireq->opt       = NULL;
 550
 551         /*
 552          * Step 3: Process LISTEN state
 553          *
 554          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 555          *
 556          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 557          * dccp_create_openreq_child.
 558          */
 559         dreq = dccp_rsk(req);
 560         dreq->dreq_isr     = dcb->dccpd_seq;
 561         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 562         dreq->dreq_service = service;
 563
 564         if (dccp_v4_send_response(sk, req, NULL))
 565                 goto drop_and_free;
 566
 567         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 568         return 0;
 569
 570 drop_and_free:
 571         reqsk_free(req);
 572 drop:
 573         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 574         dcb->dccpd_reset_code = reset_code;
 575         return -1;
 576 }
 577
 578 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 579
 580 /*
 581  * The three way handshake has completed - we got a valid ACK or DATAACK -
 582  * now create the new socket.
 583  *
 584  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 585  */
 586 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 587                                        struct request_sock *req,
 588                                        struct dst_entry *dst)
 589 {
 590         struct inet_request_sock *ireq;
 591         struct inet_sock *newinet;
 592         struct dccp_sock *newdp;
 593         struct sock *newsk;
 594
 595         if (sk_acceptq_is_full(sk))
 596                 goto exit_overflow;
 597
 598         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 599                 goto exit;
 600
 601         newsk = dccp_create_openreq_child(sk, req, skb);
 602         if (newsk == NULL)
 603                 goto exit;
 604
 605         sk_setup_caps(newsk, dst);
 606
 607         newdp              = dccp_sk(newsk);
 608         newinet            = inet_sk(newsk);
 609         ireq               = inet_rsk(req);
 610         newinet->daddr     = ireq->rmt_addr;
 611         newinet->rcv_saddr = ireq->loc_addr;
 612         newinet->saddr     = ireq->loc_addr;
 613         newinet->opt       = ireq->opt;
 614         ireq->opt          = NULL;
 615         newinet->mc_index  = inet_iif(skb);
 616         newinet->mc_ttl    = skb->nh.iph->ttl;
 617         newinet->id        = jiffies;
 618
 619         dccp_sync_mss(newsk, dst_mtu(dst));
 620
 621         __inet_hash(&dccp_hashinfo, newsk, 0);
 622         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 623
 624         return newsk;
 625
 626 exit_overflow:
 627         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 628 exit:
 629         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 630         dst_release(dst);
 631         return NULL;
 632 }
 633
 634 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 635
 636 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 637 {
 638         const struct dccp_hdr *dh = dccp_hdr(skb);
 639         const struct iphdr *iph = skb->nh.iph;
 640         struct sock *nsk;
 641         struct request_sock **prev;
 642         /* Find possible connection requests. */
 643         struct request_sock *req = inet_csk_search_req(sk, &prev,
 644                                                        dh->dccph_sport,
 645                                                        iph->saddr, iph->daddr);
 646         if (req != NULL)
 647                 return dccp_check_req(sk, skb, req, prev);
 648
 649         nsk = __inet_lookup_established(&dccp_hashinfo,
 650                                         iph->saddr, dh->dccph_sport,
 651                                         iph->daddr, ntohs(dh->dccph_dport),
 652                                         inet_iif(skb));
 653         if (nsk != NULL) {
 654                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 655                         bh_lock_sock(nsk);
 656                         return nsk;
 657                 }
 658                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 659                 return NULL;
 660         }
 661
 662         return sk;
 663 }
 664
 665 int dccp_v4_checksum(const struct sk_buff *skb, const u32 saddr,
 666                      const u32 daddr)
 667 {
 668         const struct dccp_hdr* dh = dccp_hdr(skb);
 669         int checksum_len;
 670         u32 tmp;
 671
 672         if (dh->dccph_cscov == 0)
 673                 checksum_len = skb->len;
 674         else {
 675                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 676                 checksum_len = checksum_len < skb->len ? checksum_len :
 677                                                          skb->len;
 678         }
 679
 680         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 681         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 682                                  IPPROTO_DCCP, tmp);
 683 }
 684
 685 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 686                                    const u32 saddr, const u32 daddr)
 687 {
 688         struct dccp_hdr *dh = dccp_hdr(skb);
 689         int checksum_len;
 690         u32 tmp;
 691
 692         if (dh->dccph_cscov == 0)
 693                 checksum_len = skb->len;
 694         else {
 695                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 696                 checksum_len = checksum_len < skb->len ? checksum_len :
 697                                                          skb->len;
 698         }
 699         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 700         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 701                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 702 }
 703
 704 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 705                                            struct sk_buff *skb)
 706 {
 707         struct rtable *rt;
 708         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 709                             .nl_u = { .ip4_u =
 710                                       { .daddr = skb->nh.iph->saddr,
 711                                         .saddr = skb->nh.iph->daddr,
 712                                         .tos = RT_CONN_FLAGS(sk) } },
 713                             .proto = sk->sk_protocol,
 714                             .uli_u = { .ports =
 715                                        { .sport = dccp_hdr(skb)->dccph_dport,
 716                                          .dport = dccp_hdr(skb)->dccph_sport }
 717                                      }
 718                           };
 719
 720         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 721                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 722                 return NULL;
 723         }
 724
 725         return &rt->u.dst;
 726 }
 727
 728 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 729 {
 730         int err;
 731         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 732         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 733                                        sizeof(struct dccp_hdr_ext) +
 734                                        sizeof(struct dccp_hdr_reset);
 735         struct sk_buff *skb;
 736         struct dst_entry *dst;
 737         u64 seqno;
 738
 739         /* Never send a reset in response to a reset. */
 740         if (rxdh->dccph_type == DCCP_PKT_RESET)
 741                 return;
 742
 743         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 744                 return;
 745
 746         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 747         if (dst == NULL)
 748                 return;
 749
 750         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 751         if (skb == NULL)
 752                 goto out;
 753
 754         /* Reserve space for headers. */
 755         skb_reserve(skb, MAX_DCCP_HEADER);
 756         skb->dst = dst_clone(dst);
 757
 758         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 759         dh = dccp_hdr(skb);
 760         memset(dh, 0, dccp_hdr_reset_len);
 761
 762         /* Build DCCP header and checksum it. */
 763         dh->dccph_type     = DCCP_PKT_RESET;
 764         dh->dccph_sport    = rxdh->dccph_dport;
 765         dh->dccph_dport    = rxdh->dccph_sport;
 766         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 767         dh->dccph_x        = 1;
 768         dccp_hdr_reset(skb)->dccph_reset_code =
 769                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 770
 771         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 772         seqno = 0;
 773         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 774                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 775
 776         dccp_hdr_set_seq(dh, seqno);
 777         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 778                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 779
 780         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 781                                               rxskb->nh.iph->daddr);
 782
 783         bh_lock_sock(dccp_ctl_socket->sk);
 784         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 785                                     rxskb->nh.iph->daddr,
 786                                     rxskb->nh.iph->saddr, NULL);
 787         bh_unlock_sock(dccp_ctl_socket->sk);
 788
 789         if (err == NET_XMIT_CN || err == 0) {
 790                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 791                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 792         }
 793 out:
 794          dst_release(dst);
 795 }
 796
 797 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 798 {
 799         struct dccp_hdr *dh = dccp_hdr(skb);
 800
 801         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 802                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 803                         goto reset;
 804                 return 0;
 805         }
 806
 807         /*
 808          *  Step 3: Process LISTEN state
 809          *     If S.state == LISTEN,
 810          *        If P.type == Request or P contains a valid Init Cookie
 811          *              option,
 812          *           * Must scan the packet's options to check for an Init
 813          *              Cookie.  Only the Init Cookie is processed here,
 814          *              however; other options are processed in Step 8.  This
 815          *              scan need only be performed if the endpoint uses Init
 816          *              Cookies *
 817          *           * Generate a new socket and switch to that socket *
 818          *           Set S := new socket for this port pair
 819          *           S.state = RESPOND
 820          *           Choose S.ISS (initial seqno) or set from Init Cookie
 821          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 822          *           Continue with S.state == RESPOND
 823          *           * A Response packet will be generated in Step 11 *
 824          *        Otherwise,
 825          *           Generate Reset(No Connection) unless P.type == Reset
 826          *           Drop packet and return
 827          *
 828          * NOTE: the check for the packet types is done in
 829          *       dccp_rcv_state_process
 830          */
 831         if (sk->sk_state == DCCP_LISTEN) {
 832                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 833
 834                 if (nsk == NULL)
 835                         goto discard;
 836
 837                 if (nsk != sk) {
 838                         if (dccp_child_process(sk, nsk, skb))
 839                                 goto reset;
 840                         return 0;
 841                 }
 842         }
 843
 844         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 845                 goto reset;
 846         return 0;
 847
 848 reset:
 849         dccp_v4_ctl_send_reset(skb);
 850 discard:
 851         kfree_skb(skb);
 852         return 0;
 853 }
 854
 855 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 856
 857 int dccp_invalid_packet(struct sk_buff *skb)
 858 {
 859         const struct dccp_hdr *dh;
 860
 861         if (skb->pkt_type != PACKET_HOST)
 862                 return 1;
 863
 864         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 865                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 866                 return 1;
 867         }
 868
 869         dh = dccp_hdr(skb);
 870
 871         /* If the packet type is not understood, drop packet and return */
 872         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 873                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 874                 return 1;
 875         }
 876
 877         /*
 878          * If P.Data Offset is too small for packet type, or too large for
 879          * packet, drop packet and return
 880          */
 881         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 882                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 883                                             "too small 1\n",
 884                                dh->dccph_doff);
 885                 return 1;
 886         }
 887
 888         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 889                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 890                                             "too small 2\n",
 891                                dh->dccph_doff);
 892                 return 1;
 893         }
 894
 895         dh = dccp_hdr(skb);
 896
 897         /*
 898          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 899          * has short sequence numbers), drop packet and return
 900          */
 901         if (dh->dccph_x == 0 &&
 902             dh->dccph_type != DCCP_PKT_DATA &&
 903             dh->dccph_type != DCCP_PKT_ACK &&
 904             dh->dccph_type != DCCP_PKT_DATAACK) {
 905                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 906                                             "nor DataAck and P.X == 0\n",
 907                                dccp_packet_name(dh->dccph_type));
 908                 return 1;
 909         }
 910
 911         return 0;
 912 }
 913
 914 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 915
 916 /* this is called when real data arrives */
 917 int dccp_v4_rcv(struct sk_buff *skb)
 918 {
 919         const struct dccp_hdr *dh;
 920         struct sock *sk;
 921
 922         /* Step 1: Check header basics: */
 923
 924         if (dccp_invalid_packet(skb))
 925                 goto discard_it;
 926
 927         /* If the header checksum is incorrect, drop packet and return */
 928         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
 929                                     skb->nh.iph->daddr) < 0) {
 930                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
 931                                __FUNCTION__);
 932                 goto discard_it;
 933         }
 934
 935         dh = dccp_hdr(skb);
 936
 937         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 938         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 939
 940         dccp_pr_debug("%8.8s "
 941                       "src=%u.%u.%u.%u@%-5d "
 942                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 943                       dccp_packet_name(dh->dccph_type),
 944                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 945                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 946                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 947
 948         if (dccp_packet_without_ack(skb)) {
 949                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 950                 dccp_pr_debug_cat("\n");
 951         } else {
 952                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 953                 dccp_pr_debug_cat(", ack=%llu\n",
 954                                   (unsigned long long)
 955                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 956         }
 957
 958         /* Step 2:
 959          *      Look up flow ID in table and get corresponding socket */
 960         sk = __inet_lookup(&dccp_hashinfo,
 961                            skb->nh.iph->saddr, dh->dccph_sport,
 962                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
 963                            inet_iif(skb));
 964
 965         /*
 966          * Step 2:
 967          *      If no socket ...
 968          *              Generate Reset(No Connection) unless P.type == Reset
 969          *              Drop packet and return
 970          */
 971         if (sk == NULL) {
 972                 dccp_pr_debug("failed to look up flow ID in table and "
 973                               "get corresponding socket\n");
 974                 goto no_dccp_socket;
 975         }
 976
 977         /*
 978          * Step 2:
 979          *      ... or S.state == TIMEWAIT,
 980          *              Generate Reset(No Connection) unless P.type == Reset
 981          *              Drop packet and return
 982          */
 983
 984         if (sk->sk_state == DCCP_TIME_WAIT) {
 985                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
 986                               "do_time_wait\n");
 987                 goto do_time_wait;
 988         }
 989
 990         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 991                 goto discard_and_relse;
 992         nf_reset(skb);
 993
 994         return sk_receive_skb(sk, skb);
 995
 996 no_dccp_socket:
 997         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 998                 goto discard_it;
 999         /*
1000          * Step 2:
1001          *              Generate Reset(No Connection) unless P.type == Reset
1002          *              Drop packet and return
1003          */
1004         if (dh->dccph_type != DCCP_PKT_RESET) {
1005                 DCCP_SKB_CB(skb)->dccpd_reset_code =
1006                                         DCCP_RESET_CODE_NO_CONNECTION;
1007                 dccp_v4_ctl_send_reset(skb);
1008         }
1009
1010 discard_it:
1011         /* Discard frame. */
1012         kfree_skb(skb);
1013         return 0;
1014
1015 discard_and_relse:
1016         sock_put(sk);
1017         goto discard_it;
1018
1019 do_time_wait:
1020         inet_twsk_put((struct inet_timewait_sock *)sk);
1021         goto no_dccp_socket;
1022 }
1023
1024 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
1025         .queue_xmit     = ip_queue_xmit,
1026         .send_check     = dccp_v4_send_check,
1027         .rebuild_header = inet_sk_rebuild_header,
1028         .conn_request   = dccp_v4_conn_request,
1029         .syn_recv_sock  = dccp_v4_request_recv_sock,
1030         .net_header_len = sizeof(struct iphdr),
1031         .setsockopt     = ip_setsockopt,
1032         .getsockopt     = ip_getsockopt,
1033         .addr2sockaddr  = inet_csk_addr2sockaddr,
1034         .sockaddr_len   = sizeof(struct sockaddr_in),
1035 };
1036
1037 int dccp_v4_init_sock(struct sock *sk)
1038 {
1039         struct dccp_sock *dp = dccp_sk(sk);
1040         struct inet_connection_sock *icsk = inet_csk(sk);
1041         static int dccp_ctl_socket_init = 1;
1042
1043         dccp_options_init(&dp->dccps_options);
1044         do_gettimeofday(&dp->dccps_epoch);
1045
1046         /*
1047          * FIXME: We're hardcoding the CCID, and doing this at this point makes
1048          * the listening (master) sock get CCID control blocks, which is not
1049          * necessary, but for now, to not mess with the test userspace apps,
1050          * lets leave it here, later the real solution is to do this in a
1051          * setsockopt(CCIDs-I-want/accept). -acme
1052          */
1053         if (likely(!dccp_ctl_socket_init)) {
1054                 int rc;
1055
1056                 if (dp->dccps_options.dccpo_send_ack_vector) {
1057                         dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(GFP_KERNEL);
1058                         if (dp->dccps_hc_rx_ackvec == NULL)
1059                                 return -ENOMEM;
1060                 }
1061                 dp->dccps_hc_rx_ccid = ccid_init(dp->dccps_options.dccpo_rx_ccid,
1062                                                  sk);
1063                 dp->dccps_hc_tx_ccid = ccid_init(dp->dccps_options.dccpo_tx_ccid,
1064                                                  sk);
1065                 if (dp->dccps_hc_rx_ccid == NULL ||
1066                     dp->dccps_hc_tx_ccid == NULL) {
1067                         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1068                         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1069                         if (dp->dccps_options.dccpo_send_ack_vector) {
1070                                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1071                                 dp->dccps_hc_rx_ackvec = NULL;
1072                         }
1073                         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1074                         return -ENOMEM;
1075                 }
1076
1077                 rc = dccp_feat_init(sk);
1078                 if (rc)
1079                         return rc;
1080         } else {
1081                 /* control socket doesn't need feat nego */
1082                 INIT_LIST_HEAD(&dp->dccps_options.dccpo_pending);
1083                 INIT_LIST_HEAD(&dp->dccps_options.dccpo_conf);
1084                 dccp_ctl_socket_init = 0;
1085         }
1086
1087         dccp_init_xmit_timers(sk);
1088         icsk->icsk_rto = DCCP_TIMEOUT_INIT;
1089         sk->sk_state = DCCP_CLOSED;
1090         sk->sk_write_space = dccp_write_space;
1091         icsk->icsk_af_ops = &dccp_ipv4_af_ops;
1092         icsk->icsk_sync_mss = dccp_sync_mss;
1093         dp->dccps_mss_cache = 536;
1094         dp->dccps_role = DCCP_ROLE_UNDEFINED;
1095         dp->dccps_service = DCCP_SERVICE_INVALID_VALUE;
1096         dp->dccps_l_ack_ratio = dp->dccps_r_ack_ratio = 1;
1097
1098         return 0;
1099 }
1100
1101 EXPORT_SYMBOL_GPL(dccp_v4_init_sock);
1102
1103 int dccp_v4_destroy_sock(struct sock *sk)
1104 {
1105         struct dccp_sock *dp = dccp_sk(sk);
1106
1107         /*
1108          * DCCP doesn't use sk_write_queue, just sk_send_head
1109          * for retransmissions
1110          */
1111         if (sk->sk_send_head != NULL) {
1112                 kfree_skb(sk->sk_send_head);
1113                 sk->sk_send_head = NULL;
1114         }
1115
1116         /* Clean up a referenced DCCP bind bucket. */
1117         if (inet_csk(sk)->icsk_bind_hash != NULL)
1118                 inet_put_port(&dccp_hashinfo, sk);
1119
1120         kfree(dp->dccps_service_list);
1121         dp->dccps_service_list = NULL;
1122
1123         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
1124         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
1125         if (dp->dccps_options.dccpo_send_ack_vector) {
1126                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1127                 dp->dccps_hc_rx_ackvec = NULL;
1128         }
1129         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1130         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1131         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1132
1133         /* clean up feature negotiation state */
1134         dccp_feat_clean(sk);
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);