net/dccp/ipv4.c

   1 /*
   2  *  net/dccp/ipv4.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/dccp.h>
  14 #include <linux/icmp.h>
  15 #include <linux/slab.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/random.h>
  19
  20 #include <net/icmp.h>
  21 #include <net/inet_common.h>
  22 #include <net/inet_hashtables.h>
  23 #include <net/inet_sock.h>
  24 #include <net/protocol.h>
  25 #include <net/sock.h>
  26 #include <net/timewait_sock.h>
  27 #include <net/tcp_states.h>
  28 #include <net/xfrm.h>
  29 #include <net/secure_seq.h>
  30
  31 #include "ackvec.h"
  32 #include "ccid.h"
  33 #include "dccp.h"
  34 #include "feat.h"
  35
  36 /*
  37  * The per-net dccp.v4_ctl_sk socket is used for responding to
  38  * the Out-of-the-blue (OOTB) packets. A control sock will be created
  39  * for this socket at the initialization time.
  40  */
  41
  42 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  43 {
  44         struct inet_sock *inet = inet_sk(sk);
  45         struct dccp_sock *dp = dccp_sk(sk);
  46         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  47         struct rtable *rt;
  48         __be32 daddr, nexthop;
  49         int tmp;
  50         int err;
  51
  52         dp->dccps_role = DCCP_ROLE_CLIENT;
  53
  54         if (addr_len < sizeof(struct sockaddr_in))
  55                 return -EINVAL;
  56
  57         if (usin->sin_family != AF_INET)
  58                 return -EAFNOSUPPORT;
  59
  60         nexthop = daddr = usin->sin_addr.s_addr;
  61         if (inet->opt != NULL && inet->opt->srr) {
  62                 if (daddr == 0)
  63                         return -EINVAL;
  64                 nexthop = inet->opt->faddr;
  65         }
  66
  67         tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr,
  68                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  69                                IPPROTO_DCCP,
  70                                inet->inet_sport, usin->sin_port, sk, 1);
  71         if (tmp < 0)
  72                 return tmp;
  73
  74         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  75                 ip_rt_put(rt);
  76                 return -ENETUNREACH;
  77         }
  78
  79         if (inet->opt == NULL || !inet->opt->srr)
  80                 daddr = rt->rt_dst;
  81
  82         if (inet->inet_saddr == 0)
  83                 inet->inet_saddr = rt->rt_src;
  84         inet->inet_rcv_saddr = inet->inet_saddr;
  85
  86         inet->inet_dport = usin->sin_port;
  87         inet->inet_daddr = daddr;
  88
  89         inet_csk(sk)->icsk_ext_hdr_len = 0;
  90         if (inet->opt != NULL)
  91                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
  92         /*
  93          * Socket identity is still unknown (sport may be zero).
  94          * However we set state to DCCP_REQUESTING and not releasing socket
  95          * lock select source port, enter ourselves into the hash tables and
  96          * complete initialization after this.
  97          */
  98         dccp_set_state(sk, DCCP_REQUESTING);
  99         err = inet_hash_connect(&dccp_death_row, sk);
 100         if (err != 0)
 101                 goto failure;
 102
 103         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->inet_sport,
 104                                 inet->inet_dport, sk);
 105         if (err != 0)
 106                 goto failure;
 107
 108         /* OK, now commit destination to socket.  */
 109         sk_setup_caps(sk, &rt->dst);
 110
 111         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 112                                                     inet->inet_daddr,
 113                                                     inet->inet_sport,
 114                                                     inet->inet_dport);
 115         inet->inet_id = dp->dccps_iss ^ jiffies;
 116
 117         err = dccp_connect(sk);
 118         rt = NULL;
 119         if (err != 0)
 120                 goto failure;
 121 out:
 122         return err;
 123 failure:
 124         /*
 125          * This unhashes the socket and releases the local port, if necessary.
 126          */
 127         dccp_set_state(sk, DCCP_CLOSED);
 128         ip_rt_put(rt);
 129         sk->sk_route_caps = 0;
 130         inet->inet_dport = 0;
 131         goto out;
 132 }
 133
 134 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 135
 136 /*
 137  * This routine does path mtu discovery as defined in RFC1191.
 138  */
 139 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 140                                           const struct iphdr *iph,
 141                                           u32 mtu)
 142 {
 143         struct dst_entry *dst;
 144         const struct inet_sock *inet = inet_sk(sk);
 145         const struct dccp_sock *dp = dccp_sk(sk);
 146
 147         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 148          * send out by Linux are always < 576bytes so they should go through
 149          * unfragmented).
 150          */
 151         if (sk->sk_state == DCCP_LISTEN)
 152                 return;
 153
 154         /* We don't check in the destentry if pmtu discovery is forbidden
 155          * on this route. We just assume that no packet_to_big packets
 156          * are send back when pmtu discovery is not active.
 157          * There is a small race when the user changes this flag in the
 158          * route, but I think that's acceptable.
 159          */
 160         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 161                 return;
 162
 163         dst->ops->update_pmtu(dst, mtu);
 164
 165         /* Something is about to be wrong... Remember soft error
 166          * for the case, if this connection will not able to recover.
 167          */
 168         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 169                 sk->sk_err_soft = EMSGSIZE;
 170
 171         mtu = dst_mtu(dst);
 172
 173         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 174             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 175                 dccp_sync_mss(sk, mtu);
 176
 177                 /*
 178                  * From RFC 4340, sec. 14.1:
 179                  *
 180                  *      DCCP-Sync packets are the best choice for upward
 181                  *      probing, since DCCP-Sync probes do not risk application
 182                  *      data loss.
 183                  */
 184                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 185         } /* else let the usual retransmit timer handle it */
 186 }
 187
 188 /*
 189  * This routine is called by the ICMP module when it gets some sort of error
 190  * condition. If err < 0 then the socket should be closed and the error
 191  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 192  * After adjustment header points to the first 8 bytes of the tcp header. We
 193  * need to find the appropriate port.
 194  *
 195  * The locking strategy used here is very "optimistic". When someone else
 196  * accesses the socket the ICMP is just dropped and for some paths there is no
 197  * check at all. A more general error queue to queue errors for later handling
 198  * is probably better.
 199  */
 200 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 201 {
 202         const struct iphdr *iph = (struct iphdr *)skb->data;
 203         const u8 offset = iph->ihl << 2;
 204         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
 205         struct dccp_sock *dp;
 206         struct inet_sock *inet;
 207         const int type = icmp_hdr(skb)->type;
 208         const int code = icmp_hdr(skb)->code;
 209         struct sock *sk;
 210         __u64 seq;
 211         int err;
 212         struct net *net = dev_net(skb->dev);
 213
 214         if (skb->len < offset + sizeof(*dh) ||
 215             skb->len < offset + __dccp_basic_hdr_len(dh)) {
 216                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 217                 return;
 218         }
 219
 220         sk = inet_lookup(net, &dccp_hashinfo,
 221                         iph->daddr, dh->dccph_dport,
 222                         iph->saddr, dh->dccph_sport, inet_iif(skb));
 223         if (sk == NULL) {
 224                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 225                 return;
 226         }
 227
 228         if (sk->sk_state == DCCP_TIME_WAIT) {
 229                 inet_twsk_put(inet_twsk(sk));
 230                 return;
 231         }
 232
 233         bh_lock_sock(sk);
 234         /* If too many ICMPs get dropped on busy
 235          * servers this needs to be solved differently.
 236          */
 237         if (sock_owned_by_user(sk))
 238                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 239
 240         if (sk->sk_state == DCCP_CLOSED)
 241                 goto out;
 242
 243         dp = dccp_sk(sk);
 244         seq = dccp_hdr_seq(dh);
 245         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 246             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 247                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 248                 goto out;
 249         }
 250
 251         switch (type) {
 252         case ICMP_SOURCE_QUENCH:
 253                 /* Just silently ignore these. */
 254                 goto out;
 255         case ICMP_PARAMETERPROB:
 256                 err = EPROTO;
 257                 break;
 258         case ICMP_DEST_UNREACH:
 259                 if (code > NR_ICMP_UNREACH)
 260                         goto out;
 261
 262                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 263                         if (!sock_owned_by_user(sk))
 264                                 dccp_do_pmtu_discovery(sk, iph, info);
 265                         goto out;
 266                 }
 267
 268                 err = icmp_err_convert[code].errno;
 269                 break;
 270         case ICMP_TIME_EXCEEDED:
 271                 err = EHOSTUNREACH;
 272                 break;
 273         default:
 274                 goto out;
 275         }
 276
 277         switch (sk->sk_state) {
 278                 struct request_sock *req , **prev;
 279         case DCCP_LISTEN:
 280                 if (sock_owned_by_user(sk))
 281                         goto out;
 282                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 283                                           iph->daddr, iph->saddr);
 284                 if (!req)
 285                         goto out;
 286
 287                 /*
 288                  * ICMPs are not backlogged, hence we cannot get an established
 289                  * socket here.
 290                  */
 291                 WARN_ON(req->sk);
 292
 293                 if (seq != dccp_rsk(req)->dreq_iss) {
 294                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 295                         goto out;
 296                 }
 297                 /*
 298                  * Still in RESPOND, just remove it silently.
 299                  * There is no good way to pass the error to the newly
 300                  * created socket, and POSIX does not want network
 301                  * errors returned from accept().
 302                  */
 303                 inet_csk_reqsk_queue_drop(sk, req, prev);
 304                 goto out;
 305
 306         case DCCP_REQUESTING:
 307         case DCCP_RESPOND:
 308                 if (!sock_owned_by_user(sk)) {
 309                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 310                         sk->sk_err = err;
 311
 312                         sk->sk_error_report(sk);
 313
 314                         dccp_done(sk);
 315                 } else
 316                         sk->sk_err_soft = err;
 317                 goto out;
 318         }
 319
 320         /* If we've already connected we will keep trying
 321          * until we time out, or the user gives up.
 322          *
 323          * rfc1122 4.2.3.9 allows to consider as hard errors
 324          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 325          * but it is obsoleted by pmtu discovery).
 326          *
 327          * Note, that in modern internet, where routing is unreliable
 328          * and in each dark corner broken firewalls sit, sending random
 329          * errors ordered by their masters even this two messages finally lose
 330          * their original sense (even Linux sends invalid PORT_UNREACHs)
 331          *
 332          * Now we are in compliance with RFCs.
 333          *                                                      --ANK (980905)
 334          */
 335
 336         inet = inet_sk(sk);
 337         if (!sock_owned_by_user(sk) && inet->recverr) {
 338                 sk->sk_err = err;
 339                 sk->sk_error_report(sk);
 340         } else /* Only an error on timeout */
 341                 sk->sk_err_soft = err;
 342 out:
 343         bh_unlock_sock(sk);
 344         sock_put(sk);
 345 }
 346
 347 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 348                                       __be32 src, __be32 dst)
 349 {
 350         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 351 }
 352
 353 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 354 {
 355         const struct inet_sock *inet = inet_sk(sk);
 356         struct dccp_hdr *dh = dccp_hdr(skb);
 357
 358         dccp_csum_outgoing(skb);
 359         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 360                                                  inet->inet_saddr,
 361                                                  inet->inet_daddr);
 362 }
 363
 364 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 365
 366 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 367 {
 368         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 369                                            ip_hdr(skb)->saddr,
 370                                            dccp_hdr(skb)->dccph_dport,
 371                                            dccp_hdr(skb)->dccph_sport);
 372 }
 373
 374 /*
 375  * The three way handshake has completed - we got a valid ACK or DATAACK -
 376  * now create the new socket.
 377  *
 378  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 379  */
 380 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 381                                        struct request_sock *req,
 382                                        struct dst_entry *dst)
 383 {
 384         struct inet_request_sock *ireq;
 385         struct inet_sock *newinet;
 386         struct sock *newsk;
 387
 388         if (sk_acceptq_is_full(sk))
 389                 goto exit_overflow;
 390
 391         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 392                 goto exit;
 393
 394         newsk = dccp_create_openreq_child(sk, req, skb);
 395         if (newsk == NULL)
 396                 goto exit_nonewsk;
 397
 398         sk_setup_caps(newsk, dst);
 399
 400         newinet            = inet_sk(newsk);
 401         ireq               = inet_rsk(req);
 402         newinet->inet_daddr     = ireq->rmt_addr;
 403         newinet->inet_rcv_saddr = ireq->loc_addr;
 404         newinet->inet_saddr     = ireq->loc_addr;
 405         newinet->opt       = ireq->opt;
 406         ireq->opt          = NULL;
 407         newinet->mc_index  = inet_iif(skb);
 408         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 409         newinet->inet_id   = jiffies;
 410
 411         dccp_sync_mss(newsk, dst_mtu(dst));
 412
 413         if (__inet_inherit_port(sk, newsk) < 0) {
 414                 sock_put(newsk);
 415                 goto exit;
 416         }
 417         __inet_hash_nolisten(newsk, NULL);
 418
 419         return newsk;
 420
 421 exit_overflow:
 422         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 423 exit_nonewsk:
 424         dst_release(dst);
 425 exit:
 426         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 427         return NULL;
 428 }
 429
 430 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 431
 432 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 433 {
 434         const struct dccp_hdr *dh = dccp_hdr(skb);
 435         const struct iphdr *iph = ip_hdr(skb);
 436         struct sock *nsk;
 437         struct request_sock **prev;
 438         /* Find possible connection requests. */
 439         struct request_sock *req = inet_csk_search_req(sk, &prev,
 440                                                        dh->dccph_sport,
 441                                                        iph->saddr, iph->daddr);
 442         if (req != NULL)
 443                 return dccp_check_req(sk, skb, req, prev);
 444
 445         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 446                                       iph->saddr, dh->dccph_sport,
 447                                       iph->daddr, dh->dccph_dport,
 448                                       inet_iif(skb));
 449         if (nsk != NULL) {
 450                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 451                         bh_lock_sock(nsk);
 452                         return nsk;
 453                 }
 454                 inet_twsk_put(inet_twsk(nsk));
 455                 return NULL;
 456         }
 457
 458         return sk;
 459 }
 460
 461 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 462                                            struct sk_buff *skb)
 463 {
 464         struct rtable *rt;
 465         struct flowi fl = { .oif = skb_rtable(skb)->rt_iif,
 466                             .fl4_dst = ip_hdr(skb)->saddr,
 467                             .fl4_src = ip_hdr(skb)->daddr,
 468                             .fl4_tos = RT_CONN_FLAGS(sk),
 469                             .proto = sk->sk_protocol,
 470                             .fl_ip_sport = dccp_hdr(skb)->dccph_dport,
 471                             .fl_ip_dport = dccp_hdr(skb)->dccph_sport
 472                           };
 473
 474         security_skb_classify_flow(skb, &fl);
 475         if (ip_route_output_flow(net, &rt, &fl, sk, 0)) {
 476                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 477                 return NULL;
 478         }
 479
 480         return &rt->dst;
 481 }
 482
 483 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 484                                  struct request_values *rv_unused)
 485 {
 486         int err = -1;
 487         struct sk_buff *skb;
 488         struct dst_entry *dst;
 489
 490         dst = inet_csk_route_req(sk, req);
 491         if (dst == NULL)
 492                 goto out;
 493
 494         skb = dccp_make_response(sk, dst, req);
 495         if (skb != NULL) {
 496                 const struct inet_request_sock *ireq = inet_rsk(req);
 497                 struct dccp_hdr *dh = dccp_hdr(skb);
 498
 499                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
 500                                                               ireq->rmt_addr);
 501                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 502                                             ireq->rmt_addr,
 503                                             ireq->opt);
 504                 err = net_xmit_eval(err);
 505         }
 506
 507 out:
 508         dst_release(dst);
 509         return err;
 510 }
 511
 512 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
 513 {
 514         int err;
 515         const struct iphdr *rxiph;
 516         struct sk_buff *skb;
 517         struct dst_entry *dst;
 518         struct net *net = dev_net(skb_dst(rxskb)->dev);
 519         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 520
 521         /* Never send a reset in response to a reset. */
 522         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 523                 return;
 524
 525         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 526                 return;
 527
 528         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 529         if (dst == NULL)
 530                 return;
 531
 532         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 533         if (skb == NULL)
 534                 goto out;
 535
 536         rxiph = ip_hdr(rxskb);
 537         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 538                                                                  rxiph->daddr);
 539         skb_dst_set(skb, dst_clone(dst));
 540
 541         bh_lock_sock(ctl_sk);
 542         err = ip_build_and_send_pkt(skb, ctl_sk,
 543                                     rxiph->daddr, rxiph->saddr, NULL);
 544         bh_unlock_sock(ctl_sk);
 545
 546         if (net_xmit_eval(err) == 0) {
 547                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 548                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 549         }
 550 out:
 551          dst_release(dst);
 552 }
 553
 554 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 555 {
 556         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 557         kfree(inet_rsk(req)->opt);
 558 }
 559
 560 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 561         .family         = PF_INET,
 562         .obj_size       = sizeof(struct dccp_request_sock),
 563         .rtx_syn_ack    = dccp_v4_send_response,
 564         .send_ack       = dccp_reqsk_send_ack,
 565         .destructor     = dccp_v4_reqsk_destructor,
 566         .send_reset     = dccp_v4_ctl_send_reset,
 567 };
 568
 569 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 570 {
 571         struct inet_request_sock *ireq;
 572         struct request_sock *req;
 573         struct dccp_request_sock *dreq;
 574         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 575         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 576
 577         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 578         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 579                 return 0;       /* discard, don't send a reset here */
 580
 581         if (dccp_bad_service_code(sk, service)) {
 582                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 583                 goto drop;
 584         }
 585         /*
 586          * TW buckets are converted to open requests without
 587          * limitations, they conserve resources and peer is
 588          * evidently real one.
 589          */
 590         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 591         if (inet_csk_reqsk_queue_is_full(sk))
 592                 goto drop;
 593
 594         /*
 595          * Accept backlog is full. If we have already queued enough
 596          * of warm entries in syn queue, drop request. It is better than
 597          * clogging syn queue with openreqs with exponentially increasing
 598          * timeout.
 599          */
 600         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 601                 goto drop;
 602
 603         req = inet_reqsk_alloc(&dccp_request_sock_ops);
 604         if (req == NULL)
 605                 goto drop;
 606
 607         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 608                 goto drop_and_free;
 609
 610         dreq = dccp_rsk(req);
 611         if (dccp_parse_options(sk, dreq, skb))
 612                 goto drop_and_free;
 613
 614         if (security_inet_conn_request(sk, skb, req))
 615                 goto drop_and_free;
 616
 617         ireq = inet_rsk(req);
 618         ireq->loc_addr = ip_hdr(skb)->daddr;
 619         ireq->rmt_addr = ip_hdr(skb)->saddr;
 620
 621         /*
 622          * Step 3: Process LISTEN state
 623          *
 624          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 625          *
 626          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 627          * dccp_create_openreq_child.
 628          */
 629         dreq->dreq_isr     = dcb->dccpd_seq;
 630         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 631         dreq->dreq_service = service;
 632
 633         if (dccp_v4_send_response(sk, req, NULL))
 634                 goto drop_and_free;
 635
 636         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 637         return 0;
 638
 639 drop_and_free:
 640         reqsk_free(req);
 641 drop:
 642         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 643         return -1;
 644 }
 645
 646 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 647
 648 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 649 {
 650         struct dccp_hdr *dh = dccp_hdr(skb);
 651
 652         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 653                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 654                         goto reset;
 655                 return 0;
 656         }
 657
 658         /*
 659          *  Step 3: Process LISTEN state
 660          *       If P.type == Request or P contains a valid Init Cookie option,
 661          *            (* Must scan the packet's options to check for Init
 662          *               Cookies.  Only Init Cookies are processed here,
 663          *               however; other options are processed in Step 8.  This
 664          *               scan need only be performed if the endpoint uses Init
 665          *               Cookies *)
 666          *            (* Generate a new socket and switch to that socket *)
 667          *            Set S := new socket for this port pair
 668          *            S.state = RESPOND
 669          *            Choose S.ISS (initial seqno) or set from Init Cookies
 670          *            Initialize S.GAR := S.ISS
 671          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 672          *            Continue with S.state == RESPOND
 673          *            (* A Response packet will be generated in Step 11 *)
 674          *       Otherwise,
 675          *            Generate Reset(No Connection) unless P.type == Reset
 676          *            Drop packet and return
 677          *
 678          * NOTE: the check for the packet types is done in
 679          *       dccp_rcv_state_process
 680          */
 681         if (sk->sk_state == DCCP_LISTEN) {
 682                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 683
 684                 if (nsk == NULL)
 685                         goto discard;
 686
 687                 if (nsk != sk) {
 688                         if (dccp_child_process(sk, nsk, skb))
 689                                 goto reset;
 690                         return 0;
 691                 }
 692         }
 693
 694         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 695                 goto reset;
 696         return 0;
 697
 698 reset:
 699         dccp_v4_ctl_send_reset(sk, skb);
 700 discard:
 701         kfree_skb(skb);
 702         return 0;
 703 }
 704
 705 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 706
 707 /**
 708  *      dccp_invalid_packet  -  check for malformed packets
 709  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 710  *      Packets that fail these checks are ignored and do not receive Resets.
 711  */
 712 int dccp_invalid_packet(struct sk_buff *skb)
 713 {
 714         const struct dccp_hdr *dh;
 715         unsigned int cscov;
 716
 717         if (skb->pkt_type != PACKET_HOST)
 718                 return 1;
 719
 720         /* If the packet is shorter than 12 bytes, drop packet and return */
 721         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 722                 DCCP_WARN("pskb_may_pull failed\n");
 723                 return 1;
 724         }
 725
 726         dh = dccp_hdr(skb);
 727
 728         /* If P.type is not understood, drop packet and return */
 729         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 730                 DCCP_WARN("invalid packet type\n");
 731                 return 1;
 732         }
 733
 734         /*
 735          * If P.Data Offset is too small for packet type, drop packet and return
 736          */
 737         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 738                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 739                 return 1;
 740         }
 741         /*
 742          * If P.Data Offset is too too large for packet, drop packet and return
 743          */
 744         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 745                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 746                 return 1;
 747         }
 748
 749         /*
 750          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 751          * has short sequence numbers), drop packet and return
 752          */
 753         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 754             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 755                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 756                           dccp_packet_name(dh->dccph_type));
 757                 return 1;
 758         }
 759
 760         /*
 761          * If P.CsCov is too large for the packet size, drop packet and return.
 762          * This must come _before_ checksumming (not as RFC 4340 suggests).
 763          */
 764         cscov = dccp_csum_coverage(skb);
 765         if (cscov > skb->len) {
 766                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 767                           dh->dccph_cscov, skb->len);
 768                 return 1;
 769         }
 770
 771         /* If header checksum is incorrect, drop packet and return.
 772          * (This step is completed in the AF-dependent functions.) */
 773         skb->csum = skb_checksum(skb, 0, cscov, 0);
 774
 775         return 0;
 776 }
 777
 778 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 779
 780 /* this is called when real data arrives */
 781 static int dccp_v4_rcv(struct sk_buff *skb)
 782 {
 783         const struct dccp_hdr *dh;
 784         const struct iphdr *iph;
 785         struct sock *sk;
 786         int min_cov;
 787
 788         /* Step 1: Check header basics */
 789
 790         if (dccp_invalid_packet(skb))
 791                 goto discard_it;
 792
 793         iph = ip_hdr(skb);
 794         /* Step 1: If header checksum is incorrect, drop packet and return */
 795         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 796                 DCCP_WARN("dropped packet with invalid checksum\n");
 797                 goto discard_it;
 798         }
 799
 800         dh = dccp_hdr(skb);
 801
 802         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 803         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 804
 805         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 806                       dccp_packet_name(dh->dccph_type),
 807                       &iph->saddr, ntohs(dh->dccph_sport),
 808                       &iph->daddr, ntohs(dh->dccph_dport),
 809                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 810
 811         if (dccp_packet_without_ack(skb)) {
 812                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 813                 dccp_pr_debug_cat("\n");
 814         } else {
 815                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 816                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 817                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 818         }
 819
 820         /* Step 2:
 821          *      Look up flow ID in table and get corresponding socket */
 822         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 823                                dh->dccph_sport, dh->dccph_dport);
 824         /*
 825          * Step 2:
 826          *      If no socket ...
 827          */
 828         if (sk == NULL) {
 829                 dccp_pr_debug("failed to look up flow ID in table and "
 830                               "get corresponding socket\n");
 831                 goto no_dccp_socket;
 832         }
 833
 834         /*
 835          * Step 2:
 836          *      ... or S.state == TIMEWAIT,
 837          *              Generate Reset(No Connection) unless P.type == Reset
 838          *              Drop packet and return
 839          */
 840         if (sk->sk_state == DCCP_TIME_WAIT) {
 841                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 842                 inet_twsk_put(inet_twsk(sk));
 843                 goto no_dccp_socket;
 844         }
 845
 846         /*
 847          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 848          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 849          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 850          */
 851         min_cov = dccp_sk(sk)->dccps_pcrlen;
 852         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 853                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 854                               dh->dccph_cscov, min_cov);
 855                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 856                  *         options (Section 11.7) with Drop Code 0, Protocol
 857                  *         Constraints."                                     */
 858                 goto discard_and_relse;
 859         }
 860
 861         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 862                 goto discard_and_relse;
 863         nf_reset(skb);
 864
 865         return sk_receive_skb(sk, skb, 1);
 866
 867 no_dccp_socket:
 868         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 869                 goto discard_it;
 870         /*
 871          * Step 2:
 872          *      If no socket ...
 873          *              Generate Reset(No Connection) unless P.type == Reset
 874          *              Drop packet and return
 875          */
 876         if (dh->dccph_type != DCCP_PKT_RESET) {
 877                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 878                                         DCCP_RESET_CODE_NO_CONNECTION;
 879                 dccp_v4_ctl_send_reset(sk, skb);
 880         }
 881
 882 discard_it:
 883         kfree_skb(skb);
 884         return 0;
 885
 886 discard_and_relse:
 887         sock_put(sk);
 888         goto discard_it;
 889 }
 890
 891 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 892         .queue_xmit        = ip_queue_xmit,
 893         .send_check        = dccp_v4_send_check,
 894         .rebuild_header    = inet_sk_rebuild_header,
 895         .conn_request      = dccp_v4_conn_request,
 896         .syn_recv_sock     = dccp_v4_request_recv_sock,
 897         .net_header_len    = sizeof(struct iphdr),
 898         .setsockopt        = ip_setsockopt,
 899         .getsockopt        = ip_getsockopt,
 900         .addr2sockaddr     = inet_csk_addr2sockaddr,
 901         .sockaddr_len      = sizeof(struct sockaddr_in),
 902         .bind_conflict     = inet_csk_bind_conflict,
 903 #ifdef CONFIG_COMPAT
 904         .compat_setsockopt = compat_ip_setsockopt,
 905         .compat_getsockopt = compat_ip_getsockopt,
 906 #endif
 907 };
 908
 909 static int dccp_v4_init_sock(struct sock *sk)
 910 {
 911         static __u8 dccp_v4_ctl_sock_initialized;
 912         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 913
 914         if (err == 0) {
 915                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 916                         dccp_v4_ctl_sock_initialized = 1;
 917                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 918         }
 919
 920         return err;
 921 }
 922
 923 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 924         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 925 };
 926
 927 static struct proto dccp_v4_prot = {
 928         .name                   = "DCCP",
 929         .owner                  = THIS_MODULE,
 930         .close                  = dccp_close,
 931         .connect                = dccp_v4_connect,
 932         .disconnect             = dccp_disconnect,
 933         .ioctl                  = dccp_ioctl,
 934         .init                   = dccp_v4_init_sock,
 935         .setsockopt             = dccp_setsockopt,
 936         .getsockopt             = dccp_getsockopt,
 937         .sendmsg                = dccp_sendmsg,
 938         .recvmsg                = dccp_recvmsg,
 939         .backlog_rcv            = dccp_v4_do_rcv,
 940         .hash                   = inet_hash,
 941         .unhash                 = inet_unhash,
 942         .accept                 = inet_csk_accept,
 943         .get_port               = inet_csk_get_port,
 944         .shutdown               = dccp_shutdown,
 945         .destroy                = dccp_destroy_sock,
 946         .orphan_count           = &dccp_orphan_count,
 947         .max_header             = MAX_DCCP_HEADER,
 948         .obj_size               = sizeof(struct dccp_sock),
 949         .slab_flags             = SLAB_DESTROY_BY_RCU,
 950         .rsk_prot               = &dccp_request_sock_ops,
 951         .twsk_prot              = &dccp_timewait_sock_ops,
 952         .h.hashinfo             = &dccp_hashinfo,
 953 #ifdef CONFIG_COMPAT
 954         .compat_setsockopt      = compat_dccp_setsockopt,
 955         .compat_getsockopt      = compat_dccp_getsockopt,
 956 #endif
 957 };
 958
 959 static const struct net_protocol dccp_v4_protocol = {
 960         .handler        = dccp_v4_rcv,
 961         .err_handler    = dccp_v4_err,
 962         .no_policy      = 1,
 963         .netns_ok       = 1,
 964 };
 965
 966 static const struct proto_ops inet_dccp_ops = {
 967         .family            = PF_INET,
 968         .owner             = THIS_MODULE,
 969         .release           = inet_release,
 970         .bind              = inet_bind,
 971         .connect           = inet_stream_connect,
 972         .socketpair        = sock_no_socketpair,
 973         .accept            = inet_accept,
 974         .getname           = inet_getname,
 975         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 976         .poll              = dccp_poll,
 977         .ioctl             = inet_ioctl,
 978         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 979         .listen            = inet_dccp_listen,
 980         .shutdown          = inet_shutdown,
 981         .setsockopt        = sock_common_setsockopt,
 982         .getsockopt        = sock_common_getsockopt,
 983         .sendmsg           = inet_sendmsg,
 984         .recvmsg           = sock_common_recvmsg,
 985         .mmap              = sock_no_mmap,
 986         .sendpage          = sock_no_sendpage,
 987 #ifdef CONFIG_COMPAT
 988         .compat_setsockopt = compat_sock_common_setsockopt,
 989         .compat_getsockopt = compat_sock_common_getsockopt,
 990 #endif
 991 };
 992
 993 static struct inet_protosw dccp_v4_protosw = {
 994         .type           = SOCK_DCCP,
 995         .protocol       = IPPROTO_DCCP,
 996         .prot           = &dccp_v4_prot,
 997         .ops            = &inet_dccp_ops,
 998         .no_check       = 0,
 999         .flags          = INET_PROTOSW_ICSK,
1000 };
1001
1002 static int __net_init dccp_v4_init_net(struct net *net)
1003 {
1004         if (dccp_hashinfo.bhash == NULL)
1005                 return -ESOCKTNOSUPPORT;
1006
1007         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1008                                     SOCK_DCCP, IPPROTO_DCCP, net);
1009 }
1010
1011 static void __net_exit dccp_v4_exit_net(struct net *net)
1012 {
1013         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1014 }
1015
1016 static struct pernet_operations dccp_v4_ops = {
1017         .init   = dccp_v4_init_net,
1018         .exit   = dccp_v4_exit_net,
1019 };
1020
1021 static int __init dccp_v4_init(void)
1022 {
1023         int err = proto_register(&dccp_v4_prot, 1);
1024
1025         if (err != 0)
1026                 goto out;
1027
1028         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1029         if (err != 0)
1030                 goto out_proto_unregister;
1031
1032         inet_register_protosw(&dccp_v4_protosw);
1033
1034         err = register_pernet_subsys(&dccp_v4_ops);
1035         if (err)
1036                 goto out_destroy_ctl_sock;
1037 out:
1038         return err;
1039 out_destroy_ctl_sock:
1040         inet_unregister_protosw(&dccp_v4_protosw);
1041         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1042 out_proto_unregister:
1043         proto_unregister(&dccp_v4_prot);
1044         goto out;
1045 }
1046
1047 static void __exit dccp_v4_exit(void)
1048 {
1049         unregister_pernet_subsys(&dccp_v4_ops);
1050         inet_unregister_protosw(&dccp_v4_protosw);
1051         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1052         proto_unregister(&dccp_v4_prot);
1053 }
1054
1055 module_init(dccp_v4_init);
1056 module_exit(dccp_v4_exit);
1057
1058 /*
1059  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1060  * values directly, Also cover the case where the protocol is not specified,
1061  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1062  */
1063 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1064 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1065 MODULE_LICENSE("GPL");
1066 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1067 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");